Phenformin-Induced Mitochondrial Dysfunction Sensitizes Hepatocellular Carcinoma for Dual Inhibition of mTOR Free

Worldwide, liver cancer is the second cause of cancer-related death, second in most years of lost life, and the most rapidly expanding malignant tumor in Western countries. Accounting for >70% of the total liver cancer burden, hepatocellular carcinoma (HCC) is its major histologic subtype. The most common liver insults leading to HCC are hepatitis B or C virus (HBV or HCV) and alcohol (1); however, in Western societies, the abrupt increase in HCC has been linked to nonalcoholic steatohepatitis (NASH) and the recent obesity epidemic (2, 3). Despite being a very heterogeneous tumor type without clear genetic drivers, the most aggressive forms of HCC are associated with the TGFβ1, PI3K/Akt/mTOR, Wnt/β-catenin, and NOTCH proliferative pathways (4, 5). The Cancer Genome Atlas (TCGA) consortium has recently performed the first large-scale multi-platform analysis of HCC, which showed a prevalence of alterations in WNT and p53 signaling as well as TERT upregulation (6). Clinically, surgical resection leads to approximately 70% disease-free survival over 5 years in HCC patients (7, 8), although it is only an option for a small portion of early-stage patients with solitary tumors and normal liver function. For advanced stage HCC, the standard of care is sorafenib (9), followed upon progression by the recently approved regorafenib, both multi-protein kinase inhibitors, which render modest life-extending benefits (10, 11). Recently, striking effects have been reported with nivolumab, an immunotherapeutic agent (12). Of note, such treatments may benefit from combination therapy with targeting agents (13).

mTOR emerged several years ago as an attractive target for HCC, as it is hyperactivated in 50%–60% of HCC and is associated with high-grade tumors and worst case prognosis (14–17). Blocking tumor growth by inhibiting mTOR has led to the approval of the rapamycin family of allosteric drugs (rapalogs) in the treatment of a number of cancers (18, 19). Nevertheless, the rapalog everolimus (RAD001) did not improve overall survival in the phase III EVOLVE-1 trial for advanced HCC after sorafenib failure (20). However, a recent retrospective analysis of the EVOLVE-1 data showed that patients exhibiting mutations in the Tuberous Sclerosis 2 (TSC2) protein, a tumor suppressor, which results in constitutive activation of mTOR complex 1 (mTORC1), tended to benefit from second-line treatment with RAD001 (21). Importantly, the same authors reported that loss of TSC2 is observed in approximately 20% of the Asian HBV(+) HCC patients (21). These findings have further underscored the necessity of stratifying patients as a function of subpopulations in liver cancer clinical trial designs (22).

The failure of rapalogs in a number of clinical trials may also be attributed in part to their incomplete inhibition of mTORC1 and mTORC2 (23), as well as the loss of the negative feedback loop from ribosomal protein S6K1 kinase to Akt (24, 25). We recently showed that these two shortcomings could be overcome by combining an allosteric (RAD001) and an ATP-site competitive (BEZ235) mTOR inhibitor (26). These observations have led to the proposal that the rapalogs should be reconsidered in the treatment of HCC. However, in the clinic mTOR inhibitors hinder glycolysis, inducing hyperglycemia (27), which may independently drive tumor growth (28, 29). Moreover, it was recently demonstrated that mTOR inhibitors decrease mitochondrial biogenesis and activity, but counterintuitively increase mitochondrial efficiency (30). This overall increase in mitochondrial efficiency may provide cancer cells the potential to survive the catabolic state induced by mTOR inhibitors, particularly a small population of stem-like cells. In the clinic, hyperglycemia induced by rapalogs is treated with the biguanide metformin (31, 32). Given that biguanides inhibit complex I (CI) and mTOR inhibitors suppress glycolysis, we reasoned that their combination or their sequential use might place a further energetic stress on HCCs. However, metformin requires the OCT1 transporter to enter tumor cells, which is largely downregulated in HCC (33), making the hydrophobic analogue phenformin an attractive alternative, as it does not require a transporter and is readily taken up by all tissues (34). Phenformin was initially withdrawn from clinical use due to deleterious lactic acidosis in a subset of renal patients (35). However, arguments that phenformin could be considered as an anticancer agent, given distinct dosing and usage (36), have aided in the recent approval of a phase I clinical trial for its use in melanoma (37).

Here we set out to determine the effects of RAD001/BEZ235 and phenformin treatment in human HCC cell lines, tumoroids, and mouse orthoxenografts. In cells, we found that both treatments inhibited mTOR signaling, but that mTOR inhibition was not augmented by the triple combination. In parallel, phenformin induced mitochondrial dysfunction and fragmentation, consistent with the activation of AMP-activated protein kinase (AMPK; ref. 38), and led to a metabolic shift toward glycolysis, increased ROS production, and cell death. Unexpectedly, a number of the phenformin responses appeared to be mitigated by the energy-sparing effects of RAD001/BEZ235 treatment. The growth inhibition of orthotopically implanted tumors was more dramatic when using mTOR inhibitors, as compared with phenformin alone, but more penetrant with the triple drug combination, an effect also observed in overall survival. Given that the mTOR inhibitors suppressed the effects of phenformin on mitochondria homeostasis in vivo, HCC-bearing mice were first treated with the biguanide, followed by the mTOR inhibitors. Such a treatment schedule dramatically sensitized tumors to mTOR inhibitors, profoundly increasing overall survival. Given the recent approval of phenformin IND in melanoma (37), this regimen should be explored clinically in HCC.

Huh7 cells, purchased from the JCRB Cell Bank, were maintained in DMEM (Gibco) containing 10% heat-inactivated FBS (Sigma-Aldrich) and 1% penicillin/streptomycin (10,000 U/mL and 10 mg/mL, respectively; Sigma-Aldrich). Cell number was measured by crystal violet staining (Sigma-Aldrich).

Four-week-old male Athymic Nude-Foxn1^nu mice were purchased from Harlan Laboratories (Envigo). Six- to 8-week-old mice were injected intrahepatically with 1–2 × 10⁶ Huh7 cells in an ice-cold PBS (Gibco)/Matrigel growth factor–reduced phenol-free (Corning, BD Biosciences) solution. After approximately four weeks, tumors were collected and equal size tumor pieces (∼0.05 g) were implanted surgically in the liver of experimental athymic mice cohorts. Two weeks after tumor implantation, mice were treated orally daily for 15 or 21 days with RAD001 (2.5 mg/kg) and/or BEZ235 (6 mg/kg) and/or INK128 (0.3 mg/kg), and intraperitoneally with the specified doses of phenformin (70 or 100 mg/kg). The mice were starved for 2 hours before sacrifice and received their last dose of drug 1 hour before sacrifice. Liver and tumor tissues were snap-frozen in liquid nitrogen for protein extraction or fixed in 10% formalin for IHC analysis. For survival experiments, mice were weaned from drug and monitored every other day, or daily when signs of stress emerged (weight-shedding >10% and blood glucose <90 mg/dL). Glucose levels <60 mg/dL, visible ascites, weight shedding >15%, and loss of body temperature were endpoint criteria. All in vivo studies were approved by the IDIBELL's Animal Care and Use Committee and the Generalitat de Catalunya.

Data are presented as mean ± SD. All the statistical analyses were performed using GraphPad Prism 6. For one-group variables, statistical significance was determined by unpaired two-tailed t test; whenever variances were different, we applied the unpaired two-tailed t test with Welch correction. For two-grouping variables, two-way ANOVA was performed. Kaplan–Meier survival curve differences were analyzed with the log-rank (Mantel–Cox) test. All statistical analyses were made in a 95% confidence interval, with P values of <0.05, <0.01, <0.001 considered to be significant and represented with *, **, and ***, respectively.

For details on tumoroid cultures, tumorigenic capacity (limiting dilution) assay, immunoblot analysis, GST-BHMT assay, oxygen consumption rate (OCR), and extracellular acidification rate (ECAR), mitochondrial probes for FACS analyzes, high-performance liquid chromatography (HPLC) analysis of nucleotides, newly synthesized RNA and protein synthesis rate, oxygen consumption of tumor mitochondria, refer to Supplementary Materials and Methods.

We initially set out to determine whether mTOR inhibition was effective in blocking the proliferation of a panel of six HCC cell lines both sensitive and resistant to sorafenib, the current first-line treatment of HCC (39). The RAD001/BEZ235 combination was more effective than either agent alone in all HCC cell types (Supplementary Fig. S1A), including the two we have previously analyzed (26). To test the sensitivity of HCC cells to mTOR inhibitors in vivo, we established orthotopic models from Huh7 and Hep3B cell lines. After 14 days of tumor engraftment, cohorts of tumor-bearing mice were treated with mTOR inhibitors alone or in combination. In all cases, suppression of tumor growth was observed within 15 days, with the strongest effect elicited by the RAD001/BEZ235 combination (Fig. 1A; Supplementary Fig. S1B), consistent with our earlier findings in DEN-induced tumors in C57BL/6 mice (26). Likewise, the combined mTOR inhibitor treatment had a small impact on the weight of tumor-bearing mice (Supplementary Fig. S1C) and induced the dephosphorylation of mTORC1 kinase target RPS6 (Fig. 1B). Strikingly the on-target effects were more pronounced in tumor tissue (Fig. 1B), possibly due to their higher vascularization. Moreover, in combination with RAD001, the effects of BEZ235 in Huh7 tumors were recapitulated by a second-generation ATP-site specific mTOR inhibitor INK128 (Supplementary Fig S1D), with similar effects on body weight (Supplementary Fig. S1E). In addition, as we previously showed for RAD001/BEZ235 (26), we found the RAD001/INK128 combination more effective than either agent alone in inhibiting mTORC1 signaling and the proliferation of Huh7 cells (Supplementary Figs. S1F and S1G). To further test this response, we generated three-dimensional tumoroids (40) from Huh7 orthotopic tumors, and treated them in vitro for 6 days with the RAD001/BEZ235 combination. The combined mTOR inhibitor treatment in vitro was as effective in inhibiting the growth of tumoroids (Fig. 1C), as it was inhibiting orthotopic tumors (Fig. 1A). These studies suggest that the combination of an mTOR allosteric and an ATP-site competitive inhibitor is efficacious in treating tumors derived from human HCC cell lines.

Others have reported that mTOR inhibition by rapalogs protects a small population of stem-like cells (41, 42), potentially due to the maintenance of high mitochondrial oxidation capacity and/or the induction of autophagy (42, 43). To determine whether RAD001/BEZ235 would impact cell's tumorigenic capacity, we first generated tumoroids from Huh7 orthotopic tumors treated with or without RAD001/BEZ235 in vivo. The results showed that cells derived from RAD001/BEZ235–treated tumors were impaired in their ability to grow as tumoroids in 3D culture, as compared with controls (Fig. 1D). To further validate this finding, we conducted tumor initiation assays in vivo under conditions of limiting dilution with fixed numbers of cells. Huh7 tumor-bearing mice were treated for 15 days, tumor cells were isolated as for tumoroids, but instead were subcutaneously implanted in nude mice and monitored for tumor growth (Fig. 1E). In contrast to the effects reported for the rapalogs as single agents (41), the time of tumor appearance is reduced by the dual inhibition of mTOR as compared with the vehicle (Fig. 1E). Together, our results demonstrate that combined mTOR inhibition in vivo diminishes, rather than enhances, the tumorigenic potential of Huh7 cells.

Clinically, persistent hyperglycemia is most commonly treated with the biguanide metformin (44), including those patients whose hyperglycemia is induced by rapalogs (31, 32). This raised the possibility of adding a biguanide together, or potentially sequentially, with the RAD001/BEZ235 combination, which would effectively inhibit both glycolysis and mitochondrial oxidative phosphorylation (OXPHOS), directing tumors toward energetic stress. As previously reported for mTOR inhibition (27), treatment of Huh7 tumor–bearing animals with the RAD001/BEZ235 combination (RB) leads to hyperglycemia due to lower peripheral glucose uptake, an effect blunted by the addition of phenformin, which is known to decrease gluconeogenesis (Fig. 2A). To assess the mechanism of phenformin action, we examined a number of proliferative and signaling parameters in Huh7 cells in vitro. We found that cell proliferation is inhibited by either phenformin or RAD001/BEZ235 alone and that the inhibitory response is further enhanced when all three drugs are used in combination (Fig. 2B). Nevertheless, the kinetics of inhibition of mTORC1 signaling were much slower with phenformin than RAD001/BEZ235, as measured by S6K1, RPS6, and 4EBP1 phosphorylation (Fig. 2C). Moreover, phenformin in combination with RAD001/BEZ235 appeared to have no further effect on mTORC1 signaling (Fig. 2C), suggesting that the effects of phenformin on cell proliferation are mediated through mTORC1-dependent and -independent pathways. Consistent with this observation, in contrast to treatment with RAD001/BEZ235, phenformin strongly induced AMPK activation, as measured by increased AMPK T172 phosphorylation, with concomitant increases in the phosphorylation of AMPK substrate ACC S79 (Fig. 2C). Unexpectedly, these responses are blunted in the presence of RAD001/BEZ235, even at later times (Fig. 2C). Despite the latter findings, the combined effect of the three drugs in impeding proliferation may be explained by the inhibition of mTORC1 signaling and the activation of AMPK.

The induction of AMPK activation and ACC phosphorylation by phenformin (Fig. 2C) supports a change in cellular energy usage. As phenformin inhibits OXPHOS, we hypothesized that it would induce a metabolic shift toward glycolysis. To test this possibility, we followed ECAR, a surrogate for lactate production and glucose utilization. The results show that glucose addition to culture media in the presence of phenformin led to sharp increase in ECAR, whereas the combined mTOR inhibitors had a small inhibitory effect alone and blunted the effect of phenformin (Fig. 2D; Supplementary Fig. S2A). We predicted that, if phenformin-treated cells rely on glycolysis for survival, they should be sensitive to external glucose levels. In agreement with this hypothesis, lowering glucose levels in the culture media from 25 mmol/L to either 12.5 mmol/L or 6.25 mmol/L for 72 hours led to the death of all the Huh7 cells subjected to 1 mmol/L phenformin alone, but had no impact on the survival of RAD001/BEZ235–treated cells (compare both panels in Fig. 2E). The effect of phenformin on cell growth is blunted by the RAD001/BEZ235 combination at the higher concentrations of glucose, an effect that is almost completely lost at the lowest concentration of glucose (compare both panels in Fig. 2E). Phenformin-induced cell death appears to occur through apoptosis, as measured by the induction of caspase 3/7 activation (Supplementary Fig. S2B) and the inhibition of cell death by pan-caspase inhibitors (Fig. 2F). More strikingly, in tumoroids derived from Huh7 tumors, we found that when we lowered glucose levels in culture medium from 17.5 mmol/L to physiologic levels of 5 and 1 mmol/L, we achieved inhibition of proliferation at phenformin concentrations as low as 30 μmol/L (Fig. 2G). The results demonstrate that in the presence of phenformin, cells become more sensitive to the glycolytic input.

Recent studies have shown that mTORC1 inhibition also induces a reduction in the mitochondrial O₂ consumption rate (OCR) and mass, leading to an approximately 30% decrease in cellular ATP levels, responses attributed to inhibition of translation of nucleus-encoded mitochondrial gene transcripts (30). Given this observation, it was important to distinguish the impact of RAD001/BEZ235 versus phenformin on mitochondrial activity. We found that both phenformin and RAD001/BEZ235 reduced mitochondrial OCR (Fig. 3A; Supplementary Fig. S3A; Supplementary Table S1). By 3 hours of treatment, the combined mTOR inhibitors caused a 25% reduction in OCR, while phenformin led to a striking 75% reduction in OCR (Fig. 3A; Supplementary Fig. S3A–S3C), an effect that is established within minutes of treatment, consistent with phenformin directly inhibiting CI (Supplementary Fig. S3B). When combined with mTOR inhibitors, phenformin induced a further decrease in OCR (Fig. 3A; Supplementary Fig. S3C). Furthermore, phenformin caused a dramatic reduction in mitochondrial mass measured by MitoTracker Green FM probe (Fig. 3B) and although ATP levels do not drop (Fig. 3C), the AMP/ATP and ADP/ATP ratios were markedly increased (Fig. 3D; Supplementary Fig. S3D–S3E), consistent with phenformin inducing AMPK activation (Fig. 2C). Taken together, these results suggest that the phenformin-induced shift to glycolysis is not sufficient to relieve the energetic stress, with cells striving to maintain ATP levels.

In agreement with recent findings (30) we observe a small effect of RAD001/BEZ235 treatment on mitochondrial mass (Fig. 3B), but, in contrast, we observe no effect on ATP levels (Fig. 3C), consistent with no changes in the AMP/ATP or ADP/ATP ratios (Fig. 3D; Supplementary Figs. S3D–S3E). The latter findings agree with the lack of AMPK activation by RAD001/BEZ235 treatment (Fig. 2C). Given that RAD001/BEZ235 treatment reduces OCR, but does not alter AMP/ATP or ADP/ATP ratios, we asked whether this could result from lower energy needs. Inhibition of mTOR negatively affects ribosome biogenesis and protein synthesis, two of the most energy consuming catabolic cellular processes (45). We first analyzed steady-state levels of 18S and 28S rRNA by ethidium bromide (EB) staining, and nascent rRNA levels by pulse labeling with ³H-uridine. The EB-stained agarose gel shows that that there is no selective impact on the ratio of mature 18S to 28S rRNA (Fig. 3E). However, ³H-Uridine labeling of 47S precursor rRNA and the newly synthesized 18S and 28S mature rRNA species were strongly impaired by the RAD001/BEZ235 combination, with no apparent damage to processing, but an overall reduction in rRNA biogenesis (Fig. 3E). These effects are paralleled by a strong reduction of ³H-Leucine incorporation into nascent protein by RAD001/BEZ235 treatment (Fig. 3F). In contrast, the effects of phenformin on either response are not as profound (Fig. 3E and F). Taken together, the results suggest that mTOR inhibitors sustain a healthy cellular energetic state where energy production is balanced by controlling the rate of major energy-consuming biochemical processes, and such effects suppress AMPK activation by phenformin.

The inability of cells to maintain the AMP/ATP ratio in the presence of phenformin raised the possibility that, beyond lowering mitochondrial activity, the biguanide may also cause mitochondrial dysfunction and fragmentation, similar to other mitochondrial electron transport chain inhibitors and AMPK activators (38). Although treatment with RAD001/BEZ235 decreases mitochondrial activity (Fig. 3A; Supplementary Fig. S3A), it did not perturb mitochondrial efficiency, as measured by O₂ consumption in the presence of oligomycin, a CV ATPase inhibitor (Fig. 4A). In contrast, the dramatic drop in mitochondrial activity induced by phenformin (Fig. 3A) is accompanied by a decrease in mitochondrial O₂ consumption coupled to ATP production (Fig. 4A), indicative of mitochondrial dysfunction (46). Although RAD001/BEZ235 does not negatively affect mitochondrial efficiency, it slightly augments the effects of phenformin (Fig. 4A). Consistent with the biguanide inducing mitochondrial damage, by confocal imaging of a MitoTracker probe, we could observe that the fusion/fission equilibrium of the mitochondrial network was progressing towards fragmentation in the presence of phenformin (Fig. 4B), consistent with the activation of AMPK (38). This was not the case for RAD001/BEZ235 alone, where mitochondria appeared to be more fused (Fig. 4B), as recently reported for mTOR inhibitors (47). In agreement with these findings, phenformin, but not RAD001/BEZ235, induced cleavage of optic atrophy 1 (OPA1; Fig. 4C), an inner mitochondrial membrane GTPase required for fusion, whose cleavage is associated with extensive mitochondrial damage (38). Phenformin-induced mitochondrial damage has been shown to stimulate mitophagy and ROS production, leading to apoptosis (48). Unexpectedly, we found that phenformin treatment alone did not induce autophagy, and even blocked autophagy initiated by RAD001/BEZ235 treatment, as measured by the cargo-based endpoint GST-BHMT assay (Fig. 4D; ref. 49). This would argue that in the presence of phenformin, damaged mitochondria are not being cleared by autophagic mechanisms. In parallel, phenformin induced mitochondrial ROS as early as 3 hours posttreatment, at levels significantly higher than that of RAD001/BEZ235 treatment, with no further effect of the triple combination (Fig. 4E). Critically, by applying the mitochondrial ROS scavenger MitoTempo together with pan-caspase inhibitors, we rescued phenformin-induced cell death at drug doses as high as 1 mmol/L (Fig. 4F). Thus phenformin, either alone or in combination with RAD001/BEZ235, inhibits mitochondrial efficiency, evoking mitochondrial network fragmentation, ROS production, and apoptosis.

To determine the tumor growth–suppressing efficacy of phenformin alone, or in combination with RAD001/BEZ235, we used the orthotopic Huh7 tumor–bearing mouse model (Fig. 1A). Establishing the highest tolerated dose of phenformin in combination with RAD001/BEZ235 required lowering its dosage as a single agent in tumor-bearing nude mice from 100 mg/kg/day to 70 mg/kg/day. Although not as effective as the RAD001/BEZ235 combination, phenformin alone at 70 mg/kg had a significant inhibitory effect on tumor weight (Fig. 5A) and minimal effect on body weight (Supplementary Fig. S4A). Phenformin treatment in combination with RAD001/BEZ235, appeared to improve tumor suppression beyond that of the RAD001/BEZ235 combination (Fig. 5A). In addition, phenfomin combined with either mTOR inhibitor alone, did not reach the same extent of tumor suppression as the triple combination (Supplementary Fig. S4B). In four independent experiments, the tumor suppressor effect exerted by RBP led to more homogeneously small tumors when compared with the RAD001/BEZ235 combination alone. Using the Bartlett test of homogeneity of variances, we determined that the variance within the RBP group versus the RAD001/BEZ235 combination had a p value 0.76 ± 0.033, statistically different in a 90% confidence interval. Analyses of treated tumor tissues revealed strong inhibition of phosphorylation of mTORC1 targets RPS6 and 4E-BP1 by RAD001/BEZ235 alone, or in combination with phenformin, with a less pronounced effect of phenformin alone (Supplementary Fig. S4C). In parallel, RAD001/BEZ235, in the absence or presence of phenformin, had little effect on mTORC2 signaling, as measured by Akt phosphorylation (Supplementary Fig. S4C). This was most likely due to the low concentration of RAD001/BEZ235 administered to nude mice versus the higher doses, which we employed earlier in C57BL/6 mice (26). Consistent with the inhibition of mTORC1 signaling, RAD001/BEZ235, alone or in combination with phenformin, causes a sharp drop in cyclin D1 levels, indicative of G₁ cell-cycle arrest (Supplementary Fig. S4D). As described above, RAD001/BEZ235–treated mice showed a significant increase in glycemia (Fig. 2A), whereas phenformin treatment suppressed the glycemic effect of the mTOR inhibitors (Fig. 5B). Although, RAD001/BEZ235–treated mice did not display a significant inhibition of lactate production, the combined mTOR inhibitors blunted the strong proglycolitic effect of phenformin on lactate production (Fig. 5C). These results show that RAD001/BEZ235 and phenformin treatment is highly efficacious in suppressing tumor growth (Fig. 5A; Supplementary Fig. S4B). Moreover, phenformin decreases circulating glucose and increases lactic acid production, suggesting that, as in cells, it is acting to impede mitochondrial activity and favoring glycolysis.

Given the findings above, we asked whether the extent of tumor suppression induced by phenformin when added to RAD001/BEZ235 translated into a greater overall survival benefit after drug exposure for three weeks. RAD001/BEZ235 treatment alone significantly increased median survival by 26% over vehicle-treated control as shown by Kaplan–Meier analysis (Fig. 5D), with little effect on body weight (Supplementary Fig. S5A). In combination with RAD001/BEZ235, phenformin treatment further increased the median overall survival by 33% (Fig. 5D), with similar tumor weights at sacrifice (Supplementary Fig. S5B), suggesting that mice eventually perish due to primary tumor burden. Likewise, in a second experiment, the addition of phenformin to RAD001/BEZ235 increased median survival from 26% to 37%, with animals dying with similar tumor weights (Supplementary Figs. S5C and S5D, respectively). In vivo, lactic acid production indicated a systemic inhibition of mitochondrial oxidation and a shift toward glycolysis under phenformin administration. To ascertain the extent to which the treatments affected tumor mitochondrial function in vivo, we isolated mitochondria from treated tumors and measured their ability to consume O₂ employing a succession of OXPHOS substrates. In brief, the addition of glutamate, malate, and ADP (GM_P) shows that RAD001/BEZ235 treatment improves the usage of CI substrates, as compared with phenformin treatment alone or in the triple combination (Fig. 5E). However, the addition of succinate led to higher substrate control ratio after treatment with phenformin alone (Fig. 5F; Supplementary Fig. S5E), suggesting that, unlike under the RAD001/BEZ235 combination, such mitochondria are relying more on CII versus CI to maintain mitochondrial oxidation. When combined with phenformin, mTOR inhibition appears to decrease CII usage forced by phenformin inhibition of CI. These mitochondrial alterations did not appear due to altered OXPHOS complex expression (Supplementary Fig. S5F). The data support the notion that, in tumors treated with phenformin, CI function is hindered, evoking a compensatory mechanism by which mitochondrial electron transport is channeled through CII.

Although we observed benefit from adding phenformin to the RAD001/BEZ combination, the results above suggested that the phenformin concentrations we were achieving in vivo were not sufficient to attain a more dramatic effect on tumor mitochondrial activity. Moreover, we find that mTOR inhibitors appear to increase mitochondria efficiency and fusion, in agreement with others (30, 47), events that could potentially counteract phenformin-induced mitochondrial dysfunction and fragmentation. Therefore, we reasoned that if we first treated tumors with phenformin alone, we could potentially drive mitochondrial dysfunction, such that they would be more sensitive to the subsequent addition of mTOR inhibitors. In a sequential design, we treated tumor-bearing mice with phenformin 100 mg/kg for 8 days, and then either maintained them on phenformin or switched them to the RAD001/BEZ235 combination for 14 days. After 8 days treatment, phenformin at 100 mg/kg, had to be administered every other day (QOD), to be tolerated. Treatment with phenformin alone showed no enhancement in survival, whereas sequential addition of RAD001/BEZ235 had a dramatic effect (Fig. 5G). The combination of phenformin followed by mTOR inhibitors increased median overall survival by 70% from 28 days to 48 days (Fig. 5G), without effecting body weight (Supplementary Fig. S5G), effectively doubling the response observed by adding all three agents together (Fig. 5D). Critically, at the time of sacrifice, the tumor size was similar in each treatment group, suggesting that the mice succumb to primary tumor burden (Fig. 5H). Although we cannot discount a systemic effect, the results are consistent with the recent findings in cell culture suggesting that mTOR inhibitors are hindered by their capacity to promote mitochondrial fusion and increase mitochondrial efficiency (30, 47). Here we propose that in vivo pretreatment of HCC-bearing mice with phenformin compromises mTOR inhibitors undesirable effects, by inducing mitochondrial dysfunction and fragmentation while generating high ROS levels, such that subsequent mTOR inhibitor treatment results in a striking increase in overall mouse survival (see model in Fig. 6).

Sorafenib has shown an unprecedented clinical activity in advanced HCC and is presently the standard of care (50). However, its effects are not durable, with patients eventually succumbing to disease (51). Given these observations and that the mTOR pathway is hyperactivated in 50%–60% of HCC patients, the mTOR inhibitor Everolimus phase III trial EVOLVE was initiated. Although no benefits were observed (20), retrospective stratification of patients suggested those with loss of TSC2 benefited from RAD001 treatment (21). Interestingly, conditional depletion of TSC1 in the liver leads to HCC in the mouse (52). Such mice were cleared of disease if treated with rapamycin before the onset of tumorigenesis (52). In contrast, others have found that chronic treatment with rapamycin accelerates HCC in the DEN-induced mouse model, potentially due to IL6 production (53), or the protection of a specific subpopulation of HCC cells (42, 43). In apparent agreement with these findings, using carcinogen-induced (26) or orthotopically implanted HCC human cell line mouse models, we found little protection against tumor progression with RAD001 alone. However, in combination with the mTOR ATP-site inhibitor BEZ235, we observed strong tumor suppression with no apparent overall enhancement of tumorigenic capacity. Despite the fact that the development of BEZ235 in cancer has been discontinued, the successor of INK128, TAK228, has had encouraging results in phase I/II clinical trials (54); and, in our preliminary studies, when combined with RAD001, is as effective as BEZ235 in repressing mTORC1 signaling and tumor growth. In support of this approach, others have developed a dual inhibitor which can bind to both the allosteric- and ATP- binding sites of mTOR, which is as effective as targeting the two sites in combination (55, 56). The approach of combining allosteric and catalytic site inhibitors to increase efficacy is not confined to mTOR, as such an approach has been recently reported for Abl/Bcr–positive chronic myeloid leukemia (57). On the basis of the synergistic activity of the new allosteric inhibitor ABL001 with nilotinib, imatinib, or dasatinib, this combination is being tested in clinical trials (58).

The utility of targeting mitochondrial metabolism in cancer has been recognized for some time (59). In addition, clinically, biguanides have been used to resolve hyperglycemic states associated with mTOR inhibitors (31, 32). We chose phenformin over metformin because the hydrophobic biguanide does not require cell surface organic cation transporter 1 (OCT1; ref. 60), and because OCT1 appears downregulated in advanced HCC (33). Both in vitro and in vivo studies show that phenformin acts mainly through inhibition of the mitochondrial OXPHOS CI (34, 48), and we find that phenformin recapitulates the effects of other mitochondrial electron transport chain inhibitors by inducing acute activation of AMPK (38), as a consequence of the increased AMP/ATP ratio. Moreover, even though phenformin is a potent inhibitor of mTORC1 through AMPK-dependent and -independent mechanisms (61, 62), it seems to have little additional effect on this pathway in presence of RAD001 and BEZ235. Therefore, it appears to contribute to tumor suppression by distinct mechanisms. This may be through the activation of AMPK, whose role as tumor suppressor is complex, but which is largely thought to stem from its function as a suppressor of major anabolic process and the activation of catabolic responses (63, 64). It is clear from our study that activation of AMPK by phenformin suppresses both ribosome biogenesis and protein synthesis, two of the most major anabolic processes in the cell. However, we find that the activation of AMPK by phenformin is strongly suppressed in the presence of RAD001/BEZ235, potentially due to the energy-sparing effects of mTOR inhibition, underscoring the antagonist effects that such therapies may exert.

The rationale for pursuing phenformin in combination with RAD001/BEZ235 was that together they would significantly impact cell metabolism by attacking both mitochondrial and glycolytic capacity, eventually leading to energetic catastrophe. Cellular low energy-state and mitochondrial dysfunction are known to further stimulate the generation of mitochondrial ROS (65), consistent with our findings that the phenformin-induced apoptotic effects appear to be ROS-mediated. Moreover, the effects on ROS may be further exacerbated by a lack of autophagic clearance of damaged-mitochondria in the presence of phenformin. The mitochondrial impact of phenformin is highlighted by the apparent effort of treated cells to use alternative ATP-generating processes to survive, including glycolysis and potentially fatty acid oxidation (66). Nevertheless, the observed glycolitic shift does not seem to resolve the energetic stress imposed by phenformin in vitro, because the AMP/ATP ratios are not normalized, and cells become strikingly reliant on glucose levels.

Although energetic catastrophe, leading to apoptosis, appears to be ultimately dictated by ROS levels, the managing of energetic stress seems to be largely mediated by the regulation of mitochondrial OXPHOS capacity, efficiency, and dynamics (67). Here we show that phenformin induces a low energy-state characterized by an increase in AMPK activation, mitochondrial dysfunction, and mitochondrial fragmentation, all potentially contributing to phenformin's tumor-suppressive effects. Recent studies have shown that the electron transport inhibitors rotenone and antimycin A induce mitochondrial fragmentation by abrogating the mitochondrial localization of the fission protein DRP1, in an AMPK-dependent manner (38). However, these agents do not induce OPA1 cleavage, a fragmentation-related event that is only triggered after more extensive mitochondrial damage (38, 68), such as we observe here with phenfomin. This more extensive damage may also be a consequence of phenformin's inhibitory effect on autophagy. From a bioenergetics point of view, one would have predicted that the initiation of autophagy and mitochondrial biosynthetic pathways would be activated to reestablish an efficient mitochondrial network and decrease ROS levels (69, 70). However, it has been reported that autophagy initiates at mitochondria-associated ER membranes (MAM; ref. 71), such that phenformin-induced fragmentation may structurally compromise this contact sites, blocking the initiation of autophagy.

Consistent with our data, mTOR inhibitors were previously found to decrease cell respiration and mitochondrial mass, while maintaining coupled ATP production, although we did not observe a drop in ATP levels, as reported by Morita and colleagues (30). This finding was also consistent with RAD001/BEZ235 treatment having no effects on the AMP/ATP and ADP/ATP ratios or on AMPK activation. Instead, the maintenance of ATP levels appears to be the result of increased mitochondrial efficiency and lower cellular ATP demand. In agreement, we observed an increased mitochondrial network fusion phenotype following RAD001/BEZ235 treatment, which was recently reported for mTOR inhibitors (47), and that may be responsible for the increase in OXPHOS efficiency (67). The data suggest that mTOR control of mitochondrial dynamics may partially be responsible for the cytostatic effects of mTOR inhibitors; inducing a hyperfusion state that protects mitochondria against apoptosis, reminiscent of nutrient deprivation (72). Indeed, Morita and colleagues have demonstrated that inducing mitochondrial fragmentation could render mTOR inhibitors more cytotoxic (47). However, in vivo, the potential impact of phenformin on mitochondrial inhibition of CI is apparently much less than that achieved in vitro, suggesting that it might be insufficient to overcome the protective effects of the mTOR inhibitors when added in combination.

Given the findings above, our goal was to maximize each drugs effect in suppressing tumor growth, leading to the hypothesis that their sequential use may be more efficacious. This reasoning suggested that an initial treatment with phenformin followed by mTOR inhibitors would favor tumor growth suppression by limiting the capacity of cells to respond to mTOR inhibition by hyperfusion of mitochondria and increased mitochondrial efficiency. Concomitantly, phenformin-induced mitochondrial dysfunction forces tumor cells to use glycolysis, which might render them more prone to the mTOR inhibitors, which impair this response. Such a strategy of metabolic synthetic lethality was recently reported in a breast cancer model using a tyrosine kinase receptor inhibitor and either phenformin or ME344, a CI inhibitor in clinical trials (73). The administration of phenformin prior the use of mTOR inhibitors dramatically increased the overall survival of tumor-bearing mice, supporting the notion that there is a metabolic synthetic lethality that could be exploited when combining these agents. The paradigm of combining phenformin with the mTOR inhibitors needs to be further developed, in terms of dose, frequency, and time of drug exposure. In this context, FDG-PET could be used to ascertain the effects of phenformin on the induction of glucose uptake when cells shift to glycolysis, which should recede when mTOR inhibitors are administered, as recently proposed by Navarro and colleagues in a similar context (73). This will be of interest given the recent preclinical studies showing the benefit of combining phenformin with BRAF inhibitors in melanoma (74, 75), which have led to the initiation of a phase I clinical trial with phenformin in melanoma (37). It will be of interest to determine its efficacy in such a setting and whether phenformin with the dual mTOR inhibitors can be expanded to HCC and other cancers.

No potential conflicts of interest were disclosed.

Conception and design: S.R. Veiga, C.A. Mercer, H.E. Thomas, G. Thomas, S.C. Kozma

Development of methodology: S.R. Veiga, X. Ge, C.A. Mercer, J. Hernandez-Losa, G. Thomas, S.C. Kozma

Acquisition of data (provided animals, acquired and managed patients, provided facilities, etc.): S.R. Veiga, C.A. Mercer, M.I. Hernández-Álvarez, J. Hernandez-Losa, S.R. y Cajal, A. Zorzano

Analysis and interpretation of data (e.g., statistical analysis, biostatistics, computational analysis): S.R. Veiga, C.A. Mercer, M.I. Hernández-Álvarez, S.R. y Cajal, G. Thomas, S.C. Kozma

Writing, review, and/or revision of the manuscript: C.A. Mercer, H.E. Thomas, J. Hernandez-Losa, G. Thomas, S.C. Kozma

Administrative, technical, or material support (i.e., reporting or organizing data, constructing databases): G. Thomas, S.C. Kozma

Study supervision: G. Thomas, S.C. Kozma

These studies were supported by grants to S.C. Kozma from the Spanish Ministry of Economy and Competitiveness (BFU2012-38867), the Instituto de Salud Carlos III (II12/00002 and PI16/01762); to C.A. Mercer from the NIH/NCI National Cancer Institute (R21-CA158768); and to G. Thomas from the Spanish Ministry of Economy and Competitiveness (SAF2014-52162-P), the CIG European Commission (PCIG10-GA-2011-304160), the NIH/NCI National Cancer Institute (R01-CA191814), the Asociación Española Contra el Cáncer (GCB-142035THOM), the ISCIII-RTICC (RD12/0036/0049), the AGAUR (2014SGR-870) and by shared funding from the IDIBELL and the Vall d'Hebron Institute of Oncology (VHIO). In addition, S.C. Kozma and G. Thomas are co-funded by a grant from ISCIII (PIE13/00022) to M. Esteller. The Spanish Ministry grants to G. Thomas and S.C. Kozma are co-funded by FEDER – a way to build Europe. We are grateful to F. Riaño for assisting with the in vivo models and A. Gentilella for discussions in planning and executing the nascent rRNA and protein experiments. We also thank E. Battle for advice on the culturing of tumoroids, M. Sanchez-Feutrie and D. Sebastian for providing help in the mitochondrial respirometry experiments and G. Doerman with the preparation of figures. In addition we thank A. Tauler for his insightful discussions and I. Fabregat for providing HCC cell lines and critical consultations. Finally, we are indebted to Novartis Oncology for supplying RAD001 and NVP-BEZ235.

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.