Respiratory Capacity and Reserve Predict Cell Sensitivity to Mitochondria Inhibitors: Mechanism-Based Markers to Identify Metformin-Responsive Cancers

Most cancer cells have adapted to a high rate of glycolysis independent of hypoxia—a phenomenon known as aerobic glycolysis or the Warburg effect (1, 2). Despite the early postulation that mitochondrial dysfunction accounts for aerobic glycolysis, studies revealed that most cancer cells retained a considerable level of mitochondrial respiration. Like normal cells and tissues, the magnitude of respiration in cancers varies between tumors (3–5). Aerobic glycolysis prevents complete oxidation of metabolic fuels, allowing the accumulation of intermediates important for cancer growth and proliferation (6, 7). Equally important for cancer cells is the generation of high-energy molecules such as ATP that are needed for biosynthesis, cell growth and division, and cell movement, all of which are highly upregulated in malignant cells.

There are several key measurements of mitochondrial respiration, including basal respiration and maximum respiration/respiratory capacity (RC), which can be assessed by oximetry analysis. Respiratory reserve (RR), the difference between the respiratory capacity and basal respiration, is the “spare” respiratory capacity that is important for cellular responses to stress (8). Under normal/unstressed conditions, the cell operates at a basal respiration level that is only a fraction of its mitochondrial respiratory capacity; hence, there is RR. Under situations of increasing demand, basal respiration rises at the expense of RR to meet energy need. Therefore, the level of RR is postulated to be proportional to the capacity of cells to survive periods of high-energy expenditure, such as when during cell division/proliferation and cell migration. Indeed, past studies have demonstrated not only that inhibiting electron transport chain function induces cell death (9), but also that increasing RR enables cells to resist cell death (8). When cell respiration needs exceed the maximum capacity, i.e., RR reaches zero, mitochondrial energy production fails to meet the minimal needs of the cell, cell growth and proliferation are inhibited, and cell death ultimately ensues.

There are a number of reasons why cancer cells likely have lower RR than their normal counterparts. First, a high constitutive level of energy-consuming activities, such as proliferation and migration, dictates that cancer cells have higher basal respiration. Indeed, a recent study demonstrated that, at least in some epithelial cancers, mitochondria numbers and activities are elevated compared with surrounding normal cells (10). Second, cancer cells amass metabolic intermediates by elevating glycolysis at the expense of oxidative respiration. This leads to the hypothesis that cancer cells are more vulnerable metabolically, in comparison with their normal counterparts, to inhibition of mitochondria respiration. In this regard, several drugs targeting the mitochondrial electron transport chain have been evaluated as anticancer agents (11, 12). Among these, extensive attention has been paid to the widely prescribed antidiabetic drug metformin. Although the molecular mechanism(s) behind metformin's anticancer effect need to be further clarified, recent studies have provided evidence that metformin reduces tumorigenesis via its action to inhibit mitochondrial complex I activity (13–16).

Repurposing metformin for cancer treatment has attracted much attention, leading to extensive evaluation of data in treated diabetic patients (17–19). These studies have shown convincing associations between diabetes treatment with metformin and a reduced risk of developing cancer, in comparison with other treatment regimens (20–22). Indeed, two meta-analyses conducted in 2010 and 2014 showed that cancer incidence and cancer mortality in patients taking metformin were significantly reduced by 31% and 34%, respectively (23, 24). In addition, many in vitro studies have demonstrated the inhibitory effects of metformin on cancer cells (25, 26).

Based on these findings, a number of prospective cancer trials have been initiated, and some of these trials have completed. To date, the outcomes have not allowed a clear conclusion. Looking for clues to account for the discrepancy between population studies and clinical trials of metformin treatment, we examined patient selection criteria in the completed NIH-NCI phase II trials. The cancer-related inclusion criteria consisted mostly of surgery status and disease stage; the only molecular identifier was HER2 status in one of the trials. This approach is different from that of recent cancer trials for targeted therapy, which require patient stratification strategies based on markers that predict mechanism-based vulnerability to the trial compound. Hence, stratification of cancers based on molecular markers that predict sensitivity to metformin could enhance outcomes in drug-treated patients.

To this end, we have studied a large panel of cancer cells and found that their sensitivities to metformin strongly correlate with their mitochondrial RC and RR, independent of tissue origin. This finding supports the notion that metformin inhibition of ETC function is central to its anticancer effect (13, 16). Further investigation demonstrates a causal relationship between RC/RR of cancer cells and tumor tissues and their responsiveness to metformin, in both in vitro and in vivo studies. This new understanding of metformin mechanism of action led us to identify an effective combination of an ETC inhibitor with an AMP-activated protein kinase (AMPK) inhibitor that exhibited synergistic therapeutic efficacy.

Cancer cell lines H1299, H460, PC9, H522, HPAFII, PANC1, MiaPaca2, MCF7, MDAMB231, MDAMB436, Colo205, HCT116, HCT15, HT29, HepG2, Huh7, SK-Mel28, SK-Mel2, UACC62, M14, Sn12C, and OVCAR-8 were obtained from ATCC. Benign human prostate epithelial cell lines RWPE-1 was purchased from ATCC. hTERT immortalized BJ fibroblast cell line was a gift from Dr. P. Mathijs Voorhoeve (27). Cell lines were evaluated for Mycoplasma at the time of acquisition. Cell lines were frozen and stored as low passage store in multiple aliquots in liquid nitrogen. Experiments were performed within 8 passages with each subsequent thaw. Cells were cultured in RPMI or DMEM (Sigma-Aldrich) supplemented with 10% fetal bovine serum (FBS; HyClone), 50 units/mL penicillin and 50 μg/mL streptomycin (Gibco), at 37°C with 5% CO₂. However, in all treatments, RPMI or DMEM used was supplemented with 5% FBS instead.

Cell proteins were extracted using RIPA buffer (Cell Signaling) supplemented with protease inhibitor and phosphatase inhibitor (Sigma-Aldrich). Around 5 to 10 μg protein was loaded onto each lane of polyacrylamide (Bio-Rad) gel for SDS-PAGE. Proteins were transferred to PVDF membranes (Bio-Rad), which were then blocked with 5% nonfat milk in TBS-T at room temperature for 1 hour, and incubated with primary antibodies at 4°C overnight. Antibodies for phospho-AMPKα (Thr172), phospho-ACC (Ser79), and GAPDH were all from Cell Signaling Technology and used at 1: 1,000 dilution. Membranes were washed for 5 minutes, 4 times with TBS-T before and after secondary antibody (1: 5,000 dilution) blotting for 1 hour at room temperature. Protein bands were visualized with enhanced chemiluminescence (Thermo Scientific).

BJ cells were seeded at low density (6,000 cells/well) in a 24-well plate. Cells were allowed to grow for 1 week with the drug—metformin (Santa Cruz) or antimycin (Sigma-Aldrich)—at the concentrations indicated in the appropriate figure legend; media were changed every 2 to 3 days. Cell viability was then assessed using MTS CellTiter 96 Aqueous One Solution cell proliferation assay (Promega).

A bottom layer of 0.5% noble agar (Sigma-Aldrich) in RPMI or DMEM supplemented with 5% FBS and different concentrations of drugs was placed in each well of a 24-well culture plate. Cancer cells were suspended in mixture of 0.25% noble agar in DMEM supplemented with 5% FBS and the concentrations of drug(s) as stated in the respective figure legends. Each cell line was seeded at a different density, which was based on the intrinsic colony growth rate of the cell lines. These mixtures were placed on top of the bottom noble agar layers. Then, to each well, DMEM with 5% FBS media was added, which contained drugs including metformin (Santa Cruz), antimycin A (PubChem CID: 14957, Sigma-Aldrich), rotenone (PubChem CID: 6758, Sigma-Aldrich) or compound C (PubChem CID: 49761481, Santa Cruz) as stated in the respective figure legends. The drug-containing culture media were changed every 3 days. The colonies formed were stained by adding 50 μL of 5 mg/mL MTT (Sigma-Aldrich) into each well at weeks 3–4 after seeding. OpenCFU software (28) was used for colony quantitation.

Total RNA was extracted from cells using the RNeasy Mini Kit (Qiagen), followed by cDNA synthesis by reverse transcription reaction using Superscript First-strand Synthesis System (Invitrogen). Quantitative real-time PCR was then carried out with Sybr Green (Thermo Scientific) using the iCyler iQ5 Real-time Detection (Bio-Rad). The primers for UQCRFS1 were synthesized based on a previous publication (29); forward primer: 5′-GGAAATTGAGCAGGAAGCTG-3′, and reverse primer: 5′-GGCAAGGGCAGTAATAACCA-3′. Primers targeting the other genes were designed in-house; they were NDUFAF7 forward: 5′-AGCAGAAGCCTTCATACAACATGAC-3′, NDUFAF7 reverse: 5′-GTCGCAAAACCCTCTGAA GGTATCT-3′; KRAS forward: 5′- GCAAGAGTGCCTTGACGATAC-3, KRAS reverse: 5′-TCC AAGAGACAGGTTTCTCCA-3′; p53 forward: 5′-GAGGTTGGCTCTGACTGTACC-3′, p53 reverse: 5′-TCCGTCCCAGTAGATTACCAC-3′. Quantitative analysis was done by normalizing mRNA levels of these genes to that of 18S ribosome. Sequence of 18S forward primer was 5′-AAGTTCGACCGTCTTCTCAGC-3′ and the reverse primer sequence was 5′-GTTGATTAA GTCCCTGCCCTTTG-3′.

Cells were seeded, cultured, and treated with drugs in XF24 cell culture plates as detailed in the respective figure legends. One hour before performing the extracellular acidification rate (ECAR) assays, media were replaced by XF assay medium (Seahorse Bioscience) and incubated at 37°C in a CO₂-free environment. The subsequent glycolytic stress test assay (ECAR) was performed per XF24 analyzer standard protocol (Seahorse Bioscience). Glucose and oligomycin were purchased from Sigma, although 2-deoxy-D-glucose were from Santa Cruz. ECAR were measured using the XF24 analyzer (Seahorse Bioscience).

One or two million cells were assayed in 2 mL MiR05 buffer (3 mmol/L MgCl_2, 0.5 mmol/L EGTA, 20 mmol/L taurine, 10 mmol/L KH₂PO₄, 60 mmol/L K-lactobionate, 110 mmol/L sucrose, 20 mmol/L HEPES, and 1 g/L bovine serum albumin) using the Oroboros Oxygraph-2k respirometer (Oroboros Instruments). Respiratory oxygen consumption was assessed in real time as pmol of O₂ per second per million cells. Routine cellular respiration rate was first measured before the addition of any ETC stimulator or inhibitor. Oligomycin (2.5 μmol/L) was then injected and the drop in respiration was measured as ATP production rate of the cells. Maximal respiratory capacity was achieved by FCCP stimulation. Lastly, antimycin A and rotenone were added to the final concentration of 1 μmol/L for each compound to shut down the ETC to get the non–ETC-contributed oxygen consumption. To measure mitochondrial complexes' activities, digitonin was first added to permeabilize the cells. Subsequently, activities of the complexes were measured according to the protocol described in our previous study (30). Complex I, II, and III activities were measured as respiratory rate stimulated by pyruvate/malate, succinate, and duroquinol (prereduced by sodium borohydride), respectively, and normalized to the residual respiration after the addition of their specific inhibitors—rotenone, malonate, and antimycin. All chemicals were from Sigma-Aldrich.

Silencing of human UQCRFS1, NDUFAF7, AMPKα1, and AMPKα2 was achieved using lentiviral transduction of H1299 or MiaPaca2 cells with pLL3.7 vector expressing the corresponding shRNAs. shRNA targeting NDUFAF7 and UQCRFS1 were modified referring to papers of Rendon and Tormos, respectively (31, 32): sh-UQCRFS1 #1: 5′-GTACCCATTGCAAATGCAG-3′, sh-UQCRFS1 #2: 5′- GGTAACTGGAGTAACTACT-3′, sh-NDUFAF7 #1: 5′-GTGGACTTCAGTTATTTGC-3′, sh-NDUFAF7 #2: 5′-GAGACTTCAAGGTG GAAGA-3′. Sequences targeting AMPKα1 and α2 subunits were synthesized based on a previous report (33): sh-AMPKα1: 5′-GAGGAGAGCTA TTTGATTA-3′, sh-AMPKα2: 5′-GCTGTTTGGTGTAGGTAAA-3′. In order to achieve desired knockdown efficiency, transduction of H1299 or MiaPaca2 cells was conducted 1 to 3 times or with different titers of lentivirus produced in HEK293T packaging cells.

All mouse studies were done in accordance with the institutional IACUC guidelines. H1299 (7 × 10⁶) and MiaPaca2 (10 × 10⁶) cells were mixed with Matrigel (BD Biosciences) to a final concentration of 40% Matrigel in volume of 200 μL. These cell preparations were then injected subcutaneously into the left and right flanks of 6–8-week-old female SCID mice. When the average tumor size reached 300 mm³, the mice were randomized into control and treatment groups (5 mice in each group). For treatment, each mouse was dosed with metformin dissolved in sterilized water at 100 mg/kg by once daily oral gavage. Sterilized water was used as placebo in the control group. Notably, the dose ranges of metformin in mouse studies have been well established, ranging from 50 to 400 mg/kg (34, 35). Tumor volume was calculated as (length × width²/2) and measured every 2 days using calipers. Treatment and tumor size determinations were continued for 20 and 32 days for H1299 and MiaPaca2, respectively, before all the mice had to be sacrificed per protocol.

To obtain Colo205, H1299, HepG2, HT29, OVCAR8, and MiaPaca2 tumors, 10 × 10⁶ cells of each cell line were mixed with Matrigel (BD Biosciences) to a final concentration of 40% Matrigel in volume and injected subcutaneously into the left and right flanks of 6–8-week-old SCID mice. The mice were sacrificed when the tumors reached size of around 1,500 mm³, whereupon tumors were harvested and dissected away from the surrounding connective tissues. The tumor tissues were then minced with dissecting blade. Approximately 15 mg of minced tissues were weighted out and suspended in MiRO5 buffer for homogenization using PBI-shredder (OROBOROS). Respiratory functions of the homogenized tissues were then measured by Oroboros Oxygraph-2k respirometer in MiRO5 buffer supplemented with 5 mmol/L pyruvate, 2 mmol/L malate, and 5 mmol/L succinate. Maximal respiratory capacity was measured by stepwise stimulation of FCCP to reach the highest OCR, followed by antimycin (2.5 μmol/L) and rotenone (1 μmol/L) inhibition as detailed above.

GraphPad Prism software (GraphPad Software) was used in graph plotting and IC₅₀ calculation for soft-agar clonogenic assay, long-term two-dimensional assay, and short-term cell viability assay. For each correlation curve, Pearson correlation coefficient of determination (R² value) was derived using Microsoft Excel software to assess the correlation strength. To assess statistical significance of the correlation, P value was determined using Student t distribution model. To test statistical significance of difference between two correlation coefficients, Fisher z-transformation (http://vassarstats.net/rdiff.html) was performed, which gives a P value. For all studies, P values less than 0.05 were considered statistically significant. For drug combination studies, interactions of drugs were analyzed based on the Loewe additivity model and illustrated using isobologram (36–39), wherein the diagonal line joining IC₅₀ points of single drug treatments is the additivity line and IC₅₀ points of combination drug treatments below this diagonal line correspond to a synergistic effect.

Phase II cancer trials for metformin have failed to show clear efficacy so far, despite a significant association of metformin therapy with reduction of cancer incidence and mortality in diabetes patients. Identification of mechanism-based predictors/markers for responsiveness, and application of such markers in patient selection, are clearly needed for further development of metformin as an anticancer agent. To this end, we have developed the method to assess cancer cell sensitivities to metformin by soft-agar colony formation assay. Using this approach, we compared a pair of lung cancer cell lines H1299 and H522, and a pair of pancreatic cancer cell lines HPAFII and MiaPaca2, for their sensitivities to metformin. We found that H1299 cells are more resistant to metformin than H522 cells (Supplementary Fig. S1A), and HPAFII cells are more resistant than MiaPaCa2 cells (Supplementary Fig. S1B). We then expanded the study to 22 cancer cell lines of diverse tissue origins to include lung, pancreas, colon, liver, breast, skin, and ovarian cancer cells. Inhibition curves for these cell lines were plotted (Supplementary Fig. S2) to generate IC₅₀ values for metformin. The cell lines exhibited a wide range of sensitivities (35-fold) to metformin inhibition of proliferation, and their sensitivity appeared to be independent of tissue origin.

Inhibition of mitochondrial complex I function has been suggested to account for the anticancer effect of metformin. Hence, we postulated that the intrinsic respiratory properties of a cancer cell line might determine its responses/sensitivities to metformin treatment. To evaluate this hypothesis, we profiled mitochondrial respiratory properties of various cancer cell lines, including basal respiration, RC and the calculated RR. As with their responses to metformin, we found that cancer cell lines of the same tissue origin can have diverse respiratory profiles (Table 1), and further that a cell's basal respiration had no correlation with its RC (Table 1). For instance, HPAFII and MiaPaca2 cells have similar basal oxygen consumption rates of 19 and 17 pmol O₂/s/million cells, respectively, but very different RC and RR. Concomitantly, they also exhibited very different sensitivities to metformin treatment (Table 1), in that MiaPaCa2 cells, which had lower RC and RR, were more sensitive to metformin inhibition of proliferation than HPAFII cells. We observed the same association between RC/RR and sensitivity to metformin in the lung cancer cell pair H1299 and H522 (Table 1). Further, complex I, II, and III respiration capacities of HPAFII were 2.5-fold higher than basal respiration levels when stimulated with respective substrates or electron donors, indicating an unutilized respiratory potential under normal condition, whereas MiaPaca2 displayed hardly any increase in respiration in response to the same stimuli (Supplementary Fig. S3A).

To further analyze the relationship between cancer cell vulnerability to metformin treatment and its respiratory properties, we expanded respiratory profiling to the 22 cell lines for which we had metformin sensitivity data. Rank ordering of the cancer cell lines by either RC or RR showed very high correlation to their IC₅₀ for metformin inhibition of proliferation, with R² of 0.601 and 0.669, respectively (Fig. 1A and B). In contrast, there was little correlation between basal mitochondria respiration and IC₅₀ for metformin, with R² at 0.216 (Fig. 1C). Basal respiration level is that of the cell under a specific condition, and hence does not represent its capacity to response to stress/challenges. RC and RR, on the other hand, are the properties that convey the ability of the cell to withstand energy stress. By this logic, cells that have RC near the basal respiration level, i.e., with little reserve, are expected to be sensitive to the inhibition of ETC function. Indeed, the correlations suggest that cells with higher RC and RR are better able to withstand the inhibitory effect of metformin, which is consistent with the mechanism of metformin to inhibit ETC function. These results also suggest that the lower the remaining mitochondria capacity, the more limited is the capacity for cell proliferation and survival. We also analyzed the basal glycolytic rate and glycolytic capacity of the pancreatic and lung cancer cell line panels. Interestingly, we observed an inverse relationship between RC/RR and the rate of glycolysis of these cancer cells, i.e., the lower a cell's RC and RR, the higher the rate of glycolysis (Supplementary Fig. S3B and S3C). These data support the notion that cancer cells upregulate glycolysis at the expense of oxidative phosphorylation, and that different cancer cells have different balances between glycolysis and oxidative phosphorylation.

Metformin is transported in and out of the cells primarily by the OCT1 and MATE1 transporters, respectively (40, 41). Hence, we also evaluated the relationship between metformin sensitivity and the expression levels of OCT1 and MATE1. We analyzed the expression levels of OCT1 and MATE1 transporters in 12 cancer cell lines, covering a broad range of respiratory capacity and metformin sensitivity. We found that there existed no correlation between IC₅₀ for metformin and the expression level of OCT1 or MATE1 (Supplementary Fig. S4), in contrast to the high correlation level of IC₅₀ to respiration capacity and reserve. This result suggests that the level of metformin influx and efflux is not the limiting factor in its inhibition of cell proliferation and survival. These data provide further support the utility of using respiratory capacity as a reliable predictor for a cancer cell's sensitivity to metformin and other ETC inhibitors.

Having established a strong correlation between the RC/RR of a cancer cell and its response to metformin in vitro, it is important to determine whether our findings apply to samples obtained from an in vivo model. To this end, we compared RCs determined from cancer cells grown in vitro with those from isolated tumor tissues derived from the same cell lines (Fig. 2A; see Materials and Methods). Six cancer cell lines covering a range of metformin sensitivities that consistently form xenograft tumors were selected for the study; these cell lines are Colo205, H1299, HepG2, HT29, OVCAR-8, and MiaPaCa2 (Table 1). This analysis showed high levels of correlation between the RC values obtained from tumor tissues and the RC and RR values from the corresponding cancer cells grown in tissue culture conditions, with R² values of 0.844 and 0.942, respectively (Fig. 2B and C). Notably, the procedures of sample preparation for the measurement of respiration for cultured cells and tumor samples are different. For the tumor samples, tissue homogenization disrupts the plasma membrane integrity and exposes the mitochondria to the assay buffer, making basal respiration measurement unreliable. However, the respiratory capacity obtained from in vivo and in vitro samples of the same cancer cell lines is highly consistent, suggesting that substrate utilization does not contribute to this respiratory property.

We then tested the predictive value of RC for in vivo efficacy of metformin in the xenograft mouse model. H1299 and MiaPaCa2 were selected for the study because of their comparable proliferation rates in vivo, their significantly different in vitro response to metformin treatment, and importantly, their differences in RC values measured in either cultured cells or tumor tissue. We found that, while metformin treatment significantly inhibited the growth of MiaPaCa2 xenografts, the same treatment had no impact on tumors derived from H1299 cells (Fig. 2D and E). Taken together, these studies indicate that in vivo efficacy of metformin treatment for a particular cancer can be predicted from the RC assessed from the tumor tissue before initiating treatment, supporting the potential use of RC measurements in patient selection in a clinical setting.

Having established the correlation between RC/RR and cancer cell sensitivity to metformin, we next investigated if the levels of RC/RR determine or control the levels of responsiveness of cancer cells to ETC inhibitors such as metformin. To this end, we genetically downregulated RC and RR of H1299 cells to study the changes of sensitivity to metformin. H1299 cells were chosen because they exhibit high RC and RR and relative resistance to metformin. To manipulate RC and RR in these cells, we knocked down Ubiquinol-Cytochrome C Reductase Rieske Iron-Sulfur Polypeptide 1, UQCRFS1, an essential subunit of mitochondrial complex III, using two independent shRNAs (Fig. 3A). Both shRNAs targeting UQCRFS1 led to decreased levels of RC and RR in H1299 cells in a dose-dependent manner in comparison with control shRNA (Fig. 3B). More importantly, the ability of H1299 cells to form colonies in soft agar was reduced by UQCRFS1 knockdown in a titer-dependent manner for both targeting shRNAs (Fig. 3C). From several experiments using the two targeting shRNAs at different titers, we were able to demonstrate a strong correlation between RR reduction and colony-forming capacity, with an R² value of 0.771 (Fig. 3D). We also reduced RC and RR by knocking down NADH: Ubiquinone Oxidoreductase Complex Assembly Factor 7 (NDUFAF7), the assembly factor of mitochondrial complex I, using two targeting shRNA sequences, in H1299 cells (Fig. 3E). As observed in UQCRFS1 studies, NDUFAF7 knockdown significantly reduced RC and RR in H1299 cells (Fig. 3F), resulting in reduction of colony-forming abilities (Fig. 3G). These data strongly support the existence of a causal and predictive relationship between RC and RR and the sensitivity of a cancer cell to ETC inhibition.

The findings of an inverse relationship between a cancer cell's sensitivity to ETC inhibition and its RC and RR suggested that lowering the RC/RR of a metformin-resistant cell should render it more sensitive to the drug. Such a demonstration would strength the relationship between cell-intrinsic respiratory properties and sensitivity to ETC inhibitors. To this end, we again used the approach of knocking down the expression of ETC proteins. Complex I protein NDUFAF7 expression was suppressed in H1299 cells by 40% and 60%, respectively, using low viral titers of the afore-mentioned shRNAs to achieve moderate knockdown and reduction of RC and RR (Fig. 4A). Indeed, reducing NDUFAF7 expression rendered H1299 cells more sensitive to metformin inhibition of cell proliferation, as assessed by colony formation assay (Fig. 4B). We also performed similar studies by knocking down the complex III protein UQCRFS1. Consistently, knockdown of UQCRFS1 also rendered the cells more sensitive to metformin (Fig. 4C and D).

In addition to genetic knockdown of ETC essential proteins, we also used a pharmacologic approach to reduce RC/RR, by treating H1299 cells with rotenone, a known ETC inhibitor. Rotenone-treated H1299 cells were rendered more sensitive to metformin and, vice versa, metformin-treated cells were more sensitive to rotenone (Fig. 4E). The interaction of metformin with rotenone was analyzed with isobologram, a method to identify drug synergy (42). The concentrations of the two drugs in combination to inhibit 50% colony formation were below the diagonal lines joining the IC₅₀ values for metformin and rotenone as single agents for H1299 cells (Fig. 4F), which is characteristic for synergistic interaction between two drugs. Similar experiments were performed on another metformin-resistant cell line HPAFII, and a similar interplay between metformin and rotenone sensitivity was observed (Supplementary Fig. S5A and S5B). Together with the knockdown studies, these data provide compelling evidence for the direct relationship between the cell respiratory properties and cell vulnerability to ETC inhibitors. Further, these studies support the notion that anticancer effect of metformin is mediated through the inhibition of mitochondrial respiration.

Knowledge has been growing steadily about metabolic differences between cancer cells and benign cells. Although the molecular mechanisms for the metabolic adaptation are multifaceted and likely variable in different cancer cells, several common tumor suppressors or oncogenes, such as p53 and RAS, have been linked to the regulation of glycolysis and mitochondrial activity (43–46). We postulated that cancer cells have lower RC/RC than their normal counterparts as a direct consequence of the transformation process. To test this hypothesis, we transformed immortalized benign human fibroblast BJ cells using an established method of knocking down the tumor suppressor p53 and introducing an oncogenic mutant of KRAS (47). Specifically, retroviruses encoding shRNA targeting p53 shRNA and KRAS (G12V) were introduced sequentially into BJ cells immortalized by stable expression of the telomerase reverse transcriptase. Analysis of mRNA levels of p53 and mutant KRAS confirmed the identity of the cells (Fig. 5A). Comparing the transformed isogenic BJ cells with parental BJ cells, the former were found to have significant reductions in RC and RR (Fig. 5B). Consistent with the above findings, the transformed cells were more sensitive to metformin treatment than control cells, with IC₅₀ values of 1.7 mmol/L and 13.3 mmol/L, respectively (Fig. 5C). This transformation model, generated using two of the most frequently-mutated genes in human cancers, supports the hypothesis that malignant transformation programs cells to have lower mitochondria RC and RR, thereby becoming more responsive to ETC inhibition than their normal counterparts.

AMPK is an important regulatory protein in cell metabolism that is activated by nutrient and energy depletion. Indeed, we found that cells treated with concentrations of metformin that suppressed respiration exhibited increased phosphorylation of AMPK (Thr172) and acetyl-coA carboxylase (ACC, a direct substrate of AMPK; Fig. 6A), suggesting AMPK signaling is activated in response to energy depletion elicited by metformin. Activation of AMPK has been described as a double-edged sword in cancer cells that can both promote tumorigenesis and suppress cancer cell proliferation in a context-dependent manner (48, 49). To clarify the roles of AMPK activation in metformin-induced inhibition of cancer cell growth and proliferation, we knocked down AMPKα1 and α2 subunits in MiaPaCa2 cells using shRNA-expressing lentiviruses (Fig. 6B). We found that knockdown of AMPK sensitized MiaPaCa2 to metformin treatment, shifting the dose–response curve markedly to the left (Fig. 6C and D). Similarly, suppression of AMPK activation with the inhibitor compound C also sensitized the cells to metformin (Fig. 6E); isobologram analysis of combination treatment confirmed the synergistic suppression of cell proliferation in MiaPaCa2 cells by the combination of metformin and AMPK inhibition (Fig. 6F). This synergistic interaction suggests that AMPK activation, which promotes mitochondrial biogenesis and catabolic processes such as autophagy, is an adaptive response to metformin-induced energy deficiency that facilitates short-term cell survival (Fig. 6G). These results extend our understanding of cell signaling events downstream of ETC inhibition and identify a potential strategy of combined treatment with ETC and AMPK inhibitors to treat specific cancers.

It is well established that cancer cells change their metabolic program during oncogenesis. However, the relative importance of mitochondrial respiration in cancer cells remains a point of debate (1–3, 5, 7). Our studies have demonstrated that cancer cells are more vulnerable to the inhibition of mitochondrial respiration than nontransformed cells. This notion makes biological sense and offers new opportunities for therapeutic targeting.

Cancer cells need high-energy molecules for the processes of biosynthesis/anabolic activities, such as cell growth, proliferation, and cell migration, which are significantly upregulated compared with normal cells. Besides generating ATP, ETC is important for its essential role in pyrimidine biosynthesis through mitochondrial dihydroorotate dehydrogenase (DHODH), a component of the mitochondrial electron transport chain (ETC; refs. 50–52). The Warburg effect, i.e., elevation of glycolysis at the expense of mitochondrial respiration, causes remaining mitochondria respiration to be less dispensable.

Abundant experimental evidence suggests that downregulation of mitochondrial respiration capacity is intimately linked to the process of malignant transformation (43–46). Indeed, we found that the process of transforming BJ fibroblast cells led to the reduction of RC and RR, providing direct evidence connecting malignant transformation to changes in respiratory properties, and increased vulnerability to antimitochondria agents such as metformin. Although more comprehensively studies are necessary to evaluate the impact of transformation on cell respiratory capacity, our model of manipulating the expression of a frequently mutated tumor suppressor and a similarly common oncogene provides a clear and relevant demonstration of the principle. This approach also revealed differential responses of benign cells and transformed cells to ETC inhibitors and provided support for new potential therapeutic strategies.

There have been significant efforts in developing effective mitochondria inhibitors for cancer treatment (16, 19, 53), but clinical trials of one mitochondrial inhibitor, metformin, an FDA-approved antidiabetic drug, have yielded disappointing results (17, 19, 54). The lack of mechanism-based patient stratification strategies could certainly be a major factor for inconclusive outcomes of metformin trials. Recent studies have suggested that cancer cells harboring certain mutations display increased sensitivity to biguanides (55, 56). However, these studies have also highlighted that other factors, mostly undefined, and the complicated interplay of these factors, contribute to the sensitivity to biguanides. There has been recent effort to analyze metabolic signatures associated with metformin treatment in ovarian cancers using the approach of integrated metabolomics, which yielded interesting results (57). However, no straightforward predictive metabolite markers have been clearly defined using this sophisticated approach. In contrast, our model establishes a direct functional link between measurable intrinsic respiratory property of cancer cells and their responsiveness to metformin. Indeed, in the study on 22 cancer cell lines of diverse tissue origin, we found that there is a broad spectrum of cancer cell sensitivities to the antiproliferative effect of metformin, and that a cancer cell's sensitivity to ETC inhibition strongly correlates with its RC and RR. The current study also established the correlation between the RC measured in isolated tumor tissue samples and the RC obtained from the same cells under normal tissue culture conditions, clearing the way for clinical application using small biopsy tumor samples. Practically, the oximetry analysis method can be done efficiently—within an hour of obtaining of fresh tumor sample, which requires no expensive equipment and reagents. These findings suggest that respiration-based stratification could be applied to improve efficacy of not just metformin, but ETC targeting agents in general, in cancer clinical trials and potentially in future clinical practice.

The mechanism of action of metformin in the inhibition of cancer cell proliferation and survival is not without debate, despite recent evidence connecting its role in suppressing complex I function with its antitumorigenesis effects (13, 16). In this regard, the current study not only demonstrates a strong correlation between the RC and cell sensitivity to metformin, but also establishes the causal relationship between RC and sensitivity to metformin. Altering cancer cell sensitivity to metformin through genetic and pharmacologic manipulation of ETC and RC provides important evidence for this concept. Further support for this conclusion comes from the demonstration that, under ETC inhibition, simultaneous suppression of AMPK activation synergistically inhibits cancer cell proliferation. The advances in understanding the mechanism of action, in developing methods to stratify patients with cancer, and in designing effective combination strategies will, hopefully, facilitate the clinical application of ETC inhibitors in the treatment of human cancers.

No potential conflicts of interest were disclosed.

Conception and design: J.T. Teh, W.L. Zhu, C.B. Newgard, M. Wang

Development of methodology: J.T. Teh, W.L. Zhu, M. Wang

Acquisition of data (provided animals, acquired and managed patients, provided facilities, etc.): J.T. Teh, W.L. Zhu, M. Wang

Analysis and interpretation of data (e.g., statistical analysis, biostatistics, computational analysis): J.T. Teh, W.L. Zhu, M. Wang

Writing, review, and/or revision of the manuscript: J.T. Teh, C.B. Newgard, P.J. Casey, M. Wang

Administrative, technical, or material support (i.e., reporting or organizing data, constructing databases): W.L. Zhu, M. Wang

Study supervision: M. Wang

Other (applied and provided grant support for the studies reported): M. Wang

The funding for the project is provided by the grants awarded to Mei Wang by the Singapore Ministry of Education Tier2 grant (MOE2013-T2-2-170) and Singapore National Medical Research Council (NMRC) Individual Grant (NMRC/CIRG/1486/2018).

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.