The BRAF Inhibitor Vemurafenib Activates Mitochondrial Metabolism and Inhibits Hyperpolarized Pyruvate–Lactate Exchange in BRAF-Mutant Human Melanoma Cells Free

The RAS–RAF–MEK–ERK is one of the most important signaling cascades in cancer (1). Growth factor receptor stimulation activates RAS, leading to recruitment of RAF kinases (ARAF, BRAF, and CRAF), which in turn activate MEK1/2, ERK1/2, and a range of target proteins, including transcription factors regulating proliferation, differentiation, survival, and invasion.

Aberrant ERK1/2 signaling occurs in many cancers through mutation or overexpression of components of the pathway (2). For example, activating mutations in BRAF, most often involving the V600E substitution, lead to malignant transformation and occur in about half of all cases of malignant melanoma (3).

The importance of ERK1/2 signaling in driving melanoma has prompted interest in blocking this pathway for mechanism-based therapy, with several BRAF and MEK inhibitors now approved for the treatment of BRAF-driven melanoma (e.g., the BRAF inhibitors vemurafenib and debrafenib, and the MEK inhibitor trametinib; ref. 1) and many more undergoing clinical testing.

These drugs have shown remarkable activity in BRAF-mutant melanoma patients (4, 5), but responses are invariably short-lived with tumor relapse observed within few months of treatment initiation (6). This is due to mechanisms such as reactivation of ERK1/2 signaling (e.g., via mutation in MEK, overexpression of COT) or activation of ERK1/2-independent signaling pathways (e.g., through receptor tyrosine kinase overexpression; refs. 7–9), an understanding that has informed combination therapeutic strategies targeting the compensatory oncogenic activity (10) that are now being evaluated. Understanding the consequences of treatment with BRAF and MEK inhibitors on fundamental cellular processes will enable the identification of additional combinatorial treatment options to refine the use of these drugs and achieve better disease control in the clinic.

Cancer cells exhibit altered metabolism relative to normal tissues, characterized by increased dependency on aerobic glycolysis, fatty acid and nucleotide synthesis, and glutaminolysis (11). This “metabolic transformation” is considered an enabling hallmark for cancer maintenance and progression that is tightly linked to oncogenic signaling, and as such is being pursued as a promising therapeutic strategy (12).

In the context of BRAF–MEK–ERK signaling, mutant BRAF stimulates glycolytic activity and inhibits oxidative phosphorylation (13). We and others have also shown that inhibitors of MEK and BRAF reverse this metabolic phenotype by attenuating the glycolytic activity of BRAF-mutant human melanoma cells (14, 15) and reactivating mitochondrial oxidative phosphorylation (OxPhos; ref. 16) linked to reduced expression of hexokinase 2 (HK2) and glucose transporters 1 and 3, following the downregulation of CMYC (14) and HIF1α (15) downstream of the ERK1/2 pathway. Indeed, reduced uptake of the radioactive glucose analogue 2 [¹⁸F]fluoro-2-deoxy-d-glucose (FDG), as monitored by positron emission tomography (PET) in preclinical models as well as BRAF-driven melanoma patients, has proved to be very useful for monitoring response to BRAF/MEK targeted drugs (17) but relatively nonspecific.

The reprogramming of glucose metabolism following BRAF/MEK inhibition could be considered as an adaptive response necessary to mitigate drug-induced metabolic stress (13). How such alterations are brought about in terms of glycolytic pathway flux changes, their significance for cell survival, and potential as metabolic imaging biomarkers of drug action, besides the previously described and relatively nonspecific FDG-PET uptake (18), remains largely unclear.

This work is centered on the metabolic aspects of BRAF-mutant melanoma cell response to BRAF inhibition with vemurafenib. Our aims are to characterize the metabolic and molecular response of BRAF-mutant melanoma to BRAF inhibitors and investigate the potential of the changes induced by treatment as noninvasive imaging biomarkers of response. Accordingly, we investigate the effects of the BRAF inhibitor vemurafenib on cellular metabolism as well as glycolytic pathway fluxes in BRAF-mutant human melanoma cells using NMR spectroscopy, a technique that enables the steady state as well as dynamic study of metabolism in cells and whole tissues both in vitro and in vivo (19).

We show that vemurafenib decreases glycolytic activity and reactivates TCA cycle metabolism by increasing oxidative and anaplerotic flux through pyruvate carboxylase (PC), reducing cell dependency on glucose and glutamine metabolism. We also show that vemurafenib depletes monocarboxylate transporter 1 (MCT1) protein expression resulting in decreased hyperpolarized ¹³C-pyruvate-lactate exchange, thus providing support for investigating this process as a new biomarker for noninvasive monitoring of BRAF signaling inhibitor action.

The following human melanoma cell lines were used and acquired at the American Tissue Type Collection: WM266.4 (BRAF^V600D/RAS^WT), SKMEL28 (BRAF^V600E/RAS^WT, STR profiled in house (LGC Standards) on October 16, 2015), and CHL-1 (BRAF^WT/RAS^WT). D04 (BRAF^WT/RAS^Q61L) cells were a kind gift from Dr. Amine Sadok (Cancer Biology Division, The Institute of Cancer Research, London) and were tested by STR profiling on the June 13, 2014. Vemurafenib and ¹³C-glucose were purchased from Chemietek and Sigma-Aldrich, respectively.

Cells were grown as monolayers and routinely cultured as previously described (14). For steady-state metabolic investigations, the following vemurafenib concentrations were used with WM266.4 cells: 0.5×, 1.25×, 2.5× and 5 × GI₅₀ (0.2, 0.5, 1, and 2 μmol/L, respectively). CHL-1 cells were treated with 0.02×, 0.05×, 0.1×, 0.2, 1×, 2.5×, and 5 × GI₅₀ (0.2, 0.5, 1, 2, 9, 22.5, and 45 μmol/L) vemurafenib, while SKMEL28 and D04 cells were treated with an equimolar concentration of 2 μmol/L [under these conditions, ERK signaling was effectively inhibited in SKMEL28 (BRAF^V600E) but not in D04 (BRAF^WT) cells]. Cell counts and viability were monitored with trypan blue staining using Vi-CELL Cell Viability Analyzer (Beckman Coulter).

For ¹³C-glucose flux analyses, WM266.4 cells were incubated in media containing 5 mmol/L [1-¹³C]glucose, as this is physiologically relevant and provided similar results to the standard medium used in the ¹H NMR experiments (25 mmol/L glucose; Supplementary Fig. S1). Either 0.01% DMSO or 5 × GI₅₀ vemurafenib (2 μmol/L) was added for 24 hours.

For nutrient deprivation experiments, cells were seeded in four different media conditions: 5 mmol/L glucose, 1 mmol/L glucose, 1 mmol/L glucose without glutamine, 1 mmol/L glucose without glutamine and pyruvate (48 hours before treatment) and were then exposed to either 0.01% DMSO or 2 μmol/L vemurafenib for 24 hours, 48 hours, or 72 hours in the presence of these media.

Control and vemurafenib-treated WM266.4 cells were extracted with a methanol–chloroform–water method as previously described (20). The aqueous fraction was reconstituted in D₂O using 3-(trimethylsilyl) propionic-2,2,3,3-d4 acid and methylenediphosphonic acid as ¹H and ³¹P NMR standards, respectively. Lipid fractions were resuspended after chloroform evaporation in a d-chloroform solution with tetramethylsilane as reference. Further details on this section are provided in the Supplementary Material.

¹³C-pyruvate–lactate exchange was monitored in intact WM266.4 human melanoma cells (∼8.5 × 10⁶ cells/sample) following exposure to DMSO or vemurafenib for 24 hours as previously described (21). Dynamic ¹³C spectra were acquired every 2 seconds for 4 minutes immediately after the addition of 10 mmol/L hyperpolarised [1-¹³C]pyruvic acid and 10 mmol/L unlabeled lactate in a total volume of 500 μL. For data analysis, the ratio of the area under the curve for the summed lactate and pyruvate signals (lactate_AUC/pyruvate_AUC) from the dynamic spectra was determined to estimate pyruvate–lactate exchange (21).

NMR data were acquired on a Bruker Avance III 500-MHz NMR spectrometer (Bruker Biospin). Spectra were processed using MestRe-C version 4.9.9.6 (University of Santiago de Compostela, Spain), and metabolite content was measured by peak integration relative internal standards and corrected for cell number per sample. Further details on acquisition parameters are provided in the Supplementary Material.

¹H NMR data from WM266.4 cells were subjected to unbiased metabolic profiling using partial least-squares discriminant analysis (PLS-DA), a method performed after principal component analysis (PCA) to sharpen the separation between groups of observations, determining the variables carrying the class separation information. For this, spectra were processed as previously described (22) and data were analyzed in SIMCA v13.0 (Umetrics-Umeå) using a PLS-DA model.

Target protein expression and phosphorylation levels following BRAF inhibition were assessed by Western blotting using standard conditions as previously described (23). Antibody information is provided in the Supplementary Material.

Total RNA was extracted using the RNAeasy kit (Qiagen), and 1 μg was reverse transcribed using the High Capacity cDNA Reverse Transcription Kit (Applied Biosystems). Samples were diluted 1:10 and 1 μl used in the Taqman assay, using Taqman universal master mix. Primer information is provided in Supplementary Material.

Cells (2 × 10⁶) were lyzed in Ripa buffer (Cell Signaling Technology) containing PhosSTOP Phosphatase Inhibitor Cocktail Tablets (Roche) and cOmplete ULTRA Tablets (Roche), and processed as previously described (24). A spectrophotometric reading in kinetic mode at 412 nm was taken for 10 minutes at 30°C (Ultrospec 2100pro, GE Healthcare Life Sciences) and data normalized for protein content.

Flow cytometry was performed to analyze the effect of drug on cell-cycle distributions as previously described (20).

For metabolite analysis, the Student t test with Sidak–Bonferroni correction for multiple comparisons (P ≤ 0.05) was applied. mRNA levels, cell number, and PC activity were analyzed using a single-comparison two-tailed unpaired Student t test with P ≤ 0.05 considered statistically significant. Results are expressed as mean ± standard deviation (SD).

Treatment with vemurafnib (2 μmol/L, 24 hours) led to inhibition of BRAF signaling, as evidenced by the reduced phosphorylation of ERK1/2 and MEK, in BRAF^V600D WM266.4 and BRAF^V600E SKMEL28 but not in BRAF^WT CHL-1 and D04 human melanoma cells. These effects were concomitant with a decrease in extracellular lactate (Lactate^E) levels exclusively in BRAF-mutant cells (Fig. 1A), consistent with previous reports (13, 14). The vemurafenib-induced reduction in Lactate^E was concentration-dependent in WM266.4 cells being observed with as little as 0.2 μmol/L (Fig. 1B). In contrast, BRAF^WT CHL-1 cells showed no significant changes in Lactate^E even with exposure to concentrations of vemurafenib up to 45 μmol/L (Fig. 1B).

After confirming that vemurafenib induces a significant reduction in Lactate^E in BRAF-mutant cells, comparable with that previously reported using MEK inhibitors (14), we next assessed the effect of BRAF inhibition on additional metabolic processes by investigating the changes in cellular metabolic profiles induced by vemurafenib in WM266.4 cells. As shown in Fig. 1C, PLS-DA unbiased multivariate analysis of the ¹H NMR spectral data from the aqueous phase of WM266.4 melanoma cell extracts indicated separate clustering of control- and vemrafenib-treated (2 μmol/L, 24 hours) cell data, consistent with a shift in metabolic phenotype. The score scatter plot indicated that 40.1% of total data was explained by two main principal components (PC) in the model (PC1, 13.5%; PC2, 26.6%). The high R2 and Q2 values (90.1% and 66.7%, respectively) indicated that the classification has good reproducibility and predictivity. The resonances with the highest contribution to the classification model are shown in Fig. 1C and include branched-chain amino acids (BCAAs; 0.91–1.06 ppm), lactate (1.33 and 4.12 ppm), acetate (1.92 ppm), creatine (Cr) + phosphocreatine (PCr) (3.03 ppm), glycine (3.56 ppm), and myo-inositol (4.07 ppm).

The individual ¹H NMR resonances in the control and treated spectra were manually integrated and included in a univariate analysis to corroborate significant metabolic differences identified in the PLS-DA. Table 1 shows data from the main metabolites analyzed, with a significant increase in glycine and myo-inositol, and a significant decrease in lactate and acetate in vemurafenib-treated compared with control cells. The effect sizes described are in the range of relevant findings described in previous publications using this methodology (14).

³¹P NMR analysis revealed no significant differences in the levels of measured ³¹P-containing metabolites, including NTP and PCr, between control- and vemurafenib-treated samples (Supplementary Fig. S2). Furthermore, vemurafenib treatment in WM266.4 cells had no significant effect on the ADP/ATP ratio assessed using a bioluminescence assay (Supplementary Fig. S2). Thus, WM266.4 cells are able to maintain their bioenergetic status during BRAF inhibition despite reduced glycolytic metabolism.

¹H NMR analysis of the lipid phase obtained from the same cell extracts, showed that BRAF inhibition with vemurafenib was associated with a decrease in the fatty acyl chain signal at 0.9 ppm (-CH₃) (% change within the range of previous studies on tumor lipids; refs. 25 and 26), whilst the remaining signals were mostly unchanged (Fig. 1D).

Taken together, our data suggest that vemurafenib reduces glycolytic activity and alters glycine, myo-inositol, acetate, and lipid metabolism without compromising cellular bioenergetics.

We next investigated the alterations in glucose metabolic pathway activity that could underpin the observed metabolic changes. BRAF inhibitor treatment has previously been shown to reactivate mitochondrial OxPhos, leading to increased reactive oxygen species (ROS) levels (13). We thus assessed ROS in WM266.4 cells following exposure to vemurafenib and found that, consistent with a previous report (16), BRAF inhibition in BRAF^V600D WM266.4 cells for 24 hours led to a concentration-dependent increase in ROS production (up to 197.8% ± 62.8% of controls; P = 0.03), indicating that OxPhos may also be increased in our cells (Supplementary Fig. S3).

Next, and to further explore the effect of vemurafenib on metabolic fluxes and investigate if the ROS changes are related to altered mitochondrial activity, we monitored the fate of [1-¹³C]glucose in BRAF^V600D WM266.4 human melanoma cells and growth media using ¹³C NMR (Fig. 2A). Analysis of culture media following a 24-hour incubation revealed a reduction in [3-¹³C]lactate^E levels in vemurafenib-treated cells relative to controls (down to 62.9% ± 13.1%; P = 0.01), consistent with reduced de novo lactate production being responsible for the fall in steady-state lactate^E. A trend toward decreased glucose consumption was also observed but did not reach statistical significance (59.5% ± 30.1%; P = 0.07). Analysis of intracellular ¹³C-labeled metabolites also showed a reduction in [3-¹³C]lactate (to 51.9% ± 16.3%, P = 0.003) concomitant with a significant increase in [1-¹³C]glucose (up to 245.8% ± 21.9%, P = 0.03), and myo-inositol (up to 561.8% ± 250.1% P = 0.01) in vemurafenib-treated relative to control BRAF^V600D WM266.4 cells, indicating decreased glycolysis and glucose utilization after internalization and increased routing toward myo-inositol production (Fig. 2C).

The relative contribution of the oxidative [pyruvate dehydrogenase (PDH)] versus anaplerotic [pyruvate carboxylase (PC)] glucose metabolism was assessed using the resonances of ¹³C-labeled glutamate and glutamine (labeled in carbons 2, 3, and 4; Fig. 2B). Our data showed that the [4-¹³C]glutamate signal was always far greater than that from [2-¹³C] and [3-¹³C]glutamate, in keeping with PDH flux being the primary route for pyruvate entry into the TCA cycle in melanoma cells (27). [4-¹³C]glutamate levels were not altered with treatment (103.3% ± 12.5% of controls; P = 0.6) despite the significant reduction in ¹³C-label incorporation in the glycolytic pathway (following intracellular ¹³C-glucose accumulation), suggesting a relative increase in PDH flux. In addition, a significant elevation in [2-¹³C] and [3-¹³C]glutamate (to 190.5% ± 39.7% and 160.9% ± 41.4% of controls, respectively, P < 0.05), [2-¹³C]glutamine (134.2 ± 13.5; P = 0.01) and [2-¹³C] aspartate (351.2 ± 206.4) was also observed, indicating increased PC flux. The [2-¹³C]/[4-¹³C] glutamate ratio rose from 0.13 ± 0.03 to 0.25 ± 0.06 (P = 0.01), consistent with an increase in the PC/PDH flux in vemurafenib-treated compared with control WM266.4 cells (Fig. 2B and C).

To investigate the metabolic basis for the ¹³C NMR findings, we performed an independent analysis of PC enzyme activity. This showed a significant increase (159 ± 67 %, P = 0.04) in treated samples relative to controls (Fig. 2D), providing independent confirmation of the increase in PC flux following vemurafenib treatment.

Analysis of the ¹³C-labeled lipid phase of treated cells revealed a decrease in the fatty acyl chain signal (-CH₃ and –CH₂ at 14.15 and 29.7 ppm, respectively; Fig. 2E) after 24 hours of treatment (70% ± 18% and 72% ± 19% of controls, respectively, P = 0.05), in agreement with the ¹H NMR data, indicating reduced glucose routing toward lipid biosynthesis following vemurafenib treatment.

In summary, the ¹³C flux analysis indicates that vemurafenib treatment leads to decreased glucose utilization coupled with diversion toward myo-inositol and TCA cycle metabolism (particularly via PC flux) at the expense of lactate and lipid synthesis.

To investigate the molecular mechanisms underlying the metabolic shift observed with vemurafenib in BRAF-mutant WM266.4 human melanoma cells, we assessed the expression of key enzymes in the glycose metabolism pathway, initially using qRT-PCR. Our data showed a significant decrease in the mRNA expression of HK2 (14.7% ± 2% of controls), in line with our previous findings with MEK inhibition (14). Further, we observed a reduction in the mRNA level of LDH-A (to 18.5% ± 6.5%), PDK-1 (28.9% ± 6.6% with respect to control), 3-PHGDH (to 46 ± 2.4%), and GCAT (to 62.1% ± 12.3%) in treated relative to controls. The mRNA expression of PSAT-1, GLDC, PC, IDH-1, GLS, and ISYNA-1 remained unchanged (Fig. 3A).

Protein expression changes in genes that showed a significant difference in mRNA abundance were next assessed by Western blotting in the same cell line (BRAF^V600D WM266.4; Fig. 3B) and in three additional human melanoma cell lines: BRAF^V600E SKMEL28 cells, BRAF^WT D04, and BRAF^WT CHL-1 cells; Fig. 3C). Our data show that both BRAF-mutant cell lines (but not BRAF^WTcells) exhibited a decrease in HK2 and 3-PHGDH protein levels, in agreement with qRT-PCR. However, the decrease in LDH-A expression was not as pronounced as observed with qRT-PCR, probably due to a longer protein half-life.

We additionally probed for monocarboxylate transporter (MCT) 1 and 4 and, interestingly, observed depletion of both proteins in vemurafenib-treated compared with control in BRAF-mutant WM266.4 and SKMEL28 cells but not BRAF^WT D04 or CHL-1 cells, suggesting inhibition of lactate transport in BRAF-mutant cells (28). Given the observed changes in lipid metabolism, we further assessed the levels of ACAD9 (fatty acid breakdown), ACC and P-ACL (lipid synthesis). Our data showed that vemurafenib treatment was associated with a reduction in ACAD9 and P-ACL levels in both BRAF-mutant WM266.4 and SKMEL28 cell lines but not in BRAF^WT CHL-1 and D04 cells, while no consistent trends were observed with ACC expression following exposure to vemurafenib (Fig. 3B and C).

Overall, these data show that BRAF inhibition produces a metabolic enzyme expression profile suggestive of inhibition of glycolysis, lactate transport, glycine synthesis/breakdown as well as lipid synthesis and catabolism.

Next, and to evaluate the biological significance of the metabolic shift observed following exposure to vemurafenib, and examine cell dependency on the various metabolic routes, we assessed the growth of BRAF^V600E SKMEL28 and BRAF^V600D WM266.4 melanoma cells under different nutrient-restricted conditions in the presence or absence of vemurafenib for 24 hours (WM266.4 cells), 48 hours (WM266.4 and SKMEL28 cells), and 72 hours (WM266.4 cells). The conditions were: control (5 mmol/L glucose), low glucose (1 mmol/L glucose), low glucose with glutamine deprivation (1 mmol/L glucose/no glutamine), and low glucose with glutamine and pyruvate deprivation (1 mmol/L glucose/no glutamine/no pyruvate). These conditions tested the dependence of cells on glycolysis, glutamine, and TCA metabolism, respectively. Cell numbers for both BRAF mutant cell lines relative to the seeding density are presented in Supplementary Fig. S4.

As shown in Fig. 4A and B, both control and treated samples exhibited significant reduction in cell counts when grown in low glucose (1 mmol/L) media relative to control conditions (5 mmol/L glucose) and even a greater fall when glutamine was removed after 24 hours (WM266.4 cells) and 48 hours (WM266.4 and SKEML28) of treatment. Importantly, however, the effect of nutrient deprivation was less dramatic in vemurafenib-treated cells, indicating that vemurafenib reduces the dependency of these cells on glucose and glutamine. There was no evidence for overt apoptosis (as indicated by the absence of cleaved PARP; Supplementary Fig. S4) following cell exposure to the nutrient-limited media with and without vemurafenib, indicating that the differences in cell counts observed here are related to growth rather than cell kill.

These results were corroborated for WM266.4 cells after 72 hours of treatment (Fig. 4A), confirming the growth advantage with vemurafenib under low glucose/no glutamine conditions. When pyruvate was removed in addition to glutamine under low glucose, higher cell counts were also observed in vemurafenib-treated WM266.4 compared with control cells at 24 hours, but this was abolished with prolonged exposure (48 hours) for both melanoma cell lines, consistent with the dependency of vemurafenib-treated cells on mitochondrial metabolism, with PC flux requiring pyruvate availability (29). The increased cell counts in treated relative to control WM266.4 cells under glucose and glutamine deprivation at 48 hours was abolished by the co-addition of phenylacetic acid (PAA), an inhibitor of PC (30) that led to 35.8% ± 16.4% reduction in PC activity (ref. 30; Fig. 4C), confirming the involvement of anaplerotic PC metabolism in the growth advantage conferred by vemurafenib.

Interestingly, the effect of nutrient deprivation on cell-cycle profiles, characterized in WM266.4 cells, was different in control- and vemurafenib-treated cells. Under control conditions (5 mmol/L glucose), treated cells showed a G₁ phase arrest, as expected (P = 0.042; ref. 31). Sequential removal of nutrient from control cells lead to a G₁ arrest coupled with a gradual decrease in the S phase population (Fig. 4D) relative to complete media conditions (5 mmol/L glucose). In contrast, BRAF inhibitor–treated cells (already G₁ arrested at baseline) showed a gradual increase in the G₂ phase with sequential nutrient removal (Fig. 4D; Supplementary Fig. S5), consistent with inhibition of the G₂–M cell-cycle checkpoint.

Taken together, these data suggest that vemurafenib reduces BRAF-mutant cell dependency on glucose (by downregulating glycolytic metabolism) and glutamine (by increasing PC anaplerotic flux) and imposes a G₂–M cell-cycle block under nutrient deprivation.

BRAF signaling inhibition with vemurafenib in BRAF-mutant WM266.4 and SKMEL28 cells led to depletion of MCT1, a transmembrane protein that mediates the bidirectional movement of monocarboxylic acids, such as lactate and pyruvate (28). We thus hypothesized that this effect should translate to a fall in hyperpolarized ¹³C-pyruvate–lactate exchange that can be detectable by ¹³C NMR spectroscopy and which could have potential as a new noninvasive biomarker of BRAF inhibition. This hypothesis was tested in live WM266.4 cells following 24-hour treatment with 2 μmol/L vemurafenib. As expected, based on MCT1 depletion after treatment, our data showed a significant decrease in the ratio of ¹³C-lactate_AUC/¹³C-pyruvate_AUC in vemurafenib-treated compared with control cells (to 64.3% ± 10.2%, P = 0.008), consistent with a fall in the ¹³C label exchange between pyruvate and lactate (ref. 21; Fig. 5A–C). In contrast, in BRAF^WT D04 cells, where MCT1 expression remained unchanged with vemurafenib treatment, there were no significant changes in hyperpolarized ¹³C-pyruvate–lactate exchange (Fig. 5D).

Thus, ¹³C-pyruvate–lactate exchange could serve as a noninvasive imaging biomarker for monitoring the downstream metabolic effects of vemurafenib in BRAF-mutant human melanoma cells.

BRAF and MEK inhibitors have shown unprecedented clinical responses in BRAF-mutant malignant melanoma (4, 5); however, the emergence of drug resistance remains inevitable. This stresses the need for a better understanding of the consequences of BRAF inhibition on key disease mechanisms in melanoma cells in order to develop biomarkers of response as well as combination strategies that will improve long-term disease control.

We and others have shown that inhibition of BRAF–MEK–ERK signaling in BRAF-mutant melanoma models activates mitochondrial metabolism and decreases lactate production through inhibition of HK2 and glucose transporter expression downstream of CMYC and HIF1α (14–16, 32). In this study, we sought to characterize the downstream alterations in metabolic pathways and fluxes triggered by BRAF inhibition and evaluate their significance for drug antiproliferative activity and potential as noninvasive biomarkers of response to treatment.

As expected, vemurafenib treatment in BRAF^V600D WM266.4 human melanoma cells led to a significant fall in Lactate^E that was concentration-dependent, and also recorded in an additional BRAF-mutant melanoma cell line (SKMEL28) but not in BRAF^WT CHL-1 or D04 human melanoma cells. Our previous work with a MEK inhibitor indicates that this effect is present only in mutant BRAF-driven cancer cells, being absent in mutant BRAF-expressing, but independent, cells and in nontransformed cells (14).

NMR metabolic profiling of WM266.4 cells indicated that, in addition to reduced lactate levels, vemurafenib treatment was associated with decreased acetate, increased glycine and myo-inositol, and a significant reduction in the fatty acyl chain content (0.9 ppm). The bioenergetic status of treated cells, as assessed by ³¹P NMR analysis of cellular NTP and PCr levels and bioluminescence-measured ATP/ADP, remained unaffected.

These findings indicate that, in addition to downregulation of glycolytic metabolism, BRAF inhibition alters glycine, myo-inositol, and lipid metabolism, inducing a metabolic shift that is able to maintain cellular energetic status probably by means of activating compensatory pathways, for example, OxPhos (16). Indeed, we observed increased ROS production (to a similar extent as in previous publications; ref. 33) following treatment with vemurafenib, indicating that the drug may also be activating OxPhos in our model, in agreement with earlier findings (16, 34).

Next, and to better understand the downstream alterations in metabolic flux involved in the vemurafenib-induced metabolic reprogramming, we evaluated cellular glycolytic flux with a widely reported method (35, 36), using a ¹³C-labeled glucose analogue. ¹³C NMR confirmed inhibition of de novo lactate formation and glucose utilization, as revealed by the fall in intracellular and extracellular [3-¹³C]lactate and accumulation in intracellular [1-¹³C]glucose in vemurafenib-treated compared with control cells, indicating decreased glucose utilization. Taking into account the decreased ¹³C label incorporated downstream of the glycolytic pathway, our data show a relative increase in the labeling of glutamate at position 4 (PDH flux) in treated relative to control cells. Furthermore, we observed an increase in [2-¹³C] and [3-¹³C]glutamate as well as the ratio of [2-¹³C]/[4-¹³C]glutamate in treated versus control cells, indicating increased mitochondrial metabolism via anaplerotic PC flux following exposure to vemurafenib. This metabolic shift was concomitant with significantly increased PC enzymatic activity (with a magnitude in the range of other reports in the literature; ref. 37) under BRAF inhibition.

It is noteworthy that the steady-state metabolite levels measured by ¹H NMR in WM266.4 cells (Table 1) are maintained following vemurafenib treatment. These metabolites represent total levels present in the cell, which are governed by many reactions and pathways (e.g., uptake from media and protein breakdown) as well as the net change between de novo synthesis and breakdown/utilization. In contrast, the metabolites detectable by ¹³C NMR are derived from de novo synthesis from ¹³C-glucose, which may not necessarily lead to changes in the total ¹H NMR-measured metabolite pool. Further, and despite the fall in glucose consumption and glycolytic activity, WM266.4 cells were able to maintain their energetic status, consistent with more efficient metabolism of glucose through the TCA cycle.

Molecular analysis of metabolic enzyme expression indicated that the most significant alterations observed with vemurafenib were, in addition to the previously reported decrease in HK2 expression (14, 15), a decrease in MCT1, MCT4 (involved in glycolytic and lactate metabolism), 3-PHGDH3 (serine–glycine metabolism), ACAD9 (fatty acid β-oxidation), and P-ACL (lipid biosynthesis). Although PC mRNA levels remained unchanged with BRAF inhibitor treatment, we cannot rule out changes in PC protein expression (resulting from posttranscriptional regulation) or allosteric regulation as potential contributing factors to the elevated PC activity, as previously described (38).

The ¹³C flux and molecular findings provide key insights into the mechanisms underlying the changes observed in steady-state metabolites (Table 1). Our results are consistent with a model (summarized in Fig. 5E) in which the increase in myo-inositol observed by both ¹H and ¹³C NMR suggests increased de novo synthesis from glucose following the reduction in HK2 flux. Further, the significant decrease in ¹³C-lactate and rise in ¹³C-labeled aspartate and glutamine/glutamate and ¹³C position labeling patterns observed indicate diversion of glucose from glycolysis to TCA cycle metabolism (primarily via PC, but also PDH flux). Under these conditions, acetyl-CoA utilization would be accelerated leading to reduced acetate pool observed by ¹H NMR. The tracing of glycine synthesis was not possible with [1,¹³C]glucose; however, it is unlikely that its accumulation in vemurafenib-treated cells is due to increased de novo synthesis from serine, as metabolic precursors derived from glycolysis (including the first intermediate in serine–glycine synthesis 3-phosphoglycerate) are reduced by treatment. Accordingly, the build-up in glycine is more likely due to inhibition of its breakdown, which would be consistent with the decrease in GCAT mRNA expression (39).

The upregulation of mitochondrial PC flux is of interest because melanoma cells are known to have a functional TCA cycle but with negligible PC anaplerotic metabolism (40). Activation of PC flux in glioblastoma and non–small cell lung cancer cells has previously been linked to reduced dependency on glutamine (35, 37). Indeed, we observed that vemurafenib reduces BRAF-mutant cell dependency on glucose and glutamine but commits them to consume pyruvate, that becomes essential under BRAF inhibition, as previously described for cells dependent on upregulated PC flux (29). The growth advantage conferred by vemurafenib under nutrient-depleted conditions was abolished with pharmacologic inhibition of PC activity, consistent with PC involvement in this effect.

The fact that treated cells grow better in nutrient-restricted media, which are relevant to tumor growth conditions in vivo, may facilitate the emergence of drug-resistant clones in vivo that can lead to tumor relapse under treatment (41, 42). Such differences could determine whether vemurafenib has a growth inhibitory effect in a tumor or not, which is clearly of huge clinical importance. Thus, targeting this metabolic adaptation in combination with BRAF signaling may offer a promising strategy for counteracting drug-induced growth advantage. In fact, inhibiting OxPhos with metformin (43) and mitochondrial respiration inhibitors (44) has already been shown to potentiate the therapeutic efficacy of BRAF inhibitors in human melanoma models. With regard to our findings, inhibition of PC in glutamine-independent glioblastoma models was able to inhibit tumor growth, demonstrating the importance of this metabolic route for cell survival (28). However, it remains to be established whether such effects would be applicable in melanoma tumor models, and whether PC blockade using selective pharmacologic or genetic approaches can enhance the potency of BRAF-targeted therapies in vivo.

Finally, vemurafenib depletes MCT1, a bidirectional transporter for monocarboxylic acids such as lactate and pyruvate, resulting in reduced hyperpolarized ¹³C-pyruvate–lactate exchange in live BRAF^V600D WM266.4 human melanoma cells but not in BRAF^WT D04 cells. These experiments are translatable to in vivo imaging studies (45) and provide proof of principle for developing ¹³C-pyruvate–lactate exchange as a noninvasive metabolic imaging biomarker of the molecular consequences of BRAF signaling inhibition. Hyperpolarized ¹³C-pyruvate–lactate exchange has been shown to occur at a very low rate in normal compared with cancer tissues (46), and given the results obtained with BRAF^WT D04 cells, we anticipate that this assay will be most useful for monitoring therapeutic response to BRAF inhibition in BRAF-mutant melanoma. Future work will aim to assess the translatability of our findings to in vivo tumor models.

Dynamic imaging of metabolic processes using hyperpolarized ¹³C NMR has recently entered the clinic with ¹³C-pyruvate–lactate exchange measurements being the first to be reported in prostate cancer patients, and many more studies are ongoing to assess the value of this approach for cancer imaging (47). This technique can be combined with multiparametric magnetic resonance imaging of the tumor microenvironment (e.g., cellularity, vascularity, and pH) and complemented with FDG-PET to provide information on different steps of the glucose metabolic pathway. The availability of several response biomarkers can be valuable in applying the Pharmacological Audit Trail to drug development preclinically and in patients as in the case of vemurafenib (48), providing a more robust means for assessing drug effects in patients with a greater degree of confidence (49), allowing better evaluation of the downstream metabolic consequences of BRAF signaling inhibition in cancer and more effective monitoring of therapeutic response (48, 49).

In conclusion, we show that BRAF inhibition with vemurafenib in BRAF-mutant human melanoma cells alters glucose utilization, leading to inhibition of lipid synthesis (and breakdown) and activation of oxidative and anaplerotic mitochondrial metabolism with consequences that confer a growth advantage under nutrient-deprived conditions. We also show that BRAF inhibition in BRAF-mutant cells leads to depletion of MCT1 and inhibition of hyperpolarized ¹³C-pyruvate–lactate exchange, providing support and a rationale for exploring this metabolic process as a potential noninvasive metabolic imaging biomarker of therapeutic response to BRAF signaling–targeted drugs.

R. Marais has received expert testimony from ICR rewards to inventors scheme. P. Workman reports receiving other commercial research support from Basilea and has received expert testimony/research funding to cancer therapeutics Unit and ICR of which he is CEO and President—also potential IP revenues to ICR. No potential conflicts of interest were disclosed by the other authors.

Conception and design: T. Delgado-Goni, M.F. Miniotis, P. Workman, M.O. Leach, M. Beloueche-Babari

Development of methodology: T. Delgado-Goni, M.F. Miniotis, M. Beloueche-Babari

Acquisition of data (provided animals, acquired and managed patients, provided facilities, etc.): T. Delgado-Goni, M.F. Miniotis, S. Wantuch, H.G. Parkes

Analysis and interpretation of data (e.g., statistical analysis, biostatistics, computational analysis): T. Delgado-Goni, M.F. Miniotis, H.G. Parkes, M. Beloueche-Babari

Writing, review, and/or revision of the manuscript: T. Delgado-Goni, M.F. Miniotis, H.G. Parkes, P. Workman, M.O. Leach, M. Beloueche-Babari

Study supervision: P. Workman, M.O. Leach, M. Beloueche-Babari

Other (provision of reagents and intellectual input): R. Marais

T. Delgado-Goni and S. Wantuch are supported by MRC project grant (MR/K011057/1), H.G. Parkes, M. O. Leach, and M. Beloueche-Babari are supported by a CRUK Centre for Cancer Imaging grant C1090/A16464. P. Workman is supported by CRUK program grant (C309/A11566). M. Falck Miniotis was funded by an EPSRC Platform grant EP/H046526/1. We also acknowledge grant C1060/A10334 from CRUK and EPSRC Cancer Imaging Centre in association with the MRC and Department of Health (England). P. Workman is a Cancer Research UK Life Fellow (C309/A8992). M.O. Leach is a NIHR Biomedicine Research Senior Investigator.

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.