Acetylsalicylic Acid Governs the Effect of Sorafenib in RAS-Mutant Cancers

This article is featured in Highlights of This Issue, p. 987

Mutant neuroblastoma RAS viral (v-ras) oncogene homolog (NRAS) was the first oncogene identified in melanoma (1) and it is now known that approximately 20% of all melanomas harbor mutations in NRAS, 2% in KRAS and 1% in HRAS (2). While KRAS and HRAS only play a minor role for melanoma, KRAS in particular is frequently mutated in other cancers, including lung, colon, and pancreatic carcinomas (3). Mechanistically, RAS proteins are GTPases that activate downstream signaling pathways involved in proliferation and cell survival upon GTP binding (3). Genetic mutations are located in codons 12, 13, and 61, and more than 80% of mutant NRAS harbor a mutation at codon 61 (4). Mutations result in reduced GTPase activity, which causes preferential binding of GTP and therefore constitutive activation of RAS signaling (5). In recent years, the advent of targeted therapies has advanced melanoma treatment but focused on mutant BRAF, whereas no strategies directly targeting mutant NRAS have been approved (6, 7). Inhibitors for RAS are particularly difficult to develop (8, 9) leaving low-response chemotherapy (10) or immunotherapy (11) with high toxicity rates as therapeutic strategies for patients with NRAS-mutant melanoma. Attempts to directly target RAS mutations include farnesyl transferase inhibitors, which are supposed to prevent posttranslational modifications required for the integration of RAS into the plasma membrane (12) thus preventing the interaction of RAS and the prenyl-binding protein PDEδ (PDE6D) (13). Unfortunately, farnesyl transferase inhibitors showed disappointing results in clinical settings (12) and inhibitors of PDEδ binding to farnesylated KRAS still require more development to optimize the drugs (14). Other attempts to target RAS-mutant cancers include blocking RAS downstream targets. Inhibition of MEK (MAP2K1, MAP2K2) (6, 15) or MEK in combination with PI3K (PIK3CA)/MTOR inhibitors have shown promising results (10). However, inhibition of MEK leads to the development of resistance, similar to strategies in BRAF-mutant melanoma (16). Several mechanisms of resistance have been proposed for NRAS-mutant melanoma including PDGF receptor β signaling (PDGFRB) (17) emphasizing the importance of novel single or combination therapies for sustained treatment. Sorafenib (Nexavar, BAY 43–9006; Bayer Healthcare Pharmaceuticals) is a multikinase inhibitor that targets both CRAF (RAF1) and BRAF as well as the VEGFR family (KDR and FLT4) and platelet-derived growth factor receptor family (PDGFRB; ref. 18), among others. Sorafenib is FDA-approved for the treatment of advanced renal cell carcinoma and patients with unresectable hepatocellular cancer (19) with most common adverse events being skin rashes, diarrhea, and alopecia (20). In addition, the development of squamous cell carcinomas and keratoacanthomas has been reported (21, 22). These side effects resulted in patients requiring dose reductions, interrupting or discontinuing therapy raising concerns about the toxicity, efficacy, and safety of sorafenib (20). One strategy to overcome drug-induced toxicity is to combine sorafenib with other drugs to reduce the effective dose required to trigger a tumor-specific response without inducing systemic toxicity. Several studies have explored possible anticancer effects of sorafenib in combination with other targeted inhibitors or radiation, which showed limited efficacy (reviewed in ref. 23).

Nonsteroidal anti-inflammatory drugs (NSAID) have been reported to reduce overall cancer risk including prostate (24), colorectal (25), and skin cancer (26). The association between NSAIDs and melanoma risk is less clear, and several studies yielded conflicting results (27). Acetylsalicylic acid (Aspirin) is an intriguing agent as it is one of the most widely used drugs. Synergistic effects of sorafenib and aspirin have already been described in RAS wild-type hepatocellular carcinoma where it has been reported to reduce the prometastatic effect of sorafenib monotherapy (28).

Here we show that combined sorafenib and aspirin resulted in synergistic cytotoxicity in RAS/RAF–mutant cancers including NRAS-mutant melanoma by simultaneously activating the AMP-activated protein kinase (AMPK) and mitogen-activated protein kinase (MAPK/ERK) pathways. The combined treatment, which is effective in vivo, allows the concentration of sorafenib to be substantially reduced with the likelihood of less tissue toxicity. These data provide a rationale for the application of this combination in RAS-mutant cancer patients.

KRAS-mutant human lung adenocarcinoma cell lines A549 and H358 as well as RAS/RAF–wild-type immortalized human keratinocytes HaCaT were kindly provided Dr. Gerald Hoefler (Institute of Pathology, Medical University of Graz, Graz, Austria). The RAS/RAF-wild type human breast cancer cell line SkBr3 was kindly provided by Dr Fiona Simpson (The University of Queensland Diamantina Institute, Brisbane, Queensland, Australia). All cell lines are routinely tested for mycoplasma as described previously (29, 30) and were authenticated in 2016 by the analytic facility of the QIMR Berghofer Medical Research Institute (Brisbane, Australia) via STR fingerprinting. All experiments were performed within 3 months after thawing the respective cell lines. Cells were grown in RPMI1640 medium (Sigma-Aldrich), supplemented with 5% FBS (Assay Matrix) and 2% l-glutamine (Life Technologies) and maintained at 37°C in a humidified atmosphere containing 5% CO₂. Cells were harvested for individual experiments after washing with PBS (pH 7.4; Life Technologies) using trypsin (Life Technologies).

Lentiviral vectors containing shRNA targeting AMPK1/2 (PRKAA1/PRKAA2) (shPRKAA1, NM_006251, CloneID TRCN0000000861; shPRKAA2 NM_006252, CloneID TRCN0000002171), or BRAF (NM_004333, CloneID TRCN0000231130) were purchased from Sigma-Aldrich. Cells were prepared in 12-well plates to reach confluence of 50%–80%, pretreated with 8 μg/mL polybrene (Sigma-Aldrich) for 2 hours followed by the addition of 25 μL of the viral supernatant. Transduced cells were subjected to selection with puromycin (Sigma-Aldrich) at a concentration of 5 μg/mL 72 hours post transfection. Cells were then maintained in medium containing 1.5 μg/mL.

Whole-cell lysates were generated using RIPA buffer (Sigma-Aldrich) supplemented with 1% protease inhibitor cocktail (Active Motive). Protein lysates from frozen tissue samples were generated using 500-μL RIPA buffer per 10 mg of tissue and sonicated at 180 watts 10 × 10 seconds. The protein concentration was measured using Bradford Protein Assay (Bio-Rad). Fifteen micrograms of protein were separated on a 6%–10% SDS-polyacrylamide gel followed by transfer to a polyvinylidene difluoride membrane (Bio-Rad). The membrane was then probed for the protein of interest with the specific primary and the corresponding peroxidase-conjugated secondary antibodies (Supplementary Table S1). Proteins were visualized using Amersham ECL Prime Western Blotting Detection Reagent (GE Healthcare) and scanned on an LI-COR C-DiGit Blot Scanner. The membranes were stripped and reprobed using Restore Plus Western Blot stripping buffer (Life Technologies) as required in the individual experiments. Immunoblots were quantified using ImageJ and the ratio of phosphorylated to total protein was calculated and normalized to control of the same immunoblot.

Cells were exposed to drugs for 24 hours as indicated in the respective experiments, lysed using 1× Cell Lysis Buffer (Cell Signaling Technology) and sonicated on ice three times for 5 seconds. Cell lysate was subject to a precleaning step by incubating with Protein G Agarose Beads (Cell Signaling Technology) for 1 hour at 4°C with gentle rocking followed by the addition of 2 μL of c-Raf-1 antibody (BD Transduction Laboratories) and incubated overnight at 4°C. Immunoprecipitation was performed by adding 10μL Protein G Agarose Beads (Cell Signaling Technology) per 100-μL cell lysate and incubated for 3 hours. Coimmunoprecipitation was then assessed by immunoblotting using specific antibodies for CRAF and BRAF (Supplementary Table S1)

Sorafenib, trametinib, dabrafenib, and LY3009120 were purchased from Selleck Chemicals. Dorsomorphin (compound C), A-769662, SC-560, and metformin were purchased from Cayman Chemical. Celecoxib and isobutylphenylpropanoic acid were purchased from Sigma Aldrich.

Caspase-3 activation was assessed using the Active Caspase 3 apoptosis kit (BD Biosciences) following the manufacturer's protocol. The samples were analyzed with a BD ACCURI C6 PLUS from the Translational Research Institute (TRI) FACS core facility.

A total of 1 × 10⁴ cells were seeded in 96-well culture plates for allocated times as mentioned in the experiments. Cells were treated with drugs depending on the experimental setup 48 hours after the initial seeding and subsequently incubated with MTT (3-(4, 5-dimethylthiazolyl-2)-2,5 diphenyltetrazolium bromide, Thermo Fisher Scientific) reagent (1/10 dilution in full growth medium) at 37°C for 4 hours. After incubation, 100 μL DMSO was added and incubated for 10 minutes in the incubator. Absorbance was measured at 540 nm using a microtiter plate reader. All experiments were done in triplicate or duplicate and a final concentration of 1% DMSO was used as control.

Cells, which have been exposed to drugs for various time points, were fixed with 4% paraformaldehyde, followed by 30-minute incubation with 0.1% crystal violet in 4% paraformaldehyde. The plates were washed and imaged using a Chemi Doc TM XRS Universal Hood (Bio-Rad). A final concentration of 1% DMSO was used as control.

Cell viability was assessed using the AlamarBlue assay and synergy was assessed using the Bliss independence model as described previously (31). Briefly, data were normalized to doxorubicin and DMSO controls and converted to fraction-affected values (F). Next, the predicted inhibition values (P) were calculated: [P = Fa + Fb – Fab 0<P > 1]. Predicted F equals the fraction affected by compound “a” (Fa) at concentration × plus the fraction affected by compound “b” (Fb) at concentration y minus the product of the two (Fab). The difference between the predicted additive fraction affected and the experimentally observed fraction affected is the Bliss number. Positive value indicates synergy, a negative value indicates antagonism, and an overlap of predicted and observed combination effects gives a Bliss number of zero and indicates additivity. Because of the nature of the assay, we detect a background noise level of ± 15%.

All animal experiments were performed in accordance with institutional guidelines under Wistar IACUC protocol 111954 in NSG (NOD.Cg-Prkdcscid Il2rgtm1Wjl/SzJ, Jackson Laboratory) mice or in accordance with institutional guidelines of UQ animal ethics committee ethics number: SOM/TRI/197/15/DRC in C.B-17/IcrHanHsdArcPrkdcscid mice (Animal Resources Centre Canning Vale). Animals were inoculated subcutaneously with 1 × 10⁶ human WM1366 melanoma cells in a 100 μL suspension of Matrigel (BD Matrigel)/complete media at a ratio of 1:1 or with 2.5 × 10⁶ human A549 lung adenocarcinoma cells suspended in complete media. After tumors had reached a volume of 100–200 mm³, mice were randomized into groups of 4 mice as indicated in the respective experiments: vehicle (0.75% hydroxyl methyl cellulose/25% ethanol/10% DMSO), sorafenib 30 mg/kg or 15 mg/kg, aspirin 100 mg/kg, or 200 mg/kg, metformin 100 mg/kg, sorafenib + aspirin and sorafenib + metformin. Tumor size was assessed multiple times per week using a caliper. Animals were sacrificed at the experiment endpoint or when tumor volume exceeded the ethical limit. Tumors were then harvested, snap frozen in liquid nitrogen, and stored at −80°C until further processing.

Tumors were formalin-fixed, paraffin-embedded, and stained using a Discovery Ultra Ventana in the TRI Histology Core Facility. Deparaffinzation and antigen retrieval was performed before using prediluted antibodies (Supplementary Table S1) or rabbit IgG as negative control (Invitrogen). Hematoxylin and eosin (H&E) staining was performed according to common methods. Staining intensity was assessed and immunoreactivity was calculated as described previously (32). Slides were scanned using the Olympus VS120 slide scanner (20×).

Represented data are expressed as arithmetic mean ± SD of three independent experiments. Unpaired t test has been used to determine statistical significance. N.s. indicates a P value > 0.05; *, P, ≤ 0.05; **, indicates a P ≤ 0.01; ***, P ≤ 0.001

The NRAS^Q61K-mutant melanoma cell line WM1366 was nonresponsive to sorafenib at low concentrations (0.25–1 μmol/L) but showed an almost complete loss of viability at 5 μmol/L (Fig. 1A). Exposure to aspirin alone at concentrations up to 2 mmol/L did not affect cell viability (Fig. 1A). The combination of sorafenib and aspirin, however, showed a dose-dependent toxicity, with effects of the combination observed with sorafenib concentrations as low as 250 nmol/L and 2 mmol/L aspirin (Fig. 1B). The efficacy of this combination was confirmed by MTT assay, which showed a significant reduction of cell proliferation after 72 hours of drug exposure compared with single exposure (Supplementary Fig. S1A). Next, we investigated the combination of sorafenib and aspirin using the Bliss Independence Model (33) and found synergistic efficacy in a wide range of combined concentrations of both drugs (Fig. 1C). The combination of sorafenib- and aspirin-induced apoptosis at a similar level to sorafenib 5 μmol/L, indicated by the level of active caspase-3 (CASP3) (Fig. 1D). The efficacy of combined sorafenib and aspirin was confirmed in three other NRAS-mutant melanoma cell lines (Supplementary Fig. S1B). Combining sorafenib with other NSAIDs, the specific COX2 (PTGS2) inhibitor celecoxib, the specific COX1 (PTGS1) inhibitor SC-560, or the nonspecific COX1 and COX2 inhibitor isobutylphenylpropanoic acid (Ibuprofen) showed no synergistic toxicity (Supplementary Fig. S1C–S1E), suggesting that the effect of sorafenib and aspirin is independent of the known aspirin targets COX1 and COX2. Taken together, sorafenib at low concentrations in combination with aspirin synergistically induces apoptosis in NRAS-mutant melanoma independently of COX1 and COX2 inhibition.

We expanded our investigation to cell lines harboring KRAS^G12 mutations, investigating the effects of combined sorafenib and aspirin treatment in the lung carcinoma cell line A549. The dose–response profile showed that A549 is insensitive to single-agent sorafenib and aspirin (Fig. 1E); however, the combination produced a dose-dependent response with an effective sorafenib concentration as low as 500 nmol/L when combined with 2 mmol/L aspirin, and a strong cytotoxic effect at 1 μmol/L sorafenib and 2 mmol/L aspirin (Fig. 1F). We confirmed the efficacy of the combination by MTT assay, which showed a significantly reduced cell proliferation over 72 hours (Supplementary Fig. S1F). The combination was also effective in the KRAS^G12-mutant lung cancer cell line H358 (Supplementary Fig. S1G), suggesting that RAS-mutant cell lines in general are susceptible to this combination. Furthermore, sorafenib/aspirin strongly reduced cell proliferation and viability in BRAF-mutant melanoma cell lines WM164 and WM983B as measured by MTT (Fig. 1G) and crystal violet staining (Supplementary Fig. S1H). In contrast, RAS/RAF wild-type keratinocytes (HaCaT) or the RAS/RAF–wild-type breast cancer cell line SkBr3 only showed a minor response to combined sorafenib/aspirin (Fig. 1H). Similarly the BRAF/NRAS–wild-type melanoma cell line D24 was only moderately affected by the combination (Supplementary Fig. S1I), suggesting that the cytotoxic effects of sorafenib and aspirin are specific for RAS/RAF–mutant cancers. The mutation status and sensitivity to combined sorafenib/aspirin of all tested cell lines is summarized in Table 1.

Mutant RAS results in increased RAF/MEK/ERK signaling, while sorafenib inhibits BRAF and CRAF among others, thereby inhibiting MAPK signaling (34). In addition to inhibiting COX1/2, aspirin has been shown to increase AMPK signaling (35). We therefore investigated these pathways in sorafenib/aspirin–exposed cells. Contrary to our expectations, the combination treatment resulted in the increased activation of ERK1 (MAPK3) and ERK2 (MAPK1) and the phosphorylation of the AMPK substrate acetyl-coenzyme A carboxylase (ACC/ACACA/ACACB) in NRAS- (Fig. 2A), KRAS- (Fig. 2B), and BRAF-mutant (Fig. 2C and D) cells. In contrast, RAS/RAF-wild-type keratinocytes (HaCaT; Fig. 2E) and BRAF/NRAS–wild-type melanoma cells (D24; Fig. 2F), which are both less sensitive to the combination of sorafenib and aspirin, showed no or only a subtle ERK and/or AMPK pathway activation, suggesting that these pathways could be crucial for the cytotoxic effects of the combination treatment. Sorafenib at a concentration of 5 μmol/L, which showed toxicity in WM1366 cells, also resulted in an increase of pERK and pACC (Fig. 2A), suggesting that single-agent–mediated toxicity affects similar pathways in these cells. Furthermore, activation of the AMPK and MAPK pathways was already observed after 4 hours of treatment (Supplementary Fig. S2). Because there is a significant lack of specific treatment options for RAS-mutant–driven cancers, and RAS- and RAF-mutant cancer cells showed the same pathway activation pattern, we conducted all further experiments in NRAS- and KRAS-mutant cell lines. To test for the contribution of both pathways, we blocked them using the selective MEK inhibitor trametinib and the AMPK inhibitor compound C (dorsomorphin). While single treatment with the MEK inhibitor resulted in a dose-dependent increase in toxicity, the inhibition of MEK reduced sorafenib and aspirin-induced toxicity in NRAS- and KRAS-mutant cells (Fig. 3A). Similarly, compound C showed a dose-dependent toxicity while rescuing the cells if combined with sorafenib and aspirin (Fig. 3B) confirming the importance of activated AMPK and ERK signaling for the toxicity of the combination treatment. Simultaneous inhibition of AMPK and MEK in cells treated with the combination therapy increased cell viability over inhibition of either pathway alone (Supplementary Fig. S3A and S3C). The decreased toxicity of sorafenib and aspirin after inhibition of MEK and/or AMPK was also confirmed by MTT assays (Supplementary Fig. S3B and S3D), even though no additional benefit of combining trametinib and compound C was detected using this assay. The observations of the crystal violet stainings suggest that AMPK and ERK1/2 activation are independent. Indeed, we found that treatment with sorafenib, aspirin, and trametinib resulted in activated AMPK signaling but blocked ERK signaling, whereas treatment with sorafenib, aspirin and compound C resulted in decreased AMPK signaling but activated ERK1/2 (Fig. 3C). This indicates that AMPK and ERK pathway activation are independent events that synergistically mediate cytotoxicity of the combination treatment. To confirm the specificity of the observations using pharmacologic inhibitors, we used sequence-specific shRNAs to silence BRAF and AMPKα1/2 (PRKAA1/PRKAA2) in NRAS-mutant melanoma cells (WM1366). BRAF- or AMPKα1/2-silenced cells showed decreased sensitivity to sorafenib and aspirin compared with empty vector–transduced control cells (Fig. 3D and E) suggesting that both of these proteins are involved in sorafenib/aspirin–mediated cytotoxicity. It has been shown that BRAF inhibitors can induce paradoxical MAPK pathway activation by promoting BRAF/CRAF dimerization in a RAS-dependent manner in RAS-mutant and RAS/RAF–wild-type cancers (36). CRAF immunoprecipitation showed that sorafenib in combination with aspirin, but not single-agent treatment, induced strong BRAF/CRAF complex formation in NRAS-mutant WM1366 (Fig. 3F) and BRAF-mutant WM164 melanoma cells (Fig. 3G). Interestingly, BRAF shows an electrophoretic mobility shift in WM1366 that recently has been linked to phosphorylation and increased activity of BRAF as part of a high molecular weight complex in RAS-mutant cancer cells (37). We then tested the involvement of BRAF/CRAF complexes for the activation of the ERK pathway by using the pan-RAF inhibitor LY3009120 that has been shown to inhibit active dimers (38). Similar to trametinib, LY3009120 reduced the toxicity of sorafenib and aspirin profoundly in both NRAS- and KRAS-mutant cells (Fig. 3H), suggesting that RAF activation is required for sorafenib- and aspirin-induced cytotoxicity. Combining sorafenib, aspirin, and LY3009120 significantly rescued cell proliferation as determined by MTT assays confirming previous findings (Supplementary Fig. S3E and S3F). LY3009120 alone or in combination with aspirin resulted in moderate activation of ERK signaling, which, even though observed at a higher concentration, is in line with previous reports (38). The combination of LY3009120 with sorafenib and aspirin blocked hyperactivation of ERK signaling without inhibiting AMPK pathway activity (Fig. 3I), again suggesting that MAPK and AMPK pathway activation are independent from each other. Taken together, combined sorafenib and aspirin simultaneously hyperactivate ERK and AMPK signaling, which both contribute to decreased cell viability in RAS-mutant cancers.

As a proof of principle, we tested the combination of aspirin with the BRAF^V600E-specific drug dabrafenib, which has been reported to paradoxically activate ERK in NRAS-mutant melanoma (9, 39). Combining dabrafenib at concentrations that are 100–1,000 times higher than usually used for melanoma (40), with aspirin resulted in a profound decrease of cell viability in RAS-mutant cells (Fig. 4A; Supplementary Fig. S4A). Only the combination induced hyperactivation of ERK and AMPK signaling, whereas dabrafenib alone, which showed no effect on cell viability, only resulted in hyperactivated ERK (Fig. 4B). Similar to dabrafenib in combination with aspirin, dabrafenib combined with the antidiabetic drug metformin, which is known to be an AMPK activator, also exerted cytotoxic effects (Supplementary Fig. S4B) and activated ERK and AMPK signaling (Fig. 4B). We then tested whether sorafenib in combination with other AMPK activators is also effective. Interestingly, the combination of sorafenib with metformin showed no synergistic effect in NRAS- and KRAS-mutant cells (Fig. 4C; Supplementary Fig. S4C). On the molecular level, combined sorafenib and metformin resulted in activation of the AMPK pathway but only modest activation of ERK signaling (Fig. 4D). We then hypothesized that the mechanistic differences of AMPK activation of metformin and aspirin might influence the synergistic effects. Metformin is an indirect AMPK activator as it inhibits oxidative phosphorylation (41) while aspirin interacts directly with AMPK. To account for this mechanistic difference, we combined sorafenib with the allosteric AMPK activator A-769662, which has been shown to activate AMPK by interacting with the same protein region as aspirin (35). This combination also failed to decrease cell viability in RAS-mutant cancers (Fig. 4E). Like metformin, the combination with A-769662 resulted in AMPK pathway activation while failing to trigger hyperactivated ERK signaling (Fig. 4F). Similar to the in vitro results, the combination of sorafenib and aspirin showed a significant synergistic effect in NRAS- and KRAS-mutant xenograft mouse models (Fig. 5A and B). In comparison, the combination of sorafenib and metformin failed to decrease tumor growth (Supplementary Fig. S5A) confirming the superior efficacy of the sorafenib and aspirin combination. IHC analysis of the A549-derived tumors showed that the majority of the sorafenib or sorafenib/aspirin-treated tumors consisted of necrotic/dead cells with only the periphery of the tumors showing p-ERK positivity (Supplementary Fig. S5B–S5F). By applying an immunoreactivity score, it was found that sorafenib/aspirin–treated tumors displayed a higher staining intensity compared with sorafenib-treated tumors (Fig. 5C). Immunoblotting of A549-derived xenografts tumors treated with three doses of combined sorafenib/aspirin or vehicle (n = 3) further showed significant activation of AMPK signaling, indicating a similar pathway activation pattern as observed in vitro (Fig. 5D). These data demonstrate that sorafenib in combination with aspirin is a suitable strategy to target RAS-mutant cancers, whereas other AMPK activators such as metformin or A-769662 show no synergistic antitumorigenic effects in combination with sorafenib (Fig. 5E).

The identification of the prominent melanoma driver mutation BRAF^V600E was followed by the development of clinically effective drugs specifically targeting this mutation. Despite the emergence of resistance to these drugs, treatment significantly improves patient survival and is viewed as a model for the development of oncogene-directed targeted therapies (42). Intensive research efforts have been undertaken to develop drugs specifically targeting mutant NRAS, the second-most common driver mutation of melanoma, but no agents have been approved by the FDA to date (6, 7). Besides the lack of effective targeted therapies, mutant NRAS is also correlated with shorter survival after diagnosis of late-stage disease compared with NRAS/BRAF–wild-type or BRAF-mutant melanoma (4), highlighting the aggressive nature of melanomas driven by this mutation. Here we report the novel combination of sorafenib and aspirin to target RAS-mutant cancers. Considering that the clinically achievable sorafenib plasma concentration of approximately 5 μmol/L is associated with severe adverse effects (43), reducing the effective sorafenib dosage by combining it with clinically achievable aspirin concentrations (44, 45) could be a promising strategy to overcome toxicity issues. However, adverse effects triggered by aspirin and strategies to prevent these adverse effects must also be considered (46). The best characterized mechanism of action of aspirin is the inhibition of the cyclooxygenase enzymes COX1 and COX2, both of which are acetylated at a serine residue in the active site of the enzyme, abolishing enzyme activity (47). Interestingly, the combination with other NSAIDs, SC-560, ibuprofen, or celecoxib which inhibit COX1 and/or COX2 (48, 49), showed no synergistic toxicity suggesting a mechanism independent of COX1/2 inhibition. While epidemiologic studies provide compelling evidence that NSAIDs are associated with reduced risk of cancer (50), some reports also suggest that aspirin rather than other NSAIDs show a protective effect in some cancer types (51). Indeed, several reports have suggested that the cancer preventive action of the selective COX2 inhibitor celecoxib is the result of COX2-independent effects (52, 53). Mechanistically, inhibition of COX1/2 by aspirin is unlike other NSAIDs and the result of a covalent irreversible modification (49). While acetylation of COX1 completely blocks enzyme activity, acetylation of COX2 modifies the enzyme activity leading to the generation of lipoxins (54), which inhibit cell proliferation and angiogenesis of colorectal cancer (55). While a contribution of such lipoxins to the observed toxicity of sorafenib and aspirin cannot be excluded, the lack of synergy of sorafenib with other NSAIDs that target COX1/2 strongly suggests that the effect of aspirin is independent of inhibition of cyclooxygenase enzymes. Recently it has been shown that aspirin-derived salicylate activates AMPK by directly binding to and allosterically inhibiting dephosphorylation of AMPK (35). This aspect of aspirin is intriguing as activation of AMPK has been suggested to have antitumorigenic effects in certain contexts (56), especially in melanoma (57). Accordingly, constitutive activation of MAPK signaling by mutant BRAF has been shown to inhibit LKB1 (STK11) via ERK and p90Rsk (RPS6KA1), inhibiting AMPK activation in response to energy stress and promoting melanoma cell proliferation (58, 59). Furthermore, the combination of AMPK activators, like metformin or phenformin, with MAPK pathway inhibitors has been reported to be an effective strategy to increase treatment response in BRAF-mutant melanoma (60). Thus, it was surprising to see that neither metformin nor A-769662 showed synergistic effects in combination with sorafenib. In our experiments, aspirin alone did not activate AMPK signaling, which could be the direct result of LKB1-AMPK uncoupling. Sorafenib on the other hand, has been reported to activate AMPK in an LKB1 and/or CAMKK2-dependent manner (61), which, together with the aspirin-mediated stabilization of the active state of AMPK could result in the strongly increased activation of the AMPK pathway. shRNA-mediated silencing of AMPK profoundly reduced the cytotoxic effects of sorafenib and aspirin cotreatment, suggesting that AMPK activation is necessary but not sufficient to increase the cytotoxicity of sorafenib. Inhibition of ERK signaling also reduced the cytotoxicity of the combination, and the ability of the pan-RAF inhibitor LY3009120 and BRAF silencing to block the cytotoxicity of the combination, demonstrates the importance of BRAF in the observed hyperactivation of ERK signaling. Considering that both dominant melanoma driver mutations result in constitutive activation of ERK signaling and that sorafenib is a multikinase inhibitor that is supposed to inhibit the MAPK pathway (62), the involvement of hyperactivated MAPK signaling for the cytotoxicity of the combination treatment was unexpected. However, it is well known that sorafenib induces the observed BRAF/CRAF dimerization in BRAF-wild-type cells, which can lead to paradoxical activation of the ERK signaling pathway (36, 39, 63). We also observed sorafenib/aspirin but not sorafenib or aspirin alone, mediated BRAF/CRAF dimerization in BRAF-mutant cells which has not been described for more specific BRAF inhibitors like PLX4720 (36). It is noteworthy that unlike other ATP-competitive RAF inhibitors, sorafenib binds and stabilizes RAF in a different conformation (64). It is possible that this difference in the binding mode is important for the observed synergy between sorafenib and aspirin and the resulting increased BRAF/CRAF dimerization in BRAF-mutant cells. It is noteworthy that other mechanisms like dysregulation of the complex negative feedback regulation of Dual-specificity phosphatases (DUSP; ref. 65) could also be involved in the hyperactivation of ERK signaling and therefore contribute to sorafenib/aspirin induced antitumorigenic effects. While the detailed downstream effects of sorafenib/aspirin mediated hyperactivation of ERK and AMPK signaling remain elusive, it has been recently shown that hyperactivated ERK signaling induces apoptosis in BRAF–mutant cancers (66), resulting in an oncogene-induced senescence and negative selection of RAF and RAS double mutations (67) or inhibit melanoma growth and induce autophagy in vitro and in vivo (68). Beside the effectiveness of sorafenib in combination with aspirin to target NRAS-mutant melanoma, KRAS-mutant lung adenocarcinoma cell lines also proved to be sensitive to this combination. Sorafenib has recently been tested in a phase II clinical trial for the treatment of non–small cell lung cancer (NSCLC) with KRAS mutations, which showed relevant but modest clinical activity in patients (69). Our results suggest that the use of sorafenib in combination with aspirin is a promising strategy for the treatment of KRAS-mutant NSCLC. Taken together, we identified molecular drivers mediating the cytotoxic effects of sorafenib in combination with aspirin. This work provides a strong rationale to repurpose two clinically approved drugs to treat cancer types that lack specific therapeutic options.

No potential conflicts of interest were disclosed.

Conception and design: H. Hammerlindl, D. Ravindran Menon, D. Thakkar, H. Schaider

Development of methodology: H. Hammerlindl, D. Ravindran Menon, S. Hammerlindl, D. Thakkar

Acquisition of data (provided animals, acquired and managed patients, provided facilities, etc.): H. Hammerlindl, D. Ravindran Menon, S. Hammerlindl, A.A. Emran, J. Torrano, K. Sproesser, D. Thakkar, M. Xiao, C. Krepler

Analysis and interpretation of data (e.g., statistical analysis, biostatistics, computational analysis): H. Hammerlindl, D. Ravindran Menon, A.A. Emran, J. Torrano, K. Sproesser, D. Thakkar, V.G. Atkinson, N.K. Haass, C. Krepler, H. Schaider

Writing, review, and/or revision of the manuscript: H. Hammerlindl, D. Thakkar, V.G. Atkinson, B. Gabrielli, M. Herlyn, C. Krepler, H. Schaider

Administrative, technical, or material support (i.e., reporting or organizing data, constructing databases): H. Hammerlindl, K. Sproesser, N.K. Haass, H. Schaider

Study supervision: H. Schaider

The authors thank Dr. Gerald Hoefler (Institute of Pathology, Medical University of Graz, Graz, Austria) for kindly providing A549, H358, and HaCaT cells and Dr Fiona Simpson (The University of Queensland Diamantina Institute, Brisbane, Queensland, Australia) for kindly providing SkBr3 cells. The authors also acknowledge the help provided by the Translational Research Institute (TRI) FACS, histology, and microscopy core facilities. This work was funded by the Epiderm Foundation (H. Schaider), the Princess Alexandra Research Foundation (PARSS2016_NearMiss; to H. Schaider), NIH grants PO1 CA114046 and P50 CA174523, and the Dr. Miriam and Sheldon G. Adelson Medical Research Foundation (both M. Herlyn). A.A. Emran is funded by The University of Queensland International Scholarship (UQI); H. Hammerlindl is funded by the International Postgraduate Research Scholarship (IPRS) and UQ Centennial Scholarship (UQCent). N.K. Haass is funded by the National Health and Medical Research Council (APP1084893).

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.