Cyclooxygenase-2 Influences Response to Cotargeting of MEK and CDK4/6 in a Subpopulation of Pancreatic Cancers

This article is featured in Highlights of This Issue, p. 2493

Pancreatic cancer is the third leading cause of cancer-related deaths in the United States and has the lowest 5-year relative survival rate of any cancer (1, 2). This disease is recalcitrant to chemotherapeutic approaches and recently approved therapies afford only modest improvements in survival. Consequently, the 5-year survival rate since the 1970s has only improved from 3% to 8% (2). There exists a critical unmet need for development of novel treatments for patients diagnosed with this disease.

The most commonly mutated genes in pancreatic cancer are KRAS, which occurs in over 90% of tumors, and CDKN2A (inactivated in >90% of cases), the gene encoding the endogenous cyclin dependent kinase 4/6 (CDK4/6) inhibitor p16^INK4a (3–6). KRAS is a small GTPase that activates the MAPK signaling pathway, whereas p16 is a potent suppressor of oncogenic transformation and a key mediator of RAS-induced senescence (7). While MEK inhibitors have exhibited potent in vitro activity in KRAS-mutant tumor cells (8), the in vivo activity of these agents has been disappointing due to the development of resistance (8–11).

An attractive target for MEK inhibitor–based combinations is CDK4/6, a kinase crucial for the transition from G₁–S-phase (12). In support of cotargeting MEK and CDK4/6, a synthetic lethal interaction between KRAS and CDK4 was found in non–small cell lung cancer (13). Furthermore, CDK4 was identified as a key driver of an alternative phenotype induced by MEK inhibition, but not genetic extinction of NRAS in mouse models of melanoma (14). Our laboratory as well as Kopetz and colleagues subsequently demonstrated in vivo efficacy of this combination approach in KRAS-mutant patient-derived xenograft (PDX) models of colorectal cancer (15, 16). Pancreatic cancers should also derive therapeutic benefit from this combination strategy based on their genomic features. Specifically, activating KRAS mutations have been shown to initiate formation of premalignant lesions in mouse models of pancreatic cancer, while loss of p16 has been shown to enable their malignant progression (17). Ectopic p16 expression can induce senescence and apoptosis when reintroduced into pancreatic cancer cell lines with CDKN2A deletions (18). Because CDK4 and CDK6 are the sole targets of p16, a unique opportunity is present to leverage recently approved CDK4/6 inhibitors to recapitulate this phenotype in pancreatic cancer.

The effectiveness of dual targeting of MEK and CDK4/6 to treat pancreatic cancer has been reported for high passage models (19, 20). This report extends these findings to include patient-derived xenograft (PDX) models of pancreatic cancer and concurrent phosphoproteomic profiling to identify potential prognostic biomarkers of response. We report here that two adenosquamous pancreatic models are highly responsive to dual targeting of these kinases both in vitro and in vivo. We further find that genetic manipulation of COX-2 expression, which is highly expressed in both of these models, blunts therapeutic effectiveness of combination treatment. Our results therefore provide the impetus to further explore the prognostic role of COX-2 to aid in the identification of a subpopulation of patients with pancreatic cancer who might derive the greatest therapeutic benefit from combination therapy directed against MEK and CDK4/6.

Trametinib and binimetinib (MEK162) were purchased from LC Laboratories. Ribociclib (LEE011) was purchased from Chemietek. Palbociclib isethionate was purchased from Selleckchem. Drug stocks were dissolved in DMSO at 10 mmol/L and stored at −20°C.

For growth inhibition and synergy analyses, cells were seeded in white-walled/clear bottom tissue culture–treated 96-well plates at 5,000–10,000 cells/well and allowed to adhere for 24 hours followed by addition of growth media containing serial dilutions of trametinib, palbociclib, or both drugs in combination. Cells were incubated for 5 days in the continuous presence of drug or DMSO and viability was measured using CellTiter-Glo (Promega). Viability was calculated as a percentage of DMSO-treated cells. Concentration–response curves were modeled using a nonlinear regression curve fit with a sigmoidal concentration response using GraphPad Prism 6. Synergy plot calculations were performed using Combenefit software (Cancer Research UK Cambridge Institute) and scores were generated using Chalice Bioinformatics Software (Horizon Discovery Group), both using the Loewe model of synergy.

All cell lines were cultured at 37°C in 5% CO2 in 10% FBS and 1% penicillin/streptomycin (Thermo Fisher Scientific). L3.6pl, pL45, MiaPaCa-2, Panc-1, and HS766T cell lines were grown in DMEM. HPAFII was grown in EMEM. ASPC1, Bxpc-3, and Panc10.05 were cultured in RPMI media. The PDX models UM8, UM15, UM16, UM19, UM32, UM53, UM59, UM81, UM90, UM91, and UM123 were established to grow in animals and in culture. All PDX models originated from patients providing informed written consent and undergoing surgical resection at University of Michigan (Ann Arbor, MI). All patient samples were procured with approval of the University of Michigan Institutional Review Board. All PDX lines were grown in RPMI media. All lines were negative for Mycoplasma contamination when tested with MycoAlert Mycoplasma Detection Kit (Lonza). All high passage cell lines were authenticated by short tandem repeat (STR) profiling at the University of Michigan Sequencing Core.

Cells were seeded into 6-well plates at approximately 150,000 cells/well and treated the next day at the indicated concentrations, with a maximum DMSO concentration of 0.1%. Cells were harvested with 0.05% trypsin, washed twice with PBS, and fixed with 70% ethanol at 4°C for at least 24 hours. Cells were washed twice with PBS and incubated for 30 minutes in a solution of 50 μg/mL propidium iodide (Life Technologies, P3566), 0.1% Triton X-100 (Sigma-Aldrich, T9284), 50 μg/mL RNase A (Qiagen, 1007885), and PBS. Data were collected on a Cyan ADP Analyzer (Beckman Coulter), with collection of at least 10,000 events. The analysis was performed using flow cytometry analysis software ModFit LT V4.0.5 (Verity Software House).

All procedures related to the handling, care, and treatment of animals were conducted in accordance with University of Michigan's Committee on the Use and Care of Animals guidelines. Cells (1 × 10⁶) were injected subcutaneously into the region of the right axilla of 6- to 7-week-old female NCR nude mice (NCRNU-F sp/sp CrTac:NCr-Foxn1^nu, Taconic) in a 50:50 mixture of DMEM/F12 and Matrigel. Tumors were allowed to grow until 150–300 mm³, at which time mice were randomized into different treatment groups (4–5 animals per group). Trametinib and palbociclib were administered as a fine suspension in 0.5% HPMC with 0.2% Tween-80 or saline, respectively, based on individual animal body weight (0.2 mL/20 g), once daily via oral gavage. Tumor volumes were measured using calipers and calculated using the formula: tumor volume = (length × width²)/2. Efficacy was calculated as the ratio of change in mean tumor burdens at time t (_ΔT/_ΔC) where _ΔT and _ΔC is calculated as mean tumor burden at time t minus the mean tumor burden on the first day of treatment. Percent regression is calculated as [– (_ΔT/T₀)*100], where T₀ is initial tumor weight of treated animals. Tumor growth delay was calculated on the basis of the time required for the mean to reach approximately 750 mm³. Where applicable, statistical significance between groups was calculated on the last day of treatment via one-way ANOVA with multiple comparisons between all treatment arms.

Cells were harvested by scraping in the presence of radioimmunoprecipitation (RIPA) buffer plus phosphatase and protease inhibitors (Roche). Samples were denatured and normalized to 1 μg/μL in LDS sample buffer (Thermo Fisher Scientific) and 62.5 mmol/L DTT (Thermo Fisher Scientific). For immunoblotting, multiple independent immunoblots were used to present data from single experiments. Loading controls were probed on the same blot, with a representative image shown for experiments with multiple antibodies. Ten micrograms of protein were run on precast 4%–12% polyacrylamide gels (Thermo Fisher Scientific). The following antibodies were used: pRb S780, pRb S807/11, Cdc6, total Rb, pERK, cyclin D1, COX-2, EGFR Y1068, FOXM1 (Cell Signaling Technology), Pdcd4 (Rockland), β-actin, and GAPDH (Abcam). For reverse-phase protein array (RPPA) analysis, tumors were homogenized and protein was extracted utilizing NP-40 lysis buffer plus phosphatase and protease inhibitors. Samples were denatured and normalized to 1 μg/μL in SDS sample buffer/β-mercaptoethanol (Thermo Fisher Scientific) and shipped to the MD Anderson RPPA Core Facility, where samples were profiled and processed. Per the protein loading protocol, all antibodies were median centered per antibody, then median centered per sample, and log₂ normalized. Trametinib, palbociclib, and combination-treated samples were compared with control samples using a two-sample t test for each treatment/control comparison. t test P values were adjusted for multiple testing using the FDR method and resulting q values < 0.1 were considered to represent significant changes. Significant antibodies and treatment sample replicates are clustered using agglomerative hierarchical clustering with a Euclidean distance metric.

Tissues were fixed in 10% NBF, embedded in paraffin, and sectioned according to standard procedures. The Ki67 antibody was obtained from Cell Signaling Technology. Representative images were obtained with a Nikon E-800 microscope, Olympus DP71 digital camera, and DP Controller software. To quantify the ratio of Ki67-positive nuclei to total nuclei, 5–8 snapshots of different sections of the tumor were used to determine the average Ki67 nuclear ratio using online software tool ImmunoRatio (21). Statistical significance was calculated between groups using one-way ANOVA (GraphPad Prism).

Full-length human COX-2 cDNA was excised as a BamHI/XhoI fragment from the huCOX-2 pcDNA5/FRT/TO construct, generously provided by Dr. William Smith (University of Michigan, Ann Arbor, MI), and ligated into pcDNA3.1. Transfections of L3.6pl cells were performed in 6-well dishes using huCOX-2 pcDNA3.1 linearized with Bgl II (2.5 μg plasmid/well) and Lipofectamine 2000 (Thermo Fisher Scientific) according to manufacturer's instructions. Clones stably expressing full-length human COX-2 were established after selection in medium containing 1 mg/mL G418 (Thermo Fisher Scientific). Control (sc-108080) and COX-2 (sc-29279-V) shRNA lentiviral particles (Santa Cruz Biotechnology) were transduced into cells according to manufacturer's directions and clones were established after selection with 10 μg/mL puromycin.

Total RNA was isolated from cell lines treated for 5 days using the RNeasy Mini Kit (Qiagen). First-strand cDNA was reverse-transcribed from total RNA using the SuperScript III First-Strand Synthesis System (Invitrogen) according to the manufacturer's instructions. Quantitative targeted amplification of cDNAs was performed using TaqMan Gene Expression Assays primer/probe sets for COX-2 (Hs00153133_m1), Pdcd4 (Hs00377253_m1), GAPDH (Hs02786624_g1), and Fast Advanced Master Mix (Thermo Fisher Scientific) according to the manufacturer's instructions. GAPDH was used as an endogenous control. The amplification conditions for the ViiA 7 Real-Time PCR System (Applied Biosystems) consisted of an initial step of 2 minutes at 50°C and 10 minutes at 95°C followed by 40 cycles of 15 seconds 95°C, 1 minute 60°C. For treated cells, data were analyzed using the ΔΔC_t method and expressed as fold change over control. Data from the panel of cell lines were calculated as one C_t equals a 2-fold difference in expression and are represented as relative expression to the highest expressing cell line.

Screening of high passage and PDX models of pancreatic cancer was carried out to identify models in which the combined action of trametinib and palbociclib showed the greatest degree of synergy. Cell lines showed a wide range of response to combination treatment, as depicted in Fig. 1A where models are listed in the order of a consolidated synergy score. L3.6pl cells showed the highest degree of synergy followed by UM59 cells, with both models exhibiting at least a 2-fold increase over the panel median (2.35; Fig. 1B). Synergistic response was greatest in response to the combination of trametinib and palbociclib at concentrations of 10 nmol/L and 1 μmol/L, respectively, concentrations which led to a median reduction in viability of 59% across the panel (Supplementary Fig. S1A). These optimal concentrations are consistent with our previous work in colorectal cancer models (15) as well as the studies of others carried out in lung cancer models (22). A similar degree of synergy was observed when L3.6pl cells were treated with the MEK inhibitor binimetinib (23) and the CDK4/6 inhibitor ribociclib (24), confirming that synergy is not likely attributable to off target activities (Supplementary Fig. S1B). Comparative analysis of concentration–response curves for high (L3.6pl), intermediate (UM59), and low (Bxpc-3 and Panc10.05) synergy models is shown in Fig. 1C to better understand the relationship between synergy and sensitivity. As shown here, the degree of synergy decreased as the concentration of trametinib was raised. No synergy was observed in Bxpc-3 and Panc10.05 cells at concentrations greater than approximately 20 nmol/L, where the concentration–response curves for trametinib single agent and the combination treatment plots intersect. In contrast, the concentration–response curves for L3.6pl cells never intersected, because viability could not be reduced beyond 50% despite use of high (1–10 μmol/L) concentrations of trametinib. These cells are also refractory to palbociclib treatment alone (Supplementary Fig. S2A).

MEK and CDK4/6 have both been shown to play a role in controlling cell-cycle progression, either through induction of cyclins or phosphorylation of Retinoblastoma protein (Rb), a master regulator of G₁–S progression. Cell-cycle studies carried out in the L3.6pl and UM59 models confirmed that combination treatment with trametinib and palbociclib induces a strong time-dependent G₁ arrest at concentrations previously shown to be synergistic (Fig. 2A). Immunoblotting analysis further revealed that expression levels of Cdc6, a protein critical for initiation of DNA synthesis and prereplication complex assembly (25, 26), and phosphorylated Rb, were selectively reduced in L3.6pl cells treated with the combination, consistent with G₁ arrest (Fig. 2B). These data suggest that synergy between trametinib and palbociclib to inhibit cell growth is driven at least in part by enhancement of G₁ arrest.

Immunoblotting analyses confirmed that phosphorylated ERK levels were suppressed by trametinib treatment in all lines at concentrations consistent with a reduction in viability (Fig. 3A and B). MEK inhibition additionally led to a reduction of total and phosphorylated Rb levels in a concentration-dependent manner. This result is not surprising, as MAPK signaling has been shown to contribute to control of cyclin D1 expression in response to growth factor stimulation (27), thereby indirectly reducing CDK4/6 phosphorylation of Rb. Cyclin D1 levels were recalcitrant to MEK inhibition in cell lines previously found to be most responsive to combination treatment (L3.6pl, UM59), despite reduction in pRb expression. This reduction in pRb can be explained by increased p27 expression in response to trametinib treatment (Supplementary Fig. S2B), which has been shown to bind and contribute to inhibition of the CDK4/cyclin D complex (28). In contrast, models showing a low degree of synergy (Panc10.05, Bxpc-3) exhibited significant reduction in levels of both cyclin D1 and pRb in response to trametinib single-agent treatment. While UM59 may be as sensitive to trametinib as Panc10.05 and Bxpc-3 (Fig. 3A), trametinib clearly shows cell line–dependent effects on expression of these cell-cycle proteins, likely contributing to comparative differences in synergistic potential with palbociclib. Collectively, these data suggest that tumors exhibiting MEK-dependent uncoupling of cyclin D1 and pRb expression in vitro may be most sensitive to dual targeting of MEK and CDK4/6.

On the basis of the high degree of in vitro synergy seen when MEK and CDK4/6 are both inhibited in L3.6pl cells, we evaluated the in vivo efficacy of the combination of trametinib and palbociclib in L3.6pl tumor–bearing animals. Daily treatment was initiated when tumors were advanced (∼300 mm³) for a total of 7 days. No signs of toxicity were noted at the doses administered. Neither single agent elicited a meaningful effect on _ΔT/_ΔC or tumor growth delay after cessation of treatment (Fig. 4A). In contrast, a _ΔT/_ΔC of 28% and a tumor growth delay of 10 days was observed in the combination arm. Tumors were harvested on the last day of treatment for IHC analysis of Ki67 expression (Fig. 4B and C), revealing a significant reduction in expression in tumors from the combination group compared with the control and single-agent groups. The results from this study were subsequently confirmed with less advanced L3.6pl tumors at treatment initiation, showing a _ΔT/_ΔC of 1% and a 15-day growth delay, versus 1 and 2 days for trametinib and palbociclib, respectively (Supplementary Fig. S3).

L3.6pl tumors evaluated for in vivo efficacy were further analyzed for selective proteomic changes as a consequence of combination treatment. Two proteins, COX-2 and programmed cell death 4 (Pdcd4), emerged from RPPA analyses showing inverse dysregulation in response to treatment with trametinib and palbociclib (Fig. 4D). COX-2 expression showed the highest magnitude of change in the combination arm among all the proteins measured in the RPPA platform. Palbociclib treatment caused a small decrease in COX-2, but this change was not significant. Others have noted the role of palbociclib in mediating a c-jun–dependent decrease in COX-2 expression, but the impact of this change outside of the epithelial–mesenchymal transition is unclear (29). An induction of Pdcd4 expression in response to the combination of trametinib and palbociclib was also observed, suggesting that dual targeting of MEK and CDK4/6 leads to initiation of cell death signaling. It appears that this effect is independent of apoptosis, as no significant changes were observed in PARP, caspase-3, 7, or 8 in the RPPA dataset (Fig. 4D). Subsequent immunoblotting studies were carried out to confirm that COX-2 is downregulated and Pdcd4 is upregulated in response to combination treatment with trametinib and palbociclib (Fig. 4E). Modulation of these biomarkers at the RNA level was addressed by carrying out qRT-PCR analysis of treated L3.6pl and UM59 cells, showing a reduction of COX-2 and an increase in Pdcd4 in treated samples of both cell lines (Supplementary Fig. S4A; UM59 protein expression changes shown in Supplementary Fig. S4B). Given that a decrease in COX-2 expression was the most significant change associated with activity in this study, we tested the COX-2 inhibitors celecoxib and NS-398 in the L3.6pl and UM59 models. Neither COX-2 inhibitor elicited significant antiproliferative effects at concentrations lower than 50 μmol/L. Concentrations in this range have been associated with COX-independent effects and have not been achieved in humans (Supplementary Fig. S4C; refs. 30, 31). The lack of efficacy is not surprising, as significant increases in COX-2 expression in response to these inhibitors has been shown (32).

On the basis of the reduction of COX-2 expression seen when cells were cotreated with trametinib and palbociclib, experiments were undertaken to explore a direct role for COX-2 in affecting sensitivity to dual inhibition of MEK and CDK4/6. FOXM1, a transcription factor whose stability is controlled by CDK4/6 phosphorylation (33) and whose activity and cellular localization is controlled by ERK (34), could be involved in driving this reduction in COX-2. This is possible considering FOXM1 activity has been implicated in promoting transcription of COX-2 in conjunction with Sp1 (35). The expression of FOXM1 is decreased selectively in L3.6pl cells exposed to combination treatment (Fig. 5A). Therefore, we hypothesized that combination treatment leads to synergy by decreasing expression of COX-2 through abrogation of FOXM1 activity. To test this hypothesis, a COX-2 plasmid under control of the cytomegalovirus (CMV) promoter was constructed for ectopic expression in L3.6pl cells. In this manner, removal of endogenous control of COX-2 expression would render cells unresponsive to FOXM1. After transfection, clones were selected and lysates were generated to track COX-2 expression, whereupon clone L3.6pl-C5 was found to comparatively exhibit the highest amount of COX-2 (Fig. 5B). Synergy of the parent line to the combination of trametinib and palbociclib was subsequently compared with that of the L3.6pl-C5 line and found to be significantly higher (synergy score, 7.35 vs 1.69, respectively). This finding suggests that overexpression of COX-2 by removing it from endogenous control influences the degree of synergy observed between trametinib and palbociclib. Importantly, response of the L3.6pl-C5 line to single-agent treatment remained unchanged in comparison with the parent line (Supplementary Fig. S5A). However, the L3.6pl-C5 line showed a blunted shift in the concentration–response curves when combining both agents at clinically relevant concentrations (1 to 10 nmol/L trametinib and 100 nmol/L to 1 μmol/L palbociclib; Supplementary Fig. S5B). This provides evidence that, although changes in COX-2 expression may not significantly affect response to either single agent, therapeutic efficacy of the combination is reduced upon alternatetranscriptional control of COX-2.

To confirm these findings in vivo, the parent and C5 lines were compared in tumor-bearing animals in a head-to-head study comparing efficacy from single agent versus combination therapy (Fig. 5C). On the last day of treatment, combination treated animals implanted with the parent line exhibited a _ΔT/_ΔC value of 17%, confirming the in vitro synergy seen with this combination against the L3.6pl model. In contrast, animals implanted with COX-2–overexpressing C5 tumors, exhibited a _ΔT/_ΔC value of 57%. Lysates generated from tumors harvested on the last day of treatment showed decreased FOXM1 expression in both studies, consistent with earlier in vitro studies. Furthermore, a greater ability of combination treatment to decrease COX-2 expression was observed in the parent line in comparison with L3.6pl-C5 (Supplementary Fig. S5C). These results suggest that endogenous COX-2 expression in a model with high COX-2 expression is critical for activity and removing this factor substantially reduces in vivo efficacy of combination therapy in the L3.6pl model.

Studies were designed to test the hypothesis that reduction of COX-2 expression through transcriptional control affects response to combination treatment. To explore the impact of COX-2 knockdown, L3.6pl and UM59 cells were virally transduced with COX-2 and control shRNA vectors. Despite harvesting numerous clones (>30), COX-2 was not successfully knocked down in UM59 cells, presumably due to reliance on expression for survival. In L3.6pl cells, transduction resulted in a successful knockdown of COX-2 (Fig. 5D). When these cells were evaluated for their response to combination treatment, knockdown cells showed reduced synergy in comparison with control cells (Fig. 5E). This reduction in synergy is consistent with results obtained with cells previously transfected with CMV-controlled COX-2 (Fig. 5F), confirming the impact of COX-2 expression levels on therapeutic outcome in this model.

In L3.6pl tumors, COX-2 appears to play a role in potentiating response to combination treatment. We explored the potential of COX-2 to serve as a prognostic marker of response to combination treatment across a broad panel of pancreatic cancer models. Lysates prepared from our pancreatic cell line panel were probed for expression of COX-2 (Fig. 6A). COX-2 appears as several bands, which likely represent different potential posttranslational modifications, because this protein has multiple sites for potential N-linked glycosylation, phosphorylation, and myristoylation (36). The expression levels of COX-2 in these lines were confirmed via qRT-PCR of RNA harvested from the panel (Supplementary Fig. S6A). L3.6pl and UM59 showed relatively high expression of COX-2 alongside Bxpc-3. Whereas L3.6pl and UM59 both exhibited a high degree of in vitro synergy to the combination of trametinib and palbociclib, a low synergy score was observed for the Bxpc-3 model. The lack of in vitro synergy seen here for Bxpc-3 cells is consistent with the observation that this model is exquisitely sensitive to MEK inhibition alone both in vitro and in vivo (37). Like L3.6pl, the UM59 model, which exhibited the second highest in vitro synergy score, showed improved therapeutic response when exposed to combination treatment, as evidenced by a 16-day tumor growth delay and _ΔT/_ΔC of 5% compared with ineffective single-agent therapies (Fig. 6B). Neither L3.6pl nor UM59 tumors were responsive to MEK inhibition alone. Panc-1 and Panc10.05 tumors, which exhibit low COX-2 expression (Supplementary Fig. S6B) and low in vitro synergy scores, showed somewhat improved response in the combination arms, as reflected by a _ΔT/_ΔC value of 14% and percent regression of 21%, respectively. Most notably, one tumor-bearing mouse for each of these models showed a complete regression when treated with the combination. However, the overall improvement in response of these models to combination treatment compared with either single agent, as measured by tumor growth delay or _ΔT/_ΔC, was reduced compared with the L3.6pl and UM59 models, which are characterized by high COX-2 expression. Therefore, consistent with our in vitro data, tumors exhibiting low COX-2 expression do not appear to derive as much added benefit from the combination regimen.

Novel therapeutic approaches for the treatment of pancreatic cancer are urgently needed due to the lack of significant improvements in patient survival over the past 40 years. Recent clinical approval of the CDK4/6 inhibitor palbociclib provides potential new opportunities for the treatment of cancers harboring CDKN2A (p16ink4a) aberrations, which includes the majority of pancreatic cancers. Because of the high incidence of KRAS mutations in pancreatic cancer and their cooccurrence with CDKN2A inactivation, a combination therapy approach targeting MEK and CDK4/6 was evaluated here. Cell line synergy screening was carried out in both high passage and PDX models, whereupon two lines, L3.6pl and UM59, exhibited at least a 2-fold greater response over the median. Both models, when implanted in vivo, proved to be refractory to single-agent treatment, while deriving substantial therapeutic benefit from the combination approach.

Importantly, cotargeting MEK and CDK4/6 was further found to potentiate cell-cycle arrest in both L3.6pl and UM59 cells over that with single-agent CDK4/6 inhibition. The prominent G₁ arrest observed in our studies was confirmed by synergistic reductions in total levels of Cdc6, a protein critical for initiation of DNA synthesis and implicated in response to CDK4/6 modulation and RB output (26). Furthermore, reduction was seen in the expression of FOXM1, a transcription factor involved in cell-cycle progression. It is a target of both ERK (34) and CDK4/6 (33), regulating cellular localization and stability, respectively.

Concurrent phosphoproteomic profiling of treated L3.6pl tumors revealed that COX-2 expression was downregulated in response to dual inhibition of MEK and CDK4/6. COX-2 is known for its role in mediating inflammation and promoting tumorigenesis in colorectal cancer and pancreatic cancer (38–41). While COX-2 expression can be affected by inhibition of MAPK signaling (42–44), it is unclear how inhibition of CDK4/6 synergizes with MEK to decrease expression of COX-2 in the absence of an effect by MEK inhibition. The significant reduction that we see in expression of COX-2 upon dual inhibition of MEK and CDK4/6 may ensue from reduced levels of FOXM1, as others have reported that COX-2 is in part controlled by FOXM1 activity (35, 45, 46). Another protein whose expression was altered by combination treatment in L3.6pl tumors was programmed cell death 4 (Pdcd4), which showed significant upregulation. Studies have shown that this novel tumor suppressor negatively regulates gene expression by inhibiting Sp1/Sp3 binding at important motifs (47) and may play a role in inactivating PI3K/AKT signaling and suppressing CCND1 and CDK4 expression in NCSLC (48). This finding has potential implications for this study in which FOXM1 is reduced by combination treatment, as other groups have shown FOXM1 cooperating with Sp1 to promote COX-2 expression (35). Pdcd4 may be regulated itself by direct phosphorylation through AKT (49). It is intriguing that studies have identified Pdcd4 to be in part responsible for the anticancer effects of COX-2 inhibitor NS-398 (50) in colon carcinoma. Further studies are warranted to elucidate signaling dynamics of these findings and further studies are ongoing to determine possible links. However, our studies unequivocally demonstrate that combining trametinib and palbociclib elicits a significant reduction of Ki67 staining in L3.6pl tumors, accompanied by a strong reduction in COX-2 and an increase in Pdcd4, both in vivo and in vitro.

COX-2 was expressed under CMV promoter control to test the hypothesis that transcriptional control of COX-2 was responsible for the reduction in expression seen. Ectopic expression of COX-2 increased resistance to combination therapy efficacy and blunted the reduction of COX-2 seen in response to combination treatment. However, a modest but significant reduction was still seen in L3.6pl-C5 cells. This could be explained by a low level of endogenous COX-2 that continues to be expressed in these cells. Posttranslational degradation mechanisms may also be in place that are being induced by combination treatment. Reports indicate caveolin-1 colocalizes with COX-2 at the plasma membrane (51, 52) and participates in direct degradation of COX-2 (53). As a result, studies are warranted to investigate the role of caveolin-1 in the degradation of COX-2 in cells treated with combination therapy, as a modest increase in caveolin-1 expression was observed in the RPPA dataset in the combination arm. Furthermore, knockdown of COX-2 in L3.6pl cells blunted the synergistic response in comparison with control cells, confirming a role for COX-2 in mediating response to coinhibition of MEK and CDK4/6. In summary, removing COX-2 gene expression from endogenous control, either through knockdown or expression of CMV promoter–driven COX-2, reduces synergistic response. This is expected, as modulation of COX-2 expression via control of FOXM1 is ostensibly what leads to synergy when coinhibiting MEK and CDK4/6. It is important to keep in mind that several models in our study are exceptionally sensitive to either trametinib or palbociclib alone when tested in vitro. Our data suggest that the usefulness of a synergy-based in vitro screen is biased toward models that show poor single-agent activity. In particular, the L3.6pl model scored the highest in the in vitro combination screen and was subsequently shown to elicit no benefit from either single agent in vivo, while responding favorably to the combination. This is not to say that tumors exemplified by Panc-1 and Panc10.05, which produced low in vitro synergy scores, would not benefit from the combination in vivo. In fact, both of these models showed one complete regression in the combination arm. Their low synergy scores in vitro were partly due to their high in vitro sensitivity to trametinib alone. Nonetheless, trametinib monotherapy proved to be inactive in mice in all of the models tested here. This highlights the disconnect between the in vitro and in vivo settings where tumor heterogeneity, tumor microenvironment, and adaptive signaling play a role.

Our goal was to identify models in which combination treatment disrupts signaling pathways that dictate their response. In vitro synergy screening facilitated the identification of two models, L3.6pl and UM59, which can derive substantial therapeutic benefit from dual targeting of MEK and CDK4/6. In those models, the interesting observation was made that COX-2 expression levels influence therapeutic outcome. Endogenous COX-2 expression appears to be critical for activity and its ablation substantially reduces in vivo efficacy of combination therapy. Interestingly, both models are adenosquamous carcinomas of the pancreas, a highly aggressive form of pancreatic cancer reported to show strong expression of COX-2 (54–58). Further studies are warranted to better understand the prognostic significance of high expression of COX-2, a protein implicated in pancreatic cancer development (38, 39, 59). Our collective data suggest that such studies may help guide identification of a subpopulation of patients with pancreatic cancer that could derive therapeutic benefit from cotargeting MEK and CDK4/6.

No potential conflicts of interest were disclosed.

Conception and design: J.D. Maust, J.S. Sebolt-Leopold

Development of methodology: J.D. Maust

Acquisition of data (provided animals, acquired and managed patients, provided facilities, etc.): J.D. Maust, C.L. Frankowski-McGregor, D.M. Simeone, J.S. Sebolt-Leopold

Analysis and interpretation of data (e.g., statistical analysis, biostatistics, computational analysis): J.D. Maust, C.L. Frankowski-McGregor, A. Bankhead III, J.S. Sebolt-Leopold

Writing, review, and/or revision of the manuscript: J.D. Maust, C.L. Frankowski-McGregor, A. Bankhead, J.S. Sebolt-Leopold

Administrative, technical, or material support (i.e., reporting or organizing data, constructing databases): J.D. Maust, C.L. Frankowski-McGregor, A. Bankhead III

Study supervision: J.D. Maust, J.S. Sebolt-Leopold

This research was funded by NIH R01CA155198 (to J.S. Leopold), P50CA130810 (to J.S. Leopold), T32-GM007767 (to J.D. Maust), and Luchessi Fellowship (to J.D. Maust). Research reported in this publication was further supported by the NCI of the NIH under award number P30CA046592.

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.