|Home | About | Journals | Submit | Contact Us | Français|
Specific inhibitors of mitogen-activated protein kinase/extra-cellular signal-regulated kinase (ERK) kinase (MEK) have been developed that efficiently inhibit the oncogenic RAF-MEK-ERK pathway. We used a systems-based approach to identify breast cancer subtypes particularly susceptible to MEK inhibitors and to understand molecular mechanisms conferring resistance to such compounds. Basal-type breast cancer cells were found to be particularly susceptible to growth inhibition by small-molecule MEK inhibitors. Activation of the phosphatidylinositol 3-kinase (PI3K) pathway in response to MEK inhibition through a negative MEK-epidermal growth factor receptor-PI3K feedback loop was found to limit efficacy. Interruption of this feedback mechanism by targeting MEK and PI3K produced synergistic effects, including induction of apoptosis and, in some cell lines, cell cycle arrest and protection from apoptosis induced by proapoptotic agents. These findings enhance our understanding of the interconnectivity of oncogenic signal transduction circuits and have implications for the design of future clinical trials of MEK inhibitors in breast cancer by guiding patient selection and suggesting rational combination therapies.
The RAS-RAF-mitogen-activated protein kinase (MAPK)/extra-cellular signal-regulated kinase (ERK) kinase (MEK)-ERK and phosphoinositide 3-kinase (PI3K)-PTEN-AKT signaling pathways play central roles in the signal transduction networks promoting tumor initiation and tumor progression. This is highlighted by the high frequency of mutations of RAS, RAF, PI3KCA, AKT, and PTEN as well as amplification of AKT present in a broad spectrum of human malignancies (1–5). Because signal transduction networks integrate multiple upstream inputs, targeting pathways downstream of the receptors could conceivably result in greater therapeutic efficacy and broader applicability. For example, abolishing signal transduction through MEK offers the potential advantage of inhibiting both proliferation-promoting and antiapoptotic signals originating from either activated cell surface receptors or mutant RAS and B-RAF. In breast cancer, activation of the RAF-MEK-ERK signaling cascade is associated with increased metastasis risk (6). Furthermore, cross-talk between the RAF-MEK-ERK pathway and the estrogen receptor (ERα) might mediate tamoxifen resistance (7). Although these findings validate the RAF-MEK-ERK pathway as a therapeutic target in breast cancer, clinical studies of MEK inhibitors have shown only limited antitumor activity (8, 9). The mechanisms underlying the poor clinical response to inhibition of MEK remain unclear and no markers of susceptibility or resistance to MEK inhibitors suitable for guiding patient selection have been identified.
The availability of advanced genomic and proteomic technologies has facilitated capturing the complex molecular responses to defined stimuli at a systems level. For example, reverse-phase protein array (RPPA) represents an emerging proteomic technology allowing for comprehensive assessment of signal transduction pathways at the protein and phosphoprotein level (10, 11). Here, we pursued a systems-based experimental strategy to create an in-depth understanding of the responses of breast cancer cells to MEK inhibition. We found basal-type breast cancers to be particularly sensitive to such agents. Furthermore, a potent negative feedback loop between the RAF-MEK-ERK and PI3K pathways was discovered, which counteracts effects of MEK inhibition on cell cycle and apoptosis induction. In agreement with this finding, combined treatment with MEK and PI3K inhibitors resulted in marked tumor cell growth inhibition.
Chemical synthesis and characterization of PI3K inhibitors PI103 and PIK90 are described by Knight and colleagues (12). The following other reagents were used: rapamycin (Calbiochem), U0126 (Promega), and epidermal growth factor (EGF; Millipore). CI1040 (PD 184352) was kindly provided by Dr. Jenny Bain (University of Dundee, Dundee, United Kingdom). All drugs were diluted in DMSO except for PIK90, which was diluted in 1:1 (v/v) DMSO:H2O. The list of antibodies used in this study is provided in Supplementary Table S6.
Breast cancer cell lines used in this study were described in detail previously (13). The serum used throughout the course of this study was FetalClone III (HyClone), which is not supplemented with EGF.
Cell viability was determined using the CellTiter-Glo assay (Promega). Cells were treated with MEK inhibitors CI1040 or U0126 dissolved in DMSO. Nine dilutions of each drug were made in 1:5 serial dilution. The cell growth was determined using CellTiter-Glo assay at days 0 and 3 of drug exposure. Growth inhibition was calculated as described by the National Cancer Institute/NIH Developmental Therapeutics Program7 (14). The 50% growth-inhibiting concentration (GI50), total growth-inhibiting concentration (TGI), and 50% lethal concentration (LC50) were calculated using a standardized algorithms. Drug combination studies were designed according to Chou and Talalay (15). CalcuSyn software (Biosoft) was used to calculate the combination index (CI). CI < 1 indicates synergy, a CI about 1 shows an additive effect, and CI > 1 indicates antagonism. Combined drugs were used at fixed molar ratios.
SplineMarker (a variant of the linear splines method allowing for modeling of nonlinear relationships between molecular markers and response; ref. 16) was used to identify the mRNA predictors of response to the MEK inhibitors CI1040 and UO126 using existing mRNA expression profiles from 51 breast cancer cell lines (13). A false discovery rate of 10% was used to identify transcripts associated with responsiveness to MEK inhibition. Baseline expression data for the MAPK-related gene predictors were hierarchically clustered with XCluster8 using a noncentered metric and Pearson correlation. We visualized the data with Java TreeView.9 Model variables are learned by performing least squares fit between the molecular profile and sensitivity profile for each molecular variable. In SplinMarker algorithm, the model is determined via leave-one-out cross-validation. The goodness of fit is assessed by evaluating a P value corresponding to the F statistic for the fit (16). P values are corrected for multiple hypotheses testing using the false discovery rate method (17). Ingenuity pathway enrichment analysis was performed using the Ingenuity Knowledge Base Database10 separately for the sensitivity and resistance mRNA predictors. P values were computed using Fisher’s exact test, with Affymetrix HT_Hgu133A as a reference set. We used a threshold of P < 0.01 to identify significant pathways.
Protein expression profiles were generated in 35 breast cancer cell lines using conventional Western blot analysis for the detection of 64 proteins. In addition, expression of 34 proteins was assessed by RPPA. Proteins predicting response to MEK inhibition were identified using the SplineMarker algorithm.
MDAMB231 cells were transferred to low serum conditions 24 h before treatment. The cells were pretreated with U0126 (10 µmol/L) or DMSO (control) for 30 min, after which EGF was added at a final dose of 10 ng/mL. Protein lysates were collected at 1, 4, and 24 h after EGF stimulation, denatured in SDS sample buffer, and spotted onto nitrocellulose-coated FAST slides (Whatman) using a GeneTAC G3 arrayer (Genomic Solutions). Proteins were detected using a set of validated antibodies (Supplementary Materials and Methods) and quantified as described before (18–20).
The cells were treated with drugs either in low serum or in full serum conditions as indicated in the figures. For the low serum conditions, cells were washed in the medium containing 0.1% fetal bovine serum (FBS) and incubated in this medium for 24 h. Drugs or DMSO (control) was added directly to this medium, and 30 min later, EGF was added at a final dose of 10 ng/mL. Cells were harvested at time intervals after EGF stimulation. For the full serum conditions, cells were grown in their normal growth medium and, at 48 h after plating, treated with the drugs. Protein lysates were prepared from cells at 70% to 90% confluency. The cells were washed in ice-cold PBS and then extracted in the lysis buffer containing 1% Triton X-100, 50 mmol/L HEPES (pH 7.5), 150 mmol/L NaCl, 25 mmol/L β-glycerophosphate, 25 mmol/L NaF, 5 mmol/L EGTA, 1 mmol/L EDTA, protease cocktail inhibitor set III, and phosphatase cocktail inhibitor set II (Calbiochem). The lysates were clarified by centrifugation for 13,000 rpm for 10 min on ice and frozen at −80°C. Protein concentrations were determined using the bicinchoninic acid protein assay kit (Pierce Biotechnology). Protein extract (20 µg per lane) was electrophoresed, transferred to polyvinylidene difluoride membranes (Millipore), and probed with specific antisera using standard techniques. Bound antibodies on immunoblots were detected by enhanced chemiluminescence (Amersham).
Cells were treated with drugs 24 h after plating and harvested for apoptosis or cell cycle assays at 72 h after treatment. Apoptosis was measured in live cells by Annexin V-FITC and propidium iodide (PI) labeling using Apoptosis Detection Kit I (BD Pharmingen) and quantified by flow cytometry (FACSCalibur, Becton Dickinson) with CellQuest software. For cell cycle analysis, cells were fixed in 70% ethanol, treated with PI (30 µg/mL) and RNase A (100 µg/mL), and subjected to fluorescence-activated cell sorting (FACS) analysis. Data analysis was performed using ModFit LT cell cycle analysis software (Verity Software House). All treatments were done in triplicates and at least 40,000 cells were acquired from every sample for the FACS analysis.
We ascertained the effect of pharmacologic inhibition of MEK in a large panel of breast cancer cell lines that was described before (13). Cells were treated with the specific MEK inhibitors CI1040 or U0126 and cell viability was determined. A wide spectrum of growth inhibition was observed. For U0126, the GI50 ranged from 0.8 µmol/L (HCC1187 cells) to 60 µmol/L (MDAMB231 cells), whereas the GI50 for CI1040-treated cells ranged from 0.8 µmol/L (SUM229PE) to 33 µmol/L (MDAMB231 cells). Compared with luminal-type cell lines, basal-type breast cancer cells showed more frequent growth inhibition in response to MEK inhibitor treatment, particularly following treatment with CI1040 (Fig. 1; Supplementary Materials and Methods). For the majority of cell lines, however, the LC50 was not reached. Only nine cell lines treated with U0126 and eight cell lines treated with CI1040 were killed at concentrations of equal or less than 50 and 25 µmol/L, respectively (data not shown).
To identify genes and pathways predicting sensitivity to MEK inhibitors, we performed a correlation analysis of mRNA expression data from the cell line panel with the above-mentioned growth inhibition data. The previously reported mRNA microarray data of the cell line set were used (13). Using the SplineMarker algorithm, we identified 1,548 (883) gene transcripts associated with sensitivity to CI1040 (UO126), and 790 (671) transcripts were associated with resistance. Among these, 292 and 127 predictors, associated with sensitivity and resistance, respectively, were common to both drugs. Pathway enrichment analysis revealed that the predictive mRNA markers are significantly enriched in four pathways for U0126 and eight for CI1040 (Supplementary Table S2). These included ERK/MAPK signaling, purine metabolism, p53 signaling, and metabolism of xenobiotics by cytochrome P450. Genes involved in ERK/MAPK signaling were predictive of sensitivity, whereas PI3K pathway components were associated with resistance to MEK inhibitors (Fig. 2). Hierarchical cluster analysis revealed that genes predictive of susceptibility to MEK inhibitors were relatively overexpressed in basal-type cell lines, whereas resistance markers were highly expressed in luminal cells (Fig. 2). A multivariate model of response to CI1040 consisting of 26 mRNA predictors from the Ingenuity MAPK pathway gene set was confirmed as involved in this pathway by at least one additional database (KEGG, BioCarta, GeneGo, and GSEA). We trained the model on a randomly selected set of 31 breast cancer cell lines and tested the accuracy of the model on the remaining 10 cell lines. The best model consisted of only one gene (RAC2). The sensitivity predicted by the model was strongly correlated with the measured sensitivity on the test set: r = 0.69 (P = 0.027). We applied this in vitro molecular predictor to forecast sensitivity to CI1040 of previously reported 118 tumors (21). The model predicted that the basal tumors are more sensitive to CI1040 than the luminal tumors. The difference in predicted sensitivities between these two groups was statistically significant (P = 6.7e–03).
Because the MAPK pathway is regulated predominantly through protein modification, particularly phosphorylation, we ascertained the differences in protein expression levels as predictors of susceptibility to MEK inhibition in the panel of 53 breast cancer cell lines by performing Western blot analysis for the detection of 64 proteins. In addition, expression and phosphorylation of 34 proteins was assessed by RPPA technology. Fifteen phosphoproteins common to both approaches were included in the analysis. The correlation between protein expression levels in the breast cancer cell panel and GI50 of CI1040 as well as U0126 was calculated using SplineMarker (Supplementary Table S3). Ten proteins were significantly correlated with susceptibility to MEK inhibition for both drugs (Supplementary Table S4). Cytokeratin-5/6, a well-established marker for the basal breast cancer subtype, showed the strongest correlation with GI50 of the MEK inhibitors. Expression of cytokeratin-18 and ERα, both luminal subtype markers, was correlated with resistance to MEK inhibitors. These findings clearly support the notion that basal-type breast cancer cell lines are particularly susceptible to MEK inhibitors.
To further understand the response of signal transduction pathways to MEK inhibition, we analyzed temporal changes in protein expression and phosphorylation by RPPA technology. As expected, a strong reduction in phosphorylated ERK (p-ERK) levels in response to MEK inhibitor treatment was observed, along with down-regulation of effectors of ERK, such as cyclin D1 (Fig. 3A and B). Additionally, EGF treatment induced phosphorylation of the EGF receptor (EGFR) and downstream effectors, such as AKT. Unexpectedly, MEK inhibition led to markedly enhanced phosphorylation of EGFR and activation of the PI3K pathway, as determined by phosphorylated AKT (p-AKT) at S473. We confirmed key findings by conventional Western blotting in MDAMB231 and T47D cells using two different MEK inhibitors, CI1040 and U0126 (Fig. 3C; data not shown), showing that activation of AKT in response to MEK inhibition is not limited to a single cell line or a result of off-target effects of one drug. Overactivation of EGFR and AKT was most striking in the presence of EGF in the medium, as determined by Western blot analysis (Fig. 4). To test whether the increased phosphorylation of AKT was dependent on overactivation of EGFR, we treated MDAMB231cells with the specific EGFR inhibitor gefitinib in the presence or absence of CI1040. We found that in low serum conditions in the presence of EGF, MEK inhibitor–induced activation of AKT was fully abolished by the treatment with gefitinib. In 10% serum, the effect of gefitinib was still observed, although to a lesser degree. To rule out that MEK-dependent activation of AKT being a cell line–specific phenomenon, we assessed p-AKT in eight cell lines treated with CI1040 under low serum conditions and found increased p-AKT levels in five cell lines: MDAMB231, T47D, HS578T, MDAMB175, and Sum149 (Fig. 5A; data not shown). Thus, a negative regulatory feedback loop exists between MEK and AKT that depends on activation of the EGFR and thus represents a signal-amplifying mechanism (Fig. 5B and C).
Because suppression of apoptosis is among the known biological effects mediated by the PI3K pathway, we hypothesized that the feedback activation of the pathway in response to MEK inhibition contributes to resistance of breast cancer cells to MEK inhibitors. We therefore hypothesized that inhibitors of PI3K should synergize with MEK inhibitors in cell lines showing feedback activation of AKT. We tested this using inhibitors of PI3K specific for the p110α isoform, PIK90 and PI103 (12). This PI3K isoform is frequently mutated in cancer and a crucial signal transducer in cancer cells (22–24). Synergistic growth inhibition following PI3K and MEK inhibition occurred in 4 cell lines out of 11 tested (Fig. 5D). The calculated CI values ranged from 0.259 (HS578T) to 0.742 (SUM149 cells), which are considered strong and moderate synergism, respectively. These cell lines had shown strong activation of AKT following MEK inhibition. However, only MDAMB175 and SUM149 cell lines showed an increase in apoptosis rates when treated with a combination of MEK and PI3K inhibitors (Supplementary Table S5; data not shown). In contrast, in other cell lines, including MDAMB231 cells, the combination of CI1040 or UO126 with PI3K inhibitors induced a complete G1 arrest (Fig. 6A). Interestingly, cell lines responding with apoptosis to the dual-inhibitor treatment harbor wild-type p53, whereas those showing synergistic effects on cell cycle distribution contain mutant p53.
We sought to further understand the cell cycle effects induced by inhibition of MEK and PI3K and analyzed the response of proteins downstream of these molecules. This analysis included mammalian target of rapamycin (mTOR) kinase, an important regulator of protein translation. It is required for activation of S6 kinase (S6K) and 4E-binding protein 1 (4E-BP1), which regulates cap-dependent mRNA translation. Despite the inhibition of MEK by CI1040 in MDAMB231 cells, phosphorylation of both phosphorylated S6K and phosphorylated 4E-BP1 did not change (Fig. 6B), whereas inhibition of PI3K with PIK90 led to a significant reduction in S6K phosphorylation and a moderate inhibition of 4E-BP1 phosphorylation. The combination of CI1040 and PIK90, in contrast, further inhibited activation of S6K and caused a marked decrease in 4E-BP1 phosphorylation (Fig. 6B). Expression of cyclin D1, a target of translational regulation by the mTOR pathway (25), was slightly reduced after MEK inhibition, whereas treatment with PIK90 led to increased levels at the 24-hour time point. In contrast, the combination of both inhibitors resulted in a strong down-regulation of cyclin D1 protein, which was paralleled by reduction in the corresponding mRNA levels (Fig. 6C). Thus, regulation of cyclin D1 gene expression drives the expression levels of cyclin D1 protein, whereas translational regulation plays a minor role in this context. Indeed, knockdown of cyclin D1 expression by small interfering RNAs resulted in G1 arrest (data not shown), showing the effect of cyclin D1 levels on G1-S cell cycle progression in these cells. Based on these data, it seems that cyclin D1 represents a major intersection point between the RAF-MEK-ERK and PI3K pathways, mediating cell cycle arrest in response to pharmacologic inhibition of these pathways. Interestingly, dual inhibition of PI3K and mTOR (with PIK90 and rapamycin or the dual PI3K/mTOR inhibitor PI103), which acts downstream of both the MAPK and PI3K pathways, was not sufficient to induce a significant reduction of cyclin D1 protein or mRNA levels (Fig. 6B and C). Thus, reduction of MEK-dependent and mTOR-independent regulatory factors is required for the observed repression of cyclin D1.
Because it has been described by others that cells arresting in the G1 phase of the cell cycle might be protected from apoptosis induction, we assessed whether MEK inhibition would alter sensitivity to apoptosis induced by chemotherapy in MDAMB231 cells. As shown in Fig. 6D, we treated MDAMB231 cells for 1 or 24 hours with CI1040 or DMSO (control) before adding the chemotherapeutic agent camptothecin, which strongly induces apoptosis in these cells. In agreement with our prediction, apoptosis induction was inhibited after 1 hour of CI1040 pretreatment. After 24 hours of pretreatment with CI1040, the proapoptotic activity of camptothecin was completely abolished. Thus, MEK inhibition not only did not induce apoptosis but also acted as a cytoprotective when breast cancer cells were treated with proapoptotic agents.
In this study, we examine the molecular features of breast cancer cells that determine sensitivity to pharmacologic inhibition of the MEK-ERK signal transduction pathway. We used a large set of breast cancer cell lines as a model system. Previously, we showed that these cell lines retain molecular abnormalities characteristic for human breast cancers (13) and we used this system to define molecular predictors for susceptibility to lapatinib and for EGFR and Her2/Neu. These were validated in a clinical study of lapatinib in combination with paclitaxel.11 Similar approaches were previously used to identify molecular signatures of susceptibility to conventional chemotherapeutic agents (26). Thus, the validity of cell line–based approaches for the preclinical development and optimization of treatment strategies using targeted therapeutics has been well established.
A major finding of this study is the particular sensitivity of basal-type cell lines to inhibitors of MEK. This has implications for our understanding of the role of the MAPK pathway in basal-type breast cancer as well as for the further development of MEK inhibitors clinically. The predictive nature of basal-type features was strongly supported by our analysis of correlation of protein expression patterns and sensitivity to MEK inhibitors. The most predictive protein marker was cytokeratin-5/6, a well-established marker of basal-type breast cancer (27–29). Additional subtype-defining markers were found to be correlated with response to MEK inhibitors (EGFR with sensitivity, ERα, and cytokeratin-18 with resistance). High expression levels of EGFR and lack of expression of ERα and cytokeratin-18 are characteristic for basal-type breast cancers (30). Our analysis of gene expression predictors of response revealed a strong enrichment of genes encoding for MAPK pathway components. Furthermore, we found increased expression of these gene predictors in basal-type cell lines, showing that basal-type cells use this pathway to a greater extent than luminal cells and depend on its activity for proliferation. Interestingly, gene expression of the Rho-like GTPase RAC2 was highly predictive of responsiveness to MEK inhibition in breast cancer cell lines and in human tumors. Activation of the MAPK pathway has been shown as a central feature of basal-type breast cancers (31). Overexpression of EGFR, amplification and mutation of the KRAS oncogene, and overexpression of αB-crystallin have been shown to lead to activation of this pathway in basal-type breast cancer. αB-crystallin is a MEK-activating heat shock protein that is overexpressed in basal-type breast cancers and associated with poor outcome (32). Luminal-type cells might use the MEK-ERK pathway to a lesser extent and seem to be more dependent on the PI3K pathway, shown by the preferential occurrence of PI3K mutations in this subtype (33) and our finding that AKT protein levels predict resistance to MEK inhibitors.
The role of the PI3K pathway in conferring resistance to MEK inhibitors is highlighted by our discovery of a negative feedback loop activating AKT in response to MEK inhibition in an EGFR-dependent fashion, thus amplifying EGF signals (Fig. 5B and C). Cross-talk between the MAPK and PI3K pathways has been described before. A negative feedback loop has been shown from ERK to the Grb2-associated binder 1 protein, a scaffold protein that propagates EGFR signals to the PI3K and the RAS/MAPK pathways (34, 35). Because the PI3K pathway is a known promoter of the cell cycle and negative regulator of apoptosis, we reasoned that feedback activation of AKT could counteract therapeutically relevant effects of MEK inhibition. Indeed, dual inhibition of MEK and PI3K led to synergistic down-regulation of cyclin D1 mRNA and protein levels, as well as growth inhibition in cell lines showing strong feedback activation of AKT following MEK inhibition. The role of cyclin D1 in mediating efficacy of MEK inhibitors is further highlighted by our finding that cyclin D1 protein levels were correlated with resistance to MEK inhibition, whereas levels of glycogen synthase kinase 3 (a negative regulator of cyclin D1) were correlated with sensitivity to MEK inhibitors. Only a subset of our cell lines responded with cell killing to the dual inhibition, whereas the remainder of cell lines showed synergistic inhibition of the cell cycle. Similar observations were shown by others previously (36). It is conceivable that the reduced levels of E2F1 as a consequence of loss of cyclin D1 expression could result in protection from apoptosis because known E2F1-dependent proapoptotic genes exist that are inhibited by PI3K (37). The E2F1-driven apoptotic mechanism is p53 dependent. Interestingly, wild-type p53 is present in cell lines undergoing apoptosis following MEK/PI3K inhibition, whereas p53-mutant cell lines showed synergistic cell cycle arrest. Moreover, in MDAMB231 cells, protection from apoptosis resulted from MEK inhibition in cells treated with camptothecin. These findings suggest that MEK inhibition might have antiapoptotic effects and thus counteract potential therapeutic effects of other agents administered in combination.
These results are relevant for the design of future clinical trials. The MEK inhibitors CI1040 and AZD6244 have shown only modest antitumor activity in early clinical trials (8, 9). In one study that included breast cancer patients, tumor biopsies showed reduction of p-ERK levels and reduction of the proliferation marker KI-67 in tumor biopsies (9). This finding corresponds well with our observation of predominant cell cycle effects of MEK inhibitors in breast cancer. Our findings suggest focusing further development efforts of MEK inhibitors in breast cancer on patients with basal-type cancers, which are representing ~ 15% of all sporadic and >60% of BRCA1-related breast cancers (38, 39). Currently, treatment options for these patients are limited and MEK inhibitors might represent a promising therapy. Furthermore, inhibiting the MEK-EGFR-PI3K feedback loop is likely to result in therapeutic synergism, particularly in cases with wild-type p53.
Grant support: Director, Office of Science, Office of Basic Energy Sciences, of the U.S. Department of Energy under Contract No. DE-AC02-05CH11231 and NIH, National Cancer Institute grants P50 CA 58207, U54 CA 112970 (J.W. Gray), and P30 CA82103 (F. McCormick).
Disclosure of Potential Conflicts of Interest
No potential conflicts of interest were disclosed.
Note: Supplementary data for this article are available at Cancer Research Online (http://cancerres.aacrjournals.org/).