Markers of MEK inhibitor resistance in low-grade serous ovarian cancer: EGFR is a potential therapeutic target

Advanced or recurrent LGSC samples (tumor and ascites) were obtained from the OvCaRe gynecologic tumor bank (Vancouver General Hospital/British Columbia Cancer Agency (BCCA), and the John and Mary Knight Translational Ovarian Cancer Research Unit (London Regional Cancer Program, London, Ontario, Canada). Tumor bank protocols, cell line derivation, and all research relating to this study was approved by institutional human ethics review boards at BCCA (H14-02859), the University of British Columbia (UBC; R05-0119), and the University of Western Ontario (HSREB 12668E). Clinical information was extracted retrospectively from patient records. Tumor histology was confirmed by a gynecological pathologist.

Low-grade serous ovarian cancer patient-derived cell lines were established through continuous in vitro culture of patient material obtained through OvCaRe or the John and Mary Knight Translational Ovarian Cancer Research Unit (cell line iOvCa241) tumor banks. Cultures were established and maintained in M199:MCDB105 (1:1) media (Cat. No. M5017 and M6395, Sigma-Aldrich, Oakville, Ontario, Canada) supplemented with 10% defined fetal bovine serum (dFBS; Cat. No. SH30070.03, Hyclone, GE Life Sciences, Logan, UT, USA) maintained at 37 °C and 5% CO₂. No immortalization methods were used. Doubling time of these cells ranged from 30 to 80 h, with an average of 47 h, reflecting the clinical slow growth rate of LGSC.

Microsatellite analysis of short tandem repeats (STRs) was performed on LGSC cell lines and corresponding tumor and buffy coat samples for cell line authentication. STR analyses of 10 loci were performed by Genewiz Inc. (South Plainfield, NJ) (Data available upon request). STR results confirm that all LGSC cell lines and buffy coat samples match to corresponding tumor samples.

Whole exome sequencing (WES): Agilent SureSelect RNA Library All Exons v6 protocol was performed by Beijing Genome Institute, per manufacturer’s guidelines. Quality and quantity of post-capture libraries were assessed using an Agilent 2100 Bioanalyzer. Libraries were sequenced on an Illumina Hiseq 4000 (PE 100). Copy number variation (CNV) analysis: Data analysis was performed using Nexus Copy Number Discovery Edition Version 9.0 (BioDiscovery, Inc., El Segundo, CA). Samples were processed using the Nexus NGS functionality (BAM ngCGH) with the FASST2 segmentation. The log ratio thresholds for single copy gain and single copy loss were set at + 0.18 and − 0.18, respectively. The log ratio thresholds for gain of 2 or more copies and for a homozygous loss were set at + 0.6 and − 1.0, respectively. Tumor sample BAM files were processed with corresponding normal tissue BAM files. Probes were normalized to median. Mutation analysis: Sequence alignment and mutation calling were performed in Partek Flow environment (© 2017 Partek Inc). Sequence reads were aligned to GRCh38/hg38 human genome build using bwa 0.7.2. Variants were called using Strelka 1.0.15 for all cell lines except for VOA-1312 (lacking buffy coat sample). VOA-1312 variant calling was performed using LoFreq 2.1.3.a. The called variants were annotated using the wAnnovar software (reference obtained from: http://​jmg.​bmj.​com/​content/​49/​7/​433.​citation-tools). Annotated calls were then filtered to show only protein-changing SNVs that were present in cell line DNA at allele frequencies (AF) greater than 0.1 and coverage higher than 16×. For VOA-1312, all calls not present in dbSNP (version 138) were retained, while of the calls that were present in dbSNP, only calls with (average heterozygocity + aveHet standard error) < 0.1 were retained. These were additionally filtered using the same criteria as for the Strelka calls.

Whole genome sequencing (WGS): Genomic data from LGSC tumors T7 and T11 were obtained from the personalized oncogenomics (POG) program at the BCCA. Methodology has been previously described in detail [25]. To summarize, genome and transcriptome libraries were sequenced on HiSeq instruments (Illumina, San Diego, California) using V3 or V4 chemistry and paired-end 150 or 125 base reads, respectively. Target depth was 80× coverage for the tumor genome and 40× for the normal genome.

Assessment of MEKi sensitivity using trametinib (GSK1120212; Sellekchem, Cat. No. S2673) and selumetinib (AZD6244; Cat. No. S1008), were performed as previously described [24]. Cell proliferation was monitored using IncuCyte™ real-time imaging technology using a non-labeled monolayer confluence approach (Essen Biosciences, Ann Arbor, MI, USA). LGSC cell lines were plated at 15–20% confluence in 96-well plates. After 24 h, cells were treated once with DMSO (control) or differing drug concentrations [erlotinib alone (10 μM and 2.5 μM), in combination (10, 5, 2.5, 1.25, and 0.63 μM), high and low doses of MEKi treatment (1 μM and 0.5 μM selumetinib; 0.1 μM and 0.05 μM trametinib; doses for preclinical MEKi assays as previously published)] [24]. Trametinib and selumetinib were selected as the MEKi for combination treatments. These two drugs are most commonly used clinically for treating LGSC, and binimetinib may lack efficacy based on results from the MILO clinical trial (NCT01849874). Drug doses of selumetinib and trametinib were chosen based on IC50 results from our previous experiments [24]. Selected concentrations for these experiments are in keeping with steady state serum levels (selumetinib 2 μM and trametinib 30 nM) reported for these drugs in humans [26, 27]. Phase contrast images of cells were taken every 6-h for 4–5 days. Each condition was evaluated using four technical replicates and experiments were repeated for verification. Data analysis was performed using IncuCyte™ software. Statistical analyses using the t-test on the final time point values of each assay were performed to compare treatment conditions. Differences were considered significant at a p-value < 0.05.

Cell viability was measured using MTS-Cell Titer 96R Aqueous Non-Radioactive Cell Proliferation Assay, following manufacturer recommendations (Cat. No. G5430, Promega, Madison, WI, USA) at endpoint of Incucyte™ proliferation assays. Treatment media was replaced with 100 μL of fresh media and 20 μL of MTS. Plates were incubated for 3.5 h at 37 °C in humidified 5% CO₂. Absorbance at 490 nm was measured using a microplate reader (BioTek Epoch SN257811). Viability for each treatment was compared to DMSO treated cells. Wells were subsequently stained with crystal violet (CV) to determine residual cells after treatment. Statistical analysis using t-test were used to compare treatment conditions and differences were considered significant at a p-value < 0.05.

Erlotinib (Cat. No. S7786) were purchased from Selleck Chemicals (Houston, TX, USA). Dimethylsulfoxide (DMSO; Sigma, Cat. No. D2650) was purchased from Sigma-Aldrich (Oakville, Ontario, Canada). Cells were seeded in 96-well plates at 40–50% confluence and treated after 24 h with DMSO or a range of drug concentrations. The inhibitory concentration (IC50, representing 50% of total cell viability) was determined using crystal violet staining after 72 h drug treatment.

Cell lysates were prepared according to previously published protocols [24], then 20 μg samples were separated on an 8% SDS-PAGE gel, transferred to nitrocellulose membranes and probed with primary antibodies including ERK1/2 (Millipore, Cat. No. 06-182), p-MAPK (p-ERK1/2, Cell Signaling, Cat. No. 4376S), MEK1/2 (Cell Signaling, Cat. No. 9122), p-MEK1/2 (Cell Signaling, Cat. No. 9154), PKC-alpha (Cell Signaling, Cat. No. 2056), EGFR (Santa Cruz, Cat. No. 71032), p-EGFR (Cell Signaling, Cat. No. 2234), PARP (Cell Signaling, Cat. No. 9542), and c-PARP (Cell Signaling, Cat. No. 9541S). Vinculin (V9131, Sigma) was used as a protein loading control. Horseradish peroxidase (HRP)-conjugated secondary antibodies (goat-anti-mouse or goat-anti-rabbit, Sigma Cat. No. A9917 and A0545) were used accordingly. Western blots were imaged using Immobilon HRP reagent (Cat. No. WBKLS0500, Millipore, Etobicoke, ON, Canada) and developed by autoradiograph.

Reverse-phase protein array on whole tumor and cell line lysates was performed as previously described [28, 29]. Proteomic profiles of 8 LGSC cell lines, 2 MEKi-sensitive (VOA-1312, iOvCa241) and 6 MEKi-resistant (VOA-1056, VOA-3993; VOA-3448, VOA-3723; VOA-4627, VOA-4698), were analyzed. LGSC cells were treated for 24 h with 1 μL/mL DMSO or MEKi (trametinib 0.1 μM, selumetinib 1.0 μM) in biological triplicate as previously described [24, 30]. Antibodies (n = 91) against cell surface growth factor receptors, common signaling pathway proteins, steroid hormone receptors, and other proteins involved in proliferation and apoptosis were used (Additional file 1: Table S1). Data was analyzed using SPSS software (Version 20, Chicago, Illinois). Differentially expressed proteins between cell lines and treatment conditions were determined using the t-test [31]. The Mann–Whitney U test was used for proteins with non-normally distributed expression levels. False discovery rates were not calculated as putative markers were validated by western blot.

shERWOOD-UltramiR shRNA lentiviral target gene set containing three PRKCA shRNA sequences and one non-target shRNA (Cat. No. TLHVU1401-5578) was purchased from transOMIC Technologies (Huntsville, AL). VOA-3723 and VOA-6406 were plated at 50% confluence in 6-well tissue culture dishes 24 h prior to lentiviral transduction. 199:105 media supplemented with 1% Hyclone dFBS and polybrene (2 µg/mL for VOA-3723, 0.5 µg/mL for VOA-6406) and lentivirus expressing non-targeting shRNA or PRKCA shRNA (multiplicity of infection [MOI] = 26 for VOA-3723, MOI = 1.5 for VOA-6406) in a total volume of 1.5 mL was added. After 24 h, cells were washed with PBS and complete media was added. Successful transduction was confirmed using confocal microscopy. After an additional 24-h recovery, transduced LGSC cells were selected and maintained using puromycin (1.0 µg/mL for VOA3723, 0.5 µg/mL for VOA6406).

Cell proliferation, viability and crystal violet results from in vitro drug testing (single drug and drug combinations) were used to assess drug synergism using CompuSyn software (http://​www.​combosyn.​com). This software is based on the median-effect principle and the combination index-isobologram theorem (Chou-Talalay) [32]. Drug doses (D) and effects (fa) were entered (non-constant ratios) for single drug doses and combinations, and combination indices (CI) were generated. The CI values quantitatively defined synergism (CI < 1), additive effect (CI = 1), and antagonism (CI > 1).

Our laboratory previously established a collection of LGSC cell lines derived from patients with advanced/recurrent disease. Preclinical evaluation of four MEKi in eight different LGSC cell lines resulted in the identification of two distinct phenotypes: MEKi-sensitive (MEKi-Se) cell lines (n = 2), and MEKi-resistant (MEKi-Re) cell lines (n = 6). In this first study, MEKi drug concentrations and IC50 values were reported [24]. Recognizing the challenges using IC-50 values to assess drug efficacy in-vitro, we established a stringent definition of MEKi sensitivity/resistance recognizing that only 15% of patients with advanced/recurrent LGSC will show tumor experience regression when treated with a MEKi. Thus, we classified cell lines as MEKi-Se if a single dose of MEKi resulted in complete cell death over a period of 5 days. Alternatively, cell lines were considered MEKi-Re if they continued to proliferate (even despite some degree of inhibition) under the same treatment conditions. Continuing our previous work, we established three new LGSC cell lines from three independent patients (VOA-6406, VOA-8862, VOA-9164). Tumor cells from three other LGSC patients were also grown temporarily as primary cultures (VOA-6800, VOA-6857, VOA-7604). Using our previous classification criteria, two of these new lines were classified as MEKi-Se (VOA-9164 and VOA-8862), and one as MEKi-Re (VOA-6406—see Additional file 2: Figure S1). STR analysis confirmed unique microsatellite profiles for each of these lines, matching the profiles of the original tumor tissues from which they were derived (data available upon request).

WES was performed to characterize the genomic profiles of our LGSC cell lines and primary cultures. First, we compared the copy number profiles of two of our cell cultures (VOA-4627, VOA-6857) with those of their associated tumor samples (from WGS data). As shown in Fig. 1, the copy number variation profiles of the paired samples showed a very high degree of correlation. Of note, VOA-4627 line was derived from an ascites sample taken 2 years after the tumor sample collected previously at cytoreductive surgery.

Results from the WES analysis in our LGSC cell lines and primary cultures (n = 14) show variable levels of genomic aberration and non-synonymous mutations (NsMs), ranging from 1 to 66% total genome change and 24-111 mutation calls per cell line (Additional file 3: Table S2a). Deletion of Chr9p, including loss of MTAP and CDKN2A tumor suppressor genes, was found in all samples. As expected, KRAS and NRAS non-synonymous mutations were both found most frequently. Either mutation was present in 28.6% of all cell lines/cultures, and in 36.4% (KRAS) and 27.3% (NRAS) respectively when analyzed by patient (some cell lines were derived from the same patient at different times). KRAS and NRAS mutations co-existed in only one cell line (VOA-8862). Only one BRAF mutation was detected (D594G variant; VOA-6800 culture). Additionally we analyzed gene mutations and copy-number changes affecting 61 well-known MAPK-pathway genes is shown in Additional file 3: Table S2b. A summary of all RAS mutations, copy-number variation (CNV) findings, and MEKi sensitivity in each LGSC cell culture (n = 14) is shown in Table 1.

We found that all MEKi-Se cell lines (4/4; iOvCa241, VOA-1312, VOA-9164, VOA-8862) carried oncogenic mutations in KRAS (G12D or G12V), while MEKi-Re cell lines were either NRAS mutant (3/7; VOA-1056/VOA-3993 and VOA-6406), or KRAS/NRAS wt (4/7; VOA-3448/VOA-3723 and VOA-4627/VOA-4698). Of interest, the VOA-8862 cell line (mutations in both KRAS and NRAS) was found to be sensitive to all four MEKi tested. In this line, the KRAS mutation variant detected (G12D) is known to be oncogenic, while the NRAS mutation variant detected (C118Y) was not found in the COSMIC database, therefore its oncogenic potential remains unknown. We did not observe any obvious correlation between the degree of CNV in each cell line (copy number high versus low) and MEKi response.

To identify biomarkers of MEKi response, we compared two MEKi-Se (VOA-1312, iOvCa241) and six MEKi-Re (VOA-1056/VOA-3993, VOA-3448/3723, VOA-4627/VOA-4698) LGSC cell lines using reverse phase protein array (RPPA) analysis. To do so, lines treated with DMSO, 1 μM selumetinib, or 0.1 μM trametinib were screened using a panel of 91 validated antibodies (see “Materials and methods” and Additional file 1: Table S1). We found 12 proteins that were differentially expressed between MEKi-Se and MEKi-Re cell lines (Additional file 4: Table S3). Among these proteins, EGFR and PKC-alpha were found to be overexpressed in all MEKi-Re lines independently of the treatment status. These two candidates were selected for validation and further study as they are regulators of MAPK signaling and play a role in MEKi and chemotherapy resistance in the literature [33‐37]. WB analysis confirmed these findings (Fig. 2a), and also showed that p-EGFR (Y1068) was overexpressed in the MEKi-Re lines. Subsequently we subsequently validated the same candidates in the three newly established LGSC cell lines (VOA-6406, VOA-8862, VOA-9164) (see Fig. 2b). In keeping with the discovery cohort results, the MEKi-Re line (VOA-6406) expressed much higher levels of EGFR, p-EGFR and PKC-alpha than the two MEKi-Se lines (VOA-9164, VOA-8862) (Fig. 2b). As we previously described, p-MEK and p-ERK1/2 expression were not found to distinguish sensitive and resistant lines by WB [24].

We subsequently assessed differential protein expression by RPPA between MEKi-Se and MEKi-Re cell lines after selumetinib and trametinib treatment. Twenty-one and seventeen proteins were significantly different between MEKi-Se and MEKi-Re cells after selumetinib and trametinib treatment, respectively (Additional file 4: Table S3). Confirmation of the RPPA results was assessed by WB in one representative cell line from each individual patient (VOA-3993, VOA-4627, VOA-3723). Cell lines derived from the same patients at different time points in the disease course were not included for this analysis (VOA-1056, VOA-4698, VOA-3448). As seen in the untreated cells, WB confirmed increased p-EGFR levels in the MEKi-Re cell lines (Fig. 2c, d). As expected, trametinib more effectively inhibited MAPK phosphorylation than selumetinib. Differences in GSK3B and BID protein expression were also observed between MEKi-Se versus MEKi-Re cells by RPPA, however we were unable to validate these results using mass spectrometry (MS) analysis (data not shown). Interestingly, a number of the differentially expressed proteins (MEK-Se versus MEK-Re) were found to be drug-specific. These RPPA screening results are summarized in Additional file 4: Table S3, though these findings require further validation.

To establish whether EGFR expression played a role in mediating MEKi resistance we evaluated the effects of EGFR inhibition (using erlotinib), with and without MEK inhibition (using selumetinib or trametinib), in four MEKi-Re LGSC cell lines (VOA-3723, VOA-3993, VOA-4627, and VOA-6406). IC50 values for erlotinib in these cell lines are shown in Additional file 5: Table S4. Except for VOA-3723, all MEKi-Re lines were highly resistant to single erlotinib treatment as observed in other ovarian cancer cell lines [38, 39]. Erlotinib doses chosen for the combined experiments are in keeping with erlotinib human serum levels [40]. Effects of single and combined drug treatment were evaluated using proliferation, viability, and WB assays. EGFR mutation and copy-number status were also evaluated. By WES, none of our LGSC cell lines carried activating mutations in EGFR, though some had copy-number changes affecting this gene. As summarized in Additional file 5: Table S4, we could not identify any obvious factors [EGFR CNV levels, levels of EGFR protein expression, phosphorylation, or sensitivity (IC50 values) to erlotinib treatment] that were associated with sensitivity to combination therapy.

With the highest dose of erlotinib treatment alone (2.5 μM, one dose, over 4–5 days), all four MEKi-Re cell lines continue to proliferate. Interestingly as shown in Fig. 3, a reduced dose of selumetinib (0.5 μM) in combination with erlotinib (2.5 μM) resulted in statistically significant decreases in cell proliferation and viability (p < 0.001; t-test) in 2 of the 4 cell lines tested (VOA-3723 and VOA-6406). At the end of these experiments, complete cell death of both cell lines was confirmed by image inspection on Incucyte™ (Fig. 3; and Additional file 6: Figure S2). Using Compusyn software analysis, synergistic drug effects (were demonstrable even using lower doses of erlotinib (1.25 or 0.63 μM) with selumetinib (0.5 μM) (Additional file 7: Table S5). These drug combinations were not effective in the other two lines tested (VOA-3993 and VOA-4627). Reduced dose of trametinib (0.05 μM) in combination with erlotinib (2.5 μM) resulted in similar results for the VOA-3723 cell line, but cytostatic effects for the VOA-6406 cell line. A summary of the synergistic drug effects are shown in Additional file 7: Table S5. Drug synergy was stronger with selumetinib and erlotinib combination than with trametinib and erlotinib combination.

The effects of erlotinib, with and without MEKi treatment, on EGFR and MAPK signaling pathways were evaluated using WB. Levels of total and phosphorylated EGFR and ERK1/2, as well as total and cleaved PARP (c-PARP) were measured after 24 h treatment. Results from these experiments indicated that drug effects on cell signaling were cell line dependent (Fig. 4; and Additional file 8: Figure S3). As previously reported by our group, trametinib alone (0.1 μM) caused stronger inhibitory effects on ERK1/2 phosphorylation (p-ERK1/2) than selumetinib (1 μM). Unexpectedly, selumetinib treatment increased EGFR phosphorylation (p-EGFR Y1068) in 3 out of 4 MEKi-Re cell lines (VOA-6406, VOA-3723, and VOA-4627), however these effects were less obvious with trametinib treatment. As expected, erlotinib alone inhibited EGFR phosphorylation (p-EGFR Y1068) in all cell lines. Interestingly, erlotinib alone also inhibited ERK1/2 phosphorylation in 2 out of 4 lines (VOA-3723 and VOA-4627) and activated ERK1/2 phosphorylation in another line (VOA-6406). No pathway interaction was detected in the resistant VOA-3993 cell line. In these lines, while pathway interaction was observed, none of the changes in p-EGFR Y1068, p-ERK1/2 or c-PARP correlated with sensitivity or resistance to dual EGFRi and MEKi treatment. In the two MEKi-Re lines resistant to combination therapy (MEKi and erlotinib), the trametinib and erlotinib combination resulted in more apoptosis induction than the selumetinib and erlotinib combination (as measured by c-PARP).

Genomic characterization of PRKCA by WES revealed that none of our LGSC cell lines carried activating mutations in PRKCA. It is interesting to note that two MEKi-Re cell cultures (VOA-3723 and VOA-6857) carried PRKCA copy-number gain and two MEKi-Se cells (VOA-9164 and VOA-8862) had PRKCA copy-number loss. To determine whether PKC-alpha protein expression played a role in mediating MEKi resistance we evaluated the effects of PKC-alpha knockdown using lentiviral shRNA, with and without selumetinib or trametinib in two MEKi-Re LGSC cell lines (VOA-6406 and VOA-3723). As shown in Fig. 5a, PRKCA shRNA resulted in a complete PKC-alpha protein knockdown in VOA-6406 cells and a partial knockdown in VOA-3723 cells by WB. Subsequent MEKi treatment (1.0 μM selumetinib or 0.1 μM trametinib) experiments showed no significant changes in cell viability when compared to non-target shRNA transduced lines (Fig. 5b). Proliferation assays demonstrated that PKC-alpha knockdown did not increase MEKi sensitivity in the VOA-3723 cells, but may slightly increase MEKi sensitivity to selumetinib in the VOA-6406 cell line (p = 0.048). This treatment combination was much less effective than EGFRi and MEKi combination in MEKi-Re LGSC cell lines as the cells continue to proliferate. Taken together, these results do not support PKC-alpha as a treatment target in LGSC to expand the efficacy of MEKi treatment.