ERβ localization influenced outcomes of EGFR-TKI treatment in NSCLC patients with EGFR mutations

Effects of estrogen receptorβ (ERβ) localization on epidermal growth factor receptor tyrosine kinase inhibitors (EGFR-TKIs) in advanced non-small cell lung cancer (NSCLC) are unknown. First, we analyzed the relationship between ERβ localization determined by immunohistochemistry and EGFR-TKI outcomes in 184 patients with advanced NSCLC and found that ERβ expression localized in the cytoplasm and/or nucleus. The frequency of cytoplasmic ERβ (c-ERβ) and nuclear ERβ (n-ERβ) co-expression was 12% (22/184). C-ERβ and n-ERβ co-expression was correlated with poor median progression-free survival compared to patients without co-expression. In subsequent in vitro experiments, PC9 cells transfected with ERβ isoform1 (ERβ1, strong expression of both c-ERβ and n-ERβ) were more resistant to gefitinib than PC9 cells transfected with ERβ isoform2 or 5 (ERβ2 or ERβ5, strong expression of ERβ in cytoplasm but not nucleus). Resistance was identified due to interactions between ERβ1 and other isoforms, and mediated by activation of non-genomic pathways. Moreover, gefitinib resistance was reversed by a combination treatment with gefitinib and fulvestrant, both in cell lines and in one NSCLC patient. These results suggested that c-ERβ and n-ERβ co-expression was a potential molecular indicator of EGFR-TKI resistance, which might be overcome by combining EGFR-TKI and ER antagonist.

tissue distributions and play inconsistent roles in tumor cell biology. ERβ is commonly overexpressed in human NSCLC cell lines and patients and plays an important role in lung cancer development 7 . Despite the classical model of ERs stimulating transcription of estrogen-responsive genes, non-genomic signaling pathways are also activated by estrogen, including PI3K-AKT-mTOR and MAPK, which induce cancer cell proliferation and apoptosis arrest 8,9 . These pathways are considered common downstream signaling mechanisms of EGFR. In several preclinical studies based on lung cancer cell lines and xenografts, EGFR expression was down regulated in response to estrogen and up-regulated in response to ER antagonists (i.e., fulvestrant or tamoxifen) in NSCLC cell lines. Conversely, ERβ protein expression was down-regulated in response to EGF and up-regulated in response to gefitinib (an EGFR-TKI) 10,11 . These results indicate an interaction between EGFR and ER-related pathways.
We proposed the hypothesis that ER could induce resistance to EGFR-TKIs in lung cancer and that addition of an ER antagonist could reverse the resistance. However, clinical analysis in a Japanese study showed that strong ERβ expression predicts a better clinical outcome than weak expression in patients with lung adenocarcinoma following EGFR-TKIs therapy 12 . This study did not differentiate between ERβ localization (cytoplasm vs. nuclear), which could alter non-genomic signal pathway and activate and influence clinical outcomes.
To further investigate the impact of ERβ localization on EGFR-TKI efficacy, we analyzed correlations between ERβ localization (cytoplasmic and/or nuclear) and survival after EGFR-TKI therapy in 184 Chinese patients with advanced NSCLC and confirmed the clinical results in lung cancer cell lines. In addition, we first to date illustrated that the interactions between ERβ isoforms were associated with ERβ -mediated resistance to EGFR-TKIs and also explored the rationale for using EGFR-TKIs combined with fulvestrant in EGFR-mutant NSCLC.

Results
ERb expression and correlation with clinical characteristics in patients with advanced NSCLC. A total of 184 patients with stage IV NSCLC treated with EGFR-TKIs were analyzed, and 65 patients were treated as first-line therapy. Clinicopathological characteristics of the patients are summarized in Table 1. Most patients were never/light smokers (122, 66.3%) and had adenocarcinoma (159, 86.4%). A total of 107 patients (58.2%) carried EGFR sensitizing mutations (in exon 19del or 21L858R). ERβ expression was positive in 26.6% (49/184) of the patients with different intracellular distribution patterns, including nuclear only (n-ERβ ), cytoplasmic and nuclear (c-ERβ and n-ERβ co-expression) and cytoplasmic only (c-ERβ ) (22, 22 and 5 patients, respectively) (Fig. 1A).
No significant correlations were observed between ERβ expression and EGFR mutations (P = 0.093) or gender (P = 0.37). Moreover, neither nuclear nor cytoplasmic expression of ERβ was associated with Kaplan-Meier curves illustrated that patients with co-expression of c-ERβ and n-ERβ showed a poorer PFS after EGFR-TKI treatment than those without such expression pattern (P = 0.04) in total population (N = 184). (C) Kaplan-Meier curves showed that co-expression of c-ERβ and n-ERβ predicted inferior PFS after EGFR-TKI treatment compared to those without such expression pattern (P = 0.03) in subset with EGFR mutations (n = 107). (D) indicated the localizations of different ERβ isoforms. The confocal images (upper: fluorescence, lower: bright field) of PC9 and HeLa cells transfected with different ERβ isoform plasmids (NC, ERβ 1, ERβ 2, and ERβ 5) showed that ERβ 1 mainly localizes in nuclus contrary to ERβ 2 and ERβ 5 that mainly localize in cytoplasma. gender (P = 0.586, and P = 0.105, respectively) or any other of the clinicopathological characteristics (data not shown).
Several common EGFR-TKI resistance related gene variants (including mutations in KRAS, BRAF, PIK3CA and T790M, and c-MET amplification) were detected in patients who had EGFR mutations and c-ERβ and n-ERβ co-expression. Distribution of ERβ expression and genetic variants by case are listed in Table 2. After patients with genetic variants associated with EGFR-TKI resistance were removed, the median PFS in the remaining patients with c-ERβ and n-ERβ co-expression was 4.3 months (95%CI: 1.8 to 8.8 months).

Intracellular ERb localization was associated with ERb isoforms.
Previous studies have reported that intracellular ERβ localization (c-ERβ or n-ERβ ) was due to different expression pattern of ERβ isoforms in some cancers. In lung cancers, ERβ isoforms 1, 2 and 5 are commonly expressed 13 . To check the localization of ERβ isoforms, we fused EGFP to the cDNA of these different isoforms of ERβ (1, 2 or 5) and transiently transfected them into PC9 and HeLa cells. As shown in Fig. 1D, ERβ isoform 1 (ERβ 1) mainly localized in the nucleus both in transfected PC9 and HeLa cells, while ERβ isoforms 2 and 5 (ERβ 2 and ERβ 5) majored localized in the cytoplasm.

Interactions between ERb isoform 1 and other isoforms conferred resistance to gefitinib in vitro.
To figure out the mechanism underlying the resistance to gefitinib, In vitro experiments were performed to identify whether c-ERβ and n-ERβ co-expression was a predicting factor associated with resistance to EGFR-TKI observed in clinical analyses.   As shown by real-time PCR and immunoblotting tests, PC9, a lung adenocarcinoma cell line with the EGFR 19del, expressed both ERβ isoforms 2 and 5 ( Fig. 2A-C). To mimic clinical processes, we transfected ERβ 1, 2 or 5 plasmids into PC9 cells and constructed stable cell lines. PC9/ERβ 1 cells (PC9 cell line with ERβ 1) showed strong co-expression of c-ERβ and n-ERβ compared to PC9/NC cells (PC9 cell line with control vector), which was in contrast to PC9/ERβ 2 and PC9/ERβ 5 cells (PC9 cell line with ERβ 2 or ERβ 5) that only expressed c-ERβ (Fig. 2D). Cell viability tests indicated that PC9/ERβ 1 cells had significant resistance to gefitinib compared with controls (PC9/NC) and the other two cell lines (PC9/ERβ 2 and PC9/ERβ 5) (Fig. 2E). Examination of the downstream signaling by immunoblotting test showed that the phosphorylated ERK1/2 was significantly enhanced in PC9/ERβ 1, but not PC9/ERβ 2 and PC9/ERβ 5, compared with PC9/NC cells (Fig. 2D). Considering that PC9 cells mainly expressed ERβ 2 and ERβ 5, It seemed that only co-exist of ERβ 1 and ERβ 2 or ERβ 1 and ERβ 5 could induce the resistance, suggesting the possible role of an interaction between ERβ 1 and other ERβ isoforms.

Table 2. The gene aberrances and PFS after EGFR-TKI in patients with c-ERb and n-
We further selected HeLa cells (primarily express n-ERβ as determined by immunoblotting test, Fig. 2C, and ICC tests, Fig. 2F) to identify the mechanism of PC9/ERβ 1 cell resistance to gefitinib. ERβ 1, 2 or 5 were transfected into HeLa cells stably, demonstrating that HeLa/ERβ 1 cells showed a similar sensitivity as HeLa/NC cells, but were less resistant to gefitinib than HeLa/ERβ 2 or HeLa/ERβ 5 cells (Fig. 2G). The immunoblotting test showed that phosphorylated ERK1/2 was significantly enhanced in both HeLa/ERβ 2 and HeLa/ERβ 5, but not HeLa/ERβ 1, compared to HeLa/NC cells (Fig. 2H).
Taken together, all these data demonstrated that co-expression of n-ERβ 1 and c-ERβ conferred the resistance of NSCLC to EGFR-TKI treatment, which was due to the interactions between ERβ 1 and ERβ 2 or ERβ 1 and ERβ 5.
Activation of intracellular non-genomic pathways mediated gefitinib resistance. We further examined activation of intracellular non-genomic signaling pathways in PC9/NC and PC9/ERβ 1 cells treated with gefitinib. Under various concentrations (vehicle, 30 nM and 100 nM), phosphorylation of ERK1/2 and AKT was increased in PC9/ERβ 1 cells compared with the attenuated status of PC9/NC cells when treated with 100 nM gefitinib (Fig. 3A). Together with these data and the interactions between different ERβ isoforms identified above, a diagram was fabricated (Fig. 3B).

Fulvestrant improved sensitivity to EGFR-TKI therapy in PC9/ERb1 cells and patients with EGFR mutations and c-ERb and n-ERb co-expression.
Following combined treatment with fulvestrant (1 μ M), PC9/ERβ 1 cells became sensitized to gefitinib, similar to PC9/NC cells. Fulvestrant also enhanced the antitumor activity of gefitinib in PC9/NC cells, particularly at relatively high concentrations (Fig. 4A).
To confirm the role of fulvestrant in reversing resistance to EGFR-TKIs, we enrolled one female Chinese patient with stage IV lung adenocarcinoma and an EGFR mutation. This patient underwent local progression of a primary lung lesion after 8.7 months of gefitinib treatment, and then received continuous gefitinib therapy plus localized radiation. When rapid PD was observed (primary lung lesion and bone metastasis), gefitinib combined with fulvestrant was administered based on positive c-ERβ and n-ERβ expression in sample tissue. Subsequently, 3 months of disease control was observed. CT scans showed tumor shrinkage although it failed to achieve partial remission of disease (Fig. 4B).

Discussion
As transmembrane proteins, ERs share similar intracellular non-genomic signaling pathways with EGFR, suggesting that activating ER pathways may cause resistance to EGFR-TKIs 6,8,9 . However, correlation of ER expression with EGFR-TKI efficacy remains controversial. In the present study, c-ERβ and n-ERβ co-expression was identified as a potential biomarker for predicting poor PFS with EGFR-TKI therapy, which was examined as an outcome of the interaction between different ERβ isoforms. To the best of our knowledge, this represents the first study to correlate ERβ localization and resistance following EGFR-TKI treatment.
Based on initial clinical data, c-ERβ and n-ERβ co-expression (c-ERβ + n-ERβ ) predicted inferior PFS after EGFR-TKI therapy compared to patients without this type of expression pattern. However, c-ERβ only patients also presented with a poor PFS. To identify the actual factors related to EGFR-TKI resistance, in vitro experiments mimicking clinical processes were performed. Through transfection with different ERβ isoforms, EGFR mutant lung cancer cells with c-ERβ and n-ERβ co-expression (PC9/ERβ 1)  or only c-ERβ expression (PC9/ERβ 2 and PC9/ERβ 5) were constructed as in vitro models. Significant resistance to gefitinib in PC9/ERβ 1 cells compared with PC9/ERβ 2 and PC9/ERβ 5 cells supported the important effect of c-ERβ and n-ERβ co-expression. Therefore, c-ERβ and n-ERβ co-expression could be used as a biomarker predicting poor survival after EGFR-TKI therapy.
Consistent with previous reports 15, 16 , we identified activation of intracellular pathways, such as PI3K-AKT-mTOR and MAPK, after the EGFR pathway was blocked, which indicated that the non-genomic signaling pathway mediated gefitinib resistance. Several studies have reported that ERβ 2 and ERβ 5 failed to form homodimers, but could heterodimerize with ERβ 1 and enhance transactivation in a ligand-dependent manner 13,14,17 . We speculated that heterodimerization of ERβ 1 and other isoforms activate non-genomic signaling pathways when cancer cells with both ERβ 1 and other isoforms are treated with EGFR-TKI. A recent study from Nikolos' team seemed to obtain the contradictory results which demonstrated that nuclear ERβ 1 can down-regulate the EGFR and MAPK signaling pathway 18 . There are several differences between Nikolos' and our study. First, we used different lung cancer cell line and we just focused on EGFR mutant lung cancer cells. However, Nikolos' study did not show the effects of ERβ 1 on lung cancer cells with EGFR mutation. Second, we herein explored the role of ERβ isoform interactions to gefitinib resistance. However, Nikolos' study did not show the effect of ERβ 1 after gefitinib delivery. Third, we illustrated that the activation of non-genomic signaling pathway by the interaction of ERβ isoforms in cytoplasm mediated the gefitinib resistance, which was different from nuclear ERβ 1 affecting the transcription of target genes and then regulating the ERK1/2 signaling in Nikolos' study. So, we think that ERβ 1 can down-regulate EGFR and ERK1/2 signaling in NSCLC cells just like Nikolos' study. However, after the treatment of EGFR-TKI, the interactions of ERβ isoforms will induce the EGFR-TKI resistance by activating non-genomic signaling pathway (such as ERK1/2 and AKT) especially in EGFR mutant lung cancer cells with co-expression of ERβ 1 and other ERβ isoforms (ERβ 2 or ERβ 5).
To date, several genetic variants associated with EGFR-TKI resistance have been reported, such as the KRAS mutation 19 , PIK3CA mutation/amplification 20 , T790M mutation and c-MET amplification 21,22 . To exclude the effects of these factors associated with de novo resistance to EGFR-TKIs, several common variants were further analyzed in patients with EGFR mutations who also had c-ERβ and n-ERβ co-expression. Only 2 patients had the resistance-related gene mutations, which did not change the poor PFS after EGFR-TKI treatment in this population. These results supported the concept that c-ERβ and n-ERβ co-expression might be one of the mechanisms contributing to primary EGFR-TKI resistance.
Several in vitro and in vivo studies have shown enhanced effects when combining gefitinib and an ER inhibitor (e.g. tamoxifen or fulvestrant) in NSCLC, possibly providing a rationale for combining EGFR-TKIs with anti-estrogen therapy 10,11 . A pilot clinical study of combination therapy with gefitinib and fulvestrant in NSCLC also demonstrated improved anti-tumor activity 23 . In the present study, combined therapy consisting of gefitinib and fulvestrant led to enhanced anti-proliferative activity in EGFR-mutant lung cancer cells and improved PFS in adenocarcinoma patient with an EGFR mutation. Importantly, cell models and the one enrolled patient both had concurrent c-ERβ and n-ERβ expression, which provided a type of biomarker for alternative selection. However, only 3 months of prolonged PFS was observed for the selected patient when fulvestrant was added, which seemed to be inferior to the in vitro results. Possible reasons are that the timing of fulvestrant delivery was not appropriate in the patients. Initiating gefitinib combined with fulvestrant may be a more reasonable strategy for reversing EGFR-TKI resistance induced by concurrent c-ERβ and n-ERβ expression than combination therapy given after disease progression. Second, an insufficient dosage of fulvestrant may influence PFS improvement, and administration of fulvestrant twice rather than once per month is recommended in future clinical studies.
In summary, c-ERβ and n-ERβ co-expression predicted poor PFS after EGFR-TKI treatment in advanced NSCLC patients with an EGFR mutation. ERβ co-expression might serve as a candidate biomarker for predicting prognosis following EGFR-TKI therapy and determine if combined EGFR-TKI and ER inhibitor therapy is appropriate. The innate mechanism of resistance was activation of non-genomic signaling pathways mediated by interactions between ER-β 1 and other isoforms. Further studies with larger samples to evaluate ERβ with EGFR-TKIs were warranted.

Methods
Patient selection. This study included 184 Chinese patients with advanced NSCLC who received an EGFR-TKI (gefitinib oral 250 mg/d or erlotinib oral 150 mg/d) at the Peking University Cancer Hospital between June 2005 and December 2013. All diagnoses were histologically proven and evaluated as stage IV according to the current TNM staging system (IASLC 2009). Only patients with sufficient tissue for both EGFR mutation analysis and ERβ immunohistochemistry staining were enrolled. One patient with cytoplasmic and nuclear ERβ co-expression was prospectively enrolled to receive combined fulvestrant therapy (250 mg, intramuscular injection once monthly) following disease progression after gefitinib treatment.
Specimens were stored according to protocols approved by the Institutional Review Board of Beijing Cancer Hospital, and informed consent to use biopsy tissues for sample analyses was obtained from all patients.
For all patients, medical records were reviewed to extract clinicopathological data. Responses were classified using standard Response Evaluation Criteria in Solid Tumors, version 1.1. PFS was assessed from the first day of EGFR-TKI treatment until radiologic progression or death. Overall survival (OS) was determined from the EGFR-TKI start date until the date of death. Patients without a known date of death were censored at the time of the last follow-up.

Detection of EGFR sensitive and resistance related genetic variants. Genetic variants involved
in this study included EGFR sensitizing mutations (exon 19del and 21L858R), EGFR T790M, PIK3CA, KRAS or BRAF mutations and c-MET amplification 1,23-26 . Briefly, EGFR sensitizing mutations were detected by denaturing high performance liquid chromatography (DHPLC) according to previously described methods. Amplification refractory mutation system (ARMS) was used to reevaluate the EGFR wild type patient with adenocarcinoma by DHPLC. Other mutations in EGFR T790M, PIK3CA, KRAS and BRAF were also detected by ARMS. C-MET amplification was determined by quantitative real-time PCR using the Stratagene Mx3000P Real-Time PCR System (Agilent Technologies, Santa Clara, CA, USA) with TaqMan Universal PCR Master Mix (Applied Biosystems, Foster City, CA, USA). The reference gene was RNasP, and the MET primer and probe were designed by Applied Biosystems (Hs01432482_cn). Normal human genomic DNA was used as a control. c-MET gene amplification was defined as: 2-Δ Δ CT > 2.5 (Δ CT = CTMET -CTRNasp, Δ Δ CT = Δ CTcase -Δ CTnormal).

Immunohistochemistry and immunocytochemistry. ERβ expression was analyzed in lung tissue
samples and cell lines using immunohistochemistry (IHC) and immunocytochemistry (ICC), respectively. Briefly, dried 4-micron slides with formalin-fixed, paraffin embedded tissue were prepared. Combined sodium citrate (pH 6.0) and incubation in a pressure cooker (3 min, 125 °C) was used for antigen retrieval. Slides were then incubated overnight at 4 °C with primary mouse monoclonal anti-human ERβ (ABCAM, UK) at a dilution of 1:100. A two-step polymer-HRP method (Dako, Carpinteria, CA) was used for detection. No staining was observed for negative controls, which included incubation of lung tissue with a non-immune primary antibody.
Immunoreactivity 'positive' of IHC was defined if more than 10% of cancer cells were stained. Based on the localization of 'positive' immunoreactivity in either the cytoplasm or nucleus, patients were grouped as either c-ER-and/or n-ER-positive.
IHC and ICC staining was evaluated independently by different investigators (Dr. Hua Bai and Dr. Xiaosheng Ding) and a pathologist (Yu Sun). were commercially obtained from Sigma-Aldrich. Agents (fulvestrant) administrated to patients were provided by AstraZeneca.
Immunoblotting analysis. The protein expressions in cells were evaluated with western blot. Whole cell lysates (WCL) were obtained by extraction in cell lysis buffer (cell signaling) followed by protein quantification using the bicinchoninic acid assay (Pierce) and lysis in Laemmli sample buffer. A total of 20 ug of the protein sample was run on a 10% Tris-glycine gel and transferred to nitrocellulose. Primary antibodies were added and incubated overnight at 4 °C, and secondary antibodies were conjugated to horseradish peroxidase for 2 hours at room temperature. Blots were developed by enhanced chemiluminescence and photographed using a Fujifilm Dark Box II and Image Reader LAS-1000 Plus software. Primary antibodies included ERβ , EGFR, pEGFR, ERK1/2, pERK1/2, AKT, pAKT and β -actin (Santa Cruz). Peroxidase labeled anti-rabbit or anti-mouse secondary antibodies (Amersham Pharmacia, Piscataway, NJ) were used.
Construction of pEGFP-ERb isoform1, ERb isoform 2 or ERb isoform 5 and transient transfection. Localization of different ERβ isoforms was evaluated through transient transfection of ERβ iso-form1 (ERβ 1), isoform2 (ERβ 2) or isoform5 (ERβ 5). ERβ isoform 1, 2 and 5 fragments were synthesized by Genepharma (Shanghai, China). pEGFP-C1 vector was a kind gift from Dr. Xiaodong Wang (National Institute of Biological Sciences, Beijing). Briefly, pEGFP-C1 was digested with HindIII and BamHI. ERβ isoform fragments were amplified using PCR and ligated into HindIII and BamHI sites of pEGFR-C1. For transient transfection, cells were seeded at a density of 2 × 105 in 6-well plates overnight. A mixture of 1 μ g plasmid and 3 μ l lipofectamine was prepared in opti-MEM according to the manufacturer's instructions and added to the cells. Forty-eight hours after transfection, ERβ -green fluorescent protein fusion was detected under a fluorescent microscope. Localization of various ERβ isoforms was determined by confocal imaging which was performed using a laser scanning LSM 510 confocal microscope (Carl Zeiss, Welwyn Garden City, UK).

Protein interactions by immunoprecipitation (IP).
Flag-ERβ 1 plasmid was modified from constructed EGFP-ERβ 1 plasmid. HeLa cells were transiently transfected with different plasmids separately (Flag-ERβ 1, EGFP-ERβ 2, EGFP-ERβ 5, Flag-ERβ 1+ EGFP-ERβ 2 and Flag-ERβ 1+ EGFP-ERβ 5). The transfection was performed with 2 μ l lipo2000 (Invitrogen) per microgram plasmid. After 36 hours of transfection, cells were washed twice with ice-cold phosphate-buffered saline (PBS) and lysed in ice-cold lysis buffer. Bicinchoninic acid (BCA) protein assay kit (Beyotime) was used to measure the protein concentration. Equal amount of protein was immunoprecipitated with the Anti-Flag M2-Agarose from mouse (Sigma) and then subjected to 8% SDS-polyacrylamide gel electrophoresis (SDS-PAGE) The protein was then transferred from the gels onto polyvinylidene fluoride (PVDF) membranes and the immunoblotting to ERβ was performed as above described.

Construction of stable cell lines using lentivirus transduction. ERβ 1 (including ERβ 1 and
Flag-ERβ 1), ERβ 2 and ERβ 5 were constructed into a lentivirus expression vector and packaged by Genepharma (Shanghai, China). Virus titers of the supernatants, including virus particles provided, ranged from 5 × 107 to 2 × 108. MOI of 50 were used for infection of PC9 cell lines. After 3 days of infection, 2 μ g/ml puromycin was added to the cells and a stably pooled population of cells was obtained after 5 days. Stable integration of ERβ was determined by western blot.
Briefly, cells were seeded in sextuplicate in 96-well plates containing 100 μ l medium at a density of 2 × 103 cells/well for 24 hours and cultured with increasing concentrations of indicated drugs for an additional 72 hours. Afterward, 10 μ l water soluble tetrazolium salt (WST-8) was added to each well and incubated for 3 hours. Absorbance was measured at 450 nm using a microplate reader. Relative viability was calculated as (%/control) = [A450 (treated) -A450 (blank)]/[A450 (control) -A450 (blank)].

Statistics.
Relationships between clinicopathologic factors were analyzed using Pearson's χ 2 test or Fisher's exact test. Survival time was calculated using the Kaplan-Meier method, and comparisons between groups were made using log-rank tests. All statistical tests were two-tailed, with significance