NRF2 through RPS6 Activation Is Related to Anti-HER2 Drug Resistance in HER2-Amplified Gastric Cancer

Human epidermal growth factor receptor 2 (HER2) is one of the best-characterized targetable alterations in solid tumors (1, 2). Among locally advanced or metastatic gastric cancer, HER2 is amplified in about 7% to 34% of patients (3). Despite a clear improvement in overall survival reported with the use of trastuzumab in combination with platinum-based chemotherapy (4), resistance is a common event (5). Furthermore, other anti-HER2 agents, tested as first-line treatment or beyond progression, achieved disappointing results (6–9). Both tumor heterogeneity and some biological features should be studied for their role in this lack of clinical benefit (10, 11).

Recently, nuclear factor erythroid 2-related factor 2 (NRF2) has been described as a crucial stress response mediator in solid tumors (12). Its activation confers inducible resistance to xenobiotic and oxidative stress (13). In non–small cell lung cancer and pancreatic cancer, high NRF2 levels have recently been related to poor prognosis (14, 15). This phenomenon is due not only to chemo/radioresistance but also to aggressive proliferative phenotypes probably induced by NRF2 (16, 17). In gastric cancer, NRF2 expression was retrospectively identified as a potential factor of resistance to 5-fluorouracil (18). The mechanisms by which NRF2 promotes cell proliferation are not well understood (19). Under conditions of enhanced proliferation, such as active phosphatidylinositol 3-kinase (PI3K) signaling, NRF2 was shown to augment the expression of metabolic genes that assist in driving proliferative programs (20). The PI3K/AKT/mTOR pathway is commonly deregulated in solid tumors, either by constitutive activation of receptor tyrosine kinases (21) or by inactivation of its repressor PTEN (22). mTOR, which regulates cell growth and proliferation, exists in two distinct protein complexes, mTOR complex1 (mTORC1) and mTOR complex2 (mTORC2; ref. 23). The best-characterized mTORC1 substrates are ribosomal protein S6 kinase (S6K) and the eukaryotic translation initiation factor 4E-binding protein (4E-BP; ref. 24). Activation of S6K and its substrate, ribosomal protein S6 (RPS6), is recognized as an emerging factor in the development of several tumors (25). Ribosomal proteins are moreover described as convergent points of activation of the MAPKK and PI3K/AKT/mTOR signaling pathways (26, 27).

The PI3K/AKT/mTOR pathway could occasionally promote cell survival through activation of the nuclear translocation of nuclear factor erythroid 2-related factor 2 (NRF2; ref. 28). Previous studies using several cancer cell lines reported that PI3K and AKT functions are required for NRF2 activation (29). Recently, NRF2 has been regarded as one of the main orchestrators of the cellular antioxidant response (12), inducing resistance to chemotherapy (30). Nevertheless, its role in resistance to targeted agents has not yet been as well explored as its relation with RPS6.

In the present study, we have explored possible mechanisms related to anti-HER2 drug resistance in HER2-amplified gastric cancer models in vitro and in vivo. We identified a novel role of RPS6, a downstream effector of the PI3K/AKT/mTOR pathway, and NRF2 in resistance to anti-HER2 agents in gastric cancer.

Human gastric cancer cell lines NCI-N87 and OE19 were obtained from the ATCC and the European Collection of Authenticated Cell Cultures (ECACC)-Sigma, respectively. Cells were cultured in RPMI 1640 (Thermo Fisher Scientific), supplemented with 10% FBS (Gibco) and 1× penicillin/streptomycin (Gibco). Lapatinib, pimasertib, GSK2126458 (GSK458), saracatinib, everolimus, and ly294002 were purchased from Selleck Chemicals. Stock solution was prepared in DMSO, aliquoted and stored at −20°C. Trastuzumab was obtained from Roche and diluted with phosphate-buffered saline (PBS).

Derivatives of both OE19 and NCI-N87 lines resistant to HER2-targeted therapies were obtained. Cells were treated with gradually increasing doses starting with IC₅₀ dose. Lapatinib dose was progressively increased every 2 to 3 weeks over a period of 4 to 5 months until resistant clones emerged. The final lapatinib concentration reached was 1 μmol/L for the OE19 cells and 0.05 μmol/L for the NCI N87 cells. The trastuzumab-resistant model was performed using NCI N87 cells, reaching a final concentration of 1,000 μg/mL at the end of this period. OE19 did not present a clear sensitivity to trastuzumab. Each single clone was picked with cloning cylinders (Sigma) and expanded in culture medium containing their specific resistant dose. Parental and resistant cell line authentication was performed by short tandem repeat analysis. All cells were also screened for the presence of Mycoplasma before use. The experiments were performed less than 2 months after thawing early-passage cells.

Cells (4 × 10³ cells) were seeded in 96-well plates and allowed to adhere overnight. For cell proliferation analysis, viable cells were quantified every 24 hours after transfection with siRNAs using MTT reagent 3-[4.5-dimethylthiazol-2yl]-2.5-diphenyl-tetrazolium bromide); Sigma) according to the manufacturer's protocol. In cytotoxicity analysis, cells were treated with the indicated compound with or without siRNA transfection, and incubation was continued for an additional 72 hours. Cell viability was determined by MTT. Proliferation or inhibition of proliferation was calculated relative to untreated controls. Each assay was carried out at least in triplicate. The results were finally analyzed with GraphPad to calculate viability and IC₅₀ using a log (inhibitor) versus response-variable slope model.

Whole-cell protein extracts were prepared using RIPA buffer [50 mmol/L Tris-HCl pH 7.5, 150 mmol/L NaCl, 0.1% SDS, 1% Triton x-100, 0.5% deoxycholic acid sodium salt (w/v)] supplemented with 2 μL/mL protease inhibitor cocktail (Sigma) and 10 μL/mL phosphatase inhibitor cocktail (Sigma). Samples were sonicated and centrifuged at 14,000 × g speed for 20 minutes at 4°C. The protein concentration was quantified by a BCA protein assay kit (Thermo Scientific). Nuclear fractions were prepared using Nuclear Extraction kit (Active motif) according to the manufacturer's instructions. Proteins were separated on 10% SDS-polyacrylamide gels, transferred onto a nitrocellulose membrane and incubated with primary antibodies (Supplementary Table S1). Immunoreactive bands were recognized using peroxidase-conjugated secondary antibody (DAKO). Immunoblots were visualized using the “ECL Western Blotting detection kit” reagent (GE Healthcare) and the ImageQuant LAAS 400 (Healthcare Bio-Sciences) system.

A human phospho-RTK signaling antibody array kit (Cell Signaling, 7982) was used to measure the relative levels of phosphorylation for 28 receptor tyrosine kinases and 11 important signaling nodes. Protein (250 μg) was used, and the array was handled following the manufacturer's protocol.

RNA isolation, retrotranscription to cDNA, and qPCR (RT-qPCR) were performed as previously described (31). Briefly, RNA concentration and purity was determined with Nanodrop and RNA integrity with Bioanalyzer 2100 (Agilent Technologies). A total of 500 ng of RNA was retrotranscribed to cDNA using The “High-capacity cDNA Reverse Transcription Kit” (Applied Biosystems). A real-time PCR (qPCR) was performed using iTaq Universal SYBR Green Supermix (Bio-Rad) on the “QuantStudio 5 Real-Time PCR System” (Applied Biosystems). All reactions were performed in triplicate. A commercial primer was used to analyze RPS6 gene expression (qHsaCID0038051, Bio-Rad). The rest of the primers, obtained from the PrimerBank database (32), are shown in Supplementary Table S2. Gene-expression levels for each amplicon were calculated using the 2^−ΔCt method.

cDNA from the parental OE19 cell line and its derivative resistant clones LR1OE and LR2OE was synthesized and hybridized to the Affymetrix Clariom S Human array. Data were analyzed and statistically filtered using Partek Genomics Suite v6.6 software. Input files were normalized with the robust multichip average algorithm for gene array. To narrow the list of relevant genes, a restrictive filtering algorithm was applied using a combined criterion, which required both a fold-change absolute value of two or higher and a statistical significance of P < 0.005 between subgroups. Each data set was derived from three biologically independent replicate samples. A Benjamini–Hochberg step-up false discovery rate procedure was also used with a final maximum FDR value of 0.005. Functional annotation clustering of differentially expressed genes was also performed using the Pathway Studio v10 (Elsevier) (Supplementary Table S3).

The small interfering RNA (siRNA) directed against human RPS6 (SI00708008) or Universal Negative Control (1027310) was purchased from Qiagen. siRNA against NRF2 (EHU093471) was bought from Sigma. Cells were transfected using Lipofectamine RNAiMAX (Invitrogen) according to the manufacturer's instructions. Final concentration of the siRNAs was 20 nmol/L.

Female immunodeficient SCID mice were purchased at 6 weeks old (Charles River Laboratory). Animal care was in compliance with Spanish and European Community (E.C. L358/1, December 18, 1986) guidelines on the use and protection of laboratory animals, and all procedures were approved by the local ethics committee (2018/VSC/PEA/0101). Mice were acclimatized for two weeks before being injected with cancer cells and then caged in groups of five under controlled conditions (12 hours light–dark cycle; room temperature 20°C–22°C; humidity 55%–60%). A total of 2 × 10⁶ cells in 300 μL of Matrigel (BD Biosciences):PBS (1:1) were subcutaneously injected to the dorsal flank of mice. Once tumors reached a palpable size, mice were treated with 75 mg/kg lapatinib by orogastric gavage and 1.5 mg/kg GSK458 by i.p. injection given on alternate days. Tumor size was measured twice weekly, and the tumor volume was calculated using the formula: V = 1/2(length × width²) (mm³). When the maximum diameter permitted by the animal care guidelines of our institute was achieved, the mice were euthanized, and each tumor was processed to obtain paraffin and frozen samples.

IHC staining was performed on the formalin-fixed paraffin-embedded tissue of 31 gastric cancer patients. Sections (2 μm) were cut, deparaffinized with xilol, and rehydrated using ethanol at 90%, 80%, and 70%. After washing in water, the slides were autoclaved for 3 minutes at 1.5 atmospheres in sodium citrate buffer (pH = 6) for antigen retrieval. Endogenous peroxidase activity was blocked with hydrogen peroxidase for 5 minutes at room temperature. Slides were incubated with pRPS6 ser235/236 (clone D57.2.2E, 1:400, Cell Signaling Technology) or NRF2 (clone EP1808Y, 1:100, Abcam) for 45 minutes at room temperature. Antibody detection was performed using the Dako Real EnVision HRP Rabbit/Mouse K5007 (Dako) and 3.3′ diaminobenzidine (Dako). The slides were also counterstained with hematoxylin chromogen (Dako). The resulted immunostained and H&E slides were examined by a dedicated pathologist. The pRPS6 staining intensity and the proportion of positive cells were recorded using semiquantitative scores as described by Kim and colleagues (33) and Chen and colleagues (34). In the first, the proportion of positive cells was classified into one of five categories (1: 0%–5%, 2: 5%–25%, 3: 26%–50%, 4: 51%–75%, or 5: 75%–100%). In the second, a four-tiered scale was used to classify the proportion of positive cells (0: no tumor cell positivity, 1: 0%–10%, 2: 10%–50% or 3: 50%–100%). In both protocols, the staining intensity was classified as 1 (weak), 2 (moderate), or 3 (strong). Finally, both score parameters were multiplied. According to Kim and colleagues, cases with a final score of 1 to 4 were considered “low pRPS6,” and cases with 5 to 15 “high pRPS6.” Using the Chen and colleagues protocol, cases were classified as follows: 0 to 2: negative, 3 to 5: positive moderate, and 6 to 9: positive strong. The NRF2 immunostaining was evaluated following the protocols previously described by Kawasaky and colleagues (35) and Tong and colleagues (36). In the first, the percentage of positive cells was recorded (0%–100%) and the intensity was quantified using a three-value score (0, 1, or 2). In the second protocol, a semiquantitative score for the percentage of positive cells (0: 0%–25%, 1: 25%– 50%, 2: 50%–75%, or 3: 75%–100%) and the staining intensity (0: absence of staining, 1: weak, 2: moderate; and 3: strong) were assigned. Finally, both score parameters were multiplied. According to Kawasaky and colleagues, only cases with scores above the median were considered positive. Using the scale of Tong and colleagues, cases with a final score of 9 were defined as “high expression.” The presence of nuclear NRF2 immunostaining in more than 10% of tumor cells was also regarded as “high expression.”

To perform the immunofluorescent staining of culture cells, 15 × 10³ cells were seeded per well in a μ-slide VI plate (Ibidi). After 24 hours, cells were fixed with a fresh solution of 4% formaldehyde for 12 minutes at room temperature and blocked/permeabilized with 0.5% Triton X-100 in PBS. Cells were then incubated with NRF2 antibody 1:500 (EP1808Y, Abcam) in blocking and permeabilization solution overnight at 4°C. An anti-rabbit IgG Alexa Fluor 488 (Invitrogen) was used as a secondary antibody at 1/400 dilution. To visualize the nuclei and preserve the signals generated, the cells were stained with Mounting Medium with DAPI (Qiagen). The fluorescent images were captured using appropriate filters in a confocal spectral Leica SP2 microscope (Leica Microsystems Heidelberg GmbH).

Thirty-one patients diagnosed with locally advanced or metastatic HER2-amplified gastric cancer were included in the analysis. HER2 evaluation was performed according to updated guidelines. To be considered positive and susceptible to treatment with trastuzumab, tumors must have presented IHC+++ for HER2. Pathology records were analyzed, and all reports were reviewed. Clinical data were extracted from electronic medical records and reviewed retrospectively to collect age, comorbidities, classic prognostic factors such as T, nodal invasion, and site of metastasis. IHC assay was performed to verify the presence of the alteration observed in preclinical models. Written informed consent was obtained from all patients in agreement with approved protocols from the ethics committee.

The study was conducted in accordance with recognized ethical guidelines (Declaration of Helsinki), and it was approved by the INCLIVA institutional review board (protocol number: 2017/070). Informed consent from all patients was obtained.

Data were analyzed using the GraphPad Prism 6.0 software. Results are shown as means and standard error of the mean. The cutoff for statistical significance was set as P ≤ 0.05. The sensitivity to anti-HER2 drugs of parental cell lines and their respective resistant clones was studied using parametric two-way ANOVA with interaction and Sidak correction, performing a multivariate comparison of means per dose. A parametric two-way ANOVA test was also applied to xenograft models, to evaluate the effect of drugs in combination versus control and lapatinb alone on tumor growth. In this case, a multivariate comparison of means per time point was done with a Benjamini–Yekutieli correction.

A two-way ANOVA with Dunnett multiple-comparison test was used to evaluate cell treatments. A Student t test was used to evaluate the effect of silencing on cell viability.

Kaplan–Meier was used to estimate progression-free survival (PFS) among treated patients. Independent sample t tests and Pearson correlation were used to study the association of pRPS6 with NRF2 as categorical and continuous variables, respectively. All tests were two sided, and a P value of 0.05 was considered statistically significant. Descriptive statistics and survival analysis were performed using IBM SPSS Statistics v20. A Pearson test was used to evaluate correlation. Representation of P value: *, <0.05; **, <0.01; ***, <0.001.

To explore the possible mechanisms responsible for anti-HER2 drug resistance, we developed lapatinib-resistant clones (LROE) derived from OE19, a HER2-amplified gastric cancer cell line. Several stable lapatinib-resistant clones were obtained (LROE) and certified. Two of them (LR1OE and LR2OE) were selected for this work. LR1OE and LR2OE cells were discernibly resistant to lapatinib compared with the parental OE19 cell line, as confirmed by their IC₅₀ (Fig. 1A).

To look for potential molecular changes in resistant cells, an RTK ELISA assay was performed. High activation of RPS6 and SRC in resistant cells was recognized (Fig. 1B). To further clarify those changes, protein expression and activation were tested by Western blot. In our lapatinib-resistant models, HER2 expression was preserved, but its activation was found to be slightly less intense than in sensitive cells. SRC and AKT activation was also confirmed (Fig. 1C). Despite the RTK ELISA results, increase in ERK activation was detected in resistant cells. Interestingly, RPS6 was significantly activated among all resistant clones (Fig. 1C). RPS6 is one of the components of the 40S ribosomal subunit and has been functionally regarded as a regulator of cellular metabolism, cell size, survival, and proliferation. Its activation can be related to both MAPK and PI3K/AKT/mTOR pathways, found to be hyperactivated in our resistant model (refs. 26, 27; Fig. 1C).

By performing a transcriptomic assay, we noticed statistically significant mRNA changes between the different cell lines. Principal component analysis (PCA) showed that the alteration of the two resistant models was very similar, suggesting the development of a common alteration induced by continuous exposure to lapatinib (Fig. 2A). Clear differences in the expression of 132 genes, 76 upregulated and 56 downregulated, were detected between the two resistant clones and the parental cell line (ANOVA; FDR < 0.05; Supplementary Table S3). Unsupervised hierarchical clustering revealed again a robust classification between two different groups, showing that the resistant cell lines had different gene expression to the parental cell line (Fig. 2B). A functional annotation of altered genes showed enrichment in several biological processes linked with cell metabolism and drug response (Fig. 2C). A significant proportion of the most altered genes belong to the ARE-bearing genes, involved in cellular response to oxidative and xenobiotic stress, known to be regulated by NRF2 (ref. 37; Fig. 2D). Altered gene expression was validated by RT-qPCR (Fig. 2E). This transcriptomic assay suggested NRF2 was involved in lapatinib resistance in our experimental model. Moreover, a Western blot showed an NFR2 overexpression in LR1OE and LR2OE cells, reinforcing the idea of NRF2 involvement in acquired resistance to lapatinib in our model (Fig. 2F).

To validate the previous results, we analyzed another HER2-amplified gastric cancer cell line, NCI-N87, and developed lapatinib- and trastuzumab-resistant clones (Fig. 3A). Interestingly, LRNCI cells were found to be resistant even to trastuzumab (Supplementary Fig. S1). The persistence of HER2 expression was seen, and its activation was higher in trastuzumab-resistant cells (Fig. 3B). Importantly, RPS6 expression and activation was also confirmed across these cell lines by Western blot (Fig. 3B). Its activation was higher in TR1NCI. Nevertheless, in LR1NCI, SRC and ERK activation was prominently detected, as was found in the lapatinib-resistant clones derived from OE19. Unlike these models, AKT was not activated in LR1NCI (Fig. 3B). In both lapatinib- and trastuzumab-resistant clones derived from NCI N87, NRF2 increased expression was also confirmed by Western blot (Fig. 3C). These results suggest that NRF2 plays a relevant role in resistance to anti-HER2 drugs. Once activated, NRF2 moves into the nucleus, where it triggers its target genes. An immunofluorescence assay demonstrated the nuclear presence of NRF2 in lapatinib- and trastuzumab-resistant clones, suggesting its activation across different resistant models (Fig. 3D). A subcellular fractionation confirmed the prevalence of the nuclear expression among resistant cells (Supplementary Fig. S2).

A panel of targeted drugs, such as pimasertib, everolimus, GSK458, and saracatinib, were tested to evaluate the relevance of the described alteration in conferring resistance to anti-HER2 drugs. Despite the activation of SRC, no significant effect was observed from the use of saracatinib, a potent and specific SRC inhibitor. Everolimus, an mTOR inhibitor, demonstrated a nonsignificant reduction in cell viability. When a MEK inhibitor, pimasertib, was tested, it was possible to observe a slight decrease in viability only in LR2OE and LR1NCI. Remarkably, the use of a dual PI3K/TORC1/TORC2 inhibitor, GSK458, was able to inhibit viability considerably in all models (Fig. 4A), probably due to its capability in inhibiting TORC1, which is supposed to directly activate RPS6 protein (Fig. 4A; ref. 21).

To better define the role of RPS6 on cell viability, siRNAs were used to silence its expression. Interestingly, the proliferation of knockdown cells was significantly reduced versus the control (Fig. 4B), indicating a primary role of RPS6 in cell growth and survival. When silenced RPS6-resistant cells were treated with lapatinib and trastuzumab, respectively, a statistically significant reduction of resistance was observed (Fig. 4C). Decreased viability seems to vary proportionally to each RPS6 activation level. This evidence further points to the important role of RPS6 in developing resistance to anti-HER2 agents.

It has already been reported that ribosomal proteins increase NRF2 expression through its association with RNA binding proteins. To address whether RPS6 could be involved in NRF2 increase, a knockdown of RPS6 was performed. Western blot analysis highlighted the decrease of NRF2 expression in LR1OE silenced by RPS6 (Fig. 4D). In these cells, decreased expression of the activated genes, such as AKR1C1, AKR1C2, and AKR1B10, induced by NRF2 was also detectable, suggesting the functional dependence of NRF2 and RPS6 (Fig. 4D). The impact of increased NRF2 expression was studied by siRNA silencing, observing a significant decrease of cell proliferation in lapatinib- and trastuzumab-resistant models (Fig. 4E). Interestingly, lapatinib in LR1OE NRF2 knockdown cells was able to reduce viability in a statistically significant way. A slight effect was also observed with trastuzumab when siC and siNRF2 at each doses were compared (Fig. 4F). To evaluate which compound was able to inhibit pRPS6 and NRF2, a panel of drugs were tested, such as LY294002, everolimus, GSK458, pimasertib, and lapatinib. Western blot analysis and quantification suggest an important role in inhibiting both pRPS6 and NRF2 using the dual PI3K/TORC1/TORC2 inhibitor GSK458 (Fig. 4G). This evidence confirms the link between RPS6 and NRF2, suggesting that a PI3K/TORC1/TORC2 inhibitor may be used to overcome anti-HER2 resistance.

To corroborate our results on NRF2 activation through PI3K/AKT/mTOR/RPS6 and its role in resistance to anti-HER2 therapies, we planned an in vivo experiment. LR1OE were subcutaneously implanted in SCID mice that were successively randomized and treated with vector or lapatinib. Once tumors were detectable, the mice treated with lapatinib were randomized to continue treatment or to receive GSK458. As HER2 expression was preserved in our resistant models, lapatinib was maintained over resistant cells in vivo. In a previous experiment, parental NCI-N87 and OE-19 cell lines, GSK458 was found to be mildly effective, and the combination with lapatinib was no longer able to improve the results obtained with lapatinib single agent in vitro (Supplementary Fig. S3). However, a significant difference in tumor growth of LR1OE was observed according to the different treatment administered in vivo. As expected, there were no differences in tumor growth between control and lapatinib, whereas tumor growth was significantly reduced when GSK458 was added (Fig. 5A), suggesting the relevant role of GSK48 in resistant models. When tumors achieved the maximum tolerated volume, xenografts were euthanized, and the autopsies confirmed that GSK458 was able to significantly reduce tumor growth (Fig. 5B). The analysis of samples performed by Western blot (Fig. 5C) confirmed that tumors still preserved HER2 expression. AKT activation was also detectable. The cleavage of PARP was observed when the combination was administered. As reported in our in vitro assays, pRPS6 and NRF2 were expressed in tumor tissues of xenografts treated with vector and lapatinib (Fig. 5C), suggesting their potential role in resistance, whereas when GSK458 was used, pRPS6 and NRF2 were no longer detectable. A considerable reduction in the expression of NRF2 target genes was also observed (Fig. 5D). These results highlighted the relation between RPS6 and NRF2, namely, that NRF2 expression decreases when pRPS6 is inhibited.

We retrospectively analyzed a cohort of patients diagnosed with HER2-amplified locally advanced or metastatic gastric cancer (Supplementary Table S4). The aim was to explore the clinical relevance of the findings previously identified in our cell-resistant models. HER2 IHC was evaluated by a dedicated pathologist. Only patients with a high expression by IHC of HER2 (3+) were considered for this study. Electronical medical records and pathology reports were reviewed, and finally 31 patients were considered eligible. IHC analyses of pRPS6 and NRF2 over paraffin-embedded samples were performed to verify the role of NRF2 and PIK3/AKT/mTOR/RPS6 activation in diagnosis (33–36). pRPS6 expression was high in 70.1% (22/31) patients, moderate in 12.9% (4/31), and low in 16.1% (5/31). NRF2 expression was high in 32.2% (10/31) and low in 67.7% (21/31). pRPS6 hyperexpression was not always correlated with increase in NRF2 (r = 0.21, P = 0.27). However, in 9 of 10 cases with high NRF2 expression, high expression of pRPS6 was also observed (Fig. 6A). Progression-free survival was calculated for all patients who had received first-line treatment with trastuzumab and chemotherapy (21 patients). Median time to progression was 9 months for the whole group (2–34 months). Patients with high NRF2 expression (n = 9) at baseline presented poor prognosis with a median time to progression of only 4 months, whereas patients with low NRF2 expression (n = 12) did better with a median time to progression of 10.5 months (Fig. 6B; 4 vs. 10.5; HR, 1.86; 95% CI, 1.09–3.16; P = 0.023). Due to the small size of our sample and the exploratory nature of this observation, further analyses would be needed. In conclusion, NRF2 overexpression could be associated with worse PFS deriving from platinum-based chemotherapy and trastuzumab in HER2-amplified gastric cancer patients.

Although new molecular classifications have been proposed to personalize treatment for gastric cancer patients, a precision approach is far from becoming standard. Because HER2 amplification was identified as a target, anti-HER2 drugs have been tested, and trastuzumab in combination with platinum-based chemotherapy represents the standard of care (4). Nevertheless, primary and acquired resistance to anti-HER2 treatments are frequent events. Unlike the results obtained in HER2-amplified breast cancer, no other anti-HER2 agents used in first-line treatment or beyond progression led to clinical benefit (6–9). Several causes have been proposed to justify the lack of response, such as tumor heterogeneity or molecular characteristics. The heterogeneity of gastric cancer is significant and can limit the use of targeted agents. Molecular changes should also be addressed as responsible for primary or acquired resistance to tailored drugs. Recent data pointed to the de novo appearance of PIK3CA mutations among gastric cancer patients treated with platinum-based chemotherapy and trastuzumab, indicating that hyperactivation of the PI3K/AKT pathway causes acquired resistance (38).

In our study, we sought to identify molecular mechanisms responsible for resistance of HER2-amplified gastric cancer to anti-HER2 inhibition. We generated both lapatinib- and trastuzumab-stable resistant clones provided by two different HER2-amplified gastric cancer cell lines. Both trastuzumab and lapatinib induced activation of the PI3K/AKT/mTOR/RPS6 signaling pathway, in turn sustaining tumor growth and a more aggressive phenotype. The role of RPS6 was stressed. This protein was involved in tumor progression in other solid tumors (39, 40). In our work, RPS6 was hyperactivated among all resistant models, suggesting its important role in resistance. By silencing RPS6, cell viability significantly decreased. The same result was observed when a PI3K/TORC1/TORC2 inhibitor was used, confirming the implication of the PI3/AKT/mTOR/RPS6 pathway in developing resistance.

A transcriptome analysis performed to better characterize alterations that could cause resistance found altered gene expression related to response to drugs and metabolic processes. This evidence led to the identification of NRF2 as a crucial point of anti-HER2 resistance in our models. Once activated, NRF2 moves into the nucleus and serves as a master regulator inducing the expression of several cytoprotective genes. Many studies have reported that NRF2 activation in cancer cells promotes cancer progression (41) and metastasis (42), and it also confers resistance to chemo and radiotherapy (43). Indeed, lapatinib was shown to induce oxidative stress in breast cancer cells and NRF2 acted to reduce levels of reactive oxygen species in treated cells (44, 45). Moreover, recent reports in lung and pancreatic cancers (12, 14–15) have demonstrated that NRF2 promotes metabolic changes able to maintain the aggressiveness of cancer progression (37). In these experiments, NRF2 was also observed to promote cell proliferation under the sustained activation of PI3K/AKT signaling. High NRF2 levels are also known to be associated with poor prognosis in gastric cancer (46).

In our in vitro model, we showed that RPS6 and NRF2 are involved in resistance to anti-HER2 agents. RPS6 was identified as one of the switchers of NRF2, making its activity possible. In a previous report, RPS6 was reported to interact with La/SSB, able to bind NRF2 mRNA, inducing an increase of its translation under oxidative stress condition (47). The mechanisms by which these two proteins are related should be further explored. Moreover, there is increasing evidence that many aspects of cancer behavior, including treatment sensitivity, depend on posttranscriptional gene regulation (48).

Use of a PI3K/TORC1/TORC2 inhibitor also inhibited RPS6 and consequently NRF2, ultimately decreasing cell viability. When xenografts were generated, a reduction of tumor growth with GSK458, able to decrease both RPS6 and NRF2 activation, suggested the role of these proteins in in vivo resistance.

A cohort of HER2-amplified gastric cancer patients was also retrospectively analyzed, and it was possible to observe NRF2 hyperexpression in patients with poorer PFS. This evidence suggests that NRF2 could be responsible even for primary resistance to anti-HER2 drugs.

In summary, we have shown that NRF2 via RPS6 could be considered as one of the possible mechanism inducing lapatinib and trastuzumab resistance in HER2-amplified gastric cancer models in vitro and in vivo. Beyond genomic changes, such as mutations, posttranscriptional gene regulation should be analyzed as primary determinant of resistance to target agents. RPS6 and NRF2 expression in a cohort of gastric cancer patients suggested a trend toward less benefit derived from treatment with trastuzumab. In this scenario, the role of a concomitant treatment with a PI3K/TORC1/TORC2 inhibitor combined with anti-HER2 drugs could be further investigated to evaluate its potential benefit for patients in delaying and/or limiting the emergence of acquired resistance. Although combination therapy has the potential to increase toxicity, it could provide an attractive option in this disease, where resistance will eventually develop. A rationally designed therapeutic strategy as described above could therefore be of interest.

No potential conflicts of interest were disclosed.

Conception and design: V. Gambardella, D. Roda, J. Castillo, A. Cervantes

Development of methodology: V. Gambardella, F. Gimeno-Valiente, D. Roda, J. Castillo

Acquisition of data (provided animals, acquired and managed patients, provided facilities, etc.): V. Gambardella, F. Gimeno-Valiente, C. Martinez Ciarpaglini, D. Roda, T. Fleitas, P. Tolosa, M. Huerta, S. Roselló, J. Castillo, A. Cervantes

Analysis and interpretation of data (e.g., statistical analysis, biostatistics, computational analysis): V. Gambardella, F. Gimeno-Valiente, D. Roda, T. Fleitas, P. Tolosa, J.M. Cejalvo, M. Huerta, J. Castillo, A. Cervantes

Writing, review, and/or revision of the manuscript: V. Gambardella, F. Gimeno-Valiente, N. Tarazona, D. Roda, J.M. Cejalvo, M. Huerta, S. Roselló, J. Castillo, A. Cervantes

Study supervision: D. Roda, J. Castillo, A. Cervantes

The authors thank all the patients who participated in these studies and their families. Special thanks go to Mrs. C. Mongort and Dr. S. Navarro from the Department of Pathology of the University of Valencia for gathering patient samples and for the IHC assay. The authors thank Dr. E. Serna from the Central Unit of Research in Medicine, INCLIVA, Spain, for the assistance with the multigenic analysis; and Dr. A. Diaz for the help with animal care. We would like to thank Dr. G. Ayala for the statistical analysis. This study was supported by grants from the Instituto de Salud Carlos III (Health Institute; PI18/01909; PI18/01508). V. Gambardella was funded by an ESMO fellowship program 2014 and holds Rio Hortega contract CM18/00241 from the Carlos III Health Institute; N. Tarazona was funded by Rio Hortega contract CM15/00246 from the Carlos III Health Institute; D. Roda was funded by Joan Rodes contract 16/00040 from the Carlos III Health Institute; T. Fleitas was funded by Joan Rodes contract 17/00026 from the Carlos III Health Institute.

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.