Cross-Database Analysis Reveals Sensitive Biomarkers for Combined Therapy for ERBB2+ Gastric Cancer

Exploring ERBB2-related pathways will help us finding sensitive molecules and potential combined therapeutic targets of ERBB2-targeted therapy for ERBB2+ gastric cancer (GC). In this study, we performed a cross-databases study focused on ERBB2+ GC. The data of ERBB2+ GC deposited in the cancer genome atlas (TCGA), gene expression omnibus (GEO), InBio MapTM, cancer cell line encyclopedia (CCLE), and cancer therapeutics response portal (CTRP) were analyzed. The correlation of expression levels of candidate and IC50 of candidate genes-targeted drugs were verified on NCI-N87 and MKN-45 GC cell lines. We found that RARA, THRA, CACNB1, and TOP2A are drug sensitive biomarkers of ERBB2-targeted treatment with FDA-approved drugs. All these genes act through Myc signaling pathway. Myc is the downstream hub gene of both ERBB2 and RARA. The expression of RARA, THRA, and CACNB1 were negatively correlated with Myc activation, while ERBB2 and TOP2A positively correlated with Myc activation. SH3BGRL3, SH3BGRL, and NRG2 were identified as potential ligands of ERBB2. The ERBB2+ GC with RARA amplification demonstrated better prognosis than those without RARA amplification, while overexpression of NRG2 and SH3BGRL correlated with poor prognosis in ERBB2+ GC. About 90% of ERBB2+ GC was compatible with chromosome instability (CIN) subtype of TCGA, which overlaps with intestinal-type GC in Lauren classification. In validating experiments, combination of Lapatinib and all-trans retinoic acid (ATRA) synergistically suppresses cell growth, and accompanied by decreased expression of MYC. In conclusions, we identified several predicting biomarkers for ERBB2-targeted therapy and corresponding histological features of ERBB2+ GC. Combination of ERBB2 antagonist or RARA agonist may be effective synergistic regimens for ERBB2+ GC.


INTRODUCTION
Gastric cancer (GC) is one of the most common cancers and the third leading cause of mortality worldwide (Ferlay et al., 2015). In China, both of the morbidity and mortality of GC rank the second . The major therapeutic approaches for GC are surgery, adjuvant chemotherapy and targeted therapy. The targeted drugs approved by FDA are trastuzumab (for ERBB2), cetuximab (for EGFR), and ramucirumab (for VEGF2) Wilke et al., 2014). Lapatinib, a small-molecule inhibitor of ERBB2, plus paclitaxel demonstrated activity in the second-line treatment of ERBB2+ GC (Hecht et al., 2016). In clinics, ERBB2+ GC is defined by scoring 3 + by immunohistochemistry or copy number amplification by FISH (He et al., 2013). In general, ERBB2+ GC accounts for about 13% of all GC cases (Bass et al., 2014). A phase III ToGA study showed that the incidence of ERBB2+ GC is up to 22% (Bang et al., 2010). Because the prognosis of ERBB2+ GC is poor, clarifying the mechanisms of drug sensitivity of ERBB2+ GC will be of clinical significance in ERBB2-targeted therapy (Wang et al., 2017). Currently, trastuzumab plus fluorouracil and platinum can effectively improve overall survival of ERBB2+ GC patients, but the response rate was only 32-68% (Ock et al., 2015). Finding additional therapeutic targets for combined therapy will benefit more ERBB2+ GC patients.
ERBB2 is located in chromosome 17q21.2, where some common oncogenes or tumor suppressor genes, such as TOP2A, TAU, p53, and HIC-1 are located (Zhang and Yu, 2011). Our previous studies confirmed that re-activation of tumor suppressor HIC-1 by small-activating RNAs inhibits cell division, growth and invasion (Zhao et al., 2014(Zhao et al., , 2015. Retinoic acid receptor alpha (RARA) is another gene located on chromosome 17q21.2. Paroni et al. (2012) reported that combination of RARA agonist and ERBB2-targeted drug demonstrated a synergistic anticancer activity in breast cancer. It suggested that some novel therapeutic targets for ERBB2+ GC may harbor on chromosome 17. In this study, we analyzed ERBB2-related pathways and explored potential drug sensitivity biomarkers that could be used as reference for targeted therapy of ERBB2+ GC.

Data Extraction and Data Mining
The data of gene expression, copy number variation, tissue images and clinical information of 413 GC cases was extracted from The Cancer Genome Atlas database (TCGA 1 ) and cBioPortal database 2 (Cerami et al., 2012;Gao et al., 2013;Bass et al., 2014). The data of gene expression and copy number variation of GC was also extracted from GSE62717 and GSE57302 in gene expression omnibus (GEO) database 3 (Qian et al., 2014;Cristescu et al., 2015). The data of protein-protein interaction was used in InBio Map TM database 4 (Li et al., 2017). The information of Lapatinib IC50 of 17 GC cell lines were extracted from database of Cancer Cell Line Encyclopedia (CCLE 5 ) ( Barretina et al., 2012). The information of Afatinib IC50 of 16 GC cell lines was extracted from database of Cancer Therapeutics Response Portal (CTRP 6 ) (Basu et al., 2013). Kyoto Encyclopedia of Genes and Genomes (KEGG) pathway database is used for explaining functions and biology of genes 7 . The analytic results were confirmed for proteins expression using The Human Protein Atlas 8 .

Gastric Cancer Cell Lines and Cell Culture
Human GC cell lines NCI-N87 and MKN45 were purchased from the type Culture Collection of Chinese Academy of Sciences (Shanghai, China). All cell lines were cultured in RPMI-1640 containing 10% fetal bovine serum (FBS) at 37 • C in a humidified incubator with 95% air and 5% CO 2 .

Cell Viability Assay
Firstly, 5000/well NCI-N87 or MKN45 cells were placed in 96 well plates (100 µl/well). Cells were treated by different concentrations of Lapatinib (Selleck, Houston, TX, United States) and All-Trans Retinoic Acid (ATRA, Selleck, Houston, TX, United States), respectively. CCK-8(10 µl/well) was used and OD value was measured at 450 nm by spectrophotometry (BioTek, VT, United States) at different time points.

Statistical Analysis
EdgeR package in R language was used to screen differentially expressed genes (Robinson et al., 2010). We plotted OncoPrinter map based on gene copy number data from TCGA and cBioPortal platforms. Pheatmap package in R language was utilized to plot heatmap. Receiver operating characteristic (ROC) was plotted and an optimal cutoff value was selected for biomarkers from clinical samples. The median value was used as cutoff value for cancer cell lines. Survival rate was calculated by Kaplan-Meier method and log-rank test. GSEA software was applied to perform gene set enrichment analysis (Subramanian et al., 2005). Pearson correlation between RNA levels and IC50 was analyzed by Pearson test. CompuSyn software (ComboSyn Inc., Paramus, NJ, United States) was used to calculating combination index (CI) of drugs synergistic effects. CI < 1, =1, and >1 indicate synergistic, additive and antagonistic effects, respectively (Yu et al., 2018). Other statistical analysis was performed by using GraphPad Prism 6.0 (Inc., La Jolla, CA, United States). P-value less 0.05 was considered statistically significant.

RESULTS
ERBB2+ GC Showed Amplification of RARA, THRA, CACNB1, and TOP2A Among 413 cases of GC from TCGA dataset, ERBB2+ GC accounted for 14.29% (59/413 cases), and the others were ERBB2-GC (354/413 cases). We compared the differential expressed genes and copy number variations between ERBB2+ and ERBB2-GC groups. Forty-seven differential expressed genes were selected based on the likelihood ratio (LR) and P-value (LR > 100, P < 0.0001; Figure 1A). Four out of these 47 genes are the molecular targets of FDA-approved drugs. They are CACNB1 (calcium voltage-gated channel auxiliary subunit beta 1) for amlodipine, TOP2A (DNA topoisomerase II alpha) for etoposide, RARA for ATRA and THRA (thyroid hormone receptor, alpha) for tiratricol. These genes were selected as potential targets of combination therapy with ERBB2-targeted drugs.

Amplified Genes of Chromosome 17q Were Myc Pathway-Related Genes
Gene set enrichment analysis (GSEA) is a visualized method of the gene ontology analysis, which could help researchers finding a group of functional genes in a signaling pathway. Through GSEA, we found that all amplified genes of 17q are Myc-related genes. Of those, RARA, THRA, and CACNB1 were negatively correlated with Myc activation, which revealed minus value of enrichment score (ES). For instance, the ES value of RARA is −1.93 (FDR = 0.017, P = 0.004, Figure 2A), THRA is −1.85 (FDR = 0.035, P = 0.010, Figure 2B), and CACNB1 is −1.89 (FDR = 0.028, P = 0.004, Figure 2C). TOP2A and ERBB2 were positively associated with Myc activation (ES = 2.05, FDR = 0.002, P < 0.001, Figure 2D and (ES = 1.88, FDR = 0.077, P = 0.018, Figure 2E). These results implied that agonists of RARA, THRA, and CACNB1 or inhibitors of TOP2A or ERBB2 might be combined together for the treatment of ERBB2+ GC.

The Association Between Amplified Genes and Biological Significance
The association and pathway of amplified genes were analyzed by InBio Map TM . SH3BGRL3, SH3BGRL, and Neuregulin 2 (NRG2) are three candidate ligands of ERBB2. Both RARA and ERBB2 are upstream genes in Myc pathway ( Figure 3A). To clarify the clinical significance of amplified genes in ERBB2+ GC, these genes in two datasets from TCGA and GEO were further analyzed. The tendency of better prognosis was noted in RARA amplification group, compared to non-RARA amplification group of ERBB2+ GC in both TCGA cohort (HR = 0.628, 95% CI 0.242-1.539, P = 0.295, Figure 3B), and GEO cohort (HR = 0.312, 95% CI 0.104-0.942, P = 0.04, Figure 3C). In ERBB2+ GC of TCGA cohort, high levels of NRG2 and SH3BGRL were significantly related to poor prognosis (P = 0.009 and P = 0.042, respectively, Figures 3D,E), but not for SH3BGRL3 (P = 0.267, Figure 3F). These results implied the ligand function of NRG2 and SH3BGRL in cancer progression of ERBB2+ GC.

The Relationship Between Amplified Genes and Sensitivity to ERBB2-Targeted Drugs
Since the gene copy number, mRNA level and IC50 value to ERBB2-targeted drugs of multiple cancer cell lines were deposited in CCLE database, the drug responsive status of 17 GC cell lines was analyzed. Among those, four cell lines (NCI-N87, KE39, NUGC-4, and MKN-7) were ERBB2-amplified cell lines that showed lower IC50 (µM) to drug Lapatinib than other GC cell lines (P = 0.031, Figure 4A, left). The ERBB2-amplified cell lines also showed lower IC50 (µM) to drug Afatinib than other cell lines (P = 0.027, Figure 4A, right). The CNV of ERBB2 or RARA and their responses to targeted drugs Lapatinib or Afatinib of GC cell lines were listed in Table 1. Cell line NCI-N87 were more sensitive to ERBB2-targeted drugs, which is related to dual amplification of ERBB2 and RARA, while MKN7 was not sensitive to ERBB2-targeted drugs, which is related to non-amplification of RARA.
The relationship of simultaneous overexpression of NRG2, ERBB2, and RARA with Lapatinib sensitivity was listed in Table 2. Group 1: NRG2-L/ERBB2-L/RARA-L; Group 2: high expression of any one of NRG2, ERBB2, or RARA; Group 3: lower expression of any one of NRG2, ERBB2, or RARA; and Group 4: NRG2-H/ERBB2-H/RARA-H. The IC50 to Lapatinib of Group 4 was the lowest (P = 0.001). Group 3 is better than that in Group 2 for sensitivity to Lapatinib (Figure 4D).

ERBB2+ GC Strongly Associates With CIN Subtype of TCGA and Intestinal-Type of Lauren Classification
The clinical details and histology of TCGA cohort were recorded in cBioPortal database 9 , which provided opportunity to analyze correlation between genomic information and clinicopathological characteristics. For 59 cases of ERBB2+ 9 http://www.cbioportal.org/study?id~=~stad_tcga_pub#clinical GC, the pathology reports and slide images of 46 cases were uploaded in database ( Figure 5A). Fifty-three cases out of 59 ERBB2+ GC (89.83%) belong to chromosome instability (CIN) molecular subtype, and the others include three Epstein-Bar Virus-related (EBV) subtype, two genome stable (GS) subtype, and one microsatellite instability (MSI) subtype ( Figure 5B). Based on their pathological reports of ERBB2+ GC, all of those are intestinal-type GC in traditional Lauren classification, such as tubular adenocarcinoma, papillary adenocarcinoma, and a few mucinous adenocarcinomas.

Validating Study of Lapatinib and ATRA Drugs on Different GC Cell Lines
Based on data analysis of multiple GC cell lines ( Table 2)  and Lapatinib (between 0.005 and 0.5 µM) was observed on NCI-N87 cell line, but not in MKN-45 cell line (Table 3).
To verify the effect on Myc signaling pathway, the cytoplasmic protein and nuclear protein were extracted separately, and examined expressing levels after incubating Lapatinib, ATRA or both. As results, after 2 h incubation of Lapatinib (0.05 µM), ATRA (25 µM), or both on NCI-N87 and MKN-45 cells, obvious down-regulation of MYC protein levels of cytoplasm and nucleus was found in Lapatinib treated NCI-N87 cells, but bot ATRA did. Importantly, incubation of Lapatinib and ATRA simultaneously further reduced nuclear protein level of MYC in NCI-N87 cells ( Figure 6B, right). The similar effects were also observed on NCI-N87 cells after 48 h incubation by Lapatinib and ATRA, or both ( Figure 6C, right). However, these effects were not observed in MKN-45 cells (Figures 6B,C, left).

DISCUSSION
ERBB2 plays a critical role in carcinogenesis and cancer progression and has become a targeted therapeutic molecule in breast cancer and GC. Exploring ERBB2 related pathways and mechanisms of resistance will help oncologist to overcome drug resistance and enhance therapeutic effect of ERBB2-targeted therapy (Shi et al., 2016). Clinically, not all ERBB2+ GC respond well to ERBB2-targeted treatment due to innate resistance or acquired resistance (Kim et al., 2014;Piro et al., 2016). Therefore, exploring resistant mechanism of ERBB2+ GC is desirable. Han et al found that SRC inhibitor combined with trastuzumab can synergistically inhibit proliferation of ERBB2+ GC in vitro (Han et al., 2014). Two decades passed since trastuzumab was used for treatment of ERBB2+ breast cancer (Goldenberg, 1999;Shak, 1999; Sakamoto and Mitsuyama, 2000). Some ERBB family inhibitors, such as Lapatinib and Afatinib are also developed for treatment of ERBB2+ breast cancer or GC (Geuna et al., 2012;Huang and Rizzo, 2012;Lin et al., 2012;Schuler et al., 2012;Hecht et al., 2016). Large amount of clinical information as well as their corresponding genomic changes has been submitted to open databases. Cross-database analysis is a challenging work, but it helps researchers to get an overview of interested problems. In this study, we performed a cross-database data mining for factors involved in sensitivity or resistance to ERBB2-targeted therapy in ERBB2+ GC. We found that both ERBB2 and RARA are involved in Myc signaling pathway. Myc is the important hub gene of both ERBB2 and RARA. By data mining, a total of 47 genes were outlined as potential targets of treatment for ERBB2+ GC. We noticed that RARA, THRA, CACNB1, and TOP2A are targets of FDA-approved drugs. For instance, CACNB1 is a calcium channel protein and the inhibitor amlodipine is one of the most popular medications for treatment of high blood pressure (Gregg et al., 1993;Nussberger et al., 2008). TOP2A is the target of etoposide, which has been used in cancer treatment (Slater et al., 2001;Bartlett et al., 2015). RARA, which participates in regulation of cell development, differentiation, apoptosis and  granulopoiesis. ATRA is a RARA agonist, which revealed clinical efficacy in leukemia treatment (Degos and Wang, 2001). THRA is thyroid hormone receptor alpha, whose ligand tiratricol was used in treatment of hypothyroidism (Yazdanparast et al., 2006;Tylki-Szymanska et al., 2015). GSEA is a powerful tool to find biological pathways for associated genes. By GSEA method, all amplified genes ERBB2, RARA, THRA, CACNB1, and TOP2A are identified to be related to Myc signaling pathway. Myc plays important roles in carcinogenesis and cancer progression (Sun et al., 2011;Stine et al., 2015). In gastric carcinogenesis, pathogen infection of H. pylori and EBV could activate Myc pathway (Calcagno et al., 2008). Our cross-database analysis suggests that inhibitor of TOP2A and agonist of RARA, THRA, FIGURE 7 | The crucial molecules and regulatory pathways of ERBB2+ GC. The main regulatory pathway is NRG2-ERBB2-MYC, which is cross-interfaced with RARA-MYC pathway. For ERBB2+ GC, beyond ERBB2 itself, RARA and NRG2 are potential drug sensitive biomarkers too for predicting response to ERBB2-targeted therapy. Combination of NRG2 antagonist or RARA agonist with ERBB2-targeted drug may be effective regimens for ERBB2+ GC.
or CACNB1 may have synergistic efficacy in ERBB2-targeted therapy for ERBB2+ GC. Li and coworkers found that ERBB2 activated downstream Myc and promoted cells proliferation by phosphorylating AKT1 and Erk1/2 in cervical cancer . RARA could interact with Myc and regulate RARAdependent genes expression in leukemia cells (Uribesalgo et al., 2012).
In this study, three potential ligands (SH3BGRL3, SH3BGRL, and NGR2) of ERBB2 have been proposed. The latter has been reported as a ligand for ERBB3/ERBB4 (Chang et al., 1997;Schulze et al., 2005;Chiang et al., 2015). Slattery and coworkers found that NRG2 as a growth factor was involved in progression of breast cancer (Slattery et al., 2013). Sun and colleagues found that both NRG1 and NRG2 functioned as ligands of ERBB family and involved in progression of non-small cell lung cancer (Sun et al., 2009). Hitherto, the relationship of NRG2 and ERBB2 in GC is not clear yet. Our analysis disclosed that overexpression of NRG2 is significantly associated with poor prognosis of ERBB2+ GC. Some reports suggested that NRG2 can promote phosphorylation of ERBB2, ERBB3, and ERBB4 directly or indirectly (Sliwkowski et al., 1994;Chang et al., 1997;Mill et al., 2011). Therefore, NRG2 as a potential ligand of ERBB2 is worth further explored. Regarding to SH3BGRL3 and SH3BGRL, there is no any report on the relationship of these genes with ERBB2. In the present study, the simultaneous overexpression of ERBB2, NRG2, and RARA were correlated with increased sensitivity to Lapatinib. We speculate that ERBB2, NRG2, and RARA might act together as a functional network.
It has been reported that there are some histopathological features of ERBB2+ GC . ERBB2 positivity is more often observed in intestinal-type GC (He et al., 2013;Irkkan et al., 2017). Up-to-date, there is no correlation analysis on both levels of genomics and clinical histology for ERBB2+ GC. TCGA database provides a good resource for comparative analysis on both levels because it stores perfect genomics and histological information for GC samples. In 59 cases of ERBB2+ GC, 46 cases were accompanied by whole slide tissue images. ERBB2+ GC revealed characteristic histopathology. All 46 cases were classified into intestinal-type GC using Lauren classification, mainly including tubular adenocarcinoma, papillary adenocarcinoma, and a few mucinous adenocarcinomas. The majority (near 90%) of ERBB2+ GC was classified into CIN subtype on genomic level using TCGA molecular classification (Bass et al., 2014). These findings provide an important clue for clinical selection of ERBB2-targeted therapy.
In our validating study, we confirmed that GC cell line with amplified ERBB2 and RARA (NCI-N87) was sensitive to Lapatinib relative to GC cell without amplified ERBB2 and RARA . The sensitivity to ATRA on both cell lines did not show significant difference, which may attribute to some alternative pathways for ATRA-mediated apoptosis (Patrad et al., 2018). We found that Lapatinib and ATRA could synergistically inhibit cell growth in NCI-N87 cell, but not in MKN-45 cell. These biological effects were involved in interfering Myc signaling pathway. The similar effect was reported for breast cancer (Paroni et al., 2012).
In summary, by cross-database analysis, GC with increased expression of ERBB2, RARA, and NRG2 simultaneously should be taken as sensitive category to Lapatinib therapy. The current study proposed that NRG2-ERBB2-MYC is cross-interfaced with RARA-MYC pathway. Intervention of both routes will regulate hub gene Myc, and then inhibit cell growth for ERBB2+ GC (Figure 7). Moreover, combination of RARA agonist ATRA with ERBB2-targeted drug disclosed synergistic anticancer effect for ERBB2+ GC in vitro.

AUTHOR CONTRIBUTIONS
YY, XH, and ZZ were involved in concept and design. ZX, JW, and JZ acquired the data and performed data mining. ZX, JJ, and RY performed the experiments. WC and YY supported the analysis. All authors wrote, reviewed, and revised the manuscript.