LncRNAs GIHCG and SPINT1-AS1 Are Crucial Factors for Pan-Cancer Cells Sensitivity to Lapatinib

Lapatinib is a small molecule inhibitor of EGFR (HER1) and ERBB2 (HER2) receptors, which is used for treatment of advanced or metastatic breast cancer. To find the drug resistance mechanisms of treatment for EGFR/ERBB2 positive tumors, we analyzed the possible effects of lncRNAs. In this study, using CCLE (Cancer Cell Line Encyclopedia) database, we explored the relationship between the lncRNAs and Lapatinib sensitivity/resistance, and then validated those findings through in vitro experiments. We found that the expression of EGFR/ERBB2 and activation of ERBB pathway was significantly related to Lapatinib sensitivity. GO (Gene Oncology) analysis of top 10 pathways showed that the sensitivity of Lapatinib was positively correlated with cell keratin, epithelial differentiation, and cell-cell junction, while negatively correlated with signatures of extracellular matrix. Forty-four differentially expressed lncRNAs were found between the Lapatinib sensitive and resistant groups (fold-change > 1.5, P < 0.01). Gene set variation analysis (GSVA) was performed based on 44 lncRNAs and genes in the top 10 pathways. Five lncRNAs were identified as hub molecules. Co-expression network was constructed by more than five lncRNAs and 199 genes in the top 10 pathways, and three lncRNAs (GIHCG, SPINT1-AS1, and MAGI2-AS3) and 47 genes were identified as close-related molecules. The three lncRNAs in epithelium-derived cancers were differentially expressed between sensitive and resistant groups, but no significance was found in non-epithelium-derived cancer cells. Correlation analysis showed that SPINT1-AS1 (R = −0.715, P < 0.001) and GIHCG (R = 0.557, P = 0.013) were correlated with the IC50 of epithelium-derived cancer cells. In further experiments, GIHCG knockdown enhanced cancer cell susceptibility to Lapatinib, while high level of SPINT1-AS1 was a sensitive biomarker of NCI-N87 and MCF7 cancer cells to Lapatinib. In conclusions, lncRNAs GIHCG and SPINT1-AS1 were involved in regulating Lapatinib sensitivity. Up-regulation of GIHCG was a drug-resistant biomarker, while up-regulation of SPINT1-AS1 was a sensitive indicator.


INTRODUCTION
Lapatinib is a small molecular drug that has been shown to be a dual tyrosine kinase inhibitor, which is involved in the EGFR/HER1 and ERBB2/HER2 pathways and suppresses the autophosphorylation of these receptors. Clinically, it has been used in combination therapy with capecitabine in patients with advanced or metastatic breast cancer that overexpressed ERBB2/HER2 in the cases of previous treatment with anthracyclines, taxanes, or trastuzumab (Herceptin) (Geyer et al., 2006). In addition, a satisfactory response rate has also been found with Lapatinib treatment for ERBB2-positive progressive gastric cancer (Cetin et al., 2014;Satoh et al., 2014). However, in patients with head and neck squamous cell carcinoma, Lapatinib combined with radiotherapy did not show therapeutic effects (Harrington et al., 2015). Similarly, in ERBB2/EGFR positive metastatic bladder cancer patients who underwent first-line chemotherapy didn't get benefit from Lapatinib maintenance treatment (Powles et al., 2017). Therefore, uncovering the drug-resistant mechanism of Lapatinib will help improve the therapeutic effects of Lapatinib targeted therapy and find new sensitive biomarkers.
Long non-coding RNAs (lncRNAs) are a large class of transcribed RNA molecules that are longer than 200 nucleotides but do not encode proteins. In addition to the regulation of diverse cellular processes, such as epigenetics, cell cycle, and cell differentiation, they have been found to play important roles in carcinogenesis, tumor development, and treatment resistance (Heery et al., 2017;Peng et al., 2017;Hahne and Valeri, 2018;Wang et al., 2018;Wu et al., 2018). For instance, Ma et al. found that lncRNAs CASC9 and EWAST1 were two crucial molecules associated to EGFR-TKIs resistant in non-small cell lung cancer (Ma et al., 2017).
The Cancer Cell Line Encyclopedia (CCLE) database (https:// portals.broadinstitute.org/ccle) is an open access resource with the most completely integrated datasets of cancer cells genomes and drug effectiveness. It includes the experimental datasets of drug treatment of 24 kinds of chemical compounds in almost 1,000 cancer cell lines of various human cancers (Barretina et al., 2012). Kim et al. used CCLE database in their recent publication. They found that high levels of FGFR and integrin β3 are resistant to crizotinib treatment, suggesting that FGFR, and integrin β3 could be predictive markers for Met-targeted therapy (Kim et al., 2015). To date, there is a limited number of studies (Jiang et al., 2014;Niknafs et al., 2016;Bester et al., 2018;Li D. et al., 2018;Sun et al., 2018) to explore lncRNAs by CCLE database. In this study, we analyzed the lncRNAs of whole-genome datasets of CCLE after treatment with Lapatinib on pan-cancer cell lines, and proposed crucial lncRNAs GIHCG and SPINT1-AS1 involved in regulating Lapatinib sensitivity.

Data Extraction From CCLE
There are 5,344 lncRNA probes and 49,331 non-lncRNA probes in the whole-genome gene expression profile chip used in CCLE (Barretina et al., 2012). There are 1,037 cell lines of various cancer types in the database. Among those, 504 cell lines had been treated with Lapatinib and got IC50 (half maximal inhibitory concentration) data and 501 cell lines were examined by microarrays. Since the study focused on solid tumors, we deleted cell lines of hematopoietic and lymphoid cell lines. Finally, 420 solid tumor cell lines were enrolled in the study (Table 1).

Cancer Cell Lines and Cell Culture
Nineteen cancer cell lines were used for validating experiments in vitro. Four of those were gastric cancer cell lines AGS,, three were melanoma cell lines (MuM-2C, MV3, and A-375), three were hepatocarcinoma cell lines (LM3, 97L, and Huh7), three were thyroid cancer cell lines (KHM-5M, CAL-62, and C643), two were breast cancer cell lines (MCF7 and SK-BR-3), two were pancreatic cancer cell lines (TCC-PAN2 and BxPC3), and two were colorectal cancer lines

Cancer types Count
Autonomic ganglia 10         USA) according to the manufacturer's instructions. The siRNA sequences are shown in Table S1.

RNA Extraction and Quantitative Real-Time PCR Analysis
Total RNA was isolated using the TRIzol solution (Invitrogen, California, USA). The cDNA was synthesized using Reverse Transcription kit (TOYOBO, Japan). Real-time PCR was performed in 10 µl reaction mixtures with the HT 7900 (Applied Biosystems, Foster City, USA) using SYBR TM Select Master Mix (Applied Biosystems, Foster City, USA). The sequences of primers were designed and synthesized by Sunny Biotech (Shanghai, China): The primer sequences are shown in Table S1.

Cell Viability Assay
Five thousand cells of different cancers were placed in each well of 96-well plates (100 µl/well). Different concentrations of Lapatinib (Selleck, Houston, USA) were incubated for 48 h. After adding 10 µl CCK-8 for 2 h, OD value was measured at 450 nm by spectrophotometry (BioTek, Vermont, USA).

Data Analysis
The "corrplot" and "pheatmap" package in R software were utilized for visualizing pearson correlation analysis and cluster analysis by "euclidean" method. The Benjamini and Hochberg method was used to calculate P. adjust value. By means of "clusterProfiler" package in R, GSEA (Gene Set Enrichment Analysis) and GO (Gene Ontology) analyses were carried out to explore involved gene clusters. GSEA is a computational method based on previous publication by Subramanian et al. (2005). GO analysis is a kind of gene enrichment analysis to classify gene set on three aspects: molecular function, cellular component and biological process (Ashburner et al., 2000). Differentially expressed lncRNAs and genes with difference larger than 1.5-fold were obtained by "limma" package, which is often used to explore differentially expressed genes between two phenotypes (Ritchie et al., 2015). The top 10 gene clusters of all cancer cell lines were scored using "GSVA" package (Gene Set Variation Analysis,) in R language, which utilizes non-parametric unsupervised method for evaluating gene set enrichment (GSE) in transcriptomic data (Hanzelmann et al., 2013). Cytoscape software was applied to establish co-expression network and determine hub lncRNAs. The inhibiting ratio and Lapatinib IC50 were calculated according to OD value by GraphPad Prism 6.0 (Inc., La Jolla, CA, USA). The relative RNA levels were calculated by 2 − CT ( CT = LncRNA CT value − GAPDH CT value , CT= CT− CT min , CT min : minimum CT of expression levels of lncRNA GIHCG or SPINT1-AS1 in cell line). Student's t-tests were performed by GraphPad Prism 6.0. P < 0.05 was considered statistically significant.

Lapatinib IC50 From Pan-Cancer Cell Lines Analysis
The CCLE data of Lapatinib IC50 of the selected 420 cell lines was shown in Table 2. The upper limit of IC50 was originally determined as 8 µM for those cancer cell lines in the database. There were 302 cancer cell lines with IC50 higher than 8 µM, which were insensitive to Lapatinib drug. There were 118 cancer cell lines with IC50 lower than 8 µM, which were relatively sensitive to Lapatinib drug. Taking 8 µM of IC50 as a threshold, we categorized 420 cancer cell lines into two groups, high_IC50 (n = 302) and low_IC50 (n = 118). Since EGFR and ERBB2 are the targets of the Lapatinib drug, the expression levels of EGFR, and ERBB2 in high_IC50 and low_IC50 groups were analyzed. The expression levels of EGFR and ERBB2 were significantly higher in low-IC50 group than in high_IC50 ( Figure 1A, P = 0.006 and P < 0.001, respectively). The distribution tendency of 22 types of solid cancer cell lines in high-IC50 (up to 8 µM) and low_IC50 (lower than 8 µM) groups is presented in Figure 1B. GSEA analysis showed that ERBB pathway-related genes were enriched in low_IC50 group (Figure 1C, ERBB signaling pathway NES = −1.81, P < 0.002, p. adjust = 0.064; regulation of ERBB signaling pathway NES = −1.69, P < 0.002, p. adjust = 0.064).

Pathway Analysis Involved in Lapatinib Sensitivity
To illustrate the mechanism of Lapatinib resistance, we selected genes with fold-change >1.5 times to perform GO analysis (Table  S2). In the top 10 involved pathways, Lapatinib sensitivity was positively associated with cell keratin, epithelial differentiation, The enrichment analysis of ERBB signaling pathway reveals that ERBB signaling pathway is significantly enriched in Lapatinib low_IC50 group. "Y" axis indicates the enrichment score (ES) value, and "X" axis indicates genes according to differential expression value between high_IC50 and low_IC50 groups. The blue and red dot curves represent ES value. The bottom barcodes represent the leading gene set that strongly contributed to ES value. The positive ES value represents positive correlation to Lapatinib IC50, and minus ES value represents negative correlation to Lapatinib IC50. FIGURE 2 | The network of top 10 genes by GO pathway analysis. The large spots in the center of the networks are the gene clusters, and the small spots connected with large spots are the related genes in the pathways. Red spots indicate that the genes are highly expressed in the high_IC50 group. Green spots indicate that the genes are highly expressed in the low_IC50 group. The darker red or green spot are the larger fold-change of differential genes. The black spots with different sizes and numbers on the right side indicate the gene numbers in the gene clusters. and cell-cell junction, while negatively related to signatures of extracellular matrix (Figure 2, P < 0.001, P. adjust < 0.001).

Analysis of LncRNAs Involved in Lapatinib Sensitivity
We further screened the differentially expressed lncRNAs, and 44 lncRNAs were identified between the high_IC50 group and low_IC50 group ( Figure 3A and Table 3, fold-change >1.5, P < 0.01). Then, we selected genes in the top 10 pathways and 44 differential lncRNAs for the construction of the co-expression network. The enrichment scores of the top 10 pathway genes in every cancer cell lines were calculated and determined by GSVA analysis. Five lncRNAs were highlighted as the hub factors in the top 10 regulating pathways ( Figure 3B). The association of the 5 lncRANs with 199 genes in the top 10 pathways was further analyzed, and a molecular network of coexpression was established, which included top 50 key molecules closely associated to Lapatinib sensitivity. Three crucial lncRNAs, GIHCG, SPINT1-AS1, and MAGI2-AS3, still remained in the co-expression network ( Figure 3C).

Differential Expressing Analysis of Three LncRNAs Between Epithelial and Non-epithelial Cancer Groups
We divided the 420 cancer cell lines into epithelium derived group (n = 278) and non-epithelium derived group (n = 142; including nervous system, bone, cartilage, and pleura). The differential expression levels of the three lncRNAs between the FIGURE 3 | Screening lncRNAs related to Lapatinib sensitivity. (A) The heatmap of 44 differentially expressed lncRNAs between high_IC50 group and low_IC50 groups (fold-change >1.5, P < 0.05). The red bars on the top present high_IC50 cases, and blue bars represent low_IC50 cases. The numbers of the right side are the names of lncRNAs. The numbers tagged in lncRNAs represent probe codes. (B) The co-expression molecular network of the 44 differentially expressed lncRNAs. The red ovals represent five crucial lncRNAs in the network, and the purple rectangles outside indicate the top 10 functional gene sets by GO analysis. (C) The co-expression molecular network of the top 50 differentially expressed genes and lncRNAs between the high_IC50 group and the low_IC50 group. In this network, three of differentially expressed molecules are lncRNAs (SPINT1-AS1, MAGI2-AS3, and GIHCG), which are underlined. The colors nodes of the network from red, dark yellow to light yellow indicate gradually weakened correlation to Lapatinib sensitivity. two groups are presented in Figure 4A. In the epitheliumderived group, the differential expression levels of the three lncRNAs between Lapatinib high_IC50 and low_IC50 groups were significantly different ( Figure 4B, P < 0.05). In the nonepithelium groups, there was no significant difference of the three lncRNAs between Lapatinib high_IC50 and low_IC50 groups. Higher expressing level of SPINT1-AS1 was found in epitheliumderived cancer cells, and higher expressing levels of MAGI2-AS3 and GIHCG were observed in the non-epithelium group.
Differentially expressed genes (1.5-fold change) between the Lapatinib high_IC50 and low_IC50 groups in epithelial group (Table S3) were utilized to perform GO analysis. Enhanced signatures of cell keratin, epithelial differentiation, and cell-cell junction were observed in Lapatinib low_IC50 group, and decreased signature of extracellular matrix were observed in Lapatinib low_IC50 group (Figure 5, P < 0.001, P. adjust < 0.001).

Correlation of LncRNAs SPINT1-AS1, GIHCG, or MAGI2-AS3 and Lapatinib Sensitivity in Epithelial Group
Correlation analysis revealed that Lapatinib IC50 of the nonepithelial group was higher than that of the epithelial group ( Figure 6A). Of the three critical lncRNAs, SPINT1-AS1, and GIHCG were the lncRNAs most correlated to Lapatinib sensitivity ( Figure 6B). SPINT1-AS1 and GIHCG were selected as key factors of affecting Lapatinib sensitivity of epithelial cancers. The up-regulation of SPINT1-AS1 was found in low_IC50 group and increased GIHCG was found in high_IC50 group ( Figure 6C).

Validating Study of GIHCG and SPINT1-AS1 on Regulating Lapatinib Sensitivity in vitro
In validating experiments, we examined expression levels of GIHCG and SPINT1-AS1 in seven types of cancer cell lines (thyroid cancer, pancreatic cancer, liver cancer, melanoma, gastric cancer, breast cancer, and colorectal cancer) and Lapatinib IC50 of the same cancer cell lines. Correlation analysis showed that higher expression levels of SPINT1-AS1 were significantly associated with lower Lapatinib IC50 (Figure 7A, R = −0.715, P < 0.001), while higher expression levels of GIHCG were significantly related to higher Lapatinib IC50 (Figure 7A, R = 0.557, P = 0.013).
The sensitive cancer cell lines of NCI-N87 (gastric cancer) and MCF7 (breast cancer), as well as the resistant cancer cell lines of NCIH-747(colon cancer) and BxPC3 (pancreatic cancer)   were selected for a subsequent validating study. After knockingdown expression levels of GIHCG and SPINT1-AS1 by small interfering RNAs, Lapatinib IC50, and inhibitory rate of cancer cells were detected. Among three small interference sequences of GIHCG and SPINT1-AS1 mRNAs, siRNA sequence 3 of GIHCG (Si3, Figure 7B), and siRNA sequence 1 of SPINT1-AS1 (Si1, Figure 7C) were identified as effective siRNAs for further experiments.
Knocking-down of GIHCG could significantly enhance the sensitivity to Lapatinib in MCF7 and BxPC3 cancer cell lines (Figure 7D), while down-regulation of SPINT1-AS1 could promote resistance to Lapatinib in NCI-N87 and MCF7 cancer cell lines ( Figure 7E). To clarify whether there is a mutual regulatory relationship between GIHCG and SPINT1-AS1, we detected the expression level of SPINT1-AS1 after GIHCG knockdown and vice versa. As shown in Figures 7F,G,

DISCUSSION
LncRNA is an important regulatory molecule in drug resistance during chemotherapy or gene targeted therapy (Li et al., 2016;Dong et al., 2018;Wu et al., 2018;Zhou et al., 2018). In this study, we analyzed Lapatinib sensitivity to EGFR and ERBB2 targeted therapy pan-cancer cell line wide. We noticed that Lapatinib sensitivity was not only positively correlated to the activation of EGFR and ERBB2 signaling pathways, but also positively associated to cell keratin, epithelial differentiation, and cell-cell junction. The Lapatinib sensitivity of cancer cell lines was negatively associated to extracellular matrix signature. By screening differentially expressed lncRNAs and establishing coexpression network between Lapatinib high_IC50 and low_IC50 groups, three key lncRNAs, SPINT1-AS1, GIHCG, and MAGI2-AS3, were found. Of those, GIHCG and SPINT1-AS1 were only differentially expressed in epithelial derived cancers. SPINT1-AS1 was negatively related to Lapatinib IC50, whereas GIHCG was positively associated to Lapatinib IC50. By siRNAs treatment, downregulation of SPINTA-AS1 could promote Lapatinib resistance, while downregulation of GIHCG promoted Lapatinib sensitivity. The combination of bioinformatical approach and experimental study confirmed that lncRNAs were involved in regulating sensitivity to Lapatinib targeted therapy.
PI3K/Akt, Ras/Raf/MEK/ERK1/2, and PLCγ pathways are downstream pathways of EGFR and ERBB2 and play important roles in cell proliferation and survival of multiple cancers FIGURE 5 | Pathway analysis of Lapatinib sensitivity related genes. The genes in the top 10 pathways with fold-change more than 1.5 are used between Lapatinib high_IC50 and low_IC50 groups. The middle brown dot of each network indicates the name of a gene set, and the small dots surrounding it indicate the genes of the gene set. The red dots represent the up-regulated genes in the high_IC50 group, and the green dots represent the up-regulated genes in the low_IC50 group. The darker red or green spot are the larger fold-change of differential genes. The black spots with different sizes and numbers on the right side indicate the gene numbers in the gene clusters. (Roskoski, 2014). The expression levels of EGFR and ERBB2 are positively correlated to Lapatinib sensitivity (Rusnak et al., 2007;Xiang et al., 2018). Trastuzumab (Herceptin) is a molecular targeted drug of ERBB2-positive metastatic/advanced breast cancer and gastric cancer (Bang et al., 2010;Loibl and Gianni, 2017). Lapatinib is a small molecule chemical, which proved effective for ERBB2-positive advanced or metastatic breast cancer when combined with capecitabine after previous treatment with anthracyclines, paclitaxel, or trastuzumab (Geyer et al., 2006). In gastric cancer, treatment with Lapatinib plus capecitabine and oxaliplatin also revealed anti-cancer effects on HER2amplified gastroesophageal adenocarcinoma, especially in Asian and younger patients (Hecht et al., 2016). LncRNAs emerged as one of the new resistance mechanisms to chemotherapy or molecule targeted therapy. By bioinformatics analysis, Lapatinib sensitive cancer cells exhibited enrichment of genes related to cell keratin, epithelial differentiation, and cell-cell junction. The ERBB family plays an important role in regulating cell differentiation (Pellat et al., 2017). We noticed that Lapatinib sensitivity is positively correlated to ERBB pathway activation. It means that cancer cells sensitive to Lapatinib drug often showed enrichment of cell differentiation-related genes, while Lapatinib-resistant cancer cells are often accompanied by enrichment of extracellular matrix pathway (D'Amato et al., 2015;Khan et al., 2016;Lin et al., 2017;Watson et al., 2018). Furthermore, increases of extracellular matrix could further induce epithelial-mesenchymal transition of cancer cells (Tzanakakis et al., 2018). Although the role of lncRNAs in cancer progression and Lapatinib resistance have been reported in other studies (Russell et al., 2015;Li et al., 2016;Liang et al., 2018;Ma et al., 2018), this is the first study that proved that lncRNAs GIHCG and SPINT1-AS1 are involved in regulating therapeutic sensitivity to Lapatinib. Based on pan-cancer cell lines analysis, Lapatinib IC50 is significantly different between non-epithelial cancer cell lines, and epithelial cancer cell Lines. As the inhibitor of miR-200b/200a/429, LncRNA GIHCG was shown effectively promoting the progression of liver cancer through inducing methylation of miR-200b/200a/429 promoter (Sui et al., 2016). GIHCG is also involved in promoting cancer proliferation and migration in tongue and renal cancers (D'Aniello et al., 2018;Ma et al., 2018). However, there is no study $om whether or not GIHCG could regulate Lapatinib drug sensitivity in cancers. LncRNA SPINT1-AS1 is a Kunitz type 1 antisense RNA1, belonging to serine peptidase inhibitor. An increased expression of SPINT1-AS1 has been observed in colorectal cancer . It is also the first time that lncRNA SPINT1-AS1 has been found regulating Lapatinib drug sensitivity on multiple cancer cells. In validating experiments, the knockdown of SPINT1-AS1 did not result in the up-regulation of GIHCG. We speculated that GIHCG may regulate SPINT1-AS1 expression through regulating promoter methylation or by manner of competitive endogenous RNA (ceRNA) (Zhang G. et al., 2018;Zhang L. et al., 2018). However, the mutual regulatory mechanisms of lncRNA GIHCG and SPINT1-AS1 remain to be studied in the future.

CONCLUSION
In conclusion, the current study proposed a group of lncRNAs related to Lapatinib sensitivity based on pan-cancer cell lines analysis. By subsequent experimental study, lncRNAs GIHCG and SPINT1-AS1 were firstly identified as crucial lncRNAs in regulating Lapatinib resistance or sensitivity in epithelium-derived cancer cell lines. SPINT1-AS1 is a Lapatinib sensitivity predictor, while GIHCG is a predictive molecule for Lapatinib resistance.

ETHICS STATEMENT
The protocols used in this study were approved by Rui Jin Hospital Ethics Review Boards. Written