GABRP promotes CD44s-mediated gemcitabine resistance in pancreatic cancer

GEO data mining

GEO (https://www.ncbi.nlm.nih.gov/gds) provides a platform for data collection, processing, and normalization. We performed integrated analyses of GEO datasets of our interest (GSE15471 (Badea et al., 2008), GSE28735 (Zhang et al., 2012) and GSE36563 (Van den Broeck et al., 2012)) that contained matching samples either between normal and pancreatic tumor samples or parental and gemcitabine-resistant tumor cells (Idichi et al., 2017; Zhang et al., 2013). Differentially expressed genes (DEGs) of drug resistance versus non-drug treatment crossing the tumor or normal samples were overlapped by a Venn diagram. A p-value < 0.01, and an absolute value of —logFC (fold change) — > 2 were considered as the cut-off criteria.

Cancer cell line encyclopedia database analysis

The Cancer Cell Line Encyclopedia from the Broad Institute of MIT and Harvard (https://portals.broadinstitute.org/ccle) is a public resource containing gene expression and sequencing information of nearly 1,000 human cancer cell lines (Barretina et al., 2012). We used this resource to analyze the distribution of GABRP mRNA expression in cell lines derived from different tumor types.

Overall survival analysis

The five-year survival rate was determined by a Kaplan–Meier analysis between GABRP high- and low-expression groups using 178 pancreatic cancer patient samples and 4 normal samples. The log-rank test was performed to determine survival differences between the two groups. Differences were considered statistically significant at p < 0.05. The starBase website (http://starbase.sysu.edu.cn/) offered differential gene expression and survival analyses across 32 cancer types that were integrated from The Cancer Genome Atlas (Li et al., 2013).

Cell culture

CFPAC1-CD44^high- and CD44^low-expressing cells were initially separated by flow cytometry (Zhao et al., 2016). Single clones for CD44^high expression, and stable knockdown CD44 cells, were generated in CFPAC1 cells. We used lentiviral vector-mediated transfection methods as described in our previous publication (Zhao et al., 2016). The cell lines were grown in RPMI 1640 medium supplemented with 10% fetal bovine serum (Gibco, Thermo Fisher Scientific, USA) and a 1% antibiotic mixture (100 U/mL penicillin and 100 µg/mL streptomycin) (Biotech, China). The cells were maintained in an incubator at 37 °C under 5% CO₂.

Cell viability assay

Cells were seeded in triplicate in 96-well culture plates at a density of 3.0 × 10³ cells per well, and treated with gemcitabine at different concentrations for 3 days. Cells were then incubated with MTT (Life Technologies, Thermo Fisher Scientific, China) solution (10 µL/well) and cultured at 37 °C for 3 h. Absorbance of each well was measured at 450 nm using an automatic microplate reader (Biofil, China). The cell viability percentage was calculated using the following equation: (absorbance of drug-treated sample/absorbance of untreated sample) × 100. All the experiments were repeated at least three times.

Quantitative real-time polymerase chain reaction

Total RNA was extracted from all cell lines using TRIzol (Life Technologies, Carlsbad, CA, USA) according to the manufacturer’s protocols. The RNA concentration was determined using a Nanodrop spectrometer (SMA4000 UV-VIS spectrophotometer, Merinton, China). One microgram of high-quality RNA was reverse-transcribed into cDNA (CWBio, Jiangsu, China). Then, 50 ng of cDNA were used to perform the quantitative real-time polymerase chain reaction using the SYBR Green kit (ABI Systems, Mississauga, ON, Canada). Samples were prepared in triplicate and normalized to β-actin using the 2^ΔΔCt method. The results represent means ± standard deviation. Primer sequences of GABRP and ß-actin genes were list as below: GABRP forward primer: 5′-GCCCTAACAGAGCCTCAACA-3′; Reverse primer: 5′-TTGTCACTTCTGCCGACCTC-3′. ß-actin forward primer: 5′-GACCAATCCTGTCACCTC-3′;

Reverse primer: 5′-GATCTCCGTTCCCATTAAGAG-3′.

Transfection of pancreatic cancer cells

shRNA against CD44 were designed by the company (Genecopoeia, Rockville, MD, USA). The plasmid coated with Fugene 6 lipofectin, transfected to 293t cells, accompanied with lentiviral vectors (gift from Dr. Senlin Li’s lab, University of Texas Health Science Center, San Antonio, TX, USA). After packaging the live virus, the supernatant from 293t cells was harvested and applied to CD44^high-expressing cells. Successful transfected cells were subsequently selected by puromycin for at least two weeks since the shRNA vector contained puromycin resistant gene as the selection marker. Likewise, the overexpression of CD44 vector was transfected using adenovirus and selected by puromycin resistant antibiotics for another 14 days. These cells were validated by qPCR experiments and western blotting analysis.

Colony formation assay

CD44^high-expressing pancreatic cancer cells and two knockdown of CD44 single clone cells were counted and plated at the density of 500 cells per well. After 14 days, the cells were fixed by methanol and stained by 0.1% crystal violet. The cell numbers were counted under the microscope and pictured in prism software. Three independent experiments were conducted to calculate the standard deviation.

Invasion assay

Silence of CD44 (shCD44) in CD44^high-expressing cells and its counterpart CD44^high-expressing cells were seeded at the density of 5 × 10⁴ cells in the upper chamber of the insert (Corning, USA) with the regular fresh RPMI medium (500 ul/insert) while the bottom chamber was fulfilled with the complete RPMI medium containing 10% FBS (700 ul/well) for 24 h. Then, the insert was washed by warm 1xPBS twice. And the cells were fixed by 80% ethanol for 15 min followed by 0.1% crystal violet staining for another 20 min. Fresh tap water was used to clean the stained insert and the well. At the end, the insert was wiped using the cotton swap to dry. The invasive cells were counted under the inverted microscope. Each group has three replicates and the experiments were repeated at least three times.

Western blot analyses

CD44^high-expressing cells and two short hairpin (sh)RNA-CD44 single clone cells were lysed using radioimmunoprecipitation assay buffer (Beyotime, Shanghai, China) supplemented with protease and phosphatase inhibitors (tablet mini EDTA-free, Roche, Basel, Switzerland) on ice for 30 min. After centrifugation at 10,000 g for 5 min, the lysates were centrifuged at 10,000 g at 4 °C for 5 min. The supernatant was collected and the cell debris was discarded. Subsequently, the total protein concentration was determined by the bicinchoninic acid assay (CWBio, Cambridge, MA). A total of 50 µg of protein from each sample were separated by 8% sodium dodecyl sulfate polyacrylamide gel electrophoresis and transferred onto polyvinylidene difluoride membranes (Bio-Rad, Hercules, CA, USA). Afterwards, the membranes were blocked with 0.5% bovine serum albumin for 1 h and incubated with primary antibodies against CD44, as well as β-actin as the loading control (1:2000; biodragon, China). Then, the membranes were washed three times with 1 ×, tris-buffered saline/0.1% Tween 20 and incubated with secondary antibodies conjugated with horseradish peroxidase (anti-mouse IgG, 1:2000; Beyotime, Shanghai, China) for 1 h at room temperature. The protein bands were visualized with chemiluminescence, and the images were captured on a visualization instrument (Bio-Rad, Hercules, CA, USA).

Statistical analyses

All data are presented as means ± standard deviation. Prism 7 software (GraphPad, San Diego, CA, USA) was applied to perform Student’s t-test andANOVA test. P < 0.05 was considered statistically significant.

DEGs comparing GABRP between human pancreatic cancer and normal tissues

Overcoming chemoresistance in pancreatic cancer remains the top priority among all therapies. To identify potential functional molecules selected using bioinformatics tools, we chose three independent GEO datasets: GSE15471, which contained pairs of normal and tumor tissue samples (Badea et al., 2008; Idichi et al., 2017); GSE28735, which compared the microarray gene expression profiles of 45 matching pairs of pancreatic tumors and adjacent non-tumor tissues (Zhang et al., 2013; Zhang et al., 2012); and GSE36563, which identified a side population resistant to gemcitabine (T3 vs. T4 chemoresistance) in human PDAC. The results uncovered a chemoresistant and CSC-associated phenotype (Van den Broeck et al., 2012). The microarray data were retrieved from the GEO dataset provided by the National Center for Biotechnology Information website and were initially analyzed by GEO2R software. Log2 fold change (FC) > 2 and p values < 0.01 were set as the cutoff criteria for subsequent analysis. Interestingly, GABRP was still identified as the overlapping gene GABRP gene (Fig. 1A). Therefore, we selected GABRP as our candidate gene to perform the various functional assays related to chemoresistance in PDAC.

To explore the clinical relevance of GABRP in pancreatic cancer patients, we took advantage of starBase V3.0 (http://starbase.sysu.edu.cn/) (Li et al., 2013) where 178 cancer and four normal samples were analyzed in pancreatic adenocarcinoma (PAAD). Expression data showed that GABRP expression increased in cancer patients compared to that in the normal samples presented by the box-whisker plot (Fig. 1C). Subsequently, the Kaplan–Meier overall survival curve revealed that PAAD patients with lower GABRP expression survived significantly longer than those with relatively higher expression of GABRP, with 89 patients in both groups. The hazard ratio was 1.49 and the log-rank p value was 0.056 (Fig. 1D). Collectively, the discriminatory ability of GABRP can be reflected on DEGs and the overall survival rate in PAAD. This indicated that the GABRP signature could be a prognostic biomarker for predicting the patients’ lifespan.

Functional role of GABRP in a pancreatic cancer cell lines model

To validate the previously mentioned clinical data in our cell line model, we first assessed the mRNA expression of GABRP in various organs and cancer cell lines using the Cancer Cell Line Encyclopedia: https://portals.broadinstitute.org/ccle/page?gene=GABRP. This analysis showed relatively higher mRNA levels of GABRP in the upper aerodigestive system, bile ducts, and pancreas (Fig. 2A). RNAseq data offered a comparison of GABRP among different cancer cell lines and organs. Notably, the upper aerodigestive system, bile ducts, and pancreas again showed relatively higher GABRP mRNA levels compared to other organs. This is similar to the results indicated by the Affymax data (Fig. 2B). Because both platforms indicated that the pancreas contained higher levels of GABRP, we took advantage of our pancreatic cancer CFPAC-1 cells to test whether GABRP played an indispensable role in pancreatic cancer progression. First, CFPAC-1-CD44^high and CD44^low expressing cells were separated out using flowcytometry which obtained from our previous published work (Zhao et al., 2016). Then, we established shRNA-CD44 knockdown cells as described in our previous publication (Zhao et al., 2016). Single clone (s.c.) cells of shCD44 had been generated using limited dilution over 2 months period time. To eliminate clonal variation, we selected two clones labeled cl.1 and cl.2 that are depicted by gray and blue lines, respectively (Figs. 3A & 3B). On one hand, silencing CD44 using shCD44 in CD44^high-expressing cells endowed those cells with less sensitivity to gemcitabine while CD44^high-expressing cells were gemcitabine resistant (Fig. 3A), gemcitabine inhibited shCD44 cell proliferation in a dose dependent manner which 10 ng/ml, 20 ng/ml, 30 ng/ml and 40 ng/ml had significant reduction effect on cell proliferation than those lower doses in CFPAC-1-shCD44 cells (Fig. 3A, *p < 0.05, ***p < 0.001,****p < 0.0001). On the other hand, two CD44s overexpression single clones were successfully created in CD44^low-expressing cells (Zhao et al., 2016). Cell viability assay showed that overexpression of CD44s drove higher gemcitabine resistance than those CD44^low-expressing cells (Fig. 3B, *p < 0.05, **p < 0.01, ***p < 0.001, ****p < 0.0001). Consistently, two CD44s overexpression cells promoted the gemcitabine resistance significantly in a dose dependent manner even at the small dose starting from 2.5 ng/ml to higher dose of 40ng/ml (Fig. 3B). Next, we performed a series of functional assays including colony formation assay, invasion assay, real-time quantitative PCR and western blotting techniques using two sets of cell line models: CD44^high-expressing cells versus shCD44 counterpart knockdown cells and CD44^low-expressing cells versus CD44s overexpression cells. Colony formation suggested two shRNA-CD44 single clone cells from CD44^high-expressing cells decreased colonies critically than those CD44^high-expressing cells (Fig. 3C, ***p < 0.001, ****p < 0.0001). The invasion assay displayed that both silence of CD44 single clone cells detected significantly less invasive numbers of cells than the CD44^high-expressing cells (Fig. 3D). Western blot analysis confirmed that we successfully diminished CD44s standard isoforms in shRNA-CD44 s.c cells, with ß-actin serving as the loading control (Fig. 3E). Lastly, CFPAC1 cells (CF-ctrl) were treated with increased concentrations of gemcitabine until 1,000 ng/ml over two months to generate gemcitabine resistant cell line depicted as CF-GR. Interestingly, we found both CD44 and GABRP protein levels elevated in gemcitabine resistant cells compared to the control none drug treated cells (Fig. 3G). These results indicated that CD44 and GABRP expressions were positively correlated and may contributed to gemcitabine resistance in pancreatic cancer cells. Another possibility is GABRP may potentially serve as a downstream target of CD44. Therefore, we knockdown CD44 from CD44^high-expressing cells to test whether GABRP expression changed upon CD44 inhibition. Within our expectation, the expression of GABRP mRNA was reduced dramatically in CFPAC-1- CD44 Hi -shCD44 cells compared to CD44 Hi cells (Fig. 3F). Collectively, CD44s was required for gemcitabine resistance because its knockdown in both single clone cells significantly decreased cell growth, and these cells were more drug sensitive compared to the control cells (Fig. 3A). GABRP may contribute to CD44-induced gemcitabine resistance in pancreatic cancer (Fig. 3F).

Clinical relevance of CD44 and GABRP in pancreatic cancer patients

To validate the clinical significance of CD44 and GABRP in patients with pancreatic cancer, a Spearman’s correlation analysis was conducted based on the starBase V3.0 website to analyze co-expression of CD44 and GABRP in various cancers (Fig. 4A). The top cancers with significant p-values are listed in the Fig. 4. The highest Spearman’s correlation coefficient R values of the top three cancer types were colon adenocarcinoma (R = 0.377), pancreatic adenocarcinoma (R = 0.369), and stomach adenocarcinoma (R = 0.314, Fig. 4A). All three of these cancer types are digestive tract cancers, suggesting digestive cancer tissues from patients tend to express both CD44 and GABRPmolecules. Using 178 pancreatic adenocarcinoma patient samples, CD44 and GABRP expressions were significantly co-expressed in the linear regression model. This suggests that CD44 may directly regulate GABRP expression to affect chemoresistance. However, further experiments are required to verify our hypothesis.