Pharmacogenomic Analysis Reveals CCNA2 as a Predictive Biomarker of Sensitivity to Polo-Like Kinase I Inhibitor in Gastric Cancer

Despite recent innovations and advances in early diagnosis, the prognosis of advanced gastric cancer remains poor due to a limited number of available therapeutics. Here, we employed pharmacogenomic analysis of 37 gastric cancer cell lines and 1345 small-molecule pharmacological compounds to investigate biomarkers predictive of cytotoxicity among gastric cancer cells to the tested drugs. We discovered that expression of CCNA2, encoding cyclin A2, was commonly associated with responses to polo-like kinase 1 (PLK1) inhibitors (BI-2536 and volasertib). We also found that elevated CCNA2 expression is required to confer sensitivity to PLK1 inhibitors through increased mitotic catastrophe and apoptosis. Further, we demonstrated that CCNA2 expression is elevated in KRAS mutant gastric cancer cell lines and primary tumors, resulting in an increased sensitivity to PLK1 inhibitors. Our study suggests that CCNA2 is a novel biomarker predictive of sensitivity to PLK1 inhibitors for the treatment of advanced gastric cancer, particularly cases carrying KRAS mutation.


Introduction
Gastric cancer is one of the most common malignant tumors of the gastrointestinal track [1]. Due to complex molecular mechanisms and clinical heterogeneity, clinical outcomes for patients with advanced gastric cancer remain poor, with a 5-year survival of 5-20% and a median overall survival (OS) of 10 months [2].To date, only two targeted therapies, treatment with trastuzumab (HER2 inhibitor) or ramucirumab (VEGFR2 inhibitor), have been approved for the treatment of advanced gastric cancer in patients carrying relevant biomarkers and development of more targeted therapeutic strategies for gastric cancer is needed.
Polo-like kinase 1 (PLK1), a mitotic serine/threonine protein kinase, regulates various cellular events throughout the cell cycle and has been shown to potentially be a new target in cancer treatment [3]. Increased expression of PLK1 has been observed in several types of malignant tumors and has been shown to be correlated with lower survival rates among solid tumor patients [4,5]. Meanwhile, several PLK1 kinase inhibitors have been developed as anticancer drugs and are currently being evaluated in clinical trials [6]: BI-2536, a dihydropteridinone compound and potent ATP-competitive PLK1 inhibitor [7], was found to inhibit cell proliferation in several human cancer cells, including breast, colon, lung, pancreas and prostate cancer [8]. Building on these results, BI-2536 became the first selective PLK1 inhibitor investigated in clinical trials of patients with solid tumors [9] and exhibited an acceptable safety profile in phase I clinical trials. In phase II study, however, BI-2536 showed relatively poor clinical efficacy, with only 4.2% of patients achieving a partial response in treatment of stage IIIB/IV non-small cell lung cancer [10]. Similar clinical data for BI-2536 were observed in another study of advanced solid tumors [11]. In light of these reports, further clinical studies of BI-2536 as a monotherapy have garnered little interest [10,11]. However, identifying a patient selection biomarker may help to overcome the inefficiency associated with BI-2536 monotherapy and assist with identifying patients who may better respond to treatment with PLK1 inhibitors. Indeed, research has shown that KRAS mutant cancer cells are highly sensitive to PLK1 inhibition [12], wherein cancer cells carrying the oncogenic mutation KRAS were sensitive to PLK1 depletion by shRNA or to treatment with PLK1 inhibitors. However, detailed mechanisms of the actions of PLK1 inhibitors on KRAS mutant cancers are largely unknown.
While several drugs targeting KRAS G12C mutant cancer sare under clinical trials [13], the KRASG12C mutation is very rare in gastric cancer: only 3.6% of KRAS mutant gastric cancer patients have the mutation according to combined cohort datasets in the cBioPortal (http://www.cbioportal.org). Therefore, development of alternative therapies will be significant for treatment of KRAS mutant gastric cancers.
In this study, we reviewed toxicity screens of 1345 FDA-approved, small-molecule pharmacological compounds and investigational anticancer compounds against a panel of 37 gastric cancer cell lines. Using elastic net regularization, we generated statistical models that predicted the sensitivity of gastric cancer cells to each of the tested drugs based on mRNA expression features, which allowed us to identify distinct drug-biomarker relationships. By focusing on an observed relationship between PLK1 inhibitors and CCNA2, we discovered that oncogenic KRAS mutation drives CCNA2 upregulation and consequent mitotic catastrophe and apoptosis in the presence of PLK1 inhibitors.

Pharmacogenomic Analysis Highlights Novel Drug-Biomarker Relationships Among Gastric Cancer Cells
We previously screened seven gastric cancer cell lines against 1345 pharmaceutical compounds and selected 63 compounds that induced a greater than 50% decrease in cell viability in at least four of the seven cell lines after 72 h of exposure [14]. In this study, we expanded this to 37 gastric cancer cell lines and to 75 compounds targeting cell proliferation, survival and signal transduction pathways (Figure 1a,b). Cell line-specific responses to each of the 75 drugs were calculated by estimating the mean area under survival curves in duplicate ( Figure 1c and Table S1). Using elastic net regularization, we generated statistical models that predicted the sensitivity of gastric cancer cells to each of the tested drugs according to mRNA-based gene expression features. In result, we found 23 biomarkers that predicted sensitivity among gastric cancer cells to nine drugs under bootstrapping (random sampling of cell lines with replacement) and a frequency threshold of 75% (Figure 1d and Figure S1). Intriguingly, CCNA2, encoding cyclin A2, which regulates cell cycle progression during the S phase and in G2/M transition, was commonly associated with responses to PLK1 inhibitors BI-2536 and volasertib (BI-6727) (Figure 1d). The concordant associations with CCNA2 expression (i.e., elevated CCNA2 predicts hypersensitivity) with two structurally distinct PLK1 inhibitors, but not with other drugs, were suggestive a biologically meaningful relationship. Therefore, we decided to further investigate whether CCNA2 may be a functional of differential responses to PLK1 inhibitors in gastric cancer.

CCNA2 Upregulation is Causally Linked to BI-2536 Induced Cytotoxicity in Gastric Cancer Cells
First, we sought to validate differential expression of cyclin A2 protein in gastric cancer cell lines selected from both sides of the drug response profile for PLK1 inhibitors. Compared to resistant cells, gastric cancer cells sensitive to PLK1 inhibitors showed increased expression of cyclin A2 (Figure 2a). MKN28 (sensitive) and SNU719 (resistant) cells were further evaluated in regards to multi-point dose responses to BI-2536. As expected, MKN28 cells exhibited greater sensitivity to BI-2536 than SNU719 cells (Figure 2b). In MKN28 and other sensitive cancer cell lines (AGS and SNU601), but not in SNU719 cells, BI-2536 elicited PARP1 cleavage, JNK phosphorylation and caspase-3 cleavage, all of which are indicative of apoptosis induction (Figures 2c and S2a). To determine if elevated CCNA2 is required to confer sensitivity to BI-2536 in gastric cancer cell lines, CCNA2 was transiently overexpressed in SNU719 cells and knocked down in MKN28 cells (Figure 2d). MKN28 cells transfected with CCNA2 siRNAs gained resistance to BI-2536, compared to cells transfected with negative control siRNA (siNC) (Figure 2e). Meanwhile, however, SNU719 cells transfected with CCNA2 cDNA exhibited greater sensitivity to BI-2536 than control cells (Figure 2f). To confirm that BI-2536 sensitivity indeed acts in relation to CCNA2 expression, we stably knocked down CCNA2 in MKN28 cells using viral transduction of shRNAs and tested the resultant cell viability. Therein, MKN28 isogenic cells, in which CCNA2 was knocked down by five shRNA clones, showed decreased cyclin A2 expression ( Figure 2g) and increased viability against BI-2536 ( Figure 2h). Similarly, we also observed the shCCNA2-mediated reversal of cytotoxicity to BI-2536 in AGS and SNU601 ( Figure S2b,c). Taken together, we deemed that elevated CCNA2 expression is required to confer sensitivity to PLK1inhibitors in gastric cancer cell lines.

CCNA2 is Required for BI-2536-Induced Mitotic Catastrophe and Apoptosis
Cyclin A2 regulates cell cycle progression by promoting S phase entry upon forming a complex with CDK2, as well as by facilitating mitosis through cooperation with the cyclin B1-CDK1 complex [15]. In contrast, PLK1 primarily functions in the M phase of the cell cycle [16][17][18][19] and inhibition of PLK1 causes cell cycle arrest at the G2/M-phase, followed by mitotic catastrophe, a type of apoptosis that occurs during mitosis, in cells with higher mitotic index [20,21]. Therefore, we hypothesized that aberrant upregulation of cyclin A2 in gastric cancer cells may elicit synthetic lethal vulnerability to PLK1 inhibition through failed cell cycle progression, particularly at the M phase. To test this, we performed immunocytochemistry using phospho-histone H3 antibody to detect mitotic cells and found that cyclin A2-knockdown cells show lower mitotic index values than control cells ( Figure 3a). To investigate whether elevated cyclin A2 induces sensitivity to BI-2536 due to impaired mitotic progression, we assessed changes in cell numbers in each phase of the cell cycle at 24, 48 and 72 h post treatment with BI-2536 and with or without CCNA2 knockdown in MKN28 cells. Control cells showed marked cell cycle arrest at the G2/M-phase at 24 h posttreatment with BI-2536; however, shCCNA2 cells slipped over from the G2/M-arrest and showed accumulation of polyploidy at 48 h and 72 h post BI-2536-treatment in a dose-dependent manner (Figure 3b), indicating that cancer cells characterized by high expression of cyclin A2 undergo less mitotic slippage and more apoptosis in response to BI-2536 treatment than cyclin A2 knockdown cells. We confirmed this through flow cytometricanalysis and subsequent immunoblot analysis of control MKN28 cells, which showed early apoptotic cell populations within 24 h of BI-2536 treatment ( Figure S3) and accumulation of apoptotic marker proteins (e.g., 89-kDa cleaved PARP1, JNK phosphorylation and cleaved caspase-3) upon exposure to BI-2536 ( Figure 3c). shRNA-mediated knockdown of cyclin A2 significantly reduced apoptotic marker proteins (Figure 3c), cell-fractions under mitotic catastrophe ( Figure 3d) and dead cell populations (Figure 3e) induced by BI-2536.These observations indicated that PLK1 inhibition in the context of elevated CCNA2 leads to mitotic catastrophe and apoptosis rather than to cell survival through mitotic slippage. Meanwhile, research has indicated a potential direct regulatory mechanism for cyclin A2 on PLK1 activation and phosphorylation [22]. To investigate if cyclin A2 functions through PLK1 activity, we developedphosphomimetic mutant PLK1 (T210D) and non-phosphorylatable mutant PLK1 (T210A) proteins ( Figure S4a,b) and observed their effects on BI-2536 sensitivity. Interestingly, overexpression of neither of these mutant PLK1 proteins nor wild-type PLK1 altered sensitivity to BI-2536 ( Figure S4c,d), indicating that CCNA2-induced BI-2536 sensitivity is independent of PLK1 and phospho-PLK1 levels.

CCNA2 is Required for BI-2536-Induced Mitotic Catastrophe and Apoptosis
Cyclin A2 regulates cell cycle progression by promoting S phase entry upon forming a complex with CDK2, as well as by facilitating mitosis through cooperation with the cyclin B1-CDK1 complex

KRAS Driven Upregulation of CCNA2 Confers Sensitivity to PLK1 Inhibitors Among KRAS Mutant Cancers
While it was reported that KRAS mutant cancer cells are highly sensitive to PLK1 inhibitors [12], detailed mechanisms underlying the sensitivity are largely unknown. Here, we hypothesized that oncogenic KRAS would drive aberrant upregulation of CCNA2. To test this, we compared CCNA2 expression levels between wild-type and mutant KRAS or pan-RAS (KRAS, HRAS and NRAS) tumor samples in The Cancer Genome Atlas (TCGA) cohort. CCNA2 expression was significantly higher in pan-RAS and KRAS mutant tumors than wild-type controls (p =1.01 × 10 −31 and 4.27 × 10 −20 by Wilcoxon test, respectively). KRAS mutant gastric tumor samples (TCGA-STAD) also showed significantly higher expression of CCNA2 (p = 4.76 × 10 −4 by Wilcoxon test) than wild-type gastric tumors (Figure 4a). The 37 gastric cancer cell lines, which includedeight KRAS mutant cell lines (AGS, SNU601, SK4, SNU1, NCC24, SNU668, YCC2 and NCC59) in this study also showed similar results in that KRAS mutant cell lines had higher sensitivity to BI-2536 and volasertib (Figure 4b). To test whether KRAS affects CCNA2 expression or vice versa, KRAS mutant gastric cancer cell lines were transfected with siRNAs targeting KRAS or CCNA2. Therein, the KRAS mutant gastric cancer cell lines (AGS, SK4 and SNU601) showed decreased CCNA2 expression after depletion of mutant KRAS, whereas CCNA2 knockdown did not affect expression of KRAS (Figure 4c). In addition, while depletion of mutant KRAS reversed cytotoxicity to BI-2536 and volasertib (Figure 4d), co-expression of CCNA2 in this context was sufficient to reintroduce sensitivity (Figure 4e), indicating that sensitivity to PLK1 inhibitors in KRAS mutant gastric cancer cells is mediated by CCNA2 upregulation.
Taken together, these data demonstrated that oncogenic KRAS-driven CCNA2 upregulation confers hypersensitivity to PLK1 inhibition through mitotic catastrophe and apoptosis (Figure 4f).

Discussion
The cell cycle is tightly regulated by cyclins, cyclin-dependent kinases (CDKs) and various checkpoint kinases [23]. Deregulation of the cell cycle is a hallmark of tumorigenesis [24]. Given its importance in tumorigenesis, several cell cycle inhibitors have emerged as potential therapeutic drugs for the treatment of cancers, both as single and combination therapies with traditional cytotoxic or molecular targeting agents. Currently, the most promising cell cycle inhibitors in anticancer therapeutics are orally bioavailable CDK4/6 inhibitors, which have received regulatory approval in combination with hormonal therapy for treatment of patients with metastatic hormone receptor (HR)-positive, Her2-negative breast cancer [25][26][27][28][29]. Many other compounds designed to interrupt cell cycle progression or checkpoint control have problems in demonstrating sufficient antitumor efficacy. Thus, to facilitate clinical development of this target class, proper patient selection biomarkers are urgently needed.
PLK1 is a serine/threonine kinase that regulates various cellular events during cell cycle progression, including centrosome maturation, DNA checkpoint activation, mitotic entry, spindle assembly and cytokinesis [3]. Because hyper-activation of PLK1 causes overriding checkpoints, which leads to immature cell division, several PLK1 kinase inhibitors have been developed as anticancer drugs and are currently being evaluated in clinical trials [6]. Several challenges limiting the success of these drugs are as follows: First, most PLK1 inhibitors exhibit a narrow therapeutic window, with their therapeutic effects often coupled with normal toxicity. Second, responses to PLK1 inhibitors are inconsistent. Last is the emergence of drug resistance [6]. Discovery of patient selection biomarkers can help overcome this challenge by identifying correct tumor types and by facilitating patient stratification to increase response rates to these drugs. Indeed, some studies have reported that TP53 mutation is a biomarker of sensitivity to PLK1 inhibitors [30,31]. This suggests a compensatory mechanism mediated by p53 that rescues cancer cells from mitotic arrest and subsequent apoptosis caused by PLK1 inhibition. Another potential biomarker predicting sensitivity to PLK1 inhibitors is the oncogenic KRAS mutation. Research has been shown that cancer cells with KRAS mutation are more sensitive to PLK1 inhibitors than KRAS wild-type cancers [32], suggesting that KRAS mutation induces mitotic stress in tumor cells and may underlie tumor sensitivity to anti-mitotic agents.
Cyclin A2 belongs to the highly conserved cyclin family and is expressed in almost all tissues in the human body. It plays critical roles in control of the cell cycle at G1/S and in G2/M transition. Data from the Human Protein Atlas show that CCNA2 is overexpressed in dozens of cancer types, suggesting a potential role in tumorigenesis. In this study, we demonstrated that, compared to cells with basal CCNA2 expression, cancer cells highly expressing CCNA2 are more addicted to PLK1 activity and show increased mitotic index values, leading to G2/M arrest and mitotic catastrophe, followed by apoptosis, in response to PLK1 inhibitors. We observed neither PLK1 nor phospho-PLK1 affects sensitivity to PLK1 inhibitor, suggesting that the increased sensitivity of CCNA2-elevated cancer cells to PLK1 inhibition is not due to direct perturbation of the cyclin A2-PLK1 axis, but rather, likely due to a different mechanism that needs to be further elucidated. One possibility may be that increased cyclin A2 facilitates G2/M transition by activation of the cyclin B1/CDK1 complex [33], resulting in increased mitotic cell populations that more greatly rely on proper spindle assembly checkpoint, wherein PLK1 kinase activity plays an essential role [34].
We also discovered a causal relationship between oncogenic KRAS mutation and CCNA2 upregulation. Our data suggest that aberrations in CCNA2 expression are a consequence of oncogenic KRAS mutation potentially contributing to cell cycle progression, while at the same time, conferring dependence on PLK1 function for productive mitotic progression. Although it is beyond the scope of this manuscript, several mechanistic hypotheses may explain the connections between KRAS mutation, cyclin A2 elevation and PLK1 inhibitor sensitivity: Potentially, upon glutamine (Gln) deprivation, KRAS-driven cancer cells bypass a late G1 Gln-dependent cell cycle checkpoint and enter S-phase, followed by cell cycle arrest due to insufficient nucleotide biosynthesis [35]. Meanwhile, cyclin A2 forms a complex with CDK2 at the S phase of the cell cycle to initiate and progress DNA synthesis. Thus, elevated cyclin A2 in KRAS mutant cancers may reflect anadaptation mechanism from this cell cycle stress at the S phase. Alternatively, as cyclin A2 directly phosphorylates and activates protein kinase B, also known as Akt [36,37], elevated cyclin A2 may contribute to Akt-driven tumorigenesis as well. Either way, as a consequence of CCNA2 upregulation, cancer cells mayface an unavoidable dependence on PLK1 to prevent mitotic catastrophe by disrupted spindle assembly. Thus, we considerthis relationship as a "synthetic lethality."Accordingly, synthetic lethal approaches targeting cell cycle progression with PLK1 inhibitors may prove to be effective in treating tumors characterized by increased CCNA2 expression.

Pharmacological Characterization
This study analyzed data for 37 gastric cancer cell lines treated with 75 small-molecule compounds selected from libraries of FDA-approved small-molecule pharmacological compounds (#L1300, Selleckchem, Houston, TX, USA) and investigational anti-cancer compounds (#L2000, Selleckchem). The pharmacological profiles of 29 of the 37 cell lines have previously been reported [14].

RNA Sequencing
RNA sequencing (RNA-seq) data for 29 of the 37 gastric cancer cell lines were previously reported [14]. Total RNA from the eight remaining gastric cancer cell lines were extracted with a RNeasy Plus Mini Kit (Qiagen, Hilden, Germany). RNA-seq libraries were then generated with a TruSeq RNA Sample Prep kit v2 (Illumina, San Diego, CA, USA) and sequencing with the HiSeq 2500 platform. The TopHat-Cufflinks pipeline was used to align reads to the reference genome and to calculate normalized values in FPKM (fragments per kilobase of exon per million fragments mapped).

Pharmacogenomic Analysis
We established elastic net models using a previously described method with some modifications [38]. Briefly, transcriptome and drug response data (n = 75) for the 37 gastric cancer cell lines were used to build the model. To do so, gene expression values were converted into Z-scores. Optimal values of α and λ were determined by 10-fold cross validation from 100 iterations. Bootstrapping (200×) was applied to estimate average weights (β) and selection frequency of features by the model. For drug marker selection, we chose features occurring at a frequency > 75%. Next, we applied different weight cutoffs to individual drugs because their weight spectrumsvaried greatly, which made it difficult to apply a single weight cutoff. The feature selection process was conducted using the GlmnetR package (version 2.0-8) and R (version 3.3.3).

cDNA Transfection
The Myc-DDK-tagged human CCNA2 plasmid (#RC211148L1) was purchased from OriGene (Rockville, MD, USA). Two million cells were grown in 60-mm dishes for 24 h. On the day of transfection, 5 µg of plasmids and 15 µL of Lipofectamine 2000 (Invitrogen, Carlsbad, CA, USA) were diluted in 500 µL of Opti-MEM and added onto the culture dishes. Twenty-four hourspost transfection, cell lines transfected with either target plasmid or empty vector were trypsinized and re-plated into 96-well plates (8000 cells per well) for drug toxicity assay.

The Cancer Genome Atlas Analysis
Gene expression and mutation data for 16 cancer types were downloaded from the Genomic Data Commons data portal (https://portal.gdc.cancer.gov). Gene expression data in FPKM were transformed to log2 scale values and quantile normalization was performed to remove technical biases. Samples bearing any mutations in KRAS, HRAS or NRAS were considered as RAS mutation samples. A two-sided Wilcoxon rank-sum test was used to test the difference in CCNA2 expression levels between RNA-mutant and wild-type samples.

Flow CytometricAnalysis
For cell cycle analysis using flow cytometry (FACSVerse TM , BD Biosciences, Franklin Lakes, NJ), MKN28 cells were plated at a density of 1 × 10 3 cells in 60mm culture dishes and then treated with BI-2536 (200 and 500 nM) for 72 h. Cells were then harvested and fixed in ice-cold 70% ethanol overnight at −20 • C. Afterwards, cells were centrifuged at 300× g for 5 min, re-suspended in PBS containing 10 µg/mL of propidium iodide (PI) (P4170, Sigma-Aldrich, St. Louis, MO, USA), 100 µg/mL RNase A and 0.1% (v/v) Triton X-100 and incubated at 37 • C for 10 min. For cell apoptosis assay of MKN28 cells treated with BI-2536 (200 and 500 nM), we used the Annexin V-FITC Apoptosis Kit (#640914, BioLegend, San Diego, CA, USA) according to the manufacturer's protocol. Briefly, a total of 1 × 10 3 cells was seeded in 60mm culture dishes for 24 h and then treated with BI-2536 for 72 h. Adherent and floating cells were then harvested, stained with Annexin V and PI for 15 min at room temperature and subjected to flow cytometry. The data were analyzed with FlowJo software.

Immunofluorescence Analysis
Cells were washed with ice-cold PBS and fixed with 3.7% paraformaldehyde (PFA) for 10 min, permeabilized with 0.5% Triton X-100 in PBS (PBS-T) for 5 min and incubated with blocking solution (0.1% BSA + 10% goat serum in 0.1% PBS-T) for 30 min. Cells were then incubated with primary antibodies diluted in 0.1% PBS-T for an hour. After incubation with secondary antibodies labeled with Alexa Fluor-488 or Alexa Fluor-568 (Invitrogen), coverslips were mounted on slide glasses using Prolong Gold Antifade mounting solution (Thermo Fisher Scientific) and the slides were allowed to dry at room temperature.

Conclusions
We performed pharmacogenomics analysis using a gastric cancer cell line panel and discovered a causal linkage cascade of oncogenic KRAS mutation, aberrant CCNA2 upregulation and hypersensitivity to PLK1 inhibitors. Our findings hold translational implications for the treatment of gastric cancer patients with aberrant upregulation of CCNA2 via synthetic lethal approaches targeting cell cycle progression.
Supplementary Materials: The following are available online at http://www.mdpi.com/2072-6694/12/6/1418/s1, Table S1: List of 75 compounds and their cell line-specific responses, Figure S1: Identified biomarker and drug response relationships by the elastic net regularization method, Figure S2: BI-2536 sensitivity in AGS and SNU601 cells, Figure S3: Apoptosis analysis using Annexin V-FITC/PI dual staining and flow cytometric analysis after BI-2536 (200 nM) treatment for 24 h are presented in scatter plots, Figure S4: BI-2536 sensitivity with wild-type and mutant PLK1 levels.