GSTP1 rs1695 is associated with both hematological toxicity and prognosis of ovarian cancer treated with paclitaxel plus carboplatin combination chemotherapy: a comprehensive analysis using targeted resequencing of 100 pharmacogenes

Purpose To find genetic variants that predicted toxicity and/or efficacy of paclitaxel plus carboplatin combination therapy (TC therapy). Patients and methods In a retrospective case-control study, we analyzed 320 patients who had received TC therapy for gynecological cancers (ovarian, fallopian tube, peritoneal, uterine, and cervical cancers) and collected their germline DNA. We performed a comprehensive pharmacogenomic analysis using a targeted resequencing panel of 100 pharmacogenes. For 1,013 variants passing QC, case-control association studies and survival analyses were conducted. Results GSTP1 rs1695 showed the smallest p value for hematotoxicity association, and the 105Ile wild type allele had a significantly higher risk of severe hematotoxicity (neutropenia G4, thrombocytopenia ≥ G3 and anemia ≥ G3) than the 105Val allele (p=0.00034, odds ratio=5.71 (95% confidence interval:1.77-18.44)). Next, we assessed 5-year progression-free survival (PFS) and overall survival (OS) in 56 advanced ovarian cancer patients who received tri-weekly TC as a first-line chemotherapy. Patients with the 105Ile/105Ile genotype showed significantly better PFS (p=0.00070) and OS (p=0.0012) than those with the 105Ile/105Val or 105Val/105Val genotype. Conclusion Our study indicates that the GSTP1 rs1695 105Ile/105Ile genotype is associated with both severe hematotoxicity and high efficacy of TC therapy, identifying a possible prognostic indicator for patients with TC therapy.


INTRODUCTION
Paclitaxel plus carboplatin combination chemotherapy (TC therapy) is an important treatment for gynecological malignancies, including ovarian, peritoneal, fallopian tube, uterine, and cervical cancers. In particular, for ovarian cancer, TC therapy is currently the standard first-line chemotherapy or platinum-sensitive recurrent chemotherapy [1,2], and carboplatin is considered a key drug in its activity. More recently, a modified administration schedule of TC therapy has been developed, which includes administration of paclitaxel every week (dose-dense TC therapy) [3]. However, the prognosis of patients with ovarian cancer is still poor, especially in the advanced stage [4].
In our current study, we performed a comprehensive pharmacogenomic analysis by targeted resequencing of 100 pharmacogenes, using a bench top next-generation sequencer. The aim of our study is to identify genetic variants that are associated with the toxicity and efficacy of TC therapy, so that we are able to predict the likelihood of severe toxicity or efficacy in patients before treatment.

Patient characteristics
The characteristics of the 320 patients who received TC therapy for gynecological cancers are shown in Table  1. Fifty patients were classified as ADR group and 270 patients were controls. The median age was 54.5 years (range:  in the ADR group and 54 years (range:  in the control group. Regarding possible confounding factors, including age, tumor stage by FIGO classification, regimen, and total number of cycles in the observed TC treatment period, we found no significant differences between the ADR group and control groups (Table 1).

Targeted resequencing
By the targeted resequencing of 320 subjects, we obtained an output of 35.8 Gbp of sequence data. After a data mapping to human reference sequence (build hg19), on average, 98.5% of the targeted exonic regions (159,347 bp) were covered by at least 20× depth. An average read depth of the targeted regions was 289×. After a QC filtering based on depth < 20× and allele balance < 20%, 1,029 variants were identified. After filtering variants of minor allele frequencies of ≥ 0.001, missing genotype rates of < 10% and p-values less than the cutoff value of < 1.0 × 10 -6 by the Hardy-Weinberg equilibrium test, 1,013 variants consisting of 11 frameshift deletions,  12 frameshift insertions, 5 nonframeshift deletions, 4  nonframeshift insertions, 601 nonsynonymous single  nucleotide variants (SNVs), 23 stop-gain SNVs, 1 stoploss SNV and 356 synonymous SNVs were used for the further analysis.

Case-control association study of hematological toxicity
The Manhattan plot of the case-control association analyses is shown in Figure 1, and 20 nonsynonymous variants, which showed significant association with ADR grouping, are listed in Table 2. GSTP1 rs1695 (c.A342G, p.Ile105Val) showed the lowest p-value among them. The 105 Ile allele (wild type) had a significantly higher risk of severe hematotoxicity than the 105 Val allele did (p=0.00034, odds ratio=5.71 (95%CI: 1.77-18.44)).

Survival analysis
Next, we compared the 5-year PFS (progression-free survival) and OS (overall survival) of advanced ovarian cancer patients between the GSTP1 rs1695 genotypes. The characteristics of the 56 patients who received tri-weekly TC therapy as first-line therapy are shown in Table 3 Figure 1).

DISCUSSION
To date, many researchers have been searching for SNPs associated with susceptibility to taxane-or platinum www.oncotarget.com agent-related toxicities, but, unfortunately, most results have not been replicated [6][7][8][9][10][11][12][13][14][15][16][17][18][19][20][21][22][23][24]. Consequently, there have been no established genetic markers that predict the toxicity or efficacy of these drugs, regardless of their clinical importance in the treatment of various cancers. Our current study has some advantages compared with those of previous studies. First, we only included patients receiving paclitaxel plus carboplatin combination chemotherapy, which means that it is possible to eliminate consideration of effects of other taxanes or platinum agents, such as docetaxel, cisplatin and oxaliplatin. Second, we adopted a method of targeted resequencing of 100 notable pharmacokinetics-related genes using a nextgeneration sequencer, since candidate gene approaches might miss other significant associations. Third, we recruited 320 Japanese patients with detailed clinical data in our hospital. Some previous studies were performed in small sample size, less than one hundred patients, which might have insufficient statistical power to detect significant association.
The case-control association analysis indicated that 20 variants were significantly associated with severe hematological toxicity, with GSTP1 rs1695 showing the lowest p-value among them (p=0.00034). Glutathione S-transferase P1 (GSTP1) has been shown to metabolize carboplatin in vitro [25], and in our results, the 105 Ile allele was associated with a higher risk of severe hematotoxicity than the 105 Val allele was (odds ratio=5.71 (95%CI:   1.77-18.44)). We defined severe hematological toxicity as having decreased neutrophil count, platelet count, and hemoglobin concentration. This was based on our assumption that several types of adverse events should be observed simultaneously if they occurred due to the increased drug concentration that was caused by the pharmacogenetic variant. If so, this pharmacokinetic change could also have an influence on the antitumor effect. One of the strongest findings of our study is the association of rs1695 with treatment efficacy, as well as hematological toxicity. Glutathione S-transferases (GSTs) are a multigene family of enzymes that catalyze the conjugation of glutathione (GSH) to a variety of electrophilic xenobiotics, eventually forming a mercapturic acid to be excreted into the urine [26,27]. GSTP1 is a GST isoenzyme, which plays an important role in detoxifying carcinogens, metabolizing chemotherapeutic agents, and regulating the cell cycle and apoptosis [28,29]. The GSTP1 gene locus is known to be polymorphic, and three haplotypes have been identified so far, GSTP1 * A ( 105 Ile 114 Ala), GSTP1 * B Val (rs1138272, T allele) frequency in the Japanese population is very low, and no patient had this variant in our population, the effect of 114 Val could be ignored in the current study.
Many studies have shown that GSTP1 can detoxify cisplatin and carboplatin [25,[31][32][33], and some investigators have discussed the association between GSTP1 and cisplatin resistance [34][35][36]. The functional effect of rs1695 has also been well studied [25,27,30,32,37]. Interestingly, the effect of rs1695, with an amino acid substitution in the substrate-binding site, differed depending on the substrates involved [25]. Ishimoto et al. reported that the cytotoxicities induced by cisplatin and carboplatin toward GSTP1 105 Ile-expressing Escherichia coli were higher than those of the 105 Val mutant [25]. In another study, Peklak-Scott et al. found that the conjugation of cisplatin and GSH catalyzed by GSTP1  [32]. These previous reports showed that the platinum-detoxifying ability of GSTP1 105 Ile was lower than that of 105 Val, strongly supporting our model. However, the association of GSTP1 rs1695 genotypes with the clinical outcomes of platinum-based chemotherapy is still controversial. Some research groups reported the similar findings to ours [19,38,39], while others showed the opposite direction of effects of rs1695 [40,41]. Kim [19]. Although some differences existed, overall, our findings corroborated the results of these studies, thus supporting our model strongly. Moreover, our study provides important insights into the treatment of ovarian cancer with TC therapy because there has been, as far as we know, no report that evaluated association of genetic polymorphisms with both toxicity and efficacy of platinum-based anticancer reagents using clinical information of the same patients. We believe that the present findings shed new light on safer and more appropriate chemotherapy for ovarian cancer.
Our model is interesting because it also may explain the ethnic differences in TC therapy efficacy. In the JGOG3016 study [3], the median PFS and OS of tri-weekly TC therapy in Japanese ovarian cancer patients were 17.2 months and not reached, respectively. However, in the GOG-0218 study [42], the investigators reported the median PFS and OS as 10.3 and 39.3 months, respectively, in a population that was composed of non-Hispanic white (84.2%), Asian (6.6%), non-Hispanic black (4.0%), Hispanic (3.4%), and others (1.9%). The Japanese population seems to be more responsive to TC therapy. Interestingly, ethnic differences were observed in the allele frequency of rs1695 105 Val, with the Japanese population showing the lowest frequency (0.101), according to the 1000 Genomes Project (http://www.internationalgenome.org/). The higher frequency of wild-type ( 105 Ile/ 105 Ile) in Japanese may account for the ethnic differences above. To the best of our knowledge, our study is the first to report the association of TC therapy efficacy with GSTP1 rs1695 in Japanese patients.
Our study had some limitations. First, it was designed as a retrospective case-control study, and there are inherent biases with this type of study. Second, the sample size was small in the survival analysis because only advanced ovarian cancer patients receiving triweekly TC or dose-dense TC therapy as first-line chemotherapy had been included for accurate estimation of prognosis. In order to overcome the limitations, the best way is to conduct a replication study using an independent cohort for confirmation of the associations. Therefore, further prospective analyses, including generating pharmacokinetic data or performing replication studies with larger populations, are required in order to firmly establish our model.
In conclusion, we performed a comprehensive pharmacogenomic analysis, using a targeted resequencing panel of 100 pharmacogenes. Our results reveal that GSTP1 rs1695, known as a functional variant, is associated with both paclitaxel and carboplatin hematotoxicity, and the efficacy of ovarian cancer treatment. Several prior reports in vivo and in vitro were consistent with our results. We believe that this clinically interesting association will be verified and understood more precisely in the future.

Patients
Keio Women's Health Biobank (KWB, Tokyo, Japan) was started in 2006 to reposit frozen tissue, formalin-fixed paraffin-embedded (FFPE) tissue, germline DNA, serum, and ascites samples from gynecological patients. This biobank project was approved by the ethics committee of the School of Medicine, Keio University, and written informed consent was obtained from all patients before enrollment.
We found 320 adult female patients, with germline DNA samples available in the KWB, who received at least one cycle of TC therapy for ovarian, fallopian tube, peritoneal, uterine, or cervical cancer from January 1999 to August 2016. Patients who had uncontrolled serious complications were excluded. Clinical data about patient characteristics, ADR information, and data related to survival (i.e., date of recurrence, death, and last visit to the hospital) were collected from the medical records retrospectively. Disease had been confirmed by histologic examination, and the tumor stage had been assessed according to the www.oncotarget.com International Federation of Gynecology and Obstetrics (FIGO) classification.

Chemotherapy regimens
All patients had received either tri-weekly TC or dose-dense TC therapy, and those TC therapies were performed as either first-line or recurrent treatments. Patients who had received chemotherapy regimen containing taxanes or platinum agents other than paclitaxel or carboplatin (e.g., combination therapy with docetaxel or cisplatin) were excluded. Patients with monotherapy of paclitaxel or carboplatin were also excluded from this study. Tri-weekly TC therapy included administration of paclitaxel (175-180 mg/m 2 ) and carboplatin (AUC 6.0 mg·min/mL) on day 1, which was repeated every 3 weeks. Dose-dense TC therapy, on the contrary, included administration of paclitaxel (80 mg/m 2 ) on day 1, 8, and 15, and carboplatin (AUC 6.0 mg·min/mL) on day 1, which was repeated every 3 weeks. Appropriate dose reduction was performed for febrile neutropenia, G4 neutropenia that lasted ≥7 days, G4 thrombocytopenia, and non-hematological toxicities ≥G3. Granulocyte-colony stimulating factor (G-CSF) was applied at the discretion of the attending physician, mostly for the patients with febrile neutropenia or prolonged G4 neutropenia. Red blood cell transfusion was also considered if the patient showed anemia ≥G3 and clinical symptoms. Although the standard treatment period consisted of 6 cycles, it could be discontinued if the patient experienced severe adverse events after an appropriate dose-reduction or postponement, or it could be continued beyond 6 cycles if the attending physician deemed extension beneficial.

Adverse drug reaction evaluation
Complete blood counts had been routinely performed to assess hematological toxicity during TC therapy, and their grades were scored according to the Common Terminology Criteria for Adverse Events, version 4.0. The monitoring period for ADR was from the first administration day to 6 weeks after the final administration of either paclitaxel or carboplatin. The worst grade for neutropenia, thrombocytopenia, and anemia during the treatment period was recorded. We divided the patients based on the severity of hematotoxicity as follows: patients who developed neutropenia G4, thrombocytopenia ≥ G3, and anemia ≥ G3 in one treatment period were designated the "ADR group", and the other patients were designated the "Control group". Patients exposed to several TC therapies (e.g., first-line therapy as adjuvant chemotherapy and second-line therapy for the recurrence) were only required to meet the hematological toxicity criteria indicated above during one therapy regimen to be classified into the "ADR group."

DNA extraction
Peripheral blood samples had been obtained from patients and stored at -80°C. Genomic DNA was extracted from peripheral lymphocytes using a commercial kit (QIAamp DNA Blood Mini Kit, Qiagen, Hilden, Germany), following manufacturer's instructions, and stored at 4°C until analysis.

Survival analysis
We assessed 5-year PFS and OS as measures of treatment efficacy. For PFS and OS analysis, we focused on advanced ovarian cancer (stage III+IV) from the 320 patients who received tri-weekly (n=56) or dose-dense TC (n=55) therapy as first-line chemotherapy. We analyzed tri-weekly and dose-dense TC separately because previous reports had shown significantly better PFS and OS with dose-dense TC than with tri-weekly TC [3]. PFS was defined as the interval between the first day of TC therapy administration and the day of the first relapse, progression, or death. Relapse and progression were diagnosed by imaging or clinical symptoms. OS was defined as the interval between the first day of TC therapy administration and the day of death. www.oncotarget.com

Statistical methods
For the 1,013 variants passing QC, case-control association analyses comparing the ADR group with the control group were evaluated by Fisher's exact test using PLINK version 1.9 [63], considering allelic, dominant, and recessive genetic models. P-values <0.05 were considered to be statistically significant. A Manhattan plot was generated using the minimum p-value among the three genetic models for each variant. Five-year OS and PFS curves were plotted using the Kaplan-Meier method; differences between genotypes were analyzed using the log-rank test. Patients who survived or did not show progression during the observation period were censored at the last confirmation date. The data were analyzed using the freely available statistical software R version 3.3.2. To estimate the differences in patient characteristics between the ADR and control groups, we used a Student's t-test or Fisher's exact test.

Author contributions
Tomoko Yoshihama designed the study, carried out the experiments, performed data analysis and drafted the manuscript. Koya Fukunaga and Taisei Mushiroda contributed to analysis and interpretation of data and revised the manuscript. All other authors contributed to design the conception of the work, data collection and critically reviewed the manuscript. All authors approved the final version of the manuscript.