Genetic polymorphism of SLC31A1 is associated with clinical outcomes of platinum-based chemotherapy in non-small-cell lung cancer patients through modulating microRNA-mediated regulation

SLC31A1 is the major transporter for platinum drug intake, its expression correlates with drug disposition and response. In 1004 Chinese NSCLC patients with platinum-based chemotherapy, we investigated the association between SLC31A1 polymorphisms and clinical outcomes. Heterozygotes of rs10759637 at 3′UTR was associated with severe thrombocytopenia (odds ratio [OR]: 2.69; P = 0.012) and shorter overall survival (hazard ratio [HR]: 1.24; P = 0.005). Variant homozygote of rs2233914 was correlated with longer overall survival (hazard ratio [HR]: 0.73; P = 0.008). Haplotype and diplotype of these linked SNPs were associated with hematologic toxicities. In stratification analyses, rs10759637 and rs2233914 consistently correlated with overall survival in specific subgroups such as men, smoker, patients older than 58 years, or with ECOG PS 0-1, or with squamous cell carcinoma. rs10759637 could change the local structure of 3′UTR harboring putative binding sites for hsa-miR-29, whose transfection into 16HBE cells resulted in remarkable suppression of gene expression. The rs10759637 variant significantly correlated with lowered luciferase activity in reporter assays and decreased expression of SLC31A1 transcript in tumorous tissues. The study thereby identified functional polymorphism of SLC31A1 that modulates miRNA-3′UTR interaction and gene expression as potential pharmacogenetic biomarker for clinical outcomes of platinum-based chemotherapy in NSCLC patients.


INTRODUCTION
Lung cancer is among the most common cancers in both man and woman, and is the leading cause of cancer-related death worldwide [1]. Non-small cell lung cancer (NSCLC), with two main histological types of adenocarcinoma and squamous-cell (epidermoid) carcinoma, accounts for about 80% of primary lung Research Paper www.oncotarget.com cancer, and a majority present with advanced stage (III/ IV) when they were first diagnosed. Only 10-15% newly diagnosed NSCLC patients were adopted with a potentially curative resection [2]. Despite the introduction of targeted therapy on specific mutant molecules such as EGFR, platinum agents combined with another cytotoxic compound are still the first-line chemotherapy for advanced NSCLC. The action mode of platinating agents is known to bind the DNA with intra-or interstrand crosslinks and the Pt-DNA adduct lead to DNA lesions, activation of multiple pathways and result in cell apoptosis ultimately [3]. Despite efforts to improve therapeutic efficacy, the platinum-based chemotherapy brings modest benefits but also adverse effects, with the five-year survival rates less than 15%. And an increasing evidence showed that the limited efficacy for advanced NSCLC is due to drug resistance and toxicological side effects such as thrombocytopenia, nausea/vomiting, ototoxicity, nephrotoxicity and peripheral neurotoxicity, etc [3].
Mechanistically, therapeutic efficacy and response to platinum-based chemotherapy could be linked to these biological pathways, altered cellular accumulation, cytosolic inactivation of platinum drugs, inactivation of DNA repair pathway, altered apoptosis or increased tolerance to DNA damage [3]. The aberrant regulation and dysfunction of candidate genes in these pathway, which are largely ascribed to their functional polymorphisms, may influence interindividual differences in clinical outcomes. Thus there is a growing need toward tailoring chemotherapy to identify these functional genetic polymorphisms as predictive pharmacogenetic markers for better efficacy and minimal toxicity. To this end, the influence of sequence variants in pharmacodynamics pathways, such as nucleotide excision repair pathway, on clinical outcome has been extensively investigated [4]. In contrast, the biological function and clinical relevance of genetic polymorphisms in pharmacokinetics pathway for platinum metabolism and disposition are largely elusive.
Human SLC31A1 (solute carrier family 31 member 1) gene, also known as CTR1 (copper transporter 1), encodes a high-affinity copper transporter in cell membrane that functions as a homotrimer to effect the uptake of dietary copper. Recently, an interesting possibility has clearly emerged that the copper transporter SLC31A1 also acts as the major plasma-membrane transporter for platinum drug intake, including cisplatin, oxaliplatin and carboplatin [5]. Deletion of Slc31a1 in yeast and murine cells results in reduced cisplatin accumulation and increased resistance [6]. On the contrary, transfection of small cell lung cancer cell lines with SLC31A1 gene correlates with enhanced uptake of carboplatin and oxaliplatin [7]. Consistently, Slc31a1 knockout in murine model has also been shown to completely eliminate cisplatin tumor response in vivo [8]. Furthermore, in clinical setting, NSCLC patients with undetectable SLC31A1 expression in their tumors had reduced platinum concentration and tumor response, and lower platinum concentration in clinical specimens correlates directly with reduced tumor response and shorter survival time [9,10]. Recently, in two pilot studies with relatively small sample size (two to three hundreds) in Chinese patients with NSCLC who received Pt-based therapy, Xu et al. documented that SLC31A1 polymorphisms are associated with Pt-resistance or toxicity and poor clinical outcomes [11,12]. But these results have not been validated in larger cohort of patients and functional implication of these associated variants are not clear.
In the present study, we assessed the association of tagging and potentially functional SNPs of SLC31A1 gene with toxicological phenotypes, objective response and survivals of 1004 Chinese NSCLC patients receiving platinum-based treatment. We also functionally characterized a common variant at 3ʹ untranslated region (3ʹUTR) that modulates the microRNA-3ʹUTR interaction and decreases gene expression, proposing a possible underlying mechanism for the genetic association of SLC31A1 with clinical outcomes.

Patient characteristics and clinical outcomes
A total of 1004 eligible NSCLC patients of Chinese population were recruited in the study to investigate the genetic association between SCL31A1 polymorphisms and clinical outcomes of platinum-based chemotherapy. The main characteristics and clinical outcomes of patients are summarized in Table 1. Ever smokers account for 57.5% of the patients. Adenocarcinoma was the most common histological type (62.9%). Severe gastrointestinal toxicity (nausea/vomiting) was observed in 8.3% of the evaluated patients (n = 964). Severe hematological toxicity was observed in 23.9% of the evaluated patients (n = 969), among which 29 (3.1%), 149 (15.2%), 115 (12.3%) and 34 (3.6%) patients suffered from grade 3 or 4 anemia, leukopenia, neutropenia, thrombocytopenia, respectively. Grade 3 or 4 overall toxicity was observed in 29.9% of the evaluated patients (n = 952). For survival analysis, by the time of final data collection (July 2012), the median follow-up time was 46.5 month, and death had occurred in 74.9% of enrolled patients. The median progressionfree survival (PFS) was 9.1 months, and the median overall survival (OS) was 19.3 months. The rates of clinical phenotypes such as toxicities in our study were quite comparable to those previously reported in large randomized clinical trials [13].

Association of SLC31A1 gene polymorphisms with toxicity and objective response
In the patient cohort that had been evaluated for the seven toxicological phenotypes after platinum-based chemotherapy, we compared the genotypic distributions for the eight candidate SNPs of SLC31A1 between groups with respective mild or severe toxicological outcomes (Supplementary Table 2). Interestingly, we observed that rs4979223 (P = 0.007), rs4978536 (P = 0.063, with marginal significance), rs2233914 (P = 0.039), rs10817464 (P = 0.006) and rs10759637 (P = 0.006) showed significantly divergent genotypic distribution between groups with mild or severe thrombocytopenia, some of which remained statistically significant after multiple test correction. Based on these five candidate SNPs, we then assessed the association between SLC31A1 polymorphism and toxicological outcomes with multivariate unconditional logistic regression analysis ( Table 2). As to the tightly linked rs4979223 and rs10759637, we found their heterozygous genotype (A/C), as compared to the grouped homozygous genotypes (A/A+C/C), were significantly associated with severe thrombocytopenia, with ORs (95% CI) being 2.60 (1. 20 Because these five candidate SNPs of SLC31A1 could be partitioned into one haplotype block (Figure 1), we also estimated their haplotype and diplotype frequencies in the cohort, and analyzed their association with toxicological outcomes (Table 3). We predicted only four common haplotypes for the five SNPs in the 1004 individuals, which is consistent with their strong linkage disequilibrium. Agreeing with the genotypebased association results as above, the Hap4_CGGGC, which is composed of rs4979223-variant-C, rs4978536variant-G, rs2233914-wild-G, rs10817464-variant-G and rs10759637-variant-C alleles, was significantly associated with severe toxicological outcomes such as leucopenia ( We also analyzed the association between SLC31A1 polymorphism and chemotherapy objective response. None of the eight candidate SNPs of SLC31A1 displayed statistically significant difference in genotypic distribution between the complete or partial response group and the stable or progressive disease group (Supplementary Table 3). These results do not support genetic correlation  between SLC31A1 and objective response of platinumbased chemotherapy in the patient cohort.

Association of SLC31A1 gene polymorphisms with survival
We measured genetic association of SLC31A1 polymorphism with overall survival (OS) and progression-free survival (PFS) in the patient cohort with platinumbased chemotherapy by using log-rank test and Cox proportional hazards regression model ( Table 5). As to the OS dataset, the log-rank test showed that rs4979223 (P = 0.006), rs2233914 (P = 0.016) and rs10759637 (P = 0.010) were significantly associated with survival. The median OS time of patients with heterozygous A/C of rs10759637 was significantly shorter than those patients with (C) Comparison of secondary structures, built with RNAfold program, of the local SLC31A1 3ʹUTRs harboring hsa-miR-29 binding sites and the wild or the variant alleles of rs10759637. (D) Luciferase reporter assays in 16HBE cells to analyze the regulatory role hsa-miR-29 on SLC31A1 3ʹUTR that could be modulated by rs10759637. Co-transfection with hsa-miR-29 member (a, b, or c) and SLC31A1 3ʹUTR construct consistently resulted in reduced luciferase activity. In the cases of hsa-miR-29 b or c, rs10759637 allelic state significantly affected the cis-regulation toward reporter expression. (E) Real-time quantitative RT-PCR analysis of SLC31A1 mRNA expression in lung tumorous tissues from an independent cohort of patients, which was significantly correlated with rs10759637 genotypes. homozygous A/A or C/C (17.7 vs 20.2, P = 0.004). Cox proportional hazards regression analysis further showed that rs10759637 A/C was associated with increased risk of disease progression as compared with the homozygous genotype group (HR 1.24; 95% CI 1.07−1.44). In alike manner, rs4979223 heterozygous genotype was also a risk factor for OS with HR of 1.25 (1.08−1.45) and log-rank P value of 0.002 when assuming under-dominant model. On the contrary, rs2233914 variant A/A homozygote, in recessive model, was a beneficial factor for OS with HR of 0.73 (0.58−0.92), the median OS time of patients A/A genotype was significantly longer than those with G/G or G/A genotypes (22.3 vs 18.8, log-rank P = 0.011). As to the PFS dataset, we did not observe any association between progression-free survival and the eight candidate SNPs of SLC31A1.

rs10759637 at SLC31A1 3ʹUTR affects hsa-miR-29 mediated expression regulation
In this pharmacogenetic study, clinical outcomes such as severe thrombocytopenia hematological toxicity and overall survival of NSCLC patients with platinumbased chemotherapy were associated with linked non-coding variants of SLC31A1 gene that codes the transporter for platinum agent intake. We hypothesized that the underlying mechanism for the genetic association might be linked to functional interplay between polymorphic cis-elements and putative trans-acting factors such as transcription factor and microRNA. In order to identify putative microRNA targeting sequences in SLC31A1 3ʹUTR and assess potential effect of rs10759637 on microRNA-mediated regulation, we used the miRdSNP database and TargetScanHuman 6.2 program to screen potential microRNA binding sites at SLC31A1 3ʹUTR. Interestingly, as shown in Figure 1B, we predicted two putative binding sites, 314 bp upstream and 60 bp downstream to rs10759637, respectively, for the hsa-miR-29 family that was reported to be expressed in lung tissue and implicated in lung development and carcinogenesis [14,15]. Further prediction of local secondary structure of SLC31A1 mRNA 3ʹUTR with RNAfold program revealed that rs10759637 could alter the local mRNA secondary structure in the binding sites of hsa-miR-29abc by changing the minimum free energy from -133.40 kcal/mmol at the wild A allele state to a Haplotypes were predicted with PHASE basing on rs4979223 (A/C), rs4978536 (A/G), rs2233914 (G/A), rs10817464 (A/G) and rs10759637(A/C) that were associated with toxicity outcomes as shown in Table 2. b Odds ratios (OR) and their 95% confidence intervals (CIs) were calculated with unconditional logistic regression analysis, with adjustment of gender, age, smoking status, ECOG performance status, TNM status, histological types, and treatment regimen. www.oncotarget.com  Figure  1C), suggesting that rs10759637 could modulate the regulatory function of hsa-miR-29 on SLC31A1 3ʹUTR. As to rs4979223 at 5ʹflanking region that was strictly associated with rs10759637, we did not observe putative binding sites containing it or other nearby linked SNPs in the 1000genome dataset for any transcription factor (data not shown). These results suggest that rs10759637 at SLC31A1 3ʹUTR might modulate the miRNA-mRNA interaction and affect miRNA-mediated expression regulation.
Using luciferase reporter gene assays in human bronchial epithelial cell line 16HBE, which, like lung epithelial cell, originates from primitive pluripotent pulmonary epithelial cells, we further investigated the functional modulation of rs10759637 on hsa-miR-29− SLC31A1 3ʹUTR interaction. In accordance with results of structural prediction as above, co-transfection with hsa-miR-29 member (a, b, or c) and SLC31A1 3ʹUTR construct (of rs10759637 A or C allele), as compared to the negative control microRNA mimic, consistently resulted in remarkably decreased luciferase activity, demonstrating that the hsa-miR-29 family is indeed a negative regulator for SLC31A1 expression by targeting its 3ʹUTR. Interestingly, in the setting of cotransfection with hsa-miR-29 b or c, rs10759637 A/C at 3ʹUTR significantly affected the cis-regulation toward reporter expression, the variant C allele correlated with much lowered luciferase activity as compared to the wild A allele ( Figure 1D, P < 0.05). Finally, in lung tumorous tissues from an independent cohort of patients, we analyzed the correlation between rs10759637 genotypes and transcriptional expression level for SLC31A1 using real-time quantitative RT-PCR. Consistent with the results of structure prediction and reporter assays, we found that, as compared with the wild A/A genotype, the heterozygous A/C (P = 0.018)   performance status (E), and histological type (F), the curves were plotted with SPSS software according to genotypes of two SNPs. For rs2233914, the A/A homozygote was compared to the G/G+G/A group as reference. For rs10759637, the A/C heterozygote was compared to the A/A+C/C homozygotes group as reference. www.oncotarget.com and the A/C+C/C group (P = 0.014) were significantly correlated with lowered expression of SLC31A1 transcript in lung cancer tissues, suggesting that the modulation of rs10759637 on hsa-miR-29−SLC31A1 3ʹUTR interaction observed in cell line has relevance in clinical samples. Taken together, these functional studies revealed that rs10759637 in SLC31A1 3ʹUTR could modulate the microRNA-3ʹUTR interaction and thereby decrease gene expression, thus proposing a possible underlying mechanism for the genetic association between SLC31A1 polymorphism and clinical outcomes of platinum-based chemotherapy in NSCLC patients.

DISCUSSION
In this multi-institutions based pharmacogenetic study of NSCLC patients with platinum-based chemotherapy, we found linked non-coding variants of SLC31A1 gene, which encodes the major platinum uptake transporter, are associated with severe hematological toxicities such as thrombocytopenia and shorter overall survival. Further mechanistic analysis revealed that one SNP in 3ʹUTR correlated with downregulated gene expression through modulating the microRNA−3ʹUTR interaction. The clinically relevant and biologically functional SLC31A1 polymorphism provides potential biomarker for outcome prediction of platinum-based chemotherapy and lung cancer management.
As a well-defined transporter for platinum import, the expression and function of SLC31A1 are involved in platinum disposition and clinical response to platinumbased chemotherapy. Chen et al reported that in a cohort of stage Ⅲ NSCLC patients in Chinese population receiving first-line platinum-based doublet chemotherapy, SLC31A1 protein overexpression is associated with better therapeutic response and better survival [16]. In a cohort of American patients who underwent neoadjuvant platinum-based chemotherapy, Kim et al also observed correlation between SLC31A1 expression in tumor size and tissue platinum concentration, undetectable protein expression and reduced drug load are associated with reduced tumor response and shorter survival time [9,10]. Consistently, low levels of SLC31A1 mRNA are also associated with poor clinical response to platinum-based therapy in ovarian cancer patients [17]. These clinical studies demonstrate that the expression status in situ of SLC31A1 is a predictor for platinum-based therapy response and a significant prognostic factor for advanced NSCLC patients. However, the resectable cases only account for 10-15% of diagnosed NSCLC patients, and routine measure of SLC31A1 mRNA or protein expression in lung cancer tissues is not convenient. Thus, it is of very limited clinical utility for SLC31A1 expression as a putative biomarker of response and outcome for platinum-based chemotherapy in NSCLC patients. We here present evidences that rs10759637 in SLC31A1 3ʹUTR functionally decreased microRNA-mediated gene expression, and this common non-coding variants as well as linked SNPs were associated with severe hematological toxicities such as thrombocytopenia and shorter overall survival in the Chinese cohort of NSCLC patients with platinum-based chemotherapy. Notably, the genetic associations with survival were consistently pronounced in male patients, patients ever-smoking, older than 58, or with squamous cell carcinoma. Therefore, relevant variants of SLC31A1 will be pertinent pharmacogenetic biomarkers, which is much more actionable than gene expression, for platinum-based therapy in NSCLC patients especially in these subgroups.
The association of rs10759637 in SLC31A1 3ʹUTR with clinical outcomes in NSCLC patients with platinumbased therapy is biologically plausible. In a screening in silico of potential interaction between microRNA and SNPs in 3ʹUTR of xenobiotic metabolism enzyme and transporter genes, Wei et al reported five putative miRNAs associated with rs10759637 of SLC31A1 in liver [18]. However, we could not experimentally validate these results in reporter gene assays in 16HBE cells (data not shown). Interestingly, we predicted two putative binding sites near rs10759637 for hsa-miR-29 members, and the structure of local SLC31A1 3ʹUTR harboring these sites could be affected by rs10759637. Furthermore, the C allele, which was associated with poor outcomes of platinum-based therapy in NSCLC patients, correlated with reductions in luciferase activity in reporter assay and in SLC31A1 transcript expression in clinical lung cancer tissues as well. Similarly, it was reported that one functional miRSNP, SNPs located at or near a microRNA binding site in 3ʹUTR of the target gene or in a microRNA, near the miR-24 binding site in the 3'UTR of dihydrofolate reductase gene (DHFR) interferes with miR-24 function, resulting in DHFR overexpression and methotrexate resistance [19]. A recent study also reported that the miRSNP rs1062980 may alter the expression of iron-responsive element binding protein 2 gene (IREB2) potentially through modulating the binding of miR-29a, and thereby modify risk of lung cancer [20]. The miR-29 family is expressed in lung tissues, and plays important roles in the pathogenesis of lung diseases by directly targeting genes such as those encoding extracellular matrix proteins and those associated with tissue invasion and metastasis in the contexts of fibrosis and adenocarcinoma, respectively [21,22]. The regulation mechanism for SLC31A1 expression is largely unknown, two transcription factors, Sp1 and hypoxia-induced factor 2α (HIF2α), have been shown to mediate its basal and inducible expression [23,24]. We noticed that rs4979223 and rs2233914 at the at 5ʹflanking region of SLC31A1 are not located at or near the cis-elements for Sp1 or HIF2α, their associations with toxicological phenotypes and survival could be largely due to their strong linkage disequilibrium with the functional rs10759637 at 3ʹUTR and genetic hitchhiking effect. www.oncotarget.com Therefore, the modulation of rs10759637 as miRSNP on miR-29−SLC31A1 3ʹUTR interaction not also proposes underlying mechanism for the association of SLC31A1 polymorphism with clinical outcomes of platinum-based therapy, but also provides first evidence to link SLC31A1 gene to the miR-29 regulatory network.
Another notable aspect of this study is that genotypes of rs10759637 and linked variants, as well as their combined haplotype and diplotype, were consistently associated with severe hematological toxicities especially thrombocytopenia. Thrombocytopenia is one of major common hematologic complications of cancer chemotherapy and is associated with increased morbidity, mortality and health care costs [25]. In a large cohort of solid patients (total n = 47159; NSCLC n = 7001) undergoing chemotherapy including platinum-based regimens (accounted for about 1/3 and were most common for NSCLC), NSCLC patients had the highest prevalence of severe thrombocytopenia (10.7%) cross the four major chemotherapy regimens, suggesting that the burden of thrombocytopenia remains high [26]. Although the precise mechanism for chemotherapy-induced thrombocytopenia is not well established, many cytotoxic agents including platinum are now known to cause thrombocytopenia by predominantly inducing apoptosis of the progenitors of megakaryocytes, the platelet precursor, at early stages of differentiation [27,28]. As genetic evidence supporting this toxicological mechanism, a recent whole-exome sequencing study reported that severe thrombocytopenia in NSCLC patients treated with gemcitabine/carboplatin are associated with variant of DDX53 gene, a modulator of p53 mediated apoptosis [29]. Considering that platinum agents are intravenously administered, their biotransformation and dynamic distribution in the bloodstream are likely to be directly involved inducing hematologic toxicity. Because SLC31A1 is ubiquitously expressed in all tissues of vertebrates [30], part of administered platinum agents could be imported into all tissue cells including lung cancer cells as target, the other fraction remains bound with plasma protein in circulation [31]. In this scenario, we could cautiously postulate that, through interfering miRNA mediated expression, the naturally occurring germline SNP rs10759637 in SLC31A1 3ʹUTR might render reduced expression of transporter to all tissue cells, thus result in lowered platinum uptake in overall mass tissues, and accumulated platinum agents in bone marrow and peripheral blood plasma, and consequently be associated with platinum resistance of target tissue cells and hematologic toxicity such as thrombocytopenia.
We should cautiously point out that, in the study population of Chinese NSCLC patient cohort, we observed slight deviation from HWE for the genotypic distribution of two clinically relevant SNPs of SLC31A1, rs4979223 (P = 0.026) in 5ʹflanking region and rs10759637 (P = 0.011) in 3ʹUTR, with about 43 kb distance but nearly complete linkage disequilibrium, which was due to significant underrepresentation of their putatively disadvantageous heterozygotes in patients that were pharmacogenetically associated with poor outcomes such as severe thrombocytopenia and shorter survival. These HWE tests could be statistically well powered by the large sample of the patient cohort and the common allelic frequencies of the two SNPs [32]. However, in the SNP datasets for natural populations, we found that genotypic distribution of rs10759637 complies with HWE both in the overall global population (the 1000 genome dataset) and in Chinese Han population (the HapMap CHB dataset) (data not shown). The patient ascertainment was well defined, and the two common SNPs were of independent genotyping calling but are in nearly complete linkage disequilibrium. It is therefore at least possible that the HWE departure was due to ascertainment bias or genotyping errors, but instead, possible explanations could be sampling bias, underlying genetic model at the susceptibility locus, and other unknown factors in view of population genetics and demography [33]. Consistent with this scenario, rs10759637 manifested unique association with clinical outcomes in the NSCLC cohort with heterozygote disadvantage, which is also referred to as under-dominance, a genetic model where heterozygote has a lower overall fitness than either homozygote [34]. Although being rare cases, heterozygote disadvantage was also reported to account for the deviation from HWE of a common variant (c.677C>T) of the 5,10-methylenetetrahydrofolate reductase gene (MTHFR) in subfertile patients and its association with embryo aneuploidy [35]. However, the biological mechanism by which the SLC31A1 rs10759637 A/C heterozygote could be disadvantage is unclear, especially considering that the variant C allele correlated with diminished gene expression. Furthermore, it has been reported of remarkable divergence in SLC31A1 expression in situ as well as related drug disposition and response among populations with different ethnic ancestry. In the cohort of NSCLC patients in American population who underwent neoadjuvant platinum-based chemotherapy, Kim et al. reported that the African American had significantly reduced SLC31A1 expression in tumor (P = 0.001), lowered tissue platinum concentration (P = 0.009) and decreased tumor shrinkage (P = 0.016) as compared to Caucasians [9]. We also noticed that, in public SNP database, the low-expression-related C allele of rs10759637 is the ancestral allele that is common in African population, the high-expression-related A allele is the derived allele that is dominant in Caucasian and Chinese Han population, suggesting of significant population differentiation at this locus. These observations not only support the function mode of rs10759637 on SLC31A1 expression, but also provide comparative and evolutionary pharmacogenetics implications for the association of SLC31A1 polymorphism with clinical outcomes of platinum-based chemotherapy in lung cancer.
In summary, this pharmacogenetic study on advanced NSCLC patients with platinum-based chemotherapy has identified linked variants of platinum import transporter SLC31A1 gene that were associated with severe hematological toxicities such as thrombocytopenia and shorter survival, one of which decreased gene expression through modulating microRNA-3ʹUTR interaction. Given that platinum-based chemotherapy is routinely used in clinical management of many types of cancers, and SLC31A1 is ubiquitously expressed in human tissues, validation of these findings in cohorts of lung cancer and other relevant types of cancer with larger sample size would be warranted in different ethnic populations.  [36][37][38][39]. No statistically significant difference was observed in the distribution of demographic features among the patients from the six hospitals (P gender = 0.698, P age = 0.321). In addition, previously prepared lung cancer tissues from 47 surgically resectable patients recruited at Shanghai Changzheng Hospital were used for gene expression analyses [40].

Patient recruitment and follow-up
The patients enrolled in the pharmacogenetic study were inoperable and received first-line, platinumbased chemotherapy (no prior surgery, radiotherapy, or concurrent chemoradiotherapy). The chemotherapeutic regimens were as follows: either cisplatin (75 mg/ m 2 ) or carboplatin (at an area under the curve 5), both administered on day 1 every 3 weeks, in combination with navelbine (25 mg/m 2 ) on days 1 and 8 every 3 weeks, or gemcitabine (1250 mg/m 2 ) on days 1 and 8 every 3 weeks, or paclitaxel (175 mg/m 2 ) on day 1 every 3 weeks, or docetaxel (75 mg/m 2 ) on day 1 every 3 weeks. A few patients received other platinum-based treatment (n = 49). All chemotherapeutic drugs were administered intravenously, and all treatments lasted for 2 to 6 cycles.
Clinical outcomes including toxicities, responses and survival were assessed in this study. The incidence of grade 3 or 4 chemotherapy toxicity was assessed for the patients from the end of the first two cycles of treatment according to the Common Terminology Criteria for Adverse Events version 3.0 (CTCAE v3.0), including overall toxicity, gastrointestinal toxicity (nausea and vomiting), and hematologic toxicity (leucopenia, neutropenia, anemia, and thrombocytopenia) [41]. No grade 5 toxicity (death) was observed. Responses to platinum-based chemotherapy were evaluated after the first two cycles of the course, which are classified into four categories namely complete response (CR), partial response (PR), stable disease (SD) and progressive disease (PD) according to the Response Evaluation Criteria in Solid Tumors (RECIST) guidelines version 1.0 [42]. Object response rate (ORR) is defined as the percentage of patients with CR or PR. Survival data were collected from several sources including follow-up calls, the Social Security Death Index, and clinical medical records of inpatient and outpatient. Progression-free survival (PFS) was calculated from the date of chemotherapy beginning to the date of disease progression or death (whichever occurred first) or the last progression-free follow-up. Overall survival (OS) was calculated from the date of chemotherapy beginning to the date of death. The research assistants who performed the genotyping assays were blinded to the clinical state of patients and the clinical investigators were blinded to the genotypic state of patients. The study protocol was approved by the Ethical Review Committees of Fudan University School of Life Sciences and the participating hospitals, and written informed consent was obtained from each subjects.

SNPs selection and genotyping
Eight polymorphisms were selected by an approach combining both tagging and potentially functional SNPs of SLC31A1 gene. The tagging SNPs were screened from the Han Chinese in Beijing (CHB) population dataset of the HapMap database (http://www.hapmap. org) using a minor allele frequency (MAF) cutoff of 0.05 and a correlation coefficient (r 2 ) threshold of 0.8. Genomic DNA was extracted from whole blood using the QIAamp DNA Maxi Kit (Qiagen GmbH, Hilden, Germany). Genotyping was performed using iSelect HD BeadChip (Illumina, San Diego, CA, USA) with the following quality-control criteria: genotyping call rate of SNP >0.95, GenCall score >0.2, and P value of Hardy-Weinberg equilibrium (HWE) >0.01. Genotyping of one SNP (rs10759637) at the 3ʹUTR for lung cancer tissues of the 47 surgically resectable patients was conducted using PCR-direct sequencing with forward primer 5ʹ-GGAGAGCAAGGAATGTGGACT-3ʹ and the reverse primer 5ʹ-GCAGCTCATGCTGAGATTTCTA-3ʹ. www.oncotarget.com In silico prediction of microRNA binding sites at the 3ʹUTR of SLC31A1 In order to identify putative microRNA targeting sequences in SLC31A1 3ʹUTR, and assess potential effect of rs10759637 on microRNA-mediated regulation, we used the miRdSNP database (http://mirdsnp.ccr.buffalo. edu/index.php) and TargetScanHuman 6.2 program (http:// www.targetscan.org/) to screen microRNA binding sites, and used RNAfold (http://rna.tbi.univie.ac.at/) to predict the local secondary structure of the SLC31A1 mRNA.

Real-time quantitative RT-PCR
We quantified transcript level of SLC31A1 in lung cancer tissues from 47 patients using real-time quantitative RT-PCR assay. Total RNA was extracted from tumorous lung tissues with RNeasy Plus Mini Kit (QIAGEN, Hilden, Germany) and was then reverse transcribed into cDNA with PrimeScript RT Master Mix (TaKaRa, Dalian, China) using primers 5ʹ-GAGAGAGCCTGCTGCGTAAG-3ʹ (forward) and 5ʹ-AATGCAGAGGTACCCGTTGT-3ʹ (reverse). qRT-PCR was performed with SYBR ® Premix Ex Taq Kit (Takara) on an Applied Biosystems 7900HT. We measured the expression of β-actin as internal control using primers 5ʹ-CAGAGCCTCGCCTTTGCC-3ʹ (forward) and 5ʹ-ATGCCGGAGCCGTTGTCG-3ʹ (reserve). Every sample was set in triplicate. Expression of SLC31A1 was normalized to β-actin with ∆∆C T method.

Statistical analysis
We used Pearson χ 2 tests to examine deviations of genotype frequencies from those expected under HWE, and to test differences in the distributions of genotypes between patient groups with divergent clinical phenotypes or outcomes. We used the PHASE version 2.1 program to estimate haplotypes from the genotype data [43], and constructed the linkage disequilibrium (LD) plots using Haploview (http://www.broadinstitute.org/haploview), in which pairwise linkage disequilibrium relations among SNPs were examined using D′ and r 2 . We measured the association between genetic variants and dichotomous clinical phenotypes or outcomes by calculating odds ratios (OR) and their 95% confidence intervals (CIs) in unconditional logistic regression analysis, with adjustment of gender, age, smoking status, ECOG performance status, TNM status, histological types, and treatment regimen. Haplotype-based genetic association analysis was performed using Haplo. stats package in R-plus (available at http://cran.r-project. org/web/packages/haplo.stats.index.html; Version: 1.6.8). For each SNP, three different genetic models (dominant, recessive and additive) were analyzed, and the model with lowest P values was considered the best-fitting model. Under-dominant model was also analyzed for rs4979223 and rs10759637. The association between genetic variant and survival data (overall survival and progression-freesurvival) was calculated by log-rank test with adjustment for covariates. The multivariate Cox proportional hazards regression was used for calculating the hazard ratios (HR) and 95% CI, Kaplan-Meier method was used to plot survival curve. All the statistical analysis was performed by SPSS (version 20). We use the two-side test for all P values. A P values < 0.05 was considered statistically significant. To account for the issue of multiple testing of SNPs, we used SNPSpD to correct the significance threshold taking into account LD between polymorphisms [44].