Genetic Variations in Multiple Drug Action Pathways and Survival in Advanced Stage Non–Small Cell Lung Cancer Treated with Chemotherapy

Non–small cell lung cancer (NSCLC) represents more than 80% of lung cancer diagnoses (1). For advanced stage (stage III and stage IV) NSCLC, 5-year survival varies widely (3%–50%) depending on the number of lymph nodes involved, resectability, and tumor histology (2). Established prognostic factors include stage, gender, age, and performance status (3). Chemotherapy remains the major component of the standard care in conjunction with radiation therapy and supportive care for patients with advanced stage lung cancer. For a given treatment, usually prescribed by a standard dosing protocol, survival time widely varies even after stratification by tumor stage, histology, and other clinical information, highlighting the need for improved predictive markers. In addition to clinical and biological behaviors of the tumor, multiple drug metabolism systems of the host also have an impact on the outcomes for lung cancer patients. Pharmacogenomic studies have defined gene variations responsible for varied drug efficacy. Patient-specific genetic profiles of genetic variants related to biological activity of drugs may help improve drug selection.

The glutathione metabolic pathway (GSH pathway) is directly involved in the detoxification or inactivation of multiple anticancer drugs, with evolving clinical relevance in lung cancer treatment (4). The efficacy of anticancer drugs is highly influenced by cellular DNA repair capacity; increasing evidence shows reduced DNA repair capacity resulting from genetic polymorphisms of various DNA repair genes being associated with improved survival after platinum-based chemotherapy (5–10). Taxanes (paclitaxel and docetaxel) are a class of anticancer agents, which exert their cytotoxic effects on spindle microtubules dynamics causing cell-cycle arrest and apoptosis (11, 12). These agents have been widely used as active chemotherapeutic agents in the treatment of NSCLC. Genes that regulate the cell cycle are thought to be involved in the biological activities of taxanes, although the impact of genetic variants in cell-cycle pathway genes on clinic outcomes is unclear. The epidermal growth factor receptor (EGFR) is a tyrosine kinase transmembrane receptor in the ErB family of receptors expressed on the surface of epithelial cells, including NSCLC, regulating important processes in cell survival, cell-cycle progression, tumor invasion, and angiogenesis (13, 14). The EGFR pathway has been recognized as an important drug target. The EGFR tyrosine kinase inhibitors (TKI), namely Gefitinib (IRESSA) and Erlotinib (Tarceva) are extensively studied in NSCLC and are currently incorporated into clinical practice. The EGFR pathway is associated with platinum chemotherapy sensitivity in cancer patients in a recent study (15).

A variety of chemotherapeutic drugs are used, and most advanced stage NSCLC patients receive more than 1 drug; therefore, genetic variants of genes in multiple major drug action pathways may have combined effects on outcomes. However, most studies thus far have focused on limited candidate SNPs from a few selected genes. The influence of genetic variation in multiple pathways on NSCLC prognosis remains poorly understood. In this study, genetic variants of 70 key genes in 4 candidate pathways related to major chemotherapy drug actions, that is, GSH, DNA repair, cell cycle, and EGFR pathways (Supplementary Fig. S1), were systematically investigated in a prospectively followed cohort of advanced stage NSCLC patients. Our goal was to examine whether SNP markers from the selected pathways are collectively associated with lung cancer overall survival among NSCLC patients receiving chemotherapy.

Lung cancer patients were identified and enrolled between 1997 and 2008. Detailed procedures of patient enrollment, diagnosis, data collection, and follow-up have been previously described (16–18). Briefly, new cases diagnosed with lung cancer were identified by a daily electronic pathology reporting system. Once identified, patients were invited to participate and enrolled after their consent. The overall participation and blood sample donation rates were 87%. Demographic and tobacco history data were obtained from medical records and interview. When the patient received any therapy elsewhere, authorization for the release of medical information was requested, and copies of the relevant medical records were abstracted. Treatment information was obtained from oncology records, treatment data, outside medical records, and follow-up questionnaires. Surgery was defined as a treatment when the patient had any pulmonary resection for the primary tumor, including pneumonectomy, bilobectomy, lobectomy, segmentectomy, and wedge resection. Chemotherapy agents, dosage, and date treatment started and ended were abstracted. Radiation was categorized as radiation to the chest only, radiation to another place with specific location the radiation occurred, gamma knife, and PCI; elapsed days for radiation, energy, radiation dose, times per day, and fractions were also abstracted. Clinical staging and recurrence or progression was determined by results from available chest radiography, computerized tomography, bone scans, positron emission tomography scans, and MRI. All patients were actively followed beginning within 6 months of diagnosis, with subsequent annual follow-up by mailed questionnaires. Annual verification of vital status of patients was accomplished through the use of electronic medical notes and registration database of Mayo Clinic, next-of-kin reports, death certificates, and obituary documents filed in the medical records of patients, as well as through the Mayo Clinic Tumor Registry and Social Security Death Index website. Well-trained abstractors conducted information abstraction and data entry. Research protocols were approved by Mayo Clinic's Institutional Review Board.

We focused on 4 common candidate pathways related to major chemotherapy drug actions, that is, GSH, DNA repair for platinum-based agents, cell cycle for taxanes, and EGFR pathways following a review of the literature in lung cancer therapy. Note that each of the 4 pathways may relate to multiple drug actions, particularly for individual genes; specifically, a gene can participate in multiple drug action pathways other than the one that was selected. Seventy genes were selected from 4 pathways following a review of the literature (Supplementary Table S1). Tagging single-nucleotide polymorphisms (TagSNPs; refs. 19, 20) were identified via SNPApp, a tagSNP selection program developed by the Bioinformatics Core at Mayo Clinic, Rochester, MN. SNPApp queries multiple public SNP data repositories (Hapmap, SeattleSNPs, and National Institute for Environmental Health Science SNPs) that contain information on known SNPs, using ldSelect (21) to identify tagSNPs for the genes or regions of interest. SNPs within 5 kb of each gene with a minor allele frequency (MAF) of 0.05 or more for European populations were used as candidate SNPs, and tagSNPs were identified with a pair-wise linkage disequilibrium threshold of r² ≥ 0.8. For a given gene, the SNP selection procedure used the SNP data repository with the greatest number of SNPs with a MAF of 0.05 or more and the greatest number of linkage disequilibrium bins that met Illumina Golden Gate Assay quality score thresholds. Nonsynomyous SNPs were preferentially selected as tags when they were identified, provided that their metrics were equivalent to those of alternative tagSNPs.

A total of 1,025 tagSNPs were genotyped in the Mayo Clinic Genomics Shared Resource by using a custom-designed Illumina GoldenGate panel. Concordance among the 3 genomic control DNA samples present in duplicate was 100%. Subjects with a call rate was more than 90% for all subjects; 111 SNPs failed genotyping. Of the SNPs with genotyping data, the call rate was more than 95%, and the minor allele frequency was more than 0.001 for all. SNPs with a Hardy–Weinberg equilibrium test P value less than 1 × 10⁻⁷ (n = 11) and/or monomorphic (n = 9) were excluded, resulting in 894 SNPs in the analyses.

Chemotherapy regimens mostly fall into 4 groups (Supplementary Table S2): (i) platinum agents (P) including carboplatin and cisplatin, (ii) taxane agents (T), paclitaxel and docetaxel, (iii) gemcitabine (G), and (iv) EGFR inhibitors (E), Gefitinib (IRESSA), and Erlotinib (Tarceva). Only a very small group of patients received drugs outside of these 4 groups; therefore, patients were divided into 2 subgroups on the basis of the most commonly used combinations: platinum plus taxanes with or without other agents (PT group), platinum plus gemcitabine with or without other agents (PG group). We conducted subsequent analyses on the significant SNPs in more restricted drug combinations: (i) platinum plus taxanes alone (PT-only group), (ii) platinum plus any other agents except taxanes (P/no-T group), (iii) platinum plus any other agents except gemcitabine (P/no-G group), (iv) platinum plus taxanes without EGFR inhibitors (PT/no-E), (v) platinum plus gemcitabine without EGFR inhibitors (PG/no-E), and (vi) EGFR inhibitors alone or plus any other agents (E group).

Overall survival time was used as the primary endpoint, defined as the time from lung cancer diagnosis to either death or the last known date alive. Patients known to be alive were censored at the time of last contact.

A backward selection process was used to screen potential confounders: age at diagnosis, sex, race, smoking status, stage, histologic types, comorbidity, and treatment modality. The significant variables were retained as covariates in all subsequent analyses. A Cox regression model was used to assess the associations between each genotype of the SNPs and overall survival. The primary test of association was based on a genetic model-free scheme (2 degrees of freedom). In subgroups, for SNPs in which 5 or fewer minor allele homozygotes were observed, homozygote genotypes were combined with heterozygotes. If combined frequency was still 5 or fewer, then the SNP was removed. HRs and 95% CIs were estimated, comparing patients carrying 1 and 2 minor alleles individually to patients carrying 2 major alleles. To account for multiple comparisons in the SNP-based analysis, Q values set at 0.20 were computed by using the SNP P-values to quantify the probability that a P value might be a false positive (22), accepting a false discovery rate of 20%. The significant SNPs were further tested in refined treatment groups by using the same methods outlined above. Kaplan–Meier curves and log-rank tests were used to assess the differences in survival time by individual SNPs.

To assess whether different analytical approaches resulted in consistent findings, we conducted whole-gene tests of association using a principal components analysis (PCA) approach. We utilized a genetic model-free scheme that used 2 indicator variables for each SNP, 1 for the heterozygous and 1 for the homozygous minor allele carriers and extracted sufficient principal components to capture 90% of the SNP. The resulting collection of principal component variables was used for an omnibus test of significance for the association between each gene and survival in the multivariable Cox proportional hazards regression models. P values for the global tests were obtained, along with summaries of the PCA.

All reported P values were based on a 2-sided test. All statistical analyses were done by using SAS, version 9 (SAS Institute Inc.) or R-project software.

Table 1 provides descriptive characteristics of 1,076 advanced stage NSCLC patients. The median overall survival time was 1.7 years with 83.8% deceased within the 11-year follow-up period. Of the 1,076 patients, 962 (89.4%) had chemotherapy, of which 657 (68.3%) were treated with PT, and 305 (31.7%) with PG (Supplementary Table S2). Eighty-eight percent of the patients had documented lung cancer progression or recurrence. In the PT group, gender and treatment modality were significant confounders; and in the PG group, gender, histologic types, and treatment modality were significant confounders (Supplementary Table S3).

The analytic strategy is presented in Figure 1. Significant SNP associations in each treatment group are summarized in Table 2. A poorer overall survival was observed in all patients with minor alleles of rs17309872 in GSS from the GSH pathway and rs17661089 in MAP3K1 from the EGFR pathway (HR = 1.45; 95% CI = 1.20–1.77 and HR = 1.25; 95% CI = 1.05–1.50, respectively; Table 2 and Supplementary Fig. S2). By gene-level tests, GSS and MAP3K1 remained significant (P = 3.76 × 10⁻² and 1.80 × 10⁻², respectively; Table 3).

In the stratified analysis of the PT group, 4 SNPs, rs17661089, rs16886403, rs726501 (in the MAP3K1 gene of the EGFR pathway), and rs11710163 (in RAF1 of the EGFR pathway) were significantly associated with overall survival. Patients with rare homozygote genotypes of these variants in MAP3K1 had significantly poorer overall survival, whereas patients with heterozygotes at rs11710163 (in RAF1) had better overall survival compared with having common homozygotes (Table 2 and Supplementary Fig. S3). By gene-level tests, MAP3K1 and RAF1 were the top 2 significant genes (P = 6.10 × 10⁻³ and 3.14 × 10⁻², respectively) in the PT group, consistent with the SNP level analysis (Table 3). We further analyzed the 4 significant SNPs in the PT-only, P/no-T, PT/no-E, and E groups, accordingly. All of the 4 SNPs were replicated in the PT/no-E group; 3 SNPs in MAP3K1 were replicated in the PT-only group; none showed significance in the P/no-T or E groups (Supplementary Table S4).

In the PG group, rs11710163 in RAF1 was also found to be significant; patients with a heterozygote genotype showed a nearly doubled overall survival compared with patients with a common homozygote genotype (HR = 0.54; 95% CI = 0.38–0.77). The heterozygote genotype at rs451774 in the GPX5 gene also had a protective effect on overall survival, with the HR = 0.61 (95% CI = 0.47–0.79). In contrast, heterozygote genotypes at rs1065634 in NRAS and rs12118636 in GPX7 showed a nearly doubled risk for death (HR = 1.91; 95% CI = 1.30–2.80 and HR = 1.83; 95% CI = 1.27–2.63, respectively; Table 2 and Supplementary Fig. S4). The genes, RAF1, GPX5, GPX7, and NRAS, which harbored the significant SNPs, were also confirmed to be significant (Table 3). We analyzed the 4 significant SNPs in the P/no-G, PG/no-E, and E groups, and all 4 SNPs were replicated in the PG/no-E group; however, none remained significant in the P/no-G or E groups (Supplementary Table S4).

In this study, we systematically investigated the associations between 894 SNPs in 70 key genes from 4 drug action pathways and overall survival in patient groups receiving different chemotherapy agent combinations. Genetic variations in MAP3K1 in the EGFR pathway and GSS in the GSH pathway were associated with overall survival in the analysis of all patients. In the stratified analysis on patients receiving platinum plus taxane treatment, MAP3K1 and RAF1 variations in the EGFR pathway were associated with survival. In patients receiving platinum plus gemcitabine treatment, RAF1 and GPX5 (GSH pathway), NRAS (EGFR pathway), and GPX7 (GSH pathway) showed predictive effects on overall survival.

The glutathione synthetase gene (GSS) encodes the enzyme glutathione synthetase. Mutations in GSS prevent cells from making adequate levels of glutathione, leading to glutathione synthetase deficiency (23). Genetic variations in GSS were also found to be associated with overall survival of small cell lung cancer after treatment (24). It is interesting to observe GSS variants, instead of other commonly reported genetic variations, such as GSTM1, GSTP1, and GSTT1, as the most significant predictive markers in our systematic GSH pathway SNP analysis. It is proactive to observe that genetic variations in MAP3K1 from the EGFR pathway are also significant predictive markers for survival in patients receiving the most commonly used doublet agents, cisplatin or carboplatin and paclitaxel or docetaxel, even after restricting down to patients who did not receive EGFR inhibitor agents.

The SNP and whole-gene analyses consistently indicated that genetic variations in MAP3K1 and RAF1 were significantly associated with overall survival in the PT group. Three SNPs, rs17661089, rs16886403, and rs726501 in MAP3K1, were significant. The associations of these SNPs in MAP3K1 were further replicated in the PT-only group, suggesting that polymorphic variants in the MAP3K1 were specifically associated with overall survival in patients receiving platinum and taxanes. All 4 SNPs were replicated in the PT/non-E group, but not in the E group, suggesting the associations of these SNPs with overall survival may be independent of the effect of the EGFR inhibitors, although both MAP3K1 and RAF1 regulate the EGFR pathway.

MAP3K1 has a pivotal role in a network of phosphorylating enzymes integrating cellular responses to a number of mitogenic and metabolic stimuli. MAP3K1 phosphorylates and activates MAPK kinase (MAPK2), which in turn phosphorylates MAPK/ERK to produce downstream signaling effects on a variety of cancer genes (Supplementary Fig. S1; refs. 25, 26). Researchers have recently reported an association between the EGFR pathway and platinum chemotherapy sensitivity in colorectal cancer (15). Other genes in the EGFR pathway, such as EGFR itself, also have been shown to be associated with lung cancer outcomes (13, 14). For the 3 significant SNPs in MAP3K1, rs17661089 and rs16886403 are in relatively high-linkage disequilibrium with rs726501 (r² = 0.64 and 0.72, respectively). rs17661089 is located in the flanking region of 5′-UTR (untranslated region), and the other 2 SNPs are located in intron 1 of the gene. rs17661089 was predicted to be in the transcription factor binding sites (TFBS); this SNP may be a functional SNP or other variants in high-linkage disequilibrium with it possibly influencing the expression level of the gene product, requiring further functional analyses.

The RAF family is composed of 3 related serine/threonine protein kinases—RAF1, A-RAF, and B-RAF—which act, in part, as downstream effectors of the RAS pathway (27). Activated RAS interacts directly with the amino-terminal regulatory domain of the RAF kinase, which results in a cascade of reactions including direct activation of MEK (28–30). Constitutively active mutated RAF can transform cells in vitro (31). RAF may play a broader role in tumorigenesis and promotes the expression of the multidrug resistance gene, MDR1 (30). RAF1 holds an important role in cell growth, proliferation, and cell survival (32, 33). The RAF1 protein has been found to be amplified in different lung cancer cell lines. One study showed that the RAF expression level is critical in tumor development (34). Ravi and colleagues reported that activated RAF1 causes growth arrest in human small cell lung cancer cells (35). Immunohistochemical staining indicated that RAF1 was present in 49 of 53 ovarian adenocarcinomas and high c-RAF expression correlated significantly with poor survival in ovarian cancer (36). The SNP rs11706408 that has high-linkage disequilibrium (r² = 1) with rs11710163 in RAF1 was predicted to be in the TFBS; their functional roles in NSCLC chemotherapy agent actions and patient survival need to be further investigated.

Four SNPs in RAF1, GPX5, GPX7, and NRAS were found to be significantly associated overall survival in the PG group. All 4 SNPs were replicated in the PG/non-E group but were not replicated in the E group, indicating the association of being independent of the effect of the EGFR inhibitors. None of the 4 SNPs remained significant in the P/non-G group, suggesting the association of the variations in RAF1, GPX5, GPX7, and NRAS with overall survival in the PG group may reflect the effect of gemcitabine.

The GSH pathway has been related with detoxification or inactivation of platinum drugs. Glutathione peroxidases (GPX) are a major antioxidative damage enzyme family that catalyzes the reduction of hydrogen peroxide, organic hydroperoxide, and lipid peroxides by reduced glutathione (37–39). GPX has at least 7 allozymes, distributed in different organs. In this study, GPX5 rs451774 and GPX7 rs12118636 were found to be associated with overall survival in the PG group. The association of GPX5 and GPX7 were also confirmed by whole gene-based analysis. Function prediction indicated that rs451774 located at 3′ UTR of GPX5 was in both TFBS and miRNA binding sites and is a potential functional SNP. GPX5 has structural similarity to GPX1 (40). A recent study reported GPX1 may be an inherited factor in predicting quality of life of patients (41). In previous studies, variation in GPX1 was linked tumor recurrence of bladder cancer (42). GPX7 shares a similar structural domain to GPX4. Reported findings suggest a potential role for GPX7 in alleviating oxidative stress induced by dietary consumption of fatty acids in breast cancer cells (43). Common variations in GPX4 were associated with prognosis after a diagnosis of breast cancer (44). However, no reports are currently available on the association of GPX5 and GPX7 variations and clinical outcomes. Our findings showed that genetic variations in GPX5 tend to predict a better survival, whereas variations in GPX7 were associated with worse survival. Current evidence also indicates the diversified roles of the individual GPX (45); individual GPX might have different impact on the clinical outcome of lung cancer.

The SNP, rs1065634, is located at 5′ near the gene region of NRAS. This SNP was also found to be significant in the PG group. NRAS is a member of the RAS proto-oncogene family. Point mutations of the RAS proto-oncogene family members are among the most frequent genetic alterations found in human cancer (46, 47). Until now, NRAS gene mutations were mainly associated with hematopoietic malignancies (48), melanomas (49), and bladder cancers (50). rs1065634 was predicted to be in both TFBS and miRNA binding sites, and may be an important functional SNP; its potential functional significance is required to further study.

To our knowledge, our study presents the first and largest effort to comprehensively characterize the associations of gene alterations in 4 chemotherapy drug action pathways and survival in advanced stage lung cancer patients. Platinum drugs, combined with other cytotoxic agents, continue to be the first-line chemotherapy for advanced stage lung cancers and other metastatic cancers. Our study is strengthened by using different agent combinatorial groups for stratified analysis, which more closely mimics clinic practice than studies by using single drug-based groups. We also used different analytic strategies and subgroups to confirm our results. There are several limitations. First, despite efforts to account for potential confounding factors and careful adjustment for multiple tests in the analyses, false-positive associations cannot be completely ruled out. Second, although the significant SNPs were carefully analyzed in different combinatorial agent subgroups, chemotherapy heterogeneity may also modify the true association between genetic variation and clinical outcomes. Third, although our patient cohort represents one of the largest among all published combinatorial agent pharmacogenetic studies for late-stage NSCLC, the power of this study to detect association was still modest, and it is possible that we missed some real associations with a small effect. For example, in the PG group, using a type I error level of 6 × 10⁻⁵ (0.05/894), we had 86% power to detect effects from SNPs with a minor allele frequency of 0.3, which confers a per-allele HR of 1.6. If the per-allele HR is instead 1.4, the power is only 35%. Fourth, a thorough evaluation of the gemcitabine pathway SNPs was not included in this study. An additional limitation was the lack of patients treated with first-line EGFR inhibitor drugs, so the specific SNP effects could not be detected or evaluated relevant to these drugs. Our study focused on late-stage NSCLC patients who were treated with platinum drugs with or without other treatments, which will limit the ability to compare our results to other data sets. Nonetheless, our study is useful for testing a novel hypothesis, which may be confirmed in future specifically designed clinical studies.

In summary, our results suggest that variations of MAP3K1 and RAF1 may be associated with overall survival among patients treated with platinum and taxane agents. Polymorphisms in the RAF1, NRAS, GPX5, and GPX7 genes were found to be associated with overall survival among patients treated with platinum and gemcitabine agents. In current clinical practice, oncologists usually select patients who they expect to benefit from chemotherapy based on available clinical information. Validated biomarkers, even with small effects, may have utility in optimizing patient therapies. Therefore, our findings could be considered in prospective biomarker-stratified clinical trials.

No potential conflicts of interest were disclosed.

We thank Susan Ernst, M.A. for her technical assistance with the manuscript and David N. Rider (Division of Biomedical Statistics and Informatics, Mayo Clinic) for his assistance in SNP selection.

P. Yang led the study by designing and conducting it, interpreting results, writing the manuscript, and obtaining funding. Y. Li coordinated the data analysis, results interpretation, and writing of the manuscript. Z. Sun searched the literature for the genes in the pathways, conducted tagSNP selection/custom panel design, and contributed to the final manuscript writing. V.S. Pankratz led the statistical analyses and contributed to the final manuscript writing. J.A. Wampfler and J.A. Aakre participated in the data analysis. M.C. Aubry undertook pathology verification. J.M. Cunningham led the genotyping of the Mayo Clinic samples. J. Molina and G.A. Croghan coordinated participant eligibility determination and enrollment. D. Wu and C. Johnson abstracted medical records data. L. Wang contributed to data collection and manuscript writing. All authors contributed to the final paper.

This study was supported by U.S. NIH research grants, R01-CA80127 (to P. Yang) and R01-CA84354 (to P. Yang), and Mayo Foundation funds (to P. Yang). Support was also provided by the NIH grant, P30 CA15083 (Mayo Clinic Comprehensive Cancer Center-Gene Analysis Shared Resource).

The costs of publication of this article were defrayed in part by the payment of page charges. This article must therefore be hereby marked advertisement in accordance with 18 U.S.C. Section 1734 solely to indicate this fact.