MicroRNA-Related Genetic Variants Associated with Clinical Outcomes in Early-Stage Non–Small Cell Lung Cancer Patients

Lung cancer is the leading cause of cancer related mortality in the United States (1). Most patients with early-stage non–small cell lung cancer (NSCLC) are treated with curative-intent therapy. However, 50% of surgically resected patients will relapse within 5 years. Thus, there is a strong need to identify reliable prognostic and predictive biomarkers to assist in developing personalized therapy and follow-up care. Germ line polymorphisms are characterized by their stability and accessibility. They have been identified as potential prognostic/predictive markers for NSCLC clinical outcomes and treatment response (2).

MicroRNAs (miRNA) are a class of small, noncoding RNAs approximately 22 nucleotides in length. Emerging evidence has shown that miRNAs function as oncogenes or tumor suppressor genes depending on the context (3–5) and have been shown to be potential biomarkers for cancer risk assessment, treatment response, and prognosis (6). MiRNAs undergo a complex processing procedure to produce the mature, functional unit (7), and impaired miRNA processing has been reported to reduce stable miRNA levels and promote tumorigenesis (8). Genetic variations in several miRNA processing genes have been reported to influence risk of several cancers (9–12). In addition, variations in miRNA-binding sites within 3′-untranslated regions (3′UTR) of target genes have been found to contribute to different outcomes in patients with cancer (13–15), which could be a result of altered miRNA–mRNA interactions followed by changes in target gene expression (14). For example, Zhang and colleagues found that the G allele of rs1044129 in the miR-367–binding site of RYR3 was related to poor survival in 1,125 patients with breast cancer (16). Campayo and colleagues showed that several miRNA-binding site single-nucleotide polymorphisms (SNP) in KRT81 were associated with time to recurrence in 175 surgically resected patients with NSCLCs (13).

In this study, we conducted an analysis of 77 SNPs in 8 miRNA processing genes and 163 SNPs in predicted miRNA-binding sites for 133 cancer-related genes. We evaluated associations between these variants with overall survival and time to recurrence in patients with early-stage (I and II) NSCLCs treated with curative therapy and also in subgroups of patients who received surgery-only or surgery plus chemotherapy. We also conducted luciferase reporter assays to determine the effect of selected binding site SNPs on gene regulation.

All the subjects included in the analysis were histologically confirmed patients with NSCLCs recruited at MD Anderson Cancer Center (Houston, TX) from September 1995 to February 2008, which is part of an ongoing lung cancer study initiated in 1991. Among all potential participants approached, 75% consented and were enrolled into the study (17). Blood samples were drawn from each participant. We restricted to early-stage patients (stage I and II) who received curative-intent therapy (i.e., surgical resection, chemotherapy, and/or radiation therapy). The last day of follow-up for this study was December 31, 2009. At the time of last follow-up, 38 patients were lost to follow-up and 284 patients were alive. Staging was based on American Joint Committee on Cancer (AJCC) staging system (version 6). A structured questionnaire was used to collect epidemiologic data during an in-person interview. Medical records were reviewed to collect clinical and follow-up information. Status of recurrence was ascertained by medical chart review. Vital status was ascertained by linking patient records to MD Anderson Tumor Registry that conducts annual follow-up on all patients with cancer. Deaths of patients were further confirmed by checking the social security death index. All patients signed an informed consent form and the study was approved by the Institutional Review Board of MD Anderson Cancer Center.

SNPs were genotyped on a custom Illumina iSelect Infinium II genotyping platform (Illumina) containing a comprehensive panel of approximately 10,000 SNPs from 998 cancer-related genes. The details for the design of this chip, including the SNP and gene selection, were described previously, duplicates were included for 2% of all samples; the concordance rates were greater than 99% (18). Eight miRNA processing genes (DDX20, DGCR8, DICER1, DROSHA, EIF2C1, GEMIN4, RAN, and XPO5) were among the genes on this chip with 77 tagging (10 kb flanking and within each gene, linkage disequilibrium, r² > 0.8) and potential functional SNPs genotyped. We used the PolymiRTS v1.0 database to identify SNPs in predicted miRNA-binding sites (all the 3′-UTR SNPs located within the seed region of responding miRNAs; ref. 19) for the genes included on the chip and identified a total of 163 SNPs from 133 genes. These SNPs were selected to test the hypothesis that miRNA-related genetic variations could influence NSCLC patients' clinical outcomes. All of the selected SNPs had a minor allele frequency greater than 0.01 in the Caucasian population. gDNA was extracted from peripheral blood samples using the QIAamp DNA extraction kit (Qiagen). Only SNPs with sample call rate more than 95% and samples with SNP call rate more than 95% were included in the analysis.

Luciferase reporter constructs for wild-type and variant binding site regions for FAS:rs2234978 and SP1:rs17695156 were generated. Constructs were sequenced to ensure the correct insert. NSCLC cell lines (NCI-H460 and NCI-H2444) were purchased from American Type Culture Collection in 2003 and were validated for identity by short tandem repeat DNA fingerprinting by the Characterized Cell Line Core facility at MD Anderson in April 2012. Cell lines were cultured in RPMI-1640 medium (Mediatech) supplemented with 10% FBS (Invitrogen) in 48-well plates. Cells were transfected with 0.5 mg of each reporter construct, 5 pmol of negative control (scrambled sequence), or predicted targeting miRNAs (Sigma-Aldrich) and 8 ng of pGL4 (Ambion) Renilla luciferase reporter using Lipofectamine 2000 (Invitrogen). After 36 hours of incubation, cell lysates were harvested and measured for activity using the Dual-Luciferase Reporter Assay System (Promega) on a FLUOstar Optima microplate reader (BMG Labtech). Each assay was repeated independently at least twice with 4 replicates. Firefly luciferase activity was normalized to the Renilla luciferase activity to derive the relative luciferase activity.

Cox proportional hazard modeling was used to assess the association of SNPs with overall survival (the time between recruitment and death or last follow-up) and recurrence (time from recruitment to recurrence, progression at the local/distant site/lymph nodes, or last follow-up). Patients who came to MD Anderson for treatment due to recurrence were excluded from the recurrence analysis. HRs and 95% confident intervals (CI) were estimated while adjusting for age, gender, ethnicity, stage, pack-years of smoking, and treatment regimens. Kaplan–Meier curves and log-rank tests were used to assess effect of individual SNPs on time to recurrence or overall survival. Likelihood ratio test was used to analyze the effect modifications of treatment types on SNP associations. Statistical analyses were conducted using STATA software (10.1, Stata Corporation). Survival tree analysis was conducted using the STREE program (20) to build a decision tree using recursive partitioning method. Briefly, the root node contained all the patients, and we defined the measure for goodness of split using log-rank P value to select the optimal initial split and subsequent splits of the dataset until no statistically significant split was identified (20). All test were 2-sided and associations with P < 0.05 considered significant. Multiple hypothesis testing was conducted using the “q-value” package in R (21) based on a false discovery rate (FDR) of 10%. As 240 SNPs were tested for 3 models in each outcome analysis; therefore, the P value was adjusted for multiple comparisons based on 720 tests. Bootstrap resampling was done for 500 iterations. In each resampling run, sampling with replacements was used to obtain same number of patients as the original analysis.

This study included 535 patients with early-stage NSCLCs with an overall median survival time (MST) of 90.2 months and median follow-up time of 62.1 months. At the time of analysis, 213 (40%) of the patients had died and 133 (33%) had a recurrence of their disease. The majority of the NSCLC cases were adenocarcinomas (59%). Of the 535 participants, 340 patients received surgery-only, 127 patients were treated with surgery plus neoadjuvant and/or adjuvant chemotherapy, and the remaining was treated with radiation therapy with/without surgery (Table 1).

Eleven processing and 23 binding site SNPs were significantly associated with survival. One SNP, FAS:rs2234978 (HR, 0.59; 95% CI, 0.44–0.77; P = 1.67 × 10⁻⁴), remained significant after multiple comparison correction, with the GA + AA genotype resulting in a significant increase in MST from 59 to 118 months (log-rank, P = 1.0 × 10⁻⁴; Fig. 1A).

Five SNPs in processing genes and 23 SNPs in binding sites were significantly associated with time to recurrence. The most significant association was SP1:rs17695156 (HR, 2.22; 95% CI, 1.44–3.41; P = 3.00 × 10⁻⁴). Patients with at least one variant allele had a much shorter median recurrence-free time (MRFT) than patients who had common homozygous genotype (45.3 vs. >270 months, log-rank; P = 7.0 × 10⁻⁴; Fig. 2A). However, this association did not reach significance after correcting for multiple comparisons.

We conducted subgroup analysis focusing on 2 groups of patients: surgery-only and surgery plus chemotherapy.

Eighteen SNPs were significantly associated with overall survival in surgery-only patients (Supplementary Table S1). FZD4:rs713065 (HR, 0.46; 95% CI, 0.32–0.65; P = 2 × 10⁻⁵) was the only SNP that remained significant after adjustment for multiple comparisons. Patients with at least one variant allele had significantly decreased risk of death and increased MST compared those patients with the common genotype (MST: 117 vs. 59 months, log-rank; P = 1.05 × 10⁻⁵; Fig. 1B). Similar to the results from the overall analysis, for patients who received surgery plus chemotherapy, FAS:rs2234978 displayed the most significant association with survival (Supplementary Table S1). Patients with at least one variant allele had a 81% lower risk of death (HR, 0.19; 95% CI, 0.07–0.46) and significantly longer MST (137 months) than patients who carried the homozygous common genotype (65 months; log-rank; P = 1.05 × 10⁻⁴; Fig. 1C). The association of this SNP with survival was borderline significant after correction for multiple comparisons in surgery-only patients (HR, 0.59; 95% CI, 0.42–0.84; q = 0.062), with increased MST (61 vs. 102 months, log-rank; P = 4.02 × 10⁻³; Fig. 1D) in these surgery-only patients.

By comparing the findings between 2 subgroups, 17 SNPs were found to have significant effects on survival in surgery-only patients but not in patients receiving surgery plus chemotherapy. In contrast, 28 SNPs were significantly associated with risk of death only in patients receiving surgery plus chemotherapy. Intriguingly, within each cluster of SNPs that were significantly associated with outcomes in each treatment subgroup, we identified SNPs with differential effects. A group of 29 SNPs has the same trend in both treatment subgroups (either associated with increased or decreased risk), whereas 15 SNPs conferred opposite effects (Supplementary Table S1). For example, FZD4:rs713065 was associated with significantly decreased risk of death and prolonged MST in surgery-only patients; however, in the surgery plus chemotherapy subgroup, this SNP was associated with increased risk of death and a shortened MST (P_interaction = 0.004, q = 0.030; Fig. 1B and E).

SP1:rs17695156, which was the top SNP associated with recurrence in the overall population, was the most significant SNP in surgery plus chemotherapy group (HR, 3.36; 95% CI, 1.62–6.69; P = 1.10 × 10⁻³; Supplementary Table S2). In surgery-only patients, one miRNA processing SNP, DROSHA:rs6886834, remained significant after multiple comparisons corrections; it was significantly associated with more than a 6-fold increased risk for recurrence in surgery-only patients (HR, 6.38; 95% CI, 2.49–16.31; P = 1.10 × 10⁻⁴; Table 2). Patients who carried at least one variant allele of this SNP had significant reduction in MFST compared with patients with common genotype (23 vs. >270 months, log-rank; P = 5.0 × 10⁻⁴; Fig. 2B).

When comparing results of subgroup analysis, 28 and 16 SNPs were exclusively associated with altered risk for recurrence in surgery-only patients or surgery plus chemotherapy patients, respectively. Of these, 17 SNPs were found to have opposite effects in both subgroups (Supplementary Table S2). For example, in patients receiving surgery plus chemotherapy, RRM2B:rs5005121 was associated with significantly increased risk of recurrence and a shortened MFST; but in surgery-only patients, this SNP was associated with decreased risk for recurrence with an increased MFST (P_interaction = 0.015, q = 0.170; Fig. 2C and D); however, the interaction between this SNP with treatments was not significant after multiple comparison corrections.

Figure 3 shows the survival tree structure classifying patients into subgroups with distinct risk of dying based on their risk genotype combinations. SNPs that displayed at least borderline significant association with survival in the main effect analysis after multiple comparisons (q < 0.15) were included in the analysis, and none of these SNPs were in high linkage disequilibrium. The MSTs based on these groupings varied from >86 months for the low-risk group to 41.7 months for the high-risk group in surgery-only patients, and from >118 to 36.8 months for the low- and high-risk groups, respectively, in patients receiving surgery plus chemotherapy. Moreover, the initial splits in the tree structure for each subgroup, FZD4:rs713065 and FAS:2234978, were also the 2 SNPs that remained significant after multiple comparisons in the treatment subgroup analyses. Survival tree analysis was not conducted for recurrence due to limited number of SNPs with q < 0.15.

All the SNPs that were significant after multiple comparisons at an FDR of 5% remained significant in the bootstrap analysis for at least 450 of 500 resamplings, providing internal validation to these results. Bootstrap resampling analysis was also conducted for survival tree analyses, and the results were significant in both the subgroups analysis for entire 500 resamplings at P < 0.05 (Supplementary Fig. S1).

Luciferase reporter assays were conducted to determine the effect of these predicted binding site variants on miRNA regulation of target genes. FAS:rs2234978, which was consistently associated with a favorable prognosis, was predicted to create a new miRNA-binding site for miR-561. In 2 lung cancer cell lines, H460 and H2444, a significant reduction in luciferase activity was observed when miR-561 was transfected with the variant reporter (T; H460: P = 0.0294; H2444: P = 0.025) but not with wild-type allele (C) construct (P > 0.5 for both cell lines), when compared with the scrambled sequence control transfections. Furthermore, there was a significant difference in luciferase activities between the variant and the wild-type constructs when co-transfected with miR-561 (H460: P = 0.0015; H2444: P = 0.0040; Fig. 4)

SP1:rs17695156 was predicted to disrupt a conserved miR-545–binding site; however, in our in vitro assays, suppression of luciferase activity was observed in both variant and wild-type constructs co-transfected with miR-545. There was no significant difference in reporter activities between the 2 alleles, and the extent of signal decrease varied between cell lines (data not shown).

In this study, we identified genetic variants in miRNA processing genes and miRNA-binding sites near cancer-related genes that were associated with overall survival and recurrence in patients with early-stage NSCLCs. FAS:rs2234978 was identified as a potential prognostic factor in our results, and functional data provided evidence that this SNP alters miRNA regulation of FAS. We also found evidence that some SNPs exhibited associations that were treatment-specific. These results suggest that genetic variants in miRNA processing genes and miRNA-binding sites may serve as potential prognostic markers for survival and predictive markers of response to treatment.

The most significant SNP associated with survival was FAS:rs2234978, which was consistent regardless of treatment regimens. FAS is a cell surface receptor of the tumor necrosis family that plays an important role in the regulation of apoptosis. Evidence has shown that FAS expression and polymorphisms could influence lung cancer patients' prognosis (22, 23). rs2234978 is a synonymous SNP located in the seventh exon of FAS. miRNA-binding sites for FAS are located in exon 7 instead of the typical 3′UTR. Alternative splicing of FAS results in several transcribed isoforms that are involved in nonsense-mediated mRNA decay (NMD), including a transcript where exon 7 serves as the 3′UTR. NMD plays important roles in limiting the synthesis of truncated or mutant proteins, which can negatively regulate apoptosis mediated by the full-length protein. This SNP is predicted to create a new miRNA-binding site for miR-561, which was supported in vitro by our luciferase assay, suggesting decreased expression of FAS alternative transcripts. As the NMD transcripts may negatively regulate normal FAS expression, this would ultimately result in increased level of FAS in tissues that express the targeting miRNA. It has also been reported that cisplatin treatment can increase FAS-mediated apoptosis (24). It is possible that in patients who carry the variant allele, higher expression of FAS could increase tumor cell death resulting in better overall survival independent of treatment regimen. This locus might even be synergistic with chemotherapy agents such as cisplatin, thus conferring a more extreme effect on patients' survival. Further studies are needed to confirm whether this SNP has any influence on FAS protein level and apoptotic activity in vivo.

FZD4:rs713065 is the only SNP associated with significant decreased risk of death after multiple comparison correction in surgery-only patients. FZD4 (frizzled homolog 4) is a member of the frizzled gene family of transmembrane receptors, which help to transduce WNT signals and activate downstream WNT/β-catenin pathway components in cancer stem cell homeostasis (25). This SNP may downregulate FZD4 expression by creating an miRNA-binding site, thereby inhibiting transduction of the WNT signal, leading to enhanced survival through decreased WNT signaling. However, because of difficult sequence characteristics of this region, luciferase assays were not possible.

SP1:rs17695156 is the most significant SNP associated with increased risk for death. SP1 is a transcription factor known to regulate expression of many genes, thus having a general regulatory role within the cell. This SNP is predicted to disrupt a conserved miRNA site; however, in our in vitro experiments, we did not observe any significant difference between the 2 alleles in miRNA-induced repression of reporter activity. It is possible that this 3′UTR SNP might affect SP1 expression independent of its putative role as a miRNA target site (e.g., affecting RNA stability or posttranscriptional regulation) and influence cellular components at physiologic levels. And it is also possible that rs17695156 is tagging some other miRNA-binding site polymorphism that affects transcription but has not yet been identified by binding site prediction algorithms or regulated by miRNAs that have not yet been discovered.

In this study, we identified panels of SNPs from the miRNA processing pathway and miRNA-binding site SNPs in major cancer-related pathways exclusively associated with clinical outcomes in either of the 2 treatment subgroups. Although several genome-wide association studies (GWAS) have been published for NSCLC survival (26–31), the SNPs identified in our study were not covered, indicating there is still a need for complementary pathway-based analysis to identify novel loci associated with clinical outcomes. Most microRNA-related variants are not well covered by current GWAS platforms, meaning that it is very likely that our identified potential functional SNPs would not have been identified through GWAS. In addition, the previous studies were often limited to advanced-stage NSCLCs and relatively small patient populations. In this study, the adoption of a pathway-based approach in a large, well-characterized population allowed for the discovery of novel loci of interest and enhances our understanding of the genetic mediators of clinical outcomes in NSCLCs. Our results provide supportive evidence that genetic variations could potentially interact with treatment to influence patients' clinical outcomes, especially survival, therefore highlighting the necessity of personalized treatment decisions based on patients' genetic background. Meanwhile, the tree structure identified could potentially assist the decision by classifying patients into different risk groups in a more intuitive manner. Because of the relatively favorable prognosis and long survival time, around 30% to 40% of patients with early-stage NSCLCs eventually do not die due to lung cancer (32, 33). In this study, because we focused on overall survival of patients with early-stage NSCLCs, additional investigation will be necessary to further elucidate whether or not these associations are specific to lung cancer clinical outcomes and the mechanism of the identified associations.

The strengths of the current study include the comprehensive query of SNPs from genes involved in cancer-related miRNA regulation and the evidence for biologic plausibility provided by in vitro functional assays. Although the luciferase assay, by design, does not prove that altered miRNA-binding results in changes in host gene expression or protein level, it provides evidence supporting the effect of this SNP on the function of this specific miRNA-binding sites, suggesting a downstream effect on gene expression and protein levels. The analysis also took into account the effect of treatment and adopted FDR and bootstrap resampling methods to exclude potential false-positive results.

Overall, the current study provides evidence that genetic variants in the miRNA processing pathway and miRNA-binding sites influence clinical outcomes for patients with early-stage NSCLCs. Specifically, we identified the potential prognostic role of a FAS SNP in predicting overall survival in these patients and supported this observation with in vitro functional analyses. Following validation in an independent population, our results could provide a basis for future personalized medicine whereby those patients with early-stage NSCLCs with high probability for favorable outcomes can be identified and treated with optimal regimens.

No potential conflicts of interest were disclosed.

Conception and design: X. Pu, J.D. Minna, X. Wu

Development of methodology: X. Pu, H. Wei, X. Wu

Acquisition of data (provided animals, acquired and managed patients, provided facilities, etc.): X. Pu, X. Wu

Analysis and interpretation of data (e.g., statistical analysis, biostatistics, computational analysis): X. Pu, J.A. Roth, M.A.T. Hildebrandt, Y. Ye, X. Wu

Writing, review, and/or revision of the manuscript: X. Pu, J.A. Roth, M.A.T. Hildebrandt, H. Wei, J.D. Minna, S.M. Lippman, X. Wu

Administrative, technical, or material support (i.e., reporting or organizing data, constructing databases): Y. Ye, S.M. Lippman, X. Wu

Study supervision: X. Wu

PI of NCI SPORE grant that provided financial support and overall supervision of the Project and Core resources for the publication: J.D. Minna

This research was supported in part, by National Cancer Institute (R01 CA111646, P50 CA070907, and R01 CA055769).

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.