Cytokine and Cytokine Receptor Single-Nucleotide Polymorphisms Predict Risk for Non–Small Cell Lung Cancer among Women

In the United States, lung cancer is the second leading form of cancer, is expected to affect 215,020 persons in 2008, and is the leading cause of cancer-related death (1). Whereas lung cancer incidence and mortality among men have been decreasing for several years, mortality rates among women steadily increased from 1995 to 2002 (2). Although the 5-year survival rate for lung cancer diagnosed at a distant stage is 3%, that rate increases to 49% when disease is diagnosed at a localized stage, further underscoring the need for earlier diagnosis (3). Because only 15% of smokers develop lung cancer, identification of individuals for screening efforts needs to move beyond enumeration of pack-years of smoking. Screening for single-nucleotide polymorphisms (SNPs) in candidate pathway genes is one approach to defining a high-risk subgroup.

Inflammatory pathways have been implicated in lung cancer development. Research indicating a role for inflammation in lung cancer includes a relationship between COX2 polymorphisms, COX-2 expression, and lung tumorigenesis (4-6). Aspirin and other non-steroidal anti-inflammatory drugs (NSAIDs) may be associated with a decreased risk for lung cancer (7, 8). Moreover, chronic inflammatory conditions such as chronic obstructive pulmonary disease, chronic diffuse infiltrative lung diseases, and chronic lung infections have been associated with an increased risk for developing lung cancer (9).

Studies indicate that there are differences between lung cancer cases and controls in circulating cytokine expression profiles (10-15), and polymorphisms also have been investigated to determine whether variation in cytokine genes predicts risk for non–small cell lung cancer (NSCLC). Cytokine SNPs primarily examined thus far include those in IL1B, IL1RN, TNF, IL10, IL6, and IL8. These studies suggest an association between IL1, TNF, and IL10, and risk for NSCLC (16-19). Of the 11 previously published studies, none have focused on these relationships in African-Americans and only one was conducted in the United States (20). Differences in cytokine polymorphism allele frequencies by race have been seen, suggesting that the relationship between cytokine SNPs and risk of lung cancer may vary by race (21-25).

In a large population-based case-control study, we examined the association between cytokine and cytokine receptor polymorphisms and risk of NSCLC in African-American and Caucasian women.

Population-based cases were identified via the Metropolitan Detroit Cancer Surveillance System, a member in the Surveillance, Epidemiology, and End Results program of the National Cancer Institute. Eligible participants included women ages 18 to 74 y, diagnosed with primary NSCLC in the metropolitan Detroit area between November 1, 2001, and October 31, 2005. Initially, participant eligibility was limited to adenocarcinoma cases. However, because many histologic diagnoses at the time of rapid case ascertainment were not more specific, after November 1, 2004, eligibility was expanded to include all NSCLC histologies. Women were targeted for this investigation because these data were collected as part of a study examining estrogen and other hormone exposure, and risk of NSCLC. Subsequently, men were not included in the study.

Because the questionnaire included detailed questions related to extensive family history and history of comorbidities, no proxy interviews were conducted; therefore, women deceased at ascertainment or first contact were ineligible. Five hundred seventy-seven (54.5%) women completed an interview. Because lung cancer is a rapidly fatal disease, many women were too ill (n = 129) at the time of first contact to complete the interview. Two hundred sixty-three women refused to participate. We excluded women self-reporting race other than African-American or Caucasian (n = 15) because there were too few cases of other racial backgrounds for analyses stratified by race. In total, 459 women with NSCLC who provided a blood specimen were available for analysis.

Population-based controls, identified by random-digit dialing, were frequency matched to cases on race and 5-y age group. Approximately 70% (n = 575) of the households that completed the eligibility screening questionnaire participated in the interview; 209 women refused to participate. Eleven controls who reported race other than African-American or Caucasian were excluded. In total, 483 controls who provided a blood sample were included in the analysis.

This study was approved by all local institutional review boards; informed consent was obtained from each subject. In-person interviews were conducted to collect demographic information, smoking history, health history, and environmental tobacco smoke exposure. Exsmokers were defined as women who quit smoking >2 y before diagnosis/interview. Never smokers included those who smoked <100 cigarettes in their lifetime. Demographic information included age and residence at diagnosis/interview, date of diagnosis, date and place of birth, marital status, race, and number of years of education. Medical history included self-report of physician diagnoses of asthma, emphysema, allergies, pneumonia, bronchitis, chronic obstructive pulmonary disease (COPD), and tuberculosis. Reports of diagnoses of lung diseases within 1 y of lung cancer diagnosis (for cases) or interview (for the controls) were excluded. To limit misclassification, any patient who reported a physician diagnosis of COPD, chronic bronchitis, or emphysema was coded as having COPD.

Smoking history included age began and quit smoking, years of smoking, average number of cigarettes per day, type of cigarette, and total years of smoking interruption. Smoking pack-year history was calculated by multiplying the number of packs smoked per day by the number of years smoked where a pack was defined as 20 cigarettes. Never smokers are included in this analysis as having a zero smoking pack-year history. Environmental tobacco smoke exposure at home and within the workplace, as well as hours and years of exposure for each situation, were collected. Family history of lung cancer was coded as yes or no based on the detailed first-degree family history information. Lung cancer diagnosis dates and histology information were obtained through the Metropolitan Detroit Cancer Surveillance System.

Medication history included regular use of aspirin, defined as taking at least one pill thrice per week or more for at least a month during an individual's lifetime, ages at which participants started and stopped pill use, and number of pills per week. Aspirin use was distinguished between baby/senior citizen aspirin (81 mg) and adult-strength aspirin (325 mg). We excluded from analysis pill use 1 y before diagnosis/interview.

Blood was collected in Vacutainer Plus tubes containing EDTA. DNA was isolated from whole blood with a Qiagen AutoPure LS Genomic DNA Purification System (Gentra Systems) following the manufacturers' protocols.

Genomic DNA (250 ng) was submitted to the Wayne State University Applied Genomics Technology Center for genotyping. The Illumina GoldenGate assay using the Cancer SNP Panel was utilized. The candidate SNPs on the Cancer SNP Panel were identified from the Cancer Genome Anatomy Project SNP500 Database of the National Cancer Institute, were selected to be within 10 kb of each gene, and represent pathways involved in tumorigenesis. SNPs in cytokine genes, including those for ILs, TNF, TNF receptor superfamily members, IFNs, and transforming growth factor (TGF) β, were included in our analyses. This panel includes 83 cytokine and cytokine receptor SNPs. The GoldenGate assay was run according to the manufacturer's directions. The data were analyzed using Bead Studio software (Illumina).

Cases were compared with controls on demographic factors using χ² tests for categorical variables and t tests for continuous variables. Hardy-Weinberg equilibrium was assessed by a χ² goodness-of-fit test for each SNP for African-American and Caucasian controls separately. Participating cases were compared with nonparticipating cases on race, age at diagnosis, and stage at diagnosis using t tests for continuous variables and χ² tests for categorical variables. The Kaplan-Meier method and log-rank test were used to compare median survival between the two groups. SNPs with minor allele frequencies <5% or that were in Hardy-Weinberg disequilibrium among the controls were excluded from subsequent analyses. Multivariable unconditional logistic regression models were constructed by race using SAS V9.1.3 (SAS Institute) with adjustment for age at diagnosis, smoking pack-year history, smoking history (ever/never), family history of lung cancer, history of COPD, years since COPD diagnosis, adult aspirin use, education, and body mass index (BMI). Model fit was assessed by race by calculating 1-γ̂, a measure of overfitting (26). This model was validated internally using a bootstrapping method to obtain a bias-corrected Somers' D_xy rank correlation. Heterozygotes were combined with the less frequent homozygotes in a dominant model testing for the relationship between presence of the minor allele and risk for NSCLC separately by race. To assess the associations between genes and lung cancer risk, a genetic risk score was calculated for each gene as a weighted summation of the SNPs for each gene where the weight equals the t statistical (coefficient divided by its SE) of that marker in the logistic regression model. The relationship between the minor allele and risk for lung cancer was also assessed separately by smoking history (never versus ever) and history of COPD among Caucasians. The number of African-Americans included in this study was too few for this stratified analysis. P values were adjusted for multiple comparisons according to the Benjamini and Hochberg (27) false discovery rate method. Analyses were repeated predicting risk for adenocarcinoma of the lung. Random forest analysis was also conducted by race using the randomForest package in R (version 2.7.2) as an alternative approach to model the data.

PLINK V1.01 and the haplo.ccs package in R were used for haplotype construction and testing, respectively (28, 29). Haplotypes were constructed separately by race among controls. To determine if a SNP was associated with risk for lung cancer independently of the other SNPs in the haplotype, tests of the association between a particular SNP and risk of NSCLC in the context of the haplotype were conducted. Associations between inflammatory pathway gene haplotypes and risk for NSCLC were estimated via unconditional logistic regression by race adjusting for other factors associated with lung cancer using R.

To predict the relationship between nonsynonymous SNPs in coding regions analyzed in this study and protein function, the Sorting Intolerant from Tolerant (SIFT) program was used (30). Using homologues in the protein alignment, SIFT scores <0.10 were considered to be deleterious. Median sequence information content values >3.0 indicate lower sequence diversity and higher chance for false positive error. To analyze the relationships among selected genes in a biologically informative manner through pathway analysis, we used the Ingenuity Pathway Analysis V7 program (Ingenuity Systems, Inc.).

Participant characteristics are displayed in Table 1. Approximately 20% of cases and controls were African-American. Cases did not differ from controls in age or race. Cases were significantly more likely than controls to report being current smokers and to have a history of COPD. We previously reported that cases more frequently reported having a family history of lung cancer in a first-degree relative were less likely to report having more than a high school education or general education development (GED) and had a longer smoking pack-year history and a lower BMI at the time of interview (7). Participating cases did not differ from nonparticipating cases in terms of race. Nonparticipating cases were slightly older (mean ± SD, 63.5 ± 8.1 versus 60.1 ± 9.3 years; P < 0.0001), were more likely to be diagnosed with distant disease (41% versus 32%; P = 0.03), and had a significantly shorter median survival time [33 (29-37) versus 47 (41-56) months; P < 0.0001].

SNPs not in Hardy-Weinberg equilibrium included rs9610-IL10RA, rs730690-IL12B, rs3093661-TNF, rs40401-IL3, and rs1800471-TGFB1 for African-Americans and rs1126579-IL8RB and rs11914-IFNGR1 for Caucasians. An additional six SNPs had minor allele frequencies <5%, resulting in 70 of 83 SNPs meeting criteria for analyses.

For 52 of the 70 SNPs, allele frequencies differed significantly between African-American and Caucasian controls (data not shown). Consequently, all analyses were stratified on race. The minor allele for rs1800629-TNF was associated with a decreased risk for lung cancer only among African-Americans (Table 2). For two IL10 SNPs that were in linkage disequilibrium, rs3024496 and rs3024491, presence of the minor allele was associated with an increased risk of NSCLC among African-Americans but a decreased risk of lung cancer among Caucasians. rs2857261-IL15, rs1057972-IL15, rs1800797-IL6, and rs1800795-IL6 were associated with an increased risk among African-Americans, but no association was detected among Caucasians. rs20541-IL13 was only marginally associated with a decreased risk of lung cancer among African-American women. Two SNPs in IL7R, rs1494555 and rs7737000, were associated with an increased risk among Caucasians, but only rs7737000 was associated with risk in African-Americans. This association between IL7R polymorphisms and lung cancer risk was supported by the significance of the genetic risk score for IL7R, which was statistically significant even after correction for multiple comparisons among Caucasians (P = 0.002) and African-Americans (P = 0.04). Also statistically significant after adjustment for multiple testing among African-Americans was the genetic risk score for IL15 (P = 0.04).

There are some relationships between SNPs and risk for NSCLC by smoking history among Caucasians (Table 3). The TNF SNP, rs1800630, was associated with an increased risk of NSCLC among never smokers but not among smokers. Conversely, presence of the minor allele in SNPs associated with a decreased risk of NSCLC among never smokers but not among smokers included rs2857261-IL15, rs4871857-TNFRSF10A, rs3024496-IL10, and rs3024491-IL10. The minor allele for rs1494555-IL7R was associated with an increased risk of lung cancer among smokers but not among never smokers (P value_interaction = 0.002).

For eight SNPs (rs2071374-IL1A, rs1494555-IL7R, rs7737000-IL7R, rs1057972-IL15, rs2857261-IL15, rs1799964-TNF, rs1800630-TNF, rs4871857-TNFRSF10A), presence of the minor allele was associated with an increased risk of lung cancer among Caucasian women with COPD but not among those without a history of COPD (Table 4). Two IL1RN SNPs, rs454078 and rs419598, were associated with a decreased risk of lung cancer among Caucasian women with COPD but not among women with no history of COPD. The two polymorphisms that marginally increased risk of NSCLC among Caucasian women without COPD but not among women with COPD were rs2227306-IL8 and rs2296135-IL15RA.

When analyses were restricted to predicting risk of adenocarcinoma of the lung, results were similar (data not shown).

Mean decrease accuracy variable importance measures for Caucasian and African-American women indicate that the most important predictor of lung cancer is, not surprisingly, smoking pack-year history (Table 5). rs77371000-IL7R was identified as being an important predictor of lung cancer among African-American women by both random forest analysis and logistic regression, and rs1494555-IL7R was identified by both methods as being an important predictor among Caucasian women. The second most important polymorphisms identified on random forest but not logistic regression among Caucasian and African-American women, respectively, were rs17561-IL1A and rs4871857-TNFRSF10A.

Haplotypes in IL1B, IL7R, IFNAR2, and IL13 predicted lung cancer risk differentially by self-reported race. However, the significance of each of these haplotypes determined by PLINK was driven by a single SNP (data not shown). Subsequently, haplotype analysis provided no additional risk information beyond that obtained from the single SNP analysis.

Of the 12 nonsynonymous polymorphisms included in this study, predictions of the effect of amino acid substitutions on protein function were obtained for eight SNPs (rs17561, rs1494555, rs1801275, rs1805011, rs1805012, rs1805015, rs1805016, rs4871857). No prediction was made for two of the polymorphisms, and two were not found in the database. All eight of the SNPs analyzed through the SIFT program had prediction scores >0.10 and median sequence information content values <3.00.

Of the 26 cytokines and cytokine receptors analyzed, all were incorporated into an inflammation network in addition to two other cytokines not analyzed in this study, IFN-α and IFN-β. When only interactions between cytokines expressed in human lung and known to play a role in cancer were displayed (Fig. 1), 18 cytokines and cytokine receptors remained, including IL-7R and IL-10. SNPs in these two genes were most consistently shown to be associated with NSCLC on logistic regression and random forest analyses. Reported interactions between IL-7R and TNF-α and IFN-γ and between IL-10 and IL-1α, IL-6, IL-8, IFN-γ, TNF-α, and TGF-β1 are presented.

We examined the relationship between NSCLC and 70 cytokine and cytokine receptor polymorphisms in 26 genes. Variant allele frequencies differed significantly between African-Americans and Caucasians for 74% of the SNPs included. SNPs in IL7R and IL10 were associated with increased risk of lung cancer in both African-Americans and Caucasians on logistic regression analysis. An IL10 SNP, rs3024491, was associated with an increased risk of lung cancer among African-Americans and a decreased risk among Caucasians. IL7R and IL7R and IL10 were important predictors of lung cancer on random forest among Caucasians and African-Americans, respectively. Other SNPs associated with risk of NSCLC on logistic regression among African-Americans included those in IL6, IL15, TNF, and IL13. Polymorphisms in TNF, TNFRSF10A, IL10, and IL15 were associated with risk of lung cancer among Caucasian never smokers and in IL7R among Caucasian ever smokers. When these associations were analyzed separately by history of COPD, single-nucleotide polymorphisms in IL1A, IL7R, IL15, TNF, TNFRSF10A, and IL1RN were associated with NSCLC risk in women with COPD and IL8 in women without a history of COPD. Our study is the first to analyze the association between NSCLC risk and IL7R, IL15, IL15RA, TNFRSF10A, and TNFRSF1A polymorphisms.

Distinct patterns of associations between NSCLC risk and cytokine polymorphisms were observed for Caucasians and African-Americans as evident in the relationships between IL6, IL15, TNF, and IL13 SNPs and risk of NSCLC among African-Americans but not Caucasians on logistic regression analysis. The IL10 polymorphism, rs3024491, was associated with an increased risk of NSCLC among African-Americans but a decreased risk of NSCLC among Caucasians on logistic regression analysis. These data suggest that the relationships between inflammatory pathway genes and lung cancer risk potentially differ between African-American and Caucasian women. Alternatively, particular SNPs included may not be causal but may be in linkage disequilibrium with other genetic changes that represent the observed association. This linkage disequilibrium may vary by race and account for the observed associations.

SNP associations conserved across both racial groups included those for IL7R and IL10 polymorphisms. IL-7 and IL-10 play critical roles in immune modulation and have been implicated in tumorigenesis. IL-7 is a proinflammatory cytokine involved in growth of pre–B lymphocytes and pre–T lymphocytes, is elevated in patients with T cell depletion, activates macrophages, up-regulates expression of the proangiogenic IL-8, and aids in T-lymphocyte survival. IL-7R expression has been shown in neoplastic lung cell lines (31), and IL-7 systemic therapy has been shown to increase antitumor T-cell lytic activity in mice bearing NSCLC tumors (32). One potential mechanistic link among smoking, IL-7, and lung cancer may be the up-regulation of IL-8 expression by IL-7 (33).

A key anti-inflammatory cytokine, IL-10, is produced by monocytes, tumor associated macrophages, and T_helper type 2 (T_H2) cells; acts chiefly to down-regulate expression of proinflammatory cytokines, including TNF and TNF receptors; and inhibits the antitumor cytotoxic action of macrophages. However, it has also been shown to stimulate immune activity (34). Interestingly, murine models indicate that female mice, but not male mice, hemizygous for IL10 were more susceptible to urethane induced lung adenocarcinomas (35). At least one review supports an antitumorigenic function of IL-10 and disputes the dogma that IL-10, as an anti-inflammatory cytokine, serves to suppress immunosurveillance (34). However, a study involving 154 NSCLC cases and 205 controls indicated that alleles for three IL10 promoter SNPs associated with increased IL-10 production were found more frequently among lung cancer cases than controls (18). These contradictory findings between preclinical and clinical studies highlight the complexity of the role of cytokines in tumorigenesis.

Of particular interest are the genes for which polymorphisms were associated with an increased risk of lung cancer differentially among never smokers (TNF) and among smokers (IL7R and IL15). TNF-α is produced by macrophages, CD4+ T cells, and natural killer cells in response to IL-2 and IFNs, as well as by tumor cells. TNF-α binds to TNFRSF1A and TNFRSF10A receptors, resulting in programmed cell death. Alternatively, lung cancer cells have been shown to be resistant to TNF-α–induced death through activation of the transcription factor NFκB (36). Whereas some studies have shown no effect of cigarette smoke exposure on bronchoalveolar lavage TNF-α supernatant levels (37) or in exhaled breath condensate (38), others have observed lower levels of TNF-α in bronchoalveolar lavage fluid (39) and by peripheral blood mononuclear cells from smokers in vitro (40). Similar to the latter findings, the rs1800630-TNF A allele, which we observed to be associated with an increased risk of NSCLC among never smokers, has been shown to be associated with decreased TNF-α expression (41). If these findings are replicated, TNF may serve as a gene that increases a never smoker's risk of lung cancer by mimicking the cytokine profile in a smoker's lung. Thus, TNF death receptor and IL7R genes need to be further studied for roles in the development of lung cancer among never smokers and ever smokers, respectively.

Structurally similar to IL-2, IL-15 facilitates CD8+ memory T-lymphocyte survival and stimulates proliferation of natural killer cells, dependent on IL-15RA, in the tumor environment (42, 43). Adjuvant IL-15 therapy has been shown to result in tumor regression in a murine lung tumor model (44), underscoring the importance of examining polymorphisms in the IL15 gene. Among the Caucasian women in our sample, rs2857261-IL15 was associated with a decreased risk of NSCLC among never smokers and an increased risk among smokers, highlighting a potentially important interaction between IL15 and the impact of cigarette smoke.

The relationships between cytokines and the development of lung cancer among people with and without COPD has yet to be elucidated. Our study revealed separate cytokine polymorphism signatures associated with risk of NSCLC among Caucasian women with COPD and those without COPD, even after taking into account smoking history. The increased risk of NSCLC among women without a history of COPD who carry the A allele for rs2227306-IL8, which is associated with increased IL-8 production, is expected because IL-8 serves as a chemotactic and powerful proangiogenic factor. Furthermore, the protective effect of polymorphisms in IL1RN, which codes for IL-1 receptor antagonist, among COPD patients may be related to suppression of IL-1–induced inflammation with cigarette smoke exposure; however, the effect of these two SNPs on expression remains to be determined. Although the association between TNF-α and COPD development has been well documented and anti-TNF therapies are being tested in COPD patients (45-48), we present here the association between TNF polymorphisms and risk of NSCLC among COPD patients but not among non–COPD patients. For both TNF promoter polymorphisms, the rs1799964 G allele and the rs1800630 A allele have been associated with increased expression, and both increased and decreased expression, respectively (49-51), illustrating the complexities of regulation of genetic expression. Whether the association between these SNPs and genetic expression alters the risk of NSCLC among COPD patients remains to be determined. One potential mechanism involves increased TNF-α activating NFκB and subsequently contributing to a proinflammatory environment associated with tumorigenesis. In addition, the polymorphisms in IL1A, IL7R, IL15, and TNFRSF10A associated with increased risk of NSCLC among COPD patients suggest that the role of these cytokines and cytokine receptors needs to be studied in terms of lung tumorigenesis among COPD patients.

Comparisons between logistic regression and random forest have been studied extensively for classification purposes (52-54). In general, more sophisticated statistical methods yield better classification. In our study, we used these two methods to explore risk factors rather than to classify outcomes. Logistic regression is simple and straightforward, but SNPs are considered one at a time. Random forest is a suitable complement to the logistic regression method because the importance of a SNP was calculated, given other SNPs and risk factors. Given the different statistical approaches of these methods (one is linear and the other is nonlinear with a bagging algorithm), it is not unexpected that the results differ to some degree.

In our study, IL7R polymorphisms were identified in both racial groups as being predictive of lung cancer risk on logistic regression and random forest analysis. Furthermore, identification of IL10 polymorphisms as predictive of NSCLC risk among African-American women was a consistent finding on logistic regression and random forest analysis. The remainder of the results differed between the two analytic methods. Although the findings were not statistically significant after adjusting for multiple comparisons on logistic regression analysis, IL7R and IL10 polymorphisms showed the strongest relationship with lung cancer risk using both statistical approaches.

A few key cytokine genes with polymorphisms identified on random forest analysis but not on logistic regression analysis included rs17561-IL1A and rs16944-IL1B among Caucasian women and rs4871857-TNFRSF10A among African-American women. At least one other study involving non-Hispanic Caucasians found an 18% increased risk of lung cancer for rs17561-IL1A in a dominant model [odds ratio, 1.18; 95% confidence interval (95% CI), 1.03-1.36; P = 0.02], especially among heavy smokers and participants with a history of emphysema (20). Although this particular SNP is not presumed to have an impact on IL-1α function (SIFT score, 0.81; median sequence information content, 2.59), it may be in linkage disequilibrium with another polymorphism that affects IL-1 activity. We did not observe an association between rs16944-IL1B and lung cancer risk among either racial group on single SNP logistic regression analysis; however, relationships between IL1B polymorphisms and haplotypes and lung cancer have been previously reported among Caucasian and Japanese populations (16, 17, 20). Also known as lymphocyte activating factor, IL-1 is produced in two forms (IL-1α and IL-1β) primarily by monocytes and stimulates T helper cells, B-lymphocyte proliferation, and neutrophil chemotaxis. In addition, it is involved in the production of prostaglandin E₂, colony stimulating factor, IL-6, and IL-8, which plays a role in the pathogenesis of cancer and chronic inflammatory conditions such as COPD (46, 55).

Although rs4871857-TNFRSF10A was not associated with NSCLC risk among African-Americans on logistic regression analysis, it was identified by random forest as an important predictor. TNFRSF10A codes for the DR4 TNF receptor, also referred to as “the death receptor.” The DR4 receptor is vital to induction of lung tumor cell death, but not normal cell death, by TNF-related apoptosis–inducing ligand (56) and was found to be expressed in almost 100% of NSCLC tumors in one study (57). These findings suggest that the TNF receptors may serve to reduce risk of clinically detectable lung cancer by keeping tumor cells in check. Although the rs4871857 polymorphism lies in a coding region for the ectodomain of DR4, it is not projected to have an impact on protein function (SIFT score, 0.73; median sequence information content, 1.84); however, further research into the role of TNFRSF10A genotypes in determining DR4 activity in the tumor microenvironment should be conducted.

Combined with the marginal results on logistic regression analysis, the low mean decrease accuracy for all of the SNPs, except for IL7R on random forest analysis, suggests that no single gene in the inflammatory pathway predominates in importance in predicting lung cancer risk. Rather, it is likely that the concert of cytokine signaling combined with other signaling pathways has an impact on lung cancer development.

The inflammation network identified through pathway analysis emphasizes the proinflammatory role of TNF-α in up-regulation of IL6, IL7, IL8, IL15, IL15RA, and IFNAR2 expression and/or activation and the anti-inflammatory role of IL-10 through its suppression of IL1A, IL6, IL8, and IFNG production. Mutual inhibition between proinflammatory and anti-inflammatory cytokines and cytokine receptors is evident through this immunosuppressive effect of IL-10. Further evidence lies in the inhibition of TGF-β1 by TNF-α, the relationship between IL-10 and decreased production of TNF-α, and the inhibition of IFN-γ production by TGF-β1. The question then arises how a proinflammatory state induced by exposures to lung irritants such as cigarette smoke can lead to the development of a lung tumor microenvironment where immunosuppression predominates. The answer may lie in the stimulatory effects of TNF-α, the production of which is stimulated by chemicals in cigarette smoke, on NFκB expression by epithelial cells, IL-8 expression, and IL-10 activation. A chemokine that is elevated in smokers' lungs, IL-8, draws immune system cells to the site at which it is produced and facilitates angiogenesis through up-regulation of vascular endothelial growth factor expression (58, 59). Thus, a chronic proinflammatory state created by chronic cigarette smoke exposure, driven by TNF-α/NFκB positive feedback loop and monocyte chemotaxis to the lung tissue, may also promote tumorigenesis through immunosuppression induced by IL-10, in addition to the contribution of TGF-β1 production by transformed cells.

Of particular interest are the connections in the inflammation network involving IL-7R, which we identified as being associated with NSCLC risk, and IFN-γ, which showed no relationship in our analyses. The interactions between IL-7R and TNF-α, as well as IFN-γ, shown in the network support a proinflammatory role of IL-7R in tumorigenesis and in the human lung, emphasizing the importance of the association between IL7R polymorphisms and an increased risk of NSCLC. Although SNPs in IFNG and IL8 were not identified as being predictors of NSCLC on either logistic regression or random forest analysis, they have been shown to interact with cytokine genes identified in our analyses as being associated with NSCLC risk. IFN-γ is a key proinflammatory cytokine that activates macrophages, suppresses T_H2 cytokine activity, including that of IL-10, and inhibits IL-8 secretion by human lung carcinoma cells (60). Simultaneously, IFN-γ has been shown to decrease IL-7R protein production, suggesting another regulatory mechanism (61).

This study had a number of strengths. Only in-person interviews were conducted, increasing data validity and reliability. Extensive medical histories and environmental exposure histories were collected, permitting adjustment for potential confounders and analysis of SNP-smoking interactions and COPD strata–specific analyses. Because of the proportion of African-Americans included in this study, we were able to conduct race-specific analyses. Moreover, SNPs selected for these analyses were included in the Cancer 500 Panel from the National Cancer Institute's Cancer Genome Anatomy Project, allowing for comparison with future work involving this panel by other investigators. In addition, random forest and pathway analyses were used as alternative tools for data analysis.

This study is not without limitations. The functional consequences of a number of SNPs included in our analyses are as yet unknown. It is possible that the SNPs found to be associated with NSCLC risk in this study are in linkage disequilibrium with other SNPs, not included in our study, which have a functional consequence and are associated with risk of lung cancer. Because only in-person interviews were conducted, some women were too ill to participate, potentially creating a bias toward including cases with less aggressive forms of NSCLC as is evident in the greater proportion of nonparticipating cases diagnosed at a distant stage and with a shorter survival time. Although our study included ~20% African-Americans, the number of African-American participants was still too small to stratify analyses on race and smoking history, or history of COPD with adequate power. History of COPD was determined by patient report of a physician diagnosis of emphysema, chronic bronchitis, or COPD and not by objective measures. Subsequently, the possibility exists that some participants were misclassified according to history of COPD. This study included largely adenocarcinoma cases; therefore, the results may not be generalizable to all lung cancers. Whereas an estimate of overfitting for the model was 3% for Caucasians, it was 12% for African-Americans, suggesting that the number of African-Americans included in this study might have been too small for the number of parameters included in the logistic regression model. A bootstrap estimate of Somers' D_xy, a measure of model validity, was 0.68 for Caucasians and 0.63 for African-Americans, indicating that model fit was fairly comparable for Caucasians and African-Americans. With respect to random forest analysis, the out of bag error rates for Caucasian and African-American women in this sample are 22% and 30%, respectively, further underscoring the need to conduct this study using a larger sample of African-American women. Finally, multiple comparisons were made, and some significant findings may be due to chance. However, the analysis was directed toward inflammatory pathway genes a priori, and adjusted P values are presented. Although these adjusted P values are not statistically significant, except for the association between IL7R and IL15 genetic risk scores and NSCLC risk, the false discovery rate method is an overcorrection because it assumes independent tests of association; this assumption is violated, given linkage disequilibrium between some SNPs.

Although cigarette smoking has been shown to increase lung expression of IL-1β, IL-6, TNF-α, IL-4, and IL-8, we identified IL-7 and IL-15 as additional cytokines, which may confer an increased risk of lung cancer. Moreover, the increased risk associated with IL7R and IL15 SNPs among Caucasian smokers, but not among never smokers, suggests that further research needs to be conducted about the interaction between cigarette smoke exposure and IL-7 and IL-15 expression in the prediction of lung cancer risk. Results also implicate IL1A, IL7R, IL15, TNF, and TNFRSF10A in the pathogenesis of lung cancer and IL1RN in the protection against developing lung cancer among COPD patients. The differential findings on logistic regression and random forest analyses by self-reported race indicate that further research about these cytokine SNPs in high-risk populations needs to be conducted by race. Finally, future integration of pathway analysis data with environmental risk factor data such as smoking behavior, demographic information, data on comorbidities, and disease status is necessary to understand how genetic networks in concert are modulated by other factors to affect disease risk.

No potential conflicts of interest were disclosed.

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.