Inflammatory Cytokines and Lung Cancer Risk in 3 Prospective Studies

Abstract

To further investigate the role of inflammation in lung carcinogenesis, we evaluated associations between proinflammatory cytokines and lung cancer risk. We conducted a case-control study nested within 3 prospective cohort studies—the Melbourne Collaborative Cohort Study (1990–1994), the Malmö Diet and Cancer Study (1991–1996), and the Northern Sweden Health and Disease Study (initiated in 1985)—involving 807 incident lung cancer cases and 807 smoking-matched controls. Conditional logistic regression models adjusting for serum cotinine concentrations were used to estimate odds ratios for lung cancer risk associated with concentrations of interleukin (IL)-1β, IL-2, IL-6, IL-8, IL-10, IL-12, interferon γ, tumor necrosis factor α, and granulocyte-macrophage colony-stimulating factor. We observed a higher lung cancer risk for participants with elevated concentrations of IL-6 and IL-8. These associations seemed to be stronger among former smokers (for fourth quartile vs. first quartile, odds ratio (OR) = 2.70, 95% confidence interval (CI): 1.55, 4.70) and current smokers (OR = 1.99, 95% CI: 1.15, 3.44) for IL-6 and among former smokers (OR = 2.83, 95% CI: 1.18, 6.75) and current smokers (OR = 1.30, 95% CI: 0.69, 2.44) for IL-8. No notable associations were observed among never smokers. Risk associations with IL-6 and IL-8 were observed for blood samples taken close to diagnosis (<5 years) as well as more than 15 years postdiagnosis.

cohort studies, cytokines, inflammation, lung neoplasms, risk factors

Lung cancer is the leading cause of cancer-related mortality globally (1). Overall 5-year survival remains low, reported to be approximately 15% in Western populations. While tobacco exposure accounts for the vast majority of lung cancer cases, it is of interest to identify pathways that mediate differences in risk.

Several lines of evidence suggest an etiological role for inflammation in the development and subsequent progression of lung cancer (2, 3). Proinflammatory cytokines can be released following stimulus by several factors, including infections, physical stimuli, chemical irritation, and/or chronic insult, and are believed to play an etiological role in subsequent neoplastic development. The cytokines are produced largely by T-cells and monocytes/macrophages and can elicit a myriad of downstream effects (4). Consequential increased angiogenesis, increased DNA adduct formation, and DNA damage from reactive species which are the resulting biproducts of the natural inflammatory response of the innate immune system are believed to be among the principal mechanisms (5).

Several epidemiologic studies have provided results that are consistent with a positive association between circulating concentrations of proinflammatory cytokines and risk of subsequent lung cancer (6, 7). For example, in the Prostate, Lung, Colorectal, and Ovarian (PLCO) Cancer Screening Trial, Pine et al. (7) reported that increased interleukin (IL)-6 concentrations were associated with higher lung cancer risk, but only for cases diagnosed within 2 years of blood collection, whereas increased IL-8 concentrations were associated with lung cancer diagnosed more than 2 years after blood collection. In a second analysis using a separate prospective case-control series from the PLCO, Shiels et al. (8) evaluated a broader panel of 68 individual inflammation/immune biomarkers and found several additional markers, including C-reactive protein, B-cell-attracting chemokine 1 (BCA-1)/chemokine (C-X-C motif) ligand 13 (CXCL13), macrophage-derived chemokine (MDC)/C-C motif chemokine 22 (CCL22), and IL-1 receptor antagonist, to be associated with lung cancer risk, although they did not confirm the associations with IL-6 and IL-8 noted in the previous PLCO study (7). These associations will require additional evaluation via prospective studies to clarify the evidence with respect to the time period prior to diagnosis in which these markers manifest, whether they are relevant regardless of smoking history, and whether observed associations reflect a causal role for inflammation in lung cancer etiology (9).

We aimed to address these questions by analyzing 807 cases and 807 controls from 3 cohort studies: the Melbourne Collaborative Cohort Study (MCCS), the Northern Sweden Health and Disease Study (NSHDS), and the Malmö Diet and Cancer Study (MDCS).

METHODS

Study populations

Melbourne Collaborative Cohort Study

Extensive details on recruitment and follow-up procedures within the MCCS have been published previously (10, 11). Briefly, the MCCS is a prospective cohort study of 41,514 women and men aged 40–79 years at the time of recruitment, which took place between 1990 and 1994 in Melbourne, Victoria, Australia. A total of 361 incident lung cancer cases and 361 individually matched controls were available for this analysis.

Northern Sweden Health and Disease Study

The NSHDS encompasses cohorts from several prospective studies. The current analysis involved participants from the Västerbotten Intervention Project (VIP), a subcohort within the NSHDS. The VIP is an ongoing prospective cohort and intervention study of health promotion in the general population of Västerbotten County, northern Sweden (12). The VIP was initiated in 1985, and all residents of Västerbotten County were invited to participate by attending health check-ups (1 per decade) at 40, 50, and 60 years of age. Participants were asked to complete a self-administered questionnaire including questions on various demographic factors such as education, smoking habits, physical activity, and diet. In addition, height and weight were measured and participants were asked to donate a fasting blood sample for future research. A total of 245 incident lung cancer cases and 245 individually matched controls were available for this analysis.

Malmö Diet and Cancer Study

The MDCS is a population-based prospective cohort study that between 1991 and 1996 recruited men and women aged 44–74 years living in Malmö, Sweden (13). The main goal of the MDCS is to examine the associations between diet and cancer incidence and mortality. Study measurements comprised a baseline examination including dietary assessment, a self-administered questionnaire, anthropometric measurements, and collection of blood samples. A total of 201 incident lung cancer cases and 201 individually matched controls were available for this analysis.

Lung cancer cases and control selection

For all 3 cohorts, lung cancer cases were defined on the basis of the International Classification of Diseases for Oncology, Second Edition, and included primary malignant cancers coded as C34.0–C34.9 with prediagnostic blood samples.

One control was chosen at random for each lung cancer case from appropriate risk sets consisting of all cohort members who were alive and free of cancer (except nonmelanoma skin cancer) at the time of diagnosis of the index case. Matching criteria included: date of birth (±1 year, relaxed up to ±5 years for cases without available controls); ethnicity; sex; date of blood collection (±1 month, relaxed up to ±3 months and further to ±6 months for cases without available controls); and detailed information on smoking status (never smokers, short-term former smokers (quit smoking <10 years before blood draw), long-term former smokers (quit smoking ≥10 years before blood draw), current light smokers (<15 cigarettes/day at blood draw), and current heavy smokers (≥15 cigarettes per day at blood draw)). The total combined sample size from the 3 cohorts was 1,614 (807 cases and 807 controls).

Plasma collection and measurement of inflammatory cytokines

Following blood draw, plasma fractions were stored at temperatures below −80°C. Cytokine concentrations (pg/mL) were measured at the clinical chemistry laboratory in Umeå, Sweden. We used the Human Pro-Inflammatory 9-Plex Tissue Culture Kit (K1500C; Meso Scale Discovery, Gaithersburg, Maryland), which measured levels of IL-1β, IL-2, IL-6, IL-8, IL-10, IL-12p70, tumor necrosis factor α (TNF-α), granulocyte-macrophage colony-stimulating factor (GM-CSF), and interferon γ (IFN-γ). All matched case-control pairs were analyzed in the same batch, in adjacent wells, and in random order. Fewer than 10% of samples were below the limit of detection for any of the evaluated cytokines. In instances where a value below the limit of detection was provided, the limit of detection was used in analyses. All but 2 cytokines (IL-1β and IL-12p70) showed good technical reproducibility when we evaluated marker correlations between pairs of pooled samples within each batch (see Web Table 1, available at http://aje.oxfordjournals.org/). Due to low reproducibility, we chose to remove these 2 cytokines from subsequent analyses and presentation of the results.

Statistical analyses

The distributions of cytokine concentrations were examined, and quartiles of exposure were created for risk analyses using values among all controls. Conditional logistic regression models, conditioning on the individually matched case sets, were used to estimate odds ratios and 95% confidence intervals for cancer risk by quartiles of circulating cytokine concentrations. In all models, the lowest quartile was used as the reference group. The risk models further adjusted for educational level (6 categories), body mass index (weight (kg)/height (m)²; continuous), total dietary energy intake (kcal/day; continuous), and serum cotinine concentration (nmol/L; continuous). We conducted analyses stratified by sex and smoking category to investigate potential interactions. In order to evaluate the impact of time from blood draw to diagnosis (lead time), we also conducted risk analyses stratified by 4 categories of lead time (<5 years, 5–9.9 years, 10–14.9 years, and ≥15 years). Finally, we evaluated the associations with lung cancer risk separately across the major histological subtypes, including adenocarcinoma, squamous cell carcinoma, and small-cell lung cancer. In order to examine linear associations between the cytokines and lung cancer risk, we fitted separate models containing log-linear transformations of the cytokine variables, as the concentration variables tended to be right- (positively) skewed. Analyses were conducted using SAS 9.3 (SAS Institute, Inc., Cary, North Carolina) and R, version 3.1.2 (R Foundation for Statistical Computing, Vienna, Austria).

RESULTS

Baseline characteristics of the study sample are presented in Table 1. Because of the study design, sex and smoking status were evenly distributed across cases and controls, with 47.0% of participants being current smokers, 36.8% being former smokers, and 16.2% being never smokers at baseline in the combined sample. The relatively large fraction of never-smoking cases was driven mostly by the MDCS, wherein never-smoking cases were oversampled by design in the context of a parallel study using the same study population. We observed significant differences in mean cytokine concentrations across cohorts (Web Table 2). These differences are potentially related to the difference in proportions of smokers across studies, as we also observed significant differences in cytokine concentrations across smoking groups (Web Table 3).

Table 1.

Demographic Characteristics of Lung Cancer Cases and Matched Controls From 3 Cohort Studies (the Melbourne Collaborative Cohort Study (Melbourne, Australia, 1990–1994), the Northern Sweden Health and Disease Study (Västerbotten County, Sweden, 1985), and the Malmö Diet and Cancer Study (Malmö, Sweden, 1991–1996)) Included in an Analysis of Inflammatory Cytokines and Lung Cancer Risk

Variable	Controls (n = 807)			Cases (n = 807)
	No.	%	Mean (SD)	No.	%	Mean (SD)
Melbourne Collaborative Cohort Studya
Sex
Male	218	60.4		218	60.4
Female	143	39.6		143	39.6
Smoking status at recruitment
Never smoker	50	13.9		50	13.9
Former smoker	148	41.1		148	41.1
Current smoker	162	45.0		162	45.0
Educational level
No secondary education		0.0			0.0
Secondary education	238	65.9		241	66.9
Some postsecondary education	38	10.5		43	11.9
Vocational training	33	9.1		23	6.4
University education	9	2.5		8	2.2
Graduate education	43	11.9		45	12.5
Cotinine level, nmol/L			573.7 (757.7)			698.5 (853.5)
Age at recruitment, years			59.1 (6.9)			59.1 (6.83)
Body mass indexb			27.8 (4.8)			26.56 (4.04)
Total energy intake, kcal/day			2,348.0 (1,064.1)			2,254.8 (1,162.1)
Histological type
Squamous cell carcinoma				67
Adenocarcinoma				157
Small-cell carcinoma				47
Large-cell carcinoma				30
Other				57
Missing data				58
Malmö Diet and Cancer Studyc
Sex
Male	89	44.3		89	44.3
Female	112	55.7		112	55.7
Smoking status at recruitment
Never smoker	51	25.4		51	25.4
Former smoker	72	35.8		72	35.8
Current smoker	78	38.8		78	38.8
Educational level
No secondary education	2	1.0		1	0.5
Secondary education	90	45.2		110	54.2
Some postsecondary education	51	25.6		40	19.7
Vocational training	14	7.0		15	7.4
University education	42	21.1		37	18.2
Graduate education		0.0			0.0
Cotinine level, nmol/L			517.0 (759.7)			704.8 (880.5)
Age at recruitment, years			60.7 (6.85)			60.8 (6.8)
Body mass index			25.06 (3.64)			25.02 (4.16)
Total energy intake, kcal/day			2,206.2 (644.28)			2,215.9 (700.4)
Histological type
Squamous cell carcinoma				25
Adenocarcinoma				56
Small-cell carcinoma				12
Large-cell carcinoma				1
Other				49
Missing data				3
Northern Sweden Health and Disease Studyc
Sex
Male	124	50.6		124	50.6
Female	121	49.4		121	49.4
Smoking status at recruitment
Never smoker	30	12.2		30	12.2
Former smoker	76	31.0		76	31.0
Current smoker	139	56.7		139	56.7
Educational level
No secondary education		0.0			0.0
Secondary education	110	45.3		111	45.7
Some postsecondary education	35	14.4		31	12.8
Vocational training	72	29.6		64	26.3
University education	26	10.7		37	15.2
Graduate education		0.0			0.0
Cotinine level, nmol/L			712.0 (579.8)			781.4 (630.9)
Age at recruitment, years			55.0 (6.7)			55.1 (6.7)
Body mass index			26.4 (4.2)			26.1 (4.5)
Total energy intake, kcal/day			1,760.9 (723.6)			1,661.7 (635.15)
Histological type
Squamous cell carcinoma				49
Adenocarcinoma				118
Small-cell carcinoma				31
Large-cell carcinoma				6
Other				38
Missing data				1

Abbreviation: SD, standard deviation.

^a Average follow-up time, 9.1 years.

^b Weight (kg)/height (m)².

^c Average follow-up time, 9.6 years.

Cytokine concentrations and lung cancer risk

In the overall risk analysis, we observed risk increases for subjects with elevated concentrations of IL-6 and IL-8; the odds ratio comparing the fourth (top) quartile with the first (bottom) quartile was 1.82 (95% confidence interval (CI): 1.32, 2.50) for IL-6 and 1.48 (95% CI: 0.96, 2.28) for IL-8. These associations remained evident after inclusion of additional covariates in the logistic regression models (Web Table 4): The odds ratio was 2.00 (95% CI: 1.42, 2.82) for IL-6 and 1.57 (95% CI: 1.00, 2.46) for IL-8. We hypothesized that the relationship between the cytokines and lung cancer risk might differ between never, former, and current smokers. Stratifying results by smoking status, we noted that IL-6 was positively associated with risk for former smokers (odds ratio for fourth quartile vs. first (OR_4vs1) = 2.70, 95% CI: 1.55, 4.70) and current smokers (OR_4vs1 = 1.99, 95% CI: 1.15, 3.44) but not for never smokers (OR_4vs1 = 1.16, 95% CI: 0.42, 3.19) (Table 2). IL-8 was similarly associated with risk for former smokers (OR_4vs1 = 2.83, 95% CI: 1.18, 6.75) but not for current smokers (OR_4vs1 = 1.30, 95% CI: 0.69, 2.44) or never smokers (OR_4vs1 = 1.00, 95% CI: 0.32, 3.48) (Table 2). No notable association with risk was observed for the other evaluated cytokine measures overall (Web Table 4).

Table 2.

Estimated Risk of Lung Cancer According to Quartile of Circulating Cytokine Concentration (pg/mL), by Smoking Status at Cohort Baseline, Among 807 Cases and 807 Individually Matched Controls^a From 3 Cohort Studies (the Melbourne Collaborative Cohort Study (Melbourne, Australia, 1990–1994), the Northern Sweden Health and Disease Study (Västerbotten County, Sweden, 1985), and the Malmö Diet and Cancer Study (Malmö, Sweden, 1991–1996))

Cytokine and Quartile	Never Smokers					Former Smokers					Current Smokers					P-Interactionb
	No. of Controls	No. of Cases	ORc	95% CI	P-Value	No. of Controls	No. of Cases	ORc	95% CI	P-Value	No. of Controls	No. of Cases	ORc	95% CI	P-Value
IL-2
Q1	20	27	1.00	Referent		59	66	1.00	Referent		64	74	1.00	Referent
Q2	24	23	0.98	0.37, 2.61	0.96	58	54	0.89	0.48, 1.64	0.70	68	51	0.68	0.35, 1.31	0.25
Q3	23	26	0.99	0.28, 3.53	0.99	60	53	0.76	0.40, 1.44	0.40	64	69	0.90	0.48, 1.68	0.74
Q4	28	29	0.45	0.11, 1.76	0.25	46	68	1.38	0.75, 2.53	0.31	76	79	0.96	0.50, 1.83	0.89
P-lineard	95	105			0.60	223	241			0.48	272	273			0.58	0.56
IL-6
Q1	40	45	1.00	Referent		88	62	1.00	Referent		58	45	1.00	Referent
Q2	38	40	0.91	0.46, 1.82	0.79	53	66	2.18	1.23, 3.86	0.01	100	86	1.26	0.74, 2.15	0.39
Q3	30	23	0.53	0.23, 1.21	0.13	70	64	1.70	0.96, 3.00	0.07	94	86	1.37	0.81, 2.31	0.23
Q4	16	15	1.16	0.42, 3.19	0.77	66	94	2.70	1.55, 4.70	0.0005	112	147	1.99	1.15, 3.44	0.01
P-linear	124	123			0.99	277	286			0.001	364	364			<0.0001	0.06
IL-8
Q1	25	35	1.00	Referent		65	54	1.00	Referent		90	79	1.00	Referent
Q2	43	32	0.38	0.17, 1.12	0.04	56	63	1.92	0.99, 3.71	0.05	88	82	1.09	0.68, 1.73	0.73
Q3	29	25	0.50	0.19, 1.39	0.17	67	64	2.21	1.03, 4.72	0.04	91	101	1.41	0.84, 2.35	0.19
Q4	21	26	1.00	0.32, 3.48	1.00	78	93	2.83	1.18, 6.75	0.02	88	93	1.30	0.69, 2.44	0.42
P-linear	118	118			0.94	266	274			0.22	357	355			0.43	0.85
IL-10
Q1	31	30	1.00	Referent		75	68	1.00	Referent		81	71	1.00	Referent
Q2	39	34	0.75	0.34, 1.65	0.47	65	76	1.39	0.83, 2.33	0.21	87	99	1.29	0.79, 2.10	0.30
Q3	33	32	0.86	0.35, 2.12	0.75	76	67	0.95	0.55, 1.61	0.84	87	95	1.36	0.82, 2.25	0.23
Q4	21	27	1.52	0.60, 3.86	0.38	60	74	1.33	0.78, 2.29	0.29	109	100	0.99	0.58, 1.68	0.97
P-linear	124	123			0.96	276	285			0.20	364	365			0.50	0.94
TNF-α
Q1	29	31	1.00	Referent		57	60	1.00	Referent		87	88	1.00	Referent
Q2	25	24	0.97	0.41, 2.30	0.94	69	48	0.97	0.53, 1.76	0.92	81	65	0.82	0.49, 1.37	0.44
Q3	31	27	0.86	0.33, 2.28	0.76	57	66	1.48	0.79, 2.77	0.23	85	85	1.04	0.64, 1.71	0.86
Q4	31	27	1.16	0.44, 3.03	0.77	68	82	1.62	0.89, 2.95	0.11	75	82	1.10	0.65, 1.87	0.73
P-linear	116	109			0.95	251	256			0.10	328	320			0.11	0.98
IFN-γ
Q1	21	23	1.00	Referent		54	52	1.00	Referent		61	59	1.00	Referent
Q2	24	22	1.12	0.40, 3.10	0.83	49	41	0.96	0.47, 1.95	0.91	63	62	1.17	0.63, 2.17	0.63
Q3	20	17	0.82	0.24, 2.76	0.75	54	57	1.34	0.68, 2.63	0.39	58	61	1.14	0.57, 2.30	0.71
Q4	17	25	1.54	0.43, 5.45	0.50	47	61	1.40	0.71, 2.76	0.34	72	82	1.68	0.84, 3.37	0.15
P-linear	82	87			0.38	204	211			0.96	254	264			0.34	0.20
GM-CSF
Q1	36	44	1.00	Referent		60	62	1.00	Referent		94	87	1.00	Referent
Q2	35	25	0.73	0.35, 1.52	0.41	71	71	1.06	0.62, 1.79	0.84	81	97	1.31	0.84, 2.04	0.23
Q3	31	31	1.16	0.55, 2.41	0.70	68	65	0.97	0.54, 1.74	0.91	94	82	1.00	0.62, 1.61	0.99
Q4	22	24	1.38	0.56, 3.41	0.49	77	88	1.35	0.74, 2.46	0.33	96	99	1.19	0.72, 1.98	0.50
P-linear	76	76			0.64	191	191			0.41	249	249			0.30	0.75

Abbreviations: CI, confidence interval; GM-CSF, granulocyte-macrophage colony-stimulating factor; IFN-γ, interferon γ; IL, interleukin; OR, odds ratio; Q, quartile; TNF-α, tumor necrosis factor α.

^a Cases and controls were individually matched for age, sex, and smoking status.

^bP-interaction represents the P value from a test of multiplicative interaction (log-linear transformation of cytokine concentration × smoking).

^c Estimates were obtained from conditional logistic regression models and were adjusted for education (6 categories), body mass index (weight (kg)/height (m²); continuous), total dietary energy intake (kcal/day; continuous), and serum cotinine concentration (nmol/L; continuous).

^dP-linear represents the P value from a test of the log-linear transformation of cytokine concentration (pg/mL; continuous).

We observed no systematic differences in the risk associations for IL-6 or IL-8 based on the time from blood draw to lung cancer diagnosis. Positive associations were evident for IL-6, for example, both for lung cancer diagnosed within 5 years of blood draw (OR_4vs1 = 4.09, 95% CI: 1.78, 9.39) and lung cancer diagnosed over 15 years after blood draw (OR_4vs1 = 3.15, 95% CI: 0.93, 11.46) (Web Table 5).

When comparing associations by sex, increases in risk associated with IL-6 and IL-8 were observed only for males (Figures 1 and 2). Effect estimates by sex are presented in Web Table 6. Stratified analysis also suggested heterogeneous associations with risk between the 3 studies for both IL-6 (P-interaction = 0.05) and IL-8 (P-interaction = 0.04) (Web Table 7). However, these differences could not be readily explained by differences in baseline characteristics or in the case mix according to histological subtype.

Figure 1.

Estimated effect of interleukin-6 concentration (quartile 4 vs. quartile 1) on lung cancer risk by smoking status, sex, histological type, and follow-up time at cohort baseline among 807 cases and 807 individually matched controls within the Melbourne Collaborative Cohort Study (Melbourne, Australia, 1990–1994), the Northern Sweden Health and Disease Study (Västerbotten County, Sweden, 1985), and the Malmö Diet and Cancer Study (Malmö, Sweden, 1991–1996). Estimates were obtained from a conditional logistic regression model based on age-, sex-, and smoking-matched pairs, with adjustment for education (6 categories), body mass index (weight (kg)/height (m²); continuous), total dietary energy intake (kcal/day; continuous), and serum cotinine concentration (nmol/L; continuous). CI, confidence interval; OR, odds ratio.

Figure 2.

Estimated effect of interleukin-8 concentration (quartile 4 vs. quartile 1) on lung cancer risk by smoking status, sex, histological type, and follow-up time at cohort baseline among 807 cases and 807 individually matched controls within the Melbourne Collaborative Cohort Study (Melbourne, Australia, 1990–1994), the Northern Sweden Health and Disease Study (Västerbotten County, Sweden, 1985), and the Malmö Diet and Cancer Study (Malmö, Sweden, 1991–1996). Estimates were obtained from a conditional logistic regression model based on age-, sex-, and smoking-matched pairs, with adjustment for education (6 categories), body mass index (weight (kg)/height (m²); continuous), total dietary energy intake (kcal/day; continuous), and serum cotinine concentration (nmol/L; continuous). CI, confidence interval; OR, odds ratio.

Histology-specific risk analyses

Stratified risk analyses of IL-6 and IL-8 by histological subtype (Figures 1 and 2 and Web Table 8) indicated that associations with IL-6 appeared to be restricted to cases of squamous cell carcinoma (OR_4vs1 = 3.95, 95% CI: 1.43, 10.91) and small-cell carcinoma (OR_4vs1 = 3.18, 95% CI: 1.04, 9.72). Similar associations with both squamous cell carcinoma (OR_4vs1 = 2.78, 95% CI: 0.84, 9.20) and small-cell carcinoma (OR_4vs1 = 4.01, 95% CI: 0.56, 28.71) were seen for IL-8. In contrast, no association with adenocarcinoma was seen for either IL-6 or IL-8 (for IL-6, OR_4vs1 = 1.31 (95% CI: 0.76, 2.26); for IL-8, OR_4vs1 = 1.03 (95% CI: 0.50, 2.12)). In addition, we observed an increased risk of squamous cell carcinoma for persons in the fourth quartile of IFN-γ (OR_4vs1 = 3.88, 95% CI: 1.48, 10.22) (Web Table 8) and an increased risk of adenocarcinoma for those in the fourth quartile of GM-CSF (OR_4vs1 = 2.09, 95% CI: 1.01, 4.18) (Figure 3), GM-CSF being the only cytokine having an association with adenocarcinoma (Web Table 8).

Figure 3.

Estimated effect of granulocyte-macrophage colony-stimulating factor concentration (quartile 4 vs. quartile 1) on lung cancer risk by smoking status, sex, histological type, and follow-up time at cohort baseline among 807 cases and 807 individually matched controls from the Melbourne Collaborative Cohort Study (Melbourne, Australia, 1990–1994), the Northern Sweden Health and Disease Study (Västerbotten County, Sweden, 1985), and the Malmö Diet and Cancer Study (Malmö, Sweden, 1991–1996). Estimates were obtained from a conditional logistic regression model based on age-, sex-, smoking-matched pairs, with adjustment for education (6 categories), body mass index (weight (kg)/height (m²); continuous), total dietary energy intake (kcal/day; continuous), and serum cotinine concentration (nmol/L; continuous). CI, confidence interval; OR, odds ratio.

DISCUSSION

In this study, we investigated circulating levels of proinflammatory cytokines in relation to lung cancer risk using prediagnostic plasma samples from 3 prospective cohort studies. We observed an increased risk of lung cancer among research participants with elevated concentrations of IL-6 and IL-8, particularly for squamous and small-cell carcinomas and among former or current smokers.

It may be that these cytokines are markers of a high inflammatory burden due to tobacco consumption, and when we examined the association of baseline smoking with the concentrations of cytokines, we observed positive associations between IL-6 and IL-8 and smoking at baseline, as well as smoking duration and baseline smoking intensity. In order to further examine the potential for residual confounding due to tobacco use, we evaluated models with and without adjustment for circulating cotinine levels. We did not observe any substantial change in odds ratio estimates, suggesting that these biomarkers represent effects beyond what can be explained by differences in tobacco exposure.

Similar observations, in which current smoking status was associated with multiple inflammation markers, have been made previously. An example is an analysis from the PLCO Trial by Shiels et al. (14). Increased inflammation related to smoking is thought to be one of the mechanisms by which tobacco affects lung cancer risk, in addition to the direct genotoxic effects of tobacco-derived carcinogens. In a 2011 study, Liu et al. (15) showed that adult smokers had elevated concentrations of biomarkers of potential harm related to inflammation (white blood cell counts and high-sensitivity C-reactive protein), as well as an oxidative stress marker (8-epi-prostaglandin F_2α). Our observations for IL-6 and IL-8 are also consistent with 1 previous analysis of the PLCO study population (7), which found an increased risk of lung cancer for both IL-6 and IL-8, as well as with Il'yasova et al.’s results for IL-6 (16). Where our results differ is that we observed increased risk estimates more than 5 years from diagnosis for both IL-6 and IL-8, whereas in the PLCO Trial, only IL-8 was associated with risk of a lung cancer diagnosis more than 2 years after blood draw. Interestingly, in separate analyses of a panel of 68 inflammatory biomarkers used for a nested case-control study within the PLCO Trial, neither the association with IL-6 nor the association with IL-8 was apparent, possibly due to low detectability and poor reproducibility with use of the bead-based multiplexed panels in the second analysis (8).

From a mechanistic perspective, IL-6 and IL-8 may actively promote tumorigenesis by acting directly on lung epithelial cells via signaling through the nuclear factor κβ pathway under conditions of inflammatory stress (2, 17, 18). Accordingly, there would seem to be a motivation for investigating these cytokines further in the context of lung cancer etiology, presumably as factors that mediate some of the risk increase caused by tobacco smoking.

While investigators in the US National Lung Screening Trial reported that screening for early detection of lung cancer using low-dose computed tomography can reduce lung cancer-specific mortality by 20% (19), it was associated with a large number of false-positive screens requiring clinical follow-up. The use of detailed risk prediction models may inform selection of those subjects most likely to benefit from computed tomography screening, and risk biomarkers such as IL-6 and IL-8 may provide useful risk information in addition to questionnaire information on tobacco exposure history.

As Kovalchik et al. (20) demonstrated, the use of detailed risk models may provide important improvements in screening efficiency, and risk biomarkers may provide further information on risk over and above that provided by demographic information and questionnaire information on tobacco exposure history. As Shiels et al. (21) indicated, inflammation markers are unlikely to provide sufficient added risk information on their own, but in combination with other risk markers they may be useful for risk stratification in a screening scenario. Indeed, studies combining risk biomarkers from different domains are warranted to fully evaluate the potential of improving risk stratification through the use of biomarkers.

This study had several strengths. The results were based on a larger sample than those in previous prospective studies of inflammatory markers and lung cancer risk (7, 14, 16). Additionally, our study included a large number of never-smoking lung cancer cases (n = 132), which gave us reasonable statistical power to detect any important association between cytokines and increased risk in this category of relatively rare lung cancers. Another major strength of our analysis was the use of serum cotinine concentration, a sensitive and valid measure of current smoking intensity which may better account for recent tobacco exposure than questionnaire-based measures. Additionally, adjusting for pack-years did not notably alter the odds ratio estimates (data not shown). This affords extra confidence that the observed associations in current smokers reflect relationships with risk over and above that explained by differences in tobacco exposure.

This study did have some limitations. We analyzed concentrations of inflammatory markers from a single blood specimen collected at only 1 time point. Assessment of biomarker concentrations in a prospective cohort study at multiple time points prior to cancer diagnosis might help to further clarify the underlying mechanisms through which inflammatory biomarkers affect lung cancer risk. Our analysis also relied on the assumption that prediagnostic concentrations of circulating cytokines in lung cancer cases reflect pulmonary inflammation and immune response. In previous analyses, Shiels et al. (8) observed that levels of inflammation biomarkers such as C-reactive protein were elevated in control subjects with chronic obstructive pulmonary disease, an observation that provides some support for the use of circulating biomarker measures as indicators of tissue-specific inflammation (22).

In conclusion, this study further highlights the important role of inflammation in lung cancer etiology, particularly for squamous cell and small-cell carcinoma in ever smokers. Our findings also suggest that circulating levels of the proinflammatory cytokines IL-6 and IL-8 are elevated many years prior to lung cancer diagnosis.

ACKNOWLEDGMENTS

Author affiliations: International Agency for Research on Cancer, Lyon, France (Darren R. Brenner, Anouar Fanidi, David C. Muller, Paul Brennan, Graham Byrnes, Mattias Johansson); Department of Cancer Epidemiology and Prevention Research, CancerControl Alberta, Alberta Health Services, Calgary, Alberta, Canada (Darren R. Brenner); Department of Medical Biosciences, Faculty of Medicine, Umeå University, Umeå, Sweden (Kjell Grankvist); Department of Surgery, Skåne University Hospital Malmö, Lund University, Lund, Sweden (Jonas Manjer); Cancer Council Victoria, Melbourne, Victoria, Australia (Allison Hodge, Gianluca Severi, Graham G. Giles); Human Genetics Foundation, Torino, Italy (Gianluca Severi); Institut National de la Santé et de la Recherche Médicale, Centre for Research in Epidemiology and Population Health, U1018, Villejuif, France (Gianluca Severi); Unité Mixte de Recherche 1018, Université Paris Sud, Villejuif, France (Gianluca Severi); Institut Gustave Roussy, Villejuif, France (Gianluca Severi); Centre for Epidemiology and Biostatistics, Melbourne School of Population and Global Health, University of Melbourne, Melbourne, Victoria, Australia (Gianluca Severi, Graham G. Giles); and Department of Radiation Sciences, Faculty of Medicine, Umeå University, Umeå, Sweden (Mikael Johansson).

This work was funded by the Swedish Cancer Society (Principal Investigators: T. Rasmuson, M. Johansson). D.B. was supported by a Capacity Development Award in Cancer Prevention from the Canadian Cancer Society Research Institute (grant 703917).

Conflict of interest: none declared.

Author notes

Abbreviations: CI, confidence interval; GM-CSF, granulocyte-macrophage colony-stimulating factor; IFN-γ, interferon γ; IL, interleukin; MCCS, Melbourne Collaborative Cohort Study; MDCS, Malmö Diet and Cancer Study; NSHDS, Northern Sweden Health and Disease Study; OR, odds ratio; PLCO, Prostate, Lung, Colorectal, and Ovarian; TNF-α, tumor necrosis factor α; VIP, Västerbotten Intervention Project.