Urinary Metabolite Risk Biomarkers of Lung Cancer: A Prospective Cohort Study

Abstract

Background

Lung cancer is a major health burden causing 160,000 and 1.6 million deaths annually in the United State and worldwide, respectively.

Methods

Seeking to identify stable and reproducible biomarkers in non-invasively collected biofluids, we assessed whether previously identified metabolite urinary lung cancer biomarkers, creatine riboside (CR), N-acetylneuraminic acid (NANA), cortisol sulfate and indeterminate metabolite 561+ were elevated in the urines of subjects prior to lung cancer diagnosis in a well-characterized prospective Southern Community Cohort Study (SCCS). Urine was examined from 178 patients and 351 non-diseased controls, confirming that one of four metabolites was associated with lung cancer risk in the overall case-control set, whereas two metabolites were associated with lung cancer risk in European-Americans.

Results

Odds ratio of lung cancer associated with elevated CR levels, and adjusted for smoking and other potential confounders, was 2.0 (95%CI 1.2–3.4;P=0.01). In European-Americans, both CR and NANA were significantly associated with lung cancer risk (OR=5.3 (95%CI 1.6–17.6; P=0.006 and OR=3.5 (95%CI 1.5, 8.4;P=0.004), respectively). However, race itself did not significantly modify the associations. Receiver Operating Characteristic (ROC) analysis showed that adding CR and NANA to a model containing previously established lung cancer risk factors led to a significantly improved classifier (P=0.01). Increasing urinary levels of CR and NANA displayed a positive association with increasing tumor size, strengthening a previously established link to altered tumor metabolism.

Conclusion and Impact

These replicated results provide evidence that identified urinary metabolite biomarkers have a potential utility as non-invasive, clinical screening tools for early diagnosis of lung cancer.

Introduction

Lung cancer continues to be a significant burden in the United States and worldwide (1), with still over 40 million smokers in the US, approximately 23% of men and 19% of women ages 35 to 64, and even higher percentages among lower socioeconomic groups (2). A substantial challenge with this disease is that at present, more than 61% of patients are diagnosed in later stages III and IV, when therapeutic options are limited, and the 5-year survival is only 4%. In contrast, the 5-year survival of patients with early stage and localized lung cancer is about 50%, and is thus considerably improved compared to later stages (1). These exceptionally bleak statistics provide strong motivation to search for biomarkers that could aid currently established screening methods for early detection of lung cancer. To date, no such biomarkers are clinically utilized.

A seminal report by the National Lung Screening Trial (NLST) in 2011 showed for the first time a significant 20% reduction in lung cancer mortality from low dose CT (LDCT) screening when compared to the chest X-ray (3). Since then, there have been a number of recommendations for screening high-risk populations of ages 55–79 years, with a 30-pack-year history of smoking (4–7), triggering coverage by Medicare and private health insurers under provisions of the Affordable Care Act (8). However, the major problem with LDCT is a high false positive rate of 96%, which can lead to unnecessary downstream diagnostic procedures that may result in patient anxiety, higher healthcare spending, and increased mortality (3). Despite encouraging results of the NLST, there are concerns related to the cost-benefit of LDCT for lung cancer screening, risks of radiation exposure, over-diagnosis (9), and expected epidemic rise in indeterminate pulmonary nodules (IPN) diagnoses, only some of which present a higher risk for developing lung cancer (10). The use of complementary and accurate biomarkers would be a useful strategy to ameliorate most of these issues. However, most of the available molecular factors incorporated in diagnostic algorithms that guide therapy decisions are established by using invasively collected tissue specimens (11, 12). Non-invasively collected biospecimens, such as urine and blood, are gaining a significant interest as methods for measuring robust biomarkers of risk and prognosis (13–15).

While several circulating biomarkers have been evaluated in association with lung cancer risk (16–23), a large majority are tobacco-related carcinogens (24–27), which are markers of exposure and do not present targets for potential therapeutic interventions. Other studies have evaluated demographic and smoking-related behavioral information to aid in the selection criteria for lung cancer screening, much of which is self-reported (28). Nevertheless, in order to increase the specificity of LDCT, the ultimate goal is to identify stable, reproducible and non-invasively measured biomarkers. If said biomarkers are also products of deregulated tumor metabolism, they may have an ability to distinguish benign from malignant nodules. Despite intensive research and major advances in the field, there are still a number of challenges before biomarker panels can be used routinely in clinical practice, the most important of which is a necessity to validate them. The National Cancer Institute (NCI) previously created the Early Detection Research Network to promote biomarker discovery, validation, and translation into clinical practice (29). In order to fully assess and develop reliable biomarkers for monitoring asymptomatic patients for lung cancer, there is an urgent need to evaluate their utility in pre-diagnostic samples.

To that extent, we have evaluated a previously identified panel of urinary metabolite lung cancer biomarkers (30) in a well-characterized prospective Southern Community Cohort Study (SCCS): creatine riboside (CR), N-acetylneuraminic acid (NANA), cortisol sulfate and a glucuronidated, indeterminate metabolite referred to as 561+. Previously, novel CR and NANA were found to be over-expressed in tumor when compared to adjacent non-tumor tissue, with levels positively correlated between the tissue and urine (30). Therefore, if their association with lung cancer is validated, these metabolites present promising molecular markers for clinical evaluation as a method to aid LDCT, especially considering that metabolites are more proximal to disease phenotype than other ‘omics’ markers.

Materials and Methods

Study subjects

With recruitment completed in 2009, the Southern Community Cohort Study (SCCS) comprises a population of adults aged 40–79 who reside in the southeastern U.S. and belong to historically underrepresented groups in major cancer research studies (e.g. African-Americans) (31), with a high prevalence of behaviors associated with increased disease risk (e.g. smoking). Key strengths of this cohort are that subjects come from similar socioeconomic backgrounds (lower compared to other established cohorts), thereby diminishing this factor as a potential confounder, and an established Biospecimen Repository, containing frozen stored blood, buccal cells, and urine.

Detailed procedures for data linkage, processing and quality control have been established with the 12 state cancer registries covering the SCCS catchment area, providing the primary means of identifying incident cancer diagnoses. Although reporting lags are common, the registries provide nearly complete and unbiased ascertainment of cancers diagnosed among the participants after their entry into the SCCS. Three of the registries are SEER registries, and all 12 registries have North American Association of Central Cancer Registries gold- (N=11) or silver-level (N=1) certification. Cohort member deaths are identified through annual linkages with both the National Death Index (NDI) and Social Security Administration (SSA), well-established and reliable means of identifying deaths in the US (32, 33). Additionally, social security numbers (an ideal linking criterion) were collected from 97% of SCCS participants. Informed consent was obtained from all study participants. This study was approved by the Institutional Review Boards of the involved institutions.

At the commencement of the study, there were 854 incident lung cancer cases and 178 of them had available urine samples for the metabolomics analysis, thereby forming the case group for the nested case-control analysis. Corresponding individually-matched controls with stored urine specimens were randomly selected in a 2:1 ratio to cases using incidence density sampling among cohort members matched by age (+/− 2 years), sex, race, CHC recruitment site, menopausal status (women), and date of sample collection (+/− 6 months), the latter to ensure similar specimen storage durations (Table 1). Of note, we conducted an analysis to compare the main variables used for matching or adjustment of analyses between the cases who had and those who did not have urine samples. We found no significant differences except in the race distribution (P=0.03), with a higher number of African-American subjects in those without urine samples when compared to those with (65%, vs. 54%), and self-reported prior history of chronic obstructive pulmonary disease (COPD) (P=0.05). Therefore, we believe that the results presented herein are generalizable to the entire SCCS cohort.

Table 1.

Demographic and clinical characteristics of the nested SCCS case-control sample set.

Characteristics	Cases	Controls	Total	p-valuec
n	178	351	529	0.83
Age; mean±sd	57.7 ± 8.6	57.3 ± 8.5	57.4 ± 8.6
Race; n (%)
African-American	96 (54)	186 (53)	282 (53)	0.92
European-American	74 (42)	146 (42)	220 (42)
Other	8 (4)	12 (3)	20 (4)
Refused		1 (0)	1 (0)
Missing		6 (2)	6 (1)
Gender; n(%)
Male	101(57)	194 (56)	295 (56)	0.88
Female	77 (43)	152 (44)	229 (44)
Smoking Status; n (%)
Current	127 (73)	140 (42)	267 (52)	<0.0001
Former	39 (23)	99 (29)	138 (27)
Never	7 (4)	97 (29)	104 (20)
Histology; n (%)
Adenocarcinoma	59 (33)
Squamous cell carcinoma	36 (20)
Non-small cell carcinoma	19 (11)
Small cell carcinoma	29 (16)
Large cell carcinoma	9 (5)
Othera	13 (7)
Unknown	13 (7)
Stageb; n (%)
Stage occult	2 (1)
I	16 (9)
II	13 (7)
III	41 (23)
IV	77 (43)
Unknown	29 (16)

^aOther includes: adenosquamous carcinoma, carcinoid tumor malignant, carcinoma, neoplasm, neuroendocrine carcinoma.

^bOnly cases identified through cancer registry contain staging information. The original staging was based on the 6th (N = 115) and 7th (N = 50) edition of the AJCC staging manual. Staging presented in the table is based on the 7^th edition, wherein all cases with sufficient tumor size, metastases status and lymph node involvement data previously staged based on the 6^th edition were restaged based on the 7^th edition for consistency (N total 7^th edition staging = 149).

^cTwo-sided Χ² test

Sample processing

SCCS participants were asked to provide urine samples starting in November of 2004. Participants were not required to fast before giving the samples. One-time urine specimens were collected from participants at Community Health Centers, refrigerated, and shipped overnight to Vanderbilt University for processing and frozen storage at −80°C. Participants donated approximately 60 ml of urine which was mixed with a small amount of ascorbic acid as a preservative. Samples were deproteinated using 50% acetonitrile containing 10µM chloropropamide as internal standard. Supernatants were transferred into 96-well sample plates. All pipetting and dilutions were performed using a MICROLAB STARLET automated liquid handler (Hamilton Robotics, Reno, NV), with microcentrifugation steps performed manually (14,000 × g for 15 minutes at 4°C).

Ultra Performance Liquid Chromatography-Mass Spectrometry (UPLC-MS)

Mass spectrometry was performed on a XEVO G2 ESI QTOF mass spectrometer operating in electrospray ionization (ESI) positive (monitoring CR, mass to charge ratio (m/z) =264.1196 and retention time (RT) =0.4 min, and 561+, m/z =561.3435 and RT =6.3) and negative modes (monitoring NANA, m/z =308.0982, RT =0.4 min, and cortisol sulfate, m/z =441.162, RT =5.5). Detailed methods have previously been reported [30], and are given in some detail in the Supplementary Methods.

Statistical analyses

A chi-square test was performed on the dichotomized levels of CR and NANA across smoking, gender and race characteristics, to test differences in levels across these strata. Metabolite abundances were dichotomized into high (>75th percentile) and low (<=75th percentile) groups based on the distribution of abundances in the controls from the current study population. Test for trend was performed to assess trends in metabolite levels across the tumor size categories (34).

Conditional logistic regression analyses were performed to calculate odds ratios (ORs) and accompanying 95% confidence intervals (CIs) for lung cancer associated with high metabolite levels (normalized area under the peak of corresponding chromatographic peaks). In this hypothesis-testing analysis, type I error rates of 0.05 were applied in assessing the significance of each relationship. Covariate adjustments were made by adding terms for age at enrollment, body mass index (BMI), cigarette smoking status (current/former/never), amount smoked among smokers (pack-years), education and income level, prior history of COPD, and family history of lung cancer (yes/no/don’t know). Interaction analysis to test whether race modifies observed associations was conducted using likelihood ratio test.

SCCS mortality follow up data was used to conduct survival analyses for death due to lung cancer among cases in relation to categorical variables of dichotomized metabolite abundances using Cox regression modeling. Multivariate models were computed by adjusting for cigarette smoking status (current/former/never), amount smoked among smokers (pack-years), age at diagnosis, sex, race, stage (7^th edition of the AJCC staging manual) and histology.

Subgroup analyses were performed by stratifying on race (European- vs African-American subjects), and on reported preference for smoking menthol cigarettes (yes vs. no). Sensitivity analyses were performed by removing cases diagnosed within two years of cohort enrollment (n=55), by removing cases and controls who had reported a previous cancer diagnosis (any tumor type; n=60), and by excluding those individuals who did not answer the question regarding the family history of lung cancer (n=91).

In exploratory analyses, we used receiver operating characteristics (ROC) to assess the predictive value of identified metabolites in lung cancer diagnosis using the roccomp function. Models were built using conditional logistic regression on the continuous abundances of CR and NANA, as well as on the demographic, smoking-related and behavioral risk factors, and the combinations thereof. Difference between the predictive abilities of the model containing CR and NANA, and model without the metabolites was tested using rocreg function.

Mann-Whitney (two-sample Wilcoxon rank-sum) test was used to test differences in levels of urinary menthol glucuronide between those who reported and those who did not report a preference for smoking menthol cigarettes.

All reported P values are two-sided, and all P values less than or equal to 0.05 were considered statistically significant. All analyses were conducted in STATA (Stata Statistical Software Release 13.1, College Station, TX).

Results

Risk

CR, NANA, cortisol sulfate and 561+, previously identified as diagnostic and prognostic biomarkers in a case-control study (30) were assessed for their relationship with lung cancer risk in prospectively collected urine specimens from the SCCS cohort. Of note, these specimens were collected before clinically detectable disease, and relevant demographic and clinical characteristics are given in Table 1. Out of four metabolites, only CR was associated with lung cancer in the overall case-control set, while NANA showed borderline significance; cortisol sulfate and 561+ were not associated with either risk or survival (Supplementary Tables 1 and 2). Therefore, focus of the current study is on metabolites CR and NANA.

Initially, potential correlates of CR and NANA were assessed in the control subjects alone, as the presence of the disease may introduce confounding. As illustrated in Table 2, high levels of CR and NANA (dichotomized based on the 75^th percentile of control abundances) are significantly higher in current- when compared to former- and never-smokers. Furthermore, high levels of NANA but not CR are more frequent in former- compared to never-smokers. However, correlation of CR and NANA levels with reported pack years (measure of amount smoked) in former- and current-smokers is not observed (Supplementary Figure 1), indicating that perhaps the amount smoked does not influence the metabolite levels, consistent with the previous report (30). Additionally, high levels of CR are significantly more elevated in males when compared to females, and in African-American when compared to European-American controls. These associations were not seen for NANA. Taking into account possible confounders, all subsequent analyses are adjusted for smoking status, race, and other potential confounding factors.

Table 2.

Associations of smoking, gender and race with high and low levels of A) creatine riboside and B) NANA in controls only. Levels were dichotomized into high and low based on the 75^th percentile of the control abundances.

A)
Creatine Riboside	High n (%)	Low n (%)	Χ2	p-value
Smoking
Current	53 (37.9)	87 (62.1)	21.2	<0.0001
Former	15 (15.2)	84 (84.9)
Never	16 (16.5)	81 (83.5)
Gender
Male	60 (30.9)	134 (69.1)	8.7	0.003
Female	26 (17.1)	126 (82.9)
Race
European Americans	21 (14.4)	125 (85.6)	16.4	<0.0001
African Americans	63 (33.9)	123 (66.1)

B)
NANA	High n (%)	Low n (%)	Χ2	p-value
Smoking
Current	43 (30.7)	97 (69.3)	6.2	0.05
Former	26 (26.3)	73 (73.7)
Never	16 (16.5)	81 (83.5)
Gender
Male	51 (26.3)	143 (73.7)	0.71	0.40
Female	34 (22.4)	118 (77.6)
Race
European Americans	35 (24.0)	111 (76.0)	0.15	0.70
African Americans	48 (25.8)	138 (74.2)

Next, conditional logistic regression analysis was conducted. High levels of CR are associated with increased lung cancer risk after adjusting for age at enrollment, BMI, smoking status, pack-years, income, education level, prior history of COPD, family history of lung cancer (OR_adjusted=2.0 (95% CI=1.2, 3.4; P=0.01), while NANA displays borderline significance (OR_adjusted=1.6 (95% CI=1.0, 2.6; P=0.08) (Table 3). Of note, we conducted a sensitivity analysis wherein we removed those subjects who did not answer the question regarding family history of lung cancer. This analysis indicates that both, CR and NANA are significantly associated with lung cancer risk in the overall cohort (overall: OR_adjusted=1.9 (95% CI=1.0, 3.6), P=0.4, and OR_adjusted=2.0 (95% CI=1.1, 3.6), P=0.02, respectively) (Supplementary Table 3). Stratification by race suggests that these associations are stronger in European-Americans (OR_adjusted=5.3 (95% CI=1.6, 17.6) and 3.5 (95% CI=1.5, 8.4), respectively; P≤0.004) when compared to African-Americans (OR_adjusted=1.1 (95% CI=0.6, 2.3) and 0.9 (95% CI=0.4, 1.8), respectively; P ns) (Table 3), although the likelihood ratio tests of significance for differing ORs by race are not significant (P_interaction=0.25 for CR and 0.11 for NANA). Furthermore, quartile analysis indicates that CR levels are significantly associated with lung cancer risk in the highest when compared to the lowest quartile (OR_adjusted=2.0 (95% CI=1.1, 3.8), P=0.03) (Supplementary Table 4).

Table 3.

Conditional logistic regression analysis of creatine riboside and NANA in all, European- and African-American subjects.

	Univariate				Multivariate
Metabolitea	N(%) Cases	N(%) Controls	OR (95% CI)	p-valueb	N(%) Cases	N(%) Controls	OR (95% CI)c,d	p-valueb
All
Low
Referentcr	109 (61)	262 (75)	1.00		101 (60)	237 (25)	1.00
Referentnana	109 (61)	262 (75)	1.00		101 (60)	231 (74)	1.00
High
Creatine Riboside	69 (39)	87 (25)	2.0 (1.3–3.0)	0.001	66 (40)	77 (25)	2.0 (1.2–3.4)	0.01
NANA	69 (39)	87 (25)	2.0 (1.3–3.0)	0.001	66 (40)	83 (26)	1.6 (1.0–2.6)	0.08
European-Americans
Low
Referentcr	51 (69)	123 (85)	1.00		49 (68)	116 (87)	1.00
Referentnana	40 (54)	109 (76)	1.00		38 (53)	99 (74)	1.00
High
Creatine Riboside	23 (31)	21 (15)	2.7 (1.3–5.4)	0.005	23 (32)	18 (13)	5.3 (1.6–17.6)	0.006
NANA	34 (46)	35 (24)	2.8 (1.5–5.2)	0.001	34 (47)	35 (26)	3.5 (1.5–8.4)	0.004
African-Americans
Low
Referentcr	52 (54)	122 (66)	1.00		48 (53)	113 (66)	1.00
Referentnana	64 (67)	137 (74)	1.00		59 (66)	124 (73)	1.00
High
Creatine Riboside	44 (46)	63 (34)	1.6 (1.0–2.7)	0.06	42 (47)	58 (34)	1.1 (0.6–2.3)	0.73
NANA	32 (33)	48 (26)	1.4 (0.8–2.4)	0.20	31 (34)	47 (27)	0.9 (0.4–1.8)	0.69

^aLevels dichotomized into high and low based on the 75th percentile of population control abundances (low = referent).

^bStatistically significant; p-value <0.05.

^cMultivariate conditional logistic regression adjusted for age, BMI, income, education level, prior history of COPD, family history of lung cancer, smoking status and pack years. Individually matched controls selected in 2:1 ratio to cases using incidence density sampling matched by age (+/− 2 years), sex, race, CHC recruitment site, menopausal status (women), and date of sample collection (+/− 6 months).

^dOR: odds ratio; CI: confidence interval

CR and NANA were previously shown as elevated in the stage I tumor when compared to adjacent non-tumor tissue, linking them directly to altered tumor metabolism (30). Hence it is plausible that these endogenous metabolites are involved with early processes contributing to neoplastic transformation, such as inflammation. To test this hypothesis, we evaluated whether the levels of the markers are predictive of lung cancer risk two or more years before diagnosis by excluding the cases diagnosed within two years of the cohort enrollment (Table 4). While the ORs remain in the same direction for both CR and NANA, the associations are attenuated and no longer significant in the overall cohort after the adjustment for a number of putative confounders, possibly due to insufficiently powered analysis. However, the associations with high CR and NANA remain significant in European-Americans when individuals diagnosed within two years of cohort enrollment are removed (OR_adjusted=6.7 (95% CI=1.6, 27.6) and 3.8 (95% CI=1.3, 11.5), respectively; P≤0.02) (Table 4). Next, aware of a possibility that CR and NANA may be associated with other cancer types, sensitivity analysis was performed wherein the cases (N=28) and controls (N=32) who had reported a previous cancer diagnosis at the time of the enrollment were removed. The ORs associated with high CR remain significantly elevated (OR_adjusted=2.0 (95% CI=1.1, 3.6), P=0.03), while NANA is only significant in the unadjusted model, but the effect size remains similar after the adjustment (OR_adjusted=1.4 (95% CI=0.8, 2.5, P=0.28). Both metabolites are significantly associated with lung cancer risk in European-Americans (OR_adjusted=19.3 (95% CI=2.6, 142.7) and 5.4 (95% CI=1.5, 20.1), respectively; P≤0.01) (Supplementary Table 5).

Table 4.

Conditional logistic regression analysis of creatine riboside and NANA in all, European- and African-American subjects after removal of cases diagnosed within two years (N = 55) of cohort enrollment.

	Univariate				Multivariate
Metabolitea	N(%) Cases	N(%) Controls	OR (95% CI)	p-valueb	N(%) Cases	N(%) Controls	OR (95% CI)c,d	p-valueb
All
Low
Referentcr	79 (64)	178 (74)	1.00		73 (64)	163 (74)	1.00
Referentnana	76 (62)	183 (76)	1.00		70 (61)	163 (74)	1.00
High
Creatine Riboside	44 (36)	62 (26)	1.7 (1.0–2.7)	0.04	42 (36)	56 (26)	1.6 (0.9–3.1)	0.14
NANA	22 (45)	23 (24)	2.2 (1.3–3.6)	0.003	45 (39)	56 (26)	1.7 (0.9–3.1)	0.10
European-Americans
Low
Referentcr	33 (67)	82 (85)	1.00		31 (66)	77 (87)	1.00
Referentnana	27 (55)	73 (76)	1.00		25 (53)	66 (74)	1.00
High
Creatine Riboside	16 (33)	14 (15)	3.0 (1.3–7.3)	0.01	16 (34)	12 (13)	6.7 (1.6–27.6)	0.008
NANA	22 (45)	23 (24)	3.1 (1.4–7.0)	0.007	22 (47)	23 (26)	3.8 (1.3–11.5)	0.02
African-Americans
Low
Referentcr	44 (62)	90 (65)	1.00		40 (61)	83 (65)	1.00
Referentnana	47 (66)	104 (75)	1.00		43 (65)	94 (74)	1.00
High
Creatine Riboside	27 (38)	48 (35)	1.2 (0.6–2.1)	0.64	26 (39)	44 (35)	0.6 (0.3–1.5)	0.31
NANA	24 (34)	34 (25)	1.6 (0.8–3.1)	0.15	23 (35)	33 (26)	1.0 (0.4–2.5)	0.98

^aLevels dichotomized into high and low based on the 75th percentile of population control abundances (low = referent)

^bStatistically significant; p-value <0.05.

^cMultivariate conditional logistic regression adjusted for age, BMI, income, education level, prior history of COPD, family history of lung cancer, smoking status and pack years. Individually matched controls selected in 2:1 ratio to cases using incidence density sampling matched by age (+/− 2 years), sex, race, CHC recruitment site, menopausal status (women), and date of sample collection (+/− 6 months).

^dOR: odds ratio; CI: confidence interval

Finally, in order to evaluate the ability of CR and NANA alone and in combination to classify lung cancer in this nested case-control set, conditional logistic regression models and Receiver Operating Characteristic (ROC) analysis were performed. A model containing variables previously found to be most robust in selecting high-risk individuals for downstream screening comprising smoking status, pack years, age, BMI, income and education levels, previous history of COPD, and family history of lung cancer (28), was compared to a model also comprising CR and NANA, resulting in a significant improvement indicated by the increased area under the curve (AUC) (0.78 compared to 0.80 after the addition of the metabolites, P=0.01) (Figure 1A). Of note, a selected cut-point with the most robust ability to correctly classify subjects, at 50% sensitivity and 86% specificity, leads to a positive predictive value (PPV) of 63%, a negative predictive value (NPV) of 79%, and correct classification of 75% subjects. We assessed the ability of the models to classify lung cancer cases in European-Americans also, due to observed stronger associations in this racial group. We detected a significant improvement in the AUC after the addition of CR and NANA to the model containing risk factors alone (from 0.84 to 0.90, respectively; P=0.002), with a selected cut-point leading to a correct classification of 84% of subjects (PPV=75%; NPV=89%, Figure 1B).

Figure 1.

Receiver Operating Characteristic (ROC) analysis of selected demographic, smoking and behavioral characteristics associated with lung cancer risk, CR, NANA, and combinations thereof in A) all subjects, and B) European Americans. CR: creatine riboside; NANA: N-acetylneuraminic acid; PPV: Positive Predictive Value; NPV: Negative Predictive Value; risk factors model comprises: age (age at enrollment), education (achieved level of education), income (household income), smoking (smoking status (current-, former-, never-smoker)), pack-years (amount smoked indicated by pack-years), BMI (body mass index), COPD (history of chronic obstructive pulmonary disease), family history of lung cancer.

We have previously shown that a preference for smoking menthol cigarettes was no more harmful than smoking non-menthol cigarettes, as assessed in a larger nested case-control set in the SCCS (35). Considering that urinary menthol glucuronide may be a biomarker for smoking menthol cigarettes, the association between the levels of menthol glucuronide and lung cancer risk was investigated. Menthol glucuronide was dichotomized into high and low categories based on the 75^th percentile of the control abundances. No significant association between urinary menthol glucuronide and lung cancer risk were observed in either all subjects, or across the racial strata, after adjustment for amount smoked (Supplementary Table 6). In concordance with our previous report (35), we also did not observe a significant association between a preference for smoking menthol cigarettes and lung cancer risk in this smaller nested case-control set (Supplementary Table 7). Of note, a strong association between the reported preference for smoking menthol cigarettes and urinary menthol glucuronide levels was observed (Fold change (FC)=1.9; P<0.0001 in current- and former-smokers (Supplementary Figure 2A), and FC=2.5; P<0.0001 in current-smokers only (Supplementary Figure 2B), with similar associations observed in European- and African-Americans.

Survival

Based on the previous data from the NCI-MD case-control study that showed high levels of CR and NANA to be associated with worse survival (30), the same analysis was performed in this study. Cox regression analysis adjusted for age at diagnosis, smoking status, pack-years, stage (the 7^th edition of the AJCC staging manual) and histology showed no significant associations (Supplementary Table 8, Supplementary Figure 3). This finding suggests that CR and NANA levels prior to clinically detectable disease are not associated with survival.

No significant associations between either urinary menthol glucuronide (Supplementary Table 9), nor reported preference for smoking menthol cigarettes (Supplementary Table 10) and lung cancer survival were found in this study.

Levels across tumor sizes

Considering previous evidence that CR and NANA are tumor-specific metabolites linked to altered tumor metabolism (30), it is plausible that their levels may increase with increasing tumor size. For a subset of cases with exact tumor size available (N=108), the prevalence of high CR rose with increasing tumor size (P_trend=0.03, Figure 2A). While the results are not significant for NANA, a similar trend was observed across the tumor size strata (Figure 2B).

Figure 2.

Mean levels of A) creatine riboside and B) NANA in SCCS lung cancer cases across tumor size strata.

Discussion

Currently, there are no biomarkers that can successfully complement available screening modalities for detecting early stage lung cancer. Low dose CT (LDCT) has been proven to reduce lung cancer mortality by 20%, as compared to chest X-ray (3). While no lesion goes undetected by LDCT, most are non-malignant, leading to an astonishingly high false discovery rate of 96%. A panel comprising four metabolites detected in the urine was found to be most significantly associated with both lung cancer status and survival, and reproducible and stable in stored urine over a long period of time (30). The goal of the present study was to investigate whether the four metabolites, creatine riboside (CR), N-acetylneuraminic acid (NANA), cortisol sulfate, and indeterminate compound designated as 561+, were predictive of lung cancer risk in pre-diagnostic urine specimens, one of the crucial steps for validating promising biomarkers that may have clinical utility (29). Of these four, only CR was predictive of lung cancer risk before clinically detectable disease, with NANA displaying borderline significance. In our earlier work, we did not detect elevated levels of cortisol sulfate and 561+ in tumor tissue, which also tended to lower the importance of the associations in urine. CR and NANA, however, were significantly elevated in tumor when compared to matched, adjacent non-tumor lung tissue, strengthening their importance as markers of lung cancer risk. NANA has previously been characterized and is the most common form of sialic acid that plays several roles in biology, and is thought to protect cancer cells from immunosurveilance (36). Creatine riboside is a novel metabolite previously reported for the first time by our group (30). While the characterization of this metabolite is of special importance for future studies, we speculate that CR may be a product of both, high creatine within the tumor, and high phosphate flux, as a result of a pronounced energy requirement of fast dividing tumor when compared to normal cells (37). Future studies should elucidate what role, if any, CR may play in neoplastic transformation.

While we were unable to conduct an adjusted conditional logistic analysis between CR and NANA and early stage I-II lung cancer due to a small number of patients in this group, univariate analysis suggest that these metabolites may be robust in predicting early stage lung cancer (data not shown), consistent with our previous report. The utility of these metabolites in aiding early lung cancer diagnosis remains to be established by investigating their usefulness in distinguishing benign from malignant nodules detected by LDCT.

Although the levels of the two metabolites were higher among current than non-smokers (and levels of NANA but not CR were higher among former than never smokers), the associations between CR and lung cancer risk held after adjusting for cigarette smoking status and amount smoked. In both this and in the previous report, we did not observe any correlations between the metabolites and the amount smoked (cigarettes per day and pack years, respectively). Furthermore, a previous study showed that amino acid and lipid pathways are most significantly affected by the exposure to tobacco-smoke and reversible after smoking cessation; CR and NANA, however, are not members of these pathways (38). Thus, the levels of these metabolites may not be affected by immediate exposure to tobacco smoke, and they are not likely markers of exposure. While NANA displayed borderline significance after adjustment in the overall cohort, the levels of both metabolites were significantly associated with lung cancer risk in European-Americans. Although race did not seem to significantly modify the effect of the associations between metabolites and lung cancer risk, the associations were stronger in European- than African-Americans. The lack of observed formal interaction may be a result of decreased power, or factors associated with race (such as lifestyle), which could not be accounted for in this study. In our previous case-control study, racial differences did not attenuate associations with lung cancer status, as disease presence may override any other contributing factors, such as genetic backgrounds. Future investigation may elucidate whether or to what extent racial differences in the association with lung cancer risk may exist. It is possible that genetic variants may contribute to racial differences, as they play a role in the metabolism of both exogenous (25) and endogenous metabolites (39, 40). Additionally, lifestyle and external exposures may exacerbate the observed differences, factors that could not be controlled in this study.

In order to address other causes that may contribute to health disparities in lung cancer, a preference for smoking menthol cigarettes, which is much higher among African- than European-Americans (41–43) was investigated. We have previously reported that smoking menthol cigarettes, preferred by African-Americans, does not lead to an increased risk for developing lung cancer in the SCCS (35). In the current study we took an extra step and measured urinary menthol glucuronide as a biomarker of exposure to menthol from tobacco smoke. Consistent with the previous report, no significant associations were observed in relationship with lung cancer risk. We are the first group to our knowledge to report this observation.

Metabolic markers have gained increased interest recently, as they are recognized as proximal markers to the disease phenotype in comparison to other ‘omics’ markers (44–46). The footprint that products of altered tumor metabolism may leave in non-invasively collected biospecimens serves as a great opportunity for biomarker discovery. This “liquid biopsy” practice would be a high-throughput, noninvasive and inexpensive approach to allow not only cancer detection, but also patient monitoring during treatment, and would provide an ideal therapeutic strategy for precision medicine (47). There is a great necessity to identify and validate a complementary tool to the solid biopsy, which may also allow for delivery of targeted therapy. Using very precise mass-spectrometry for measuring lung cancer risk markers, such as CR and NANA, presents a feasible opportunity to develop reproducible, inexpensive and high-throughput diagnostic tools to refine algorithms for early detection of lung cancer.

Caveats of the study include a relatively limited sample size, as well as the inability to control for the exogenous effects on metabolism. We adjusted for the effects of smoking, although residual confounding cannot be ruled out, but other unmeasured factors may have contributed to the differences in CR and NANA levels between those with and without lung cancer. A major strength of the study is a well matched nested case-control sample set with its close matching on demographic variables and in methods in and timing of collection of urine samples, as well as similar socio-economic backgrounds of the subjects enrolled in the cohort.

Overall, the results of this study suggest that CR and NANA are associated with lung cancer risk in prospective samples, more strongly in European-Americans, and therefore may have clinical utility for disease screening or diagnosis. Considering an expected forthcoming rise in the detection of indeterminate pulmonary nodules as a result of LDCT screening, it is crucial that non-invasive biomarkers be developed, with ability to distinguish malignant from indolent nodules. Future biomarker studies in LDCT screening trials are needed to determine the clinical utility of these and other biomarkers of NSCLC.