Bridging the clinical gaps: genetic, epigenetic and transcriptomic biomarkers for the early detection of lung cancer in the post-National Lung Screening Trial era

Lung cancer is the leading cause of cancer death in both men and women in the US and the world, causing more than 1 million deaths per year [1‐4]. The global cancer burden in annual cases is projected to double by 2050, and lung cancer is expected to remain the leading cause of all cancer deaths during that time. Cigarette smoke remains the main risk factor for lung cancer, with 85% to 90% percent of lung cancer cases in the US occurring in current or former smokers. However, only 10% to 20% of heavy smokers develop lung cancer [5]. While smoking cessation gradually reduces the risk of lung cancer, the majority of new lung cancer cases occur in former smokers. The high mortality in patients with lung cancer (80% to 85% in five years) results in part from our inability to predict which of the 100 million current and former smokers in the US are at greatest risk for developing lung cancer, and from the lack of effective tools to diagnose the disease at an early stage [6].

Recent results published from the National Lung Screening Trial have shown that screening high-risk smokers (based on age and cumulative exposure to tobacco smoke) with low-dose helical computed tomography (CT) can lead to a reduction in both lung cancer mortality (by 20.0%) and all-cause mortality (by 6.7%) compared to standard radiographic screening. While this landmark study is already impacting CT screening guidelines and practices across the US, there were a number of important caveats to the study, including the age (55 to 75 years old) and smoking exposure cutoffs (>30 pack-years) chosen for patient inclusion, and the duration of annual CT screening (three years). Importantly, 39.1% of all participants in the low-dose CT arm of the trial had at least one positive screen for lung cancer, and 96.4% of these initial positive screenings represented false positives for lung cancer [7]. This overabundance of false positives could lead to higher screening costs and unnecessary invasive procedures on many smokers who do not actually have lung cancer. Thus, there is a critical need to develop biomarkers that can determine which of the frequently detected lung nodules on CT scan are malignant (that is, diagnostic markers), and to further define the large high-risk population that would be eligible for screening by CT to increase the efficacy of screening and to reduce the cost and morbidity associated with it (that is, screening markers; Figure 1).

The sequencing of the human genome together with the technological advances that enabled this accomplishment have ushered in a new era of molecular biomarker development that promises to help address these unmet needs. This review will summarize recent genetic, transcriptomic and epigenomic biomarkers that are emerging as tools for the early detection of lung cancer (Figure 2), both in the diagnostic and the screening setting (prognostic and predictive biomarkers will not be covered). The review will focus on genome-wide studies in clinical biospecimens (no animal models or cell line studies) that leverage these emerging high-throughput technologies, and will review commonality of variants between lung cancer and chronic obstructive airways disease. Although there are a number of promising metabolic and proteomic biomarkers for early lung cancer detection, these fall outside the scope of this review [8].

Initial genome-wide associations in lung cancer robustly implicated SNPs (Table 1) spanning the chromosome 15q25 region encoding the gene cluster of nicotinic receptors, CHRNA3/A5/B4[9‐12]. Subsequent multi-investigator consortia analyses confirmed the association of SNPs spanning this region with heavy smoking, nicotine dependence, craving and related endophenotypes [11, 13, 14]. Saccone et al. [13] conducted a meta-analysis across 34 datasets of European-ancestry participants (Table 1), including a diverse group of 38,617 smokers, and demonstrated that rs16969968, a nonsynonymous coding polymorphism of the CHRNA5 gene, correlated highly significantly with smoking behavior (odds ratio = 1.33, P = 5.96 × 10⁻³¹). Three other large smoking genetics consortia confirmed this locus as that most associated with smoking quantity [11, 14, 15].

Therefore, the challenging question was the degree to which the associations between these chromosome 15q25 variants and lung cancer were due to their effects on smoking intensity, rather than a direct carcinogenic effect. The lung cancer association, though statistically robust, and initially not altered by adjustment for smoking, increasingly appears to be mediated through smoking. However, there is still uncertainty as to the degree with which the association for lung cancer is mediated through genetic risk beyond that attributed to smoking intensity. Saccone et al. [13] showed that locus 1 was associated with lung cancer even when controlling for amount smoked per day (odds ratio = 1.31, P = 1.99 × 10⁻²¹), suggesting possible direct genetic effects of locus 1 on this cancer, at least in the presence of smoking. Spitz et al. [16] noted that the lung cancer risk associated with the variant genotype was highest in the lightest smokers (<20 cigarettes per day) and younger patients (<61 years), arguing a role for genetic susceptibility in these lesser-exposed groups. Furthermore, they [16] were not able to implicate this locus as a risk factor in other smoking-related cancers (bladder and renal), suggesting genetic effects on both smoking behavior and lung cancer risk.

Wang et al. [17] demonstrated that each copy of chromosome 15q risk alleles was associated with increased cigarette consumption of 1.0 cigarette per day at rs12914385 and 0.9 cigarettes per day at rs8042374 with, and concluded that these modest differences in smoking behavior were sufficient to account for the 15q25 association with lung cancer risk. However, it could also be argued that cigarettes per day is not a sufficient proxy for carcinogen exposure [18].

Truong et al. [19] used data from 21 case-control studies (nine in North America, eight in Europe and four in Asia) and replicated the association between chromosome 15q25 SNPs and lung cancer risk (Table 1) in white ever smokers (odds ratio = 1.26, CI:1.21-1.32, P-trend = 2 × 10^(-26)) and also confirmed that this association was higher at younger age of onset (P-trend = 0.002), whereas no association was found in never smokers or in Asian participants. Spitz et al. [16] found no elevated risk associated with these variants in over 547 lifetime never smoking patients with lung cancer. Subsequent meta-analyses of never smokers with lung cancer (Galvan and Dragani [20] in >1,000 never smokers and >1,800 controls; and Wang et al. [17] in 2,405 patients and 7,622 controls) replicated the lack of any statistically significant association with this locus in never smokers.

Other top hits identified in the GWAS have also been replicated. A number of well-designed GWAS and meta-analyses have implicated variants at the 5p15.33 locus in cancer risk at several different sites, including lung cancer in both white and Asian patients [21]. Truong et al. [19] confirmed the significant association in white patients for rs2736100 in the chromosome 5p15 locus. Both Troung et al.[19] and Landi et al.[12] noted a histology-specific role of rs2736100 in adenocarcinoma. This locus was also recently implicated in lung cancer risk in African American patients [22]. There is biologic plausibility for this finding because mean relative telomere length has been associated with four genetic variants of the hTERT gene, including rs2736100 [23], and TERT gene amplification is responsible for TERT mRNA overexpression in a majority of lung adenocarcinomas [24]. Cleft lip and palate transmembrane protein 1-like (CLPTM1L) gene also resides in this region of chromosome 5 for which copy number gain has been found to be the most frequent genetic event in early stages of non-small cell lung cancer. James et al. [25] demonstrated increased CLPTM1L expression in lung adenocarcinomas and protection from genotoxic stress-induced apoptosis and concluded that anti-apoptotic CLPTM1L function could be another mechanism of susceptibility to lung tumorigenesis. A third region implicated by GWAS in susceptibility to lung cancer in Caucasians is the human leukocyte antigen region at chromosome 6p21 [10, 26].

The association with SNPs in the 5p15 and 15q25 regions was confirmed in a Korean population with similar magnitude of effect as reported for other ethnic groups, but there was no association with the 6p locus [27]. Likewise, the effect of the 5p15 SNP was significant only for adenocarcinoma. Truong et al.[19] noted no effect for the chromosome 15q locus, but replicated the association with the 5p locus in an Asian population. A Japanese study [28] confirmed the finding at 5p15.33. There have been several GWAS in Chinese populations. Hu et al. [29] replicated findings of significance in both 3q28 (TP53) and at the 5p13 locus (Table 1). They also reported significance at two additional loci, 12q12 and 22q12 (Table 1). In an attempt to identify additional susceptibility loci in Chinese patients with lung cancer, Dong et al. [30] reported genome-wide significance for three additional lung cancer susceptibility loci at 10p14 (close to GATA3), 5q32 in PPP2R2B-STK32A-DPYSL3, and 20q13.2 in CYP24A1. They also found additional associations for rs247008 at 5q31.1 (IL3-CSF2-P4HA2), and rs9439519 at 1p36.32 (AJAP1-NPHP4). There was suggestive evidence for interactions with smoking dose. Jin et al. [31] noted that genetic variants at 6p21.1 and 7p15.3 were associated with risk of multiple cancers in Han Chinese patients, including lung cancer. Finally, Shi et al. [32] reported that a locus on RAD52, involved in DNA double-strand break repair and homologous recombination, influenced risk of squamous cell lung cancer but not other cell types.

It is likely that many more common variants can be anticipated to contribute to lung cancer risk, although with effect sizes too small to reach significance in genome-wide analyses. It has been argued that there are diminishing returns in predicting disease risk from common marker SNPs, and greater effort should be spent investigating functional relevance of the GWAS findings. For example, evaluating the effect that SNP variation has upon expression and activity of nicotinic receptors can be explored by taking advantage of animal and cellular models of CHRNA3 and CHRNA5 knock-out animals [33, 34]. Studies of cell lines and primary lung cancers can provide insights into the effects of these variants on proliferation and apoptosis; one such study suggested a role of a proteosome gene in this region beyond the effects of nicotinic receptors [35]. Emerging metabolomic markers may provide useful biomarker dosimeters of smoking damage relative to carcinogenesis. Certainly, multiple strategies are needed to further tease apart these complex relationships [18].

Lung cancer and chronic obstructive pulmonary disease (COPD) result from the combined effects of smoking exposure and genetic susceptibility. Tobacco smoke exposure has been responsible for 80% of lung cancers, however only 15% to 20% of chronic smokers develop lung cancer or COPD. Approximately 50% to 90% of smokers with lung cancer also have COPD. Studies have shown that COPD is an independent risk factor for lung cancer among Caucasians and African Americans, conferring a four- to six-fold increased risk. Over the past few years, several lung cancer risk models have been developed [36‐40], some of which included pulmonary diseases such as COPD and pneumonia. Consistently, the inclusion of COPD in the models leads to improvement of the discriminatory power and good calibration [41]. The model with the highest discriminatory power reported to date is the extended Prostate, Lung, Colorectal and Ovarian lung cancer risk model [37], which also includes COPD. This dual susceptibility indicates a link between the processes that induce COPD and lung cancer.

Results from recent GWAS suggest a possible overlap in the genetic risk factors predisposing smokers to lung cancer and COPD. Several regions in the genome associated with lung cancer and/or COPD have been identified, including chromosome 1q21, 4q22, 4q24, 4q31, 5p15, 5q32, 6p21, 6q24, 15q25 and19q13 [9, 10, 41‐49]. Several important genes mapping to those regions have also been identified as significant players in the pathogenesis of lung cancer and/or COPD (Table 1), and many of these loci overlap. For example, a variant in the FAM13A gene has been reported to have a protective effect in COPD and lung cancer [49]. CHRNA3/5 (15q25) was reported to be associated with both COPD and lung cancer [10, 48, 49] through its effects on both smoking exposure and COPD. Using mediation analysis, Wang et al. [50] reported that COPD is a mediating phenotype that could partially explain the effect of smoking exposure on lung cancer. These findings suggest the presence of shared susceptibility mechanisms for these two smoking-related diseases. Such susceptibility may also be mediated through receptors expressed on the bronchial epithelium that implicate molecular pathways underlying both COPD and lung cancer [51]. To date, most of the lung cancer and COPD genetic studies have been conducted independent of each other, which has contributed to the mediating effect of one disease over the other being overlooked [52].

Epigenetics is classically defined as the study of changes in downstream phenotypes or gene expression that cannot be attributed to changes in DNA and is heritable. Another refined definition is that epigenetics concerns structural changes in chromosomal regions that are not related to changes in DNA that mark altered activity states [53]. Two major types of epigenetic regulation are DNA methylation and histone modification, both of which are known to modulate gene expression. Given that the abundance of molecular biomarkers in this field have been DNA methylation-based, this section will focus on DNA methylation studies that hold the potential to impact early lung cancer detection (Table 2a).

DNA methylation is an epigenetic mechanism marked by the joining of a methyl group to a cytosine base to form 5-methylcytosine, typically at a CpG dinucleotide near or within a CpG island. When CpG dinucleotides are methylated to a high degree in the promoter region of a gene, that gene’s expression is usually down-regulated as a result. This is one way that cells can regulate which genes are expressed (Figure 2) and is a mechanism utilized during cell and tissue differentiation during development [54]. Aberrant hypermethylation of oncogenes or hypomethylation of tumor suppressor genes (Table 2a) is one way that transcriptional regulation can spiral out of control in cancer cells [55].

Genome-wide methylation profiling has been used to identify altered methylation patterns in lung cancer tissue (including genes such as CDKN2A, RASSF1A, ARHI, MGMT and RARβ) [56, 57], but so far only one larger scale study has shown the possibilities of identifying methylation biomarkers for the diagnostic or screening setting in noninvasive biospecimens utilizing microarray-based technologies. In this study, nine CpGs were able to discriminate between lung cancer cases and controls with an area under the receiver operator characteristic curve (AUC) of 0.86 [58]. The vast majority of current methylation studies that could be useful for screening and diagnostic tests remain at a candidate gene or gene panel level analysis (Table 2a).

Belinsky et al. [59] originally identified the hypermethylation of CDKN2A in lung tumors but within the same study also examined the sputum of 33 people who smoked. In this small initial study, eight patients had sputum with methylated CDKN2A detected by methylation-specific polymerase chain reaction (MSP). Of those, three were diagnosed with lung cancer at the time of sputum collection and one other would develop lung cancer a year later [59]. Work on identifying CDKN2A, as well as MGMT, as a measure of cancer risk and diagnosis was expanded in a 21-patient study of matched sputum and squamous cell carcinoma (SCC) samples as well as sputum samples from 32 patients evaluated for possible lung cancer. This study was able to significantly improve cancer detection and risk using the methylation status of the two genes compared to cytology alone, with 100% of patients with SCC displaying methylation of one or both of these genes. More importantly, these genes were aberrantly methylated up to three years before diagnosis [60]. By looking at the sputum of lung cancer-surviving smokers, cancer-free smokers and never smokers, then adjusting for age and smoking duration, MGMT, RASSF1A, DAPK and PAX5α were also identified as being significantly differently methylated in the lung cancer survivors. This indicates that aberrant methylation of a panel of candidate genes could identify patients with higher risk of lung cancer (lung cancer-surviving smokers had a 6.2-fold higher odds of having three or more of these genes methylated in sputum) [61]. Other genes that have been identified in the sputum with aberrant methylation associated with increased risk for lung cancer include ASC/TMS1 (increased odds in cancer patients from 7.2 to 28.6) [62], GATA4, GATA5 and PAX5β (6.5-fold increase in cancer risk with methylation of three or more genes) [63]. Recently, a larger panel of 31 genes in the sputum was used to identify signatures of stage I lung cancer. It had >70% accuracy and could predict which smokers had cancer between 3 and 18 months before clinical diagnosis (AUC of 0.71 and 0.77 for the two cohorts in the study) [64].

Other potential distal sites for assessing lung cancer risk using methylation markers include the serum, plasma and blood leukocytes. Based on evidence that DNA from tumor cells can be found freely in circulating serum [65], Esteller et al. [66] examined the serum, normal lung tissue and tumor tissue from 22 patients with non-small cell lung cancer (NSCLC). They found that 73% of patients had serum DNA that reflected hypermethylation events found in their tumors. Specifically using MSP, they looked at methylation of CDKN2A, MGMT, DAPK and GSTP1, genes whose aberrant methylation profiles have already been shown to associate with lung cancer risk or diagnosis [66]. A larger study with a cross-section case-control design looked at the serum from 200 patients, 91 of whom had lung cancer, 100 had non-malignant lung disease, and nine had some other malignant disease. RARβ, CDKN2A, DAPK, RASSF1A and MGMT were examined, and the analysis showed that a patient having methylation of just one gene had an odds ratio of 5.08, meaning they were approximately five times likely to have lung cancer than patients without any methylated genes. This odds ratio increased in patients with two or more genes being aberrantly methylated [67]. Overall, just looking at this limited candidate gene list, almost 50% of patients with lung cancer displayed at least one case of aberrant methylation in their serum. Other genes with aberrant methylation in serum DNA have been found to associate with lung cancer risk, including TMEFF2[68], RUNX3[69] and CDH13 [70], suggesting that many genes in the serum could signify lung cancer risk and that a larger profile of aberrant methylation could produce a more accurate biomarker for lung cancer risk. The work by Begum et al. [71], who looked at methylation profiles of a slightly larger set of 15 genes and then selected the six most sensitive and specific genes for predicting lung cancer risk (APC, CDH1, MGMT, DCC, RASSF1A and AIM1), clearly shows evidence that a more global methylome approach could lead to a more sensitive (75%) and specific (73%) biomarker of lung cancer risk from serum DNA [71]. Methylation events in plasma, specifically in CDKN2A, MGMT and RASSF1A[61], as well as in peripheral blood leukocytes [58] and lymphocytes [72, 73], are promising less invasive sites for assessing lung cancer risk through measuring DNA methylation differences.

Gene-expression profiling or transcriptomics has been used to delineate disease classification, improve diagnostic accuracy, identify new molecular targets for drugs and provide new biological insights into lung cancer. High-throughput technologies, such as microarray, and sequencing platforms allow the measurement of thousands of genes simultaneously, to look for different pattern changes across subsets that help characterize a particular physiological state or clinical phenotype. In this section, we will review the diagnostic and screening transcriptomic biomarkers that have been developed in the airway and blood of at-risk smokers (Table 2b).

MicroRNAs are recently discovered small molecules that play an important role in regulating gene expression. These noncoding RNAs, in their final active form, are usually 22 nucleotides in length and target specific parts or mRNA sequences, usually found in the 3′ untranslated regions of mRNA, which either prevent translation or promote mRNA degradation, and lead to down-regulation of specific genes [89]. Because miRNA are relatively more stable than mRNA [90], any miRNA profiles of lung cancer risk or diagnosis are likely to be more accurate when moving from the bench to the clinic. This review will focus on large-scale miRNA studies that have been performed in airway, sputum and blood for early lung cancer detection (Table 2c).

Circulating cell-free DNA (cf-DNA) is a known marker of cancer cell death and an increase in the prevalence of cf-DNA in the blood has been used as a marker to distinguish patients with cancer patients from patients with no cancer [110, 111]. In this section, we will review recent efforts to utilize cf-DNA as a diagnostic and screening biomarker for differentiating patients with lung cancer from those without, especially some studies that have been able to identify changes in cf-DNA that can distinguish patients with early stage lung cancer from patients with no cancer.

Utilizing polymerase chain reaction (PCR), Sozzi et al. [112] examined the levels of plasma DNA in 84 patients with NSCLC and 43 healthy blood donor controls, and showed over a range of cutoff points that healthy controls could be distinguished from patients with lung tumors with an AUC of 0.844. They showed that, even in patients with stage 1a cancer, the amount of circulating plasma DNA was significantly higher than in the control patients [112]. Although this study showed the use of quantifying cf-DNA to distinguish patients with cancer from healthy controls, only 8 of the 43 controls were smokers, so smoking may have been a confounding effect in the cancer diagnosis. Sozzi et al. [113] addressed this by matching 93 control individuals who smoked with 100 patients with NSCLC, and were able to discriminate the patients from controls by concentration of cf-DNA with an AUC of 0.94. They also quantified the risk such that a unit increase in plasma DNA led to a 21% increase in NSCLC risk [113]. Other studies have shown that cf-DNA can distinguish patients with benign lung disease from lung cancer with an AUC of 0.73 [114]. However, in a screening cohort of >1,000 higher risk smoking volunteers, Sozzi et al. [115] found that quantification of cf-DNA could not be used to distinguish the individuals who would develop lung cancer from those who did not [115]. Other studies have also confirmed this finding [116].

In terms of the emerging clinical needs (Figure 1), these data argue that cf-DNA may not be an effective marker for screening high-risk smokers, but based on these and many other cf-DNA studies [117‐120], it could still play a role in diagnosing whether nodules identified by low-dose CT are either benign or malignant. The field is also progressing towards the identification of screening- or diagnostic-specific markers within lung tumor circulating cf-DNA including methylation markers [68] and genetic mutations such as epidermal growth factor receptor mutations [121‐123]. Although these developments are relatively new, the Dawson et al. study [124], in which genetic alterations could be identified in cf-DNA that corresponded dynamically with metastatic breast cancer, shows the potential for identifying novel genomic and genetic biomarkers within cf-DNA to better stratify patients [124].