AHRR (cg05575921) Methylation Safely Improves Specificity of Lung Cancer Screening Eligibility Criteria: A Cohort Study

This article is featured in Highlights of This Issue, p. 693

Lung cancer is the leading cause of cancer-related death, and smoking accounts for 70% to 80% of lung cancer causality (1). The overall 5 years survival after a diagnosis lung cancer is below 20% (2–4), mainly because more than half of patients have advanced disease at the time of diagnosis (4). With the recent results from the NELSON trial (5), it is clear that screening with low-dose computed tomography (LDCT) reduces mortality from lung cancer in the eligible population. It is now essential to refine the eligibility criteria to further increase benefits and reduce harms of lung cancer screening.

In lung cancer screening trials, eligible individuals have been identified mainly based on age and smoking history (5–11). Self-reported smoking information provides valuable risk information, but it is difficult to capture subtle, but important, variation in smoking behavior due to variation of number of inhalations per cigarette, depth, and duration of tobacco smoke inhalation in the lung (12–15). Therefore, even very detailed smoking records underestimate the association of smoking history with future risk of lung cancer (16). Smoking is associated with epigenetic modifications, with DNA methylation of the aryl hydrocarbon receptor repressor (AHRR) at CpG site cg05575921 showing the strongest association (17–25). A Danish population study demonstrated that variation in leukocyte AHRR (cg05575921) methylation provided information beyond that of self-reported smoking on lung cancer risk up until 23 years after the measurement (26).

In this study, we tested the hypothesis that AHRR (cg05575921) methylation has the potential to improve the specificity of current lung cancer screening criteria, by excluding those with the lowest lung cancer risk from the risk population. Using eligibility criteria from seven lung cancer screening trials (5–11) to identify individuals at risk in a population-based cohort in Denmark, we compared the performance without and with inclusion of AHRR (cg05575921) methylation. The model with the lowest number of included individuals per lung cancer event within 5 years after examination was considered the best and was validated in another examination of the same cohort.

The background study populations consisted of 10,135 and 6,238 individuals from the 1991 to 1994 examination and the 2001 to 2003 examination of The Copenhagen City Heart Study. From the background populations, we excluded participants without methylation measurements, or with lung cancer before examination or without data on smoking history, leaving 9,206 and 5,334 individuals for analysis (Supplementary Fig. S1). The Copenhagen City Heart Study is a prospective study of the general Copenhagen population, inviting residents above age 20 to complete a questionnaire and undergo a physical examination. Lung cancer was not and still is not screened for in Denmark. Detailed study rationale, design and procedures have been previously published (27).

Baseline information on lifestyle, including smoking, was collected using a self-administered questionnaire, reviewed together with an examiner at the day of examination. All individuals were asked about smoking status, and smokers and former smokers were asked about number of daily-consumed cigarettes, cheroots, cigars, or pipes, age at smoking initiation and—for former smokers—age at cessation. For former smokers, we calculated duration of smoking as the difference of age at smoking cessation and initiation. For current smokers, duration of smoking was calculated as the difference between age at examination and at smoking initiation. Also, cumulative smoking was calculated in pack-years corresponding to the consumption of the equivalent of 20 cigarettes per day for 1 year.

Body mass index (BMI) was calculated as measured weight in kilos divided by height in meters squared, and blood samples were drawn. The study was approved by the Danish ethics committee (KF100.2039/91) and was conducted according to the Declaration of Helsinki. Written informed consent was obtained from all individuals.

The AHRR (cg05575921) methylation extent was measured in duplicate samples using a Taqman assay developed in our own laboratory. Bisulphite treated leukocyte DNA drawn from peripheral blood, was amplified using forward and reverse PCR primers, which were designed to bind to DNA around the cg05575921 site on sequences without genetic or possible CpG methylation variation. Details of the AHRR (cg05575921) methylation measurements has been published previously (26). DNA methylation extent was expressed as the percentage of methylated of all AHRR cg05575921 alleles.

Data on lung cancer (ICD10 code C34) were found by merging the unique personal ID of individuals with the Danish Cancer Registry. Date of death or emigration was found by linkage with the Danish Civil Registration System. Follow-up time for lung cancer began at the time of the 1991 to 1994 examination or at the time of the 2001 to 2003, and ended at the day of lung cancer diagnosis, death, emigration, or 5 years after examination, whatever came first. Follow up for lung cancer was terminated at 5 years after examination to simulate the screening situation, and thus excluding events originating from very small or nonexisting lung cancers not detectable by screening at the day of examination. Individuals with lung cancer before date of examination were excluded.

In this study, we externally validated eligibility criteria from seven lung cancer screening trials (5–11). In most trials, the high-risk population was identified uniquely based on age and smoking history. The exact age range varied between 50 and 69 years (7, 10), 50 and 74 years (5), 55 and 69 years (8), 55 and 74 years (6), and 60 and 74 years (9). Minimum pack-years varied between 15 (5, 10), 20 (7–9), and 30 (6), quitting years for former smokers varied between 10 years (5, 7–8, 10) and 15 years (6). One eligibility criterion (PLCO_M2012) was based on a risk prediction model (11), which includes a detailed modeling of risk in relation to age and smoking history, and prior diagnoses of malignant tumor, COPD, family history of lung cancer, BMI, education (Supplementary Table S1). We did not examine the Bach model (28), the Spritz model (29), and the LLP model (30), as they include variables as asbestos exposure, dust exposure, pneumonia, emphysema, and hay fewer, which are not available for our population.

The population was categorized by their baseline smoking status (never, former, or current; Supplementary Table S2). STATA/SE 16.0 was used for all calculations. We used Kruskal–Wallis test and Pearsons X² test to test for trend of smoking status through all included variables.

Among the individuals of the 1991 to 1994 examination, we identified seven eligible populations based on the eligibility criteria used in lung cancer screening trials (DANTE, DLCST, ITALUNG, LUSI, NELSON, NLST, PLCO_M2012; refs. 5–11; Supplementary Table S2). We estimated the risk of developing lung cancer over the next 5 years, using Kaplan–Meier plots for individuals fulfilling versus not fulfilling the specific set of criteria. To test for equality of lung cancer risk, we used log-rank tests. We calculated sensitivity, specificity, positive predictive values (PPV), and negative predictive values (NPV) of each set of criteria. Sensitivity is defined as the proportion of individuals with lung cancer within 5 years, who have a positive result. Specificity is defined as the proportion of individuals without lung cancer within 5 years, who have a negative result. Positive predictive value is defined as the probability that an individual with a positive test develops lung cancer within 5 years. NPV is defined as the probability that an individual with a negative test does not develop lung cancer within 5 years. Cox proportional hazard regression model using time since examination as the time metric was used to estimate HR and 95% confidence intervals (95% CI) for the association between AHRR (cg05575921) methylation extent and risk of lung cancer. Restricted cubic spline models were used to assess nonlinear relationships and the median value of AHRR (cg05575921) methylation extent among never smokers was chosen as reference. We used a model controlled for age, sex, smoking status, and cumulative smoking. Among eligible individuals, identified by the eligibility criteria, using AHRR (cg05575921) methylation cut-off values in the range from 50% to 60%, we calculated the reduction of screening burden and fraction of undetected 5-year lung cancers compared with solely use of each eligibility criteria without AHRR (cg05575921) methylation to discriminate. The AHRR (cg05575921) methylation cut point by which a maximum of 5% of 5-year lung cancers were undetected, was chosen. Not detecting 5% of lung cancers was considered as the upper tolerable limit for a screening program. The seven criteria combined with the identified AHRR (cg05575921) methylation cut point were ranked according to number of included individuals per lung cancer event within 5 years. Among participants of the third examination, with AHRR (cg05575921) methylation extent fixed at 60%, we calculated our power to improve specificity above 95% for each of the criteria: DANTE (96%), DLCST (81%), ITALUNG (55%), LUSI (89%), NELSON (92%), NLST (54%), and PLCO_M2012 (46%).

The model with the lowest number of included individuals per lung cancer event within 5 years was validated among 5,370 individuals from the 2001 to 2003 examination.

The characteristics of the individuals of the 1991 to 1994 examination with available DNA are shown in the Supplementary Table S1. The average follow-up time was 4.8 years for lung cancer, with a total of 43,999.9 person-years. A total of 92 individuals (1 per 100) had a clinical diagnosis of lung cancer within the first 5 years of follow-up. The average time to lung cancer diagnosis for these individuals was 2.9 years.

For all seven eligibility criteria there was a higher lung cancer incidence within the eligible populations compared with the rest of the individuals (P < 0.05) with Kaplan–Meier estimates varying from 2.2% (LUSI) to 3.5% (PLCO_M2012; Fig. 1).

Applying the seven eligibility criteria led to populations ranging from 3,143 (DANTE) to 1,621 (ITALUNG; Table 1) individuals. The number of individuals per 5-year lung cancer varied from 31 (PLCO_M2012) to 47 (LUSI). The DANTE criteria included the highest number of individuals and showed the highest sensitivity 77.2% (95% CI, 67.2–85.3), but the lowest specificity 67.0% (95% CI, 65.1–68.0). The ITALUNG and LUSI, in contrast, included fewer individuals (n = 1,621 and 1,832) and were associated with significantly higher specificities (83.1% and 81.9%), but also significantly lower sensitivities (51.1% and 42.4%). The positive predictive value (PPV) was highest for the PLCO_M2012 criteria, 3.2% (95% CI, 2.5–4.1), whereas the DANTE showed the highest NPV, 99.7% (95% CI, 99.5–99.8).

In the whole population, there was a strong inverse association between AHRR methylation and 5-year risk of lung cancer after adjusting for smoking status and cumulative smoking (Fig. 2). When the population was stratified by smoking status, the AHRR methylation distributions differed across never, former, and current smokers (Supplementary Fig. S2).

Adding AHRR (cg05575921) methylation improved performance parameters of the criteria at varying AHRR (cg05575921) methylation levels (Table 2). The AHRR (cg05575921) methylation cut-off corresponding to 5% undetected lung cancers to the eligibility criteria, led to inclusion of fewer individuals and higher specificities (ranging from 86.3% ITALUNG to 75.6% DANTE), but lower sensitivities (ranging from 73.9% DANTE to 40.2% LUSI). The number of included individuals per 5-year lung cancer varied from 38 [NELSON and AHRR (cg05575921) methylation <55%] to 27 [NLST and AHRR (cg05575921) methylation <55%; Fig. 3]. NLST and AHRR (cg05575921) methylation <55%, was considered the best model, and led to a reduction of the screening burden of 21.9% and a significantly higher specificity (84.0%; P < 0.05) and only slightly reduced sensitivity (62.0%) compared with the use of the NLST alone.

This model was validated in the 2001 to 2003 population. Among 5,334 individuals, 54 developed lung cancer within 5 years after examination. Among the 745 individuals fulfilling the NLST criteria and with AHRR (cg05575921) methylation extent <55%, 33 developed lung cancer within 5 years. This was higher than for the individuals not fulfilling these criteria (Fig. 4, P = 3 × 10^–24, log-rank). The sensitivity was 64.8 (95% CI, 50.6–77.3) for the NLST criteria alone and 61.1% (95% CI, 46.9–74.1) for the NLST criteria and AHRR (cg05575921) methylation extent <55%, whereas the specificities were 82.1 (95% CI, 81.1–83.1) and 86.9% (95% CI, 86.0–87.8), respectively. The reduction of the screening burden was 24.3%, when comparing the NLST criteria and AHRR (cg05575921) methylation extent <55% with the use of the NLST criteria alone.

In this sample from the Copenhagen general population, we found that different eligibility criteria for selecting individuals for lung cancer screening led to substantial variations in the number of eligible individuals identified and various specificities (77.6%–82.9%) and sensitivities (42.4%–77.2%), and that the NLST criteria led to the lowest number of included individuals per 5-year lung cancer. Combining the NLST criteria with AHRR (cg05575921) methylation <55% led to a further reduction of the screening burden of 21.9% compared with the sole use of the NLST criteria alone. These results were largely validated in the 2001 to 2003 examination of The Copenhagen City Heart Study.

These novel results points towards a potential clinical use of AHRR (cg05575921) methylation to provide additional predictive risk information to identify eligible smokers for lung cancer screening and support prior evidence on the association of AHRR (cg05575921) methylation and lung cancer risk (26).

Improving the NLST screening specificity by additional use of AHRR (cg05575921), methylation only missed two individuals who did develop lung cancer within 5 years. This corresponds to 6% of the cancers, which would have been included using the criteria NLST alone. Before implementing such a scheme, its value must be confirmed prospectively within an actual screening program.

More narrow eligibility criteria likely improve the balance between benefits and harms, which in turn might raise the general acceptance of lung cancer screening and hence participation (31). This will impact quality and value of any screening program. An important dimension of benefits and harms considerations is cost effectiveness. AHRR (cg05575921) methylation extent is a cost-effective measurement compared with lung CT scanning: blood sample could be drawn at the general practitioner, and the method is easily standardized and implemented in laboratory medicine departments. AHRR (cg05575921) methylation could supplement eligibility criteria to identify those who will benefit the most from screening. Knowledge on how fluctuations of AHRR (cg05575921) methylation correspond to changes in smoking behavior are scarce (32–33), but our results indicate that a single AHRR (cg05575921) methylation measurement provides relevant 5-year lung cancer risk information on top of self-reported smoking and age. However, longitudinal studies of AHRR (cg05575921) methylation and smoking behavior are warranted, to further understand the stability of AHRR (cg05575921) methylation and its reversion after smoking cessation, as a long-term biomarker smoking behavior, and thus of lung cancer risk.

The mechanism on how smoking reduces AHRR methylation is not fully understood (34). When cells are exposed to toxins, for example, polycyclic aromatic hydrocarbons (PAH) from cigarette smoke, the AHRR gene interacts with the aryl hydrocarbon receptor (AHR) that mediates the metabolism of toxins (35). Hypomethylation in the cg05575921 site results in an upregulation of the AHRR gene expression in blood (36) and lung tissue (19) that inhibits AHR, and thereby the metabolism of toxins. Whether cg05575921 hypomethylation has any direct causal role in the development of lung cancer is unknown (37).

Strengths of this study was the prospective cohort from the Danish general population with detailed and almost complete information on various covariates, complete follow-up on lung cancer, and no misclassification of mortality. Lung cancer diagnoses in this unscreened population were based on histologic confirmation by a fully trained pathologist and registration is compulsory by law, thus excluding false-positive findings. Further, the best model was reproduced within the 2001 to 2003 examination. Some limitations must, however, be acknowledged. AHRR (cg05575921) methylation levels are affected by other intrinsic and environmental factors than tobacco smoking exposures (38). We did not include measurements of other methylation loci affected by smoking (17–25), as these data are not available. Thus, it is unknown if adding those to the AHRR (cg05575921) methylation provide an even better indicator of the future risk of lung cancer. Nine hundred twenty-four (9.1%) individuals from the 1991 to 1994 population did not have an available methylation measurement. However, these represented a nondifferential sample according to lung cancer. Also, individuals were not actually screened with low-dose CT scan, thus we do not know how specificities and sensitivities would have been in a real-world setting with ongoing screening. Presented results are therefore very conservative since screening would not be expected to detect all lung cancers, which surface clinically within 5 years after examination. Also, possible fluctuations in AHRR (cg05575921) methylation during the 5-year follow-up could not been taken into account. Performance evaluation of screening procedures is population dependent as more individuals must be screened to identify one lung cancer in a lower risk population, for example, a younger population (39). On the other hand, extending the upper age limit for eligibility can be associated with a lack of screening participation (40) and reduced screening benefit due to competing comorbidities (41). In this study, we examined a cohort from the 1990s whose overall lifetime duration of smoking and smoking intensity is somewhat higher compared with current populations, especially among women (42). The overall cumulative lung cancer incidence was 1% within 5 years. The 1991 to 1994 study population represent 56% of the actual invited population, and for example, smoking-related death and severe diseases may have prevented some from being examined. The potential self-selection in our study can affect the balance between sensitivity and specificity, but not explain the additional predictive risk information from AHRR (cg05575921) methylation since all eligible populations were obtained from the same sample.

Implementation of CT based lung cancer screening is now recommended throughout Europe (43). Adding AHRR (cg05575921) methylation on top of current eligibility criteria for lung cancer screening improves the specificity of lung cancer screening, by excluding those individuals with the lowest lung cancer risk from the eligible population.

No disclosures were reported.

K.K. Jacobsen: Conceptualization, data curation, formal analysis, supervision, funding acquisition, validation, methodology, writing–original draft, project administration, writing–review and editing. P. Schnohr: Writing–original draft, writing–review and editing. G.B. Jensen: Writing–original draft, writing–review and editing. S.E. Bojesen: Conceptualization, data curation, formal analysis, supervision, funding acquisition, validation, methodology, writing–original draft, project administration, writing–review and editing.

We acknowledge participants and teams of the Copenhagen City Heart Study. This work was supported by the Independent Research Fund Denmark, Medical Sciences, Region Hovedstaden, Chief Physician Johan Boserup and Lise Boserup's Fund, and Frederiksberg Hospital, all grants awarded the Copenhagen City Heart Study. The salary of K.K. Jacobsen was supported by grants from Dagmar Marshalls Fund and the Harboe Foundation awarded to K.K. Jacobsen. Grant/award numbers are not applicable.

The costs of publication of this article were defrayed in part by the payment of page charges. This article must therefore be hereby marked advertisement in accordance with 18 U.S.C. Section 1734 solely to indicate this fact.