In this longitudinal study, we demonstrate that cg05575921 demethylates in response to smoking in adolescents. For all three waves, the tobacco use status (defined as serum cotinine ≥ 2 ng/mL) and quantitative cotinine levels were significantly negatively correlated with methylation of cg05575921. Most notably, using data from the 19 subjects in W3 whom reported smoking cigarettes within the past week and had detectable cotinine values, we showed a negative relationship between quantitative cotinine levels and the degree of demethylation of cg05575921. Limitations of the findings include the high rate of unreliable self-report observed in the study, heterogeneity in the type of tobacco product consumed, and the single site, largely European ancestry cohort.

The high rate of unreliable self-report of nicotine use is not only consistent with some prior studies but also is informative for prospective study design. In this study, whose overarching goal was to understand the relationship between smoking and demethylation, 67% (W2) and 55% (W3) of the subjects positive for serum cotinine denied any use of nicotine containing products in the past 6 months. Since the bogus pipeline principle stipulates that telling subjects that they will be tested for substances should decrease the rate of unreliable self-report, we expected that informing each subject during the consenting procedures that we would be using testing for tobacco and cannabinoid metabolites; and testing for exhaled carbon monoxide levels at each phlebotomy visit would increase the level of reliable self-report [25, 26]. Nevertheless, we had a high, but not entirely unprecedented, rate of unreliable self-report. Using data from NHANES population collected over a decade ago, Caraballo and associates found that about 1/3 of the adolescents (12-17 years old) with serum cotinine levels > 10 ng/ml self-reported smoking [10]. Conceivably, if they would have used a lower cutoff level, even higher levels would have been observed. Taking the low cutoff into consideration, some of the false positives that we observed could have come from second hand smoke. In that regard, others have reported high rates of low background nicotine exposure in adolescents [27, 28]. It is also important to appreciate that many students who are vaping may not know that they were consuming nicotine. At the same time, we will note that among those in our study with positive serum levels, there is no relationship between the level of cotinine and veracity of self-report, which suggests that second hand smoke is not causing the majority of these discrepancies. Furthermore, in those with positive levels in earlier waves, there is a marked tendency for their cotinine levels to increase in subsequent waves which suggests that what we are detecting in those subjects is the progressive onset of a nicotine use disorder. Finally, we have not had a problem with false positives when using this ELISA methodology in our studies of adults. Therefore, we feel confident that the vast majority of the positive cotinine values that we observed are secondary to nicotine consumption by the subject. Still, no matter what explanation(s) for the high rate of unreliable self-report is correct, our findings suggest that projects such as ours whose goal is to understand the initial response thresholds of biomarker response in adolescents, may have difficulty obtaining accurate initial self-report of substance consumption.

The finding that the subjects were much more forthcoming with their self-report of cannabis use than their use of nicotine products was surprising. In contrast to the problems encountered with respect to nicotine use self-report, only 30% (W1) and 27% (W3), of those subjects with a positive cannabinoid level denied use of cannabis. Conceivably, this better rate of cannabis self-report could be due to shifting attitudes. Over the past decade, the use of marijuana for medical purposes has become progressively more socially acceptable. Conversely, the messaging with regards to nicotine use and, in particular, to smoking, has continued to increase in negativity. In a 2016 study of 786 California teenagers, Roditis and colleagues found that cannabis was consistently more positively viewed than cigarettes [29]. Unfortunately, these more positive attitudes towards cannabis may be misplaced [30] and our findings support the need for additional steps to educate adolescents on the potential harms of cannabis use [31].

These data suggest that the demethylation response to PAH occurs early in the course of smoking. In prior work in the Strong and Healthy African Americans Project (SHAPE) cohort, we showed clear demethylation at cg05575921 in a group of 42 eighteen-year-old subjects who reported less than a ½ pack-year of consumption (3600 cigarettes) [32]. The objective and subjective data from the current study support those prior findings. With respect to a purely objective approach, a simple linear fit of the objective data from the nineteen W3 subjects who reported some smoking in 6 months prior to phlebotomy shows that a serum cotinine level of 100 ng/ml corresponds to a methylation level of about 72%. According to the data of Caraballo and associates, this level of cotinine reflects the consumption of about 4 cigarettes per day. Assuming that these subjects smoked at that rate for a year, this would correspond to an annual average consumption of 0.2 pack years for those 5 subjects with cotinine levels ≥ 100 ng/ml. Still, the actual consumption of cigarettes that these cotinine levels represent may be lower. In our studies of adults who only smoked cigarettes that used this same ELISA assay, the same methylation level corresponded to ~ 2 cigarettes per day [15]. Furthermore, the majority of the 19 W3 subjects, whose data is given in Fig. 1, also reported using other sources of nicotine, it is likely that a substantial portion of the cotinine in these subjects may be from non-combustible sources. Subjectively, the number of cigarettes reported by those subjects was remarkably low. All totaled together, the 19 subjects report a combined consumption of less than 3400 cigarettes (n.b., 1 pack year = 7300 cigarettes in a year) over the 3 years prior to the W3 blood draw. If that figure is even remotely reflective of actual cigarette consumption that indicates that the actual threshold for demethylation may be relatively small, as little as a carton of cigarettes for some individuals. Given the challenges that we experienced obtaining reliable information, resolving the exact threshold for demethylation in adolescents may be difficult and thus achieving an accurate understanding of the relative effects of acute versus chronic smoke exposure on demethylation in adolescents may be impossible. If this is so, establishing an exact understanding of the dose and time dependency of initial demethylation may be a task best accomplished through use of a rodent or primate model.

An important finding of this study is that in the absence of smoking, the baseline methylation of subjects of both genders and all ethnicities at study entry (age ~ 15 years) is about the same (86.7% ± 3.2). This may be confusing to some because of Joubert and colleague’s prior demonstration that the newborns from mothers who smoked had lower methylation at cg05575921 [33]. However, we and others have shown that methylation at cg05575921 reverts as a function of smoking cessation [34–37]. Bauer has shown that in the absence of smoking, the smoking induced methylation changes at this locus in newborns fully reverts back to normal within 2 years [36]. The reversion speed in adults may be even faster. In 2016, we showed in a small group of subjects that the methylation levels at cg05575921 increased ~ 5% after 1 month of smoking cessation. Recently, we have confirmed and extended this finding in a larger set of subjects [38].

Our findings also support some prior studies that suggest that CO detectors are limited tools for screening for smoking in adolescent or adult smokers [39, 40]. In our study of smoking cessation in adults, we found that the same model of CO detector used in this study gave reading of 8 ppm or more in less than ½ of the subjects who smoked 5 or less cigarettes per day [34]. Hence, if those data are correct, using the default adult cutoff for CO in adult smokers [41], it is unlikely that the CO monitor would detect smokers in the earliest phases of smoking where smokers consume only 1 or two cigarettes per day. If a lower cutoff such as the 3 ppm suggested by Javors and colleagues is used, [42] many more smokers would be detected. But the rate of false positive detections would rise dramatically. At W3, 100 of the 287 cotinine negative subjects had readings of 4 ppm or greater. Therefore, we do not believe that altering the cutoff point is a viable solution for the use of this CO monitor in our hands.

However, we may never know the true utility of CO monitors or methylation tests in adolescents due of the challenges imposed by the increasing use of e-cigarettes or vape pens. In brief, to understand the sensitivity and specificity of a given device, one must have a gold standard for the trait being examined. In the past, cotinine determinations served as a gold standard for smoking. However, vaping produces much lower levels of PAH and CO than conventional cigarettes [43–45] and the rate of self-reported use of e-cigarettes/vape pens markedly increased from W1 to W3 in our population and in the nation overall during the time this study was conducted (2015-2018) [46]. It should also be noted that adolescents often do not know the content of the solutions that they are vaping. Because of these developments, cotinine determinations can no longer be considered as a gold standard for smoking in adolescents and we believe that many of the positive cotinine levels detected during this study do not reflect the use of combustible sources of nicotine. Therefore, many of the “positive cotinines” that we observed may not represent smoking and therefore could not be detected by a CO monitor or the methylation test.

These data also highlight the tendency of prolonged use for both nicotine and cannabis. Sixty percent (15 of 25) of the subjects who had a positive cotinine value at W1 and participated in W3, also had a positive cotinine value at W3. Similarly, 54% of the 13 subjects who had a positive cannabinoid level at W1 and participated in W3 had a positive W3 cannabinoid level. These data make it even more clear that once use of either of these substances has been initiated, it is likely to continue which highlights the need for prevention of initiation.

A large number of studies have examined the relationship between depression and substance use initiation [47]. Consistent with some but in contrast to others [48–51], our results show a modest yet significant relationship between depression and the later onset of both nicotine and cannabis use, but not the vice versa. However, the failure to show a relationship between smoking or cannabis consumption and the later onset of depression does not mean a relationship does not exist. Because there were only 38 subjects positive for cotinine and 20 cannabinoids at W1 power to detect an effect is rather limited. Instead, what our results highlight are the complex yet significant interrelationships between depression and substance initiation and the need to be vigilant for both of these highly prevalent conditions in adolescents.

Another unresolved issue is the relationship of vaping to the initiation of smoking. Conceivably, by testing the serum samples for combusted forms of tobacco metabolites such as 2-cyanoethylmercapturic acid (CEMA) or looking for non-nicotine-derived tobacco metabolites, we could gather additional insight as to which positive cotinine values result from tobacco use and which are secondary to vaping [52]. However, urine is the preferred substrate for this type of mass spectroscopy testing and it should be noted that even this cumbersome and expensive form of testing is not particularly sensitive to tobacco use in the early stages of smoking. The unreliability of self-report of nicotine initiation in this age group will pose a formidable challenge to those seeking to understand the initiation of nicotine use at its earliest stages. In future studies, these limitations could be overcome using a battery of biological measures that include metabolite testing and epigenetic measures, minimizing the effects of unreliable self-report on the development of a more comprehensive understanding of the relationship of vaping to the onset of smoking.

In conclusion, we demonstrated that demethylation of cg05575921 occurs early in the trajectory of smoking and that it may serve as a useful tool for understanding factors associated with the initiation or maintenance of smoking behaviors.