Saliva DNA Methylation Detects Nascent Smoking in Adolescents

Abstract

Objectives: Early identification of smoking, essential for the successful implementation of interventions, arrests the escalation of smoking and smoking-associated risk behaviors in adolescents. However, because nascent smoking is typically episodic and infrequent, enzyme-linked immunoassay reagent-based approaches that detect cotinine, a key nicotine metabolite, are not effective in identifying adolescents in the earliest stages of smoking. Epigenetic methods may offer an alternative approach for detecting early-stage smokers. In prior work, we and others have shown that the methylation status of cg05575921 of whole-blood DNA accurately predicts smoking status in regularly smoking adults and is sensitive to nascent smoking. Yet, the blood draws necessary to obtain DNA for this method may be poorly accepted by adolescents. Saliva could be an alternative source of DNA. However, the ability of saliva DNA methylation status to predict smoking status among adolescents is unknown.

Methods: To explore the possibility of using salivary DNA for screening purposes, we examined the DNA methylation status at cg05575921 in saliva DNA samples from 162 high school aged subjects for whom we also had paired serum cotinine values.

Results: Overall, the reliability of self-report of nicotine/tobacco use in these adolescents was poor with 67% of all subjects whose serum levels of cotinine was ≥2 ng/mL (n = 75) denying any use of nicotine-containing products in the past 6 months. However, the correspondence of the two biological measures of smoking was high, with serum cotinine positivity being strongly correlated with cg05575921 methylation (p < 0.0001). Receiver operating characteristic (ROC) analyses showed that cg05575921 methylation status could be used to classify those with positive serum cotinine values (≥2 ng/mL) from those denying smoking and have undetectable levels of cotinine.

Conclusions: We conclude that saliva DNA methylation assessments hold promise as a means of detecting nascent smoking.

Introduction

Smoking is the leading preventable cause of death in the United States, and is responsible for the majority of the 25-year decrease in life expectancy of those with major mental illnesses (Parks et al. 2006; DeHay et al. 2012; Centers for Disease Control and Prevention 2015). In adolescents, both externalizing and internalizing problems predict the onset of tobacco use (Green et al. 2018). Efforts to change that dynamics include the prevention of the initiation of smoking, and the promotion of interventions to arrest the escalation of smoking and treatment of regular dependent smoking. Because genetic factors play a key role in shaping smoking behaviors (Sartor et al. 2015) and smoking cessation is often ineffective (Halpern et al. 2018), many have come to believe that little can be done to alter the onset of adolescent smoking. However, the same genetically informed designs that demonstrate the importance of genetic factors in smoking have also shown interactions between social environment and genetic influences (DeHay et al. 2012). This suggests the likely mutability of many heritable effects on the development of smoking through carefully targeted family-based prevention programs.

Preventing the escalation of periodic cigarette use to regular dependent smoking is likely to be particularly valuable because this would interrupt smoking and smoking-related behaviors before illness or economic consequences occur. Once regular dependent smoking has begun, it is very difficult to stop, with one recent study showing that treatment as usual has only a 0.1% chance of achieving abstinence at 6 months (Halpern et al. 2018). As a result of this high failure rate, once fully established, smoking behaviors are stable across many years, leading smokers to experience multiple cycles of quitting and relapsing even if they finally achieve complete abstinence (Berry et al. 2000). Because efforts to quit smoking often fail, individuals may develop irreversible medical consequences before they are able to quit. These considerations support the belief that the best way to stop smoking is to intervene early in its course before it becomes established, that is, during the phase of nascent, irregular smoking. Thus, the development of detection methods for identifying individuals during this earliest smoking phase is of critical importance to future prevention efforts.

The most efficient way to identify those early-stage smokers is by screening adolescents. Approximately 90% of all adult smokers report that they initiated smoking in their youth (Bonnie and Kwan 2015). Typically, their journey of addiction began with initial puffs on cigarettes, followed by experimental smoking (irregular, contextually dependent periods of smoking), followed by regular smoking (regular, but context-specific smoking), and then finally dependent smoking (daily smoking in response driven by psychological or physiological cues) (Shadel et al. 2000). Given tobacco industry pressure and the availability of nicotine-containing products in the adolescent environment, nearly all adolescents are potentially at risk for dependent smoking. Therefore, all children should be targets for educational measures to prevent initiation of tobacco product use.

Fortunately, the majority of adolescents will not advance to regular cigarette use and thus will not suffer the deleterious health consequences of smoking. However, medical sequelae are not the only concern for teenagers who initiate smoking. Prospective studies have shown that teen onset of smoking is associated with a constellation of other risk behaviors. For example, early smoking is the most powerful predictor of drunk driving in early adulthood, and early smokers have a 50% greater likelihood of engaging in early high-risk sexual behavior resulting in pregnancy (Berry et al. 2000; Hanna et al. 2001; Riala et al. 2004). For adolescents with attention deficit/hyperactivity disorder, cigarette smoking is also highly correlated with cannabis use (Gray et al. 2011). Literature reviews have concluded that interventions targeting adolescent smokers can double the cessation rate compared with no intervention control groups and maintain this efficacy for more than 1 year, while pharmacotherapies for nicotine dependence in adolescents also show promise (Sussman et al. 2006; Hammond and Gray 2016). Unfortunately, identifying adolescents in the earliest stages of tobacco use is difficult.

Currently, the preferred method for ascertaining adolescent smokers for potential inclusion in prevention programs is through self-report. Although many adolescents may truthfully report use, research has documented the high rate of unreliable self-reports of tobacco use (Caraballo et al. 2004; Jarvis et al. 2008). For example, using data from the NHANES study and serum cotinine value of 10 ng/mL as a cutoff, Caraballo et al. (2016) reported that only two-thirds of the youth who had positive serum cotinine values reported smoking in the past 5 days. Hence, using self-report as the method for ascertaining clients for treatment is not likely to identify the majority of adolescents who are beginning to smoke.

Unfortunately, there is no consensus on the best biological method through which to identify early-stage smokers. However, in adults, two approaches, carbon monoxide (CO) assessments and cotinine determinations, have been commonly used to assess smoking status (Florescu et al. 2009). The first, CO assessment, suffers from a lack of sensitivity in periodic or light smokers because of the short half-life of CO in blood, which limits their detection window to between 2 and 6 h after last use (Florescu et al. 2009). Hence, CO assessments have very limited utility in detecting the initial periods of smoking in adolescents.

In contrast, serum, urine, or salivary levels of cotinine, a metabolite of nicotine, have found greater utility (Gorber et al. 2009). In adults, cotinine assessments for the detection of regular adult smokers are extremely sensitive. However, because of the use of nicotine for cessation therapy and relapse prevention, as well as from false positives from the increasingly common practice of “vaping” (or the use of noncombustible forms of tobacco), these tests often have poor specificity.

In adolescents, false positives from the use of nicotine replacement therapies for smoking cessation treatment are uncommon. Since any use of a nicotine-containing product by adolescents should be of concern for clinicians, false positives from vaping or chew are more of a problem for researchers. Instead, the biggest limitation to the clinical use of cotinine assessments in adolescents is their lack of sensitivity. In one study, 37.4% of subjects who reported that they smoke less than weekly had detectable levels of cotinine (Jennifer et al. 2008). This lack of sensitivity is a consequence of the 20-h half-life of cotinine in blood and the low total number of cigarettes smoked in this earliest stage of smoking (Gorber et al. 2009). As a result of these challenges, tandem use of complementary technologies that can indicate smoking in the absence of recent smoking could lead to increased identification of adolescents who could benefit from therapy.

The use of DNA methylation technologies may offer this complementary approach to the assessment of smoking status in adolescents. It has become increasingly clear that methylation status at a cytosine phospho guanine (CpG) dinucleotide pair in the aryl hydrocarbon receptor repressor gene referred to as cg05575921 is a sensitive and specific indicator of smoking status in adults (Andersen et al. 2015; Gao et al. 2015). Recently, using a newly developed droplet digital polymerase chain reaction (ddPCR) assay of methylation at this locus and DNA from whole blood, we have shown a excellent sensitivity and specificity in classifying adult smokers versus nonsmokers, with a receiver operating characteristic (ROC) area under the curve (AUC) of 0.99 as well as a dose-dependent relationship of cg05575921 to average daily cigarette consumption (Philibert et al. 2018).

For the detection of smoking in adolescents, the use of saliva DNA for a ddPCR assay of cg05575921 methylation would be an attractive possibility, given the reluctance of many adolescents to undergo venipuncture. Indeed, results from other groups using genome-wide methylation assays and multilocus pyrosequencing techniques in salivary and buccal samples suggest the feasibility, although results of those studies indicated lower sensitivity in detecting smoking than our blood-based ddPCR assay (Hussain et al. 2018; Teschendorff et al. 2015). However, technical challenges in the use of saliva DNA for ddPCR assays present obstacles in the translation of this technology to a point-of-care test suitable for screening adolescent populations.

First, admixture of buccal cells and white blood cells (WBCs) in saliva requires that the proportion of each cell type be controlled for because methylation set points for loci such as cg05575921 may be markedly different in buccal cells compared with white blood cells (Genotek 2013; Lowe et al. 2013; Cianga et al. 2016). Fortunately, as Jones et al. (2017) have shown, it is possible to calculate the relative tissue contribution of a DNA sample of mixed cellular origins, such as saliva, by assessing methylation status at differentially methylated regions (DMRs), whose epigenetic status in specific tissues has been determined.

A second obstacle in the use of salivary DNA is degradation due to the acidic conditions of the saliva and exposure to extracellular DNase (Philibert et al. 2008; Zhang et al. 2016). Because methylation-sensitive ddPCR assays depend on the presence of large numbers of intact DNA contigs of 150–250 bp, determination of DNA quality through the use of intercalating dyes such as PicoGreen has been recommended (Ahn et al. 1996). However, in the case of DNA prepared from saliva samples, admixture with bacterial cells containing double strands renders the results of intercalating dye assessments more difficult to interpret. To date, there has been limited information on the suitability of saliva DNA for ddPCR assays for loci such as cg05575921 that could be used to detect smoking.

In this communication, we demonstrate the ability to overcome the above challenges in the use of salivary DNA in detecting smoking in adolescents using a newly developed ddPCR assay. We demonstrate that adolescent smoking, as indicated by serum cotinine levels, can be detected with a high degree of sensitivity and specificity using a ddPCR assay for methylation at cg05575921.

Methods

Participants and clinical characterization

The Healthy Iowans Study is a longitudinal study of substance use and other risk behaviors in high school sophomores designed to further our understanding of the trajectories of tobacco and cannabis use in adolescence. All procedures and protocols used in the study, which included the acquisition of an NIH Certificate of Confidentiality and the use of bilingual interview procedures, were approved by the University of Iowa Institutional Review Board (IRB 201409705).

The subjects in this study were drawn from seven high schools in Eastern Iowa (Table 1). As a first step, a complete description of the study was provided to prospective high schools in the Iowa City region and an appointment was made with the designated school official to discuss the study. If the district was willing to participate, the publically available names, addresses, and phone numbers of each high school sophomore in that district were then obtained. Using that information, an introductory letter and a study summary were sent to each student. Within 2–5 days following the arrival of the letter, each student was called by study staff to inquire about the student's interest in participating in the study. If interested, an appointment was scheduled for the student and at least one parent/guardian. At that visit, a full presentation of the study was made, parents or guardians provided written consent for themselves and their child, and the adolescents provided written assent. In the case of subjects whose parents were not fluent in English, written consent was obtained in Spanish by a bilingual staff member. As part of the consent procedures, adolescents and their parents or guardians were informed that results of blood tests indicating the presence or absence of substance would only be available to study staff in a deidentified form, and that any results would not be shared with the adolescent, their parent or guardian, or any health care providers.

Table 1.

Key Characteristics of Subjects

	Nonsmokers	Newly positive	Previously positive
Number of subjects	90	57	15
Age at intake	15.6 ± 0.6	15.9 ± 0.8	15.9 ± 0.6
Saliva sample used
Intake	90	26	—
One year	—	21	9
Two year	—	10	6
Gender
Male	51	25	8
Female	39	32	7
Ethnicity
White	65	40	9
AA	9	8	2
White Hispanic	12	4	2
Other	4	5	2
Self-report
No nicotine	90	39	9
Smoking	0	15	6
Other nicotine onlya	0	3	0
Cotinine level at sampling time
<1 ng/mL	90	—	—
<2–5 ng/mL	—	34	6
5–25 ng/mL	—	10	1
>25 ng/mL	—	13	8

^aIncludes e-cigarettes, vaping, and noncombustible consumption of tobacco.

After consent, each student was interviewed with an abbreviated child version of the Semistructured Assessment for the Genetics of Alcoholism and the Substance Use Interviews in private by a trained research assistant (Kuperman et al. 1999; Philibert et al. 2014) (Supplementary File 1). After the interview was complete, samples of blood and saliva were taken to provide the biomaterials for this study. Then, every 6 months, each subject was reinterviewed either via phone (at 6 and 18 months) or in person (at 1 and 2 years) with the Substance Use Questionnaire that examines consumption of substances over the past year and specifically asks whether an individual has consumed any of the given substance over the past 6 months.

Biological measures

Venipuncture for whole blood and serum samples and saliva sampling were performed at each of the intake, 1-, and 2-year time points. Saliva DNA was prepared as per the directions of the manufacturer of the saliva kit (Isohelix, United Kingdom) and then frozen at −20°. Sera and DNA samples were prepared from blood specimens as previously described and then frozen at −80°C until usage (Philibert et al. 2016). Serum cotinine values were determined as previously described using enzyme-linked immunoassay reagents from Abnova (Taiwan) and a Molecular Devices (Sunnydale) EMax spectrophotometer (Philibert et al. 2016). Methylation status at cg05575921 and DMR16 (DMR on Chromosome 16) was determined using methylation-sensitive ddPCR (Philibert et al. 2018). Digital PCR techniques such as ddPCR quantify DNA variants, for example, the proportion of alleles of a certain genotype or the proportion of CpGs that are methylated versus unmethylated, by fractionating the DNA sample into a large (>10,000) number of discrete compartments, amplifying the DNA in each fraction by PCR, and conducting statistical analysis on the distribution of fractions positive or negative for a given probe corresponding to the variant of interest to determine the starting quantity of that variant (Hindson et al. 2013).

In brief, 1 μg of DNA from each subject was bisulfite converted using an EpiTect Fast 96 Bisulfite Conversion kit (Qiagen, Germany). An aliquot of the resulting nucleic acid solution was preamped, diluted 1:3000, and then PCR amplified using fluorescent, dual-labeled primer probe sets specific for cg05575921 and DMR16 from Behavioral Diagnostics (Coralville, IA) through their distributor IBI Scientific (Dubuque, Iowa) and Universal Digital PCR™ reagents and protocols from Bio-Rad (Carlsberg, CA). The number of droplets containing amplicons that have at least one “C” allele (corresponding to the methylated cytosine residue), one “T” allele, one “C” and “T” allele, or no amplifiable alleles was determined using a QX-200 droplet counter and Bio-Rad's proprietary QuantaSoft™ software, and expressed as a percent methylation.

To correct for admixed buccal and WBC cells in saliva, WBC proportion was determined using a ddPCR assay for methylation status at DMR16, according to the manufacturer's instructions. The proprietary DMR16 assay measures methylation at a chromosome 16 locus that is highly methylated in WBC (95%) but not in buccal cells (18%). Relative contribution of white blood cell contribution (X) to the total DNA sample was determined by solving the equation of DMR16^obs = [0.95X + 0.18(1-X)] where DMR16^obs is the observed methylation signal in the saliva sample, and 0.95 and 0.18 are the fractional methylation values of DMR16 in white blood cells and buccal cells, respectively.

Using a standard RNAse P copy number assay, we assessed the effective concentration of PCRable contigs present in saliva DNA compared with those in whole blood as previously described (Philibert et al. 2008). The resulting data were then compared with genomic DNA standards and relative copy number/effective DNA concentration of the saliva DNA solutions imputed as described by Livak and Schmittgen (2001).

Subject classification

The goal of classification was to distinguish between subjects who had no evidence of smoking across the 2-year study period (“controls”) from those who had evidence of smoking at one time point (“newly positive” and more than one time point (“previously positive”). At study intake, subjects were classified as “controls” if they denied any lifetime use of tobacco products, other nicotine-containing products such as e-cigarettes, or cannabis products and had negative (≤1 ng/mL) serum cotinine value. Subjects positive for any self-reported history of tobacco, nicotine, or cannabis products or who had a positive serum cotinine (≥2 ng/mL) at intake were classified as “newly positive.” At the 1- and 2-year follow-up visits, any subjects who were previously classified as controls before that visit, but had a positive (≥2 ng/mL) serum cotinine value, were reclassified as “newly positive.” Subjects with a prior classification of a “newly positive” who demonstrated a subsequent positive (≥2 ng/mL) serum cotinine at the 1- or 2-year follow-up visits were reclassified as “previously positive.” In total, each subject received a single classification of either “control,” “newly positive,” or “previously positive” reflecting the overall status across the study period, with no subjects being shared between groups.

Data handling and plan of analysis

Data for the study were double entered into a secure database with field restrictors. The resulting entries were then inspected and cleaned to ensure accuracy. Please see Supplementary File 2 for all data used in this study.

Saliva samples chosen for analysis were based on the time the subjects converted to their final classification, either intake, year 1, or 2. For controls, these were all at the intake time point by definition, whereas “newly positive” subjects could have converted to that status at any of the three time points. For “previously positive” subjects, conversion to that status could only occur at the 1- and 2-year time points. For each subject, the serum samples selected for cotinine assay were those taken from the same time as the saliva.

All data analyses were conducted using JMP Version 10 (SAS Institute, Cary, NC). The significance of differences in group means was calculated by analysis of variance (ANOVA). The primary analyses were conducted using logistic regression in which the outcome variable was group status, either controls versus “newly positive” subjects or controls versus “previously positive” subjects, with and without the addition of controls for cellular heterogeneity and gender.

The relative utility of logistic regression findings in classifying group status was quantified by ROC AUC (Fawcett 2006). ROC analysis is a method of analyzing the performance of classifier, for example, a biomarker with continuous values such as glycosylated hemoglobin in the diagnosis of diabetes mellitus. ROC curves can show the effect that changing the cutoff value for a classifier has its sensitivity and specificity, with a greater total AUC generally indicating better sensitivity and/or specificity for the classifier. Biomarkers with AUCs of over 0.90 are generally held to be “excellent,” over 0.80 “good,” over 0.70 “fair,” and under 0.70 “poor” (Youngstrom 2013).

Results

Subject classification and serum cotinine assays

Clinical and demographic characteristics of the subjects are given in Table 1. At intake, subjects were mostly White and between 15 and 17 years of age, with a slight preponderance (52%) of subjects being male. A total of ninety subjects were classified as controls at intake and retained that classification over the 2-year study period. Of the 57 subjects who received a classification of “newly positive,” 26 were positive at intake, 21 became positive at the 1-year follow-up visit, and 10 became positive at the 2-year follow-up visit. Of the 15 subjects classified as “previously positive,” 9 had their second positive serum cotinine occur at the 1-year follow-up visit and 6 at their 2-year visit. The majority of both the “newly positive” (68%) and “previously positive” (60%) groups denied any smoking or use of nicotine-containing products throughout the study period.

By definition, subjects classified as controls had negative (≤1 ng/mL) serum cotinine values at each of the three study time points. The mean cotinine value in the “newly positive” group was 29 ± 49 ng/mL, with the substantial majority (60%) of the subjects having a serum level between 2 and 5 ng/mL. In the 15 subjects classified as “previously positive,” serum cotinines showed an upward trend over time, with an average of 36 ± 59 ng/mL at the point of the first positive sampling and 76 ± 73 ng/mL at the second positive sampling.

Nearly half (46%) of “newly positive” saliva samples were obtained from subjects at intake, with the rest obtained at their 1-year (n = 21) or 2-year follow-up visits (n = 10), depending on which visit demonstrated a positive serum cotinine. With respect to the “previously positive” group, the majority (n = 9) of the saliva DNA samples were obtained at their second visit with the remainder (n = 6) obtained at their third visit.

Salivary DNA quality control and cell mixture analysis

The availability of intact DNA contigs in the saliva for PCR amplification was adequate and did not significantly vary between classification groups. As shown in Figure 1, both the “nonsmoking” and “newly positive” groups had ∼90% of the effective copy number of the genomic whole-blood comparator, with the “previously positive” groups having ∼75% of copy number of the whole-blood standard (all comparisons p > 0.05, Tukey–Kramer honest significant difference). The high level of PCRable contigs seen across groups demonstrated a more than adequate quantity of nondegraded DNA available in the saliva samples for bisulfite conversion and amplification, equal to ∼13,000 complete genomes per subject available for assay.

FIG. 1.

The percentage of amplifiable DNA in the saliva DNA samples by group compared with a whole-blood genomic DNA standard. The center bar of the “diamond” for each sample represents the sample mean, the tips of each diamond represent the bounds of the 95% confidence interval of the sample mean, while the bars between the tips and the center bar represent the region bounded by two standard deviations of the sample. Sample size for each group: newly positive = 57, previously positive n = 15, control n = 90. Color images are available online.

The proportion of DNA of salivary WBC versus buccal cell origin was calculated to allow for correction of cellular heterogeneity in subsequent analyses. Overall, ∼65.5% ± 17.5% of all amplifiable human DNA was contributed by WBCs. There was a trend for smokers to have lower WBC contribution compared with controls (62.5% ± 19.4% vs. 67.9 ± 15.5, t-test p < 0.06; see Fig. 2).

FIG. 2.

The percentage of the human DNA that is contributed by WBCs for each DNA sample by group as inferred by DMR16 methylation status. Overall, WBCs contributed 65.5% ± 17.5% of the total amplifiable DNA present in the saliva samples. Modal contribution was 69%. Total sample size: n = 162. WBCs, white blood cells. Color images are available online.

Relationship of saliva DNA methylation to serum cotinine and classification status

Next, we analyzed the relationship of DNA methylation to group classification to assess its utility in detecting smoking status, as indicated by serum cotinine. As shown in Figure 3, control subjects had significantly greater average saliva DNA methylation at cg05575921 than subjects in the newly positive or previously positive groups (p < 0.0001, ANOVA n = 162). Not surprisingly, unadjusted saliva cg05575921 DNA methylation (i.e., not corrected for differences in cellular composition) was lowest in those samples from the previously positive group, presumably corresponding to those individuals with the greatest cumulative smoke exposure.

FIG. 3.

The percent methylation at cg05575921 of each saliva DNA sample by group unadjusted for cellular heterogeneity. The center bar of the “diamond” for each sample represents the sample mean, the tips of each diamond represent the bounds of the 95% confidence interval of the sample mean, while the bars between the tips and the center bar represent the region bounded by two standard deviations of the sample. Sample size for each group: newly positive = 57, previously positive n = 15, control n = 90. ANOVA, analysis of variance. Color images are available online.

The relationship of saliva DNA methylation status at cg05575921 to classification status was analyzed using logistic regression and receiver operator curve techniques. Figure 4A provides a visual representation of a logistic regression analysis of methylation at cg05575921 as a highly significant (p < 0.0008) predictor of classification in the control and “newly positive groups.” Using just cg05575921 as a classifier, the ROC AUC with respect to classification status was 0.66 (Fig. 4B). Adjustment for cellular heterogeneity using the DMR16 signal, controlling for gender and exclusion of the three “newly positive” subjects who stated that they only used noncombustible sources of nicotine, did not improve the AUC.

FIG. 4.

Results of comparison of methylation between the control (n = 90) and newly positive (N = 57) groups. (A) Logistic plot of data. (B) ROC plot using only cg05575921 as a classifier. ROC analysis is a method of analyzing the performance of a classifier that can show the effects that different cutoff values for the classifier have sensitivity and specificity. The ROC curve is indicated by the black line. A greater total AUC of the ROC plot generally indicates better sensitivity and/or specificity for the classifier. Points where the ROC curve intercepts the green tangent line indicate the highest ratio of true positives to false positives for the classifier. ROC, receiver operating characteristic; AUC, area under the curve. Color images are available online.

Figure 5A–C illustrates the results of similar logistic regression analysis using cg05575921 saliva DNA methylation to predict classification status in the control and “previously positive groups,” with and without controls for cellular heterogeneity and gender. Once again, cg05575921 was significantly associated with classification status (p < 0.002). The AUC for classifying just the “previously positive” subjects with respect to controls was markedly better compared with controls versus “newly positive subjects,” as would be expected given the presumably greater smoke exposure in subjects with positive serum cotinine at multiple time points. Results of ROC analysis resulted in an AUC of 0.70 using just cg05575921 as a classifier. However, in contrast to the prior comparison, inclusion of DMR16 methylation status to compensate for cellular heterogeneity did provide an increase in the AUC to 0.79. Inclusion of gender had no effects on classification performance.

FIG. 5.

Results of comparison of methylation between the control (n = 90) and the previously positive (N = 15) groups. (A) Logistic plot of data. (B) ROC plot using only cg05575921 as a classifier. (C) ROC curve plot using both cg05575921 and DMR16 methylation as classifiers. ROC analysis is a method of analyzing the performance of a classifier that can show the effects that different cutoff values for the classifier have sensitivity and specificity. The ROC curve is indicated by the black line. A greater total AUC of the ROC plot generally indicates better sensitivity and/or specificity for the classifier. Points where the ROC curve intercepts the green tangent line indicate the highest ratio of true positives to false positives for the classifier. ROC, receiver operating characteristic; AUC, area under the curve. Color images are available online.

Discussion

In this study, we demonstrated the ability to overcome technical challenges involved in the use of salivary DNA for the detection of epigenetic changes due to smoking in adolescents, and “good” performance characteristics using a novel digital PCR assay for methylation of cg05575921, an emerging epigenetic biomarker for smoking. Receiver operating curve analysis yielded an AUC of 0.79 in distinguishing between nonsmoking controls and subjects with serum evidence of nicotine exposure at multiple time points after correction for cellular heterogeneity, with weaker results seen when distinguishing between controls and subjects with positive serum cotinine at only one time point. Our results are particularly interesting given that the majority of subjects with serum evidence of nicotine exposure denied any use of tobacco or nicotine-containing products such as e-cigarettes, indicating the need for better screening tools in adolescent populations.

Although previous experiences with saliva demonstrated significant degradation of genetic material to the point that some samples had no detectable intact DNA (Philibert et al. 2008), our results indicate that the quality of DNA available using newer kits was more than adequate. Similarly, we were able to overcome technical issues related to the cellular heterogeneity of saliva samples using a cell-type-specific digital PCR marker as a control in our analyses, leading to an improved AUC in distinguishing controls from “previously positive” subjects. Our finding that ∼70% of the usable DNA in adolescent saliva comes from WBC is consistent with prior findings from Cianga et al. (2016), Smith et al. (2015), as well as findings from DNAGenotek (Genotek 2013), a major manufacturer of saliva DNA kits. Going forward, the use of a larger number of markers to correct for cellular heterogeneity could offer further improvements in the performance of our assay (Houseman 2015; Smith et al. 2015).

The techniques demonstrated above offer multiple advantages over current screening methods for adolescent smoking. In contrast to blood-based testing, adolescents and their parents are more likely to agree to a saliva-based screen for nicotine use. Similarly, a test that be performed by a clinician during a routine office visit offers logistical advantages over a test requiring venipuncture performed by a trained phlebotomist. In contrast to the use of cotinine and other established biomarkers, smoking-induced demethylation of cg05575921 may be particularly better suited due to the detection of adolescent smoking due to its significantly longer half-life (Guida et al. 2015) since adolescent smoking is typically characterized by lighter and more irregular use patterns than dependent adult smoking. It is also worth noting that because cg05575921 demethylation is indicative of exposure to toxins contained in both tobacco and cannabis smoke (Moir et al. 2008; Tsay et al. 2013; Wei et al. 2016), this tool is an effective screening measure for both types of substance use, with a positive test indicating the need for further clinical evaluation, and potentially the use of more specific confirmatory biomarkers such as urine THC and/or cotinine.

From a technical standpoint, we note as an additional strength that digital PCR offers advantages over older methods of quantifying genetic material. For many years, quantitative PCR (qPCR) has been the laboratory standard for the determination of copy number of RNA or DNA. However, because the imputation of copy number via qPCR necessitates the need for external standards, qPCR approaches have an unavoidable source of error. In contrast, digital PCR approaches used by manufacturers such as Bio-Rad, Stilla, Fluidigm, and Applied Biosystems, generally partition the reaction solution into smaller volumes (such as chambers or droplets), which are subsequently individually PCR amplified, and then assessed via an “end read” for the presence or absence of a given allele (Wong et al. 2017). Because of the inherent ability of this approach to “count” the number of alleles in a given volume, these digital PCR techniques are commonly used as “reference-free” assays for library titration for high-throughput sequencing and are being used clinically for the initial diagnosis and detection of recurrence of cancer (Hindson et al. 2011; Lund et al. 2016; Sacher et al. 2016). In addition, the same Poisson statistical techniques used for copy number imputation also allow the “reference-free” calculation of relative ratios of alleles in a solution as we have done herein.

Finally, like qPCR, dPCR is also quick and can be relatively affordable. Because the estimated turnaround time of this test is anticipated to be less than 24 h when it is fully developed, and avoids the need for venipuncture or urine sampling, the ability to implement this test in a primary care setting is feasible, and because its anticipated cost is anticipated to be equal to or less than current biomarkers such as urinary cotinine, it is likely that the test will be covered by insurance companies in the future.

We also wish to note that discrepancies between adolescent self-report and smoking biomarkers do not necessarily indicate intentional deception. Although not formally assessed as part of our study protocols, study personnel noted that some adolescents endorsed the use of vaporizer devices without knowing that such devices contained nicotine. It is possible that some of the adolescents who participated in this study along with their peers in the community do not understand what substances they are consuming or misunderstand the constantly evolving jargon surrounding the use of nicotine atomizers. These experiences also serve to highlight the need to incorporate biomarkers in studies of adolescent substance use as much as possible as terminology, technology, and use patterns continue to evolve.

From a clinical standpoint, the question of how best to handle medical results that contradict a patient's self-report is a challenging one, no less so in adolescents than in adults. As in all cases, this situation demands a sensitive approach by the treating clinician with an adequate discussion of the limitations of diagnostic tests, as well as the need for repeat or confirmatory testing in some situations. Nonetheless, because adolescent smoking is common and conveys substantial future risks for smoking-related morbidity and mortality and other risk behaviors, positive results should prompt a heightened level of monitoring and referral to treatment as soon as warranted, and ideally before the development of nicotine dependence.

We note several potential study limitations and considerations for alternate methodologies. First, the number of subjects in the “previously positive” sample group is small (n = 15). We anticipate that on completion of the full study, including 450 subjects being followed over a 2-year period, we will be better powered to demonstrate the effects presented above and refine our estimates of sensitivity and specificity.

Second, the use of a relatively low 2 ng/mL cutoff to positively identify nicotine exposure conveys both advantages and disadvantages. The cutoff chosen is because it is consistent with the low and irregular smoking patterns of nascent smokers. A recent review of the literature and results from the National Health and Nutrition Examination Survey suggest a cutoff of 3 ng/mL for background smoking (Benowitz et al. 2009; Kim 2016). We opted for a slightly less stringent cutoff point to maximize sensitivity, although in the future, other studies wishing to replicate our results may choose different cutoffs. Nonetheless, the observation that our “previously positive” subjects had roughly double the mean serum cotinine as those in the “newly positive” group suggests that our cutoff was a reasonable one.

Third, specificity issues exist, as noted above, both in our use of cotinine and cg05575921 methylation as biomarkers for smoking. Serum cotinine is an indicator of exposure to any source of nicotine, whether from cigarette smoking, e-cigarettes, or even tobacco products admixed with cannabis such as “blunts,” while cg05575921 demethylation indicates exposure to smoke of any source, but is not sensitive to e-cigarettes, which do not contain the polyaromatic hydrocarbons that cue cg05575921 demethylation (Philibert et al. 2013). Although it is unlikely that second-hand smoke would lead to high levels of serum cotinine or more significant demethylation of cg05575921 seen in many subjects, controlling for household exposure in future analyses may improve prediction. We plan to address this by including environmental measures of our subjects' households, including parent reports of smokers in the home in future analyses. To distinguish between tobacco and other sources of nicotine exposure, such as e-cigarettes, we plan to incorporate the use of biomarkers more specific to tobacco such as anabasine, an oxidation product of nicotine only formed by smoking (Xu et al. 2004).

Finally, to increase the probability that subjects classified as “controls” are truly nonsmokers, in future analyses we may choose to restrict analysis of “control” subjects to only those subjects with three consecutive negative screens for both serum cotinine and serum THC. We anticipate that combining the above strategies will lead to increases in the mode-based estimates of the sensitivity and specificity of our saliva-based test in detecting adolescent smoking.

Conclusions

In conclusion, with replication of the findings above and further refinements in our technological approaches to assaying salivary DNA, we believe that our digital PCR assay for cg05575921 methylation may offer clinicians and researchers a sensitive and specific tool to identify individuals at the earliest stage of smoking. The ability to assay this biomarker in saliva offers significant advantages in screening adolescent populations for smoking, an important preventable cause of morbidity and mortality and a marker for other risk behaviors.

Clinical Significance

Because of the significant morbidity and mortality associated with smoking, particularly among those with mental illnesses, and limitations in current treatments for nicotine dependence, better tools to identify nascent smoking are needed to allow preventative interventions before the onset of nicotine dependence. Unfortunately, adolescents are often inaccurate when reporting on their smoking habits, and limitations in the currently available biomarkers, particularly low sensitivity in nascent smokers, make it difficult to identify those in need of interventions. Here we demonstrate the feasibility of using a saliva-based epigenetic assay to detect adolescent smoking, as confirmed by serum cotinine. Further development of this tool may allow clinicians to detect nascent smoking, interrupt the trajectory of escalating nicotine use and eventual dependence, and identify adolescents at risk for smoking-associated risk behaviors.

Authors' Contributions

R.P. obtained funding, conducted some of the analyses, and wrote the initial draft of the article. M.G. and F.X.G. also helped to obtain funding, edited the article, and developed key concepts. S.R.H.B. edited the article extensively and originated key concepts. K.D. wrote some parts of the article and performed a good chunk of the initial assays and measurements. E.P. supervised the execution of the project and contributed to editing of the article. W.P. and K.V. performed many of the laboratory procedures and assays, and contributed to editing of the article. A.A. provided technical support, clinical coverage, and finalized editing of the article.

Disclosures

Dr. R.P. is the Chief Executive Officer of Behavioral Diagnostics and inventor on a number of granted and pending patent applications with respect to both alcohol and tobacco consumption related to the material discussed herein. The use of cg05575921 status to determine smoking status is protected by US Patents 8,637,652 and 9,273,358.

Supplementary Material

Supplementary File S1

Supplementary File S2