Saliva and Lung Microbiome Associations with Electronic Cigarette Use and Smoking

Smoking causes at least 18 different types of cancer and numerous other diseases (1). Electronic cigarettes (EC) have been purported to be a less harmful alternative nicotine delivery device for adult smokers who completely switch to ECs, although there are substantial concerns regarding an adverse impact on youth uptake (2, 3). Even though EC use has been associated with substantially lower exposure to cigarette smoke toxicants (4, 5), EC use has been hypothesized to have similar and also unique toxic effects compared to smoking, including an inflammation effect (6, 7). Understanding the effects of EC use on target organs including the lung and oral cavity is critically important.

There is a growing interest on the relationship of the human microbiome with carcinogenesis and other diseases (8, 9). Microbial dysbiosis (altered relative abundance of bacteria within bacterial communities—number and diversity of species) and increases in specific bacteria genera/species are hallmarks of numerous diseases, including lung disease and lung cancer (10–13). How microbial dysbiosis might affect lung disease and cancer risk is currently poorly understood, but hypothesized to occur through alterations in inflammation pathways (14–18).

Smoking has been well demonstrated to affect the oral microbiome (18–20), which are reversible with smoking cessation (21, 22). There are associations with changes in the oral microbiome affecting lung cancer and lung disease risk (13, 23). Many of the same oral microbiota have been reported as present in the lung (24, 25). It is hypothesized that oral microbiota migrate into the lung through microaspiration, and also via inhalation and dispersion along the mucosa (26–28). However, the relative abundance of bacteria may differ between the oral cavity and lung (25). Several studies indicated an impact of EC use on the oral cavity microbiome, but the impact of EC use has not been studied in the human lung (29–32).

We hypothesized that smoking cigarettes and EC use impact the lung and oral microbiome in similar ways. Accordingly, we conducted a microbial metatranscriptome study in the oral cavity and lung of 28 individuals to seek associations with smokers (SM) compared with never-smokers (NS), and for the first time report associations with EC use. Also compared with earlier studies of SM/NS, we used metatranscriptome RNA sequencing (RNA-seq) that allows for bacterial identification to the species level and detects live bacteria (33).

A cross-sectional study of 28 healthy participants, age 21–30, was conducted from 2015–2017 (Table 1). The participants were a subset of a larger study previously described chosen to balance gender, age, and sample availability (34). There were 10 NS, 10 EC users (eight were former SM), and eight SM. Using the Centers for Disease Controls guidelines of smoking, NS were classified as smoking fewer than 100 cigarettes in their lifetime (35) and not have smoked a single cigarette or used EC for at least 1 year. EC users were defined as those who reported daily and frequent EC use for ≥ 6 months and had not smoked a single cigarette for at least 6 months. SM were defined as those smoking ≥100 cigarettes in their lifetime, have been smoking at least five cigarettes per day and for at least 6 months, and had not used ECs within 1 year.

This study was approved by the Ohio State University (OSU) Comprehensive Cancer Center Clinical Scientific Research Committee and the OSU Institutional Review Board (IRB).

Participants underwent bronchoscopy at the OSU Research Center after an orientation visit and consent. Bronchoalveolar lavage (BAL) was performed with 5–6 sequential 20 mL aliquots of saline. Aspirated samples were collected at each aliquot, combined and transported to lab on ice. Samples were promptly aliquoted into six 1 mL aliquots, within 30 minutes of BAL sample collection and immediately spun at 10,000 g. The BAL supernatant was then removed, and the cell pellets for RNA-seq were resuspended in 1 mL of QIAgen RNAprotect cell reagent and stored at −80°C. To prepare for saliva collection, study participants were asked to refrain from brushing their teeth or chewing gum at least 90 minutes before their visit. Saliva samples were collected by having study participants spit into a 50 mL conical tube. The conical tube was then transported to lab on ice and the saliva transferred to a 15 mL conical tube with QIAGEN RNAprotect reagent. Samples were then promptly stored at −80°C.

RNA from saliva and BAL cell pellets were extracted using miRNeasy kit (Qiagen). After extraction, RNA concentration was measured using Nanodrop spectrophotometer (Thermo Fisher Scientific) and Qubit fluorimeter (Life Technologies). RNA concentrations ranged from 4 to 267.8 (ng/μL) when measured by Nanodrop, and 4.4 to 280 (ng/μL) when measured by Qubit (Supplementary Table S1). All samples were prepped for library generation.

Adequate RNA integrity was determined using the Bioanalyzer (Agilent). Total RNA was then treated with TruSeq Stranded Total RNA Gold (Illumina) to remove rRNA; cDNA was then synthesized from the remaining total RNA and the library was generated. Final concentration measurements were measured using Qubit, and concentrations ranged from 0.27 to 10.8 (ng/μL). RNA-seq was done on the Illumina Hi-seq 4,000 at 2×150 bps. Samples were sequenced to at least 35 million total reads per sample. Two saliva samples failed to produce enough total sequenced reads due to low concentration and could not produce minimum reads required. BaseSpace was used to demultiplexed reads, which produced read 1 and read 2 FASTQ files for each participant.

FASTQC was applied to FASTQ files to assess quality of reads. Reads with PHRED quality scores > = 30 were retained. Cutadapt (36) was applied to trim adapter sequence. The remaining reads were then aligned to the human genome (Hg19) by TopHat (37) version (v.) 2.1.0, using the following parameters: -p 8 –no-coverage-search –library-type fr-firststrand hg19.genome Read1.fastq.gz Read2.fastq.gz.

Reads that were not mapped to the human genome were aligned to complete genomes of microbes downloaded from the NCBI RefSeq database (https://ftp.ncbi.nlm.nih.gov/genomes/refseq/; dated March 2018). Kraken v.1 (38) was used for alignment with the following with the following parameters: kraken –preload –Kraken_Database –fastq-input –gzip-compressed –paired Microbiome_Read1.fastq.gz Microbiome_Read2.fastq.gz –output Microbial_alignment. Resulting alignment files were then converted into a read count table: kraken-report -Kraken_Database Microbial_alignment.

Aligned reads by Kraken were then provided to Bracken v.1 to align to bacterial species using the following parameters: python est_abundance.py -i BAL_Samples_Full_Kraken_subanalysis -k KMER_DISTR.txt. Bracken (39) was used because Kraken may classify some microbial reads to genus level or higher level taxonomic classification. Bracken reclassified all nonspecies classifications from Kraken to species level classifications. All bacterial counts were used to assess outlier subjects by principal component analysis (PCA). One lung sample was identified as an outlier from the PCA plot and removed. (Analysis by including the one outlier did not appreciably affect the results.)

Raw counts from Bracken alignment were used to calculate bacterial diversity. Alpha diversity (including number of observed species, Shannon diversity index and Gini–Simpson index) was calculated using microbiome v. 1.9.1 (40) in R v. 3.5.3 using the alpha function. Beta diversity was calculated using vegan v.2.5–6 (41) vegdist function with the Bray–Curtis dissimilarity index. PERMANOVA from the adonis2 function of vegan was applied to the vegdist results and 1,000 permutations were run to determine significance of beta diversity. Alpha diversity was compared using Kruskal–Wallis test, and beta diversity was compared using PERMANOVA.

Bacteria species were retained if they met the following two criteria: (i) species present in at least two study participants, and (ii) aligned reads of that species is ≥0.005% total aligned microbial reads in a study participant. After filtering, in the oral microbiome, 763 bacteria species were retained out of 3,345 and in the lung microbiome, 1,016 bacteria species were retained out of 3,073. After preprocessing, remaining reads were assessed for batch effects (library generation, extraction batch) and possible outliers subjects by PCA plot.

Remaining bacteria were normalized to library size using the calcNormFactor function in the edgeR package v.3.26.8 (42). log₂ counts per million were then calculated using the voom (43) function in the limma package v. 3.40.6. Corrections for library generation batch were made using combat function in the sva package v. 3.32.1 (44).

Log₂-transformed and library generation batch corrected bacterial reads were then analyzed in three pairwise comparisons to identify bacteria species that were differentially abundant by transcriptional load based on smoking status. The three pairwise comparisons were SM versus NS (SM/NS), SM versus EC users (SM/EC), and EC users versus NS (EC/NS). The limma (45) function contrasts.fit was employed for pairwise comparisons of smoking groups. Empirical bayes (eBayes) was applied to results from contrasts.fit to calculate log₂ fold change and P values using a moderated t-statistic. TopTable was applied to the results from eBayes and P values were FDR corrected (46). Altered bacteria species with FDR-corrected P < 0.1 were considered to be significant.

This study was approved by the OSU Comprehensive Cancer Center Clinical Scientific Research Committee and the OSU IRB (the IRB approval number: 2015C0088, ClinicalTrials.gov: NCT02596685) with written informed consent obtained from participants and was conducted in accordance with the U.S. Common Rule.

The data generated in this study are publicly available in Gene Expression Omnibus GSE180785.

Metatranscriptome data were collected for 28 BAL samples and 26 saliva samples. Each sample was sequenced to a minimum of 35 million total reads (microbial and human reads). In the 28 BAL samples, at least 78,000 sequenced reads mapped to 3,073 bacteria species, and in the 26 saliva samples, at least 17 million sequenced reads mapped to 3,345 bacteria species (Supplementary Table S1). Quality (e.g., possible outliers, batch effects) of the metatranscriptome data was evaluated using PCA (see Supplementary Fig. S1 for a detailed data analysis workflow). Accordingly, one outlier sample from a smoking subject was removed, and data were corrected for observed batch effects caused by library generation (Supplementary Fig. S2 and S3).

Analysis of the lung microbiome began with overall bacterial composition using both alpha diversity (number of observed species and evenness of abundance within subjects) and beta diversity (measured dissimilarity of bacterial communities between groups). The alpha diversity metrics analyzed included the Shannon H-index (showing the similarity for the number of microbes and relative abundance were similar; Fig. 1A), the number of observed species (Supplementary Fig. S4A) and the Gini–Simpson index (which measures possible dominance of bacteria species in a given sample; Supplementary Fig. S4B). The Bray–Curtis dissimilarity index (which measures differences in bacteria composition using the quantitative bacterial counts, instead of presence or absence of bacteria), was used to assess beta diversity and no statistical differences were identified among the groups. When we plotted the Bray–Curtis dissimilarity index on a multi-dimensional scaling (MDS) plot, EC users and NS tended to cluster together compared with SM (Fig. 1B).

Three pairwise differential analyses (SM/NS, SM/EC, and EC/NS) were conducted to identify bacteria species with differentially abundant transcriptional loads. Figure 1C shows the highly abundant bacteria species in the lung, which was defined as bacteria species which were present with >1% abundance in the lung of any smoking subject. None of these species, however, were differentially abundant species that were identified between groups. There were 44 bacteria species with differentially abundant transcriptional loads when comparing SM/NS, 12 bacteria species for the SM/EC comparisons and none for the EC/NS comparison (FDR-adjusted P < 0.1; Supplementary Table S2). Ten bacteria species were identified in both SM/NS and SM/EC comparison (Fig. 1D). Almost all of the differentially abundant bacteria in the SM/NS comparison, and all of the SM/EC comparisons, represented species which were significantly decreased in smokers (Fig. 1E; Supplementary Table S2). Of the top five decreased species, Neisseria elongate, Neisseria sicca, Haemophilus parainfluenzae, which were identified in both SM/NS and SM/EC, while Fusobacterium periodonticum and Leptotrichia sp. oral taxon 212 were identified in SM/NS, and Neisseria zoodegmatis and Ottowia sp. oral taxon 894 were identified in SM/EC comparison; N. zoodegmatis and O. sp. oral taxon 894 were also decreased in the SM/NS comparisons, but were not in the top five decreased species (Fig. 2; Table 2). Neisseria sp. KEM232 and Curvibacter sp. AEP1–3 were two bacteria species identified in only the SM/EC comparison (Fig. 2A). There were eight bacterial species that increased in the SM/NS comparisons, which were Niastella koreensis, Polaribacter sp. SA4-12, Myroides odoratimimus, Vibrio tasmaniensis, Paraburkholderia rhizoxinica, Stigmatella aurantiaca, Lactobacillus kefiranofaciens, and Paraburkholderia terrae (Fig. 2B and C).

The alpha diversity measured by the Shannon H-index for assessing variance within a sample (number and relative abundances of the microbes) within groups indicated that each group of subjects were similar (Fig. 3A). Similar results were found for the number of microbes (Supplementary Fig. S4C) and by the Gini-Simpson index (Supplementary Fig. S4D). The beta diversity measured by Bray–Curtis dissimilarity index (which used quantitative bacterial counts to measure differences in bacteria composition), among the three groups were statistically different (PERMANOVA P < 0.05). We observed in the MDS plot of Bray–Curtis dissimilarity index that NS and EC users tended to cluster together compared with the SM (Fig. 3B).

Figure 3C shows the 28 bacterial species identified in the oral cavity with >1% abundance. There were 152 bacteria species that were statistically significantly different for the SM/NS comparison (Fig. 3D). A total of 83 were increased and 69 were decreased in smokers (Fig. 3E; Supplementary Table S2). Only 17 were found for the SM/EC comparisons, and all but three were decreased. Common oral microbiome species such as Haemophilus parainfluenzae, Prevotella melaninogenica, Ottowia sp. oral taxon 894, Campylobacter concisus, and Neisseria elongata were observed as decreased in the SM oral microbiome compared with EC users’ oral microbiome (Supplementary Table S2). The top five species that were identified in the oral microbiome comparison of SM/NS were Lactobacillus gasseri, Lactobacillus fermentum, Streptococcus lutetiensis, Lactobacillus crispatus, and Lactobacillus helveticus, all of which were increased in SM (Table 2). In SM/EC the top five species were, Herbaspirillum sp. Meg3, Candidatus Kinetoplastibacterium desouzaii, Neisseria sicca, Neisseria meningitidis, and Neisseria gonorrhoeae which were all decreased in SM (Table 2). Examples of bacteria species with altered transcriptional loads which were also observed to be differentially abundant in the lung microbiome include Neisseria elongata, Haemophilus parainfluenzae, and Fusobacterium periodonticum (Fig. 3F). Notably, no oral bacteria species were found to be different when comparing EC/NS.

Combining SM/EC/NS, there were only four bacterial species found in both the lung and oral cavity with >1% abundance (Figs. 1C and 3C). These species were Haemophilus parainfluenzae, Prevotella melaninogenica, Prevotella jejuni, and Haemophilus influenza which overlapped with the 28 highly abundant bacteria species in the oral microbiome. These species were found in a higher proportion of reads in the saliva samples (>5% abundance) compared with the lung samples (<5% abundance). The most abundant phyla were Proteobacteria, Firmicutes, Actinobacteria, and Bacteroidetes (N = 5) mostly overlap for the lung and oral cavity, except for Tenericutes in the lung and Fusobacteria in the oral cavity (Supplementary Fig. S5A and S5B). There was less concordance between the lung and oral cavity when analyzed for most abundant genera (N = 10; Supplementary Fig. S5C and S5D) and species (N = 10; Supplementary Fig. S6).

For the pairwise comparisons between the SM, EC users, and NS showing differential abundance, Figs. 4A and 4B show the number of nonoverlapping and overlapping species observed in the oral cavity and lung in SM/NS and SM/EC comparison, respectively. A total of 21 overlapping species were observed for the SM/NS comparisons (Supplementary Table S3), and for these 20 were decreased in both the lung and oral cavities. There was little overlap among the various groups (Supplementary Fig. S7). Only four bacteria species were identified across all comparisons in the lung and oral cavity (Neisseria elongata, Neisseria sicca, Neisseria zoodegmatis, and Snodgrassella alvi). Heatmaps were generated using the differentially abundant bacteria species of the lung and oral cavity microbiomes, as shown in Fig. 4C and D. Smokers clustered together in one cluster in the lung; however, in the oral microbiome, the same clustering was not observed.

This observational study was designed to compare the lung and oral microbiome population comparing SM, EC users, and NS using metatranscriptomics. This study was the first to study the lung microbiome among EC users. The overall bacteria composition across the three study groups were similar in the lung (alpha and beta diversity), whereas differences were observed in the beta diversity in the oral microbiome. In the lung pairwise comparisons, the greatest difference in bacterial species abundance was between SM/NS, and almost all of the differences for SM/EC overlapped with the SM/NS comparisons. There were no differences between EC/NS, despite most of the EC users being former smokers. For comparisons of both the SM/NS and SM/EC users, almost all, or all, of the bacterial species were decreased in SM. Comparing the lung and oral microbiomes, many of the same bacterial species were present in both, but the overall profiles of the lung and oral microbiomes were different.

Prior studies of healthy participants have identified the presence of a core pulmonary microbiome (25, 47, 48). Candidate genera present in at least 50% of subjects included Pseudomonas, Streptococcus, Prevotella, Fusobacterium, Haemophilus, Veillonella, and Porphyromonas. In this study, 15 bacterial species had >1% abundance. These core bacteria included Pseudomonas, Streptococcus, Prevotella, and Haemophilus, but not Veillonella, Porphyromonas or Fusobacterium.

Lung dysbiosis by smoking status has previously been studied for SM/NS by sequencing the 16S rRNA bacterial subunit (16S sequencing) rather than metatranscriptomics (25, 47, 49). Metatranscriptomics analysis like 16S sequencing identified no differences for lung alpha or beta diversity. However, the three studies observed no differences in the lung microbiome between SM and NS. This result may be due to resolution limitations in 16S sequencing which studied higher order classifications of bacteria.

In this study, 44 lung bacterial species were differentially abundant when comparing SM/NS, and 12 bacteria species were differentially abundant when comparing SM/EC (FDR-adjusted P < 0.1). Of those, 80% and 100% of the bacteria species abundances were decreased in SM, respectively. The bacteria species with decreased transcriptional load among smokers were predominately gram-negative bacteria in the phyla Proteobacteria and Bacteroidetes. This negative association could be due to the toxic effects of smoking cigarettes on the lung which may include change in pH, humidity, heat, salivary flow, or xenobiotic degradation. There were eight differentially abundant bacterial species that increased in SM compared with NS. These included bacteria species with unclear pathogenic relevance to lung disease or cancer pathways, although Myroides odoratimimus is a rare opportunistic pathogen in humans (50) and the Vibrio genus may cause foodborne infections (51).

There were negligible differences in the lung microbiome associated with EC use; no significant differences in transcriptional load were observed in comparisons of EC users with NS. In the SM/EC comparisons, all but two of the differentially abundant bacteria overlapped with the SM/NS comparisons, and thus these two may be a residual smoking effect. The two bacteria that were differentially abundant only in the SM/EC comparisons were Neisseria sp. KEM232 and Curvibacter sp. AEP1-3, both of which were decreased. The former is an oral organism that was isolated from dental caries (52). The latter is identified as an aerobic bacteria species of unclear clinical significance. Overall, these data do not support the hypothesis that smoking-related lung disease or cancer is mediated by a specific bacterial agent; however, the disease risk by microbiome is unknown at this time. Although the decrease in bacterial content may hypothetically enable an environment for pathogenic bacteria, none was seen in this study.

Oral microbiome bacterial diversity for SM/NS comparisons as found herein has been previously studied using 16S sequencing in six studies, and the results have generally indicated dysbiosis caused by smoking (25, 32, 53–56), although one study observed no dysbiosis (56). In a large study, Wu and colleagues (54) observed increases in SM for the Actinobacteria phylum and Coriobacteriia class, and the genera Lactobacillus and Streptococcus. They also reported decreases in Proteobacteria phylum which included the genera Haemophilus and Neisseria, although not part of Proteobacteria phylum, the genus Capnocytophaga was also decreased in SM. Morris and colleagues (25) studied 64 participants and found differences only for Porphyromonas, Neisseria, and Gemella. Using metatranscriptomics, we found 152 bacterial species for the SM/NS comparisons, and confirmed the above study results for increased Lactobacillus and Streptococcus, and decreased Haemophilus, Neisseria, and Capnocytophaga genera. The increased genera in the oral microbiome were mostly anaerobic bacteria, implying that oxygen availability may have a role in bacterial populations’ response to smoking in the oral microbiome. The increase in Lactobacillus and Streptococcus genera also implied that smoking may select for gram-positive bacteria.

Three prior studies have assessed associations of EC use with differences in the oral microbiome in humans (31, 32, 56). While we observed no differentially abundant bacterial species for the EC/NS comparisons, Pushalkar and colleagues (32), with 119 participants, reported an increased relative abundance for EC use compared with NS. They further identified five bacteria genera that were increased in EC users over SM, including Actinomyces and Campylobacter, and five bacteria genera that were decreased. In addition, Ganesan and colleagues (31) studied gingival plaques with metatranscriptomic sequencing in 70 participants. The authors observed that EC use was associated with higher levels of gram-negative facultative anaerobes including Aggregatibacter, Haemophilus, and Rothia compared with NS (31). Although their sequencing and sampling methods differed, both studies found EC use was associated with increase in the bacteria genera Leptotrichia, Selenomonas and Veillonella. Similar to our own findings, Stewart and colleagues (56) reported no between-group differences in a study of 30 participants. It is possible that the reason we failed to observe a significant difference in our study may be due to the small sample size which resulted in limited statistical power to observe small differences.

Considering bacteria present in >1% in any subject, only four species were found in both the lung and oral microbiome. There were 21 bacterial species that were differentially abundant in the SM/NS comparisons in both the oral cavity and the lung, of which 20 were decreased in SM. Examples included Neisseria elongata, Neisseria sicca, Fusobacterium periodonticum, and Haemophilus parainfluenza. Meanwhile, Lactobacillus kefiranofaciens was the only bacteria species with increased transcriptional load in both the lung and oral microbiome of SM/NS. Thus, while the lung microbiome may derive from oral microbiome via microaspiration and dispersion along the mucosa (26–28), this was not observed herein. Therefore, assessing the oral cavity as a surrogate for the lung microbiome is not supported by this study.

There were several limitations in our study. This was a small study with a mostly male study population in smokers, so generalizability to smoking and EC user populations will need to be affirmed in additional studies. Second, because of this study's cross-sectional design which evaluated the metatranscriptome at a single timepoint, these results may not accurately reflect the dynamic nature of microbial populations and their transcriptional activity, which we assume vary over time and length of exposure to EC use or smoking combustible cigarettes. In addition, the study design and small sample size makes the results susceptible to confounding such as seasonality, air pollution, and occupational exposures, which would need to be explored in a larger study. However, we believe that the effects of these potential confounders would be low compared with the direct inhalation and lung effects of smoking and vaping. Also, there were more male White SM than the NS and EC groups, which could be a source of bias. However, in our other studies for gene expression and other ‘omics biomarkers, we did not find that race and gender were confounders. Larger studies would be needed to explore this potential confounding. Another limitation is the possibility that the bronchoscope traveling through the nasopharynx or oropharynx may get contaminated with upper airway bacteria affecting BAL results. Given that the differentially abundant bacteria had little overlap lessens this concern, and other studies that have examined this found little evidence for contamination by upper airway bacteria (28, 57–59). In this study, we did not examine the amount of cigarette smoking or EC use given the small number of study subjects, but should be considered in larger studies. Finally, the oral microbiome results may be confounded by the presence of periodontal disease, which was not assessed herein (31, 32).

Lung microbiome dysbiosis may be associated with lung disease including lung cancer, and dysbiotic lung microbiome populated by oral species have been observed to drive lung cancer progression (60). In this study, the hypothesis that the pathogenesis for smoking-related lung disease or cancer includes a toxic effect on the lung microbiome was not supported because of the similar bacterial diversity for SM and NS, and because most differentially abundant bacteria are either decreased, or have an unclear clinical significance. No pathogenic bacteria were identified which may have overgrown due to decreases in other bacteria. The lung microbiome for EC users was similar to NS, which indicated that EC use does not have a strong pathogenetic effect on the lung mediated by the microbiome. Our results for the oral microbiome in particular the SM/NS and SM/EC comparisons were consistent with prior studies. However, when the oral and lung microbiome are compared, the former does not appear to be a surrogate biospecimen for assessing smoking and EC effects on the lung microbiome. Given the small study size, the results should be considered hypothesis generating and replicated in larger studies.

M.-A. Song reports grants from NIH during the conduct of the study. P.G. Shields reports grants from NIH grants during the conduct of the study; personal fees from Law firms in tobacco litigation on behalf of plaintiffs outside the submitted work. No disclosures were reported by the other authors.

K.L. Ying: Resources, data curation, software, formal analysis, investigation, visualization, writing–original draft, writing–review and editing. T.M. Brasky: Writing–review and editing. J.L. Freudenheim: Conceptualization, supervision, writing–review and editing. J.P. McElroy: Supervision, writing–review and editing. Q.A. Nickerson: Resources, writing–review and editing. M.-A. Song: Conceptualization, supervision, writing–review and editing. D.Y. Weng: Resources, supervision, writing–review and editing. M.D. Wewers: Conceptualization, resources, writing–review and editing. N.B. Whiteman: Data curation, writing–review and editing. E.A. Mathe: Resources, supervision, writing–review and editing. P.G. Shields: Conceptualization, resources, supervision, funding acquisition, writing–review and editing.

Research reported in this publication was supported by funding from the NCI (P30 CA016058) awarded to P.G. Shields, the FDA Center for Tobacco Products (CTP; P50CA180908 and R21HL147401) both awarded to P.G. Shields, the National Center for Advancing Translational Sciences (UL1TR001070), Pelotonia Intramural Research Funds awarded to M-A. Song, and the Prevent Cancer Foundation awarded to P.G. Shields. This work was supported in part by the Intramural/Extramural research program of the National Center for Advancing Translational Sciences, NIH.

The authors acknowledge the support of the OSU Comprehensive Cancer Center Genomics Shared Resource, Bioinformatics Shared Resource, and the Biostatistics Shared Resource. The authors also thank the study participants, the staff and nurses of the OSU Clinical Research Center, and Sahar Kamel for assisting in recruiting participants.

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.