Untargeted Metabolomics to Characterize the Urinary Chemical Landscape of E-cigarette Users

Abstract

The health and safety of using e-cigarette products (vaping) have been challenging to assess and further regulate due to their complexity. Inhaled e-cigarette aerosols contain chemicals with under-recognized toxicological profiles, which could influence endogenous processes once inhaled. We urgently need more understanding on the metabolic effects of e-cigarette exposure and how they compare to combustible cigarettes. To date, the metabolic landscape of inhaled e-cigarette aerosols, including chemicals originated from vaping and perturbed endogenous metabolites in vapers, is poorly characterized. To better understand the metabolic landscape and potential health consequences of vaping, we applied liquid chromatography-mass spectrometry (LC-MS) based nontargeted metabolomics to analyze compounds in the urine of vapers, cigarette smokers, and nonusers. Urine from vapers (n = 34), smokers (n = 38), and nonusers (n = 45) was collected for verified LC-HRMS nontargeted chemical analysis. The altered features (839, 396, and 426 when compared smoker and control, vaper and control, and smoker and vaper, respectively) among exposure groups were deciphered for their structural identities, chemical similarities, and biochemical relationships. Chemicals originating from e-cigarettes and altered endogenous metabolites were characterized. There were similar levels of nicotine biomarkers of exposure among vapers and smokers. Vapers had higher urinary levels of diethyl phthalate and flavoring agents (e.g., delta-decalactone). The metabolic profiles featured clusters of acylcarnitines and fatty acid derivatives. More consistent trends of elevated acylcarnitines and acylglycines in vapers were observed, which may suggest higher lipid peroxidation. Our approach in monitoring shifts of urinary chemical landscape captured distinctive alterations resulting from vaping. Our results suggest similar nicotine metabolites in vapers and cigarette smokers. Acylcarnitines are biomarkers of inflammatory status and fatty acid oxidation, which were dysregulated in vapers. With higher lipid peroxidation, radical-forming flavoring, and higher level of specific nitrosamine, we observed a trend of elevated cancer-related biomarkers in vapers as well. Together, these data present a comprehensive profiling of urinary biochemicals that were dysregulated due to vaping.

Graphical abstract

Introduction

E-cigarette use is a public health crisis due to potential harm similar to combustible tobacco, limited toxicological evaluations, but rapid adoption by the public.^1–3 The evolution of e-cigarette products and high prevalence of dual users with combustible tobacco have added to the challenging nature of characterizing the health effects of the use and defining whether e-cigarettes can serve as a less harmful alternative to conventional cigarettes. In this study, we define an e-cigarette as a collective combination of the delivery hardware and the “e-liquid” – the fluid stored in the delivery hardware for aerosolization – and we sought to compare biomarkers that differed in e-cigarette usage (vaping) to non-usage as well as to conventional cigarette smoking. With the safety profile identified to date, e-cigarette usage has been suggested as risk factor for adverse outcomes such as asthma^{1, 4, 5}, cardiovascular diseases⁶, and different classes of cancer⁷. With the short time since its inception, it may take decades to observe the eventual incidences of these chronic health impacts; however, monitoring e-cigarette exposures and related biomarkers may reflect development of these adverse outcomes, and thus provide references for e-cigarette research and regulations.

To study the health effects of e-cigarettes, the most current research has focused on profiling compounds in e-liquids, and epidemiological relationships between vaping and specific disease outcomes. Nearly 300 unique chemicals have been identified, by GC-MS,⁸ in 49 e-liquids with different flavors and several other studies have shown similar complexity in e-liquids.^{9, 10} Detected compounds include oxidative stress stimulants (e.g., polycyclic aromatic hydrocarbons “PAHs”)^{10, 11}, and psychoactive substances (e.g., cannabinoids)¹². Researchers have speculated what the downstream health consequences of exposure to these chemicals may be. It is even more concerning that many identities of the e-cigarette chemicals remain unknown, and the majority of the known and unknown chemicals lack toxicological assessments through inhalational exposures. Other studies evaluated e-cigarettes with rodent models and resulted in widely varying conclusions, with some suggesting cancer inductions and disrupted lipid homeostasis.¹³ In order to better understand the health implications and metabolic pathway alterations of vaping and overcome limitations of prior studies, we chose to focus on the metabolites/toxicants/carcinogens in the urine of vapers, as urine is non-invasive and convenient to collect, has been associated with early diagnosis of disease through biomarkers, and has abundant biological information. There is a paucity of data on this approach, and our work showed novelty and importance by harnessing high-resolution mass spectrometry (HRMS).

After the introduction of HRMS, the field of nontargeted chemical analysis has rapidly evolved, nurturing new fields such as metabolomics and exposomics.¹⁴ Both of these fields can benefit from HRMS, as metabolomics focuses on profiling endogenous metabolites, and exposomics focuses on xenobiotics, exposures, and contaminants. The manner in which HRMS facilitates chemical profiling has been described in detail previously.^15–17 In brief, HRMS scans in an unbiased fashion across a wide range of m/z, usually with separation technologies such as liquid chromatography (LC) that elute compounds in different retention time (RT). With simple and nonselective sample preparation, LC-HRMS can record the levels of the molecular features (i.e., m/z-RT combinations), and some features of interest can then undergo tandem mass scans to collect structural information that allow compound characterization.

Vaping, as aforementioned, introduces a repertoire of chemicals mainly by inhalation, which previous health endpoint-based studies have suggested associations with disease outcomes. The power of HRMS-based nontargeted chemical analysis can help explore the chemical landscape that e-cigarette use introduces, and how vaping can affect human physiology. The objective of this study was to capture the shift of the metabolic landscape (both chemicals originating from e-cigarette and endogenous metabolites) brought about by vaping and to compare against combustible smoking, using LC-HRMS. Our data showed similar levels between vapers and smokers for nicotine metabolites, but elevated and more diverse exposure against non-nicotine alkaloids in vapers. We also observed alterations in the fatty acids, esters, and amides, as well as different carnitine species in both smokers and vapers (than in controls). Of which, vapers have more consistent elevation in the carnitine profiles, suggesting dysregulation of fatty acid oxidation. We highlighted some other detected compounds including flavoring agents, exposure biomarkers (i.e., diethyl phthalate), and cancer biomarkers that showed considerable changes associated with vaping. The results of this study may have widespread implications including guidance on Food and Drug Administration regulations of e-cigarette use, education of vapers on the potential health risks, specifically that of young adults, and inform health care providers on counseling regarding risks of e-cigarette use.

Experimental Procedures

Chemicals and reagents

Chemicals and reagents were purchased from Sigma-Aldrich (St. Louis, MO) unless otherwise specified. [D₂]-indole propionic acid and [D₃]-tryptophan were provided by CDN Isotopes (Pointe-Claire, Quebec, Canada). Solvents used (water, methanol, acetonitrile, and formic acid) were all Optimal LC-MS grade obtained from Thermo Fisher Scientific (Rockford, IL).

Recruitment of participants

Urine samples were previously collected through New York University School of Medicine¹⁸ and the University of North Carolina School of Medicine research studies as previously described.⁵ Briefly, subjects were healthy young adults 18–50 years of age in three groups: 1) nonsmokers not regularly exposed to second hand smoke; 2) self- described active cigarette smokers; and 3) self- described, active e-cigarette users/vapers who had been using e-cigarettes regularly for at least 6 months. The demographics of the subjects were listed in detail in Table S1. The recruitments and studies at New York University School of Medicine and the University of North Carolina School of Medicine were respectively approved by the Institutional Review Board at New York University School of Medicine (s16–02226 and s17–01143) and University of North Carolina at Chapel Hill Biomedical Institutional Review Board (13–2246). All research was performed in accordance with relevant guidelines/regulations, and informed consent was obtained from all participants.

For both studies, urine was collected in a sterile urine cup, placed on ice during transportation, and then frozen until assayed for biomarkers of exposure. Unlike other biological samples, urine was not collected immediately after the smoking/vaping session because expected biomarkers are not likely to be present yet.

Sample preparation

The extraction of compound from urine followed the procedures provided by previous studies with minor modification.^{16, 19} Urine samples from subjects were thawed on ice and a 50 μL aliquot from each sample were used for further analysis. Protein precipitation and metabolite extraction were facilitated by adding 200 μL methanol and incubating under −20°C for 1 hr. Supernatants after centrifugation under 15,000 ×g for 10 minutes were collected and dried with SpeedVac^®. The dried samples were reconstituted with 50 μL of 2% acetonitrile and were ready for LC-ESI-MS analysis.

LC-ESI-MS(/MS) analysis

Extracted urine metabolites were analyzed on a high resolution LC-MS platform as previously described.^{16, 19} The instrumental analysis was composed of two sequences, with the first sequence recording feature profiles for alignment and quantitation; and the second sequence providing tandem mass spectrums for compound identification. Both runs utilize a Thermo Fisher Scientific Vanquish UHPLC coupled to a Q Exactive mass spectrometer with a heated electrospray ionization source (HESI) as the interface. Processed urine sample was injected (3 μL) into a Waters Acquity UPLC HSS T3 (reverse phase C₁₈, 100 Å, 1.8 μm, 2.1 mm × 100 mm) analytical column controlled at 40 °C. The mobile phase was composed of water (A) and acetonitrile (B) both acidified to 0.1% formic acid, with the 15-min-gradient set as the following: 2% B from 0 to 1 min; 2% to 15% B from 1 to 3 min; 15% to 50% B from 3 to 6 min; 50% to 98% B from 6 to 7.5 min; 98% B held from 7.5 to 11.5 min; 98% to 2% B from 11.5 to 11.6 min; and 2% B held from 11.6 to 15 min for a final re-equilibration. The mass spectrometer monitored in positive mode with the sheath gas, auxiliary gas, and sweep gas set to flow rates of 50 psi, 13 (arbitrary unit), and 3 (arbitrary unit), respectively. The spray voltage of HESI was set at 3.50 kV; with the capillary and auxiliary gas heating temperature were respectively controlled to 263°C and 425°C. The full scan monitored from m/z 70–1000 at the resolution of 70,000 FWHM (m/z 200) with the automatic gas control (AGC) and maximum injection time (MIT) set to 3×10⁶ and 225 msec, respectively. If MS² data were to be collected, the parallel reaction monitoring (PRM) scan will be added along with the full scan, with the isolation width, AGC, and MIT set to 1.2, 3×10⁵ and 100 msec, respectively, at the resolution of 17,500 FWHM (m/z 200).

Urine samples were injected in a blocked randomized fashion. Spotted injections of solvent blanks, quality control (a pool of all prepared urinary samples), and a standard mixture composed of [D₃]-tryptophan, [D₅]-glutamic acid, and [D₂]-indole propionic acid each in 500 nM (10 μL injection) were inserted in the sequence. Stabilities in the retention time (RT) and mass accuracies of the standards, as well as the total signal sum of each injection were monitored for the performance of the instrument. For LC-MS/MS analysis, the quality control was used to collected MS² spectrums to decipher compounds.

Data processing

The data processing of our nontargeted chemical analysis include two stages: feature identification and compound characterization, which rely on R (ver 4.1.3, R Core Team, Vienna, Austria), R Studio (ver 2022.02.1+461, R Studio Team, Boston, MA), and Compound Discoverer (CD, ver 3.2, Thermo Fisher Scientific). For feature identification, XCMS (ver 3.16.1) was used in assistance of the IPO package (ver 1.20.0) to optimize parameters, which the methods referenced previous studies.²⁰ The LC-MS data collected from the instrument were in the proprietary format of .raw, and were firstly converted into the open data format .mzXML by ProteoWizard (ver 3).²¹ The quality control data were then used to optimize the parameters in the algorithms of feature detection (centWave), retention time adjustment (obiwarp), and feature alignment (density) in XCMS, using the IPO package. Then, all LC-MS data were processed with XCMS with the optimized parameters to obtain a list of molecular features with distinct m/z-RT combinations. The criteria of detected proportion in aligning features across samples was set to 50% in at least one group. Missing values in the measurement matrix were imputated with random forest algorithm by the missForest package in R as suggested in previous comparison studies.²² After log-transforming the data, group-wise F-tests were conducted for each of the feature to reveal the overall difference among controls, vapers, and smokers. Additionally, a pair-wise Student’s t-test for any two groups were also practiced. The p-values for both the F- and t-tests were used to calculate q-values to address the false discovery by the Benjamini-Hochberg procedure. In order to find a larger potential set of features that could distinguish among our study groups, a criterion of mean-fold-change larger than 1.5 and p-value for t-test lesser than 0.05 in any two comparisons among controls, vapers, and smokers was set to select MS² targets for capturing tandem mass spectrums that allows further compound characterization.

The LC-MS/MS data were imported to Compound Discoverer to facilitate compound annotation with a workflow previously described with minor modification.²³ In brief, chromatogram correspondences (that implied isotopic patterns and different adduct forms) were used to group features into unknown compounds. The isotopic patterns were used to propose chemical formula of the unknown compounds, and the formula is then queried against four annotation databases: our in-house LC-MS library, mzCloud Advanced Mass Spectral Database (mzCloud, Thermo Fisher Scientific), NIST 2020 LC-MS/MS library (NIST library, National Institute of Standards and Technology, MD), and ChemSpider chemical structure database (ChemSpider). Our in-house library was composed by testing authenticated standards of 734 common metabolites, drugs, and environmental exposures, which were matched to the unknown compound list for accurate mass (± 5 ppm) and retention time (± 0.25 mi), in parallel, which was considered as the most reliable annotating source for this study. The second and third annotating sources, mzCloud and NIST library, were both commercially or publicly available spectrum databases with high resolution tandem mass spectrums to match against our unknown compound list. ChemSpider unites multiple data sources to provide informative and curated structural data. Molecules recorded in the Human Metabolome Database (HMDB), Kyoto Encyclopedia of Genes and Genomes (KEGG), and BioCyc were subset from ChemSpider and matched against the unknown compound list. The in silico fragment prediction in CD was applied and the compound with the highest matching score was proposed as the most possible structure.

Data interpretation

Annotated chemical structures from the four annotating sources were further matched with multiple identifiers in PubChem, CAS registry, HMDB, and KEGG databases to validate, cross-reference, and maximize capability to reveal relationships among these molecules. MetaMapp were used for chemical/biochemical similarity network derivation.²⁴ Cytoscape (ver 3.8.2) was used to illustrate network files exported from MetaMapp. To account for the sex disproportion in our study participants, sex-stratified statistical analysis with the same method describe previously was used to confirm the effect of vaping and smoking regardless of sex.

Results

Method validation and exploratory data analysis

To uncover metabolites and exposure biomarkers in the urinary component of vapers, smokers, and nonusers, we conducted nontargeted LC-HRMS analysis after the non-selective compound extraction from the urine samples of participants. The workflow for experiments and data analysis for our urinary nontargeted chemical analysis are illustrated in Figure 1A. Besides the biological samples (non-smokers/non-vapers n = 45, vapers n = 38, and smokers n = 34), solvent blank samples (n = 11), standard mixtures (n = 11) and pooled quality control samples (n = 11) were routinely analyzed in parallel to examine background signals, and to assure instrumental performance and stability. With little fluctuation in RT (Figure S1A) and stability in measurement (Figure S1B), our method was able to record informative molecular features in the urine samples (Figure 1B). Under the detected proportion set to 50% in at least one group (of the controls/vapers/smokers), we were able to detect and align 27,869 molecular features (Figure S2A). These features also corresponded to consistent detections in the quality control samples (Figure S2B). The levels of the features were tested by principal component analysis (PCA), and showed no apparent clustering (Figure 1C), which is expected due to the intricate physiologies and exposures the participants had encountered.

Figure 1.

Workflow and exploratory data analysis of nontargeted chemical analysis for the urine of vapers, smokers, and the controls. (A) Schematic illustration of the experimental, instrumental, and data processing workflow of the extraction and analysis of urinary components. (B) Distribution of base peak intensity for the LC-ESI-MS detection for one of the representative pooled quality control samples. (C) Principal component analysis (PCA) of signal distribution for the pooled quality controls (red, with 11 total injections across the single analytical batch), vapers (yellow, n = 34), smokers (green, n = 38), and controls (grey, n = 45).

Differential analysis and compound annotations

The 27,869 molecular features (i.e., RT-m/z signatures) detected in the urine samples of the controls, vapers, and smokers were analyzed by both group-wise F-tests and pair-wise t-tests (of any two-group-comparison) to understand the alterations contributed by vaping and smoking. In order to maximize the coverage of altered features, a criterion of > 1.5 fold-change and < 0.05 p-value in any pair-wise t-test was used as a filter to compose a list of MS² targets to collect tandem mass spectrums that allows the deciphering of features to compounds with structural information. This criterion resulted in 839, 396, and 426 altered features in the comparison of urine samples between smokers and controls, vapers and controls, and vapers and smokers, respectively (Figure 2A). The union of these altered features built an inclusion list of 1,277 PRM targets to collect MS² spectrums. It is noted that 1,260 of the 1,277 altered features (98.6%) also showed p-values lower than 0.05 in the group-wise F-test. The ratio of upregulated:downregulated urinary metabolites was greater in both smoker (2.6, 607:232) and vapers (4.6, 326:70) as respectively compared to the controls. There were 101 in common up-regulated features by smoking and by vaping, suggesting that smoking and vaping share some similar metabolic alterations or xenobiotic exposures (Figure 2B). However, they also showed apparent differences, as the up-regulations of 506 features in smokers were absent in vapers, and the elevations of 225 features in vapers were not observed in smokers. The difference in the feature profiles between vapers and smokers was also shown in their pair-wise comparison (Figure 2A), as smokers showed 1.8 times (276:150) elevated than lowered features.

Figure 2.

Differential analysis for detected molecular features and (bio-)chemical network analysis for the annotated compounds in the urine of vapers, smokers, and the controls. (A) Volcano plot with the x-axis as the binary logarithm of fold changes and the y-axis as the negative common logarithm of p-value from pairwise Student t-test between the urine samples of smokers and controls (upper panel), vapers and controls (middle panel), and vapers and smokers (lower panel). (B) Venn diagram summary of significantly altered features in the urine samples of smokers and vapers against the controls, respectively. The number of up-regulated features in urine samples of smokers (left) or in vapers (right) are in annotated in red ellipses, and the number of down-regulated features in urine samples of smokers or in vapers are annotated in blue ellipses. (C) Networks of structural similarities (Tanimoto) and biochemical pathways (KEGG) calculated by MetaMapp. The node colors indicate the highest group detected for the compound, with orange, green, and grey as vaper, smoker, and control, respectively. The node shapes indicate the annotation sources, with triangles as our in-house library; round rectangles as the NIST 2020 LC-MS/MS spectrum library, diamonds as the mzCloud Advanced Mass Spectral Database; and circles as ChemSpider (selected sub-database: HMDB, KEGG, and BioCyc) chemical structure database.

The 1,277 altered features sent to capture MS² spectrums were further queried for proposal of chemical formulas and compound annotations against four databases, which by the rank of annotating confidence were: (1) our in-house library, (2) mzCloud, (3) NIST library, and (4) ChemSpider. Of the 1,277 features, 506 passed examination on isotopic patterns and were assigned a chemical formula. Among these formulas, 336 were successfully annotated with a chemical structure, with 22, 26, 67, and 221 respectively annotated by our in-house library, mzCloud, NIST library, and ChemSpider. The 336 annotated compounds varied in structure, and covered a wide field of endogenous metabolites and xenobiotics. To better understand the functions and chemical characteristics of these compounds, a MetaMapp approach was applied to network the chemical similarities and biochemical relationships of the compounds, which is illustrated in Figure 2C. We were able to characterize clusters of similar compounds, which included steroids and derivatives of amino acids. We were most interested in those clusters that demonstrated vaping/smoking differences, consistent effects or modified effects in control of confounding factors (e.g., sex, study group), and disease biomarkers. This led us to focus on the nicotine metabolites, alkaloids, fatty acids and derivatives, and carnitines, which are discussed in detail below. Chemical information on the focused compounds in this study were provided in Additional File 1.

Nicotine related metabolites and non-nicotine alkaloids detection

Whether e-cigarette products and vaping behavior differ nicotine intake and absorption from combustible cigarette has been debatable. Our data showed that all six identified nicotine metabolites did not differ between smokers and vapers (Figure 3A), and both of these groups resulted in elevations compared to the nonexposed controls. For example, the major proximate metabolite of nicotine, cotinine, had mean fold-changes of 37.5 (q = 3.7×10⁻⁷) and 35.9 (q = 6.0×10⁻⁷) in the urine samples when comparing vapers and smokers against controls, respectively, with little difference between vapers and smokers (p = 0.46, q =0.99).

Figure 3.

Levels of nicotine-related metabolites, and non-nicotine alkaloids detected in the urines of vapers, smokers, and controls. (A) Boxplots of the detected nicotine-related metabolites. (B) Boxplots of the detected non-nicotine alkaloids. Vapers, smokers, and controls were grouped and annotated by colors of yellow, green, and grey, respectively. Group-wise F-tests and pair-wise Student t-tests were used for comparison, and were respectively annotated with straight lines and bracket lines. q-values adjusted for false-discovery rate and p-values were calculated, and if significant (i.e., q-value or p-value < 0.05), were annotated in blue or green colors. *** p-/q-value < 0.001; ** p-/q-value < 0.01; * p-/q-value < 0.05.

Several non-nicotine alkaloids were also detected in our analysis (Figure 3B). For example, anatabine is commonly found in tobacco, and has served as an indicator of tobacco use, and its level in smokers was significantly higher than in the controls (fold-change 2.9, q = 0.03). Vapers showed a small but insignificant increase in urinary anatabine (fold-change 1.3, p = 0.07) compared to the controls.. We were able to characterize some urinary alkaloids which were altered by vaping but not smoking, which includes ankorine, ecgonine, and heliotrine. Of which, heliotrine, a monoester pyrrolizidine alkaloid that can be found in plants such as the Heliotropium spp. that can induce diarrhea, was 2.1 times higher level in the urine samples of vapers than in the controls (p = 0.007).

Alteration of fatty acid oxidation observed for vapers and smokers

Our nontargeted profiling on the urinary component detected endogenous metabolites with differences among the urine samples from vapers, smokers, and controls, which include a variety of biochemicals with diverse functions, including endocrines (e.g., hydrocortisone, 5α-andeost-1-en-3-one) and amino acid derivatives (e.g., N-acetylhistamine, Ile-Leu). Among the diverse profile of the altered metabolites, we found more enriched clusters of fatty acids, esters, and amides (FAEAs) as well as carnitine species (Figure 2C).

Profiles of the (acyl-)carnitine species as well as the fatty acids, esters, and amides (FAEAs) can reveal the status of energy synthesis and fatty acid metabolism. In the urine samples, we observed a total of 21 altered species of the FAEAs, with 6 of them elevated in both smokers and vapers compared to the controls (Figure 4A). On the other hand, 6 FAEAs were significantly different in the comparison of vapers, but not smokers, against controls; similarly, 6 other FAEAs showed significant difference between smokers and controls, but not between vapers and controls, suggesting differential effects of smoking and vaping on fatty acid metabolism. The carnitine species, which are essential in utilizing and transporting the FAEAs, showed that more urinary acylcarnitines were dysregulated by vaping than by smoking, as shown in Figure 4B. Specifically, when comparing the 10 acylcarnitines detected in this study against the controls, tiglycarnitine was elevated in both vapers and smokers; in addition, 6 and 2 different acylcarnitines were elevated only in vapers or smokers, respectively. The three top-elevated urinary acylcarnitines increased by vaping were pimelylcarnitine, 3-methylglutarylcarnitine, and 3-hydroxy-5,8-tetradecadiencarnitine, with respective fold-changes of 7.6 (p = 0.003), 2.5 (p = 0.009), and 2.0 (p = 0.001). Overall, vaping and smoking increased levels of different acylcarnitines in urine, with vaping altering more species. When the levels of the urinary FAEAs and carnitine species among vapers, smokers, and controls were combined, PCA analysis (21 FAEAs and 10 acylcarnitines, 31 compounds in total) showed improved clustering compared to the total 27,869 features detected (Figure 4C and 1C, also see Figure S3, S4). This further supported that both vaping and smoking changed the levels of the FAEAs and carnitines, but vaping has a greater impact on the carnitine profile.

Figure 4.

Levels and clustering results of the fatty acids, esters, and amides, as well as the carnitines detected in the urines of vapers, smokers, and the controls. (A) Boxplots of the detected fatty acids, esters, and amides. (B) Boxplots of the detected carnitines. Vapers, smokers, and controls were grouped and annotated with colors of yellow, green, and grey, respectively. Group-wise F-tests and pair-wise Student t-tests were used for comparison, and were respectively annotated by straight lines and bracket lines. q-values adjusted for false-discovery rate and p-values were calculated, and if significant (i.e., q-value or p-value < 0.05), were annotated in blue or green colors. *** p-/q-value < 0.001; ** p-/q-value < 0.01; * p-/q-value < 0.05. (C) Principal component analysis (PCA) of signal distribution for the vapers (yellow, n = 34), smokers (green, n = 38), and controls (grey, n = 45) for their profiles of detected fatty acids/esters/amides and carnitines.

Other exposure and cancer biomarkers

Thus far, we have reported profile differences of nicotine metabolites, alkaloids, FAEAs, and carnitines in urine contributed by vaping or smoking. Several other compounds that showed differences among the groups caught our attention as they may represent the special but under-researched additives or impurities from e-cigarettes. In our nontargeted chemical analysis of urinary samples, we characterized three potential flavoring compounds that showed higher levels in vapers than in smokers and controls: ethyl levulinate (EL), delta-decalactone (δ-DL), and gamma-heptalactone (γ-HL) (Figure 5A). The mean fold-changes were 1.7, 1.7, and 1.5 with the p-values being 0.001, 0.002, and 0.004 between vapers and controls for EL, δ-DL, and γ-HL, respectively.

Figure 5.

Level of the potential flavoring chemicals detected in the urines of vapers, smokers, and the controls. (A) Chemical structures, information, and identifier the potential flavoring chemicals characterized: ethyl levulinate, delta-decalactone, and gamma-heptalactone (B) Boxplot of the potential flavoring chemicals in vapers (yellow), smokers (green), and controls (grey). Group-wise F-tests and pair-wise Student-t tests were used for comparison, and were respectively annotated by straight lines and bracket lines. q-values adjusted for false-discovery rate and p-values were calculated, and if significant (i.e., q-value or p-value < 0.05) were annotated in blue or green colors. *** p-/q-value < 0.001; ** p-/q-value < 0.01; * p-/q-value < 0.05.

In addition, vapers had the highest urinary level of diethyl phthalate (DP) among the 3 groups, while smokers also showed elevation of DP exposure in their urine. DP, as a member of plasticizers (Figure 6A) can be released from heated plastic, which can happen when e-cigarette devices heat and deliver e-liquids. The phthalate family has been considered as an endocrine disrupter which can result in reproductive toxicity.²⁵ Vapers and smokers had fold-changes of 1.7 (p = 4.6×10⁻⁶) and 1.4 (p = 0.007) in urinary DP compared to the controls (Figure 6B).

Figure 6.

Level of the diethyl phthalate detected in the urines of vapers, smokers, and the controls. (A) Chemical structures, information, and identifier of diethyl phthalate. (B) Boxplot of diethyl phthalate in vapers (yellow), smokers (green), and controls (grey). Group-wise F-tests and pair-wise Student-t tests were used for comparison, and were respectively annotated by straight lines and bracket lines. q-values adjusted for false-discovery rate and p-values were calculated, and if significant (i.e., q-value or p-value < 0.05) were annotated in blue or green colors. *** p-/q-value < 0.001; ** p-/q-value < 0.01; * p-/q-value < 0.05.

Since the debate continues whether e-cigarette products and vaping contribute to carcinogenesis, we examined the presence of carcinogens and cancer-related biomarkers in this study. N(5)-methyl-N(5)-formyl-2,5,6-triamino-4-hydroxypyrimidine (Me-Fapy) is one of the by-products of the repair or degradation of 7-methyl-deoxyguanosine (7-Me-dG, the detailed mechanism is provided in Figure 7A).²⁶ The monitoring of Me-Fapy may imply the level of alkylating DNA damage (at least for DNA adduct damage).²⁷ The level showed a significant elevation in the urine samples of smokers compared to control (fold-change = 2.0, p = 0.008). Increased urinary Me-Fapy in vapers had a weak statistical trend (fold-change: 1.5, p = 0.10). However, we detected significantly higher levels of a known genotoxic nitrosamine, 1-(methylnitrosoamino)-4-(3-pyridinyl)-1,4-butanediol (MNPB, Figure 7C) in the urine of vapers, with the fold-change of 1.2 (p = 0.02) compared to the controls (Figure 7D). These results advocate some cancer induction effected by e-cigarette and vaping. It is also noted that for xenobiotics (i.e., nicotine, alkaloids, flavoring chemicals) and endogenous metabolites (FAEAs, carnitines) focused on in this study, we observed higher susceptibility to change in female than in male, for both smokers and vapers, in an extended sex-stratified analyses (Table S2-S7). We discuss possible sex-specific effects and how our study limitation can confound these interpretations later, but we would like to acknowledge our prior objective of this study in understanding the overall contribution of vaping or smoking to our metabolomic and exposure landscape, regardless of sex.

Figure 7.

Cancer related biomarkers detected in the urines of vapers, smokers, and the controls. (A) Mechanistic illustration of the formation of the cancer biomarker N(5)-methyl-N(5)-formyl-2,5,6-triamino-4-hydroxypyrimidine (Me-Fapy) (B) Boxplot of Me-Fapy in vapers (yellow), smokers (green), and controls (grey). (C) Chemical structure, information, and identifier of known carcinogenic nitrosoamine detected, 1-(Methylnitrosoamino)-4-(3-pyridinyl)-1,4-butanediol (MNPB). (D) Boxplot of MNPB in vapers (yellow), smokers (green), and controls (grey). For the boxplots, group-wise F-tests and pair-wise Student-t tests were used for comparison, and were respectively annotated by straight lines and bracket lines. q-values adjusted for false-discovery rate and p-values were calculated, and if significant (i.e., q-value or p-value < 0.05) were annotated in blue or green colors. *** p-/q-value < 0.001; ** p-/q-value < 0.01; * p-/q-value < 0.05.

Discussion

With the goal of characterizing shifts in the chemical landscape of urinary biomolecules attributed to vaping, we collected urine samples from vapers, smokers, and controls in this study, and differentially analyzed the 27,869 features detected with LC-HRMS. We were able to filter then characterize 336 chemicals with structural information that showed difference in the urine samples among the vapers, smokers, and controls. These altered chemicals composed of e-cigarette originated compounds. Besides members of nicotine and non-nicotine alkaloids, we highlighted potential flavoring agents (e.g., δ-DL), plasticizer (DP), and nitrosamine (MNPB). For these xenobiotics and their metabolites, we observed similar urinary level of nicotine species between vapers and smokers. The flavoring agents and DP were highest in the urine of vapers. In addition, we observed distinctive changes in the urinary endogenous metabolites between vaping and smoking, which includes multiple species of FAEAs and acylcarnitines, as well as a specific cancer biomarker (Me-Fapy). The dysregulations observed for urinary FAEAs and especially acylcarnitines in vapers compared to both smokers and controls, suggested that fatty acid metabolism may be critically impacted by vaping. The findings in this study served as a pioneering profiling of absorbed e-cigarette xenobiotics and the metabolic alterations induced by vaping. The results reflect key information on e-cigarette chemicals that underwent the toxicokinetics, and how vaping can influence the human metabolome thus impact our physiological functions.

The matching of LC-HRMS information to chemical structure and the scrutiny of resulted annotations have high influences on the quality and reliability of an unbiased nontargeted analysis.^{15, 28} The annotation process and curation effort followed the standards defined by the Metabolomic Standard Initiative and were applied in previous studies.^{28, 29} From a scale of Level 1 to 5, our in-house library, as described in previous research, provided Level 1(highest) annotation as validated structures. Annotations by mzCloud and NIST library qualified as Level 2 (putative structures) with spectrum library matching, and ChemSpider annotations obtained Level 2 confidence as well with in silico fragment matching. These annotation sources allowed powerful deciphering on urinary compounds of our interest in understanding the health effects of vaping.

The exposure level of nicotine and its comparison against combustible cigarette smoking has been investigated ever since e-cigarettes were introduced. Numerous studies have sought in profiling levels of nicotine and derivatives in vaping from the exposure and biomonitoring aspects, however consistent conclusion has not yet been obtained. For instance, some exposure assessments on monitoring nicotine in e-cigarette generated aerosols showed lower levels than in combustible cigarette,³⁰ while others showed equivalent concentrations under certain conditions.^{31, 32} In biomonitoring studies, our results were consistent to those detecting similar levels of nicotine metabolites between vapers and smokers;^{33, 34} for instance, Etter et al. in monitoring saliva samples and Hecht et al. in analyzing urinary samples from vapers both resulted in comparable levels against smokers in nicotine related biomarkers.³³ All 6 nicotine biomarkers detected in this study, including the unmetabolized nicotine and the major metabolites, cotinine and hydroxycotinine, showed no significant differences in the urine of vapers and smokers (Figure 3A). E-cigarette has been marketed as an alternative to assist in cigarette cessation and further lower nicotine dependence. Our data was consistent to those showing similar levels of nicotine biomarkers. As a cross-sectional analysis and limited control on the lag time after e-cigarette usage, the levels of nicotine biomarkers may not fully represent nicotine dependence. However, our data still showed that vaping does not alleviate the nicotine exposure. A composite of factors can play in nicotine exposure in vaping. Studies have shown that properties of the e-liquid (including the base PG/VG ratio, flavor, nicotine salts versus freebase nicotine, other humectants) and the design of the e-cigarette delivering device (e.g., disposable or prefilled or tank-type) can significantly affect nicotine delivery even under the same nicotine concentration in e-liquids.^{31, 35} Also, the improved palatability and presumed perception of less harmless (than cigarette) have been thought to increase the self-administration of e-cigarette.³ Our data, by monitoring urinary nicotine biomarkers, suggested similar nicotine intake in vaping and smoking. It is also noted that the newer generations of e-cigarette devices improved in higher capacity of nicotine bioavailability, which is not yet clear on how this may affect nicotine exposure and dependence.³⁶

The miscellaneous flavors introduced in e-liquids improve palatability and the ability to meet personal preferences. In multiple behavioral surveys, the flavor was the prior motivators for initial and continuous use of e-cigarette.³⁷ We were able to identify three potential flavoring agents which expressed higher levels in vapers’ urine (Figure 5). δ-DL, a sweetener with apricot and butter tasting, has been detected in previous study on profiling constituents of e-liquids, and data supported its promotion in radical formation.^{8, 38} EL and γ-HL were two other flavoring chemicals detected in this study, which can provide sweetness with berry/pear/floral, and caramel/coconut/nut tasting, respectively.³⁹ It is note that levulinic salts are common ingredients in converting alkaline free-base nicotine to protonated nicotine salts in e-liquids, which enhances perceptions of smoothness and mildness in taste.⁴⁰ EL may be the metabolite of levulinic salts after endogenous alkylation, though there is lack of proposed and evidenced metabolic pathway.⁴¹ The urinary biomolecules were abundant with excreted endogenous metabolites, thus characterizing flavoring agents in e-cigarette was difficult. Our confidence on δ-DL, EL, and γ-HL as flavoring chemicals originating from e-cigarettes based on existing literatures of e-liquid chemical analysis or wide industrial flagrance use. We also detected higher level of DP, a plasticizer with endocrine disrupting abilities, in vapers.⁴² The majority of e-cigarette hardware were made of plastics, and were heated in aerosolizing e-liquids, and this explanation has been supported by other studies.⁴³

The comparisons of the endogenous metabolome among the vapers, smokers, and controls were motivated by established knowledge of the vast dysregulation on endogenous processes smoking can contribute, such as elevation of oxidative stress, induction of DNA damages, and augmentation of sympathetic nervous system activities.⁴⁴ These mechanisms may subsequently increase risk to disorders such as cardiovascular disease, kidney failure, and cancer.⁴⁵ The altered and characterized metabolites in this study contained considerable number of FAEAs (n = 20) and acylcarnitines (n = 10), which play main roles in the lipid metabolism and mitochondrial energy production (Figure 4A, 4B). Medium- or long-chain fatty acids can be metabolized to acylcarnitines to enter the mitochondria for generating energy, or can be converted to fatty acid amides for other biological functions. The carnitine profiles in our study showed more consistent trends, as vapers had higher levels in 7 acylcarnitines (e.g., 3-hydroxy-5,8-tetradecadiencarnitine, valerylcarnitine), while smokers only showed increases in 3 acylcarnitines (Figure 4B). This may suggest more lipids were metabolize for the mitochondria to produce energy in vapers than in smokers or nonusers, which may imply higher lipid peroxidation. Studies have showed interactions between proinflammatory signaling and acylcarnitines elevations, which may support our derivation on a higher inflammatory status caused by vaping.⁴⁶

On the other hand, FAEAs are precursors, intermediates, and facilitators of miscellaneous biological functions (e.g., cell signaling, cell membrane stability).⁴⁷ In addition, many of the functions of fatty amides in physiology were uncharacterized, making it challenging to conclude specific mechanisms on the alterations of observed FAEAs. However, the three N-acylglycines (N-undecanoylglycine, N-tridecanoylglycine, N-myristoylglycine) showed consistent higher levels in vapers than in smokers and controls (Figure 4A). Higher levels of urinary N-acylglycines were known as diagnostic biomarkers for innate mitochondrial fatty acid β-oxidation disorders (FAOD).⁴⁸ N-acylglycines were also considered to modulate calcium mobilization, control of cell apoptosis/migration, (anti-)inflammatory action, and many other mechanisms.⁴⁹ The underlying mechanisms for elevated N-acylglycines remain uncertain, but our observation in their dysregulation due to vaping is concerning and should be further investigated in future studies.

Several of our observations in vaping were cancer-related, which include detection of carcinogenic nitrosamine (MNPB) (Figure 7C, 7D), flavoring agents that can generate radicals (δ-DL), and induced lipid peroxidation (that can increase oxidative stresses). The carcinogenicity of e-cigarette has been uncertain and difficult to determine due to its complexity in constituents and effects. Our effect biomarker, i.e., Me-Fapy, as a by-product of repairment or degradation of DNA adduct related to alkylating DNA damages, though with moderate statistical evidence showed trend of elevation in vapers (Figure 7B). The potential of e-cigarette in contributing cancer risk remain ambiguous, but our data provide possible biomarkers and pathways for future research to reference.

This study has several limitations; for instance, the components (e.g., e-liquid or hardware type) of e-cigarettes were not recorded nor controlled for this study, and the duration from the last smoking/vaping activity to the collection of urine was not recorded. Also, the sex profiles among our groups were disproportional: there were more male in smokers and vapers, but more female in the controls. This limitation confounded some of the detected compounds that showed general difference among the exposure groups. Sex-specific differences were not the primary focus for our study, but were carefully managed when measuring the effects on exposures. A smaller sample size of females in the exposed group also made us tentative in interpreting sex-specific differences observed in the comparisons of vaping and smoking. However, a consistent observation that females were more susceptible to the exposure of nicotine, alkaloids, and flavoring can be made (Table S2, S3, S6). We suggest that sex can be involved in the exposomic and metabolomic (e.g., the FAEAs) differences under vaping (and smoking), and a more unambiguous understanding can benefit from a dedicated study design in future research. Nevertheless, compound databases useful in annotating human urinary compounds were majorly metabolite-oriented, and lack in recording those (metabolized) xenobiotics, which may be a predominating reason of a lower rate of success in annotating unknown compounds (336/1,277). In addition, an increased sample size for this study may improve the reliability and confidence of the results in comparing the molecular features among vapers, smokers and nonusers, and can help elucidate how factors such as sex, age, and behaviors modulate these differences. There have been research that investigated sex-specific difference in the health outcomes of tobacco use,⁵⁰ yet there is limited investigation done on vaping, which requires future efforts.

E-cigarette introduces human exposure to a mixture of chemicals, and their interactions can result in unexpected and unknown synergistic effects, which challenges health science research in assessing safety. Our data obtained from state-of-art nontargeted chemical analysis with HRMS provided insights on how vaping can introduce xenobiotics and perturb biological functioning. These findings, including the remained nicotine intake (compared to cigarette), elevated lipid catabolism that can be inflammatory, and concerning absorption on harmful chemicals (flavoring, plasticizer) can drive new hypothesis for evaluating the health safety and toxicology of e-cigarettes.

Data Availability

Data are available from the corresponding author upon reasonable request.