Solvent Chemistry in the Electronic Cigarette Reaction Vessel

Knowledge of the mechanism of formation, levels and toxicological profiles of the chemical products in the aerosols (i.e., vapor plus particulate phases) of e-cigarettes is needed in order to better inform basic research as well as the general public, regulators, and industry. To date, studies of e-cigarette emissions have mainly focused on chromatographic techniques for quantifying and comparing the levels of selected e-cigarette aerosol components to those found in traditional cigarettes. E-cigarettes heat and aerosolize the solvents propylene glycol (PG) and glycerol (GLY), thereby affording unique product profiles as compared to traditional cigarettes. The chemical literature strongly suggests that there should be more compounds produced by PG and GLY than have been reported in e-cigarette aerosols to date. Herein we report an extensive investigation of the products derived from vaporizing PG and GLY under mild, single puff conditions. This has led to the discovery of several new compounds produced under vaping conditions. Prior reports on e-cigarette toxin production have emphasized temperature as the primary variable in solvent degradation. In the current study, the molecular pathways leading to enhanced PG/GLY reactivity are described, along with the most impactful chemical conditions promoting byproduct production.

The chemistry of PG and GLY has a rich history. The preparation of GLY in 1779 by Scheele, and his determination that it was susceptible to thermal decomposition during simple distillation [15][16][17][18][19] , predated even Wöhler's urea synthesis by half a century. By the mid-19 th century, acrolein 20 and acetic acid 21 had been identified as products of GLY decomposition. Wurtz synthesized PG in 1859, and determined that it could be oxidized to lactic acid in air in the presence of catalysts 22 . In 1904, Nef provided the foundation for the current understanding of GLY and PG chemistry 23 . He reported that heating GLY afforded hydroxyacetone, acetaldehyde, formaldehyde, acrolein, 3-hydroxypropanal, and a series of acetals. He discovered that GLY formed glycidol upon gentle heating in the presence of acetic acid. Nef also discovered that the decomposition of PG gave propanal. The literature moreover strongly suggests that there should be more compounds produced by PG and GLY than have been investigated in the majority of e-cigarette aerosol studies to date. We propose that since NMR is a technique that can enable relatively broad profiling of compound classes with limited sample perturbation, it will allow the detection and study of overlooked e-cigarette aerosol products as compared to studies that to date have relied nearly exclusively on chromatographic-based methods.
Recently, we reported the discovery of hemiformals 1a and 1b (major isomers observed are shown in Fig. 1), products of the reaction of PG/GLY with HCHO that are formed during PG/GLY aerosol generation 24 . Herein, we describe an extensive investigation of the products derived from vaporizing PG and GLY under relatively mild e-cigarette conditions, along with a detailed description of the molecular pathways leading to product formation. Products identified by NMR include glycidol (2), an International Agency for Research on Cancer (IARC) Group 2A probable human carcinogen 25 . In addition, we report several products identified for the first time in e-cigarette aerosols such as reactive vinyl alcohol isomers (3a and 3b) and dihydroxyacetone (4), the main ingredient in spray tan products that has raised concerns as an inhalation hazard.

Results and Discussion
General Methods. Aerosol sample production and initial analysis. The e-cigarettes used in this study consisted of two main components, a variable voltage/variable wattage (VV/VW) battery, the Innokin ® iTaste VV4, fitted with a KangerTech ® Protank-II clearomizer. The clearomizer contained a replaceable bottom heating coil (coils with resistances from 1.8-2.5 Ohms were provided by the manufacturer) embedded in a wick that was covered with the PG/GLY e-liquid during usage. E-liquids were composed of 1:1 v/v PG:GLY solutions, except where indicated. Single puffs of aerosolized e-liquid (50 mL) were drawn via a syringe directly into a DMSO-d 6 solution for analysis by NMR spectroscopy. Between 6-22 mg of aerosol were collected per puff. Such data from single puff samples was obtained because NMR is non-destructive, meaning that one can signal-average until sufficient signal-to-noise is obtained for the required analyses. Typical parameters included: a 30° observation pulse angle, a 6.2 sec repetition rate, and 64 k data point acquisitions for between 64 and 2048 acquisitions (between 0.12 and 2 hr). Line broadening of 0.3 Hz and a final data size of 64 k real data points were used for data processing. Structure assignments were confirmed by minute addition of authentic standards. This was accomplished by first  Figure 2 illustrates that more PG/GLY was consumed as a function of increasing device power. In addition to raising the power settings to directly elevate heating coil temperatures, elongating the puff duration likewise promoted PG/GLY consumption. It is well-known that e-cigarette operating temperatures modulate the extent of PG/GLY degradation 11 . In the context of the investigation herein, the experiments illustrated in Fig. 2 show that the ability of the device to produce aerosol mass and the intensities of novel aerosol product peaks in an NMR spectrum are proportional to the power delivered to the e-cigarette coil.

Results.
Avoidance of dry coils and burnt e-liquid. Conditions were chosen to avoid drying the heating coil and associated overheating of the e-liquid; it has been proposed that users can detect and self-regulate toxin intake (including HCHO) based on taste 10 . However, it is known that in traditional cigarettes the harsh taste from formaldehyde and other aldehydes is overcome due to the nicotine drive 26 and to cross-desensitization of transient receptor potential ankyrin subtype 1 (TRPA1) channels in sensory neurons 27,28 . Moreover, it has not been shown to what degree the use of flavorants in e-cigarettes dulls or overcomes any harsh taste from toxins.
Nonetheless, several steps were taken to avoid overheating of PG/GLY. First puffs were sampled as single acquisitions, or with 5 min puff intervals or longer between any two puffs, and no primer puffs were used. Extra wicking and cooling of the coil were performed by pulling the syringe (cold puff) without activating the device, again 5 min before the sampling puff. These methods ensured that aerosols were never generated from dried coils, since e-liquid had cooled between puffs and had completely covered the coils prior to drawing any aerosol samples. Also, an average 3-5 s puff duration was used. In addition, the Innokin ® iTaste VV4 battery (1000 mAh) that was used herein possesses a variable output of 6.0-15 W and variable voltages of 3.0-6.0 V, and is compatible with a variety of clearomizers. This device is described by the manufacturer as accurate to within 0.1 W with "no fluctuation and unexpected dry hits or burned e-liquid", and has its highest 15 W setting precisely maintained between 0.8 to 2.5 ohms 29 . Both the Innokin battery and KangerTech ® Protank-II clearomizer had received positive reviews from the online vaping community.
Identification of aerosol products. The compounds that have been identified in this study are shown in Fig. 1; all of these compounds could be predicted based on the existing PG and GLY literature. Several are reported here for the first time in electronic cigarette aerosols (3, 4, 6-8). The determination of others (e.g., 2, 15) validates their recent discovery in e-cigarette aerosols, and moreover shows that they can form under milder conditions than those reported 25 . Structure assignments for all compounds were validated via spiking with authentic samples when available (Supporting Information). Formaldehyde hemiacetals. Glycerol formal was described by Nef in 1904 23 , and propylene glycol formal was described by Trillat and Cambier in 1894 30 . The production of 1a and 1b in e-cigarette aerosols 24 was thus not surprising, particularly since hemiformals are relatively stable compared to other acyclic hemicacetals. The toxicity of 1a and 1b has not been investigated. They reverted to HCHO and PG over a period of hours in aqueous solution. Since 1a and 1b have not been separable to date we validated the structure assignment using isotopically labeled CH 3 OCH 2 OH (methoxymethanol) as a model hemiformal. As depicted in Fig. 3, labeled CH 3 OCH 2 OH exhibited analogous peak position as well as the characteristic O-H proton splitting pattern corresponding to those reported for 1a and 1b. The 1 H-13 C decoupled spectrum B corresponded to that of the product formed by bubbling 12 CH 2 O in CH 3 OH (spectrum C). Upon spiking with D 2 O (spectrum D) the hydroxyl resonance at 6.14 ppm diminished and the methylene protons at 4.53 ppm appeared as a singlet. The full 1 H spectrum (E) of CH 3 OCH 2 OH/CH 3 OH in DMSO-d 6 is shown for completeness. The data are consistent with the predominant NMR peaks as arising from the isomers of 1a and 1b shown in Fig. 1.
Glycidol (2) and enols (3). NMR spectra derived from aerosols produced at increasingly higher wattage settings revealed that, in addition to greater overall PG/GLY consumption and product formation at higher coil temperatures (Fig. 2), specific products arise at different settings. An expansion of the regions in the NMR containing the proton resonances of glycidol (2) as well as those of the cis and trans isomers of the propanal enols (3) is shown in Fig. 4. At 10 W, the peaks from 2 are present; however, the enol resonances are observable with adequate S/N only beginning at 12 W. This observation is in keeping with observations by Nef in 1904 when he described glycidol (2) as a "low temperature" product 23 . More recently, Laino and co-workers reported that the dehydration of glycerol to form 2 is the rate limiting step of a GLY degradation pathway 31 . Compound 2 has been shown to react with DNA 32 It appears to be a relatively minor aerosol component based our studies to date (vide infra). Enols 3a and 3b have been previously found to persist for up to two weeks in dilute acetone at room temperature 33 . Their potential inhalation toxicology is not clear, though enol reactivity is well-known.
Aldehydes. An expansion of the aldehyde region of the 1 H NMR spectrum of an aerosol generated from PG/ GLY at 15 W, along with corresponding structure assignments, is shown in Fig. 5. Acrolein, compound 5, has long been known as a decomposition product of GLY 20 . In fact, it is the target molecule of the most common qualitative chemical test for the detection of GLY 19 . Several routes have been proposed for its formation, including one from a recent study showing its formation via the GLY dehydration product 2 31 . Acrolein is a well-known hazardous air pollutant 11 . Importantly, the doublet at 9.56 ppm corresponding to the aldehyde proton of 5 is  highly prominent relative to those of the other products. This is significant because DNPH trapping cartridges used in prior chromatographic studies of e-cigarette aldehydes 13 have been reported as unreliable for the determination of acrolein levels 34 , affording low recoveries. Prior reports of acrolein levels approaching those of other aldehydes have been characterized as having been attributed to overheating conditions 11 . Of the remaining aldehydes 6-10, acetaldehyde (9) has garnered significant attention in e-cigarette aerosols because it is a possible human carcinogen (IARC Group 2B) 35 .
Reaction pathways. The reactions of PG and GLY under thermal conditions are predominantly dehydrations and oxidations (Figs 6 and 7). As mentioned previously, the conversion of PG to propanal (10) was reported in 1904 by Nef 23 . The oxidation of PG has also been previously shown to afford acetone (11), acetaldehyde (9), HCHO, acetol (12) and its tautomer lactaldehdye (6) along with dehydration product (5) 36 .
At the lowest device power setting, 12 was the major detectable aerosol product in the 1 H NMR shown in Fig. 2. Compounds 12 and 6 can result from O-H proton abstraction via a higher temperature pathway owing to the relatively higher O-H bond dissociation energies. Cleavage of C-C bonds from the oxygen radicals would result in the formation of 13 and 14, which can also form as oxidation or retro-aldol products of 12. Compounds 13 and 14 were each observed herein at power settings between 10-15 W. A recent report on PG dehydration has shown that propylene oxide serves as an intermediate towards the formation of 10 and 11, or can alternatively form allyl alcohol (15) 25,37 GLY is oxidized to form dihydroxyacetone (4) and glyceraldehyde (8) by H atom abstraction, C-H bond cleavage and tautomerization. Compound 4 has been shown to possess genotoxic 38 and mutagenic 39 properties. Compound 12 is produced by the dehydration of GLY 40,41 . Compound 12 can further degrade to HCHO and acetaldehyde (9) 41 . Compound 8 can furnish 7 via a retro-aldol reaction. An ensuing retro-aldol affords HCHO, the abundance of which increases while that of 7 diminishes at high temperatures. Dehydration of GLY affords 2 and subsequently 5 31 .
Physical factors modulating e-cigarette aerosol composition. As noted above, there is general agreement that increasing battery output, and thus the temperature of the heating coils, enhances the levels of PG/GLY degradation products in e-cigarettes. However, this does not clearly explain inter-laboratory differences in toxin levels. A recent report summarized HCHO (and other carbonyl) levels reported from five independent studies, each performed in 2014, and showed that the lowest HCHO range found was 3.2-3.9 ng/puff and the highest was 660-3400 ng/puff. The differences were attributed mainly to the different types of DNPH trapping cartridges used in each laboratory 11 . In addition to an inability to effectively recover specific compounds such as 5, DNPH trapping columns were designed for gas-phase rather than aerosol compounds.
To begin to understand the unique differences that may arise between devices, such as a poor electrical connection or a manufacturing flaw, we investigated the vaping product yields of hydroxyacetone (12) and acetaldehyde (9) (Supporting Information) using three identical KangerTech ® Protank II clearomizers. Each clearomizer was equipped with a unique, identical 2.2 Ω coil and each fitted to one of three identical Innokin ® iTaste V4 batteries. Samples were run in triplicate. At power settings of 12 W and 15 W, one of the three devices afforded yields of both products that were several-fold higher compared to the other two devices. However, the mass of PG/GLY consumed from the reservoir of the outlier device during collection of the samples presented above was comparable to the other devices (Supporting Information).
One issue that will affect the degree of product degradation is the efficiency of latent heat transfer from the heating coils to the e-liquid. The bottom coil configuration of the clearomizer used herein is more efficient than the typically shorter top load coils of relatively older clearomizers and cartomizers. In addition, multiple coils should be the most efficient at dispersing heat over the entire solvent volume, and would be expected to afford the least PG/GLY degradation. Figure 8 shows a representative comparison of expansions of the NMR spectra of aerosols produced by each of four clearomizers. All aerosols contained 1a and 1b at 6.2 ppm; however, in the case of the CE4 clearomizer (top spectrum) peak broadening was significant due to enhanced exchange due to the high concentration. It is apparent by visual inspection, as expected, that the top coil clearomizer afforded the most degradation products, whereas one of the dual coil clearomizers afforded the least. The issue of heat transfer is thus not only significant in designing safer devices, but will contribute to inter-laboratory variability in reported e-cigarette product levels. For instance, one would expect that differences in coil gauges, the number of turns and morphological defects would be among factors influencing heat transfer and product formation.
To investigate the role of inter-coil differences in product formation, we examined eight replacement coils for a clearomizer. Three or more coils were tested at each of two resistances for their effect on PG/GLY product levels, using the same battery. However, there was some statistically significant variability in the product levels observed from different coils with the same resistance values (Fig. 9). Consistent trends were also observed, such as glycidol (2) formed at the lowest levels as compared to the other three products shown. Hydroxyacetone (acetol, 12) was the major product using the lower resistance coils. Dihydroxyacetone (4) was observed as a relatively abundant product formed via the majority of the eight coils. Overall, the fact that there are statistical differences (t critical = 4.303, DF = 2, p < 0.05) in product levels using the same device and power levels while varying only coils of the same model and resistance levels, exacerbates the challenges inherent in controlling device and inter-laboratory consistency.
Chemical factors modulating e-cigarette aerosol composition. Many of the reaction sequences shown in Figs 6 and 7 have rate limiting steps with activation energies well over 50 kcal/mol determined under pyrolysis conditions. For example, Laino et al. reported that heating PG to 537 °C for 30 s without O 2 led to a nearly 99.9% recovery of unreacted compound, and that GLY had very similar thermal stability 37 . However, since e-cigarettes are used under aerobic conditions, the presence of O 2 will play a significant role in promoting oxidation.
Although the chemistry of PG and GLY has been studied for centuries, there is still a relative lack of understanding of their reactions under aerobic conditions, as recently noted by Diaz 36 and Hemings et al. 42 The evidence reported to date, however, clearly shows that O 2 initiates the thermal degradation of PG and GLY at significantly lower temperatures as compared to anaerobic (pyrolysis) conditions. Diaz et al. demonstrated that O 2 -promoted hydrogen abstraction from PG (Fig. 6) to form products derived from carbon-centered radicals at temperatures as low as 127 °C over the course of 6-14 seconds in the presence of O 2 36 No reaction was observed under the same conditions under an inert atmosphere.
Based on the available evidence, GLY is relatively more stable to oxidation as compared to PG, with polymerization and decomposition reportedly initiating at 200 °C 43 . At temperatures close to the boiling point of GLY (290 °C), Sabatier and Gaudin produced glyceraldehyde (8) as a main intermediate in the formation of CO 2 and ethanol, 9 and 10 44 . Stein and co-workers reported no evidence of GLY degradation at temperatures of up to 200 °C in the presence of O 2 , but observed discoloration after heating samples to 250 °C 45 . They ascribed H-atom abstraction as the initial free radical reaction (Fig. 7), analogous to that observed for PG by Diaz 36 , as leading to the production of acrolein, acetaldehyde, and formaldehyde.
In order to confirm that O 2 promotes product formation, we compared product yields obtained under ambient vs. reduced-O 2 conditions. An obvious decrease in the intensity of many 1 H NMR peaks -corresponding to decomposition products was observed when samples of aerosolized PG/GLY were collected in a sealed glove-bag that had been flushed with N 2 (Supporting Information). Interestingly, specific product yields associated with decomposition products in the anaerobic spectra, including glycolaldehyde (7) and hydroxyacetone (12) were  clearomizer, varying only the replaceable single-coil heating element. Three single-puff samples were collected at a modest power setting of 10 W from eight different replacement coils, three of which were labeled 2.2 Ω resistance by the manufacturer, while the other five were labeled 1.8 Ω resistance. (A) The intensity of NMR signals from several degradation products were compared by relative integration to the intensity of un-degraded PG/GLY peaks; the relative intensity of the thermal degradation products formic acid (Plot B, compound 14), hydroxyacetone (Plot C, compound 12), dihydroxyacetone (Plot D, compound 4), and glycidol (Plot E, compound 2) are plotted as percentages of the intensity of residual (PG/GLY). The average value among the eight replacement coils is plotted as a grey dashed line in plots B-E. All error bars denote a 90% confidence interval (n = 3). The asterisk (*) in plot C denotes p < 0.05 as determined by a two-variable, unpaired t-test. In addition to oxidation, the other main thermal reaction of PG and GLY is dehydration. It is well-known that dehydration reactions are catalyzed by acid. In 1985 Rossiter described the degradation of aqueous glycol solutions, including PG, in the presence and absence of air and metals, noting that the acidic products that form as a result of thermal breakdown decrease the solution pH, catalyzing further degradation 46 . More recently, Nimlos and co-workers showed that activation energies for the dehydration of neutral GLY ranged from 65.2-79.5 kcal/ mol, but were lowered to 20-25 kcal/mol when GLY was protonated 47 . In addition to protonation, metal catalysis is well-known to promote PG and GLY reactivity. Laino has also modeled the interactions of PG and GLY with various metal surfaces which would have relevance to e-cigarettes due, for instance, not only to interaction with device components but also with metals and metal nanoparticles found in e-cigarette aerosols 48 .
Although acidic products have been observed in e-cigarette aerosols ( Fig. 2b; the peak downfield of 10 ppm, for example), their contribution to the overall pH will not generally be as significant as relatively more abundant acidic additives, such as certain flavorants. The study of the effects of acidic flavorants on product profiles during vaping is currently under investigation in our laboratories.
Limitations of this study include the relative lack of sensitivity of NMR as compared, for instance, to mass spectrometry. Moreover, the NMR spectra of e-cigarette aerosols have overlapping peaks, hindering complete product profiling. In addition, studies were performed on single puff samples, which depresses the average temperature of the heating coils as compared to multiple puff real-world usage, thereby likely underestimating realistic aerosol product levels. However, the primary goals of this investigation were (i) analytical target discovery, namely identifying aerosol components that were overlooked or under-investigated to date in the e-cigarette field, as well as (ii) clarifying the reasons for some of the discrepancies in the results reported by various labs, and (iii) an extensive description of the chemical pathways of PG and GLY degradation in the context of e-cigarette usage.

Conclusion
In this investigation we used NMR for e-cigarette aerosol product identification with no aerosol sample processing, apart from dilution in DMSO-d 6 . As had been proposed, a main finding was that the e-cigarette solvents PG and GLY afford products that are fully consistent with prior studies of their pyrolysis and combustion. In addition, the results herein suggest NMR as a viable alternative to DNPH trapping cartridges for monitoring challenging, reactive toxins such as acrolein. Finally, (i) the sensitivity of PG and GLY to thermal oxidation, (ii) the catalysis of their dehydration reactions by acids and/or metals, and (iii) the variability in the heat transfer efficiencies of individual clearomizers and heating coils should be taken into account when considering strategies to minimize toxin production and inter-laboratory inconsistencies in evaluating these devices.