Flavoring Compound Chemical Class and Vaping Conditions Determine Toxic Carbonyl Emissions from E‑Cigarettes

Abstract

The increasing use of flavored electronic cigarettes (e-cigarettes) raises concerns about their potential impact on the emission of harmful chemicals, particularly carbonyl compounds. This study systematically examines the effects of eight representative flavoring chemicals from four major classes including esters (ethyl acetate, ethyl butyrate), alcohols (menthol, ethyl maltol), aromatic aldehydes (benzaldehyde, vanillin), and terpenes (limonene, linalool) under varying power outputs (50 and 90 W), base liquid composition [propylene glycol (PG)/vegetable glycerin (VG) ratios (80:20, 50:50, and 20:80)], and flavor concentrations (1 and 5 mg/mL). Across all conditions, flavored e-liquids tend to produce carbonyl emissions that are higher than those of unflavored controls. Terpene-based flavors showed the strongest effects, with formaldehyde emissions being up to 2-fold higher and acrolein emissions up to 8-fold higher, frequently exceeding short-term exposure limits. Aromatic aldehydes and alcohols also increased emissions, though to a lesser extent, while esters showed smaller or inconsistent effects. The influence of flavors was further modulated by their concentration, PG/VG ratio, and device power, with higher concentration, VG content, and power amplifying emissions. These results highlight the complex interactions among e-liquid composition, flavor class, and vaping conditions, demonstrating that certain flavorings substantially elevate toxicant emissions. These findings underscore the importance of considering flavor composition, device power, and base material in evaluating the potential health risks associated with e-cigarette use.

1. Introduction

The rapidly growing popularity of electronic cigarettes (e-cigarettes) has raised concerns about potential adverse health effects for users. − E-cigarette aerosols contain several harmful and potentially harmful constituents (HPHCs), including dangerously high levels of carbonyls. − Among the HPHCs found in e-cigarette aerosols, carbonyl compounds such as formaldehyde, acetaldehyde, acetone, acrolein, propionaldehyde, and benzaldehyde are particularly prominent, with concentrations of formaldehyde often exceeding occupational exposure limits. ,, These compounds are known to exhibit pulmonary toxicity and may have vasoactive and broncho-reactive effects.

Various factors influence carbonyl production during e-cigarette use, including device construction (e.g., top/bottom coil, mod/pod devices, etc.), power settings, coil material, e-liquid composition (i.e., flavor composition and concentration, base material), and puff topography (i.e., puff duration and puff flow rate). ,,− These factors can be broadly categorized into two groups: (i) those that are a “chemical reaction-driven factor” involving the thermal decomposition of flavoring compounds and their interactions with the base material constituents [propylene glycol (PG) and vegetable glycerin (VG)] and the e-cigarette coil materials, and (ii) those that are a “transfer-based factor” (i.e., e-cigarette construction, power settings, e-liquid viscosity, and puff topography), which affect mass and heat transfer within the device.

Understanding how these factors and their various permutations influence carbonyl emissions is crucial for advising e-cigarette users about potential health risks associated with different devices and configurations. It is also essential for informing policies on regulating certain e-cigarette designs and e-liquid constituents. The impact of physical parameters on carbonyl emissions has been well-documented and consistent across most published research, ,,− , though variations in absolute values are often attributed to the wide variety of e-cigarette devices and differences in sampling techniques, such as variations in gas-particle distribution of carbonyls and differences in particle collection efficiencies. However, significant discrepancies persist among studies regarding the role of flavoring compounds in carbonyl emissions, even in aerosols produced by the same or similar devices. ,,,−

We have shown that flavoring compounds can have a substantial impact on carbonyl formation during e-cigarette use. For example, unflavored e-liquids have been observed to produce significantly lower amounts of carbonyls compared with flavored e-liquids. The relative abundance of carbonyls also varies among different flavors, indicating that flavoring compounds can strongly influence carbonyl production, and this effect is dependent on the specific flavor. Furthermore, it has been noted that carbonyl emissions increase with the concentration of flavor in the e-liquid. These findings have been corroborated by several studies, , although some researchers continue to debate the extent of the impact of flavoring compounds on carbonyl formation. ,

Flavorings can contribute to carbonyl production through direct thermal decomposition or by chemically catalyzing the thermal decomposition of the main e-liquid constituents, PG and VG. The use of flavoring in e-cigarette base material has become a central aspect of e-cigarette marketing, and while FDA regulations limit the sale of flavors to tobacco and menthol among major brands, a wide variety of flavored e-liquids continues to be offered by smaller manufacturers. This complexity is further compounded by the large number of flavors and their varying concentrations in the e-liquids. To gain a deeper understanding of the role that flavors play in carbonyl production, it is necessary to reduce this complexity by focusing on the basic chemical constituents and classes of flavoring compounds. It is logical to expect that a flavoring’s thermal stability and its interaction with other e-liquid constituents and the coil will depend on its chemical properties, particularly its chemical class. By examining the main chemical classes of flavoring compounds commonly found in e-liquids, such as esters, alcohols, aromatic aldehydes, and terpenes, and selecting representative compounds from each class, such as ethyl acetate, ethyl butyrate, menthol, ethyl maltol, benzaldehyde, vanillin, limonene, and linalool, , this study primarily aims to determine whether the chemical class of these chemical compounds can predict their effect on carbonyl production during e-cigarette use.

Previous studies on the effect of flavoring compounds and other additives on carbonyl production have generally been limited to a single combination of PG/VG ratio and e-cigarette power settings, , with few studies varying the concentration of flavorings in e-liquids. , Given that chemical reaction pathways, such as thermal decomposition and interactions with the base e-liquid constituents, depend on temperature and reactant concentrations, variations in e-liquids and e-cigarette power/temperature settings used in earlier studies complicate efforts to resolve discrepancies and provide clear guidance to the public and policymakers on the relative health risks associated with various flavoring additives in e-liquids. By examining how the interplay of power, PG/VG ratio, and flavor concentration affects the role of flavorings in carbonyl production during vaping, this study seeks to provide a clearer understanding of the significance of flavors in carbonyl emissions. This comprehensive approach will yield critical information on which types of flavoring compounds, and under which conditions, contribute most to carbonyl emissions, thereby clarifying the health risks associated with e-cigarette use and supporting more effective regulation.

2. Materials and Methods

2.1. Chemicals and Reagents

Acetonitrile (HPLC grade; CAS 75-05-8) was obtained from Fisher Chemical. Tetrahydrofuran (non-UV, HPLC grade, 99.7%, stabilized with 250 ppm of BHT; CAS 109-99-9) was purchased from Alfa Aesar (Ward Hill, MA). Propylene glycol (PG; 1,2-propaneidiol, ACS, 99.5%; CAS 57-55-6) and glycerol (VG; 99%+; CAS 56-81-5) were obtained from Thermo Fisher Scientific (Ward Hill, MA). A certified carbonyl calibration mixture (TO11/IP-6A Aldehyde/Ketone-DNPH Mix; Sigma-Aldrich, CRM4M7285) containing formaldehyde-DNPH, acetaldehyde-DNPH, acetone-DNPH, acrolein-DNPH, propionaldehyde-DNPH, crotonaldehyde-DNPH, n-butyraldehyde-DNPH, benzaldehyde-DNPH, isovaleraldehyde-DNPH, valeraldehyde-DNPH, o-tolualdehyde-DNPH, m + p-tolualdehyde-DNPH, hexaldehyde-DNPH, and 2,5-dimethylbenzaldehyde-DNPH was obtained from Sigma-Aldrich. DNPH-silica cartridges (Sep-Pak Plus Short Body, 350 mg; part no. WAT047205) were purchased from Waters (Milford, MA, USA). Flavoring compounds, including ethyl acetate (99.5%; CAS 141-78-6), ethyl butyrate (99%; CAS 105-54-4), menthol (99%; CAS 89-78-1), ethyl maltol (99%; CAS 4940-11-8), benzaldehyde (99%; CAS 100-52-7), vanillin (99%; CAS 121–33–5), limonene (97%; CAS 5989-27-5), and linalool (97%; CAS 78-70-6) were all obtained from Sigma-Aldrich.

2.2. E-Cigarette E-Liquids and Experimental Matrix

Table outlines the e-cigarette and e-liquid types and vaping conditions used in this study. Due to the extensive variability in PG/VG ratios, flavors, and e-cigarette coil temperature/power settings, testing every possible combination was impractical. To address this challenge, we adopted a strategic approach as follows: (1) selection of e-cigarette type: we chose a “mod” e-cigarette equipped with a Kanthal (FeCrAl, 0.15 ohm dual mesh) coil, a currently popular material/type for its flexibility and power control capabilities. This device represents the most adaptable current e-cigarette designs, ensuring relevance and versatility in our experiments. (2) Simplification of flavor variables: to gain fundamental insight into the role of flavors in carbonyl production, we simplified the vast array of e-liquid flavors by focusing on their primary chemical classes and compounds. Since a flavor’s thermal stability and interactions with other e-liquid components and the coil are likely influenced by its chemical properties, we categorized flavorings into four chemical classes: esters, alcohols, aromatic aldehydes, and terpenes. From each class, two of the most common flavoring compounds were selected based on literature prevalence: ethyl acetate, ethyl butyrate, menthol, ethyl maltol, benzaldehyde, vanillin, limonene, and linalool, respectively. Each chemical species was collected individually. (3) Investigation of combined effects: we further examined the combined effects of flavoring concentration, e-cigarette power, and liquid viscosity. The experimental design included the following parameters: PG/VG ratios: 80:20, 50:50, and 20:80; power settings: 90 W and 50 W; flavoring concentrations: low (1 mg/mL) and high (5 mg/mL). Each flavored liquid (i.e., a combination of PG/VG, flavoring chemical, and concentration) was prepared as a separate liquid and tested at the two power settings.

1. Experimental Matrix Used in This Study .

flavor Chemical		flavor concentration (mg/mL)		power (W)		PG/VG ratio
ethyl acetate	×	1	×	50	×	80/20
ethyl butyrate
menthol
ethyl maltol						50/50
benzaldehyde		5		90
vanillin
limonene						20/80
linalool
unflavored

^aSix samples (except unflavored liquid with 50% PG at 90 W, for which only 4 samples were collected) for each experimental variable were collected, where each sample was obtained from 5 puffs at a flow rate of 1.5 L/min, with a 4 s puff duration and a 30 s puff interval.

2.3. E-Cigarette Aerosol Generation

The entire experimental matrix listed in Table was tested using a custom-built puffing machine developed at the Desert Research Institute (DRI). This machine (Figure ) provides full control over puff volume, duration, interpuff intervals, and number of puffs and has been successfully utilized in a pilot study examining the effects of puff topography on carbonyl and carbon monoxide emissions. After an initial set of ten warm-up puffs, five additional puffs were collected using 2,4-dinitrophenylhydrazine (DNPH) cartridges (Sep-Pak XPoSure Plus Short Cartridge, WAT047205, Waters, Milford, MA, USA) for analysis. The tank of the “mod” was filled with 5 mL of e-liquid, and the liquid consumption was approximately 0.2 g after 15 puffs, which remained consistent across all tested experimental settings. Every precaution was taken to avoid “dry puffs”, such as ensuring that the tank had sufficient liquid and the wick was properly wet. All measurements were conducted six times (except unflavored liquid with 50% PG at 90 W, for which only 4 samples were collected), resulting in 610 observations (576 flavored samples and 34 unflavored). All measurements were done at a flow rate of 1.5 L/min, with a puff duration of 4 s and a puff interval of 30 s.

1.

Scheme of the DRI e-cigarette puffing machine.

The sampling method was validated by sampling air with a known amount of formaldehyde, one of the primary carbonyl compounds in e-cigarette emission. Formaldehyde from a formalin solution was introduced into a 1.8 m³ Tedlar bag, yielding concentrations of 503 ± 5 μg/m³ and 838 ± 8 μg/m³ after complete mixing. Also, blanks were collected prior to sampling to check the presence of carbonyls in the collection bag.

2.4. Carbonyl Analysis

DNPH cartridges were extracted with acetonitrile (ACN) following standard protocols. ,, The ACN extracts were analyzed by using high-performance liquid chromatography with photodiode array detection (HPLC-PDA). A 10 μL aliquot was injected into an HPLC system (Waters ARC HPLC with a 2998 PDA detector, Waters, MA, USA) equipped with an XBridge BEH C18 column (BEH C18, 2.5 μm, 75 × 3.0 mm; Waters, MA, USA). The HPLC was operated at a flow rate of 1.5 mL/min with ultrapure water (solvent A), acetonitrile (ACN, solvent B), and tetrahydrofuran (THF, solvent C) as the mobile phases. The gradient program was set to 65% A, 30% B, and 5% C for 12 min. Detection was performed at a wavelength of 360 nm. The HPLC response was calibrated in micrograms per milliliter using a certified calibration mixture (TO11/IP-6A Aldehyde/Ketone-DNPH Mix, CRM47285, Supelco, Bellefonte, PA), containing all six DNPH derivatives listed in Table S1. An eight-point external calibration was performed before analysis, with a calibration check every 24 h. The system was recalibrated if the response of any compound deviated more than 10%. Calibration curves for all carbonyls were linear, with R ² values exceeding 0.995. Detailed information regarding the analytes, including retention times, calibration parameters, and limits of detection, can be found in Table S1. Such losses were reported to occur during collection times exceeding 1 h, which is much longer than sampling times used in this study (about 3 min). Therefore, the acrolein values reported here could be underestimated, but this underestimation is likely to be negligible.

2.5. Statistical Analysis

The HPLC-PDA data was converted to carbonyl emissions (μg/puff) using the concentration in the extract (μg/mL), the extraction volume (mL), and the number of puffs. The emissions per puff were then analyzed by using the Mann–Whitney U rank test to determine the statistical significance of differences between the collection methods. The statistical test was performed using Python (version 3), with “mannwhitneyu” in the SciPy statistical functions. A p-value of less than 0.05 was considered indicative of a statistically significant difference.

To systematically investigate the effect of different parameters on carbonyl emissions, linear regression modeling and multivariate analysis of variance (MANOVA) were performed. Three separate linear regression models were constructed for formaldehyde, acrolein, and acetaldehyde emissions. All models used the following specification:

$$\mathrm{R}\mathrm{e}\mathrm{s}\mathrm{p}\mathrm{o}\mathrm{n}\mathrm{s}\mathrm{e}\sim \mathrm{p}\mathrm{o}\mathrm{w}\mathrm{e}\mathrm{r}+\mathrm{P}\mathrm{G}+\mathrm{c}\mathrm{o}\mathrm{n}\mathrm{c}\mathrm{e}\mathrm{n}\mathrm{t}\mathrm{r}\mathrm{a}\mathrm{t}\mathrm{i}\mathrm{o}\mathrm{n}+C(\mathrm{f}\mathrm{l}\mathrm{a}\mathrm{v}\mathrm{o}\mathrm{r}\text{\_}\mathrm{c}\mathrm{l}\mathrm{a}\mathrm{s}\mathrm{s})$$

where power, PG content, and concentration were modeled as continuous linear predictors, and flavor class was entered as a categorical factor. Models were fitted using ordinary least-squares (OLS) estimation via the statsmodels library in Python. Response variable transformations were selected individually for each aldehyde based on residual diagnostics and model fit criteria. For formaldehyde and acetaldehyde, the log transformation of responses was applied to improve residual normality and stabilize variance across the measurement range. Log-transformed coefficients can be directly interpreted as percent changes in the original response. For acrolein, the raw concentration scale was retained despite residual non-normality (Omnibus p < 0.001, kurtosis = 18.09) because log transformation substantially degraded model fit (ΔR ² = −0.14, ΔAIC = +4400) without improving residual properties. This reflects the inherent measurement noise as acrolein concentrations were low, which cannot be corrected by transformation. Model adequacy was assessed through residual diagnostics (Omnibus and Jarque–Bera tests for normality and the Durbin–Watson statistic for autocorrelation), condition numbers to assess multicollinearity, and coefficient of determination (R ²) statistics. Regression models with two-way interactions and polynomial terms were tested but exhibited severe multicollinearity (condition numbers >1 × 10¹⁶) and were rejected in favor of the main-effects-only models. Potential confounding between flavor class and concentration predictors was evaluated, given the factorial design structure (all unflavored have concentration = 0 only, while flavored have concentrations ≥1 only). The correlation between the concentration and flavor class was found to be negligible (r = 0.045). Variance inflation factors (VIFs) were also acceptable for all predictors (concentration: 3.25; flavor: 3.61; PG: 4.52; power: 8.61). This indicates that the concentration and flavor effects are reliably separable in the regression models despite the experimental design constraints.

To evaluate whether the three emitted aldehydes responded to experimental factors as a coordinated multivariate system versus independently, we conducted a Multivariate Analysis of Variance (MANOVA). MANOVA tests the null hypothesis that mean vectors of the response variables are equal across factor levels, using multivariate test statistics (Wilks’ Lambda, Pillai’s trace, Hotelling–Lawley trace, and Roy’s greatest root). All three log-transformed aldehydes were entered as joint response variables with the same predictor structure as the univariate regressions. Data were aggregated by computing means across replicates within each treatment combination prior to MANOVA to ensure independence of observations (n = 54 aggregated conditions: 6 unflavored baseline conditions +48 flavored conditions). Univariate follow-up F-tests were conducted for each aldehyde separately to identify which compounds drove the multivariate significance.

3. Results

The emissions of the main carbonyl compounds (i.e., formaldehyde, acetaldehyde, acetone, acrolein, propionaldehyde, and benzaldehyde) measured for the 8 flavors at different flavor concentrations, PG/VG ratios, and power settings are provided in Tables S2–S7. Figure shows formaldehyde emissions measured for a PG/VG ratio of 80:20. All flavors, except ethyl maltol and vanillin at 50 W and low concentration, produced higher formaldehyde emissions than the unflavored liquid at the same power. Not all measured increases in formaldehyde emission relative to the unflavored liquid are statistically significant due to a considerable puff-to-puff variability. However, several flavors, especially those of the terpene class (limonene and linalool), exhibit a substantial and statistically significant increase in formaldehyde emissions. At low flavor concentration and 50 W, limonene- and linalool-flavored e-liquids produce ∼30% more formaldehyde than the unflavored liquid. At low concentration and 90 W, these terpene-flavored liquids produce ∼64% more than the unflavored liquid. At the high flavor concentration, benzaldehyde, menthol, and vanillin (only at 90 W), limonene, and linalool demonstrate a statistically significant increase in formaldehyde emissions relative to the unflavored liquid. This increase for benzaldehyde, vanillin, and menthol is relatively modest, 12–33%. On the other hand, limonene and linalool show a much larger increase (71% and 64% at 50 W and 182% and 161% at 90 W, respectively).

2.

Formaldehyde emissions (μg/puff) from unflavored e-liquid and eight flavored e-liquids at a PG/VG ratio of 80:20 with corresponding p-values across different experimental conditions. Error bars represent standard deviations based on six measurements.

Similar plots showing comparisons among flavored and unflavored liquids for other PG/VG ratios and for acetaldehyde and acrolein are shown in Figures S1–S8. The results demonstrate that different flavoring compounds affect carbonyl emissions relative to the unflavored liquid of the same PG/VG ratio and vaped at the same power settings. There are also differences in emissions among individual flavors, with many of such differences being statistically significant (Figures S9–S11).

A comparison of emissions from the same flavor liquids but different flavor concentrations provides additional confirmation that flavors affect e-cigarette carbonyl emissions. As an example, Figure shows a comparison of formaldehyde emissions between low- and high-concentration pairs for individual flavors. Even though not all concentration pair differences are statistically significant at 95%, all high-concentration liquids produce higher formaldehyde emissions than the low-concentration liquids. Similar trends are observed for acetaldehyde and acrolein (Figures S12 and S13).

3.

Formaldehyde emissions (μg/puff) from eight flavored e-liquids, comparing low and high flavor concentrations with corresponding p-values across different experimental conditions. Error bars represent standard deviations based on six measurements.

Table shows the percentage of average increases in emissions of formaldehyde, acetaldehyde, and acrolein relative to unflavored liquids for different PG/VG ratios and flavor compounds. The increase in emissions was calculated by averaging emissions for each compound at both concentrations and both power settings. Clear differences are observed not only among individual compounds but also among compound classes. There are also notable differences in how emissions of formaldehyde, acetaldehyde, and acrolein depend on the flavor class. For example, the terpenes class (limonene and linalool) produce the highest increases in emissions relative to the unflavored liquid for all three compounds. On the other hand, the esters (ethyl acetate and ethyl butyrate) cause a medium increase in formaldehyde and acetaldehyde emissions while having the smallest effect on acrolein emissions. The effect of flavors also appears to depend somewhat on the PG/VG ratio. The increase in formaldehyde emissions tends to be smaller for higher VG content, while the increase in acetaldehyde shows the opposite trend. This suggests that carbonyl emissions are likely caused by complex chemical processes that need to be investigated further.

2. Percentage of Average Increase in Emissions of Formaldehyde, Acetaldehyde, and Acrolein Relative to Unflavored Liquids Emissions at Different PG/VG Ratios and Flavor Compounds^a .

		formaldehyde			acetaldehyde			acrolein
chemical class	flavor compound	80:20	50:50	20:80	80:20	50:50	20:80	80:20	50:50	20:80
esters	ethyl acetate	36.0%	22.5%	11.3%	45.8%	49.0%	10.4%	–7.98%	–14.1%	35.5%
	ethyl butyrate	35.7%	27.8%	6.10%	42.7%	49.9%	9.5%	–0.779%	–22.3%	54.9%
aromatic aldehydes	vanillin	25.9%	16.0%	11.0%	–19.8%	33.1%	149%	24.3%	122%	158%
	benzaldehyde	49.2%	40.8%	12.2%	–5.46%	162%	156%	61.8%	257%	169%
alcohols	ethyl maltol	3.89%	9.50%	3.16%	25.2%	–0.823%	–3.79%	46.7%	28.2%	151%
	menthol	37.3%	12.3%	12.9%	14.8%	86.8%	–12.4%	31.4%	194%	–4.72%
terpenes	limonene	116%	92.6%	87.1%	117%	231%	177%	172%	491%	719%
	linalool	114%	87.0%	80.0%	124%	188%	206%	156%	450%	765%

^aThe values represent an average for each compound and PG/VG ratio for all flavor concentrations and power settings.

It has been demonstrated by numerous previous studies that e-cigarette power has a profound effect on carbonyl emissions. ,, The composition of the base material (PG and VG combinations) was also shown to affect carbonyl emissions. − It is not known, however, how these parameters affect the emissions of the flavored liquids. Table shows the percentage of average increase in emissions of formaldehyde, acetaldehyde, and acrolein relative to the unflavored liquids across all compounds for different power settings, flavor concentrations, and PG/VG ratios. As shown above, higher concentration liquids cause a stronger increase in carbonyl emissions than low concentration liquids for both power settings and all PG/VG ratios. The increase in e-cigarette power tends to increase the contribution of flavors to emissions, except for liquids with a high PG content. Similar to the discussion of Table , the increase in formaldehyde emissions relative to the unflavored liquids tends to be smaller for higher VG content liquids, while acrolein emissions tend to be higher.

3. Percentage of Average Increase in Emissions of Formaldehyde, Acetaldehyde, and Acrolein Relative to Unflavored Liquids at Different Power Settings, Flavor Concentrations, and Different PG/VG Ratios^a .

		formaldehyde			acetaldehyde			acrolein
power output	concentration	80:20	50:50	20:80	80:20	50:50	20:80	80:20	50:50	20:80
50 W	low	31.5%	7.72%	12.0%	2.31%	69.4%	2.05%	34.4%	115%	43.8%
	high	78.8%	41.0%	29.9%	84.2%	151%	57.2%	86.6%	204%	241%
90 W	low	29.0%	33.9%	15.0%	4.70%	51.1%	88.6%	28.1%	68.4%	213%
	high	69.9%	71.6%	55.1%	80.9%	128%	198%	92.8%	366%	526%

^aThe values represent an average for each combination of power and concentration.

These findings were further confirmed by linear regression modeling. Three separate linear regression models evaluated formaldehyde, acetaldehyde, and acrolein emissions as functions of power, PG content, concentration, and flavor class (610 observations per model). Formaldehyde and acetaldehyde were modeled with log-transformed responses; acrolein was modeled on the raw concentration scale (see Materials and Methods). Model fit quality differed substantially across aldehydes (Table S8), likely reflecting differences in measurement precision.

The regression coefficients, their standard error, and statistical significance are summarized in Table . Device power was a significant predictor for all three aldehydes (p < 0.001). For formaldehyde, a 40 W increase (from 50 to 90 W) corresponds to 1.93-fold higher emissions (95% CI: 1.88–1.98). Acetaldehyde shows a similar but slightly weaker power effect (∼1.79-fold). Acrolein exhibits the weakest power effect (∼1.024-fold). PG content has a strong reduction effect for all three aldehydes (p < 0.001). A 60 percentage-point increase in PG (from 20% to 80%) leads to a 45.8% formaldehyde reduction (95% CI: 43.2–48.1%). Acetaldehyde reduction is more modest (36.2% over the same PG range), while acrolein shows the strongest effect (66.5% reduction). Flavor concentration was universally significant (p < 0.001) and showed the largest relative effects in acetaldehyde. With a per unit mg/mL increase in concentration, formaldehyde increases ∼5.9%, acetaldehyde increases ∼12%, and acrolein increases ∼0.7%.

4. Linear Regression Coefficient Estimates (Standard Error) and Statistical Significance (n = 610) .

parameter	formaldehyde (log)	acetaldehyde (log)	acrolein (raw)
intercept	–1.35 (0.0543)	–3.20 (0.0995)	0.0755 (0.00991)
alcohol	–0.0456 (0.0462)	–0.250 (0.0852)	–0.00513 (0.00843)
aromatic aldehyde	+0.0493 (0.0462)	+0.0156 (0.0852)	–0.000562 (0.00843)
ester	+0.0341 (0.0462)	–0.0811 (0.0852)	–0.0232 (0.00843)
terpene	+0.469 (0.0462)	+0.603 (0.0852)	+0.0496 (0.00843)
power (per watt)	+0.0164 (0.000466)	+0.0146 (0.000860)	+0.000583 (0.0000851)
PG (per % point)	–0.0130 (0.000380)	–0.00603 (0.000701)	–0.00169 (0.0000694)
concentration (per mg/mL)	+0.0587 (0.00480)	+0.114 (0.00884)	+0.00686 (0.000876)

^aUnflavored liquids were used as the reference for flavored classes.

^b p < 0.01.

^c p < 0.001.

As expected from the experimental results discussed above, terpenes show the strongest and most consistent flavor class effect relative to the unflavored baseline across all three emitted aldehydes (p < 0.001). The other three flavor classes do not show statistically significant effects relative to unflavored liquids, except for alcohol-flavored liquids, showing a significant reducing effect on acetaldehyde (p = 0.003) and a weak reducing effect of ester-flavored liquids on acrolein (p = 0.008).

To determine whether the three aldehydes responded to experimental factors as a coordinated system, Multivariate Analysis of Variance (MANOVA) was conducted on aggregated data (n = 54 experimental conditions). All four multivariate test statistics (Wilks’ Lambda, Pillai’s trace, Hotelling–Lawley trace, and Roy’s greatest root) confirmed that all factors significantly affect the multivariate aldehyde response profile (p < 0.0001; Table S9). PG and device power exhibit equally strong multivariate effects (Wilks’ Λ = 0.163 and 0.181, respectively). Flavor class produced a moderate multivariate effect (Λ = 0.322), while concentration showed the weakest but still highly significant multivariate influence (Λ = 0.570). The multivariate significance across all factors indicates that the three aldehydes respond as a coordinated system driven by shared chemical mechanisms, despite their compound-specific sensitivities.

Univariate follow-ups on aggregated data revealed substantially higher explanatory power than raw data models (R ² increased by 0.08 to 0.15), which is expected due to noise reduction through aggregation. All univariate models on aggregated data remained highly significant (F > 16.92, p < 0.001), confirming that the multivariate significance was not driven by a single outlier compound and that all three aldehydes respond significantly to the experimental factors regardless of whether analyzed individually or in the multivariate framework. All univariate regression models showed acceptable residual properties (Omnibus test p > 0.004 for log-transformed responses; condition numbers <1000). Formaldehyde and acetaldehyde models achieved near-normal residuals following log transformation (kurtosis ≈3.3–3.9). Acrolein’s residual non-normality (kurtosis = 18.09) reflects measurement noise at low concentrations but did not warrant transformation given the severe penalty to model fit (see Materials and Methods). MANOVA residuals on aggregated data showed adequate normality consistent with the reduced noise of condition means.

4. Discussion

The results of this study highlight the significant role that flavors play in the production of carbonyl compounds during e-cigarette use. The addition of a flavor compound tends to increase carbonyl emissions relative to the unflavored e-liquid of the same PG/VG ratio and vaped at the same power setting. This increase in emissions depends on the flavor class, with the two terpene compounds (limonene and linalool) showing the strongest effect. For formaldehyde, the smallest effect was observed for ethyl maltol, while for acrolein, the ester compounds showed the smallest effect. The contribution of different compounds to carbonyl emissions was also found to depend on the PG/VG ratio and power setting. All of this demonstrates that carbonyl production during e-cigarette use is a result of complex processes that depend on the flavor compound, e-liquid base composition, and the e-cigarette power setting.

The effects of power and PG/VG ratio on carbonyl emissions have been reported by numerous previous studies. A higher power leads to higher temperatures, leading to thermal decomposition of both the base liquid constituents and the flavoring compounds, yielding higher levels of carbonyls and other degradation products. ,,, A higher VG (or lower PG) increases e-liquid viscosity, which slows the liquid transfer to the heating element (the coil) and leads to the increase in temperature and higher emissions. −

The observed increase of emissions with increasing power and VG content suggests that thermal degradation is a leading cause of carbonyl emissions. This observation can help with understanding the differences between the tested compound classes. Terpenes were observed to have the strongest effect on emissions, which is likely explained by their chemical structure. Terpenes are oligomers of the 5-carbon isoprene unit containing several CC double bonds. Under vaping conditions (hot coil, O₂, and PG- and VG-derived radicals), terpenes can form formaldehyde and other carbonyls mainly via radical and oxidative cleavage of their CC-containing backbones, especially at allylic and tertiary sites, followed by β-scission and CC cleavage that release small carbonyl fragments such as formaldehyde. − Other flavoring chemical classes can also undergo chemical transformations under e-cigarette conditions, which warrants further dedicated studies of chemical mechanisms of reactions occurring in e-cigarettes.

The observed increase in carbonyl emissions due to the addition of flavoring compounds is substantial. For example, the terpene compounds lead to up to a factor of 2 higher formaldehyde emissions and a factor of 8 higher acrolein emissions. This is noteworthy considering that flavor concentrations in our experiments were about 0.5% by weight, while total flavor concentrations in commercially available e-liquids could reach 10–20%. While our experiments show that not all flavor compounds show as high emissions as terpene-class flavors, the cumulative contribution of other flavor compounds could also be substantial.

Among the carbonyl compounds studied, formaldehyde and acrolein are of particular concern due to their low threshold limit values (TLVs). The TLV-C (ceiling) or TLV-STEL (short-term exposure limit) values set by the American Conference of Governmental Industrial Hygienists (ACGIH) were used to estimate the risks associated with e-cigarette use. The TLV-C for acrolein is 0.11 mg/m³, while for formaldehyde, the TLV-STEL is 0.37 mg/m³ (15 min time-weighted average, not to be exceeded). Given that the average tidal volume of a healthy adult is approximately 500 mL, these limits correspond to approximately 0.055 μg/puff for acrolein and 0.185 μg/puff for formaldehyde. Our findings indicate that formaldehyde emission from unflavored as well as all flavored e-liquids across all tested experimental settings exceeds its threshold, raising concerns about potential health risks associated with this emission. In the case of acrolein, the results indicate that emissions exceed the threshold primarily under conditions of high VG ratios. Figure illustrates the distribution of acrolein emissions from all tested e-liquids, at both high and low concentrations and at power settings of 90 and 50 W with a PG/VG ratio of 20:80. The data confirm that acrolein emissions frequently surpass the TLV-C in this e-liquid base configuration for some flavors, especially terpene and aromatic aldehyde (benzaldehyde) classes.

4.

Acrolein emissions (μg/puff) from unflavored and flavored e-liquids at a PG/VG ratio of 20:80 at 90 and 50 W power output experimental settings. Error bars represent standard deviations based on six measurements.

While acetaldehyde and acetone are also hazardous, and their emissions are the highest after formaldehyde, their TLVs of 45 mg/m³ and 594 mg/m³, respectively (corresponding to 22.5 μg/puff and 297 μg/puff), are significantly higher than the levels detected in our study (Tables S2–S7), making them less of an immediate concern. However, cumulative exposure to multiple carbonyl compounds, even at lower individual levels, could pose a combined risk that should not be overlooked. Benzaldehyde was predominantly detected in the benzaldehyde-flavored e-liquid at levels approximately a thousand times higher than those in other flavors, which is expected and indicates that most of this compound vaporizes rather than undergoing thermal decomposition.

Overall, this study underscores the complex interactions among e-liquid composition, flavor additives, and device settings in determining carbonyl emissions. While most flavors, by themselves, may not contribute significantly to carbonyl emissions under standard conditions, in line with previous results, certain classes, particularly terpenes, can play a substantial role under specific scenarios. These findings highlight the need for more stringent regulation and further research into the safety of flavored e-liquids, especially as vaping devices become more powerful and diverse. It should also be noted that this study examined individual flavor compounds; interactions between multiple flavors, as well as nicotine and acidifying agents, in real-world mixtures could alter carbonyl emissions and warrant further investigation.

5. Conclusions

Our investigation into the role of e-liquid flavors, base composition, and power settings in carbonyl emissions from e-cigarettes reveals the significant impact these factors have on the emission of harmful compounds, such as carbonyls. An addition of a flavor compound tends to increase carbonyl emissions relative to the unflavored e-liquid of the same PG/VG ratio and vaped at the same power setting. Terpene-based flavors, such as limonene and linalool, were identified as the strongest contributors to elevated carbonyl emissions, particularly when used with e-liquids with high VG content and at higher power outputs. Different flavors affect emissions of individual carbonyl compounds differently, with some compounds affecting all main carbonyls, while some affect some carbonyls with little effect on other compounds. The contribution of different compounds to carbonyl emissions was also found to also depend on the PG/VG ratio and power setting.

Our results suggest that careful selection of flavoring compounds, reducing flavor concentration and VG content, and using lower power settings could mitigate the health risks associated with carbonyl emissions, particularly for formaldehyde and acrolein, which in our experiments exceeded their short-term exposure limits under certain conditions. While emissions of acetaldehyde and acetone were below immediate concern levels, the potential for cumulative exposure to multiple carbonyl compounds warrants further investigation. These insights underscore the need for more stringent regulations and consumer guidelines to ensure the safer use of e-cigarettes, with particular attention given to flavor composition, device settings, and e-liquid formulation.

The Supporting Information is available free of charge at https://pubs.acs.org/doi/10.1021/acs.chemrestox.5c00404.

• Additional tables related to chromatographic parameters, carbonyl emissions, and regression model and multivariate test statistics and figures showing comparisons of formaldehyde emissions at different conditions, p-value matrixes, acetaldehyde emissions at different conditions, and acrolein emissions at different conditions (PDF)

CRediT: Elham Fazeli data curation, formal analysis, investigation, visualization, writing - original draft; Bianca Martinez data curation, formal analysis, investigation, writing - review & editing; Yeongkwon Son conceptualization, funding acquisition, methodology, supervision, validation, writing - review & editing; Andrey Khlystov conceptualization, funding acquisition, methodology, project administration, resources, supervision, validation, visualization, writing - review & editing.

The authors declare no competing financial interest.