Development of a Flavor Ingredient Wheel Linking E-Liquid Additives to the Labeled Flavor of Vaping Products

E-liquids contain combinations of chemicals, with many enhancing the sensory attractiveness of the product. Studies are needed to understand and characterize e-liquid ingredients, particularly flavorings, to inform future research and regulations of these products. We identified common flavor ingredients in a convenience sample of commercial e-liquids using gas chromatography–mass spectrometry. E-liquid flavors were categorized by flavor descriptors provided on the product packaging. A Flavor Ingredient Wheel was developed to link e-liquid flavor ingredients with flavor categories. An analysis of 109 samples identified 48 flavor ingredients. Consistency between the labeled flavor and ingredients used to produce such flavor was found. Our novel Flavor Ingredient Wheel organizes e-liquids by flavor and ingredients, enabling efficient analysis of the link between ingredients and their flavor profiles and allowing for quick assessment of an e-liquid ingredient’s flavor profile. Investigating ingredient profiles and identifying and classifying commonly used chemicals in e-liquids may assist with future studies and improve the ability to regulate these products.


Introduction
Many flavoring chemicals are added to e-liquids to increase attractiveness in taste and smell.Although oral ingestion of these compounds may be generally recognized as safe (GRAS) [1], inhalation safety is unknown because GRAS designation does not include an assessment of inhalation risks [2].Inhalation of complex mixtures, like flavored e-cigarette aerosols, may potentially cause adverse health effects, including respiratory tract inflammation through a variety of mechanisms [1,3,4].Observed respiratory effects associated with flavored e-liquids may be less than with tobacco smoke but may depend on the specific flavoring chemicals used [1,3,4].
Numerous chemicals have been reported in e-cigarette products [1].Many previous studies have quantified the number of chemicals, although they all suffered limitations on the sampling or analytical methods [5].Hutzler et al. [6] found 141 flavor chemicals in 28 e-liquids, and Girvalki et al. [7] found 171 chemicals in 122 samples; however, others have criticized these works, as Hutzler [6] included some samples with ethylene glycol as a solvent and Girvalki [7] only used a quantitative analysis [8].A review [9] found 9 studies investigating 670 flavored e-liquids and detected between 1 and 47 chemicals per sample, with the most common being ethyl maltol, ethyl vanillin, vanillin, cinnamaldehyde, and menthol.Most recently, Omaiye et.al. found 126 flavor chemicals in 103 bottles of refill liquid manufactured by one company [10].
Thus, it is important to have a better understanding of e-liquid chemicals.This study aimed to analyze e-liquid samples purchased within North Dakota, and to identify flavor chemicals used in the various retail flavored e-liquids to inform future research and potential regulations of these products.This work was an exploratory effort to determine as many ingredients as possible in a wide range of products without prior knowledge of these compounds.We then mapped the relationship between flavor descriptors and flavor chemicals to create a new classification system for ingredients, which we call the Flavor Ingredient Wheel.

Sampling of E-Liquids for Analysis
As part of a more extensive study, e-liquids (n = 285; free-base, salts, and nicotine-free) were purchased in 2019 from 35 licensed and unlicensed tobacco specialty stores (vape shops, 45.7%; head shops [shops selling paraphernalia related to recreational drug use, music, countercultural art, etc.] [11], 28.6%; tobacco shops, 17.1%; and other shops, 8.6%) across North Dakota, USA.Shop sampling is described elsewhere [12].Excluded were retail shops located on American Indian reservations, convenience stores, gas stations, grocery stores, and other similar venues.
Data collectors purchased common strengths of free-base e-liquids (0, 6, 12, 18, 24, and 36 mg/mL).Less common concentrations (3 or 9 mg/mL), uniquely packaged eliquids, and e-liquids compounded in-shop were purchased when possible.Data collectors purchased different brands for each concentration, with six e-liquids per shop minimum.Data collectors purchased nicotine salts in three varying concentrations.Samples were stored in the original packaging in a resealable bag in a dark space at room temperature.
Due to financial constraints, we needed to limit the number of samples for chemical analysis.Of the 285 total e-liquids purchased, 75 e-liquids were randomly sampled.Additionally, some of the 285 samples that were compounded in-shop (n = 34) or uniquely packaged (n = 14) that were not already in the random sample were also included (n = 24 and n = 10, respectively).Uniquely packaged samples were defined as those not in containers resembling small bottles, with outer packaging resembling everyday items, or were not standard rectangular packages, resulting in 109 samples being analyzed.

Overall Sample Analysis Process
Figure 1 provides a schematic for the e-liquid and flavor analysis and the development of our Flavor Ingredient Wheel.
menthol.Most recently, Omaiye et.al. found 126 flavor chemicals in 103 bottles of refill liquid manufactured by one company [10].
Thus, it is important to have a better understanding of e-liquid chemicals.This study aimed to analyze e-liquid samples purchased within North Dakota, and to identify flavor chemicals used in the various retail flavored e-liquids to inform future research and potential regulations of these products.This work was an exploratory effort to determine as many ingredients as possible in a wide range of products without prior knowledge of these compounds.We then mapped the relationship between flavor descriptors and flavor chemicals to create a new classification system for ingredients, which we call the Flavor Ingredient Wheel.

Sampling of E-Liquids for Analysis
As part of a more extensive study, e-liquids (n = 285; free-base, salts, and nicotinefree) were purchased in 2019 from 35 licensed and unlicensed tobacco specialty stores (vape shops, 45.7%; head shops [shops selling paraphernalia related to recreational drug use, music, countercultural art, etc.] [11], 28.6%; tobacco shops, 17.1%; and other shops, 8.6%) across North Dakota, USA.Shop sampling is described elsewhere [12].Excluded were retail shops located on American Indian reservations, convenience stores, gas stations, grocery stores, and other similar venues.
Data collectors purchased common strengths of free-base e-liquids (0, 6, 12, 18, 24, and 36 mg/mL).Less common concentrations (3 or 9 mg/mL), uniquely packaged e-liquids, and e-liquids compounded in-shop were purchased when possible.Data collectors purchased different brands for each concentration, with six e-liquids per shop minimum.Data collectors purchased nicotine salts in three varying concentrations.Samples were stored in the original packaging in a resealable bag in a dark space at room temperature.
Due to financial constraints, we needed to limit the number of samples for chemical analysis.Of the 285 total e-liquids purchased, 75 e-liquids were randomly sampled.Additionally, some of the 285 samples that were compounded in-shop (n = 34) or uniquely packaged (n = 14) that were not already in the random sample were also included (n = 24 and n = 10, respectively).Uniquely packaged samples were defined as those not in containers resembling small bottles, with outer packaging resembling everyday items, or were not standard rectangular packages, resulting in 109 samples being analyzed.

Overall Sample Analysis Process
Figure 1 provides a schematic for the e-liquid and flavor analysis and the development of our Flavor Ingredient Wheel.

Step 1: Identification of E-Liquid Flavor Ingredients
Flavoring chemicals and other organic substances were quantitatively measured within three months after purchase.The derivatization method, based on published methods [13,14], was used as follows: methanol (500 µL) was added to 100 µL of 1:100

Step 1: Identification of E-Liquid Flavor Ingredients
Flavoring chemicals and other organic substances were quantitatively measured within three months after purchase.The derivatization method, based on published methods [13,14], was used as follows: methanol (500 µL) was added to 100 µL of 1:100 methanol-diluted e-liquid samples, and the samples were vortexed vigorously for 20 s.N-methyl-N-tert-butyldimethylsilyl-trifluoroacetamide (MTBSTFA; 500 µL) was added to the samples to help ionization of the different molecules, which were vortex mixed for 15 s.Samples were then heated at 80 • C for 3 h.Samples were injected on the Agilent 7890 Gas Chromatography-Mass Spectrometry (GC-MS) system using the 30-m, 0.25-µm film thickness Agilent HP-5ms column (Agilent, product #19091S-433, Santa Clara, CA, USA).Two µL of the material were injected using a 10:1 split method, with instrument conditions as follows: inlet held at 325 • C, with a gas flow rate of 1.5 mL/min.The oven temperature profile was 35 • C for 0 min, then a ramp of 20 • C/min to 320 • C, with a hold for 1 min.The transfer line to the MS was held at 325 • C. The MS scan method was from 35 to 450 m/z.Samples were analyzed using the Agilent MassHunter Quantitative Analysis software tool (Version 4).Data collection from MS started after three minutes post-injection as the peak of the propylene glycol with short retention time interfered with the collection of the other smaller peaks.Peaks were identified using the 2008 National Institute of Standards and Technology (NIST) [15] standard library spectrum matching.The following was used for matching criteria: molecules had to match at a probability score of 50% using the NIST 08 software, which utilized major ion fragment patterns for comparison [16,17].Due to the number of mass features present in the samples compared to the reference spectra, this wider tolerance was utilized.Major and unique ions were utilized from butterfly plots to assess match quality.Data were manually curated for NIST matches and identifications that were less than 50% were excluded from the analysis.Ingredients were identified using NIST information and orthogonal data, including retention time, ion fragmentation patterns, and exact mass matching, and were quantified using peak area integration.To determine the limit of detection (LOD) and limit of quantitation (LOQ), US Food and Drug Administration (FDA) and Environmental Protection Agency guidelines and recommendations were used [18,19].Because this was an untargeted approach, no standards with calibration curves were injected, and the samples were curated manually for LOD.The peaks with the lowest abundance but still 10× the baseline (manual assessment, chromatograms available upon request) for a given analyte were used as the basis for the 3× LOD calculation to reach LOQ.These 3× LOD values are listed in Supplemental Table S1.
To expand the approach to this analysis, global trends were also evaluated.To achieve this, the relative abundance for each flavoring chemical was compared between samples.This approach assessed trends in the samples by characterizing the ingredients that were present but not seen at our determined LOQ.We wanted to capture all ingredients in each sample, even low concentrations because we acknowledge that the analysis techniques utilized here likely needed to include low-abundant ingredients.This work was intended as an exploratory effort to determine as many ingredients as possible without prior knowledge of these compounds.The MTBSTFA derivatization method selects for active hydrogen molecules and thus yields results for the more polar molecules in the samples.This precluded utilizing standards for absolute quantitation and exact calculations for LOD or LOQ, undoubtedly resulting in ingredients not captured in this analysis.
Chemical ingredients were first identified using the NIST database [15]; the naming schema of NIST was used for original annotation.Chemical names, chemical functional groups, and properties of each ingredient were annotated using databases such as Pub-Chem [20], the Human Metabolome Database [21], the Good Scents Company Information System (Good Scents) [22], and the Flavor Extract Manufacturers Association Flavor Ingredient Library (FEMA) [23].Functional group classifications were based on super class information from these databases, mostly HMDB.For example, Vanillin acetate's super class chemical taxonomy is listed as benzenoid under the kingdom of organic compounds, so it is listed as a benzenoid and a cyclic molecule in our table [24].Ingredients were categorized and defined using the International Union of Pure and Applied Chemistry (IUPAC) [25] name, common name, Chemical Abstracts Services (CAS) [26] number, FEMA [23] number, HMDB number [21] and flavor profile details as described from these sources.IUPAC standardizes nomenclature and terminology for presenting scientific results [25].

2.2.2.
Step 2: Categorization of E-Liquid Flavors According to Flavor Descriptors Provided on Product Packaging E-liquids were categorized as purchased from vape shops, head shops, tobacco shops, and "other" shops.E-liquids were categorized as free-base nicotine salts or nicotine-free based on labels or information provided during purchase.E-liquids were also categorized as compounded or not, either made entirely on-site or when the shop staff added nicotine [12].
The labeled flavor categories were used to report class frequencies, percentages, and groups for the statistical analysis regarding the number of ingredients and relative abundance of identified chemicals and ingredients.Labeled flavor categories included alcohol, candy, coffee/tea, dessert, fruit, menthol/mint, and nuts.Additional categories included other beverages, other flavors, and other sweets.Spices, tobacco, and unflavored were assessed; no samples had these categories.Labels without specific ingredients listed were classified as "unknown flavors".E-liquid sample labels were also evaluated for references about "cooling" or menthol in their name and/or description or no cooling claim.Cooling claim terms included menthol, mint, ice, freeze, blast, mist, cooler, frost, frozen drink (e.g., daiquiri), mojito, popsicle, slushie (drink), and blue razz.

Step 3: Development of the Flavor Ingredient Wheel
We developed a novel Flavor Ingredient Wheel to organize the reporting of chemical ingredients and the flavor categories of the products.The Flavor Ingredient Wheel provides a visual interpretation of the chemical analysis (Step 1) and flavor product descriptors (Step 2).The outer layer of the wheel includes common flavor ingredients identified in analyzed products (see Step 1 above).The middle layer provides ingredient-based flavor descriptors provided in FEMA [23] and Good Scents [22] databases.The inner (central) layer of the wheel includes six primary flavor categories, as identified in Step 2. The primary flavor categories included in the inner layer of newly developed the Flavor Ingredient Wheel were compared with the previously published Flavor Wheel [27].

Step 4: Characterizing E-Liquids Using the Proposed Flavor Ingredient Wheel
We conducted two functional tests of the Flavor Ingredient Wheel.First, we confirmed the label flavor description with the chemical ingredient properties using a random sample of five collected e-liquids.Second, we chose two e-liquid samples without an identifiable flavor profile based on their labels and estimated a flavor profile.
For a confirmation test, we assessed whether label flavor descriptions were reflected by the flavor ingredients in five samples.The chemical composition of each sample was evaluated, and a potential e-liquid flavor was identified using our Flavor Ingredient Wheel.Each flavor ingredient was placed on our Flavor Ingredient Wheel's outer layer, which relates to the additional flavor descriptors found in the second layer of the Flavor Ingredient Wheel and were compared with the e-liquid product labels.From there, a primary categorization of an overall flavor is noted by referencing the inner layer of the Flavor Ingredient Wheel.
We tested whether flavor categories of two unknown flavor samples could potentially be identified based on detected flavor ingredients.For this test, we chose two products with unknown flavors, with names such as "Original Bold", that did not have a discernible flavor profile on the label.

Statistical Analysis
To evaluate how well the 109 e-liquid samples used in this study represent the larger pool of 285 e-liquid samples from which they were drawn, chi-square tests of association and Fisher's exact tests were implemented to compare the distributions of each sample categorization between the e-liquid samples used in this study and the remaining 176 e-liquid samples.An initial analysis was performed on the number of chemical ingredients detected.Then, using the relative abundances (of any amount) for each ingredient, further comparative statistical analysis was performed to parse the more subtle trends and changes in the data due to the categorical factors of interest.For the study of the number of ingredients, one-and two-way analysis of variance (ANOVA) was performed to determine significant differences in group mean among groups in various sample categorizations.The global analysis of abundance data was performed using principal component analysis (PCA), heatmaps with hierarchical clustering, partial least squares discriminant analysis (PLS-DA), and one-way ANOVA.Group comparisons of the sample categorizations previously mentioned were performed in the global study.Statistical significance was set at p < 0.05.Statistical analysis was performed in R Version 3.6.2(using the base and MetaboAnalyst [28] packages) and Excel 2016.

Sample Characterization
Figure 2 (also see Table A1) provides the complete sample characteristics of the 109 e-liquid samples analyzed for flavor ingredients in this study and the pool of all 285 e-liquid samples from which the study samples were randomly selected.Except for the compounded categorization, no significant differences were found in the distributions of each sample categorization between the e-liquid samples used in this study and the remaining 176 e-liquid samples from the larger pool of samples.The distributions of the two groups differed significantly for the compounded categorization (p < 0.001) due to all compounded samples from the original collection being selected for ingredient analysis as described above.
The majority of e-liquids were categorized as purchased from vape shops (50.5%), followed by head shops (22.0%), tobacco shops (19.3%), and "other" shops (8.3%).The majority of e-liquids were categorized as free-base (50.5%), followed by nicotine salts (35.8%) and nicotine-free (13.8%) based upon labels or information provided during purchase.About one-quarter (25.7%) of e-liquids were categorized as compounded.Approximately one-fifth (20.2%) of e-liquid sample labels referenced "cooling" or menthol in their name and/or description.

Step 1: Common E-Liquid Flavor Chemicals Identified in Analyzed E-Liquids
After analysis using GC-MS and NIST [15] compound matching, 48 ingredients were identified from the 109 samples.Table 1 lists their IUPAC chemical names [25], common names, HMDB numbers [21], chemical super classes (as annotated by databases), chemical functional groups (as annotated by databases), flavor profiles (as annotated by databases), CAS numbers [26], FEMA numbers [23], and substance descriptions (as annotated by databases).Two ingredients were categorized as derivatives of a parent compound because they had the same FEMA number [23], although they had different CAS numbers [26].These ingredients are denoted on the table and were not annotated in detail as the flavor profile was not unique to the other identified ingredients.A supplementary Excel document (Revised Table 1 with Full Reference Hyperlinks) also denotes this information and includes hyperlinks to each individual ingredient reference.

Step 1: Common E-Liquid Flavor Chemicals Identified in Analyzed E-Liquids
After analysis using GC-MS and NIST [15] compound matching, 48 ingredients were identified from the 109 samples.Of all samples tested, the mean number of flavor ingredients was 11.7 (standard deviation = 4.6).The mean number of ingredients in salt-based e-liquids (15.2 ± 3.4) was significantly greater than either free-base (9.8 ± 4.1; p < 0.001) or nicotine-free samples (9.5 ± 3.4; p < 0.001) but did not differ significantly among the groups for any other categorical factors (Figure 3 and Table A2).Of all samples tested, the mean number of flavor ingredients was 11.7 (standard deviation = 4.6).The mean number of ingredients in salt-based e-liquids (15.2 ± 3.4) was significantly greater than either free-base (9.8 ± 4.1; p < 0.001) or nicotine-free samples (9.5 ± 3.4; p < 0.001) but did not differ significantly among the groups for any other categorical factors (Figure 3 and Table A2).To ascertain the average chemical makeup of each flavor group based on GC-MS quantitative analysis, the mean number of flavor ingredients, categorized by the identified functional group, was summed, and the distribution was plotted (Supplemental Figure S1) relative to the number of samples in the flavor group, allowing visualization of the most common chemicals per flavor.The fruit-based flavor and unknown-flavor profiles had similar ratios of chemical groups.The candy and alcohol/coffee/tea flavor groups had a similar ratio of aldehydes to terpenoid ingredients.The menthol/mint/herbal and other beverage groups had the most terpenoid ingredients relative to other ingredient classes.The single nut flavor sample, combined with different flavors because there was just one sample in the nut category, was the only sample without an acid-classified ingredient.
The ANOVA showed 15 ingredients with significantly different mean relative abundances among the flavor groups (Supplemental Figure S2 and Supplemental Table S2).To ascertain the average chemical makeup of each flavor group based on GC-MS quantitative analysis, the mean number of flavor ingredients, categorized by the identified functional group, was summed, and the distribution was plotted (Supplemental Figure S1) relative to the number of samples in the flavor group, allowing visualization of the most common chemicals per flavor.The fruit-based flavor and unknown-flavor profiles had similar ratios of chemical groups.The candy and alcohol/coffee/tea flavor groups had a similar ratio of aldehydes to terpenoid ingredients.The menthol/mint/herbal and other beverage groups had the most terpenoid ingredients relative to other ingredient classes.The single nut flavor sample, combined with different flavors because there was just one sample in the nut category, was the only sample without an acid-classified ingredient.
The ANOVA showed 15 ingredients with significantly different mean relative abundances among the flavor groups (Supplemental Figure S2 and Supplemental Table S2).The PCA and PLS-DA did not show clear differentiation among the flavor groups, but the heatmap of all the samples and ingredients (Supplemental Figure S3) showed some trends when using only the top 15 ingredients that ANOVA differentiated.When considering only these ingredients, a higher abundance of vanilla (benzenoid and aldehyde compounds) and nut-flavored (alcohol and aldehydes) ingredients in the dessert and coffee/tea category were observed, and herbal and terpenoid ingredients were more abundant in the other beverage and menthol/mint/herbal categories.
The label claim of a cooling sensation revealed patterns among the samples.For the global abundance comparison, ANOVA showed nine ingredients, with significant differences between samples that claimed to cool and samples that did not.The ingredient with the most significant difference in mean relative abundance was cyclohexanol, 1methyl-4-(1-methylethyl), or menthol, which is consistent with the claims because the cooling sensation comes from menthol.Following menthol, hexanoic acid, butyl ester (glycol, with a fruit flavor profile), and vanillin acetate were the following most significantly differentiated ingredients between the cooling sample types (Supplemental Table S3), with mean relative abundance lower in e-liquids with a cooling claim.The PLS-DA showed separation in component distributions between the cooling and non-cooling samples, although there was some overlap (Supplemental Figure S4).As with the ANOVA, the VIP scores showed that menthol was the ingredient that most discriminated the groups, followed by vanillin, acetate (benzenoid), and 4H-pyran-4-one, 2-ethyl-3-hydroxy (ethyl maltol) (aldehyde).In summary, samples that had cooling ingredients had fewer fruit or floral flavor ingredients and more additives known to provide a cooling sensation, including cyclohexanol, 1-methyl-4-(1-methylethyl) (menthol), and less of other dessert flavors, such as vanillin acetate and the fruity flavor 4H-pyran-4-one, 2-ethyl-3-hydroxy (ethyl maltol).
Supplemental Figure S5 shows the percentage of samples in which other ingredients (non-flavor related) were detected.Nicotine was the most common, seen in 86.2% of the samples, matching the number of samples labeled as containing nicotine.Supplemental Table S4 provides the number of samples with acid type (used to create nicotine salts) and related proportions.Other common ingredients were acetic acid pentyl ester (59.6%), ethyl maltol (58.7%), and benzyl alcohol (56.9%).It should be noted that these data did not characterize propylene glycol, a common solvent used in e-liquids.This ingredient was excluded from the analysis because it overloaded the GC column.

Step 2: Categorization of E-Liquid Flavors Based on Flavor Descriptors Provided on Packaging
Seven flavor categories were identified and annotated from database entries: alcohol/coffee/tea, candy, dessert, fruit, menthol/mint/herbal, other beverages, and other flavors/nuts.We also included an "unknown" flavor category if an ingredient had no flavor reference."Herbal" was added to menthol/mint based on FEMA [23] and Good Scents [22] information.Some groupings were collapsed.Alcohol was combined with coffee/tea."Other flavors" and nuts were combined for statistical analyses due to a small sample size.

Step 3: Flavor Ingredient Wheel
To illustrate the results, we developed a Flavor Ingredient Wheel (Figure 4), as described in Section 2.2.Our Flavor Ingredient Wheel has three levels.The outer level identified chemicals by common name, or by chemical name if there was no identified common name (see Section 2.2.1).The middle level contains the flavor descriptor as assigned in Sections 2.2.1 and 2.2.2.The inner level categorizes the flavor ingredients similar to Krüsemann's flavor categories and includes six primary categories: fruit, candy, dessert, floral, herbal/mint/menthol, and other additives (see Section 2.2.3).The Flavor Ingredient Wheel excludes the following solvents: propylene glycol and glycerin.Tables 1 and A3 provide the Flavor Ingredient Wheel detailed information.
Section 2.2.1 and Section 2.2.2.The inner level categorizes the flavor ingredients similar to Krüsemann's flavor categories and includes six primary categories: fruit, candy, dessert, floral, herbal/mint/menthol, and other additives (see Section 2.2.3).The Flavor Ingredient Wheel excludes the following solvents: propylene glycol and glycerin.Tables 1 and A3 provide the Flavor Ingredient Wheel detailed information.
We compared our Flavor Ingredient Wheel with the Flavor Wheel [27] to assist with the organization of the Flavor Ingredient Wheel's inner level, and we designed it to be somewhat parallel to the Flavor Wheel [27].See Supplemental Table S5, Comparison of Primary Flavor Categories.Note.The e-liquid Flavor Ingredient Wheel categorizes and provides a link between chemistry of the products (flavor ingredients identified in products) and product labelling (flavor category provided on packaging).The outer level includes the common or chemical name of the identified e-liquid ingredients.The middle level identifies the flavor descriptor of each ingredient as described by FEMA Library [23] and Good Scents System database [22].The inner level categorizes the flavor ingredients similar to Krüsemann's Flavor Wheel [27].We compared our Flavor Ingredient Wheel with the Flavor Wheel [27] to assist with the organization of the Flavor Ingredient Wheel's inner level, and we designed it to be somewhat parallel to the Flavor Wheel [27].See Supplemental Table S5, Comparison of Primary Flavor Categories.

Step 4: Characterizing E-Liquids Using the Flavor Ingredient Wheel
As shown in Supplemental Table S6, Characterizing Product Flavors Using the Flavor Ingredient Wheel, a dessert sample with a cooling claim contained menthol and other ingredients consistent with a dessert profile, including vanilla flavors and a fruit, herbal, and nutty profile.Without a cooling claim, a second dessert sample had floral, fruit, and nutty ingredients.A fruit-berry sample contained fruit ingredients, and two other fruitother samples contained primarily fruit ingredients.We tested two samples whose labels provided no flavor profile indication."Original Bold" had positive detections for acetic acid, pentyl ester (fruit), and ethyl maltol (fruit and cooked sugar), indicating a likely fruity flavor profile.Another unknown flavor, "Zerbert", could be categorized as a combination of fruit and dessert flavor by evaluating the ingredients of ethyl maltol, ethyl vanillin, and three fruit-flavored ingredients.Using the flavor categories, determined from each ingredient on our Flavor Ingredient Wheel, we could ascertain a rough characterization of these two unknown-flavor samples.

Additives Not Associated with Flavors
In addition to the flavor ingredients, we identified major solvents, glyceryl acetate, triacetin, nicotine and organic acids used to create the nicotine salts.The details about acid ingredients are located in the supplementary materials (Supplemental Table S4).Carrier solvents like propylene glycol were excluded from this analysis, as the chromatographic peak was large and interfered with quantitation of other molecules.

Discussion
The unique Flavor Ingredient Wheel, described for the first time in this paper, helps visualize chemicals in e-liquids and their flavor profiles according to packaging claims.Our Flavor Ingredient Wheel expands on and highlights common flavor ingredients associated with label claims as originally outlined in Krüsemann's work [27].Our innovative Flavor Ingredient Wheel is database-sourced and can be applied to methods for categorizing other e-liquids.Investigating ingredient profiles and the frequency of flavor ingredients used in various products has already been demonstrated as a valuable tool when classifying samples and tracking trends in ENDS products.The Flavor Ingredient Wheel allows for a quick assessment of an ingredient's flavor profile, common name and relationship to package claims.FEMA only provides information about a single compound [23]; thus, it is time-consuming to look up multiple ingredients.The Flavor Ingredient Wheel organizes this information to reference ingredient information quickly but should not replace the detailed information in other databases.We recognize the complexity of identifying flavoring ingredients.Although some ingredients alone may contribute to a single flavor (ex: cherry, menthol), many other flavors are derived from a complex combination of ingredients.A major limitation of the Flavor Ingredient Wheel is that it cannot completely predict a product's flavor.We strongly recommend that further research be conducted to further explore the complexity of flavoring and revise the Flavor Ingredient Wheel.As the first version of the Flavor Ingredient Wheel, we also recommend that it be updated frequently as ENDS products continue to evolve and as policy impacts the regulation of ingredients and flavors.We encourage other authors to update this figure with their findings.
Overall, we detected 48 different ingredients in 109 e-liquid samples.Two-thirds of our samples had ten or more ingredients.Other studies have shown a range of 1 to 171 ingredients [6,7,10].In fruit and floral-flavored e-liquids the most identified ingredients were esters and benzenoids, while most dessert-flavored products contained fatty acid derivatives.The most identified ingredients in menthol/mint/herbal were monoterpenes and monoterpenoids.
Vanilla, sweet, and floral ingredients were abundant in the nicotine-free samples.Vanilla and fruit-flavored ingredients were less abundant in e-liquids with a cooling claim.Lactic acid and lactide were more abundant in nicotine salts (Supplemental Table S4).
We used multiple factors to categorize label claims and associated flavor ingredients.Flavor categories were annotated from the detected ingredients based on FEMA [23] and Good Scents [22] descriptions, allowing for a comprehensive evaluation of the eliquids.Other work, like Scieszka et al. [29], utilized a similar approach to categorize ENDS product small molecules detected in lung tissue using the HMDB and Hsiao et al., 2023 [30] utilized the NIST library to identify unknown molecules and further annotate molecules using the HMDB, a public database that aids other researchers in classifying small molecules [31].Recent work on user preference based on social media references shows that fruit, sweet, and floral ingredients are users' most popular flavor profiles [32].Other evidence indicates that cooling and sweet e-liquids are popular among users, regardless of age [33].Although our data cannot be extrapolated to all e-liquids, the flavor profile with the most representation was fruit, aligning with these documented e-liquid user preferences [32].Importantly, we found consistency between labeled flavors and actual ingredients in e-liquids.
Samples that claimed a cooling effect contained menthol, which produces a cooling sensation for the user [33,34].Mint-flavored tobacco products are preferred where other flavors are difficult to obtain or are prohibited [35].Calls have been made to remove menthol for over two decades, and a recent proposed rule makes this a reality [36].Although menthol has not been regulated e-cigarettes yet, menthol is banned in tobacco cigarettes in Canada, the United Kingdom, the European Union, and several other jurisdictions [37,38].Vanillin acetate and ethyl vanillin were abundant and contributed to the differentiation of samples based on the label's cooling and flavor claims.Thus, one might hypothesize that the presence of menthol and lack of vanillin suggests that the e-liquid has a "cooling" flavor.
Identifying commonly used flavoring chemicals in e-liquids may assist with future studies and evaluating their inhalation toxicity.All flavor ingredients detected in analyzed products were included on the FDA Generally Recognized as Safe (GRAS) list.It is essential to recognize that the GRAS designation is based on the chemical's safety profile when ingested.For many ingredients identified in our study, inhalation safety is largely unknown.The FDA has also established a list of harmful and potentially harmful constituents (HPHC) in tobacco products [39,40].We used the HPHC list to cross-check against our list of identified e-liquid ingredients.The only chemical ingredient our study identified that matched the HPHC list was nicotine.However, we did find several ingredients which are derivatives of chemicals listed on the HPHC list.Those include derivatives of coumarin, benzene, and acetamide.We did not find diacetyl and cinnamaldehyde, two flavoring ingredients previously identified in e-liquids with potential respiratory concerns [10], in any of the products analyzed in our study.The Dutch National Institute for Public Health and the Environment also published an advisory list of substances that should not be added to tobacco products and e-cigarettes [41].Among the 48 ingredients in our study, there were some chemical classes that overlapped with the advisory list including coumarins, nicotine molecules and sugars.
A study limitation was that all samples were purchased in one state and thus may not include e-liquids available nationwide or globally.The samples were purchased and analyzed in 2019 and therefore may not be representative of those currently on the market.However, it is very likely that the same chemicals are currently being used, as many of the identified ingredients are common chemicals in e-liquid products.Additionally, we used purposeful sampling rather than random sampling.Because this was part of a larger study focused on e-liquids, it did not include pod-style or other ENDS products.All products were analyzed within three months of purchase and unknown/variable manufacturing time.The e-liquid composition may have changed from the manufacturing time (e.g., due to extreme storage conditions), which we could not evaluate in our study.Other significant limitations of our study are that sampling for chemical analysis was not based on label flavors and our analysis of the random sample did not include tobacco flavored products.
Furthermore, because of budgetary constraints, we conducted chemical research using the MTBSTFA derivatization method, which optimized for polar compounds with reactive hydrogen moieties and did not explore other approaches to evaluate unknown chemicals, degradation, and aerosolized products.This approach, while untargeted, does offer bias towards molecules with reactive hydrogens and is a limitation of our analysis.Only chemicals noted in the NIST database were utilized; new compounds not included in the NIST database may have been missed.For instance, synthetic cooling agents like WS-3, WS-23, and Evercool 180 and 190 have been recently reported in e-liquids as substitutes for menthol.Still, the NIST database needs to include their mass spectra data [42][43][44].In addition, the NIST database does not list hemiacetals that are byproducts of condensation reactions between flavoring chemicals and solvents in e-liquids [45][46][47].Notably, the LOQ of the method used in our study can vary significantly based on batch run day effects.Here, we utilized an estimation of LOQ, which was manually determined, not based on a neat standard calibration curve, to ensure that molecules in very low abundance would not be classified as present and calibration curve data were not collected, considering this was an exploratory analysis.Despite these limitations, this study provides a necessary step in the future characterization and regulation of ENDS products.Determining these samples' chemical flavor ingredients, flavor profiles, and trends will help inform future policy on efficient methods for analyzing these products.One avenue for future research might include analyzing the proportion of ingredient integrated areas from chromatograms compared to overall total integrated area of each sample.This is one approach that can speak to the proportion of ingredients within the products [48,49], but was not applicable here, as we excluded the carrier solvent from our analysis and thus a large peak that should be included in such an approach.
The development of straightforward tools such as our Flavor Ingredient Wheel demonstrates that tracking use and trends in ENDS products can be simplified on some levels.Future research recommendations include replication/expansion of this study with more samples, using a random-sampling protocol, include information about the e-liquid brands and their market share, and include additional tobacco products and emerging products to confirm or revise the Flavor Ingredient Wheel.

Conclusions
In conclusion, this study shows the diversity of flavor ingredients and labeled flavor profiles in e-liquids and suggests a method of organizing them.Trends prevail despite the ambiguity and multiplicity of flavors when analyzing these samples by flavor categories or cooling status and the variety of ingredient profiles.Recent FDA regulations [50] are starting to prohibit the use of fruit-and dessert-flavored e-cigarette liquids.However, the effect of e-liquids on health has yet to be fully understood, and studies such as this one offer tools to better understand the implications of using these products.

Supplementary Materials:
The following supporting information can be downloaded at: https://www.mdpi.com/article/10.3390/toxics12050372/s1,Supplemental Table S1 displays integrated peak area abundances for LOQ estimation.Supplemental Figure S1 provides the average chemical classification breakdown of ingredients by labeled flavor category.Supplemental Figure S2 illustrates ingredients by flavor classification.Supplemental Table S2 lists ingredients with significant mean abundance differences noted among flavor categories.Supplemental Figure S3 is a heatmap showing the relative abundances of ingredients detected in e-liquid samples.Supplemental Table S3 lists ingredients significantly different between groups based on cooling claim.Supplemental Figure S4 provides PLS-DA and VIP scores in the projection of ingredients based on cooling claims on packaging.Supplemental Figure S5 illustrates the distribution of detected e-liquid ingredients.Supplemental   Note.Max = maximum; min = minimum; SD = standard deviation; a Mean number of ingredients significantly higher for salt samples compared with free-base and nicotine-free samples (p < 0.05).
Table A3.Chemical functional group ingredients and flavor profile.

Figure 1 .
Figure 1.Overall schematic of the study.

Figure 1 .
Figure 1.Overall schematic of the study.

Figure 2 .
Figure 2. Characterization of e-liquids products analyzed for flavor ingredients.

Figure 2 .
Figure 2. Characterization of e-liquids products analyzed for flavor ingredients.

Figure 3 .
Figure 3.The mean number of ingredients with standard deviation error bars by e-liquid categories.

Figure 3 .
Figure 3.The mean number of ingredients with standard deviation error bars by e-liquid categories.

Figure 4 .
Figure 4. E-Liquid Flavor Ingredient Wheel (Version 1).Note.The e-liquid Flavor Ingredient Wheel categorizes and provides a link between chemistry of the products (flavor ingredients identified in products) and product labelling (flavor category provided on packaging).The outer level includes the common or chemical name of the identified e-liquid ingredients.The middle level identifies the flavor descriptor of each ingredient as described by FEMA Library [23] and Good

Figure 4 .
Figure 4. E-Liquid Flavor Ingredient Wheel (Version 1).Note.The e-liquid Flavor Ingredient Wheel categorizes and provides a link between chemistry of the products (flavor ingredients identified in products) and product labelling (flavor category provided on packaging).The outer level includes the common or chemical name of the identified e-liquid ingredients.The middle level identifies the flavor descriptor of each ingredient as described by FEMA Library [23] and Good Scents System database[22].The inner level categorizes the flavor ingredients similar to Krüsemann's Flavor Wheel[27].

Table 1 .
Identification of E-Liquid Chemicals.
Note. # = number.CAS,Chemical Abstracts Services; FEMA, Flavor Extract Manufacturers Association Flavor Ingredient; IUPAC, International Union of Pure and Applied Chemistry; *** = the first asterisk indicates the "parent" ingredient: the next ones are derivatives of that one; NA = no information provided by a database source.Superscript numbers indicate the Chemical Class annotation source: HDMB = 1 ; IFA = 2 ; PubChem = 3 ; LMBD = 4 ; EBI = 5 ; NP-MRD = 6 ; YMDB = 7 ; ContaminantDB = 8 .Full citation links for each ingredient and chemical class can be found in the Supplemental Excel Document S1 titled Revised Table1with Full Reference Hyperlinks.

Table S4
lists acids identified in e-liquid samples.Supplemental TableS5provides an identification of e-liquid chemicals.Supplemental TableS6characterizes product flavors using the Ingredient Wheel.Supplemental Excel Document S1 (Revised Table1with Full Reference Hyperlinks) denotes references to all of the databases used to create Table1and includes hyperlinks to each individual ingredient reference.

Table A2 .
Summary statistics for number of ingredients by e-liquid categories.