Characterization of Korean Distilled Liquor, Soju, Using Chemical, HS-SPME-GC-MS, and Sensory Descriptive Analysis

The volatile compounds and sensory profiles of 18 different types of distilled soju, chosen with regard to various raw materials and distillation methods (atmospheric vs. vacuum), were explored using headspace solid-phase microextraction (HS-SPME) with gas chromatography-mass spectrometry (GC-MS) and descriptive analysis. General chemical properties such as pH, total acidity (TA), total soluble solids (°Brix), and lactic acid concentration were also determined. A total of 56 volatile compounds, comprising 31 esters, 11 alcohols, 1 acid, 4 aldehydes, 3 ketones, and 6 miscellaneous compounds, were identified. From the principal component analysis (PCA) of the volatile data, samples made using atmospheric distillation such as MSO and PJU showed a clear difference from decompressed distillation samples. Based on the PCA of the sensory data, there was also a clear distinction between samples by their distillation method. To explore relationships among chemical, volatile, and sensory data sets, multiple factor analysis (MFA) was applied. Yeasty and earthy flavors showed a close relationship with 1-nonanol, octatonic acid, and longer-chain esters such as ethyl phenylacetate and ethyl tetradecanoate, and with chemical parameters such as TA, °Brix, and lactic acid.


Introduction
Soju is one of the most popular alcoholic beverages in Korea. Soju is a distilled liquor manufactured using a pot still or continuous distillation after saccharification of grains or starchy raw materials [1]. Traditionally, distilled soju is made through a pot still at atmospheric pressure, but currently, modern vacuum distillation is widely applied and gaining more popularity in the industry. The vacuum distillation method, which is a method of distilling by lowering the pressure without applying direct heat by using a stainless-steel concentrator, has the advantage of maintaining high volatile aroma components with less heated and burnt flavors of the distilled liquor [2].
The headspace solid-phase microextraction (HS-SPME) methodology, which has the advantage of shortening the extraction time and increasing the sensitivity by using a variety of fibers, is generally applied to volatile analysis in alcoholic beverages as well as general foods [3]. HS-SPME is widely used to detect volatile compounds in whiskey [4], brandy [5], Chinese Baijiu [6], and other alcoholic beverages. The distribution and chemical composition of aroma compounds result from the fermentation raw material, fermentation and distillation conditions, aging containers, and durations of distilled liquor [7].
Traditional soju manufactured by atmospheric distillation leaks a lot of middle-highboiling components compared to vacuum distilled liquor [8]. It was reported that the higher the distillation temperature, the more the esterification reaction occurs during the distillation, resulting in longer-chain ester compounds [9]. Soju distilled under atmospheric pressure has heated, burnt, and grainy flavors with a strong bitter taste. These flavors are

Materials and Chemicals
Various commercial distilled soju samples were screened using online markets, postal shopping services for traditional Korean liquors, or liquor wholesale stores. Initially, twentyeight samples from different manufacturers were purchased from the above sources. The samples were informally evaluated using blind taste tests by experienced drinkers, who also described the sensory characteristics and identified defective samples. Defective samples were eliminated from further consideration and among samples that were extremely similar. Finally, twenty samples were selected for descriptive analysis and subsequent instrumental tests. However, two samples were discontinued during the sensory tastings. Eventually, eighteen samples with regard to various raw materials, distillation methods, or aging methods were used in this study. Detailed information about the samples is presented in Table 1. 2-Methyl-1-pentanol (internal standard), n-alkane standards (C9-C25), and sodium chloride were purchased from Sigma-Aldrich Chemical Co. (St. Louis, MO, USA) for HS-SPME and GC-MS. Sodium sulphate (Junsei, Tokyo, Japan) and HPLC-grade solvents were used (J.T. Baker, NJ, USA). Other reagents used were purchased from Sigma-Aldrich (St. Louis, MO, USA). All chemical standards and reagents were analytical grade with at least 97% purity.

General Chemical Analysis
The pH of the samples was measured with a pH meter (520A; Orion Research Inc., Franklin, MA, USA). The titratable acidity (acetic acid in grams per liter) was measured by adding 10 g of sample to 50 mL of deionized water and titrating with 0.1 mol/L sodium hydroxide to an endpoint of pH 8.3. Total soluble solids (ºBx) were measured using an ATAGO hand refractometer (model N-1E, ATAGO, Tokyo, Japan). The lactic acid concentration was analyzed using high-performance liquid chromatography (HPLC) procedures adapted from Lee et al. [24]. The HPLC (Waters UV-2487, Miliford, MA, USA) was equipped with a Bio-Rad HPLC Organic Acid Analysis Aminex HPX-87H ion exclusion column (300 × 7.8 mm). Lactic acids were detected at 210 nm using a UV detector. A calibration curve was prepared, and the results were expressed as mg L-lactic acids/L. All chemical measurements were repeated three times, and the average values were reported.

Identification and Quantitation of Volatile Compounds
Volatile compounds were tentatively identified by comparison of the Kovats retention index (KI) [26] and the MS fragmentation patterns with those of reference compounds or with mass spectra in the Wiley 275 mass spectral database (Hewlett-Packard, Palo Alto, CA, USA). The KI of unknown compounds was determined via injection with a homologous series of alkanes (C9-C25). The GC/MS conditions were the same as described in Section 2.3.1. The relative concentrations of the volatiles were determined by the averaged integrated areas based on the TIC from the duplicate run, normalized to the area of the internal standard (2-methyl-1-pentanol), assuming a response factor of 1.

Sensory Descriptive Analysis
For the descriptive analysis of eighteen distilled soju samples, six females and four males (age 25-42 years) were recruited based on interest and availability, as well as their liking of liquors, from Sejong University, Seoul, Korea. Seven 1.5 h training sessions were conducted, and consensus was reached for thirteen aroma, ten flavor/taste, and five mouthfeel attributes, as shown in Table 2. Standards were presented to deliberate each sensory attribute during the training and formal sessions. Five samples per session were evaluated, in duplicate, and a total of 8 sessions were conducted. The presentation order of each sample was randomized for each session by a Williams Latin square design [27]. To minimize the differences in alcohol levels among the presented samples, the alcohol level of the samples was adjusted to 20% w/v with distilled water. Then, 50 mL aliquots of samples were presented in clear plastic cups marked with three-digit numbers and covered with Petri dishes. The intensity of each attribute was evaluated on a scale of 0 to 9, where 9 was the highest possible score (i.e., most intense). Water and white bread were used to cleanse the palates of the panelists between samples [28]. All sensory testing was conducted in sensory booths at room temperature.

Statistical Analysis
Analyses of variance (ANOVAs) were run on the sensory descriptive data using SAS ver.6.12 (SAS Institute, Cary, NC, USA) by employing a three-way mixed model (judges, samples, and replications), with all two-way interactions with judges treated as random. Individual product differences were identified by Fisher's least significant difference (LSD) test. The mean intensities of 22 significantly different sensory attributes were used to perform principal component analysis (PCA) using the covariance matrix with no rotation on XLSTAT ver. 2022.1 (Addinsoft, New York, NY, USA). PCA was also performed on the mean concentrations of 56 volatiles detected in more than 6 samples using the correlation matrix. These volatiles were significantly different among samples by two-way ANOVA (sample, duplicate injections). To evaluate any relationships among the chemical, volatile, and sensory data, multiple factor analysis (MFA) was conducted with XLSTAT using 4 chemical compositions, 22 sensory attributes, and 56 volatile compounds.

General Chemical Compositions of 18 Distilled Soju Samples
The general chemical compositions of the 18 distilled soju samples are presented in Table 3. While Korean traditional distilled soju samples have been reported to have an acidic pH level of 3.40-4.99 [29], the pH ranges of the 18 commercial samples in this study showed large differences from 3.62 to 7.42. The pH level of MSO using atmospheric distillation was the lowest at 3.62, while R25 using blends both from atmospheric and vacuum distillation, MIR, and JRO showed the highest levels of 7.42, 7.30, and 7.24, respectively. According to esterification studies of organic acids [1], JRO, MIR, and R25 seem to have the highest outflow of organic acids during distillation. The total acidity also showed a large difference of 0.02-0.31%. MSO and HBI had total outputs of 0.31% and 0.29%, which were much higher than those of other samples. HGO, JRO, MIR, and R25 had the lowest total output of 0.02%. Studies have shown that the acidity of alcoholic beverages greatly affects the flavor and preservation of products, and it is reported that the lower the acidity, the less perceived the pungent acidity of alcohols [2]. The total soluble solids had levels of 8.10-16.20 • Brix, with the lowest level of 8.10 found in DJA made under vacuum distillation and the highest level of 16.20 found in CTO, which was a 15-month oak-aged sample. Through organic acid analysis, only lactic acid was detected in the 18 samples. No other organic acids were detected. The content of lactic acid was the highest in MSO at 1610.70 µg/mL, which was also the highest in total acidity. The organic acid contents of liquors fermented with traditional nuruk using wild microflora were higher than those using selected enzymes [8]. Accordingly, MSO made with traditional nuruk showed a high acidity due to the higher production of organic acids.
A total of 31 esters were detected among the 18 samples. Esters have a big influence on the aroma characteristics of alcoholic beverages with strong fruit-related notes [30]. Ethyl octanoate described as a sweet and fruity aroma [31] and ethyl decanoate described as a floral and sweet odor [32] were the two most frequently detected compounds among the 18 samples, and in particular, MSO and R40, which were atmospherically distilled, showed much higher levels of these compounds. In addition, isoamyl acetate, 2-phenylethyl acetate, ethyl heptanoate, and diethyl butanedioate were identified as major compounds. Among the various esters detected in this study, isoamyl acetate with a banana-like flavor [33], ethyl acetate having a sweet and fruity hint [34], and 2-phenylethyl acetate described as a rose or apple-like odor [35] are considered potent fruity compounds, which are also found in various distilled soju, beers, cheongju, and rice wines [35][36][37][38].
Alcohols are major compounds that give various aroma characteristics to liquor along with esters [36]. In the case of alcohol compounds, unlike the esters, the differences among samples were not large. GSO (41.38 mg/L) made under atmospheric distillation showed the highest contents of alcohols, and MBA (20.04 mg/L) made under reduced pressure distillation using millet, sorghum, and rice showed the lowest. Isoamyl alcohol, isobutyl alcohol having a fruity flavor [35], and phenylethyl alcohol having a rose odor [37] were detected as major alcohol components. Isoamyl alcohol is known to give off a harmonious flavor when there is an appropriate amount in the liquor, but when there is a large amount, it has an unpleasant, musty flavor [38]. These alcohols are commonly found in various alcoholic beverages such as soju, wine, whiskey, and brandy [30][31][32][33][34][35][36][37][38]. Volatile acids are known to be inappropriate in alcoholic beverages [37]. Octanoic acid was detected in six samples.
Principal component analysis (PCA) was performed using 56 volatile compounds detected in 6 or more samples to determine the overall distribution of volatile substances according to sample separation. The first principal component (PC1) showed 29.40% explanatory power, while the second principal component (PC2) showed 13.49%, as shown in Figure 1. The volatile compounds found on the positive side of PC1 ( Figure 1) were ethyl lactate (es16), ethyl undecanoate (es41), ethyl dodecanoate (es48), ethyl tetradecanoate (es56), 1-nonanol (al15), octanoic acid (ac2), nonanal (ad2), and benzenaldehyde (ad5). These compounds are considered to be the major compounds of the atmospherically distilled samples rather than the vacuum distilled samples, as also determined in other studies [7,15]. Moreover, across the second principal component (PC2), there was a distinction among samples based on the major ingredients used for brewing. The sample brewed using only sweet potatoes (R40) was detected to have higher concentrations of esters such as ethyl 2-methylbutanoate (es6), propyl octanoate (es24), 2-methylbutyl octanoate (es35), and 2-decanone (ke2) along the positive side of PC2.    Table 1; (5) volatiles were identified based on the following criteria: A, mass spectrum and retention index consistent with those of an authentic standard; B, mass spectrum consistent with that of the Wiley 275 mass spectrum database.  Table 3.
By examining the sample distribution in Figure 1, MSO, an atmospheric pressure distillation product, showed a significant difference along the positive side of PC1, compared to the other samples. PJU and R40 made under atmospheric distillation were also positioned on the positive side of PC1. Most of the other samples except MBA, HBI, and CTO were distributed on the negative side of PC1, which were vacuum distilled samples. Across PC2, there was a distinct separation between R40 and MBA, which were manufactured using sweet potatoes or various grains such as sorghum and millet, respectively. As a result, this plot demonstrates that the composition of volatile compounds in these samples differed according to the raw materials and the distillation method.

Sensory Characteristics of Distilled soju Samples by Descriptive Analysis (DA)
To depict the sensory characteristics of the distilled soju samples, sensory descriptive analysis was performed. The mean intensity ratings of the 18 samples are presented in Table 5. The ANOVA performed on the 28 sensory attributes of the 18 samples revealed that all attributes except 'metal', 'bitter', 'alcohol flavor', 'astringent', 'spicy', and 'swallow' were significantly different across the samples (p < 0.05). Because the alcohol concentration of each sample was adjusted to the same level, a similar intensity of alcohol and mouthfeel-related sensory attributes could be expected. This outcome was also reported in the descriptive analysis of eleven distilled spirits also adjusted to the same alcohol level [15].  Table 3.
By examining the sample distribution in Figure 1, MSO, an atmospheric pressure distillation product, showed a significant difference along the positive side of PC1, compared to the other samples. PJU and R40 made under atmospheric distillation were also positioned on the positive side of PC1. Most of the other samples except MBA, HBI, and CTO were distributed on the negative side of PC1, which were vacuum distilled samples. Across PC2, there was a distinct separation between R40 and MBA, which were manufactured using sweet potatoes or various grains such as sorghum and millet, respectively. As a result, this plot demonstrates that the composition of volatile compounds in these samples differed according to the raw materials and the distillation method.

Sensory Characteristics of Distilled Soju Samples by Descriptive Analysis (DA)
To depict the sensory characteristics of the distilled soju samples, sensory descriptive analysis was performed. The mean intensity ratings of the 18 samples are presented in Table 5. The ANOVA performed on the 28 sensory attributes of the 18 samples revealed that all attributes except 'metal', 'bitter', 'alcohol flavor', 'astringent', 'spicy', and 'swallow' were significantly different across the samples (p < 0.05). Because the alcohol concentration of each sample was adjusted to the same level, a similar intensity of alcohol and mouthfeelrelated sensory attributes could be expected. This outcome was also reported in the descriptive analysis of eleven distilled spirits also adjusted to the same alcohol level [15]. a-k Mean values with the same letter in a row are not significantly different, with significance set at p < 0.05 by Fisher's least significant difference test. The intensity of the attributes ranged from 0 to 9 (0, none; 1, very weak, 5: moderate, 9: very strong). A,B The sensory attribute codes and samples are defined in Tables 1 and 2. To examine the relationships among the sensory terms and separations of samples at a glance, principal component analysis (PCA) was performed using the mean ratings of 22 sensory characteristics showing significant differences across the 18 samples (Figure 2). PC1 and PC2 explained 69% and 13% of the variance across the samples, respectively. PC1 showed a contrast between 'earthy'/'yeasty'/'barley'/'soy_A' and 'sweet aroma'/'fruity'/fruit-related aroma attributes, as shown in Figure 2. PC2 seemed to show a contrast between 'woody_A'/'bleach_A' and 'sour_A'. Examining the sample configuration (Figure 2), along PC1, there was a strong separation between samples shown by DA to be fruity and sweet and those which were low in fruitiness but high in sensory attributes related to yeasty and earthy. MSO, PJU, MIR, and GSO were located to the far right along PC1, indicating high levels in 'earthy', 'yeasty', 'barley', and 'soy_A'. Likewise, these samples were made using atmospheric distillation, except for MIR. These typical sensory characteristics of atmospherically distilled liquors were also revealed in sensory studies of commercial distilled liquor products [15,17]. Conversely, DJA, R25, and R40 appeared on the negative side of PC1 with strong fruit-and sweet-related sensory characteristics. Along PC2, CTO and OAK, both aged in oak, were located on the far positive side, which showed significantly high intensities in 'woody_A' and 'bleach_A'. To examine the relationships among the sensory terms and separations of samples at a glance, principal component analysis (PCA) was performed using the mean ratings of 22 sensory characteristics showing significant differences across the 18 samples (Figure 2). PC1 and PC2 explained 69% and 13% of the variance across the samples, respectively. PC1 showed a contrast between 'earthy'/'yeasty'/'barley'/'soy_A' and 'sweet aroma'/'fruity'/fruit-related aroma attributes, as shown in Figure 2. PC2 seemed to show a contrast between 'woody_A'/'bleach_A' and 'sour_A'. Examining the sample configuration (Figure 2), along PC1, there was a strong separation between samples shown by DA to be fruity and sweet and those which were low in fruitiness but high in sensory attributes related to yeasty and earthy. MSO, PJU, MIR, and GSO were located to the far right along PC1, indicating high levels in 'earthy', 'yeasty', 'barley', and 'soy_A'. Likewise, these samples were made using atmospheric distillation, except for MIR. These typical sensory characteristics of atmospherically distilled liquors were also revealed in sensory studies of commercial distilled liquor products [15,17]. Conversely, DJA, R25, and R40 appeared on the negative side of PC1 with strong fruit-and sweet-related sensory characteristics. Along PC2, CTO and OAK, both aged in oak, were located on the far positive side, which showed significantly high intensities in 'woody_A' and 'bleach_A'.

Relationships among Chemical, Volatile, and Sensory Profiles Using Multiple Factor Analysis
To understand the relationships among the chemical, volatile, and sensory data sets including 4 chemical compositions, 56 volatile compounds, and 22 sensory attributes, MFA was applied to the 18 distilled samples. From the global analysis of the MFA using the three data sets, the first two factorial axes accounted for 36.9% and 11.7% of the total variance, respectively. The first eigenvalue, 2.47, of the global MFA was very close to the maximum eigenvalue that could be reached, while the eigenvalue for the second factor was 0.78. In this sense, the first factor was a major direction for the interpretation of the MFA. The coordinates, respective contributions, and squared cosines of the groups of variables of the first two factors of the global MFA are presented in Table 6. The coordinates of the three data sets were highly related to the first factor as shown by the values of 0.84, 0.76, and 0.86 for the chemical, sensory, and volatile groups, respectively. The three data sets showed a similar contribution (30.9-35.1%) to the formation of F1, while volatile (42.5%) and sensory (25.6%) showed a difference in the formation of F2 (Table 3). The RV coefficient, which is a multivariate analogue of the correlation coefficient in MFA, was 0.59 between the chemical and volatile data, which is stronger than the RV coefficients of the chemical and sensory data or the volatile and sensory data, with values of 0.45 and 0.53, respectively.  Figure 3 shows the projections of the active variables in the global MFA. The four chemical parameters, namely, lactic acid, TA, Brix, and pH, showed a strong contribution to F1 with values of 12.9, 10.1, 6.0, and 4.9%, respectively. Secondly, the sensory variables such as 'earthy', 'yeasty', 'barley', 'soy_A', 'salty', and 'sour' on the positive side of F1, with 'sweet', 'fruity', and fruit-related attributes on the opposite side, showed a strong contribution to the first factor. This distinction also appeared in the PCA of the sensory data ( Figure 2). However, 'woody_A', 'sour_A', 'bleach_A', 'body', and 'alcohol_A' were found to show a weak contribution to F1 and F2. Because the volatile data consisted of 56 compounds, many compounds showed little contribution to the formation of factors 1 and 2, especially those positioned near the centroid. However, ethyl trans-4-decenoate (es36), ethyl nonanoate (es25), 1-nonanol (al15), ethyl 3-phenylpropanoate (es54), ethyl undecanoate (es41), ethyl tetradecanoate (es56), octanoic acid (ac2), ethyl lactate (es16), 3-methylbutyl octanoate (es23), diethyl succinate (es37), and ethyl phenylacetate (es45) showed a strong contribution to F1, with a clear association with the 'earthy', 'yeasty', 'barley', 'soy_A', 'salty', and 'sour' attributes. Furthermore, all of these compounds also showed strong positive correlations with TA and lactic acid, as well as with the 'earthy', 'earthy_A', 'salty', and 'sour' attributes (p < 0.05). Conversely, fruit-and sweet-related sensory attributes were highly correlated with isobutyl acetate (es4), isoamyl acetate (es8), and 2-phenylethyl acetate (es47) (p < 0.05).
The distribution of the 18 samples in the global MFA is presented in Figure 4. This plot shows an overall similar pattern to the sensory PCA. MSO, which showed prominent yeasty and earthy flavors under atmospheric pressure distillation, was located on the far-right side, with a dominant contribution (56.8%) to F1. Other atmospherically distilled samples such as PJU were also positioned on the right side of F1. However, atmospherically distilled GSO and R40 did not show a similar grouping. Other samples such as JRO, DJA, R25, and HBI showed higher levels of contributions to F1 in the range from 5.14 to 7.63%. CTO, IDO, SRE, IPU, GSO, and MIR showed a minimal contribution to F1. The DJA and R40 samples with strong fruit flavors and sweetness showed a higher contribution to F2, with 29.93 and 7.58%, respectively. The apparent grouping of the samples by their raw materials was not determined by MFA. The distribution of the 18 samples in the global MFA is presented in Figure 4. This plot shows an overall similar pattern to the sensory PCA. MSO, which showed prominent yeasty and earthy flavors under atmospheric pressure distillation, was located on the farright side, with a dominant contribution (56.8%) to F1. Other atmospherically distilled samples such as PJU were also positioned on the right side of F1. However, atmospherically distilled GSO and R40 did not show a similar grouping. Other samples such as JRO, DJA, R25, and HBI showed higher levels of contributions to F1 in the range from 5.14 to 7.63%. CTO, IDO, SRE, IPU, GSO, and MIR showed a minimal contribution to F1. The DJA   Tables 1-4. and R40 samples with strong fruit flavors and sweetness showed a higher contribution to F2, with 29.93 and 7.58%, respectively. The apparent grouping of the samples by their raw materials was not determined by MFA.  Table 1.

Conclusion
The volatile compounds and sensory characteristics of 18 distilled soju samples made with different types of materials and distillation methods were examined. The volatile composition of the samples was primarily determined by distillation methods (atmospheric vs. reduced pressure) and major raw ingredients such as sweet potato or rice. Soju samples manufactured using atmospheric distillation such as MSO and PJU had higher levels of longer-chain esters, 1-nonanol, and furfural, while the other samples produced by vacuum distillation showed much lower levels of those compounds. In addition, R40 made from sweet potatoes showed a distinctive sweet fruitiness in the volatile and sensory profiles compared to other samples made of grains and rice. As a result of sensory descriptive analysis, earthy and yeasty flavors with a barley hint were the main characteristics of the atmospherically distilled samples. From the MFA using chemical, sensory, and volatile data, the selected volatiles and chemical parameters showed strong interrelations with the sensory data. Volatiles such as 1-nonanol, octatonic acid, and longer-chain esters such as diethyl succinate, ethyl phenylacetate, and ethyl tetradecanoate and chemical parameters such as TA, Brix, and lactic acid were highly associated with earthy and yeasty sensory characteristics, while isoamyl acetate, isobutyl acetate, and 2-phenylethyl   Table 1.

Conclusions
The volatile compounds and sensory characteristics of 18 distilled soju samples made with different types of materials and distillation methods were examined. The volatile composition of the samples was primarily determined by distillation methods (atmospheric vs. reduced pressure) and major raw ingredients such as sweet potato or rice. Soju samples manufactured using atmospheric distillation such as MSO and PJU had higher levels of longer-chain esters, 1-nonanol, and furfural, while the other samples produced by vacuum distillation showed much lower levels of those compounds. In addition, R40 made from sweet potatoes showed a distinctive sweet fruitiness in the volatile and sensory profiles compared to other samples made of grains and rice. As a result of sensory descriptive analysis, earthy and yeasty flavors with a barley hint were the main characteristics of the atmospherically distilled samples. From the MFA using chemical, sensory, and volatile data, the selected volatiles and chemical parameters showed strong interrelations with the sensory data. Volatiles such as 1-nonanol, octatonic acid, and longer-chain esters such as diethyl succinate, ethyl phenylacetate, and ethyl tetradecanoate and chemical parameters such as TA, • Brix, and lactic acid were highly associated with earthy and yeasty sensory characteristics, while isoamyl acetate, isobutyl acetate, and 2-phenylethyl acetate were related to sweet fruitiness. These components can be selected as preliminary sensory quality indicators of distilled soju, and it is expected that more accurate volatile compounds contributing to sensory characteristics can be identified through the analysis of trace volatile compounds.