Diversity of flavour characteristics of table grapes and their contributing volatile compounds analysed by the solvent-assisted flavour evaporation method

Abstract To identify the compounds that contribute to the diverse flavours of table grapes, the flavours and volatile compounds of 38 grape cultivars harvested over 3 years are evaluated through sensory analysis and solvent-assisted flavour evaporation (SAFE). The cultivars are characterized and grouped into seven clusters by hierarchical cluster analysis (HCA) using sensory evaluation data with a flavour wheel specific to table grapes. These clusters were similar to conventional flavour classifications, except that the foxy and neutral cultivars form multiple clusters, highlighting the flavour diversity of table grapes. The SAFE method provides a comprehensive profile of the volatile compounds, including slightly volatile compounds whose profiles are lacking in hybrid grapes and Vitis rotundifolia. The sensory evaluation is supported by the volatile compound profiles, and relationships between the datasets are clarified by multivariate analysis. Specific accumulations and combinations of compounds (α-pinene, β-pinene, phenylethyl alcohol, furaneol, mesifurane, methyl N-formylanthranilate, and mixed ethyl ester and monoterpenoid) were also identified that contribute to the diversity of flavours (fresh green, floral, fruity, fatty green, sweet, fermented/sour) in table grapes, including linalool and linalool analogues (muscat flavour) along with ethyl ester and hydroxyethyl esters (foxy flavour). The accumulation of these compounds was positively related to a higher flavour intensity. Their specific accumulation and combination supported the flavour diversity of table grapes. This study identified novel flavour-associated compound profiles in table grapes through in-depth volatile compound analysis and non-conventional multivariate analysis.


Introduction
Vitis vinifera are the most commonly cultivated grapes, with a global cultivation area of 7.5 million ha, half of which is used to produce wine [1].However, interspecific hybrids (especially the hybrids obtained from Vitis labrusca L. and V. vinifera L.) and the related genus Muscadinia (Vitis rotundifolia) are grown for wine and table grapes in regions with an unfavourable climate or disease pressure [2].
The taste and aroma of various fruits give rise to distinct flavours [3].Flavour is an integral property of both wine and table grapes and has long been a subject of cultural and scientific interest.Flavour and aroma wheels are popular tools in evaluation the f lavour and aroma profiles of various foods and beverages, including wines, for which they provide a visual representation of the most common aroma [4].This allows producers, distributers, sellers, and consumers of wine to evaluate its quality using a shared language.In contrast, the f lavour of table grapes is not commonly evaluated using f lavour and aroma wheels, likely because aroma is less important than other characteristics.
Many grape cultivars are classified based on their f lavour profile, such as muscat, foxy, muscadine, and neutral (no or weak f lavour).The f lavours of table grapes can be found in the literature and various databases [5][6][7][8][9][10][11][12][13][14].Information on the f lavour characteristics of most table grape cultivars is limited, with the exceptions of muscat and foxy; however, hybrid grapes have a variety of f lavours that are difficult to express only in terms of muscat and foxy.Additionally, technical terms such as foxy and muscat may be unfamiliar to general consumers.
Aroma derives mainly from complex combinations of volatile compounds [3]; thus, the f lavour of grapes is inf luenced by the composition and concentrations of volatile compounds, which depend on several factors, including genetics, species, location, climate, cultivation practices, and ripeness [15,16].The volatile compounds found in grapes include monoterpenes, C 13 -norisoprenoids, alcohols, esters, and carbonyls [17].Esters and terpenes provide fruity and f loral aroma characteristics [15], while the green leafy aroma of grapes is derived from C 6aldehydes and -alcohols [18].Grape berries of the muscat cultivar (V.vinifera L.) exhibit a distinctive aroma associated with their high levels of monoterpenes [19].Similarly, V. labrusca-derived hybrids and V. rotundifolia often contain methyl anthranilate, 2aminoacetophenone, and furaneol, which contribute to their foxy aroma [20,21].Although distinctive f lavour compounds of certain grape species have been identified, the f lavour compounds of hybrid grape cultivars are poorly understood.
The volatile compound profiles of several table grapes have been identified using headspace-solid-phase microextraction (HS-SPME) [9,16,[22][23][24]; however, this technique may be unable to identify volatile compounds with relatively high boiling points and low volatilities, which can contribute significantly to the overall aroma of table grapes.Solvent-assisted f lavour evaporation (SAFE) is widely used to separate volatile compounds from complex matrices, and has proven effective in recovering trace compounds and high-boiling-point compounds at low temperatures [25][26][27].SAFE has been applied in the analysis of volatiles in horticultural crops since its development in 1999, but its use in the analysis of the volatiles in grapes is limited.
Both orthonasal olfaction (perceiving odours through the nostrils) and retronasal olfaction (perceiving odours through the throat while eating or drinking) are crucial in perceiving f lavour.Odour activity values (OAVs; defined as the ratio of concentration to odour threshold) and gas chromatography-olfactometry (GC-O) are common screening tools to evaluate the potential importance of aroma compounds in foods.Several studies have used GC-O and OAVs to identify important f lavours in table grapes [21][22][23]; however, most of these studies were based solely on orthonasal olfaction.In recent years, integrated analysis of largescale sensory and chemical data, such as multivariate analysis, has enabled the identification of f lavour-and taste-related volatiles.This provides novel insights into f lavour chemistry, the interactions between taste and retronasal olfaction, and establishes a paradigm for enhancing the appeal of natural products [28].
Existing studies on the f lavour and f lavour compounds of table grapes have primarily examined foxy and muscat f lavours, along with limited volatile component analysis and screening methods.Knowledge of other f lavours and the compounds that contribute to them is comparatively lacking.In this study, we conducted sensory analysis to evaluate the f lavour profiles of 38 table grape cultivars, including muscadine (V.rotundifolia).Volatile compounds in the grape berries were analysed using a combination of SAFE and gas chromatography-mass spectrometry (GC-MS).The relationship between the f lavour and volatile compound profiles was examined using multivariate data analysis to identify the volatile compounds that contribute to grape f lavours.

Sensory characterisation
The f lavour characteristics of hybrid table grapes were evaluated using a f lavour wheel (Fig. S1), which was then used to investigate the f lavour characteristics of a total of 102 samples from 38 table grape cultivars with various f lavour types harvested over three years (Table 1, Table S1).Hierarchical cluster analysis (HCA) of the sensory evaluation data from the f lavour intensity, foxy f lavour and muscat f lavour of 102 samples was applied to classify their f lavour characteristics and categorise the samples accordingly (Fig. 1).Furthermore, fresh green, fatty green, fermented/sour, f loral, sweet, and fruity f lavours were also represented by heat maps (Fig. 1).The mean sensory evaluation scores of the clusters shown in Fig. 1 are listed in Table S2.The codes in Table 1 representing the abbreviated names of the cultivars are hereafter used throughout the manuscript.
Cluster analysis was carried out on the sensory evaluation data of 38 grape cultivars for the conventional elements of f lavour classification, including foxy f lavour, muscat f lavour, and f lavour intensity.The 102 samples were categorised into seven clusters (Fig. 1).Six cultivars (HA, HI, DE, SR, NP, and TR) occasionally clustered differently, whereas 27 cultivars displayed similar clustering across the years (Fig. 1).All samples in cluster 1 are of the muscat f lavour type (Table 1).The grapes in this cluster displayed muscat, f loral, fresh green, and sweet f lavours with f lavour intensities of at least (Table S2).With the exception of HA19, the samples in cluster 2 have a neutral f lavour (Table 1).Indeed, cluster 2 showed the lowest f lavour intensities among the clusters, with a fresh green value of 2.66, but no characteristic flavour (Table S2).Clusters 3 and 4 included a mixture of cultivars previously classified as having a foxy f lavour or other f lavour types (Table 1).Samples HV17 and HV18 exhibited mixed foxy and muscat f lavours (Fig. 1).Cluster 3 exhibited sweet, f loral, and fruity f lavours with a high f lavour intensity above three (Table S2).Cluster 4 was characterised by sweet, foxy, fruity, f loral, fatty green, and fermented/sour f lavours with a f lavour intensity greater than three (Table S2).Clusters 5 and 6 were characterised by foxy, fruity, fermented/sour, fatty green, f loral, and sweet f lavour characteristics with a f lavour intensity greater than three (Table S2).Cultivars in clusters 5 and 6 (with the exceptions of BB and TR17) have a foxy f lavour (Table 1); however, cluster 5 showed considerably lower values in foxy, fatty green, and fermented/sour f lavours than cluster 6 (Table S2).All samples in cluster 7 have muscadine f lavours (Table 1).Although clusters 6 and 7 had similar f lavour characteristics (Fig. 1), the intensity of the foxy f lavour was lower in cluster 7 than in cluster 6 (which had the highest value of 6.25 for foxy f lavour).Conversely, cluster 7 had a higher value for sweet f lavour compared to cluster 6 (5.00 vs. 3.56).

Analysis of volatile compounds by the SAFE method and GC-MS
Following SAFE extraction, several hundred peaks were detected in the GC-MS data, and 98 volatile compounds were identified (Table 2).

Profiling of volatile compounds by principal component analysis
The log-transformed quantitative data obtained from the 98 volatile compounds in 102 fresh table grape samples were analysed using principal component analysis (PCA).The first four principal components accounted for 49.8% of the total variation in the data (Fig. 2).
In the plot plane, the samples fell into the same clusters identified in the sensory-sorting task.PC1 shows that monoterpenoids contributed to cluster 1, and ethyl esters to clusters 5-7 (Fig. 2A and 2D), thereby explaining the separation of cluster 1 from clusters 5, 6, and 7. PC2 explained the separation of cluster 7 from the other clusters; Hexyl, butyl, and phenylethyl esters and their corresponding alcohols contributed to the  formation of other clusters, but not cluster 7 (Fig. 2A and 2D).Although clusters 2, 3, and 4 overlapped in PC1 and PC2 (Fig. 2A), the volatile compound profiles of these clusters were characterised by PC3 and PC4 (Fig. 2).PC1 and PC3 indicated the presence of mixed ethyl esters and monoterpenoids in cultivars such as 'Keuka' (Fig. 2B and 2E).PC1 and PC4 separated clusters 3, 4, and 6.The samples in these clusters contained several compounds such as farnesene, furaneol, and mesifurane (Fig. 2C and 2F).

Content of volatile compounds for each grape flavour classification based on sensory evaluation
The content of volatile compounds in each grape f lavour classification based on Fig 1 is represented in the heat map (Fig. 3), and the content of volatile compounds in each cluster is summarised in   1).Samples denoted 1 and 2 were harvested from different trees in the same year.Compound code: compound type and compound number (Table 2).
(Table 3).Cluster 1 had a higher monoterpenoid content than the other clusters, while cluster 2 was dominated by C 6 compounds (Table 3).Clusters 3 and 4 tended to have the same volatile profile as cluster 2, but typically contained more ethyl esters, alcohols, monoterpenoids, lactones, and furanones (Table 3).Cluster 3 contained more sesquiterpenes, while cluster 4 contained more The contents of 98 volatile compounds were quantified as 3-heptanol equivalents, log-transformed, and plotted in a heat map, in which lower (higher) volatile compound contents are presented in green (red), as shown on the scale.The samples are indicated by a code of abbreviated cultivar name (Table 1) followed by harvest year (e.g., FR17 = Fry harvested in 2017). 1 and 2 denote samples harvested from different trees in the same year.The cluster classification of the samples is based on the classification by sensory evaluation in Fig. 1.
lactones and furanones such as furaneol and γ-decanolactone (Table 3).Cluster 5 was characterised by abundant alcohols, ethyl esters, and aldehydes and ketones (Table 3).Cluster 6 was dominated by alcohols, ethyl esters, lactones and furanones, and methyl esters (Table 3).Conversely, the total volatile compound content in cluster 7 was higher than that of all the other clusters (Table 3).Cluster 7 was dominated by esters, including ethyl esters and alcohols such as butyl, hexyl, and phenethyl alcohol (Table 3).The content of these classes of compounds is significantly higher than that of other clusters, including C 13 norisoprenoids (Table 3).

Volatile compounds associated with flavour characteristics through partial least squares analysis
Quantitative data from the analysis of 98 volatile compounds in 102 fresh grape samples were log-transformed and the relationship between volatile compound profiles and f lavour characteristics was determined using the partial least squares (PLS) method.
A model was developed for each f lavour.Table 4 summarises the parameters of the PLS regression model.The cumulative R 2 Y and Q 2 values of all f lavour models except fresh green and f loral exceeded the threshold of 0.5.Table 5 lists the 10 compounds with the highest VIP values for projection among foxy, muscat, and f lavour intensity, as well as the compounds that were previously considered important for foxy and muscat f lavours.In addition to volatile compounds with an OAV > 1, other compounds also exhibited high VIP values (Table 5).This study identified 27 compounds as key variables in the muscat f lavour with VIP values greater than 1 (Table 4).Eighteen monoterpenoids (including linalool, geraniol, nerol, αterpineol, and cis-rose oxide) and α-caryophyllene were positively associated with the muscat f lavour (Table S3).Linalool derivatives (including linalool oxide and 8-hydroxylinalool) showed the strongest association with the muscat f lavour (Table 5A), while nine ethyl esters (including ethyl octanoate, ethyl decanoate, ethyl-3-hydroxybutanoate, ethyl-2-butenoate, ethyl hexanoate, and ethyl butanoate) exhibited a negative association with the muscat f lavour (Table S3).Forty compounds were identified as key variables in the foxy f lavour with VIP values greater than 1 (Table 4), whereas the VIP values of furaneol and methyl anthranilate did not exceed 1 (Table 5B).Seventeen ethyl esters (including ethyl-3-hydroxyhexanoate, ethyl octanoate, ethyl decanoate, ethyl-3-hydroxybutanoate, ethyl hexanoate, and ethyl (E,Z)-2,4-decadienoate) and 11 other compounds (including methyl-3-hydroxybutanoate, β-phenethyl acetate, acetoin, methyl salicylate, and β-pinene) were positively associated with the foxy flavour (Table S3).Methyl N-formylanthranilate, which has a green f loral-like aroma similar to that of Concord grapes, and mesifurane, described as having a burned, sherry-like, or fusty aroma, were also associated with the foxy f lavour (Table 5B).Eleven monoterpenoids (including trans-linalooloxide (pyranoid), hotrienol, linalool, and cis-rose oxide, etc.) and 3-hexenal showed a negative associated with the foxy f lavour (Table S3).Compounds related to muscat and foxy f lavours also inf luenced the f lavour intensity and several other f lavour characteristics (Table 5A and 5B).Compounds that were strongly associated with the muscat f lavour (linalool, linalool derivatives, and monoterpenoids) were positively related to f loral and fresh green f lavours, and negatively relate to foxy and fatty green f lavours (Table 5A).Compounds that were strongly associated with the foxy f lavour (ethyl esters) were positively related to f lavour intensity and the fruity, sweet, f loral, fermented/sour, and fatty green f lavours, but were negatively associated with muscat and fresh green f lavours (Table 5B).Compounds found only in muscadines, such as butyl, hexyl, and phenethyl esters, were related to f lavour intensity, positively associated with the fruity, sweet, f loral, fermented/sour, and fatty green f lavours, and negatively associated with the fresh green f lavour (Table 5C, Table S3).Methyl-3-hydroxybutanoate, ethyl cinnamate, phenylethyl alcohol, mesifurane, furaneol, αpinene, and β-pinene were not strongly associated with the foxy and muscat f lavours; however, these compounds were strongly associated with the f lavour intensity and fruity, fatty green, sweet, f loral, and fermented/sour f lavours (Table 5C, Table S3).

Discussion
A large-scale sensory and chemical analysis of 102 grape samples was conducted over three years to identify the volatile compounds responsible for grape f lavour.Our study analysed grape f lavour by sensory evaluation and volatile compound profiles obtained using the SAFE method.Furthermore, this study builds upon previous research [9] by using a large number of samples incorporating a greater variety of f lavours, and provides new insights into the volatile compounds that contribute to grape flavour.

Profiles of a wide variety of flavour characteristics in table grapes
Flavour characteristics are typically determined by sensory evaluation.The f lavour characteristics of ripe table grapes include muscat, foxy, muscadine, and neutral [5,7,14].In this study, a f lavour wheel including foxy, muscat, fresh green, fatty green, fermented/sour, f loral, sweet, and fruity f lavours, among others, was created for table grapes (Fig. S1).This f lavour wheel includes categories similar to those used by Sasaki et al. in the sensory evaluation of table grapes [22], along with additional categories such as fresh green, fatty green, fermented, and muscat f lavours.Our findings demonstrate that the developed f lavour wheel facilitates the classification and evaluation of the f lavour characteristics of table grapes and will therefore expedite the future analysis of grape f lavour profiles.
Cluster analysis of the sensory evaluation data of the conventional elements of f lavour classification, i.e., foxy f lavour, muscat f lavour, and f lavour intensity, produced seven clusters (Fig. 1).The classification of the conventional f lavour characteristics of muscat, none, and muscadine formed clusters; however, the remaining f lavours, including foxy, formed multiple clusters, indicating their diversity (Fig. 1).A possible factor in the formation of multiple clusters was the diversity of foxy f lavour and f lavour intensity among the samples is a contributing factor in the formation of multiple clusters.In particular, certain samples in cluster 3 exhibited sweet, f loral, and fruity f lavours rather than muscat and foxy f lavours (Fig. 1, Table S2).Interestingly, grapes of the same cultivar grown in different years tended to cluster together, highlighting the significant inf luence of genetic factors on f lavour characteristics (Fig. 1).

Analysis of volatile compounds using SAFE
The application of SAFE in the analysis of the volatile compounds of grapes is limited.To the best of our knowledge, the present study represents the first use of SAFE to obtain volatile profiles in table grapes, despite its use for this purpose in the analysis of wine [29].We detected C 6 compounds (including hexanal,    (E)-2-hexenal, and 3-hexenol), abundant esters (such as ethyl butanoate and ethyl hexanoate), and monoterpenoids (such as linalool, geraniol, nerol, citronellol, and α-terpineol) in table grapes (Table 2).While earlier volatile profiling studies of table grapes also identified these compounds [16,18], we also observed intermediate-molecular-weight esters and acids, hydroxyl esters, hydroxyl monoterpenoids and polyols, furaneol, methyl anthranilate, methyl N-formylanthranilate, acetoin, and vanillin (Table 2), which have rarely been detected by the HS-SPME analysis of the volatile compounds in table grapes [16,18].Further, to the best of our knowledge, no other study has detected methyl N-formylanthranilate in grapes, which may be produced by enzymatic or chemical reactions involving the formylation/deformylation of methyl anthranilate and its analogues in the fruit.These observations suggest that the SAFE method provides a comprehensive analysis of the volatile compounds, including trace compounds and those with high boiling points, in table grapes and in other horticultural crops [25,26,29].

Relationship between diversities of volatile compounds and flavour characteristics of table grapes
Several esters (particularly ethyl esters) and monoterpenoids contributed to the differences in main volatile compounds found in V. vinifera, V. rotundifolia, and V. interspecific crossing (Fig. 2A and 2D).These results were partially consistent with those of a previous study [16], which found that terpenoids are prevalent in V. vinifera cultivars with muscat aroma, while esters are the most abundant in V. labrusca and its hybrids with V. vinifera or V. amurensis.Furthermore, both V. rotundifolia and V. interspecific crosses exhibited varied volatile profiles that include more than ethyl esters and monoterpenoids (Fig. 2).V. rotundifolia contained hexyl, butyl, and phenylethyl esters and the corresponding alcohols (Fig. 2A and 2D), while V. interspecific crossing exhibited a wide variety of volatile profiles: one volatile profile was similar to that of V. vinifera varieties with muscat f lavour, such as 'Shine Muscat' (Fig. 2A and 2D, Fig 3).Certain varieties such as Keuka had a mixed volatile profile of ethyl esters and monoterpenoids (Fig. 2B and 2E, Fig 3).In contrast, other varieties such as 'Muscat Berry A' and 'Delaware', which are V. interspecific crossing with the genetic backgrounds of V. lincecomii and V. amurensis, respectively, had lower levels of esters and monoterpenoids but higher levels of farnesene, furaneol, and mesifurane (Fig. 2C and 2F, Fig 3).These results suggest that the diversity of volatile components in table grapes arises from differences in the content and composition of not only ethyl esters and monoterpenoids, but also those of other compounds like hexyl, butyl, phenylethyl esters, their corresponding alcohols, and furanones, such as furaneol and mesifurane.The diversity of volatile profiles in table grapes is likely related to the genetic background.Interestingly, the first four principal components (PC1, PC2, PC3, and PC4) were consistent with the f lavour classification obtained by sensory evaluation (Fig. 2), suggesting that the diversities in f lavour characteristics and volatile compounds are interrelated.

Search for volatile compounds associated with flavour characteristics using PLS analysis
The relationship between sensory descriptors and volatile compounds was determined using SAFE-GC-MS and PLS.For the Q 2 parameter, a significance threshold of 0.5 generally indicates that the model has a high explanatory power [30].All f lavour models developed here, except for those of fresh green and f loral, exhibited high explanatory power (Table 4).The models of foxy f lavour, muscat f lavour, and f lavour intensity were particularly effective; thus, volatile compounds analysed by SAFE-GC/MS are expected to accurately predict the grape f lavour, particularly in the case of grapes with a foxy or muscat f lavour.PLS-VIP identified the volatile compounds in grapes that are responsible for f lavour characteristics.A parameter with a higher VIP value is more important to the model; parameters with VIP values greater than 1 are important compound f lavour characteristic [9,24,31].Analysis of each f lavour characteristic identified between 27 and 46 compounds with VIP values greater than 1, many of which showed association with f lavour characteristics (Table 4).This indicates that multiple compounds are involved in the formation of f lavours.In addition, compounds can be associated with different f lavours; for example, the ethyl esters that contribute to the foxy f lavour and monoterpenoids that contribute to the muscat f lavour were overlappingly associated with other f lavours, including fruity and f loral (Table 5, Table S3).This suggests that f lavours, such as foxy and muscat, are supported by the complex combination of f lavours.We therefore summarize the volatile compound profiles of representative f lavour categories based on sensory evaluation.S5.

Characteristics of flavour and volatile compounds in cluster 1, which primarily contains Muscat varieties
'Muscat' is known for its distinct f loral f lavour [19].Cluster 1, which primarily includes muscat varieties, had muscat, f loral, fresh green, and sweet f lavours with f lavour intensities above three (Table S2).The most abundant monoterpenoids linalool, nerol, and geraniol are the primary f lavour compounds responsible for the muscat f lavour [32,33].Monoterpenoids were present at approximately 3.12 mg•kg −1 FW and account for 58.3% of the composition in cluster 1 (Table 3).Twenty-seven compounds including linalool, nerol, and geraniol exhibited a positive association with the muscat f lavour (Table S3).
In particular, linalool and its derivatives (cis-linalooloxide (pyranoid), 8-hydroxylinalool, trans-linalooloxide (pyranoid), and 2,6-dimethyl-1,7-octadiene-3,6-diol) exhibited the strongest association (Table 5).Interestingly, linalool and its derivatives showed a positive relationship with the fresh green f lavour, while linalool derivatives, α-terpineol, and cis/trans-roseoxide exhibited positive relationship with the f loral f lavour (Table 5).Cluster 1 showed significantly higher linalool, hotrienol, trans-linalooloxide (pyranoid), and cis-rose oxide contents than the other f lavour characteristic clusters (Table 3).These compounds can therefore be considered as markers of muscat, fresh green, and f loral f lavours in table grapes.Ruiz-García al. [34] also suggested that the correlation between rose oxide and muscat f lavour could be a useful tool in the identification and selection of table grapes.Furthermore, linalool and cis-rose oxide have OAV values greater than 1 (Table 5, Table S5), suggesting their potential inf luence on the perception of muscat, fresh green, and f loral f lavours in muscat varieties.

Characteristics of flavour and volatile compounds in clusters 5 and 6, which primarily contain foxy varieties
The term 'foxy' describes the unique, earthy, and sweet muskiness present in most V. labrusca and V. rotundifolia species [35].Clusters 5 and 6 primarily contained foxy varieties and had high levels of foxy, fruity, fermented/sour, fatty green, f loral, and sweet f lavours with f lavour intensities above three (Table S2).
At least three compounds contribute to the foxy f lavour: methyl anthranilate, 2-aminoacetophenone, and furaneol [20,21].In this study, methyl anthranilate was detected at a frequency of 6.25% in cluster 5 and 12.5% in cluster 6, while furaneol was detected at 100% in cluster 6 but was absent in cluster 5 (Table S4).Baek et al. detected 2-aminoacetophenone with in 'Carlos' samples with the foxy f lavour [21], but this compound was not identified in the current study.Interestingly, 28 compounds, including 17 ethyl esters (Table S3), were more positively associated with the foxy f lavour than either methyl anthranilate and furaneol, neither of which showed associations with the foxy f lavour (Table 5).Some of these compounds, including ethyl 3-hydroxyhexanoate, ethyl octanoate, ethyl hexanoate, ethyl butanoate, and ethyl 3-hydroxybutanoate, are known to inf luence the fruity aroma of pears, strawberries, pineapple, and wine [36,37].Because those compounds were detected in most samples of clusters 5 and 6, they are likely important contributing factors to the foxy f lavour of hybrid grapes.
Despite both being considered foxy varieties, clusters 5 and 6 exhibit some differences (Fig. 1).Cluster 5 includes varieties described as favourable foxy, while samples in cluster 6 have significantly higher values for foxy and fatty green f lavours (Table S2).The contents of 18 compounds, including methyl N-formylanthranilate, mesifurane, methyl 3-hydroxybutanoate, acetic acid, and several ethyl esters in clusters 5 and 6 differed significantly (Table 3).All compounds except β-phellandrene were more abundant in cluster 6 than cluster 5. Like methyl anthranilate and furaneol, methyl N-formylanthranilate and mesifurane were detected at low frequencies (6.25%) in cluster 5 but high frequencies (75%-100%) in cluster 6 (Table S4).Further, methyl N-formylanthranilate has green f loral aromas reminiscent of 'Concord' grapes, while mesifurane has aromas reminiscent of strawberries, burned sugar, and caramel.The relative abundance of these 18 compounds may contribute to the f lavour diversity and preference of the foxy varieties.

Characteristics of flavour and volatile compounds in cluster 7, which primarily contains muscadine varieties
Muscadine grapes are known for their dense, foxy, candy like, and uniquely fruity aromas [21,24].Cluster 7, which includes most muscadine varieties, showed foxy, fruity, fermented/sour, fatty green, f loral, and sweet f lavours, which were more intense and richer than those of the other clusters (Table S2).The total compound content of cluster 6 was approximately 5-100 times higher than that of other clusters (Table 3).Some compounds identified by Baek et al. [21] as the most abundant aroma compounds in muscadine juice (furaneol, ethyl butanoate, and phenylethyl alcohol) accumulated and thus increased the f lavour intensity (Table 5, Table S3); however, the most abundant compounds were alcohols (35.2 mg•kg −1 FW, 42.0%) and esters other than ethyl esters (45.1 mg•kg −1 FW, 53.8%) (Table 3).Ten compounds (i.e., 1-hexanol, octanol, phenylethyl alcohol, hexyl hexanoate, β-phenethyl acetate, hexyl octanoate, phenethyl hexanoate, phenylethyl octanoate, benzyl acetate, and phenylacetaldehyde) were significantly more abundant in cluster 7 than in the other clusters and associated with f lavour intensity (Table 3, Table S3).In addition, isoamyl alcohol, benzyl alcohol, γ-hexalactone, and β-ionone were significantly more abundant in cluster 7 than in the other clusters, although they were not strongly associated with f lavour intensity (Table 3, Table S3).These compounds have green, fruity, f loral, and waxy aromas, which may contribute to the unique f lavour of muscadine.

Characteristics of flavour and volatile compounds in clusters 2, 3, and 4, which primarily contain neutral varieties
Cluster 2 primarily contains cultivars classified as 'neutral' f lavour, and has the lowest abundance and number of detected compounds; 46 compounds with a total content of 860 mg•kg −1 FW were detected (Table 3).The C 6 compounds, aldehydes, ketones, and alcohols accounted for 72.8% of the volatile compounds present in cluster 2 (Table 3).C 6 compounds (hexanal, (E)-2-hexenal, and 3-hexenol), also known as 'green leaf volatiles', are basic background volatiles in table grape berries [16,23].Like the C 6 compounds, acetic, hexanoic, nonanoic, and benzoic acids, along with benzyl alcohol, benzaldehyde, and vanillin are considered basic background volatiles because they were detected in 90% of all samples (Table 2).The contents of volatile compounds, including the basic background volatiles, differ significantly among table grape samples [23]; however, we did not observe any significant differences in the content of these compounds among the clusters (Table 3).These basic background compounds were present in cluster 2 and might contribute to the grape f lavour; however, the lack of any distinctive relationship between the compounds present and the f lavours may have resulted in a less f lavourful cluster relative to the other clusters.
Clusters 3 and 4, consisting mostly of specific-f lavoured varieties, were characterized by f lavours other than foxy and muscat, including sweet, fruity, f loral, fatty green, and fermented/sour flavours (Fig. 1, Table S2).Clusters 3 and 4 exhibited more diverse volatile compound profiles than the other clusters owing to the higher number of compounds in these two clusters (75 and 71 compounds, respectively).Additionally, clusters 3 and 4 contained a higher number of compounds with a frequency ratio of less than 50% (48 and 37 compounds, respectively) than the other clusters (5-27 compounds, Table S4).Some specific cultivars, such as 'Keuka', 'Buffalo', 'Delaware', and 'Muscat Bailey A', exhibited distinct compound profiles.For example, 'Keuka' contained a mixture of ethyl esters and monoterpenoids, while 'Buffalo' and 'Delaware' specifically accumulated farnesene, furaneol, mesifurane, and γ-decalactone (Fig. 3).Like 'Keuka', the accumulation of both monoterpenoids and ester compounds produced a distinct f lavour without intensifying the muscat or foxy f lavour owing to the negative relationships between muscat f lavour and esters and between foxy f lavour and monoterpenoids (Table 5).Along with mesifuran, furaneol, which is responsible for the strawberry-like aroma of 'Muscat Bailey A', contributes to the typical caramellike, sweet, f loral, and fruity aroma [36,38].Additionally, fatty green, fruity, and sweet f lavours were associated with the unique f lavour profiles in clusters 3 and 4 (Table 5, Table S3).Although farnesene and γ-decalactone were less strongly associated with the f lavour intensity (Table S3), the sweet or peachy f lavour of γ-decalactone and the woody odour of α-farnesene have been proposed as descriptors for classifying apple cultivars [36].Thus, sensory evaluations of grape f lavours in clusters 3 and 4 correspond to specific compound profiles, suggesting that these cultivars can be used to efficiently breed novel f lavours and preferred cultivars owing to the diversity of their compound profiles and flavours.

Interaction of aroma compounds
Fruits contain numerous volatile compounds that contribute to their diverse sensory perceptions.The qualitative (odour quality) and quantitative (odour intensity) sensory perceptions of these compounds interact in various ways to create the overall sensory experience of the fruit [39][40][41][42].
The OAV, defined as the concentration-odour threshold ratio, is commonly used to assess the contribution of each compound to fruit aroma.Compounds with OAVs greater than one are considered active contributors [9,19,23].Compounds identified as active contributors, including linalool, ethyl hexanoate, ethyl (E,Z)-2,4-decadienoate, β-phenethyl acetate, phenethyl alcohol, furaneol, mesifurane, and other compounds, were also associated with f lavour (Table 5, Table S5).Some linalool derivatives, hydroxyl esters, and other compounds exhibited even stronger associations (Table 5), but the OAVs of these compounds are either below the threshold or without available threshold information.Despite showing OAVs below the threshold, linalool oxides enhance the perception of muscat aroma, and ethyl 3hydroxybutanoate enhances that of fruity f lavour in wine and juice [32,37,41].This demonstrates that approaches based solely on odour thresholds are insufficient, because compounds at or below the threshold may still affect the f lavour of table grapes.The role of volatiles in grape f lavour therefore depends on both their OAVs and their interactions with other compounds.Furthermore, to identify compounds that contribute to grape f lavour, the complex interactions of f lavour-related compounds must be evaluated, including taste, texture, and threshold information.

Conclusion
This study provided detailed f lavour characteristics and volatile compound profiles of the main grape varieties grown as table grapes and genetic resources.Sensory evaluation using a f lavour wheel including f lavours other than foxy and muscat revealed that table grapes have a variety of f lavour characteristics such as sweet, f loral, and fruity.SAFE provided comprehensive profiles of volatile compounds in grapes, including those of slightly volatile substances that have not been reported before.Multivariate analysis using f lavour characterization and volatile compound data identified compounds strongly correlated with f lavour in addition to those identified based on the concentration of individual volatiles and odour thresholds.Both volatile compound analysis with SAFE and multivariate analysis not based on conventional odour units can identify novel classes of f lavour compounds in table grapes.This fundamental knowledge is expected to enable more efficient creation of novel grape cultivars with highly preferred f lavours and more accurate methods to assess the f lavour quality.

Grape samples
Grapes (Vitis spp.) were grown in vineyards at the Grape and Persimmon Research Station (NARO, Japan).The vine ages of the evaluated cultivars were between three and 30 years as of 2017.Vine management such as fertilization and canopy management were performed according to the usual practices in Japan, and the vines were cane-pruned.Kober (Teleki) 5BB was used as the rootstock.A total of 102 fruit samples were harvested from 38 table grape cultivars in 2017, 2018, and 2019 when the acid content was judged to have decreased enough by sensory evaluation (Table S1).Bunches of grapes from each sample were prepared for sensory evaluation and chemical analysis.The harvested grapes were transported to the laboratory at the Grape and Persimmon Research Station for sensory evaluation and sugar and acidity analysis.For volatile compound analysis, the fruits were immediately transported under refrigeration to the University of Tsukuba and stored at 5 • C. The storage period from harvest to extraction was approximately three days.

Sensory analysis by expert panels
Eleven individuals experienced in f lavour development (six males and five females) provided a list of 203 terms to describe the f lavour of grapes ('Pione', 'Kyoho', 'Delaware', 'Shine Muscat', and 'Seto Giants').A f lavour wheel was created by defining and sharing evaluation terms as classifications using the KJ method with six individuals experienced in f lavour development and four sensory evaluators (seven males and three females) (Fig. S1).Over the course of 3 years, evaluation was conducted by four grape breeders individually each year.In the first 2 years (2017 and 2018), evaluation was conducted by the same four panels.In the third year (2019), one panel was replaced and the other three panels were unchanged.Evaluation was performed at the appropriate harvest time for each variety, and no specific time interval was set between evaluations of different varieties.The grapes of each variety were prepared in a tray in the research room, and each panel performed sensory evaluation using multiple cultivars and multiple grape berries.Each panel evaluated each f lavour descriptor (f lavour intensity, foxy, muscat, fresh green, fatty green, fermented/sour, f loral, sweet, and fruity) on a seven-point scale (extremely weak, very weak, weak, moderate, strong, very strong, and extremely strong) using the format shown in Fig. S2, on the degree of each f lavour by reference to the evaluation term shown in Fig. S1.Examples of the f lavour intensity of certain varieties were provided in Fig. S2 as a guide.For example, the f lavour intensity of 'Kyoho' (the leading Japanese variety) is set as four (Moderate).

Figure 1 .
Figure 1.Classification of f lavour characteristics by hierarchical cluster analysis (HCA) using Ward's cluster algorithm for the dataset of f lavour intensity values and foxy and muscat f lavours obtained from the sensory evaluation of 102 samples.The heat maps indicate the mean sensory evaluation values of the nine types of f lavour (n = 4).The samples are indicated by a code of abbreviated cultivar name (Table1) followed by harvest year (e.g., FR17 = Fry harvested in 2017).Table1gives the reference f lavour type. 1 and 2 denote samples harvested from different trees in the same year.Lower (higher) sensory evaluation scores in the heat map are presented in green (red), as shown on the scale.

Figure 2 .
Figure 2. PCA of volatile compounds obtained from 102 grape samples.The colour of each sample represents its classification based on the f lavour characteristics shown in Fig. 1. 2A-2C: Scatter plots of the PCA scores of all the samples.2D-2F: The corresponding loading plots indicating the relative importance of the variables.Sample codes: cultivar, harvest year, and reference f lavour (Table1).Samples denoted 1 and 2 were harvested from different trees in the same year.Compound code: compound type and compound number (Table2).

Figure 3 .
Figure 3. Cluster analysis of 102 grape samples and quantified volatile compounds.The contents of 98 volatile compounds were quantified as 3-heptanol equivalents, log-transformed, and plotted in a heat map, in which lower (higher) volatile compound contents are presented in green (red), as shown on the scale.The samples are indicated by a code of abbreviated cultivar name (Table1) followed by harvest year (e.g., FR17 = Fry harvested in 2017).1 and 2 denote samples harvested from different trees in the same year.The cluster classification of the samples is based on the classification by sensory evaluation in Fig.1.

− 1 FW
) of volatile compounds between clusters determined based on Fig.1 5 and positive contribution, +++: VIP > 2.0 and positive contribution, −: VIP > 1 and negative contribution, -: VIP > 1.5 and negative contribution, -: VIP > 2.0 and negative contribution Positive and negative contributions are shaded in pink and blue, respectively, with the intensity of the colour indicating the value of VIP, and VIP < 1.0 is indicated by a white background.Bold text in (A) indicates the active volatile compounds (OAVs >1) identified in cluster 1. Bold text in (B) indicates the active volatile compounds (OAVs >1) identified in clusters 5 or 6.Bold text in (C) indicates the active volatile compounds (OAVs >1) identified in any of clusters 1-7.The respective active volatile compounds were obtained from Supplemental Table

Table 1 .
Variety and f lavour characteristics of 38 table grape cultivars based on previous reports and databases −1 FW (methyl salicylate) to 591 μg•kg −1 FW (phenylethyl alcohol), as shown in Table2.

Table 2 .
Variation in volatile compounds within table grape

cultivars Codes Compounds RI a ID b Content μg•kg −1 FW (range) c IQR d DFR e
(Continued)

Table 2 .
Continued a Retention indices were determined using a DB-WAX UI column.b Identification of volatile compounds: A, MS and RI were consistent with standards; B, MS and RI were consistent with MS database and literature data; C, identified by MS and RI agreed with MS database.c Mean volatile concentrations (μg•kg −1 FW) and their distribution ranges (in parentheses) in grape berries.d Interquartile range (μg•kg −1 FW).'-' indicates that the majority of compounds were not detected in the samples and have a value of zero.e Detection frequency of compounds in all samples (%).

Table 3
. No significant difference was observed in the total content of clusters 1-6 (p > 0.05), with cluster 2 having the lowest total volatile content, and cluster 6 having the highest

Table 4 .
Partial least squares (PLS) regression model summary for each f lavour All flavour categories were regressed on the log-transformed quantitative data of 98 volatile compounds in 102 samples of fresh grapes using PLS. a Number of factors is the number that provides the best predictive performance on the validation set.b Cumulative R 2 X and Cumulative R 2 Y report the percentage of X and Y variations explained by the model with the given number of factors.c Cumulative Q 2 is an indicator of the predictive ability of models with the given number of factors or fewer.d Number of compounds (VIP > 1) indicates the number of compounds that are considered important for prediction.VIP = importance of the variable for projection.

Table 5 .
The 10 compounds with the highest VIP values in (A) muscat, (B) foxy, and (C) f lavour intensity based on PLS analysis, and those compounds considered important for foxy and muscat f lavour based on previous studies