Comprehensive analysis of volatile compounds in cold‐pressed safflower seed oil from Xinjiang, China

Abstract Three varieties of safflower seed oil (SSO) from Xinjiang Autonomous Region, China, were analyzed by headspace solid‐phase micro‐extraction gas chromatography coupled with mass spectrometry (HS‐SPME‐GC‐MS) to reveal volatile components. Overall, 67 volatile components were determined and four compounds including isoamyl alcohol, caproic acid, n‐pentanal, and heptanal were newly identified in SSO as aroma‐active components. Meanwhile, 16 compounds were selected by relative odor activity value (ROAV) to evaluate contributions of single compounds to the overall odor (ROAV > 1), in which nonanal, (Z)‐6‐nonenal, and (E)‐2,4‐decadienal were the top three contributed substances (ROAV > 70). The sensory panel was described as eight definition terms (grassy, fruity, almond, mushroom, fatty, sweet, paddy, and overall fragrance). Principal component analysis (PCA) revealed a significant separation of three cultivars with the first principal component (PC‐1) and the second principal component (PC‐2) expressing 73.9% and 23.1%, respectively. Both PCA and ROAV allowed identifying the compounds positively correlated to sensory evaluation.

up to about 360,00 thousand tonnes, and northwest of Xinjiang Autonomous Region is the top growing area with a share of 80% (Hussain, Lyra, Farooq, Nikoloudakis, & Khalid, 2016;Lu et al., 2013). Additionally, SSO in this area also has a highest content of linoleic acid (85.04%) compared with other districts in China (Han, Cheng, Zhang, & Bi, 2009). The three most popular cultivars of safflower (Tacheng HH-4, Bole HH-1, and Changji HH-2) devoted to SSO were collected in different locations from Xinjiang, China (Liu et al., 2009).
Flavor is considered as the key driver of consumer's appreciation and is closely related to both qualitative and quantitative composition of volatile and nonvolatile compounds (Sabatini & Marsilio, 2008). Romero, García-González, Aparicio-Ruiz, and Morales (2015) showed that high potency of some volatile compounds were responsible for sensory defects and different mixing ratios generated different sensations. Erten and Cadwallader (2017) indicated that yield area, processing technology, and environment could be known by analyzing aroma profiling of oil-roasted almonds. In fact, although many complete characterization of sensory profiles have been reported for other oils such as olive oil, peanut oil, and palm kernel oil (Lim et al., 2017;Marina, Che Man, & Amin, 2010;Morales, Luna, & Aparicio, 2000), few investigation of SSO relating sensory attributes to volatile components was carried out, especially the SSO from Xinjiang Autonomous Region, China. Moreover, although SSO from different sources has been characterized for their sensory properties (Baiano, Terracone, Viggiani, & Nobile, 2013), few cultivars have received in-depth investigation of their volatile compounds. To the best of our knowledge, there is only one literature published concerning the sensation description and volatile compounds of SSO from Xinjiang Autonomous Region, China, in which only one type had been detected (Aydeniz, Güneşer, & Yılmaz, 2014). Therefore, it is necessary to conduct a study for remedying this aspect.
Currently, several techniques have continuously investigated to achieve quick and efficient extracting for volatile components including supercritical fluid extraction (SFE), enzymatic oil extraction, and pressurized liquid extraction (Conte et al., 2016;Gibbins, Aksoy, & Ustun, 2012;Han et al., 2009;Hu et al., 2012), among them, headspace solid-phase micro-extraction (HS-SPME) is a common extraction method beneficial for isolation and preconcentration of volatiles with advantages of fast, simple, sensitive, and no solvent, prior to gas chromatographic analysis (Mesquita et al., 2017).
However, high content of volatile compounds not necessarily has a high contribution value; thus, a parameter named "relative odor activity value (ROAV)," which was frequently combined with electronic nose and GC-MS technology to assign the key odor compounds, was widely applied to evaluate the contributions of the single compound to the overall odor (Multari, Vall, Yang, & Suomela, 2018). Sun et al. related key odor compound data of ROAV with principal component analysis (PCA) to reveal a clear classification of aroma profiling of star anise, and the cluster analysis showed a clear consistency with ROAV analysis results (Sun, Chen, Li, Liu, & Zhao, 2014). It is known that in general, the perceived odors in foods are the result of a mixture of odorant; hence, a multivariate analysis is suited to explore the relationship between sensory attributes and volatile compounds (Wang, Zou, Shi, & Shi, 2018). Despite extensive studies on other plants' oil (Sun et al., 2014), no comparative study using ROAV has been published regarding aroma contributors of SSO.
The objectives of the present study were as follows: (a) to study the volatile composition and sensory attributes of different cultivated samples of SSO in Xinjiang Autonomous Region, China, using headspace solid-phase micro-extraction gas chromatography coupled with mass spectrometry (HS-SPME-GC-MS) analysis, respectively, (b) to elucidate the relationship between volatile compounds and sensory attributes by ROAV and identify the compounds that could potentially contribute to odor perception, and (c) to apply PCA for visualizing their cluster trends of SSO of various origins characterized by their volatile profiles.

| Samples
Safflower seeds from the three most popular cultivars (Tacheng HH-4, Bole HH-1, and Changji HH-2) were harvested in early September 2017 from the Xinjiang Autonomous Region, China. The safflower seeds were cold-pressed by oil press (ZY-22A) at temperature of 45°C. The safflower seed oils extracted from the seeds of HH-1, HH-2, and HH-4 safflower varieties were called SSO-1, SSO-2, and SSO-3, respectively. All other chemicals and standards were analytical grade and purchased from Zhengzhou Wanbo Chemical Products Co., Ltd.

| Sensory and quality analysis
The sensory profile is evaluated by well-trained panel of 10 personnel (five females and five males without smoking history) between the age-group of 25-50. They were recruited because of their major role in the implementation of the method for the sensory analysis of SSO and professionally trained by Food College of Shihezi University in accordance with ISO standards. Analysis of safflower seed oil samples is according to the requirements of GB/T22465-2008. Initial draft of descriptors through current references was meticulously discussed under suitable laboratory environment. In total, there were eight descriptions, including grassy, fruity, almond, mushroom, fatty, sweet, paddy, and total acceptance. After training and once the judges agreed to be familiar with the new set of descriptors, the panel was used for this study. Oil samples were served in cups coded with a 3-digit random number in a balanced order for presentation. Each sample was re-evaluated three times, and the code was rescrambled each time.
Samples were scored using a ten-point scale ranging from 1.0 (very slight) to 10.0 (very intense), at intervals of 1.0.

| HS-SPME-GC-MS analysis
Volatile compounds were analyzed by HS-SPME-GC-MS referencing the procedure described in a previous work (Durant et al., 2014;Wei, Zhou, Han, Chen, & Liu, 2018). Ten gram of SSO samples was inserted in a 125-ml SPME vial. Then, the vial was closed with a Teflon/ silicone septum. The fiber coated with polydimethylsiloxane/polydimethylsiloxane/divinylbenzene (PDMS/PDMS/DVB) 50/30 μm fiber (model 57348-U, Supelco Inc.) was inserted through the septum and exposed to the headspace of the SPME vial after conditioning for 2 hr at 250°C. Then, extraction was carried out at 60°C for 15 min.
Volatile compounds adsorbed by the SPME fiber were desorbed in the injector port of the GC for 3 min at 250°C and identified by GC-MS (Trace 2000 DSQ, Finnigan Co.). Separation was achieved on a DB-WAX capillary column (30 m × 0.25 mm × 0.25 μm film thickness, Supelco Co.). The carrier gas was high-purity helium at a constant flow of 0.8 ml/min with the injector temperature of 250°C. Oven temperature programming was as follows: initiated at 40°C, held for 3 min, then rose at 6°C/min to 120°C, and then 120-240°C at 10°C/min, and held there for 8 min. The temperature of transfer line is 250°C. The mass spectra were obtained using a mass selective detector by electronic impact at 70 eV with a scan range from m/z 33 to 373, emission current

| ROAV determination
The relative odor activity value (ROAV) is used to evaluate the contribution of individual compounds to the overall aroma (Cui, Liu, & Li, 2010). The formula of ROAVA is defined as the equation: where C A is the concentration of the compound A in the sample and T A is its odor detection threshold concentration found in the literature.
T MAX and C MAX are the maximum of C A /T A among all the compounds in the sample.
Relative odor activity value ranges from 0 to 100. Volatile compound with ROAV ≥ 1 is considered as key odor compounds, of which >0.1 and smaller than 1 play a embellish role in aroma.

| Multivariate statistical analysis
Cluster analysis (CA) is the unsupervised classification of patterns (feature vectors) into groups, so that individuals within the same group are more similar to each other than those belonging to different groups (Cebi, Dogan, Develioglu, Yayla, & Sagdic, 2017). Principal component analysis (PCA) and hierarchical cluster analysis (HCA) were carried out on the entire data obtained from the HS-SPME-GC-MS in an attempt to identify the similarity of the samples. HCA dendrogram was calculated by Euclidean distance with Ward's method. All these analyses were made by means of the SPSS 8.0 package.

| Overall volatile compounds profile analysis
As shown in Table 1, a significant disparity was found on the grouped chemical families, the major chemical families that were identified TA B L E 1 Volatile compounds identified by HS-SPME-GC-MS in safflower seed oil from different regions proportion of ketones with 2.80% (Figure 1).

| Sensory evaluation
The average scores of sensory descriptive in the spider web form for SSO are shown in Figure 2. Eight panel sensory terms were used   to assess flavor characteristics including grassy, fruity, almond, mushroom, fatty, sweet, paddy, and overall fragrance. As shown in Table 3, by comparing all average score for the sensory evaluation among three SSO samples of different varieties, "grassy," "almond," and "fatty" descriptors were of comparable intensity (and all gained high scores) while the remaining odor profiles were differed significantly (p < .05). It is worth mentioning that despite species different, all SSO samples represented characteristically "grassy" aroma in a highest level (score > 9.7). Besides, SSO-1 was also abundant in intense of "almond" and "fatty" with a score of 9.0 and 9.5, while "mushroom" (8.9) and "paddy" (8.9) were highest in the SSO-2; SSO-3 was characterized by "fatty" aroma (score = 9.1).
It is noteworthy that the terms of SSO-2 and SSO-3 showed obviously higher olfaction scores in "fruity" (8.2), "mushroom" (8.6), and "sweet" aroma (8.4) than SSO-1. Sensory odor map also indicated that except paddy term, all of the sensory scores gave similar intensity values for SSO-2 and SSO-3. SSO-3 represented a best level with score 9.5 in all fragrance.

| Multivariate statistical analysis
The different nature of volatile compounds prompted the need of a reliable analytical method that allowed their proper quantification to explain the method of panel test. In this study, all the volatile compounds extracted by HS-SPME-GC-MS were analyzed by PCA to reduce the dimensionality of multivariate data. The score and loading plot of PCA are shown in Figure 3. 97.0% of variation in the data was explained by both PC-1 (73.9%) and PC-2 (23.1%), respectively (Figure 3a). SSO-2 and SSO-3 specimens were clustered together, while there was clear separation from SSO-1. The major volatile compounds that positively correlated to PC-1 were 2-octene, n-hexanal, acetic acid, and (Z)-2heptenal, and those that positively correlated to PC-2 were n-hexanol, caproic acid (Figure 3b). Such clustering pattern of SSO was also revealed in hierarchical clustering analysis (Figure 4).
Hierarchical cluster analysis was further applied to creates groups of SSO based on their degree of association which reflected by distance between the samples. The results of HCA based on entire data of volatile components extracted by HS-SPME-GC-MS were represented as dendrogram in Figure 4. Euclidean distance with Ward's linkage method was employed, and the distance between the samples was described by Euclidean in y-axis. The SSO samples were divided into two groups, and in the first group, SSO-2 and SSO-3 were included while the second group included SSO-1. SSO-2 and SSO-3 were closed, and these two samples were separated from SSO-1.

| Volatile compounds by HS-SPME-GC-MS analysis
According to HS-SPME-GC-MS analysis, there was a wide variety in proportion and composition of volatile compounds among different varieties of SSO as shown in Table 1. It is also the first time that n-valeraldehyde, isoamyl alcohol, n-pentanal, and heptanal were identified in SSO as aroma-active components. 44, 41, and 47 volatile compounds were detected in SSO-1, SSO-2, and SSO-3, respectively. This result related to cultivar, climate, and different environments of safflower.
One hundred ninety-three safflower accessions from forty countries were assessed for investigating the relationship between patterns of geographic diversity and agro-morphological traits, showing that safflower seed diversity pattern related barely with DNA agricultural variety but geographic location (Khan, Witzke-Ehbrecht, Maass, & Becker, 2008 Aydeniz et al. reported no ether compounds detected in SSO, and this result may be due to the ether residue considering extraction solvent in HS-SPME; in addition, differences in variety, environment, and SPME coating are also considered to be the possibly main cause of conflict (Hashemi, Zohrabi, & Shamsipur, 2018).
All volatile compounds extracted were classified into nine cat- F I G U R E 4 Hierarchical clustering analyses of aroma components of safflower seed oil from different regions also play an important role in composition of aldehydes (Benincasa et al., 2003). Some researches indicated that aldehydes can be used as one of the oil quality management indicators, for example, (Z)-2hexenal could partly explain quality differences in olive oils due to the increasing properties in postharvest storage (Kalua et al., 2007;Zheng et al., 2018). Alcohols also contributed significantly to volatile profile, of which n-hexanol accounts for the largest proportion (4.90%-29.93%).
Sánchez-Ortiz, Pérez, and Sanz (2013) disclosed that alcohols were mainly produced by self-oxidation pathway of fatty acids, while the enzyme route also represented a certain impact on it during the seed storage process. The relatively high level of n-hexanol (4.90%-29.93%) also might be attributed to the activity of the lipid oxygenase and alcohol dehydrogenase enzyme (Sánchez-Ortiz, Bejaoui, Quintero-Flores, Beltrán, 2018). This study provides the first data for comparing volatile compound composition of SSO of Xinjiang Autonomous Region, China. Twelve key aroma components screened by ROAV > 1 from all volatile compounds through HS-SPME-GC-MS were accounted by aldehydes, which were the largest group followed by alcohols (2), furan (1), and alkene compounds (1). Aldehydes, generally degradation products of fatty acids, were the most important key odor compounds with high peak area and low odor threshold (Genovese et al., 2018). Some studies also showed that there were certain correlations between C5-C6 aldehydes and high-quality vegetable oils (Kalua, Bedgood, Bishop, & Prenzler, 2013). It is known that grass, fatty, and fragrance were the major intensive flavors of SSO as shown in Table 2, in accordance with the grassy and fatty taste in the sensory panel, which might largely be related to flavor of certain aldehydes. (Z)-2-hexenal is often considered to be a volatile component in high-quality olive oil with contributions to the characteristic flavor of sweet and vanilla (Veneziani et al., 2018). Nonanal was found at relatively high content (0.53%-1.51%) in three samples, which closely related to high linoleic acid in safflower seed oil. It is likely that in these samples, nonanal provides the highest contribution to the general odor of grass and almond, obtaining the ROAV of 98.23, 100.00, and 100.00, respectively (SSO-1, SSO-2, and SSO-3). (Z)-6-nonenal also was a possible major odor compound providing a flavor of grass and fatty, which had a high contribution to the characteristic flavor with an extreme low odor threshold of 0.08 ppb in SSO-1 and SSO-3. Alcohols also produced an important fraction of volatile compounds of SSO from Xinjiang Autonomous Region.

| Main odor-active compounds
1-Octen-3-ol was previously reported to act as the primary flavor compound in mushroom (Deveci, Tel-Cayan, Emin Duru, & Turkoglu, 2017). 1-Octen-3-ol was found at minor relative concentration; nevertheless, it exhibited strong odor activity of mushroom and sweet having an odor threshold as low as 1 ppb and can influence markedly the aroma of SSO. N-hexanol constituted an abundant content due to its high sensory threshold of 250 ppb and therefore is considered for a lesser extent of modification.
Although only 1 furan and 1 alkene were selected as key aroma-active component with ROAV > 1, they also produced an essential fraction of volatiles. Limonene was a natural monoterpenoid compound. It is widely reported in lemon and orange products and also is major volatile component (ca 52% w/w of the total volatiles) in peel of Libyaone Libyan oranges (Macleod, Macleod, & Subramanian, 1988). Limonene, as the only alkenes selected, might be the only significant source of citrus flavor providing contribution to fruit and orange. As one of the heterocyclic compounds, furaneol has been identified previously as a product of the Maillard reaction (Zhang, Wang, Yuan, Yang, & Liu, 2016). 2-N-pentylfuran was perceived with a favorable paddy-like note and has been intensively demonstrated to be the primary flavor compound in cereals (Multari et al., 2018).
In a previous study about SSO, 2-n-pentylfuran was regarded as a key odor-active compound and its role of the flavor has only been identified after seed roasting while cold-pressed sample had none of it (Aydeniz et al., 2014). However, in this study, 2-n-pentylfuran was firstly indicated as the major odorant of cold-pressed SSO of SSO-2 variety. This was consistent with sensory panel.
Besides the components shown in Table 2, some potential odorants might also necessary to aroma profile but not detected by HS-SPME-GC-MS. Though they represented at both low levels of threshold and concentration, these compounds may construct the delicate background aromas through interactions between each other (Yang et al., 2009). Thus, a further identification was needed to explore whether and how they contributed to the flavor.

| Sensory evaluation of SSO
To obtain more intuitive sense result, a sensory evaluation was applied to reveal the further relationship between volatile components and sensory descriptors of different SSO (Figure 2). Apparently, all three samples were found to exhibit relative higher score in grass and fatty aroma than other sensory profiles, which could be regarded as basic flavor of SSO. In a previous study, Sakač et al. reported that aldehydes were generally described as grass and fatty aroma (Sakač et al., 2016). Most aldehydes including hexanal, (Z)-6-nonenal, (Z)-2,4-decadienal, and 2,4-decadienal all showed positive impact on the contribution to grass and fatty aroma and made these two aroma profiles greatly strengthened. The almond odor was another basic flavor that had similar intensity among three samples, which perhaps is associated mainly with nonanal. Beltrán, Ramos, Grané, Martín, and Garrigós (2011)

monitored the oxidative process of Spanish and
American almond oils and indicated nonanal as a volatile compound that could be obtained in all samples.
SSO-1 presented a statistically lower intensity (p < .05) of fruity odor (score = 6.1) than other two samples (8.1 and 8.2, respectively) as shown in Table 3. This result was probably correlated with lower content of limonene and octanal in SSO-1; in a previous study, limonene and octanal also were reported widely in citrus fruit as major for mustiness-humidity, which is quite like the mushroom descriptor in our study. SSO-3 gain highest score with 9.5 of the overall fragrance, which are responsible for the mixture of key aroma-active compounds.

| CON CLUS IONS
A comprehensive characterization of the aroma profile of SSO from Xinjiang Autonomous Region, China, was analyzed by HS-SPME-GC-MS. There were 67 volatile substances tentatively identified in three samples. ROAV analysis further indicated that the flavor of SSO was more correlated with 12 aldehydes, two alcohols, one alkene, and one heterocyclic compounds according to cultivated varieties. Meanwhile, limonene, octanal, 2-octenal, and 1-octen-3-ol were major volatile compounds providing fruity, mushroom, and sweet aroma. The sensory variance analysis indicated that the odor profiles differed significantly with regard to the "fruity," "mushroom," "sweet," and "paddy" aroma (p < .05), which largely in accordance with the key odor compounds. In addition, these aroma descriptions of SSO can play an essential role in the possibility of using volatile constituents to identify the source of safflower varieties. The result of PCA revealed a clear discrimination and demonstrated that PCA was a potential and useful tool for SSO primary evaluation of category similarity.

CO N FLI C T O F I NTE R E S T
There are no conflicts of interest.

E TH I C A L A PPROVA L
This study does not involve any human or animal testing.