Key-Marker Volatile Compounds in Aromatic Rice (Oryza sativa) Grains: An HS-SPME Extraction Method Combined with GC×GC-TOFMS

The aroma of rice essentially contributes to the quality of rice grains. For some varieties, their aroma properties really drive consumer preferences. In this paper, using a dynamic headspace solid-phase microextraction (HS-SPME) system coupled to a two-dimensional gas chromatography (GC×GC) using a time-of-flight mass spectrometric detector (TOFMS) and multivariate analysis, the volatile compounds of aromatic and non-aromatic rice grains were contrasted to define some chemical markers. Fifty-one volatile compounds were selected for principal component analysis resulting in eight key-marker volatile compounds (i.e., pentanal, hexanal, 2-pentyl-furan, 2,4-nonadienal, pyridine, 1-octen-3-ol and (E)-2-octenal) as responsible for the differences between aromatic and non-aromatic rice varieties. The factors that are most likely to affect the HS-SPME efficiency for the aforementioned key-marker compounds were evaluated using a 2III5−2 fractional factorial design in conjunction with multi-response optimisation. The method precision values, expressed as % of coefficient of variation (CV), were ranging from 1.91% to 26.90% for repeatability (n = 9) and 7.32% to 37.36% for intermediate precision (n = 3 × 3). Furthermore, the method was successfully applied to evaluate the volatile compounds of rice varieties from some Asian countries.


Introduction
Indonesia is the world's third-largest rice producer in addition to one of the world's major rice consumers [1]. Within this region, rice dominates not only food security but also national economies. Rice has been cultivated in Indonesia from the time between 2000 and 1400 B.C., while the production has considerably increased since 1925, thereby giving rise to a number of rice varieties. There are two groups of the grains based on their aroma (i.e., aromatic and non-aromatic) [2].
Some rice varieties are known as aromatic rice. They contain some typical volatile compounds released from the grain that discriminate these rice varieties from the ordinary ones [3]. These varieties have become more widely appreciated in the current market for their specific aroma properties in addition to their appearance and taste. Since the grain aroma is a primary sensory attribute of Henceforth, to achieve the aforementioned objective of developing the optimised HS-SMPE conditions for key-marker volatile compounds in rice, FFD in conjunction with MRO was used in this study.
The essential objective of this particular research is the identification of marker compounds, indicating the existence of quality features sought after for the studied rice samples. Therefore, among the volatiles identified by GC×GC-TOFMS, fifty-one odour-active compounds were selected since the compounds were known to contribute to the unique flavour of a cross-section of rice cultivars [27,[37][38][39][40], besides having a variability of the levels in the tested rice samples. These compounds were then further studied.
PCA was performed on the data of odour-active compounds concentration in aromatic and non-aromatic rice varieties, to assess the possibility of defining the key-marker compounds in aromatic grains. From the analysis, five components were extracted due to having eigenvalues ≥ 1.0 that account for 99.99% of the variability in the original data.
Meant for appropriate assessment of the regression analysis, a biplot of correlation loadings is preferable to conventional loading plots, as it provides easier interpretation of the relationships between volatile compounds and rice varieties ( Figure 1). The technique described here permits an effective tool to define the key-marker compounds of Indonesian aromatic rice varieties.
Meant for appropriate assessment of the regression analysis, a biplot of correlation loadings is preferable to conventional loading plots, as it provides easier interpretation of the relationships between volatile compounds and rice varieties (Figure 1). The technique described here permits an effective tool to define the key-marker compounds of Indonesian aromatic rice varieties.  Table 1).
The PCA 3D biplot accounted for 83.98% of the total variance, with principal component 1 (PC1), PC2, and PC3 explaining 48.07%, 24.09% and 11.82%, respectively. The six rice varieties were alienated revealing the probability of distinctive volatile compounds profiles ( Figure 1). The group of non-aromatic rice varieties was plotted on the positive axis of PC3, while aromatic varieties were laid on the opposite coordinate along the PC3 axis.
The scent of both aromatic and non-aromatic rice involved the combination of odour-active compounds [23,31]. In aromatic rice, two compounds in negative PC3, 2-acetyl-1-pyrroline (C13) and 2,4-Nonadienal (C43), were considered remarkably essential. Particularly, 2-acetyl-1-pyrroline occurred in relatively lower concentration compared with other volatile compounds, but it is presented in aromatic rice varieties with different levels.
Eucalyptol (C27), linalool (C37), and 1H-indole (C49) were much more dominant in nonaromatic than in aromatic cultivars (positive PC3). The relative content of linalool (C37) has been reported to be increased with drought stress [41] as a result of quality improvement for some nonaromatic cultivars. The compounds with a value near zero in PC3, such as 2-butylfuran (C10), guaiacol (C33), o-cymene (C38) and trans-2-nonenal (C39), did not produce clear distinctions between aromatic and non-aromatic rice varieties due to the similar levels of these compounds in the grains.
Nonetheless, in regards to the PC2 axis, the non-aromatic rice varieties can be separated. IR64 (positive PC2) can be noticeably discriminated to the C4 varieties (negative PC2). The non-aromatic rice samples studied here were developed in a major advance in rice production, as it provided higher yield potential for their specific land assignments. IR64, also known as Sentra Ramos, is the most common rice in the Indonesian market attributable to its massive production within the region. In contrary, C4 Raja and C4 Dewi Sri are only produced in extreme land, as the plant was designed to  Table 1).
The PCA 3D biplot accounted for 83.98% of the total variance, with principal component 1 (PC1), PC2, and PC3 explaining 48.07%, 24.09% and 11.82%, respectively. The six rice varieties were alienated revealing the probability of distinctive volatile compounds profiles (Figure 1). The group of non-aromatic rice varieties was plotted on the positive axis of PC3, while aromatic varieties were laid on the opposite coordinate along the PC3 axis.
The scent of both aromatic and non-aromatic rice involved the combination of odour-active compounds [23,31]. In aromatic rice, two compounds in negative PC3, 2-acetyl-1-pyrroline (C13) and 2,4-Nonadienal (C43), were considered remarkably essential. Particularly, 2-acetyl-1-pyrroline occurred in relatively lower concentration compared with other volatile compounds, but it is presented in aromatic rice varieties with different levels.
Eucalyptol (C27), linalool (C37), and 1H-indole (C49) were much more dominant in non-aromatic than in aromatic cultivars (positive PC3). The relative content of linalool (C37) has been reported to be increased with drought stress [41] as a result of quality improvement for some non-aromatic cultivars. The compounds with a value near zero in PC3, such as 2-butylfuran (C10), guaiacol (C33), o-cymene (C38) and trans-2-nonenal (C39), did not produce clear distinctions between aromatic and non-aromatic rice varieties due to the similar levels of these compounds in the grains.
Nonetheless, in regards to the PC2 axis, the non-aromatic rice varieties can be separated. IR64 (positive PC2) can be noticeably discriminated to the C4 varieties (negative PC2). The non-aromatic rice samples studied here were developed in a major advance in rice production, as it provided higher yield potential for their specific land assignments. IR64, also known as Sentra Ramos, is the most common rice in the Indonesian market attributable to its massive production within the region. In contrary, C4 Raja and C4 Dewi Sri are only produced in extreme land, as the plant was designed to adapt to the heat and drought in some regions [42]. This fact may explain the distinctive aroma profile of these varieties with the other non-aromatic variety, viz., IR64.
Likewise, PC2 also distinguished the rice within the aromatic group. Additionally, specific volatile compounds characterised specific aromatic rice varieties. Mentik Wangi was principally explained by 2-acetyl-1-pyrroline (C13), while pentanal (C1) largely described Pandan Wangi. In contrast, Rojolele is depicted by more than one volatile compound and emanates a stronger aroma than other aromatic rice. It is; therefore, recognised as an elite grain in the Indonesian rice market.
In addition to being considered as aromatic rice, together with Mentik Wangi, Pandan Wangi is described as a round-shaped and relatively thick grain [43]. Rojolele rice is characterised by long slender grains with a high elongation ratio. The differences in physical characteristics endorse some expectations of discrepancies in chemical markers.

Optimisation of HS-SPME for the Key-Markers in Aromatic Rice
The variables that were likely to influence the extraction of key-marker compounds from aromatic rice were optimised. The factors considered were the amount of the sample (x 1 ), the volume of water (x 2 ), adsorption temperature (x 3 ), pre-incubation time (x 4 ), and adsorption time (x 5 ). Based on the experimental design generated by the 2 5−2 III FFD with two centre points, 11 extraction processes were completed to extract the key-marker compounds from rice ( Table 2). The response for each extraction in the experimental design generated by the 2 5−2 III FFD was calculated and expressed as the value relative to the maximum yield obtained (%) for the individual level of key-marker aroma compounds in rice (i.e., pentanal (C1), pyridine (C3), hexanal (C6), 2-acetyl-1-pyrroline (C13), 1-octen-3-ol (C19), 2-pentylfuran (C22), (E)-2-octenal (C28) and 2,4-nonadienal (C43)). The responses were simultaneously optimized using MRO, wherein the optimization target for each response was considered equivalently important. The importance of the responses for computational analysis was indicated by the impact coefficient given to the responses in the MRO. By default, values of the impact coefficients were set to three (STATGRAPHICS Centurion XVI, Warrenton, VA, USA) with medium sensitivity.
Prior to MRO, the response surface methodology (RSM) data were formerly analysed to generate a model for each response separately. The efficiency of the model was checked by ANOVA and the suitability of the model was judged by considering coefficient of determination (R 2 ). The values of the R 2 statistic ranged from 68.05% (2PF) to 95.96% (OCA). Henceforth, the RSM for each response was confirmed to provide a high degree of correlation between the experimental and predicted values.
As the response surface equation constructed by the software for each response was plotted, the model provides the variable effects on the response over the studied range of the 2 5−2 III FFD. Subsequently, the desirability function d(y) was then constructed based on the values obtained for each optimized response. The MRO approach assumes the response values equal to (y) can be modelled through the d(y), where the desirability ranges from di(ŷi) = 0 for an undesirable response and di(ŷi) = 1 represents a completely desirable value. The target optimization defined by MRO was to maximize the HS-SPME recovery (100% extraction yield) of each key-marker aroma compound simultaneously. To obtain these optimum values, the d(y) was plotted as a 3D contour plot, which illustrated the optimum point of the simultaneous optimization ( Figure S2 Supplementary Material).
The proposed ordinates and optimal conditions for HS-SPME by MRO were as follows: Amount of the sample (x 1 , −1.00, 0.5 g), volume of water (x 2 , 1.00, 5 mL), adsorption temperature (x 3 , 0.36, 80.73 • C), pre-incubation time (x 4 , −1.00, 5 min), and adsorption time (x 5 , 1.00, 50 min). Because the value for adsorption time was in the corner of the studied range for this extraction variable, it was decided to study values above the highest assayed level.
The results of extraction yields by different adsorption times are shown in Figure S3 in Supplementary Material. A single-factor ANOVA was used to evaluate the significance of adsorption time in the extraction yield. The adsorption time of 70 min was found to have a significant effect on the extraction yield because the Fcalculated for adsorption time (5.21) was higher than Fcritical (2.84). A longer extraction time results in a decrease of the extracted compounds, attributable to a longer process, and applying relatively high temperature may ruin the stability of the target compounds. As a result, 70 min was defined as the optimum adsorption time.

Method Validation of HS-SPME GC×GC-TOFMS
The analytical procedure for the extraction of volatile compounds was validated according to the recommendations of ISO 17,025 and the International Council for Harmonisation (ICH) Guideline Q2 (R1) [44,45]. Under the optimum experimental conditions, the validation of the proposed HS-SPME GC×GC-TOFMS method involving HS-SPME followed by GC×GC-TOFMS was accomplished.
The precision of the method was evaluated by assessing repeatability (intra-day) and intermediate precision (extra-day). Precision was expressed as the coefficient of variation (CV). The method precision values, expressed as % CV, of the developed method ranged from 1.91% (2PF) to 26.90% (PYR) for repeatability (n = 9), and 7.32% (OCA) to 37.36% (PEN) for intermediate precision (n = 3 × 3). The result confirmed that acceptable precision for the extraction method had been achieved.
A certified reference material was not available for the studied compounds in rice matrices; consequently, definitive statements cannot be made with regard to accuracy. Nonetheless, the extraction recovery (%R) was determined after evaluating the results from spiked rice samples with standards. The recoveries related to the spiked standards on rice samples ranged from 78.79% (2PF) to 96.86% (OCT). These results show that the developed extraction method is applicable for the assessment of studied volatile compounds.

Real Rice Samples Application of HS-SPME
To evaluate the efficiency of the proposed method in real samples, the developed HS-SPME was applied to assay the key-marker volatile compounds in several aromatic rice samples, including aromatic rice from Indonesia (Pandan Wangi and Mentik Wangi), India (Basmati) and Thailand (Jasmine). Volatile profiles were obtained from these samples, then compared in order to establish differences. The results of real sample application experiments is shown in Figure 2. The four tested rice samples are considered as aromatic rice varieties in the national and international market [46,47]. Pandan wangi and Basmati had the highest proportion of 2-acetyl-1pyrroline, whilst hexanal and 2-pentylfuran were the most prominent volatile compounds for Jasmine and Mentik Wangi. The different levels of key-marker volatile compounds in aromatic rice samples could be due to different regions for cultivation [48]. The four tested rice samples are considered as aromatic rice varieties in the national and international market [46,47]. Pandan wangi and Basmati had the highest proportion of 2-acetyl-1pyrroline, whilst hexanal and 2-pentylfuran were the most prominent volatile compounds for Jasmine and Mentik Wangi. The different levels of key-marker volatile compounds in aromatic rice samples could be due to different regions for cultivation [48].
Since 1983, 2-acetyl-1pyrroline is regarded as the solely most important compound in rice, especially fragrant or aromatic rice [7]. However, it was not the case for Kao Dok Mali 105 or the so-called Thai Jasmine rice and Mentik Wangi. Apart from 2-acetyl-1-pyrroline, other key-marker volatile compounds were also counted as important compounds that affect the quality of aromatic rice, including hexanal and 2-pentylfuran. The result also disclosed that Jasmine rice has a markedly higher amount of key-marker compounds compared with other tested aromatic rice samples.

Natural Source of 2-Acetyl-1-Pyrroline
There is not a commercially available standard for this compound. Therefore, Pandan (Pandanus amaryllifolius) leaf was selected as a natural source of 2-acetyl-1-pyrroline (2AP) as the abundant amount of this compound in the leaves has been previously described [10,13,49]. Fresh Pandan leaves were acquired from a local supplier in Yogyakarta, Indonesia. The leaves were cut into pieces ±1 mm in size and stored in a sealed vial at 4 • C. The identity of 2AP in Pandan leaves was confirmed by HS-SPME GC×GC-TOFMS using the NIST 2011 mass spectral library ( Figure S1 Supplementary Material). It was used only for identification purposes.

Rice Grains and Sample Preparation
In the initial study, three non-aromatic rice (IR64, C4 Raja and C4 Dewi Sri) were used to contrast with three aromatic varieties (Rojolele, Mentik Wangi and Pandan Wangi) to define the key-marker volatile compounds in grain [46]. The samples used in this study were fully polished grains of the white rice variety. The rice sample (2.5 g) and Milli-Q water (5 mL) was placed in a 15 mL vial, which was then tightly capped with an open top closure with PTFE/silicone septa.
An aromatic rice variety of Pandan Wangi was selected for the study to develop an optimised extraction method of key-marker compounds. Subsequently, the final extraction method was applied to a number of aromatic rice products available in the international market (Basmati and Jasmine) and the Indonesian national market (Rojolele, Pandan Wangi and Mentik Wangi) from a different region of origins in Java Island. Several samples (IR64, C4 Raja, C4 Dewi Sri, Rojolele, Mentik Wangi and Pandan Wangi) were acquired from a smallholder rice distributor in the Central Java area, Indonesia. These samples were harvested no more than 6 months before being used. Some samples (Basmati and Jasmine) were obtained from a commercial market in Spain, no information about the harvest period was found about these samples. A rice sample (20 g) was placed in a plastic cylinder and the rice grains were milled with an Ultraturrax homogenizer (IKA ® T25 Digital, Staufen, Germany) for 10 min prior to extraction. Every 1 min, the milling process was stopped to avoid excessive heating of the sample. The fine powder of rice grain was then homogenized by stirring and the sample was stored in a closed container in a refrigerator before being used for analysis. Samples were analysed over a period of two weeks.

Headspace Solid-Phase Microextraction (HS-SPME)
Volatile compounds from the rice samples were extracted using a dynamic headspace solid-phase microextraction (HS-SPME) attached with divinylbenzene/carboxen/polydimethylsiloxane (DVB/CAR/PDMS) StableFlex fibre of 50/30 µm thickness and 2 cm length (Sigma-Aldrich, Saint Louis, MO, USA). According to the experimental design, rice grains were accurately weighed at either 0.5, 1.5 or 2.5 g and Milli-Q water was loaded at either 0, 2.5 or 5.0 mL into a 15 mL screw top vial, then 100 µL of aqueous solution containing 5 ng of 2,4,6-trimethylpyridine (TMP) (Sigma-Aldrich, Saint Louis, MO, USA) as the internal standard was added and the vial was sealed with PTFE/silicone septa. The HS-SPME was carried out according to the design of experiment (DOE), varying the extraction factors of equilibration time (5-15 min), adsorption temperature (40-100 • C) and adsorption time (10-50 min). Thermal desorption of the analytes from the SPME fibre was done at 250 • C. Before starting the extraction, 0.1 mL of TMP standard solution was added into the sample. Every peak area in the chromatograms were standardized by the resulting area for the TMP peak. The volatile compounds were separated using the following temperature gradient program for the primary GC oven: Initial temperature of 40 • C maintained for 1 min, then ramped at 8 • C/min to 250 • C, and finally kept for 10 min. The temperature program for the secondary GC oven was with the shift of +40 • C according to the program of primary GC oven. The total analysis time was 37 min. The injector was carried out in splitless mode at 250 • C. Helium was used as the carrier gas at a constant flow of 1.0 mL/min. The temperatures for the transfer line and ion source were maintained at 250 • C. The detector voltage was set to 1600 V. Ions in the m/z 40-500 range were analysed with a data acquisition rate of 125 spectra/s.

Experimental Design and Optimisation
The effect of the tested independent factors on the response within the studied range was evaluated by performing a fractional factorial design (FFD) (i.e., a 2 5−2 (quarter fraction) with two central points of analysis). The extraction factors included in the design were amount of the sample (x 1 , 0.5-2.5 g), volume of water (x 2 , 0-5 mL), adsorption temperature (x 3 , 40-100 • C), pre-incubation time (x 4 , 5-15 min), and adsorption time (x 5 , 10-50 min). Since the variables have different units and ranges, each of the variables was first normalised and forced to range from −1 to +1 in order to obtain a more even response. Therefore, the factor levels were denoted as −1 (low), 0 (central point) and +1 (high) according to the following equation: where x i is the coded value of the factor x i , x 0 is the value of x at the centre point, and ∆x is the increment of x i corresponding to a variation per unit of x i . The factors included in the design are shown in Table 3 along with their respective levels.
The design of experiment (DOE) matrix was established with resolution (R) of III, wherein every main effect is confounded (aliased) with at least one first-order interaction. The 2 5−2 III fractional factorial design allowed the first three variables (x 1 to x 3 ) to be set and thus the DOE was obtained by establishing the full 2 3 factorials as the basic design (with the three factors x 1 , x 2 and x 3 ) and factors x 4 and x 5 were subsequently equated to the x 1 x 2 and x 1 x 3 interactions, respectively. This particular design produced the following defining relationships: The linear model for this fractional factorial design is: where β i (i = 1, 2, ..., 5) is the parameter estimated for the factor i, β ij (i = 1, 2, ..., 5; j = 1, 2, ..., 5) is the parameter estimated for the interaction between variables i and j; x i is the coded form of factor i that influences the response y; and x i is the coded form of factor i that influences the response y. The whole design consisted of 11 runs carried out in random order and these are presented in Table 3.  Principal component analysis (PCA) and multi-response optimisation (MRO) were performed with the trial version of STATGRAPHICS Centurion XVI (Statpoint Technologies, Inc., Warrenton, VA, USA) to define and optimise the key-marker compounds of aromatic rice grains. The experimental results in single factor experiments were analysed using Gnumeric 1.12.17. The analysis of variance (ANOVA) and least significant difference (LSD) test were used to determine the significance of differences between the means.

Conclusions
Eight volatile compounds were found as chemical key-markers for different rice grains varieties using HS-SPME GC×GC-TOFMS and chemometric analysis. These compounds were effectively extracted using HS-SPME under the following optimised conditions: Amount of the sample (0.5 g), volume of water (5 mL), adsorption temperature (80.73 • C), pre-incubation time (5 min), and adsorption time (50 min). The validation of HS-SPME ensured acceptable precision and accuracy of the method. In addition, the method developed based on HS-SPME GC×GC-TOFMS was successfully applied to evaluate the volatile compounds of four aromatic rice varieties, thus considered as a reliable analytical method for the key-marker compounds in rice grains.