2019 年 25 巻 6 号 p. 775-784
Japanese sake reportedly contains several D-amino acids that could contribute to its characteristic flavor. In this study, component profiling of Japanese sake using combined data of gas chromatography/mass spectrometry (GC/MS)-based metabolomics and liquid chromatography/mass spectrometry (LC/MS)-based enantioselective amino acid analysis, and orthogonal partial least squares (OPLS) regression analysis was conducted. GC/MS-based metabolomics is typically the first choice in constructing taste prediction models because it can perform a comprehensive analysis of hydrophilic compounds to obtain informative, explanatory variables. Here, enantioselective amino acid profiles obtained by LC/MS were also used as explanatory variables instead of non-enantioselective amino acid data, which are part of the GC/MS results. The evaluation of prediction models by linearity (R2), predictability (Q2), and root mean square error estimation (RMSEE) indicated that enantioselective amino acid profiles improved metabolomics-based sensory prediction.
Japanese sake, a traditional Japanese alcoholic beverage, is brewed from rice and water via a fermentation process involving the microorganisms koji-mold (Aspergillus oryzae) and sake-yeast (Saccharomyces cerevisiae) (Steinkraus, 2002). Starch in the rice is hydrolyzed by enzymes produced by the koji-mold, such as amylase and glucoamylase, to yield sugars. The resultant sugars are subsequently consumed as a carbon source by the yeast. Hydrolysis of starch and sugar consumption proceed in parallel to prevent an extreme rise in osmotic pressure. Finally, rice is converted to ethanol, amino acids, organic acids, and aromatic compounds, among others. The component profile produced by microorganism metabolism is important for producing the particular flavors of Japanese sake.
Much attention has been paid to the use of metabolomics, omics study of the metabolome, to investigate the relationship between component profile and sensory score (Ochi et al., 2012; Yamamoto et al., 2012; Shiga et al., 2014). This technology enables association of the differences in component profiles with sensory scores of food, and reveals the presence and magnitude of the correlation between each compound and the sensory score. Furthermore, it has been previously demonstrated that component profiling and multivariable analysis are useful for sensory studies in Japanese sake (Sugimoto et al., 2010; Mimura et al., 2014). Generally, mass spectrometry (MS) is used in metabolomics studies as a detector because it enables high-resolution separation via the differences of mass to charge ratio. Furthermore, liquid chromatography (LC), gas chromatography (GC), and capillary electrophoresis (CE) are also used prior to MS to achieve higher separability. Among these methods, GC/MS is the most frequently used analytical method. This is because, in GC/MS, the repeatability of retention time is high and the mass spectrum is specific for each compound due to the electron ionization (EI) method, making the identification of compounds comparatively easy. Although GC/MS can only analyze volatiles, the derivatization procedure before GC/MS analysis to induce volatility and heat stability enables analysis of hydrophilic compounds. Therefore, GC/MS-based non-target metabolomics can be used to target hydrophilic compounds such as sugars, organic acids, and amino acids, which are thought to be taste substances.
Specifically, amino acids are of interest in food component analysis because they are known to be important taste-stimulating substances. Moreover, profiling of amino acid concentrations and subsequent multivariate analysis have demonstrated that amino acids contribute to the flavor of food. As a result, the function of amino acids as food-flavoring agents has been attributed to the naturally abundant L-forms without any rational proof. However, recently, several D-amino acids have been detected in some foods due to improvements in the enantioselective amino acid profiling method (Brückner and Hausch, 1989). Gogami et al. developed a method to analyze the enantiomers of proteinogenic amino acids using a liquid chromatography-fluorescence detector (LC-FLD), which revealed the presence of D-amino acids in Japanese sake (Gogami et al., 2011). Comparatively high amounts of D-amino acids have been found in ‘kimoto-sake’, in which lactic acid bacteria contribute significantly to lactic acid production in the initial stages of ‘seed mash’ preparation. Further, D-amino acid racemase genes were identified from Lactobacillus sakei and Lactobacillus mesenteroides isolated from ‘sake seed mash’ (Kato and Oikawa, 2017a; Kato and Oikawa, 2017b). Moreover, some D-amino acids were reported to play a role as taste substances in Japanese sake (Okada et al., 2013). Thus far, the profiling of amino acids as food flavoring agents has been conducted without conventional considerat ion of amino acid chiral i ty. However, enantioselectivity may be important for amino acid profiling in foods. In order to construct a good prediction model, without the problem of overfitting, via multivariate analysis using component profile data as explanatory variables and sensory score as the response variable, the diversity of food samples and analytically targeted compounds must also be considered. Conventional enantioselective amino acid analytical methods are beset with disadvantages associated with throughput, resolution, the accuracy of compound identification, and the variety of analytes (Tanwar, 2015). In our previous study, a novel enantioselective amino acid analytical method was developed using liquid chromatography-time of flight mass spectrometry (LC-TOF/MS) (Konya et al., 2017). In this method, enantioseparation was performed via the chiral column CROWNPAK CR-I. In the column, optically active BINAP derivatives bound to crown ether are installed as the chiral selector. Chirality of alpha amino carbons affects the preference of primary amino acid retention to the crown ether motif by steric clash derived from BINAP. Based on the aforementioned principle, chiral primary amines can be enantioselectively separated. We employed high resolution TOF/MS to separate co-eluted compounds with the same nominal molecular mass for realization of simultaneous analysis of various targets. A previous study indicated that approximately 100 types of amino acids and amines could be analyzed within 10 min by distinguishing their enantiomers. Therefore, this analytical system was expected to be effective for sensory studies using multivariate analysis. In the present study, to construct the sensory prediction model, the combination of component profiles obtained from both GC/MS-based non-target metabolomics of hydrophilic compounds and LC/MS-based enantioselective amino acid analysis was used as explanatory variables. The present study is the first report of the use of sensory scores of foods with D-amino acid profiles as explanatory variables for sensory prediction.
Sake samples Fifteen types of commercially available Japanese sake were used. The sample information is listed in Table 1.
| No. | Designation | Region | Feature of brewing method | Rice-polish rate (%) | OD430 |
|---|---|---|---|---|---|
| 1 | Originary | Nara | The use of sake instead of a apart of water in moromi brewing | 70 | 0.25 |
| 2 | Originary | Toyama | 0.17 | ||
| 3 | Originary | Hyogo | 0.17 | ||
| 4 | Originary | Niigata | 60 | 0.08 | |
| 5 | Originary | Fukushima | 65 | 0.15 | |
| 6 | Originary | Hiroshima | 65 | 2.47 | |
| 7 | Originary | Hyogo | The use of less amount of water in moromi brewing | 88 | 0.38 |
| 8 | Junmai | Hokkaido | 65 | 0.08 | |
| 9 | Originary | Kyoto | 88 | 0.40 | |
| 10 | Originary | Hyogo | The use of more amount of rice | 0.01 | |
| 11 | Originary | Miyagi | The use of ethanol instead of a apart of water in moromi brewing | 65 | 0.71 |
| 12 | Originary | Akita | Unknown | 70 | 0.69 |
| 13 | Junmai | Kyoto | 68 | 0.06 | |
| 14 | Originary | Nagano | 60 | 0.01 | |
| 15 | Junmai | Niigata | 70 | 0.01 |
Blank column: Unknown information
LC-MS analysis
Amino acids and amines DL-2,4-Diaminobutyric acid dihydrochloride, DL-2-aminopimelic acid, DL-alanine, DL-arginine hydrochloride, DL-asparagine monohydrate, DL-aspartic acid, DL-citrulline, DL-cysteine hydrochloride monohydrate, cystine (mixture of DL- and meso-form), DL-glutamic acid, DL-glutamine, DL-histidine, DL-homophenylalanine, DL-homoserine, DL-isoleucine (mixture of DL- and DL-allo-form), DL-leucine, DL-lysine monohydrochloride, DL-methionine, N-acetyl-DL-alanine, DL-norvaline, DL-ornithine monohydrochloride, DL-phenylalanine, DL-proline, DL-serine, DL-threonine, DL-tryptophan, DL-tyrosine, and DL-valine were purchased from Tokyo Chemical Industry Co., Ltd. (Tokyo, Japan). DL-α-Aminobutyric acid, β-alanine, γ-aminobutyric acid, anthranilic acid, betaine, creatine, cytosine, N,N-dimethylglycine, dopamine hydrochloride, glycine, guanine, histamine, nicotinamide, hypotaurine, nicotinic acid, sarcosine, thymidine, thymine, and tryptamine were purchased from FUJIFILM Wako Pure Chemical Industries, Ltd. (Osaka, Japan). DL-Ethionine was purchased from Acros Organics (Geel, Belgium). Adenine was purchased from Kishida Chemical Co., Ltd. (Osaka, Japan). Uracil was purchased from Nacalai Tesque, Inc. (Kyoto, Japan). Creatinine, cystathionine (mixture of DL- and DL-allo-form), cytidine, DL-kynurenine, and DL-pipecolic acid were purchased from Sigma-Aldrich Japan K.K. (Tokyo, Japan).
Reagents for sample pretreatment and mobile phase Ultrapure water was obtained using GENPURE (Thermo Scientific, Osaka, Japan). Ethanol for HPLC, 0.1 mol/L hydrochloric acid, and trifluoroacetic acid were purchased from FUJIFILM Wako Pure Chemical Industries, Ltd. Methanol- Plus for LC/MS and Acetonitrile-Plus for LC/MS were purchased from Kanto Chemical Co., Inc. (Tokyo, Japan). Chloroform for HPLC was purchased from Kishida Chemical Co., Ltd. DL-Alanine-2,3,3,3-d4 was purchased from Santa Cruz Biotechnology (Dallas, TX, USA).
Instruments Enantioselective amino acid analysis was performed using the Nexera system (pump: LC-30AD, autosampler: SIL-30AC, degasser: DGU-20A5, column oven: CTO-30A, Shimadzu, Kyoto, Japan) connected to a TripleTOF 5600 System (SCIEX, Concord, Canada). Chromatographic separation was performed using an enantioselective column, CROWNPAK CR-I(+) (3.0 mm i.d. × 150 mm, and 5 µm particle size, Daicel CPI, Osaka, Japan). Data acquisition was performed using Analyst 1.6 software (SCIEX).
Sample preparation One hundred microliters of each Japanese sake sample diluted 2-fold with water was mixed with 100 µL of 50%-methanol (a mixture of methanol and water at the rate of 1/1(v/v)), 300 µL of methanol, and 20 µL of 20 µmol/mL d4-DL-alanine dissolved in 50%-methanol as internal standard (IS-solution). After vortexing and centrifugation (16,000 ×g, 4 °C, 10 min), 360 µL of the supernatant was added in a new tube containing 180 µL of water and 360 µL of chloroform. After vortexing and centrifugation (16,000 ×g, 4 °C, 10 min), the supernatant was diluted 5-fold in a solvent mixture of acetonitrile and ethanol (8/2(v/v)). After vortexing and centrifugation (16,000 ×g, 4 °C, 10 min), 100 µL of the supernatant was transferred into a vial for LC, and 1 µL was injected into the LC-MS system.
Standard solution preparation Amino acids and amines were diluted with either 50%-methanol or 50%-methanol including hydrochloric acid, to make a 1000 µmol/mL solution. Subsequently, liquid-liquid extraction was performed on amino acid and amine standard solutions following the same method for sake samples. One hundred microliters of standard solutions was mixed with 100 µL of water, 300 µL of methanol, and 20 µL of IS-solution. After vortexing and centrifugation (16,000 ×g, 4 °C, 10 min), 360 µL of the supernatant was transferred to a new tube containing 180 µL of water and 360 µL of chloroform. After vortexing and centrifugation (16,000 ×g, 4 °C, 10 min), the supernatant was diluted 5-fold in a solvent mixture of acetonitrile and ethanol (8/2(v/v)). After vortexing and centrifugation (16,000 ×g, 4 °C, 10 min), 100 µL of the supernatant was transferred into a vial for LC, and 1 µL was injected into the LC-MS system.
LC condition The mobile phase was acetonitrile/ethanol/water/trifluoroacetic acid (80/15/5/0.5 (v/v/v/v)), the flow rate was 0.4 mL/min, the autosampler temperature and the column oven temperature were maintained at 4 °C and 30 °C, respectively. All analyses were performed under isocratic conditions.
MS condition The TripleTOF 5600 System was used under the following conditions: ionization mode (positive), ion source gas 1 (50 psi), ion source gas 2 (50 psi), curtain gas 1 (30 psi), temperature (600 °C ), ion spray voltage floating (5500 V), declustering potential (60 V), collision energy (5 V), and mass range (60–600 m/z).
Data analysis Data processing was performed using MultiQuant (SCIEX). The extraction chromatogram of each compound was performed with [M+H]+ ± 5 mDa ([M+H]+: molecular weight of hydrogen ion adducted compound). Compound identification was performed by comparing retention time with that of its standard solution. The peak area was normalized against the internal standard. The relative data are summarized in the Supplementary information.
GC-MS analysis
Reagents for sample preparation Ultrapure water was obtained using GENPURE (Thermo Scientific). Pyridine (infinity pure grade) was purchased from FUJIFILM Wako Pure Chemical Industries, Ltd. Methoxyamine hydrochloride was purchased from Sigma Aldrich (Milwaukee, WI, USA). N-Methyl-N-(trimethylsilyl) trifluoroacetamide (MSTFA) and n-alkanes (C8-C40) for annotation were purchased from GL Sciences, Inc. (Tokyo, Japan).
Instruments Non-targeted metabolomics was performed using GCMSQP2010 Ultra coupled to an AOC-20s autosampler and AOC-20i auto injector (Shimadzu). Chromatographic separation was performed using an InertCap 5 MS/NP (0.25 mm i.d. × 30 m, and df = 0.25 µm, GL Sciences). Data acquisition was performed using GCMS solution software (Shimadzu).
Sample preparation The mixture of 10 mg/mL α-ketoglutaric acid dissolved in 50%-methanol as IS-solution and 20 µL of each Japanese sake sample diluted 5-fold with water was lyophilized at 22 °C for 15 h. Thereafter, the derivatization procedure was performed. For the methoxylation reaction, 100 µL of 20 mg/mL methoxyamine hydrochloride in pyridine was added to the freeze-dried sample. The mixture was subsequently incubated in a Thermomixer comfort (Eppendorf Ltd., Hamburg, Germany) at 30 °C for 90 min. For the silylation reaction, 50 µL of MSTFA was added. The mixture was incubated at 37 °C for 30 min. After vortexing and centrifugation (16,000 ×g, 4 °C, 10 min), 100 µL of the supernatant was transferred into a vial for LC, and 1 µL was injected into the GC-MS system.
GC condition The injection temperature was maintained at 230 °C. Helium was flowed at a linear velocity of 39 cm/s as the carrier gas. The column temperature was maintained at 80 °C for 2 min and then raised at 15 °C /min to 330 °C and maintained for 6 min.
MS condition The interface and ion source temperatures were maintained at 250 °C and 200 °C, respectively. Ionization voltage was 70 eV. Detection voltage was automatically determined by auto tuning. Mass spectra were recorded at 10 scans per second over the mass range of m/z 85–500.
Data analysis After converting data to net CDF format, peak finding and alignment were performed using MetAlign (Lommen, 2009). Peak annotation was performed using AIoutput (Tsugawa et al., 2011). The peak area was normalized against the internal standard. The relative data were summarized in the Supplementary information.
Multivariable analysis SIMCA-P+ version 13 (Umetrics, Umeå, Sweden) was used. For data pretreatment, auto scaling was used such that each variable had a unit variance and zero mean.
Sensory evaluation The sensory evaluation experiments were performed by a sake expert assessor, certified by the National Research Institute of Brewing. The criteria for certification involved passing the 5 tests conducted during the sake assessment seminar (1. Recognition of basic tastes and smells; 2. Distinction between the different levels of sourness and sweetness; 3. Ranking flavor intensities; 4. The description of smell and its origin; 5. Descriptive analysis). Quantitative data of 5 sensory attributes, including sweetness, sourness, bitterness, strong taste (Nojun), and aftertaste length, were acquired using a 7-point scaling method (none 0–6 strong). Seventy mL of each sake sample was stored at 22 °C for 1 h prior to evaluation by the assessor. Natural light coming through the window was prevented by a curtain, and only LED light was used as the light source. The sensory evaluation of the 15 samples was performed within 1 h to prevent intra-day or inter-day variation.
Component profiling of Japanese sake samples Fifteen kinds of commercially available Japanese sake were analyzed by GC/MS and LC/MS. GC/MS detected 109 kinds of compounds including sugars, organic acids, and amino acids. The amino acid profiles based on GC/MS analysis were not enantioselective, indicating that the information of each amino acid included information from both D- and L-amino acids. Several D-amino acids are known to have a different taste from the corresponding L-amino acids. However, GC/MS-based amino acid analysis does not consider amino acid chirality, which is important information. On the other hand, LC/MS detected 48 kinds of amino acids and amines including 11 kinds of D-amino acids. The ratio of detected D-amino acids to the corresponding L-amino acids is shown in Table 2. Although several D-amino acids showed levels of more than 10% of the corresponding L-isomers, the level of most D-amino acids was very low. However, according to a previous report, it is possible that trace components can contribute to the construction of food sensory prediction models (Yamamoto et al., 2014). Therefore, it is necessary to assess whether explanatory variables including D-amino acid profiles might be effective in the construction of sensory prediction models of food. Next, the component profiles obtained from GC/MS or LC/MS were subjected to principal component analysis (PCA). Figures 1-A and B show the score and loading plots produced using LC/MS data, respectively. The score plot successfully showed the differences in each sake sample. In addition, the loading plot indicated that the D-amino acid cluster is separated from the corresponding L-enantiomer cluster. Although the loading information of most L-amino acids is noticeably biased toward the positive side along PC1, D-amino acids were positively correlated with PC2. The loading plot along PC2 is related to coloration, and thus sake samples with high OD430 contained higher amounts of D-amino acids. To further study the relationship between D-amino acid profile and sake coloration, OPLS regression analysis was performed using component profiling data obtained via CROWNPAK CR-I and OD430 as explanatory and response variables, respectively. The regression model was evaluated for the following statistics: linearity (R2), predictability (Q2), and root mean square error estimation (RMSEE), which show the residual error in predictions. R2 values closer to 1 indicate models of higher accuracy, and Q2 values closer to 1 indicate higher predictive quality. Conventionally, a good regression model generally requires an R2 value higher than 0.9 and a Q2 value higher than 0.5. The regression model constructed in this study yielded values of 0.982, 0.976, and 0.087 for R2, Q2, and RMSEE, respectively. Hence, the R2 and Q2 values were sufficiently high and RMSEE was sufficiently small compared with the scale of response variables. OD430 is used as an index of maturation in the sake industry, since the coloration measured by OD430 is derived from melanoidin, which is the brown pigment produced by the Maillard reaction during aging. Therefore, the high correlation of the component profile with coloration was largely caused by melanoidin production and enabled the construction of a prediction model of maturation (Maillard reaction frequency). Furthermore, in order to identify compounds that were particularly highly correlated, the variable importance in projection (VIP) value of each compound was evaluated. The VIP value reveals the contribution to prediction model construction, and explanatory variables with a VIP >1 are generally thought to be important for constructing the prediction model. A list of the 10 compounds with the highest VIP values in this study is shown (Table 3). Consequently, many D-amino acids exhibited high VIP values; therefore, their profiles were thought to vary depending on generation of the Maillard reaction. This tendency appeared to be reasonable because naturally abundant L-amino acids may be racemized during the Maillard reaction (Amadori reaction). Amadori compounds, which are composed of amino acids and reducing sugars, are intermediate products of the Maillard reaction. Such Amadori compounds might be reversed to L- or D-amino acids and sugars in non-enantioselective reverse reactions (Inoue et al., 2016). In the score plot shown in Fig. 1-B, the plot that indicates OD430=0.15 was at PC2=ca. 1, indicating that the sample contained moderately high D-amino acids in spite of the low OD430. This result suggests that D-amino acids are not derived from racemization of L-amino acids via chemical reactions but from lactic acid bacteria involved in the traditional seed mash, ‘kimoto’. In the comparison of LC/MS and GC/MS data using PCA, the locations of each sake sample toward PC1 of the score plot were almost the same, and PC1 might indicate the differences in manufacturing features. On the other hand, the locations toward PC2 were different between LC/MS and GC/MS results. The addition of the LC/MS data helped explain the component profile changes by maturation and obtain more informative, explanatory variables.
| Sample No. | |||||||||||||||
|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|
| Compound | 1 | 2 | 3 | 4 | 5 | 6 | 7 | 8 | 9 | 10 | 11 | 12 | 13 | 14 | 15 |
| d-Ala | 0.26±0.04 | 0.61±0.20 | 3.71±0.04 | 0.07±0.12 | 2.85±0.15 | 3.79±0.07 | 2.52±0.09 | 0.98±0.05 | 1.56±0.02 | 3.34±0.32 | 2.47±0.38 | 0.56±0.07 | 0.05±0.08 | 0.13±0.11 | 0.49±0.06 |
| d-Ser | 0.68±0.04 | 1.31±0.27 | 0.61±0.05 | 0.15±0.02 | 0.62±0.03 | 3.86±0.11 | 0.6±0.03 | 0.25±0.08 | 0.24±0.03 | 0.09±0.15 | 5.34±0.68 | 3.00±0.08 | N.D. | N.D. | 0.20±0.07 |
| d-Ile | 0.14±0.01 | 0.21±0.02 | 0.05±0.04 | 0.08±0.03 | 0.26±0.01 | 0.34±0.03 | 0.07±0.01 | 0.04±0.01 | 0.03±0.01 | N.D. | 1.03±0.92 | 0.36±0.03 | N.D. | N.D. | N.D. |
| d-Leu | 0.05±0.01 | 0.09±0.01 | 0.07±0.001 | 0.03±0.01 | 0.15±0.02 | 0.68±0.01 | 0.05±0.003 | 0.05±0.004 | 0.01±0.001 | N.D. | 1.44±0.26 | 0.28±0.02 | N.D. | N.D. | N.D. |
| d-Asn | 1.86±0.13 | 2.65±0.05 | 1.49±0.07 | 0.48±0.05 | 1.43±0.12 | 8.99±0.05 | 1.40±0.07 | 0.62±0.02 | 2.11±0.12 | 0.45±0.1 | 13.31±0.61 | 8.59±0.24 | 0.27±0.05 | N.D. | 0.56±0.07 |
| d-Asp | 0.84±0.04 | 1.82±0.15 | 1.77±0.07 | 0.26±0.06 | 4.68±0.26 | 7.54±0.29 | 1.41±0.13 | 0.69±0.04 | 0.66±0.05 | 1.26±0.08 | 7.92±0.38 | 6.05±0.43 | 0.23±0.01 | 0.93±0.03 | 0.34±0.04 |
| d-Glu | 1.49±0.01 | 1.94±0.10 | 1.51±0.02 | 0.33±0.01 | 3.33±0.14 | 13.92±0.25 | 1.57±0.07 | 0.59±0.02 | 0.7±0.03 | 1.62±0.16 | 17.72±1.07 | 13.45±0.33 | 0.28±0.02 | 0.95±0.23 | 0.35±0.02 |
| d-His | 1.23±0.05 | 1.19±0.24 | 0.73±0.13 | 0.29±0.09 | 0.98±0.08 | 5.00±0.11 | 0.73±0.05 | 0.3±0.12 | 0.51±0.06 | 0.12±0.21 | 9.96±0.37 | 4.91±0.09 | 0.12±0.1 | N.D. | N.D. |
| d-Phe | 0.17±0.01 | 0.18±0.16 | 0.11±0.03 | N.D. | 0.21±0.001 | 1.67±0.10 | 0.09±0.01 | N.D. | 0.05±0.01 | N.D. | 5.7±0.11 | 1.01±0.03 | N.D. | N.D. | N.D. |
| d-Arg | 0.12±0.03 | N.D. | 0.11±0.004 | N.D. | 0.78±0.16 | 0.88±0.06 | 0.24±0.02 | 0.03±0.05 | 0.19±0.01 | N.D. | 0.96±0.03 | 0.68±0.02 | N.D. | N.D. | N.D. |
| d-Tyr | 0.06±0.002 | 0.08±0.01 | 0.03±0.004 | 0.01±0.006 | 0.05±0.009 | 0.68±0.034 | 0.03±0.002 | 0.01±0.01 | 0.02±0.003 | N.D. | 0.88±0.05 | 0.44±0.01 | N.D. | N.D. | N.D. |
N.D.: d-amino acid was not detected.
PCA score plot (A) and loading plot (B) obtained by LC/MS-based enantioselective amino acid analysis, and PCA score plot (C) and loading plot (D) obtained by GC/MS-based metabolomics. The number in score plot indicates the sample number, which links to that shown in Table 1.
| Index | Compound | VIP value | Coefficient |
|---|---|---|---|
| 1 | D-Phenylalanine | 2.232 | 0.156 |
| 2 | D-Tyrosine | 2.213 | 0.156 |
| 3 | D-Leucine | 2.193 | 0.169 |
| 4 | D-Serine | 2.028 | 0.085 |
| 5 | Histamine | 1.917 | 0.216 |
| 6 | D-Aspargine | 1.864 | 0.024 |
| 7 | D-Aspartic acid | 1.825 | 0.039 |
| 8 | D-Histidine | 1.757 | 0.004 |
| 9 | D-Isoleucine | 1.710 | 0.020 |
| 10 | D-Arginine | 1.655 | 0.051 |
| 11 | D-Glutamic acid | 1.482 | 0.020 |
| 12 | Creatine | 1.162 | 0.028 |
| 13 | D-Alanine | 1.132 | 0.081 |
| 14 | Nicotinamide | 1.130 | 0.111 |
Correlation analysis between component profiles and sensory score In order to obtain sensory data on Japanese sake, five kinds of sensory attributes, sweetness, sourness, bitterness, strong taste (Nojun), and aftertaste length, were evaluated on a 7-point integer scale from 0 to 6. An OPLS regression analysis was conducted, in which sensory scores were used as the response variable and component profiles were used as the explanatory variable to create a sensory prediction model for each sensory attribute (Figure 2). In this study, three kinds of component profiles obtained by LC-MS analysis, GC-MS analysis, and the combination of GC-MS and LC-MS analyses were used as explanatory variables. When the LC/MS data were integrated with the GC/MS data, some GC/MS-based amino acid information overlapped with the LC/MS-based information, and the overlapping GC/MS amino acid information was excluded from further multivariate analysis. First, to confirm the reliability of prediction models, R2, Q2 values, and RMSEE were examined. RMSEE was defined to be less than 0.7, indicating a prediction error of less than 10% in the 7-point scale of the response variable. Consequently, R2, Q2, and RMSEE of all sensory attributes were sufficient to successfully construct a prediction model via enantioselective amino acid profiles, obtained by the combination of LC-MS and GC-MS data. When only GC-MS data and the combination of GC-MS and LC-MS data were compared, the combination of GC-MS and LC-MS data showed higher R2 and Q2 values and smaller RMSEE in the prediction model of sweetness, sourness, bitterness, and aftertaste length (Table 4). This result indicates that the usage of enantioselective amino acid analysis contributed to the construction of better food sensory prediction models. Moreover, to investigate the contribution of each compound towards the construction of the model, VIP values were assessed. Compounds with high VIP values are shown (Table 5). Trehalose, glucose (or galactose), and xylitol, compounds with high VIP values in the prediction model of sweetness, are sweeteners. Furthermore, D-amino acid profiles showed a strong positive correlation with ‘sweetness’. D-Alanine exists in Japanese sake in high concentrations (Gogami et al., 2011) and the taste of D-alanine standard solution is known to be sweet (Schiffman and Sennewald, 1981). Therefore, this suggests the possibility that D-alanine plays a direct role as a component of sweetness. Lactic acid, a sour compound, showed a high VIP value in the prediction model of sourness, demonstrating that lactic acid is related to the sour taste of Japanese sake (Shimazu et al., 2009). In addition, several D-amino acids, such as D-aspartic acid, D-histidine, D-isoleucine, D-arginine, D-glutamic acid, D-serine, D-asparagine, and D-phenylalanine, showed a strong correlation with ‘bitterness’. Among these D-amino acids, D-arginine is known to indicate ‘bitterness’ (Schiffman and Sennewald, 1981). However, the other D-amino acids might not be directly associated with ‘bitterness’ taste. There is a possibility that other compounds that have been reported to increase during maturation are associated with a bitter taste (Takahashi, 1974). D-Amino acid profiles, which were correlated with maturation (as shown in Table 3), might be co-correlated with those compounds having a bitter taste. Therefore, these D-amino acid profiles appear to be very important components for predicting ‘bitterness’ intensity, although it remains unclear whether they play a direct role as bitter components. In the prediction model of strong taste, many kinds of sugar alcohols had high VIP values. Although no D-amino acid was ranked in the top 10 of the compounds with the highest VIP values, the VIP values of D-alanine and D-glutamic acid were sufficiently high at 1.49 and 1.31, respectively. This result suggests that D-alanine and D-glutamic acid might have a novel synergistic function in ‘strong taste’. Okada et al. revealed the relationship between D-alanine, D-aspartic acid, and D-glutamic acid and the ‘strong taste’ of Japanese sake (Okada et al., 2013). Basically, the present result is consistent with the previous report by Okada et al. Moreover, D-amino acids, which were important for the models of ‘bitterness’ and ‘strong taste’, might contribute to the model of ‘aftertaste length’. The sensory score of ‘aftertaste length’ was correlated with that of ‘bitterness’ and ‘strong taste’ (Correlation coefficients were 0.67 and 0.79, respectively). Therefore, high VIP value compounds that were related to ‘bitterness’ and ‘strong taste’ were also important for ‘aftertaste length’.
OPLS regression model using component profiles obtained by GC/MS and LC/MS analysis as explanatory variables and the sensory evaluation score as the response variable.
| GC-MS | LC-MS | GC-MS+LC-MS | ||||||||||
|---|---|---|---|---|---|---|---|---|---|---|---|---|
| Latent variables | R2 | Q2 | RMSEE | Latent variables | R2 | Q2 | RMSEE | Latent variables | R2 | Q2 | RMSEE | |
| Sweetness | 1+3+0 | 0.967 | 0.921 | 0.208 | 1+3+0 | 0.936 | 0.896 | 0.289 | 1+3+0 | 0.979 | 0.948 | 0.166 |
| Sourness | 1+3+0 | 0.973 | 0.944 | 0.202 | 1+3+0 | 0.923 | 0.838 | 0.339 | 1+3+0 | 0.977 | 0.962 | 0.187 |
| Bitterness | 1+2+0 | 0.925 | 0.886 | 0.330 | 1+2+0 | 0.936 | 0.913 | 0.305 | 1+2+0 | 0.962 | 0.934 | 0.235 |
| Strong taste | 1+3+0 | 0.981 | 0.960 | 0.221 | 1+3+0 | 0.931 | 0.894 | 0.424 | 1+3+0 | 0.976 | 0.955 | 0.250 |
| Aftertaste length | 1+4+0 | 0.979 | 0.954 | 0.277 | 1+4+0 | 0.937 | 0.897 | 0.476 | 1+4+0 | 0.984 | 0.959 | 0.240 |
| Sensory | Compound | Method | VIP | Coefficient |
|---|---|---|---|---|
| Sweetness | Unknown_657 | GC | 2.09 | 0.11 |
| Unknown_529 | GC | 1.87 | 0.07 | |
| Propyleneglycol | GC | 1.73 | 0.04 | |
| Trehalose | GC | 1.7 | 0.04 | |
| Galactose&Glucose | GC | 1.65 | 0.05 | |
| D-Alanine | LC | 1.64 | 0.08 | |
| Xylitol | GC | 1.61 | 0.05 | |
| Succinic acid | GC | 1.61 | 0.03 | |
| Heptanoic acid | GC | 1.59 | 0.01 | |
| Cadaverine | GC | 1.57 | 0.04 | |
| Sourness | 3-Phenyllactic acid | GC | 2.86 | 0.09 |
| Unknown_493 | GC | 2.85 | −0.07 | |
| Lactic acid | GC | 2.63 | 0.06 | |
| Gentiobiose | GC | 2.56 | −0.07 | |
| Unknown_894 | GC | 2.37 | −0.06 | |
| Unknown_1112 | GC | 2.35 | −0.05 | |
| Unknown_1135 | GC | 2.22 | −0.05 | |
| Xylulose&Ribulose | GC | 2.18 | 0.06 | |
| Unknown_657 | GC | 2.03 | 0.06 | |
| Maltose | GC | 2.02 | 0.04 | |
| Bitterness | D-Aspartic acid | LC | 2.33 | 0.06 |
| D-Histidine | LC | 2.29 | 0.07 | |
| D-Isoleucine | LC | 2.11 | 0.06 | |
| D-Arginine | LC | 2.07 | 0.07 | |
| D-Glutamic acid | LC | 2.05 | 0.04 | |
| Glyceric acid | GC | 2.04 | 0.04 | |
| D-Serine | LC | 2.04 | 0.04 | |
| D-Aspargine | LC | 2.00 | 0.06 | |
| D-Phenylalanine | LC | 1.86 | 0.04 | |
| Creatine | LC | 1.80 | 0.03 | |
| Strong taste | Trehalose | GC | 1.74 | 0.05 |
| (Nojun) | Psicose&Tagatose | GC | 1.73 | 0.04 |
| Propyleneglycol | GC | 1.70 | 0.05 | |
| Unknown_499 | GC | 1.68 | 0.04 | |
| Xylitol | GC | 1.65 | 0.06 | |
| Unknown_198 | GC | 1.6 | 0.05 | |
| Sorbitol | GC | 1.59 | 0.03 | |
| Glycerol | GC | 1.57 | 0.03 | |
| Unknown_430 | GC | 1.54 | 0.07 | |
| Unknown_762 | GC | 1.54 | 0.03 | |
| Aftertaste length | D-Histidine | LC | 1.84 | 0.07 |
| D-Glutamic acid | LC | 1.79 | 0.01 | |
| Glyceric acid | GC | 1.78 | 0.04 | |
| D-Isoleucine | LC | 1.72 | 0.01 | |
| Unknown_430 | GC | 1.70 | 0.06 | |
| D-Aspartic acid | LC | 1.69 | 0.01 | |
| Psicose&Tagatose | GC | 1.68 | 0.06 | |
| D-Serine | LC | 1.68 | 0.04 | |
| Unknown_499 | GC | 1.67 | 0.06 | |
| D-Aspargine | LC | 1.65 | 0.07 |
In summary, the enantioselective amino acid analytical method and subsequent OPLS regression analysis successfully demonstrated the correlation between D-amino acids and sensory attributes in Japanese sake. VIP values obtained by OPLS regression analysis revealed that D-amino acids were strongly correlated with sweetness, bitterness, strong taste, and aftertaste length. The addition of the LC/MS-based enantioselective amino acid profile to the GC/MS-based non-target metabolomics profile improved the predictability of sensory prediction models compared with GC/MS data alone. Regression analysis using D-amino acid profile and quantitative data of food functions is expected to promote further studies on the effect of D-amino acids on the sensory attributes of food. The present study is the first trial of regression to sensory scores of food items with D-amino acid profiles as explanatory variables.
Acknowledgements This work was supported by JSPS KAKENHI Grant Number 18J20256 (MT) and 17K19235 (EF) and by grants from the Project of the NARO Bio-oriented Technology Research Advancement Institution (R&D matching funds on the field for Knowledge Integration and innovation). The study represents a portion of the dissertation submitted by Moyu Taniguchi to Osaka University in partial fulfillment of the requirement for her Ph.D.
