MetabolomicBiomarkersDifferentiate SoySauceFreshnessunder Conditions of Accelerated Storage

Department of Biomolecular Chemistry, University of Wisconsin-Madison, Madison, WI 53706, USA Morgridge Institute for Research, Madison, WI 53715, USA Department of Chemistry, University of Wisconsin-Madison, Madison, WI 53706, USA Department of Food,Bioprocessing,& Nutrition Sciences, North Carolina State University, Raleigh, NC 27695, USA Kikkoman USA, R & D Laboratory,Inc., Madison, WI 53719, USA Department of Food Science, University of Wisconsin-Madison, Madison, WI 53528, USA


Introduction
Analytical technologies that enable effective food quality characterization, differentiation, and management are based on accurate and reliable data generation and have been increasing in type and performance, including emerging methodologies, such as high-resolution mass spectrometry metabolomic-based approaches [1,2]. Different types of analytical platforms are often applied in combination with nontargeted analyses to maximize and broaden detection capabilities for improved sensitivity and molecule identification. Such approaches generate discriminatory patterns of complex chemical components from different samples through the creation and comparison of metabolic fingerprints, thus providing the capability to differentiate chemically complex biological samples [3,4]. In fact, several studies have reported attempts to relate these chemical fingerprints to the discriminatory sensory qualities of food samples by examining these analytical data with various multivariate statistical designs [5][6][7].
In the case of exploratory studies for pursuing compounds in foods responsible for a particular sensory quality, sensory-guided techniques in conjunction with novel, advanced chromatographic and mass spectrometry analyses are applied. Such approaches can be useful for characterizing changes in sensory attributes with changes in chemical profiles resulting from independent experimental variables, such as product storage time [8]. However, such approaches include the risk of missing the more complicated collective or cumulative effects of biologically active substances that occur in authentic food systems. Yet, these exploratory approaches provide a critical foundation for further discoveries and can provide a more comprehensive understanding of the changing chemistries involved for the purposes of product differentiation. However, the elucidation of key, flavor-impact compound identities is not the universal goal of flavor assessments such as is the case with electronic nose technologies [9], where the compound-dependent signal can be used to differentiate treatment effects rather than elucidating the causative chemistries. Here, we present a similar approach using metabolomics to differentiate soy sauce (SS) samples exposed to an extended storage treatment.
Previous studies that characterize critical SS sensory attributes have been conducted, and some lexicons have been consequently developed by several research teams [10,11]. ese lexicons allow a common language for the evaluation of native SS sensory qualities, thus sharing new insights and applied discoveries among consumers, researchers, and manufacturers. e sensory attribute of freshness in SS is absent from existing SS lexicons, despite the fact that the loss of freshness has been anecdotally recognized as a key sensory attribute that determines SS quality as perceived by consumers, especially in markets that have high standards for SS sensory performance and value [12]. Furthermore, freshness as a key characteristic of foods has gained attention given its critical importance to consumers. However, it requires a complex and multisensory, crossmodal affective assessment [13].
As a condiment, SS is applied to a wide variety of foods ranging from common food to some of the most sophisticated meals crafted. After manufacturing and as a function of storage parameters, such as time, light exposure, and oxygen concentration, SS is anecdotally thought to lose its fresh character, manifested as a darkened color and altered sensory attributes, suggestive of spontaneous reactions that are especially detrimental, where SS is applied in fine culinary applications [12]. ese adverse changes over storage time are typically expressed in the industry as a loss of freshness and are thought to be catalyzed when SS is exposed to oxygen, high temperatures, or excessive storage time. Chemical reactions generally associated with oxidative or Maillard reaction pathways have been suggested to contribute to these undesirable changes [12]; however, there is little information detailing discernable shifts in chemistry, and thus, there is no rationale to objectively mark its occurrence. In Japan, the trade association, called the Japan Soy Sauce Brewers Association (JSSB), proposed a standard "best before date" shelf life determination, in which freshness is a factor that must be sensorially determined. Yet, even in this JSSB method, there is no definitive sensory characterization for freshness aside from a general subjective loss of desirable character relative to a freshly prepared control anchor.
is lack of definitive sensory and chemical change prevents the application of a means to prevent the loss of SS freshness. We further hypothesize that the native chemistries in flux over the course of a loss of freshness from storage can be elucidated through the comparison of differential metabolic fingerprints and multivariate analyses. us, the objectives for this work include the assessment of SS samples aged under conditions noted later as a means of elucidating sensory attribute changes with concomitant changes in chemical profiles. Study objectives were achieved based on changes in a limited set of metabolites, including organic acids, amino acids, and various glycosylated compounds, resulting from the aging process using a metabolomic approach.

Soy Sauce Samples.
Traditionally brewed SS was directly obtained from the manufacturer (Kikkoman, Chiba, Japan) bottled in sealed, 1-liter plastic containers for storage treatments. Samples were stored in a dark incubator held at 30°C, and triplicate samples were randomly removed at twomonth intervals up to and including the eight-month storage event. Upon removal from incubation, samples were stored at − 80°C until analyses were completed.

Sensory Analyses.
A protocol for sensory analysis of SS was filed and approved (North Carolina State University, Institutional Review Board) prior to initiation of study. A trained panel (n � 6 panelists, each with at least 100 h of experience in the descriptive analysis of foods and beverages) assessed SS attributes from the literature [10,14] and novel attributes from preliminary studies using a 15-point universal intensity scale consistent with the Spectrum TM method. Based on this documented complexity and other recently published literature, eighteen aroma and flavor attributes were selected for the study (Table 1), from which eleven attributes generated responses in the study ( Table 2). Each SS sample was evaluated by each panelist in triplicate. Paper ballots were used for data collection and were manually transferred into statistical software. SS samples were diluted at 1 : 1 with deionized water prior to the sensory evaluation. Training sessions and preliminary evaluations indicated that dilution at this level minimized fatigue with no effect (p > 0.05 ) on perceived sensory attributes. Attribute intensities were scored on a 0-to 15-point universal intensity scale consistent with the Spectrum method; panelists were allowed to score beyond this range if warranted by a particular attribute intensity [19].

High-Resolution Gas Chromatography (HR-GC-MS)
A preliminary study revealed several hundred polar compounds in SS by HR-GC-MS to determine if the instrumental analyses were able to detect relative changes in compound concentrations and refine sample preparation methodologies and instrument run parameters. We further note that traditional volatile chemistries associated with aroma character, such as esters or aldehydes, were not discoverable using the sample extraction and derivatization methods employed. Yet, this methodology allowed a relatively simple, rapid approach that discovered of a host of compounds as noted later, and thus, we deemed this approach suitable for the intention of this study, which is the ability to differentiate sample aging treatments. In general, classes of compounds shifted by the aging treatment included organic acids, amino acids, and various glycosylated metabolites.

Sample Preparation for HR-GC-MS Analyses. SS samples
were kept on ice along with all other laboratory reagents used. Aliquots of SS samples were diluted 10-fold with DI water. Performed in triplicate, 20 μL of diluted sample was pipetted into a 1.5 mL Eppendorf tube and 225 μL MeOH was added. e mixture was vortexed for 10 s. Next, 750 μL methyl tert-butyl ether was added, and the mixture was vortexed for 10 s. e solution was mixed using an orbital Table 1: Initially trained panel sensory attributes for soy sauce (terms adopted from previous studies are cited).

Term
Definition Reference Ashy [10] Dry, dusty, smoky aromatics associated with the residual of burnt products Ashes from burnt wood (from fireplace or outdoor fire pit) Beefy/brothy [17] Aromatics associated with beef stock or bouillon cubes Bouillon cubes rehydrated in water, and then, 20 mL was placed in a 58 mL lidded soufflé cup Brown fruity/ prune [14] A sweet, floral aromatic associated with prunes Mariani brand prunes Brown spice clove [10] Aromatics associated with nutmeg, clove, and cinnamon Nutmeg, clove, and cinnamon Caramel/sweet aromatic [10] Sweet aromatics associated with vanilla, vanillin, burnt sugar, caramel, and molasses Vanilla extract, vanillin, burnt sugar, caramel, and molasses Chemical [10] e aromatics associated with plastic and burnt plastics Clear PET bottle Fruity ester [16] A sweet, floral aromatic blend of a variety of ripe fruits Juicy juice Nestle all natural 100% kiwi strawberry ethyl hexanoate Fruity fermented/ beer/malty [10] A sour, sweet fermented aromatic associated with fermented fruit, beer, fruits, and malt 225 mL amber sniff jar with 10 ppm of 2-ethyl-3methylbutanal, grape nuts soaked in water and beer Fruity/grape A sweet aromatic associated with grape juice Welch's grape juice Maple Aromatic associated with maple syrup Maple syrup Methional/potato [17] Aromatics associated with vegetable soup stock, bouillon cubes, or canned potatoes (methional) Bouillon cubes rehydrated in water, and then, 20 mL was placed in a 58 mL lidded soufflé cup or a 225 mL amber sniff jar with 10 ppm of methional or potato flakes Mushroom e earthy aromatic associated with fresh mushrooms 1-octen-3-one shiitake mushroom Nutty/sesame [10] e nonspecific nutlike flavors that are characteristic of several different nuts, e.g., peanuts, hazelnuts, pecans, and almonds Shamrock mixed crushed nuts or sesame seeds Olive Aromatic associated with green olives Green olives Phenolic [18] Aromatic associated with phenol 225 mL amber sniff jar with 10 ppm of phenol or Band-Aids Pyrazine green/ raw10 Aromatic associated with pyrazine or raw peanuts 225 mL amber sniff jar with 10 ppm of pyrazine Smoky [10] A sweet, pungent, dry, and woody aroma/aromatic associated with smoke Liquid smoke with hickory barbecue flavor (Tone's brand) Sour aromatic/ vinegar [10] Sour aromatic associated with acetic acid 225 mL amber sniff jar with 10 ppm of acetic acid Surimi (sweet aroma) Sweet aromatic associated with imitation crab Imitation crab (Fisherman's market) Astringent mouthfeel [15] Chemical feeling factor on the tongue or oral cavity described as puckering or dry Alum (1% in water) Mouth burn [15] Trigeminal pain response to the activation of neural receptors on the tongue and soft palate Soda water or ethanol Sweet taste [15] Fundamental taste sensation elicited by sugars Sucrose (5% in water) Sour taste [15] Fundamental taste sensation elicited by citric acid Citric acid (0.05% in water) Salty taste [15] Fundamental taste sensation elicited by salts Sodium chloride (5% in water) Bitter taste [15] Fundamental taste sensation elicited by caffeine and quinine Caffeine (0.05% in water) Umami taste [15] Fundamental taste sensation elicited by certain peptides and nucleotides MSG (1% in water) shaker for 6 min. To induce phase separation, 187.5 μL water was added, and the solution was vortexed again for 20 s. e sample was then centrifuged for 2 min at 12,000×g and 4°C. e upper phase in the 1.5 mL Eppendorf tube was discarded, and 250 μL of the lower (aqueous) phase was removed and placed in a separate 1.5 mL Eppendorf tube. To this, 250 μL ACN was added to precipitate protein.
e mixture was vortexed for 15 s and centrifuged at 13,000×g for 5 min at 4°C. en, 300 μL of the supernatant was aliquoted into glass autosampler vials. e mixture was dried in a vacuum concentrator. Once dry, samples were resuspended in 50 μL methoxyamine hydrochloride solution (20 mg/mL, pyridine solution), vortexed for 15 s, and heated at 37°C for 90 min.
en, 100 μL of N-methyl-N-trimethylsilyltrifluoroacetamide (MSTFA) was added and vortexed for 15 s, and the mixture was heated at 60°C for 60 min.

HR-GC-MS Analysis.
Samples were analyzed using a GC-MS instrument (Trace 1310 GC, ermo Scientific, Waltham, MA) coupled to a mass spectrometer (Q Exactive Orbitrap, ermo Scientific, Waltham, MA). A temperature gradient ranging from 100°C to 320°C was employed spanning a total runtime of 25 min. Analytes were injected onto a 30 m × 0.25 mm ID × 1 μm thickness column (TraceGOLD TG-5SILMS, ermo Scientific, Waltham, MA) using a 1 : 10 split at a temperature of 275°C and ionized using electron ionization (EI). e mass spectrometer was operated in full scan mode using a resolution of 30,000 [3].

HR-GC-MS Data
Processing. Data processing was done using a software suite developed in-house that is available at https://github.com/coongroup. Following data acquisition, raw EI-GC/MS spectral data were deconvolved into "features" and then grouped into individual spectra containing only product ions stemming from a singular parent molecule. Feature groups from samples and background were compared and those found in both were removed from further analyses. Compound identifications for the metabolites analyzed were assigned by comparing deconvolved high-resolution spectra against unit-resolution reference spectra present in the NIST 12 MS/EI library and authentic standards run in-house. To calculate spectral similarity between experimental and reference spectra, a weighted dot product calculation was used. Metabolites lacking a confident identification were classified as "unknown." Peak heights of specified quantified m/z ratios were used to represent metabolite abundance. e data set was further processed using a linear regression approach (non-log 2 transformed intensity values versus run order) to normalize for run order effects on signal.

Data Analysis.
All MS measurements were integrated with sensory information and processed with data analysis and visualization software (see https://coonlabdatadev.com). p values displayed in figures and tables were calculated using Student's t-test comparing time points 2, 4, 6, and 8 months with 0 months. We performed covariant analysis in R using the "cor.test" function and method "Kendall"; with this method, we identified compounds from MS datasets correlating sensory descriptor scores (p < 0.05). From this subset of molecules which correlate with sensory terms, we performed hierarchical cluster analysis on the correlation matrix of molecule-sensory pairs using the R function "hclust" and k-means equal to 5. A heatmap of this matrix was generated using the R function "pheatmap" [20][21][22]. Both chemical and sensory data were further analyzed for the purpose of generating predictive models of their correlative value using multiple, stepwise linear regression (JMP vs. 14, SAS Institute Inc., Cary, NC, 1989-2020).

Results and Discussion
Overall, sensory and HR-GC-MS analyses provided novel analyte discoveries resulting from the aging or storage of SS as described by the aforementioned conditions. Correspondingly, there were limited, but significant age-based sensory changes in the SS including the loss of fruity/grape and nutty/sesame aroma characteristics and the increase of methional/potato aroma. Changes in biomarker profiles varied across the aging variable yet intensified in number and quantity of compounds captured as described in detail later. We note that the analyses conducted were designed to discern the most notable changes in polar biomarker profiles discernable by the instrumental analyses as a function of the storage time variable. We further acknowledge that such an approach is not designed to or sufficient for establishing causal sensory changes; it is rather a means of discerning the influence of storage-based aging through specific biomarkers.

Sensory Profiles.
Although eighteen attributes were initially considered by sensory panelists, eleven terms were selected for further consideration in that the other seven terms did not generate changes in sensory responses over the course of the study. Of the terms that generated responses, five were not affected by the treatment variable of storage time, leaving six that were affected. Two terms, namely, fruity/grape and nutty/sesame, displayed significant decreases in aroma intensity with increases in storage time. e term methional/potato increased with storage time, suggesting an increase in aroma resulting from sulfurcontaining volatiles. Although the term caramel/sweet aromatic was influenced by storage time, the results were not consistent. e samples also displayed slight changes in the astringent and mouth burn sensations, yet the results were not consistent across storage time.
Changes in freshness in foods and beverages, such as spoilage in fluid milk, are complex and involve changes in various classes of compounds, which affect color, texture, aroma, and taste attributes created by spontaneous, enzymatic, and microbial activities. In isolation, none of these effects may be discernable by sensory assessment, yet collectively they can affect a notable sensory departure from the native state. A notable observation in this study is that newly manufactured or "fresh" SS is differentiated from aged SS by multiple distinguishable sensory attributes. Fresh SS was higher in caramel and fruity and nutty attributes and lower in methional notes. is latter aroma has been associated with deteriorative changes in other foods with complex compositional profiles [15,21]. We further noted no changes in other SS attributes, such as umami or sweet tastes. e following is a regression model derived from sensory factors affected by age in months as a means of defining SS freshness loss: and we propose that this model infers that SS freshness is characterized from a sensory standpoint from the collective changes in the sensory attributes of caramel and fruity and nutty attributes and it is lower in methional notes as weighted by the correlation coefficients derived from regression analysis. Although the sensory character is complex, these four attributes were able to predict 79% of the changes manifested across the storage time assessed based on correlation analysis. A complementary model was created using instrumental data wherein compounds were first selected based on their relative strength of correlation to the storage variable and then assessed using a stepwise regression model. e final model for the prediction of SS freshness loss using instrumental data is shown as

(2)
Upon validation of the assessment using the aforementioned correlations, these five chemical variables were able to predict >99% of the variation in the storage variable.

HR-GC-MS Metabolomics.
As anticipated, chromatograms were complex and yielded several thousand resolved analytes. While many analytes were tentatively identified, a significant number were not, namely, those with sugar moieties where the type and degree of ligand substitution were not differentiable through molecular weight or mass spectra database comparison (Table 3). A graphical depiction of the type and complexity of variation from the t � 0 control SS sample is presented in Figures 1(a)-1(d). To better understand which metabolites are associated with the critical sensory parameters, we performed a correlation analysis between metabolite abundance and sensory measurements for the 15 samples (0-8 mo). e Kendall Tau correlation coefficient was used to assess strength of association and was chosen to better account for the nonparametric nature of the sensory data. e resulting significant correlations are visualized using a heatmap (Figure 2(a)), where the strength of the correlation is indicated by heat color ranging from blue (indicating a strong negative correlation) to red color (strong positive correlation).
After applying hierarchical clustering and k-means (k � 5) clustering techniques, we were able to identify distinct patterns of associations between metabolites (Figure 2(a)) and sensory attributes. For example, in the yellow cluster, there are strong positive correlations between cluster member metabolites, like L-tryptophan, and sensory attributes, nutty/sesame and fruity grape. In contrast, the orange cluster metabolites, such as glyceric acid, show more negative correlations with nutty/ sesame and positive correlations with astringent mouthfeel scores. Next, we explored how these cluster members (Figure 2(b)) changed over time, and we plotted the identified metabolite features abundance relative to the time zero months (Figures 2(c)-2(g)). e orange cluster members positively associated with astringent mouthfeel scores increased between 0 and 8 months, while the yellow cluster members associated with nutty/sesame decreased between 0 and 8 months. e green cluster members also decreased with time, and cluster members of blue and gray clusters showed less change with time. Lastly, we wanted to integrate these results to the model using metabolite features, sugar RT 13.6, sugar RT 19.4, sugar RT 19.5, arabinose, and L-tryptophan, to quantify "freshness." Notably, the first three components of the model are members of the orange cluster, and the last two components of the model, arabinose and L-tryptophan, are members of the yellow cluster.
is work provides a depiction of modern metabolomic technologies applied to the subject of assessing storagebased aging of SS with time and sensory-based variables. Metabolite profiling such as the one applied in this study offers a promising technology to unravel complex, yet Journal of Food Quality      Figure 2: Heat map of correlations between sensory terms affected by storage time and metabolites. Heat coloring denotes the strength and direction of the correlation based on the Kendall correlation coefficient scale. Metabolites with significant correlations (p < 0.05) to one or more sensory terms were split into five groups based on hierarchical clustering denoted by the color bars orange, yellow, green, blue, and gray. Sensory variables were similarly grouped by hierarchal clustering based on correlations with metabolites (a). Identified metabolites associated within each hierarchical cluster (b). Unidentified features are not listed and sugar molecules are only confidently assigned at a molecular class level and are differentiated by chromatographic retention time. For each cluster, the changes in abundance for cluster members (molecules) are plotted over time as log2 of the fold change from time 0 month (c-g). e members of the yellow cluster have the greatest change over time.
critical aroma-based attributes in foods such as SS. We further suggest that SS freshness, as a sensory attribute, is a complex term defined by the presence of specific aromatic attributes of fruity/grape and nutty/sesame and absence of methional/potato aroma. More quantitatively, it can be expressed as the models derived from the regression analyses mentioned earlier. Furthermore, we recognize that consumer perceptions of freshness and value of finished food products such as SS may vary based on prior experiences, familiarization, and exposure factors [23]. However, we suggest that a sensory basis for quality assessment provides a key means of discriminating important quality parameters such as those influenced by extended storage or product spoilage. We acknowledge that the changes in sensory attributes were not rationally associated with the changes noted in the chemical profiles, whereas changes in the fruity attribute did not correlate with compounds noted for fruity character, such as ethyl esters. We propose that these finer discoveries, at least on the volatile fraction, are the subject of future investigations such as those done in other works [23], wherein more comprehensive analytical examinations are conducted complemented by sensory validation using the addition of authentic standards to induce specific sensory attributes.
Yet, three novel contributions are served in this work. First, we have established that changes in sensory character are discernable in SS as a function of age or storage time. Second, the sensory character of SS freshness is indeed a multidimensional or meta-term attribute affected by changes in several key aromatic attributes as noted in other foods [24].
ird, a rapid, analytical assessment of SS compounds was developed that can predict or discern aging in SS across storage time with high precision using changes in polar metabolite profiles.

Data Availability
Data can be accessed through contacting author's institution at https://www.minds.wisconsin.edu.

Conflicts of Interest
Author Coon is a consultant for ermo Fisher Scientific. e remaining authors declare no conflicts of interest related to this work.