Fingerprints of volatile flavor compounds from southern stinky tofu brine with headspace solid‐phase microextraction/gas chromatography–mass spectrometry and chemometric methods

Abstract It is difficult to produce southern stinky tofu, a famous traditional Chinese snack, at industry scale due to the complex composition of its brine. In this study, the fingerprints of organic volatile flavor compounds in the southern stinky tofu brine samples from five manufacturers were studied using headspace solid‐phase microextraction/gas chromatography–mass spectrometry (HS‐SPME/GC‐MS) with the aid of chemometric methods. The fingerprints were obtained by HS‐SPME/GC‐MS and analyzed with the time shift alignment method, Shannon entropy, correlation coefficient, and principal component analysis. The results show that the time shifts in the samples can be accurately corrected by the time shift alignment method despite unexpected interferences. The fingerprint information was evaluated by Shannon entropy, while the similarities and differences in the fingerprints were investigated by correlation coefficient. Moreover, the identification of stinky tofu manufacturers can be achieved by principal component analysis. The predominant volatile compounds in southern stinky tofu brines were indole, 3‐methylindole, phenol, and 4‐methylphenol. Therefore, the established fingerprinting of volatile compounds for the brines by combining HS‐SPME/GC‐MS with chemometric methods was a simple and reliable method.

However, without accurate quantitative analytical technology, it is difficult for stinky tofu brine to be industrialized and commercialized. The crucial process parameters of brine manufacturing are not yet identified, making it difficult to satisfy batch repeatability and to scale up in food industry (Xu & Jiang, 2014).
Different odor characteristics of the brine samples from different fermentation periods and manufacturers can be observed, and the research of the fingerprints of volatile flavor compounds in the brine samples helps to optimize quality control of these products.
As an advanced analytical technique to analyze flavor compounds in food samples, headspace solid-phase microextraction/gas chromatography-mass spectrometry (HS-SPME/GC-MS) has many advantages, such as easy to perform, solvent free, sensitive, and selective (Canellas, Vera, & Nerín, 2016;Lv et al., 2017;Wang et al., 2017;Xiao et al., 2017;Xu & Jiang, 2014). However, the use of this method in the analysis of organic volatile flavor compounds in southern stinky tofu brine has been very limited, because the research of organic volatile flavor compounds is a very difficult task (Chao, Tomii, Watanabe, & Tsai, 2008;Liu, Chen, Sun, & Huang, 2009). The evaluation of these compounds in the complex sample is a challenging to GC methods due to the overlapping signals and the high number of compounds. Besides, many factors can influence the chemical composition, including different raw materials and process parameters.
Different instruments or conditions for any particular product may also lead to differences between samples of the same product (Zeng, Liang, & Xu, 2005). Furthermore, it is a difficult task to obtain information about the presence or absence of specific components in the brine samples by comparing the mass spectra with those in the mass spectrometry library.
Identifying and validating all the components in the brine samples are very time consuming, and it is not mandatory by quality control. One option to resolve this problem is to study the chromatographic fingerprints without determining all the components in each brine sample (Ding, Ni, & Kokot, 2015;Pripdeevech & Machan, 2011;Wan, Stevenson, Chen, & Melton, 1999;Xia, Mei, Yu, & Li, 2017). The fingerprint technique, which characterizes the integral and local features of the brine samples, can be used to make comprehensive quality assessments of southern stinky tofu. Due to the highly complex GC-MS datasets obtained from brines, chemometric techniques have become essential to analyze the chemical variability and to detect slightly and almost imperceptible composition changes (Arisseto, Vicente, Furlani, Pereira, & de Figueiredo Toledo, 2013;Li, Cai, & Shao, 2015;Lv et al., 2015;Zhang et al., 2017). Therefore, in this study, the fingerprints of organic volatile flavor compounds in the brine samples of southern stinky tofu from five manufacturers were studied using HS-SPME/GC-MS with the aid of chemometric methods. The fingerprints were obtained by HS-SPME/GC-MS and analyzed with the time shift alignment

| Materials
Brine samples manufactured at five production sites, referred to as Cheng, Huo, Wang, Bai, and Luo, respectively, were analyzed.
Two kinds of SPME fibers with different coatings were pur- and polyella (85 μm in thickness, white). They were preconditioned prior to the analysis in the injection port of GC according to the instructions suggested by the manufacturer.

| Headspace solid-phase microextraction/gas chromatography-mass spectrometry
The brine sample (5 ml) and a magnetic stir bar were placed in a 15ml vial. Before the insertion of SPME fiber, the vial was sealed with one Teflon cover and equilibrated for 20 min in a 60°C water bath.
After that, the fiber was exposed in the upper space of the sealed vial to extract compounds for 40 min.

×0.25 μm;
Restek, Bellefonte, PA, USA) was employed. In the experiment, the electron impact ionization was tuned at 70 eV and helium (99.999%, BOC) was used as carrier gas with an average linear velocity of 1.0 ml/min. The temperatures of the GC injector and the ion sources were 250°C and 200°C, respectively. The mass range of the MS detector was from 45 to 450 m.u. The oven temperature was initially at 45°C for 2 min; then increased at 5°C/min to 150°C, which was held for 2 min; and finally raised to 290°C at 15°C/min, which was held for 10 min. The injection port was in splitless mode.
Take the analysis of Huo sample as an example. Figure

| Time shift correction
For the analysis of the brine samples, it might be difficult to separate the analytes from the interferences with good resolution. Moreover, the interferences of peak shifts are also serious for the chromatograms of the other four brine samples. However, with the COW method, similar results can be obtained for the analysis of them, and the run-to-run retention time shifts can be accurately corrected, too. Therefore, the corrected fingerprints can be used for further discussion.

| Shannon entropy
It is very important to reasonably evaluate whether a chromatographic fingerprint carries enough information. In the work, the fingerprint information was evaluated by Shannon entropy, as shown in

| Correlation coefficient
The similarities and differences in the fingerprints were investigated by correlation coefficient, as shown in Table 2. Each correlation coefficient was averaged from measurements of three samples. parameters between them. The other correlation coefficients are <0.3000, indicating little similarity of raw materials and process parameters among the other three samples.

| Principal component analysis
In order to discriminate the samples from the five manufacturers, principal component analysis was performed. Figure   Furthermore, the identified compounds were determined by comprising the mass spectra with those in the mass spectrometry library.
The match ratios are above 80%, giving the positive answer of the existence of the compound. The types of compounds identified are similar to those obtained from fermented stinky tofu (Liu et al., 2009). A total of 24 typical volatile compounds were identified in Luo sample by comparing the mass spectra with those in the mass spectrometry library, while 23 typical volatile compounds were identified in Wang sample. There are 9 same common components in the two samples, including ethanol, acetic acid, propionic acid, butyric acid, phenol, 4-methylphenol, diethylene glycol ethyl ether, indole, and 3-methylindole, determining the great similarity between the fingerprints of the two samples. From Table 3, which lists the common components in the five samples, it can be found that indole and 3-methylindole, with very strong unpleasant odors, typical volatile flavor compounds of southern stinky tofu brine, exist in all the brine samples. Phenol and 4-methylphenol, as both flavor compounds and bactericides, can also be found in all the five brine samples.
The different compounds were summarized in Table 4, which may be due to the differences between the manufacturing processes. The different compounds are esters, alcohols, sulfides, organic acids, aldehydes, and ketones. The ester compounds can impart bines with fruity notes and make the odor of brine lifting and diffusive. The formations of alcohol compounds may be due to the fermentation of carbohydrates from soybean during the ripening step, when the sulfide compounds arise from the degradation of amino acids containing sulfur. The ester, alcohol, aldehyde, and ketone components may give the different brands of the southern stinky tofu brines different fruity and sweet odors.
However, the aroma intensities of indole and sulfides exceed their aroma intensities, and they give the brine its very strong unpleasant odor.

| CON CLUS ION
The fingerprints of organic volatile flavor compounds in southern stinky tofu brine samples from five manufacturers were studied using HS-SPME/GC-MS with the aid of chemometric methods. The fingerprints were obtained by HS-SPME/GC-MS and analyzed with the time shift alignment method, Shannon entropy, correlation coefficient, and principal component analysis. The results show that the time shifts in the samples can be accurately corrected with the time shift alignment method despite unexpected interferences. The fingerprint information was evaluated by Shannon entropy, while the similarities and differences in the fingerprints were investigated by correlation coefficient. Moreover, the identification of manufacturers was achieved by principal component analysis. The predominant volatile compounds in southern stinky tofu brine were indole, 3-methylindole, phenol, and 4-methylphenol.

ACK N OWLED G EM ENTS
This study was supported by National Natural Science Foundation of China (No. 31571819, 31601551, and 31671931) and the "1515 Talent Project" of Hunan Agricultural University.

CO N FLI C T O F I NTE R E S T
The authors notify that there are no conflicts of interest.

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