Microbial Community Succession and Metabolite Changes During Fermentation of BS Sufu, the Fermented Black Soybean Curd by Rhizopus microsporus, Rhizopus oryzae, and Actinomucor elegans

BS Sufu is a fermented food that is made by mixed black soybeans and soybeans. Microbial communities and metabolites play an important role for the final product. We characterized microbial diversity of BS Sufu during fermentation by high-throughput DNA sequencing. Meanwhile, volatile compounds were investigated by solid-phase microextraction (SPME) coupled with gas chromatography–mass spectrometry (GC-MS). The results showed that bacterial diversity was higher than that of fungi in BS Sufu. We found the existence of bacterial and fungal core communities, including Enterococcus, Enterobacter, Rhizopus, and Monascus. Network analysis indicated that bacterial and fungal communities maintain positive and negative interactions, which are important to shape the resident microbial communities in Sufu. In addition, 17 free amino acids (FAAs) were detected at the post-fermentation stage, and umami amino acid mainly contributed to taste of BS Sufu. Furtherly, a total of 79 volatile constituents in BS Sufu, including nine alcohols, 31 esters, and four aldehydes, form synergistically the unique odor of Sufu. Additionally, the correlations between microbiota and metabolites were analyzed. Our results suggested that these microbial taxa and metabolites contribute to the taste and flavor of BS Sufu. This study provided information for analysis of BS Sufu at different fermentation periods in terms of the microbial diversity and metabolites, and this information was important to understand the properties of mixed soybeans Sufu.


INTRODUCTION
Sufu is a traditional fermented soybean food in China that shares similar shapes and fermentation mechanisms with cheese, but the taste and flavor are different . Sufu is normally consumed as a flavor enhancer and appetizer owing to its characteristic flavor, pleasant taste, and nutritional value in China and other Asian countries (Xie et al., 2018). Based on the strain used as a starter, Sufu can be classified into three different types, including mold-fermented Sufu, bacteria-fermented Sufu, and naturally fermented Sufu (Feng et al., 2013). The nutrition and taste of Sufu are affected by many factors, for example, the type of raw materials and the production process. Current studies on Sufu mainly focus on optimization of technological conditions, determination of flavor components, and microbial analysis during fermentation process, but there are few studies on the production of Sufu with various beans as raw materials. The content of protein and essential amino acids (EAAs) in black beans is higher than that of soybeans, and the content of phenolic substances is very rich (Meng et al., 2020). Therefore, we decided to use mixed black beans and soybeans (BS) as raw materials to make Sufu by processing technology of mold-fermented Sufu with minor modifications, which can promote nutrition diversification of traditional Sufu.
Sufu is produced under open or semi-open conditions via the actions of various microbes, including starter and indigenous microorganisms, which play important roles in flavor generation (Huang et al., 2018). At present, highthroughput sequencing (HTS) provides a reliable method to comprehensively describe the dynamics of microbial communities, and these studies have expanded our knowledge of the microbial community structure in Sufu. Recent published data indicate that Sufu microbiota showed a prevalence of bacteria genus belonging to Lactococcus, Streptococcus, Enterobacter, and Acinetobacter (Wan H. F. et al., 2020); and fungi such as Trichosporon and Aspergillus were also described. In addition, microbial abundances are mostly correlated with chemical characteristics, such as salinity, ethanol content, and environmental factor. Moreover, metabolomic approaches, such as gas chromatography (GC), mass spectrometry (MS), and high-performance liquid chromatography (HPLC) have been applied to determine the metabolite profiles of Sufu (Moy and Chou, 2010). However, few systematic studies have evaluated the correlation on the microbial compositions and metabolites of Sufu at different fermentation stages; so far, there is no study on the characteristics of mixed beans Sufu. Accordingly, in this study, we aimed to explore the diversity and composition of the microbial community, the profile of metabolites, including amino acids and flavor compounds, and the correlation between microbes and metabolites in BS Sufu. The results will help us to improve understanding of the role of microorganisms in the production of Sufu, which will provide a certain reference for research and development of a new type of Sufu.

BS Sufu Preparation and Sample Collection
The isolated strains Rhizopus microsporus SE-3 and Rhizopus oryzae CD-1 (derived from spontaneous fermented Sufu in our previous study) combined with Actinomucor elegans CCIC3093 (China Center of Industrial Culture Collection) of 1:1:1 ratio were used as the starter for inoculation and fermentation. Black beans (Jihei 6) and soybeans (Heihe 43) (Jiansanjiang, Heilongjiang, China) of 1:1 ratio were used as raw materials to prepare Sufu, which is called BS Sufu. Simply, pehtze is prepared by inoculating mixed strain on the surface of tofu cubes, then the pehtze is salted for 24 h, and the salted pehtze is dispensed into wide-mouthed bottles and ripened in the dressing mixture for 60 days. The dressing mixture of BS Sufu mainly consists of food-grade monascus pigment, yellow rice wine, high-concentration liquor, and other spices (Supplementary Figure 1). Sufu samples were collected periodically at different stages of fermentation: pehtze (BS_M), salted pehtze (BS_Y), and post-fermentation 10 days (BS_10), 30 days (BS_30), and 60 days (BS_60). The samples were collected and stored at −80 • C.

Volatile Flavor Composition Analysis of BS Sufu Through Headspace Solid-Phase Microextraction-Gas Chromatography-Mass Spectrometry
Individual samples (5 g) were mashed and placed into 20ml headspace vial. Then, the vial was sealed after adding isoamyl phenylacetate (Solarbio, Beijing, China) and heated in water bath at 60 • C for 20 min. After that, the solidphase microextraction needle (57348-U, Supelco, Bellefonte, Pennsylvania, United States) was inserted into the vial to extract for 30 min (the needle being aged at 250 • C before use) and immediately inserted into the GC inlet, followed by 3 min of desorption. Subsequently, the GC-MS analysis of volatile compounds was performed on Agilent 7890A-5975C GC-MS equipped with HP-5MS chromatographic column (60 m × 250 µm × 0.25 µm). High-purity helium (purity > 99.999%) was used as the carrier gas at a constant flow of 1.5 ml/min, and the inlet temperature was set at 250 • C. Oven starting temperature was 50 • C, programmed at 5 • C/min to 100 • C for 2 min, programmed at 4 • C/min to 140 • C for 1 min, programmed at 4 • C/min to 180 • C for 2 min, and finally programmed to 250 • C at 5 • C/min and held at 250 • C for 5 min. The conditions of MS were ion source temperature 230 • C, quadrupole temperature 150 • C, electron ionization (EI) 70 eV, and full scan 35∼550 da.

DNA Extraction and High-Throughput Sequencing
Total genomic DNA was extracted using an Ezup Column Genomic DNA Purification Kit (Sangon Biotech, Shanghai, China), according to the manufacturer's protocol. The quality and quantity of the extracted DNA were assessed by the ratios of 260 nm/280 nm and 260 nm/230 nm with a NanoDrop spectrophotometer (A360, AOE, Shanghai, China). The V3-V4 region of the bacterial 16S rRNA gene and ITS1 region of the fungi were amplified using TransStart Fastpfu DNA Polymerase (TransGen, Beijing, China). Then PCR products were purified using the SanPerp Column PCR Product Purification Kit (Sangon Biotech, Shanghai, China). The 16S rRNA and ITS1 gene amplicons were sequenced using Illumina HiSeq deep sequencing (Majorbio, Shanghai, China).

Bioinformatics Analysis
The double-end sequence data obtained by HiSeq sequencing were spliced or merged into a sequence tag, and quality control filtering was performed on the quality of reads and the effect of the merging. Then, the stitching sequence obtained is the original tag data. The read sequences were aligned with species annotation databases through the UCHIME Algorithm. Then, the chimera sequence was identified and removed to obtain the final valid data. Usearch software was used to cluster the tags at a similarity level of 97% to obtain operational taxonomic units (OTUs) (Hang et al., 2014). Alpha diversity was evaluated using the Mothur v.1.11.0 program, and Kyoto Encyclopedia of Genes and Genomes (KEGG) function of 16S RNA gene was annotated by Tax4Fun and Silva database. Ultimately, redundancy analysis (RDA) was performed using Canoco software to reflect the relationship between the core microorganisms and flavor compounds (Barka, 2019).

Statistics Analysis
SPSS 22.0 was used to calculate the significance of difference by ANOVA; the results were expressed as the mean ± SD.

Sequence Accession
The datasets (raw sequences) generated and analyzed during this study have been submitted to National Center for Biotechnology Information (NCBI) Sequence Read Archive (SRA) repository with accession number PRJNA693985 and can be accessed through the following link: http://www.ncbi.nlm.nih.gov/sra/ PRJNA693985.

Free Amino Acid Composition and Content of BS Sufu
Seventeen FAAs were detected in BS Sufu at the postfermentation stage ( Table 1). The content of all FAAs reached the highest in BS_60, except for ARG. FAAs have the characteristics of low taste threshold and strong taste ability. According to the taste characteristics of FAAs, they are grouped into three categories, including umami amino acids (UAAs), sweet amino acids (SAAs), and bitter amino acids (BAAs). The content of total SAAs and UAAs accounted for 62.5%, and total BAAs accounted for 37.5%. The content of seven EAAs was relatively stable, maintaining about 37% in post-fermentation stages. The taste intensity is affected by both the content of the taste substance and the threshold value. TAV is the ratio between the concentration of the taste substance and its threshold value. The value (TAV > 1) can be considered as the food taste contribution: the greater the TAV value, the greater the contribution of FAAs to taste. In the BS_10 stage, FAAs with TAV > 1 included ASP, GLU, ALA, VAL, and HIS; FAAs in the BS_30 stage includes ASP, GLU, PHE, ALA, VAL, LYS, and HIS; the BS_60 stage includes ASP, GLU, PHE, ALA, VAL, MET, ILE, TYR, LYS, and HIS. In total, the kinds of BAAs were significantly higher than those of UAAs and SAAs, but the TAV value of UAAs was significantly higher than that of other FAAs, indicating that UAAs contribute to the taste of BS Sufu.

Volatile Flavor Composition Analysis of BS Sufu
A total of 79 volatile constituents in BS Sufu at the postfermentation stage, including nine alcohols, 31 esters, four aldehydes, seven ketones, two acids, eight olefins, seven alkanes, three ethers, and eight other organic compounds, form synergistically the unique odor of Sufu ( Table 2). These volatile compounds in three fermentation stages were different. Compared with BS_30, BS_60 had fewer kinds of flavor compounds, but the content was higher. For example, in the BS_60 stage, the total amount of alcohols and esters compounds was significantly higher than that of the BS_30 stage. Esters accounted for the highest proportion in all compounds identified of BS Sufu, which mainly contribute to the complex flavor of BS Sufu. The relative content of ethyl caproate, isoamyl phenylacetate, ethyl palmitate, ethyl linoleate, and ethyl alcohol was higher than that of other flavor compounds. The relative content of ethyl alcohol and phenethyl alcohol was higher among the alcohol components. In addition, the aldehyde, ketone, alkane, and olefin compounds were also identified, but the content was extremely low, which contributes less to the flavor of BS Sufu.

Microbial Community Dynamics of BS Sufu During Fermentation
After sequencing and quality control, the clean reads of all samples were clustered, revealing 346 and 168 OTUs of bacteria and fungi at 97% similarity level, respectively. Furthermore, alpha diversity was analyzed ( Table 3). The Shannon index of bacteria was generally higher than that of fungi (P < 0.05), which proved that bacterial diversity was higher than that of fungi in BS Sufu. At the end of fermentation, the Shannon index of bacteria and fungi were reduced. Likewise, abundance-based coverage estimator (ACE) index indicated that the abundance of bacterial species was significantly higher than that of fungi. The coverage of samples at different fermentation stages is greater than 99%, indicating that sequencing has a large coverage of species, and showed that the sequencing depth was sufficient, which proved the reliability of the sequencing results and could represent the true composition of microorganisms in BS Sufu.
Bacterial community dynamics were evaluated based on changes in relative abundance at phylum and genus levels ( Figures 1A,B). Firmicutes, Proteobacteria, and Bacteroides were the main phyla, which exist in the whole fermentation process. At the genus level, the abundances of Enterococcus and Enterobacter were higher. At the stage of post-fermentation, the abundance of Acinetobacter decreased from 23.91% to 5.74%, while the abundance of Enterobacter and Enterococcus increased to 19.92% and 53.67%, respectively. During the whole postfermentation process, the abundance of Enterococcus was the highest, the abundance of Bacillus increased, and Acinetobacter was almost undetectable.
The phylum of fungi included Ascomycota, Mucoromycota, and Basidiomycota ( Figure 1C). The Mucoromycota was mainly concentrated in the pre-fermentation stage, but the Ascomycota and Basidiomycota were concentrated in the post-fermentation stage. The distribution of fungi genus is shown in Figure 1D. The abundance of Rhizopus in the pre-fermentation stage is 77.96% in the BS_M stage and 93.35% in the BS_Y stage, while the abundances of Mucor and Actinomucor declined in the postfermentation stage. In general, the diversity of fungi elevated in the post-fermentation stage; for instance, Wallemia, Millerozyma, and Xeromyces begin to appear, and the abundance of Monascus was the highest in entire post-fermentation process. In summary, BS Sufu had significantly more bacterial species than fungi during the pre-fermentation stage, but the identified fungal species are slightly higher than bacteria during the post-fermentation stage.

Comparison of Microbial Communities in BS Sufu at Different Fermentation Stages
The Lefse analysis was performed to uncover differentially abundant biomarkers by an Linear discriminant analysis (LDA) score (log 10) > 4 (Supplementary Figure 2). The unclassified_Bacteria was significantly enriched in BS_M, with an LDA value of 4.8. Acinetobacter (5.1) and Chryseobacterium (4.5) were significantly enriched in BS_Y, which contributed to the difference between BS_Y and other samples. Only the LDA score of Pseudomonas (3.4) at BS_10 was calculated, indicating a Different letters indicate the significant difference (P < 0.05) of BS Sufu at different fermentation times; Threshold is the lowest taste concentration of amino acids; TVA, amino acid content/amino acid taste threshold; TFAA, total free amino acids; TUAA, total umami amino acid; TSAA, total sweet amino acid; TBAA, total bitter amino acid; EAA (*), essential amino acid.

Network Inference of Species in BS Sufu
The relationship between BS Sufu microorganisms based on the relevant network diagram was studied, and the correlation coefficient in different species was calculated (Figure 2). The correlation network diagram of genus level revealed that there were significant interactions between different genera of the BS microorganisms. In the bacterial network diagram, Acinetobacter Different letters indicate the significant difference between bacteria and fungi in a certain fermentation stage (P < 0.05). and Enterobacteriaceae had a significantly positive correlation, and the clustering coefficient was 0.504. The number of genera connected to Enterobacteriaceae and Staphylococcus was the largest (degree = 12) (degree represents the number of one node connected to other node), followed by Weissella, but the clustering coefficients of Weissella were small (0.348), and Pseudomonas and Lactococcus had the largest clustering coefficients of 0.669. There was an obviously positive correlation between Lactococcus and other bacterial genera. The clustering coefficient indicated the importance of this genus in the entire microbial environment.
In the fungi network diagram, Mucor has the biggest positive correlation with Actinomucor and Rhizopus in the pre-fermentation stage. The number of fungi genera interconnected with Millerozyma was the largest (degree = 15). Interestingly, Monascus added in the post-fermentation process is positively correlated with Rhizopus and Mucor. In addition, Zygosaccharomyces had a significantly positive correlation with other genera.

Correlation Analysis of Microbial Diversity and Metabolites of BS Sufu
Bacteria Pearson correlation heat map ( Figure 3A) indicated that all FAAs were positively correlated with Kurthia, unclassified_Enterobacteriaceae, Staphylococcus, Weissella, and   and (B) fungi. The figure shows 14 free amino acids with high content; * represents the correlation between free amino acids and species (***P < 0.001, **P < 0.01, *P < 0.05). Colors closer to red mean stronger positive correlation; colors close to blue means stronger negative correlation.
Klebsiella. Besides, all FAAs had obviously negative correlation with Acinetobacter, Gammaproteobacteria, Paenibacillus, and Streptococcus. The fungal correlation is shown in Figure 3B. All amino acids had an obviously positive correlation with Monascus, unclassified_Saccharomycetales, Wallemia, Millerozyma, and Xeromyces. Furthermore, Aspergillus and Zygoascus were also positively correlated with GLU, SER, PRO, etc.
Redundancy analysis revealed the correlation between dominant genus and flavor compounds (Figure 4). Enterococcus had a significantly positive correlation with esters, alcohols, ketones, alkanes, and ethers. Stenotrophomonas and Bacillus were significantly positively correlated with organic acids and aldehydes. Enterobacter and unclassified-Enterobacteriaceae were positively correlated with organic acids but negatively correlated with the other types of compounds. Unclassified-fungi had a significantly positive correlation with alcohols, esters, ketones, and ethers, which were concentrated in the end of postfermentation. Rhizopus and Apiotrichum were positively related with organic acids and aldehydes, but the two substances were concentrated in the early stage of post-fermentation. Monascus  was also positively correlated with alcohols, ethers, and ketones and negatively correlated with esters and organic acids.

Functional Prediction of Bacteria Gene
As shown in Figure 5, the function of bacterial genes was mostly annotated in metabolism (59.9%), genetic information processing (15.1%), environmental information processing (18.9%), cellular processes (2.8%), human diseases (2.5%), and biological systems (0.7%). Among them, carbohydrate metabolism accounted for 25.8% of metabolism, lipids 6.7%, amino acids 21.4%, and metabolism of terpenoids and polyketides 3.9%. The carbohydrate metabolism was more abundant in BS_M, BS_30, and BS_60. The amino acid metabolism was enriched in BS_Y, BS_10, and BS_60. In general, the genes involved in carbohydrate metabolism were more than those in amino acid metabolism. The function prediction showed that majority of the functional information of bacterial community in BS Sufu was meaningful.

DISCUSSION
The fermentation process of Sufu is complicated and affected by various factors (Liu et al., 2018). As a fermented food rich in protein, protein degradation and amino acid metabolism are extremely key in the formation of flavor substances. The content of FAAs depends on the ratio of their generation and degradation (Tang et al., 2018). ARG has the function of anti-tumor and regulating immunity (Hui et al., 2017;Mondanelli et al., 2017). LYS can increase the absorption and accumulation of calcium, accelerate bone growth, enhance immunity, and promote development (Korompokis et al., 2016). PHE can synthesize important neurotransmitters and hormones to improve memory in the body (Kumar et al., 2017;Souza et al., 2017). LEU can improve the recovery ability of muscle fibers and make thinking more agile; MET is involved in fat reduction and detoxification, as well as the synthesis of choline; THR has the function of enhancing immunity and transforming certain amino acids to achieve balance (Zhang et al., 2017). FAAs not only give BS Sufu a unique taste but also make the BS Sufu possess a certain physiological effect. The total amount of FAAs increases with the extension of BS Sufu fermentation time. In addition, through comparative studies, we found that the types and contents of various FAAs in BS Sufu are not significantly different from those of previous studies . However, due to the different fermentation strains used, there are certain differences in the enzyme systems during the fermentation process, which will lead to differences in the total amount of amino acids.
We identified 79 flavor constituents in BS Sufu. Ji et al., identified 35 volatile compounds in nine Sufu (Ji, 2020). Since the flavor components in Sufu are affected by raw materials, production technology, fermentation microorganisms, and environmental conditions, we believe that the effects of raw materials (black beans and soybeans) and microorganisms (three strains) make the BS Sufu rich in flavor substance diversity and higher content. Lipid oxidation gives a large number of volatile compounds or aroma precursors, which can subsequently react with amino acids derived from proteolysis to form the characteristic aroma of Sufu. The esters are the most common volatile compounds detected in BS Sufu. There are 31 kinds of ester, and their content increased with the extension of fermentation time. Esters are mainly synthesized by two pathways. One is the reaction of acyl-CoA with the corresponding higher alcohols under the action of ester synthase, and the other is the formation of organic acids and higher alcohols through esterification, but the synthesis speed of this pathway is very slow. Besides, acetyl-CoA is the precursor for the synthesis of fatty substances. Therefore, there is a certain competitive relationship between ester metabolism and lipid metabolism; but during the fermentation of BS Sufu, the abundance of carbohydrate metabolism is much higher than that of lipid metabolism (Figure 5). So, acetyl-CoA is mainly used for ester metabolism. Studies have also shown that hexanal can generate hexanol under the action of alcohol dehydrogenase, hexanol can generate hexyl acetate, and hexanal can be oxidized to hexanoic acid and can further generate esters, such as methyl hexanoate (Egea et al., 2014). Esters usually have a special aroma (Moy et al., 2012). The relative content of ethyl caprylate is 8.67% with brandy flavor, being mainly used for flavoring and spices. Ethyl phenylacetate is often found in liquor with a noticeable rose fragrance and honey-like smell. Ethyl myristate has the scent of iris oil, which is generated by esterification of myristic acid and ethanol. Ethyl caproate has a strong fruity and wine aroma and has the aroma of pineapple and banana (Kai et al., 2013). In addition, some esters also have very good pharmacological effects, such as linoleic acid ethyl ester, which has the same effect as linoleic acid in lowering cholesterol and blood lipid and can prevent or reduce atherosclerosis. Moreover, the side effect of ethyl linoleate is much lower than that of linoleic acid. BS_60 sample is rich in long-chain esters such as ethyl stearate, methyl palmitate, and ethyl linoleate. These esters improve the flavor of BS Sufu as well as health benefits. There are nine kinds of alcohols detected in BS Sufu. After 60 days of fermentation, the relative content of alcohols is significantly reduced, which may be the result of esterification or oxidation of alcohols. Alcohols often have botanical, aromatic, and earthy odors. Although their thresholds are high, they can form esters with organic acids to facilitate the formation of Sufu flavor (Wu et al., 2009). Several kinds of acids, aldehydes, and ketones were detected in BS Sufu. The aldehydes possess aromatic characteristics such as fresh fragrance, fruity and nutty fragrance, and low flavor threshold (Qi et al., 2012). Ketones are generally formed by fat degradation and oxidation reactions (Jonsdottir et al., 2008). The other volatile components detected in BS Sufu are mainly furans and pyrazines. Among them, 2-pentylfuran is mainly derived from the oxidation of linoleic acid or 2,4-decadienal, with a low threshold value and has a bean aroma (Lu et al., 2014).
The bacterial diversity of BS Sufu is higher than that of fungi. The composition of dominant bacteria in BS Sufu at different fermentation stages has a certain similarity, yet the composition of dominant fungi is quite different. Enterococcus and Enterobacter are present in the whole fermentation process, and their abundance is relatively high. Enterococcus with higher abundance was also found in Jiajiang Sufu (Wan H. F. et al., 2020). Similarly, Enterobacter and Enterococcus are widely present in various fermented foods, such as soy sauce, fruit juice, and tea. The abundance of the two genera in BS Sufu is relatively high, but we have measured the total number of coliforms, and the result is <3.0 MPN/g, which is lower than the national standard, indicating that the Sufu products are safe. Previous reports have shown that Lactococcus, Acinetobacter, Tetragenococcus, Pseudomonas, and Lactobacillus are the dominant bacteria and crucial contributors during the production of fermented foods, such as cheese, Sufu, soy sauce, liquor, and tea (Tanaka et al., 2012;Liu and Qiao, 2019). Similarly, Lactobacillus, Acinetobacter, and Pseudomonas were detected in BS Sufu. Acinetobacter and Chryseobacterium are abundantly enriched in BS_Y, but the amounts are significantly reduced in post-fermentation. Acinetobacter is also abundant in other fermented foods (Li et al., 2017;Huang et al., 2018). Previous studies have indicated that Acinetobacter is well known for its capacity to secrete esterolytic enzymes and is positively related to flavor compounds (Hassan et al., 2010), particularly esters.
Lefse difference analysis showed that Lactobacillus and Streptococcus were enriched in BS_M samples.
Lactobacillus was often used in cheese fermentation  and existed in other fermented products . Streptococcus had a strong ability to metabolize carbohydrates (Gefei et al., 2020). The genus is enriched in the pre-fermentation stage, which may be related to carbohydrates and protein metabolism. The Pantoea is higher during the pre-fermentation period, and 43% of Pantoea is distributed in the salted pehtze (Figure 1). Pantoea can produce volatile compounds such as p-hydroxyphenylethyl alcohol, methyl 4-hydroxybenzeneacetate, and certain antibacterial properties (Mingxi et al., 2012).
Compared with bacteria, the fungal community composition is quite different. The main genera in the pre-fermentation stage are the inoculated Rhizopus and Mucor. Inoculated strain requires the accumulation of a large amount of protease and other enzymes for fermentation (Guiliang et al., 2020). Rhizopus and Mucor can secrete a large amount of protease to decompose large molecular proteins. In BS-Y, Rhizopus rose to 93%, which indicated that Rhizopus has a strong resistance to high salt environments. In the post-fermentation stage, Monascus is the dominant genus because Monascus pigment is added. Millerozyma is a yeast that can degrade bio-amine amines and has a certain heat resistance. It is often found in fermented wines (Jiangang et al., 2018) and also used to degrade bio-amines purposefully during the fermentation of soy sauce (Cheng et al., 2019). Wallemia is a salt-tolerant fungus with metabolites that have an antibacterial activity, and its fermentation products are mainly pyrrolidones and sesquiterpenoids.
Kurthia has a significantly positive correlation with almost all amino acids (P < 0.001). Current research shows that Kurthia can consume glucose to produce lactic acid and can produce ALA, GLU, and GLY through autotrophic pathways (Fang et al., 2015). In addition, Staphylococcus and Enterobacteriaceae also showed a significantly positive correlation with all amino acids, which indicated their important roles in the production and transformation of amino acids in Sufu fermentation. Enterobacteriaceae comprises various genera at the taxonomic level, appearing dominantly in fermented foods, such as doenjang, Sufu, and soy sauce, which is essential for the production of EAAs and volatile compounds (Kim et al., 2009;Devanthi et al., 2018). All amino acids have a clearly positive correlation with Monascus (Figure 3), which proves that Monascus not only provides dyes in the post-fermentation stage but also plays a prominent role during the formation of amino acids. Mucor and Rhizopus are inoculated for pre-fermentation of BS Sufu; the enzymes produced by the two molds can act directly on protein of materials involved in the production of amino acids (Zhang Y. T. et al., 2020). In addition, Bacillus has obviously a positive correlation with organic acids and aldehyde compounds (Figure 3). Bacillus can produce highly active extracellular enzymes, such as protease, amylase, and glucanase. These complex enzyme systems provide a key impetus for the formation of primary and secondary metabolites. So Bacillus can produce unique characteristic flavors during fermentation process, such as spicy, sweet, honey, acetic, aromatic, ethereal, and creamy (Quan et al., 2005). Therefore, Bacillus played a role in the production of volatile compounds during fermentation of BS Sufu. Stenotrophomonas can use nitrogen sources to produce peroxidase and lignolytic enzymes during growth, which can decompose macromolecular substances such as cellulose (Olajuyigbe et al., 2018). The microorganisms that metabolize aromatic amino acids and form ester flavors in the postfermentation stage mainly include Staphylococcus, Enterococcus, Lactobacillus, and Pseudomonas. Monascus also has a certain correlation with amino acids, esters, alcohols, etc., while Rhizopus and Apiotrichum have a certain positive correlation with acids and ketone compounds.
Some research indicated that the number of genes in glycolysis/gluconeogenesis pathway is the most abundant in fermented foods, followed by citric acid cycle and oxidative phosphorylation (Procopio et al., 2013). In addition, carbohydrate and amino acid metabolism accounted for the highest proportion, followed by lipid and other metabolism. It proved that Enterococcus, Enterobacter, and other microorganisms with high abundance participate in the production of metabolites through different pathways. Although ester compounds have the highest correlation with Enterococcus, the flavor formation of BS Sufu is the result of joint action of all microorganisms in the fermentation system.

CONCLUSION
In summary, we examined the microbial community structure and flavor compounds of BS Sufu at different fermentation stages by coupling HTS and GC-MS. The microbial community and metabolites in BS Sufu at different fermentation stages are significantly different. The correlation between dominant microorganism and metabolites is extremely complex, which contributes to the taste and flavor of BS Sufu. Our work provides insights into the role of microbial communities in the production of mixed soybean Sufu, and this information is important to develop a new type of Sufu.

DATA AVAILABILITY STATEMENT
The datasets (raw sequences) generated and analyzed during this study has been submitted to NCBI Sequence Read Archive (SRA) repository with accession number PRJNA693985 and can be access through following links: http://www.ncbi.nlm.nih.gov/ sra/PRJNA693985.

AUTHOR CONTRIBUTIONS
DY and LX designed the experiments, performed most of the experiments, and wrote the manuscript. MW and XW performed the statistical analysis. KW and ZL provided valuable advice on experiments design and data analysis tools. DZ was the coordinator and guarantor, overseeing all aspects of this study.