Biogenic amines analysis and microbial contribution in traditional fermented food of Douchi

Biogenic amines (BAs) have been reported to threaten the Douchi safety, while the BAs formation mechanism and corresponding control method have not been clarified for Douchi. The present study aims to investigate the microbial contribution to BAs in Douchi, and to find the beneficial strain for BAs control. Firstly, the BAs profiles of 15 Douchi samples were analyzed, and common 6 kinds of BAs were detected from different samples. All the samples showed the total BAs contents within the safe dosage range, while the histamine concentrations in 2 samples and β-phenethylamine in 6 samples were above the toxic level. Then, the bacterial and fungal communities were investigated by high-throughput sequencing analysis, and Bacillus and Candida were identified as the dominant bacteria and fungi genus, respectively. Furthermore, nineteen strains were selected from the dominant species of Douchi samples, including 14 Bacillus strains, 2 Staphylococcus strains, 1 Enterococcus strain and 2 Candida strains, and their BAs formation and degradation abilities were evaluated. B. subtilis HB-1 and S. pasteuri JX-2 showed no BAs producing ability, and B. subtilis GD-4 and Candida sp. JX-3 exhibited high BAs degradation ability. Finally, fermented soybean model analysis further verified that B. subtilis HB-1 and S. pasteuri JX-2 could significantly reduce BAs. This study not only contributed to understanding the BAs formation mechanism in Douchi, but also provided potential candidates to control the BAs in fermented soybean products.

complicated. Using the high BAs-degrading strains or the low BAs-producing strains as the starter cultures will be beneficial to control the Douchi BAs.
Generally, Douchi is produced by spontaneous fermentation with an uncontrolled manner, and various microbial communities are formed at different environment conditions [14][15][16] . The microbial diversities in commercial Douchi have been investigated using traditional culture method, denaturing gradient gel electrophoresis (DGGE) and high-throughput sequencing 16,[26][27][28] , and the Bacillus species were identified as the dominant bacteria in Douchi 11,14,15,27 . However, the microbial contributions on BAs formation in Douchi have not been clarified, and the efficient starter cultures for BAs control have not developed. In this study, BAs contents and microbial communities in Douchi samples from different regions were investigated, and the BAs producing and degradation abilities of representative strains were evaluated to elucidate the microbial contribution and select beneficial candidate for BAs control.
Illumina MiSeq sequencing. The total DNA was extracted from each Douchi sample by a SDS-based DNA extraction method previously described 29 . DNA quality was monitored by agarose gel electrophoresis, and stored at −20 °C for subsequent analysis. Primers of 515 F (GTGCCAGCMGCCGCGGTAA) and 806 R (GGACTACHVGGGTWTCTAAT) were designed according to the V4 region of bacterial 16S rRNA gene, and ITS1 (TCCGTAGGTGAACCTGCGG) and ITS2 (GCTGCGTTCTTCATCGATGC) were designed based on the ITS1 region of the fungal internal transcribed spacer (ITS).
The Miseq sequencing for bacteria and fungi followed the protocols described by Caporaso et al. and Kozich et al. 30,31 , respectively. The bacterial 16S rDNA and fungal ITS genes fragments were amplified using the total DNA, primers, and Thermo Scientific ® Phusion High-Fidelity PCR Master Mix (New England Biolabs, UK), purified by the QIAquick PCR Purification Kit (QIAGEN, Germany). Then, purified amplicons were employed for DNA library construction using the TruSeq ® DNA PCR-Free Sample Preparation Kit (Illumina). At last, DNA libraries were evaluated using Qubit@ 2.0 Fluorometer (Thermo Scientific) and real-time PCR, then sequenced on the MiSeq platform at Beijing Novogene Bioinformatics Technology Co., Ltd. All experiments were conducted in three replicates.
Bioinformatics analysis. Raw sequences generated through MiSeq sequencing were merged using fast length adjustment of short reads (FLASH) 32 , and low-quality sequences were discarded using QIIME. Clean paired sequences retained for each sample were analyzed using the UPARSE pipeline to generate operational taxonomic units (OTUs) and select representative sequences at 97% similarity 33 . PyNAST alignment and ribosomal database project (RDP) assignment were carried out based on the latest Green genes database 34 . Resampling was performed according to the minimum sequence numbers across all samples before the downstream analysis. In addition, the community compositions were provided at different taxonomic levels.
Isolation of strains from Douchi samples. Douchi samples (5 g) were added into 45 mL sterile water, rotated at 37 °C and 140 rpm for 40 min. The liquid mixtures were diluted, and spread onto the LB plates, MRS plates added with 20 g/L CaCO 3 and SDB plates added with 50 mg/L rifampicin, respectively. The LB and SDB plates were cultured at 37 °C and 30 °C to isolate the dominant bacteria and fungus strains, respectively, and the MRS plates were incubated at 37 °C in a vacuum bag to select the lactic acid bacteria.
Evaluation of BAs production and degradation ability. The BAs production ability was assessed by culturing the strains in 5 mL LB (Bacillus and Staphylococcus), MRS (Enterococcus) or SDB (Candida) medium added with 1 g/L of histidine, tyrosine, tryptophan, phenylalanine, ornithine monohydrochloride, lysine or agmatine sulfate salt. After incubating for 48 h, the resulting BAs were determined. To measure the BAs-degrading ability, the cells were cultured and collected by centrifugation at 6000 × g for 5 min. After washing with 0.05 mol/L phosphate buffer (pH = 7), the cell pellets were diluted to OD600 = 0.4 in phosphate buffer (0.05 mol/L) containing 100 mg/L of histamine, tyramine, tryptamine, β-phenethylamine, putrescine and cadaverine, cultured for Fermented soybean product model analysis. The fermented soybean model was applied to compare the BAs-controlling capacities of different strains. The cells (OD600 = 2) were inoculated into the soaked soybeans (30 g) with the inoculum size of 5% (v/w), incubated for 7 days, and the total BAs contents were measured by HPLC. The soybeans were fermented by inoculating with B. subtilis HB-1, S. pasteuri JX-2, B. subtilis GD-4 and Candida sp. JX-3, respectively, and the soybeans added with equivalent sterile water was set as the control.
BAs determination. The BAs contents were detected based on a previous report 20 . Briefly, 1.5 mL of 0.4 M HClO 4 was added into 0.5 g solid sample or 0.5 mL liquid sample, extracted for 1 h. After centrifugation at 12,000 × g for 10 min, the supernatant (250 μL) was added with 25 μL of 2 M NaOH and 75 μL of saturated NaHCO 3 , then reacted with 500 μL of 5 mg/mL dansyl chloride at 50 °C for 45 min. After that, the reactant was mixed with 25 μL of 25% NH 4 OH and incubated at 50 °C for 15 min to remove the residual dansyl chloride. Then the mixture was adjusted to 1.5 mL with acetonitrile and centrifuged at 2500 × g for 5 min. The supernatant was filtered through a 0.22 μm membrane for HPLC (high-performance liquid chromatography) analysis, which was carried out using an Agilent 1260 HPLC system with an Agilent column Zorbax Eclipse XDB-C18 (4.6 mm × 250 mm, 5 μm) at 30 °C. The separation was achieved using a linear gradient of mobile phase A (acetonitrile) and B (H 2 O) at a flow rate of 1 mL/min. The solvent gradient was as follows: 50% A (0-3 min), 50-90% A (3-20 min), 90% A (20-29 min), 90-50% A (29-32 min), 50% A (32-35 min). The detection was carried out at 254 nm. Each BA in the samples was quantified with a calibration curve generating by analyzing standard BA solution. Typical chromatograms are shown in Figure S1. Statistical analysis. Alpha diversity including Chao 1, Shannon diversity indices and Goods coverage were subjected to statistical analysis using Qiime (Version 1.7.0). The community structure was statistically analyzed at different classification levels. The significance was analyzed by one-way ANOVA using the statistical software SPSS 20.0, and means were compared by Duncan's multiple range test at 5%.

Results and Discussion
BAs contents in Douchi samples. BAs were widely detected in various soybean foods. Kim et al. detected six BAs in fermented soybean products, natto 4 , and Gong et al. reported that tryptamine, putrescine and cadaverine were detected in Douchi samples 19 . In this study, six common BAs with different contents were found among the Douchi samples (Table 1), including histamine (1.43-213.13 mg/kg), tyramine (1.37-64.33 mg/kg), tryptamine (0.86-99.31 mg/kg), β-phenethylamine (1.47-191.24 mg/kg), putrescine (2.65-171.00 mg/kg), and cadaverine (0.49-397.70 mg/kg), and the total BA contents ranged from 5.09 to 679.46 mg/kg. Generally, total BA contents in food are expected to be not more than 900 mg/kg 3,35,36 , and they may cause serious harm to human health when the total contents exceed 1000 mg/kg 24 . Our results indicated that no sample was above the harmful level (900 mg/kg) in the total content level, and as-detected samples were in the relative safe level on the whole. However, different BA showed different toxic level for human, thus each BA is needed to be further analyzed for comprehensive safety assessment.
As shown in Table 1, histamine was detected in five Douchi samples, the histamine concentration of two samples were higher than the allowable limit (50 mg/kg) suggested by the US Food and Drug Administration 3,13 . And 100-800 mg/kg of tyramine and 30 mg/kg of β-phenylethylamine have been reported to be the toxic doses   attributed to natural occurrence of amine in soybean products 7 . Meanwhile, cadaverine was detected in twelve products, while the concentrations of cadaverine in two samples were higher than 200 mg/kg. The toxicity of putrescine and cadaverine is not obvious relatively, but they can interact with nitrite to form carcinogenic nitrosamine 38 .
Microbial communities among the Douchi samples. As described above, BAs contents in Douchi samples showed relatively large variations, which was probably due to the different microbiological compositions. Therefore, it is important to analysis the microbial communities among the Douchi samples. By high-throughput sequencing, 3,514,240 16S rRNA (V3-V4 regions) and 3,402,964 ITS2 reads were generated from 45 sequencing samples. After quality filtering and chimera removal, 62,603 bacterial and 74,327 fungal sequences were obtained from each sample, with the average read length of 426 bases for bacteria and 211 bases for fungi. Community richness and diversity were assessed using four alpha-diversity metrics (observed OTU, Chao1, Shannon and Good's sample coverage) ( Table 2). Chao1 is an estimator of species richness, and Chao1 values indicated that all the Douchi samples possessed good sample richness. The Shannon diversity index, a measurement of overall diversity, indicated the diversified microbiota throughout the samples, and the Good's coverage result, an estimator of completeness of sampling 39 , highlighted good overall sampling, with levels of 99%. As to bacteria, Firmicutes was the predominant phylum in most Douchi samples, while Proteobacteria was only dominant in the sample of SCCD (Fig. 1a). Yang et al. found Firmicutes and Proteobacteria were major phyla in Douchi fermentation 16 . At the genus level, Bacillus was the dominant bacterial genus in most Douchi samples (Fig. 1a), and it was also reported as the predominant genus in various fermented soybean products [40][41][42] . In addition, the genera of Weissella, Enterococcus, Pantoea and Tetragenococcus were also the major genera in some Douchi samples, which was consistent with previous reports 8, 16,43,44 . In many fermented soybean products, Bacillus and Tetragenococcus played the crucial role in production of nutrients and flavors 26,45 . Figure 1b showed the difference in fungal abundance among samples at phylum level. Ascomycota was the most popular phylum in twelve Douchi samples, and this result was similar to previous studies 16,44 . The heatmap revealed differences in fungal genera among different Douchi samples (Fig. 2b). Candida was the most prevailing genus in JXJJ, SXHZ and CQ Douchi samples. Zhang et al. reported Candida was the dominant genus in Douchi 44 . During Douchi fermentation, Candida species grew more quickly than other fungi under high contents of organic acids and salts, then became the dominant fungal species 46 . Additionally, the genera of Millerozyma, Xeromyces, Lichtheimia, Issatchenkia, Agaricus, Alternaria, Wallemia and Blastobotrys were dominant in some Douchi samples. In a previous study, Lichtheimia was also detected as the major genus in Douchi 16 . Garcia et al. found that Lichtheimia produced β-glucosidase during solid-state fermentation 47 , which contributed to the release of glucose to improve the flavour of Douchi.
To further investigate the microbial contribution to BAs accumulation in Douchi, we evaluated the BAs producing and degradation abilities of these isolated strains. The BAs producing abilities of all isolated strains were detected in medium supplemented with corresponding precursors, and the BAs produced by different strains were shown in Table 4. Most of the isolated strains (79%) could produce putrescine via agmatine, while none could synthesize putrescine from ornithine. Only a few strains could generate histamine, tyramine, tryptamine, β-phenethylamine or cadaverine from corresponding substrates. Thereinto, B. subtilis HB-1 and S. pasteuri JX-2  Table 3. 16S rDNA and ITS sequences similarities of isolated strains with representative microbe.   showed no ability to produce six common BAs, while E. faecium GX-1 produced the highest concentration of total BAs, especially the relative high level of tyramine. In addition, E. faecium also exhibited the ability of generating β-phenethylamine, which was consistent with a previous study 53 .
The BAs-controlling capacity of selected strains in fermented soybean model. For the 19 strains, B. subtilis HB-1 and S. pasteuri JX-2 showed no BAs producing ability (Table 4), and B. subtilis GD-4 and Candida sp. JX-3 revealed high degradation rates for BAs (Table 5). Those four strains were probably the beneficial candidates for controlling the BAs, and thus their BAs-controlling capacities were evaluated in a fermented soybean model. After fermentation for 7 days, the total BAs content of the control sample without inoculation reached    284.88 mg/Kg (Fig. 3), which might be due to the native microbes in soybean. After inoculation with B. subtilis HB-1 and S. pasteuri JX-2, the total BAs concentrations were much lower than that of control (Fig. 3). Meanwhile, the highest BAs content (432.72 mg/kg) was detected in the soybean inoculated with Candida sp. JX-3. These results might be due to their different BAs producing ability. Among the four strains, B. subtilis HB-1 and S. pasteuri JX-2 had no BAs formation ability, with Candida sp. JX-3 exhibiting the highest BAs formation ability (367.2 mg/L). The B. subtilis HB-1 and S. pasteuri JX-2 with no BAs producing ability were more beneficial to reduce the BAs accumulation in fermented soybean product, and their existence probably inhibited the growth of BAs-producing strains by competition for nutrients, which might be the main reason for the decreased BAs. Eom et al. also observed that B. subtilis isolated from a Korean traditional fermented food was able to inhibit BA production 50 . In addition, Staphylococcus xylosus was used as a starter culture to inhibit BA production in a salted and fermented anchovy 55 . Similarly, the B. subtilis HB-1 and S. pasteuri JX-2 isolated in this study can be used as potential candidates for BA control in fermented soybean.

Conclusion
In summary, the BAs profiles of 15 representative Douchi samples were analyzed and the correlation between microbes and BAs were also investigated. The total BAs contents of all the Douchi samples were within the acceptable dose level, while histamine and β-phenethylamine showed above the toxic level in some samples, indicating that evaluation and control of specific amine are important for Douchi safety. High-throughput sequencing showed that Bacillus and Candida were the dominant genera. Nineteen strains were isolated from Douchi samples to investigate microbial contribution to BAs accumulation. B. subtilis HB-1 and S. pasteuri JX-2 showed no BAs formation ability, and B. subtilis GD-4 and Candida sp. JX-3 revealed high BAs degradation abilities. Moreover, fermented soybean model further confirmed that B. subtilis HB-1 and S. pasteuri JX-2 would be the potential candidates for BAs control. This study not only explained the microbial contribution to BAs in Douchi, but also provided potential strains to reduce the BAs in fermented soybean products.
Data availability. All relevant data are within the paper.