Tomato Sauce were provided by Xinjiang University (Xinjiang, China). Methanol, toluene, from Tianjin Shengao Chemical Reagent Co., Ltd. (Tianjin, China). MRS from Hangzhou Best Biotechnology Co, Ltd. (Hangzhou, China). Lactobacillus fermentum (L. fermentum) CICC 21800, Lactobacillus acidophilus (L. acidophilus) CICC 6085, Lactobacillus reuteri (L. reuteri) CICC 6126, Lactobacillus plantarum (L. plantarum) CICC 21797, Lactobacillus paracasei (L. paracasei) CICC 22709, Lactobacillus casei (L. casei) CICC 6114, Lactobacillus delbrueckii subsp bulgaricus (L. bulgaricus) CICC 20247, Pediococcus pentosaceus (P. pentosaceus) CICC 21862, Pediococcus acidilacriti (P. acidilacriti) CICC 20720, Lactobacillus rhamnosus (L. rhamnosus) CICC 6135 were obtained from the China Center of Industrial Culture Collection (CICC) (Beijing, China). Pectinase, cellulase, and hemicellulase were purchased from Fibrochem Biochemicals Co., Ltd. (Shanghai, China). All strains were activated twice and incubated at 37°C for 24 h before use. Methanol and acetonitrile were supplied by Meguiar’s. The SOD kit was purchased from Beijing Solebo Technology Co., Ltd. (Beijing, China).

First, tomato paste was mixed well with water 1:2.6. The Total Soluble Solid (TSS) was adjusted to 12.5^°Brix, Then, sterilized in water at 90°C for 10 min and cooled down subsequently. Add 1.5% (w/v) pectinase, 1% (w/v) cellulase, and 1% (w/v) hemicellulase in the sample and enzymatically digest for 3 h at 57°C. Following, the samples were sterilized at 90°C for 30 min. After cooling, the strains were accessed into samples and placed under a Constant temperature oscillation incubator (ZD-85A, Changzhou, China, Setting 37°C) for 22 h of fermentation.

Ten strains (L. fermentum, L. acidophilus, L. reuteri, L. plantarum, L. paracasei, L. casei, L. bulgaricus, P. pentosaceus, P. acidilacriti, L. rhamnosus) of fermented tomato homogenate were used and fermented in an incubator at 37°C for 22 h. After fermentation, the best four dominant strains were selected based on pH, SOD, LYC, TSS and sensory scores. The optimal ratio of these 4 strains were determined by four factors, a10-level, uniform design table. The results were analyzed and calculated through SPSS 26 and Excel (Table 1).

After determining the fermentation strain category and ratio, the SOD activity value and LYC value obtained after fermentation were used as indicators. The one-way experimental conditions were set as follows Soluble solids content (10.5, 11.5, 12.5, 13.5, 14.5°Brix), Inoculum volume (1.2.3.4.5*10⁶ CFU/mL), Fermentation temperature (27.32.37.42.47°C), Fermentation time (14.18.22.26.30 h), with the above four factors as the independent variables, and SOD (Y1) and LYC (Y2) as the response values, to design a Box–Behnken test. Each factor has 3 levels (−1, 0, 1) and the design includes 5 centroids for a total of 29 sets of experiments. The factors and levels of Box–Behnken are tabulated in Table 2. Response surface results are shown in Table 3.

The activity of SOD was determined according to the instructions of the kit. LYC was determined using the national standard method. The soluble solid solids content was determined using a digital refractometer (TD-45, Zhejiang, China). pH was determined by a digital desktop acidimeter (PHS-3C, Shanghai, China). The sensory evaluation was performed on the fermented samples. Samples of the fermentation broth produced under different conditions were numbered, and then 10 people trained in the sensory evaluation were asked to rate the fermentation broth product’s sensory, flavor, acceptability, and hypothetical purchase intent for a total score of 90. The details were shown in Table 4.

Each sample (200 μL) was mixed with 800 μL of extraction solution (methanol: acetonitrile = 1:1, containing an isotope-labeled internal standard mixture) in an EP tube. The samples were then vortexed for 30 s, extracted by low-temperature ultrasound for 30 min (5°C, 40 KHz), and allowed to stand for 30 min at −20°C. The samples were then centrifuged again for 15 min (13,000 g, 4°C), and the supernatant was dried with nitrogen, followed by the addition of 120 μL of the compound solution (acetonitrile: water = 1:1). The sample was vortexed again for 30s, extracted by low-temperature ultrasonic extraction for 5 min (5°C, 40 KHz), centrifuged for 10 min (13,000 g, 4°C), and the supernatant was put into vials to be analyzed on the machine. In addition, 20 μL of supernatant from each sample was mixed as quality control (QC).

Metabolite detection was performed on an ACQUITY UPLC HSST3 column (100 mm × 2.1 mm i.d., 1.8 μm, Waters, Milford, USA) with two mobile phases: 95% ultrapure water +5% acetonitrile (containing 0.1% formic acid) and 5% ultrapure water +47.5% acetonitrile +47.5% isopropanol +5% ultrapure water (containing 0.1% formic acid). The two mobile phases were 0.1% formic acid, the temperature of the autosampler was 40°C, and the injection volume was 3 μL. The ion source was electrospray ionization (ESI), and the ionization modes: positive (ESI+) and negative (ESI-) scanning modes were used to collect mass spectrometry signals. The scanning range was 70 ~ 1,050 m/z; the intrathecal gas flow rate was 50 Arb, the auxiliary gas flow rate was 13 Arb, and the capillary temperature was 325°C. The resolution of the full mass spectrum was 60,000, and the resolution of the MS2 was 7,500. The collision energies were 20, 40, and 60 eV.

Data were expressed as mean plus or minus standard deviation and analyzed using SPSS26 software [In SPSS 26, we use the one-way ANOVA test to compare means. Statistical significance (p < 0.05; Duncan test)]. Plotting was done using Origin 2021 software. Multivariate analysis was performed using SIMCA 16.0 software. Principal Component Analysis (PCA) and Partial Least Squares Discriminant Analysis (PLS-DA) were performed to reduce the dimensionality of the data by visualizing sample distribution and grouping. Subsequently, supervised Orthogonal Projection-Least Squares Discriminant and Applied Analysis (OPLS-DA) of the underlying structure was performed, and additionally, to assess the importance of the variables in the projection, the importance of the projected variables of the initial principal components (VIP) was determined in Orthogonal Partial Least Squares Discriminant Analysis (OPLS-DA). This technique allows for summarizing and determining the contribution of each variable. Metabolites with VIP > 1 and p < 0.05 were considered as significantly changed metabolites.

Each strain has its characteristics during the fermentation process, so it is important to consider the characteristics of each strain when selecting the appropriate strain for tomato juice. The effect of multi- strain association on fermentation is more beneficial than single-train fermentation due to the reciprocal symbiotic interaction between microorganisms and products derived from mixed fermentation containing more metabolites (19). In this experiment, 10 strains were used for fermentation, SOD, LYC, pH, TSS, and sensory during fermentation were determined, and finally four strains were selected for mixed fermentation. The results were shown in Table 5.

SOD is an important component of the antioxidant enzyme system in biological systems and has a strong antioxidant ability. SOD is not only naturally present in fruit and vegetable juices, it is also produced during the metabolism of LAB (20). After the fermentation of tomato juice, the SOD activity increased in all groups. Compared to the original solution producing, the L. casei, L. plantarum, L. fermentum, L. paracasei increased the SOD by 41.44, 35.63, 49.02, and 36.85 U/g, respectively. Explanations for these differences may include strain-specific functions, differences in growth rates and environments, and differences in carbon source utilization.

Xinjiang tomato fruits contain higher levels of LYC than those from other regions (21). LYC has been shown to have beneficial effects on human health. The LYC content produced by different strains varied. P. pentosaceus, L. fermentum, P. acidilactici,. L. plantarum was increased by 9.7, 8.2, 6.0, and 5.8%, respectively, compared to the unfermented samples. It may be due to fermentation hydrolysis and the destruction of the cell walls of tomato cells by LAB during the fermentation process. The destruction of the cells leads to a decrease in the pectin content and reduces the size of the tissue fragments, which reduces gravitational separation and promotes the release of LYC, thereby increasing the content (22).

The proportion of flavoring substances influences affects the sensation. The decrease in pH during fermentation may be due to the production of organic acids, and the absorption of a portion of glycogen by microorganisms, which leads to a decrease in TSS and destroys the flavor of the fermented juice. The highest sensory evaluation scores were given by L. fermentum, L. plantarum, P. pentosaceus, and L. paracasei, with scores of 90.33, 89.00, 88.33, and 87.66, respectively. For these results, the differences are due to the unique properties of each strain resulting in fermented samples with different acidic flavors and different organic acid compositions that give each sample a unique flavor. Besides, the volatile compound composition of the FTJ varies from strain to strain, and the senses will be different (23).

Combined fermentation of multiple strains can make up for the defects of other single strains in fermentation. Based on the three factors of high SOD, LYC, the best four strains, P. pentosaceus, L. casei, L. plantarum, and L. fermentum were selected for mixed fermentation. The optimum percentage of each strain was determined using a homogeneous design and the results are shown in Table 6, and the equation follows:

R² = 0.999 and p < 0.01, indicating that the equation is well-fitted and accurately predicts optimal conditions. In conclusion, the optimal inoculation rates of FTJ were predicted to be P. pentosaceus (53.79%), L. casei (13.17%), L. plantarum (19.87%), L. fermentum (13.17%), with a corresponding SOD activity of 525.76 U/g. Based on the optimal inoculation rate described above, the total strain inoculum of the four strains was maintained at 2 × 10⁶ CFU/mL and the TSS was maintained at 13.5^°Brix. Fermentation for 22 h at 37°C. In the validation test, SOD activity reached 486.35 U/g.

Different fermentation environments have different effects on the results and to achieve the best product, it is necessary to optimize the fermentation conditions. According to the preliminary experiments showed no significant changes in pH and TSS content. Therefore, SOD and LYC were used as dual response values for the optimization of fermentation conditions.

The metabolic activity of LAB is affected by fermentation temperature, thereby affecting the fermentation quality (24). As shown in Figure 1A, the SOD activity is 488.78 U/g at 37°C, then decreased with increasing temperature. LYC also reached a maximum value of 73.13 μg/g at 37°C, after which it gradually decreased. The appropriate growth temperature for LAB is generally around 37°C, so the decrease in both SOD activity and LYC content may be due to over-high temperature. The intolerance of bacteria to heat can lead to a decrease in enzyme metabolism and activity (25). Hence, the optimal fermentation temperature was set at 37°C.

The SOD activity and LYC content were affected by the fermentation time as shown in Figure 1B. When the fermentation time is less than 14 h, the growth and reproduction of LAB were insufficient, metabolites were not accumulated enough, and the SOD activity and LYC content were low. If the time is too long, it will lead to a decrease in SOD activity and LYC content because insufficient nutrients will lead to an increase in the number of dead bacteria and an excessive accumulation of metabolic wastes. SOD activity increased rapidly between 14 h and 22 h and was 493.06 U/g at 22 h, afterwards decreased. The LYC reached a maximum value of 72.20 μg/g at 26 h and then decreased with increasing fermentation time. Therefore, the optimal fermentation time is 22–26 h.

As shown in Figure 1C, the SOD activity was (486.95 U/g) when the TSS content was adjusted to 13.5^°Brix, and the LYC reached 71.14 μg/g at a TSS content of 11.5^°Brix. From these results, we chose the range of 11.5^°Brix-13.5^°Brix.

From Figure 1D, it can be observed that When the strain number concentration was 10⁶ CFU/mL, the bacterial population was insufficient and the bacterial metabolic rate was slow. It induced a decrease of SOD activity and LYC. However, the values of the indicators increased with the increase of the inoculum. The LYC reached 72.43 μg/g when the strain number concentration was 2 × 10⁶ CFU/mL. the maximum SOD activity was 486.92 U/g. When the strain number con-centration was 3 × 10⁶ CFU/mL. If the inoculum exceeds 3 × 10⁶ CFU/mL, it may result in excessive fermentation, which can result in destabilization of terpenoids and loss of biological enzyme activity. As a result, the optimal inoculum amount ranges from 2 × 10⁶–3 × 10⁶ CFU/mL.

Experiments showed that four factors had effects on both SOD activity and LYC. The design and results of the Box–Behnken test were shown in Table 3. The regression equations are modeled below:

Anova and significance test for SOD activity and LYC. R², adjusted R², final results for flow rate flow %, and proper accuracy are shown in Table 7.

The test of significance of the coefficients of the model regression equation for Y1 (Figure 2A) shows that the linear coefficients (B, D, AB, AD, BD, BC, A², B², C², and D²) are highly significant and the rest of the terms are not significant. The test of significance of coefficients of regression equation of the Y2 model (Figure 2B) shows that linear coefficients (A, B, C, AC, BD, A², B², C², and D²) are highly significant and the rest of the terms are not significant (p < 0.01).

As can be seen from Table 7, the correlation coefficient R² value for SOD activity was 0.9877 and for LYC it was 0.9605, indicating that the model can be used for the analysis and prediction of the fermentation process. By predicting the regression model, the best treatment for FTJ was obtained as follows: fermentation temperature of 37.09°C, LAB inoculum of 2.95 × 10⁶ CFU/mL, fermentation time of 22.20 h, and TSS of 12.45^°Brix. In these conditions, the predicted value of FTJ SOD activity was 505.79 U/g, and the predicted value of LYC was 77.33 μg/g.

After comprehensive consideration, the optimal conditions were modified to a fermentation time of 22 h, fermentation temperature of 37°C, LAB inoculum of 3 × 10⁶ CFU/mL, and sugar addition of 12.5^°Brix. Under these conditions, we conducted experiments and verified that the SOD activity of the tomato enzyme was 496. 67 U/g and the LYC content was 77.12 μg/g.

LC–MS technique was used to characterize the metabolites in the fermentation broth to gain insight into the metabolites at different time stages during the fermentation process. A total of 1867 ionic metabolite signatures were generated at five stages, A, B, C, D, and E, during the fermentation process, including 1,020 positive ion patterns and 847 negative ion patterns. The metabolites were further analyzed by applying PCA and OPLS-DA models to the processed metabolite lists, which can effectively reduce the data dimensionality and improve the interpretability and validity of the data. PCA responds to the overall temporal and within-sample variability, reveals the distribution trends among various samples, and identifies possible discrete dispersal points, which can intuitively reflect the similarities or differences among the samples. As shown in Figure 3, fermentation samples at 0 h–10 h, 0 h–16 h, 0 h–20 h, and 0 h–22 h were separated by PCA principal component differentiation. All samples and QCs were within the 95% confidence interval, indicating good stability and reproducibility of the experiment.

In this work, we also analyzed the metabolite variation using the OPLS-DA model. OPLS-DA is an orthogonal correction that filters out uncorrelated orthogonal signals, ensuring a valid and more reliable model. In the model, there were four groups, and their OPLS-DA score plots (Figure 4) showed excellent model parameters (0 h–10 h: R²Y = 1, Q2 = 0.963; 0 h–16 h: R²Y = 1, Q² = 0.983; 0 h–20 h: R²Y = 1, Q² = 0.990; 0 h–22 h, R²Y = 1, Q² = 0.993). In cross-validation and response alignment, tests showed that no overfitting occurred in the OPLS-DA model (Supplementary Figure S1). Therefore, the OPLS-DA model is valid with high predictive power and can be used to explore metabolic differences at different stages during the fermentation of FTJ.