The reference strain used in this study was the commercial S. cerevisiae X16 obtained from Laffort Company (France). A total of 80 strains of non-Saccharomyces yeasts, isolated from spontaneous fermentation of R. roxburghii were screened for ethanol-tolerant strains. All yeasts cells were cultured on yeast extract peptone dextrose (YEPD) solid medium (1% yeast extract, 2% peptone, 2% glucose, and 2% agar) containing 100 mg/L of Chloramphenicol at 28°C for 72 h and then stored at 4°C for later use.

Ethanol-tolerant strains were screened by culturing them in YEPD broth (1% yeast extract, 2% peptone, and 2% glucose) containing 12% (v/v) ethanol with the initial concentration of 10⁸ cfu/ml, and the yeast cells were cultured at 28°C with shaking at 180 rpm for 36 h. The optical density (OD) values were measured at a wavelength of 600 nm using a spectrophotometer (Hitachi, Tokyo, Japan).

Yeast strains were identified using both morphological and molecular methods. Firstly, cells were scraped onto Wallerstein Laboratory nutrient agar and cultured for 72 h. The characteristics of the colony and cellular morphology were examined and photographed with a microscope (Olympus, Tokyo, Japan). Next, genomic DNA was extracted from three ethanol-tolerant strains (C6, F112, and F15) using a DNA extraction kit (B518257; Sangon Biotech, China) following the manufacturer’s instructions. The D1/D2 domain within the 26S rDNA was amplified using the polymerase chain reaction (PCR). The yeast species were then determined by comparing the 26S rDNA D1/D2 domain sequences in the GenBank database.

The C6, F112, and F15 strains were inoculated into YEPD broth at a concentration of 10⁸ cfu/ml and cultured under agitation at 180 rpm and 28°C for 48 h. The OD of the cultures was measured at the wavelength of 600 nm every 4 h, and a growth curve was plotted based on the time and OD_{600 nm} values.

Strains C6, F112, and F15 were inoculated with a concentration of 10⁸ cfu/ml into a 0.6% yeast powder solution containing 2% final concentration of glucose, sucrose, maltose, lactose, and galactose, respectively. The yeast powder solution was placed in test tubes containing Durham tubes and incubated at 28°C for 48 h. The formation of gas bubbles in the Durham tubes was observed. A positive reaction was recorded as “+” if bubbles formed. Otherwise, a negative reaction was recorded as “−.”

Strains of C6, F112, and F15 were inoculated in YEPD broth at a concentration of 10⁸ cfu/ml with (1) glucose concentrations of 100, 150, 200, 250, or 300 g/L; (2) citric acid mass fractions of 1, 1.5, 2, 2.5, or 3% (w%); and (3) a sulfur dioxide contents of 50, 100, 150, 200, or 300 mg/L. All groups were cultured at 28°C and 180 rpm for 36 h with three replicates, and then OD_{600 nm} values were measured.

The hydrogen sulfide (H₂S) production activities of C6, F112, and F15 were investigated using BiGGY agar by comparing the depth of colony color (Caridi et al., 2022).

The ability of the strains to produce β-glucosidase was analyzed using the p-nitrophenyl-β-D-glucopyranoside (p-NPG) method. Strains C6, F112, and F15 were inoculated into YEPD medium and shaken at 180 rpm at 28°C for 72 h. The supernatant was obtained after centrifugation at 3,000 g for 10 min and used for the determination of enzyme activity. Enzyme activity units (U) were defined as the amount of enzyme required to produce 1 μmol of p-nitrophenol (p-NP) by hydrolyzing 1 μmol of p-NPG under conditions of pH 5.0 and 50°C for 1 min.

Fresh, mature, and non-rotten R. roxburghii (Supplementary Figure 1) was juiced with juice extractor (Midea, WJE2802D, China) and then treated with 100 mg/L of potassium metabisulfite and 20 mg/L of pectinase at room temperature for 12 h. The juice was then adjusted to 24°Brix with crystalline sucrose and divided into four groups, with each group replicated in triplicate in 2 L sterile triangular flasks. For the C6 + S. cerevisiae X16 group, F112 + S. cerevisiae X16 group, and F15 + S. cerevisiae X16 group, each group was inoculated with 10⁸ cfu/ml of the C6, F112, or F15 strain and 10⁷ cfu/ml of S. cerevisiae X16, with a control group that was only inoculated with 10⁷ cfu/ml of S. cerevisiae X16. The flasks were left to ferment statically at 26°C until fermentation was completed. After fermentation, the R. roxburghii fruit wine from each group was centrifuged at 4,000 rpm for 10 min, and the supernatant was used for the determination of the quality indicators of R. roxburghii fruit wine.

The alcohol content, total sugar, total acidity, and volatile acid content of the R. roxburghii wine were determined following the methods described by Liu et al. (2021b). The sensory characteristics of the R. roxburghii fruit wine were analyzed using an electronic tongue system. For this, 80 ml of each group of R. roxburghii fruit wine was taken and added to a dedicated beaker for the electronic tongue system. The electronic tongue system was used according to the instructions in the user manual to test each group of R. roxburghii fruit wine. The sampling time was 120 s, the sampling speed was 1/s, each sample was measured in triplicate, and each replicate was collected four times.

The headspace solid-phase microextraction-gas chromatography-mass spectrometry system (TQ8040, Agilent, USA) was used to analyze the aroma characteristics of the R. roxburghii fruit wine. The aroma components of R. roxburghii fruit wine were extracted at 40°C for 30 min, with cyclohexanone used as the internal standard for determining the aroma components of R. roxburghii fruit wine. The odor activity value (OAV) of each aroma component was calculated by referring to the threshold values of each volatile aroma component.

Data results were presented as mean ± SD. Principal component analysis (PCA) and one-way ANOVA were performed using SPSS 21.0 to test for significant differences among the groups. A p-value of less than 0.05 was considered statistically significant. Each experiment was repeated in triplicate.

When the native non-Saccharomyces yeasts isolated from R. roxburghii were treated with 12% (v/v) ethanol, most of them died. However, three yeast strains (designated as C6, F112, and F15) exhibited satisfactory growth with OD_600nm values of 0.56 ± 0.02, 0.47 ± 0.01, and 0.48 ± 0.01, respectively. Therefore, C6, F112, and F15 were selected as ethanol-tolerant strains for further analysis.

The identification of the three ethanol-tolerant yeast strains was initially based on morphological characteristics on WL agar. As shown in Figure 1, the colony color of C6, F112, and F15 was white, and their colony topography was convex and opaque.

To confirm the identity of the ethanol-tolerant yeast strains, their 26S rDNA D1/D2 domain sequences were compared. The analysis revealed that the 26S rDNA sequences of C6, F112, and F15 had the highest similarity to C. tropicalis, Pichia guilliermondii, and W. anomalus, respectively. Therefore, these three strains of ethanol-tolerant yeasts (C6, F112, and F15) were identified and named C. tropicalis C6, P. guilliermondii F112, and W. anomalus F15 based on the results of morphological characteristics and sequence alignment.

The growth curves of the strains are shown in Figure 2, with a lag phase from 0 to 4 h, a logarithmic growth phase from 4 to 20 h, and a stationary phase after 20 h. During the logarithmic growth phase, the OD_{600 nm} values of C. tropicalis C6, P. guilliermondii F112, and W. anomalus F15 were all lower than those of the commercial S. cerevisiae X16. During the stationary phase, the OD_{600 nm} of P. guilliermondii F112 was lower than that of S. cerevisiae X16. Throughout the entire growth period, the growth of P. guilliermondii F112 was lower than that of S. cerevisiae X16, while the growth of C. tropicalis C6 and W. anomalus F15 was basically consistent with that of S. cerevisiae X16 in the later stages of the stationary phase.

To assess the tolerance of the selected yeasts to winemaking conditions, their OD_{600 nm} values were measured after exposure to different concentrations of glucose, SO₂, and citric acid. Results demonstrated that all three ethanol-tolerant yeast strains exhibited excellent sugar tolerance, as they were able to grow in all glucose concentrations tested (100–300 mg/L) (Figure 3A). Furthermore, C. tropicalis C6, P. guilliermondii F112, and W. anomalus F15 displayed similar sulfur dioxide and acid tolerance to S. cerevisiae X16 within the tested ranges of sulfur dioxide (50–300 mg/L) and citric acid (2.5–4.0%) concentrations, respectively (Figures 3B, C). Therefore, C. tropicalis C6, P. guilliermondii F112, and W. anomalus F15 showed perfect tolerance to the winemaking environment.

As shown in Table 1, different strains have different utilization characteristics for different sugars. W. anomalus F15 can only metabolize glucose with the least number of sugars that it can utilize. On the other hand, C. tropicalis C6 can metabolize all types of sugars except for galactose, with the broadest range of sugar utilization. P. guilliermondii F112 can utilize three types of sugars (glucose, sucrose, and maltose). Therefore, C. tropicalis C6 has the widest range of sugar metabolism, and its sugar utilization characteristics are similar to those of S. cerevisiae X16, except for galactose.

The ability of ethanol-tolerant yeasts to produce H₂S production ability was evaluated by observing the color depth on BiGGY agar (Caridi et al., 2022). As shown in Figure 4, C. tropicalis C6 had a similar H₂S production ability to the control (S. cerevisiae X16), while P. guilliermondii F112 and W. anomalus F15 exhibited stronger H₂S production abilities than the control, as evidenced by the colony depth color on the filter paper.

Flavor compounds in fruit are often present in the form of glycoconjugate, making them flavorless (Gueguen et al., 1996). β-Glucosidase are enzymes that could hydrolyze these glycosyl bonds, thereby releasing the flavor compounds to wines (Haslbeck et al., 2017). To investigate the β-glucosidases of the selected strains, namely, C. tropicalis C6, P. guilliermondii F112, and W. anomalus F15, p-NPG colorimetry was used. The result showed that the β-glucosidase production abilities of C. tropicalis C6 and P. guilliermondii F112 were similar to those of S. cerevisiae X16. However, the strain of W. anomalus F15 exhibited significantly lower β-glucosidase production ability than S. cerevisiae X16 (Table 2).

The combination of non-saccharomyces yeasts with S. cerevisiae as fermentation starters has been widely studied and accepted in wine production (Comitini et al., 2011). To further analyze the fermentative properties of the ethanol-tolerant yeast strains, R. roxburghii wine was fermented by co-inoculating the C. tropicalis C6, P. guilliermondii F112, or W. anomalus F15 together with S. cerevisiae X16. Dynamic changes of the non-Saccharomyces yeasts population during R. roxburghii wine fermentation were monitored by colony counting method, and the results showed that the proportion of C. tropicalis C6, P. guilliermondii F112, and W. anomalus F15 gradually decreased, in contrast, the proportion of S. cerevisiae X16 gradually increased and dominate at the middle and later periods of fermentation (Supplementary Figure 2).

The physicochemical parameters of the fermented R. roxburghii wines are listed in Table 3. Ethanol degrees of wines fermented by P. guilliermondii F112 or W. anomalus F15 were lower than the wine produced by S. cerevisiae X16 alone, while the C. tropicalis C6 fermented wine was similar to that produced by S. cerevisiae X16. The pH and volatile acidity parameters of the four groups of R. roxburghii wines were similar, with no differences found among them. The total acidity were lower in the C. tropicalis C6 and P. guilliermondii F112 groups compared to the S. cerevisiae X16 group.

In addition, an electronic tongue system was used to perform sensory analysis and differentiate the sensory characteristics of the R. roxburghii wines fermented with different yeast strains. However, no significant differences in the sensory characteristics including in sourness, bitterness, astringency, aftertaste-A, aftertaste-B, umami, richness, and saltiness were found among the three types of R. roxburghii wines (Figure 5).

The volatile aroma profiles of the R. roxburghii wines fermented with the selected yeast strains were further examined by using GC-MS analysis. A total of 66 volatile compounds, including 32 volatile esters, 10 volatile alcohols, 6 volatile acids, 3 volatile aldoketones, and 15 other volatile chemicals, were identified in the four groups of fermented R. roxburghii wines (Table 4). The R. roxburghii wines co-fermented with the three ethanol-tolerant yeast strains contained 50, 46, and 55 volatile compounds, respectively, whereas only 44 volatile compounds were detected in the R. roxburghii wine inoculated with S. cerevisiae X16 alone. Additionally, 22 chemicals, including 10 esters, 1 alcohol, 1 acid, and 10 other compounds, were specifically detected in the R. roxburghii wines inoculated with the ethanol-tolerant yeasts. On the other hand, octyl acetate, decanoic acid, 1-nonanal, and 2,4-di-tert-butylphenol were specifically discovered in the S. cerevisiae X16 group. Overall, the co-inoculation of these ethanol-tolerant yeast strains isolated from R. roxburghii along with S. cerevisiae increased the types of volatile compounds in the wine (Supplementary Figure 3).

When checking the volatile compound contents of R. roxburghii wines with the different ethanol-tolerant yeast strains, variations in the levels of volatile esters, alcohols, acids, aldoketones, and other chemicals were observed (Table 4). Co-inoculation with the three ethanol-tolerant yeasts resulted in an increase in volatile esters and alcohols, as well as a decrease in volatile aldoketones and other compounds (Table 4). In addition, R. roxburghii wines co-fermented with W. anomalus F15 and S. cerevisiae X16 exhibited higher levels of volatile acids compared to those produced with S. cerevisiae X16 alone.

The OAV was used to further evaluate the contribution of the main aromatic compounds to the aromatic characteristics of R. roxburghii wine. Compounds with OAV greater than 1 were considered to have a significant impact on the aroma, while those with OAV less than 1 were considered less important. Table 5 shows the calculated OAVs of twenty aromatic compounds in R. roxburghii wine. Thirteen compounds had OAVs greater than 1, while only 1 had an OAV less than 1 across all 4 groups of R. roxburghii wine. Specifically, ethyl 9-decenoate had an OAV greater than 1 only in the S. cerevisiae X16 group, whereas isobutanol had the only OAV less than 1 in the same group. The OAVs of ethyl caprate, isoamyl acetate, ethyl hexanoate, and ethyl octanoate were high in all four groups, suggesting that these compounds strongly contribute to the aroma of R. roxburghii wine.

Principal component analysis was used to further assess the impact of the main aromatic compounds on the characteristics of R. roxburghii wine. As depicted in Figure 6, the three principal components, PC1, PC2, and PC3, accounted for 62.84, 25.02, and 12.14% of the total variance, respectively, explaining 100.00% of the total variance. Most of the compounds were clustered in the positive axis of PC1 and PC2, and significant differences in distribution were observed among the four groups of fermented R. roxburghii wine. Ethyl palmitate might be closely related to the mixed fermentation of C. tropicalis C6 and S. cerevisiae X16, while W. anomalus F15 and S. cerevisiae X16 produced wines were characterized by compounds located in the positive PC1 and PC2, such as fruity and rosy chemicals (e.g., isobutyl acetate and ethyl 9-decenoate), which may contribute to the aroma of S. cerevisiae X16 fermented wine. However, it was challenging to identify the main volatile characteristics of P. guilliermondii F112 and S. cerevisiae X16 fermented R. roxburghii wines. Moreover, the mixed fermented wine using W. anomalus F15 and S. cerevisiae X16 was closely clustered to many esters, alcohols and acids, such as ethyl laurate, phenethyl alcohol, and octanoic acid, which may endow the wine more complex aroma characteristics.