Abstract Different spatial positions lead to inconsistent fermentation effects and flavors, however, the spatial heterogeneity of Qingxiangxing (QXX) Baijiu remains unknown. We investigated the microbes, flavors, and physicochemical properties of different layers in fermented grains of QXX Baijiu using Illumina HiSeq sequencing, two-dimensional gas chromatography–mass spectrometry (GC × GC–MS) and ultra-high performance liquid chromatography-mass (UHPLC-MS). A total of 79 volatiles, 1596 metabolites, 50 bacterial genera, and 52 fungal genera were identified. The contents distribution followed the order: upper layer > bottom layer > middle layer. Organic acids and derivatives were the main differential metabolites across the three layers. Starch, pH, and reducing sugar levels increased from the upper to bottom layer. Saccharomyces and Lactobacillus were dominant microbes. Pediococcus, the biomarker of upper layer, showed positive correlations with formic acid, ethyl lactate, acetic acid, ethyl linoleate, and ethyl oleate. These findings deepen our understanding of the fermentation and flavor formation mechanisms of QXX Baijiu. Keywords: Qingxiangxing baijiu, Different layers of fermented grains, Microbial composition, Volatile compounds, Metabolites, Physicochemical properties Highlights * • The spatial heterogeneity of Qingxiangxing Baijiu fermentation was firstly explored. * • The microbiota, flavor and physicochemical properties were comprehensively studied. * • Upper-layer fermented grains played an important role in the flavor formation. * • The Pediococcus promoted the production of flavor compounds in upper fermented grain. * • Differential flavor compounds were closely correlated with microbial communities. 1. Introduction Baijiu is a traditional distilled liquor in China with a long history. The production of Baijiu encompasses four key processes: traditional solid-state fermentation, distillation, aging, and blending. Qingxiangxing (QXX) is one of the four major aroma types of Baijiu in China, which is fermented primarily with sorghum as the raw material ([41]Huang et al., 2020). It is characterized by a clear and mellow aroma, sweetness, softness, and a long aftertaste. Ethyl acetate and ethyl lactate are the principal aroma compounds of QXX Baijiu ([42]Li et al., 2023). As the carriers of microbes, fermented grains (FG) play an important role in the formation of Baijiu flavor. As a relatively independent ecosystem, the micro-ecology in the pit is different in different space environments ([43]Qian et al., 2021). During microbial growth, reproduction, and metabolism, the physicochemical properties undergo changes during fermentation, and the fermentation effects in different layers of FG within pit vary ([44]Qian et al., 2021). Most studies ([45]Ao et al., 2022; [46]Ding, Wu, Huang, & Zhou, 2016; [47]Guan et al., 2023; [48]Qian et al., 2021) suggest that the nutrients gradually sink with the water under gravity during the fermentation process, and the fermentation happening in the upper layer (UL) of FG is not as effective as that of the the bottom layer (BL), and consequently impacting the formation of flavor substances ([49]Ding et al., 2016). [50]Guan et al. (2023) demonstrated that concentrations of ethyl hexadecanoate, ethyl acetate, and ethyl butyrate progressively increased in the UL, middle layer (ML), and BL of Nongxiangxing (NXX) Baijiu ([51]Guan et al., 2023). Temperature, oxygen, and acidity are closely associated with the composition of the microbial community ([52]Guan et al., 2023). However, some studies ([53]Ao et al., 2022; [54]Ding et al., 2016) have found that the highest content of flavor substances occurs in the UL. [55]Ao et al. (2022) indicated that at the end of Lu-flavor liquor fermentation, the content of major volatile flavor compounds (VFCs) in FG was highest in the UL, followed by the BL, and were lowest in the ML ([56]Ao et al., 2022). Xiaolong [57]Hu et al. (2020) employed 16S rRNA gene sequencing to analyze the bacterial communities in the FG of Luzhou-flavor liquor across various spatial locations during the fermentation. Their findings revealed that the bacterial community diversity in the ML's FG exceeded that in both in UL and BL at the same time in fermentation ([58]Hu et al., 2020). However, existing studies have primarily focused on the spatial heterogeneity of NXX Baijiu and Jiangxiangxing (JXX) Baijiu ([59]Ao et al., 2022; [60]Guan et al., 2023; [61]Hu et al., 2020). For QXX Baijiu, research has mainly been conducted on the entire FG microbial community, physicochemical characteristics, and flavor or changes during the fermentation process ([62]Huang et al., 2020). Research on the distinct layers of FG in QXX Baijiu is limited, and the non-volatile flavor compounds (non-VFCs) are infrequently detected by metabonomics, hindering comprehensive analysis. Consequently, the limited understanding of the brewing mechanism of QXX Baijiu impedes the development and optimization of traditional brewing technologies. In this study, we comprehensively investigated the physicochemical properties, microbial community composition, flavor substances, and metabolic pathways of the FG from UL, ML, and BL of QXX Baijiu using Illumina HiSeq sequencing, ultra-high performance liquid chromatography-mass (UHPLC-MS) for non-targeted metabolomics, and two-dimensional gas chromatography–mass spectrometry (GC × GC–MS). The relationship between key microorganisms and other factors was also explored. Differences between the layers of FG and their potential formation mechanism were analyzed. The results of this study will enhance our understanding of the spatial heterogeneity of FG and the brewing mechanism of QXX Baijiu. This study provides a theoretical basis for subsequent production optimization. 2. Materials and methods 2.1. Samples collection QXX Baijiu was fermented in a stainless steel tank for 15 days at 25 °C using red glutinous sorghum as the raw material and Xiaoqu as the starter. The sorghum, Xiaoqu, and fermented grain sample used in our study were sourced from Sichuan Tujiu Liquor Co., Ltd.. Three tanks were selected for replicates. The mixed FG samples were collected at the beginning of fermentation (day 0, before fermentation) as the BF group. Samples collected from UL, ML, and BL (each layer was approximately 50 cm high) were collected at the end of fermentation (day 15, after fermentation) as AF group. Subsequently, the samples from each layer were blended. One portion of the FG samples was stored at 4 °C for the analyses of physicochemical properties and VFCs analysis, while the other portion was stored at −80 °C for microbial and non-VFCs extraction and further analyses. 2.2. The physicochemical properties analysis The physicochemical properties of FG, including acidity, pH, starch content, and reducing sugar levels were assessed using standard analytical methods ([63]Guan et al., 2023). The moisture content in the various FG samples was determined by assessing the dry/wet weight at 105 °C. 2.3. DNA extraction, PCR, and Illumina HiSeq sequencing The total DNA of FG samples was extracted using the Wizard® DNA kit (Promega, USA), following the manufacturer's protocol. The concentration and purity of the extracted DNA were measured using a Nanodrop 2000 ultra-micro spectrophotometer (Thermo Scientific, Illkirch, France) and agarose gel electrophoresis (Beijing Liuyi biotechnology Co.,Ltd.). The V3-V4 hypervariable regions of the 16S rRNA genes were amplified using the primer set 338F (5′-ACTCCTACGGGAGGCAGCAG-3′) and 806R (5′-GGACTACHVGGGTWTCTAAT-3′) for bacteria. The ITS3-ITS4 region of the ITS rRNA genes were amplified using the ITS3R (5′-GCATCGATGAAGAACGCAGC-3′) and ITS4F (5′-TCCTCCGCTTATTGATATGC-3′) primers for fungi. Library construction was carried out according to the instructions of the NextFlex Rapid DNA-SEQ Kit (BIOO Scientific, USA). High-throughput sequencing was conducted by the Illumina HiSeq X Ten (Illumina, San Diego, CA, USA) platform. 2.4. Extraction of volatile compounds Volatile compounds from samples were extracted using the head space-solid phase microextraction (HS-SPME) method ([64]Guan et al., 2023). Briefly, 2 g of the sample was placed into a 15 mL headspace vial with 5 mL sterile distilled water and 1.5 g NaCl, followed by performed ultrasonic treatment for 30 min. Subsequently, 20 μL of 0.822 mg/L 2-Octanol was added as internal standard. The vial was then sealed and equilibrated at 60 °C for 20 min. The 50/30 μm DVB/CAR/PDMS fibers (Supelco, Bellefonte, PA, USA) were inserted into the headspace of the vial to absorb the volatiles at 60 °C for 45 min. Afterward, the fiber was inserted into the GC injection port for desorption at 250 °C for 5 min. 2.5. GC × GC–MS detection and analysis of VFCs VFCs in FG were analyzed using GC × GC–MS instrument (GCMS-QP2020 NX, Shimadzu, Japan) following previously reported methods ([65]Fan et al., 2018). The one-dimensional (1D) column used was a DB wax column (30 m × 0.25 mm × 0.25 μm). The two-dimensional (2D) column comprised a DB-17MS column (1.2 m × 0.18 mm × 0.18 μm) and was equipped a solid-state thermal modulator HV (c720–21,005). Helium gas (purity >99.9999%) was used as a carrier gas for splitless injections at a flow rate of 1.0 mL/min. Additionally, 2D analysis time was synchronized with the 1D column, featuring a 4 s modulation period. The GC–MS test parameters were as previously described ([66]Gao, Zhang, Regenstein, Yin, & Zhou, 2018). VFCs were identified in the samples by matching their retention indices and mass spectra against the National Institute of Standards and Technology (NIST 20) spectral database. Relative quantification was performed by normalizing peak areas. The odor activity values (OAVs) were calculated based on the quantitative results and corresponding odor threshold values. The OAV of volatile substances was calculated by dividing concentration by their reported threshold value (Table S4). 2.6. Untargeted UHPLC-MS Metabolomic analysis A UHPLC-Q Exactive HF-X system (Thermo Fisher Scientific, USA) was employed to separate and analyze the metabolites. A UHPLC system equipped with an ACQUITY HSS T3 column (100 mm × 2.1 mm i.d. × 1.8 μm; Waters, Milford, USA) facilitated the chromatographic separation of metabolites. Mobile phase A consisted of 0.1% formic acid in water: acetonitrile (95:5, v/v). Mobile phase B was composed of acetonitrile, isopropanol, and water in a ratio of 47.5:47.5:5 (v/v). All reagents were purchased from Fisher Chemical Co., Ltd. (USA). The flow rate was 0.40 mL/min, the injection volume was 2 μL, and the temperature of the chromatographic column was 40 °C. LC-MS was performed as previously described ([67]Zeng et al., 2022). The mass spectrometric data were collected with an electrospray ionization (ESI) source operating in both positive and negative ion mode. Fermented grains (50 mg) were accurately weighed, and the metabolites were extracted using a 400 μL methanol:water (4:1, v/v) solution containing 0.02 mg/mL L-2-chlorophenylalanin as internal standard. The sample solution was placed in a high-throughput tissue crusher (Wonbio-96c, Shanghai Wanbo Biotechnology Co., LTD) and ground for 6 min at −10 °C, followed by ultrasonication at 40 kHz for 30 min at 5 °C. The samples were incubated at −20 °C for 30 min to precipitate proteins. Following centrifugation at 13000 g at 4 °C for 15 min, the supernatant was transferred to sample vials for LC-MS/MS analysis. All sample-derived supernatants were mixed in equivalent volumes to prepare the quality control sample (QC). The metabolites were identified and annotated by referencing the Human Metabolome (HMDB) database and the Kyoto Encyclopedia of Genes and Genomes (KEGG) database, pathway enrichment analysis was conducted by MetaboAnalyst 4.0. 2.7. Statistical analysis SIMCA-P software (Version 14.1, Umetrics, Sweden) was used for principal component analysis (PCA) of the relative content of VFCs, and partial least squares discriminant analysis (PLS-DA). Histogram, venn diagram, and cluster analysis of the relative content of VFCs were conducted using Origin software (Version 7.00, OriginLab Corporation, Northampton, MA, USA). GraphPad Prism software (Version 8.2.1, Inc. San Diego, CA, USA) was utilized for significant difference analysis by Tukey test (P < 0.05 was considered significant). Shannon, Simpson, Chao, and Sobs indices were calculated with Mothur v1.30.1(version v.1.30.2). The Wilcoxon rank-sum test was used to analyze the differences between groups of Alpha diversity. Metabolites showing significant differences were selected based on variable importance in projection (VIP) scores (VIP > 1) from orthogonal partial least-squares discriminant analysis (OPLS-DA) and P-value (P < 0.05) from t-tests. R software (version1.6.2) was used to draw the heatmap. All measurements were carried out in triplicate, and the results were expressed as the means ± standard deviation. 3. Results 3.1. Microbial community composition of FG To understand the characteristics of microbial communities in different layers of FG, the diversity and community composition were analyzed. The PCA plots ([68]Fig. 1A, B) demonstrated that samples from the beginning and end of fermentation were distinct. While, FG samples from different layers were clustered together, but the distribution of microbial composition varied across layers. In bacterial and fungal PCA, samples from the BL and ML were closely aligned ([69]Fig. 1A, B), indicating similar community compositions between these two layers. Fig. 1. [70]Fig. 1 [71]Open in a new tab Microbial community characteristics analysis of different FG. Beta diversity of the (A) bacterial and (B) fungal communities assessed by PCA analysis. The circos plots of the (C) bacterial and (D) fungal communities at the phylum level. The bar plots of the (E) bacterial and (F) fungal communities at the genus level. The Shannon, Simpson, Chao, and Sobs indices of the microbial community were analyzed. The results showed that the bacterial community diversity and richness in FG significantly decreased after fermentation (Fig. S1A, S1B, S1C, S1D), indicating that the fermentation process optimized the bacterial structure. This phenomenon is the same as what has been reported ([72]Huang et al., 2020). Disadvantaged microorganisms in the competition and the environment gradually disappeared after fermentation, while dominant microorganisms became more predominant. The Chao and Sobs indices were higher in the FG's ML compared to the other two layers (Fig. S1B, S1D), indicating greater community richness in the ML. After fermentation, the Chao, Simpson, and Sobs indices of the fungi in the FG gradually decreased in the UL, ML, and BL (Fig. S1E, S1F, S1G, S1H), although the diversity index remained significantly higher in the UL than before fermentation. The fungal community diversity and richness of fungi from the BL of FG were lower than pre-fermentation samples. This suggests that the high acidity and high water content of the BL may unsuitable for the survival of many fungi, whereas the UL provides an optimal environment for the growth and reproduction of aerobic fungi. Venn diagram analysis revealed 18 unique bacterial genera in the ML (Fig. S1I, Table S3). Almost all of them were reported to be aerobic or facultative anaerobes ([73]Rainey et al., 2015). Among them (Table S3), Chryseobacterium, as a cellulose-degrading bacterium, has high glucanase and xylanase activity ([74]Tan et al., 2018). Sphingomonas has a special cell structure, and its cell membrane replaces lipopolysaccharide with sphingomonas ([75]Aso et al., 2006), which has a metabolic mechanism to tolerate poor nutrition ([76]Li et al., 2022). The characteristics of these genera may play important roles in the fermentation. Thus they survive in the ML with higher alcohol and acid content and lower oxygen content. Six genera (Fibrobacter, Cloacibacterium, Corynebacterium, Bifidobacterium, Propionibacterium, Cetobacterium) and two genera (Lactococcus, Roseburia) were unique to the UL and BL (Table S3), respectively. There were more unique genera in the ML than in the UL and BL (Fig. S1I). This could be attributed to the state of complete anaerobic fermentation in the middle and late stages of the fermentation, which inhibited the growth of aerobic microorganisms. The increase in alcohol and acid contents inhibited the growth of intolerant microorganisms in FG, especially in the BL. Thus, the environment in the ML may be more suitable for the survival of most facultative anaerobic bacteria, as well as acid-resistant microorganisms. A total of 10 phyla and genera with relative abundance >0.01% were identified by composition analysis of bacterial community in FG. Firmicutes and Proteobacteria were the main phyla of FG before fermentation (BF), with relative abundance of 44.7% and 53.5%, respectively ([77]Fig. 1C). At the end of fermentation, Firmicutes became the main phylum in all layers, with relative abundance of >99% and increasing from top to bottom. At the genus level, the bacterial community structure of FG tends to be simplified ([78]Fig. 1E), which was consistent with our diversity analysis and also found in previous studies ([79]Xue et al., 2023). Gluconobacter and Acetobacter were the dominant genera in the FG at the beginning of fermentation ([80]Fig. 1E), with relative abundances of 37.36% and 20.39%, respectively. At the end of fermentation, Lactobacillus became the dominant genus in the FG ([81]Fig. 1E), consistent with previous study ([82]Luo et al., 2023; [83]Xue et al., 2022). The relative abundances of Lactobacillus in the UL, ML, and BL were 91.58%, 95.69%, and 98.46% ([84]Fig. 1E), respectively. The deeper the depth, the simpler the microbial community structure, which may be related to the high acidity of the BL ([85]Fig. 2B). On the one hand, it may be due to the metabolism of acid-producing bacteria ([86]Luo et al., 2023). On the other hand, the high-acid environment further promoted the replacement of microorganisms, and intolerant microorganisms were gradually eliminated ([87]Xue et al., 2022). The 27 genera showed significant differences among the 4 groups (Fig. S2A). The first few genera with high abundance were Gluconobacter, Acetobacter, Achromobacter, and Bacillus. Except for the genera Lactobacillus and Bacillus, which showed a significant increase in abundance after fermentation, all other genera decreased significantly or even disappeared. More and more studies have shown that Bacillus plays an important role in the brewing process of Baijiu ([88]Li, Lian, Ding, Nie & Zhang, 2014; [89]Yang et al., 2020). The metabolites of Bacillus can promote the formation of Baijiu flavor ([90]Li, Lian, Ding, Nie, & Zhang, 2014). Lefse analysis of FG in different layers (Fig. S1K) at the end of fermentation revealed that Pediococcus and unclassified__f__Lactobacillaceae were the biomarkers in the UL, while Weeksellaceae and Flavobacteriales were the biomarkers in the ML (Fig. S1K). The results showed that the relative abundance of Pediococcus increased at the beginning of fermentation, remained stable in the middle stage, and decreased at the later stage ([91]Xue et al., 2023). Pediococcus is a type of lactic acid bacteria (LAB) with low pH resistance. It has been found to be the dominant bacteria in Baijiu and the main functional contributor in Baijiu fermentation ([92]Hu et al., 2021), which can promote the synthesis of flavor compounds. Fig. 2. [93]Fig. 2 [94]Open in a new tab Physicochemical factors in different FG. (A) Moisture content; (B) Acidity; (C) pH value; (D) Starch content; (E) Reducing sugar content. The results revealed that the diversity and richness of fungi in FG samples after fermentation (AF) (i.e., UL, ML, BL) were significantly different from those before fermentation (BF), and the diversity of the UL of FG was significantly higher than that of the other layers (Fig. S1E, S1F, S1G, S1H). Fourteen genera of fungi were unique to the UL of FG (Fig. S1J, Table S3), among which the genus Kloeckera was found to have a significant effect on the flavor and sensory qualities of wines ([95]Tang, Zhao, Cui, Lai, & Zhang, 2023). The genus Thermoascus was reported to produce amylase, pectinases, cellulase, and xylanase ([96]Wang et al., 2023), making it a potential functional genus for QXX Baijiu brewing. There were 9 unique genera (Cystobasidium, Neopestalotiopsis, Papiliotrema, Rhodotorula, Rhodosporidiobolus, Cryptococcus, Kregervanrija, Trichosporon, Neocucurbitaria) in the ML and 1 genus (Dekkera) in the BL (Table S3). Dekkera has been reported to have a higher ethanol yield than Saccharomyces cerevisiae under oxygen limitation ([97]De Barros Pita et al., 2019). The diversity and abundance of BL in FG were lower than those before fermentation (Fig. S1E, S1F, S1G, S1H). This indicates that the high acidity and high moisture content of FG in the BL might not be suitable for the survival of most fungi. While the ML environment was more suitable. Ascomycota was the sole fungal phylum before and after fermentation of FG ([98]Fig. 1D), and the structure was relatively simple. Previous studies have shown that Ascomycota was the predominant fungus in the FG of Maotai-flavor Baijiu, NXX Baijiu, and QXX Baijiu, indicating that Ascomycetes are the key fungi in Baijiu brewing ([99]Xue et al., 2023). The dominant genus of fungi in FG was Saccharomyces (68.70%) both before and after fermentation ([100]Fig. 1F). At the end of fermentation, Saccharomyces decreased in the UL (40.56%), but increased in the ML (83.43%) and BL (83.06%), followed by the genera Cyberlindnera and unclassified_f_Metschnikowiaceae ([101]Fig. 1F). The abundance of fungi exceeded that of bacteria at the end of fermentation, and there was no significant reduction. The composition of fungi in the FG of QXX Baijiu varied across multiple studies ([102]Xue et al., 2023). [103]Xue et al. (2023) showed that Issatchenkia and Saccharomycopsis were the dominant fungi in the fermentation process. [104]Li, Fan, Huang, and Han (2022) found that Pichia was the dominant genus in the fermentation process. The analysis of differences between groups revealed that eight genera including Saccharomyces, Clavispora, and unclassified__f__Metschnikowiaceae exhibited significantly differences among the four groups (Fig. S2B). The abundance of Saccharomyces in the UL of FG was significantly lower than that in the other two groups, while the abundance of Clavispora, unclassified__f__Metschnikowiaceae was significantly higher than that in the other three groups. Lefse analysis revealed that Cutaneotrichosporon, Trichosporonaceae, and Trichosporonaless were biomarkers of the ML of FG at the end of fermentation (Fig. S1L). 3.2. The physicochemical properties of FG The physicochemical properties of FG represent the growth environment of the microbial community, which is crucial to the fermentation of Baijiu brewing. The moisture content of FG at the end of fermentation was higher than that before fermentation (66.02 g/100 g) and increased gradually from top to bottom (67.39 g/100 g, 71.39 g/100 g, and 78.78 g/100 g, respectively) ([105]Fig. 2A). These moisture contents may result from various physiological and biochemical reactions during fermentation and microbial metabolism. For instance, certain lactic acid bacteria naturally produce 2-butanol through anaerobic fermentation, utilizing acetyllactic acid as a substrate ([106]Speranza et al., 1997). The presence of moisture lays the foundation for the smooth progress of saccharification and fermentation. During fermentation, the moisture content of FG affects the metabolism and community structure of microorganisms, and the growth metabolism of microorganisms also regulates the moisture content, which affects each other inseparably. Therefore, the difference in moisture content among different layers may be attributed to variations in microbial metabolic activity across spatial positions. Additionally, the moisture of FG gradually descended under gravity in the late stage of fermentation, with the order eventually becoming BL > ML > UL. At the end of fermentation, the acidity of FG in the UL, ML, and BL increased gradually to 0.725, 0.75, and 0.835 mmol/10 g, respectively ([107]Fig. 2B). Additionally, the acidity of post-fermentation in the BL was higher than that of the pre-fermentation samples. This is consistent with the existing studies ([108]De Vuyst & Leroy, 2020). The pH decreased from top to bottom (4.115, 4.005, and 3.970 mmol/10 g, respectively) ([109]Fig. 2C), all of which were lower than that of the pre-fermentation samples. As the oxygen content of the BL was lower than that of the UL and ML ([110]Liu et al., 2023; [111]Yang et al., 2023), facilitating the growth of anaerobic or facultative anaerobic bacteria in FG, the acidity of the FG in BL was the highest. Variations in microbial composition, acidity changes, and esterification effects may contribute to differences among FG layers. The starch content of FG in the UL, ML, and BL showed an increasing trend, with values of 11.2, 11.6, and 14.9 g/100 g, respectively, which were lower than that of pre-fermentation samples (15.6 g/100 g) ([112]Fig. 2D). The reducing sugar contents of FG in the UL, ML, and BL were 0.267, 0.5, and 0.53 g/100 g ([113]Fig. 2E), respectively, and they also increased from the top to bottom. Reducing sugar content in FG was closely related to starch content. Starch is the material base of alcohol fermentation. After saccharification, starch and other polysaccharides were enzymatically hydrolyzed to glucose and other fermentable sugars. During the fermentation stage, glucose was converted to alcohol and trace flavor compounds. These substances provide the material basis and energy source for the growth and metabolism of microorganisms and make the fermentation proceed normally. The lowest content of starch and reduced sugar was found in the UL at the end of fermentation. This may be attributed to the decomposition of starch into reducing sugar in the UL of FG, with most of it converting into alcohol and esterifying into flavor substances. Therefore, its reduced sugar content is significantly lower than the other two layers. 3.3. The correlation of physicochemical properties and microbial community To investigate the effects of various physicochemical environmental factors on microbial community distribution during Baijiu fermentation, we analyzed the correlation between environmental factors and dominant genera using canonical correspondence analysis (CCA). The result showed that CCA1 and CCA2 explained 98.93% of the bacterial species distribution ([114]Fig. 3A), with CCA1 explaining 95.56% and CCA2 explaining 3.37%. This indicated that bacterial community is correlated with physicochemical properties. The CCA1 is corelated with most of the bacterial genus except Bacillus. Acidity and moisture exhibited a positive correlation with Lactobacillus ([115]Fig. 3A). Acidity and moisture had a strong effect on Lactobacillus associated with CCA1. Lactobacillus metabolizes carbohydrates to produce acetic acid, lactic acid, ethanol, and other important flavor substances during of fermentation ([116]Luo et al., 2023; [117]Xue et al., 2022). The rapid accumulation of these substances can lead to the change of acidity and ethanol concentration in the fermentation environment. Due to the lower oxygen content of FG in the ML and BL compared to the UL ([118]Liu et al., 2023; [119]Yang et al., 2023), which promotes the growth of anaerobic or facultative anaerobic bacteria ([120]Jiao et al., 2022), the acidity of FG in the ML and BL was higher. A negative correlation was observed between Bacillus and starch content ([121]Fig. 3A). Bacillus has been reported to produce amylase and protease, which breaks down macromolecules such as starch and protein ([122]Yang et al., 2020). The anaerobic environment and higher acidity in the ML and BL of FG led to a gradual decrease in amylase activity, potentially affecting the decomposition of macromolecules. Consequently, the BL exhibited the highest strach content. Fig. 3. [123]Fig. 3 [124]Open in a new tab The correlation network analysis between the microbial communities and physicochemical properties as well as flavor compounds of the FG. The CCA analysis of the correlation between the microbial communities (bacteria A, fungi B) and physicochemical properties at the genus level (P < 0.05). (C) The correlation between the microbial communities and VFCs by network correlation analysis. The line thickness indicates the correlation strength, with a yellow line denoting a positive correlation and a gray line signifying a negative correlation. (For interpretation of the references to colour in this figure legend, the