Multi-Omics Reveals Microbial Roles and Metabolic Functions at the Spatiotemporal Niche in Pit Mud


 Background
 Popular distilled Chinese strong-flavoured liquor (CSFL) is produced by solid fermentation in ground pits. Pit mud (PM), as a habitat for microbes, plays an important role in the production of CSFL. However, understanding the taxonomic composition, metabolic potential, and functional diversity of core microbes of PM in spatiotemporal niche remains a major challenge.
Results
 Using a multi-omic approach of high-throughput full-length 16S rRNA, ITS sequencing, and metagenomics, we identified bacteria such as Caproiciproducens, Clostridium, Lactobacillus, Bellilinea, Petrimonas, Proteiniphilum, and Prevotella and the archaea Methanobrevibacter and Methanobacterium as the core microbiota in 30-, 100-, and 300-year-old cellars. Significant correlations between microbial communities and environmental factors showed that lactic, caproic, butyric, and acetic acids were the core driving forces for microbiota succession in different spatial locations and were mainly correlated with Caproiciproducens, Clostridium, Lactobacillus, Methanobrevibacter, and Methanobacterium. A total of 982 metabolites were detected using GC/MS and LC/MS, mainly including amino acids, peptides, and fatty acids, and correlations were shown between seven microorganisms and 12 amino acids and fatty acid metabolites. The crucial genes of flavour-relevant and substrate degradation pathways mainly included nine microorganism orders, with the most important being Clostridiales, Bacteroidales, and Methanobacteriales. However, only Clostridiales and Bacillales were potentially involved in alcohol–aldehyde–carboxylate–ester metabolism.
Conclusions
 The 30-, 100-, and 300-year-old cellars provided an ideal opportunity to understand the effect of cellar age on the microbial composition and functional diversity of the microorganisms in PM. Our detailed metagenomic analyses across the continuously used cells extend the known diversity of microorganisms involved in flavour generation and substrate degradation over a wide range of environmental conditions. The results indicate that Clostridiales and Bacteroidales are major microbiota orders for flavour generation and substrate degradation pathways, while archaea also play an important role in cooperation with bacteria in various flavour generation pathways. This study helps elucidate the core microbiota composition, different metabolic roles of microorganisms, and the formation mechanism of PM partial functions, thus providing a basic theory to support the regulation of Baijiu production.

moisture content in the pit can directly affect the pH, humus, and growth of the microbial ora. The proper pH of PM not only promotes CSFL fermentation but also facilitates the formation of aroma precursor substances and improves the quality of the original liquor. Microorganisms are continuously enriched, and additional populations are derived during the long-term reciprocating fermentation process. In fact, the interaction between these microorganisms and physicochemical indices has formed a complex microbial ecosystem [26] .
Although many studies have elaborated the microbiota composition and the factors driving the microbiota succession in PM, previous studies have only focused on the microbiota composition in the whole cell without considering the spatial locations. Moreover, only limited knowledge of the microbial composition in different spatial locations of PM is available, and the factors that drive the microbiota succession in PM with hundreds of year intervals need to be clari ed. Concurrently, little is known about the metabolic potential and functional properties of the PM microbiota owing to the lack of knowledge of complex microbiota interactions. Therefore, understanding the relationship among microbes, metabolites, and functional enzymes in PM remains a major challenge.
To address some of these problems, using a multi-omic approach, this study aimed to (1) characterise the microbiota community composition, analyse the driving factors for microbial succession, and reveal the relationships between chemical properties and microbial community in different PMs; and (2) investigate the differential metabolites and functional enzymes produced by microorganisms related to avour generation and substrate degradation pathways.

Analysis of PM physicochemical factors
The levels of PM moisture, TN, AP, pH, and acetic, caproic, butyric, and lactic acid content from different cellar samples changed signi cantly with increasing age of the cellar (Table 1, Fig. 2). pH is an important factor that has a cumulative effect on the properties of PM; however, it did not show differences in different spatial locations in the 30-, 100-, and 300-year-old cellars (p>0.05). The relative moisture content showed a signi cant difference in the 100-and 300-year-old cellars. The moisture content in position 4 was signi cantly lower than that in positions 1 and 2 (p<0.05). For the same positions, the moisture content in the 300-year-old cellars was slightly higher than that in the 100-and 30-year-old cellars, but the difference was not signi cant (p>0.05). The moisture content produced by microbial metabolism gradually accumulated and sank to the bottom of the cellar during fermentation, resulting in higher moisture content at the bottom of the cellar compared with that at other spatial positions. Nutrients are also important contributors to the biochemical properties of PM. The concentration of TN at position 4 was signi cantly lower than that at other positions in the 30-, 100-, and 300-year-old cellars (p<0.05).
Notably, the concentration of TN at position 1 in the 300-year-old cellar was signi cantly lower than that in the 30-and 100-year-old cellars. The concentration of AP at position 1 in the 100-and 300-year-old cellars was signi cantly higher than that at positions 2, 3, and 4 (p<0.05) and at positions 2 and 4, respectively.
Organic acids are important precursors of esters in CSFL and inevitably affect the quality of CSFL. We compared the organic acid contents spatiotemporally in the PM. The acetic acid content ( Fig. 2A) was signi cantly higher at position 3 (p<0.05) and position 1 (p<0.05) than that at other positions in the 30and 100-year-old cellars and 300-year cellars, respectively. The butyric acid content (Fig. 2B) showed no signi cant spatial difference in the 30-year-old cells (p> 0.05) but was signi cantly higher at position 1 than that at other spatial locations (p<0.05) in the 100-and 300-year-old cellars. The caproic acid (CA) content (Fig. 2C) showed different trends in the different cellar types. The CA content at positions 2 and 1 was signi cantly higher than that at other positions (p <0.05) in the 30-and 300-year-old cellars, respectively. As for lactic acid content (Fig. 2D), positions 4 and 1 showed signi cantly higher levels than other spatial locations in the 30-(p>0.05) and in the 100-and 300-year-old cellars (p<0.05), respectively. In summary, organic acids accumulated at the bottom of the pit with continuous CSFL fermentation, leading to a high content of organic acid at the bottom of the pit, when compared with other spatial locations. In addition, the total organic acid content increased with the continuous use of pits.

Microbial succession in different cellars
Spatiotemporal differences of microbial diversity in PM Succession patterns of PM microbial communities in 30-, 100-, and 300-year-old cellars were investigated. High-throughput sequencing was applied to obtain information on the active microbial community in the three cellar types at four positions, and sequence similarity thresholds were used to de ne the OTUs. Based on the sequencing results, the total reads were correlated with bacterial and archaeal phyla.  (Fig. 3). In this study, we de ned core microbiota genera as those detected in all PM samples and with relative abundance higher than 1.0% in all samples. Based on this, the core microbiota genera were Lactobacillus, Caproiciproducens, Clostridium, Proteiniphilum, Prevotella, Bellilinea, Petrimonas, Methanobacterium, and Methanobrevibacter and accounted for 60.28%, 84.39%, and 80.65% in the 30-, 100-and 300-year-old PM samples, respectively. Notably, Prevotella, Bellilinea, Petrimonas, Methanobacterium, and Methanobrevibacter were increased in the 100-and 300year-old PM samples when compared with those in the 30-year-old PM samples, especially for Methanobacterium and Methanobrevibacter. We further investigated the differential microbes among the three PM samples. The results (Fig. 3D) indicated that Proteiniphilum, Caproiciproducens, Methanobacterium, and Methanobrevibacter were the main differential microorganisms among the 30-, 100-, and 300-year-old samples. With the continuous use of the cellars, microbes will inevitably undergo succession progress. We calculated the beta nearest taxonomic index (βNTI) value to compare the in uence of the two (deterministic and stochastic) assembly processes of the 30-, 100-, and 300-yearold cellars (Fig. 3E) to compare and analyse the differences in the construction of microbial communities in the different PM cellars. The results showed that both deterministic and stochastic processes promoted the assembly of the microbiome in the PM cellars, with the pit used for a long time, especially for 300-year-old cellars, the microbial composition follows a deterministic process.

Microbial diversity in different spatial locations
The cellar environment changed with the CSFL fermentation process, such as the Huangshui generation.
It gradually changed from a micro-anaerobic to an anaerobic environment, leading to a diversity of microbial communities in different spatial locations. At position 1 (Additional le: Figure S1-A), Lactobacillus, Bellilinea, Methanobrevibacter, Petrimonas, and Prevotella were found in the 30-, 100-, and 300-year-old cellars. Petrimonas (13.18%) showed a higher relative abundance in the 100-yearold PM samples than that in the 30-and 300-year-old PM samples. Moreover, Lactobacillus (41.82%) and Prevotella (1.78%) were higher in the 300-year-old PM samples than in the 30-and 100-year-old PM samples, but the relative abundance of Bellilinea (0.7%) for the 300-year cellar was lower than that for the 30-and 100-year samples. In contrast, Bacillus, Weissella, Brachybacterium, Staphylococcus, Anaerocella, Caloramator, Caldicoprobacter, and Caloribacterium were only detected in the 30-year, 100-year, and 300year-old PM samples, respectively.
Differences in the physicochemical properties of different spatial positions in the pit, such as oxygen, moisture content, nutrients, and the contact between the jiupei (fermented grains) and PM, will inevitably lead to differences in microbial composition and abundance. Notably, the relative abundance of the same microorganism is also different in the use of the same year pit. In summary, the microbiota composition among the four spatial locations were different; for example, Caproiciproducens and Lactobacillus were concentrated at the bottom of the cellar, while Proteiniphilum and Clostridium were highly abundant at positions 3 and 4. In addition, some genera were only detected in speci c spatial locations, indicating the complex interactions between the microorganisms. Moreover, the microbial composition also correlated with the physicochemical environment in the cellars, and the diversity of microbes could in uence the avour substances in CSFL.
After studying the microbial communities of different PM samples, we further evaluated the similarities and differences in the microbiota using a Bray-Curtis approach. A non-metric multi-dimensional scaling analysis revealed that all samples in bacteria were clustered into two parts and showed year-dependent clustering (Fig. 4A). In summary, samples from the 30-, 100-, and 300-year-old samples were centralised and formed clusters, especially for 30-and 100-year-old samples, suggesting that samples from different years are possibly in different transitional states and bacterial communities in the 30-and 100-yearold cellars had similar microbial compositions. The microbial community continues to evolve over time, stabilises, and gradually forms a stable microbial environment. Therefore, the age of the cellar is the main factor driving microbial succession.

Relationships of microbial community with physicochemical properties
Environmental factors in fermentation can affect the growth and metabolism of microorganisms. Hence, physicochemical factors (pH, moisture content, TN, AP, acetate, caproate, butyrate, and lactate) were selected and analysed to reveal their connection with the microbiota community (Fig. 4B). The results indicate that the contents of lactate, butyrate, acetate, and caproate signi cantly affected microbial succession change (p<0.05) and that lactate and caproate contents were positively correlated with prokaryotic communities in the 30-, 100-, and 300-yearold samples at positions 1 and 2. Moreover, butyrate and acetate contents were positively correlated with prokaryotic communities in the 30-and 100-year-old PM samples at position 2.
To clarify the relationships between speci c genera and driving factors, we analysed their correlation using Spearman's coe cient (p<0.05). Organic acids signi cantly affected the core microbiota Lactobacillus, Caproiciproducens, Clostridium, Proteiniphilum, Prevotella, Bellilinea, Petrimonas, Methanobacterium, and Methanobrevibacter, while TN, AP, and moisture content affected a few families. Speci cally, at position 1 (Additional le: Figure S2-A), lactic acid was positively correlated with levels of Caproiciproducens. In addition, butyric acid and acetic acid were positively correlated with Methanobacterium and Caproiciproducens, Caproic acid was negatively correlated with Petrimonas, and TN was negatively correlated with Prevotella and Lactobacillus. At position 2 (Additional le: Figure S2-B), lactic acid was negatively correlated with Clostridium yet positively correlated with Lactobacillus, butyric acid was negatively correlated with Petrimonas yet positively correlated with Prevotella, acetic acid was positively correlated with Lactobacillus and Caproiciproducens, CA was negatively correlated with Petrimonas yet positively correlated with Lactobacillus, and moisture content was negatively correlated with Prevotella. At position 3 (Additional le: Figure S2-C), lactic acid and AP were the two important factors in uencing microbial community composition. For example, lactic acid was positively correlated with Methanobrevibacter, but AP was negatively correlated with Ralstonia, Bellilinea, and Proteiniphilum and negatively correlated with Clostridium. At position 4 (Additional le: Figure S2-D), CA and butyric acid were negatively correlated with Methanobrevibacter, while butyric acid and lactic acid were positively correlated with Methanobacterium.
In summary, physicochemical factors such as TN, AP, and organic acids played an important role in determining the microbiota composition, and microbes in different spatial locations were affected by different factors, because the microbial composition differed spatially.

Metabolomic analysis of PM
In the untargeted metabolomics analysis, we identi ed 5991 metabolites, of which 714 were identi ed by gas chromatography (GC-MS), and 5277 were identi ed by liquid chromatography (LC-MS) from PM samples collected over different years that are continuously used. After ltering and quality control (deletion of unknown compounds), 982 metabolites were left for further analysis.

Classi cation of total metabolites
We rst classi ed the metabolites detected by GC-MS and LC-MS. The results showed that amino acids, peptides, and analogues, fatty acids, and conjugates were the main metabolites and accounted for approximately half of the metabolites (Additional le: Figure S3). Metabolites from PM also contribute to the liquor avour. Fatty acids, amino acids, and peptides are decisive compounds in CSFL because they can produce avour compounds as precursors in CSFL production.

Analysis of differential metabolites
We selected the metabolites according to the standard of P value <0.1 and VIP value> 1 to obtain the differential metabolites in different cell types.
In the position under the Huangshui uid, there were 129 differential metabolites in the 30-and 100-yearold cellars (Additional le: Figure S4-A), of which 105 were up-regulated and 24 were down-regulated.
There were 86 differential metabolites in the 30-and 300-year-old cellars (Additional le: Figure S4-B), of which 60 were up-regulated and 26 were down-regulated. Of 100 differential metabolites found in the 100-and 300-year-old cellars (Additional le: Figure S4-C), 41 were up-regulated and 59 were downregulated.
In the position of the Huangshui uid, 68 differential metabolites were found in the 30-and 100-yearold PM samples (Additional le: Figure S4-D), of which 29 were up-regulated and 39 were down-regulated, and of the 247 differential metabolites in the 30-and 300-year-old cellars (Additional le: Figure S4-E), 95 were up-regulated and 152 were down-regulated. There were 178 differential metabolites in the 100-and 300-year-old cellars (Additional le: Figure S4-F), of which 77 were up-regulated and 101 were downregulated.

Analysis of differential metabolic pathways between different cellars
The main avour compounds in CSFL are esters, alcohols, aldehydes, ketones, phenol, and pyrazine compounds. According to our results, differential metabolites were related to a variety of metabolic pathways. We selected differential metabolites for their potential participation in the substrate degradation and avour development of PM. Dodecanoic, caprylic, myristic, arachidonic, γ-linolenic, and eicosapentaenoic acids were related to the biosynthesis of fatty acids and unsaturated fatty acid pathways, while threonine, asparagine, phenylalanine, isoleucine, aspartic acid, valine, proline, and phenylpyruvic acid were related to the biosynthesis of amino acids. Malic acid, succinic acid, and γaminobutyric acid were related to the butanoate and pyruvate metabolism pathways. The relative concentrations of the metabolites were also compared. Under the Huangshui uid (Table S1), compared to the 30-year-old samples, the compounds γ-aminobutyric acid, threonine, proline, phenylalanine, dodecanoic acid, succinic acid, and valine were increased in the 100-year-old PM samples and decreased in the 300-year-old PM samples. The concentrations of arachidonic acid, eicosapentaenoic acid, caprylic acid, malic acid, myristic acid, isoleucine, glyceric acid, and β-hydroxypyruvate in the 30-year-old PM samples were higher than in the 100-and 300-year-old PM samples. The concentrations of phenylpyruvic acid and γ-linolenic acid in the 300-year-old samples were higher than in the 30-and 100-year-old PM samples. At the position of the Huangshui uid (Table S2), compared to the 30-year-old samples, the compounds dodecanoic acid and α-linolenic acid were increased in the 100-year-old PM samples and decreased in the 300-year-old samples. The concentrations of homoserine, saccharopine, threonine, succinic acid, and aminoadipic acid in the 30-year-old PM samples were higher than in the 100-and 300year-old PM samples. The concentrations of γ-linolenic acid, aspartic acid, caprylic acid, glucuronic acid, myristic acid, asparagine, and arachidonic acid in the 300-year-old PM samples were higher than in the 30-and 100-year-old PM samples. Figure 6A shows the reciprocal connections between the differential metabolites and KEGG pathways.
We further analysed the relationship between core microbiota and PM metabolites (Additional le: Figure  S5). The correlation coe cients indicated a strong relation (0.6<|Spearman's|<1). Seven of the nine main genera were involved in the distribution of certain metabolites. Positive correlations were observed between Caproiciproducens and isoleucine, gluconic lactone, and glyceric acid, Lactobacillus and CA, Methanobrevibacter and lactic acid, and Bellilinea and Proteiniphilum and caprylic acid.  Figure S6).

Functional gene pro les of the PM metagenome
To investigate the metabolic potential of the PM microbiome, microbial genes aligned against the KEGG database among all PM samples were categorised into three levels of pathways based on metagenomic sequencing data. The functional gene pro le at level 2 was mainly composed of carbohydrate metabolism, amino acid metabolism, energy metabolism, and translation metabolism in PM samples (Additional le: Figure S6-C).
We selected 135 enzymes from the KEGG database for their potential participation in the substrate degradation and avour development of PM, and they were then grouped into 20 functional assemblies.
Starch and sucrose metabolism are the primary sources of all compounds in the process of fermentation. According to starch and sucrose metabolism, the key enzymes alpha-amylase (EC 3.2.1.1) and glucan 1,4-alpha-maltohydrolase (EC 3.2.1.133) can catalyse starch hydrolysis to dextrin and maltose. The major producers of these two enzymes were Clostridiales (mainly Clostridium), Anaerolineales (mainly Bellilinea), Methanomicrobiales (mainly Methanoculleus), and Methanosarcinales (mainly Methanosarcina). Glucose is generally generated from starch during fermentation and is converted to pyruvate through the activities of aldose 1-epimerase (EC 5.1.3.3) and glucokinase (EC 2.7.1.2), while aldose 1-epimerase catalyses the interconversion of α-and β-terminal isomers of hexose, which are mainly produced by Petrimonas, Prevotella, and Clostridium. Moreover, glucokinase is related to the degradation of glucose and can catalyse glucose phosphorylation, which is mainly produced by Petrimonas, Clostridium, Prevotella, Methanoculleus, and Syntrophomonas. The above-mentioned microbiota genera were considered the major users of polysaccharides, and the generated glucose could be utilised by other taxa.
As for the avour-generation metabolism pathways, CA is generally produced by anaerobic bacteria via the chain elongation pathway. The oxidation of ethanol can provide energy (acetyl-CoA) to sequentially elongate the carbon chain of carboxylic acids (acetic acid to n-butyric acid to CA Metabolic network related to substrate-avour metabolism in PM microbiota To better explain the link among the 20 functional assemblies summarised in Fig. 5, a speci c metabolic network in PM microbiota among the nine orders and 30 microbial genera is presented schematically in Fig. 6, showing the pattern of enzyme-coding genes involved in the substrate-avour metabolic pathway.

Discussion
CSFL, a fermented and traditional distilled alcoholic beverage, is produced by spontaneous solid-state fermentation containing various microbes and their complex interactions. The standard production of Chinese liquor involves four steps, including cooking, sacchari cation, alcohol fermentation, and distillation, using grain as the main raw material. In CSFL production, the core microorganisms in PM play important roles in the characteristics of avour metabolites and determine the quality and safety of CSFL to a large degree.
PM cellar is a microbial ecosystem that provides a natural model for a broader understanding of microbiota composition and succession. Combining 16S rRNA, ITS full-length sequencing, metabonomics, and metagenomic analysis may be valuable for elucidating the microbial community composition, differential metabolites, and the relationship between the microorganisms and the metabolites in PM samples. In our study, sequencing data revealed that PM microbiomes are commonly dominated by bacteria such as Firmicutes, Euryarchaeota, Bacteroidetes, Chloro exi, Petrimonas, Proteiniphilum, Prevotella, Bellilinea, Caproiciproducens, Clostridium, and Lactobacillus and by archaea such as Methanobacterium and Methanobrevibacter (Fig. 3). Previous studies have shown that prokaryotic communities of 30-, 300-, 40-, and 400-year-old PMs were only found to have prokaryotic abundance differences, without prokaryotic structural differences between them [27] , and Clostridium, Caproiciproducens, Syntrophomonas, Sedimentibacter, and Methanobrevibacter were clearly the predominant microorganisms in mature PM [28,29] . The difference in the PM bacterial groups in different studies could be explained by a combination of methodological and environmental factors [30] . Along with the relatively steady eco-environment and microbiota structure formed through long-term fermentation progress, Caproiciproducens, Prevotella, and Lactobacillus were identi ed as the most important bacteria at the bottom of the pit, while Clostridium, Bellilinea, and Proteiniphilum showed a higher relative abundance in the Huangshui uid of the pit. The accumulation of Lactobacillus, Clostridium, and other acid-producing microorganisms at the bottom of the pit leads to an increase in organic acid content (Fig. 2). It is notable that the relative abundance of Methanobacterium and Methanobrevibacter was 1.44% to 4.58% and 2.54% to 34.42% in our PM samples, respectively, but the average abundance of Methanobrevibacter was less than 0.05% in new PM samples in a previous study [31] . Low pH inhibits methanogenic activity in anaerobic biological processes [32] . The archaeal genera in PM can utilise H 2 , acetate, methanol, and methylamine [33] . In addition to fermentative bacteria, methanogens can enhance organic acid production through syntrophic interactions [34] .
For fungal microbiota, we did not detect fungi in the 30-, 100-, and 300-year-old PM samples. Previous studies have shown that fungi were not detected in the older PM and were only detected in the new cellars that were used for 2 years [18,35] . There are several possible reasons for this: (1) the long fermentation environment is not suitable for the growth of fungi, and oxygen de ciency in the inner compartment partially leads to a low survival rate of fungi, especially Aspergillus [36] . (2) The presence of fungi in PM can also be attributed to the addition of the Daqu starter. It is known that the Daqu starter contains abundant microbes such as Lactobacillus, Pseudomonas, Aspergillus, Saccharomyces, Wickerhamomyces, and Pichia, and enzymes such as α-amylase, β-amylase, glucoamylase, and proteases [37,38] . During long-term use of the pit, microbes may undergo exchanges between the PM and fermented grains during the fermentation process of the raw material. Therefore, bacteria are more dominant than fungi in older PM samples [9] .
Function enzymes related to substrate degradation pathways Glycolysis is a central pathway in carbohydrate metabolism and is regarded as a feeder that prepares glucose for further catabolism and energy conservation [39] and has many steps that lead to the catabolism of glucose into pyruvate [40] , which diverges according to the availability of oxygen. Thus, the most dominant raw material synthesising the four dominant acids in PM is starch. In our study, the nine orders were all related to starch and sucrose metabolism and glucose utilisation pathway. Moreover, the absolute abundance of Clostridiales, Bacillales, and Lactobacillales were highest in the 300-year-old PM, suggesting that the amount of starch in the 300-year-old PM samples, which turned into the four dominant acids, was much greater than in the 30-and 100-year-old PM samples. The potential capacity for the comprehensive utilisation of different substrates may contribute to their dominance in the entire microbial community.
Function enzymes related to avour generation pathways Like beer and wine, the synergistic effect of microorganisms is thought to be responsible for the production of various avour compounds, such as CA, butyric acid, acetic acid, lactic acid, and alcohols [3,41] , in CSFL production [42] . Previous studies have shown that pyruvate, ethanol, glutamate, and 4aminobutanoate are related to the four dominant acid syntheses, especially pyruvate and ethanol [14,43] .
Thus, investigating the ethanol and organic acid metabolism pathways and the core enzymes is vital for improving CSFL quality. Ethanol is the main component of CSFL and is mainly generated through the conversion of pyruvate to acetaldehyde and reduction to ethanol. Some genera such as Bacteroides, Porphyromonas, and Sedimentibacter have been reported to produce succinic acid, propionic acid, and alcohols by using starch and glucose [44] . Increasing the content of ethyl CA is an effective method for improving the quality of CSFL. The common synthesis method of CA can occur via carboxylic acid chain elongation using ethanol as an electron donor [45] . Acyl-CoA hydrolase is the nal enzyme in CA generation and has been previously observed to be enriched within Clostridium spp., indicating that the Clostridium genus was the most important microbiota participating in CA production. Generally, some of the most CA-producing Clostridium species include C. kluyveri [6] and Clostridium sp. BS-1 [46] . In addition, the CA production is a process of H 2 and CO 2 production [47] , and C. ljungdahlii could synthesize acetic acid and ethanol from H 2 and CO 2 produced by C. clarkii [48] . Concurrently, the presence of hydromethanogens and CA-producing bacteria can lead to interspeci c hydrogen transfer, which is bene cial to increase CA production [14] . Thus, the co-occurrence of the high abundance of Clostridium, Methanobacterium, and Methanobrevibacter may indicate e cient CA production.
Lactic acid not only can inhibit some bacteria but also is an ethyl lactate precursor [49] , and it contributes to the formation of other avour substances [27] . Clostridiales (mainly Clostridium) and Lactobacillales (mainly Lactobacillus) were the important potential orders for producing lactic acid in our study. Lactobacillus produces lactic acids from sugar by homofermentative metabolism or produces alcohols in addition to lactic acid via heterofermentative metabolism [50] . In addition, ldh (EC 1. catalyse too much lactic acid to pyruvate and then generate acetic acid, butyric acid, and CA [51] , which could contribute to the recovery of pH of PM and enhance the production of organic acids. Moreover, the decrease in lactic acid is due to its catalysis not only to pyruvate, but also to acetic acid through EC 1.13.12.4. Excess lactate can lead to high levels of ethyl lactate, which can deteriorate CSFL quality [8] . Key genes (e.g., alcohol dehydrogenase and L-lactate dehydrogenase) involved in reverse β-oxidation with both ethanol and lactate as substrates were found in PM samples, indicating that the PM microbiome possesses metabolic potential for CA production from ethanol and lactate. Our results veri ed the importance of lactic acid in CA production. Lactobacillus was positively correlated with CA (Additional le: Figure S5), because lactic acid is mainly produced by Lactobacillus and converted into CA by Caproiciproducens [52] .
Butyric acid, one of the crucial avour substances in CSFL, is synthesised through two alternative pathways, namely butyrate kinase (EC 2.7.2.7) and butyryl-CoA:acetate CoA-transferase (EC 2.8.3.8) [53] . A previous study showed that Clostridium was most likely involved in butyrate production through the buk pathway in PM [14] . In our study, besides Clostridiales and Bacteroidales, we also found that Synergistales (mainly Aminobacterium) have the ability to produce enzymes of butyrate kinase (EC 2.7.2.7) and butyryl-CoA: acetate CoA-transferase (EC 2.8.3.8). In addition, butyric acid can provide more precursors for CA synthesis [14] . For example, Sedimentibacter and Aminobacterium can ferment amino acids to acetic and butyric acids [54,55] . Butanoate metabolism and pyruvate metabolism pathways were related to the production of CA, butyric acid, and lactic acid, and two metabolites, malic acid and succinic acid, were enriched, indicating their importance for the production of organic acids.
Ethyl acetate produced by acetic acid was also important for the formation of avour in CSFL, and in addition to L-lactic acid, more acetic acid was produced by acetyl-CoA through the catalytic action of pct (EC 2.8.3.1) and acdA (EC 6.2.1.13). The major genus sources of butyrate kinase (EC 2.8.3.1) were Clostridium and Lactobacillus, which had a high absolute abundance in the 300-year-old cell, while the major genus sources of acdA (EC 6.2.1.13) were Methanosarcina, Methanobacterium, Methanobrevibacter, Bellilinea, and Syntrophomonas. Syntrophomonas can degrade long-chain fatty acids (C 4 -C 8 ) into acetic acid, propionic acid, and H 2 [56] . Furthermore, speci c members of the Clostridium species were also identi ed with acetate biosynthesis from various substrates, such as sugar, starch, and cellulose.
Methane metabolism is also an important pathway during pit fermentation because it can enhance the production of organic acids during fermentation. Methanosarcina co-cultured with bacteria can metabolise carbon dioxide to methane [57] . Methanobrevibacter can produce CH 4 from H 2 and CO 2 [58] , and the interspecies hydrogen transfer between archaea and Clostridiaceae (mainly Clostridium) plays an important role in improving the quality of CSFL [59] . In our study, the Methanomicrobiales (mainly Methanoculleus) and Methanobacteriales (mainly Methanobrevibacter and Methanobacterium) were the most abundant microbiota orders participating in methane metabolism. Further, the genera Caloramator and Clostridium, belonging to Clostridiaceae, mainly produced the enzymes EC 5.4.2.12 and 1.17.1.9. In our study, Methanobrevibacter was positively correlated with lactic acid by metabonomics (Additional le: Figure S5), indicating that Methanobrevibacter played an important role in the production of lactic acid.

Environmental factors drive microbiota succession in PM
Microbial community succession is a complex long-term process. Physicochemical factors are believed to be the main determinants that in uence the shift of the microbiota community and structure in the soil and PM [19] . This study revealed that organic acids such as acetic acid, CA, butyric acid, and lactic acid were the most important indicators in uencing microbial community structure in the 30-, 100-, and 300year-old PM samples (Fig. 4B). The organic acid content was highest at position 1 of the cellar, especially in the 300-year-old PM samples (Fig. 2). Previous studies revealed that lactic acid and pH were the two important factors in uencing microbiota composition in 1-, 10-, 25-, and 50-year-old PM samples.
Hartman et al. [60] also found that pH was the most essential environmental factor in uencing the bacterial community structure in soil, and pH broadly altered the prokaryotic diversity and affected speci c taxa, including Acidobacteria and Actinobacteria. In our study, pH and moisture were stable in the cellars and in uenced a few bacteria. The moisture content can affect the microbiota composition, which may be related to the lack of free water available to PM microorganisms. In our study, the moisture content gradually increased with cellar use and was higher in positions 1 and 2 than in other spatial locations, especially for the 100-and 300-year-old cellars. This may be because PM was soaked in Huangshui for a long time during long-term CSFL fermentation, thus leading to a high moisture content. Simultaneously, ammonium nitrogen can provide a source of nitrogen for microbial growth and has an important in uence on the abundance of PM microbiota [19] . The concentrations of TN and AP increased with the cellar age and showed certain spatial rules; the content of AP and TN in position 1 was higher than in other spatial locations. Therefore, the microbial community structure can be adjusted by adjusting environmental factors such as moisture and pH of the PM to achieve the purpose of increasing CA and decreasing lactic acid content [61] . Consequently, a challenge in Chinese liquor production is to achieve a reduction in ethyl lactate synthesis during CSFL production.
In our study, we investigated the core microbiota composition in PM using sequencing technologies and identi ed metabolites using GC-MS and LC-MS. Amplicon analysis revealed fundamental information about microbial succession, and RDA revealed positive correlations between the core microbiota and major endogenous factors. The metagenomic data illustrated the effectiveness of the analysis in increasing our understanding of the possible mechanisms responsible for the microbial community structure in PM, providing a reference for further utilisation of PM microbiota genetics and paving the way for the optimisation of CSFL production. However, a limitation of this study is that some microorganisms present throughout the process may have died and left their DNA in the samples. This process requires further exploration via isolation of microbes and functional genomic analysis.

Conclusions
These cellars have been used for several years without interruption, providing an ideal opportunity to uncover the effect of cellar age on the microbial succession and functional diversity of the microorganisms in PM. Although the different cellars harboured the taxonomically and functionally diverse and abundant microbial community, the functional microbiota related to avour generation as a hotspot for microbial diversity, genes for avour generation, and substrate degradation were identi ed throughout the 30-, 100-, and 300-year-old cellars. However, metagenomic datasets revealed that the pathways and taxa involved in avour generation and substrate degradation were complex and diverse. Clostridiales (mainly Clostridium) and Bacteroidales (mainly Petrimonas and Proteiniphilum) are major microbiota orders for avour generation and substrate degradation pathways, while archaea also play an important role, in cooperation with bacteria, in various avour generation pathways. Our data help elucidate the core microbiota composition, different metabolic roles of microorganisms, and disclose the formation mechanism of PM partial functions, providing a basic theory to support the regulation of Baijiu production.

Detection of PM physicochemical properties and organic acids
The moisture content of PM was measured using a dry/wet weight measurement method after drying at 105°C for 6 h. The pH of fresh PM was determined in a 1:10 (w/v) ratio in deionised water using a pH meter (PB10; Sartorius, Gottingen, Germany). The concentration of available phosphorus (AP) was detected by extracting with ammonium uoride and hydrochloric acid [62] . Total nitrogen was extracted with Seignette salt and ammonium chloride and its concentration measured using UV-visible spectrophotometry. Organic acids such as Caproic acid, butyric acid, and acetic acid in the PM were detected by GC-MS: 1 g PM samples and 2 mL 60% ethanol were placed in a 5-mL centrifuge tube with a vortex mixer. After 30 minutes of extraction, the mixture was centrifuged at 10000 rpm for 10 minutes.
Afterwards, the supernatant was collected through the membrane for GC-MS analysis. The conditions were as follows: the inlet temperature was 230℃, the carrier gas was high-purity helium, and the ow rate was 1.0 mL/min. Moreover, the following gradient conditions were used: a temperature of 35℃ maintained for 10 min and then increased to 60℃ at 4℃/min and maintained for 4 min. The temperature was further increased to 195°C at 6°C/min for 20 min. Lactic acid in the PM was quanti ed using HPLC: 1 g PM samples and 2 mL ultra-pure water were placed in a 5-mL centrifuge tube, and after 30 min of extraction, the supernatant was collected through the membrane for HPLC analysis, as described previously [63] .
DNA extraction, 16S rRNA gene, and ITS amplicon sequencing 16S rDNA and ITS amplicon data analyses The original FASTQ le was processed using QIIME software [64] , and the low-quality sequences were ltered out. After analysing the sequencing data, the SILVA database (v13.2) was used to compare the 16S rRNA and ITS gene sequences to determine the taxonomic status of the corresponding microbes. QIIME software was used to de ne sequence similarity >97% as an operational taxonomic unit (OTU) [65] .
Only OTUs containing at least ve reads were considered valid in this study, and we selected a representative sequence of each OTU for taxonomic analysis. The alpha diversity index was calculated to analyse species richness and uniformity in the samples, after which the Shannon and Chao1 indices of the samples were determined to obtain the diversity of the community and the number of OTUs in the samples. The genus-level clustering results for each sample were used as inputs for LEfSe analysis. Finally, the beta diversity index was determined to analyse the heterogeneity of community composition between samples.
Metagenomic sample preparation, processing, and assembly of metagenomic sequencing data