Unraveling the Contribution of High Temperature Stage to Jiang-Flavor Daqu, a Liquor Starter for Production of Chinese Jiang-Flavor Baijiu, With Special Reference to Metatranscriptomics

Jiang-flavor (JF) daqu is a liquor starter used for production of JF baijiu, a well-known distilled liquor in China. Although a high temperature stage (70°C) is necessary for qualifying JF daqu, little is known regarding its active microbial community and functional enzymes, along with its role in generating flavor precursors for JF baijiu aroma. In this investigation, based on metatranscriptomics, fungi, such as Aspergillus and Penicillium, were identified as the most active microbial members and 230 carbohydrate-active enzymes were identified as potential saccharifying enzymes at 70°C of JF daqu. Notably, most of enzymes in identified carbohydrate and energy pathways showed lower expression levels at 70°C of JF daqu than those at the high temperature stage (62°C) of Nong-flavor (NF) daqu, indicating lowering capacities of saccharification and fermentation by high temperature stage. Moreover, many enzymes, especially those related to the degradation of aromatic compounds, were only detected with low expression levels at 70°C of JF daqu albeit not at 62°C of NF daqu, indicating enhancing capacities of generating special trace aroma compounds in JF daqu by high temperature stage. Additionally, most of enzymes related to those capacities were highly expressed at 70°C by fungal genus of Aspergillus, Coccidioides, Paracoccidioides, Penicillium, and Rasamsonia. Therefore, this study not only sheds light on the crucial functions of high temperature stage but also paves the way to improve the quality of JF baijiu and provide active community and functional enzymes for other fermentation industries.


INTRODUCTION
Baijiu (Chinese liquor), one of the oldest known distilled liquors with an approximate 2000-year history, is the largest consumed spirit globally (over 13 billion liters in 2016) (Liu and Sun, 2018). Compared with whisky and brandy, baijiu is well known for its taste with more flavor compounds (>1870 volatile compounds) in liquor, including alcohols, aldehydes, organic acids, esters, phenols, lactones, heterocycles, terpenes, aromatic compounds, amino acids, and peptides, which leads to the final special aroma and health of baijiu (Jin et al., 2017;Liu and Sun, 2018). Thus, based on its distinctive flavor characteristics, baijiu can be divided into three major categories [i.e., Jiang-flavor (JF, also called sauce-flavor) baijiu, Nong-flavor (NF) baijiu and Qingflavor (QF) baijiu] and nine minor categories, among which JF baijiu is with a full-bodied long-lasting aroma (Zheng and Han, 2016;Liu and Sun, 2018). The representative JF baijiu is moutai, the most famous baijiu, having the distinction as "the national liquor" and largely dominating the market in China (Zheng and Han, 2016;Jin et al., 2017). JF baijiu is fermented and distilled under solid-state conditions with a production process that mainly includes four distinct stages; i.e., daqu preparation (approximately 4 months), stacking fermentation (2-4 days), alcoholic fermentation and distillation processes (Figure 1) (Fan et al., 2012;Zheng and Han, 2016).
Typically, daqu, a liquor starter used to initiate the alcoholic fermentation process, constitutes the most essential component for alcoholic fermentation, not only providing the microbial community and enzymes (as a saccharifying and fermenting agent) for alcoholic fermentation but also significantly contributing to the final liquor flavor (Zheng and Han, 2016;Liu and Sun, 2018). Similar liquor starters can be found in many Asian countries, e.g., xiaoqu/fuqu in China (Zheng and Han, 2016;Jin et al., 2017), meju in Korea (Kim et al., 2011), ragi in Indonesia (Fibri and Frøst, 2019), marcha/thiat/dawdim/hamei/chowan in India (Sha et al., 2018), bubod in Philippines (Tamang et al., 2016), and banh men in Vietnam (Thanh et al., 2008). All those liquor starters are prepared in an open system with starchy materials (wheat, rice, etc), shaped into different sizes and shapes, and cultured under different conditions (temperature and time) (Liu and Sun, 2018;Sha et al., 2018;Waché et al., 2018). Among them, JF daqu is made from wheat, shaped into brick, and produced with two steps of spontaneous fermentation for approximately 1 month in a Qu-room and drying for another 3 months without ventilation in a storage room (Huang et al., 2017b;Jin et al., 2017). During the spontaneous fermentation process, the cultivation of JF daqu is controlled by manually turning over the bricks and opening/closing the windows to change the ventilation and temperature, with the special microbial community being enriched from raw materials and the working environments by environmental variables (temperature and moisture), among which temperature serves as a key driving force (Huang et al., 2017a;Xiao et al., 2017;Liu and Sun, 2018). According to the maximum temperature in the daqu preparation process, JF daqu is grouped into high-temperature (60-70 • C) daqu and requires cultivation at the high temperature stage for approximately 7-8 days (Huang et al., 2017b;Liu and Sun, 2018). Owing to this high temperature stage, the thermophilic microbial community may be enriched in JF daqu and various thermostable enzymes (i.e., proteinase, glucoamylase, cellulase, alpha-amylase, and esterase) may also be produced to degrade materials and generate special flavor compounds.
Recently, the daqu microbial community has been studied throughout fermentation by culture-dependent and -independent methods, and their diversity and dynamics are well understood (Yan et al., 2013;Wang and Xu, 2015;Huang et al., 2017b;Xiao et al., 2017). However, little is known regarding the active microbial community and their metabolic functions. In addition, although numerous crude enzymes have been identified in daqu (Li et al., 2015;, active enzymes and their relationships with the microbial community are yet unknown. Metatranscriptomics constitutes an ideal tool for studying daqu microbial ecology, as it directly analyzes mRNA from environments and provides information not only on the microbial community composition but also on active members and their specifically expressed enzymes (Bokulich et al., 2016). This technology has been successfully applied in microbial ecological systems; e.g., compost (Mello et al., 2017), mouse gut (Just et al., 2018), cattle rumen (Pandit et al., 2018), sludge (Xia et al., 2018), ocean (Yoshida et al., 2018), and human feces (Abu-Ali et al., 2018). Nevertheless, owing to the complicated conditions in baijiu brewing systems, such as the high content of starch and fermentation products along with strongly colored materials, it remains challenging to extract high-quality RNA from baijiu fermentation samples, especially from the high temperature stage (70 • C) of JF daqu, in which greater amounts of fermentation products were generated with strong colors than in all the other daqus' making stages. Thus, to our knowledge, only samples from the JF alcoholic fermentation process (42.8 • C) (Song et al., 2017) and Nong-flavor (NF) daqu (a mediumtemperature daqu) preparation process (62 • C) (Huang et al., 2017a) have previously been studied using metatranscriptomics.
The cultivation temperatures in the production process for JF daqu, a typical high-temperature daqu, are largely higher than those in other daqus including NF daqu (Huang et al., 2017b). The high temperature condition constitutes the most striking difference among the daqu production processes of JF daqu and other daqus, as well as their subsequent alcoholic fermentation processes, which results in unique microbial community, enzymes, and aroma compounds being generated in the JF daqu and fermented feedstock (Wu et al., 2009;Wang et al., 2014;Xiao et al., 2016;Liu and Sun, 2018). Compared with NF daqu, JF daqu has a lower capacity for saccharification, liquefaction, and fermentation (Zheng and Han, 2016;Liu and Sun, 2018), thus requiring the use of a large amount of JF daqu (nearly 1:0.9 ratio of daqu versus feedstock) in the alcoholic fermentation process, which is higher than that of NF daqu (approximately 1:2 ratio). Thus, the flavor precursors, enzymes, and microbial community enriched in JF daqu would likely be more strongly associated with the final liquor flavor than those in NF daqu. Recently, we have published breakthrough research wherein significant differences were predictively shown in energy, carbohydrate metabolism, FIGURE 1 | Process diagram of traditional production of Jiang-flavor (JF) baijiu. JF daqu was produced with two steps of spontaneous fermentation in a Qu-room and drying without ventilation in a storage room, and its maximum cultural temperature was usually between 60 and 70 • C. Besides stage of daqu preparation, JF baijiu was produced with other three distinct stages of stacking fermentation, multiple alternations of alcoholic fermentation and distillation processes. and degradation of aromatic compounds between the JF daqu and NF daqu bacterial community (Huang et al., 2017b), and the active microbial community was found to highly express pivotal enzymes at the high temperature stage of NF daqu making process (Huang et al., 2017a). However, the active microbial community and important enzymes, as well as their functional correlations in JF daqu remain to be identified. More specific understanding regarding differences of the high temperature stage between JF and NF daqu have not been clarified. Therefore, in this study, we first employed metatranscriptomics to explain the structure and function of the actual microbial community and its pivotal enzymes at the high temperature stage of JF daqu making process. Moreover, a comprehensive and global comparison was performed between JF and NF daqu to shed light on functions of the high temperature stage with regard to saccharification and fermentation along with flavor compound generation. This study provides fundamental information related to the active microbial community and functional enzymes and may facilitate a comparative understanding of the pivotal role of the high temperature stage in the JF daqu making process and JF baijiu brewing.

Sample Collection
JF daqu samples were collected at different time points from a fermentation workshop of Kweichow Hanwang Group Co., Ltd. in Renhuai, Guizhou, China, as described previously (Huang et al., 2017b). Briefly, Sample J1 was harvested at the beginning of daqu production (30 • C); J2 was harvested after 3 days of daqu preparation (55 • C); J3 was harvested after 8 days of daqu preparation (70 • C); and J4 was harvested from the mature daqu after fermentation for 20 days (25 • C) (Figure 1). In addition, all samples were selected and mixed from three locations in the same Qu-room at each time point. For RNA extraction, the daqu samples were frozen in liquid nitrogen immediately after collection, transferred to the Chengdu Biology Institute, Chinese Academy of Sciences on that day and stored in a −80 • C freezer. For enzyme analysis, all the samples were suspended in 0.1% (v/v) Tween 80 solution and transferred to the institute at room temperature (Huang et al., 2017a).

Carbohydrate-Degrading Enzyme Activities
A total of 18 polymer analogs of insoluble chromogenic AZurine Cross-Linked (AZCL) polysaccharides (Megazyme, Ireland) were selected for detecting enzyme activities on cellulose, hemicellulose, starch, chitin, and glucan degradation ( Table 1). As in our prior study (Huang et al., 2017a), all daqu samples in 0.1% (v/v) Tween 80 solution were incubated at 25 • C and 100 rpm overnight, then their supernatants were added directly onto the wells of solid plates with AZCL polysaccharides according to the manufacturer's protocol. After incubation at 35, 45, or 55 • C for 22 h, carbohydrate-degrading enzyme activities were determined by measuring the diameter of the blue haloes, which were recorded in millimeters.

RNA Extraction and Sequencing
Similar to the RNA extraction from NF daqu (Huang et al., 2017a), total RNA was extracted from JF daqu samples using borate buffer, cleaned with the RNeasy Midi Kit (Qiagen #75142, Venlo, Netherlands) and treated with DNase I (Fermentas, Waltham, MA, United States) according to the manufacturer's protocols. The RNA integrity was evaluated by gel electrophoresis and RNA integrity number (RIN) was checked using an Agilent2100 Bioanalyzer (Santa Clara, CA, United States). RNA samples with RIN value greater than 7.0 and OD260/OD280 ratio greater than 1.8 were selected for deep sequencing.
Total RNA (approximately 20 µg) from J3 was used for the RNA sequencing. Prior to metatranscriptomic library construction, using a previously reported method (Huang et al., 2017a), mRNA was isolated using magnetic beads with Oligo (dT) for eukaryotes, and for prokaryotes, mRNA was obtained after removing ribosomal RNA. The isolated mRNA was first fragmented and then used as template for subsequent first-and second-strand cDNA synthesis with random primers. Short cDNA fragments were purified and resolved with EB buffer for end reparation and poly(A) addition. Thereafter, the short cDNA fragments were ligated to sequencing adapters and suitable sized cDNA fragments were purified as templates for polymerase chain reaction amplification. RNA sequencing of the library was performed using platform (Illumina, San Diego, CA, United States) at the Beijing Genomics Institute (BGI)-the HiSeqTM 2000 Shenzhen, China.

Metatranscriptomics Assembly and Annotation
As for our previous metatranscriptomics assembly of NF daqu samples (N1-4), raw sequenced reads of J3 were first filtered by removing adaptors, low quality reads, and the rRNA sequences (Li et al., 2009). The clean reads of J3 were then de novo assembled using Trinity 1 (Grabherr et al., 2011), by which unigene sequences were generated. To annotate the metatranscriptome, the unigene sequences were aligned using Blastx (version 2.5.0) with protein and nucleotide databases including Nonredundant protein (NR), Non-redundant nucleotide (NT), Swiss-Prot, Kyoto Encyclopedia of Genes and Genomes (KEGG), Clusters of Orthologous Groups (COG), and Gene Ontology (GO) (e-value < 10 −5 ), and identified according to the highest similarity to known sequence. In cases of the non-alignment of unigenes against one of the listed databases, ESTScan was used to determine their coding directions. Thereafter, according to the standard codon usage, coding DNA sequences (CDSs) were translated into protein sequences. KEGG pathways were extracted from the KEGG web server 2 (Kanehisa et al., 2017). WEGO software 3 was used for GO classification (Ye et al., 2006). Carbohydrate-active enzymes (CAZymes) were retrieved from the Carbohydrate-Active Enzymes database (CAZy) 4 (Lombard et al., 2014).

Identification of Differentially Expressed Genes (DEGs) and Pathway Analysis
To compare the gene expression levels among J3 and NF daqu samples (N1-4), the predicted ORFs were combined after removing redundancy using cd-hit (Version 4.6.1) 5 (Li and Godzik, 2006). Gene expression levels were calculated using the Reads Per Kilobase per Million mapped reads (RPKM) method (Mortazavi et al., 2008). DEGs among J3 and NF daqu samples were identified using a method based on the Poisson distribution (Audic and Claverie, 1997). DEGs between two samples were identified using p-value ≤ 0.05, Log 2 (RPKM ratio) ≥ 1, and false discovery rate (FDR) value ≤ 0.001 (Benjamini and Yekutieli, 2001). To analyze GO enrichment, all DEGs were mapped to terms of the GO database.

Accession Number
The raw and assembled metatranscriptomics data of J3 have been deposited to the GenBank database under accession numbers SRR7785758 and GGWC00000000, respectively.

RNA Sequencing and Metatranscriptomics Assembly
After RNA sequencing of the J3 sample, 5.882 Gbp of raw data was generated, from which 5.663 Gbp of clean data was then obtained by filtering (Supplementary Table S1). These clean data were de novo assembled, from which 38,899 unigenes were identified with a total length of 46,187,298 nucleotides (nt) and N50 length of 2232 bp (Supplementary Table S2). As shown in Supplementary Figure S1, there were 3585 unigenes with sequence size > 3000 nt.

Functional Annotation and Classification of Unigenes
To annotate the unigenes of J3, blastx alignment against the protein and nucleotide databases of NR, NT, Swiss-Prot, KEGG COG and GO was performed; the results are shown in Supplementary Table S3. The CDSs that mapped to the protein database and were predicted by ESTscan numbered 30,615 and 1041, respectively. A total of 31,279 known unigenes were identified by blastx, among which 14,912 genes were annotated by COG classification. There were 25 classes in the COG classification with the largest number of unigenes being found solely in the class of "general function prediction" (15.1%; Supplementary Figure S2). In addition, 19,468 unigenes were also annotated by the GO database, which accounted for 50.1% of all the unigenes, with the annotations grouped into three categories (biological process; cellular component; and molecular function) (Supplementary Figure S3). "Metabolic processes, " "cell" and "catalytic activities" were dominant in the categories of biological processes, cellular components, and molecular functions, respectively.
Overall, 30,793 genes (e-value < 10 −5 ) were annotated using the NR database (Supplementary Table S3), which is far higher than those by other databases; the composition of active bacterial and fungal taxa in J3 is presented in Table 1. Based on their gene numbers, the active fungal community was more prevalent than the bacterial community and accounted for 97.7% in J3. In the fungal component, Aspergillus and Penicillium were the pivotal genera with high relative abundances of 53.2 and 29.2%, respectively. In addition, the active yeast showed low relative abundances of 0.2% at this high temperature stage.
As shown in Figure 2, starch and sucrose metabolism had the highest number of unigenes in J3, and except for the citrate cycle (TCA cycle) and oxidative phosphorylation, all of the 30 most abundant KEGG pathways showed higher numbers of unigenes in J3 than those in the high temperature stage (N3) of NF daqu. Moreover, large differences were found between J3 and N3 in basic metabolisms (i.e., purine metabolism, RNA degradation, RNA transport, meiosis-yeast, MAPK signal pathway-yeast, cell cycle-yeast, spliceosome, and mRNA surveillance pathway), degradation of aromatic compounds (aminobenzoate degradation, naphthalene degradation, benzoate degradation and bisphenol degradation), starch and sucrose metabolism, and amino sugar and nucleotide sugar metabolism. Conversely, comparable numbers of unigenes were found between J3 and N3 in oxidative phosphorylation, FIGURE 2 | The 30 most abundant KEGG pathways in high temperature stage samples of J3 and N3. J3 was harvested after 8 days of JF daqu preparation and N3 was harvested after 9 days of NF daqu preparation. The temperatures of J3 and N3 were 70 and 62 • C, respectively. glycolysis/gluconeogenesis, butanoate metabolism, and pyruvate metabolism. Additionally, metabolism of amino acids; i.e., tyrosine, glycine, serine, threonine, arginine, and proline, were also ranked in the top 30 of both J3 and N3.

DEGs Among J3 and NF Daqu Samples
The DEGs between the J3 and NF daqu samples were identified and a heatmap of hierarchical clustering of DEGs was constructed using log 2 (RPKM ratio) to visualize the respective patterns of DEGs. As shown in Figure 4, numerous DEGs (union) in J3, N2, N3, and N4 were clearly up-regulated with high log 2 (rations) values when compared with N1; thus, J3 together with N2-4 exhibit the largest differences in DEGs compared with N1. Alternatively, J3 presented the smallest differences in DEGs with N3. Similar results among J3 and NF daqu samples were also observed by analysis of hierarchical clustering of inter DEGs (Supplementary Figure S4). Furthermore, a comprehensive comparison performed between J3 and N3 identified a total of 14,149 unigenes as significant DEGs including 506 up-and 13,642 down-regulated genes (Supplementary Figure S5). In addition, for the GO functional classification (J3/N3), numerous DEGs were grouped into four dominant categories: "cellular processes, " "metabolic processes, " "binding, " and "catalytic activities" (Supplementary Figure S6).

Pathway Comparisons of Starch and
Sucrose Metabolism, Glycolysis, Pyruvate Metabolism, and the Citrate Cycle Between J3 and N3 For further functional comparison of DEGs between J3 and N3, metabolic pathways were analyzed based on the KEGG database. Moreover, several key carbohydrate and energy metabolisms that were associated with relatively high numbers of unigenes were selected for comparative analysis including starch and sucrose metabolism, glycolysis, pyruvate metabolism, and citrate cycle pathways. As shown in Figure 5, enzymes related to these selected pathways were mainly present, and a complete metabolic process of converting polymers into end-products was apparent in both J3 and N3. In addition, the majority of enzymes in these four key pathways exhibited lower expression levels in J3 than in N3, with the exception of e.g., aldehyde reductase (1.  (Figure 5 and Supplementary Tables S5-S8).

DISCUSSION
The daqus of Chinese JF and NF liquor, the most consumed liquors in China, undergo markedly different production processes that make large contributions to their special flavors.
To ascertain the underlying factors, in comparison with our previous work of NF daqu, the present study comprehensively revealed the active microbial community and enzymes at the high temperature stage (J3) of JF daqu, and comparatively analyzed the active enzyme profiles at high temperature stages of JF and NF daqus. The active fungal community produced more diverse enzymes than those of the bacterial community, with Aspergillus and Penicillium representing the dominant genera at J3. This finding was complementary to the previous microbial diversity revealed for JF daqu by 16S rRNA and ITS sequencing, which indicated that the bacterial community was more diverse than the fungal community at J3 (Huang et al., 2017b). Meanwhile, low abundances of active yeast might be due to high temperature condition at J3, which may be well consistent with previous finding that yeast decreased quickly from J2 (55 • C) to J3 (70 • C) (Huang et al., 2017b). Additionally, the prevailing role of the active fungal community was also revealed in NF daqu samples by metatranscriptomics analysis (Huang et al., 2017a). Therefore, the present study further confirmed the suitability of metatranscriptomics for obtaining the active microbial community profiles in daqus. JF daqu exhibited lower numbers and expression levels of CAZymes at the high temperature stage of J3 than those at the high temperature stage of N3 (Huang et al., 2017a). In addition, except for the initial stage of J1, most CAZymes were detected with lower activities and less diversities in the production process of JF daqu samples than those of NF daqu samples ( Table 2) (Huang et al., 2017a). Lower activities and diversities of amylases (one kind of CAZymes) were similarly found in JF than in NF daqu via activities assay and protein electrophoresis . Thus, these findings might to some extent be consistent with the lower capacities of saccharification and liquefaction in JF daqu than those in NF daqu (Liu and Sun, 2018;. Additionally, several thermostable CAZymes were detected in special stages of JF daqu samples, such as α-amylase in J1 and J4, endo-β-1,3-1,4-glucanase in J3 and J4, endo-proteases in J2 and J4, and endo-1,4-β -D-xylanase in J4, which suggests the feasibility of mining thermostable enzymes from special stages in the future. Based on the functional annotation, starch and sucrose metabolism was the most abundant pathway in J3, which might imply that the microbial community has full capacity FIGURE 6 | Relative abundances of enzymes related to the degradation of aromatic compounds in J3 and N3. A total of 11 abundant pathways associated with the degradation of aromatic compounds were analyzed: aminobenzoate degradation, benzoate degradation, fluorobenzoate degradation, chlorobenzene degradation, ethylbenzene degradation, naphthalene degradation, bisphenol degradation, styrene degradation, xylene degradation, polycyclic aromatic hydrocarbon degradation, and toluene degradation. In these pathways, only the portion of enzymes with relatively high expression levels is presented by EC number and total RPKM. Relative expression [log 10 RPKM)] is shown in red for J3 and blue for N3. The key products are highlighted with black closed circles.
for degrading different polymers into glucose in J3. Upon comprehensive comparison between J3 and N3, most pathways showed higher diversities with more unigenes in J3 than in N3, which indicated more complicated metabolism for the microbial community in J3. Moreover, large differences of diversities were observed in basic metabolisms, degradation of aromatic compounds, starch and sucrose metabolism, and amino sugar and nucleotide sugar metabolism between J3 and N3, which further suggested that the microbial community of J3 might produce higher diversities of metabolites, some of which, such as phenol, benzaldehyde, and phenylethanol, might serve as precursors for aroma compounds (Fan et al., 2012;Wang et al., 2014;Xiao et al., 2016). In contrast, similar diversities were found in oxidative phosphorylation and glycolysis between J3 and N3, which indicated that the microbial communities released considerable bio-heat to maintain the high temperatures of 62 and 70 • C for several days in NF and JF daqu, respectively (Huang et al., 2017b). In addition, butanoate metabolism and pyruvate metabolism were also similarly active between J3 and N3, suggesting that their intermediates, such as butanoate and acetate, represented important substrates for flavor compounds of e.g., butanol, acetic acid, butanoic acid, ethyl hexanoate, hexyl acetate, and isopentyl butanoate in JF and NF liquor (Fan et al., 2012;Wang et al., 2014;Xiao et al., 2016). Furthermore, six amino acid metabolisms were dominant in both J3 and N3, the products of which, i.e., amino acids, a-keto acids, and aldehydes, may serve as pre-substrates for important flavor precursors such as pyrazine, alcohol, and acids (Badrinarayanan and Sperry, 2012;Nashalian and Yaylayan, 2014;Scalone et al., 2015;Xu et al., 2017). Therefore, the microbial community was both active at the high temperature stages of JF and NF daqu for generating bioheat (Huang et al., 2017a,b;Xiao et al., 2017) and releasing flavor precursors (Wu et al., 2009;Zheng et al., 2011), and JF daqu could provide larger diversities of flavor precursors than NF daqu from most of the active pathways, in particular from the degradation of aromatic compounds. Similar DEG profiles were observed between the high temperature stages of J3 and N3; thus, detailed functional comparisons of DEGs were performed between these stages with regard to four key carbohydrate and energy metabolisms: starch and sucrose metabolism, glycolysis, pyruvate metabolism, and the citrate cycle, as their intermediates are essential for ethanol and flavor generation. The results showed that both J3 and N3 contained an intact process for converting polymers into glucose, pyruvate, acetyl-coA, and ethanol, indicating a complete system for saccharification, liquefaction, and fermentation. In general, the majority of enzymes related to these four key pathways showed lower expression levels in J3 than in N3, indicating lower activities for enzymes in J3 than in N3 to a degree that is consistent with the lower capacities in saccharification, liquefaction, and fermentation exhibited by high-temperature JF daqu than those by medium-temperature NF daqu (Liu and Sun, 2018;. Low expression levels of enzymes might result from the inhibition caused by the high temperature (70 • C) in J3. However, some enzymes were only detected in J3, albeit with relative low expression levels, indicating that a large number of minor intermediates would likely be specifically generated in J3. Notably, among enzymes related to saccharification and liquefaction in J3, glucoamylases were clearly active with high expression levels, indicating their collaborative roles along with high temperature in degrading starches, which would be spontaneously decomposed under high temperature, as well as suggesting a feasible way to mine thermostable glucoamylases from J3. The majority of enzymes related to saccharification and liquefaction in J3 were highly expressed by fungal species of R. emersonii, A. oryzae, A. fumigatus, and C. immitis, some of which have been found to secrete numerous carbohydrateactive enzymes and show high capacities toward degrading polymers, such as Aspergillus (Culleton et al., 2013;de Vries et al., 2017;Cologna et al., 2018) and R. emersonii (Hua et al., 2014;Martínez et al., 2016). In addition, J3 showed considerable potential for converting glucose to pivotal intermediates, such as acetate, ethanol, pyruvate, and acetyl-coA, which might then serve as direct or indirect substrates for JF flavor compounds including ethyl acetate, ethyl butanoate, ethyl propanoate, ethyl 2-hydroxypropanoate, ethyl 2-hydroxyhexanoate, acetic acid, 2-acetylpyridine, hexyl acetate, benzyl acetate ethyl, ethyl 3-methylbutanoate, ethyl benzeneacetate, and 3-methylbutyl acetate (Fan et al., 2012;Wang et al., 2014;Xiao et al., 2016;Gao et al., 2017). The highly expressed enzymes related to glycolysis and pyruvate metabolism were mostly derived from fungal species, some of which have been applied to the production of fermented foods and drugs, such as A. fumigatus (Qin et al., 2012;Wakefield et al., 2017), A. clavatus (Mo et al., 2008;Zutz et al., 2013;Li et al., 2017), and A. oryzae (Park et al., 2018;Son et al., 2018;Zhong et al., 2018). Furthermore, low concentration of ethanol might be generated by several fungi in J3, which to some extent agreed with the earlier finding that a small amount of ethanol could be directly produced by co-culture of fungi (Takano and Hoshino, 2012). Additionally, relatively high expression levels of D-lactate dehydrogenase (cytochrome) might indicate high concentration of lactate in J3, which may be consistent with the high level of lactate in the subsequent mature JF daqu (Wu et al., 2009). Moreover, intermediates of the citrate cycle also serve as pre-substrates for flavor compounds, and the highly expressed enzymes related to this pathway also originated from the fungal community, some of which have been applied to the saccharification and fermentation process of foods and drugs, including A. oryzae, A. clavatus, and A. terreus.
Numerous aromatic compounds, such as tannin, ferulic acid, and lignin have been identified in the materials of cereals, the degradation of which is strongly related to liquor flavor generation (Liu and Sun, 2018). Several laccases, feruloyl esterase and ferulic acid decarboxylase were detected with low expression levels from A. clavatus, C. posadasii, P. marneffei, A. terreus or Pseudomonas aeruginosa in J3 (data not shown), which might clearly confirm the degradations of ferulic acid and lignin during high temperature stage of JF daqu. Similarly, many aromatic compounds and phenols were identified in both JF and NF liquors Xiao et al., 2016); consistent with this, in the present study some enzymes related to the degradation of aromatic compounds were also found to be expressed in JF and NF liquor starters, with most showing lower expression in the former. However, the remainder constituted those enzymes that were only detected (at low levels) in J3, indicating that trace aroma compounds were likely particularly associated with JF liquor flavor, such as ethyl benzeneacetate and benzaldehyde (Xiao et al., 2016). Highly expressed members of these enzymes were mostly derived from fungal species in J3, which appears consistent with the contributions of some fungi toward the degradation of aromatic compounds (Godoy et al., 2016;Sun et al., 2016;Vieira et al., 2018). Alternatively, enzymes expressed at low levels from bacteria may also substantively contribute to degradation of aromatic compounds (Pérez-Pantoja et al., 2015;Van der Waals et al., 2017). Therefore, both the fungal and bacterial communities appear to have an active role in degrading aromatic compounds in JF daqu (Bhattacharya et al., 2017;Wang et al., 2017;Kamyabi et al., 2018;Ma et al., 2018), especially in the high temperature and mature stage (Huang et al., 2017b).
In addition to the microbial community, temperature also makes large contributions to generate flavor compounds in JF daqu, such as pyrazines and their derivatives, which comprise pivotal impact aroma compounds of JF liquor (Zhu et al., 2007;Fan et al., 2012). In particular, their generation may be thermally induced from microbial metabolites by nonenzymatic browning via the Maillard reaction at 70 • C in J3 (Richards et al., 2011;Nashalian and Yaylayan, 2014). Overall, JF liquor flavor thus appears to be determined by a highly complicated process and further analysis of the active microbial community, enzymes, and metabolites from the daqu preparation in addition to stacking fermentation and alcoholic fermentation processes are required to unravel the mystery of JF liquor flavor generation.

CONCLUSION
In the present study, fungi including Aspergillus and Penicillium, were identified as the most active microbial community members at the high temperature stage (J3: 70 • C) of JF daqu by metatranscriptomics. Furthermore, the high temperature stage was found to not only lower the capacities of JF daqu toward saccharification and fermentation, but also enhance its ability in generating diverse minor flavor compounds, e.g., derivatives of aromatic compounds. Additionally, most of enzymes related to those capacities were highly expressed at 70 • C by fungal genus of Aspergillus, Coccidioides, Paracoccidioides, Penicillium, and Rasamsonia. These exploratory findings shed light on our understanding of the JF baijiu fermentation system, in which the high temperature stage plays key roles in improving JF daqu by providing unique active microbiota and enzymes, and strongly contributing to the final distinctive aroma and taste of JF baijiu.

AUTHOR CONTRIBUTIONS
HZ, ZY, YF, YJ, LT, and KH designed the experiment. ZY performed the experiments and analyzed the data. ZY, YX, DL, and HL collected samples and communicated with the liquor factory. ZY and LC wrote the main manuscript. ZY, AD, YF, and HZ revised the manuscript. All authors revised and approved the final version of the manuscript.