Characterization and discrimination of microbial community and co‐occurrence patterns in fresh and strong flavor style flue‐cured tobacco leaves

Abstract Fermentation, also known as aging, is vital for enhancing the quality of flue‐cured tobacco leaves (FTLs). Aged FTLs demonstrate high‐quality sensory characteristics, while unaged FTLs do not. Microbes play important roles in the FTL fermentation process. However, the eukaryotic microbial community diversity is poorly understood, as are microbial associations within FTLs. We aimed to characterize and compare the microbiota associated with two important categories, fresh and strong flavor style FTLs, and to reveal correlations between the microbial taxa within them. Based on 16S and 18S rRNA Illumina MiSeq sequencing, the community richness and diversity of prokaryotes were almost as high as that of eukaryotes. The dominant microbes of FTLs belonged to seven genera, including Pseudomonas, Bacillus, Methylobacterium, Acinetobacter, Sphingomonas, Neophaeosphaeria, and Cladosporium, of the Proteobacteria, Firmicutes, and Ascomycota phyla. According to partial least square discriminant analysis (PLS‐DA), Xanthomonas, Franconibacter, Massilia, Quadrisphaera, Staphylococcus, Cladosporium, Lodderomyces, Symmetrospora, Golovinomyces, and Dioszegia were significantly positively correlated with fresh flavor style FTLs, while Xenophilus, Fusarium, unclassified Ustilaginaceae, Tilletiopsis, Cryphonectria, Colletotrichum, and Cyanodermella were significantly positively correlated with strong flavor style FTLs. Network analysis identified seven hubs, Aureimonas, Kocuria, Massilia, Brachybacterium, Clostridium, Dietzia, and Vishniacozyma, that may play important roles in FTL ecosystem stability, which may be destroyed by Myrmecridium. FTL microbiota was found to be correlated with flavor style. Species present in lower numbers than the dominant microbes might be used as microbial markers to discriminate different flavor style samples and to stabilize FTL microbial communities. This research advances our understanding of FTL microbiota and describes a means of discriminating between fresh and strong flavor FTLs based on their respective stable microbiota.


| INTRODUC TI ON
Tobacco (Nicotiana tabacum L.) is one of the largest economic nonfood crops in the world. In China, the most important type of tobacco is the flue-cured tobacco (Su et al., 2011;Zhao et al., 2007). The flavor of flue-cured tobacco leaf (FTL) changes throughout the process of fermentation, gradually aging over long periods (typically at least 12 months). The aging process results in FTL of a high commercial quality and causes a change in color to a darker yellow, elimination of harmful odors, degradation of harmful substances, reduction of incentive odor, and development of tobacco-specific flavors (Yu & Gong, 2009). According to flavor styles, Chinese aged FTL could be traditionally divided into three categories: fresh flavor style, middle flavor style, and strong flavor style.
Microbes have been found to play important roles during the FTL fermentation process (Reid, McKinstry, & Haley, 1937), which include the production of tobacco-specific flavors (English, Bell, & Berger, 1967) and degradation of nicotine and tobacco-specific nitrosamines Liu, He, et al., 2015;Liu, Ma, et al., 2015). Studies based on traditional culture-dependent methods have isolated Bacillus, Streptomyces, Aspergillus, and Penicillium from FTLs, identifying these species as dominant microbes in FTL fermentations . To better characterize the microbial community associated with FTLs, prokaryotic diversity has been investigated using culture-independent molecular biology techniques such as polymerase chain reaction denaturing gradient gel electrophoresis (PCR-DGGE) (Huang et al., 2010;Zhao et al., 2007), 16S rRNA gene libraries (Su et al., 2011), Roche 454 bar-coded pyrosequencing (Liu, He, et al., 2015;Liu, Ma, et al., 2015), and Illumina MiSeq sequencing (Wang et al., 2018). However, little is known about the eukaryotic community structure of FTL based on culture-independent molecular biology techniques.
PLS-DA has been used in distinguishing different kinds of Chinese liquors (Zhang, Yuan, Zeng, et al., 2017) and pit muds (Zhang, Yuan, Liao, & Zhang, 2017). It may therefore be of use in distinguishing different FTL flavor styles and in identifying microbes, which contribute significantly to desirable tobacco characteristics.
The network interface, in the form of a set of nodes and edges, carries meaningful statistical and structural features that shed light on the underlying rules guiding the community components and functions of the system being described (Newman, 2006). Recently, network analysis has been widely applied to reveal ecological linkages among microorganisms in complex ecosystems, such as marine water (Steele et al., 2011), soil (Barberán, Bates, Casamayor, & Fierer, 2012), and pit mud (Hu, Du, Ren, & Xu, 2016). To our knowledge, the existence of direct or indirect interactions among microbial taxa coexisting in FTLs has not been reported. Identifying and describing such interactions could clarify the ecological rules guiding community assembly within the FTL ecosystem.

| FTLs sampling
Samples of FTL were collected from a tobacco warehouse located in Shifang city, Sichuan province of China. The FTLs labeled as fresh flavor style and strong flavor style by sensory assessors were marked accordingly (F1, F2, F3, S1, S2, and S3). FTLs from three wellknown planting origins located in China were randomly selected for each style, and triplicate subsamples were collected and placed into each tobacco leaf storage box. Tobacco leaves approximately 20 cm from the top of the tobacco leaf storage box was removed and discarded. In total, 2 kg of leaf samples were taken from the four corners and the center of the storage box using the five-point method. All samples were well-mixed, transferred into sterile bags, and stored at −20°C.

| DNA extraction and illumina MiSeq sequencing
Tobacco (25 g) was suspended in 500 ml of sterile saline and shaken for 2 hr at 200 rpm, after which the supernatant was centrifuged at 10,000 g for 20 min. Genomic DNA was extracted from the resulting pellet using an EZNA® Soil DNA Kit (Omega). The genomic DNA was sent to GENEWIZ Inc. for PCR amplification and sequencing of the V3-V4 hypervariable region of 16S rRNA genes (primers: 5′-CCT ACG GRR BGC ASC AGK VRV GAA/T3′ and 5′-GGA CTA CNV GGG TWT CTA ATC C-3′) and the V7-V8 hypervariable region of 18S rRNA genes (forward primers containing the sequence: 5′-CGW TAA CGA ACG AG-3′ and reverse primers containing 5′-AIC CAT TCA ATC GG-3′).

| Sequence processing and data analysis
The sequencing data were processed using the QIIME platform version 1.9.1 (http://qiime.org/). The forward and reverse reads were merged according to the unique sample barcode sequence, followed by quality control processing (for 16S rRNA gene lengths between 430 and 470 bp, average length 455 bp, and for 18S rRNA gene lengths between 340 and 380 bp, average length 355 bp), and then truncated by removing the barcode and primer sequences. Qualified sequences were classified into operational taxonomic units (OTUs) at a 97% sequence identity using the clustering program VSEARCH version 1.9.6 and the Silva 132 database (https ://www.arb-silva.de/) (Rognes, Flouri, Nichols, Quince, & Mahé, 2016). The phylogenetic affiliation of each sequence was analyzed by Ribosomal Database Program (RDP) Classifier at the 80% confidence level. Significant differences between the fresh flavor style and strong flavor style FTL groups were determined using SPSS software, version 19 (IBM), by one-way analysis of variance and Duncan's multiple comparison test (p < .05). Alpha diversity, including Chao1 and Shannon values, was analyzed using QIIME version 1.9.1 (Caporaso et al., 2010). To reduce potential confounding effects due to uneven sampling, we randomly rarefied the OTU table to an even depth for alpha diversity analysis. Pairwise Spearman's rank correlations among genera with relative abundances higher than 0.1% were performed using SPSS Statistics version 19 (IBM, America). Spearman's correlation coefficients with statistical significance (p < .01) were considered valid co-occurrence (or negative) events for a robust correlation (Barberán et al., 2012;Hu et al., 2016;Zhao et al., 2014). Networks were explored and visualized using the interactive platform Gephi (Bastian, Heymann, & Jacomy, 2009) based on the correlation matrix constructed by Spearman's correlations, with each node and edge representing one genus and a strong and significant correlation, respectively.

| Nucleotide sequence accession number
The MiSeq sequences determined in this study have been deposited in the GenBank under the following accession number: PRJNA498896 and release date: 2019-11-19.

| Prokaryotic Community Diversity and Structure
In total, 344,929 qualified reads were obtained from all FTLs. Each sample contained 102 to 122 OTUs, based on 97% similarity of 16S rRNA sequences (Table 1)

| Eukaryotic community diversity and structure
In total, 367,337 qualified reads were obtained from all FTLs (Table 1)  and Basidiomycota, which had average relative abundances of 58.16 ± 9.06% and 19.43 ± 6.53%, respectively. Ascomycota had higher relative abundances (p < .05) in fresh flavor style FTLs (F1, F2, and F3) than in strong flavor style FTLs (S1, S2, and S3). There were 53 eukaryotic genera across all samples, with 17 genera having a relative abundance of higher than 1.0% in at least one sample

| PLS-DA and HCA
PLS-DA was used to construct a statistical model for FTL discrimination and classification, and two significant principal components of the total variance in data matrix were extracted. The R2 and Q2 were 0.997 and 0.775, respectively, which meant that a total of 99.7% dummy Y variable per class, and 77.5% overall cross-validated  Table 2 HCA also identified two groups of samples (Figure 2b), the first consisting of samples F1, F2, and F3, and the second of samples S1, S2, and S3, which was consistent with the PLS-DA.

Nicotine degradation pathways in
Pseudomonas species can be classified into two categories, depending on whether metabolites are directed into the tricarboxylic acid cycle (TAC). Nicotine can be converted into N-methylmyosmine, cotinine, or nornicotine, which is then converted into maleamic acid, and finally fumaric acid in the TAC. Alternatively, nicotine may be converted into nicotyrine, which is not directed into the TAC (Li, Duan, Zhang, & Yang, 2010;Ruan et al., 2005;Tang et al., 2008Tang et al., , 2009Tang et al., , 2011Wang, Yang, Min, & Lv, 2009).
Bacillus species fulfill different functions in FTLs. Some species are considered endophytic and/or beneficial to plants, including tobacco, and have been reported to be the functional microorganism in the promotion of tobacco fermentation and formation of aged flavor compounds (English et al., 1967;Huang et al., 2010). English et al. (1967) revealed that B. subtilis and B. circulans could hasten the development of desirable flavors and improve the smoking qualities of cigar tobacco. B. thuringiensis, which produces bipyramidal crystals, is present on the tobacco leaf surface and can control insect pests that affect stored tobacco (Kaelin & Gadani, 2000). B. subtilis has a strong ability to control the effects of tobacco black shank (Han et al., 2016) and displayed an antagonistic effect against Verticillium dahliae, which causes verticillium wilt (Berg & Ballin, 1994).

Similar to B. subtilis, Sphingomonas has an antagonistic effect on
Verticillium dahliae (Berg & Ballin, 1994), but has the additional ability to degrade a wide variety of dimeric lignin compounds into a series and CoeffCS[2] represent significant principal component 1 and 2, respectively. DA(1) denotes the fresh flavor style FTL group (F1 to F3); DA(2) denotes strong flavor style FTL group (S1 to S3). ‡, * Coefficient is significant (above the noise).
Sphingomonas abundance was 4% in the 16S rRNA clone library of Zimbabwe tobacco (Su et al., 2011). Acinetobacter and Sphingomonas isolated from soil tobacco waste are able to degrade nicotine . Methylobacterium strains, which are frequently encountered as endophytes, degrade one-carbon compounds such as methanol and methylamine, and are capable of forming biofilms, producing quorum-sensing signals, and resisting heavy metal and other stresses (Ardanov, Sessitsch, Häggman, Kozyrovska, & Pirttila, 2012;Rossetto et al., 2011).
Neophaeosphaeria and Cladosporium, from the Ascomycota, were dominant fungal genera in FTLs, although Lodderomyces, Candida, unclassified Ustilaginaceae, and Tilletiopsis were also present at high relative abundances (>10%) in some samples. However, the specific functions of these genera during tobacco leaf fermentation and flavor formation are not well-understood. Cladosporium can produce γ-decalactone (Berger, 2015). It can also contribute to lignin and cellulose-degradation, as it produces laccase and cellulase , and may consequently play a flavor-enhancing role during FTL fermentation. Tilletiopsis produce hydrolytic enzymes and antifungal compounds, which are effective against powdery mildew fungi (Urquhart & Punja, 2002); this genus may therefore serve as a biocontrol agent. FTLs. However, the boundary separating fresh and strong flavor style FTLs was not obvious and did not allow for straightforward discrimination or classification. Therefore, in order to distinguish between fresh and strong flavor style FTLs, the microbial data had to be analyzed using metrology tools. Analysis of the microbial relative abundances by PLS-DA and HCA indicated that samples could be separated into fresh and strong flavor groups and that some genera could be used as markers for the discrimination of samples. The results of our study agree with those of the sensory evaluation methods traditionally used to classify FTLs.
Although the number of samples (n = 3) was statistically significant and results of previous studies (Huang et al., 2010;Su et al., 2011;Wang et al., 2018) were corroborated, the sample size is still small in this study. More samples of fresh and strong flavor style FTLs from different batches and sources should be considered to verify the reliability and reproducibility. Illumina MiSeq sequencing based on 16S rRNA and 18S rRNA genes provided the diversity and structure of the prokaryotic and eukaryotic communities associated with FTLs; however, amplicon-based studies could not identify viable organisms. Thus, omics technologies, including genomics, transcriptomics, proteomics, metabolomics, and fluxomics, should be utilized to intensively study genetic, protein, and product information related to the microbial metabolism, and the mechanisms producing patterns of community coexistence, as the complex structure and community of microbes in FTLs is complex, and several correlations of the microbial taxa. Moreover, some specific species, including endophytes, plant-growth-promoting rhizobacteria, and other abundance microbes isolated from FTLs, should be studied for their function in FTLs, as they might play essential roles in stabilizing inhabiting microbial community and producing beneficial substances for FTLs.

ACK N OWLED G EM ENTS
This work was funded by Sichuan Science and Technology Program (No. 2019YJ0264), and China Tobacco Sichuan Industrial Co., Ltd.

CO N FLI C T O F I NTE R E S T
None declared.

E TH I C S S TATEM ENT
None required.

DATA AVA I L A B I L I T Y S TAT E M E N T
The datasets generated for this study can be found in GenBank under the accession number PRJNA498896: https ://www.ncbi.nlm.