Multi‐omics dissection of phyllospheric microbial succession and volatile flavor compound formation in cigar tobacco during fermentation

Abstract

BACKGROUND

Fermentation plays a critical role in determining the quality and aroma of cigar tobacco; however, the interactions between phyllospheric microbial succession and volatile flavor compound (VFC) formation remain insufficiently understood. This study aimed to elucidate the associations between microbial community dynamics and flavor development during tobacco fermentation using a multi‐omics approach.

RESULTS

Distinct stage‐dependent shifts in microbial communities were observed across unfermented (F1), mid‐fermentation (F2), and final‐fermentation (F3) stages. Bacterial diversity showed a ‘first‐decrease‐then‐increase’ trend, with Staphylococcus dominating mid‐fermentation (relative abundance > 97%) and keystone genera (Methylobacterium‐Methylorubrum, Aerococcus) enriching at the final stage. Fungal communities were persistently dominated by Aspergillus throughout fermentation. Volatile metabolomics identified 47 differentially abundant flavor compounds, with aldehydes (e.g. benzaldehyde, nonanal) and alcohols (e.g. 1‐pentanol) significantly enriched post‐fermentation. Correlation analyses revealed stronger associations between bacterial communities and VFCs compared to fungi, with Aerococcus showing significant positive correlations with key upregulated compounds (P < 0.05). Functional prediction suggested that carbohydrate, amino acid, and lipid metabolism pathways may contribute to VFC formation.

CONCLUSION

This study demonstrates that microbial succession is closely associated with chemical transformation and flavor development during cigar tobacco fermentation. Bacterial taxa, particularly Aerococcus, may play important roles in shaping volatile profiles. These findings provide a theoretical basis for future microbial regulation strategies aimed at improving fermentation quality. © 2026 Society of Chemical Industry.

INTRODUCTION

Fermentation constitutes a pivotal bioprocessing methodology extensively employed in the industrial production of plant‐derived commodities, including leaf‐based materials (tobacco, tea, Allium spp.), leguminous derivatives, and fruit matrices.^{1, 2, 3} In leaf processing systems, fermentation orchestrates an intricate cascade of catabolic and anabolic reactions – encompassing polysaccharide depolymerization, chlorogenic acid deconjugation, proteolysis, nonenzymatic glycosylation, Strecker‐type amino acid decarboxylation, and sugar caramelization. ⁴ These synergistic pathways generate signature volatile organic compounds that define product‐specific organoleptic profiles. Empirical studies demonstrate that during black tea fermentation, lipolytic cleavage of fatty acids, proteolytic liberation of amino acids, oxidative fragmentation of carotenoids, and glycosidase‐mediated release of bound volatiles collectively yield volatile organic compound ensembles associated with sweet, floral, and fruity sensory attributes. ⁵ Parallel mechanisms operate in tobacco curing, where amylolytic degradation of starch reserves, cellulolytic breakdown of structural polysaccharides, and pectin demethoxylation generate aromatic precursors that critically determine the molecular complexity of pyrolytic aroma profiles in cured leaves. ⁶

Phyllosphere microorganisms serve as biocatalytic mediators in these transformations, enzymatically converting macromolecular substrates (starch, proteins, lipids) into low‐molecular‐weight aroma compounds such as aldehydes, ketones, and organic acids. ⁷ For instance, Bacillus subtilis and Bacillus marinus enhance polysaccharide and protein degradation to solanone, β‐damascone, and neophytadiene, significantly improving overall aroma quality. ⁸ Acinetobacter, exhibiting metabolic synergy with Bacillus, demonstrate strong correlations with solanone and megastigmatrienone production. ⁹ Taxa including Sphingobacterium and Enterobacter elevate carotenoid‐derived aroma compounds, thereby enriching the olfactory complexity of fermented products. ¹⁰

Emerging research demonstrates that phyllospheric microbial metabolic networks and volatile flavor formation mechanisms exhibit critical dependence on dynamic substrate availability and coordinated microenvironmental regulation.^{11, 12} High‐sensitivity volatile metabolomics platforms enable precise delineation of spatiotemporal dynamics in low‐boiling‐point aroma compounds (e.g. benzaldehyde, megastigmatrienone, β‐ionone) during fermentation through targeted capture of aldehydes, ketones, and terpenoids. ¹³ In cigar leaf fermentation, volatile metabolomics profiling reveals Bacillus spp.‐mediated proteolytic liberation of phenylalanine during mid‐fermentation (days 14–21), with subsequent Strecker degradation yielding benzaldehyde – a process exhibiting linear correlation with microbial abundance.^{14, 15} Concurrently, actinobacterial taxa (e.g. Corynebacterium) upregulate carotenoid cleavage pathways, achieving 2.3‐fold β‐ionone enrichment via asymmetric oxidative scission. ¹⁶ This microbe–substrate–aroma precursor cascade demonstrates phylogenetic conservation across fermented taxa, as evidenced by Saccharomyces cerevisiae–Rhodotorula mucilaginosa co‐cultures in cider fermentation, which elevate ethyl lactate and 1‐butanol titers while accelerating glucose consumption kinetics relative to axenic controls. ¹⁷

While recent studies have elucidated associations between core microbiota (e.g. Bacillus, Actinobacteria) and specific aroma compounds (e.g. benzaldehyde, β‐ionone), the causal linkages between microbial successional dynamics and volatile compound biosynthesis during fermentation remain systematically unresolved. The study reported here employed integrated multi‐omics network analysis combining metagenomics and volatile metabolomics to identify keystone microbial biomarkers and establish microbe–metabolite correlation frameworks. Compared to single‐omics approaches, this strategy enables synchronous tracking of dynamic chemical shifts in tobacco leaves and their regulatory impacts on microbial metabolic pathways, thereby facilitating causal inference from community structure to functional phenotypes. We further aimed to delineate the cascading interactions among microbial consortia, substrate turnover, and aroma precursor flux, advancing mechanistic understanding of self‐organizing principles in fermentation ecosystems and identifying actionable targets for precision modulation of tobacco fermentation protocols.

MATERIALS AND METHODS

Sample preparation

Cigar tobacco leaf samples were obtained from the variety ‘QX 107’ grown in Mengyin County, Shandong Province. After drying, three distinct sample sets were collected: unfermented leaves (F1), leaves at the midpoint of fermentation (F2), and leaves at the conclusion of fermentation (F3). At each sampling stage, three leaves were taken from the upper, middle, and lower layers, respectively. From these, six leaves exhibiting uniform appearance and quality were carefully selected. Approximately 30 g of plant leaf material from the designated portions was excised using sterile scissors and placed into centrifuge tubes. These tubes were appropriately labeled and stored in a refrigerator maintained at −20 °C.

Determination of major chemical components

Cured tobacco leaf samples were collected and prepared according to YC/T 31‐1996. After removing impurities and separating stems from leaves, the samples were dried at 45 °C to constant weight, ground into fine powder using a cyclone mill, and sieved through a 100‐mesh (0.15 mm) screen. The contents of total sugar (TS), reducing sugar (RS), nicotine (NIC), chloride (NL), total nitrogen (TN) and potassium (K) were determined using a Flowsys III continuous flow analyzer (SYSTEA S.p.A., Anagni, Italy). The analytical procedures followed the Chinese tobacco industry standards: YC/T 159‐2019 for TS and RS, YC/T 468‐2013 for NIC, YC/T 161‐2002 for TN, YC/T 217‐2007 for K, and YC/T 162‐2011 for NL. All results were expressed as mass percentage on a dry weight basis (% w/w). Three independent biological replicates were collected for each treatment to ensure analytical reproducibility.

Sample DNA extraction and high‐throughput sequencing

Sample DNA extraction

An amount of 5 g of leaf samples was weighed into a sterilized conical flask. Three replicates were performed for each sample. Subsequently, 250 mL of 10 g L⁻¹ phosphate‐buffered saline buffer solution was added until the samples are completely submerged in the liquid. The conical flask was sealed and placed in a constant‐temperature shaker maintained at 14 °C for a duration of 150 min, with the rotational speed set at 220 rpm. Sterile gauze was used to filter the mixture in order to obtain the extract. The extract was transferred into a centrifuge, adjusting the ambient temperature to 4 °C and setting the rotational speed to 12 000 × g. The extract was centrifuged at 4 °C and 12 000 × g for 30 min. To further purify, the lower precipitate portion was centrifuged two additional times. The resulting crude precipitate was then stored in a refrigerator at −80 °C for subsequent experimental procedures.

DNA extraction and quality assessment

After DNA extraction, a two‐step quality assessment was performed to ensure the suitability of genomic DNA (gDNA) for downstream applications. First, the structural integrity of the extracted DNA was visually inspected by loading aliquots onto a 10 g L⁻¹ agarose gel and performing electrophoresis followed by visualization under UV light. Second, the concentration and purity of gDNA were determined using a NanoDrop 2000 spectrophotometer (Thermo Fisher Scientific, USA). Purity was assessed based on the A ₂₆₀/A ₂₈₀ and A ₂₆₀/A ₂₃₀ absorbance ratios.

PCR amplification, library preparation, and sequencing

The V3–V4 hypervariable region of the bacterial 16S rRNA gene was selected for amplification using the primer pair 338F (5′‐ACTCCTACGGGAGGCAGCAG‐3′) and 806R (5′‐GGACTACHVGGGTWTCTAAT‐3′), as this region provides sufficient sequence polymorphism for taxonomic classification of complex bacterial communities. This dual‐zone approach enhances the accuracy of taxonomic and species‐level classification, ¹⁸ owing to its excellent compatibility with the Illumina MiSeq platform, high‐resolution dual‐zone sequencing, and mainstream databases such as SILVA. Each 20 μL PCR reaction comprised 4 μL of 5× TransStart FastPfu buffer, 2 μL of 2.5 mmol L⁻¹ dNTPs, 0.8 μL each of forward and reverse primers (5 μmol L⁻¹), 0.4 μL of TransStart FastPfu DNA polymerase, and 10 ng of template DNA. The thermal cycling conditions were as follows: the initial denaturation step was conducted at 95 °C for 3 min, followed by 27 cycles of denaturation at 95 °C for 30 s, at 55 °C for 30 s, and at 72 °C for 30 s. This was followed by a final extension at 72 °C for 10 min. The purification of amplified products was conducted using a PCR Clean‐Up Kit (YuHua, China), following separation on a 20 g L⁻¹ agarose gel. Quantification of the samples was performed using a Qubit 4.0 fluorometer (Thermo Fisher Scientific, USA). The construction of sequencing libraries was undertaken using a NEXTFLEX Rapid DNA‐Seq Kit (Bioo Scientific, USA), a process which encompassed adapter ligation, magnetic bead‐based size selection, PCR enrichment, and final library purification. Paired‐end sequencing (PE250/PE300) was performed on an Illumina platform by Shanghai Majorbio Bio‐pharm Technology Co. Ltd.

Bioinformatics analysis

The raw reads were then filtered using fastp (v0.19.6) to remove low‐quality bases (Phred score < 20), reads shorter than 50 base pairs, and those containing ambiguous nucleotides. Overlapping paired‐end reads were then merged with FLASH (v1.2.11) using a minimum overlap of 10 base pairs and a maximum mismatch ratio of 0.1. Chimeric sequences were identified and removed via reference‐based filtering against the SILVA database (v138) using UPARSE (v7.1). Operational taxonomic units (OTUs) were then subjected to clustering, with the 97% similarity threshold applied to define clusters. Furthermore, sequences originating from chloroplasts or mitochondria were excluded from further analysis. It was essential to standardize the sequencing depth, so all samples were rarefied to 20 000 reads. This achieved a Good's coverage of 99.09%. Taxonomic annotation was performed using the RDP Classifier (v2.11) against the SILVA database, with a 70% confidence threshold. We predicted functional profiles using the Phylogenetic Investigation of Communities by Reconstruction of Unobserved States 2 (PICRUSt2, V2.2.0) method, which is based on KEGG pathways.

Volatile metabolomic analysis

The metabolomic analysis of the samples was conducted using an Agilent 7697A‐8890‐7000D headspace gas chromatography–mass spectrometry (HS‐GC–MS) system. For the identification and quantification of volatile organic compounds, 20 μL of 2‐octanol (10 mg L⁻¹) was added to each sample as an internal standard prior to HS‐GC–MS analysis. The headspace conditions were as follows: the heating box temperature was set at 130 °C, the quantification loop temperature at 150 °C, the transfer line temperature at 170 °C, the sample vial equilibration time at 20 min, the injection duration at 0.5 min, the GC cycle time at 35 min, the sample vial volume at 20 mL, and the final pressure of the quantification loop at 10 psi. The samples were analyzed using a GC–MS system in split mode, with an injection volume of 1 μL and a shunt ratio of 10:1. The sample was separated on a VF‐WAXms capillary column (25 m × 0.25 mm × 0.2 μm, Agilent CP9204) at a flow rate of 1 μL, with a split ratio of 10:1, and detected by MS. The inlet temperature was set at 180 °C, and the carrier gas was high‐purity helium with a carrier gas flow rate of 2 mL min⁻¹ and a spacer purge flow rate of 3 mL min⁻¹. The heating procedure was initiated at an initial temperature of 40 °C, followed by a 2 min equilibration period. Thereafter, the temperature was increased to 100 °C at a rate of 5 °C min⁻¹, subsequently reaching 230 °C at a rate of 15 °C min⁻¹. This temperature was maintained for a duration of 5 min. Thereafter, the temperature was set to 230 °C and maintained for 2 min. MS conditions: the electron bombardment ion source (EI) operated at a transmission line temperature of 310 °C, an ion source temperature of 230 °C, a quadrupole temperature of 150 °C, and an electron energy of 70 eV. The mass spectrometer operated in full‐scan mode (SCAN), with a mass scanning range of m/z 30–1000 and a scanning frequency of 3.2 scans per second.

Data processing

The bioinformatic analysis of the phyllosphere microbiota of cigar leaves was conducted using the Majorbio Cloud platform (https://cloud.majorbio.com). In accordance with the amplicon sequence variants (ASVs) provided, rarefaction curves and alpha diversity indices, incorporating observed ASVs, Chao1 richness, Shannon index, and Pielou index, were calculated with Mothur v1.30.1. The similarity among the microbial communities in different samples was determined by principal coordinate analysis based on Bray–Curtis dissimilarity using the Vegan v2.5‐3 package. The PERMANOVA test was employed to evaluate the proportion of variation attributable to the treatment, in conjunction with its statistical significance, utilizing the Vegan v2.5‐3 package. Linear discriminant analysis effect size (LEfSe) analysis (http://huttenhower.sph.harvard.edu/LEfSe) was performed to identify significantly abundant taxa (from phylum to genera level) of bacteria among the different groups, thereby screening for characteristic dominant microorganisms (CDM), with a LDA score threshold >3 and P < 0.05. The construction of co‐occurrence networks was undertaken to facilitate the exploration of internal community relationships across the samples. The existence of a statistically robust correlation between two nodes was indicated by a Spearman's correlation coefficient that lay between −0.6 and 0.6 (absolute value ≥ 0.6), with P < 0.01.

For the volatile metabolomics data, metabolites were identified by searching against public databases such as NIST (version 2017), Fiehn (version 2013), and MS‐DIAL (version 2021). A match factor threshold of ≥80 was used for NIST library identification. An internal standard was used during sample processing to monitor instrument stability. The data matrix was pre‐processed by retaining metabolic features detected in at least 80% of samples, estimating minimum values for features below the quantification limit, and normalizing by sum. Variables with a relative standard deviation > 30% in quality control samples were excluded. Metabolites demonstrating variable importance in projection (VIP) scores exceeding 1, adjusted P below 0.05, and absolute fold‐change values greater than 1 were classified as differentially accumulated metabolites. Differential metabolites among two groups were mapped into their biochemical pathways through metabolic enrichment and pathway analysis based on the KEGG database (http://www.genome.jp/kegg/). Python packages ‘scipy.stats’ (https://docs.scipy.org/doc/scipy/) was used to perform enrichment analysis to obtain the most relevant biological pathways for experimental treatments.

RESULTS AND DISCUSSION

Primary chemical composition dynamics during fermentation

Quantitative analysis of primary chemical components in cigar leaves across fermentation stages revealed dynamic changes in TS, RS, NIC, NL, TN, and K contents (Fig. 1). During fermentation, TS (Δ = 75.53%), RS (Δ = 94.29%), TN (Δ = 16.50%), and NL (Δ = 55.62%) showed significant decreases, with NIC exhibiting a marked reduction by the final stage, while K levels remained relatively stable (Δ = 0.30%) – trends consistent with previous studies.^{2, 19, 20} This significant decline in chemical components is not merely a passive response to microbial nutrient consumption, but rather an active process involving the formation of flavor precursors.

Figure 1.

Contents of total sugars (TS), reducing sugars (RS), nicotine (NIC), chloride (NL), total nitrogen (TN), potassium (K), and the ratios of TS/NIC, TN/NIC, K/NL at pre‐fermentation, mid‐fermentation and post‐fermentation stages. Statistical analysis was performed using one‐way ANOVA followed by Tukey's post hoc test (P < 0.05). Different lowercase letters indicate significant differences among treatments.

The observed progressive decline in TS and RS aligns with microbial‐mediated polysaccharide hydrolysis. For instance, Aspergillus fumigatus has been reported to produce highly active α‐amylase and glucoamylase, ²¹ and Bacillus species are known to secrete amylases and proteases. ²² These enzymes degrade macromolecular carbohydrates into reducing sugars, which serve as essential precursors that enter the Maillard reaction, combine with free amino acids, and ultimately generate Strecker aldehydes such as benzaldehyde and nonanal.^{23, 24} Notably, similar metabolic behaviors have also been observed in other fermented plant systems. During dark tea fermentation, Bacillus species (e.g. B. oleronius) significantly increase soluble sugar content and promote interactions between sugars and amino acids, ²⁵ while solid‐state fermentation by Eurotium cristatum demonstrates a tight metabolic coupling between carbohydrate consumption and terpenoid aroma compound accumulation. ²⁶ These parallels support the notion that microbial‐driven carbohydrate turnover is a conserved mechanism across plant fermentations, consistent with our observations in cigar tobacco, where sugar depletion correlates with the production of key volatile compounds.

In addition, the significant decline in total nitrogen indicates active proteolysis, whereby microbial proteases degrade proteins into free amino acids. Free amino acids are important precursors of volatile flavor compounds (VFCs) in tobacco and are primarily transformed via two key metabolic pathways: first, the Ehrlich pathway, in which amino acids undergo transamination to form α‐keto acids, followed by decarboxylation to generate aldehydes, which are ultimately reduced to form alcohols ²⁷ ; second, the Strecker degradation pathway, in which amino acids, mediated by α‐dicarbonyl compounds generated from the Maillard reaction, undergo deamination and decarboxylation to produce characteristic aldehydes with one fewer carbon atoms. ²³ This microbe‐driven nitrogen metabolism is further corroborated by analogous processes in dark tea fermentation, where a sharp decrease in total free amino acids directly correlates with the accumulation of aroma volatiles like phenylethyl alcohol and methyl salicylate by Debaryomyces hansenii. ²⁷

The decrease in NIC could be attributed to microbial degradation by Pseudomonas (via pyrrolidine pathways) and Sphingomonas (via VVP pathways). ²⁸ The reduction in NIC at the later fermentation stage helps alleviate the irritation of tobacco smoke. Nitrogen compounds likely serve as microbial nutrient sources, with released volatile NH₃ potentially contributing to pyridine derivative formation. NL loss may correlate with moisture activity gradients during pile‐turning operations, a process hypothesized to reduce leaf hygroscopicity and mitigate combustion irregularities common in under‐fermented tobacco. Potassium content did not change significantly during the three fermentation stages, suggesting that fermentation had a minor impact on leaf K levels. In the late fermentation stage, the sugar‐to‐nicotine ratio decreased and the potassium‐to‐chlorine ratio increased, reflecting a general trend of flavor precursor depletion. This establishes a basis for further processing and enhanced sensory quality of tobacco leaves.

Microorganisms in fermentation

Characteristics of microbial communities

16S rRNA sequencing revealed stage‐specific shifts in microbial diversity and composition, with alpha diversity assessed via Ace index (species richness), Shannon index (diversity), and Pielou index (evenness). Bacterial communities exhibited marked mid‐fermentation declines in all three indices, followed by recovery to peak values exceeding pre‐fermentation levels (Fig. 2(A)). This pattern suggests phase‐dependent community restructuring, characterized by reduced richness and diversity during mid‐fermentation (dominance phase) and subsequent stabilization through late fermentation (equilibrium phase). Notably, Staphylococcus dominated mid‐fermentation bacterial consortia (>99% relative abundance), declining significantly by the final stage (Fig. 2(F)). This transient dominance likely reflects Staphylococcus's enhanced adaptability to low pH and elevated temperatures during active fermentation,^{29, 30} coupled with its antimicrobial activity suppressing competitor growth. ³¹ Fungal communities demonstrated contrasting dynamics, with Shannon and Ace indices initially increasing, peaking at mid‐fermentation, then declining – yet remaining significantly higher than pre‐fermentation levels (Fig. 2(C)). This trajectory may reflect sequential niche colonization: early environmental shifts promote fungal proliferation, while late‐stage physicochemical stabilization favors dominant taxa establishment.

Figure 2.

Alpha diversity of microbial communities: (A) bacteria, (C) fungi; and beta diversity: (B) bacteria, (D) fungi. Composition of microbial communities during fermentation at the phylum level: (E) bacteria, (G) fungi; and at the genus level: (F) bacteria, (H) fungi.

Principal coordinate analysis revealed distinct clustering of bacterial and fungal communities across fermentation stages (pre‐, mid‐, last stage). Bacterial communities showed partial overlap between early and mid‐phase samples, whereas fungal assemblages exhibited greater similarity between initial and final stages (Fig. 2(B),(D)). These patterns demonstrate significant structural shifts aligned with fermentation progression in bacteria (analysis of similarities, ANOSIM: R = 0.6790, P < 0.001) and fungi (R = 0.5638, P < 0.001), which is consistent with documented microbial successional patterns in leaf fermentation ecosystems. ³²

Structural composition of microbial communities in fermentation

Comparative analysis of the dominant microbial taxa identified the top 10 phylum‐level and top 15 genus‐level organisms across fermentation stages. Bacterial communities displayed substantial structural divergence while maintaining conserved core taxa (Fig. 2). The phylum Firmicutes dominated initial and mid‐fermentation phases, reaching 97.53% relative abundance at mid‐stage. Proteobacteria prevailed during late fermentation at 48.79%, whereas Actinobacteriota remained minor throughout with less than 10.20% abundance (Fig. 2(E)). At the genus level, the predominant bacterial taxa throughout the fermentation process included Staphylococcus, Methyobacterium‐Methylorubrum, Pseudomonas, Sphingomonas, and Aerococcus, consistent with findings from prior studies.^{33, 34} Staphylococcus maintained dominant prevalence across all stages, reaching peak activity (97.51%) during mid‐fermentation, followed by a significant decline in the late phase. With the exception of Pseudomonas and Sphingomonas, which decreased in abundance during late fermentation, all other genera exhibited progressive enrichment (Fig. 2(F)). Fungal communities were primarily dominated by the phyla Ascomycota and Basidiomycota (Fig. 2(G)). Mirroring bacterial dynamics, the genus Aspergillus displayed gradual abundance reduction during fermentation yet retained absolute dominance throughout the process (Fig. 2(H)), aligning with previously reported patterns.^{35, 36} Additional dominant fungal genera, including Wallemia, Sampaiozyma, Alternaria, Rhodotorula, and Plectosphaerella, demonstrated consistent enrichment trends across fermentation stages.

Venn diagram analysis delineated phase‐specific microbial consortium dynamics, revealing 76 persistent bacterial OTUs across all fermentation stages (Fig. 3(A)). F2–F3 exhibited the highest proportion of unique bacterial OTUs (41.10%), while F1 contained the lowest (5.15%). Fungal communities comprised 41 conserved genera, with F1 harboring the most unique fungal OTUs (37.38%), contrasting starkly with F2, which displayed no unique taxa (Fig. 3(B)). These patterns align with alpha diversity fluctuations: bacterial communities underwent mid‐phase homogenization (extreme dominance of select taxa) followed by late‐phase diversification, whereas fungal assemblages displayed progressive richness expansion.

Figure 3.

Venn diagrams of bacterial (A) and fungal (B) OTUs. LEfSe analysis identifying CDM at the genus level: (C) bacteria, (D) fungi. Abundances of bacterial (E) and fungal (F) CDM across different samples.

LEfSe analysis identified Methylobacterium‐Methylorubrum, Allorhizobium‐Neorhizobium‐Pararhizobium‐Rhizobium, Brachybacterium, and Aerococcus as characteristic dominant bacteria (CDB) in late fermentation (Fig. 3(C)). Early‐phase communities were typified by Paracoccus, while Staphylococcus monopolized the mid‐fermentation phase with relative abundance exceeding 97%, suppressing other CDB emergence. Fungal signatures diverged markedly: no statistically significant characteristic dominant fungi (CDF) emerged in early stages, whereas late fermentation hosted Alternaria, Coprinellus, Schizophyllum, and Albifimbria. Rhodotorula served as the sole mid‐phase CDF (Fig. 3(D)). This community succession pattern, shifting from mono‐dominance to multi‐species coexistence, is crucial for the development of flavor complexity. The metabolic activities of Staphylococcus during the mid‐fermentation stage establish a foundational pool of flavor metabolites dominated by organic acids and simple alcohols. Subsequently, the various CDB enriched in the final fermentation stage perform refined modification and transformation of these precursors using their distinct enzymatic systems. Notably, we discovered a unique bacterial genus, Methylobacterium‐Methylorubrum, which demonstrated dual industrial relevance. This methylotrophic genus synthesizes β‐ionone and megastigmatrienone through carotenoid cleavage oxygenases, ³⁷ directly enhancing tobacco aroma profiles. Concurrently, its production of novel l‐methionine analogs exhibits dose‐dependent tumor suppression via p53 pathway activation, suggesting biomedical potential. ³⁸ Aerococcus, through its amino acid aminotransferases and short‐chain fatty acid synthases, further converts the free amino acids accumulated during the mid‐fermentation stage into Strecker aldehydes such as benzaldehyde and nonanal, as well as sulfur‐containing aroma compounds such as 2‐mercaptopropanoic acid, ⁶ thereby shaping a richly layered final aroma profile.

Microbial interaction networks during fermentation

Microbial interactions also significantly influence the fermentation process. Figure 4 depicts the topological networks of bacteria and fungi during fermentation. Staphylococcus showed negative correlations with 23 nodes, including dominant genera Methylobacterium‐Methylorubrum, Allorhizobium‐Neorhizobium‐Pararhizobium‐Rhizobium, and Aerococcus. This antagonistic effect with other bacterial genera is consistent with previous reports. ³⁹ Despite serving as an antagonistic hub, Staphylococcus positively impacts fermentation by inhibiting pathogenic microbes. ⁴⁰ Co‐culture with salt‐tolerant yeasts and Pediococcus pentosaceus is reported to enhance flavor diversity in fermented products.^{41, 42} In contrast, Pseudomonas acted as a core hub in the synergistic network with a connectivity of 30, being a widespread bacterium in fermented products that promotes fermentation. ⁴³ In the fungal network, Aspergillus exhibited negative correlations with 17 nodes as the antagonistic hub, while Fusarium formed positive correlations with 27 nodes as the synergistic hub, consistent with its high co‐occurrence with other fungi reported previously. ⁴⁴

Figure 4.

Co‐occurrence networks of bacterial (A) and fungal (B) genera.

Dynamic profiles of VFCs in fermentation

GC–MS was employed to characterize VFCs in cigar tobacco leaves at the unfermented (UF) and post‐fermented (AF) stages. The major VFC classes comprised alcohols, esters, acids, aldehydes, and heterocyclic compounds, whereas phenols, amides, hydrocarbons, amines, ketones, and organic nitrogen compounds were detected at relatively low levels (Fig. 5(A)). Principal component analysis revealed significant separation of UF and AF samples along the first (PC1, R ²X = 77.9%) and second (PC2, R ²X = 17.6%) principal components (Fig. 5(B)), indicating fermentation‐induced remodeling of aroma profiles. Volcano plot analysis identified 47 differentially abundant VFCs (Fig. 5(C)). Among the differentially expressed compounds, five compounds were significantly upregulated during the late fermentation stage, including two aldehydes (benzaldehyde, nonanal), two alcohols (1‐pentanol, propargyl alcohol), and one acid (2‐mercaptopropionic acid). Heatmap clustering of differential VFCs further validated stage‐specific accumulation patterns, demonstrating fermentation‐driven enrichment of alcohols and aldehydes (Fig. 5(D)).

Figure 5.

Analysis of VFCs in plant leaves before and after fermentation: (A) major categories and abundances of VFCs; (B) principal component analysis results of VFCs before and after fermentation; (C) volcano plot showing the number of significantly upregulated and downregulated VFCs before and after fermentation; (D) variation trends of significantly differential VFCs during fermentation and their VIP values.

Alcohols serve as key aroma contributors in plant‐derived products (e.g. black tea, dark tea), imparting herbal, alcoholic, sweet, and woody notes.^{45, 46} This study identified 1‐pentanol and propargyl alcohol as fermentation‐induced alcohols. The formation of 1‐pentanol is closely related to the activity of microbial esterases. For example, genera such as Bacillus and Aeromonas secrete esterases that catalyze the hydrolysis of esters (e.g. pentyl acetate), releasing 1‐pentanol. ⁴⁷ This compound can enhance the aroma produced during fermentation in plant‐derived products. ⁴⁸ Propargyl alcohol – commonly used in chemical engineering and known for its antibacterial properties ⁴⁹ – represents a novel aroma contributor in fermentation systems, and its microbial biosynthetic pathway requires further investigation. In addition, the significant accumulation of 2‐mercaptopropionic acid is also noteworthy. This organic acid, characterized by a roasted aroma, is widely used as a food additive ⁵⁰ ; its formation may be associated with the microbial degradation of sulfur‐containing amino acids (e.g. cysteine and methionine). The genus Pseudobacillus is known to possess the ability to metabolize short‐chain fatty acids and may be directly involved in this conversion process. ⁶

Strecker aldehydes, formed via oxidative deamination and decarboxylation of α‐dicarbonyl compounds during Maillard reactions with amino acids, ²³ are critical flavor volatiles in processed foods. ⁵¹ The marked enrichment of benzaldehyde and nonanal after fermentation serves as direct evidence of active microbial‐mediated carbon and nitrogen metabolism. Nonanal, characterized by typical fatty and citrus notes, is a major aroma component in fermented tea ⁵² and also contributes to the fatty aroma of jujube fruit. ⁵³ The biosynthesis of nonanal proceeds primarily through the lipoxygenase (LOX) pathway: microbial lipases first release unsaturated fatty acids (e.g. oleic acid and linoleic acid) from membrane lipids; subsequently, LOX catalyzes their oxidation to hydroperoxides, which are then cleaved by hydroperoxide lyase (HPL) to form C9 aldehydes.^{5, 24} Benzaldehyde, which imparts bread‐like, almond, and cooked meat aroma characteristics, is generated mainly via the phenylalanine ammonia‐lyase (PAL) pathway: microbial proteases degrade proteins to release phenylalanine, PAL deaminates it to form trans‐cinnamic acid, and oxidative cleavage of this intermediate ultimately yields benzaldehyde.^{14, 15}

Associations of CDM with tobacco chemical components and differentially abundant VFCs during fermentation

The present study elucidated the tripartite correlation network among microbial communities, chemical components, and VFCs through Mantel test analysis (Fig. 6). The results revealed robust negative correlations among TN, TS, RS, and NL contents, indicating coupled carbon–nitrogen metabolism during fermentation. Bacterial communities showed an extremely significant negative association with NIC (P < 0.001) and significant correlations with TS/NL (P < 0.05), while fungi exhibited strong links to TN (P < 0.001) and moderate associations with TS/RS/NL (P < 0.01). These findings suggest that bacteria and fungi play distinct roles in NIC degradation and protein hydrolysis, respectively.

Figure 6.

(A) Mantel test heatmap of correlations between bacterial/fungal CDM and cigar tobacco leaf chemical components during fermentation. (B) Spearman correlation analysis (P < 0.05) between key CDM (relative abundance > 1%) and chemical components. (C) Mantel test heatmap of correlations between bacterial/fungal CDM and significantly differential VFCs. (D) Spearman correlation analysis (P < 0.05) between key CDM (relative abundance > 1%) and significantly differential VFCs.

By screening core microbial taxa (Fig. 3(E),(F), relative abundance > 1%, including five bacterial and four fungal genera) and performing correlation analysis with chemical components (Fig. 6(B)), Aerococcus emerged as the most prominent taxon, showing significant negative correlations with all components except RS and K – strongest for NIC – underscoring its multifaceted metabolic influence on tobacco quality. Among the VFCs, the upregulated VFCs (1‐pentanol, benzaldehyde, nonanal, propargyl alcohol, and 2‐mercaptopropanoic acid) were significantly and positively correlated with each other. Furthermore, bacterial taxa were more closely associated with VFCs than fungal taxa; among these, Aerococcus was the only genus significantly correlated with all metabolites (P < 0.05). Its abundance was positively associated with upregulated compounds and negatively with downregulated ones, whereas core fungal taxa showed no significant metabolic associations (Fig. 6(C),(D)).

Microbial‐driven metabolic mechanisms underlying the formation of VFCs

The dynamic changes in the primary chemical components of cigar tobacco leaves during fermentation are not an isolated passive consumption but rather a systematic process of flavor precursor remodeling driven by active microbial metabolism. To characterize the potential metabolic functions of bacterial communities, we employed PICRUSt analysis, which revealed that ‘metabolism’ dominated the level 1 pathways (accounting for >70% of functional annotations), followed by ‘environmental information processing’ and ‘genetic information processing’ (Fig. 7(A)). In terms of carbohydrate metabolism, PICRUSt functional prediction analysis revealed a prominent abundance of carbohydrate metabolism pathways at level 2, peaking at the mid‐fermentation stage (F2) (Fig. 7(B),(C)), indicating that glycolysis (EMP pathway) and the pentose phosphate pathway may play important roles during this phase, participating in carbon source decomposition and the generation of reducing equivalents (NAD(P)H), thereby providing energy and precursors for subsequent metabolic reactions. Previous studies have shown that certain fungi (e.g. Aspergillus) and bacteria (e.g. Bacillus) secrete hydrolases including α‐amylase, glucoamylase, and cellulase, which can degrade polysaccharides (e.g. starch and cellulose) in tobacco leaves into utilizable reducing sugars.^{21, 22} On the one hand, the generated reducing sugars are rapidly taken up by microorganisms as readily available carbon sources and consumed via respiratory metabolism; on the other hand, they can undergo the Maillard reaction with free amino acids in tobacco leaves, followed by subsequent Strecker degradation, to generate Strecker aldehyde flavor compounds such as benzaldehyde and nonanal.^{23, 24} Furthermore, during the mid‐fermentation stage, Staphylococcus dominated with a relative abundance exceeding 97%, becoming the absolute dominant bacterial group (Fig. 2(F)). This genus possesses efficient sugar transport and catabolic capabilities.^{29, 30} In the early fermentation stage, fungi and Bacillus dominated polysaccharide degradation and reducing sugar release, while in the mid‐stage, Staphylococcus proliferates massively and rapidly consumed reducing sugars. This synergistic succession of microbial communities drove the dramatic loss of carbohydrate materials in tobacco leaves.

Figure 7.

Dynamics of bacterial functional profiles in different fermentation stages analyzed by PICRUSt (n = 3). (A) Level 1 metabolic pathway. (B) Level 2 KEGG ortholog functional predictions of the relative abundances in the top 20 functions. (C) Level 3 KEGG ortholog functional predictions of the relative abundances of the top 20 metabolic functions.

The continuous transformation and consumption of nitrogen‐containing compounds during fermentation are typically associated with microbially mediated protein hydrolysis and subsequent amino acid metabolism, but may also be jointly influenced by factors such as ammonia volatilization and nonenzymatic reactions. Extracellular proteases secreted by bacteria and fungi progressively degrade macromolecular proteins in tobacco leaves into peptides, which are further hydrolyzed to generate free amino acids.^{54, 55} PICRUSt functional prediction results showed significant enrichment of the ‘amino acid metabolism’ pathway at the mid‐fermentation stage F2 (Fig. 7(B),(C)), providing functional evidence that supports the continuous degradation of tobacco proteins and amino acid transformation. After fermentation, two typical Strecker aldehydes, benzaldehyde and nonanal, accumulated significantly in the samples (Fig. 5(C),(D)). Among them, benzaldehyde primarily originates from the metabolic transformation of phenylalanine: microorganisms such as Aspergillus and Aerococcus can secrete PAL, which catalyzes the deamination of phenylalanine to trans‐cinnamic acid, and this acid is then degraded via β‐oxidative cleavage to generate benzaldehyde.^{14, 15} Nonanal can be generated either from the Strecker degradation of branched‐chain amino acids such as methionine and leucine, or through the lipid peroxidation pathway of tobacco leaves. It is worth emphasizing that, in addition to participating in the LOX–HPL pathway, Methylobacterium‐Methylorubrum can also synthesize aroma compounds such as β‐ionone and megastigmatrienone by cleaving carotenoids via its secreted carotenoid cleavage oxygenase. ³⁷ These compounds can directly impart woody, floral, and fruity notes to tobacco, further enriching the aroma complexity and sensory quality of tobacco leaves.

In lipid metabolism pathways, the degradation of membrane lipids (e.g. phospholipids, glycolipids) and neutral lipids (e.g. triglycerides) in tobacco leaves is an important source of characteristic aroma compounds. Microorganisms in the fermentation system secrete lipases and oxidases. Lipases catalyze the hydrolysis of lipids to generate free fatty acids (e.g. unsaturated fatty acids such as linoleic acid and linolenic acid). ⁵⁶ These free fatty acids are further converted into VFCs such as aldehydes, ketones, and alcohols via β‐oxidation or the LOX‐mediated oxidative pathway, contributing to tobacco leaf aroma. ⁵⁷ Although this study did not directly measure changes in lipid content during fermentation, two key pieces of evidence jointly confirm the active state of lipid oxidation metabolism during fermentation: first, the significant accumulation of nonanal (Fig. 5(C),(D)), and second, the significant enrichment of the ‘lipid metabolism’ pathway at the mid‐fermentation stage (F2) as shown by PICRUSt functional prediction (Fig. 7(B)). Specifically, the LOX–HPL pathway is a core biochemical mechanism in plants and microorganisms for converting unsaturated fatty acids (e.g. linoleic acid, linolenic acid) into C6–C9 aldehydes.^{24, 58} In this pathway, LOX secreted by microorganisms (e.g. Bacillus, Aspergillus, and Methylobacterium‐Methylorubrum) first catalyzes the oxidation of linoleic acid to generate 9‐hydroperoxide or 13‐hydroperoxide, which is then specifically cleaved by HPL to produce aliphatic aldehydes such as hexanal, nonanal, or nonenal.

Through multi‐omics data integration, this study identified that core microbial taxa act as a critical ‘bridge’ between chemical components and volatile compounds. Specifically, on the one hand, chemical components (e.g. sugars and amino acids) serve as substrates that significantly drive microbial community succession; on the other hand, specific functional microorganisms regulate the generation pathways and accumulation levels of volatile compounds through their metabolic networks. This process of ‘substrate‐driven microbial response–metabolic output’ constitutes the core mechanism of flavor formation during tobacco leaf fermentation.

Limitations of functional prediction and future validation strategies

Importantly, these functional inferences are predictions based on 16S rRNA gene‐derived taxonomic profiles rather than direct measurements of enzymatic activities or gene expression. Accordingly, our results should be viewed as a robust hypothesis‐generating platform rather than conclusive evidence of in situ metabolic functions. The co‐occurrence of the predicted peaks in carbohydrate and amino acid metabolism with the accumulation of their downstream products (e.g. Strecker aldehydes) in F2 provides compelling correlative evidence, but experimental validation is still required. Future studies should integrate a ‘multi‐omics analysis plus in vitro validation’ strategy to overcome the limitations of single correlational analyses. First, shotgun metagenomic sequencing should be prioritized to resolve species‐specific key functional genes, such as those encoding nicotine dehydrogenase, lipoxygenase, glycosidases, and esterases. Second, metatranscriptomics should be employed for deciphering the temporal transcriptional expression profiles of the aforementioned pathways during the critical fermentation phases (with particular emphasis on the F2 stage). Finally, pure culture experiments (targeting taxa such as Bacillus spp. and Staphylococcus spp.) or the construction of directed synthetic microbial communities (e.g. Aerococcus + Methylobacterium‐Methylorubrum) should be performed to track their metabolic activities on sterilized tobacco leaves in gnotobiotic experiments. This transition from theoretical hypotheses to causal verification will provide the most direct evidence for elucidating the molecular mechanisms by which specific microbial consortia regulate substrate degradation and the biosynthesis of characteristic aromas during cigar fermentation.

CONCLUSION

This study systematically revealed the dynamic relationships between phyllospheric microbial succession and volatile flavor formation during cigar tobacco fermentation using integrated multi‐omics analysis. Bacterial diversity exhibited a characteristic ‘decrease–increase’ trend, with Staphylococcus dominating the mid‐fermentation stage (relative abundance > 97%), while Aerococcus and Methylobacterium‐Methylorubrum became enriched in the final stage. Fermentation significantly reduced total sugars (75.53%), reducing sugars (94.29%), total nitrogen (16.50%), and chloride (55.62%), accompanied by 47 differentially abundant VFCs, among which benzaldehyde, nonanal, 1‐pentanol, propargyl alcohol, and 2‐mercaptopropanoic acid were significantly enriched. Bacterial communities showed stronger associations with volatile compounds than fungal communities; with Aerococcus correlated positively with all key upregulated aroma compounds, suggesting its potential role in flavor development. Functional prediction further indicated that carbohydrate, amino acid, and lipid metabolism pathways may contribute to volatile flavor formation through microbial‐mediated substrate transformation. Overall, this study provides new insights into microbial succession–aroma relationships in cigar fermentation and identifies potential functional microorganisms, offering a theoretical basis for targeted microbial inoculants and precision fermentation strategies. Future studies integrating metagenomics, metatranscriptomics, and synthetic community validation will clarify the causal roles of key microorganisms and enable more controllable fermentation systems.

FUNDING INFORMATION

This work was supported by ‘Study and application of microbial dynamics in cigar fermentation’ (202312).

AUTHOR CONTRIBUTIONS

Jiaxin Liu and Lei Tian: conceptualization, methodology, data curation and writing – original draft. Youqing Dai, Mingming Sun, and Gao Qiang: formal analysis and writing – review & editing. Lili Wang and Liang Wen: software and investigation. Dianjun Wu and Limin Kong: visualization and supervision. Xiaoyu Wang and Long Yang: validation and resources. Xin Hou and Li Zhang: project administration and funding acquisition. All authors have read and agreed to the published version of the paper.

DATA AVAILABILITY STATEMENT

The datasets generated during and/or analyzed during the current study are available from the corresponding author on reasonable request.The raw sequence data that support the findings of this study have been deposited in the NCBI Sequence Read Archive (SRA) under BioProject accession number PRJNA1452187.