In this study, we used our previously characterized K. variicola H8 strain for MTA analysis after applying it to RTLC fermentation. Our research group has reported its role in the overproduction of several NAECs, nicotine degradation, and sensory quality improvement. Our previous studies have reported optimized growth conditions for the optimal growth of this strain and a method for quantification of sensory quality (Huang et al., 2024a; Huang et al., 2024b). Since MTA adds another layer to metagenomics analysis for this current study, we decided to stick with the same strain and culture conditions. We collected samples at five different time points, specifically at 0, 8, 16, 24, and 36 h (labeled CK, H8H, H16H, H24H, and H36H, respectively, and took three replicates for each. All samples were then sent to the Shanghai Paisenore Biological Co., Ltd. (China) for metatranscriptomic sequencing.

We also used the same protocol for the extraction, GC-MS analysis, and quantification of NAECs and nicotine in fermented RTLC (Huang et al., 2024b). Statistical significance of correlations was evaluated using Pearson’s test with Benjamini-Hochberg false discovery rate (FDR) correction for multiple testing. Adjusted p-values <0.05 were considered significant.

In a previous study, we found that fermenting RTLC with K. variicola H8 led to a noticeable increase in NAEC production, specifically, 34 compounds were enhanced, making up about 45% of the total. At the same time, nicotine levels dropped by 25%, and sensory quality scores improved by 5.71% (Huang et al., 2024b). However, one concern with prolonged fermentation is the formation of tobacco-specific nitrosamines (TSNAs), which are known carcinogens (Li et al., 2020). Because of that, we wanted to figure out the exact fermentation time to avoid the production of TSNAs.

To do this, we ran a time-course analysis, tracking NAEC and nicotine levels over several key points of fermentation. According to the GC-MS results, the most significant rise in NAECs and the most effective reduction in nicotine was observed by the 24-h mark (Figure 1). This spike likely stems from how K. variicola H8 utilizes a range of nutrients, including carbohydrates, amino acids, lipids, and even nicotine itself, as fermentation progresses (Ardö, 2006; Liang et al., 2024; Ning et al., 2023; Rodríguez-Bustamante and Sánchez, 2007; Yvon and Rijnen, 2001).

Interestingly, our earlier metagenomic analysis supports this, showing that the RTLC microbiome carries the genetic tools needed to break down those same compounds (Huang et al., 2024b). Other studies have reported similar patterns. For instance, microbes like Paenarthrobacter nicotinovorans (Zhang et al., 2022), Ochrobactrum intermedium DN2 (Yuan et al., 2006), and Pseudomonas sp. Nic22 has all been used to degrade nicotine and enhance tobacco quality (Li et al., 2024). According to Z.-J. Li and colleagues, microbes can use nicotine as a source of both carbon and nitrogen to generate the energy they need for growth (Li et al., 2024).

Based on our findings, the ideal fermentation time when using K. variicola H8 appears to be around the 24-h mark. At this point, NAEC production is maximized, nicotine levels are significantly reduced, and harmful TSNAs are not yet a concern (Figure 1). In this study, TSNAs were not quantified, as their formation during fermentation has already been reported in our previous work (Huang et al., 2024b).

The relative abundance of RNA was analyzed to evaluate the development of microbial community structure during the fermentation of RTLC; subsequently, their role in shaping the quality of tobacco products is inferred (Figures 2, 3). Figures 2, 3 present the relative abundance of the top 20 microbial species and the differential gene expression of each organism, respectively, at 0 h, 8 h, 16 h, 24 h, and 36 h of RTLC fermentation. Figure 3 complements the results in Figure 2.

Figure 2 shows K. variicola H8, Bacillus coagulans, and Lactobacillus formosensis as dominant microbial strains. In contrast, Figure 3 indicates that these species, along with K. variicola, Klebsiella pneumoniae, and Serratia marcescens, are transcriptionally active microbial species at 0 h of the fermentation. These species are commonly found in the raw material of plants, and they generally kickstart the fermentation process when they see the opportunity (Suresh, 2023; Hleba et al., 2021; Jia et al., 2008; Rodríguez-Medina et al., 2019).

The results in Figure 2 show that K. variicola H8, Debaryomyces hansenii, Citreicella sp. 357, Lactobacillus farraginis, and Klebsiella pneumoniae were relatively dominant strains and therefore the relative abundance of their RNA was high (Figure 3) at the beginning of the fermentation (8 h), which indicates that these strains also play a critical role in the fermentation of RTLC. The decomposition of plant material with microbial consortia for the production of high-value compounds has been reported by other studies as well (Gentzke et al., 2022; L. Zou et al., 2024). The cigar fermentation studies have also reported the role of the abovementioned dominant bacterial and yeast species on the production of flavor and metabolic activity (Si et al., 2023; Tao et al., 2024).

It is evident from the results depicted in Figure 2 that the structure of the microbial community involved in the fermentation of RTLC is tending toward equilibrium state in which microbial species adapt to the changing culture environment (16 h). The microbial species such as Debaryomyces hansenii, Lactobacillus farraginis, and Citreicella sp. 357 remain relatively abundant (Figure 2) and transcriptionally active (Figure 3). The relative RNA abundance of K. variicola H8 was highest at the 16th hour of RTLC fermentation (Figure 3). In contrast, the relative RNA abundance of early contributors to the RTLC fermentation, such as Serratia marcescens, and K. Pneumoniae, along with Citrobacter sp., Lactobacillus sp., and Aromaticobacter was also simultaneously decreased (Figure 3) (16 h).

The change in the microbial structure and prevalence of yeast and lactic acid bacterial species has been reported as crucial for RTLC fermentation and the production of flavor compounds in the subsequent stages (Huang et al., 2024b; Pan et al., 2022). The relative species and RNA abundance of S. cerevisiae and acid-tolerant lactic acid bacteria like L. Pobuzihii increased at 36 h of RTLC fermentation (Figures 2, 3), which explains their role in the carbohydrate metabolism and organic acid production (Tao et al., 2024). On the contrary, the relative species and RNA abundance of Escherichia coli decreased (Figures 2, 3), which can be attributed to the following factors, including changes in the pH of the media, depletion of resources, and outnumbered (Li et al., 2020) by S. cerevisiae and acid-tolerant lactic acid bacteria like L. Pobuzihii.

The results in Figures 2, 3 highlight microbial succession and cooperation, particularly the role of K. variicola H8 in initiating RTLC fermentation. This initiation subsequently creates an opportunity for the growth of opportunistic members commonly found in unfermented RTLC. As a result, K. variicola H8 strongly influences NAECs production and the overall quality of tobacco products.

The relationship between the changes in conventional chemical components and microbial transcription after RTLC fermentation was evaluated (Figure 4). The correlation analysis reveals a strong association between microbial activity and variations in nicotine, soluble sugar, and potassium (K) content. K. variicola exhibited a significant positive correlation with soluble sugar, indicating its potential role in sugar metabolism during fermentation. This aligns with previous studies suggesting that Klebsiella species actively participate in carbohydrate metabolism and contribute to the breakdown of complex sugars in fermentation systems (Duran-Bedolla et al., 2021).

Furthermore, Escherichia sp. R8 and Acinetobacter baumannii showed a negative correlation with nicotine content, suggesting their potential involvement in nicotine degradation (Figure 4). Similar findings have been reported in studies where Acinetobacter and Pseudomonas species possess nicotine-catabolizing enzymes, contributing to their biotransformation during microbial fermentation (Wang et al., 2007). The weak correlation of K. variicola with nicotine suggests that its primary metabolic activity is centered around sugar utilization rather than nicotine degradation (Figure 4) (Huang et al., 2024b). Interestingly, the correlation analysis also revealed that K. variicola exhibited a negative correlation with potassium content (Figure 4). The underlying mechanism remains unclear; however, previous research suggests that potassium plays a crucial role in bacterial osmoregulation, stress responses, and metabolic activity, which could indirectly influence microbial interactions in the fermentation system (Stautz et al., 2021).

Additionally, Lactobacillus acidipiscis demonstrated a positive correlation with soluble sugar content, reinforcing its known role in lactic acid fermentation and carbohydrate metabolism (Figure 4). This is consistent with studies showing that Lactobacillus species are key players in sugar fermentation and organic acid production (Cufaoglu and Erdinc, 2023; Hedberg et al., 2008; Stautz et al., 2021).

The findings highlight the complex interactions between microbial communities and chemical composition changes during fermentation. K. variicola is a dominant sugar-fermenting bacterium, whereas Acinetobacter and Escherichia contribute to nicotine degradation.

The correlation analysis between relative RNA abundance and NAECs during the RTLC fermentation reveals significant relationships between specific microbes and volatile aroma compounds (Figure 5). The results in Figure 5 demonstrate a strong positive correlation between K. variicola H8 and key NAECs (2,4-Di-tert-butylphenol, dihydroactinidiolide, phenylethyl alcohol, benzyl alcohol, linalool, β-damascenone, 1-cyclohexyl-2-butenol, nerylacetone, solanone, α-cyperone, farnesyl acetone, and 2-acetyl-1H-pyrrole). Different studies have reported that these NAECs add aroma to tobacco products, such as β-damascenone and linalool, which add floral and fruity flavor (Gong et al., 2023; Pan et al., 2022). In addition, significant positive correlation between K. variicola H8 and compounds like phytol, phytone, megastigmatrienone, and saffron aldehyde was observed which indicates that this strain plays a vital role in the biosynthesis of these compounds (Figure 5). These findings align with previous studies that demonstrate how Klebsiella species contribute to the transformation of precursor molecules into aromatic volatiles (Chen et al., 2021; Huang et al., 2024b).

The correlation analysis of microbial species and aroma compounds exhibited that the Westpac and vitamin A are negatively correlated with K. variicola H8 (Figure 5), which indicates that both compounds are either used by these organisms for their growth or their production was inhibited during the RTLC fermentation. Similar outcomes have been reported by other studies, which state that fermentations with microbial organisms result in reduced concentrations of Westpac and vitamin A (Denter et al., 1998; Whited et al., 2002). The aroma characteristics and sensory attributes of 2,4-di-tert-butylphenol and dihydrokiwifolactone are well-registered in the tobacco and food industry (Wang et al., 2023). In this, we also discovered a positive correlation between these compounds and L. acidipiscis (Figure 5).

The correlation analysis between microbial species and NAECs production demonstrates that K. variicola H8 and other microbial species play a critical role in enriching the RTLC with NAECs.

The transcriptional analysis of flavor-enhancing bacteria during the fermentation of RTLC was performed using the CAZy enzyme family annotation (Lombard et al., 2014). The distribution of CAZy-related transcripts is shown in Figure 6A. Among the enzyme families, glycoside hydrolases (GH) exhibited the highest transcriptional abundance (955 transcripts), followed closely by glycosyltransferases (GT) (931 transcripts). Polysaccharide lyases (PL) were the least represented (42 transcripts).

Other enzyme families, such as carbohydrate esterases (CE), auxiliary activity (AA), and carbohydrate-binding modules (CBMs), which are associated with carbohydrate metabolism, also played a vital role in the RTLC fermentation, and their contribution was recorded in terms of 137, 118, and 293 transcripts, respectively (Figure 6A). The presence of CBM transcripts highlights their role in the hydrolysis of carbohydrate, which involves the facilitation of the enzyme-substrate interactions (Boraston et al., 2004). The presence of high levels of GH and GT transcripts indicates that carbohydrate molecules in the RTLC were mainly degraded through hydrolysis and glycosylation reactions (Chen et al., 2025; Muradova et al., 2023; Pan et al., 2022; Parapouli et al., 2019). In addition, the difference in the relative abundance of GH, GT, PL, CE, AA, and CBM in this study (Figure 6A) endorses the findings of previously published studies (Cantarel et al., 2009; Lombard et al., 2014). Furthermore, these studies have also reported that GH, GT, PL, CE, AA, and CBM catalyze the breakdown of carbohydrates into NAECs (Cantarel et al., 2009; Lombard et al., 2014).

The change in the transcriptional profile of GH, GT, PL, CE, AA, and CBM over time is presented in Figure 6B, which shows that the expression level of GH was severalfold increased (Figure 6B). Among these, GH130 exhibited a striking 4.2-fold increase, followed by GH43-30 (3.7-fold), GH78 (2.5-fold), and GH31 (3.8-fold) by 36 h. These results suggest that these glycoside hydrolases are crucial in carbohydrate degradation and aroma formation. GH47, GH32, GH39, GH76, GH3, GH13-25, and GH28 also displayed transcriptional upregulation, ranging from 1.6 to 4.5-fold. Interestingly, these observations align with findings from other fermentation studies. For instance, research on Debaryomyces hansenii Y4 during Sichuan South-road Dark Tea fermentation identified the upregulation of GH families such as GH17, GH18, GH76, GH31, GH47, and GH2, where enzymes like β-galactosidase and mannosidase influenced the tea’s flavour by degrading polysaccharides and oligosaccharides (Zou et al., 2023). Similarly, comparative genomics of lactic acid bacteria emphasized the genetic basis for flavour compound biosynthesis, including the role of GHs in forming flavour-active metabolites (Liu et al., 2008).

The findings on the expression patterns of the GH gene suggest that enhancing the expression of specific glycoside hydrolases can improve the efficiency of the production of NAECs and improve the flavor quality of the product. Therefore, this approach can be applied to the food and tobacco industries, where controlled microbial fermentation could be optimized to enhance sensory attributes.

A correlation analysis was performed to investigate the relationship between GH family transcript abundance and the production of NAECs during the fermentation of RTLC by K. variicola H8. The resulting heatmap (Figure 7) illustrates positive and negative associations between specific GH families and various NAECs, underscoring the critical role of microbial enzymatic activity in modulating tobacco aroma profiles.

Several GH families, particularly GH78, GH13_25, GH31, GH28, GH16_18, and GH76, exhibited strong positive correlations with a broad range of aroma compounds (Figure 7). These findings suggest that the overexpression of these enzymes during fermentation may facilitate the enzymatic release of volatile compounds from glycosidically bound precursors, a mechanism widely supported in the literature (Hu et al., 2016a; Hu et al., 2016b). Among the aroma compounds, megastigmatrienone, an essential contributor to the sweet and woody aroma, showed exceptionally high positive correlations with GH78 and GH13_25 (Figure 7), consistent with reports that it is released via microbial deglycosylation of carotenoid-derived precursors (Hu et al., 2016a; Hu et al., 2016b). Similarly, the fruity aroma compound dihydrokiwi lactone is strongly associated with GH31, GH28, and GH16_18 (Figure 7), further supporting their involvement in lactone biotransformation. This aligns with prior findings where microbial strains like Yarrowia lipolytica converted hydroxy fatty acids into aroma-active lactones such as γ-decalactone through enzymatic processes (AL Mualad et al., 2022; Silva et al., 2021).

Additional volatiles such as benzyl alcohol, farnesyl acetone, and linalool, known for their floral and woody aromatic profiles, were also positively associated with GH31 and GH28 (Figure 7). Previous research has shown that these compounds often occur in plants as glycosidically bound forms, which GH enzymes can hydrolyze to release the free aroma-active compounds (Sarry and Günata, 2004; Zheng et al., 2019). While direct evidence for GH-mediated release of farnesyl acetone remains limited, the general role of glycoside hydrolases in liberating bound volatiles supports this hypothesis. The antioxidant compound 2,4-di-tert-butylphenol, which contributes woody notes, also showed positive correlations with GH16_18 and GH76 (Figure 7), consistent with their potential role in producing phenolic volatiles (Leonard et al., 2021).

Other key NAECs, such as 1-cyclohexyl-2-butanone, orange peptone, 4-cyclopentene-3-one, and 6-methyl-5-hepten-2-one, exhibited significant positive correlations with GH31, GH78, GH13_25, and GH115 (Figure 7). These compounds produce sweet, citrus, smoky, and fruity sensory qualities in the tobacco products (Maldonado-Robledo et al., 2003; Yan et al., 2022). Likewise, kamakui yeoh alcohol displayed a positive correlation with GH31 and GH16_18 (Figure 7), highlighting the broad substrate specificity of these enzymes (Lindsay et al., 2022).

The GH31 demonstrated a negative correlation with Westpacene and vitamin A (Figure 7). A similar trend has also been observed between these compounds and K. variicola H8 in Figure 5. Therefore, these results confirm the degradation of Westpacene/vitamin A through hydrolysis during the RTLC fermentation, which is perhaps for microbial growth or NAECs overproduction or both. Other studies have associated GH31 activity with terpene metabolism and glycoside conversion (Caffall and Mohnen, 2009; Cai et al., 2023). Additionally, further transcriptomic correlations revealed that GH78, GH13_25, GH28, and GH31 were also associated with increases in secondary aroma compounds such as kauri ketones, α-hydroxybenzoin, lycopene, and macadamia trienone.

The above results unveil the central role of GH, particularly GH78, GH13_25, GH31, GH28, GH16_18, and GH76, in producing NAECs during the fermentation of RTLC.