Keystone microbial taxa in the formation of stale aroma during pile fermentation of ripened Pu-erh tea Authors: Yiwei Weng, Shiqiang He, Zhengfei Luo, Jin Sun, Qinghua Cheng, Yingjuan Chen, Huarong Tong Published: 2026 ABSTRACT AbstractRipened Pu-erh tea is prized for its distinctive stale aroma. Methoxy-phenolic compounds, key contributors to this aroma, are produced during pile fermentation; however, the specific microorganisms responsible for their synthesis remain unclear. In this study, we identified the dominant taxa (Aspergillus luchuensis, A. fumigatus, Staphylococcus gallinarum, and S. kloosii) during pile fermentation through morphological analysis and metagenomic profiling. Gas Chromatography-Mass Spectrometry (GC-MS) analysis demonstrated the pivotal role of methoxy-phenolic compounds in the stale aroma. Moreover, using a metagenomic-based Weighted Gene Co-expression Network Analysis (WGCNA) combined with bivariate correlation network analysis, we identified key microbial taxa (Trichomonascus ciferrii, Heyndrickxia coagulans and Enterococcus sp.) involved in the generation of these compounds. Finally, we found that solid-state fermentation involving both dominant and keystone microbial taxa produced the highest levels of methoxy-phenolic compounds. Our findings reveal an inconsistency between dominant high-abundance taxa and keystone microbial taxa responsible for methoxy-phenolic compound synthesis during pile fermentation.Subject terms: Biotechnology, Microbiology Introduction Ripened Pu-erh tea from Yunnan Province, China, is well-liked due to its unique stale aroma. The development of stale aroma is primarily attributed to the formation of numerous aromatic compounds, with methoxy-phenolic compounds playing a pivotal role in contributing to the overall aromatic profile1,2. The content of methoxy-phenolic compounds in ripened Pu-erh tea is significantly higher than that in green, black, oolong, and white teas, establishing them as diagnostic chemical markers for ripened Pu-erh tea among other tea varieties3. The odor activity value (OAV) is defined as the ratio of a compound’s concentration to its sensory threshold, wherein a higher OAV signifies a more pronounced contribution to the overall aromatic profile4. In the analysis of volatile compounds in ripened Pu-erh tea, methoxy-phenolic compounds exhibit the highest OAV5–10. The formation of methoxy-phenolic compounds occurs during pile fermentation of sun-dried green tea into ripened Pu-erh tea3,11. Microorganisms are essential in this process, facilitating the microbial-mediated biotransformation of the intrinsic components of tea. This transformation is driven by the synergistic interaction of extracellular enzymes secreted by microorganisms, the thermal energy generated from microbial metabolic respiration, and the moist heat from the tea’s inherent moisture12,13. Methoxy-phenolic compounds are derived from gallic acid (GA) via microbial methylation, catalyzed by oxygen-methyltransferases secreted by microorganisms. These enzymes transfer a methyl group to the hydroxyl group of GA14,15. Although this pathway is widely acknowledged as the principal route for the biosynthesis of methoxy-phenolic compounds, no research has yet identified the specific microbes responsible for secreting oxygen-methyltransferases during pile fermentation of Pu-erh tea. Microbial communities serve as critical drivers of numerous ecosystem functions, including climate warming, soil nutrient cycling, and fermented foods, among others16,17. Weighted Gene Co-expression Network Analysis (WGCNA) utilizes metagenomic data derived from microbial communities to construct a gene co-expression network18. By analyzing gene-gene correlations, WGCNA identifies keystone taxa (i.e., highly interconnected microbial groups that significantly influence microbiome structure and function regardless of their abundance) and explores their associations with relevant phenotypes. Their removal can cause profound changes in the microbiome’s structure and function16,19–21. In Chinese strong-flavor Baijiu, Yuan et al. identified keystone taxa associated with ethyl acetate and higher alcohols that, while non-dominant, significantly influenced the liquor’s flavor profile22. Similarly, Cheng et al. demonstrated that seven bacterial keystone taxa played pivotal regulatory roles in both flavor formation and biocontrol during the industrial pile fermentation of Qingzhuan tea23. Dominant microbial taxa refer to those with high relative abundance whose substantial population size allows them to exert influence over ecosystem functions or specific processes, primarily by virtue of their numerical predominance within the community16. Although significant research has been done on the dominant taxa in the pile fermentation of ripened Pu-erh tea, studies focusing on keystone taxa remain limited. In this study, we identified the keystone taxa involved in the formation of stale aroma during the pile fermentation of ripened Pu-erh tea. First, we confirmed the significance of methoxy-phenolic compounds in the stale aroma through sensory evaluation and gas chromatography-mass spectrometry (GC-MS) analysis. Subsequently, by combining microbial isolation and identification techniques with metagenomic analysis and bioinformatics approaches, we characterized the dominant taxa and identified the keystone taxa involved in the biosynthesis of methoxy-phenolic compounds through an in-depth analysis of tea samples from the pile fermentation process. Finally, we validated these findings through solid-state fermentation simulation experiments. Results Volatile compounds in ripened Pu-erh tea at different pile fermentation stages We initially conducted a comprehensive evaluation of the aroma of tea infusions from the eight tea samples. The aroma of ripened Pu-erh tea was described using a sensory panel established through a 0–7 point scale. The selected characteristic aromas included grass, stale, sweet, woody, jujube, and herbal24 (Fig. 1a). We found that with the progression of fermentation time, the grass aroma gradually decreased and even disappeared, while the stale, herbal, woody, and jujube aromas progressively increased, particularly the stale aroma. The sweet aroma initially increased but then decreased. The stale aroma was proven to be an essential component of the overall aroma profile; consequently, we also assessed the purity and persistence of the stale aroma using a 0–7 point scale. As depicted in Fig. 1b, the purity and persistence of the stale aroma for all eight tea samples gradually increased. Both P5 and RP exhibited the same high level of purity, surpassing that of P6, while the persistence of P6 and RP was comparable, both being greater than that of P5. Therefore, through sensory evaluation, we observed that the stale aroma not only plays an important role in the overall aroma profile but also forms during the pile fermentation process. Volatile compounds in ripened Pu-erh tea at different pile fermentation stages. a Aroma sensory evaluation of ripened Pu-erh tea at different fermentation stages (radar plot); b Evaluation of the purity and persistence of stale aroma in ripened Pu-erh tea at different fermentation stages; c PCA analysis of ripened Pu-erh tea at different fermentation stages; d Composition of volatile compounds in ripened Pu-erh tea at different fermentation stages; e Heatmap of the changes in key volatile compounds with OAV > 1 at different fermentation stages; f Analysis of 1,2-dimethoxybenzene and 1,2,3-trimethoxybenzene variation among eight tea samples using external standard method. The bar charts were generated using OmicStudio (https://www.omicstudio.cn/tool), while the radar plot, PCA plot, and heatmap were created with the CNSknowall online tools (https://cnsknowall.com). The final layout was assembled using Adobe Illustrator (version 2024). Volatile compounds in eight tea samples were analyzed using GC-MS with ethyl decanoate as an internal standard (Table S1). The results revealed that RM and RP could be distinctly separated from the tea samples undergoing pile fermentation (Fig. 1c). As illustrated in Fig. 1c, the observed overlaps between P1 and P2, P3 and P4, as well as P5 and P6, indicate that the tea samples from the pile fermentation process can be classified into three distinct stages—early (P1, P2), mid (P3, P4), and late (P5, P6)—consistent with the staging criteria established by Li et al.15. This classification is further corroborated by hierarchical clustering analysis, which clearly groups the samples into three corresponding clusters corresponding precisely to the early, mid, and late fermentation stages (Fig. S1), thereby reinforcing the robustness and methodological convergence of the proposed stage delineation. Additionally, a classification of the volatile compounds in the eight tea samples was conducted, revealing that the relative abundance of alcohols was highest in both the raw material and early stages of pile fermentation, followed by ketones, esters, aldehydes, and alkanes (Fig. 1d). However, with the extension of fermentation time, methoxy-phenolic compounds progressively increased from the mid-stage of pile fermentation, eventually surpassing alcohols in abundance. Given that methoxy-phenolic compounds were absent in both the RM and early fermentation stages, their formation was attributed to the mid-to-late stages of pile fermentation. Volatile compounds with an OAV greater than 1 were considered key aroma compounds, contributing significantly to the overall aroma profile. Figure 1e illustrates the variation of volatile compounds with an OAV greater than 1 across the tea samples. Methoxy-phenolic compounds include 1,2-dimethoxybenzene and 1,2,3-trimethoxybenzene. As pile fermentation progressed, the OAV of methoxy-phenolic compounds exhibited a continuous increase, while the OAV of linalool, linalool oxide II, and β-ionone demonstrated a concomitant decrease. Both the content of methoxy-phenolic compounds in Fig. 1d and their OAV in Fig. 1e exhibit an increasing trend as the pile fermentation time lengthened. To ensure the accuracy of the results, we also conducted quantitative analysis of methoxy-phenolic compounds using the external standard method, with the standard curve data provided in Table S2. Figure 1f illustrates the concentration variations of 1,2-dimethoxybenzene and 1,2,3-trimethoxybenzene, calculated via the external standard method, across six tea samples (Table S2). Since methoxy-phenolic compounds were undetected in RM, P1, and P2, only the six samples were displayed. As fermentation time increased, the concentrations of 1,2-dimethoxybenzene and 1,2,3-trimethoxybenzene gradually rose, with the highest concentrations observed in P6. Moreover, the content of 1,2,3-trimethoxybenzene exceeded that of 1,2-dimethoxybenzene, consistent with the results obtained using the internal standard method (Table S1). Culturable dominant microbial taxa isolated via culture medium Based on the preceding experiments, it was determined that methoxy-phenolic compounds were formed during pile fermentation, and therefore, their formation mechanism was intricately linked to microbial activity. Figure 2a depicts six tea samples (P1, P2, P3, P4, P5, and P6) collected at different stages of the pile fermentation process. The first row of Fig. 2a presents photographs of the six tea samples. The second row displays images of the tea samples plated on PDA medium after being diluted to 10−4. The final row shows images of the tea samples plated on BEPA medium after being diluted to 10−4. We observed that the color of the tea leaves from the six samples in the fermentation room deepened progressively, and the moisture content of the leaves decreased (Fig. 2a and Table S3). Observations on PDA reveal that in the P1, P2, and P3, fungi with white mycelium and black spores predominated, while yeasts in the form of round colonies were relatively scarce. In contrast, in the P4, P5, and P6, yeasts became the dominant microorganism. On BEPA, the number of bacteria gradually increased. Therefore, we hypothesize that as fermentation progressed, the abundance of fungi decreased, while the populations of bacteria and yeast increased. To investigate the variation in microbial abundance across different layers of the tea pile, we isolated and quantified the dominant microorganisms in each layer. Figure 2b presents the abundance of culturable dominant fungi and bacteria in six tea samples. The data indicate that fungi predominated during the early stages of fermentation, whereas bacteria and yeast became the dominant microorganisms in the later stages, a trend that is consistent with the findings shown in Fig. 2a. Table 1 lists all cultivable microorganisms isolated and identified from the six tea samples during the pile fermentation process based on ITS and 16S rDNA sequences (Figs. S2 and S3). From Table 1, we observed that among the cultivable fungi, Aspergillus species appeared most frequently, particularly A. niger, A. tubingensis, and A. luchuensis. Among the cultivable bacteria, Staphylococcus species were the most prevalent, especially S. kloosii, S. saprophyticus, and S. gallinarum. Furthermore, as indicated in Table 1 and Fig. 2a, Aspergillus species, which were most frequently observed during pile fermentation, also appeared to be the most abundant fungi in the early stages, suggesting that Aspergillus likely served as the dominant microbial group during this phase. Similarly, Staphylococcus species, which showed the highest prevalence during pile fermentation, were observed as the most abundant white, round bacteria in the later stages, indicating that Staphylococcus may have represented the dominant bacterial group toward the end of the process. Only one yeast species, T. ciferrii, was isolated and identified, which could imply its potential role as a dominant microbial component in the later stages of pile fermentation. Number of dominant microorganisms in ripened Pu-erh tea at different pile fermentation stages. a Photographs of six tea samples, tea leaf dilutions (10-4) plated on agar, and micrographs of the dominant microorganism on the plates; b Bar chart illustrating the number of culturable dominant fungi and bacteria in the six tea samples. The photographs presented in this figure were captured by the authors; the bar charts were generated using OmicStudio (https://www.omicstudio.cn/tool). The final composite layout was assembled using Adobe Illustrator (version 2024). Table 1Cultivable microorganisms isolated and identified from six tea samples during pile fermentationSampleFungiBacteriaYeastP1Aspergillus nigerAspergillus tubingensisAspergillus luchuensisIsotericola variabilisMammaliicoccus sciuriAchromobacter xylosoxidansStaphylococcus saprophyticus/P2Aspergillus nigerAspergillus tubingensisAspergillus luchuensisAspergillus fumigatusAchromobacter xylosoxidansBordetella petriiHeyndrixia coagulansOrnithinibacillus bavariensisStaphylococcus kloosiiTrichomonascus ciferriiP3Aspergillus nigerAspergillus tubingensisAspergillus luchuensisAspergillus fumigatusBipolaris subramanianiiTrichoderma atrovirideTrichoderma harzianumMammaliicoccus sciuriHeyndrixia coagulansStaphylococcus kloosiiBacillus albusStaphylococcus gallinarumBrevibacterium atlanticumTrichomonascus ciferriiP4Aspergillus nigerAspergillus tubingensisAspergillus luchuensisAspergillus fumigatusStagonosporopsis citrulliPhlebiopsis magnicystidiataPleosporales sp.Paraconiothyrium sp.Penicillium brevicompactumAchromobacter xylosoxidansHeyndrixia coagulansStaphylococcus kloosiiStaphylococcus gallinarumBrevibacterium atlanticumEnterococcus sp.Trichomonascus ciferriiP5Aspergillus nigerAspergillus tubingensisAspergillus luchuensisAlternaria destruensCladosporium penidielloidesRhizomucor pusillusLichtheimia corymbiferaMammaliicoccus sciuriStaphylococcus saprophyticusStaphylococcus kloosiiBacillus albusStaphylococcus gallinarumTrichomonascus ciferriiP6Aspergillus nigerAspergillus tubingensisAspergillus luchuensisRhizomucor pusillusApiotrichum montevideenseStaphylococcus saprophyticusStaphylococcus kloosiiStaphylococcus gallinarumHeyndrixia coagulansBrevibacterium atlanticumTrichomonascus ciferrii Cultivable microorganisms isolated and identified from six tea samples during pile fermentation Aspergillus niger Aspergillus tubingensis Aspergillus luchuensis Isotericola variabilis Mammaliicoccus sciuri Achromobacter xylosoxidans Staphylococcus saprophyticus Aspergillus niger Aspergillus tubingensis Aspergillus luchuensis Aspergillus fumigatus Achromobacter xylosoxidans Bordetella petrii Heyndrixia coagulans Ornithinibacillus bavariensis Staphylococcus kloosii Aspergillus niger Aspergillus tubingensis Aspergillus luchuensis Aspergillus fumigatus Bipolaris subramanianii Trichoderma atroviride Trichoderma harzianum Mammaliicoccus sciuri Heyndrixia coagulans Staphylococcus kloosii Bacillus albus Staphylococcus gallinarum Brevibacterium atlanticum Aspergillus niger Aspergillus tubingensis Aspergillus luchuensis Aspergillus fumigatus Stagonosporopsis citrulli Phlebiopsis magnicystidiata Pleosporales sp. Paraconiothyrium sp. Penicillium brevicompactum Achromobacter xylosoxidans Heyndrixia coagulans Staphylococcus kloosii Staphylococcus gallinarum Brevibacterium atlanticum Enterococcus sp. Aspergillus niger Aspergillus tubingensis Aspergillus luchuensis Alternaria destruens Cladosporium penidielloides Rhizomucor pusillus Lichtheimia corymbifera Mammaliicoccus sciuri Staphylococcus saprophyticus Staphylococcus kloosii Bacillus albus Staphylococcus gallinarum Aspergillus niger Aspergillus tubingensis Aspergillus luchuensis Rhizomucor pusillus Apiotrichum montevideense Staphylococcus saprophyticus Staphylococcus kloosii Staphylococcus gallinarum Heyndrixia coagulans Brevibacterium atlanticum Dominant microbial taxa identified through metagenomic analysis Cultivable microorganisms were isolated and identified to obtain pure strains for direct experimental validation of methoxy-phenolic compound production. To complement this targeted approach and identify keystone taxa within the complex community, we performed metagenomic sequencing, enabling the application of WGCNA for the systematic screening of microbial taxa associated with the synthesis of these compounds. We selected P2, P3, P4, P5, and RP for metagenomic analysis, as the significant variation in the content of methoxy-phenolic compounds across these samples offers greater potential for elucidating their formation mechanism (Table S1). Based on Illumina Novaseq sequencing, the sequencing results demonstrated a high degree of reliability and accurately reflected the true microbial composition within the tea samples (Table S4). Figure 3a illustrates the β-diversity and α-diversity of the five tea samples. The left panel of Fig. 3a shows the β-diversity analysis of the five tea samples, revealing overlapping data distributions despite differences in median values. We observed a significant microbial difference between the two-turn and three-turn stages, while the microbial variation between the three-turn and four-turn stages was relatively minimal. Similarly, the microbial differences between the five turn stages and the ripened Pu-erh tea were also relatively small. The right panel presents α-diversity, with the sobs, chao, and ace indices reflecting community richness, and the shannon and simpson indices indicating diversity. The sobs, chao, ace, and Shannon indices first increased and then decreased, while the Simpson index followed a decline-then-increase pattern. This suggested an initial rise followed by a decline in both microbial richness and diversity. Dominant taxa of microorganisms in ripened Pu-erh tea at different pile fermentation stages through metagenomic analysis. a On the left, β-diversity of the five tea samples, and on the right, α-diversity of the five tea samples (P2, P3, P4, P5, and RP); b Venn diagram of the five tea samples; c Microbial composition of the five tea samples (P2, P3, P4, P5, and RP) at the family level; d On the left, variation of the top ten dominant fungi ranked by their relative abundance across the five tea samples, and on the right, variation of the top ten dominant bacteria ranked by their relative abundance across the five tea samples (P2, P3, P4, P5, and RP). The box plot and Venn diagram were generated using the CNSknowall online tools (https://cnsknowall.com); the bar chart was created with Majorbio Cloud (www.majorbio.com). The final composite layout was assembled using Adobe Illustrator (version 2024). The left panel of Fig. 3b presents the Venn diagram of the five tea samples. A total of 12,892 shared microbial species were identified across the five tea samples. Additionally, P3, P4, P5, and RP exhibited 2628 identical microbial species, while P2, P3, P4, and P5 exhibited 956 identical microbial species. The right panel of Fig. 3c presents the microbial composition during the pile fermentation process. With the progression of fermentation time, a noticeable decline in Aspergillaceae and Trichocomaceae was observed, while Staphylococcaceae and Micrococcaceae gradually increased. The Bacillaceae family initially increased and then decreased. During the early stages of pile fermentation, fungi were the predominant microorganisms, while bacteria became the dominant microorganisms in the later stages of fermentation. Figure 3d presents the top ten dominant fungi and bacteria ranked by their relative abundance. The left illustrates that with the progression of fermentation time, the relative abundance of A. luchuensis, A. fumigatus, A. tubingensis, A. niger, and A. piperis decreased, while the relative abundance of Rasamsonia emersonii, T. ciferrii, Fennellomyces sp. T0311, Talaromyces proteolyticus, and Sugiyamaella lignohabitans increased. The right demonstrates that with the increase in fermentation time, the relative abundance of S. gallinarum, Rothia koreensis, S. kloosii, H.coagulans, and Brevibacterium oceani increased, while that of Klebsiella pneumoniae and Listeria gray decreased. Additionally, the relative abundance of Caldibacillus thermoamylovorans, K. sp., and C. sp. initially increased and then decreased. The changes in the relative abundance of microbial composition obtained through metagenomic data (Fig. 3c) align with the variation in the number of cultivable microorganisms determined on the culture medium (Fig. 2a), both indicating that fungal numbers decrease while bacterial and yeast numbers increase during fermentation. However, the results for dominant microorganisms showed both similarities and discrepancies (Table 1). Some dominant microorganisms identified through metagenomic analysis, such as the fungi R. emersonii, F. sp., and the bacteria R. koreensis, K. sp., were not isolated on culture media. Nonetheless, a significant portion of the dominant microorganisms identified through metagenomics corresponded to those accessible to cultivation, including representatives of Aspergillus (e.g., A. luchuensis, A. fumigatus, A. tubingensis, A. niger), Staphylococcus (e.g., S. gallinarum, S. kloosii), and the yeast T. ciferrii. Therefore, for subsequent functional validation, we selected cultivable dominant taxa based on their relative abundance ranking. This included the two highest-ranking fungi (A. luchuensis and A. fumigatus) and the two highest-ranking cultivable bacteria (S. gallinarum and S. kloosii), all of which exhibited a relative abundance greater than 2%. Keystone taxa in microbiota linked to methoxy-phenolic compounds A taxon co-abundance network was constructed from metagenomic data using WGCNA based on Pearson correlation. Topological interrogation revealed hub modules with high interconnectivity, where nodes spanned phylogenetically distant taxa. Through integrative module-trait association analysis correlating specific network modules with methoxy-phenolic compound profiles, we delineated keystone microbial lineages putatively involved in the biosynthesis and metabolic regulation of these aromatic compounds. Figure 4a delineates the node membership distribution across hub modules in the reconstructed co-expression network, with Fig. 4b characterizing intermodular connectivity patterns. A total of 17 modules were identified in the network, with the module members consisting of microorganisms (Fig. 4a). Among these, the MEturquoise module had the highest number of members, followed by MEblue, MEbrown, and MEyellow. The modules with the fewest members were MElightcyan, MEcyan, and MEidnightblue. As illustrated in Fig. 4b, the interrelationships between the modules are depicted. A strong positive correlation was observed between MEturquoise and MEtan, as well as between MEgreenyellow. Furthermore, a significant positive correlation existed between MEblue, MEblack, and MEred. To elucidate module-compound associations, Fig. 4c presents covariant relationships between topological modules and methoxy-phenolic compound profiles, where positive/negative associations are encoded in red/green chromatic scales, respectively. The correlations between 1,2-dimethoxybenzene, 1,2,3-trimethoxybenzene, and the modules followed similar trends. Notably, MEblue, MEblack, and MEred all exhibited significant inverse associations with methoxy-phenolic compounds, while MEgreen demonstrated positive covariation trends. Furthermore, other modules displayed both positive and negative correlations with microorganisms associated with methoxy-phenolic compounds. Figure 4d illustrates the correlations between methoxy-phenolic compound concentrations and the top fifty microorganisms ranked by relative abundance. Among these, T. ciferrii, E. sp., Microbacterium suaedae, S. lignohabitans, H. sp., H. coagulans, M. sediminis, and Saccharopolyspora rectivirgula exhibited the strongest positive correlations. Notably, 50% of these synergistic taxa were assigned to MEgreen, with the remainder distributed across MEbrown (25%), MEturquoise (12.5%), and MEblue (12.5%) modules. Conversely, pronounced anticorrelations were identified with Sphingobacterium composti, Pseudomonas aeruginosa, Pantoea sp. Acro-835, A. tubingensis, and A. eucalypticola. Intriguingly, 90% of these negatively associated taxa clustered within MEblue, while the residual 10% mapped to MEpink. Significantly, these findings were consistent with the results presented in Fig. 4c. The bivariate correlation network analysis presented in Fig. 4e reveals distinct interaction patterns between aromatic compounds and microbial taxa. Green lines show negative correlations, and red lines show positive correlations. The size of the circular nodes corresponding to microorganisms reflects the strength of these correlations. The left panel depicts 1,2-dimethoxybenzene demonstrating significant positive correlations with bacterial species H. coagulans and E. sp., while showing inhibitory associations with fungal taxa including A. luchuensis, A. tubingensis, A. piperis, and A. niger. In contrast, the right panel illustrates that 1,2,3-trimethoxybenzene establishes synergistic relationships with both yeast (T. ciferrii) and prokaryotic species (H. coagulans, E. sp., and H. sp.), concurrently exhibiting antagonistic effects specifically against A. luchuensis and A. piperis. Quantitative criteria were applied to identify keystone taxa: a correlation coefficient threshold of >0.5 was used for WGCNA-based screening (Fig. 4d and Table S5), while a more stringent threshold of >0.7 was adopted for the bivariate correlation network analysis. Through these integrated approaches, H. coagulans, E. sp., and T. ciferrii were consistently identified as keystone taxa. Therefore, the findings from the bivariate correlation network analysis in Fig. 4e were consistent with those from the WGCNA analysis in Fig. 4d, both revealing that methoxy-phenolic compounds exhibited negative correlations with many fungi ranked among the top fifty in relative abundance, while showing positive correlations with numerous bacterial species and a specific yeast. Furthermore, comparison between the keystone taxa identified via metagenomics and those listed in Table 1 revealed that all were cultivable. Keystone taxa of microorganisms linked to methoxy-phenolic compounds in ripened Pu-erh tea at different pile fermentation stages. a Bar plot of the hub modules’ member numbers in WGCNA; b Correlation-based clustered heatmap of hub modules in WGCNA; c Hierarchical dendrogram of methoxy-phenolic compounds and hub module correlations; d Heatmap of correlations between methoxy-phenolic compounds and the top 50 microorganisms by relative abundance; e Bivariate correlation network analysis of methoxy-phenolic compounds and microorganisms. The heatmap was generated using CNSknowall online tools (https://cnsknowall.com). The bar chart, hierarchical clustering diagram, and network diagram were created with Majorbio Cloud (www.majorbio.com). The final composite layout was assembled using Adobe Illustrator (version 2024). The dominant taxa and keystone microbial taxa were confirmed by solid-state fermentation After identifying the dominant taxa and keystone taxa in the microbiota associated with methoxy-phenolic compounds during the pile fermentation, we designed solid-state fermentation experiments to validate which taxa play the most significant role in the production of methoxy-phenolic compounds. Figure 5 illustrates validation of the role of keystone taxa. We initially sterilized sun-dried green tea leaves that had not undergone fermentation using autoclaving. The samples were then inoculated with either dominant taxa (A. luchuensis, A. fumigatus, S. gallinarum, and S. kloosii), keystone taxa (T. ciferrii, H. coagulans, and E. sp.), or a combination of both. Fermentation was carried out at 37 °C for 20 days. Upon completion of fermentation, the content of methoxy-phenolic compounds was analyzed through GC-MS (Fig. S4). Figure 5a demonstrates a noticeable difference in the tea leaves before and after fermentation. Tea leaves fermented solely with dominant taxa showed black spots and turned black, while those fermented with keystone taxa turned dark brown with flask condensation. Co-fermented tea appeared black but was less dark and compact than dominant-only fermentation. This may be due to dominant taxa being Aspergillus spp., while keystone taxa are mainly bacteria and yeast, explaining the absence of black spots. The tea leaves fermented after co-inoculation with both dominant and keystone taxa exhibited a higher content of 1,2-dimethoxybenzene compared to those fermented with dominant taxa alone or keystone taxa alone, with the latter two showing comparable levels of 1,2-dimethoxybenzene. In contrast, regarding the content of 1,2,3-trimethoxybenzene, the levels in the tea leaves fermented with both dominant and keystone taxa, keystone taxa alone, and dominant taxa alone decreased sequentially (Fig. 5b). Validation of the role of keystone taxa. a Comparison of tea fermentation before and after using dominant taxa or keystone taxa alone, or co-inoculated with both dominant and keystone taxa; b Content of methoxy-phenolic compounds after 20 days of fermentation. The photographs presented in this figure were captured by the authors; the bar charts were generated using CNSknowall online tools (https://cnsknowall.com). The final composite layout was assembled using Adobe Illustrator (version 2024). Discussion Methoxy-phenolic compounds are key constituents responsible for the distinctive stale aroma of ripened Pu-erh tea. With the prolonged fermentation time, the stale aroma became more prominent, with increased purity and persistence. Furthermore, with the extended pile fermentation time, the OAV and content of methoxy-phenolic compounds, determined through both internal and external standard methods, exhibited significant increases. These findings exhibited strong congruence with our research group’s prior characterization of aroma compounds in ripened Pu-erh tea, establishing conclusive evidence that methoxy-phenolic compounds critically defined the stale aroma profile8. The formation of methoxy-phenolic compounds during pile fermentation is intrinsically mediated by microbial enzymatic activity. As pile fermentation progressed, the abundance of fungi (Aspergillaceae, Trichocomaceae) declined, while the abundance of bacteria (Staphylococcaceae, Micrococcaceae) increased. This observation aligned with previous studies15,25. The temporal shift in the composition of dominant microorganisms during pile fermentation indicated that the environmental conditions, particularly temperature and moisture content, favored fungal proliferation in the early stages, while promoting bacterial growth in the later stages. The current mechanistic consensus identifies a canonical pathway for the formation of methoxy-phenolic compounds through a two-step process. In the first step, tannins are degraded by tannase, an enzyme secreted by microorganisms, forming gallic acid. In the second step, gallic acid undergoes methylation via microbial O-methyltransferases to form methoxy-phenolic compounds. This methylation, known as microbial methylation, involves a methyl group from a tea methyl donor (S-Adenosylmethionine, SAM) replacing the hydrogen atom on the hydroxyl group of gallic acid, resulting in the formation of methoxy-phenolic compounds3,14,26. During the pile fermentation process of ripened Pu-erh tea, although the relative abundance of Aspergillus species decreased in the later stages, they remained the dominant microorganisms in terms of quantity throughout the entire process2,12,27–29. Given the substantial production of methoxy-phenolic compounds during fermentation, Aspergillus fungi were generally considered key contributors to their formation. However, through WGCNA and bivariate correlation network analysis, we identified that microorganisms positively correlated with methoxy-phenolic compounds were primarily bacteria (e.g., H. coagulan) and yeast (T. ciferrii), while those negatively correlated were predominantly fungi (A. luchuensis, A. tubingensis, A. piperis, and A. niger). The formation of methoxy-phenolic compounds requires the involvement of tannase and O-methyltransferase. While extensive research has confirmed the abundant presence of tannase in Aspergillus and Penicillium species, studies on microorganisms associated with O-methyltransferase remain limited26. Our study not only revealed a negative correlation between Aspergillus fungi and methoxy-phenolic compounds, and a positive correlation between bacteria and yeast, but also demonstrated that the co-fermentation of dominant and keystone microorganisms results in higher methoxy-phenolic compound production. We hypothesize that the abundant dominant microorganisms (Aspergillus) secrete tannase to degrade tannins into gallic acid, while the key microorganisms (bacteria and yeast) secrete O-methyltransferases to form methoxy-phenolic compounds. Some studies also suggested that the keystone taxa responsible for the production of methoxy-phenolic compounds during pile fermentation were not exclusively fungi. For example, the inoculation of A. niger spore suspension into sun-dried green tea for six days of solid-state fermentation did not yield methoxy-phenolic compounds30. Additionally, shotgun metagenomic and metabolomic analyses of the microbial community during the pile fermentation of ripened Pu-erh tea identified Pseudomonas and Achromobacter as the key microorganisms producing O-methyltransferases15. Ripened Pu-erh tea is valued for its distinctive stale aroma. Methoxy-phenolic compounds, key contributors to this aroma, are produced during pile fermentation; however, the specific microorganisms responsible for their synthesis remain unclear. Here, we identified the keystone taxa involved in the formation of stale aroma during the pile fermentation of ripened Pu-erh tea. First, we demonstrated that methoxy-phenolic compounds significantly contribute to the stale aroma, as confirmed by GC-MS. Second, through morphological analysis and metagenomic profiling, we identified the dominant taxa (A. luchuensis, A. fumigatus, S. gallinarum, and S. kloosii) during pile fermentation. Subsequently, by employing a metagenomic-based WGCNA in combination with bivariate correlation network analysis, we further identified the keystone taxa (T. ciferrii, H. coagulans, and E. sp.) involved. Finally, we validated these findings through solid-state fermentation simulation experiments. Our study identifies keystone taxa involved in methoxy-phenolic compound formation, yet the functional output of these microbial networks is primarily governed by abiotic factors in pile fermentation. Temperature drives microbial succession and enzyme activity, moisture regulates growth and metabolite diffusion, and pile heterogeneity creates micro-environments with varying oxygen and nutrient availability. These factors collectively shape microbial interactions and the efficiency of aroma synthesis. Thus, stabilizing these physical parameters is as critical as managing microbial composition in industrial practice. We further find that dominant microorganisms are not necessarily the main producers of methoxy-phenolic compounds. This has important industrial implications, as the prolonged, large-scale fermentation of ripe Pu-erh tea relies heavily on microbial activity. An imbalanced microbiota often results in erratic fermentation, including undesirable odors and inconsistent “stale” aroma quality. Therefore, we propose that targeted inoculation of key taxa—especially methoxy-phenolic producers—as defined starter cultures could enhance the controllability and stability of the desired aroma profile in industrial fermentation. Notwithstanding these insights, this study has limitations. Although we linked specific taxa to methoxy-phenolic formation and observed suggestive production trends, the detailed biochemical mechanisms—including the roles of key precursors like gallic acid and SAM, and the regulation of relevant enzymatic systems—remain unresolved. Additionally, while our validation experiments showed cooperative potential among taxa, more targeted designs are needed to dissect the precise metabolic pathways. Future work addressing these aspects will be essential to establish a verified biosynthetic pathway for these compounds in pile-fermented tea. Methods Sample collection All tea samples were collected from the pile fermentation of ripened Pu-erh tea in Mengku Town, Lincang City, Yunnan Province, China, during the period from November 2023 to January 2024. Our tea samples were classified into three categories: raw material for pile fermentation (RM), samples in the pile fermentation process (P1, P2, P3, P4, P5, and P6), and ripened Pu-erh tea product (RP). The entire pile-fermentation process typically lasts 45 to 60 days, and it is advisable to turn the pile over when the temperature at the core of the tea pile reaches 60 °C31. We defined six sampling time points during the pile fermentation (days 14, 21, 28, 34, 39, and 42), corresponding to the first through sixth turnings. The samples collected at the six time points were designated as P1, P2, P3, P4, P5, and P6. The tea fermentation pile (9 × 9 × 0.5 m3) was stratified into upper, middle, and lower layers. At each time point, composite samples from each layer were collected using a five-point diagonal sampling method (Fig. S5): the central point at the diagonal intersection with four peripheral points equidistant from the center (50 g per sampling point, 250 g total per layer). All samples were divided into two aliquots: one immediately transferred to the laboratory for microbial isolation, and the other stored at −80 °C for subsequent analysis. Sensory evaluation The aroma characteristics of the tea samples were evaluated by 10 sensory panelists (5 males and 5 females, aged between 20 and 30 years) based on national standards after smelling the samples32. All panelists had undergone formal sensory training prior to the assessment, and each possessed over one year of practical experience in professional sensory evaluation of tea. The panelists assessed the aroma purity and persistence of the tea samples using a 0–7 point scale, where 0 indicates no perception, 1–4 represents weak intensity, and 5–7 represents moderate to strong intensity, with 7 indicating the strongest perception. Each panelist also rated the overall acceptability of the tea samples. Ethical approval was obtained from the Institutional Review Board of the College of Food Science at Southwest University, under approval number HF20250602. Additionally, each panelist signed an informed consent statement before the sensory evaluation. Characterization of volatile compounds Volatile compounds were extracted using solid-phase microextraction (SPME) fibers (50/30 μm i.d., 2 cm length; Supelco, Darmstadt, Germany) and analyzed via GC-MS (GC-MS-QP2010nc plus system; Shimadzu). A 40 mL extraction vial was prepared by adding 1 g of tea powder and 2 g of NaCl, and thoroughly mixing with a vortex shaker. After adding a magnetic stir bar, 5 mL of deionized water at 100 °C and 50 μL of decyl acetate (100 ppm) were added. The vial was quickly sealed and placed in a magnetic stirrer with a temperature-controlled water bath set to 60 °C, rotating at 300 rpm. After a 5-min equilibration, the SPME fiber was inserted for extraction. The fiber had been preconditioned in the GC injection port at 250 °C for 30 min under a high-purity helium flow (1.0 mL/min) in splitless mode to remove any volatile residues. The extraction lasted for 55 min. After extraction, the fiber was removed and placed in the GC injection port for 5 min analysis, followed by GC-MS detection. The samples were analyzed on a DB-5MS capillary column (30 m × 0.25 mm i.d., 0.25 μm film thickness) with the injector temperature maintained at 250 °C. Helium was used as the carrier gas at a constant flow rate of 1.0 mL/min in split mode (split ratio 10:1). For mass spectrometric detection, the ion source temperature was set at 220 °C, and electron ionization (EI) mode was employed with an electron energy of 70 eV. The electron multiplier voltage was operated at 1780 V, and data were acquired in full scan mode at a rate of 5.27 scans per second, over a mass range of m/z 30–600. The initial column temperature was set at 50°C for 3 min, then increased to 125 °C at a rate of 2 °C/min and held for 3 min. The temperature was then ramped to 180 °C at a rate of 6 °C/min and held for 3 min; finally, the temperature was raised to 250 °C at a rate of 25 °C/min and maintained for 1 min. The qualitative identification of volatile compounds was initially performed using the NIST014S mass spectral library for preliminary identification, followed by confirmation based on the calculated retention index (RI)33. The RI of each compound was calculated using a C8-C30 n-alkane mixture under identical conditions, based on formula (1). In this, ti represents the retention time of the compound, while tn and tn+1 denote the retention times of n-carbon and (n + 1)-carbon n-alkane standards, respectively, with the condition tn < ti < tn+1. Volatile compounds were semi-quantified using decyl acetate as the internal standard, based on formula (2). Where Wi is the concentration of the compound, Ai is the compound’s peak area, As is the internal standard’s peak area, and Ws is the internal standard concentration. The OAV of volatile compounds, with a threshold of ≥1, is considered a reliable indicator of key aroma compounds. The OAV is calculated using formula (3). In this, Ci represents the concentration of the compound, and ODTi represents its olfactory threshold. The compound thresholds used were obtained from the references listed in Table S1. The concentration of methoxy-phenolic compounds was determined using their corresponding standard curves, as shown in Table S2. Isolation and identification of culturable microorganisms The microorganisms during the pile fermentation process were isolated by the serial dilution technique on Potato Dextrose Agar (PDA), which supports the fungal growth, and Beef Extract Peptone Agar (BEPA), which supports bacterial growth. While these media support a wide range of fungi and bacteria, they may not accommodate all species present in the complex fermentation environment. A sample suspension was prepared by adding a 1.5 g tea sample to 15 mL of distilled water in an Erlenmeyer flask, then shaking on a shaker at 200 rpm for 15 min. Each suspension was serially diluted from 10−1 to 10−6. 0.1 mL suspension was pipetted onto plates with PDA and BEPA media, spread with a glass spreader, and incubated at 28 °C for fungal and 37 °C for bacterial observation. Each colony that appeared on the plate was considered as one colony-forming unit (cfu)34. The colony count of microbial strains was determined following the method for total microbial colony count outlined in national standards35. The colonies obtained through gradient dilution were purified by streaking on agar plates to isolate single pure colonies. The microorganisms in our tea samples were initially identified through morphological analysis, followed by bacterial identification via 16S rDNA, and fungal and yeast identification through ITS sequencing. Morphological identification involved a combination of macroscopic observation and microscopic examination of colony morphology, hyphal structure, and conidia. DNA was extracted using the CTAB method for both ITS and 16S rDNA, followed by PCR amplification with specific primers. The specific primer sequences utilized for PCR amplification were as follows: for fungi, ITS4 (5′-TCCTCCGCTTATTGATATGC-3′) and ITS5 (5′-GGAAGTAAAAGTCGTAACAAGG-3′); for yeasts, ITS1 (5′-TCCGTAGGTGAACCTGCGG-3′) and ITS4 (5′-TCCTCCGCTTATTGATATGC-3′); and for bacteria, F27 (5′-AGAGTTTGACCTGGCTCAG-3′) and R1492 (5′-TACGGTTCCTTGTTACGACTT-3′), which target the 16S rRNA36–38. The PCR products were then purified and sent to Shanghai Sangon Biotech Co., Ltd. for sequencing and analysis. Metagenome sequencing and analysis Genomic DNA was extracted from the samples using the Mag-Bind® Soil DNA Kit (Omega Bio-tek, USA). Following extraction, the concentration and purity of the DNA were assessed, and its integrity was verified by 1% agarose gel electrophoresis. A Covaris M220 (Gene Company Limited, China) was used to fragment DNA extracts, producing fragments that were roughly 350 bp in size on average. NEXTFLEX Rapid DNA-Seq (Bioo Scientific, Austin, TX, USA) was used to create a paired-end library, which was then sequenced on an Illumina NovaSeq 6000 (Illumina Inc., San Diego, CA, USA). The sequenced reads were subjected to quality control and adapter trimming using Fastp v0.20.0. After quality control, we obtained 98.2 Gb of metagenomic clean reads, with an average data volume of 6.55 Gb per sample (ranging from 5.88 Gb to 7.08 Gb). MEGAHIT v1.1.2 was used to assemble the data, and Prodigal v2.6.3 was used to predict the ORFs within the contigs from the assembly. Genes that were translated into amino acid sequences had nucleotide lengths of at least 100 base pairs. A non-redundant gene set was then obtained by clustering the predicted gene sequences using CD-HIT v4.6.1. The abundance of each gene was determined by aligning and annotating the non-redundant gene set against the NR database using DIAMOND software. The sequencing and related work were conducted at Majorbio Bio-Pharm Technology Co., Ltd. (Shanghai, China). WGCNA and bivariate correlation network analysis β-Diversity reveals the microbial dissimilarity among the five tea samples, while α-diversity provides insights into the richness and evenness of the microbial communities within each tea sample. To identify keystone microorganisms associated with methoxy-phenolic compounds, two network analysis methods were employed. The first method was WGCNA analysis. Gene abundance data were first normalized via the variance-stabilizing transformation (VST) implemented in the DESeq2 R package, after which outliers in the transformed dataset were further examined and addressed. Next, a soft threshold of 7 was selected based on the scale-free topology fit and mean connectivity curves to assess whether two genes exhibited similar expression patterns (Fig. S6). Genes were then classified into modules based on their weighted correlation coefficients, with genes showing similar expression patterns grouped together. Each module was rigorously analyzed, and Pearson’s rank correlation analysis was used to characterize inter-module relationships and correlations between modules and methoxy-phenolic compound levels, thereby identifying biologically significant hub modules. Following hub module identification, critical keystone species are discerned through systematic evaluation of two correlation metrics: the module membership (MM) representing the correlation between species expression levels and module eigengenes, and the gene significance (GS) quantifying species-trait correlations. This comprehensive analytical workflow was implemented using the WGCNA package (Version 3.3.1) within R. The second method was Bivariate correlation network analysis. Pearson’s rank correlation analysis was applied to characterize the relationships between microorganisms and methoxy-phenolic compound levels. Statistical significance was assessed using false discovery rate (FDR) correction for multiple comparisons, with a correlation coefficient ≥ 0.7 and an FDR-adjusted p < 0.05 set as the inclusion thresholds. This comprehensive analytical workflow was implemented in R, with visualization of the co-occurrence network using NetworkX (Version 1.11) and Cytoscape (v3.9.1). In this network, node size reflects degree centrality, and edge width represents the strength of correlation. Fermentation experiments Following the identification of dominant and keystone taxa during the pile fermentation, a validation fermentation experiment was systematically conducted. In a 500 mL conical flask, 80 g of sun-dried raw tea material was supplemented with 30 mL of distilled water, followed by autoclave sterilization. In order to validate microbial functions, a pure culture fermentation approach was employed, which involved the aseptic introduction of one or more exogenous microorganisms into the sterilized substrate for fermentation, thereby ensuring that the fermentation environment consisted solely of purified microbial strains39,40. Dominant taxa referred to the top four microorganisms ranked by relative abundance during the pile fermentation process (Aspergillus luchuensis, A. fumigatus, Staphylococcus gallinarum, and S. kloosii), while keystone taxa were the three microorganisms identified through WGCNA and bivariate correlation network analysis, which were associated with the production of methoxy-phenolic compounds during the pile fermentation (Trichomonascus ciferrii, Heyndrickxia coagulans and Enterococcus sp.). The seven microbial strains were individually subcultured onto fresh solid agar plates for activation, after which the cells were collected and washed with sterile saline solution, inoculated into a tea broth medium, and incubated with shaking at an appropriate temperature to prepare single-strain suspensions at a concentration of approximately 1 × 10⁸ CFU/mL for subsequent use. Subsequently, equal volumes of the individual suspensions from the four dominant taxa were pooled and inoculated into one batch of tea leaves, while an equal volume of the keystone taxa suspension was inoculated into a second batch. A third batch received equal-volume inoculations of both the pooled dominant taxa suspension and the keystone taxa suspension. The preparation of the tea leaf medium involved adding 100 g of tea to 2 L of water at 100 °C and steeping for 30 min. After filtration through four layers of gauze, the medium was sterilized at 121 °C for 20 min. Subsequently, the microorganisms from the three media were inoculated into the sterilized sun-dried raw tea and incubated in an incubator under controlled conditions (30 °C) (Fig. S4). After 20 days of fermentation, methoxy-phenolic compounds were quantitatively analyzed via Gas Chromatography-Flame Ionization Detector (GC-FID TRACE 1300, Thermo Fisher, Waltham, USA) to evaluate their accumulation patterns. Following fermentation, the tea leaves were freeze-dried and ground into powder. Methoxy-phenolic compounds were extracted using the method described in Section 2.2, with ethyl decanoate omitted, as the external standard method was applied directly. The extracted volatile compounds and standard solutions were analyzed using a gas chromatography-flame ionization detector GC-FID, equipped with a non-polar DB-5 capillary column (30 m × 0.25 mm × 0.25 μm). Hydrogen and air served as the carrier gases at flow rates of 30.0 mL/min and 300.0 mL/min, respectively, with a make-up gas flow of 30.0 mL/min. The injector and detector temperatures were set at 250 °C and 280 °C, respectively, under splitless injection mode. The GC temperature program was as follows: initial temperature 40 °C (held for 2 min), increased at 2 °C/min to 80 °C (held 2 min), then at 5 °C/min to 150 °C (held 1 min), followed by 4 °C/min to 180 °C (held 1 min), and finally 10 °C/min to 260 °C (held 5 min). The concentration of methoxy-phenolic compounds was determined using their corresponding standard curves, as shown in Table S2. Statistical analyses All experiments on volatile compound analysis, metagenomics, and solid-state fermentation were conducted in triplicate. For statistical analysis, comparisons between two groups were performed using Student’s t test, while comparisons across more than two groups were assessed by one-way ANOVA, followed by pairwise post-hoc tests with Bonferroni correction. All analyses were carried out using IBM SPSS Statistics (version 27.0.1). To determine the taxonomic affiliations, the obtained DNA sequences were compared against known sequences in the NCBI database, followed by similarity analysis and phylogenetic tree construction using Clustal X 1.83 and MEGA to confirm their taxonomic positions. The α-diversity indices were calculated to assess the microbial diversity within the samples using the Mothur (version v.1.30.2) and R software (version 3.3.1). For β-diversity, the Bray-Curtis dissimilarity metric was employed to evaluate the microbial diversity between samples, based on distance matrices computed using R (version 3.3.1) and Qiime (version 1.9.1). The bivariate correlation network analysis was conducted using NetworkX (version 1.11). Data visualization was performed using CNSknowall (https://cnsknowall.com; for radar, PCA, heatmap, box, and Venn plots), OmicStudio (https://www.omicstudio.cn/tool; for bar, line, and column charts), and Majorbio Cloud (www.majorbio.com; for hierarchical clustering and network diagrams). All figures were finalized in Adobe Illustrator (version 2024). Supplementary information Supplementary information. Supplementary Table.S5. Acknowledgements This work was supported by the National Natural Science Foundation of China (32272764), the Yunnan Provincial Department of Education (2024J1110), “Xinglin Rencai” Program of Lincang City-Special Project for Young Scientific and Technological Talents, the Open Fund of Yunnan Key Laboratory of Tea Germplasm Conservation and Utilization in the Lancang River Basin, and the Chongqing Graduate Student Research Innovation Project (CYB23127). Author contributions Yiwei Weng: Conceptualization; Data curation; Funding acquisition; Validation; Writing-original draft; Writing review & editing. Shiqiang He: Data curation; Validation; Visualization; Writing-review & editing. Zhengfei Luo: Formal analysis; Funding acquisition; Investigation. Jin Sun: Investigation; Validation; Visualization. Qinghua Cheng: Software; Supervision; Methodology. Yingjuan Chen: Conceptualization; Funding acquisition; Writing review & editing. Huarong Tong: Conceptualization; Funding acquisition; Supervision; Writing review & editing. Data availability The data that support the findings of this study are available from the corresponding author, Huarong Tong, upon reasonable request. Code availability The underlying code for this study is not publicly available but may be made available to qualified researchers on reasonable request from the corresponding author. Competing interests The authors declare no competing interests. Footnotes Publisher’s note Springer Nature remains neutral with regard to jurisdictional claims in published maps and institutional affiliations. Contributor Information Yingjuan Chen, Email: cyj8@swu.edu.cn. Huarong Tong, Email: thuarong@126.com. Supplementary information The online version contains supplementary material available at 10.1038/s41538-026-00794-8. References Associated Data Supplementary Materials Supplementary information. Supplementary Table.S5. Data Availability Statement The data that support the findings of this study are available from the corresponding author, Huarong Tong, upon reasonable request. The underlying code for this study is not publicly available but may be made available to qualified researchers on reasonable request from the corresponding author.