https://orcid.org/0000-0002-6606-6884
Abstract
Kombucha is a unique, naturally fermented sweetened tea produced for thousands of years, relying on a symbiotic microbiota in a floating biofilm, used for successive fermentations. The microbial communities consist of yeast and bacteria species, distributed across two phases: the liquid and the biofilm fractions. In the fermentation of kombucha, various starters of different shapes and origins are used, and there are multiple brewing practices. By metabarcoding, we explored here the consortia and their evolution from a collection of 23 starters coming from various origins summarizing the diversity of kombucha fermentation processes. A core microbiota of yeast and bacteria has been identified in these diverse kombucha symbiotic consortia, revealing consistent core taxa across symbiotic consortium of bacteria and yeasts from different starters. The common core consists of five taxa: two yeast species from the Brettanomyces genus (B. bruxellensis and B. anomalus) and bacterial taxa Komagataeibacter, Lactobacillus, and Acetobacteraceae, including the Acetobacter genus. The distribution of yeast and bacteria core taxa differs between the liquid and biofilm fractions, as well as between the “mother” and “daughter” biofilms used in successive fermentations. In terms of microbial composition, the diversity is relatively low, with only a few accessory taxa identified. Overall, our study provides a deeper understanding of the core and accessory taxa involved in kombucha fermentation.
Introduction
Kombucha fermentation involves multi-species microbial communities known as symbiotic consortium of bacteria and yeast (SCOBYs), which provide a valuable model for understanding how microbial communities are shaped through interactions between microorganisms and their environment (Wolfe and Dutton 2015, May et al. 2019, Huang et al. 2022, Landis et al. 2022). Kombucha, a popular effervescent beverage, is produced by fermenting a sweetened tea infusion over a period of 10–20 days (Jayabalan et al. 2014, Coton et al. 2017, de Miranda et al. 2022). It has low residual sugar and alcohol content as well as a mildly acidic taste. A key feature of kombucha SCOBYs is the division of microorganisms into two fractions: the liquid fraction, which becomes the beverage, and the solid fraction, a translucent biofilm. The biofilm, synthesized during fermentation, consists of reticulated cellulose and embedded microbial cells (Savary et al. 2021, Tran et al. 2021, Hamed et al. 2022, Wang et al. 2022).
Unlike many fermented products, where fermentation can be initiated by the spontaneous colonization of environmental microorganisms (Callanan et al. 2021), kombucha fermentation relies entirely on human-driven conservation and propagation of SCOBYs that are passed from one fermentation cycle to the next as mandatory starters. The traditional method of initiating kombucha fermentation involves a bipartite starter, made up of both the liquid and biofilm fractions from a previous batch (Jayabalan et al. 2014). The biofilm fraction is known as the “mother.” During fermentation, a new cellulosic biofilm forms at the air–liquid interface, referred to as the “daughter.” For the next fermentation cycle, either biofilm can be used as part of the starter: when the original mother biofilm is reused, the process is called backslopping; when the daughter biofilm is used, it is known as repitching (Leroy and De Vuyst 2004).
The liquid fraction alone can also serve as a starter to regenerate bipartite SCOBYs. This diversity in starter forms enables the process to continue indefinitely and facilitates the accumulation of multiple SCOBYs across successive cycles. Kombucha enthusiasts often share and distribute SCOBYs, further promoting the practice. As a result, kombucha SCOBYs are now distributed worldwide, used to ferment under a wide range of conditions, including different tea types, sugar types, and concentrations (Laavanya et al. 2021, Cohen et al. 2023, Phung et al. 2023). Historically, they have been used in households to produce small batches of the beverage (1–2 l), but in recent decades, their use has expanded to large-scale production, with brewers now utilizing fermentation tanks ranging from 1000 to 2000 l (Coton et al. 2017, Harrison and Curtin 2021).
Although many independent studies have described the yeast and bacteria taxa present in kombucha SCOBYs (Harrison and Curtin 2021, Laavanya et al. 2021, Huang et al. 2022, and references therein), a comprehensive study defining the core and accessory microbiota, and how they shape SCOBY diversity, remains lacking. In this study, we collected 23 starters from diverse origins, representing the variety of kombucha fermentation and propagation practices. These starters were used to generate SCOBYs, which we then investigated and compared for their microbial compositions to identify patterns of variation. The starters were fermented under identical conditions, and the resulting SCOBYs were propagated through the traditional repitching method. Using a culture-independent metabarcoding approach, we analyzed the yeast and bacteria compositions across successive cycles of propagation. Our results revealed a core set of yeast and bacteria taxa present across this diverse collection of SCOBYs. Furthermore, both core and accessory taxa were found to play a key role in shaping the microbial diversity within the SCOBYs.
Material and methods
The SCOBY collection
We gathered a collection of 23 kombucha starters. The various starters were either purchased on French commercial platforms or obtained from private households or kombucha brewers located in France. In order to generate a collection of comparable SCOBYs for further propagation by consecutive repitching cycles, and as the starters display disparate stages and shapes (Table S1), these latter were handled according to one of the three following protocols (Fig. 1):
When the initial starters had both the biofilm and liquid fractions available, they were subjected to a single fermentation step to be ready for further propagation.
In order to generate a SCOBY from unpasteurized marketed kombucha beverages sold by various brands, we left the purchased bottles to settle for 2 h after opening and release of gas pressure. We then inoculate five volumes of a sweet tea infusion with one volume of the bottle bottom phase and incubate the culture for 20–30 days at 27°C. When this step resulted in the formation of a floating biofilm, it was used to re-inoculate a fresh sweet tea infusion, along with a fraction of the liquid phase. After 14 days of incubation at 27°C, these cultures resulted in a de novo biofilm suitable for further propagation, along with a fraction of the liquid phase.
When the initial starter was a concentrated microbial suspension, it was subjected to a single fermentation step in five to ten volumes of sweet tea infusion. After 14–21 days of incubation at 27°C, the de novo biofilm was ready for further propagation, along with a fraction of the fermented liquid.
Generation of the SCOBYs and the fermentation series and collection of the samples analyzed by metabarcoding. Starters are sorted in the table at left, according to the presence or not of a cellulosic biofilm. They are denominated by acronyms in line with the provider names, letter colors indicate the technological origin of the starters (red: industrial brewery; blue: household; black: unpasteurized commercial beverage; gold: commercial starter). The steps for generating comparable initial SCOBYs and the three sequential repitchings to establish the fermentation series are represented. The samples are denominated (red labels) according to the fraction from which the cells were harvested (fermented liquid, L; daughter biofilm, D; mother biofilm, M) and according to the repitching cycles 1–3..
Routine repitching of the fermentation series and sampling
A volume of 20 l of tea infusion was routinely prepared in a 30 l heating container. A mix of clean tap water and demineralized water (1:9 v/v) was brought to the boil and maintained for 10 min. A total of 120 g of green tea leaves (0.6% w/v, Fannings vert, Jardins de GAÏA) were added and let to steep for 15 min. After removal of the tea leaves, 800 g of sugarcane (4% w/v, ActiBio) was added and stirred until dissolution. Immediately after dissolution, 1.1 l fractions of the hot sweet tea infusion were distributed in 1.7 l glass containers that were previously rinsed twice with simmering water and covered by a paper towel fixed with a rubber band. Once infusion had cooled to room temperature, handlings were carried out in the sterile area surrounding the flame of a Bunsen burner. Inoculation was performed using the traditional repitching method for kombucha brewing, with a starter consisting of 100 ml of the liquid fraction and the intact “daughter” biofilm from a previous kombucha fermentation. Fermentations were carried out at 27°C for 14 days under static conditions, and the starter was re-inoculated every 14 days into a fresh sweetened tea infusion.
Samples were taken on day 14: 50 ml of the liquid phase was collected after gentle homogenization, and 5 g of the solid phase (mother or daughter from the last repitching) were collected by cutting a 2-cm wide band from the biofilm on the edge of the disc using a sterile blade. To release the cells from the biofilm, the pieces of biofilm were treated with 2 ml of cellulase solution (240 mg/ml, Sigma-Aldrich, Denmark) for 4 h at 37°C. The cells of the liquid or the solid phase were passed through a 100 µm cell strainer (Corning, USA) to remove remaining small cellulose aggregates. The filtrates were centrifuged at 4500 g for 10 min at 4°C, and the drained cell pellets were kept frozen at −20°C until performing the DNA extraction. A volume of 10 ml of the supernatant of the liquid phase was filtered via a 0.45 µm sterile membrane and preserved at –20°C until performing the biochemical dosages.
Biochemical composition
The liquid fraction supernatants were defrosted at ambient temperature for 2 h. pH was measured with a pH meter (Bio-Rad). Ethanol contents were determined using the K-ETOH enzymatic kit (Megazyme, LIBIOS, France). Sugar contents (sucrose, fructose, and glucose) were determined using the K-SUFRG enzymatic kit (Megazyme, LIBIOS, France). Acetic acid contents were determined using the K-ACETRM enzymatic kit (Megazyme, LIBIOS, France). Sample dilutions and concentration calculations were done following the manufacturer’s instructions, and all the analyses were carried out in triplicate.
DNA extraction
The Qiagen DNeasy PowerSoil Kit (Qiagen, France) was used to extract the total genomic DNA according to the manufacturer’s instructions with two major modifications. (i) The Power Beads were replaced by 25 mg of 425–600 µm glass beads (Sigma, USA), and (ii) the lysing procedure was split into three steps in order to perform sequential lysis as follows. A fraction of only 40% of the volume of the lysis buffer was added to the lysing tube, supplied with 1/100 (w/v) of sodium dodecyl sulfate. A first homogenization treatment was applied for 30 s at a 6 m/s speed with a FastPrep instrument (MP Biomedicals). The lysing tube was centrifuged for 2 min at 15 000 g, the supernatant was collected and kept on ice. In a second step, a fraction of 30% of the volume of the original lysis buffer was added to the lysing tube, and a new homogenization treatment, centrifugation, and supernatant collection were performed in the same conditions. The second step was repeated: in total, three serial homogenization steps were performed to ensure breakdown of the most solid cells and preservation of the DNA extracted from the most delicate cells. The pooled supernatants of each step were subjected to the next steps of the supplier protocol for DNA purification.
Metabarcoding library preparation and sequencing
For the analysis of fungal communities, the D2 hypervariable region of the 28S rRNA gene was amplified using forward and reverse primers based on NL-D2 (this study) and NL-4 (O'Donnell and Cigelnik 1997). For the analysis of bacterial communities, the 16S V3–V4 region was amplified using forward and reverse primers based on Pro341F and Pro805R (Takahashi et al. 2014). All locus-specific Polymerase Chain Reaction (PCR) primers were extended with overhangs suitable for the sequencing platform; primer sequences are listed in Table S2.
The multiplexed Illumina libraries were prepared following a two-step PCR approach: a first PCR using the locus-specific primers (PCR1, with 30 cycles) and a second PCR (PCR2) for the incorporation of Illumina dual-indexed adapters.
The PCR1 reaction was performed using 50 ng of DNA template and 1.25 units of DreamTaq polymerase (Thermo Scientific) in a final volume of 50 µl. For each forward or reverse primer, an equimolar mix was added to the PCR mix to a final concentration of 0.5 µM. The amplification was performed with first a denaturation step at 95°C for 2 min followed by 7 cycles of [denaturation (94°C, 30 s), annealing (55°C, 15 s), and extension (72°C, 40 s)] and 23 cycles of [denaturation (94°C, 30 s), annealing (62°C, 20 s), and extension (72°C, 40 s)]. Bead purification was carried out after PCR1. PCR2 was performed at the sequencing platform.
The fungal and bacterial libraries were either pooled or sequenced separately according to the samples (see Table S6) using MiSeq 2 × 300 bp.
Bioinformatic and diversity analyses
Sequence processing was performed using the FROGS pipelines as developed by Escudié et al. (2018) under Galaxy (Galaxy Community 2024) on the migale platform (https://galaxy.migale.inra.fr/). Paired reads were assembled by VSEARCH with a rate of mismatches in overlapping regions set at 10% (Magoč and Salzberg 2011). Primers were removed using Cutadapt 1.18 (Martin 2011). Sequences were clustered into operational taxonomic units (OTUs) using the Swarm algorithm (Mahé et al. 2021) with an aggregation distance of 3. Chimera removal was conducted using VSEARCH with the de novo UCHIME method (Edgar et al. 2011). OTUs were filtered on cluster size with a threshold of 0.00005. The taxonomic assignment was performed using BLASTn as a similarity search tool. For the 16S OTUs, the taxonomic assignment was performed using the non-redundant SILVA nucleotide database, version 138 (Quast et al. 2013). Multi-affiliations of OTUs were dealt with by assigning the lowest common taxonomy level to multi-affiliated OTUs. For the 28S OTUs, the taxonomic assignment was performed using a custom database of 527 Ascomycete and Basidiomycete species sequences established according to Kurtzman and Robnett (1998) (File S1). To overcome the limitation caused by the low number of taxa in this database, the taxonomic assignments were manually refined as follows: if the identity score given by BLASTn was lower than 97%, the OTU consensus sequence was subjected to BLASTn search against NCBI Standard and rRNA/ITS databases (release: January 2024), and if a better BLAST result was obtained, the corresponding taxa were used for assignment. As an example, two OTUs were re-assigned to Starmerella davenportii instead of S. stellata. This procedure was also applied to refine the taxonomy of Pichia clades Pc1 to Pc5. Data were normalized based on the sample that had the lowest number of sequences using the rarefy_even_depth function of the R (v. 4.1.0) phyloseq package (v. 1.24.2) (McMurdie and Holmes 2013). Diversity indexes and multidimensional scaling (MDS) were computed using the R (v. 4.1.0) phyloseq package (v. 1.24.2) (McMurdie and Holmes 2013).
Results
The collection of samples and their diversity
To create a SCOBY collection, we gathered 23 starters from various sources and types (Table S1). These included two samples from industrial production tanks (BB1 and MK), three household “ready-to-use” starters (AK, HK, and PK), ten commercial unpasteurized beverages (BB2, BK, CP, EB, FZ, KK, OK, RI, UR, and VK), and eight commercial “ready-to-use” starters (DK, FE1, FE2, KF1, KF2, KOK, LO, and LK). Each starter, based on its shape (Table S1), was cultivated in a sweet tea infusion under appropriate conditions to generate an initial SCOBY, consisting of a fresh biofilm and part of the fermentation liquid (Fig. 1, see the “Materials and methods” section). This SCOBY was then used to initiate a series of three consecutive fermentation cycles, referred to as a “fermentation series.” Of the 23 fermentation series, 18 met the following criteria for further analysis (Table S1): (i) The SCOBY was divided between the liquid and solid fractions required for traditional repitchings, (ii) a daughter biofilm was produced at the end of each fermentation cycle, (iii) no visible mold contamination occurred during repitching, and (iv) the biochemical composition of the liquid fraction met kombucha beverage standards [pH below 3.5, sucrose consumption, ethanol content <1% (g/v), and acetate production] (Fig. S1).
During the fermentation series, intact cells were collected from the liquid phase at the end of each fermentation cycle, as well as from the mother and daughter biofilms of the third cycle. This resulted in a total of 90 samples, with five samples per fermentation series. For each sample, total DNA was extracted, and two amplicon libraries were generated: one targeting the D2 hypervariable region of the 28S rRNA gene of fungi and the other targeting the V3–V4 region of the 16S rRNA gene of bacteria. After sequencing and processing the reads with FROGS (see the “Materials and methods” section), the fungal library yielded an average of 100 078 sequences per sample across 76 samples. These sequences were clustered into 99 OTUs, primarily within Ascomycota (98 OTUs) and Basidiomycota (1 OTU). These OTUs were further assigned to 11 yeast genera, representing 21 species (Table S3). After rarefaction to equalize sample depth, 12 051 sequences per sample were retained for analysis. The bacterial library showed an average of 42 726 sequences per sample across 69 samples, which were clustered into OTUs corresponding to 25 bacterial families or genera (Table S5). Following rarefaction, 7457 sequences per sample were retained for further analysis.
The diversity in yeast compositions underscores the presence of core yeast taxa
The yeast composition, as determined by the D2 amplicon library, was analyzed for 76 samples, including 46 from the liquid fractions and 30 from the biofilm fractions. The detected OTUs were identified at both the genus and species levels, with taxon abundance distributions represented (Fig. S2) and taxonomic profiles established (Fig. 2). At the genus level, 11 genera were identified, with Brettanomyces being the dominant genus, accounting for an average of 83.6% of the sequences across samples, ranging from 9% to over 99%, depending on the sample (Fig. S2A). Brettanomyces was the only genus present in all samples (Fig. 2A). In addition to Brettanomyces, five other genera were commonly represented. Two genera were shared by multiple samples from diverse original starters: Pichia, found in 24 samples from 12 different starters, representing an average of 21.5% of the sequences, and Zygosaccharomyces, found in 20 samples from 12 starters, representing an average of 4.5% of the sequences. Three other genera were found in variable proportions across five to six samples from different starters: Kregervanrija (five samples from two starters), Starmerella (six samples from four starters), and Saccharomyces (six samples from various starters). Additionally, sequences from Hanseniaspora, Kluyveromyces, Malassezia, Torulaspora, and Schizosaccharomyces were detected in very low proportions in only a few samples. The alpha diversity, measured by the “Observed” index, ranged from 1 to 11 taxa per sample, with a mean of 4.64. The “Gini-Simpson” index ranged from 0 to 0.66, with a mean of 0.17 and a median of 0.048 (Fig. S3A), indicating a generally low diversity in the yeast compositions. This low diversity is largely due to the dominance of Brettanomyces, which was the predominant genus in 66 samples, including three liquid samples from the KOK and KF1 starters, and seven biofilm samples.
Yeast taxonomic profiles represented by the normalized sequence count of taxa in each sample. Compositions were resolved at the genus level (A) or at the species level (B) and are represented for 46 samples from the liquid fractions across the three cycles of the fermentation series and in 30 samples from the mother and daughter biofilms of cycle 3. Sequence counts are normalized based on the sample that had the lowest number of sequences. Taxa represented by more than 0.2% of the total sequences are distinguished by the color scheme on the right of the plots. Taxa represented by <0.2% of the total sequences were merged in the “Other” categories. Empty columns correspond to the absence of data for the corresponding samples. Pc1 to Pc5 correspond to the clades defined from the analyses of sequences affiliated to the Pichia genus. Samples were sorted according to (i) the starter name, (ii) the fermentation cycle, (iii) the fraction, and (iv) the profile similarities in the major taxa resolved at the species level (visual assessment). The colored circles beneath the starter names represent their technological origin (red: industrial brewery; blue: household; black: unpasteurized commercial beverage; gold: commercial starter).
At the species level (Fig. S2B and Fig. 2B), the number of taxa increased to 18, with only the three most abundant genera represented by multiple species. This higher taxonomic resolution significantly increased the mean diversity indices (Fig. S3B): the mean number of taxa per sample rose to 6.93, and the mean Gini–Simpson index increased to 0.32 (P-values < 1.10−5, Wilcoxon test). In all samples, the Brettanomyces genus was represented by two species, B. anomalus and B. bruxellensis, which appeared in varying proportions. Among the 66 samples where Brettanomyces was the dominant genus, the two species had similar proportions in only six samples. Brettanomyces anomalus predominated in 27 samples, while B. bruxellensis was dominant in 33. In the remaining 10 samples, where Brettanomyces was not the major genus, B. anomalus was more prevalent in five samples, while B. bruxellensis dominated the other five (Figs 2B and 3). The genus Zygosaccharomyces was represented by two dominant species, each specific to certain starters: Z. lentus was found in samples from the KF1, KF2, and RI starters, while Z. bailii appeared in samples from the AK, KOK, FE2, and VK starters. It is worth noting the absence of a clear relationship between yeast content diversity and the technological origin of the initial starters. However, the limited sample size poses a constraint in establishing such a link.
Similarities in the yeast compositions across successive repitchings. (A) The abundances of yeast taxa in the samples along Axis 1 are represented by the normalized sequence counts for B. anomalus (yellow, maximum = 12 006 sequences), B. bruxellensis (red, maximum = 11 959 sequences), aggregated accessory genera (black, maximum = 9172 sequences). (B) and (C) Spatial ordination along Axis 1 and Axis 2 based on Bray–Curtis dissimilarity matrices computed on the sample compositions in yeast species. For better visibility, the samples were displayed separately: (B) samples from AK, BB1, BB2, BK, EB, FE1, FE2, FZ, HK, UR, and VK starters, and (C) samples from CP, KF1, KF2, KOK, LK, MK, and RI starters. The yellow and red dotted ellipses encircle the samples whose compositions are dominated by B. anomalus or B. bruxellensis, respectively. Samples are colored according to the original starter name as defined by the color schemes on the right, and the marker shape indicates the fermentation cycle. The colored lines and polygons connect the samples from the same starter. The colored circles next to the starter names represent their technological origin (red: industrial brewery; blue: household; black: unpasteurized commercial beverage; gold: commercial starter).
To resolve the distribution of the 33 OTUs affiliated with the Pichia genus, a manual curation step was necessary. Alignment of the corresponding sequences allowed them to be grouped into five distinct clades, labeled Pc1 to Pc5. Interestingly, each sample predominantly contained OTUs from a single Pichia clade, and these clades were largely specific to certain original starters. For example, in the BB2, HK, and FE1 starters, Pichia was represented exclusively by sequences from clade Pc1; in the KOK and LK starters, it was represented by sequences from clade Pc2; and in the FE1 and BB1 starters, it was represented by sequences from clade Pc5. We then compared these OTU sequences to annotated yeast sequences in the GenBank database using BLAST (Table S4). This comparison revealed that OTUs from Pc1 corresponded to Candida californica, OTUs from Pc2 to Pichia deserticola, OTUs from Pc4 to Pichia chibodasensis, and OTUs from Pc5 to Pichia membranifaciens. OTUs from Pc3 were rare, mostly found in the EB starters, where they accounted for 0.01% to 1% of the sequences. These OTUs appear to correspond to an uncharacterized Pichia species or a candidate for a novel species. The limited representation of Pichia in current databases has been previously noted (Gallegos-Casillas et al. 2024).
Yeast species exhibit different patterns of variation across fermentation cycles
To examine how yeast species compositions evolve over successive repitchings, we focused on the liquid fractions collected during consecutive fermentation cycles. Microbial composition data were available for three cycles from 10 starters, two cycles from FZ, HK, KOK, and UR, and for the first and last cycles from AK, CP, KF2, and VK. Bray–Curtis similarities were computed to compare the yeast compositions, and these were visualized in an MDS plot (Fig. 3).
Yeast compositions remained closely related across repitchings from six initial starters, regardless of whether the compositions were dominated by B. bruxellensis (FE1, HK, and VK) or B. anomala (BK, FZ, and UR). For samples from five other starters, dominated by either B. bruxellensis (AK, BB1, and EB) or B. anomala (BB2 and FE2), the compositions showed limited similarity, indicating fluctuations across cycles. However, we cannot rule out the possibility that the observed changes are simply due to the absence of replicates. For the SCOBYs from the seven remaining starters, one of their samples was notably distinct, indicating significant compositional variation across repitchings. For two starters (CP and MK), this variation was characterized by an increasing proportion of B. anomalus relative to B. bruxellensis. For five other starters (KF1, KF2, KOK, LK, and RI), samples showed higher proportions of one of the non-Brettanomyces taxa, with most of them exhibiting a decrease in the corresponding taxa proportions over repitchings, except for samples of KF1.
The Pichia genus is notably abundant in the mother biofilms
The solid fractions, represented by the two cellulosic biofilms (mother and daughter) formed at the end of fermentation, are a defining feature of kombucha. To compare their microbial compositions with those of the corresponding liquid fractions, cells from the mother and daughter biofilms of the third fermentation cycle in each series were harvested after enzymatic digestion of the cellulose network. These cells were then subjected to DNA extraction and metabarcoding analysis.
For the samples of 13 starters, fungal libraries were available to compare the compositions of the three fractions (liquid L3, mother, and daughter) within the same fermentation cycle (Fig. 2). At the genus level, two distinct patterns emerged. For eight of the starters, the microbial compositions were quite similar across the three fractions, as indicated by short distances on the MDS plot based on Bray–Curtis similarity (Fig. 4, upper panel). In contrast, for the five other starters, the composition of the mother biofilm was significantly different from the liquid and daughter fractions, as seen in the longer distances on the plot (Fig. 4, lower panel).
Similarities in the yeast compositions in the different SCOBY fractions of the third fermentation cycle. The spatial ordination along Axis 1 and Axis 2 was based on Bray–Curtis dissimilarity matrices computed on the sample compositions in yeast genera. For better visibility, the samples from BB1, BB2, FE1, HK, and RI starters are displayed on the separate lower panel. Samples are colored according to the original starter name as defined by the color schemes on the right, and the marker shape indicates the SCOBY fraction (L: liquid; D: daughter biofilm; M: mother biofilm). The colored lines and polygons connect the samples from the same starter. The colored circles next to the starter names represent their technological origin (red: industrial brewery; blue: household; black: unpasteurized commercial beverage; gold: commercial starter).
This difference was mainly due to the dominance of the Pichia genus in the mother biofilms from BB2 and HK, its high abundance in BB1 and FE1 mother biofilms, and the prevalence of Starmerella in the RI mother biofilm. The mother biofilms from KOK and FE2 (which were excluded from these comparisons due to missing data for the other fractions) also exhibited high proportions of Pichia (Fig. 2A).
Bacterial composition diversity and its evolution
The bacterial compositions of 42 liquid fractions across three successive repitchings, as well as 27 biofilm fractions from the third fermentation cycle, were analyzed and resolved to the family or genus level (Fig. 5). Six taxa, including five genera and one family, are predominant and together account for a mean of 98.3% of the sequences. Among these, four taxa are highly shared across samples (Fig. 5C). The Komagataeibacter genus is the most abundant, present in all samples with sequence abundances ranging from 0.6% to 99.4%, with a mean of 45.2% (Fig. 5A). In all but one sample, the next most abundant genus is Lactobacillus, averaging 20% of the sequences, with abundance ranging from 0.001% to 94%. All samples also contain sequences affiliated with the Acetobacteraceae family, which make up an average of 8.2% of the sequences, with a range from 0.1% to 54%.
Bacteria contents and taxonomic profiles of 69 samples. Taxa were revealed by the V3–V4 libraries resolved at the genus level. Sequence counts were normalized based on the sample that had the lowest number of sequences. Taxa represented by more than 0.2% of the total sequences are distinguished by the color scheme on the right. Taxa overall represented by <0.2% of the sequences were merged in the “Other” category. (A) Abundance of each taxa in 69 available samples. (B) Abundances of the six major taxa (Acetobacter, Acetobacteraceae, Gluconobacter, Komagataeibacter, Lactobacillus, and Oenococcus) in the samples of the liquid fractions (left) or in the mother and daughter biofilms (right). In this plot, the data of the samples of AK, FE2, LK, and UR starters were excluded because they lack detection of the Lactobacillus genus. (C) Taxonomic profiles are represented by the normalized sequence count of taxa in 42 samples from the liquid fractions across the three cycles of each fermentation series and 27 samples of the mother and daughter biofilms of cycle 3. Empty columns correspond to the absence of data for the corresponding samples. Samples were sorted according to (i) the starter name, (ii) the fermentation cycle, (iii) the fraction, and (iv) the profile similarities in the major taxa resolved at the species level (visual assessment). The colored circles beneath the starter names represent their technological origin (red: industrial brewery; blue: household; black: unpasteurized commercial beverage; gold: commercial starter).
The Acetobacter genus, the next most common genus, is present in nearly all samples, with abundance varying from 0% to 71% and a mean of 14.4%. The Gluconobacter genus is found in 56 out of 69 samples, with a range of 0% to 78% abundance and a mean of 6.8%. An additional taxon, Oenococcus, is detected in 43 samples, but its abundance exceeds 1% in only eight samples from four starters (BB2, HK, KF2, and KOK), where it has a mean abundance of 29.3% (Fig. 5A and C). The remaining 19 genera together contribute <1.7% of the total sequences, with individual abundances ranging from 0% to 17.3%, and 76% of these values being <0.01%.
The Bray–Curtis index analysis clustered the samples into two main groups (Fig. 6), primarily based on the presence and dominance of either Lactobacillus or Acetobacter. Compositional data for the liquid fractions were available for three consecutive fermentation cycles from six original starters, for two consecutive cycles from eight starters, and for the first and last cycles from four starters. The abundances of Lactobacillus and Acetobacter genera also showed the greatest variability between samples from the same starter, suggesting fluctuation across repitchings. For samples characterized by Lactobacillus, its abundance increased from the first to the second cycle and remained relatively stable from the second to the third cycle. In contrast, Acetobacter abundance increased over repitchings in the samples from four starters (FE1, KF1, KF2, and LK), remained high for the samples of three other starters (FZ, HK, and UR), and stayed low for the rest. Only the samples from one starter (FE2) exhibited close bacterial compositions across all three fermentation cycles.
Comparisons of the bacteria genus compositions in the liquid fractions across successive repitchings. The spatial ordination along Axis 1 and Axis 2 was based on Bray–Curtis dissimilarity matrices. For better visibility, the samples from AK, BK, EB, FE2, FZ, HK, and UR starters are displayed on the separate left panel. Samples are colored according to the original starter as defined by the color schemes on the right, and the marker shape distinguishes the samples according to the fermentation cycle. The colored lines and polygons connect the samples from the same starter. The barplots at left and above indicate the distribution along Axis 2 or Axis 1 of the abundances represented by the normalized sequence counts in the corresponding samples for Acetobacter (green, maximum = 5369 sequences) and Lactobacillus (purple, maximum = 7015 sequences) genera, respectively. The colored circles next to the starter names represent their technological origin (red: industrial brewery; blue: household; black: unpasteurized commercial beverage; gold: commercial starter).
The genus Lactobacillus shows a highly diverse distribution across samples. In 42 samples, its abundance is below 5%, while in 11 samples, it exceeds 70%, with a median abundance of 1%. The low abundance of Lactobacillus in samples from 10 starters (AK, BK, FE1, FE2, FZ, HK, KF1, KF2, LK, and UR) and in nearly all biofilm fractions, no matter the starter, contributes to this variability. This contrasting distribution is especially evident when comparing the liquid and solid fractions (Fig. 5B). For the samples from 10 original starters, data were available for the composition of the liquid fraction (L3) and both biofilm fractions (D3 and M3). Additionally, six more starters had data for either L3 and D3 or L3 and M3. Bray–Curtis index-based profiles revealed samples from seven starters where Lactobacillus dominated the liquid fraction but was absent or less abundant in the biofilm fractions (Fig. S4). In these cases, the dominance of Lactobacillus in the liquid fraction was balanced by a very low proportion of Komagataeibacter. This suggests a preferential distribution of Komagataeibacter in the biofilm fractions and Lactobacillus in the liquid phase, though further analysis would be needed to confirm the exact distribution patterns. Additionally, the samples from three starters (FE1, LK, and UR) displayed a high abundance of Acetobacter in the liquid fraction but not in the biofilm fractions (Fig. S4).
Discussion
Our study compares the microbial compositions of SCOBYs from diverse natural kombucha starters. After additional repitchings under controlled conditions, we assessed the stability and spatial distribution of microbial taxa between liquid and biofilm compartments. To our knowledge, among the available studies on the microbial compositions of kombucha (Harrison and Curtin 2021, Laavanya et al. 2021, Huang et al. 2022, and references therein), this is the first study to apply such a design to this scale.
To decipher yeast compositions in our sample collection, we developed a new, shorter taxonomic marker (∼330 bp) compared to the conventional D1/D2 marker. Unlike ITS markers, this marker maintains a consistent amplicon size and higher interspecies similarity (Ercolini 2004). While a recent study suggests a minimal impact of ITS1 size on species quantification (Rué et al. 2023), our own limited comparison, analyzing six SCOBY samples not included in this study with both D2 and ITS1 primer pairs, revealed that low-abundance species with longer ITS1 regions were barely detected (data not shown). Manual curation helped overcome the limitations of low D2 sequence representation in public databases for accurate taxonomic assignment.
We detected all the common yeast and bacterial taxa reported for kombucha (Harrison and Curtin 2021, Laavanya et al. 2021, Huang et al. 2022, Landis et al. 2022). Within our SCOBY collections, certain taxa are present across all samples (or nearly all, and at least in samples from each initial starter): the two Brettanomyces species, as well as bacteria from the Komagataeibacter and Lactobacillus genera, and from the Acetobacteraceae family. Therefore, using the “common core” model criterion (Risely 2020), these taxa constitute the taxonomic core microbiota, regardless of their abundance. The remaining detected taxa are classified as accessory taxa, as defined by Vos (2023). Since the number of accessory taxa exceeds that of core taxa, the defined core aligns with the “minimal core” model proposed by Hamady and Knight (2009).
The common core defined here aligns well with the kombucha functional core characterized through approaches that reconstitute SCOBYs from isolated bacterial and yeast strains, including Komagataeibacter species and B. brettanomyces (Ferremi Leali et al. 2022, Landis et al. 2022). Compared to previous studies, the taxonomic core identified here underscores the importance of the B. anomala yeast species, whose functional role remains largely underexplored. Additionally, this is the first study to incorporate the Lactobacillus genus as part of the kombucha taxonomic core, as it is present in all samples from the SCOBY collection, albeit in highly variable proportions. Previous studies have reported inconsistent presence of Lactobacillus, likely due to its compartmentalized distribution when analyses focus on the solid compartment of SCOBYs (Harrison and Curtin 2021). Investigating its role in the functional core will also have to consider the potential influence of the fermentation medium (green or black tea infusion), as it has been suggested to have opposite effects on Lactobacillus and Acetobacteraceae, as reported by Andreson et al. (2022).
Our analyses revealed distinct patterns of variation in SCOBY microbial compositions, reflecting the diversity of the initial starter communities. Yeast compositions, particularly the ratios of the core yeast species B. anomalus and B. bruxellensis, vary significantly between samples and may serve as a signature of SCOBY structural diversity, especially when these ratios remain consistent within samples from the initial starter across at least three consecutive repitchings. Additionally, investigating microbial community dynamics over longer fermentation series would be valuable for detecting cyclic variations, as described for sourdough microbiota (Oshiro et al. 2023).
The bacterial compositions in the liquid compartment showed variability, with sample clustering based on the dominance of Lactobacillus or of the widely spread Acetobacter. However, variations in bacterial proportions across repitchings were less understood, likely due to the genus-level resolution of our metabarcoding approach, which could benefit from more detailed metagenomic methods.
Our analysis revealed that SCOBYs vary in the presence and proportions of accessory taxa. Common accessory taxa include Pichia, Zygosaccharomyces, and Gluconobacter, while others, such as Starmerella and Kregervanrija, are more specific. We observed that accessory taxa fluctuate across fermentation cycles and compartments, with some yeast taxa decreasing in the liquid fraction across repitchings. This suggests that it might be worth investigating whether certain taxa do not permanently colonize the kombucha niche, but instead require recurrent reintroduction. Given that repitching steps were performed under sterile conditions, these findings underscore the influence of environmental and repitching strategies on SCOBY microbial diversity. Our results also highlight compartmental differences in microbial compositions, with biofilms possibly acting as reservoirs for accessory taxa. Further research into these factors could provide valuable insights into the dynamics of kombucha microbial populations (Harrison and Curtin 2021, Laavanya et al. 2021, Huang et al. 2022).
Acknowledgments
The authors thank the professional and amateur brewers who provided kombucha SCOBYs. The authors would also like to thank Olivier Rué for his advice in using FROGS and Delphine Sicard, Sylvain Santoni, and Audrey Weber for helping with the Illumina MiSeq sequencing. We also thank all the members of the HaploTeam (https://www.haploteam.org/) for the fruitful discussions and their valuable advice.
Conflict of interest
F.F. and O.C. are the owners of the company Bio Brasseurs. The remaining authors declare that the research was conducted in the absence of any commercial or financial relationships that could be construed as a potential conflict of interest.
Funding
This work was supported by a CIFRE grant (no. 2018/1442).
Data availability
The sequencing data generated in this study were deposited in the Sequence Read Archive (SRA) under accession number PRJEB72042. Correspondence between samples and SRA files is provided in Table S6.
