Comparative phylobiomic analysis of the bacterial community of water kefir by 16S rRNA gene amplicon sequencing and ARDRA analysis

Authors


Correspondence

Matthias Ehrmann, Technische Universität München, Lehrstuhl für Technische Mikrobiologie, Weihenstephaner Steig 16, 85350 Freising, Germany. E-mail: M.Ehrmann@wzw.tum.de

Abstract

Aims

The aim of this study was to analyse the bacterial microbiota of water kefir using culture-independent methods.

Methods and Results

We compared four water kefirs of different origins using 16S rDNA amplicon sequencing and ARDRA. The microbiota consisted of different proportions of the genera Lactobacillus (Lact.), Leuconostoc (Leuc.), Acetobacter (Acet.) and Gluconobacter. Surprisingly, varying but consistently high numbers of sequences representing members of the genus Bifidobacterium (Bif.) were found in all kefirs. Whereas part of the bifidobacterial sequences could be assigned to Bifidobacterium psychraerophilum, a majority of sequences identical to each other could not be assigned to any known species. A nearly full-length sequence of the latter exhibited a beyond-species similarity (96·4%) with the sequence from the closest relative species Bif. psychraerophilum. A Bifidobacterium-specific ARDRA analysis reflected the abundance of the novel Bifidobacterium species by revealing its unique MboI restriction profile. Attempts to isolate the bifidobacteria were successful for Bif. psychraerophilum only.

Conclusions

The complexity of the water kefir microbiota has been underestimated in previously studies. The occurrence of bifidobacteria as part of the consortium is novel.

Significance and Impact of the Study

These data give new insights into the understanding of the complexity of food fermentations and underline the need for approaches detecting noncultivable organisms.

Introduction

Traditionally, water kefir is cultured by placing kefir grains in an 8% sucrose solution in tap water with dried fruits, such as figs and lemon slices. After one or two days of fermentation at room temperature, a cloudy, carbonated, straw-coloured, acidic, sugar-poor and slightly alcoholic beverage is produced.

The origin of water kefir grains, also known as ‘ginger beer plants’, ‘Tibi grains’ (Lutz 1899), ‘California bees’, ‘African bees’, ‘Ale nuts’, ‘Balm of Gilead’, ‘Japanese Beer Seeds’ (Kebler 1921) and ‘Sugary kefir grains’ (Pidoux 1989; Pidoux et al. 1990), is still unclear.

The microbiota of water kefir grains have been described to contain different lactic acid bacteria, acetic acid bacteria and yeasts in a stable symbiotic system (Ward 1892; Lutz 1899; Horisberger 1969; Pidoux 1989; Galli et al. 1995; Neve and Heller 2002). In the literature describing the composition of the microbiota of water kefir, the isolation and reliable identification of species has been rare (Waldherr et al. 2010). Recently, we identified Lactobacillus nagelii, Lactobacillus hordei, Lactobacillus hilgardii, Lactobacillus casei, Leuconostoc mesenteroides and Leuconostoc citreum, as well as two Acetobacter species (Acetobacter fabarum and Acetobacter orientalis), as stable components of this microbiome using culture-dependent techniques (Gulitz et al. 2011). As there are known limitations to the cultivability of many micro-organisms, technologies that do not require cultivation have proven useful to identify nonculturable micro-organisms. Recent studies analysing the microbial communities of fermented foods clearly indicate that diversities have been routinely underestimated by standard techniques and that culture-independent methods allow much more insight into their complexities (Fleet 1999; Roh et al. 2010). Sequence-based molecular technologies for investigating the diversity and functionality of microbiota in fermented foods become mainstream for fermentation phylobiome analysis (Van Hijum et al. 2012).

The application of high-throughput pyrosequencing techniques employing 16S rRNA amplicon sequencing was successfully applied to investigate the microbial diversity of kimchi (Jung et al. 2011) and milk kefir (Dobson et al. 2011), providing novel data describing the ‘phylobiomes’ of the microbiota involved. Thus, 16S rDNA amplicon sequencing might be a sensible approach for the in-depth exploration of the microbial ecology of water kefir. Rapid amplified ribosomal DNA restriction analysis (ARDRA) is a well-established method for exploring the microbial consortia in fermented foods (Spiegelman et al. 2005). We used ARDRA as a second culture-independent technique to evaluate the data obtained using the sequencing approach.

Materials and methods

Preparation of water kefir

Four water kefirs (A, F, I and W) were obtained from home-made preparations made in different regions in Germany. Three kefirs (A, F and W) were propagated under standardized conditions at least two times prior to analysis. Water kefir I was analysed directly after arrival from the supplier by post.

The water kefirs were propagated in a medium containing 100 ml l−1 of fig extract and 80 g l−1 sucrose. To guarantee a constant mineral content, we used mineral water (Residenz Quelle, natural, Bad Windsheim, Germany). The fig extract was prepared by manually shaking 480 g of dried figs cut into small pieces in 1 l of mineral water for 20 min. This extract was centrifuged for 3 h at 17 000 g and sterilized using a 0·22-μm membrane filter unit (Sarstedt, Germany). Dried figs that had not been treated with any preservative agents (Seeberger, Germany) were obtained from a local supermarket.

The fermentation was performed at 21°C for 72 h. The supernatant was discarded, and the kefir grains were washed with tap water. Lemon slices were not added to minimize potential contamination from handling the water kefir under aseptic conditions.

DNA Isolation

For DNA isolation, 10 g of grains of each water kefir was diluted with 90 ml ¼ Ringer's reagent (Merck, Germany) and mechanically homogenized with a bag mixer for 60 s. Then, 5 ml of this solution was centrifuged at 5000 g for 10 min. The pellet was washed with 2 ml TE buffer containing 1 mmol l−1 EDTA, 10 mmol l−1 Tris, pH 8 and centrifuged again. The pellets were then stored at −20°C until further use.

The DNA isolation was performed with the E.Z.N.A. Bacterial DNA kit (Omega Bio-Tek, Norcross, GA, USA) according to the manufacturer′s instructions. The pellet from each water kefir was resuspended in 200 μl TE buffer containing lysozyme (10 mg ml−1), and the mixture was incubated at 37°C for 60 min. The DNA was eluted twice with 50 μl of elution buffer.

16S rDNA amplification and sequencing

The DNA isolated from each water kefir was used as a template for the amplification of the V1 to V4 hypervariable regions of the bacterial 16S rRNA gene with the Ba27f and Ba519r primers (Lane 1991). The reaction mixture (50 μl) consisted of 0·1 mmol l−1 of each deoxynucleoside triphosphate, 0·75 U Taq polymerase, 5 pmol of each primer and 1 μl of the genomic DNA, and amplification was carried out using a PCR program of 94°C for 2 min; 32 cycles of 94°C for 45 s, 52°C for 90 s and 72°C for 2 min; and a final step at 72°C for 5 min.

After the PCR, the quality of the amplified PCR products (approximate length: 520 nt) was confirmed by electrophoresis, and the products were purified using a Cycle Pure Kit (Omega Bio-Tek) according to the manufacturer's instructions. A 20 μg sample of each amplicon was sequenced in a 454 Genome Sequencer FLX Titanium by LGC Genomics (Berlin, Germany).

Data processing and sequence analysis

The initial sequence analysis was performed using the classifier software (RDP Naïve Bayesian rRNA Classifier Version 2·4 (Wang et al. 2007)) provided in Ribosomal Database project II (Cole et al. 2009) to estimate the microbial diversity at the genus level. Sequences shorter than 150 nt were removed. The confidence threshold was at least 80%. Sequences with less than 97% similarity to the sequences deposited in RDP II were classified as unidentified.

Taxon assignment for the 16S rDNA sequences was performed in the ARB software package (http://www.arbhome.de; (Ludwig et al. 2004; Westram et al. 2011) and using the SILVA (http://www.arb-silva.de; (Pruesse et al. 2007) and LTP (all-species Living Tree Project; http://www.arb-silva.de/projects/living-tree. (Yarza et al. 2008; Munoz et al. 2011)) databases. Only sequence reads of at least 400 nucleotides were included for comparative data analysis. Preliminary taxon assignment above the species level was roughly achieved by applying the classifier software (Wang et al. 2007) provided by the RDP (Ribosomal Database Project; (Cole et al. 2009) and the PT-server (Positional Tree server; (Ludwig et al. 2004; Westram et al. 2011))-based ‘next relative search’ of the ARB package.

For a more detailed taxon assignment, the species differentiation capacity of the 16S rRNA marker for the relevant taxonomic groups extracted from the LTP database was evaluated. The current species threshold is set at 98·7% identity over the entire 16S rRNA sequence (Stackebrandt and Ebers 2006). As there are many type strains that share higher similarities, 16S rRNA-based assignment is only possible for closely related species groups in such cases. The species threshold value cannot be applied directly to analyse partial rDNA sequence data, such as that derived in this study. This analysis required that the (nonlinear) correlation of full and partial sequence-derived similarities be determined for the individual type strain groups and that the respective species-group thresholds be delineated. These similarity-based thresholds were used to define the respective score value ranges obtained by applying the ARB PT-server approach (Ludwig et al. 2004; Westram et al. 2011). This suffix tree-based search for common heptanucleotide occurrences in the source (pyrosequencing) and reference (type strain) data allows a phylogenetic similarity analysis without the need for sequence alignment. The underlying principle is based on the finding that multiple occurrences of identical heptanucleotide stretches are highly unlikely in 16S rRNA primary structures (our unpublished data). Thus, sequences from diverse sources that contain identical heptanucleotides can be regarded as homologous. PT-server score thresholds were defined for two levels: the species or (lowest) species-group level and the lowest level that clearly separates the respective species (group) from all other type species. The resulting three score ranges allow a sequence to be assigned to its respective species or species group, its phylogenetic lineage or to higher taxa, respectively (Table 2). The high error rate in sequence data obtained by high-throughput methods (Balzer et al. 2010) must be taken into account when interpreting these score ranges. Lower scores may represent as-yet undescribed species or simply low-quality data.

ARDRA analysis

DNA samples from the four water kefirs were used as templates for amplification with the specific bifidobacteria primers Bif164_mod_for and LM3_mod_rev (Table 1). The reaction mixture (50 μl) consisted of 0·1 mmol l−1 of each deoxynucleoside triphosphate, 0·75 U Taq polymerase, 5 pmol of each primer and 1 μl of the genomic DNA, and the amplification was carried out using a PCR program of 94°C for 5 min; 32 cycles of 94°C for 1 min, 45 s of 60°C and 72°C for 2 min; and a final step at 72°C for 5 min.

Table 1. Primer sequences and binding positions
 Sequence 5′–3′Reference16S binding positionaSpecificity/comment
  1. a

    Target region for hybridization in the 16S rRNA molecule, numbered according to the homologous Escherichia coli sequence (Brimacombe et al. 1990).

Ba27fAGAGTTTGATYMTGGCTCAGLane 1991;8–27Bacteria universal/Pyrosequecing
Ba519rTATTACCGCGGCKGCTGLane 1991;519–534Bacteria universal/Pyrosequencing
Bif164_mod_forGGGTGGTAATGCCGRATGLangendijk et al. 1995;164–181G replaced by R, bifidobacteria genus
LM3_mod_revGGTGCTNCCCACTTTCATGKaufmann et al. 1997;1412–1432Del 3′CG and I replaced N, bifidobacteria genus
Bif_spec_forGGATGTGGGACCCATTCThis study829–845Bif. spec.kefir
Bif_spec_revGAACCCGTGGAATGGGTCThis study837–856Bif. spec.kefir
630RCAKAAAGGAGGTGATCCJuretschko et al. 2002;15282Bacteria universal
616VAGAGTTTGATYMTGGCTCAGBrosius et al. 19817Bacteria universal

The amplified DNA had a length of approx. 1400 nt and was purified using the PeqGOLD Gel Extraction Kit (PeqLab, Germany) according to the manufacturer's instructions. The DNA was eluted in 30 μl elution buffer and stored at 4°C until further use.

Amplicons were cloned in Escherichia coli Top 10 using the AccepTor Vector Kit (Novagen Inc., Madison, WI, USA).

Restriction analysis with MboI (GATC, Fermentas) according to manufacturer′s instructions was performed on 100 clones from each kefir sample after colony PCR using the above primers with MboI restriction enzyme (GATC, Fermentas) according to the manufacturer's instructions. Fragments were electrophoresed in a 1·3% (wt/vol) agarose gel (0·5× Tris-borate–EDTA buffer [45 mmol l−1 Tris-borate, 1 mmol l−1 EDTA]), followed by staining with dimidium bromide.

Amplicons from the DNA of ‘Bifidobacterium crudilactis’ LMG 21775, Bifidobacterium psychraerophilum DSM 22366T, Bifidobacterium longum DSM 20088T, Bifidobacterium animalis subsp. lactis DSM 10140T, Bifidobacterium lactis BB12 and Bifidobacterium breve DSM 20213T were also included in this analysis.

Isolation and identification of bifidobacteria

Analysis of the full 16S rDNA sequence of the novel Bifidobacterium species

To amplify and analyse the full 16S rRNA gene sequence of the novel uncultivated bifidobacteria species, the universal 16S primers 616V and 630R were combined with the newly designed specific primers Bif_spec_for and Bif_spec_rev (Table 1). DNA from water kefir A was used as a template. From this amplicon, 1525 nt were sequenced and submitted to the EMBL database under the accession number HE804184.

Isolation and growth of bifidobacteria

For the selective isolation of bifidobacteria, 10 g of water kefir grains (water kefir I and water kefir W as representatives) was diluted in 90 g Ringer's reagent and then mechanically homogenized with a bag mixer for 60 s. Serial dilutions were plated on modified tryptone–phytone agar plates containing 10 g l−1 peptone from casein, 10 g l−1 peptone from soya, 6 g l−1 NaCl, 5 g l−1 K2HPO4, 2 g l−1 glucose, 2 g l−1 raftilose (ORAFTI, Active Food Ingredients, Belgium, Oreye) and 15 g l−1 agar. The pH was adjusted to 6·1, and 400 μg ml−1 cycloheximide and 400 μg ml−1 kanamycin sulfate were added after the agar was cooled to 50°C. The agar plates were incubated under anaerobic conditions (70% N2 and 30% CO2) at 30°C for 3–5 days.

Identification of Bifidobacterium psychraerophilum

The primers LM3_mod_rev and Bif164_mod_for were used to identify bifidobacteria. The amplified PCR product was purified with the Cycle Pure Kit (Omega Bio-Tek, USA) according to the manufacturer's instructions. The Sanger sequencing was performed by a commercial provider (GATC Biotech, Germany). The identification of the bifidobacteria was performed with the BLAST program. The accession number of the Bif. psychraerophilum sequence used as reference is NR_029065.

Results

Bacterial identification and distribution determined by high-throughput sequence-based analysis

The bacterial microbiota of four independently grown water kefir grains from different origins were analysed by high-throughput sequence-based analysis. Three water kefirs (A, F and W) were grown at least two times under standardized conditions to ensure that they all are propagated under identical conditions and to eliminate possible influences of the microbiota resulting from cultivation procedures of the supplier. Water kefir I was analysed directly after arrival from the supplier to detect any kind of difference in the microflora resulting from growing conditions. A total of 43293 (kefir A), 33079 (kefir F), 25951 (kefir W) and 41404 (kefir I) reads of the V4 region of the 16S rRNA were analysed. After quality trimming, a total of 31795 (kefir A), 23909 (kefir F), 18305 (kefir W) and 25319 (kefir I) reads were used for taxon assignments. The distribution of the different bacteria at the family level was evaluated by the software tools provided by the Ribosomal Database project II (Cole et al. 2009), demonstrating the presence of Acetobacteriaceae, Bifidobacteriaceae, Clostridiaceae, Leuconostocaceae, Lactobacillaceae and unclassified bacteria (Fig. 1). The most abundant bacteria were Lactobacillaceae (A 69%, F 67·9%, W 57·4% and I 27·3%), followed by Bifidobacteriaceae (A 21·5%, F 18·7% and I 21·9%), with the exception of water kefir W, which had only 3·8%. Kefir W showed a higher percentage of Leuconostocaceae (29·1%) than the other kefirs. Only kefir I had a high amount of Clostridiaceae (37·2%). Acetobacteraceae were present in low amounts (A 2·8%, F 5·6%, W 3·3% and I 5·8%).

Figure 1.

Microbial diversity of the four water kefirs on family level. Each colour represents a different bacteria family (×) Clostridiaceae, (image_n/jam12124-gra-0002.png) Bifidobacteriaceae, (image_n/jam12124-gra-0001.png) Leuconostocaceae, (image_n/jam12124-gra-0003.png) Acetobacteraceae, (image_n/jam12124-gra-0004.png) unclassified Bacteria and (image_n/jam12124-gra-0005.png) Lactobacillaceae.

The suffix tree-based search for the most similar type strain sequences was used to assign the sequences to genera, species groups or species (Table 2). With this method, we were able to identify 12 bacterial species and 12 bacterial groups. The main species of water kefirs A and F consisted of members of the Lact. hordei group, Lact. nagelii and bifidobacteria spp.; water kefir A also contained species from the Bif. psychraerophilum group; and water kefir F had a greater amount of Acet. fabarum group members. The most abundant sequences in water kefir I were from members of the Bif. psychraerophilum group, Lact. hordei group, Lact. hilgardii, Lactobacillus satsumensis, Acet. orientalis group and Clostridium tyrobutyricum. The core species of water kefir W were from the Lact. hordei group, Lact. nagelii, Lact. hilgardii, Leuc. citreum group and Leuc. mesenteroides group.

Table 2. Taxonomic assignments of 99328 analysed sequences of four water kefir granules to genus, species group or species. Only taxa with >1 reads are shown. Species previously identified with culture-dependent methods (Gulitz et al. 2011) are marked in bold letters. Numbers in parentheses show sum of reads of kefirs A, F, W and I assigned at genus level
Genus Species(group)aNumber of sequence reads
Kefir A (31795)Kefir F (23909)Kefir I (25319)Kefir W (18305)
Genus b GroupcSp/gpd Genus GroupSp/gp Genus GroupSp/gp Genus GroupSp/gp
  1. a

    The species groups (combined in the respective cells) unify valid species sharing over all 16S rDNA sequence similarities beyond the species threshold (Stackebrandt and Ebers 2006). The species of a given genus are only listed if they are members of the groups represented in the table or newly determined data could be assigned to them.

  2. b

    Number of 16S rDNA sequences that could be assigned to valid taxa at (not below) the genus level.

  3. c

    Number of 16S rDNA sequences that could be assigned to the (phylogenetic) periphery of valid species or species groups (seea). The respective sequence similarities are within a range defined by the species threshold value and the sequence similarity for the respective species (group) and its closest type strain relative.

  4. d

    Number of 16S rDNA sequences that could be assigned to valid species or species group (seea).

Acetobacter (3715) cerevisiae, cibinongensis, indonesiensis, malorum, orientalis, orleanensis, tropicalis 127021145057260428130036
aceti 0203000 
fabarum, iovaniensis, ghanensis, syzygii 18315749254418592264163
peroxydans 20005140
Gluconobacter (359) albidus, oxydans, roseus, kanchanburiensis, sphaericus 125221010844509411
cerinus, frateuri, japonicus, thailandicus wancheriae, 01013910510
Gluconacetobacter (504) entanii, europaeus, hansenii, intermedius, kombuchae, nataicola, oboediens, rhaeticus, sacharivorans, sucrofermentans, swingsii, xylinus 3802771459761109411
liquefaciens, sacchari 000000057
johannae 200000000
Bifidobacterium (18759) crudilactis, psychraerophilum 1220327837556972261655659303316137107
subtile 3100051000
species 2540214218812016408326285164
Lactobacillus (60561) aquaticus, sucicola, uvarum, capillatus 1257400918420326620273290
cacaonum, hordei, mali 4051266421149231612178774612
ghanensis 042105800320289
nagelii 080570501801450738
satsumensis 11989136509243
casei, paracasei, zeae 50222131916698606552
hilgardii 18801280355908390
Leuconostoc (7293) citreum, holzapfeli, lactis, palmae 82001640046013227803106
holzapfelii 0020000255
mesenteroides, pseudomesenteroides 2486461571627755228
kimchii 10000070
Clostridium (8137) arbusti 00000003370000
tyrobutyricum 02057787060

ARDRA analysis

With ARDRA technique, we focused on bifidobacteria because they were identified the first time in water kefir, and we wanted to elucidate whether there are different bifidobacteria species in the water kefir. Using Bifidobacteria genus-specific primers (LM3_mod_rev, Bif164_mod_for), we were able to amplify a 1253 nucleotide 16S rDNA sequences from all four water kefirs and from the reference strains. A clone library of 100 ARDRA profiles of each water kefir was also analysed. MboI restriction fragments of reference strains correspond to those as predicted by in silico analysis of the respective 16S amplicons. ‘Bif. crudilactis’ (a hitherto not validated species) and Bif. psychraerophilum shared identical restriction patterns, as well as Bif. animalis subsp. lactis and Bif. lactis that showed also same restriction patterns. Bif. breve shows a slightly different pattern, with one additional band in the 150-bp range and a high molecular weight band smaller than those of ‘Bif. crudilactis’ and Bif. psychraerophilum of 480 bp. Bif. longum shows a unique restriction pattern (Fig. 2).

Figure 2.

16S rDNA ARDRA profiles of Bifidobacterium species after restriction with MboI. The agarose gel shows the restriction pattern of Bifidobacterium crudilactis LMG 21775 (lane 1), Bifidobacterium psychraerophilum DSM 22366T (lane 2), Bifidobacterium psychraerophilum isolated from water kefir (lane 3), Bifidobacterium longum DSM 20088T (lane 4), Bifidobacterium animalis subsp. lactis DSM 10140T (lane 5), Bifidobacterium lactis DSM 58 (lane 6) and Bifidobacterium breve DSM 20213T (lane 7). Lanes 8–11 show the ARDRA profiles of cloned 16S rDNA amplicons derived from PCR amplification using genus Bifidobacterium primers from water kefirs A (lane 8), F (lane 9), W (lane 10) and I (lane 11). Each lane represents the pattern of 100 randomly chosen colonies. A Generuler 100-bp ladder was used as size marker.

Interestingly, in all kefirs, a unique profile was obtained consisting of 2 characteristic bands of 446 and 235 nt (lanes 8 to 11 in Fig. 2) in common with Bif. crudilactis and Bif. psychraerophilum strains (lanes 1–3) and one of 572 bp., indicating the dominant presence of one or more Bifidobacterium species sharing this pattern, nonidentified by the ARDRA profile, in all four kefirs. This identical profile from all four water kefirs was different from that of other bifidobacteria, including the closest 16S relatives ‘Bif. crudilactis’, Bif. psychraerophilum DSM 22366T and the Bif. psychraerophilum strains isolated in this study.

The phylogeny of the Bifidobacterium sp

The 16S rRNA gene sequence was amplified from the DNA samples isolated from kefir A in two fragments using a combination of a species-specific primer for the ‘new’ water kefir species (Bif_spec_for, Bif_spec_rev) and universal primers (616V, 630R). The almost complete 16S rRNA sequence (1525 nt) was determined. Comparison of this sequence with those available in public databases revealed Bif. psychraerophilum as the closest relative, with a sequence similarity of 96·4%. The subsequent phylogenetic classification supports a new candidate species of the genus Bifidobacterium. The 16S rRNA gene sequence was submitted to the EMBL database under the accession number HE804184.

The cultivation and quantification of bifidobacteria

The water kefirs W and I were analysed representatively for Bifidobacteriaceae. The bacteria were cultured on a modified tryptone–phytone (TP) agar, and the viable cell numbers (cfu/g) were enumerated. The number of Bifidobacteriaceae cells in water kefir I was 4·9 × 105 cfu/g and that in water kefir W 1·5 × 106 cfu/g. An analysis of representative colonies (at least 100 colonies of each water kefir) by PCR with the species-specific primer (Bif_spec_for and Bif_spec_rev) (data not shown) showed that all isolates grown on these plates belonged to the species Bif. psychraerophilum. Bifidobacteria cell numbers were not analysed for the water kefirs F and A because they were not physically available at that time.

Discussion

In this study, the bacterial communities of four different water kefirs were analysed by a culture-independent, high-throughput sequencing procedure employing 16S rRNA gene amplicon sequencing. The microbiota of all kefirs consisted of bacteria from three phyla: Firmicutes (79%), Actinobacteria (17·1%) and Proteobacteria (3·7%).

The Firmicutes were the dominant organisms in all kefirs, represented mainly by the two genera Lactobacillus and Leuconostoc.

Kefir I, which was directly analysed upon arrival without any propagation, showed high numbers of clostridia (37·2%), while clostridial sequences represented less than 0·03% of the sequences in the other kefirs. Interestingly, Clostridiaceae assignments accounted for 0·82% of reads in the starter grains of milk kefir (Dobson et al. 2011). The high clostridia content and a noticeable off flavour were obviously consequences of improper propagation or contamination during preparation and/or shipment. Clostridium tyrobutyricum, which was the most abundant Clostridiaceae species, is known as a food-spoiling bacterium and a major cause of late blowing of cheese (Klijn et al. 1995) but is also found in environments related to water kefir, such as in fruit pulp (Mayer et al. 2010). Therefore, we do not consider kefir I to harbour a typical kefir consortium and did not include the data from this sample in the subsequent analysis.

The phylum Actinobacteria is represented in the water kefir consortium exclusively by the genus Bifidobacterium, and Proteobacteria are represented by Acetobacter, Gluconobacter and Gluconacetobacter in all four water kefirs. Kefirs A, F and W differ in their composition only marginally.

Due to sequencing artefacts such as sequence-specific errors, miscounted homopolymeric runs, etc., the use of 454 pyrosequencing data for community profiling is known to occasionally inflate estimates of actual diversity when taxon assignments are made using alignment-based approaches (Kunin et al. 2010). To prevent overestimates of diversity using regions of the 16S rDNA, a clustering threshold of 97% identity is widely used. This limits the assignment of sequences at the current species threshold of 98·7%. In this study, we used a suffix tree-based search for common heptanucleotide occurrences in the pyrosequencing data and reference data allowing similarity analyses without the need for sequence alignments of large data sets. This allows a more reliable assignment of reads to a single species or species group.

The data from this study corroborate the identification of species that have been described in previous studies of water kefir using culture-dependent techniques (Gulitz et al. 2011). The main core of bacteria described here is similar to that shown in our previous investigation with culture-dependent methods, in which Lact. hordei, Lact. nagelii, Leuc. mesenteroides and Leuc. citreum, as well as Acet. fabarum, represented the core group of organisms (Gulitz et al. 2011). In this study, kefir W had the highest amount of Leuconostocaceae, and a high number of Lact. nagelii and Lact. hordei were found in water kefirs A and F. We were able to detect only small amounts of Lact. casei in all four water kefirs, while Lact. casei (Franzetti et al. 1998) and Lact. hilgardii were identified as the predominant bacterial species in other water kefirs (Pidoux 1989). This might be due to the fact that the two species Lact. hordei and Lactnagelii were described in 2000 and 2008, respectively. A low percentage of the sequences in water kefir A and F were derived from Lact. hilgardii, whereas water kefirs I and W showed a slightly higher percentage. This is consistent with the data obtained using the culture-dependent approach (Gulitz et al. 2011).

The detection of bifidobacteria in water kefir is unexpected and has not been described previously. In this context, it is worth mentioning that we were not able to observe cells with bifidobacterial morphology using light microscopy. Notably, previous studies using scanning electron microscopy have not identified bifidobacteria as elements of the water kefir consortium (Neve and Heller 2002).

This might be due to the atypical morphology of Bif. psychraerophilum, which is described to show bifurcation only occasionally (Simpson et al. 2004).

In our hands, neither the type strain of Bif. psychraerophilum nor strains isolated from water kefir in the present study showed bifidobacterial morphology, even in pure cultures (data not shown).

Recently, Dobson et al. (2011) reported the finding of bifidobacterial sequences in a culture-independent analysis of milk kefir. In this consortium, the Bifidobacterium population, containing the species Bif. breve, Bifidobacterium choerinum, Bif. longum and Bifidobacterium pseudolongum, comprised just 0·2% of total assigned taxa in the collective starter grain and 0·4% in the kefir milk. However, culture-dependent methods failed to detect Bifidobacterium species in either sample. This underscores the fact that culture-independent analysis is a powerful tool for the better understanding of microbial consortia and that bifidobacteria with unknown taxonomy and physiology may contribute to various extents to such consortia.

Interestingly, a large number of bifidobacterial reads from water kefir could not be assigned to any known species.

The bifidobacterium-specific 16S ARDRA analysis of 100 clones per kefir showed a single restriction profile in all of the kefirs that could be clearly differentiated from that of the type strains Bif. crudilactis FR62/b/3T and Bif. psychraerophilum LMG 21775T.

Moreover, the sequences of all corresponding amplicons showed a similarity of less than 97% to its closest relatives, Bif. crudilactis FR62/b/3T and Bif. psychraerophilum LMG 21775T. This result was unexpected because the metagenomic data showed ratios of 3 : 1, 9·4 : 1, 1 : 6·8 and 1 : 2·1 of Bifidobacterium spec. to the bifidobacteria Bif. psychraerophilum, Bif. crudilactis and Bifidobacterium subtile, respectively. Because the template DNA samples used for pyrosequencing and ARDRA and primer binding sites were identical, this discrepancy may be ascribed to an artificial cloning bias.

The selective cultivation of bifidobacteria from natural habitats harbouring lactobacilli and other lactic acid bacteria is difficult because the cultural and biochemical properties of both genera overlap (Thitaram et al. 2005). Many attempts to isolate bifidobacteria from water kefir with previously described media, including mupirocin-containing media, failed in our hands because the bifidobacteria were overgrown by lactic acid bacteria or yeasts (data not shown). Our modified selective medium suppresses yeast and LAB growth due to the high amounts of cycloheximide and kanamycin, and the addition of oligofructose as bifidogenic carbohydrate allowed the selective cultivation of up to 1·5 × 106 cfu g−1 Bif. psychraerophilum alone from kefir granules. However, the isolation of this novel bifidobacterium species, the 16S sequence of which was highly abundant in the pyrosequencing and ARDRA data, was not successful, suggesting that either we were unable to meet its optimal growth conditions or the pyrosequencing data do not necessarily reflect the true numbers of living cells.

Nevertheless, this medium may also be helpful to allow the cultivation of bifidobacteria from other sources from which the isolation of bifidobacteria has not been successful to date.

Our study confirms previous data and broadens the available knowledge about the microbial consortium of water kefir. The unexpected presence of bifidobacteria in all samples and the difficulty of cultivating these species indicate that the role of bifidobacteria in other fermented foods may also be underestimated. The phylogenetic position of bifidobacterial 16S rRNA gene sequences with similarities of less than 96·4% to known species suggests the identification of a new candidate species of the genus of Bifidobacterium. The large number of bifidobacteria sequences found in all samples indicates that these bacteria are part of the core species of water kefir.

Acknowledgements

This project was financially supported by the German Ministry of Economics and Technology and the WiFö (Wissenschaftsförderung der Deutschen Brauwirtschaft, Berlin, Germany) in project AiF 16454 and the DFG (Deutsche Forschungsgemeinschaft) grant Lu 421/7-1. Authors have no conflict of interest to declare.

Ancillary