Presence of an ultra-small microbiome in fermented cabbages

Fermented vegetable samples

To explore the diversity of USM in fermented cabbage, kimchi, a fermented cabbage made in South Korea; suancai, a fermented cabbage made in China; and sauerkraut, a fermented cabbage made in Germany, were purchased (8 kg of kimchi, 4 kg of sauerkraut, and 5 kg of suancai) from an online market in June 2019. Moreover, the expiration date of kimchi, sauerkraut, and suancai from the date of manufacture was 1, 36, and 18 months, respectively. Both kimchi and suancai samples were non-sterile, while sauerkraut was sterile. Notably, suancai contained a preservative (potassium sorbate). Kimchi was made from kimchi cabbage (Brassica rapa subsp. pekinensis), suancai was made from Chinese cabbage (B. rapa subsp. pekinensis), and sauerkraut was made from white cabbage (Brassica oleracea var. capitata). B. rapa subsp. pekinensis, produced in Korea and bred for kimchi production, is called kimchi cabbage in order to be easily distinguished from B. rapa subsp. pekinensis, which is used to produce suancai (CAC, 2013). In addition, the fermented cabbages used in this study were not artificially produced in a laboratory, but were representative commercial products commonly consumed by the general public. The detailed ingredients of each fermented cabbage are shown in Fig. S1. The fermented cabbage samples were stored in chilled conditions at approximately 4 °C until filtration (Lee et al., 2022a).

Pre-filtration

Pre-filtration was performed on the broths of samples to facilitate TFF using a polypropylene capsule filter (GVS Filter Technology, Morecambe, UK) with a pore size of 10 µm. A vacuum pump (2546C-10; Welch Materials Inc., Shanghai, China) was used to aid the filtration process by reducing the pressure during the pre-filtration step. Before pre-filtration, the polypropylene capsule filter and tubing were sterilized with a solution of sodium hypochlorite 0.1% (v/v) (Lee et al., 2022a).

TFF

TFF was performed using a TFF system (Cogent µScale TFF System; Millipore, Sigma-Aldrich, St. Louis, IL, USA) in two phases. In the first phase, a TFF cartridge (Pellicon 2 Mini Cassette, Media: Durapore 0.22 µm; Millipore, Sigma-Aldrich) was used to cut off particles > 0.2 µm for the isolation of normal size bacteria as well as to remove macromolecules, including plasmid DNA. In addition, because UMB can pass through a 0.2-µm pore size filter due to its small size (Cavicchioli & Ostrowski, 2003; Duda et al., 2012; Nakai, 2020), a 0.2-µm pore size filter was used in many previous studies to identify UMB present in various environments (Silbaq, 2009; Loveland-Curtze, Miteva & Brenchley, 2010; Sahin et al., 2010; Maejima et al., 2018). Therefore, we set the pore size of the filter separating USM from NM to 0.2 µm to confirm USM based on UMB. Although USM encompasses endospores and OMVs including UMB, separation standard for the microbiome was set as UMB. In the second phase, a TFF cartridge (Pellicon 2 Mini Cassette, Media: Biomax 100 kDa; Millipore, Sigma-Aldrich) with a molecular weight cut-off (MWCO) of 100 kDa was used to remove small molecules below the USM size. Six samples, including (1) normal microbiome (NM) > 0.2 µm from kimchi (Kimchi_NM), (2) USM below 0.2 µm from kimchi (Kimchi_USM), (3) NM > 0.2 µm from sauerkraut (Sauerkraut_NM), (4) USM < 0.2 µm from sauerkraut (Sauerkraut_USM), (5) NM > 0.2 µm from suancai (Suancai_NM), and (6) USM < 0.2 µm from suancai (Suancai_USM), were subjected to further evaluation.

The samples were concentrated to approximately 25–50 fold via TFF followed by TFF system sterilization by recirculation with 0.1% (v/v) sodium hypochlorite for 30 min and cleaning by recirculation with sterilized ultrapure water for 2 h. The detailed specifications of the two phases of TFF are shown in Fig. S2. Additionally, Cai and colleagues (Cai et al., 2015) used TFF to efficiently separate bacteria and viruses from the marine environment and found that the adsorption of bacterial cells in a filter made of polyvinylidene fluoride (PVDF) was low (Cai et al., 2015). Based on their results, we used a cassette filter based on PVDF (Durapore 0.22 µm) for TFF to separate the USM < 0.2 µm and the NM > 0.2 µm. The USM and NM isolated via TFF were stored in a refrigerator at 4 °C until DNA extraction (Lee et al., 2022a).

DNA extraction and PCR amplification

Nucleic acids were extracted from 5 to 10 mL of concentrated NM and USM sample solutions using a DNeasy PowerSoil kit (Qiagen, Hilden, Germany) and quantified using the Quant-IT PicoGreen assay kit (Invitrogen, Thermo Fisher Scientific, Waltham, MA, USA) following the manufacturer’s instructions. Libraries were prepared via PCR amplification using the PacBio RS II. The nucleic acids were amplified with a primer set (27F, 5′-AGRGTTYGATYMTGGCTCAG-3′; 1492R, 5′-GGTTACCTTGTTACGACTT-3′) for the full-length 16S rRNA gene. The PCR conditions were as follows: initial denaturation at 94 °C for 5 min, followed by 35 cycles of denaturation at 94 °C for 30 s, annealing at 53 °C for 30 s, extension at 72 °C for 90 s, and a final extension at 72 °C for 5 min. Purification of the PCR amplicons was carried out using AMPure beads (Agencourt Bioscience, Beverly, MA, USA). To verify the amount and size of PCR products, fluorescence was measured using the Quant-IT PicoGreen assay kit (Thermo Fisher Scientific), and the template size distribution was measured using an Agilent DNA 12000 kit (Agilent Technologies, Santa Clara, CA, USA). Pooled amplicons were used for library preparation PacBio Sequel sequencing. A library was prepared using the PacBio DNA template prep kit 1.0 (Pacific Biosciences, Menlo Park, CA, USA). The PacBio DNA sequencing kit 4.0 and 8 SMRT cells (Pacific Biosciences) were used for sequencing (Lee et al., 2022a).

SMRT sequencing

SMRT sequencing was performed using a PacBio RSII system (Pacific Biosciences) according to the manufacturer’s instructions. In brief, 10-h movies were captured for each SMRT cell (Pacific Biosciences). Subsequent steps were carried out based on the PacBio Sample Net-Shared Protocol (https://www.pacb.com/). Circular consensus sequencing (CCS) reads, such as raw sequence reads, were processed using the SMRT analysis software (version 2.3; Pacific Biosciences). Short CCS reads and those with zero quality bases, considered as sequencing errors, were removed (Lee et al., 2022a).

Taxonomic and statistical analysis

Taxonomic analysis of the CCS reads was performed using the MG-RAST server (Meyer et al., 2008) with the SILVA SSU database (Quast et al., 2013). Then, in the pipeline version 4.0.3, the pipeline options were set as follows: assembled, dereplication, screening, length filtering, length filter deviation multiplicator, ambiguous base filtering, and maximum ambiguous base pairs were set to ‘yes,’ ‘no,’ ‘E. coli NCBI st. 536′, ‘2.00′, ‘no’, and ‘5′, respectively. Moreover, the e-value, percent identity, minimal alignment length, and minimal abundance values were set to ‘5′, ‘90′, ‘15′, and ‘1′, respectively. Additionally, ‘representative hit’ was selected. Statistical analyses were performed using MicrobiomeAnalyst (Dhariwal et al., 2017). Data normalization for each sample was performed for total sum scaling. Abundance profiling was presented as a stacked bar chart by calculating the percentage abundance, and less than 10 taxa were omitted. Species richness based on the alpha diversity of samples was determined via rarefaction curve analysis (McMurdie & Holmes, 2013), and the diversity of operational taxonomic units (OTUs) was indicated by the Shannon index. In addition, the evenness of OTUs was indicated by the pielou index (Jost, 2010). A heat tree was constructed using the non-parametric Wilcoxon rank sum test and was used to statistically quantify the hierarchical structure of taxonomic classification (Foster, Sharpton & Grünwald, 2017). The distance method and principal coordinate analysis (PCoA) was used for determining the beta diversity of samples and were set to unweighted UniFrac distance and permutational multivariate analysis of variance (PERMANOVA), respectively.

In addition, the distance measure and clustering algorithm of the hierarchical clustering algorithm (HCA) were set to the Bray–Cutis index and Ward, respectively. A heat tree was used to compare and sum the microbial communities and ultramicrobial communities for each sample. As a high-dimensional data analysis performed using a supervised machine learning algorithm, random forest classification analysis was conducted to identify the variability of strains in samples (Liaw & Wiener, 2002; Lee et al., 2022a).

Transmission electron microscopy (TEM)

For TEM observations via negative staining, droplets of the samples were mounted on a carbon support film on a 150-mesh nickel grid, stained with 4% uranyl acetate for 10 min and 0.4% lead citrate for 6 min, washed three times with deionized water, and air-dried. In addition, the samples were prepared in the form of ultrathin sections. To this end, samples were fixed by the addition of glutaraldehyde and paraformaldehyde adjusted to 2% in 0.05 M phosphate buffer (pH 7.2) and then incubated at room temperature (15–25 °C) for 4.5 h under vacuum. The fixed samples were washed three times for 15 min each with 0.05 M phosphate buffer at pH 7.2. The washed samples were post-fixed with osmium tetroxide adjusted to 1% in 0.05 M phosphate buffer (pH 7.2) at room temperature for 1 h. The post-fixed samples were washed three times for 15 min each with 0.05 M phosphate buffer at pH 7.2 and then dehydrated by passing through an ethanol gradient from 50 to 100%. After dehydration, the samples were precipitated to resin (LR white resin; EMS, Hatfield, PA, USA), placed in a disposable mold and embedded for 24 h at 60 °C. After the sample was hardened, ultrathin sections were prepared using an ultramicrotome equipped with a diamond knife and stained with 4% uranyl acetate for 10 min and 0.4% lead citrate for 6 min to complete sample preparation for TEM observation. The prepared samples were observed using a field-emission TEM (FE-TEM) (JEM-2100F; JEOL Ltd., Tokyo, Japan) at 200 kV accelerating voltage (Lee et al., 2022a).

Data availability

The sequencing reads of fermented cabbages were deposited to the NCBI under BioProject ID PRJNA684410, the metadata for each sample can be found in SRR13260135, SRR13260137, SRR13260138, SRR13260156, SRR13260177, SRR13260178.

SMRT sequencing

The NM and USM in fermented cabbage were separated using TFF and analyzed via SMRT sequencing. SMRT sequencing read quality indices are shown in Table S1. Sequence read information for each sample is presented in Table 1. A total of 19,356 sequence reads from Kimchi_NM, 1,383 from Kimchi_USM, 1,208 from Sauerkraut_NM, 1,610 from Sauerkraut_USM, 11,446 from Suancai_NM, and 1,570 from Suancai_USM were generated. The total number of reads from NM was much greater than that of USM. Notably, the number of reads from the Sauerkraut_USM was slightly higher than that from the Sauerkraut_NM. Kimchi_NM had the greatest number of sequence bases at 28,690,199 bp, while Sauerkraut_NM had the lowest at 1,741,058 bp. The average read lengths of the samples ranged from 1,425 to 1,482 bp, and were almost identical between groups (Lee et al., 2022a).

Taxonomic and statistical analyses

Community richness in the samples was expressed using the rarefaction curve (Fig. 1A), while alpha diversity was expressed using the Shannon index (Fig. 1B). Each sample in the rarefaction curve plateaued. Suancai_NM had the highest richness (OTUs of 84), while Sauerkraut_NM had the lowest (OTUs of 11). However, generally, the OTUs in Kimchi_NM and Suancai_NM were more abundant than those in the Kimchi_USM, Sauerkraut_USM, and Suancai_USM, but not than that in Sauerkraut_NM. Since the rarefaction curve of Kimchi_NM was gently inclined, richness was high, while diversity was low. Based on the Shannon index, Kimchi_NM had the lowest diversity at 0.26, while Kimchi_USM had 1.62. Comparing fermented cabbage types, Sauerkraut_NM had a diversity of 1.45, Sauerkraut_USM had 1.13, Suancai_NM displayed the highest diversity at 1.95, and Suancai_USM had 1.23. Suancai_NM had the highest microbial diversity, whereas Kimchi_NM had the lowest. The alpha diversity of suancai was higher than that of kimchi and sauerkraut when diversity was assessed between fermented cabbage types. Further, USM had higher alpha diversity on an average when diversity was assessed based on community type and regardless of the sample type. Based on the Pielou index, Kimchi_NM had the lowest evenness at 0.03, while Kimchi_USM displayed the highest evenness at 0.43. Comparing fermented cabbage types, Sauerkraut_NM had an evenness index of 0.31, Sauerkraut_USM had 0.23, Suancai_NM had 0.26, and Suancai_USM had 0.26. In terms of evenness among fermented cabbage types, sauerkraut and suancai showed relatively similar values. While assessing evenness based on community type and regardless of sample type, USM samples generally demonstrated higher evenness compared with NM samples, with the exception of sauerkraut, where NM samples had higher evenness than USM samples (Lee et al., 2022a).

The NM and USM in kimchi and suancai are shown in Fig. 2A and Fig. S3. At the phylum level, Firmicutes were dominant in Kimchi_NM and Suancai_NM. In the former, Firmicutes accounted for 100% of the microbial community. Actinobacteria dominated Kimchi_USM, and uncultured bacteria dominated Sauerkraut_USM as well as Suancai_USM. At the species level, Weissella koreensis was dominant at 94% in Kimchi_NM, while Weissella cibaria was also present, but in minor quantities. Cellulomonas uda and Cupriavidus pauculus were predominant in Sauerkraut_NM, at 32% and 26%, respectively. However, in Sauerkraut_NM, uncultured bacteria accounted for a significant proportion (32%). Lactobacillus acetotolerans was dominant (53%) in Suancai_NM. Lactobacillus similis was also predominant in Suancai_NM, accounting for 11%. Cellulomonas biazotea dominated at 42% in Kimchi_USM. In particular, candidate division TM7 single-cell isolate TM7a (TM7a), known as Saccharibacteria, was predominant at 35%. Cellulomonas uda predominated in Sauerkraut_USM at 36%. However, uncultured soil bacteria dominated Sauerkraut_USM at 54% and Suancai_USM at 48% (Lee et al., 2022a).

The OTU values indicated that the species richness of NM is generally higher than that of USM (Fig. 1). In particular, Suancai_NM exhibited the highest abundance and alpha diversity, as indicated by significantly higher OTU and Shannon index values of the microbiome compared to that of the other samples. However, the Pielou index value of NM was 0.26, which was relatively low compared with other samples, and was the same as that of USM. Therefore, Suancai’s USM and NM were almost equal in evenness, and evenness was inferior to that of Sauerkraut_NM with a Pielou index value of 0.31. For Kimchi_NM, the OTU value was relatively higher than that for other samples, however, its Shannon index value was the lowest. In addition, the rarefaction curve of Kimchi_NM only slightly increased, suggesting low diversity. The Shannon index was lower than the OTU value because Weissella koreensis showed > 94% dominance. In contrast, L. acetotolerans showed > 53% dominance in Suancai_NM, while other species showed minor dominance, between 1 and 11%. Therefore, the Shannon index of Suancai_NM was estimated to be the highest. In addition, The OTU values of USM were relatively low, with the one for Kimchi_USM being the lowest. However, Shannon index of Kimchi_USM was the second highest, > 1.6, possibly due to the even distribution of species. As proof of this, the Pielou index value of Kimchi_USM was 0.43, indicating higher evenness than other samples. While the difference in alpha diversity values between the NM and USM of kimchi and suancai was high, the difference in alpha diversity value between the NM and USM of sauerkraut was small. Since sterilization was performed during the manufacturing process of sauerkraut, most of the normal microorganisms did not survive. Therefore, the difference in alpha diversity between Sauerkraut_NM and Sauerkraut_USM was not expected to be high. The species richness of USM was low, yet the diversity was higher than that of NM. The microbial distribution of the fraction that passed through the 0.2-µm filter, which is sterile, was more diverse than expected (Nakai, 2020).

The beta diversity of NM and USM in fermented cabbage samples was compared via PCoA (Fig. 2B) and HCA (Fig. 2C). The plot and dendrogram showed that Suancai_NM and Kimchi_USM were closely related as also observed for Sauerkraut_NM, Sauerkraut_USM, and Suancai_USM. In addition, PCoA indicated that Kimchi_NM was separated from the other samples. However, Suancai_NM, Kimchi_USM, and Kimchi_NM were grouped in the HCA dendrogram, as were Suancai_USM, Sauerkraut_NM, and Sauerkraut_USM (Lee et al., 2022a).

A heat tree was used to compare and sum the NM and USM at the genus level for each fermented cabbage (Fig. 3). The sum of NM and USM per fermented cabbage type is shown in Figs. 3A–3C. Obtaining this sum was equivalent to combining the NM and USM of each fermented cabbage in the bar chart of Fig. 2A. Figures 3D–3F compare the NM and USM in each fermented cabbage. In kimchi, several genera belonging to the phylum Firmicutes had a relatively high ratio in the NM compared to the USM (Fig. 3D). Unlike in kimchi, in sauerkraut, several genera belonging to the phylum Proteobacteria had a relatively higher ratio in the NM compared to that in the USM (Fig. 3E). Suancai was similar to kimchi as several genera belonging to the phylum Firmicutes had a relatively higher ratio in the NM compared to that in the USM (Fig. 3F) (Lee et al., 2022a).

The random forest algorithm was used to confirm the top 15 OTUs with large variability between the microbiomes (Fig. 4). The mean reduced accuracy represents the accuracy that the microbiome loses by excluding each variable. Thus, the lower the accuracy, the more important the variable species is for a successful classification. Species are displayed in descending order of importance, that is, the higher the mean reduction accuracy, the higher the importance of the species in the microbiome (Martinez-Taboada & Redondo, 2020). Since the mean decrease accuracy of uncultured bacteria was the highest (0.0279), the microbiome was likely to be divided based on the content of these strains. The mean decrease in accuracy of uncultured Azospira sp. and Lactobacillus paracollinoides was 0.0231 and 0.0230, respectively, which were the second and third highest, respectively (Lee et al., 2022a).

Morphological observations via TEM

TEM was used to confirm the presence of microorganisms in the USM of fermented cabbage samples (Fig. 5 and Fig. S4). Most of the USM were cocci with both outer and inner membranes observed in the USM isolated from all fermented cabbages. In addition, the periplasmic space between the outer and inner membranes was observed (Figs. 5C, 5G and 5K). The size of the microorganism in USM was approximately 100–200 nm. In addition, USM isolated from fermented cabbages had multiplied via dichotomy (Fig. 5B, Figs. S4H and S4K) (Lee et al., 2022a).