Phylotype-Level Characterization of Complex Communities of Lactobacilli Using a High-Throughput, High-Resolution Phenylalanyl-tRNA Synthetase (pheS) Gene Amplicon Sequencing Approach

Species formerly classified within the genera Lactobacillus and Pediococcus have been studied extensively at the genomic level. To accommodate their exceptional functional diversity, the over 270 species were recently reclassified into 26 distinct genera. Despite their relevance to both academia and industry, methods that allow detailed exploration of their ecology are still limited by low resolution, high cost, or copy number variations. The approach described here makes use of a single-copy marker gene which outperforms other markers with regard to species-level resolution and availability of reference sequences (98% coverage). The tool was validated against a mock community and used to address diversity of lactobacilli and community structure in various environmental matrices. Such analyses can now be performed at a broader scale to assess and monitor the assembly, structure, and function of communities of lactobacilli at the species level (and, in some cases, even at the subspecies level) across a wide range of academic and commercial applications.

ABSTRACT

The lactobacilli identified to date encompass more than 270 closely related species that were recently reclassified into 26 genera. Because of their relevance to industry, there is a need to distinguish between closely related and yet metabolically and regulatory distinct species, e.g., during monitoring of biotechnological processes or screening of samples of unknown composition. Current available methods, such as shotgun metagenomics or rRNA gene-based amplicon sequencing, have significant limitations (high cost, low resolution, etc.). Here, we generated a phylogeny of lactobacilli based on phenylalanyl-tRNA synthetase (pheS) genes and, from it, developed a high-resolution taxonomic framework which allows for comprehensive and confident characterization of the community diversity and structure of lactobacilli at the species level. This framework is based on a total of 445 pheS gene sequences, including sequences of 276 validly described species and subspecies (of a total of 282, including the proposed L. timonensis species and the reproposed L. zeae species; coverage of 98%), and allows differentiation between 265 species-level clades of lactobacilli and the subspecies of L. sakei. The methodology was validated through next-generation sequencing of mock communities. At a sequencing depth of ∼30,000 sequences, the minimum level of detection was approximately 0.02 pg per μl DNA (equaling approximately 10 genome copies per μl template DNA). The pheS approach, along with parallel sequencing of partial 16S rRNA genes, revealed considerable diversity of lactobacilli and distinct community structures across a broad range of samples from different environmental niches. This novel complementary approach may be applicable to industry and academia alike.

IMPORTANCE Species formerly classified within the genera Lactobacillus and Pediococcus have been studied extensively at the genomic level. To accommodate their exceptional functional diversity, the over 270 species were recently reclassified into 26 distinct genera. Despite their relevance to both academia and industry, methods that allow detailed exploration of their ecology are still limited by low resolution, high cost, or copy number variations. The approach described here makes use of a single-copy marker gene which outperforms other markers with regard to species-level resolution and availability of reference sequences (98% coverage). The tool was validated against a mock community and used to address diversity of lactobacilli and community structure in various environmental matrices. Such analyses can now be performed at a broader scale to assess and monitor the assembly, structure, and function of communities of lactobacilli at the species level (and, in some cases, even at the subspecies level) across a wide range of academic and commercial applications.

INTRODUCTION

The generic term “lactobacilli” encompasses all organisms described as belonging to the family Lactobacillaceae until 2020, a highly diverse range of bacteria within the phylum Firmicutes (1). Species within this group are Gram-positive, facultative anaerobic or microaerophilic, rod-shaped, non-spore-forming bacteria. Due to their beneficial effects on food during fermentation, lactobacilli play an important role for the food industry and have thus been studied extensively both for their phenotypic and genomic characteristics over the last decades (2). Several species have received generally recognized as safe (GRAS) status for use in human food and animal feed. Besides their naturally high abundances in fermented foods, they are frequently observed as part of the healthy microbiomes of animals, humans, and plants (2–4). Formerly, this group consisted of only three genera, Lactobacillus, Paralactobacillus, and Pediococcus. However, these traditionally defined genera alone represent a range of genetic diversity wider than that of a typical bacterial family (5, 6). A tremendous recent effort of reclassification of the lactobacilli resulted in the establishment of 26 genera (encompassing 272 species plus the proposed species L. timonensis) and reduction of the original genus Lactobacillus to only 38 species around its type species, Lactobacillus delbrueckii (1). This refined taxonomic structure will in future facilitate the detection and description of functional properties that are shared within each genus.

In the past 2 decades, numerous studies aimed at elucidating the bacterial communities involved in the fermentation of specific traditional foods or beverages by using both culture-dependent and culture-independent methods (reviewed in references 7 and 8). These communities are most often dominated by lactobacilli. While cultivation-based approaches allow isolation and detailed study of individual strains, they are biased toward microorganisms that are culturable under laboratory conditions (9). The molecular techniques used for bacterial community structure analysis of fermented-food samples so far have largely relied on amplification and (high-throughput) sequencing of 16S rRNA marker genes (recently reviewed in reference 10). Comparatively few studies have used shotgun metagenomic sequencing (10–12) despite the availability of highly resolving bioinformatics tools (13), likely because this approach is still rather costly at a sequencing depth that allows for confident species or even strain level taxonomic classification and, instead, is more suited to derive functional information (14). Furthermore, the accuracy of taxonomic assignment relies on the availability of a curated genome database and of at least one representative genome for every species present in the sample. In the case of the lactobacilli, genome-based taxonomies have been published and updated on a regular basis (5, 6, 15–17). However, new species within this group are even more frequently discovered and described (1). While draft genomes may not be immediately available, deposition of corresponding marker gene sequences in public databases for reference is required for the valid description of new species. The 16S rRNA gene is a widely accepted marker gene used to analyze bacterial and archaeal community diversity and structure in many habitats and to understand evolutionary distances between species. Sequence databases and taxonomic frameworks for bacterial and archaeal 16S rRNA genes such as Greengenes and SILVA are highly curated (18–20) and compatible with next-generation-sequencing bioinformatics pipelines such as mothur (14, 21) or QIIME/QIIME2 (22, 23). However, at the 16S rRNA gene level and even if almost the entire 16S rRNA gene (∼1,500 bp) is analyzed, different type species of lactobacilli display sequence similarities greater than the accepted species cutoff of 98.7% and up to 100% sequence identity across the shorter regions commonly used for next-generation amplicon sequencing protocols (24, 25). In consequence, high-throughput 16S rRNA gene amplicon sequencing provides insufficient taxonomic resolution for understanding the composition of communities of lactobacilli in complex environmental samples (26, 27). It is reasonable to speculate that the studies that have used partial 16S rRNA genes as a marker so far merely touched the surface of the vast lactic acid bacterial diversity that may exist in host-associated and fermented-food-associated matrices and that contribute to their potential beneficial effects. If the aim is not to derive evolutionary relationships but to obtain fine-scale species-level or even strain-level structural information for a particular subcommunity, then other marker genes are available that provide greater taxonomic resolution. Recently, the use of lacS and serB genes for highly specific strain-level monitoring of Streptococcus thermophilus during cheese manufacturing was explored (28–30). For species-level characterization of complex communities of lactobacilli across a large number of samples from different matrices, the use of the internal transcribed spacer (ITS) region (31) or the groEL gene (32) has been suggested by different groups. The ITS region has the advantage of harboring highly conserved primer binding sites due to the presence of flanking 16S and 23S rRNA genes. However, difficulties encountered in efforts to extract ITS sequence information from publicly available (draft) genomes have hampered the taxonomic assignment of the majority of species of lactobacilli, and the use of a multicopy gene region such as the ITS presents an additional challenge (31). Moreover, the ITS region and the groEL gene do not seem to provide a clear separation between several closely related species (e.g., species belonging to the L. casei, L. plantarum, L. sakei, L. delbrueckii, and L. buchneri phylogroups as defined earlier [1, 6]).

One of the best-characterized marker genes of lactobacilli is the phenylalanyl-tRNA synthetase (pheS) gene. For the genera Enterococcus, Lactobacillus, and Pediococcus, Naser et al. (26, 33) reported a pheS-based interspecies gap which, in most cases, exceeded 10% sequence dissimilarity and an intraspecies variation of up to 3% across an alignment consisting of approximately 450 bp (26, 33). Currently available next-generation-sequencing chemistry easily achieves full coverage of amplicons of this size, thus allowing researchers to make full use of the resolving power of the pheS gene. These characteristics, along with the general practice of depositing the pheS gene sequence along with valid descriptions of novel species, make the pheS gene a highly suitable candidate for analysis of diversity and composition of lactobacilli in fermented foods or other host-associated or non-host-associated environmental samples.

Here, we introduce a new approach for high-throughput next-generation sequencing based on the use of pheS marker genes and a curated pheS-based taxonomic framework to characterize the diversity and structure of communities of lactobacilli down to the species level. We demonstrate the validity of this method by analysis of mock communities of lactobacilli (containing DNA of 13 different species of lactobacilli belonging to seven different genera, which differed in the known relative amounts of each taxon). The robust taxonomy allowed us to assess the community structure of lactobacilli in samples from complex host- and food-associated environments at high throughput and high resolution.

RESULTS AND DISCUSSION

Evaluation of pheS primer binding sites and coverage within the lactobacilli.

Here, we assessed the pheS gene as an alternative marker gene for high-resolution profiling of lactobacilli. Primer pairs pheS21F/pheS22R and pheS21F/pheS23R had been reported previously to successfully target a wide diversity of Enterococcus species and of species formerly assigned to the Lactobacillaceae in vitro (26, 33). Since the time of those earlier publications, numerous additional species of lactobacilli have been validly described. While the amplicon of primer pair pheS21F/pheS22R was considerably longer (494 bp), the use of primer pair pheS21F/pheS23R resulted in an amplicon of 431 bp in length, which makes it suitable for high-throughput next-generation sequencing performed with currently available technologies (Fig. 1A and B). In view of the increased diversity of lactobacilli that has been unraveled in recent years, we aimed to reevaluate sequence conservation along the primer binding regions and coverage of the primers among the target species. The pheS gene region was excised from a total of 266 type strains of lactobacilli for which genomes are publicly available (272 of 282 type strains, including all subspecies and the proposed species L. timonensis; for Bombilactobacillus bombi, the type strain genome was unavailable and that of strain BI-2.5 was used instead). Extracted pheS genes were aligned, and representations of the consensus sequences along the primer binding regions were visualized through the use of sequence Web logos (34) (Fig. 1B). As expected, the degrees of conservation differed between primer binding positions. For pheS21F, none of the tested species showed >2 mismatches with the primer sequence, suggesting comprehensive coverage for the target community. Importantly, primer pheS21F had no mismatches with the crucial five consecutive bases counted from the 3′ end of the primer sequence. A higher degree of sequence variability was observed for the binding region of primer pheS23R (Fig. 1B), and 9% of the sequences showed a mismatch at position 3 from the 3′ end. However, the majority (82%) of type strains of lactobacilli had ≤2 mismatches with the primer sequence. In future, it may be possible and desirable to increase the degeneracy of the primer sequences to test if the range of coverage is more extensive than it is currently thought to be. A higher degree of degeneracy, however, may result in more nonspecific primer binding and may need to be balanced by the use of deeper sequencing to achieve similar coverage of lactobacilli.

FIG 1.

Primary structure of the complete phenylalanyl-tRNA synthetase subunit alpha (pheS) and binding sites of the primers used in this study. (A) Protein sequence of the Lactobacillus type species, L. delbrueckii subsp. delbrueckii DSM 20074, with amino acid numbering. Approximate binding sites of primers pheS21F and pheS23R are indicated. (B) Graphical map of the complete pheS nucleotide sequence depicting the positions of primer binding and PCR amplicon length. Primer sequence conservation was tested across 266 type strain sequences of lactobacilli and is represented through the use of a sequence Web logo. The overall height of each stack indicates the sequence conservation at that position, the height of symbols within each stack indicates the relative frequency of nucleic acids at that position, and error bars indicate approximate Bayesian 95% credible intervals. Black arrows denote positions at which mismatches to the primers were observed.

pheS gene phylogeny of the lactobacilli.

Our comprehensive pheS phylogeny encompasses sequences from 277 of the total 283 species and subspecies of lactobacilli validly described by March 2020 (coverage of 98%) (Fig. 1; see also Table S1 in the supplemental material). Representative pheS gene sequences of the remaining six type species could not be derived in this study either, because efforts to perform in silico PCR and BLAST analyses of publicly available draft genomes were unsuccessful (2 type species) or because the genome of a particular described species has not yet been sequenced or deposited in public databases (4 type species; Table S1 in the supplemental material). Our analysis revealed several inconsistencies in the species assignments of previously deposited pheS gene sequences. Thus, seven previously published sequences were reclassified based on determinations of pheS gene sequence similarities to the type strain of the previous species assignment and to other closely related type strains (Table 1; see also Fig. 2).

TABLE 1.

Previous species assignment and revised species assignment of several pheS gene sequences of lactobacilli retrieved from GenBank^a

Accession no.	Previous assignment	% similarity to type strain (previous assignment)	Reclassified assignment	Similarity to type strain (reclassified assignment)
AM087760	Lactobacillus murinus	91.3	Ligilactobacillus animalis	98.3%
AM087771T	Lactobacillus homohiochii		Fructilactobacillus fructivorans	100%
AM284237	Lactobacillus gasseri	95.4	Lactobacillus paragasseri	100%
AM284241	Lactobacillus gasseri	95.4	Lactobacillus paragasseri	100%
AM284196	Lactobacillus gasseri	95.6	Lactobacillus paragasseri	99.3%
AM284194	Lactobacillus gasseri	95.4	Lactobacillus paragasseri	99.1%
AM284246	Lactobacillus farciminis	93.6	Companilactobacillus formosensis	99.1%

^aFor the “previous assignment,” the original genus name “Lactobacillus” was kept for clarity, while the reclassified species-level assignments include the new genus names according to Zheng et al. (1).

FIG 2.

Phylogenetic tree showing 265 species-level clades derived from a total of 445 phenylalanyl-tRNA synthetase (pheS) gene sequences of lactobacilli. The backbone, consisting of 384 sequences (alignment positions 487 to 884), was calculated in ARB using the neighbor-joining algorithm with Jukes-Cantor correction. The remaining 61 partial sequences (alignment positions 556 to 802) were added using the Fast Parsimony tool. The scale bar indicates 0.1 nucleotide substitution per nucleotide position. For clarity, some coherent groups of sequences are indicated only as triangles or trapezoids, with the number of sequences clustering into these groups given in parentheses behind the clade name. Sequences in green were derived from isPCR, while those in blue were obtained by BLAST-based extraction from the draft genomes. The pheS gene sequences of 16 Enterococcus faecium strains (including the LMG 11423^T type strain; GenBank accession number AJ843428) served as the outgroup. New genus-level affiliations and former Lactobacillus and Pediococcus phylogroups (in brackets) are indicated to the right of each clade. Generally, the former phylogroup name matches exactly the type species of the new genus. However, in the case of the former “algidus” phylogroup, the type species has been renamed Dellaglioa algida. Genus affiliations of sequences belonging to Ligilactobacillus (Lg.) and Liquorilactobacillus (Lq.) are abbreviated with two letters for clarity.

The average observed level of pheS gene interspecies sequence similarity was 69.7% ± 4.9% (± standard deviation), with the interspecies gap of neighboring species exceeding 10% in most cases. The average intraspecies sequence similarity was 99.2% ± 1.6% (StDev), which is comparable to the intraspecies dissimilarity of up to 3% reported earlier (33). Most type strains shared <97% pheS gene sequence similarities. The chosen arbitrary threshold allowed for clear species calling for all but six validly described species: L. amylovorus and L. kitasatonis shared 98.5% pheS gene sequence similarity, while Apilactobacillus micheneri and A. timberlakei shared 98.9% pheS gene sequence similarity (Table 2). Here, we named the clades “L. amylovorus/kitasatonis” and “A. micheneri/timberlakei” (Table 2). On the basis of our analyses, the species Companilactobacillus mishanensis and C. salsicarnum shared 100% 16S rRNA gene sequence similarity and 100% pheS gene sequence similarity and showed an average nucleotide identity (ANI) of 99.9%. Since both species were described between October 2019 (35) and November 2019 (36), they may not have been compared at the genomic level prior to this study. Hence, it is likely that the C. salsicarnum species characterized by Schuster et al. in 2019 (36) represents a later heterotypic synonym of the C. mishanensis species characterized by Wei et al. in 2019 (35). Here, we assigned the clade name “C. mishanensis.”

TABLE 2.

Sequence similarities based on pheS genes (pheS-based) and whole genomes (ANI-based) between closely related strains of different species or distantly related strains within the same species^a

(Type) strain(s) present in clade	Given clade name	% sequence similarity between strains
		pheS-based	ANI-based
A. micheneri DSM 104126T	A. micheneri/timberlakei	98.9	94.3
A. timberlakei DSM 104128T	A. micheneri/timberlakei
C. mishanensis LMG 31048T	C. mishanensis	100	99.9
C. salsicarnum LMG 31401T	C. mishanensis
L. amylovorus DSM 20531T	L. amylovorus/kitasatonis	98.5	92.4
L. kitasatonis JCM 1039T	L. amylovorus/kitasatonis
C. crustorum LMG 23699T	C. crustorum	89.9	Only type strain sequence available
C. crustorum LMG 23702	C. crustorum 2
L. curieae CCTCC M 2011381T	L. curieae	84.4	83.8
L. curieae NBRC111893	L. curieae 2

^aClade names refer to the species-level (L7) assignment as provided in pheS-DB. The same clade name is given if type strains of the species cannot be confidently distinguished based on pheS genes. Species-level clades are divided into subclades, if strains of the same species are only distantly related to the type strain (pheS gene sequence similarity <90%, ANI <95%). For the two C. crustorum strains, only the type strain genome is publicly available, and no comparison can be done at the genomic level.

The recent reproposal of the species L. zeae (37) is in line with our findings: L. zeae LMG 17315T could be confidently distinguished from the most closely related type strains, L. casei LMG 6904T and L. chiayiensis BCRC 81062T, using our pheS taxonomic framework database (pheS-DB). Our pheS gene sequence analysis further revealed that strain NBRC 111893, previously deposited as Lentilactobacillus curieae NBRC 111893, shared only 84.4% pheS gene sequence similarity with the type strain L. curieae CCTCC M 2011381^T (pheS-DB clade name “L. curieae”) (38) and an ANI of only 83.8% (Table 2). This suggests that strain NBRC 111893 may represent a novel species. Hence, it was given the clade name “L. curieae 2”. Similarly, pheS gene sequence analysis confirmed a previous observation by Scheirlinck and colleagues that Companilactobacillus crustorum strain LMG 23702 is considerably different from the type strain C. crustorum LMG 23699^T (≤89.9% pheS gene sequence similarity) (39). Unfortunately, no genome is available for strain LMG 23702 for analysis at the genomic level. Until more data become available, we allow the pheS approach to differentiate between the members of the C. crustorum clade that includes the type strain (“C. crustorum”) and strain LMG 23702 (“C. crustorum 2”). In future, a polyphasic approach should be used to characterize strains NBRC 111893 and LMG 23702 at the genomic and physiological levels, thereby providing clarity on their potential roles as the type strains of novel species.

Previously, seven species of lactobacilli were subdivided into subspecies. Two species formerly classified as subspecies of Ligilactobacillus aviarius (L. aviarius subsp. aviarius and L. aviarius subsp. araffinosus) have now been reclassified into different species, namely, L. aviarius and L. araffinosus (1, 17). This split was supported by ANI data as well as pheS gene analysis, in which the genomes and pheS genes of the type strains shared only 90% and 94.2% sequence similarity, respectively. The two subspecies of Latilactobacillus sakei, L. sakei subsp. sakei and L. sakei subsp. carnosus, shared only 93.2% pheS gene sequence similarity. Similarly, the two subspecies of Lactiplantibacillus plantarum, L. plantarum subsp. plantarum and L. plantarum subsp. argentoratensis, shared only 91.7% pheS gene sequence similarity. While L. plantarum subsp. argentoratensis was indeed elevated to species level recently (40), ANI analysis of the two L. sakei subspecies confirmed the standing of L. sakei subsp. carnosus as a subspecies (ANI: 97.3%). The classification of the species L. sakei can be resolved to the subspecies level by our pheS taxonomic framework. For the remaining subspecies containing lactobacilli, namely, Loigolactobacillus coryniformis (subsp. torquens), Lactobacillus delbrueckii (subsp. bulgaricus, subsp. indicus, subsp. jacobsenii, subsp. lactis, and subsp. sunkii), Lactobacillus kefiranofaciens (subsp. kefirgranum), and Lacticaseibacillus paracasei (subsp. tolerans), all subspecies within the respective species share ≥97% pheS gene sequence similarity, supporting their standing in the nomenclature as subspecies. These cannot be resolved to the subspecies level by our current pheS taxonomic framework. In summary, pheS-DB allows differentiation between 265 species-level clades and L. sakei subsp. carnosus (Fig. 2).

The lactobacilli were previously divided into 24 phylogroups based on sequence information representing the concatenated protein sequences of single-copy core genes of 174 type strains (6). A later whole-genome-based taxonomy of the Lactobacillus genus complex represented a total of 208 Lactobacillus and Pediococcus species, all of which clustered within the 24 phylogroups (17). In a recent keystone study that followed an analysis that used a comprehensive polyphasic approach, one additional phylogroup (genus Acetilactobacillus) was detected through the inclusion of the latest validly described species, and one phylogroup (L. salivarius group) was split into two (1). All 26 phylogroups were then converted into genera with unique genus names (1). In our analysis, a reference sequence for Acetilactobacillus jinshanensis, the type species of the genus, is missing. Hence, the 265 species-level clades and 9 subspecies that were included in the pheS phylogenetic tree in this study fell into 25 of the 26 genera. All genera formed monophyletic clades apart from the following three exceptions. (a) The sequence of Paucilactobacillus kaifaensis clustered basally to the clade containing the genera Paucilactobacillus, Limosilactobacillus, Lactiplantibacillus and Furfurilactobacillus. (b) The genus Schleiferilactobacillus (former perolens phylogroup) formed a cluster within the genus Lacticaseibacillus (former casei phylogroup). (c) The genera Secundilactobacillus (former collinoides phylogroup) and Ligilactobacillus (former salivarius phylogroup) were paraphyletic, with three species of Ligilactobacillus (L. hayakitensis, L. ruminis, and L. salivarius) clustering into the Liquorilactobacillus clade (also former salivarius phylogroup; Fig. 2). These difficulties may result from the fact that pheS sequences covering the entire amplicon length were not available for all species. The information derived from shorter sequences does not seem to be sufficient in all cases to group the respective species into the correct genus-level clades. This underpins the necessity for whole-genome-based approaches to define existing and erect new genera within the lactobacilli. However, it does not hamper the use of the pheS gene to assign sequence data to particular species.

Development of a taxonomic framework (pheS-DB).

On the basis of our pheS gene phylogeny of lactobacilli, we compiled a comprehensive pheS database (pheS-DB). PheS-DB comprises a sequence .fasta file with a total of 445 pheS gene sequences encompassing 265 species-level clades of lactobacilli (File S1 in the supplemental material) and the corresponding taxonomy .txt file (File S2 in the supplemental material). In addition, we included reference sequences and taxonomic strings of the type strains of the genera Bacillus, Enterococcus, Fructobacillus, Lactococcus, Leuconostoc, Oenococcus, Streptococcus, and Weissella due to their frequent occurrence in fermented foods and beverages and the likelihood that these species would be captured by the pheS primers.

This framework is compatible with commonly used next-generation amplicon sequencing pipelines such as mothur (21) or QIIME2 (23) and can be used for taxonomic assignment of pheS gene sequence data. The taxonomy .txt file comprises a total of eight taxonomic levels: domain, phylum, class, order, family, genus, species (and, in some cases, subspecies, as described above), and strain (see File S2). We recommend its use at the species/subspecies level (pass option “-L 7” with QIIME script “qiime taxa collapse”) for high-resolution community structure analysis of lactobacilli.

Evaluation of detection sensitivity and accuracy of taxonomic assignment through analysis of mock communities.

High-throughput 16S rRNA and pheS gene sequencing of two synthetic (mock) communities allowed us to validate the results obtained with the newly developed framework of lactobacilli. For these two samples, after quality filtering, averages of 20,261 ± 11,603 and 32,285 ± 9,016 paired-end sequence reads were obtained for the 16S rRNA and pheS genes, respectively.

While the use of the 16S rRNA gene allows coverage of the vast majority of bacterial biodiversity detected to date, non-full-length 16S rRNA gene sequences are generally accurate only for taxonomic assignment down to the genus level (25). In our study, on the basis of analyses performed with 16S rRNA genes, only 3 of a total of 13 species that comprised the mock community were identified correctly at the species level (Lactiplantibacillus plantarum, Levilactobacillus brevis, and Pediococcus acidilactici). At the genus level, of the 7 major genera represented in the two samples, 5 genera, Lacticaseibacillus, Lactiplantibacillus, Lactobacillus, Levilactobacillus, and Pediococcus, were correctly identified by 16S rRNA gene profiling, while 2 genera, Lentilactobacillus and Limosilactobacillus, remained unidentified. PheS gene sequence data revealed a considerably higher level of community structure resolution. All 13 species included in the artificial community were successfully detected and correctly identified, including even those that were spiked into the mock community at levels as low as 0.02 pg per μl DNA. Assuming an average genome size of 2.2 Mbp (3), this detection limit corresponds to ∼10 genome copies per μl template DNA. Closely related species such as L. argentoratensis, L. plantarum, and L. pentosus were successfully distinguished. For species that were represented in the sample at a relative abundance of ≥0.1%, sequence data showed relative abundances that corresponded well with their expected abundances in the mock community (Fig. 3). Below this threshold, species tended to be overrepresented. This observed deviation from that expected for low template concentrations may have been caused, e.g., by primer bias and preferential amplification of certain DNA templates. As this trend was observed independently of the species, another possible explanation is that pipette error more severely affected highly diluted DNA of individual species of lactobacilli and thus may have caused consistent overrepresentation of those species that were added into the mock communities at very low abundances. Importantly, primer bias, if present, is expected to similarly affect samples that have been treated in the same way. Under these circumstances, while the relative abundance data may not be exact, the presence or absence of genera and species, as well as trends in the relative abundances of genera and species between different samples (β-diversity), can be determined reliably. In future, the use of mock communities of lactobacilli cells rather than DNAs may provide better insights into the assessment of complex communities and into potential biases (e.g., DNA extraction bias) influencing these analyses.

FIG 3.

Evaluation of accuracy of the pheS gene sequencing and taxonomic assignment approach using two individual mock communities of lactobacilli (STD1 and STD2) consisting of DNAs of 13 different species, each at different proportions (Table S2 in the supplemental material). The dotted black line indicates the correlation between the expected relative abundance and the observed relative abundance, while the dotted gray line indicates the theoretically expected perfect correlation. Only species that were represented in either of the mock communities at ≥0.1% relative abundance are shown in the graph.

Comparative analysis of diversity and community structure of lactobacilli in environmental samples based on pheS and 16S rRNA marker genes.

To demonstrate the application of our high-resolution pheS-based community profiling method, we collected 24 environmental samples from a variety of fermented foods as well as different host-associated matrices. Our aims were, first, to identify samples with greatly differing communities of lactobacilli for the purpose of isolation, and second, to assess similarities and differences between the community structures of lactobacilli in similar food matrices. The samples were analyzed by using both 16S rRNA and pheS marker gene amplicon sequencing. After filtering, denoising, merging, and chimera removal using DADA2, a total of 723,525 paired-end 16S rRNA gene sequences (30,147 ± 13,339 sequences per sample [average ± standard deviation]) and 810,805 paired-end pheS gene sequences were obtained (33,784 ± 11,448 sequences per sample) across the 24 environmental samples (Table S3 in the supplemental material). Analysis of 16S rRNA gene sequence data resulted in complete bacterial community profiles obtained from all environmental samples. Members of the lactobacilli accounted for an average of 26.8% ± 15.5% of total 16S rRNA gene sequence reads across all samples. Of these, an average of 56.0% ± 11.2% achieved species-level assignment; however, the accuracy of these species-level assignments may be compromised as discussed above. In the pheS approach, lactobacilli made up an average of 55.1% ± 24.3% of total sequence reads across all samples, 100% of which achieved species-level assignment. An average of 35.5% ± 23.8% of pheS gene sequence reads remained unassigned (represented by 1,083 representative sequences). Of these representative sequences, 78.2% could be assigned at least to the genus level via the use of Kraken2 (Fig. S2 in the supplemental material), with the highest proportions of representative sequences observed for the genera Enterobacter (14.4%), Bacillus (6.1%), and Lactococcus (4.4%). Only 45 (4.2%) representative sequences were assigned to the lactobacilli by Kraken2. Upon closer inspection of these sequences by BLAST and alignment, we found that 21 represented true pheS sequences, while the remaining 24 matched other genomic regions and likely resulted from unspecific primer binding. Nonspecific binding to pheS genes of nontarget organisms or to other genomic regions of target or nontarget organisms reduces the number of total pheS sequence reads per sample and thus should be considered when deciding on sequencing depth. Importantly, however, the 21 representative sequences of true pheS gene sequences of lactobacilli that remained unassigned with pheS-DB together represented sequences that accounted for only 0.2% and 0.07% of unassigned and total sequence reads, respectively. As such, any pheS-DB unassigned sequences did not contribute noticeably to the apparent community structure of lactobacilli in the environmental samples and were excluded from subsequent analyses in this study.

Results obtained from pheS gene profiling were compared to profiles reconstructed via partial 16S rRNA gene amplicon sequencing at both the genus level (Fig. 4) and the species level (Fig. S1 in the supplemental material). Importantly, only reads classified as belonging to the lactobacilli were included in this analysis. Reads that were classified into a designated species within the lactobacilli but that accounted for a relative abundance of <1% in all samples were collapsed as “lactobacilli <1%.” Reads that were classified to the genus level only (16S rRNA gene data only) were collapsed as “lactobacilli, other.”

FIG 4.

Genus-level community structure of the lactobacilli based on the pheS gene (left) and the 16S rRNA marker gene (right) in samples from five broad environments. Bar plots show the relative abundances of lactobacilli genera in (A) 3 animal-associated fecal or jejunum (ANIM) samples, (B) 3 dairy and nondairy alternative (DNDA) products, (C) 3 fermented vegetable (FVEG) samples, (D) 5 fermented tofu (FTOF) samples, and (E) 10 tempeh (TEMP) samples. Only genera with a relative abundance of ≥1% in at least 1 of the 24 samples are reported. Designated genera that showed relative abundances of <1% in all samples are collapsed as “Lactobacilli <1%,” while sequences that could not be assigned at the genus level or species level are collapsed as “Lactobacilli, other.”

The pheS approach allowed all sequence reads of lactobacilli to be assigned down to the species level. Furthermore, the number of detected taxa that accounted for ≥1% in at least one sample according to the Chao-1 index on the basis of pheS analysis was significantly higher than that obtained on the basis of 16S rRNA gene analysis (12 versus 7 [taking into account undefined clades], P < 0.001). Across all analyzed samples, a total of 19 different species of lactobacilli (≥1% in at least one sample) were detected by pheS gene profiling compared to only 5 by 16S rRNA gene profiling (Fig. 5). Beta-diversity as calculated with the Bray-Curtis metric showed that communities of lactobacilli were generally less similar between samples based on pheS gene sequence data (within-method similarity 29.3% ± 23.5%, average ± standard deviation) than on 16S rRNA gene sequence data (within-method similarity 63.3% ± 13.8%, average ± standard deviation, P < 0.001). Between-method similarity was low at only 13.0% ± 10.9% (average ± standard deviation). The differences between methods were corroborated by sample clustering using correlation distance and average linkage (Fig. 5) and may be attributed to (i) the limited number of species of lactobacilli that are distinguishable based on the partial 16S rRNA gene compared to the higher-resolving pheS gene and (ii) the lower number of species of lactobacilli that are represented by 16S rRNA gene sequences in the SILVA database.

FIG 5.

Heat map and dendrogram representation of sample similarity within and between methods based on 16S rRNA and pheS gene community structure analysis of lactobacilli at the species level. Samples are labeled according to the data listed in Table 3.

The heat map representation, in particular, of the samples analyzed with the pheS-based methodology, allowed us to identify samples that contained the largest proportions of a particular species. This may be useful for the purpose of isolating additional and potentially novel strains from the analyzed samples (Fig. 5).

Furthermore, principal-component analysis (PCA) revealed a strong influence of sample type on clustering of tempeh (TEMP) samples, with fermented vegetable (FVEG) samples also grouping relatively closely together (Fig. 6). Interestingly, the remaining sample types, including animal-associated fecal or jejunum (ANIM) samples, dairy and nondairy alternative (DNDA) products, and fermented tofu (FTOF) samples, were distributed across two clusters (Fig. 5; see also Fig. 6). Cluster 1 was characterized by the prevalence of L. plantarum and included two fermented tofu samples, one kefir sample, one wild boar sample, and one guinea pig fecal sample. Samples in cluster 2 included three fermented tofu samples, one mouse jejunum content sample, one coconut milk sample, and one Greek yogurt sample. This cluster was characterized by the presence of L. paracasei in all six samples and relatively high abundances of Lactobacillus pobuzihii, L. johnsonii, or L. acidophilus. To some degree, these samples shared similar communities of lactobacilli; however, the broad clustering observed here may have been a result of the small sample size per sample type. Taking the data together, the pheS approach allowed confirmation of populations that had previously been reported from specific sample types, as well as identification of populations of lower but physiologically relevant relative abundances that had not previously been linked to particular intestinal environments or food fermentation processes.

FIG 6.

Principal-component analysis of the community structure of lactobacilli in all analyzed samples based on pheS genes. Sample identifiers refer to Table 3, with symbols representing the five different sample types as follows: ●, ANIM; ■, DNDA; ♦, FVEG; □, TEMP; ▴, FTOF.

With third-generation technology becoming more available and affordable, high-throughput sequencing of full-length 16S rRNA genes may soon be achievable routinely (25). However, sophisticated denoising algorithms are required to correct for PCR and sequencing errors. Moreover, in the case of some species of lactobacilli, even full-length 16S rRNA gene sequencing may not be able to resolve sequences at the species level (41). The pheS approach may be improved in future by extending primer coverage and by obtaining pheS gene sequences for the remaining six type strains, since the lack of availability of these missing species currently renders them elusive with respect to detection and relative quantification in environmental samples. Despite these limitations, our new pheS gene sequencing approach, combined with taxonomic assignment through the use of pheS-DB, provides highly accurate and highly resolved species-level reconstruction of communities of lactobacilli and may complement 16S rRNA gene profiling of a broad range of complex environmental and host-associated matrices.

Conclusions.

The newly established pheS gene sequencing approach, together with taxonomic assignment by pheS-DB, provides comprehensive and highly resolving community structure information representing lactobacilli. The framework currently enables exact species-level taxonomic assignment of a total of 265 species of lactobacilli and L. sakei subsp. carnosus. The pheS gene amplicon sequencing approach is highly sensitive, with a limit of detection of approximately 0.02 pg per μl DNA (corresponding to ∼10 genome copies). We applied the method to 24 diverse samples obtained from host-associated environments, dairy and nondairy alternative products, and fermented Asian soybean products (tempeh and tofu), as well as from fermented vegetables, and were able to evaluate the diversity and community composition of the resident lactobacilli at great depth. As such, the newly described methodology represents a valuable tool for understanding the diversity of lactobacilli in a broad range of sample types. In future, it may also be used to monitor communities of lactobacilli during industrial fermentation processes, where it could potentially assist in linking distinct populations to desirable or nondesirable food or feed rheological and organoleptic properties or metabolites.

MATERIALS AND METHODS

In silico assessment of variability and coverage of pheS gene primers.

For assessment of primer binding site variability and primer coverage, draft genomes of 266 species and subspecies of lactobacilli were downloaded from NCBI (Table S1 in the supplemental material). Available pheS sequences were used for BLAST searches of these genomes for candidate regions with an E value of 5 to filter spurious hits. Subsequently, these regions were extended by 400 bp at either end and excised from the genomes for subsequent primer binding site analysis. The obtained sequences were aligned in MUSCLE (42) and subjected to the WebLogo 3 tool for visualization of sequence variability of the pheS gene along the primer binding sites (34). The percentage of primer coverage was calculated from the alignment by establishing the number of mismatches of each type strain with the primer sequences.

Establishment of pheS gene sequence database and taxonomic framework.

A total of 445 pheS gene sequences were collected for this study. They were either directly downloaded from NCBI or obtained in this study via in silico PCR or via BLAST searches against whole-genome sequences of type strains. These sequences cover the type strain sequences of 267 Lactobacillus and Pediococcus species and 9 subspecies. Of these, 162 were available from previous publications; however, 3 were replaced with sequences derived by in silico PCR of the type strains (this study; see also Table S1 in the supplemental material). Reference sequences from a total of 114 species were previously unavailable and were obtained in this study by using in silico PCR (78 sequences, not including that of L. curieae 2) or BLAST searches (36 sequences) against the genomes of the type strains. Briefly, whole-genome assemblies of the type strains were downloaded from NCBI. In silico PCR was performed using Cutadapt software (43) and primers as described earlier (26). Species for which the in silico PCR approach failed were subjected to BLAST searches and extraction of the pheS sequence of the closest relative at the whole-genome level (if known) or at the marker gene level by using an in-house Python script. Sequences of the six remaining type species could not be obtained in this study (Table S1 in the supplemental material). All 445 pheS gene sequences were imported into the ARB software package (44) and aligned. A phylogenetic tree that included 384 sequences (pheS gene alignment of positions 487 to 884) was constructed using the neighbor-joining method and Jukes-Cantor correction. The remaining 61 sequences were added to the tree using the FastParsimony tool in ARB (pheS gene alignment positions 556 to 802). Type strain sequences that shared ≥97% sequence similarity were regarded as nonresolvable based on pheS and were given the same species-level clade name. Non-type strain sequences that shared ≥90% pheS gene sequence similarity with the most closely related type strain were assigned to the same species-level clade. However, the sequences are still individually traceable based on their unique strain-level identifiers (i.e., their accession numbers). Non-type-strain sequences that shared <90% pheS gene sequence similarity with the most closely related type strain were subjected to further investigations to assess their standing in the nomenclature based on whole-genome sequence similarity analysis as follows. Draft genomes of the most closely related type strains were downloaded from NCBI, and average nucleotide identity (ANI) values were calculated using the average_nucleotide_identity.py script from the pyani package (45). We used the typical percentage threshold of 95% for the ANI-based species boundary (46). Non-type strains that shared <90% pheS gene sequence similarity and <95% ANI with the most closely related type strain were given the clade name of the most closely related type strain but were assigned a separate clade number (e.g., L. curieae 2).

Subsequently, each sequence was assigned from the phylum level down to the species level (or to the subspecies level, where applicable) and strain level (= accession number) using the Greengenes scheme (18) according to the list of prokaryotic names with standing in nomenclature (http://www.bacterio.net/index.html). In addition to sequences formerly belonging to the genera Lactobacillus, Paralactobacillus, and Pediococcus, pheS gene sequences of the type species of each of the genera Enterococcus (E. faecalis, GenBank accession no. AJ843387), Fructobacillus (F. fructosus, AM711194), Lactococcus (L. lactis, JN226442), Leuconostoc (L. mesenteroides, AM711145), Oenococcus (O. oeni, FM202093), Streptococcus (S. pyogenes, AM269572), and Weissella (W. viridescens, FM202120) as well as the pheS gene sequence of Bacillus subtilis (extracted by a BLAST search against the type strain ATCC 6051; CP003329) were also included in the taxonomic framework.

Construction of mock communities of lactobacilli.

A total of 13 strains of lactobacilli were individually grown in MRS media (Table S2 in the supplemental material). Nucleic acids were extracted by using bead beating in combination with a Maxwell DNA extraction system. Briefly, 30 μl of pellet from the overnight culture was transferred into a bead-beating tube (MP Biomedicals LLC, Santa Ana, CA, USA) together with 550 μl buffer A (NaCl 0.2 M, Tris 0.2 M, EDTA 0.02 M; pH 8), 200 μl 20% SDS, and 20 μl proteinase K (Promega, Madison, WI USA). The sample was incubated for 30 min at 56°C. Bead beating was performed at 6.0 m/s for 40 s using a FastPrep-24 system (MP Biomedicals LLC). The sample was centrifuged at 16,000 × g for 6 min. The supernatant was transferred into the first well of the Maxwell cartridge containing 300 μl of lysis buffer, and all subsequent steps were done as described in the manufacturer’s protocol (Maxwell 16 FFS nucleic acid extraction system, Promega). DNA was eluted into a total volume of 80 μl elution buffer (EB, 10 mM Tris [pH 8.5] with HCl). Subsequently, nucleic acids from individual strains were serially diluted and mixed at known concentrations to generate two different mock communities (Table S2 in the supplemental material). The two final pools, each containing a total of 100 ng DNA, were evaporated and reeluted in 50 μl of RNase-free water. Per PCR volume of 20 μl, 1 μl of the mock mix DNA was used.

Collection and processing of fermented-food and host-associated samples and extraction of nucleic acids.

In this study, 24 samples from various environments were analyzed (Table 3). Fermented-food samples were purchased from local markets or from home-fermenting individuals in three countries in Southeast Asia: Vietnam, Indonesia, and Malaysia. Coconut milk yogurt, Greek yogurt, and kefir, all labeled to contain live cultures, were obtained from an organic store in Singapore. Aliquots of a total volume of approximately 20 ml were freeze-dried using a VirTis lyophilizer (SP Industries, Gardiner, NY, USA) and subsequently homogenized using a coffee grinder (CoffeeMill, Severin, Germany), which was thoroughly sterilized after each use. Nucleic acids were extracted from 30 mg of food-associated sample as described above for isolates of lactobacilli by using the protocol that combined bead beating and Maxwell DNA extraction.

TABLE 3.

Fermented-food and host-associated samples analyzed in this study for bacterial community composition using 16S rRNA and pheS gene amplicon sequencing

Sample name	Sample type	Sample origin
DNDA01	Coconut milk yogurt	Singapore, Singapore
DNDA02	Greek yogurt	Singapore, Singapore
DNDA03	Kefir	Singapore, Singapore
ANIM01	Guinea pig feces	Singapore, Singapore
ANIM02	Mouse jejunum content	Singapore, Singapore
ANIM03	Wild boar feces	Singapore, Singapore
FVEG01	Fermented mustard greens	Ha Long City, Vietnam
FVEG02	Kimchi	Ha Long City, Vietnam
FVEG03	Fermented mustard greens	Jakarta, Indonesia
TEMP01	Tempeh	Pasar Tegal Danas, Cikarang, Jakarta, Indonesia
TEMP02	Tempeh	Lemahebang, Cikarang, Jakarta, Indonesia
TEMP03	Tempeh	Bojong Market, Karawang, Jakarta, Indonesia
TEMP04	Tempeh	Bojong Market, Karawang, Jakarta, Indonesia
TEMP05	Tempeh	Roxy Market, Cikarang, Jakarta, Indonesia
TEMP06	Tempeh	Cikarang Puteh, Cikarang, Jakarta, Indonesia
TEMP07	Tempeh	Rusa Market, Cikarang, Jakarta, Indonesia
TEMP08	Tempeh	Rusa Market, Cikarang, Jakarta, Indonesia
TEMP09	Tempeh	Rusa Market, Cikarang, Jakarta, Indonesia
TEMP10	Tempeh	Niaga Haji Abdul Malik Market, Cikarang, Jakarta, Indonesia
FTOF01	Fermented tofu	Kuala Lumpur, Malaysia
FTOF02	Fermented tofu	Kuala Lumpur, Malaysia
FTOF03	Fermented tofu	Kuala Lumpur, Malaysia
FTOF04	Fermented tofu	Kuala Lumpur, Malaysia
FTOF05	Fermented tofu	Kuala Lumpur, Malaysia

To test applicability of the method to host-associated samples, one fresh fecal sample each was collected from a free-ranging wild boar (Singapore, NParks research permit number 18-075) and a pet guinea pig (Singapore). Nucleic acids were extracted with a Qiagen Power Fecal Pro kit (Qiagen, Hilden, Germany) according to the manufacturer’s protocol. In addition, DNA from the jejunum content of a laboratory mouse fed on a standard chow diet (kindly provided by Henning Seedorf, Temasek Life Sciences Laboratory, Singapore) was extracted as described previously by Rius et al. (47).

Amplification and next-generation sequencing of 16S rRNA and pheS marker genes.

PCR amplification of 16S rRNA and phenylalanyl-tRNA synthetase α-subunit (pheS) genes was carried out using previously described primer pair 515F (5′-GTGYCAGCMGCCGCGGTAA-3′) (48) and 806R (5′-GGACTACNVGGGTWTCTAAT-3′) (49) and previously described primer pair pheS21F (5′-CAYCCNGCHCGYGAYATGC-3′) (26) and pheS23R (5′-GGRTGRACCATVCCNGCHCC-3′) (26), respectively, and PCR cycling conditions as described previously (26, 50) on a Bio-Rad T100 thermal cycler (Bio-Rad Laboratories Inc, Hercules, CA, USA). The expected amplicon sizes were approximately 330 bp for the 16S rRNA gene fragment and 431 bp for the pheS gene fragment. The PCR mixture in each assay contained 30 μl REDiant 2× PCR master mix (1st BASE Biochemicals; Axil Scientific Pte Ltd., Singapore), 6 μl forward primer (final concentration 0.2 μM), 6 μl reverse primer (0.2 μM), 14 μl RNase-free water (Life Technologies, Grand Island, NY, USA), and 3 μl of DNA (samples with DNA concentrations above 60 ng per μl were diluted). The assay was split into triplicate reaction mixtures per DNA sample. Three reaction mixtures per 96-well plate did not contain any DNA and served as negative controls. Successful amplification and absence of amplification products from the no-template negative controls were verified on 1% (wt/vol) agarose gels. Subsequently, amplicons were sent to NovogeneAIT (Singapore, Singapore) for multiplexing and sequencing on a NovaSeq 6000 sequencer using PE250 chemistry (Illumina, San Diego, CA, USA).

Bioinformatic and statistical analyses.

High-throughput 16S rRNA and pheS gene amplicon sequence data were processed using the QIIME2 pipeline (23). Paired-end reads were filtered, denoised, merged, and subjected to chimera checking by using DADA2 (51). Parameters for 16S rRNA gene sequences were “--p-trim-left-f 27 --p-trim-left-r 28 --p-trunc-len-f 200 --p-trunc-len-r 150 --p-chimera-method pooled,” while the parameters for pheS gene sequences were “--p-trim-left-f 27 --p-trim-left-r 28 --p-trunc-len-f 0 --p-trunc-len-r 0 --p-chimera-method pooled.” DADA2-derived representative sequences of bacterial 16S rRNA genes were taxonomically assigned with SILVA (version 138; https://www.arb-silva.de/silva-license-information/) (19) using the QIIME2 compatible files produced according to the Github pipeline (https://github.com/mikerobeson/make_SILVA_db). DADA2-derived representative sequences of pheS genes were taxonomically assigned using pheS-DB (this study). For 16S rRNA gene sequence data, classification was done using the vsearch algorithm with default parameters (“--p-maxaccepts 10” and “--p-perc-identity 0.80”) (52). For pheS gene sequence data, the vsearch algorithm was applied using default parameters apart from “--p-maxaccepts 1” and “--p-perc-identity 0.90” (52). For downstream analyses, all abundance tables were collapsed at the genus level (-L 6) and species level (-L 7), exported to .tsv format, and further processed in Excel (Microsoft Corp., Redmond, WA, USA).

The fraction of representative sequence reads that was classified as “unassigned” by pheS-DB analysis was dissected further using Kraken2 (reference database version 2019) (53). PheS-DB-unassigned representative sequence reads that were assigned to the lactobacilli by Kraken2 were subjected to BLAST searches against the NCBI nucleotide collection (nr/nt) for differentiation between true pheS sequences and nonspecific sequences. True pheS hits were imported into ARB, aligned, and calculated into the phylogenetic tree for additional verification. The total percentage of sequence reads that these representative sequences accounted for was calculated from the corresponding feature table in QIIME2.

To compare levels of species richness between samples analyzed based on 16S rRNA and pheS genes, Chao-1 indices were calculated using the software PAST (54). The similarity of environmental samples within and between methods was assessed using the Bray-Curtis similarity metric in PAST (54, 55).

Student's t test (two-tailed with unequal variance) was performed in Excel (Microsoft) and used to test for significant differences between Chao-1 indices based on 16S rRNA and pheS genes.

Visualization of sample similarity by clustering was done by generation of a heat map and dendrograms in ClustVis, where row centering and unit variance scaling were applied and both rows and columns were clustered using correlation distance and average linkage (56). Principal-component analysis (PCA) was also performed in ClustVis. For this, unit variance scaling was applied to the rows, and principal components were calculated using singular value decomposition with imputation.

Data availability.

Partial pheS gene sequences from isolated lactobacilli were deposited with GenBank under accession numbers MT415913.1 to MT415925.1. Partial 16S rRNA and pheS gene sequence data generated by high-throughput amplicon sequencing were deposited in the NCBI BioProject database under accession number PRJNA629775.