Streptococcus pseudopneumoniae: Use of Whole-Genome Sequences To Validate Species Identification Methods

A correct identification of Streptococcus pseudopneumoniae is a prerequisite for investigating the clinical impact of the bacterium. The identification has traditionally relied on phenotypic methods.

ABSTRACT

A correct identification of Streptococcus pseudopneumoniae is a prerequisite for investigating the clinical impact of the bacterium. The identification has traditionally relied on phenotypic methods. However, these phenotypic traits have been shown to be unreliable, with some S. pseudopneumoniae strains giving conflicting results. Therefore, sequence-based identification methods have increasingly been used for identification of S. pseudopneumoniae. In this study, we used 64 S. pseudopneumoniae strains, 59 S. pneumoniae strains, 22 S. mitis strains, 24 S. oralis strains, 6 S. infantis strains, and 1 S. peroris strain to test the capability of three single genes (rpoB, gyrB, and recA), two multilocus sequence analysis (MLSA) schemes, the single nucleotide polymorphism (SNP)-based phylogeny tool CSI phylogeny, a k-mer-based identification method (KmerFinder), average nucleotide identity (ANI) using fastANI, and core genome analysis to identify S. pseudopneumoniae. Core genome analysis and CSI phylogeny were able to cluster all strains into distinct clusters related to their respective species. It was not possible to identify all S. pseudopneumoniae strains correctly using only one of the single genes. The MLSA schemes were unable to identify some of the S. pseudopneumoniae strains, which could be misidentified. KmerFinder identified all S. pseudopneumoniae strains but misidentified one S. mitis strain as S. pseudopneumoniae, and fastANI differentiated between S. pseudopneumoniae and S. pneumoniae using an ANI cutoff of 96%.

INTRODUCTION

The mitis group of the genus Streptococcus (MGS) consists of approximately 20 species (1). Most MGS strains are normal constituents of the human oral microbiome and rarely cause disease. One exception is Streptococcus pneumoniae, which is one of the most common causes of pneumoniae, meningitis, and otitis media (1).

The discrimination of S. pneumoniae from the remaining MGS strains has relied on phenotypic tests, with S. pneumoniae being optochin sensitive and bile soluble. In 2004, a new species, Streptococcus pseudopneumoniae, was described and characterized as being bile insoluble and sensitive to optochin when grown in ambient air but optochin resistant when grown in a CO₂-enriched atmosphere (2). Later studies have questioned the validity of these phenotypic traits for identification of S. pseudopneumoniae (3).

Sequencing of the 16S rRNA gene has been the standard reference identification method for most bacterial species. However, due to a high degree of interspecies similarity, 16S rRNA gene sequence analysis is not sufficient for the identification of MGS strains to the species level (4). Instead, multilocus sequence analysis (MLSA) schemes, consisting of several concatenated housekeeping genes, have been used. Two MLSA schemes have been reported for MGS, comprising six and seven housekeeping genes each (5–7). Due to the workload of doing PCR of and sequencing of 6 or 7 genes, these schemes are not feasible in clinical laboratories. Therefore, several single genes have been used for MGS species identification, including recA (8, 9), rpoB (10), and gyrB (11).

Next-generation sequencing (NGS) technology has rapidly expanded the sequences available (12). By using the long stretches of DNA obtained using NGS, it is possible to obtain a high degree of resolution between the strains tested (1, 13, 14). Furthermore, easy and accessible analysis methods for NGS data have emerged, making the phylogenetic analysis of data from unknown species faster (15–17).

The aim of this study was to elucidate the applicability of sequence-based identification methods within MGS, with the primary focus on identification of S. pseudopneumoniae. We tested single gene phylogeny, MLSA and multilocus sequence typing (MLST) schemes, average nucleotide identity (ANI), core genome analysis, and available online tools, with the use of CSI phylogeny as the inclusion identification method.

MATERIALS AND METHODS

Strains.

Twenty Streptococcus pseudopneumoniae strains previously described by Arbique et al. (2) were obtained from the Centers for Disease Control and Prevention (CDC, Atlanta, GA, USA), and five additional clinical strains were tested and identified as suspected S. pseudopneumoniae strains with routine matrix-assisted laser desorption ionization–time of flight mass spectrometry (MALDI-TOF MS) at the Regional Department of Clinical Microbiology, Slagelse, Region Zealand, Denmark.

For all strains, DNA was extracted using the Nextera DNA Flex microbial colony extraction protocol (Illumina, San Diego, USA). Library preparation for the 20 strains obtained from the CDC was with the Nextera XT DNA library preparation kit, the libraries were sequenced on an Illumina NextSeq 500 generating 150-bp paired-end reads and quality ensured using the Bifrost platform (https://github.com/ssi-dk/bifrost). Library preparation for the five clinical strains was with the Nextera DNA flex library preparation kit (Illumina); the libraries were sequenced on an Illumina MiSeq generating 150-bp paired-end reads and quality ensured using the CLC Genomics Workbench (Qiagen, Denmark). The raw data were assembled using SPAdes v. 3.13.0 (18), and assembly metrics were calculated using QUAST 5.0.2 (19).

Furthermore, we downloaded an additional 151 sequences (S. pseudopneumoniae, n = 39; S. pneumoniae, n = 59; S. mitis, n = 22; S. oralis, n = 24; S. infantis, n = 6; S. peroris, n = 1) from the NCBI database, see Data Set S1 in the supplemental material for details. Sequences were collected with an emphasis on having diversity regarding isolation sites and serotypes (20) and included the species type strains. To summarize, this study includes 176 genome sequences representing six species: S. pseudopneumoniae (n = 64), S. pneumoniae (n = 59), S. mitis (n = 22), S. oralis (n =24), S. infantis (n = 6), and S. peroris (n = 1).

CSI phylogeny.

The single nucleotide polymorphism (SNP)-based CSI phylogeny Web application (https://cge.cbs.dtu.dk/services/CSIPhylogeny) was applied with default settings for species identification. In short, CSI phylogeny maps each contig against a chosen reference genome using BWA v. 0.7.2 (20). Afterwards, SNPs are called using SAMtools v. 0.1.18 (21), leaving out all SNPs in a 10-base vicinity of each other. CSI phylogeny returns a newick file for tree visualization. All strains were tested using their full assembled genomes, and sequences were uploaded in a fasta format (16). To avoid any bias toward a single species used in this study, we used Streptococcus sanguinis ATCC 10556^T as the reference genome.

Gene sequence extraction for single gene analysis and MLSA.

From each genome, the following single genes were extracted for phylogenetic comparison: recA, gyrB, and rpoB for single gene phylogeny; map, pfl, ppaC, pyk, rpoB, sodA, and tuf for a seven-gene MLSA scheme (here, “MLSA scheme”) (7); and aroE, gdh, gki, recP, spi, and xpt for a six-gene MLSA scheme (derived from the S. pneumoniae MLST scheme and therefore termed “MLST scheme”) (5, 6).

To find and extract the genes from the genomes, we used an in-house Python script that uses previously published gene sequences as query sequences (see Data set S2). In short, the script makes a BLAST database for each strain, performs a dc-MEGABLAST search, and extracts the gene. The script returns a fasta file containing the hits present in the genome file tested. For the MLSA and MLST schemes, the Python script also returns the concatenated sequence for each strain.

All genes were aligned using MUSCLE v. 3.8.425 (22), and all gene alignments were visually inspected to ensure homology.

Core genome analysis.

For the core genome analysis, genes were predicted and translated into amino acid sequences using prodigal v. 2.6.2 (21). The Bacterial Pan Genome Analysis tool (BPGA) version 1.3 (using USEARCH with a sequence identity cutoff of 0.5) was used to find the genes present in all strains (the core genome) used in this study (23). Genes were aligned using MUSCLE v. 3.8.425 (22). To ensure homology, all genes with <35% conserved sites were excluded from the analysis. All the remaining core genes were concatenated for further analysis.

Phylogenetic trees.

A maximum likelihood tree based on the nucleotide sequences was constructed for each of the three single genes and the MLSA and MLST schemes using PhyML v 3.1, with SMS and bootstrapping (×100) (24).

For the core genome analysis, a tree based on the amino acid sequences was built using IQ-TREE, with the optimized maximum likelihood model, the gamma model, the -fast option applied, and bootstrapping (×100) (25).

All trees were visualized using the online tool iTOL (26).

KmerFinder.

To perform KmerFinder, we used the Web application available at https://cge.cbs.dtu.dk/services/KmerFinder/ (software v. 3.0.2, accessed 29 October 2020) (15). The “standard output” was used, and the species of the “template” with the highest score was considered the result. If the template with the highest score was designated “streptococcus species,” the result was categorized as “identification to genus level.”

Average nucleotide identity.

ANI was calculated using the tool fastANI v. 1.32 (17). All strain pairs were tested using the “many to many” method in fastANI and by using the “–matrix” option; results were obtained as a phylip-formatted lower triangular matrix. Earlier studies have shown that an ANI value of 94% to 96% corresponds to the recommended DNA-DNA hybridization species cutoff of 70%.

Data availability.

The sequences of the 25 S. pseudopneumoniae strains sequenced during this study have been deposited in the NCBI database with BioSample numbers SAMN15921647 to SAMN15921671, under BioProject number PRJNA659631.

RESULTS

CSI phylogeny.

CSI phylogeny was able to cluster all S. pneumoniae, S. pseudopneumoniae, and S. oralis strains in clearly distinct clusters together with a type strain, see Fig. 1. The S. infantis strains and the single S. peroris strain clustered together. The strains downloaded as S. mitis (except for one strain, see below), formed several subclusters. The subclustering of S. mitis has been described earlier (1). One strain, downloaded as S. mitis (S. oralis ATCC 6249) from the NCBI Web server, clustered in the S. oralis cluster. Further review of this strain confirmed that it was correctly identified as S. oralis by CSI phylogeny and that the identification had been corrected on the American Type Culture Collection (ATCC) website (https://www.lgcstandards-atcc.org/Products/All/6249.aspx) but not on the NCBI Web server (GenBank accession AEEN00000000.1). All other strains achieved the same species identification using CSI phylogeny as on the NCBI Web server.

FIG 1.

Phylogenetic tree made with 176 genomes belonging to the mitis group of the genus Streptococcus using the CSI phylogeny Web server (https://cge.cbs.dtu.dk/services/CSIPhylogeny). All strains were analyzed using whole-genome sequences and with the S. sanguinis type strain NCTC 7863 as the reference genome. Strains are colored according to the type strain contained in each cluster. The species designations were in concordance with the species annotation on the NCBI website, except for S. oralis ATCC 6249, which was annotated as a S. mitis (see text). Confidence scores of >0.9 are shown.

Single gene analysis.

We made phylogenetic trees based on the recA, gyrB, and rpoB gene sequences shown in Fig. 2 (see also Fig. S1 and S2 in the supplemental material). None of the single gene analyses were able to separate the strains into the phylogenetic groups defined by CSI phylogeny.

FIG 2.

Maximum likelihood phylogenetic tree made using the rpoB gene. The gene sequences were extracted from whole-genome sequencing (WGS) and aligned using muscle; the tree was made using PhyML. Strains are colored according to the species obtained using the CSI phylogeny (Fig. 1).

MLSA and MLST schemes.

The phylogenetic tree based on the MLSA scheme showed clustering of S. oralis, S. infantis, and S. pneumoniae strains with their respective type strains (Fig. 3). The S. peroris type strain (ATCC 700780^T) clustered in the S. infantis cluster. Twenty of the 23 S. mitis strains formed one cluster around the type strain, and the remaining three S. mitis strains clustered in the S. pseudopneumoniae branch.

FIG 3.

Maximum likelihood phylogenetic tree made using seven concatenated housekeeping genes included in the MLSA scheme (map-pfl-ppaC-pyk-rpoB-tuf) (7). The gene sequences were extracted from WGS and concatenated using an in-house Python script. A tree was made using PhyML with bootstrapping (×100). Strains are colored according to the species obtained using the CSI phylogeny (Fig. 1). Bootstrap values >90 are shown.

In contrast, the phylogenetic tree based on the MLST scheme showed that the S. pseudopneumoniae strains did not form a distinct cluster but had strains spreading to the S. pneumoniae and S. mitis branches (Fig. 4). S. pneumoniae, S. mitis, S. oralis, and S. infantis formed distinct clusters with their type strain using the MLST scheme.

FIG 4.

Maximum likelihood phylogenetic tree made using six concatenated housekeeping genes included in the MLST scheme (aroE-gdh-gki-recP-spi-xpt). The gene sequences were extracted from WGS and concatenated using an in-house Python script. A tree was made using PhyML with bootstrapping (×100). Strains are colored according to the species obtained using the CSI phylogeny (Fig. 1). Bootstrap values >90 are shown.

Core genome analysis.

Using BPGA and discarding all genes with <35% conserved sites, we found the core genome of the 176 strains to be 590 genes, with a total alignment length of 180,460 amino acids. As shown in Fig. 5, all S. pneumoniae, S. pseudopneumoniae, S. mitis, and S. oralis strains formed clusters together with their type strains; the S. peroris type strain lies inside the S. infantis cluster.

FIG 5.

Maximum likelihood phylogenetic tree made with IQ-TREE using the concatenated sequences of 590 core genes, with a total amino acid alignment length of 180,460. Strains are colored according to the species obtained using the CSI phylogeny (Fig. 1). Bootstrap values >90 are shown.

KmerFinder.

We used the online platform of KmerFinder to identify our strains (Table 1) (16). All S. pseudopneumoniae and S. pneumoniae strains were correctly identified using KmerFinder. In contrast, one S. mitis strain (S. mitis SK1080) was misidentified as S. pseudopneumoniae, and one S. mitis strain and two S. oralis strains were only identified to the genus level. All S. infantis and S. peroris strains were only identified to the genus level.

TABLE 1.

KmerFinder results for 176 mitis group streptococci

Species (no. tested)	No. identified:
	At the species level	At the genus level	Incorrectlya
S. pseudopneumoniae (64)	64	0	0
S. pneumoniae (59)	59	0	0
S. mitis (22)	20	1	1b
S. oralis (24)	22	2	0
S. infantis (6)	0	6	0
S. peroris (1)	0	1	0

^aSpecies identification different from identification with CSI phylogeny.

^bIdentified as S. pseudopneumoniae.

ANI.

The means and ranges of ANI values between different species can be found in Table 2, and all ANI values can be found in Data Set S3. Importantly, when testing the S. pseudopneumoniae strains, we did not see any overlap between the results from S. pseudopneumoniae versus S. pseudopneumoniae and the results from S. pseudopneumoniae versus S. pneumoniae. Despite this nonoverlap, the distances between the lowest value when testing S. pseudopneumoniae versus S. pneumoniae and the highest value when testing S. pseudopneumoniae versus S. pneumoniae were 95.87% and 95.09%, respectively, which clearly reflects the close relationship between the two species. When testing S. pseudopneumoniae against S. mitis, S. oralis, S. infantis, and S. peroris, all had an ANI of <93.25%.

TABLE 2.

Pairwise ANI between the species in the study

Species	% ANI (mean [range])a
	S. pseudopneumoniae	S. pneumoniae	S. oralis	S. mitis	S. infantis
S. pseudopneumoniae	97.54 (95.87–99.96)
S. pneumoniae	94.48 (94.04–95.09)	98.45 (97.27–100.0)
S. oralis	86.44 (85.97–86.94)	86.55 (86.01–87.2)	92.91 (90.42–98.82)
S. mitis	92.12 (91.32–93.25)	91.73 (91.12–92.83)	86.84 (86.13–87.51)	93.31 (92.02–96.79)
S. infantis	83.13 (82.39–83.92)	82.92 (82.3–83.7)	83.44 (82.94–84.09)	83.4 (82.76–84.3)	90.8 (88.16–94.92)
S. perorisb	82.64 (82.34–82.96)	82.4 (82.14–82.88)	82.58 (82.34–82.96)	82.72 (82.41–83.14)	88.34 (87.1–89.0)

^aANIs between all strains in one species against all strains in another species. The individual ANI values can be found in Data Set S3 in the supplemental material.

^bOnly one S. peroris strain was used in this study, and so an S. peroris versus S. peroris result is left out in the table.

DISCUSSION

We used NGS sequences and bioinformatic tools to investigate identification schemes used to identify MGS to the species level, with an emphasis on S. pseudopneumoniae. Earlier studies have included up to 44 whole-genome sequenced S. pseudopneumoniae strains (3). In this study, we have included 64 whole-genome sequenced S. pseudopneumoniae strains, which, to our knowledge, is the largest collection to date. As an inclusion identification method, we used CSI phylogeny, which has previously been used for MGS with good results (23, 24). The sequence for one strain (S. oralis ATCC 6249) which was downloaded from the NCBI Web server as S. mitis, clustered in the S. oralis cluster and should be considered S. oralis, which is in agreement with earlier studies (1) and the nomenclature on the ATCC website when we checked (https://www.lgcstandards-atcc.org/Products/All/6249.aspx).

Some studies have used single-gene sequence analyses to achieve a species identification of S. pseudopneumoniae. In a study including 11 S. pseudopneumoniae strains identified using phenotypic methods, recA was found to be able to discriminate S. pneumoniae, S. pseudopneumoniae, and S. mitis (9). Later, the recA gene was used to differentiate between S. pseudopneumoniae and the remaining MGS (8). A MALDI-TOF validation study used the rpoB gene to identify MGS, including 17 S. pseudopneumoniae strains. The use of the rpoB gene as the primary identification method was based on the gene having the best overall correlation between two MALDI-TOF systems and phenotypic methods (10). Zhou et al. used gyrB and 16S rRNA gene sequence analysis to identify 9 S. pseudopneumoniae strains and an additional 87 non-S. pneumoniae MGS strains and were able to cluster all MGS strains into distinct species clusters using gyrB (11). In our study, phylogenetic trees based on recA, rpoB, and gyrB gene sequences revealed that none of the genes could be used to identify S. pseudopneumoniae, since no distinct clusters were formed.

Earlier studies have tested other genes for species identification within the MGS. Rasmussen et al. tested the genes gdh, recA, and sodA on a collection of 13 S. mitis, 26 S. oralis, 2 S. infantis, 18 S. gordonii, 20 S. sanguinis, and 1 S. cristatus strain. The study found that, especially, phylogeny inferred by the gdh gene often misidentified S. oralis strains (27). The lytA gene combined with restriction fragment length polymorphism analysis has been used for the identification of S. pneumoniae (28). Simões et al. found that some S. pneumoniae strains contained lytA genes that were different from the “typical” S. pneumoniae lytA gene and had the highest sequence similarity with the gene for the S. pseudopneumoniae type strain. Interestingly, Simões et al. also found two strains that clustered together with the S. pseudopneumoniae type strain using MLSA but contained a lytA gene with the highest sequence similarity to that for S. pneumoniae TIGR4 (29).

The results presented in this study and earlier studies emphasize that the identification of MGS should not be inferred from single genes, as it is often inaccurate, which is probably due to a high degree of interspecies horizontal gene transfer in this group (30).

To overcome the problem with horizontal gene transfer in MGS, an MLSA approach using concatenated sequences of six or seven housekeeping genes has been utilized. However, even the MLSA approach has been reported to be insufficient to identify some MGS strains. One study testing 132 MGS strains was unable to infer phylogeny on two strains clustering at the root of the S. pneumoniae branch (31), and in another study, 28 of 634 strains did not cluster according to any recognized species (32). Our results are in concordance with these results, since some strains did not cluster together with their type strains. S. pseudopneumoniae can be especially difficult to identify using the MLST scheme, and some S. mitis strains could be misidentified as S. pseudopneumoniae using the MLSA scheme. Furthermore, when comparing with the phylogeny inferred by CSI phylogeny, some strains would probably be misidentified using the MLSA and MLST schemes.

Using core genome analysis, we were able to place all strains into the same defined clusters as when using CSI phylogeny. We used an amino acid-based approach for core genome analysis, which is very similar to the one used by Rasmussen et al. (27).

Two studies have used a nucleotide-based core genome approach. Garriss et al. (3) used a SNP-based method to infer phylogeny on 147 MGS strains, and Jensen et al. (1) used a method that relied on the alignment of 200-bp fragments of the S. oralis strain Uo5 against the 195 MGS strains tested. Despite the use of nucleotide sequences, different analysis methods, and different strain collections, the trees from the studies by Garriss et al. (3) and Jensen et al. (1) are remarkably similar to the tree obtained in our study. For example, all three core genome trees cluster the two S. pseudopneumoniae 338-14 and 61-14 strains and the two S. mitis SK575 and SK564 strains together, revealing the robustness of all three methods.

Interestingly, three S. mitis strains (321a, SK564, and SK575) were found to cluster outside the S. mitis cluster in the phylogenetic tree based on the MLSA scheme using the maximum likelihood method. These three strains have previously been found to cluster inside the S. mitis cluster using the minimum evolution method (1). The minimum evolution method is less computationally demanding than the maximum likelihood method but is reported to disregard much of the potential evolutionary information contained in sequences (33). To clarify whether the choice of phylogeny inference method will influence the clustering pattern, we used MEGAX to construct minimum evolution trees with bootstrapping (×100) and compared them with those trees constructed using the maximum likelihood method (34) (see Fig. S3 to S8 in the supplemental material). We did not find any significant differences between the two methods using either single gene or core genome analysis.

KmerFinder compares k-mers of a query genome with k-mers from genomes contained in the database. It has been tested on a variety of publicly accessible genomes, including MGS, and was found to occasionally misidentify S. mitis as S. pneumoniae (15). The only misidentification experienced in the present study was S. mitis SK1080 being misidentified as a S. pseudopneumoniae. Interestingly, S. mitis SK1080 was an outlier in the S. mitis group using both core genome analysis and CSI phylogeny. Often, KmerFinder returns several matches with lower scores than the best match, and in our strain collection, three S. mitis strains have S. pseudopneumoniae as the second-best taxon match. Due to only one S. pseudopneumoniae strain being present in the database, most S. pseudopneumoniae strains have a second-best taxon match with S. pneumoniae. However, the score difference between best match and second-best taxon match was higher for S. pseudopneumoniae than for S. mitis, approximately 40,000 and 12,000, respectively. Interestingly, the three S. mitis strains (S. mitis SK564, S. mitis SK575, and S. mitis 321a) with S. pseudopneumoniae as second-best taxon match all cluster together in both the tree made by core genome analysis and the CSI phylogeny. As previously described, S. mitis contains several subclusters, and the misidentification and the lower distance to the second matched taxon probably reflects the lack of representatives within all S. mitis subclusters in the KmerFinder database.

ANI values between 94% and 96% have been shown to reflect the generally accepted DNA-DNA hybridization species boundary of 70% (1, 35). Previously, Jensen et al. used a BLAST-based method to calculate the ANI in a collection of 195 MGS strains, finding that only S. pneumoniae, S. pseudopneumoniae, and S. gordonii had a pairwise ANI of >96% (1). These results are comparable to the results found in the present study. However, we found two ANI values that were <96% when testing S. pseudopneumoniae versus S. pseudopneumoniae. However, this was a rare instance, since there were more than 2,000 ANI values calculated when testing S. pseudopneumoniae versus S. pseudopneumoniae. As shown before, most intraspecies ANI values of S. oralis, S. mitis, and S. infantis were <94%, making the 94% to 95% ANI value cutoff too high for species identification of these species. From our findings, an ANI cutoff of 96% would not misidentify an S. pseudopneumoniae strain as S. pneumoniae or as any other of the species tested in the present study.

In conclusion, we found that (i) core genome analysis clustered the S. pseudopneumoniae strains and the other examined MGS strains into distinct species clusters, similar to those found by CSI phylogeny, (ii) by using single gene analysis, misidentification of strains often occurs, so the method cannot be used for identification of S. pseudopneumoniae, (iii) MLSA and MLST schemes did not cluster all S. pseudopneumoniae strains into a single cluster, and misidentifications as S. mitis and S. pneumoniae may occur, (iv) KmerFinder was able to correctly identify all S. pseudopneumoniae strains, though one S. mitis strain was misidentified as S. pseudopneumoniae, and (v) fastANI can be used to discriminate S. pseudopneumoniae and S. pneumoniae and both from the remaining MGS using an ANI cutoff of approximately 96%.