Background

Staphylococcus aureus is a bacterial species often associated with primate hosts, and specific sequence types (ST) can be found frequently living on different domestic animals [1]. In humans, S. aureus is frequently isolated from nasal membranes and skin as residents or transients [2]. Because of its association with clinically significant infections and foodborne diseases [3, 4], S. aureus has been extensively investigated, leading to massive datasets on its species diversity. After the genetically divergent ST 75 was first reported in Australia in 2002 [5], many other isolates genetically related to clonal complex 75 (CC 75) were described in Belgium [6], Cambodia [7], China [8], Fiji [9, 10], France [11], French Guiana [12], New Zealand [9], Thailand [13, 14], Trinidad & Tobago [15], and the UK [16]. Genomic data tracking showed that this lineage also appeared in the United States [17]. Meanwhile, another S. aureus lineage has recently been recovered from nonhuman primates [18, 19] and bats [20] in Africa. Recently, these two genetically divergent lineages have received formal taxonomic classification and were recognized as S. argenteus and S. schweitzeri, respectively, two novel species within the S. aureus complex (SAC) [16].

S. argenteus cannot be distinguished from S. aureus using routine diagnostic microbiology identification methods [7, 8, 10, 12, 13], and PCR amplification of the gene nucA, which is used as a standard confirmatory marker for S. aureus, may be positive in S. argenteus [16]. However, considerable difficulties were observed in amplification of some multilocus sequence typing (MLST) gene loci from S. argenteus using standard MLST primers used for typing S. aureus [7, 21]. These difficulties may result in S. argenteus isolates being excluded from or misidentified as S. aureus. Data on clinical features of S. argenteus infection are limited, but these studies indicated that S. argenteus was associated with skin and soft tissue infections, nosocomial infections, invasive staphylococcal sepsis, and even death [13, 14, 22]. Genes encoding the Panton-Valentine leukocidin (PVL) cytotoxin and staphylococcal enterotoxin B (SEB) were detected in S. argenteus isolates [11, 12, 14]. Community-acquired methicillin-resistant “S. aureus” (CA-MRSA) isolates, which were subsequently characterized as S. argenteus, were predominant in remote aboriginal communities of Australia [23]. Therefore, there is no doubt that S. argenteus is a threat to human health.

The emergence of S. argenteus and S. schweitzeri results in the need to determine whether they should be distinguished from S. aureus in routine practice. Chantratita et al. suggested that this might be necessary if the infection is associated with different clinical manifestations, and/or requires different antimicrobial regimens [14]. According to this suggestion, S. schweitzeri is clearly separated from the other two species though one isolate was obtained from human (ST 1822, cause no infection) [24], but S. argenteus is difficult to make a judgment based on current available data. In this study, we sequenced the genomes of five S. argenteus isolates, and performed a genomic comparison among SAC, so as to clarify their evolutionary relationships, to evaluate their pathogenic potential, and to identify genomic differences. This study is expected to establish an exploratory evolutionary framework regarding the evolution of pathogenicity of SAC.

Results

Genomic features of SAC species

The sequences of 51 SAC genomes were used in this study and shown in Fig. 1. The draft genome sequences of five S. argenteus strains previously identified in our laboratory [8] were sequenced in this study. Thirty S. aureus genomes, which represent a genetically diverse collection of strains of several different STs (Additional file 1: Figure S1), were selected from the complete genome sequences available in the NCBI genome database. Ten S. argenteus and six S. schweitzeri genomes (complete or draft) were also downloaded from NCBI.

Fig. 1
figure 1

Phylogenetic relationship, genome information, and relevant typing information of SAC species. A maximum likelihood phylogenetic tree of SAC was constructed based on concatenated deduced amino acid sequences of 1375 single copy core genes of the 51 strains used in this study. Bootstrap values (expressed as percentages of 1000 replications) greater than 70% are shown at the branch points and the asterisk (*) indicates a bootstrap value of 100. The color-coded clades denote different species: red, S. aureus; green, S. argenteus; blue, S. schweitzeri. S. simiae was used as an outgroup (not shown). S. argenteus strains, SJTU F20124 (GenBank accession number: LWAN00000000), F20419 (LWAO00000000), F21164 (LWAP00000000), F21224 (LWAQ00000000), and F21285 (LWAR00000000) were sequenced as part of this study. For Sequence type, CRISPR, SCCmec, agr type, Serotype, and Prophage data: -, not detected/absent; +, detected/present; NA, data not available in the MLST database; and #, the type is close but divergent from the indicated subtype

Full size image

For S. aureus, the genome sizes varied from 2.74 Mb (strain ST398/S0385) to 2.99 Mb (TW20), the GC contents varied from 32.75% (MN8) to 32.96% (LGA251), and the number of coding sequences (CDS) per genome varied from 2399 (RKI4) to 3009 (TW20) (Fig. 1). For S. argenteus, the genome sizes varied from 2.71 Mb (M051_MSHR) to 2.87 Mb (SJTU F21285), the GC contents varied from 32.20% (M051_MSHR) to 32.42% (SJTU F21285), and the number of CDSs per genome varied from 2490 (M051_MSHR) to 2662 (SH3) (Fig. 1). For S. schweitzeri, the genome sizes varied from 2.67 Mb (FSCB1B) to 2.81 Mb (FSA096), the GC contents varied from 32.21% (FSA096) to 32.28% (FSCB5), and the number of CDSs per genome varied from 2434 (FSCB1B) to 2587 (FSA096) (Fig. 1).

The genome size, GC content and number of CDSs were significantly different between S. aureus and the two new species (p < 0.01), while no significant differences were observed between S. argenteus and S. schweitzeri (P > 0.01; Additional file 2: Figure S2). The number of genomes used in this study was limited, particularly for S. argenteus and S. schweitzeri. Furthermore, some genome sizes and CDS numbers were estimated from draft genome assemblies and remained incomplete. Nevertheless, the difference in GC content, which is less affected by the quality of genome assembly, suggests an apparent divergence between S. aureus and the other two members of the SAC.

Pan-genome of SAC species

To facilitate a genomic comparison and achieve a sound comparison at the whole genome scale, the pan-genome of SAC was defined. A pan-genome was previously described to include two distinct components, the core and variable genomes, which represent the essence and the diversity of the population, respectively [25]. The core genome consists of genes that are common to all strains while the variable genome is composed of genes absent or present at least in one strain, due to either gene loss or acquisition [26]. Here we expanded the classification of the SAC pan-genome into four genomic components: 1) core genes (CR), present in all strains; 2) core variable genes (CV), present in all strains with at least one pseudogene; 3) variable genes (VR), present in at least two strains and absent at least in one strain; and 4) unique genes (UQ), present in only one strain. The pan-genome of the 51 SAC genomes was grouped into 4249 homologous gene families, including 1671 CRs, 328 CVs, 1634 VRs, and 616 UQs (Additional file 3: Table S1; Additional file 4: Figure S3). The pan-genome of the 30 S. aureus genomes was grouped into 3966 homologous gene families, including 1752 CRs, 330 CVs, 1329 VRs, and 555 UQs. The sizes of CRs, CVs, VRs and UQ were also calculated for S. argenteus and S. schweitzeri (Additional file 5: Figure S4). The category analysis of the Cluster of Orthologous Groups (COGs) [27] for the pan-genome of the SAC and each species showed similar profiles.

Clearly, the pan-genomes for each group were all open (i.e., not complete), due to the limited number of genomes included in the analyses. The accumulation curves and pan-genome sizes largely depend on sampling size and diversity [25]. In this study, S. aureus was better sampled than S. argenteus and S. schweitzeri in both the number of strains and strain diversity. For example, of the fifteen S. argenteus genomes examined, nine were ST2250 demonstrating little genome diversity (see below). Therefore, a brief comparison of the pan-genome is shown in Additional files 6: Figure S5, and the details will not be discussed here. However, it is worth noting that the 1999 common gene families (CR plus CV) in the SAC pan-genome were approximately three quarters of the size of an individual SAC genome, and the 2578 sharing gene families (genes present in at least one genome of each species, including CR, CV and partial VR) were approximately equivalent to the CDS number of an individual SAC species genome. This is indicative of high level of sharing of homologous gene families between the species and strains of the SAC. It is predictable that S. aureus, S. argenteus and S. schweitzeri share many biological characteristics, many of which have been demonstrated in previous studies as a common phenotypes [16].

Core genome of the SAC species

Similar metabolic pathways and phenotypes between species may be the result of frequent genetic exchanges, such as horizontal gene transfer (HGT) mediated by bacteriophage [28, 29], or limited evolutionary time for divergence from a common ancestor. HGT mediated by mobilome mainly contributes to the VRs and UQs, respectively as follows (Additional file 4: Figure S3; Additional file 5: Figure S4, code X): SAC, VRs and UQs accounting for 62.7% and 34.7% of the genes assigned to X class, respectively; S. aureus, 63.5% and 33.6%; S. argenteus, 61.5% and 32.3%; and S. schweitzeri, 64.8% and 27.8%. Also, it is believed that distinct species are separated by apparent genetic distances that can act as barriers to recombination [30,31,32]. To better understand the genetic distances, the average nucleotide identities based on BLAST (bANI) were calculated using the core genes shared among SAC species (Fig. 2). Most of the intraspecific sequence identity values were greater than 95%, which was consistent with the species definition of an ANI cutoff of 95–96% [33]. The interspecific values from different groups formed three peaks similar to normal distributions and exhibited apparently differences: the peak of S. argenteus versus S. schweitzeri was near 95% while those of S. argenteus versus S. aureus and S. schweitzeri versus S. aureus were 86–89%. The core genome of S. schweitzeri (90.92 ± 4.77%) was closer related to that of S. aureus than S. argenteus (89.79 ± 4.63%, Fig. 2). Notably, in pairwise comparisons between species, bANI values greater than 95% were a small but notable part of the total. To determine whether these identity values (> 95%) were generated from highly conserved house-keeping genes (evolving very slowly), the pairwise distance among alleles of each gene family was tested. It was found that the number of gene families with an average interspecific identity greater than 95% was 188 between S. aureus and S. argenteus, 307 between S. aureus and S. schweitzeri, and 698 between S. argenteus and S. schweitzeri (161 in common). Meanwhile, 34 gene families between S. aureus and S. argenteus, 33 between S. aureus and S. schweitzeri, and 80 between S. argenteus and S. schweitzeri (28 in common) had the same amino acid sequence. It indicated that the bANI values greater than 95% were mainly due to slowly evolving genes, but horizontal gene transfer between species may also be present.

Fig. 2
figure 2

Pairwise bANI of SAC species. Pairwise bANI of S. simiae, S. aureus, S. argenteus and S. schweitzeri were determined based on the 1375 common single copy genes. For comparison of S. simiae to S. aureus, S. argenteus and S. schweitzeri, the mean bANI values were 84.63 ± 5.18% (median 84.14%), 84.89 ± 5.25% (84.26%), and 84.85 ± 5.19% (84.17%), respectively. Among the members of the SAC 1596 common genes were used determine the bANI. Interspecific mean bANI values were as follows: S. aureus, 98.76 ± 1.92% (median 99.23%); S. argenteus, 99.6 ± 0.85% (99.79%); and S. schweitzeri, 98.87 ± 1.74% (99.38%). The intraspecific mean bANI values were as follows: S. aureus vs. S. argenteus, 89.79 ± 4.63% (median 89.39%); S. aureus vs. S. schweitzeri, 90.92 ± 4.77% (90.64%); and S. argenteus vs. S. schweitzeri, 94.08 ± 3.69% (94.81%). The group interval was 0.1%

Full size image

The species of SAC have identical 16S rRNA gene while MLST and other house-keeping genes can be used to distinguish them from each other [16, 21]. Nevertheless, the MLST loci are just a small part of the genome. Here, an upgraded MLST method [34] was used to compare 1375 single copy core genes of SAC and S. simiae (used as an outgroup), which was previously considered as the closest taxa to S. aureus [35], were used to infer phylogeny. The Maximum Likelihood (ML) algorithm was used and a well-supported topology was obtained (Fig. 1). S. argenteus and S. schweitzeri clustered together (3.05% in distance based on amino acid sequence) and then clustered with S. aureus (4.42% in distance) in the core-genome tree, with perfect supporting bootstrap values (100%). S. simiae grouped far away from SAC with 21.34% in distance. This suggests that SAC members separated from each other much later than from other staphylococci, resulting in species divergence from a shared global core genome of a common ancestor.

Population structure

The program Structure version 2.3.4 [36], which implements a model-based clustering method, was used to infer the SAC population structure using genotype data of the 1596 single copy core genes. Structure Harvester was used for collating results generated by the program Structure [37]. The ΔK values, an ad hoc quantity related to the second order rate of change of the log probability of data with respect to the number of clusters, were calculated by Structure Harvester to detecting the number of K populations that best fit the data. A higher ΔK means a better fitness of the K value. The highest ΔK value (= 5840) emerged when K = 2 (Fig. 3). It indicated that staphylococcal strains investigated here fall into two distinct populations corresponding to S. aureus and S. argenteus, respectively (Fig. 3a). Six individuals of S. schweitzeri seemed to be hybrids between S. aureus (24–25%) and S. argenteus (75–76%). Five S. aureus strains, JKD6159, CA-347, H-EMRSA-15, JS395 and ST398/S0385, appeared to be hybrids among the S. aureus (red) and the S. argenteus (green) populations, with all individuals showing < 3.2% ancestry from S. argenteus (green) population. None of S. argenteus strains was found to be a hybrid. We also observed a higher ΔK value (= 579) when K = 3, and the three distinct populations corresponded to S. aureus, S. argenteus and S. schweitzeri, respectively (Fig. 3a). None of S. schweitzeri strains was found to be a hybrid, but four strains of S. argenteus were hybrids showing < 1.5% ancestry from S. aureus (red) and S. schweitzeri (blue) populations. And five strains of S. aureus were hybrids showing < 2.3% ancestry from S. argenteus (green) and S. schweitzeri (blue) populations. These results suggested that S. argenteus had an independent population structure with rare recombination occurred between the core genomes of S. argenteus and the other two species during the speciation. Notably, S. schweitzeri was found to be hybrids when K = 2, and it was independent population when K = 3, which implied this species might be the descendant of an ancient hybrid between S. aureus and S. argenteus. Nevertheless, it was not further discussed in this study, considering limited genomes and knowledge on this species.

Fig. 3
figure 3

Population structure of Staphylococcus aureus complex. a The population memberships of the inspected species for a priori defined number of clusters K = 2–4 inferred by the Structure software. Each individual is represented by a thin vertical line divided into K colored segments that represent the individual’s estimated membership fractions in K clusters. Populations are labeled below the figure. b The detection of the true number of clusters inferred by the Structure software and set ΔK = mean(|L"(K)|)/ sd(L(K)) as a function of K. ΔK attains its highest value when K = 2

Full size image

SAC species-specific genes

The species-specific genes (present or absent in all strains of one species) among SAC species were identified and shown in Table 1. We arbitrarily defined one gene is acquired in one species if the gene (including related pseudogene) is not found in the other species, and one gene is lost in one species if the gene is found in the other species. It shoud be a gene loss event if a pseudogene is found in one species and the related gene is present in the other species. Based on this definition, S. aureus acquired 24 unique genes while S. argenteus and S. schweitzeri acquired six and zero, respectively. S. aureus lost four genes while S. argenteus and S. schweitzeri lost 16 and ten, respectively. Some newly acquired genes were not required for growth under some conditions and became pseudogenes in some strains, for example, a gene encoding a protein N-acetyltransferase was acquired by but became pseudogenes in some strains (gene family 2832 in Table 1). Meanwhile, some genes were inherited from the common ancestor but were kept as pseudogenes in all strains of one species (Table 1). These pairwise species-specific genes and single species-specific genes were reannotated in COG, UniPro database, and KEGG Automatic Annotation Server [27, 38, 39], but most of their functions were unknown, especially those of S. aureus (Table 1). S. aureus and S. schweitzeri are different from S. argenteus in that S. argenteus lacks the crtOPQMN operon encoding the staphyloxanthin pathway (resulting in yellow colonies) [8, 16, 22]. Our data also indicated that white-colony forming S. aureus isolates might have resulted from inactivation of crtM, crtN, and crtO (Table 1) or a non-functional sigB operon [40]. Some species-specific genes are also prevalent in coagulase-negative staphylococci (CoNS) with low identities (< 90%), indicating that they might be inherited from the common staphylococcal ancestor, with some species of SAC losing them during speciation. The other species-specific genes, which were absent or present in a limited number of CoNS, probably were gained by HGT. Interestingly, all S. aureus-specific genes, most encoding small proteins (< 80 aa) with unknown functions, were absent in other staphylococci.

Table 1 Species-specific gene list of S. aureus complex (SAC)
Full size table

Mobile genetic elements (MGE), agr and the capsular polysaccharide gene cluster

Temperate bacteriophages of the major Siphoviridae family play an important role in the pathogenicity of S. aureus by mediating the HGT of virulence factors [29]. The detection of genes encoding these prophages’ integrases, integrase groups Sa1–3, indicated the presence of the prophages in the genomes of S. argenteus and S. schweitzeri (Fig. 1), which take many virulence genes along with them [29]. Methicillin-resistant S. aureus (MRSA) strains are of particular importance because they are a leading cause of nosocomial infections worldwide [41]. Methicillin resistance in MRSA is due to an acquisition of the staphylococcal cassette chromosome mec (SCCmec) element. Evidence for the presence of SCCmec in S. argenteus has been reported previously [6, 23] and is further substantiated in this study (Additional file 3: Table S1; mecA, gene family 2509; ccrA, family 2525; ccrB/C, family 2583). The clustered regularly interspaced short palindromic repeats-CRISPR-associated proteins (CRISPR-Cas) modules are adaptive immunity systems that are present in many archaea and bacteria [42], but are not very common in S. aureus (Fig. 1). Some strains of S. argenteus and S. aureus harbor a characteristic cas10/csm1 (gene family 2896 in Additional file 3: Table S1), and a subtype of CRISPR-Cas system III was present in them according to previous classification [42], especially strains of ST2250. Related genes of S. argenteus CRISPR-Cas system are a somewhat divergent from those of S. aureus strain JS395 (nucleotide sequence similarity 93.8%). However, further comparison with other S. aureus strains in NCBI showed that some S. aureus strains, such as M1169 (JEKD00000000) possessed a CRISPR-cas system with high similarity to that of S. argenteus (> 99%), suggesting that CRISPR loci can be mobilized and can transfer between different but closely related species.

The staphylococcal agr locus encodes a quorum sensing system (QS) that controls the expression of virulence and other accessory genes by a classical two-component signaling module, and it is distinct among staphylococcal species [43]. However, S. argenteus harbors S. aureus arg type I and IV while S. schweitzeri type I, II, and IV (Fig. 1). Further analysis suggested that agrA was highly conserved within species and had no common sequence differences among agr types. The agrB gene may suffer ancient HGT among species, especially that of S. schweitzeri agrI and ‘agrIII’ (maybe novel type), while the agrCD genes cluster primarily by arg type and secondarily by species (Additional file 7: Figure S6). The agrD alignment is shown in Fig. 4. It looks like that the evolution of agr locus among SAC species was polyphyletic. Based on this conjecture, the same agr types in different SAC species would have evolved independently. Gene agrD produces a ribosomal propeptide of which the middle section encodes the seven to nine residue autoinducing peptide (AIP) used as a QS signal molecule [43]. We found two possible novel types of agr most closely related to type III agr (Fig. 4). These novel agr types were present in three strains of S. argenteus (SJTU F20124, H115100079 and M051_MSHR) and one strain of S. schweitzeri (FSA090) (Fig. 1).

Fig. 4
figure 4

Comparisons of the predicted AgrD amino acid sequences from SAC species and agr types. Strains of the same agr type for each species have a same common sequence and are represented by a single sequence in this figure. Known autoinducing peptide (AIP) molecule sequences are marked in red rectangles. Potential novel AIP molecule sequences are marked in green rectangles

Full size image

Like many other bacterial pathogens, S. aureus produces capsular polysaccharide (CP) that enhances its resistance to clearance by host innate immune defenses, with strains of serotypes 5 and 8 being the most among clinical isolates [44]. We compared the CP gene clusters among SAC members (Additional file 8: Figure S7). For most of the genes common region of the CP gene cluster (capA-P, see reference 4 for a review), S. aureus, S. argenteus and S. schweitzeri were all phylogenetically distinct from each other. The exceptions were for the capM and capN genes where the phylogenetic trees showed that S. aureus and S. schweitzeri clustered together and HGT is likely to have occurred. In the serotype specific gene region (capHIJK), S. aureus was clearly divided into two groups consisting of strains of serotype 5 and 8, respectively. Strain S. schweitzeri FSA090 clustered with 14 S. argenteus strains in a group more closely related to S. aureus strains of serotype 5. Similarly, one strain S. argenteus, H115100079, clustered with five S. schweitzeri strains in a group most closely related to S. aureus strains of serotype 8. These results suggest that S. argenteus and S. schweitzeri may express their own serotype(s) that could have been derived from the ancestors of S. aureus serotype 5 and 8, respectively.

Virulence gene distribution

A hallmark of S. aureus infections is their frequent recurrence, which results from the manipulation of host immune responses by dozens of virulence genes [45]. Several virulence genes have been reported in S. argenteus and S. schweizeri, such as nuc m , pvl, sak, scn, seb, sec, selk, selq, and tsst [6, 11, 14, 18, 46], but the presence of many other potential virulence genes have not been tested for yet. Here, we performed a systematic investigation of genome sequences for the virulence gene distribution among SAC. The results indicated that (Additional file 9: Table S2): of 111 virulence genes previously reported in S. aureus, only five were not found in the genome sequences of any strain examined in this study, 92 (82.9%) were present in at least two SAC species, 85 (76.6%) in S. argenteus, and 86 (77.5%) in S. schweizeri. Of the 19 (17.1%) genes absent in S. argenteus or S. schweizeri, 11 encode enterotoxins or related superantigen-like proteins, and four encode leukocidins related proteins. Besides, of the 71 (64.0%) genes that were present in at least two species and at least three strains per species, 56 (50.5%) showed a significant difference in nucleotide sequence identity in a pairwise comparison of genes between species (p < 0.01). It is notable that nine genes (clfA-B, fnbA-B, sdrC-E, spa and coa) have repeat regions of uneven numbers in some strains, of which a subtyping method targeting spa gene (coding Staphylococcus Protein A) is widely used to characterize S. aureus isolates [47]. This could result in inaccurate identification of the genetic divergence based on these genes because of difficulties to align homologous nucleotide loci and a rapid change in the number of repeats within these regions. The principal component analysis (PCA) was employed to evaluate the differences in gene content and overall divergence of virulence genes among SAC members. The results suggested that the virulence gene content of each SAC species did not differ from each other while the divergence in core virulence gene is noteworthy (Fig. 5). In fact, most virulence genes were divergent at nucleotide sequence level among S. aureus, S. argenteus and S. schweitzeri (please find the ML trees in Additional file 9: Table S2). However, HGT may occur in some non-MGE-associated virulence genes, such as nucA of S. aureus JS395.

Fig. 5
figure 5

Principal Component Analysis of SAC virulence genes. a PCA based on the presence (1) and absence (0) of 106 virulence genes; b PCA based on the mean distance of one strain against the rest 50 strains among 30 core virulence genes. Five genes, sed, see, sej, selr and etb, were excluded from the gene because they were not present in any strain. Species are marked using different colored dots: S. aureus (red), S. argenteus (green) and S. schweitzeri (blue). The cluster of dots representing S. argenteus ST2250 strains is indicated by the arrow. The genes that provide the best strain differentiation are shown in the biplot

Full size image

Overall, the S. argenteus and S. schweitzeri pan-genome harbor all the virulence genes expressing the essential functions required for the pathogenicity in S. aureus. For examples, icaA-D encodes polysaccharide biosynthesis, which is critical to biofilm elaboration [48]; esaA-C, essA-C, and esxAB encodes the ESAT-6 system for secretion of exoenzymes or exotoxins into the extracellular milieu [49]; isdA-G and srtB code for heme uptake [50]. The genomic islands νSaα and νSaβ are also present in these two species, which carry many virulence determine genes [4]. Most of the virulence genes absent in the S. argenteus and S. schweitzeri pan-genome code for enterotoxins and other exotoxins (Additional file 9: Table S2), which are usually located in MGE and easily acquired or lost.

Geographic characteristics of S. argenteus ST2250

S. argenteus ST2250 was the most frequent lineage isolated, and strains of this ST have a broad geographic distribution [6, 8, 11, 13,14,15,16,17]. The widespread distribution of S. argenteus ST2250 may have occurred long ago or may have occurred only recently. Signatures from core housekeeping genes and variable genes suggest that geographic barriers promote divergence among microbial populations in the environment [26, 51]. However, increased mobility of human populations may break these geographical barriers for pathogenic bacteria [52]. To clarify the question as to why S. argenteus ST2250 was well distributed, the core and variable genes were further analyzed. The core genome of ST2250 contained 2348 single copy gene families, of which the concatenated alignment had a length of 2.07 M bp and 1177 single nucleotide polymorphisms (SNP). However, more than a half of the SNPs (653 bp) were due to indel variation located in the tandem repeat regions of three genes, including a hypothetic serine protease gene (gene family 1742 in Additional file 3: Table S1), a serine-rich adhesin for platelets gene (sraP, gene family 2098), and a hypothetic membrane anchored protein gene (gene family 2132). Due to the high rate of mutation, these loci were excluded from the analysis. The remaining 524 SNPs were used to infer maximum likelihood phylogenies (Fig. 6a). This phylogenetic tree matches the core gene nucleotide phylogeny (Fig. 1) very well with significant bootstrap support, but does not resolve a distinct population in each geographical region. Strains SH3, SJTU F21164 and F21224 were isolated from China (CN) and clustered with LBSA043 from Australia (AU) and 1299_SAUR from America (US). The presence/absence of gene content matrix based on variable genes also showed similar clustering (Fig. 6b). Notably, the number of SNPs was very small compared with S. aureus ST239 lineage, which were also distributed intercontinentally and from which 4310 SNPs were found among the isolates from different countries [52]. By comparison, only 382 VR and UQ gene families were identified in the S. argenteus pan-genome, of which strain LSBA043 was present in the least gene families (43) and strain SH3 in the most (167). All 382 gene families were annotated by ACLAME web site [53] and homologues of 247 were reported to be located in MGE. These results demonstrated that isolates of S. argenteus ST2250 lineage were very closely related and might have spread internationally by human hosts.

Fig. 6
figure 6

Maximum likelihood phylogenies of nine S. argenteus ST2250 genomes. Phylogenies were inferred from (a) the concatenated SNP-containing genes of core genome and (b) a presence/absence of gene content matrix. Numbers near each node correspond to bootstrap values in percentage of 1000 replications and the asterisk (*) indicates a bootstrap value of 100%. Geographical origins of the strains are marked on the branches: US, United States; AU, Australia; CN, China

Full size image

Discussion

Isolates of S. argenteus and S. schweitzeri were originally classified as parts of lineages of S. aureus because they share many phenotype and genotype properties [8, 16]. However, when using some molecular subtyping methods, considerable difficulties were observed, such as PCR amplification of aroE and glpf in MLST [8], and nucA identification of S. schweitzeri [16]. Few large-scale virulence gene investigations on S. argenteus and S. schweitzeri have been reported, and studies relying on PCR amplification of virulence targets should be regarded with skepticism because many S. argenteus and S. schweitzeri virulence genes are significantly divergent from those in S. aureus (Additional file 9: Table S2). Nonetheless, there is a great deal of genetic relatedness among SAC species, and homologous genes have been leveraged to develop many broadly recognized descriptive indexes among SAC, such as MLST, SCCmec, spa, agr, and prophage typing. Obviously, S. aureus, S. argenteus and S. schweitzeri are more closely related to each other than they are to other Staphylococcus species, of which these descriptive indexes are very distinct, for example, agr and prophage types of SAC never reported to be present in other staphylococci [29, 43]. Therefore, as previously suggested [16], the term SAC can be used to describe the closely related group of species including S. aureus, S. argenteus and S. schweitzeri.

Taking SAC as a starting point, many descriptions should be re-examined, for example, MRSA could refer to any SAC isolate harboring a SCCmec element. There is also a need to unify other descriptive indexes as well as diagnostic and typing methods, especially virulence gene detection and molecular subtyping. For example, we recently reported redesigned aroE and glpf primers for MLST of SAC and proposed a simple PCR method for identification and differentiation of SAC isolates [8]. The gene family list generated in the current study will also accelerate the development of uniform methods for the detection of many virulence genes (Additional file 3: Table S1; Additional file 9: Table S2). However, some virulence genes (especially core genes) may evolve quickly, with several very divergent subtypes occurring among or within SAC species. For example, alleles of S. schweitzeri nuc2 formed two distant clades, one closely related to S. aureus and the other closely to S. argenteus. Furthermore, divergent subtypes appear in many virulence genes of S. aureus and S. schweitzeri (note the high intraspecific average distance in Additional file 9: Table S2). The nuc M recently identified in S. schweitzeri [46] is a divergent homolog of the S. aureus nucA, and S. argenteus harbors an additional nucA variant. Thus, the current nomenclature of nuc M does not properly reflect its relationships to nucA (ortholog) and nuc2 (paralog). Therefore, much effort needs to be devoted to properly name genes among species of SAC.

We demonstrated here that, S. schweitzeri and S. argenteus have a similar distance to S. aureus, and that they are much closer to each other than to S. aureus (Fig. 2; ML trees in Additional file 9: Table S2). The relationships of these two species to S. aureus were reflected by whole genome ANI and DNA–DNA hybridization analysis in previous study [16]. Therefore, the most recent common ancestor of all three was earlier than that of S. schweitzeri and S. argenteus. However, this relationship is in contrast to host association for SAC infection. That is, S. aureus and S. argenteus are commonly associated with human disease, while S. schweitzeri rarely. S. argenteus and S. schweitzeri have lower GC content compared to S. aureus, which may be the result of mutation and selection involving multiple factors, such as the environment, symbiotic lifestyle and aerobiosis [54]. Additionally, S. argenteus shows a low occurrence in many regions where it has been reported [6, 8, 13]. It seems unlikely that S. argenteus emerged as a human pathogen and can escape clinical monitoring until the last decade. It is more likely that S. argenteus may have originally been a different ecotype from S. aureus and a host adaption occurred to allow it to infect humans, which is also implied by a very recent report [55]. Nonethelessly, we can not deny another possibility that S. argenteus have been previously mistaken as S. aureus and failed to spread and draw enough attention.

Although MRSA of S. argenteus was predominant in community-acquired isolates in some regions [23], S. argenteus showed resistances to fewer antimicrobials than is typical in S. aureus [12, 14]. Thus it seems that there is currently no need to modify therapeutic regimes for S. aureus and S. argenteus infections. Nevertheless, genes responsible for antimicrobial resistance are often located in MGEs and easily acquired, enhancement of S. argenteus antimicrobial resistance may be just a matter of time and the rapid international spread of S. aureus [52], and S. argenteus (this study) may aggravate the situation. In this study, it was suggested that S. argenteus harbors all the core virulence genes of S. aureus (Additional file 9: Table S2), but the expression and regulations of most of these virulence factors have not yet been characterized. In addition, many species-specific genes were recognized that presently have unknown functions (Table 1). The population structure analysis (Fig. 3) suggested distinct evolution background of the core genome between S. aureus and S. argenteus. These differences may result in different mechanisms of virulence to S. argenteus and S. aureus and different invasiveness between S. schweitzeri and S. aureus, which will impact the development of approaches to targeted drug design and therapeutic regimens in the post-antibiotic era [56]. For example, it was recently shown that diapophytoene desaturase (CrtN, the second enzyme of the staphyloxanthin biosynthesis pathway) is a potential target for drug development against S. aureus infections [57], but, to date, this drug targets are absent in all S. argenteus isolates. The future development of targeted drug therapies and diagnostic tests that distinguish these two species should allow for improved patient outcomes.

S. argenteus was found to possess most of the virulence genes of S. aureus (Fig. 5; Additional file 9: Table S2), which implied a pathogenic potential similar to S. aureus at a genomic level, and the international spread of S. argenteus ST2250, most likely anthropogenically, may worsen the situation. Furthermore, species-specific genes among SAC members recognized in this study may be responsible for the different ecotypes. Further investigation on the function of these unique gene products will help determine their contribution to speciation and ecotype. It is fascinating that the diversity of the agr locus indicates a polyphyletic relationship among SAC, suggesting that each species of SAC was derived from more than one common ancestor [58]. However, the number of available genome sequences for S. argenteus and S. schweitzeri are still very limited and their diversity is unclear, so it was not discussed in this study.

Conclusions

We proposed to use the term SAC to cover S. aureus, S. argenteus and S. schweitzeri to indicate their close relationships. Considering difficulties in subtyping and virulence gene detecting using the methods designed for S. aureus, much effort needs to be devoted to developing universal and robust methods among SAC species. It is found in this study that S. argenteus harbored most virulence genes of S. aureus and had spread internationally, which suggested that S. argenteus may have a similar pathogenic potential as S. aureus. However, genomic divergence was also observed, especially regions of virulence genes, which draw necessary to distinguish S. argenteus from S. aureus in routine practice. Data from this study and previous ones draws a hypothesis that S. argenteus should have originally been a different ecotype from S. aureus and a host adaption occurred to allow it to infect humans. Finally, the clinical importance of S. argenteus underscores the need for broader genomic epidemiological investigations. Such studies would be expected to provide clarification on the origin of S. argenteus and the evolution of its infectivity and pathogenicity. Additional microbiological studies are also needed to determine the species environmental niche and further elaborate mechanisms of pathogenicity.

Methods

Strains and genomes

The five S. argenteus strains were isolated from China and characterized previously [8]. The genomes were sequenced using a MiSeq 300PE sequencer (Illumina, San Diego, CA, USA) at GENEWIZ (Suzhou, China). Velvet version 1.2.10 [59], SSPACE version 3.0 [60], and GapFiller version 1–10 [61] software packages were used for genome assembly. Annotation was performed by the NCBI Prokaryotic Genome Annotation Pipeline (https://www.ncbi.nlm.nih.gov/genome/annotation_prok/). An additional 46 genomes, including ten of S. argenteus, six of S. schweitzeri, and 30 of S. aureus, were obtained from the NCBI public database (ftp://ftp.ncbi.nlm.nih.gov/genomes/refseq/bacteria/Staphylococcus_aureus/) and Whole Genome Shotgun (WGS) sequencing projects databases (http://www.ncbi.nlm.nih.gov/Traces/wgs/) in May 2016. The genomic data was generated from the genome sequences and related files from NCBI. The STs were assigned by the public MLST database of S. aureus (http://saureus.mlst.net/), using the seven homologous fragments from the genome sequences.

Assignment of homologous gene families

The deduced amino acid sequences of all CDSs from the 51 genomes were adjusted to a prescribed format and were grouped into homologous gene families using OrthoMCL version 2.0.9 [62] based on sequence similarity. The BLAST reciprocal best hit algorithm [63] was employed with a 70% match cutoff and 1e-5 e-value cutoff, and Markov Cluster Algorithms (MCL) [64] were applied with an inflation index of 1.5.

To account for pseudogenes and cases where genes were missed in the gene-calling step, BLAST (blast + package, version 2.2.29) [65] was used to align all genes of one family (assigned by OrthoMCL) against the genome sequence of the strains in which the family was not previously identified. If an alignment with at least 90% nucleotide sequence identity covering at least 90% of the sequence length was detected in the genome, the gene/pseudogene was considered present and the related location was recorded in the complete homologous gene families list using a different font (Additional file 3: Table S1). This produced a more robust pan-genome outcome and accommodated the differences in the gene prediction. For example, 137 and 140 genes were newly detected in genomes of strains S. aureus DAR4 and RKI4, respectively. The sizes of the four components of the pan-genome (CR, CV, VR and UQ) were simulated as has been done in previous studies [66] using Perl scripts. For example, pan-genome analysis of SAC was conducted starting from one single genome to 51 genomes. Genomes were added 1000 times in a randomized order without replacement for each fixed number of genomes, and the gene reservoir was accumulated. The functional category of each homologous gene family was determined by using the BLAST program locally or on the web server against the Cluster of Orthologous Groups (COGs), UniProt, KEGG Automatic Annotation Server (KAAS), and ACLAME database [27, 38, 39, 53], with 50% identity cutoff and 1e-5 e-value.

Phylogenetic analysis and bANI

To determine the phylogenetic relationships of SAC members based on genomic data, both supermatrix and gene content methods were applied to infer phylogenetic trees. For the supermatrix approach, 1375 single-copy orthologous genes shared by all 51 SAC strains and S. simiae CCM 7213 (AEUN00000000.1) were selected from the homologous gene families. For each orthologous family, protein sequences were aligned using Clustal Omega version 1.2.1 [67] and the resulting alignments of individual proteins were concatenated to infer the phylogeny using the maximum likelihood algorithm (ML) in RAxML version 8.1.2 [68] under the substitution matrix JTTDCMUT which was selected by the Perl script in the software package. The gene content matrix was parsed for the gene content method using a phyletic pattern indicating the presence (1) or absence (0) of the respective non-core genes of all strains. The Neighbor-Joining distance between pairwise genomes was calculated based on the gene content matrix to reconstruct the gene content dendrogram using package Splits Tree version 4.13.1 [69].

The SNPs among S. argenteus ST2250 strains were detected, extracted and concatenated from nucleotide sequence alignments of 2348 orthologous gene families using Perl scripts. The maximum likelihood algorithm (ML) in RAxML version 8.1.2 [68] was used to infer the phylogeny under the substitution matrix GTR. The Neighbor-Joining phylogenetic trees of 16 genes of capsular polysaccharide (CP) were constructed based on amino acid sequence using the Poisson correction method in MEGA6 [70]. The bANI tests were simulated as done in previous studies [26]. Briefly, the core genes’ nucleotide sequences of each of the genomes were extracted and used in BLAST against the other whole genome sequences, and all the identity values were grouped into six groups, including three intraspecific groups and three interspecific groups. Then the data of each group were counted and presented as percentages in a histogram.

Population structure

To investigate the population structure of the SAC and the relationships among species groups, the Markov chain Monte Carlo (MCMC) based program Structure version 2.3.4 [36] was used to cluster individuals into populations. Initially, we treated core genes as MLST sequence data from Extended FASTA Format into the Structure Format using xmfa2struct (available from http://www.xavierdidelot.xtreemhost.com/clonalframe.htm). The admixture ancestry model with the assumption of correlated allele frequencies among populations was used. We ran the simulation five times under a burn-in period of 10,000 and a run length of 20,000 MCMC, without prior population information. K values from one to seven were tested to identify the best K value, represented by the highest value of K and ΔK [36]. Results of the five independent runs were averaged for each K value was used to determine the most likely model, i.e., the one with the highest likelihood, and they were subsequently plotted using Distruct version 1.1 [71]. The identification of the best K was evaluated following the ΔK-method through local program Structure Harvester [37].

Species-specific genes and virulence genes

Species-specific genes were inferred from the homologous gene families list (Additional file 3: Table S1) and reanalyzed against the missing species genome using BLAST algorithm to confirm that they were lost or became pseudogenes. The 111 virulence genes were identified and collected from a virulence factor database (VRprofile, http://bioinfo-mml.sjtu.edu.cn/VRprofile/) [72] and a recent review of the literature [45]. Reference sequences of these genes were used with BLAST algorithm to find the homologous gene families presented in Additional file 3: Table S1. Some virulence genes, which have a paralogous relationship and may be assigned to one gene family by OrthoMcl software (Additional file 9: Table S2), were separated for subsequent analysis. The pairwise k-tuple distance among each of these virulence genes was calculated and outputted from the Clustal Omega (version 1.2.1) created alignment [67]. The distance matrix was then parsed using Perl script. Principal component analysis (PCA) were performed using software Past software version 3.13 (http://folk.uio.no/ohammer/past/), and other statistical analyses were performed using the R package (version 3.1.1, http://www.R-project.org).