General features of the sequenced genomes

Among the 33 known serotypes, serotypes 1, 2, 3, 7, 9 and 1/2 are the most prevalent in pigs, and the strains causing human infections were also found among these serotypes [24–28]. Although 8 genome sequences of strains from serotype 2 were available, there was little information about the other serotypes, except for our recently updated genome sequences for serotypes 3 [34] and 14 [35]. In this study, whole genome sequencing was performed on 4 prevalent Chinese S. suis strains belonging to serotypes 1, 7, 9 and 1/2. Each of the 4 genomes was sequenced to a high level of redundancy (sequencing depth was 722 to 1627 fold). We filtered low-quality reads and used only high-quality reads for assembly. Reads for each genome were assembled into scaffolds, with 26 to 94 large scaffolds (>500bp) obtained per genome. Then scaffolds were aligned to the published genomes of S. suis to obtain linkage information for gap closure. All 4 annotated complete genomes were deposited in GenBank.

Every genome consisted of a single circular chromosome with an approximate size of 2 Mb (Figure 1). Genome and assembly statistics for each strain were summarized in Table 1. The number of predicted ORFs for the 4 sequenced genomes was ranged from 2030 to 2136, and approximately 71% of the ORFs was assigned biological functions. The average gene length varied among different strains and was related to the number of pseudogenes and truncated genes presented. The genome of strain D9 (serotype 7) carried the greatest number of disrupted genes (126, or 5.9%), and conversely, SS12 (serotype 1/2) carried the lowest number (63, 3.0%) (Additional file 1). The average GC content was 41.2%, which was consistent with previous studies [8, 31], and the genomic regions exhibiting an aberrant GC content may be the sites of horizontal gene transfer in different strains. Additionally, several IS elements were identified, and the number of IS elements was found to be similar to that of previously sequenced strains, such as P1/7, SC84 and BM407.

Identification of gene clusters

All CDSs from the 13 completely sequenced S. suis genomes used for clustering were available in multi-FASTA format in the Supplemental Material (Additional file 2). There were 2374 S. suis orthologous gene clusters and 1211 unique genes, and the observed pan-genome shared by the 13 strains consisted of 3585 genes. The core genome of these strains comprised 1343 genes, accounting for 66.5% of total CDSs, and 28.9% of the genes were "dispensable" because they were shared by at least 2 strains, but not by all. All of the unique genes from these genomes only accounted for 4.6% of genes, but the percentage in each strain varied considerably (Figure 2). Non-core genes, including both dispensable genes and unique genes, usually play roles in nonessential metabolism and are more associated with virulence, environmental adaptation or serotype determination than core genes. Strain D9 possessed the highest proportion of non-core genes (38.6%), and strain P1/7 had the lowest proportion (26.4%). This may reflect different levels of gene gain and loss during the evolution of these strains or serotypes. Pairwise gene content comparisons among the 13 genomes indicated that the number of genes involved in gain and loss events between the strains was 587 on average. The largest number of gene difference between the strains was 1090, which was identified between strains D12 and 05ZYH33, and the minimum was 88, identified between strains P1/7 and SC84. A COG functional classification for core and non-core genes was performed, and the results showed that non-core genes were most likely to be assigned to categories, such as carbohydrate transport and metabolism, replication, recombination and repair, whereas core genes were more often associated with translation and ribosomal structure and biogenesis (Figure 3). Genes involved in translation, ribosomal structure and biogenesis lipid transport and metabolic functions were much less prevalent among non-core genes, while defense-related genes were more likely to be found among the core genes.

Core and pan-genome analysis of S. suis

S. suis core genome

To determine the core genome of S. suis, the number of conserved genes found upon sequential addition of each new genome was extrapolated by fitting a decaying function that was considered to provide the best fit to the dataset (Figure 4A). Although the number of core genes initially decreased with the addition of each new genome, the core genome appeared to reach a plateau at approximately 1126 genes for S. suis species. The core gene number in each genome varied slightly because of the involvement of duplicated genes and paralogs in the shared clusters.

Phylogenetic relationships among different serotype strains

We used two methods to investigate the phylogenetic relationships among different serotypes, one of which was based on gene presence or absence among different strains, while the other one utilized the concatenated sequence of all single-copy core genes with exactly identical lengths from the 13 complete genomes. Figure 5A displays the phylogenetic relationships among the different strains based on the large sequence alignment of 522 core genes with the same length in each cluster. With the exception of serotype 14 strain JS14 and serotype 1/2 strain SS12, the non-serotype 2 strains appeared to be phylogenetically distinct from the serotype 2 strains, and could be assigned to a common clade. In this clade, serotype 7 strain D9 and serotype 3 strain ST3 were more closely related than any other pair. All serotype 2 strains presented an extremely short evolutionary distance from each other, indicating that these strains were probably derived from a recent common ancestor. It can also be inferred that phylogenetically serotype 1/2 and serotype 14 may be more closely related to the serotype 2 strains than the other 4 serotypes. However, to confirm this, more strains of other serotypes need to be sequenced. Figure 5B shows that the UPGMA (Unweighted Pair Group Method with Arithmetic Mean) phylogenetic tree reflect the number of gene gains and losses between all pairs of the 13 strains. The topologies of the MrBayes tree and the UPGMA tree bore some similarities. The 4 strains from serotypes 1, 3, 7 and 9 were also included in a common branch and differed greatly from the serotype 2 strains, whereas the serotype 1/2 and 14 strains were grouped into the same clade with A7 and GZ1 from serotype 2. The main difference between the two trees was that the two Chinese isolates, 05ZYH33 and 98HAH33, were more evolutionarily distant from the other serotype 2 strains indicated in the UPGMA tree.

Genomic arrangement of S. suis strains

A global multi-genome alignment of all 13 complete genomes was performed, and the results showed that some rearrangement occurred (Figure 6). These genomes could be classified into 3 categories according to their collinearity. All serotype 2 strains except for BM407, as well as the serotype 14 strain JS14 and serotype 1/2 strain SS12, were quite similar with respect to genome structure, with the exception of some small insertions. The genomes of BM407, D12 and ST1 shared a similar synteny with each other, and they displayed a large inversion when compared to that of other serotype 2 strains. The D9 and ST3 genomes were collinear along their length, with the exception of an insertion in the D9 genome. This is interesting because these synteny types were similar to some extent to the phylogenetic relationships seen among these strains.

Genes involved in CPS biosynthesis

S. suis is surrounded by a capsule that has been shown to be essential for its virulence [37, 38]. It has been demonstrated that the presence of the capsule can decrease the activation of the PI-3K/Akt/PKCα signaling pathway involved in phagocytosis processes [39] and allows the bacterium to escape being killed by both macrophages and neutrophils [37, 38, 40]. The antigenic properties of capsular polysaccharides are the basis for serotype characterization. Only the structure of the serotype 2 capsular polysaccharide had previously been determined, and the genes involved in the biosynthesis of capsular polysaccharides had been found to be clustered in a single locus [41]. Orf2Y and Orf2Z are located upstream of the operon and may be involved in the regulation of these cps genes. Most of the serotypes include these two genes, as determined by hybrid assays [27], and the genome sequences indicate that these genes share very high sequence similarity. At the cps locus, the orfX and cpsA to cpsD genes could be hybridized in most serotypes [27], and these genes presented in all sequenced serotypes strains and highly conserved, indicating that they were involved in common functions related to the biosynthesis of the capsular polysaccharides, such as regulation, chain length determination and export, which were the functions of the homologous genes in Streptococcus pneumoniae[38]. The other 7 genes at the locus may be responsible for its specific CPS structure (Figure 7). The agglutination test indicated that the serotype 1/2 strains could react with hyperimmune sera against serotype 1 and serotype 2, and the sequenced genes encoding the CPS biosynthetic enzymes of serotype 1/2 showed high uniformity compared to serotype 2, suggesting that the modifications of the cps of serotype 1/2 were similar to those of serotype 1. Corresponding to the findings of a previous report [27], the genes coding for the CPS biosynthetic enzymes of serotypes 1 and 14 were highly conserved, indicating that the determinants for both serotypes include not only CPS structure, but also the modifications of polysaccharides (Figure 7).

The prevalent serotypes supply a potential recombinant site for a pathogenic island (89 K)

The two large-scale outbreaks in China in 1998 and 2005 prompted researchers to determine which changes in the S. suis genome make it so highly virulent. Using comparative genomic analysis, an 89-kb sequence was identified only in the Chinese epidemic strain [31]. The subsequent investigation indicated that the 89-kb represented a GI-type T4SS-mediated horizontal transfer of a pathogenicity island that could be transferred to the recipient strain through a 15-bp sequence specific recombination event, although the transfer could be successfully observed only to serotype 2 [33]. Because the 89-kb harbored necessary elements for horizontal transfer, such as integrase, excisionase, DNA relaxase and so on, suggesting that this pathogenicity island maintained the potential to transfer to the recipient strain harboring the 15-bp sequence. The Genomic analysis indicated that the pathogenicity island did not exist in the other sequenced prevalent serotypes and such a 15-bp sequence could be found in the genomes of sequenced serotypes 1/2, 2, 3, 7 and 9, but not in serotype 1. More surprisingly, the flanking sequence structure of the 89 K region in the epidemic strain SC84 showed high similarity with the other sequenced serotypes, suggesting that these prevalent serotypes harboring the site for homologous recombination (the 15-bp sequence) would have the potential to act as recipient strains for the pathogenic island from the epidemic strain.

Bacterial strains

Sequencing and assembly

Bacterial genomes were sequenced at the Beijing Institute of Genomics (China) using a whole-genome shotgun sequencing strategy and Illumina Genome Analyzer sequencing technology. For each sample, a paired-end sequencing library containing fragments of approximate 500 bp was constructed. The short reads were filtered for quality and assembled with SOAPdenovo (http://soap.genomics.org.cn/soapdenovo.html). To fill the intra-scaffolds gaps, we used paired-end information to retrieve read pairs that had one read that was aligned to the contigs and another read that was located in the gap region. With this information, we did a local assembly for the collected reads. Then, these scaffolds were ordered relative to the genome of S. suis strain 05ZYH33 (deposited in the NCBI database; GenBank accession number CP000407) using MUMmer3 [42]. Gaps were closed by primer walking and sequencing of PCR products. Possible misassemblies were corrected using PCR amplification and direct sequencing. Sequences were edited in Consed [43].

Genome annotation

Initially, Open Reading Frame (ORF) prediction was performed using Glimmer3 [44] and Genemarks [45], and the results were amalgamated. To avoid possible missing coding sequences, entire DNA sequences were compared to all known protein sequences from other published S. suis strains using BLAST searches. Then, all predicted ORFs were translated into amino acid sequences and compared against the non-redundant protein (nr) database using the BLASTp program, with a maximum expectation value of 1 × 10^-6. ORFs with no BLAST hit to any other protein were automatically annotated as "hypothetical proteins." tRNAs and rRNAs were identified using tRNAscan-SE [46] and RNAmmer1.2 [47], respectively. Insertion sequence (IS) elements were found with IS Finder [48]. Genome islands (GIs) were identified using IslandViewer [49], which integrates three different genomic island prediction methods, followed by manual inspection.

The four annotated complete genome sequences have been deposited in GenBank with the accession numbers CP002640 (SS12), CP002641 (D9), CP002644 (D12) and CP002651 (ST1).

Whole genome alignment and ortholog identification

Multiple genome alignments for 13 completely sequenced strains were constructed and visualized using the progressive Mauve program in Mauve v2.3.1 [50] at default settings.

All CDSs were extracted from the 13 S. suis genomes, and they were grouped into homologous clusters using InParanoid4 [51–53], which employs a BLAST reciprocal best hit algorithm, with default parameters.

Core and pan-genome analysis

Tables of homologous clusters from InParanoid4 were compiled for identifying shared and unique genes. The numbers of conserved genes and unique genes depend on how many strains are taken into account. Thirteen strains with complete genome sequences were simulated in all possible combinations. The sizes of the core genome and novel gene set were calculated for each combination and then extrapolated using several functions to find a best fit from the mean number at each sampling point [54].

Phylogenetic analysis

Phylogenetic trees of S. suis strains were constructed using two different methods [55]. The first utilized multiple sequence alignments of 522 single-copy core genes with nearly identical lengths and exactly one member in each of the compared strains. The alignments of these genes were concatenated into one large sequence alignment with a length of 457779 bp, and a phylogenetic tree was reconstructed using MrBayes 3 [56, 57] (200,000 generations, sampled every 100 generations with a gamma distribution model and invariant class). The second method was based on the presence or absence of genes in the pan-genome. Genetic distances were defined as Σ _n | g _{n, i} - g _{n, k} |, where g _{n, i} is 1 if gene n is present in strain i and is zero otherwise. A dendrogram was generated using the UPGMA (unweighted pair group method with arithmetic mean) method implemented in the Phylip package [58].