The strain collection analyzed in this study included 26 GBS clinical isolates recovered in China during 2013–2014 from neonates with invasive disease, mostly in Guangzhou hospital and one isolate in Changsha. Eight strains were prospectively isolated from cases enrolled during a former observational study (Liu et al., 2015) while 16 isolates were retrospectively collected in the same region. GBS COH1, a clinical isolate obtained from an infected newborn with sepsis at the Children’s Orthopedic Hospital (COH; Wilson and Weaver, 1985) was used as reference serotype III-ST17 strain in all performed assays. All GBS isolates used in this study and their relative information are listed in Table 1. A GBS case was identified when an infant younger than 3 months was GBS-culture positive from a sterile site (blood or cerebral spinal fluid), with any of the signs of clinical disease (e.g., sepsis, pneumonia, or meningitis). The presence of GBS in samples was determined within routine internal hospital laboratory services using the BD BACTEC^TM 9210 Culture System and API 20STREP (Guangzhou) and BD BACTEC^TM 9120 Blood Culture System and CAMP test (Changsha). GBS isolates were collected and stored in STGG stock medium at –70°C.

S. agalactiae strains were grown at 37°C in 5% CO₂ in Todd Hewitt Broth (Difco Laboratories), in trypticase soy-agar or Muller-Hinton-agar supplemented with 5% sheep blood.

The study protocols were approved by the Novartis Vaccines Research Institutional Board and by the Medical Ethical Committees of Guangzhou Women and Children’s Medical Center and of Changsha Hospital for Maternal and Child Health. The study was conducted in accordance with guidelines for Good Pharmacoepidemiology Practices and the Declaration of Helsinki. Informed consent was obtained from all subjects.

Serotyping was performed using the standard Strep-B-Latex rapid agglutination method (Strep-B-Latex kit; Statens Serum Institut, Denmark). A heavy suspension of the test organism was prepared by harvesting bacteria from the agar plate in 250 μl of phosphate-buffered saline (PBS, pH 7.4). A 20 μl aliquot of the bacterial suspension was mixed with 5 μl of each of the ten Latex reagents (rabbit antisera Ia, Ib, and II–IX) onto a disposable card. The reaction card was rotated slowly and a positive reaction was indicated by agglutination appearing within 30 s.

The E test strips (Biomiereux) method was used to assess the antimicrobial susceptibility by determining the minimal inhibitory concentrations (MICs) of these antibiotics: azithromycin, benzylpenicillin, clindamycin, erythromycin, levofloxacin, ceftriaxone, claritromycin, tetracycline, vancomycin. The breakpoints adopted were in accordance with the 2015 criteria set by the European committee on antimicrobial susceptibility testing based on tables for interpretation of MICs version 5.0¹. The susceptibility for kanamycin, gentamycin, streptomycin, lincomycin, and spectinomycin was defined by Société Française de Microbiologie².

Genomic DNA was prepared from overnight GBS cultures by a standard protocol for Gram-Positive bacteria, using Mutanolysin-treatment of bacterial cells and a Gene Elute Bacterial Genomic DNA kit (Sigma–Aldrich) according to the manufacturer’s instructions.

A molecular capsular genotyping was used to confirm the Latex serotype assignment and to identify the capsular type of GBS isolates resulting serologically non-typeable (NT). The assay was performed by multiplex PCRs using a previously published method (Poyart et al., 2007) modified in house by including a new primer pairs specific for serotype IX detection (5′-AGATCATTTGTTCCTGATTCCCTAAAG-3′; 5′-AATCATCTTCATAATTTATCTCCCATT-3′). The pattern for type IX strains detected two amplicons (371 and 1,132 base pairs). The protocol is based on two multiplex PCRs, one containing the primer pairs specific for serotypes Ia, Ib, II, III, and IV and the other containing the primer pairs specific for serotypes V, VI, VII, VIII, and IX.

Pilus genes were PCR-amplified using primers specifically annealed to conserved genomic regions external to the coding sequences as already described (Margarit et al., 2009).

The presence of the Gbs2018/hvgA gene was assessed as previously described (Lamy et al., 2006; Tazi et al., 2010).

Whole-genome sequencing was performed on libraries generated using 1 ng of purified bacterial DNA processed by Nextera XT DNA Library Preparation Kit (Illumina). Samples were normalized with a bead-based approach, pooled in an equal amount and finally diluted for the deep sequencing run, following Nextera XT protocol. Sequencing was carried out on a HiSeq 2500 platform in a 100 bp paired-end run with the TruSeq SBS version 3 chemistry (Illumina).

The newly sequenced genomes were assembled using SPAdes 3.0.0 (Bankevich et al., 2012). The contigs number ranged between 38 and 54, with an average of 45. All obtained draft genome assemblies were annotated with Prokka (Seemann, 2014) using a genus specific non-redundant BLAST database, built from all the translated annotations retrieved from RefSeq (Pruitt et al., 2009) for TaxId 1311, clustered with CD-HIT (Fu et al., 2012). The newly sequenced genomes were deposited in the NCBI database under the BioProject ID PRJNA324749.

The complete genome of GBS COH1 (Accession number HG939456) was used as reference sequence for genome analysis. Genome comparison and visualization was performed by BRIG software (Alikhan et al., 2011).

Sequence types were assigned by comparing the genomic sequences with the seven housekeeping genes of the GBS MLST (Multi Locus Sequence Type) system (Jones et al., 2003)³ using the Bio-MLST-Check-2.0.1510612 suite⁴. Clonal Complexes (CCs) were assigned as previously reported (Rosini et al., 2015). ICEs sequence alignments and comparison were performed using Easyfig (Sullivan et al., 2011). ICE sequence similarities were searched on the ICEberg server (Bi et al., 2012). ICESag(RR1) from CH-2 isolate was used as reference sequence for reads mapping and subsequent sequence comparison of the ICEs from all GBS genomes. Reads mapping and multiple alignments were performed using Geneious R9 with default settings⁵. The distribution of the antibiotic resistance genes present in the Antibiotic Resistance Database (Liu and Pop, 2009) and the RESFAMS database (Gibson et al., 2015) within the genomes of the 14 newly sequenced isolates was investigated using custom scripts based on the BLAST and HMMer algorithms.

To investigate the genetic characteristics of a clinically relevant GBS population from Southern China, we collected 26 isolates from infants affected by sepsis and meningitis in Guangzhou/Changsha area. Eight strains were obtained from EOD cases and 18 from neonates with LOD.

We first assessed the CPS type and the pilus islands (PIs) distribution by serological and PCR-based assays. Serotype III was the most prevalent accounting for 58% of all isolates (15/26) among both EOD (4/8) and LOD (11/18), followed by type Ib (19%; 5/26), type V (11%; 3/26), and type Ia (8%; 2/26). One genotype II strain was NT by serological analysis (4%; 1/26) (Table 2).

To verify whether the prevalent serotype III was associated to the hypervirulent ST-17 lineage as in previous observations related to the same region (Wang et al., 2015a), the isolates were screened for the presence of the hvgA gene, a characteristic feature of the ST-17 strains (Lamy et al., 2006). PCR results showed that 14 strains, all CPS type III, were hvgA positive and thus ST-17-associated. Only one type III isolate was negative for the presence of the gene and, as expected, the same was true for all strains belonging to other serotypes.

Concerning PIs distribution, half of the analyzed isolates only harbored PI-2b in their genomes (13 out of 26 strains). Moreover, all these isolates presenting the PI-2b-alone genotype belonged to CPS type III-ST-17. The PI-1+2a combination was present in eight strains distributed among four CPS types (three for type Ib and V each, one for type II and III). Finally, PI-2a alone was found in the four strains belonging to CPS types Ia or Ib (Table 2).

In conclusion, the vast majority of type III strains (14/15) belonged to the ST-17 lineage, and all but one carried the PI-2b alone, an uncommon genotype among type III-ST-17 isolates characterized so far (Springman et al., 2014).

To further investigate the molecular characteristics of the 14 identified hvgA-positive serotype III strains we performed WGS. In silico MLST analysis and CC assignment of the newly sequenced genomes confirmed that all strains belonged to CC17, and specifically 12 were strictly ST-17 while two had novel ST profiles that were single locus variants (s.l.v.) of ST-17. Further sequence analysis confirmed both the presence of the type III cps operon and the PI-2b in all strains and the absence of the otherwise well-conserved PI-1 in 13 of these 14 CC17 strains. To assess DNA sequence similarity, we performed BLAST pairwise sequence comparisons between each genome and the complete chromosome of the ST-17 reference strain COH1. Figure 1 provides a comprehensive view of all single comparisons by displaying similarity between the central COH1 reference sequence and the other draft genomes as concentric rings. All analyzed genomes shared more than 99% sequence identity to the reference genome across at least 94% of its whole length. We also identified three distinct genomic loci in the COH1 reference chromosome (locus 1, 2, and 3) that could not be found in some of the newly sequenced strains. These regions are indicated by interruptions in the circular map that represent each genome (Figure 1A). Locus 1 is absent in all analyzed genomes except for one, while locus 2 and locus 3 are absent in 9 and 7 genomes, respectively. According to the curated annotation of the COH1 genome, locus 1 includes the entire pilus island 1 plus 15 genes belonging to different functional categories (Supplementary Table 1). Locus 2 and 3 mainly contain predicted open reading frames belonging to mobile elements along with poorly characterized hypothetical genes. Particularly, locus 2 includes a well-known large integrative conjugative element (ICE) Tn916 carrying the tetM gene conferring resistance to tetracycline (Da Cunha et al., 2014), while locus 3 hosts a putative prophage carrying different genes of unknown function (Figure 1B).

Horizontal gene transfer events determining exchange of large genome tracts are commonly considered an important mechanism in the diversification of S. agalactiae (Brochet et al., 2008). Therefore, we hypothesized that the loss in 13 out of 14 sequenced isolates of the genomic region corresponding to locus 1 could be counterbalanced by the acquisition of new genetic material, giving a potential selective advantage to these pathogenic strains. To test this hypothesis, the sequenced draft genomes were analyzed to look for new genes absent in the reference strain. The assembly of the sequenced short reads permitted to partially rebuild the regions corresponding to locus 1 for all 13 strains. We identified a large putative integrative and conjugative element (ICE), ICESag(RR1) in the PI-1 minus genomes replacing the entire locus 1. For one genome, namely CH-2, we were able to isolate the full-length ICE since it was harbored in a single sequence contig. This mobile element presented a typical mosaic organization, composed by different modules that showed a high sequence similarity to other ICEs found in different other streptococcal species such as S. suis, S. dysgalactiae, S. pyogenes, S. pneumoniae. Among all the ICEs present in the ICEberg database (Bi et al., 2012), the newly identified ICESag(RR1) displayed the highest sequence similarity (91% identity in a 5772 bp region) to S. suis ICESsu(SC84). More detailed sequence analysis revealed that the central module of this large mobile element contained several ORFs annotated as putative antibiotic resistance genes, including tetO, ermB, ant6, and aphA, known to confer multiple resistances to tetracycline, erythromycin and aminoglycosides (Figure 2A). Mapping of the reads of the remaining 12 strains lacking locus 1 on the extracted ICE revealed that the sequence identity of this tract was highly conserved in each strain (>98%) (Figure 2B). Finally, in strains lacking locus 2 or locus 3 we observed no additional genetic elements in the corresponding regions.

The previous analysis showed that the invasive strains of our collection could be resistant to multiple antimicrobial compounds due to the acquisition of an uncommonly large number of resistance genes hosted on mobile elements. To gain a clearer picture of the antibiotic resistance gene repertoire of these strains we performed a deep search of their genomes using a custom approach integrating different antimicrobial resistance gene databases. As shown in Figure 3, all sequenced genomes harbored genes predicted to confer resistance against aminoglycosides and tetracycline antibiotics. Moreover, with the exception of GZ13, all sequenced isolates hosted at least one macrolide resistance gene. In most cases a single copy of each antibiotic resistance gene was detected, while two or three copies of ant6 were found in all strains, except for GZ02 that carried a single copy. Moreover two additional genes predicted to confer resistance to spectinomycin and lincomycin, ant9 and lnuB, were identified in a subgroup of strains (GZ02, GZ20, GZ06, GZ11, and CS01). These genes were contained in a new gene module similar to the previously described ICE SGB76 (Montilla et al., 2014), that was found inserted in the ICE Sag(RR1), thus resulting in a previously undescribed ICE configuration that was named Sag(RR2) (Figure 4).

To verify whether the identified genes were able to confer antibiotic resistance to the strains harboring them, we tested susceptibility on plates to 14 different chemical compounds belonging to different antibiotic classes (benzylpenicillin, ceftriaxone, streptomycin, spectinomycin, kanamycin, azithromycin, claritromycin, erythromycin, clindamycin, lincomycin, tetracycline, levofloxacin, gentamycin, and vancomycin). The results confirmed that the antibiotic resistance profiles predicted in silico closely matched the resistance phenotype observed in vivo (Figure 3). Overall, these results revealed a direct correlation between the presence of a specific antibiotic resistance gene and a resistance phenotype. Compared to COH1 phenotype, 13 CC17 Chinese strains showed additional antibiotic resistance to macrolides and lincosamides. Moreover, the subgroup including the strains carrying the ant9 and lnuB genes showed also resistance to lincomycin and spectinomycin. Conversely, benzylpenicillin, ceftriaxone, levofloxacin, gentamycin, and vancomycin proved to be effective in preventing growth of all isolates.