δ-Haemolysin assay

The S. aureus accessory gene regulator (agr) activity was measured by δ-haemolysin production on Trypticase soy agar containing 5% sheep blood using S. aureus strain RN4220, as described previously (Traber et al., 2008). Evidence of enhanced haemolysis between RN4220 and a test isolate was considered positive for δ-haemolysin production. Repression of agr was defined as an absence of δ-haemolysin production.

Antimicrobial susceptibility testing

Isolates were subcultured onto HBA for 18–24h before performing any antimicrobial susceptibility testing. Vancomycin and teicoplanin macro E-tests were performed by applying 200μl 2.0 McFarlane bacterial suspension onto brain heart infusion (BHI) agar. The plates were incubated for 48h before readings were taken. An oxacillin E-test was performed using 0.5 McFarlane suspension onto 2% NaCl BBL Mueller–Hinton II agar according to the manufacturer's instructions. Agar dilution was used for all other susceptibility testing, according to the Clinical Laboratory Standards Institute. Vancomycin population analysis profile testing was performed on all isolates to detect hVISA, as described previously (Howden et al., 2011).

Antimicrobial peptide susceptibility testing

Flow cytometry and propidium iodide-stained bacterial cells (excitation, 535nm; emission, 620nm; Sigma) were used to determine membrane permeabilization after treatment with antimicrobial peptides (hNP-1 or hBD-2, 20μg ml⁻¹) as described previously (Gao et al., 2013). Heat-killed bacteria (95°C for 15min) and non-peptide-exposed bacteria were used as positive and negative controls, respectively. Flow cytometry was performed (LSRFortessa; Becton Dickinson) in 10mM K⁺ minimal essential medium, pH7.2. Fluorescence of a minimum of 5×10³ cells was acquired for statistical analysis. The analysis was performed in biological triplicates for each peptide/strain combination and data were analysed using FACSDiva software (BD Biosciences).

Closed genome sequences

The complete genome assemblies for Sa_JKD6210 and Sa_JKD6229 were based on Ion Torrent mate-pair libraries. The libraries were prepared using a demonstration protocol (Ion Torrent p/n 4472004, release date 27 September 2011) with the 5500 SOLiD mate-pair library kit. Libraries were sequenced on an Ion Torrent Personal Genome Machine with a 316D chip and 200bp chemistry, running TorrentSuite version 3.2.1 (Life Technologies). After processing, 1629524 reads for JKD6210 and 2760456 reads for JKD6229 contained the mate-pair linker sequence with a mean insert size of 5211 and 5152bp, respectively. Primary de novo assembly was performed using Newbler (version 2.8), resulting in three scaffolds (2762210bp) and two scaffolds (2727771bp), respectively. Genome finishing was managed using Geneious Pro (version 7.2) with gaps closed by PCR, primer walking and Sanger sequencing. Final assemblies were validated by reference to high-resolution NcoI chromosome optical maps using MapSolver (version 3.2.0; OpGen). Ion Torrent sequencing errors were corrected by mapping Illumina MiSeq 250bp paired-end reads from each genome back to the finished sequences (see below). Annotation was performed using Prokka version 1.10 (Seemann, 2014). Chromosome sequences were also compared by blastn analysis and visualized using act (Artemis Comparison Tool; Sanger). Sequences were screened for bacteriophage-associated sequences using phast (Zhou et al., 2011).

Whole-genome sequencing and analysis

Whole-genome sequencing was performed on the Illumina MiSeq platform, with Nextera XT library preparation and paired-end 250bp sequencing chemistry. The raw reads were filtered using Trimmomatic (Bolger et al., 2014) to remove ambiguous base calls, Nextera adaptors, bases at Phred quality 10 and reads <50bp. Resulting FASTQ sequence read files were read-mapped to the completed JKD6210 genome using Bowtie version 2.1.0 (Langmead & Salzberg, 2012) with default parameters and consensus calling to identify SNPs (indels excluded) using Nesoni version 0.109 (https://github.com/Victorian-Bioinformatics-Consortium/nesoni). A list of the isolates sequenced and summaries of the output for each are shown in Table S1. An alignment file based on the variable nucleotide positions amongst all isolates was used to infer a whole-genome haplotype network using the median-joining algorithm as implemented in SplitsTree version 4.13.1 (Huson & Bryant, 2006).

Quantitative PCR (qPCR)

Real-time qPCR was used to assay the copy number of selected genes within the chromosomal repeats. The primers are listed in Table S2. Phire Hot Start II DNA polymerase (Thermo Scientific) was used in the qPCR and Evagreen dye (Biotium) was added into reactions in 1:20 ratio. The PCR was cycled at 98°C for 15s, 52°C for 18s and 72°C for 15s. The dye signals were sampled during the annealing steps. Standard curves were generated for each pair of primers, using 10-fold serial dilutions of S. aureus genomic DNA. The housekeeping gene gyrB was used as a reference gene to normalize the results.

Allelic exchange

By allelic exchange with pKOR1 (Bae & Schneewind, 2006), we introduced the alr1 15bp deletion from BPH1123 into JKD6229. A 2kb amplicon centred on the deletion was amplified from BPH1123 genomic DNA with primers listed in Table S2 and cloned into pKOR1. The plasmid was electroporated into JKD6229 after isolation from RN4220, with subsequent allelic exchange as described previously (Bae & Schneewind, 2006). Sanger sequencing of the region confirmed the authenticity of the mutation.

Inducible ParC expression

The full-length parC genes from JKD6210 (parC WT), SCV JKD6229 (parC V463) and BPH1116G20 (parC G296D, V463) were amplified by PCR (oligonucleotides used are listed in Table S2). The PCR products were digested with KpnI/EcoRI, ligated into plasmid pRAB11 (Helle et al., 2011) and transformed into Escherichia coli strain IM08B (Monk et al., 2015). The sequences of the parC alleles cloned into pRAB11 were confirmed by Sanger sequencing. The plasmids were directly transformed into JKD6210 with parC expression induced by addition of anhydrotetracycline. JKD6210 was also transformed with empty vector as a negative control. The resulting plasmids and strains are listed in Table S3. To assay the impact of inducible parC expression, the transformed isolates were grown on Mueller–Hinton agar containing chloramphenicol (10μg ml⁻¹) with anhydrotetracycline (5ng ml⁻¹) and incubated for 24h.

Comparison of fully assembled chromosomes reveals a large chromosomal duplication

To enable a thorough analysis of the 15 additional draft genomes sequences, we fully assembled the chromosome for the initial and the final clinical isolates, JKD6210 (single circular chromosome 2748898bp, circular plasmid 25003bp) and JKD6229 (single circular chromosome 2847260bp). We utilized a combination of Ion Torrent mate-pair and Illumina paired-end libraries to produce high-quality closed genomes. High-resolution NcoI optical maps were additionally generated for both isolates to guide and validate in silico scaffolding. Unexpectedly, the optical maps suggested an ~100kbp repeated region in the SCV JKD6229 chromosome (Fig. 2a). The in silico assembly of JKD6229 confirmed the optical mapping and demonstrated a 98231bp sequence in a tandem duplication from recA nucleotide position 1356561 to trpB position 1454925 (JKD6210 numbering). The duplication created a hybrid trpB/recA gene (Fig. 2). The impact of the duplication on phenotype is assessed in a later section.

Open in a separate window

Fig. 2.

Complete and correct assembly of JKD6210 and JKD6229. (a) Comparison of high-resolution NcoI optical maps with in silicoNcoI predictions confirming correct assembly of the chromosome for each isolate. Depicted also is the 98kb duplicated region. (b) Expanded view of the 98kb duplication showing the flanking protein-coding DNA sequence impacted by the mutation, and the location of parC and mprF within a 20kb internal segment. Shown also is 50bp of sequence either side of the trpB/recA junction.

Analysis of the two closed chromosomes also uncovered previously unrecognized mutations in SCV JKD6229, including three small codon deletions in rplC, spoU and entS, and one in-frame insertion in rocA (Fig. 3, Table S4).

Open in a separate window

Fig. 3.

Population genotype and phenotype summary of all sequenced isolates. (a) Median-joining network graph showing the genetic relationship between the 17 clinical S. aureus isolates obtained over the duration of treatment, inferred from whole-genome sequence alignments against JKD6210.Numbers on graph edges indicate SNPs. (b) Mutation summary, culminating in SCV JKD6229 as shown by purple blocks. Lower row contains a heatmap summary of changing vancomycin susceptibility. (c) Subset of the phenotype data for each isolate presented in Fig. 1 for ease of comparison with the corresponding genotype, including heatmap of mean colony size (mm) fluctuations, haemolysis and antibiogram results generated by agar dilution. Red blocks denote resistance.

Genome sequence comparisons of the intermediate isolates confirm a heterogeneous population

Fifteen intermediate isolates were selected for whole-genome sequencing. Sequence reads were mapped to the JKD6210 completed chromosome and a full list of variants was produced (Figs 3 and S1, Table S4). A haplotype network was inferred using the median-joining algorithm from alignment of the resulting variable nucleotides (indels excluded). This network demonstrated the heterogeneous structure of the S. aureus population within the patient, as it evolved from the single progenitor strain (Fig. 3a). The majority of mutations arose after the exposure to rifampicin, ciprofloxacin and vancomycin from day 14 onwards, and this was also reflected in dramatically diverse colony morphotypes. There were eight persisting mutations that were absent in JKD6210 and accumulated in one or more intermediate strains to persist through to JKD6229 (Fig. 3). The critical time period for the emergence of the persistent SCV was between days 14 and 45, when the mutations in rpoB, relA, parC and rplC first appeared in BPH1116. This was the first hVISA strain isolated during the infection (Fig. 3a). All four of these mutations were sustained in isolates recovered during the infection except for the absence of the rplC mutation in two vancomycin-sensitive S. aureus (VSSA) strains (Fig. 3a, b, Table S3). The mutations in rlmN and rocA were detected at day 57 after linezolid was first employed, with the rlmN mutation shown to inactivate RlmN methylase activity and directly associate with linezolid resistance (Gao et al., 2010; LaMarre et al., 2011). Mutations in spoU and entS appeared in isolates from the day 113 blood culture.

An analysis of the mutations that appeared in one or more of the intermediate isolates but did not persist through to JKD6229 was also undertaken (Fig. S1, Table S4). Some of these changes helped explain the varying phenotypes of the intermediate isolates. For example, BPH1118 had five additional changes that included an agrA loss-of-function mutation that explained the absence of δ-haemolysis in this isolate (Fig. 1a). A 15bp deletion in alr1 was found in the VSSA isolate BPH1123,. This gene encodes an alanine racemase. Transfer of the alr1 allele from BPH1123 into the hVISA JKD6229 increased the susceptibility of the strain to vancomycin, which fits with the sensitivity of BPH1123 to vancomycin and oxacillin (Fig. S2).

Large chromosomal duplications contribute to antibiotic and antimicrobial peptide resistance

Given the discovery of the chromosomal duplication in JKD6229, we examined changing the read-coverage density across the JKD6210 chromosome to investigate possible chromosomal duplications amongst the intervening isolates (Fig. 4). Two distinct high-coverage regions suggestive of chromosomal repeats were noted, first detected in BPH1116 at day 45 (Fig. 4a). Region 1 was variable in length, up to ~98kb and spanned the genes SA1128–SA1204 (S. aureus N315 locus_tag nomenclature). This was the same region duplicated in JKD6229 (Fig. 2). Region 1 also appeared to be duplicated in five other isolates (Fig. 4). We used the unique sequence in JKD6229 across the trpB/recA junction (Fig. 2b) to screen for the same duplication site within the sequence reads of the intermediary isolates. Only JKD6229 contained this sequence, indicating that this region was independently duplicated amongst the intermediary isolates.

Open in a separate window

Fig. 4.

Detection of chromosomal expansions in intermediate isolates. (a) Heatmap sequence coverage summary of the 15 isolates against the chromosome of JKD6210, showing a region of likely chromosomal amplification of variable length from 10 to 98kb (region 1) present in six isolates and a ~35kb region (region 2), corresponding to phiSA3. (b) Three coding sequences from region 1 and 2 were selected as qPCR targets for verifying the copy number of the amplified regions shown in (a). The copy numbers of these coding sequences were aligned with coverage summary and confirm the expected copy number changes. Isolates that demonstrated reduced vancomycin susceptibility are highlighted in red, whilst green indicates vancomycin susceptibility. All isolates containing chromosomal expansions had reduced vancomycin susceptibility. P1–P3, region 1; P4–P6, region 2 (phiSA3).

Region 2 was ~35kbp and spanned the staphylococcal bacteriophage phiSA3. Higher read-coverage across this region was suggestive of mixed cell populations with integrated and excised circular phage forms, as the strains BPH1117 and BPH1136 were still producing α-haemolysin, which is inhibited if β-haemolysin is made (Fig. 1a). The β-haemolysin gene is disrupted on phiSA3 integration. We therefore conducted an in silico screen of the reads from all isolates for the presence of the sequence across the junction of the circular form of phiSA3. Reads from S. aureus isolate BPH1116 onwards indicated the presence of excised, circular forms of phiSA3. This result explains the higher read-coverage across region 2. We further confirmed the read-coverage results by qPCR targeting three protein-coding DNA sequences within each of region 1 and region 2 (containing phiSA3) (Fig. 4a). Read-coverage and mapping locations across region 1 indicated there were duplications of variable length (20–98kb) and copy number (Fig. 4a).

Intriguingly, the read-coverage and confirmatory qPCR showed that only isolates with reduced vancomycin susceptibility contained chromosome expansions, suggesting a possible contribution of this DNA amplification to antibiotic resistance. Isolate BPH1122 contained the shortest example of the region 1 chromosome duplication, covering 11 coding sequences from SA1183 to SA1193 (Fig. 4). These coding sequences included genes encoding proteins involved in DNA replication, in particular chromosome segregation (parC, parE), and mprF, encoding multiple peptide resistance factor (MprF). In S. aureus, mutations leading to increased activity of MprF are associated with enhanced resistance to host defence peptides and reduced susceptibility to daptomycin (Bayer et al., 2014). Only one mutation was detected across this region compared with JKD6210 and this was the V463 codon insertion that was present in all copies of parC, first observed in the day 45 isolate BPH1116 (Figs 3a and 4). Mutations in parC are associated with fluoroquinolone resistance, but usually occur at or around position 80 (Schmitz et al., 1998).

To investigate the functional impact of the region 1 chromosomal duplication in S. aureus, isolate BPH1116 (3×60kb tandem repeated sections) was serially subcultured for 20days, generating BPH1116G20 which was shown by qPCR and genome sequencing to have lost the chromosomal expansion (Fig. S3). When BPH1116 (day 45) was compared with JKD6210 (day 0), it exhibited a reduced colony size and susceptibility to vancomycin (Fig. 1a), with 3×60kb copies of the region 1 chromosomal duplication present (Fig. 4). BPH1116G20 was tested for the above phenotypes and additionally screened for the resistance to host defence peptides, due to the presence of mprF in the amplified region. BPH1116G20 displayed increased colony size, susceptibility to vancomycin and host defence peptides compared with BPH1116 (Fig. 5).

Open in a separate window

Fig. 5.

Impact of chromosomal expansion on key phenotypes. (a) Comparisons of colony morphology, showing partial restoration of colony size after loss of the chromosomal amplification spanning parC and mprF in isolate BPH1116G20.(b) Increased susceptibility to vancomycin upon loss of the chromosomal amplification as measured by population analysis profile. (c) Increased sensitivity to antimicrobial peptide killing upon loss of the chromosomal amplification as indicated by enhanced propidium iodide staining, as measured by flow cytometry. The y-axis shows the increase in propidium iodide-mediated fluorescence for antimicrobial peptide-exposed versus non-exposed S. aureus. Results indicate means and 95% confidence intervals for biological triplicates. **Significantly different, P<0.001, unpaired t-test.

In addition to the loss of the chromosomal duplication, genome sequencing of BPH1116G20 identified two additional non-synonymous point mutations, one each in the genes parC and abgT (AbgT^G326R). The gene abgT encodes a putative p-aminobenzoyl-glutamate transport protein, which may be involved in uptake of folate degradation products (Green et al., 2010). The additional mutation in parC that yielded ParC^G296D/V463 (compared with JKD6210) was particularly interesting given the essential nature of parC and that three copies of ParC^V463 were present in the chromosomal duplication within BPH1116 (Fig. 4). Despite numerous attempts, we were unable to introduce the V463 allele into the WT parC of JKD6210.

This research was supported by the National Health and Medical Research Council (NHMRC) of Australia (1008973). B.P.H. and T.P.S. are recipients of NHMRC fellowships (1023526 and 1008549, respectively). S.G. and T.S. were supported in part by the VLSCI Life Sciences Computation Centre, a collaboration between Melbourne, Monash and La Trobe Universities, and an initiative of the Victorian Government, Australia.