Pneumococcal study isolates (Table 1) were assigned to CCs based on multilocus sequence type (MLST) data. Isolates within CCs are considered to have shared ancestry, while those in different CCs are more distantly related. Within our collection, identical or highly similar, 'nonsusceptible' pbp alleles and/or sequence regions were identified repeatedly among pneumococci belonging to different CCs. These pbp sequences were those associated with the PMEN1 reference strain (GenBank accession: NC_011900.1, hereafter simply called 'PMEN1').

Two isolates, CGSP14 (GenBank accession NC_010582.1) and the PMEN3 reference strain (GenBank accession ABGE00000000.1, hereafter called 'PMEN3'), representing independent CCs (Table 2), shared identical pbp2x, pbp1a and pbp2b alleles with PMEN1. Analysis of the genomic regions flanking these pbp genes showed that, in fact, much larger regions of the CGSP14 and PMEN3 genomes (5,362 to 22,104 bp; Figure 1) were identical or highly similar to that of PMEN1, suggesting these regions, which conferred penicillin nonsusceptibility, were acquired from a PMEN1 representative(s) in both cases. The PMEN2 reference strain, representing another CC (Table 2), also shared identical regions (225 to 1,034 bp in length) of all three of its pbp alleles with PMEN1 (the intervening regions of sequence were different). A total of 12 distinct CCs were represented by isolates sharing identical pbp sequences in whole or part (119 to 1,376 bp) in one or two of their pbp alleles to PMEN1, suggesting that they too were acquired from a PMEN1 representative. Nine additional CCs were represented by isolates sharing pbp2x and/or pbp2b regions (199 to 1,436 bp) that differed from those of PMEN1 by only one or two nucleotide substitutions, again suggesting possible acquisition from a PMEN1 representative (see Additional file 1 for pbp allele and CC information). No other pbp alleles were as widely distributed as those of PMEN1.

Given that the CGSP14 and PMEN3 genomes each contained three large, independent regions that were putatively acquired from PMEN1, we used whole genome sequence data generated on the Illumina platform to assess whether other regions of the PMEN1 genome were also shared by CGSP14 and PMEN3 (both of which represent multidrug-resistant clones, the latter of which is also globally distributed). We found that, in fact, 15 regions (1,663 to 32,312 bp, totalling 211,963 bp, approximately 9.5% of the genome) of the CGSP14 chromosome were identical to those of PMEN1 or differed by only 1 to 2 nucleotides across a minimum of 7,537 bp (Figure 2; Additional file 2), and were not identical in any of four CGSP14 ancestral representative genomes. The ancestral representatives (14/5, the PMEN9 reference strain, ICE13 and ICE570; Table 2) were chosen from our collection because they represented older and/or penicillin-susceptible members of the CGSP14 CC and so likely resembled the true ancestor of the penicillin-resistant CGSP14. Importantly, this allowed us to identify and exclude identical sequence that was shared by descent, whereas regions of the CGSP14 genome that differed from its ancestral representatives but were identical to those of PMEN1 were most likely acquired from PMEN1 through the process of recombination. The ancestral isolates were generally all identical or highly similar to each other at the genomic regions in question; sequences from these isolates were also highly similar to CGSP14 in the regions flanking the putative recombination imports, confirming their suitability as ancestral representatives.

Following a similar comparison of PMEN3 and its ancestral representatives (9A/1, 9V/5 and 9V/6; Table 2), we identified nine PMEN3 genomic regions (4,662 to 22,104 bp, totalling 117,689 bp, approximately 5.3% of the genome) that were identical to sequences from PMEN1 or differed by only two nucleotides across 22,104 bp, and were different from those of the ancestral representatives (Figure 2; Additional file 2). We suggest that these regions were also acquired by PMEN3 from PMEN1 through the process of recombination.

The CGSP14 and PMEN3 putative recombinogenic regions contained 213 and 108 whole or part coding sequences, respectively (Additional file 2). In addition to the penicillin resistance-conferring pbp2x, pbp1a and pbp2b genes, the following coding sequences associated with enhanced virulence or colonization potential were included: the CGSP14 bgaA, clpX, SPN23F_05810 and SPN23F_20890 genes associated with enhanced host colonization ability [24]; the CGSP14 virulence-associated cbpJ, msmK, chloramphenicol acetyltransferase and zeta toxin/epsilon antitoxin genes [25-27] (the latter two of which are situated on the CGSP14 conjugative transposon); and the PMEN3 virulence-associated cbpD, cbpJ and clpC genes [26].

The possibility that the putative imported regions were acquired by CGSP14 and PMEN3 from non-PMEN1 representative pneumococci cannot be ignored. However, the majority, 13 and 6, respectively, of regions described here were not identical among any of 96 additional pneumococcal genomes, representing a diverse range of 31 independent CCs (listed in Additional file 3, excluding 23F/4 and CC66 isolates). Two and three of the regions, respectively (Figure 2), were identical or highly similar among isolates representing CC66, and so it is possible that those regions may have been acquired from a CC66 representative (also see Additional file 4). Although our collection cannot capture the full complement of pneumococcal genetic diversity, given these results and the extremely high similarity of the putative import regions to the PMEN1 genome, we believe that they originated from a PMEN1 representative. Thus, these data suggest that DNA originating from the PMEN1 clone has greatly influenced the genomic evolution of at least two unrelated pneumococcal clones.

The Genome Comparator tool [28] was used to screen 104 pneumococcal genomes, representing a very diverse range of CCs and serotypes (Additional file 3), for regions of DNA (aside from the pbps) that were identical to PMEN1. We identified 1,990 coding sequences from the PMEN1 GenBank Accession NC_011900.1 file (totaling 1,833,672 bp, approximately 83% of the genome). Comparable coding sequences in the query genomes were identified by BLASTn search if they shared ≥ 70% nucleotide sequence identity and could be aligned at 100% length. Consequently, coding sequences spanning the ends of assembly contigs could not be identified.

Surprisingly and uniquely, the analyses showed that 23F/4, the first PNSP isolated from Australia (Table 2), was very likely to be an ancestral representative of PMEN1, since approximately 95% of the identified coding sequences were ≥ 98% similar to those of PMEN1. Approximately 75% of identified coding sequences in 23F/4 were identical to PMEN1 and over one-third of the coding sequences that were not identical differed by only one or two nucleotide substitutions. The pbp2x, pbp1a, and pbp2b genes and three of the seven MLST loci were not identical. Additionally, a comparison of the PMEN1 and 23F/4 genomes, performed using the Artemis Comparison Tool [29] using all available sequence (both coding and non-coding sequence; Figure 3), clearly showed three large regions of the PMEN1 genome that were partially or completely missing from the 23F/4 genome and were presumably acquired by a more recent ancestor of PMEN1. These were the ICESp23FST81 element, the ɸMMI phage and a Na⁺-dependent ATPase island, all associated with increased colonization and virulence potential [30-33].

The Genome Comparator analyses also revealed that the majority (87, 83.6%) of genomes analyzed - representing 31 unique CCs and 30 serotypes - shared 15.1% ± 2 standard deviations (excluding 12 outliers with > 30% identical coding sequences) of the identified coding sequences identically with PMEN1 (Figure 4). We suggest that these 'baseline' coding sequences were shared through ancestral descent, that is, they were identical in the most recent common ancestor of these clones and have not undergone subsequent change. (For purposes of comparison, the same analysis was performed using pneumococcal genome TIGR4 (GenBank accession AE005672.3) as the reference and a similar baseline mean level of identical sequence of 14.5% was observed (Additional file 5).

Isolates belonging to CC66 shared between 44.5% and 58.4% coding sequences identically with PMEN1. These isolates included PSP and were dated from 1952, preceding the estimated origin of PMEN1. Since the proportion of identical coding sequences between CC66 and PMEN1 was much greater than between those of PMEN1 and the other CCs represented here, we conclude that PMEN1 was ancestrally more closely related to CC66 than to other CCs, which supports the findings of Donati and colleagues [34] (also see Additional file 4).

A unique collection of 426 global PSP and PNSP dated 1937 to 2007 was studied, as summarized in Table 1. Isolates included one of the first PNSP reported from Australia [6], the first high-level PNSP reported from South Africa [8,9] and the 43 PMEN reference strains [15]. The Statens Serum Institut, Denmark provided 211 pneumococci dated 1937 to 1996. These isolates included both PSP (n = 203), PNSP (n = 7) and an isolate of unknown penicillin susceptibility (n = 1). Following the analyses of MLST and pbp sequence data from these isolates, the 43 PMEN clones (including both PSP (n = 18) and PNSP (n = 25)) and additional modern (1990s to 2000s) PSP (n = 38) and PNSP (n = 9) that had been recently genotyped by A Brueggemann/L McGee and represented CCs of interest, were selected for inclusion. In addition, a literature search was performed to identify studies that reported early PNSP. The authors of those papers were contacted and isolates were requested; the majority of isolates were still available and were added to our collection. The available isolates comprised PNSP dated 1969 and 1972 from Papua New Guinea (n = 2), PSP (n = 8) and PNSP (n = 71) dated 1977 to 1989 from South Africa, PSP (n = 1) and PNSP (n = 26) dated 1981 to 1991 from the USA, PNSP dated 1987 to 1989 (n = 9) from Spain and PNSP dated 1986 to 1994 from Germany (n = 8). Finally, genomic sequence data for 16 isolates were retrieved from the GenBank database. These isolates included both PSP (n = 5), PNSP (n = 4) and seven pneumococci of unknown penicillin susceptibility. Serotype and penicillin susceptibility data are summarized by decade in Additional file 6. Where not previously published, isolates were serotyped by Quellung reaction and penicillin minimum inhibitory concentrations were obtained by Etest (bioMérieux Basingstoke, United Kingdom) as per the manufacturer's guidelines.

Single pneumococcal colonies were cultured on Columbia agar with sheep blood (Oxoid Basingstoke, United Kingdom) and incubated overnight at 37°C + 5% CO₂. Three to four sweeps of growth were subcultured onto tryptic soy agar (Oxoid) with 1,000 U ml^-1 catalase (Oxoid) and spread with a swab before further overnight incubation at 37°C and 5% CO₂ (at Oxford); or cultured in serum broth for 7 h (at the Statens Serum Institut). Pneumococcal growth was suspended in 1 ml phosphate buffered saline (pH 7.4; Fisher Bioreagents Loughborough, United Kingdom) and centrifuged at 7,600 rpm for 10 minutes. Extractions were completed using the DNeasy kit (Qiagen, West Sussex, United Kingdom) as per the manufacturer's instructions.

Where not previously published (n = 352) multilocus sequence types were determined for all isolates [49]. MLST data for all of our isolates and those deposited in the pneumococcal MLST database [50] were analyzed by the goeBURST [51] method to predict CC group and sub-group founders. Isolates were assigned to CCs so that only single/double locus variants of the group founder and any additional single locus variants of large (n ≥ 5 STs) sub-group founders were included. CCs were named according to the predicted group founder(s) ST(s). When goeBURST could not distinguish a group founder the group was designated NoneX, where X was the ST of lowest numerical value in the group. When isolates could not be assigned to a CC due to a lack of closely related STs, they were designated SingletonX, where X was the isolate ST.

pbp2x (1,673 bp), pbp1a (901 bp) and pbp2b (1,272 bp) nucleotide sequences for 389 selected isolates were retrieved from GenBank (n = 36 sequences) or determined by conventional PCR and Sanger sequencing (n = 1,131 sequences). We chose 174 of 211 isolates from the Statens Serum Institut for pbp nucleotide sequencing because they represented a wide range of serotypes, STs and isolation dates. All isolates from each of the other sources were also included. PCR reactions containing 1 μl DNA extract, 13.5 μl Hotstar Taq DNA Polymerase master mix (Qiagen, West Sussex, United Kingdom), 0.5 μl each of the forward and reverse primers (100 μM) (Additional file 7) were prepared to a final volume of 25 μl. Cycling parameters were as follows: initial denaturation of 95°C for 180 s, followed by 35 (40 for pbp2x) cycles of 95°C for 30 s, 55°C (58°C for pbp1a) for 30 s and 72°C for 60 s followed by a final extension period of 72°C for 10 minutes. Products were precipitated by 20% polyethylene glycolate/2.5 M NaCl, washed in 70% ethanol and rehydrated in 10 to 20 μl nuclease-free H₂O (Sigma Gillingham, United Kingdom). Sequencing reactions containing 2 μl rehydrated PCR product, 1.75 μl 5x sequencing buffer, 4 μl primer (1 μM) (Additional file 7) and 0.5 μl BigDye^® (Applied Biosystems, Paisley, United Kingdom) were prepared in a final volume of 10 μl. Cycling parameters were as follows: 25 cycles of 96°C for 10 s, 50°C for 5 s and 60°C for 2 minutes. Sequencing products were precipitated with 5 M sodium acetate (Sigma) solution, washed with 70% ethanol and subsequently separated and detected with an ABI Prism 3730 automated DNA sequencer (Applied Biosystems, Paisley, United Kingdom). Sequences were aligned by MUSCLE [52] and imported to MEGA5 [53] for visual comparison. Unique pbp sequences were assigned unique allele numbers.

We selected 96 isolates for whole genome sequencing (Additional file 3) using the following criteria. First, isolates that were not closely related to PMEN1 based on MLST data but shared identical pbp alleles or partial pbp alleles to those of the PMEN1-reference strain. The selected isolates were distributed through time, by geographic location and by genotype in order to capture the greatest possible amount of genetic diversity. Second, isolates representing CCs of interest (for example, major international CCs) for which both historical and modern isolate representatives were available. Inclusion of the historical representatives was crucial, as it allowed us to differentiate identical/highly similar sequence imports obtained exogenously versus identical sequence due to common ancestry. Within these CCs the selected isolates were distributed through time, by geographic location, by serotype and by pbp allele combination, in order to capture the greatest possible amount of genetic diversity.

DNA was extracted to a final concentration of 20 ng/μl as described above. Sequencing was via the Illumina platform as previously described [17]. As per standard protocols, multiplex library construction with a 200 bp insertion size was followed by 54 nucleotide paired-end sequencing. Illumina reads were assembled to contigs using the Velvet software [54]. Data for seven genomes were rejected based on technical problems. K-mer lengths were optimized to the maximum length achieving an average of ≥ 20× read coverage across assemblies. Resulting contigs were between 41 bp and 328,742 bp length (mean = 3,797 bp; see Additional file 4 for further details). Contigs were deposited in a BIGSdb database [28] and in the European Nucleotide Archive (see Additional file 8 for accession numbers). Contigs were ordered to the PMEN1 reference genome using ABACAS [55] and viewed with the Artemis Comparison Tool [29], allowing for the extraction of selected sequence regions spanning > 1 contig.

The BIGSdb Genome Comparator tool, which utilizes the BLAST algorithm, was used to compare PMEN1 coding sequences to those of CGSP14, PMEN3 and their respective predicted ancestral representatives (Table 2). Comparable coding sequences in the query genomes were identified by BLASTn search if they shared ≥ 70% nucleotide sequence identity (to allow for the identification of variable sequences among conserved protein classes [28] and to avoid bias towards sequence variants specific to PMEN1) and could be aligned at 100% length (to ensure that any coding sequences identified as identical to PMEN1 were indeed identical across their entire length). Potential recombination regions were indicated by coding sequences shared identically by PMEN1 and CGSP14 or PMEN3, but not by any of the respective ancestral representatives. Sequence regions spanning the coding sequences of interest plus 5 kb flanking either side were extracted from the PMEN1, CGSP14, PMEN3 and ancestral representative genomes, aligned and imported into MEGA5 [53] to produce variable site maps for visual inspection (Additional file 9). Recombination regions were classified as the maximum length of sequence over which CGSP14 or PMEN3 were identical to PMEN1 but different from their ancestral representatives.

Genome comparator was used to screen 104 isolates (Additional file 3) representing 34 CCs and 31 serotypes for coding sequences identical to those of the PMEN1 reference (parameters as above). 23F/4 was identified as sharing a high percentage of coding sequences identically with PMEN1 and was thus subjected to further analysis. PMEN1 coding sequences that differed in 23F/4 were extracted from the PMEN1 whole-genome sequence and BLASTed against the 23F/4 Velvet contigs in the BIGSdb. Sequences were aligned and percentage nucleotide difference was calculated. 23F/4 contigs were ordered against the PMEN1 chromosome by ABACAS [55] and viewed with the Artemis Comparison Tool [29] to identify genomic regions present in PMEN1 but absent from 23F/4.