Isolate characterisation

In total, 497 isolates were obtained from 346 patients. Of these, 21 were identified as non-pneumococci by rMLST (S. mitis, S. parasanguinis, S. pseudopneumoniae and S. australis). Furthermore, after MIC determination, serotyping and core genome alignment, 86 isolates were found to be phenotypic and genotypically identical to isolates obtained from the same patient, leaving 390 unique pneumococcal isolates obtained from 335 patients. The isolate distribution across treatment arms and time-points was similar (Fig. 1). It had been previously reported that there were no differences in susceptibility levels between trial arms and between D1 and D29 [8], and serotype distribution across arms was homogenous (Table S3, Supplementary information), thus results described below refer to the entire isolate population.

Fig. 1

Flow chart depicting isolate collection and filtering process

Overall, 33 different serotypes were detected (Table 1). The most prevalent were all non-PCV13 serotypes, 15B/C (12.3%, 48/390), 11 A (10.3%, 40/390), 15 A (7.7%, 30/390) and 23B1 (6.9%, 27/390) (Table 1). PCV13 serotypes 3 (3.3%, 13/390), 19A (1.8%, 7/390) and 19F (1.3%, 5/390) were detected at a low prevalence, and only one serotype 3 isolate was found in a non-vaccinated patient (Table S1). Only 3.3% (13/390) of the isolates were labelled as non-typeable due to the low (< 30%) coverage of the capsular operon (Table 1), and upon examination of the cps locus, nine were classified as Null Capsule Clade (NCC) 2a and the other 4 as NCC2b (Table S1) [28]. No regional clustering of serotypes was found (data not shown).

Table 1 Serotype distribution and penicillin and amoxicillin susceptibility among study isolates (n = 390). VTs are marked in bold. Shannon diversity index calculated for the STs found in each serotype

Overall, 103 different STs were detected, of which six were novel. Most STs clustered within one serotype (Table S1), however, 5 STs (ST156, ST162, ST177, ST193, ST199) were distributed in 2 or more different serotypes (Fig. 2), indicative of potential capsular switching events. Serotype 15B/C isolates were remarkable in their ST heterogeneity, as nine genomically divergent STs were associated to this serotype (Fig. 2; Table 1, Table S1). A total of 59 local clusters were identified using PopPUNK, and the isolates were classified into 45 GPSCs when using PopPUNK with the GPSC database, indicating some variability within the GPSC clusters. This is a more robust clustering method and related STs can be engulfed into one GPSC, hence the lower number of clusters compared to STs.

Fig. 2

Phylogenetic tree generated from the 390 study isolates’ assemblies. Inner squares represent the presence of amino acid changes in (from inner to outer) penicillin-binding protein (PBP) 1a, PBP2b and PBP2x protein sequences described in the Comprehensive Antibiotic Resistance Database (CARD) to provide non-susceptibility to penicillin. Filled squares represent the presence of all described mutations for PBP1a and PBP2b, and > 4 of the 7 mutations described for PBP2x in CARD. Empty squares depict < 4 PBP2x mutations. Penicillin and amoxicillin susceptibilities are colour-coded (green for susceptible, orange for non-susceptible and red for resistant). Acquired resistance genes are depicted as coloured squares when present in the isolate, potentially conferring resistance to macrolides (brown, blue and purple squares), tetracycline (grey square) and aminoglycosides (light blue square). Outer circles represent serotype of the isolate and Global Pneumococcal Sequence Clusters (only GPSCs presenting > 3 isolates are shown). Finally, pink blocks indicate isolates belonging to the same ST and presenting with different serotypes. Serotype 15B/C was present across 8 STs and is highlighted in blue

In total, 15.6% (61/390) and 2.6% (10/390) of all isolates were non-susceptible to penicillin and to amoxicillin, respectively. None of the isolates showed resistance to penicillin and 5 were amoxicillin resistant. Penicillin non-susceptible isolates (n = 61) were distributed in 13 different serotypes. In four serotypes (11 A, 15 A, 19F, 35B), six isolates (6.3%, 6/96) were penicillin and amoxicillin non-susceptible (Table 1). Additionally, non-typeable isolates presented a high prevalence of penicillin (7/13) and amoxicillin (4/13) non-susceptibility (Table 1). All genotype 23B1 (n = 27) and serotype 15 F (n = 1) isolates were found to be non-susceptible to penicillin, and susceptible to amoxicillin (Table 1). Seventeen GPSCs showed penicillin non-susceptibility, and of these six (6, 44, 60, 9, 81, 59) also showed amoxicillin non-susceptible isolates (Fig. 3). Remarkably, all isolates in GPSCs 5 (genotype 23B1) and 9 (serotype 15 A) were penicillin non-susceptible.

Fig. 3

Local popPUNK clustering. Node filling and node border represent penicillin and amoxicillin susceptibility, respectively. Node labels refer to Global Pneumococcal Sequence Clusters

From the 39 patients presenting different serotypes at D1 and D29, only in four instances a penicillin susceptible isolate was replaced by a penicillin non-susceptible one at D29 after antibiotic treatment, and, remarkably, in all these instances the replacing serotype was 23B1 (Fig. 4). In three other instances, a penicillin non-susceptible isolate was replaced by a susceptible one, while in the remaining 32 cases, both unique isolates at D1 and D29 were susceptible to penicillin (Fig. 4). None of the replacing isolates was resistant to amoxicillin.

Fig. 4

Flow chart depicting isolate distribution within patients across samples, and table showing in detail the serotypes and penicillin non-susceptibility observed in patients where two different serotypes were found at D1 and D29. Green denotes penicillin susceptibility and orange penicillin non-susceptibility

In total, 73 acquired resistance genes were detected in 38 isolates (9.7%, 38/390), conferring resistance primarily to macrolides and tetracycline, although susceptibility testing was not performed for these antibiotics (Table S1). The most prevalent genes were tet(M) (n = 28) and erm(B) (n = 25), and were mostly co-integrated in the chromosome (5.9%, 23/390). In 21 isolates, these genes were flanked by a Tn916 family transposon (Table S1). Similarly, mef(A) and msr(D) were present together in 7 isolates. Only 4/390 (1%) of the isolates carried an aminoglycoside resistance gene (aph (3’)-III), which was always found to be integrated along with erm(B) (Fig. 2).

Geno-pheno-type correlation

From the generated cgMLST scheme, sequences from PBP amino acid variants were extracted and analysed from the study population, totalling 65 PBP1a, 70 PBP2b and 78 PBP2x variants, respectively. These alleles presented an average of 19 (2-108/720, 97.3% identity), 7 (0–40/681, 99.0% identity) and 19 (0–78/751, 97.5% identity) amino acid modifications compared to the reference strain R6, respectively (Figs. 5 and S2).

Fig. 5

Phylogenetic tree generated from penicillin-binding protein 2b amino acid sequences. Coloured squares represent the presence of mutations previously described to confer an increase in MIC. Number of total amino acid modifications, penicillin and amoxicillin susceptibility, serotype, and number of isolates are also depicted. Variants highlighted in blue present a low number of overall amino acid modification while containing the three key modifications described in the CARD database to reduce susceptibility to beta-lactams

The presence of specific PBP mutations, as described in the CARD database (derived from Stanhope et al.) [29], explained penicillin non-susceptibility, especially in the case of PBP2b (Fig. 5). Upon comparison to non-S. pneumoniae isolates found in our collection, the PBP variants conferring beta-lactam non-susceptibility were found to be genetically similar to those in S. mitis and S. pseudopneumoniae (data not shown).

All identified PBP variants were subjected to protein-based phylogenetic analysis, which showed clustering of mutated variants and a correlation between key mutations and overall number of changes in comparison to the reference sequence, that is, the variants presenting the key mutations described in the CARD database contained a higher number of total amino acid modifications compared to the reference, indicating that these variants arose from recombination (i.e., mosaic genes), and the total number of mutations found in the three PBPs correlated with increasingly higher MICs to both penicillin and amoxicillin (Supplementary information, Results).

All identified PBP2b variants clustered into three clades of which the penicillin non-susceptible isolates clustered exclusively in two clades (Fig. 5). Of note, 3 PBP2b variants (n = 8) were unique in that they harboured 3 predicted amino acid changes that are linked to beta-lactam non-susceptibility (T445A, E475G, T488A), however, the rest of the gene was highly conserved (6, 7 and 9 amino acid modifications, i.e. >98.5% identity), indicating that these variants were the result of de novo mutations instead of recombination events and mosaicism as observed for other PBPs.

We found 64 different combinations of PBP variants (PBP types) in our collection, of which 38 had already been described and associated to MIC values that match the ones we obtained [15]. Twenty-four novel variants were observed, of which 18 were found in penicillin non-susceptible isolates. In general, PBP type was consistent within STs and variation within ST 156, 162, 177 and 199 was explained by serotype (Table S1).

Genomic analysis of persistent vaccine typesLow genomic heterogeneity in serotype 3 and predominance of clade Ia

Genome stability varied greatly between serotypes, indicated by the within-serotype ST diversity (Table 1). All serotype 3 isolates belonged to ST180, and genomic differences were only observed between clades, suggesting that a serotype 3 clade is a temporally stable unit (Fig. 6A). In our study collection, 10/13 (76.9%) of serotype 3 isolates clustered in clade Ia and only three isolates belonged to clade II, which contained 14 non-study UK isolates from 2010 onwards (Fig. 6A). Additionally, no genomic changes were observed surrounding the cps locus (Figure S3A).

Fig. 6

Phylogenetic trees generated from whole genome alignments of serotype 3 (A), serotype 19A (B) and serotype 19F (C) isolates derived from this study (orange shading) and older isolates from the same serotypes. References used were CAPIT119_D1-1, CAPIT226_D1-1 and CAPIT214_D1-1 (marked in red shading), respectively. In red, are shown the large genomic changes predicted to have occurred earlier in evolution, thus being present in a cluster of isolates. In blue, genomic changes predicted only in the branch leaves, that is, in only one isolate. The position of the cps locus in the core genome is indicated with a black rectangle. For serotype 3, the clade to which the isolates belong is also depicted. All serotype 3 isolates belonged to sequence type 180

High genomic heterogeneity in serotypes 19A and 19F

Serotypes 19A and 19F presented a higher genomic complexity, even within the same ST. Serotype 19 A isolates belonging to ST2062 formed a separate clade when compared to the rest of serotype 19A isolates, which were clustered in three sub-clades (Fig. 6B). The biggest sub-clade was composed of isolates belonging to ST199 and ST450, while the two smaller sub-clades were composed of ST667 and ST199 isolates (Fig. 6B). Genomic differences surrounding the cps locus in ST199 and ST450 were more pronounced than in ST667, wherein the region upstream of the cps locus was less divergent compared to the other STs, although the region between pbp1A and dexB was highly variable in all three STs (Figure S3B).

Finally, serotype 19F isolates presented higher within-ST genome variability (Fig. 6C). Isolates belonging to ST162, ST420 and ST422 were not detected after 2009, except for one ST162 isolate in 2015. Overall, 19F isolated from 2006 onwards were observed to have undergone more genomic changes (Figure S3C).

Within ST serotype variability might facilitate vaccine escape

In order to observe possible capsular switch events leading to vaccine escape, isolates belonging to ST162, ST177 and ST199 were also studied, as these STs were detected in our collection to be associated with several different serotypes. In the case of ST162, isolates presenting serotype 9V were not detected after PCV7 implementation, and were very divergent from the rest of isolates (Fig. 7A). However, isolates presenting serotype 19F could still be detected after PCV7, but in a lower proportion, and only one isolate was observed after PCV13 implementation (Fig. 7A). Isolates presenting serogroups 15 and 24 only started to be detected after PCV13 implementation, and they were more related to serotype 19F isolates than to 9V, but presenting lower genomic divergence, especially around the cps locus (Figure S4A).

Fig. 7

Phylogenetic trees generated from whole genome alignments of sequence type (ST) 162 (A), ST177 (B) and ST199 (C) isolates derived from this study (orange shading) and older isolates from the same serotypes. References used were CAPIT086_D1-1, CAPIT292_D1-1 and CAPIT226_D1-1 (marked in red shading), respectively. In red are shown the large genomic changes predicted to have occurred earlier in evolution, thus being present in a cluster of isolates. In blue, genomic changes predicted only in the branch leaves, that is, in only one isolate. The position of the cps locus in the core genome is indicated with a black rectangle

ST177 was less represented in the pubMLST database, although 3.7% (11/390) belonged to this ST in our collection. We show here a serotype divergence in ST177: two clades were clearly differentiated, one including serotype 19F isolates pre-PCV7, and one including isolates presenting different serotypes and only detected after PCV13 introduction (Fig. 7B). Remarkably, three ST177 isolates from our collection that were genomically very similar presented with serotypes 21, 23B and 23B1 (Fig. 7B). These showed slight divergence from the closely-related serogroup 24 isolates, which might indicate a higher potential for these isolates to switch capsule, as can be observed by the genomic variability surrounding the cps locus (Figure S4B). ST177 presenting serotype 7C were only detected after 2015 and presented a similar additional genomic variability compared to the genome structure of serotype 19F isolates before PCV13 implementation (Fig. 7B).

Finally, isolates belonging to ST199 only presented serotypes 15B/C or 19A. In general, isolates clustered according to their serotype (Fig. 7C). Remarkably, one serotype 19A isolate from 2010 clustered together with serotype 15B/C isolates, and the biggest observed divergence was only surrounding the cps locus, indicating a potential capsule switch event to 19A before the PCV13 implementation (Figure S4C).