National invasive pneumococcal disease surveillanceAmong the 54,199 IPD episodes reported during the 16-year study period (2005–2020), age was captured for 96% (51,917/54,199) of which 33% (17,121/51,917) were among children <18 years. Of these, 12,283 (72%) cultures were submitted to the National Institute for Communicable Diseases (NICD); 96% (11,833/12,283) had viable isolates available for sequencing. A total of 3141 (27%) viable isolates were sequenced.Among the 3141 genomes, we determined concordance between the routine phenotypic serotyping results and in silico serotypes. Isolates reacting in the pool G antisera serogroup (n = 32) that could not be resolved to the serotype level phenotypically were distinguished in silico to serotype 29 (n = 1), 34 (n = 4), 35A (n = 1), 35B (n = 19), 35D (n = 3), and 35F (n = 4). Serogroup 16, 22, and 25 (n = 5) were resolved to serotype 16F (n = 3), 22F (n = 1), and 25F (n = 1), respectively. Among 64 of 3141 (2%) isolates that had discordant serotyping results, 27 isolates were viable and upon repeat testing of both phenotypic and in silico serotyping, all were concordant. The remaining 37 that could not be repeated due to loss of viability were excluded. Concordance of 99% (3071/3104) was achieved between phenotypic and in silico serotypes; 1% (33/3104) was concordant to the serogroup level. Finally, a total of 3104 isolates were included in the current study. Isolate characteristics are shown in Supplementary Table 1.Serotype distributionAmong the 3104 genomes, we identified 57 serotypes and one unencapsulated isolate (Supplementary Data 1). Overall, the ten most predominant serotypes in our selection, in order of prevalence, were serotype 19F (244/3104, 8%), 6A (241/3104, 8%), 8 (231/3104, 7%), 14 (231/3104, 7%), 1 (228/3104, 7%), 23F (224/3104, 7%), 19A (217/3104, 7%), 6B (181/3104, 6%), 15B/C (127/3104, 4%), and 12F (117/3104, 4%; Supplementary Data 1). The serotype distribution among sequenced isolates was consistent with the serotype distribution among all 11,789 IPD cases with known phenotypic serotype data from our IPD surveillance during the same period (Supplementary Figs. 1 and 2).Before the introduction of PCVs in South Africa, IPD in children was largely caused by PCV7 serotypes 14 (122/934, 13%), 23F (112/934, 12%), 6B (95/934, 10%), and PCV13 unique serotypes 6A (108/934, 12%) and 1 (105/934, 11%). Similar serotypes predominated after the introduction of PCV7. Changes in serotype predominance were observed during the early-PCV13 period where PCV13 serotypes 19A (81/891, 9%), 1 (67/891, 8%), 23F (58/891, 7%), 6A (57/891, 6%), and non-vaccine serotype 8 (58/891, 7%) became the five most common serotypes causing disease. In the late-PCV13 period, disease-causing serotypes were largely non-vaccine serotypes, including serotype 8 (154/855, 18%), 12F (50/855, 6%), 15B/C (46/855, 5%), 16F (40/855, 5%), and PCV7 serotype 19F (66/855, 8%). Similar serotypes circulated among the <5 and 5–17-year-olds before and after PCV introduction (Supplementary Fig. 3). Serotype 8 was the leading cause of IPD in children <5 years in the late-PCV13 period accounting for 20% (146 of 739) of invasive disease whilst serotype 19F was the most common cause of IPD in the 5–17-year-olds with a frequency of 14% (14 of 102) in the same period. Overall, we observed a strong association (Cramer’s V = 0.5126) between PCV periods (pre-PCV, PCV7, early-PCV13, and late-PCV13) and the frequency of observing vaccine and non-vaccine serotypes (Supplementary Table 2). Vaccine and non-vaccine serotype frequencies in this collection reflected similar trends in our surveillance database across all four PCV periods (Supplementary Table 3).Genomic population structure and lineage-specific incidence changesOf the 3104 isolates in our study population, 3098 (99.8) were classified into 152 existing GPSC lineages and 3032 (98%) into 629 STs. We identified six new GPSCs all of which occurred in the late-PCV13 period. Among the 152 GPSCs, 22% (n = 33), 22% (n = 33), 23% (n = 35), and 34% (n = 51) expressed only vaccine serotypes, only non-vaccine serotypes, both vaccine and non-vaccine serotypes, and rare lineages, respectively. Lineages expressing both vaccine and non-vaccine serotypes accounted for 55% (1719/3104) of the isolates in our collection. The ten most common lineages in this study accounted for 50% (1561/3104) of the isolates (Fig. 1). An interactive visualisation of lineage data is available online (https://microreact.org/ZA_snp_analysis). The most commonly expressed serotypes among the ten predominant lineages were: GPSC1 (serotype 19F, n = 100/104 [96%]), GPSC2 (serotype 1, 226/226 [100%]), GPSC3 (serotype 8, 220/255 [86%]), GPSC5 (serotype 35B, 60/142 [42%]), GPSC10 (serotype 14, 120/155 [77%]), GPSC13 (serotype 6A, 118/126 [94%]), GPSC14 (serotype 23F, 153/167 [92%]), GPSC17 (serotype 19A, 199/202 [99%]), GPSC21 (serotype 19F, 90/91 [99%]), and GPSC41 (serotype 6A, 88/93 [95%]) and were those largely included in the PCV13 formulation. With reference to the GPS database, GPSC17, 21, and 41 were more prevalent among African isolates at frequencies of 92% (349/380), 100% (275/275), and 100% (129/129), respectively (Supplementary Table 4).Fig. 1: Single nucleotide polymorphism (SNP) maximum-likelihood phylogeny (N = 3104) of invasive pneumococcal disease isolates from South African children <18 years.The SNP tree is midpoint rooted. The 10 most predominant GPSC taxa are highlighted and listed in descending order: GPSC3 (n = 255/3104 [8%], GPSC2 (n = 226/3104 [7%], GPSC17 (n = 202/3104 [7%], GPSC14 (n = 167/3104 [5%], GPSC10 (n = 155/3104 [5%], GPSC5 (n = 142/3104 [5%], GPSC13 (n = 126/3104 [4%], GPSC1 (n = 104/3104 [3%], GPSC41 (n = 93/3104 [3%], and GPSC21 (n = 91/ 3104 [3%]. Antibiotic nonsceptibility by GPSC lineage is represented by coloured circular strips: PEN penicillin, AMO amoxicillin, CFT ceftriaxone, TAX cefotaxime, CFX cefuroxime, MER meropenem, CHL chloramphenicol, ERY erythromycin, CLI clindamycin, FLU fluoroquinolones, TET tetracycline, COT cotrimoxazole, GPSC global pneumococcal sequence clusters.Among children <5 years, the average incidence of vaccine-type lineages decreased significantly between the pre-PCV and PCV7 periods for GPSC2 (incidence rate ratio [IRR] 0.7, 95% confidence interval [CI] 0.9–0.5, p = 0.04), GPSC10 (0.6, 0.9–0.5, p = 0.03), GPSC11 (0.3, 0.7–0.1, p = 0.03), GPSC14 (0.7, 0.9–0.5, p = 0.03), GPSC52 (0.3, 0.8–0.1, p = 0.04), and GPSC77 (0.2, 0.4–0.09, p < 0.001, Fig. 2a and Supplementary Data 2). Similar to the <5-year-olds, the incidence of GPSC9 (0.05, 0.4–0.007, p < 0.001) and GPSC13 (0.2, 0.4–0.08, p < 0.001) declined significantly in the 5–17 year age group (Fig. 2b and Supplementary Data 3). Comparing the pre-PCV and late-PCV13 periods, both age groups showed similar incidence reduction patterns for vaccine-type lineages (Fig. 2a, b; Supplementary Data 4 and 5). For non-vaccine-type lineages, reductions in invasive disease caused by GPSC3 (0.4, 0.8–0.3, p = 0.03) and GPSC48 (0.3, 0.6–0.1, p = 0.03) were observed when comparing the pre-PCV and PCV7 periods among the <5 year age group. In contrast, significant increases in IPD caused by GPSC3 (1.8, 1.2–2.7, p = 0.008) were observed in the same age group when comparing the pre-PCV and late-PCV13 periods (Fig. 2a).Fig. 2: Global Pneumococcal Sequence Cluster (GPSC) dynamics among invasive disease isolates from children in South Africa (N = 3104).a children <5 years (n = 2680) and b children 5–17 years (n = 424). Invasive pneumococcal disease incidence per 100,000 population is plotted by GPSC lineages stratified into four vaccine periods (pre-PCV [2005–2008], PCV7 [2009–2010], early-PCV13 [2011–2014], and late-PCV13 [2015–2020]). Serotypes included in the 13-valent pneumococcal conjugate vaccine are shown by solid colour fills and non-vaccine serotypes by hatched pattern colour fills. Lineages with <40 isolates are not shown. A vaccine-type (VT) lineage was defined as having ≥50% vaccine serotype(s) in the pre-PCV period; non-vaccine-type (NVT) lineage as having >50% non-vaccine serotype(s) in the pre-PCV period. Lineage-specific clonal complexes (CC) and sequence types (ST) are shown underneath the graph. Either Poisson regression or negative binomial regression was used to calculate the incidence rate ratios (IRR) of GPSCs by vaccine period. If neither model fit, the IRR was calculated using the average annual incidence (Supplementary Data 2–5). Significant changes in the average incidence between the pre-PCV and PCV7 or late-PCV13 periods are represented by an asterisk or triangle, respectively. Significance was determined at <0.05 using a two-sided p-value where applicable. Multiple testing (>10 tests) was corrected using the Benjamini–Hochberg false discovery rate of 5%. Antibiotic nonsusceptibility proportions by lineage are shown in red horizontal bars. Penicillin (PEN) nonsusceptibility was predicted based on the penicillin-binding protein types: pbp1a, pbp2x, pbp2b; chloramphenicol (CHL), based on the presence of chloramphenicol acetyltransferase gene, cat; erythromycin (ERY), based on the presence of erythromycin resistance methylase gene ermB or macrolide efflux pump gene mefA; tetracycline (TET), by the presence of tetM or tetS/M gene with no promoter region interruptions; cotrimoxazole (COT), by the presence of mutation I100L in folA and/or indel within amino acid residue 56–67 in folP.Before PCV introduction GPSC5, 9, 10 and 11 predominantly expressed PCV13 serotypes. After PCV introduction, especially in the early- and late-PCV13 periods, these lineages largely expressed non-PCV13 serotypes (Fig. 2). GPSC5 (CC172) largely expressed vaccine serotype 23F (n = 29/33, 88%) with no observation of non-vaccine serotype 35B in the pre-PCV period in this collection. In the late-PCV13 period, replacement serotype 35B (CC172) was the commonly expressed serotype within this lineage (34/49, 69%) with no observation of serotype 23F (Supplementary Fig. 4). Similarly, both GPSC9 (CC63) and GPSC10 (CC230) expressed vaccine serotype 14 (27/27, 100% and 59/60, 98%) in the pre-PCV and PCV7 periods (9/9, 100%; 26/27, 96%), respectively (Supplementary Figs. 5 and 6). However, in the late-PCV13 period, GPSC9 largely expressed non-vaccine serotype 15A (13/19, 68%); GPSC10 expressed a combined proportion of 49% (20/41) of non-vaccine serotype 10A (11/41, 27%) and serogroup 24 (9/41, 22%), in addition to serotype 14 (12/41, 29%, Supplementary Figs. 5 and 6). Similar patterns were observed for GPSC11 (Supplementary Fig. 7).Phylodynamic analysisWe observed significant temporal signals (p < 0.001) for the persistent vaccine-type lineages of interest (Supplementary Data 6) that continued to circulate in the late-PCV13 period, allowing for coalescent analysis. Effective sample sizes (ESS) of greater than 200 were also obtained for these lineages. Past population dynamics for GPSC5 (CC172, serotype 35B) and GPSC17 (CC2062, serotype 19A) had comparable temporal trends with steady increases in the effective population size in the years after 1990 to the end of the study period in 2020 (Fig. 3a, d). This observation was supported by clonal expansion findings where the time-calibrated phylogeny for GPSC5 showed the highest probability for clonal expansion in 1992 (credible interval: 1981–1998). Similarly, GPSC17 had several clonal expansion events with the earliest expansion occurring in 1953 (1936-1965) followed by several others that coincided with the increasing effective population size in 1989 (1982–1994) and 1991 (1984–1996, Supplementary Fig. 8a, d). Although the effective population sizes for GPSC9 (serotype 14 and 15A) and GPSC10 (serotype 14, serogroup 24) showed similar increases around the 1990s, a decreasing trend in the effective population size for GPSC9 was observed before the introduction of PCV7 followed by a slight increase after PCV13 introduction (Fig. 3b). For GPSC10, however, this decrease coincided with the introduction of PCV7 in 2009 (Fig. 3c). Clonal expansion analysis for GPSC9 and GPSC10 revealed that the vaccine and non-vaccine-serotype variants in these lineages diverged before PCV introduction with non-vaccine serotype 15A prominently expanding in 2005 (2001–2009), shortly before PCV7 introduction (Fig. 4a, b).Fig. 3: Temporal trend analysis of important vaccine-type lineages causing persistent invasive disease in the late-PCV13 period inferred using skygrowth.a–d Past population size dynamics as a function of time inclusive of the pneumococcal conjugate vaccine (PCV) introduction years shown by the grey arrows (PCV7 in 2009 and PCV13 in 2011) in South Africa. Time-calibrated phylogenies were used for inference with root dates as start time. Root probabilities and the number of analysed genomes per lineage are shown in Supplementary Data 6. The clonal complexes (CC) and serotypes expressed by each lineage are shown in each panel title. The solid line represents the posterior median population size. The shaded purple area represents the 95% credible intervals. The effective population size on the y-axis is expressed in log scale. GPSC global pneumococcal sequence cluster. *Only serotypes that had 5 or more genomes are included in the (a–d) panel titles.Fig. 4: Clonal expansion dynamics of Global Pneumococcal Sequence Cluster (GPSC) lineages inferred using CaveDive.Time-resolved phylogenies for GPSC9 (a) and GPSC10 (b) showing within lineage clonal expansions with clades coloured by serotype. GPSC references are coloured in grey. The parental nodes with high probabilities for clonal expansions are highlighted by the clonal expansion frequency colour scale and the estimated expansion year with 95% credible intervals is included. Other parental nodes are highlighted in grey including the root date. GPSC global pneumococcal sequence cluster, CC clonal complex.Simpson’s diversity index trendsWe further explored serotype and lineage diversity indices which were generally high across the study period, ranging between 0.903–0.993 and 0.917–0.971, respectively (Fig. 5a, c). Compared to the pre-PCV and PCV7 years, significant increases in serotype diversity were observed in the early-PCV13 years, with the highest increase occurring in the PCV13 introduction year, 2011 (Simpson’s diversity index [SDI] 0.993, 95% CI 0.991–0.995). Declining serotype diversity trends were evident when comparing 2011 through to the late-PCV13 years, with diversity approaching pre-PCV estimates (Fig. 5a). However, significant increases in lineage diversity were observed between 2011 and 2013 (0.948, 0.937–0.960 vs 0.971, 0.963–0.979, Fig. 5c). Pooled lineage diversity trends reflected those of pooled serotype diversity trends both with significant differences when comparing the PCV7 and early-PCV13 periods (pooled lineage SDI 0.954, 95% CI 0.948–0.961 vs 0.965, 0.962–0.969 and pooled serotype SDI 0.915, 95% CI 0.907–0.924 vs 0.953, 0.950–0.957). Similarly, significant differences were also observed when comparing the early- and late-PCV13 periods (0.965, 0.962–0.969 vs 0.946, 0.937–0.956 and 0.953, 0.950–0.957 vs 0.938, 0.930–0.947, Fig. 5b, d).Fig. 5: Simpson’s diversity index (SDI) estimates for invasive disease serotypes and lineages.a Serotype diversity trends by year showing fluctuations before and after the introduction of 7- and 13-valent pneumococcal conjugate vaccines (PCV7 and PCV13) in 2009 and 2011, respectively (highlighted in light grey); the COVID-19 pandemic year (2020) is highlighted in dark grey. b Pooled serotype SDI estimates by vaccine period showing statistically significant differences between the PCV7 and early-PCV13 periods and early-PCV13 and late-PCV13 periods. c Lineage diversity trends by year before and after the introduction of PCV7 and PCV13 in 2009 and 2011, respectively (highlighted in light grey); the COVID-19 pandemic year (2020) is highlighted in dark grey. d Pooled lineage SDI estimates by vaccine period showing statistically significant differences between the PCV7 and early-PCV13 period and early-PCV13 and late-PCV13 periods. Estimates were derived at a 95% confidence interval among invasive disease isolates (N = 3104) from children <18 years in South Africa during the pre-PCV (n = 924), PCV7 (n = 457), early-PCV13 (n = 882), and late-PCV13 (n = 841) periods. SDI estimates represent the measure of diversity with standard error included at each data point.Antimicrobial resistanceAMR concordances comparing phenotypic antimicrobial susceptibility testing (AST) and in silico AMR predictions are shown in Supplementary Table 5. Among the 3104 IPD isolates, 2058 (66%) had at least one resistance determinant for the antibiotics investigated (Fig. 1). Overall, reductions in the prevalence of predicted antibiotic resistance among all classes of antibiotics, except chloramphenicol, were observed when comparing the pre-PCV and late-PCV13 periods (Table 1). In contrast, predicted antibiotic resistance to penicillin, erythromycin and MDR increased significantly in invasive disease caused by non-vaccine serotypes (Table 1). Among the 1249 non-vaccine serotypes, 232 (19%) had PBP gene combinations that conferred penicillin resistance. Of these, serotype 35B (GPSC5, n = 60), 15B/C (GPSC48, n = 46), 15A (GPSC9/GPSC168, n = 40), 10A (GPSC10, n = 12), 23B (GPSC102, n = 12) and serogroup 24 (GPSC10, n = 9) accounted for 77% (179/232) of the isolates and these serotypes were among the predominant non-vaccine serotypes circulating in the late-PCV13 period (Supplementary Figs. 1 and 2). Erythromycin resistance was identified in 50 (4%) of 1249 non-vaccine serotype isolates with ermB-mediated resistance occurring in 35 (70%) of 50 cases compared to mefA-mediated resistance. Of the 120 MDR non-vaccine serotype isolates, 73 (61%) were from the late-PCV13 period (Table 1). The three most common lineages driving MDR increase were GPSC9 (serotype 15A; 13/73, 18%), GPSC10 (serotype 10A; 11/73, 15% and serogroup 24; 9/73, 12%), and GPSC26 (serotype 12F, 30/73, 41%, Supplementary Table 6). GPSC9 and GPSC10 were both resistant to penicillin, tetracycline, and cotrimoxazole, with GPSC10 additionally resistant to erythromycin. All of serogroup 24 MDR isolates were from the late-PCV13 period in this study.Table 1 Changes in the prevalence of antibiotic nonsusceptibility among South African children, comparing the pre-PCV and late-PCV13 periods
