Introduction Streptococcus pneumoniae is a bacterium that frequently colonises the human nasopharynx. Apart from disease outcomes such as sinusitis, otitis media, and community-acquired pneumonia, which result from direct spread from the nasopharynx, the pneumococcus can invade the bloodstream and cause septicaemia, meningitis, and invasive pneumonia. Most carriage episodes, however, do not result in either local or systemic disease. It is believed that the propensity to cause invasive disease in healthy individuals—termed invasiveness—is largely determined by the characteristics of the pneumococcus' polysaccharide capsule, although the explicit underlying mechanisms are yet to be identified ,. On the basis of the immune response to differences in capsular polysaccharide structure, more than 90 serotypes causing invasive disease have been described . A pneumococcal conjugate vaccine (PCV7) that induces anticapsular antibodies against the seven serotypes, which at that time were responsible for most of the pneumococcal invasive disease in the United States (US), was introduced into the US childhood immunisation schedule in 2000 and the majority of the developed world subsequently. Since PCV7 is protective against invasive pneumococcal disease (IPD)  and carriage ,, the assumption of protection of the unvaccinated against vaccine type (VT) IPD through herd immunity played a major role in evaluating the likely impact and cost-effectiveness of vaccination . Prevention of VT carriage, however, creates a potential ecological niche in the nasopharynx for previously less prevalent serotypes to emerge (replacement). The extent to which the benefits of herd immunity will be offset by serotype replacement is hard to predict  and may vary by country depending on local factors such as differences in serotype distribution before vaccination and the population demography. Hence, there is a need for enhanced surveillance to evaluate the effect of vaccination in different epidemiological settings. Most surveillance systems focus on IPD and have shown large reductions in the numbers of VT cases in the targeted age groups, irrespective of vaccine schedule –. However, differences were observed in the indirect effect (i.e., the degree of induced herd immunity and the level of non–vaccine-type [NVT] replacement), the reasons for which remain unclear but may include vaccine coverage, time since introduction of PCV, and sensitivity of the reporting system . Monitoring disease outcomes provides little insight into the underlying mechanisms that determine herd immunity and serotype replacement. For this, carriage data are essential. Carriage studies in children from Massachusetts and Norway suggest full replacement of pneumococcus in carriage after PCV7 introduction ,. The implications of changes in serotype-specific carriage prevalence for expression as IPD will, however, depend on the invasiveness of individual serotypes, which is reflected by the case∶carrier ratio (CCR). Invasiveness has only been studied in one of these settings and was restricted to children ,. Improving our understanding of this relationship, largely determined by the invasiveness potential of the replacing NVT organisms, is essential to understanding the effect of PCV7 in different epidemiological settings. In September 2006, PCV7 was introduced into the immunisation schedule in the United Kingdom as a 2/4/13-month routine schedule with a catch-up for children up to 2 y of age. Information on carriage in England prior to PCV7 introduction is available from a longitudinal study conducted in 2001/2002 in index children and their household members. We report here the results of a cross-sectional carriage study conducted in a demographically similar population in 2008/2009. We compare our post-PCV7 findings with the pre-PCV7 baseline both for carriage and IPD to help understand the serotype-specific effects of PCV7 on both carriage and IPD and use this analysis to predict the potential impact of higher valency conjugate vaccines on herd immunity and replacement disease. Methods Study Population Children born since 4 September 2004 and thus eligible for routine or catch-up PCV were recruited along with family members from general practices in Hertfordshire and Gloucestershire. Exclusion criteria were: moderate to severe disability, cerebral palsy, neurological disorders affecting swallowing, ear, nose, and throat disorders affecting the anatomy of the ear, or immunosuppression. The NHS National Research Ethics Service approved the study protocol. Written informed consent was obtained from adult study participants and from a parent/guardian of study children prior to enrolment. Information was collected on participants' age, gender, household size, number of smokers in household, recent antibiotic treatment, hours in day-care and PCV7 vaccination history. To compare to prevaccination carriage in England, we used the results from a longitudinal study carried out in 2001/2002 in families attending the same general practices in Hertfordshire in which swabs were taken each month over a 10-mo period . At that time, serotype 6C could not be distinguished from 6A, but in 2009, 19 of the 122 serotype 6As from the earlier study were randomly retested, six of which were found to be 6C. We have assumed that this proportion (32%) holds for the rest of the 6A carriage isolates from the 2001/2002 study. Specimen Collection and Testing Nasopharyngeal swabs (calcium-alginate) were taken between April 2008 and November 2009 by trained nurses and placed directly in STGG broth. Samples collected at Hertfordshire were sent by same day courier to the Respiratory and Systemic Infection Laboratory at the Centre for Infections (RSIL). They were stored overnight in at 2–8°C and frozen the next morning at −80°C. Samples collected at Gloucestershire were stored locally at the Gloucester Vaccine Evaluation Unit at −80°C and transferred to RSIL in batches on dry ice. On receipt the batches were stored at −80°C. The sample then was thawed, vortexed, and 50 µl STGG broth was placed onto each of Columbia blood agar plate (HPA media services) with optochin disc (MAST) and Streptococcus-selective Columbia blood agar plate (HPA media services) and streaked out. The plates were incubated overnight at 35°C with 5% CO2. Any colonies resembling pneumococcus were subjected to normal identification methods and serotyped using the standard laboratory protocol . Statistical Analysis Descriptive data analysis was performed in R 2.11.0 and Generalized Estimating Equations (GEEs) models were analysed with STATA 10.1. Exact binomial 95% confidence intervals (CIs) were obtained for carriage rates in 2008/2009 by age group ( 20 y). To account for longitudinal design in the 2001/2002 study, we computed these carriage rates using a GEE model with exchangeable correlation structure. To determine the significance of changes in carriage for individual serotypes between 2001/2002 and 2008/2009, a Fisher exact test was used because of small numbers. When comparing overall carriage as well as vaccine and NVT carriage between periods, this comparison took account of the longitudinal design of the 2001/2002 along with other covariates by using a GEE model with exchangeable correlation structure and factors for study period, age in years, gender, whether the household has a smoker, and the number of children and adults in the household. For comparability with previously reported changes in carriage, the data were stratified into two age groups ( 20 y (%) 237 (49) 133 (35) n Proportion female 53.0% 56.4% n HH 121 146 Median HH size (range) 4 (2–7) 4 (3–7) Median n adults in HH (range) 2 (1–5) 2 (1–5) Median n children in HH (range) 1 (1–3) 2 (1–4) Proportion of smoke-free HH 66.9% 81.0% HH, household. A pneumococcus was grown from 127 of the 382 (33.2%) swabs and a serotype determined in 123 (97%). The most prevalent serotypes were 19A (10), 23B (9), 11C (8), 15B (8), 21 (8), and 6C (8). Compared to prevaccination levels, we found a significant reduction in carriage of VTs 6B, 14, 19F, 23F, and 6A. For the remaining PCV7 types no carriage episodes of serotypes 4 and 9V were found postvaccination, but prevaccination levels were too low to detect any significant change. VT 18C was identified in three out of 382 (0.79%) swabs in 2008/2009 and in 25 out of 3,868 (0.64%) in the 2001/2002 study. NVTs 33F, 7F, 10A, 34, 15B, 31, 21, 3, 19A, 15C, and 23A significantly increased (p 20 y Cases 2008/2009 (n = 133) 10 3 13 Proportion 2008/2009 7.5% (3–12) 2.3% (0–5.3) 9.8% (5.3–15) Proportion 2001/2002a 3.3% (2.4–4.8) 4.1% (3.0–5.5) 7.6% (6.2–9.5) All Cases 2008/2009 (n = 382) 112 10 127 Proportion 2008/2009 29.3% (24.9–34) 2.6% (1–4.5) 31.9% (27.2–36.6) Proportion 2001/2002a 8.5% (7.2–9.9) 15.2% (13.2–17.4) 24.4% (21.9–27.1) a The proportion for 2001/2002 was calculated accounting for multiple testing of the participants. 10.1371/journal.pmed.1001017.t003 Table 3 Odds ratios for comparing 2001/2002 to 2008/2009 carriage using GEE. Participants 5 y All VT 0.06 (0.03–0.16) abg *** 0.31 (0.04–2.49) a*** 0.07 (0.03–0.16) aeg ***, b** NVT 4.25 (2.81–6.43) c*,g*** 5.16 (1.95–13.66) ag** 4.40 (3.06–6.33) ag***, bc* All 1.03 (0.70–1.51) ab***, e* 2.46 (1.04–5.83) a***, g* 1.06 (0.76–1.49) ab***,e** Key for significant fixed effects: a, age; b, antibiotic treatment; c, smoking; d, gender; e, adults in household; f, children in household; g, study period. Significance codes: *≤0.05; **≤0.01; ***≤0.001. Simpson's index of diversity for the 2001/2002 samples was 0.908 95% CI (0.899–0.917); children: 0.891 95% CI (0.878–0.904) and adults: 0.936 95% CI (0.926–0.947). It increased significantly in the 2008/2009 samples to: 0.961 95% CI (0.953–0.969); children: 0.960 95% CI (0.949–0.971) and adults: 0.955 95% CI (0.928–0.982). Furthermore, the ranked frequency distribution of the serotypes, while similar in the prevaccination era in both children and adults in our study compared to children in Massachusetts, changed to become more distinct after vaccination (Figure 1). 10.1371/journal.pmed.1001017.g001 Figure 1 Top: Comparison in ranked-serotype distribution prior to vaccination in children in Massachusetts to our findings in children (left) and adults (right). For comparison with the findings with Hanage and colleagues, we aggregate 6A and 6C to 6A/C and 15B and 15C to 15B/C. Bottom: Changes in ranked serotype distribution in overall carriage in our findings from 2001/2002 to 2008/2009. Prior to its introduction, PCV7 included types responsible for similar proportions of carriage episodes (62.2%) and disease (55.9%). In 2008/2009 the additional types covered by higher valency vaccines were more prevalent in IPD than carriage, particularly the additional three in PCV10, which comprised 32.6% of IPD but only 4.7% of carried isolates (Table 4). 10.1371/journal.pmed.1001017.t004 Table 4 Carriage prevalence and IPD incidence in participants less than 60 y caused by serotypes included in PCV7, in PCV10 and not in PCV7, in PCV13 and not in PCV10, and the remaining serotypes. Serotypes 2008/2009 2001/2002 Carriage Percent Percent in IPD Carriage Percent Percent in IPD PCV7 11 (8.7) 15.2 605 (62.2) 55.9 +PCV10 6 (4.7) 32.6 2 (0.2) 10.2 +PCV13 18 (14.2) 15.8 155 (15.9) 8.9 Rest 92 (72.4) 36.4 210 (21.7) 25.0 The ranking of carried serotypes by frequency of detection in the post-PCV7 dataset and their associated CCRs as estimated from our 2008/2009 carriage prevalence data are shown in Figure 2. CCR estimates were highly correlated (p 30% of IPD cases, whereas the further three serotypes in PCV13 are more similar to the PCV7 serotypes, being similarly prevalent in carriage and disease. While changing to PCV10 has therefore less potential to prevent IPD than PCV13, it may cause fewer perturbations in the nasopharyngeal pneumococcal population. Comparative carriage studies in countries using PCV10 with those using PCV13, or with different PCV coverage of prevalent serotypes before introduction, would be informative to help understand the carriage dynamics underlying serotype replacement. These studies would ideally be repeated cross-sectional studies to monitor alterations in carriage prevalence, which could be linked to changes in serotype-specific IPD in the same population. The latter requires the continued microbiological investigation of suspected cases of invasive disease, including those in fully vaccinated children, in order to document the serotype-specific changes in IPD associated with vaccine-induced changes in carriage. The diversity of the pneumococcal carriage population in the absence of any external pressure is thought to be relatively stable . If this population is challenged by vaccination with a reduction in the dominance of a few highly prevalent types, the diversity increases and the population takes time to return to the previous level of diversity. Hanage and colleagues suggested methods of assessing these changes: Simpson's index of diversity and the concept of a typical distribution for the ranked frequency of the serotypes . Applying these to our prevaccination carriage data, we see similar diversity in children and slightly higher diversity in adults, although the significance of this difference was not consistent between both methods. However, we found an increase in overall diversity in 2008/2009 as well as in children and in adults (although not significant in adults), consistent with the PCV7-induced changes in the bacterial population still evolving at that time. Evidence for this can also be found in the ongoing changes in non-PCV7 IPD in 2009/2010, prior to introduction of PCV13. These show a continuing increase in the six additional serotypes covered by PCV13 but a decrease in non-PCV13 serotypes in children under 2 y compared with 2008/209 . With the introduction of PCV13 in the UK in March 2010 , it will not be possible to evaluate further the longer term impact of PCV7 on carriage and IPD, but it is important to note that PCV7 may continue to have an effect and therefore not all future changes will necessarily be attributable to PCV13. Recently developed molecular serotyping methods found up to nine times higher proportions of multiple carriage than detectable with standard WHO culturing methods . Using the WHO method we identified one (0.26%) multiple carriage episode in 2008/2009 and four (0.10%) in 2001/2002. Undetected episodes of multiple carriage would result in over estimation of CCRs. However, direct comparison of molecular and conventional serotyping methods have so far only been performed on specimens from developing countries where carriage prevalence is very high ,. In such settings, molecular methods might reveal more multiple carriage episodes than in countries such as England where carriage prevalence is lower. Furthermore, there is some evidence that detecting multiple serotype carriage is likely to primarily uncover carriage episodes of serotypes previously found to be less prevalent . Therefore we believe that the potential bias introduced by the WHO standard culturing methods would have little impact on our inferences from the CCR, because we focus on the serotypes more common in carriage. Our study has some limitations. First, the earlier study had a longitudinal design while the recent study was cross-sectional. However, we accounted for multiple testing of individuals in the earlier study as well as differences in age distribution within the age groups, gender, exposure to smoke, and household size by using a GEE, which is designed to fit the parameters of a generalised linear model in the presence of unknown correlation. Second, owing to the lack of power of serotype-specific carriage data in adults, we pooled data of children and adults to derive the CCR, despite different age distributions in the samples for IPD and carriage. Previously reported CCR estimates for children and adults in England and Wales  using the carriage data from the earlier study are highly correlated (Figure S1), supporting our use of pooled carriage data from children and adults in the later study. Third, secular changes in serotype distribution in IPD can occur in the absence of vaccination , which may be due to alterations in carriage prevalence. With the cross-sectional design of the 2008/2009 study, we were not able to account for these. However, in England the only major secular change in the serotypes causing IPD observed over the last decade has been in serotype 1, which was not detected in either our pre- or post-PCV7 carriage studies. Fourth, invasion is thought to follow shortly after acquisition of carriage rather than being a constant risk throughout the duration of carriage . Thus, a further potential limitation of our study is that we estimate CCRs using carriage prevalence rather than the incidence of new carriage episodes, the latter being derived using prevalence and carriage duration. Few data on serotype-specific duration of carriage are published, and for the serotypes newly emerging after introduction of PCV7, no information is available. Therefore, we used carriage prevalence to get an estimate of the CCRs. Where information on CCRs estimated using carriage incidence was available , we found a high correlation with our estimates. Furthermore, our estimates for the CCRs were consistent with those derived from 2001/2002 carriage and IPD (unpublished data), showing that this measure is stable over time. Hence we are confident that our estimates of the CCR can distinguish serotypes with lower invasiveness from those with higher invasiveness. In conclusion, our study illustrates the value of generating carriage data in parallel with IPD surveillance data to help understand the serotype-specific changes in IPD observed in different epidemiological settings and predict the effect of higher valency vaccines. We provide evidence that the incremental benefit on IPD of the recent switch from PCV7 to PCV13 in the UK, while likely to be substantial, may be somewhat offset by increases in serotypes 8, 12F, and 22F. Such emerging serotypes with high CCRs are potential candidates for inclusion in future conjugate vaccines. More research to elucidate the serotype-specific capsular properties , or other factors associated with carriage and invasiveness is needed in order to understand better the likely impact of future conjugate vaccines. Supporting Information Figure S1 Estimated CCR in children and adults from Trotter and colleagues . The grey lines represent the confidence bounds. Spearman's rank test for correlation: p = 0.01, ρ = 0.62. (TIF) Click here for additional data file. Table S1 Estimated CCR for all serotypes found in carriage in 2008/2009 with the corresponding number of isolates found in carriage and in IPD in the under 60-y-old population. Some serotypes were found in IPD but not in carriage: sertoype 1 (387 isolates), 8 (196), 12F (83), 4 (74), 9V (58), 14 (48), 5 (23), 20 (21), 15A (12), 17F (10), 16F (10), 35B (6), 27 (6), 13 (4), 28A (3), 12B (3), and one each of 9L, 7C, 7B, 7A, 7, 6, 35A, 28, 18A, 10F. *A total of 81 isolates of 6A/6C were found in IPD of which one-third was assumed to be 6C and the rest 6A. (DOC) Click here for additional data file. Table S2 Number of isolates found in carriage in 2001/2002 and 2008/2009. In 2001/2002 carriage of a second serotype was detected in four isolates; serotypes 22F (1 isolate), 3 (2), and 6B (1) were found. In 2008/2009 additional carriage of serotype 21 was detected once. *6A and 6C were not distinguished in 2001/2002. (DOC) Click here for additional data file.