Nasopharyngeal carriage of Streptococcus pneumoniae is required for transmission of the bacteria and for invasive disease. There have been conflicting reports as to whether protection against carriage is serotype specific and which immune mechanisms drive carriage. Analyzing longitudinal carriage data from Israeli toddlers in day care, we found a lower risk of colonization with types 6A, 14, and 23F after previous exposure to the homologous type. Nonsignificant trends suggesting possible protection derived from prior exposure were found for types 19A and 23A. Furthermore, we found that, for types 14 and 23F, this specific protection correlated with increased serotypespecific antibody concentration. We found no evidence of specific protection for type 6B, group 15, or type 19F. Our findings imply that at least some serotypes generate anti-capsular antibodies that can reduce the risk of carriage in unimmunized toddlers.
Streptococcus pneumoniae is a major public health concern worldwide and is an important cause of pneumonia, meningitis, otitis media, and septicemia. Pneumococcal disease most frequently affects children under the age of 2 years, the elderly, and immunocompromised individuals. Nasopharyngeal carriage precedes invasive disease and is the source of most transmission . There are 91 known serotypes of S. pneumoniae, with classification based on the structure of the capsular polysaccharide. The serotypes vary greatly in their prevalence, and a small number of types account for a large proportion of carriage isolates [1, 2].
It has generally been thought that the immune response to pneumococcus in naturally exposed, unvaccinated children depends primarily on antibody directed against the polysaccharide capsule. This notion has been supported by the fact that immunization with a polysaccharide conjugate vaccine generates an effective serotype-specific antibody response. Similarly, treatment of infected patients with serum containing typespecific anti-capsular antibodies can aid in bacterial clearance , and patients with agammaglobulinemia are at a significantly increased risk of pneumococcal disease . Nasopharyngeal carriage can lead to the production of antibodies against specific capsular polysaccharides  and against pneumococcal surface proteins [6, 7].
However, despite these suggestions that anti-capsular antibodies play a role in the development of immunity to pneumococcus, there has been little quantitative evidence that these antibodies are the primary mechanism of naturally acquired immunity in children, and several recent studies have suggested otherwise. Although there is a decline in carriage duration  and carriage prevalence  for many serotypes during the second year of life, only certain serotypes generate measurable antibody, and this occurs in a minority of children by the age of 2 years. An experimental challenge study in adults found no evidence of an association between prechallenge anti-capsular antibody concentrations and protection , whereas a longitudinal observational study in adults  found evidence for such an association for 1 of 6 serotypes evaluated. Additionally, a recent study of elderly patients with chronic obstructive pulmonary disease (COPD) found no association between anti-capsular antibody production and protection . In mice, acquired protection against pneumococcal carriage is not type specific and does not require antibodies but does depend on an effective CD4+ T cell response [11, 12].
It is unknown, then, whether the serotype-specific antibodies generated in response to natural carriage provide protection to the host and, if so, what proportion of protection can be attributed to these antibodies. In the present study, we tested the hypothesis that previous colonization with a particular serotype specifically protects against new colonization with that same type. Our data suggest that type-specific protection occurs for some serotypes. In particular, we found that colonization with types 6A, 14, 23F, and possibly 19A and 23A was associated with specific protection against future colonization with the homologous type, and we found no evidence of protection for type 6B, group 15, or type 19F. Additionally, we found that, for type 14, prior colonization correlated with increased concentrations of anti-capsular antibody. Moreover, higher concentrations of type 14– and type 23F–specific antibodies correlated with lower levels of new acquisition of these types.
Data collection. In this study, we used data collected during a randomized trial of pneumococcal conjugate vaccine among children in Israel. Details of this study have been described elsewhere [13, 14]. Subjects aged 12–35 months were followed up for 2 years—monthly for the first year and every other month for the second year. The children were randomized to receive either a 9-valent pneumococcal conjugate vaccine or a control meningococcal type C conjugate vaccine. For the present analysis, we focus on the 132 subjects (2105 samples) in the control group (table 1).
Nasopharyngeal cultures. Bacterial isolates from the nasopharynx were collected with Dacron-tipped swabs, as described elsewhere . A single colony was then isolated, and the serotype was determined on the basis of the Quellung reaction.
A patient was considered to be “at risk” for a new carriage episode with a serotype if that serotype was not detected in either of the last 2 swab samples, and only at-risk episodes were assessed in the analysis for each type. We focused on the 8 most prevalent serotypes/groups in the study population: 6A, 6B, 14, 15, 19A, 19F, 23A, and 23F. Other serotypes were detected in the population but were too rare to include in the analysis. Crude odds ratios (ORs) were calculated to compare the odds of acquiring a particular serotype between subjects who had been previously colonized with that serotype and those who had not.
Detection of a specific protective effect. We used a generalized estimating equation (GEE) with a logistic link to identify predictors of new carriage episodes in this longitudinal data set. This approach is required to account for within-subject correlation. To confirm the strength of our analytical technique, we also conducted a univariate analysis and sensitivity analyses. Predictors in the GEE included prior colonization with each of the 8 types, age at study entry (quartiles corresponding approximately to 6-month intervals), age at time of swab collection (cubic spline), and presence of a positive swab for any other type on the preceding visit. Visits of a single subject were assumed to have an exchangeable correlation structure. In sensitivity studies, we reanalyzed the data by defining a person as being at risk for a new carriage episode with a particular serotype if that serotype was not detected in the previous swab or in any of the previous 3 swabs.
To evaluate the contribution of exposure to a given serotype within a day care center to the risk of carriage acquisition, we determined how many times a particular type had been detected within 30 days of a visit in other subjects at the same day care center and how many swabs had been collected during the same period. For this analysis, we focused on the 3 serotypes that showed significant protection in our initial analysis. The ratio of positive cultures to total swabs, an estimate of prevalence, was logit transformed, and we then reanalyzed the data using the GEE described above with the addition of this new predictor for exposure. All statistical analyses were performed with Intercooled Stata software (version 9.0; StataCorp), using the uvrs regression splines command.
Serological testing. Blood samples were obtained ∼0, 12, and 24 months after enrollment in the study, and serum was stored at –70°C. Serotype-specific serum IgG concentrations (in micrograms per milliliter) were determined by ELISA, as described elsewhere [14, 15]. Antibody data were available for types 6B, 14, 19F, and 23F.
Antibody effects. To determine whether colonization affected anti-capsular antibody concentrations, we performed linear regression with log10 antibody concentration as the dependent variable, with the following predictors: prior colonization with the homologous serotype, subject age at the time of blood sampling (cubic spline), and age at entry into the study. The antilog of the regression coefficient for antibody concentration is reported as the fold increase in antibody concentration. To determine whether homologous antibody concentrations affected new acquisition of each of the types, we used a GEE with the following predictors: log10 antibody concentration at the most recent blood sampling, subject age, age at entry into the study, and colonization with any pneumococcus at the most recent visit.
Crude analyses: serotype-specific protection for types 23F, 14, and 23A. In a crude analysis, we calculated ORs for the frequency of new colonization events in previously colonized hosts compared with those for whom we had no evidence of prior colonization with the same type. We found that if a child had been previously colonized with type 23F, the odds of future colonization with 23F were approximately half that for a child who had not previously carried this type (OR, 0.49 [95% confidence interval [CI], 0.30–0.80]) (table 2). Types 14 and 23A demonstrated a nonsignificant trend toward specific protection, whereas prior colonization with group 15, type 6A, type 6B, type 19A, or type 19F did not significantly affect the odds of future acquisition of the same type. Table 2 also shows the association between the number of times an individual had previously acquired a specific type and that person's risk of reacquiring that type.
Adjusted analysis: serotype-specific protection for types 6A, 14, and 23F. Although informative, this initial analysis did not control for a number of important factors, including subject age, age at entry into the study, and recent colonization with other serotypes of pneumococcus. To estimate the association between prior carriage and risk of acquisition while adjusting for these confounders, we used a GEE with a logistic link. Similar to what was seen in the crude analysis, we found that prior colonization with type 23F was associated with significantly lower odds of future colonization with 23F (OR, 0.47 [95% CI, 0.26–0.86]). Additionally, types 6A (OR, 0.48 [95% CI, 0.27–0.84]), and 14 (OR, 0.08 [95% CI, 0.01– 0.66]) showed statistically significant evidence of specific protection in this analysis (table 3). Types 19A (OR, 0.58 [95% CI, 0.15–2.21]) and 23A (OR, 0.51 [95% CI, 0.14–1.84]) showed nonsignificant trends toward specific protection. Previous colonization with type 6B, group 15, or type 19F showed no evidence of specific protection. Interestingly, prior colonization with type 6B showed a nonsignificant trend toward protection against type 6A, and prior colonization with type 23A showed a trend toward protection against type 23F.
Effect of serotype prevalence in day care centers. If a particular serotype were to sweep through a day care center in a short period, causing a “microepidemic,” and then disappear from that day care center, a spurious association could occur between prior colonization with that serotype and reduced risk of acquiring that type, owing to temporal differences in risk. In such a case, it might be impossible by our definition to distinguish protection due to previous colonization and a decrease in new acquisition due to decreased exposure within the day care center. To control for this potential artifact and determine whether changes in exposure could account for the observed patterns of acquisition, we added a predictor to the GEE to account for the prevalence of a specific serotype in a day care center within 30 days of each at-risk visit, and we then reanalyzed the data. We focused on the 3 types that showed evidence of specific protection in the initial analysis. For types 14 and 23F, increased prevalence within a day care center increased the odds of future acquisition of that serotype, but the observed serotype-specific protective effect was not greatly affected (table 4). In particular, for each unit increase in our surrogate for the prevalence of type 14 during the previous 30 days, the odds of acquiring type 14 were 2.44 (95% CI, 1.82–3.28) times as great. Inclusion of this predictor in the logistic model had a negligible effect on the serotype-specific protective effect from prior colonization (OR, 0.07 [95% CI, 0.01–0.69]), suggesting that changes in prevalence were not an important confounder of the observed protective effect of prior colonization. Similarly, increased prevalence of type 23F during the previous 30 days was associated with increased odds of colonization (OR, 1.37 [95% CI, 1.08–1.74]) with type 23F, and again the serotype-specific protective effect was not affected by the inclusion of this term in the logistic model (OR, 0.43 [95% CI, 0.23–0.81]). For type 6A, we found no significant association between day care center prevalence and risk of acquisition of type 6A (OR, 1.16 [95% CI, 0.93–1.44]), and inclusion of this term did not strongly affect the serotype-specific protective effect (OR, 0.47 [95% CI, 0.27–0.82]).
Relationship between anti-capsular antibody and protection. To determine the effect of prior colonization with a particular serotype on log-transformed anti-capsular IgG concentration, we used linear regression, adjusting for subject age and age at entry into the study, and found that prior carriage of type 14 was associated with a 2.59-fold (95% CI, 1.69–3.97-fold) increase in type 14 antibody concentration (table 5). In contrast, no significant association was found between prior colonization and homologous antibody concentration for type 6B (1.09-fold increase [95% CI, 0.74–1.61]), type 19F (1.25-fold increase [95% CI, 0.92–1.70]), or type 23F (1.10-fold increase [95% CI, 0.83–1.46]). Finally, to determine whether the log concentration of anti-capsular antibody was associated with protection against new acquisition of the homologous type, we used a GEE and found that increased concentrations of type 14 antibody were specifically associated with protection against future colonization with type 14 (OR for a 10-fold increase in antibody concentration, 0.40 [95% CI, 0.20–0.80]) (table 5). We also found that a higher concentration of type 23F antibody was protective against future colonization with type 23F (OR, 0.44 [95% CI, 0.25–0.78]), even though our linear regression did not detect an increase in type 23F antibody in response to previous colonization. Type 6B (OR, 1.50 [95% CI, 0.60–3.75]) or type 19F (OR, 0.88 [95% CI, 0.56–1.37]) antibody concentrations did not provide any significant specific protection against colonization with the homologous type.
This study provides epidemiologic evidence that serotypespecific protection does occur after natural carriage in some cases. The strength of the evidence for this protective effect varied across the serotypes considered. In particular, type 23F demonstrated strong serotype-specific protection in all analyses, and types 6A and 14 demonstrated significant protection with the multivariate GEE regression, which adjusted for within-subject correlation in the risk of colonization with a particular serotype. We also saw some evidence of cross-protection between types within a serogroup for groups 6 and 23, although these were nonsignificant trends and the effect was not reciprocal.
To determine the robustness of our findings, we also performed a crude analysis and sensitivity analyses. For serotype 6A, we did not see any protective effect in the crude analysis, but we did see significant protection when we applied the multivariate GEE framework, which tends to reduce the weight attributed to carrying type 6A 3 or more times. The results for type 6A could be particularly sensitive to use of the GEE framework because a larger proportion of persons in this study population carried type 6A on 3 or more occasions than other types. Sensitivity analyses demonstrated that changing the risk definition to require 3 negative swabs before a person is considered to be at risk for 6A did not have a significant effect on the OR (data not shown). This suggests that the large proportion of recolonizations with type 6A was not due to misclassification of new carriage episodes.
The sampling strategy used to collect these data could also potentially bias the results. At each visit, the serotype of only 1 colony was determined. As a result, if a subject was colonized with multiple serotypes, we would be more likely to detect the dominant type, and some colonization events could be masked. Additionally, because subjects were sampled every 1–2 months, we might not detect colonization episodes that lasted <1 month.
Inclusion of the cross-serotype predictors in the GEE model (table 3) had a negligible effect on the outcome of the serotypespecific analysis but served as a comparison to demonstrate the specificity of the protective effect. There were a few instances of cross-serotype protection, and we expected this finding when performing our analysis at a type I error rate of 5% for 56 crossserotype comparisons. We believe that some of these results could be due to chance rather than a biological phenomenon. Interestingly, though, prior colonization with type 6B showed a trend for protecting against type 6A, and prior colonization with type 23A showed a trend for protecting against type 23F. This within-serogroup cross-protection was expected, although the protective effect was not reciprocal.
The study design employed here was not optimized for the detection of cross-protection between different serotypes. Because the overall incidence of new pneumococcal colonization decreases with age, we could distinguish between a general decrease in colonization due to general immune maturation and a specific decrease resulting from pneumococcus-specific crossserotype protection. It is likely, then, that additional serotypeindependent factors contribute to the decrease in pneumococcal disease and carriage as a child matures.
We cannot rule out the possibility that the apparent serotype-specific protective effect was due to the genetic background of the strains and not the capsule type. If the strains of a given serotype circulating in a day care center tended to come from a single clone or a few closely related clones, our observation of serotype-specific protection could in principle be a marker of protection mediated by immune responses to other determinants, such as PspA. In such a situation, the specific protective effect could be due to conserved epitopes linked to the genetic background rather than the capsule type
Prior colonization with group 15 or type 19F was not associated with specific protection against carriage. Because this study had a relatively small sample size with resulting low power, it is difficult to say with certainty that there was a lack of protection in these subjects. However, given the 95% CIs, if there was a specific effect, it would probably be modest and within∼50% in either direction. Our findings of protection from types 14 and 23F but not type 19F anti-capsular antibodies are consistent with some previous findings. Studies of the effectiveness of pneumococcal conjugate vaccine against carriage of type 19F have found conflicting results. One study found that type 19F antibody concentration did not affect new acquisition . Among vaccinees in the same trial for which we have here analyzed the controls, very high concentrations of antibody against type 19F capsular polysaccharide were associated with protection against colonization . It is possible that the antibody concentration (or quality) generated by natural carriage among the unimmunized toddlers studied here was insufficient to achieve the level of protection seen in high responders to the conjugate vaccine. Vaccine-induced antibodies against type 14 in the vaccine arm of this study population  and naturally produced antibodies against type 14 in adults in the United Kingdom  correlated with specific protection against colonization, which is consistent with our findings. The effect of type 23F antibody concentration among vaccinees in the vaccine arm of this study population was modest, although in the predicted direction .
A recent study of patients with COPD found no association between antibody concentration and colonization , but this was in an adult population that probably had different immune characteristics than the children in this study. It has been previously proposed, on the basis of epidemiologic data, that the protection against pneumococcal disease acquired with age is due to a factor other than anti-capsular antibody . The present study, however, addressed only carriage and not invasive disease. Additionally, for some serotypes, interpreting the data is difficult owing to the quality of the ELISA used to acquire those data. The serum was not absorbed with type 22F polysaccharide before the assay, so we would expect a high level of nonspecific binding for serotypes other than type 14 .
Despite the evidence that higher concentrations of type 23F antibody correlate with protection, our analysis did not detect an increase in type 23F antibody concentrations in response to prior colonization. This discrepancy could result from exposures to type 23F that occurred before enrollment in the study and were not included in our data.
In evaluating observational data for evidence of serotypespecific acquired immunity, the most compelling case would include at least 3 key findings. First, we would find protection resulting from prior colonization that was robust to the details of the analytical approach. Second, we would expect to find an increase in type-specific antibody in response to carriage. Third, we would expect that these antibodies would effectively decrease the risk of future colonization with the same type.Wecame close to this standard for 2 of the types examined. For type 14, we saw evidence of specific protection in the adjusted analysis, a trend toward protection in the crude analysis, the production of typespecific antibody, and evidence that these antibodies are protective against new colonization. For type 23F, we found evidence of specific protection in both the crude and adjusted analyses and found that elevated type 23F antibody was protective, although we did not detect an increase in specific antibody in response to colonization. For type 6A, we did not have data on the specific antibody, although increased concentrations of type 6B antibody were not protective (data not shown).
As with all observational studies, this analysis has several important limitations. Althoughweperformed analyses to account for the effect of transmission within day care centers, we could not completely eliminate the effects of possible differences in exposure over time and across day care centers, which could influence acquisition of new strains. Because these differences might be correlated with the history of exposure in a particular subgroup, they could lead to confounding of the association between prior carriage and the risk of further acquisition.Onthe other hand, they certainlyaddnoise to the data and may obscure associations that are real. Further studies will be required to evaluate more fully the role played by antibody in protecting against different serotypes, the generality of the serotypespecific patterns that we have found here, and the role played by the bacterial genetic background and clonality in eliciting specific protection.
We thank Justin O'Hagan and William Hanage for helpful discussions of the manuscript.