Estimating the Transmission Dynamics ofStreptococcus pneumoniaefrom Strain Prevalence Data

Pneumococcus and the stress-gradient hypothesis: A trade-off links R0 and susceptibility to co-colonization across countries

Modern molecular technologies have revolutionized our understanding of bacterial epidemiology, but reported data across studies and different geographic endemic settings remain under-integrated in common theoretical frameworks. Pneumococcus serotype co-colonization, caused by the polymorphic bacteria Streptococcus pneumoniae, has been increasingly investigated and reported in recent years. While the global genomic diversity and serotype distribution of S. pneumoniae have been well-characterized, there is limited information on how co-colonization patterns vary globally, critical for understanding the evolution and transmission dynamics of the bacteria. Gathering a rich dataset of cross-sectional pneumococcal colonization studies in the literature, we quantified patterns of transmission intensity and co-colonization prevalence variation in children populations across 17 geographic locations. Linking these data to an SIS model with cocolonization under the assumption of quasi-neutrality among multiple interacting strains, our analysis reveals strong patterns of negative co-variation between transmission intensity (R0) and susceptibility to co-colonization (k). In line with expectations from the stress-gradient-hypothesis in ecology (SGH), pneumococcus serotypes appear to compete more in co-colonization in high-transmission settings and compete less in low-transmission settings, a trade-off which ultimately leads to a conserved ratio of single to co-colonization μ=1/(R0-1)k. From the mathematical model's behavior, such conservation suggests preservation of 'stability-diversity-complexity' regimes in coexistence of similar co-colonizing strains. We find no major differences in serotype compositions across studies, pointing to adaptation of the same set of serotypes across variable environments as an explanation for their differential interaction in different transmission settings. Our work highlights that the understanding of transmission patterns of Streptococcus pneumoniae from global scale epidemiological data can benefit from simple analytical approaches that account for quasi-neutrality among strains, co-colonization, as well as variable environmental adaptation.

Flexible Bayesian inference on partially observed epidemics

Individual-based models of contagious processes are useful for predicting epidemic trajectories and informing intervention strategies. In such models, the incorporation of contact network information can capture the non-randomness and heterogeneity of realistic contact dynamics. In this article, we consider Bayesian inference on the spreading parameters of an SIR contagion on a known, static network, where information regarding individual disease status is known only from a series of tests (positive or negative disease status). When the contagion model is complex or information such as infection and removal times is missing, the posterior distribution can be difficult to sample from. Previous work has considered the use of Approximate Bayesian Computation (ABC), which allows for simulation-based Bayesian inference on complex models. However, ABC methods usually require the user to select reasonable summary statistics. Here, we consider an inference scheme based on the Mixture Density Network compressed ABC, which minimizes the expected posterior entropy in order to learn informative summary statistics. This allows us to conduct Bayesian inference on the parameters of a partially observed contagious process while also circumventing the need for manual summary statistic selection. This methodology can be extended to incorporate additional simulation complexities, including behavioural change after positive tests or false test results.

On predictive inference for intractable models via approximate Bayesian computation

Approximate Bayesian computation (ABC) is commonly used for parameter estimation and model comparison for intractable simulator-based statistical models whose likelihood function cannot be evaluated. In this paper we instead investigate the feasibility of ABC as a generic approximate method for predictive inference, in particular, for computing the posterior predictive distribution of future observations or missing data of interest. We consider three complementary ABC approaches for this goal, each based on different assumptions regarding which predictive density of the intractable model can be sampled from. The case where only simulation from the joint density of the observed and future data given the model parameters can be used for inference is given particular attention and it is shown that the ideal summary statistic in this setting is minimal predictive sufficient instead of merely minimal sufficient (in the ordinary sense). An ABC prediction approach that takes advantage of a certain latent variable representation is also investigated. We additionally show how common ABC sampling algorithms can be used in the predictive settings considered. Our main results are first illustrated by using simple time-series models that facilitate analytical treatment, and later by using two common intractable dynamic models.

SUPPLEMENTARY INFORMATION

The online version contains supplementary material available at 10.1007/s11222-022-10163-6.

It Takes Two to Tango: Combining Conventional Culture With Molecular Diagnostics Enhances Accuracy of Streptococcus pneumoniae Detection and Pneumococcal Serogroup/Serotype Determination in Carriage

Background

The specificity of molecular methods for the detection of Streptococcus pneumoniae carriage is under debate. We propose a procedure for carriage surveillance and vaccine impact studies that increases the accuracy of molecular detection of live pneumococci in polymicrobial respiratory samples.

Methods

Culture and qPCR methods were applied to detect pneumococcus and pneumococcal serotypes in 1,549 nasopharyngeal samples collected in the Netherlands (n = 972) and England (n = 577) from 946 toddlers and 603 adults, and in paired oropharyngeal samples collected exclusively from 319 Dutch adults. Samples with no live pneumococci isolated at primary diagnostic culture yet generating signal specific for pneumococcus in qPCRs were re-examined with a second, qPCR-guided culture. Optimal C_q cut-offs for positivity in qPCRs were determined via receiver operating characteristic (ROC) curve analysis using isolation of live pneumococci from the primary and qPCR-guided cultures as reference.

Results

Detection of pneumococcus and pneumococcal serotypes with qPCRs in cultured (culture-enriched) nasopharyngeal samples exhibited near-perfect agreement with conventional culture (Cohen’s kappa: 0.95). Molecular methods displayed increased sensitivity of detection for multiple serotype carriage, and implementation of qPCR-guided culturing significantly increased the proportion of nasopharyngeal and oropharyngeal samples from which live pneumococcus was recovered (p < 0.0001). For paired nasopharyngeal and oropharyngeal samples from adults none of the methods applied to a single sample type exhibited good agreement with results for primary and qPCR-guided nasopharyngeal and oropharyngeal cultures combined (Cohens kappa; 0.13–0.55). However, molecular detection of pneumococcus displayed increased sensitivity with culture-enriched oropharyngeal samples when compared with either nasopharyngeal or oropharyngeal primary cultures (p < 0.05).

Conclusion

The accuracy of pneumococcal carriage surveillance can be greatly improved by complementing conventional culture with qPCR and vice versa, by using results of conventional and qPCR-guided cultures to interpret qPCR data. The specificity of molecular methods for the detection of live pneumococci can be enhanced by incorporating statistical procedures based on ROC curve analysis. The procedure we propose for future carriage surveillance and vaccine impact studies improves detection of pneumococcal carriage in adults in particular and enhances the specificity of serotype carriage detection.

Serotype-Dependent Effects on the Dynamics of Pneumococcal Colonization and Implications for Transmission

Capsule-switch mutants were compared to analyze how serotype affects the success of Streptococcus pneumoniae (Spn) during colonization and transmission. Strains of multiple serotypes were tested in highly susceptible infant mice, both singly and in competitive assays. Our findings demonstrated a role of serotype, apart from genetic background, in competitive success of strains, but this depended on timing postinoculation. As is the case for natural carriage, there was a hierarchy of success among serotypes using capsule-switch strains. The long-term dominance of a serotype was established within the first 4 h after acquisition, suggesting an effect independent of Spn-induced host responses. The hierarchy of serotype dominance correlated with decreased clearance rather than increased growth in vivo. Competitive assays staggering the timing of challenge showed that the first strain to dominate the niche sustained its competitive advantage, potentially explaining how increased density from delayed early clearance could result in serotype-dependent success. Effector molecules of intrastrain competition (fratricide), regulated by the competence operon in a quorum-sensing mechanism, were required for early niche dominance. This suggested a winner-takes-all scenario in which serotype is a major factor in achieving early niche dominance, such that once a strain reaches a threshold density it is able to exclude competitors through fratricide. Serotype was also an important determinant of transmission dynamics, although transit to a recipient host depended on effects of serotype different from its contribution to the dominance of colonization in the donor host. IMPORTANCE Capsule is the major virulence factor and surface antigen of the opportunistic respiratory pathogen Streptococcus pneumoniae (Spn). Strains of Spn express at least 100 structurally and immunologically distinct types (serotypes) of capsule, but for unknown reasons only a few are common. The effect of serotypes during the commensal interactions of Spn and its host, colonization and transmission, was tested. This was carried out by comparing genetically modified strains differing only in serotype in infant mouse models. Results show that serotype is an important factor in a strain's success during colonization. This was attributed to the effect of serotype on early clearance of the organism in the host. Competitive factors expressed by Spn (in a mechanism referred to as fratricide) allow the strain gaining this initial advantage to then dominate the upper respiratory tract niche. Serotype also plays an important role in a strain's success during transmission from one host to another.

Widespread sharing of pneumococcal strains in a rural African setting: proximate villages are more likely to share similar strains that are carried at multiple timepoints

The transmission dynamics of Streptococcus pneumoniae in sub-Saharan Africa are poorly understood due to a lack of adequate epidemiological and genomic data. Here we leverage a longitudinal cohort from 21 neighbouring villages in rural Africa to study how closely related strains of S. pneumoniae are shared among infants. We analysed 1074 pneumococcal genomes isolated from 102 infants from 21 villages. Strains were designated for unique serotype and sequence-type combinations, and we arbitrarily defined strain sharing where the pairwise genetic distance between strains could be accounted for by the mean within host intra-strain diversity. We used non-parametric statistical tests to assess the role of spatial distance and prolonged carriage on strain sharing using a logistic regression model. We recorded 458 carriage episodes including 318 (69.4 %) where the carried strain was shared with at least one other infant. The odds of strain sharing varied significantly across villages (χ²=47.5, df=21, P-value <0.001). Infants in close proximity to each other were more likely to be involved in strain sharing, but we also show a considerable amount of strain sharing across longer distances. Close geographic proximity (<5 km) between shared strains was associated with a significantly lower pairwise SNP distance compared to strains shared over longer distances (P-value <0.005). Sustained carriage of a shared strain among the infants was significantly more likely to occur if they resided in villages within a 5 km radius of each other (P-value <0.005, OR 3.7). Conversely, where both infants were transiently colonized by the shared strain, they were more likely to reside in villages separated by over 15 km (P-value <0.05, OR 1.5). PCV7 serotypes were rare (13.5 %) and were significantly less likely to be shared (P-value <0.001, OR −1.07). Strain sharing was more likely to occur over short geographical distances, especially where accompanied by sustained colonization. Our results show that strain sharing is a useful proxy for studying transmission dynamics in an under-sampled population with limited genomic data. This article contains data hosted by Microreact.

Negative frequency-dependent selection and asymmetrical transformation stabilise multi-strain bacterial population structures

Streptococcus pneumoniae can be divided into many strains, each a distinct set of isolates sharing similar core and accessory genomes, which co-circulate within the same hosts. Previous analyses suggested the short-term vaccine-associated dynamics of S. pneumoniae strains may be mediated through multi-locus negative frequency-dependent selection (NFDS), which maintains accessory loci at equilibrium frequencies. Long-term simulations demonstrated NFDS stabilised clonally-evolving multi-strain populations through preventing the loss of variation through drift, based on polymorphism frequencies, pairwise genetic distances and phylogenies. However, allowing symmetrical recombination between isolates evolving under multi-locus NFDS generated unstructured populations of diverse genotypes. Replication of the observed data improved when multi-locus NFDS was combined with recombination that was instead asymmetrical, favouring deletion of accessory loci over insertion. This combination separated populations into strains through outbreeding depression, resulting from recombinants with reduced accessory genomes having lower fitness than their parental genotypes. Although simplistic modelling of recombination likely limited these simulations' ability to maintain some properties of genomic data as accurately as those lacking recombination, the combination of asymmetrical recombination and multi-locus NFDS could restore multi-strain population structures from randomised initial populations. As many bacteria inhibit insertions into their chromosomes, this combination may commonly underlie the co-existence of strains within a niche.

Mathematical assessment of the impact of cohort vaccination on pneumococcal carriage and serotype replacement

Although pneumococcal vaccines are quite effective in reducing disease burden, factors such as imperfect vaccine efficacy and serotype replacement present an important challenge against realizing direct and herd protection benefits of the vaccines. In this study, a novel mathematical model is designed and used to describe the dynamics of two Streptococcus pneumoniae (SP) serotypes, in response to the introduction of a cohort vaccination program which targets one of the two serotypes. The model is fitted to a pediatric SP carriage prevalence data from Atlanta, GA. The model, which is rigorously analysed to investigate the existence and asymptotic stability properties of the associated equilibria (in addition to exploring conditions for competitive exclusion), is simulated to assess the impact of vaccination under different levels of serotype-specific competition and illustrate the phenomenon of serotype replacement. The calibrated model is used to forecast the carriage prevalence in the pediatric cohort over 30 years.

Incorporating Contact Network Uncertainty in Individual Level Models of Infectious Disease using Approximate Bayesian Computation

Infectious disease transmission between individuals in a heterogeneous population is often best modelled through a contact network. However, such contact network data are often unobserved. Such missing data can be accounted for in a Bayesian data augmented framework using Markov chain Monte Carlo (MCMC). Unfortunately, fitting models in such a framework can be highly computationally intensive. We investigate the fitting of network-based infectious disease models with completely unknown contact networks using approximate Bayesian computation population Monte Carlo (ABC-PMC) methods. This is done in the context of both simulated data, and data from the UK 2001 foot-and-mouth disease epidemic. We show that ABC-PMC is able to obtain reasonable approximations of the underlying infectious disease model with huge savings in computation time when compared to a full Bayesian MCMC analysis.

A Bayesian model of acquisition and clearance of bacterial colonization incorporating within-host variation

Bacterial populations that colonize a host can play important roles in host health, including serving as a reservoir that transmits to other hosts and from which invasive strains emerge, thus emphasizing the importance of understanding rates of acquisition and clearance of colonizing populations. Studies of colonization dynamics have been based on assessment of whether serial samples represent a single population or distinct colonization events. With the use of whole genome sequencing to determine genetic distance between isolates, a common solution to estimate acquisition and clearance rates has been to assume a fixed genetic distance threshold below which isolates are considered to represent the same strain. However, this approach is often inadequate to account for the diversity of the underlying within-host evolving population, the time intervals between consecutive measurements, and the uncertainty in the estimated acquisition and clearance rates. Here, we present a fully Bayesian model that provides probabilities of whether two strains should be considered the same, allowing us to determine bacterial clearance and acquisition from genomes sampled over time. Our method explicitly models the within-host variation using population genetic simulation, and the inference is done using a combination of Approximate Bayesian Computation (ABC) and Markov Chain Monte Carlo (MCMC). We validate the method with multiple carefully conducted simulations and demonstrate its use in practice by analyzing a collection of methicillin resistant Staphylococcus aureus (MRSA) isolates from a large recently completed longitudinal clinical study. An R-code implementation of the method is freely available at: https://github.com/mjarvenpaa/bacterial-colonization-model.

Pneumococcal quorum sensing drives an asymmetric owner-intruder competitive strategy during carriage via the competence regulon

Competition among microorganisms is a key determinant of successful host colonization and persistence. For Streptococcus pneumoniae, lower than predicted rates of co-colonizing strains suggest a competitive advantage for resident bacteria over newcomers. In light of evolutionary theory, we hypothesized that S. pneumoniae use owner-intruder asymmetries to settle contests, leading to the disproportionate success of the initial resident 'owner', regardless of the genetic identity of the 'intruder'. We investigated the determinants of within-host competitive success utilizing S. pneumoniae colonization of the upper respiratory tract of infant mice. Within 6 h, colonization by the resident inhibited colonization by an isogenic challenger. The competitive advantage of the resident was dependent on quorum sensing via the competence (Com) regulon and downstream choline binding protein D (CbpD) and on the competence-induced bacteriocins A and B (CibAB) implicated in fratricide. CbpD and CibAB are highly conserved across pneumococcal lineages, indicating evolutionary advantages for asymmetric competitive strategies within the species. Mathematical modelling supported a significant role for quorum sensing via the Com regulon in competition, even for strains with different competitive advantages. Our study suggests that asymmetric owner-intruder competitive strategies do not require complex cognition and are used by a major human pathogen to determine 'ownership' of human hosts.

Pneumococcal Vaccines: Host Interactions, Population Dynamics, and Design Principles

Streptococcus pneumoniae (the pneumococcus) is a nasopharyngeal commensal and respiratory pathogen. Most isolates express a capsule, the species-wide diversity of which has been immunologically classified into ∼100 serotypes. Capsule polysaccharides have been combined into multivalent vaccines widely used in adults, but the T cell independence of the antibody response means they are not protective in infants. Polysaccharide conjugate vaccines (PCVs) trigger a T cell–dependent response through attaching a carrier protein to capsular polysaccharides. The immune response stimulated by PCVs in infants inhibits carriage of vaccine serotypes (VTs), resulting in population-wide herd immunity. These were replaced in carriage by non-VTs. Nevertheless, PCVs drove reductions in infant pneumococcal disease, due to the lower mean invasiveness of the postvaccination bacterial population; age-varying serotype invasiveness resulted in a smaller reduction in adult disease. Alternative vaccines being tested in trials are designed to provide species-wide protection through stimulating innate and cellular immune responses, alongside antibodies to conserved antigens.

Geographic variation in pneumococcal vaccine efficacy estimated from dynamic modeling of epidemiological data post-PCV7

Although mean efficacy of multivalent pneumococcus vaccines has been intensively studied, variance in vaccine efficacy (VE) has been overlooked. Different net individual protection across settings can be driven by environmental conditions, local serotype and clonal composition, as well as by socio-demographic and genetic host factors. Understanding efficacy variation has implications for population-level effectiveness and other eco-evolutionary feedbacks. Here I show that realized VE can vary across epidemiological settings, by applying a multi-site-one-model approach to data post-vaccination. I analyse serotype prevalence dynamics following PCV7, in asymptomatic carriage in children attending day care in Portugal, Norway, France, Greece, Hungary and Hong-Kong. Model fitting to each dataset provides site-specific estimates for vaccine efficacy against acquisition, and pneumococcal transmission parameters. According to this model, variable serotype replacement across sites can be explained through variable PCV7 efficacy, ranging from 40% in Norway to 10% in Hong-Kong. While the details of how this effect is achieved remain to be determined, here I report three factors negatively associated with the VE readout, including initial prevalence of serotype 19F, daily mean temperature, and the Gini index. The study warrants more attention on local modulators of vaccine performance and calls for predictive frameworks within and across populations.

Likelihood-free inference via classification

Increasingly complex generative models are being used across disciplines as they allow for realistic characterization of data, but a common difficulty with them is the prohibitively large computational cost to evaluate the likelihood function and thus to perform likelihood-based statistical inference. A likelihood-free inference framework has emerged where the parameters are identified by finding values that yield simulated data resembling the observed data. While widely applicable, a major difficulty in this framework is how to measure the discrepancy between the simulated and observed data. Transforming the original problem into a problem of classifying the data into simulated versus observed, we find that classification accuracy can be used to assess the discrepancy. The complete arsenal of classification methods becomes thereby available for inference of intractable generative models. We validate our approach using theory and simulations for both point estimation and Bayesian inference, and demonstrate its use on real data by inferring an individual-based epidemiological model for bacterial infections in child care centers.

Carriage burden, multiple colonization and antibiotic pressure promote emergence of resistant vaccine escape pneumococci

Pneumococcal conjugate vaccines target the limited subset of the more than 90 known serotypes of Streptococcus pneumoniae responsible for the greatest burden of pneumococcal disease and antibiotic resistance. Following the introduction of these vaccines, serotypes not targeted were able to expand and resistance became more common within these types. Here we use a stochastic dynamic model of pediatric pneumococcal carriage to evaluate potential influences on the emergence of new resistant lineages following the introduction of a vaccine targeting more common resistant types. Antibiotic pressure was the strongest driver, with no emergence at low levels and universal emergence at high levels. At intermediate levels of antibiotic pressure, higher carriage burden and a greater degree of dual carriage promoted emergence. This may have implications for current plans to introduce childhood pneumococcal vaccination in several high-burden countries.

Two-phase importance sampling for inference about transmission trees

There has been growing interest in the statistics community to develop methods for inferring transmission pathways of infectious pathogens from molecular sequence data. For many datasets, the computational challenge lies in the huge dimension of the missing data. Here, we introduce an importance sampling scheme in which the transmission trees and phylogenies of pathogens are both sampled from reasonable importance distributions, alleviating the inference. Using this approach, arbitrary models of transmission could be considered, contrary to many earlier proposed methods. We illustrate the scheme by analysing transmissions of Streptococcus pneumoniae from household to household within a refugee camp, using data in which only a fraction of hosts is observed, but which is still rich enough to unravel the within-household transmission dynamics and pairs of households between whom transmission is plausible. We observe that while probability of direct transmission is low even for the most prominent cases of transmission, still those pairs of households are geographically much closer to each other than expected under random proximity.

Climate induces seasonality in pneumococcal transmission

Streptococcus pneumoniae is a significant human pathogen and a leading cause of infant mortality in developing countries. Considerable global variation in the pneumococcal carriage prevalence has been observed and the ecological factors contributing to it are not yet fully understood. We use data from a cohort of infants in Asia to study the effects of climatic conditions on both acquisition and clearance rates of the bacterium, finding significantly higher transmissibility during the cooler and drier months. Conversely, the length of a colonization period is unaffected by the season. Independent carriage data from studies conducted on the African and North American continents suggest similar effects of the climate on the prevalence of this bacterium, which further validates the obtained results. Further studies could be important to replicate the findings and explain the mechanistic role of cooler and dry air in the physiological response to nasopharyngeal acquisition of the pneumococcus.