A world without bacterial meningitis: how genomic epidemiology can inform vaccination strategy

Bacterial meningitis remains an important cause of global morbidity and mortality. Although effective vaccinations exist and are being increasingly used worldwide, bacterial diversity threatens their impact and the ultimate goal of eliminating the disease. Through genomic epidemiology, we can appreciate bacterial population structure and its consequences for transmission dynamics, virulence, antimicrobial resistance, and development of new vaccines. Here, we review what we have learned through genomic epidemiological studies, following the rapid implementation of whole genome sequencing that can help to optimise preventative strategies for bacterial meningitis.


Introduction
Bacterial meningitis describes infection of the subarachnoid space with bacterial pathogens, resulting in inflammation of the brain linings (meninges), a condition that causes significant morbidity and mortality worldwide. Bacteria reach the subarachnoid space through haematogenous or direct contiguous spread, where they replicate with resultant local meningeal inflammation and potential for involvement of the brain tissue ( Figure 1). In addition, where haematogenous spread of bacteria causes meningitis, persistence of bacteria in the blood, septicaemia, can rapidly develop into multi-organ failure and death.
Depending on age, geographic location, immune system function, and vaccine implementation, the incidence rates and causative organisms of bacterial meningitis differ 1,2 . In 2013, there were an estimated 303,500 deaths globally from meningitis, attributed to Streptococcus pneumoniae (n = 79,100), Neisseria meningitidis (n = 65,700), Haemophilus influenzae type b (Hib) (n = 64,400) and other agents (n = 94,200) 3 . Despite highly effective vaccination programmes against the major pathogens, disease persists. This review will discuss what we have learned through genomic epidemiological studies, from elucidating transmission networks to describing bacterial biodiversity, with the aim of improving the use of existing vaccines and novel vaccine development.

Global burden of bacterial meningitis
Bacterial meningitis in newborns in the first seven days of life is most commonly caused by group B streptococcus (Streptococcus agalactiae) and Escherichia coli through vertical transmission from the birth canal and perineal region. After the first week of life, cases are mainly nosocomial or acquired via horizontal transmission, and Listeria monocytogenes and S. pneumoniae also contribute to disease burden 5,6 . Vaccination against Hib, N. meningitidis, and S. pneumoniae has greatly altered the epidemiology of bacterial meningitis in older children and adults over the last three decades (Table 1). H. influenzae meningitis is now extremely rare in countries with high uptake of the conjugate polysaccharide (Hib) vaccine, but cases can occur in unvaccinated individuals or with non-b serotypes. S. pneumoniae and N. meningitidis cause most disease, with meningococcus predominating in older children and adolescents, and pneumococcus predominating in adults. Other causes include: L. monocytogenes in the elderly or immunocompromised; Staphylococcus aureus co-existent with endocarditis; and H. influenzae co-existent with otitis media or sinusitis 7,8 .

Bacterial diversity: carriage and immune selection
With the exception of L. monocytogenes, the bacteria principally responsible for causing meningitis are carried asymptomatically as members of a healthy microbiota. Group B streptococcus is found in the vaginal tract of up to 20% of women, E. coli is found universally in the gut, S. aureus on the skin, and S. pneumoniae, N. meningitidis, and H. influenzae in the naso/ oropharynx. There are likely to be interactions between the host immune system and the microbiota, although these are not fully understood, that result in the structured diversity observed in bacterial populations 9 . This diversity is found amongst microorganisms of the same species, manifested as distinctive lineages (organisms that share a common ancestor and therefore exhibit genetic similarity) which persist through time. Even within these lineages, bacteria vary genotypically (that is, in their genetic constitution) and phenotypically (that is, in their observable characteristics) with an extensive capacity to alter protein expression states through phase variation. Thus, to appreciate the degree and mechanisms by which these populations are structured, it is necessary to study genome-wide variation among representative bacterial isolates. Only by understanding the host bacterial population structure can we start to identify the strains (bacteria that have a similar genotype and phenotype) that cause disease.

Persistence of bacterial meningitis despite vaccination
The diversity of S. pneumoniae and N. meningitidis challenges the continued success of vaccines and the elimination of bacterial meningitis caused by these organisms. Both bacteria exhibit high rates of horizontal genetic transfer (HGT) and comprise distinct, non-overlapping genetic lineages with varying degrees of pathogenicity. Each genetic lineage, recognised by multilocus sequence typing (MLST) as groups of sequence types (STs) called clonal complexes (ccs), can exhibit a variety of polysaccharide capsular types and undergo capsule switching ( Figure 2). Until 2013, all licenced vaccines for the prevention of bacterial meningitis pathogens were based on polysaccharide capsular antigens, key virulence factors as both acapsulate streptococci and meningococci very rarely cause invasive disease. Prevention of disease by capsular group was beneficial but allowed bacteria from hyperinvasive lineages that switched capsule to persist through carriage and ongoing transmission between hosts.
The first pneumococcal polysaccharide conjugate vaccine included seven capsular serotypes (PCV7) and subsequently increased to PCV10 and PCV13 with further iterations in development (Table 1). This was based on the serotypes most frequently causing disease, but some capsular types were antigenically related, resulting in a degree of cross-protection. With more than 90 serotypes identified worldwide, the development of a universal vaccine remains challenging. In contrast, for meningococci there are only 12 recognised capsular groups, of which six serogroups cause almost all disease (Table 1). Conjugate vaccines against serogroups A, C, W, and Y are available but not universally used 10 . Until 2013, there was no licensed vaccine against serogroup B, a major cause of meningitis in industrialised countries. Hence, non-vaccine types continue to be carried in the host nasopharynx and transmitted, potentially causing disease in susceptible populations. Furthermore, through the extensive HGT in these pathogens, newly emerging hyperinvasive genotypes can arise. The introduction of a novel antigenic combination can result in epidemic or hyperepidemic disease.

Genomic epidemiology
Genomic epidemiology aims to achieve "systematic investigation of how natural genomic variation affects the clinical outcome of disease" 11 . The utility of this methodology in the prevention of bacterial meningitis lies in understanding transmission networks, population structure of bacterial pathogens, and epidemiology. In combination, this can inform vaccine development, implementation, and post-vaccine surveillance ( Figure 3) 12 .
Next-generation sequencing technology Genomic epidemiological studies are increasingly available because of the generation of high-quality microbial genomes with benchtop sequencers, including Illumina, Ion Torrent, and Pacific Biosciences platforms. Portable sequencing devices, such as the Oxford Nanopore MinION, have been used in the field for Ebola and Zika virus epidemics, although at the time of writing they still had higher error rates than other next-generation sequencing technologies, of which the Illumina platform was predominant 13,14 . The multiplicity of platforms provides flexibility in the face of diverse scale, research or clinical questions, and settings. The cost of sequencing genomes fell rapidly since its commercial inception, but the challenge remains in developing bioinformatics techniques for systematic analyses, which are inexpensive, standardised, highly reproducible worldwide, and easily accessible to microbiologists, epidemiologists, and clinicians alike 15 .

Bioinformatics approaches
The most widely used sequence-based method for typing bacteria is MLST, which uses housekeeping genes to catalogue diversity and has been successful because of its highly discriminatory, portable, and unambiguous results 16 . Genetic lineages within bacterial populations are still most frequently defined by MLST, Prevalence varies by region, with 10 serotypes responsible for 62% of disease.

Carriage
Carried in the nasopharynx as part of the normal commensal. Prior to vaccination, Hib was carried predominantly by young children. Carried in the nasopharynx as part of the normal commensal.
Varies by age and geographical distribution, but ranges 5-60%. Carried in the nasopharynx as part of the normal commensal.
Varies by age and geographical distribution, but ranges from 1-40%.
No vaccines are available against non-b strains. Plain polysaccharide vaccines against A, C, W, Y.  Initially, microbial isolates undergo whole genome sequencing (WGS). WGS can be assembled de novo or by mapping to a reference. Bioinformatics platforms enable the uploaded WGS to be annotated and allow users (microbiologists, bioinformaticians, public health officials, and clinicians) to analyse the genes of interest by visualising phylogenetic relationships and associating these with appropriate and relevant meta data. The example of outbreak tracing is used here but this can be extrapolated to many areas of health and disease. Figure reproduced unchanged with permission 12 .
even when whole genome sequencing (WGS) data are available. By assigning unique alleles at each locus, irrespective of whether the alleles have arisen by individual mutations or HGT, one can systematically index genetic diversity, regardless of the rates of vertical or horizontal transmission 16 . However, the resolution attainable by seven MLST loci is limited by the small number of loci indexed (Figure 2a). This can be overcome by using the hierarchical and scalable gene-by-gene approach with assembled WGS data 20 . By following the principles of MLST but employing more loci, for example, in ribosomal MLST, core genome or whole genome MLST, one can successively increase resolution to identify genetic diversity (Figure 2b). Typing at other loci allows characterisation of potential vaccine components or virulence factors, including capsule loci, outer membrane proteins, and antibiotic resistance-encoding genes. Alternative approaches include single-nucleotide polymorphism (SNP) typing, where short-read data are mapped onto a reference sequence, after which the SNPs can be identified (SNP calling). The SNPs are collated to reconstruct a phylogeny or into an SNP address, identifying closely related isolate clusters within a given isolate collection, which can be interpreted with additional epidemiological data. This method can be performed rapidly, easily, and sensitively but is dependent on specialist software, reference genomes and sequencing platforms, which can limit portability among sites 21 .  Figure 4) 27,28 . N. lactamica was carried at the highest rate of 14.1% by 1-to 4-year-olds, and N. meningitidis was carried at the highest rate of 5.2% by 5-to 14-year-olds 28 . Furthermore, there was a mean 4.7-year delay in acquisition of N. meningitidis following N. lactamica carriage 29 . Although the underlying mechanisms are yet to be elucidated, this observation has implications for intervention strategies. Since 2010, the PsA-TT vaccine has been progressively implemented across the African meningitis belt, and there has been a dramatic reduction in hyperepidemic meningococcal disease and carriage in vaccinated and unvaccinated individuals 30 .
Manipulating the nasopharyngeal niche. The age-specific rate of meningococcal carriage and invasive disease is inversely proportional to the rate of colonisation with harmless N. lactamica, and alternative prophylactic strategies that exploit this observation have been proposed 31-33 . For example, nasal inoculation with live N. lactamica has been investigated in UK university students. New colonisation with N. lactamica occurred from two weeks after inoculation, and carriage of meningococci fell from 24.2% (n = 36/149) to 14.7% (n = 21/143) (P = 0.006) in those individuals carrying N. lactamica 34 . This effect may be due to displacement of resident N. meningitidis soon after colonisation with N. lactamica, or in those not colonised with either Neisseria spp. at baseline, the colonisation with N. lactamica might inhibit meningococcal acquisition. There remained a group of study participants persistently colonised with N. meningitidis despite N. lactamica challenge, suggesting that displacement can be inhibited. The serogroup distribution was not characterised, so it is not known whether this effect was related to all serogroups; however, the effect was seen across ccs 34 .

Extent of capsular group B vaccine coverage
Estimating protein-based 'serogroup B substitute' vaccine coverage. The serogroup B polysaccharide capsule is poorly immunogenic and shows structural similarity to human tissue, raising safety concerns. Protein-based 'serogroup B substitute' vaccines, including 4CMenB (Bexsero ® , GlaxoSmithKline) and bivalent rLP2086 (Trumenba ® , Pfizer), were developed to address this issue 35,36 . In September 2015, 4CMenB vaccine Bexsero ® was introduced for infants in the UK immunisation schedule at 2, 4, and 12 months of age. This vaccine contains multiple subcapsular proteins, including factor H-binding protein (fHbp), Neisserial heparin binding antigen (NHBA), Neisseria adhesin A (NadA), and an outer membrane vesicle (OMV) containing Porin A (PorA). Efficacy of these vaccines must be considered in terms of host immunogenicity and strain coverage. Owing to the practical constraints of performing multiple serum bactericidal assays (SBAs), the accepted correlate of protection for meningococcal vaccines, alternative assays were devised to estimate coverage. Meningococcal strain coverage estimates for England and Wales were 73% (95% confidence interval [CI] 57-87%) using the Meningococcal Antigen Typing System (MATS) 37 . MATS assesses potential immunological cross-reaction of meningococcal isolates but (i) can be performed by specialist laboratories only, (ii) is expensive and time-and labour-intensive, and (iii) relies on pooled infant serum. Genomic analysis can be used to measure vaccine antigen prevalence using Bexsero ® Antigen Sequence Typing (  and epiglottitis/supraglottitis in addition to septicaemia and meningitis [49][50][51] . In response to this outbreak, conjugate meningococcal ACWY vaccine was introduced into the UK immunisation schedule in August 2015 for adolescents, historically those with the highest carriage 49,52 . A modest reduction in carriage of serogroups C, W, and Y-from 36.2 to 33% (CI 15.6-51.7)was seen at least two months after conjugate ACWY vaccination amongst UK university students 45 . Although this effect on carriage is relatively limited, this may impact on disease incidence due to reduction in acquisition rates.

Conclusions
Genomic epidemiology of disease-causing bacteria has farreaching implications for promoting human health and preventing disease. The ability to perform such studies has been accelerated with increasing ease, rapidity, and affordability of WGS. This simultaneously presents challenges to develop methods for distributing and analysing these data for non-specialists.
In the case of bacterial meningitis and related diseases, S. pneumoniae and N. meningitidis have been extensively studied by WGS, and studies of the meningococcus have increased our understanding of carriage, transmission, interactions with commensal Neisseria and the distribution of vaccine antigens in national surveillance and emergent organisms. This information has helped to shape vaccination strategies worldwide, ultimately reducing the burden of this devastating disease.
Competing interests M.C.J.M has received grants and personal fees from vaccine companies, including GlaxoSmithKline and Novartis, outside the scope of the submitted work. CMCR declares that she has no competing interests.

Grant information
The authors declare that this work was supported by the Wellcome Trust (grant 109031/Z15/Z to CMCR and grant 104992/Z/14/Z to MCJM).
The funders had no role in study design, data collection and analysis, decision to publish, or preparation of the manuscript.