Introduction

Streptococcus pneumoniae is a major pathogen causing a wide spectrum of infections ranging from acute otitis media and lower respiratory tract infection to severe invasive diseases, such as septicemia and meningitis. Recent estimates of childhood mortality due to S. pneumoniae range from 700,000 to 1 million deaths yearly worldwide [1].

Among the >92 capsular serotypes of S. pneumoniae identified thus far, seven (4, 6B, 9V, 14, 18C, 19F and 23F) were responsible for 65 to 85% of invasive pneumococcal diseases (IPD) in children worldwide in the year 2000 [2]. A conjugate pneumococcal vaccine containing these serotypes (PCV7, Prevenar®, Pfizer) was licensed in 2000. Introduction of this vaccine rapidly led to a dramatic decrease in the incidence of IPD from an average of 25.1 cases/100,000 persons in 1999 to 12.6 cases/100,000 persons in 2004 in the USA [3]. Large post-licensure vaccine surveillance studies have raised concerns about an increased incidence of invasive disease due to non-vaccine serotypes, especially serotypes 3, 7F, 19A, 22 and 33 [4,5], a phenomenon known as serotype replacement. A 13-valent pneumococcal vaccine including the PCV7 serotypes as well as serotypes 1, 3, 5, 6A, 7F and 19A was licensed in 2010 (PCV13, Prevnar 13®, Pfizer). This new vaccine and others (Synflorix®, GSK) have the potential to further decrease IPD rates and prevent resurgence of pre-PCV7 IPD rates. However, ongoing serotype replacement is still of concern. Consequently, serotyping strains remains of paramount importance to both assess the effectiveness of current vaccines and to closely monitor the emergence of non-vaccine strains.

S. pneumoniae serotyping was developed at the beginning of the 20th century using panels of specific anti-sera produced in animals and directed against polysaccharides of the pneumococcal capsule. These agglutination assays are complex, costly and require highly skilled personnel [6]. Traditional agglutination assays cannot be performed directly on clinical samples since growth of S. pneumoniae on culture media is required. However, culture is often negative if patients received antibiotics before sampling of blood, cerebrospinal fluid (CSF) or other biological fluids, and in children with pneumococcal pneumonia in whom blood culture tends to be negative but PCR-positive for S. pneumoniae [7,8]. In recent years, immunological assays based on ELISA or latex-bead agglutination have been shown to work directly on clinical specimens [9–11]. For example, Sheppard and collaborators show that a multiplex immunoassay using xMAP beads can identify 14 S. pneumoniae serotypes directly from CSF [12].

Genes coding for the bacterial capsule are part of a complex gene structure where multiple and subtle polymorphisms in the capsular operon result in different serotypes. PCR-based serotyping using primers that amplify serotype-specific sequences are considered the method of choice for non-cultivable pneumococci serotyping [13,14]. A sequential multiplex PCR approach was initially proposed by Pai and collaborators [15] to identify 29 serotypes and has later been optimized for geographical locations such as South-Saharan Africa [16], Latin America [17] and Bangladesh [18]. Siira and collaborators complemented sequential multiplex PCR reactions by Quellung testing to increase confidence for 13 serogroups of the 42 serotypes/serogroups detected by PCR [6]. Selva and collaborators proposed a method that uses 3 multiplex PCR followed by fragment analysis and automated capillary electrophoresis to identify 40 serotypes or groups of serotypes [19]. Coskun-Ari and collaborators published a single multiplex that allows the identification of the 13 serotypes included in PCV13 [20]. Ahn and collaborators were able to identify 35 serotypes using as series of 8 multiplex PCR reactions [21]. Pimenta and collaborators described a series of multiplex real-time PCR that allowed the identification of 21 serogroups or serotypes, the order of which can be optimized for different regions [22]. PCR-based serotyping of S. pneumoniae has also been coupled with reverse line blot hybridization [23], restriction fragment length polymorphism [24,25] or amplicon sequencing [26,27]. This latter method, also termed sequetyping, relies on the sequencing of capsular operon genes such as cpsB in order to serotype samples by comparing the sequence obtained to those published by Bentley and collaborators [28]. Although these methods are cost-effective and specific, some serotypes are difficult to discriminate. As an example, a single nucleotide polymorphism in the wciP gene results in two distinct capsular serotypes, namely 6A and 6B, emphasizing the inability of PCR-based methods to realistically be used to differentiate serotypes from closely related serogroups [29]. Methods using PCR fragment length [30] or pyrosequencing [29] have been proposed to solve this problem, but are specific to serogroup 6. Immunological methods remain the most reliable approaches to discriminate closely related serotypes such as 6A and 6B. Microarray-based methods have also been published. Scott and coworkers used a microarray targeting capsular operon genes to serotype pneumococcal isolates [31]. Tomita and collaborators used a targeted microarray to identify 23 serotypes [32]. Both studies used conventional glass microarray supports, which require highly skilled personnel and are not amenable to clinical laboratories or to conduct large studies.

Immunological and molecular serotyping methods often include serotypes based on their prevalence, an approach that can bias serotyping studies. Moreover, most DNA-based methods allow the identification of a limited number of serotypes or serogroups. Herein, we suggest an innovative serotyping approach that, instead of detecting one target per serotype, relies on the genotyping of 12 positions located in the capsular operon to identify pneumococcal serotypes. This assay constitutes a proof-of-concept that typing a limited number of polymorphisms could allow a fast and efficient serotyping of S. pneumoniae. To facilitate large-scale clinical studies, the assay was adapted for the AutoGenomics INFINITI system and it could be readily ported to another automated platform. A flowchart exemplifying the steps of the assay and data analysis are shown in Figure 1 and Figure 2, respectively.

Download:

• PPT
  PowerPoint slide
• PNG
  larger image
• TIFF
  original image

Figure 1. Schematic representation of experimental procedure.

Multiplex PCR amplifies up to 12 genes and is performed outside of the INFINITI system. Amplification products are loaded in the INFINITI system at the primer extension step. Primer extension oligonucleotides include a tag sequence that will hybridize to the microarray and a specific detection sequence that allows for primer extension of S. pneumoniae genes. Fluorescent nucleotides are incorporated during primer extension. Once the reaction is complete, the INFINITI system automatically transfers the labeled products to the microarray for hybridization. Tag sequences hybridize to anti-tags located on the microarray. Microarrays are then washed, dried and loaded into the integrated confocal scanner where fluorescence is measured. The report generated by the instruments is analyzed off-line using the Pneumotyper software.

https://doi.org/10.1371/journal.pone.0076197.g001

Download:

• PPT
  PowerPoint slide
• PNG
  larger image
• TIFF
  original image

Figure 2. Algorithm for results analysis.

Samples were assessed for the presence of S. pneumoniae by detecting the pneumolysin and the autolysin genes. If S. pneumoniae was detected in the sample, the serotyping probes were analyzed in order to identify the serotype.

https://doi.org/10.1371/journal.pone.0076197.g002

Methods

Results

Design of the assay

A machine learning approach was used to identify genes and single nucleotide variations within the capsular locus that, in combination, allow for the serotyping of S. pneumoniae. Coding sequences (CDS) were extracted from 90 capsular polysaccharide biosynthetic clusters of S. pneumoniae [28] and CDS were compared in order to form subgroups. Initially, the optimal combination of CDS necessary to identify the 13 vaccine serotypes of interest was determined using the Id3 decision tree algorithm implemented in the Weka software [33]. Selected CDS were mined to find polymorphisms specific to different capsular serotypes. Selected polymorphisms were added to the decision tree optimization parameters and targets were again optimized with Id3 to identify additional serotypes. Several iterations of this process were performed to increase the number of serotypes detected by the assay using a minimal number of targets. Detection primers were designed to identify these polymorphisms, which, in combination, permit serotype identification. Figure 3 provides an example of the nucleotides typed for serotype 19A.

Download:

• PPT
  PowerPoint slide
• PNG
  larger image
• TIFF
  original image

Figure 3. Capsular operon of S. pneumoniae serotype 19A.

Expected amplicons are shown under the sequence, with genotyped positions marked with vertical lines. The expected genotype for serotype 19A is in large characters while the other possible genotypes are shown in smaller type. Positions are in nucleotides.

https://doi.org/10.1371/journal.pone.0076197.g003

Discussion

Accurate serotype determination of Streptococcus pneumoniae is of paramount importance to assess the impacts of pneumococcal vaccines and to monitor emergent serotypes. In order to simplify S. pneumoniae serotyping, we created a test that relies on a combination of genomic properties to identify serotypes. In this study, we validated the identification of 46 serotypes, 22 of which were precisely identified and 24 assigned to a subgroup of probable serotypes (Table 3). This is done by genotyping only 12 positions in the capsular operon. Serotypes included in the 13-valent vaccine were correctly assigned to serotypes, although serotypes 6A and 6B were sometimes not resolved. Bioinformatics analyses of available capsular operon sequences suggest that several other serotypes could be identified using this assay, but they have not been validated in the current study. In such cases, and for vaccinal serotypes 6A, 6B and 19A, it would be appropriate to confirm serotype determination using another method, such as the Quellung assay. Validation could also be performed to increase the precision of serotyping when the assay provides a subgroup of serotypes or unvalidated serotypes. We also tested the assay on clinical specimens consisting in 56 nasopharyngeal aspirates from patients with pneumonia and 43 cerebrospinal fluid samples from patient with pneumococcal meningitis, demonstrating that the assay do not require culture prior to serotyping.

The assay targets 12 polymorphic positions located in genes of the capsular operon. It was designed with sufficient redundancy to allow for the occasional false negative probe so that, in most cases, missing probes will not prevent serotyping. If several variants are possible at a single position in the known S. pneumoniae population, all polymorphisms are accounted for in the design. This provides redundancy that may counterbalance missing results, making the assay more robust. This is especially useful in specimens with low sample concentration or in specimens for which capsular gene sequences may differ from reference sequences. In some cases, a precise identification of serotypes is theoretically possible, but missing signal for some polymorphisms prevent precise serotype identification. Thus, the results for these samples are a group of serotypes. This could be caused by sequence variations in target genes and could be resolved by testing the samples using another method.

The assay was validated on 166 cultures of Streptococcus pneumoniae covering 63 different serotypes. Fourteen samples were not identified as the appropriate serotype but were correctly assigned as members of the correct serogroup. Six samples of serotype 6A were identified as serotype 6B and one serotype 6B was identified as 6A/6C. It is implicit in the literature that serotypes 6A and 6B are difficult to discriminate using DNA-based assays. Indeed, the vast majority of previously published nucleic acid serotyping methods do not discriminate serotype 6A from 6B [6,14–22,27,32]. Moreover, 6A pneumococci with a wzh sequence previously observed in 6B serotypes have been reported in the Netherlands, highlighting the genetic diversity of this serogroup [26]. Assay designs based on phylogeny can provide relevant information, but can prove unreliable in some cases, such as in distinguishing between serotypes 6A and 6B. Designs based on functional polymorphisms, such as functional wciP genetic differences, could be more robust and reliable for these serotypes [29]. Serotypes 19B and 19C have been misidentified as serotype 19A. This is unexpected since the profile observed for these samples is not concordant with the published sequences for these serotypes [28]. It is of note that most molecular assays do not identify serotypes 19B and 19C, and that only a handful of molecular assays have been validated with serotypes 19B or 19C [15,22,32]. Variant 19F capsular genes have also been reported, leading to false positive 19A identification using PCR [37]. This highlights one of the challenges involved in validating S. pneumoniae serotyping assays, since it is difficult to obtain several representative strains of all serotypes. Using our assay, serotypes 24F was serotyped as 24B and 28F as 28A. Similar results were obtained with sequetyping [27]. The other misidentified samples were of low prevalence serotypes. Discordant samples outside of the serogroup were of serotypes 21, 41F, 42, 44, 45. Negative serotyping was obtained for 18 samples, 8 of which would have been correctly serotyped if their score had not been under the threshold of 0.4. They include 2 samples of serotype 3, and one sample for each of the following serotypes : 6A, 6B, 13, 14, 15A and 19F. These false negatives could be due to sequence variations that affect either PCR amplification, leading to low signal for some genotyping probes, or to mutations affecting the nucleotides that are genotyped by the assay, leading to missing signal or genotyping that do not conform with the information available in the serotyping database. This highlights the main problems that can be observed when a limited number of reference sequences are available to design assays.

In a recent study, Scott and collaborators used a glass-based microarray to detect the presence or absence of capsular genes, the combination of which allowed the identification of serotypes [31]. Instead of targeting a large number of capsular genes, the method described herein relies on the genotyping of selected nucleotides to identify serotypes. The use of a limited number of targets allowed the development of a multiplex PCR, which improves the limit of detection of the assay and eliminate the need for culture.

An advantage of this assay is that raw data from experiments can be reanalyzed upon the addition of new entries to the serotyping database or modifications to the identification software. Notably, new DNA sequences for which the serotype is known can help validate the assay and improve its robustness. The sequencing of additional S. pneumoniae capsular operons for relevant serotypes could improve the specificity of this assay. By testing the assay on several S. pneumoniae strains of the same serotype, we observed divergences between the expected genotype based on already determined capsular sequences and the actual results. For example, serotypes 19B and 19C produced the same genotyping profile as 19A, while the previously determined sequences hinted to a different profile. The same was observed with 6A compared to 6B. This suggests that actual strain-to-strain sequence variability within a single serotype could impact on the results of not only this assay, but on other molecular assays too. Thus, the sequencing of more isolates of each serotype would vastly improve serotyping assays that rely on genome sequences. In the future, new serotypes could be added to the analysis software provided that their capsular genes have been sequenced. The bioinformatics approach used to create this assay could also be applied to a larger sequence dataset, which should include the genome of several strains of each serotype, in order to create an improved molecular serotyping assay. A second multiplex PCR could be devised to genotype supplementary positions and thus improve the specificity and serotype coverage of the assay. In the end, as with other serotyping assays based on PCR and genotyping, immunological methods, such as the Quellung assay, may be required to confirm serotypes [6,37].

To facilitate the use of this assay in a clinical setting and to perform large scale epidemiological studies, we adapted the assay for use in the AutoGenomics INFINITI analyzer, an automated molecular diagnostic system that both reduces the hands-on time and the number of steps required to perform the assay. Twenty-four samples can be processed simultaneously with an overall processing time of 16 hours, with less than one hour of hands-on time. This system has been successfully used for the diagnostic of multiple respiratory viruses [38]. The serotyping assay could be adapted for additional platforms, such as the Luminex (Millipore), the BeadExpress (Illumina), or amplicon sequencing.

As exemplified by our Cameroon study [35] and several others [39], the serotypes circulating in developing countries are different from those observed in developed countries. This stresses the need for appropriate monitoring of vaccinal strains for surveillance. In order to properly describe the serotypes found in a country where no previous serotype prevalence studies have been performed, a test that allows the identification of a great number of serotypes would be beneficial. An assay such as the one described herein constitutes a reliable tool for a first screening of the serotypes circulating in unstudied populations. An approach based on polymorphism detection instead of serotype-by-serotype targets would allow a more versatile first screening of previously unstudied regions, where a wide coverage of serotypes is required.

Conclusions

This microarray-based assay uses a single test to identify most S. pneumoniae serotypes or serogroups. It is therefore useful for many types of infections, sample sources and for different countries. The current embodiment of the assay constitutes at this time a reasonable compromise between complexity and yield. Basically, it is a proof of concept that S. pneumoniae serotype identification can be performed by typing a limited combination of carefully selected polymorphic positions. Indeed, 22 serotypes can be precisely identified and 24 can be assigned to a subgroup of serotypes by genotyping only 12 nucleotides of the capsular operon. While the assay could be improved by the addition of new targets, the concept behind the assay could be used to devise methods to solve problems of similar complexity.

Acknowledgments

The authors wish to thank Maurice Boissinot and Marie-Claude Hélie, for providing access to the Centre de recherche en Infectiologie de l’Université Laval strain collection and preparing bacterial genomic DNA, and Kathrin Mühlemann from the Swiss National Reference Laboratory for S. pneumoniae, for providing multiple serotyped strains of S. pneumoniae.