Whole genome sequencing of Streptococcus pneumoniae: development, evaluation and verification of targets for serogroup and serotype prediction using an automated pipeline

Isolate selection

Reference strains (n = 91) for all serotypes (excluding 6D) were acquired from Statens Serum Institut. Reference strain for 6D was kindly provided by The National Institute for Health and Welfare (THL), Finland. A total of 926 clinical isolates were selected from the archives of the Public Health England (PHE) National Reference Lab as a test cohort; for serotypes found to belong to a genogroup, at least 10 isolates were selected where available. Post genomic-sequence data cleansing (to remove repeat isolates from the same patient, mixed cultures, other species and MLST partial profiles or failures) resulted in 871 isolates (Development Set in Table 1). In addition, 2079 prospective or research-related isolates were sequenced as part of the UK validation cohort. This cohort covers 72 of the commonly circulating serotypes (including all vaccine serotypes), and includes prospective isolates received by PHE during 2015, isolates selected as part of research projects and epidemiological investigations (15A (n = 196) and 19A (n = 249), respectively) and archived isolates for rarer serotypes. Post genomic-sequence data cleansing of this dataset resulted in a total of 2065 isolates (Validation Set in Table 1). All isolates were serotyped on receipt as part of the PHE enhanced surveillance programme using slide agglutination with Statens Serum Institut typing sera.

Genomic data for non-UK isolates were obtained from Streptococcus pneumoniae isolate database hosted in BIGSdb (http://pubmlst.org/software/database/bigsdb/) (Jolley & Maiden, 2010) and the European Nucleotide Archive (ENA; http://www.ebi.ac.uk/ena). Specifically, three collections were used; a set of 2531 isolates from Thailand initially described by Chewapreecha et al. (2014), an Icelandic panel of 252 serogroup 6 isolates described in Van Tonder et al. (2015) and a USA panel of 181 invasive isolates available in ENA as study SRP059723.

DNA extraction and sequencing

Isolates were grown overnight on horse blood agar (PHE Media Services) with 5% CO₂. DNA was extracted from an entire plate of growth for each isolate using the QIAsymphony SP automated instrument (Qiagen) and QIAsymphony DSP DNA Mini Kit, using the manufacturer’s recommended tissue extraction protocol for Gram negative bacteria (including a 1 h pre-incubation with proteinase K in ATL buffer and RNAse A treatment). DNA concentrations were measured using the Quant-iT dsDNA Broad-Range Assay Kit (Life Technologies, Paisley, UK) and GloMax^® 96 Microplate Luminometer (Promega, Southampton, UK). DNA was sent for whole genome sequencing (WGS) by Illumina sequencing using the PHE Genomic Services and Development Unit (Colindale, London, UK) (Dallman et al., 2014). Illumina Nextera DNA libraries were constructed and sequenced using the Illumina HiSeq 2500.

Casava 1.8.2 (Illumina inc., San Diego, CA,USA) was used to deplex the samples and FASTQ reads were processed with Trimmomatic (Bolger, Lohse & Usadel, 2014) to remove bases from the trailing end that fall below a PHRED score of 30. K-mer identification software (https://github.com/phe-bioinformatics/kmerid) was used to compare the sequence reads with a panel of curated NCBI Refseq genomes to identify the species. A sample of k-mers (DNA sequences of length k) in the sequence data are compared against the k-mers of 1769 reference genomes representing 59 pathogenic genera obtained from RefSeq. The closest percentage match is identified, and provides initial confirmation of the species. This step also identifies samples containing more than one species of bacteria (i.e., mixed cultures) and any bacteria misidentified as Streptococcus pneumoniae by the sending laboratory. Further analysis continued only if S. pneumoniae was identified. FASTQ reads from all sequences in this study were submitted to ENA using the ena_submission tool (https://github.com/phe-bioinformatics/ena_submission) and can be found at the PHE Pathogens BioProject PRJEB14267 at ENA (http://www.ebi.ac.uk/ena/data/view/PRJEB14267; Tables S1–S2).

Polysaccharide capsule operon locus sequence analysis

GenBank and FASTA files for the capsular locus sequences for the 90 serotypes published by Bentley et al. (2006) were retrieved from NCBI (http://www.ncbi.nlm.nih.gov/) using accession numbers CR931632 –CR931722 (Bentley et al., 2006). For serotypes 6C and 6D accession numbers JF911515.1 and HV580364.1 were used, respectively. Capsular locus sequences for 6E and 23B1, a new 23B genetic subtype identified during the course of this project, were derived in-house using the assembly-based approach described below. Isolates for both types (6E n = 12; 23B1 n = 27) were assembled using SPAdes (Bankevich et al., 2012) and the capsular locus sequences were extracted using the sequence for 6B and 23B, respectively. The extracted sequences were aligned using ClustalW (Thompson, Higgins & Gibson, 1994) and the alignment was visualized with MEGA6 (Tamura et al., 2013). Based on the alignment, consensus sequences were derived for each type and submitted to ENA as assemblies under PHE Pathogens BioProject PRJEB14267 with accession numbers LT594598.1 and LT594599.1 for 6E and 23B1, respectively. Using the GenBank annotation, gene sequences were retrieved from all capsular types. BLAST+ (version 2.2.27) (Camacho et al., 2009) alignment was then used to detect gene-level differences within each genogroup.

Capsular typing tool implementation

PneumoCaT (Pneumococcal Capsule Typing; https://github.com/phe-bioinformatics/PneumoCaT), written in Python (version 2.7.6), utilises a two-step method to assign capsular type (Fig. 2). In the first step, reads from each readset are mapped to a multi-fasta file containing 94 capsular locus sequences (92 serologically distinct serotypes + 2 molecular types/subtypes –6E and 23B1) using bowtie2 (version 2.1.0; following options used: –fr –minins 300 –maxins 1100 -k 99999 -D 20 -R 3 -N 0 -L 20 -I S,1,0.50) (Langmead & Salzberg, 2012). Thresholds of coverage of >90% of the length of the sequence, minimum depth of 5 reads per bp and a mean depth of >20 reads over the entire length of the sequence are implemented during this step. This step is considered successful if one or more capsular locus sequences are returned.

If a match to a single capsular locus is returned and the capsular type predicted does not belong to one of our defined genogroups (Table 2), then the software terminates here and reports this as the predicted capsular type (Fig. 3). If the match belongs to a genogroup or more than one locus is matched then the software moves to the second step; a variant-based approach. If multiple serotypes are matched in step 1 and they do not correspond to the same genogroup then a ‘Mixed sample’ flag is called. Finally, if no serotypes are called due to low coverage (<90%) then a ‘Failed’ flag is called.

The capsular types matched in step 1 (genogroup) are fed into step 2 and the relevant information for this specific genogroup is retrieved, if genogroup information is available. This information is organised in a multi-fasta reference file, with complete gene sequences for all genes that display genetic diversity within the selected genogroup, and a database containing all sequence variants pertaining to the same genogroup. Paired reads are then mapped to the loaded reference file using bowtie2 and the tool tests for the variants loaded from the capsular type variant (CTV) database. Specific functions were designed to detect the different types of variants, such as presence/absence of genes, different alleles, single nucleotide polymorphisms (SNPs) and loss-of-function mutations. Base calling and insertion/deletion detection were made with SAMtools mpileup (version 0.1.19; following options used: -B -A) (Li et al., 2009), requiring a minimum-read depth of 5 and using reads with minimum base mapping quality of 30. These base calls were only included in the downstream analysis if they constituted >80% of read depth (ie, a nonmixed base call). A capsular type is assigned if a 100% match to all variants tested is achieved, else a ‘Failed’ flag is returned. In the case of serogroups 24 and 32, where no distinguishing variants are available, the software returns the serogroup.

Phylogenetic tree generation

Reads were mapped to selected reference sequence using bwa (version 0.7.12) (Li & Durbin, 2009). Variants were called using GATK 2.6.5 (McKenna et al., 2010). Variants were then parsed to retain high quality SNPs based on the following conditions: depth of coverage (DP) ≥5, AD ratio (ratio between variant base and alternative bases) ≥0.8, Mapping Quality (MQ) ≥30, ratio of reads with MQ0 to total number of reads ≤0.05. All positions that fulfilled the filtering criteria in >0.9 of the samples were joined to produce a multiple fasta format file where the sequence for each strain consists of the concatenated variants. This file was used as an input to generate a maximum likelihood (ML) tree using RAxML (Stamatakis, 2014) with the following parameters –m (substitutionModel) ASC_GTRCAT –asc_corr=lewis (Paul Lewis’s model correction for ascertainment bias when analysing variant-only data) –b (bootstrapRandomNumberSeed) 12345 -# (numberOfRuns) 1000.

Assembly-based sequence analysis

Genomic reads were assembled using SPAdes (version 2.5.1) de novo assembly software (Bankevich et al., 2012) with the following parameters ‘spades.py –careful −1 strain.1.fastq.gz −2 strain.2.fastq −t 4 −k21, 33, 55, 77, 85, 93′. The resulting contigs.fasta file was converted into a BLAST database using blast+ (Camacho et al., 2009) (version 2.2.27) and queried using selected query sequence (i.e., gene or capsular operon sequences).

Genogroups

Genetic clustering of S. pneumoniae strains based on the capsular locus sequence was previously described by Mavroidi et al. (2007). Based on their analysis, 88 pneumococcal serotypes were grouped in eight major clusters and 20 subclusters. All but six serogroups fell within the same genetic subcluster as previously described (Mavroidi et al., 2007) and 9 of the subclusters contained serotypes from different serogroups. In this study, we define genogroups as groups of serotypes with high (>90%) sequence identity within the capsular locus. Based on these criteria, 21 genogroups were identified by comparing published capsular locus sequences (Bentley et al., 2006). All genogroups contain serotypes grouped within the same genetic subcluster as defined by Mavroidi et al. (2007) (Table 2); 11 of the genogroups correspond to serogroups as defined by the SSI sera, six of which also form genetic subclusters described by Mavroidi et al. (2007). Of the remaining 10 genogroups, six are identical to previously described subclusters and three correspond to incomplete serogroups. For example, serogroup 10 assigned previously to a single genetic subcluster is split into two genogroups, 10A–10B and 10C–10F. Similarly, serogroup 19 was assigned to a single subcluster, whereas in our analysis only serotypes 19B and 19C form a genogroup. The remaining two serotypes are too genetically dissimilar to be included; 19A shares only 50% of the capsular locus with 19F, its closest genetic neighbour within the serogroup, whereas 19F shares 65% with 19B. The last genogroup is 33A, 33F and 37; serotypes 33A and 33F belong to a genetic subcluster; serotype 37 was not included in the previous analysis since it does not use the Wzx/Wzy-dependent pathway. However, previous analysis revealed that serotype 37 isolates use the tts synthase gene which is located elsewhere in the genome to define serotype, but still contain a defective 33F-like capsular locus (Llull, Munoz & Lopez, 1999).

Serotype-defining sequence variants

Serotype-defining variants were detected following gene-based alignment of published capsular locus sequences (Bentley et al., 2006). Sequences were extracted in GenBank format and for each genogroup, gene profiling and sequence variant analysis was performed to determine variant profiles that could predict to serotype level. Types of sequence variation investigated include presence/absence of capsular biosynthesis-related genes, detection of inactivating mutations such as early stop codons and frameshifts (insertions, deletions, truncations), detection of differing alleles of genes that can be distinguished by coverage alone due to low sequence similarity (<80%) and SNP differences that can be detected by variant calling. Reported variants were then evaluated with our test cohort (n = 871; Table 2) and variants with 100% incidence rate (found in all isolates of serotype tested) were incorporated into capsular type variant (CTV) database (Table 3). Many of the variants detected during this study correspond to structural and genetic differences reported previously (Kolkman, Van der Zeijst & Nuijten, 1998; Bentley et al., 2006; Mavroidi et al., 2007; Song, Baek & Ko, 2011; Yang et al., 2011; Oliver et al., 2013b).

In two cases of previously reported frameshift mutations (7A and 15C), the mutations were confirmed in our test cohort, but no consensus of the inactivated sequence was detected (20–80% of reads carry frameshift mutation). In 7As, an insertion of a thymine (T) at residue 587 of wcwD inactivates the glycosyltransferase (Table 3) leading to loss of the side branch D-Galp-(β1 − 2)-α-D-Galp (Mavroidi et al., 2007). During evaluation with a cohort of 34 isolates (7A, n = 9; 7F, n = 25—Table 2), all 7F isolates carried a wild type (wt) wcwD gene sequence. In comparison, 7/9 of the 7A isolates also carried a wt wcwD sequence, whereas 2/9 (including the SSI 7A type strain) exhibited a mixed profile at residue 587, with only a percentage of the reads carrying the insertion (60–70%). All 7A isolates were retested in the laboratory using the SSI sera and only the two isolates with the mixed profile were re-typed as 7A whereas the 7/9 with a wt wcwD were typed as 7F, suggesting that this mixed profile is indicative of 7As and can still lead to phenotypic changes. In 15Cs, differences in the length of the TA tandem repeat region of the wciZ gene lead to loss of function for the O-acetyltransferase in 15C and are responsible for the differences in acetylation between 15B and 15C isolates (Bentley et al., 2006). However, unlike the 7A/7F scenario where the two types can be easily distinguished, distinguishing 15B and 15C isolates on the bench using SSI sera in the slide agglutination method is extremely challenging. Reversible switching between the 15B and 15C serotypes was previously described in the laboratory and in natural infection (Venkateswaran, Stanton & Austrian, 1983; Van Selm et al., 2003). Thus, in practice it may not be possible to clearly distinguish between these two serotypes in the laboratory using serological methods and for the reasons above such isolates appear as 15B/C in many publications (Laufer et al., 2010). Therefore, it was unsurprising that the capsular typing tool had problems matching the serological results where isolates had previously been reported as either 15B or 15C prior to a change to the reporting; the previous result mostly likely due to the dominance of one type over the other in the mix. During evaluation with 30 isolates (15B, n = 16; 15C n = 14), we encountered 15B and mixed 15B/C isolates but no pure 15C isolates. A genetically pure 15C was found later during validation but a different frameshift mutation was responsible for the inactivation of wciZ. Overall 11/16 15B isolates were correctly predicted as 15B and 5/16 were mixed 15B/C whereas 10/14 15C isolates were predicted as 15B and 4/14 as 15B/C. Since any 15B-15C-15B/C discordances cannot be resolved in the laboratory the 15B-15B/C-15C isolates will always be reported as 15B/C.

Molecular type 6E

Serotype 6E is not a serologically distinct serotype, but has been shown to be a distinct molecular subtype (Ko, Baek & Song, 2013). Due to the low genetic similarity between the 6E capsular locus sequence and the other serogroup 6 capsular locus sequences, PneumoCaT was unable to determine a serotype for isolates with 6E capsular locus in the first step of the capsular typing tool due to coverage <90% (n = 12; 10 6B and 2 6A). In order to take account of this variation, the genomic sequences of these 12 isolates were used to determine a consensus 6E capsular locus sequence and this was introduced into the capsular locus typing tool. With this in place the tool can determine whether an isolate has a 6E molecular type and using the variant profiles described above can then determine the serological type (i.e., 6E(6B)).

Novel serogroup 7 subtype

Genogroup 7B, 7C and 40 can be distinguished using a panel of 9 SNPs (Table S3) all found within glycosyltransferase gene wcwK. Nucleotide differences at residues 46 and 385 are considered as more significant, as both result in distinct amino acids in the three serotypes. During evaluation using a cohort of 38 isolates (7B, n = 15; 7C, n = 21; 40, n = 2—Table 1 -Development Set), one isolate originally typed as serotype 7C in the laboratory, exhibited a novel codon at residue 385 which encoded for a distinct amino acid (CTT (Leu) compared to ACT (Thr) for serotype 40, TTT (Phe) for 7B and TGT (Cys) for 7C). The rest of the SNP profile matched the 7B profile. Further investigation revealed a distinct combination of factor sera reactions (Table 4) for this isolate suggesting that it might correspond to a novel serogroup 7 serotype.

Serogroup 22

Biochemical or genetic differences between serotypes 22A and 22F have not been previously described. Initial genetic analysis revealed 99% identity over 100% of the capsular locus sequences of the two types (22A: CR931681; 22F: CR931682) suggesting a possible error in submission or serotyping. To investigate this, genomic data from the two SSI type strains (SSI-22A, SSI-22F) were assembled and using the reference sequence for 22A, the capsular locus sequences were extracted. The extracted sequences were compared to the reference sequences and the capsular locus sequence of SSI-22A was identical to the two reference sequences whereas the SSI-22F had 98% identity over 91% of the capsular locus of all three of the other sequences. Specifically, the region covering genes wcwA and wcwC shares no similarity between SSI-22F and 22A. The SSI-22F capsular locus sequence was submitted to ENA as assembly under PHE Pathogens BioProject PRJEB14267 (Accession number LT594600.1).

Novel molecular subtype 23B1

Serotypes 23A, 23B and 23F have distinct capsular locus sequences which allows for serotype prediction based on mapping coverage alone. A cohort of 72 isolates (23A, n = 25; 23B, n = 25; 23F, n = 22) was used to test the coverage-based approach. All isolates typed as 23A or 23F and 21/25 isolates typed as 23B were correctly predicted; however, 4 × 23B isolates gave a failed result in the capsular typing tool with highest coverage of approximately 70%. These isolates were re-tested in the laboratory using SSI sera and the previous results were confirmed. In order to investigate this further, more 23B isolates were sequenced and analysed using the capsular typing tool. In total, 24/46 23B isolates failed the pipeline with 70% coverage. Following further investigation, these isolates were identified as a novel molecular subtype (23B1) with ∼70% homology to the 23B capsular locus. The 23B1 capsular locus sequence was introduced into PneumoCaT assuring 100% concordance with serological type.

Serogroup 24

Genetic differences between serotypes in serogroup 24 have previously been described (Mavroidi et al., 2007) and were also confirmed following genetic analysis of the reference sequences (Accession numbers CR931686, CR931687, CR931688 for 24A, B and F, respectively). Serotype 24A lacks wzy, a polymerase gene and rbsF, a putative ribofuranose biosynthetic gene. In addition, the gene sequence of the flippase gene wzx is unique to 24A as indicated by using a blast nucleotide query against the NCBI database. This could only be confirmed by the SSI type strain (SSI-24A) since no other 24A isolates were available in the PHE archive. According to Mavroidi et al. (2007) serotypes 24B and 24F are distinguished by loss of function of genes abp1 and rmlC in 24B. This was confirmed when comparing the reference sequences. However, following evaluation with a cohort of 25 isolates (6, 24B; 19, 24F) the frameshift in rmlC was not seen in any of the 24B isolates, including the SSI-24B type strain whereas the frameshift in abp1 was only seen in SSI-24B strain, but no other 24B strains. This suggested that either all 24B isolates are actually 24F or that the reference sequences are not representative of the circulating strains; the serological typing was repeated for all isolates and original serotypes were confirmed, therefore we surmised that the latter is more likely. Genomic SNP analysis was used to investigate this further by mapping to either (a) the 24F capsular locus sequence (CR931688) or (b) the whole genome of the non-capsular S. pneumoniae R6 strain (NC_003098). Subsequent variant calling and filtering produced 145 and 19,123 high quality SNPs, respectively. Evolutionary history was inferred using the Maximum-likelihood method and phylogenetic trees were drawn (Fig. S1). No serotype specific clustering was observed in either of the trees suggesting that if there is a molecular basis for the observed serological differences, it occurs using a mechanism that is not vertically inherited e.g rapid strand-slippage variation in a microsatellite within a promoter region. Further investigation is underway but in the meantime, PneumoCaT can predict isolates to serogroup level and serological analysis can distinguish further to serotype level. Although genetic differences have been described above for differentiating 24A from 24B/24F, they are not currently implemented in PneumoCaT due to incomplete evaluation.

Serogroup 32

Structures for the capsular polysaccharide repeat unit of both serotype 32A and 32F are available and differences based on the structure have previously been described (Bentley et al., 2006; Mavroidi et al., 2007). However, when the two reference capsular locus sequences (32A: CR931696; 32F: CR931697) were aligned, 99% identity was observed over 99% of the length of the sequence. The only difference was a 5 bp gap at the intergenic region between wcrN and the HG272/3 pseudogene observed in the serotype 32 capsular locus. Gene-based analysis returned no nucleotide differences within coding regions. To investigate whether there was a submission error, as described for 22F, the capsular locus sequences (lacking the tnp regions at the beginning of the capsular locus) of the 32A and 32F SSI type strains were extracted from the assembled genomes of SSI-32A and SSI-32F strains and compared to the reference strains. Both type sequences showed 100% identity with the respective reference sequences over the length of the alignment. The serogroup 32 evaluation cohort only includes 5 isolates (32A, n = 2; 32F, n = 3—Table 1; Development Set) and following analysis with the capsular typing tool all isolates were assigned to serogroup 32. Further investigation into the BAM files revealed that the 5 bp gap was seen only in the SSI-32F type strain. More serogroup 32 isolates need to be analysed before any conclusions can be reached, but no serogroup 32 isolates have been observed in the UK since 2007. At present, the predicted serogroup 32 isolates can only be differentiated into serotypes using serological analyses.

Validation of the capsular typing tool

Following development, the PneumoCaT tool was evaluated using two panels: (a) 2046 UK isolates retrieved from PHE archive covering 72/92 serotypes, including all serotypes contained in commercial vaccines (Table 5; excluding non-typeables) and (b) genomic data from 2964 non-UK isolates from three distinct, publicly available datasets (Thailand; n = 2531—65 serotypes, USA; n = 181—serotype information not available- and Iceland; n = 252—serotypes 6A, 6B, 6C and 6E).

Both panels were used to evaluate the typability of PneumoCaT based on the proportion of isolates that were assigned a type. In the UK cohort, serotype was assigned initially to 98.2% of the typeable isolates (2010/2046). Of the 36 failed isolates, 19 were called as mixed and following retesting using a culture from a single colony, 18/19 were resolved, raising the typability to 99.1%. In the non-UK cohort, a serotype was assigned to 99% of the isolates (2934/2964) and 8/20 failed ones were called as mixed, suggesting that if retesting was possible these could be resolved.

Concordance with currently used methodology (slide agglutination using SSI sera) was evaluated using the UK cohort for which further laboratory testing was possible. Concordance was estimated on the 2028 typeable isolates for which a serotype was called with both methods (2010 + 18 mixed resolved). Overall, 92.4% concordance was observed between the predicted serotypes and the serologically-derived serotypes (n = 1873). Discordance was observed in 155 isolates (Table 6) and in most cases (77%, n = 119), the discordance involved serotypes using the same factor sera set for serotype allocation (i.e., 12B/12F, 7A/7F discordances) or serotypes using the same pool sera (i.e., serotype 31 and 7F both react with pool C serum) (Statens Serum Institut, 2013) (n = 22). Repeating the serological serotyping for these two sets resolved the initial discordance in 88.7% of the cases (125/141). This suggests that discordance in these cases may be attributed to difficulty in reading slide agglutination reactions during the serological analysis with the SSI sera (i.e., auto-agglutination with all group 12 factor sera in the case of 12B/12F typing). In one case, a novel serogroup 9 serotype was predicted based on both the SNP-pattern (Table S4) and sera reaction pattern that differed from the expected 9L pattern, showing a weak reaction with factor sera 9 g. In the remaining 14/155 cases, there are no shared sera between the serotypes predicted by WGS and the serologically-derived serotypes. These isolates were retested with both methods and all but one (unable to serotype due to autoagglutination) were concordant suggesting a possible laboratory sample translocation during archiving and/or retrieval of the isolate. Overall, following retesting, we observed a final concordance of 99.3% (2013/2028), with only 18 discordant. In some of these cases, serotyping could not resolve the discordance because auto-agglutination leads to inconclusive results (n = 10), whereas other cases (n = 8) exhibit a persistent discordance even after retesting with both methods (Table S9). In these cases, the capsular operon sequence was investigated further using an assembly and blast approach to query for both capsular operons. In all cases, the capsular operon sequence detected did not match the serologically derived serotype. However, it must be noted that for each of these serotypes, serotype was accurately predicted for the majority of cases and these discordant cases are outliers (Table 5).

The serogroup 6 isolates from the Icelandic dataset (n = 252) were originally used to demonstrate the distribution of the new 6E type and molecular typing was used to characterized all isolates (Van Tonder et al., 2015). Although these isolates were not available for further investigation, they were the only complete dataset with confirmed 6E isolates (molecular confirmation since serological confirmation is not possible), so they were used to demonstrate the ability of PneumoCaT to differentiate 6E from the other serogroup 6 types. Following analysis, 100% typability and concordance was observed (n = 252).

Reproducibility of the WGS method was investigated using a subset of 292 UK isolates covering 82/94 serotypes that were randomly selected for inclusion in the reproducibility study. For this study each isolate was cultured twice and DNA extracted and sequenced at different times. Concordance was seen in all cases.

Non typeable isolates

Virulence of S. pneumoniae is classically associated with the capsule, however some pathogenic strains, often associated with non-invasive sites (i.e., respiratory isolates), are serologically non-typeable (NT) (Park et al., 2012). Nineteen NT strains were selected from the PHE archive and analysed using WGS and PneumoCaT to predict serotype. In 13/19 cases PneumoCaT failed to predict serotype with a highest coverage <50% (Fig. 4 and Table 5). This is consistent with lack of a capsular operon and these isolates would be considered non-typeable. In addition, three isolates that originally had been typed in the lab (19A and 2 × 38) but failed with <50% coverage during PneumoCaT analysis were retested and found non-typeable in the lab, suggesting that this cut off could be used an indicative of non-typeable isolates following further validation. In the remaining 6/19 cases a serotype was predicted. In three of the cases, retesting with SSI sera gave the same serotype as predicted by PneumoCaT. In three other cases, repeat serotyping by slide agglutination consistently failed to give an acceptable result. In all of these cases, the colonies had an abnormal morphology (i.e., rough).