Direct Detection and Prediction of All Pneumococcal Serogroups by Target Enrichment-Based Next-Generation Sequencing

Abstract

Despite the availability of standard methods for pneumococcal serotyping, there is room for improvement in the available methods, in terms of throughput, multiplexing capacity, and the number of serotypes identified. We describe a target enrichment-based next-generation sequencing method applied to nasopharyngeal samples for direct detection and serogroup prediction of all known serotypes of Streptococcus pneumoniae, 32 to the serotype level and the rest to the closely related serogroup level. The method was applied to detect and to predict the serogroups of pneumococci directly in clinical samples and from sweeps of primary culture DNA, with increased detection rates versus culture-based identification and agreement with the serotypes/serogroups determined by conventional serotyping methods. We propose this method, in conjunction with traditional serotyping methods, as an alternative to rapid detection and serotyping of pneumococci.

INTRODUCTION

The introduction of conjugate vaccination has dramatically altered the prevalence and community structure of pneumococcal serotypes, in both disease and carriage (1). As the serotype valency of conjugate vaccines increased from 7 to 13, the serotypes and their prevalence were subjected to dynamic changes to less common serotypes. Thus, there is a need for laboratory methods that are capable of identifying the maximum possible number of serotypes with a limited number of assays, in order to monitor serotype replacement and any emerging serotypes (2).

The Quellung reaction remains the standard method for identification of pneumococcal serotypes. This method is expensive and time-consuming and requires expertise (3). With the sequencing of the capsular biosynthesis loci of all 90 pneumococcal serotypes, new molecular methods for pneumococcal serotyping have been developed (4). The most widely used of these methods remain sequential multiplex PCRs. The Centers for Disease Control and Prevention (CDC) (Atlanta, GA) recommends a set of 8 multiplex PCRs that are capable of differentiating 40 seroidentities, 22 to the serotype level (5) (http://www.cdc.gov/streplab/pcr.html). More recently, a set of seven real-time PCR assays capable of differentiating 21 serotypes was also recommended by the CDC (6). However, the number of tests to be performed still remains relatively high, due to the limited multiplexing capabilities associated with gel-based differentiation and quencher dye combinations. Thus, there is a need for alternative serotyping methods capable of differentiating the greatest number of serotypes at least to closely related serogroups, with a minimal number of assays, in a rapid and cost-effective manner.

Next-generation sequencing (NGS) is an attractive alternative platform for the development of diagnostic methods. The high throughput, the increasingly simple and fast methods for sample preparation, and the ability to pool samples together make these platforms versatile for adaptation. Target enrichment-based sequencing through selective enrichment of the regions of interest enables the use of sequencing reads in a more cost-effective manner. While target enrichment and sequencing are commonly used for the diagnosis of cancers and hereditary diseases, their use in diagnostic microbiology is still emerging (7). We previously used a set of published primers to enrich the common pneumococcal serotypes included in the 23-valent pneumococcal polysaccharide vaccine (PPSV23) (8) and established cutoff values for the interpretation of serogroup/serotype data based on the NGS target reads. However, with the rapid changes in the serotype composition of Streptococcus pneumoniae globally, the methodology needs to be expanded to include other serotypes. We thus extended and validated the NGS protocol to enrich serotype-specific sequences from additional S. pneumoniae serotypes. This enabled the identification and detection of all current serogroups of S. pneumoniae, including 32 at the serotype level. We then applied this methodology to identify and to serotype S. pneumoniae directly from clinical samples from hospitalized children with pneumonia and from sweeps of primary cultures and compared the results with those of the conventional method of culture and serotyping of S. pneumoniae.

MATERIALS AND METHODS

Pneumococcal isolates and capsular typing.

Thirty-eight isolates of pneumococcal serotypes/serogroups included in the second enrichment PCR were used for the validation section. All isolates were serotyped as described previously, with multiplex PCR (5) (http://www.cdc.gov/streplab/pcr.html). DNA from bacterial isolates was prepared by boiling lysis of overnight cultures.

Processing of clinical samples.

Nasopharyngeal aspirate (NPA) samples collected from children hospitalized with pneumonia at a tertiary care pediatric department during a consecutive 9-month period were evaluated. NPA samples were stored in skim milk-glycerol-glucose-tryptone soy broth (STGG) at −80°C, thawed to room temperature, and vortex-mixed for 10 to 20 s. DNA was extracted from 200 μl using the Qiagen DNeasy blood and tissue kit, with modifications suggested by the CDC (http://www.cdc.gov/streplab/downloads/pcr-body-fluid-DNA-extract-strep.pdf). Ten microliters of each NPA sample was cultured on blood agar (BA) plates with gentamicin (5 μg/ml), the samples were incubated in 5% CO₂ at 37°C for 24 h, and suspected S. pneumoniae colonies were identified by routine methods described previously (9). Plates with no growth were reincubated for an additional 24 h.

DNA extraction was also performed from sweeps of the primary cultures (sweep culture) as described by Turner et al. (10), with modifications. After a single colony was picked from the primary culture plate, a sweep of the remaining bacterial colonies was suspended in 500 μl to 1 ml of ultrapure water, and the turbidity was adjusted to a McFarland standard of 1. DNA was extracted from this suspension by simple boiling lysis.

Target enrichment-based next-generation sequencing.

The multiplex PCR for target enrichment of extended pneumococcal serotypes contained 33 previously described pairs of primers (11), of which 20 were serotype specific (13, 11F, 16F, 16A, 17A, 21, 23A, 23B, 27, 29, 31, 33C, 34, 35B, 36, 39, 43, 45, 47A, and 48) and 13 were serogroup specific (7B/7C/40, 10F/10A, 11B/11C, 15A/15F, 19B/19C, 24F/24A/24B, 25F/25A/38, 28F/28A, 32F/32A, 33B/33D/33C, 35F/47F, 35A/35C/42, and 41F/41A) (see Table S1 in the supplemental material). An 18-bp nucleotide adaptor was added to the 5′ end of the primers to enable sample pooling (8), and multiplex PCRs were optimized with or without the addition of a pair of primers targeting the streptococcal autolysin gene, with an intervening segment of specific sequence signatures (8), for the identification of S. pneumoniae. The remaining 24 common serotypes (those present in PPSV23 and serotype 6A) were enriched using the previously described PCR (8). The extended enrichment PCR used a 4-μl volume from a primer mixture containing serotype/serogroup-specific primers at 1 μM concentrations, without or with lytA-specific primers at 0.25 μM concentrations, with 2 μl of identified isolate DNA, 4 μl of primary culture DNA, or 8 μl of direct sample DNA as the template in a total reaction mixture of 25 μl, using a Platinum multiplex PCR kit (Life Technologies).

A modified step-out (MSO)-PCR used the sequence of the 18-nucleotide adaptor for the primer, with 10 unique 5-nucleotide indexes selected from an online list (http://cloud.github.com/downloads/faircloth-lab/edittag/edit_metric_tags.txt) at the 5′ end, to enable sample pooling (12, 13). The MSO-PCR used 4 μl of the purified products of multiplex PCR as the template in a 50-μl reaction mixture with other constituents as recommended for ActiTaq polymerase (Life Technologies), for 20 cycles with an annealing temperature of 53°C.

Figure 1 presents the workflow for sample preparation and the methods evaluated. The multiplex PCRs described above and described previously (8) were used in conjunction to determine the identification and pneumococcal serogroups/serotypes from DNA extracted directly from samples and from sweeps of primary cultures.

FIG 1.

Workflow for three methods for S. pneumoniae identification (ID) and prediction of serogroups/serotypes.

Library preparation and sequencing.

MSO-PCR products were analyzed visually for the presence of bands regardless of size, and DNA was purified using the QIAquick PCR product purification kit (Qiagen) and quantified using a Qubit fluorometer (Life Technologies). Purified PCR products from samples with 10 unique barcodes were pooled together in equal quantities to generate a single “index sample,” and library preparation was performed using the TruSeq DNA library preparation kit (version 2; Illumina), according to the manufacturer's instructions. Sequencing was performed with a MiSeq sequencer (Illumina), using 2 by 150-bp sequencing. The paired-end reads obtained from the sequencing run were demultiplexed for Illumina indexes with MiSeq reporter software, followed by quality filtering and demultiplexing for in-house indexes using the FASTX toolkit (http://hannonlab.cshl.edu/fastx_toolkit). Amplicons were aligned with reference sequences mentioned in the articles describing the original primers (11, 14) and with atypical and typical pneumococcal lytA gene sequences (GenBank accession numbers AJ419979.1 and AJ243407.1, respectively).

Confirmation of discrepant results between culture-based and target-based NGS methods.

For samples for which there were discrepancies in S. pneumoniae isolation or multiple serotypes were detected with the NGS method and not culture, the specimens were recultured using a 50-μl inoculum, and up to 20 to 50 colonies were picked for confirmation of the pneumococcal identities and serotyping by conventional methods. Additionally, the original DNA samples were tested with the CDC-recommended PCR, with a second round in which the product of the first round was used as a template.

Interpretation of results.

In the previous study for detection of 23-valent serotypes, we evaluated different criteria with variable stringency to correctly assign serotypes. We found that identification of a serotype for which >500 reads were mapped against the given serotype, accounting for >15% of reads mapped against serotype sequences, was a stringent criterion with 100% correct prediction of serotypes (8). Thus, the same criterion was used in this study. For the identification of S. pneumoniae in samples containing S. pneumoniae with a mixture of viridians streptococci, the percentage of reads mapped against the pneumococcus-specific lytA gene was identified as >10% of the total mapped reads for the given sample. Thus, in the validation using clinical samples versus pneumococcal isolates, only samples for which >10% of total reads were mapped against the typical lytA gene and >500 reads were mapped against serotype sequences were considered to contain pneumococci.

RESULTS

Target enrichment-based NGS detection and prediction of pneumococcal serotypes.

Table 1 shows the detailed results of the pneumococcal serotype/serogroup prediction by target-based NGS. The total numbers of reads mapped against each serotype/serogroup-specific sequence ranged from 1,667 to 18,106, while totals ranged from 1,902 to 18,701 reads in the results with inclusion of the lytA primers for detection of pneumococci. For the results with lytA primers, 316 to 14,760 reads were mapped against the typical lytA gene, which accounted for >10% of total mapped reads for a given sample. Considering all samples, the mean percentages of reads mapped against the correct serotype/serogroup-specific sequence were well above the defined cutoff value of >15% of reads at 80.6% (95% confidence interval [CI], 77.7 to 84.1%) and 80.9% (95% CI, 77.8 to 84.1%) with inclusion of the lytA gene. Applying the criteria derived from the previous study (>15% of all sequence reads and >500 total reads for a particular serotype sequence), all 38 serogroups/serotypes (100%) were correctly identified to the corresponding type; in the reaction including the lytA gene for pneumococcal identification, 37 of the 38 serogroups/serotypes (97.4%) were correctly identified. The remaining sample was also correctly identified to the original serotype but gave an additional serotype match based on the cutoff criterion.

TABLE 1.

Detection of S. pneumoniae and prediction of pneumococcal serotypes by target-based NGS

Serotype	Serogroup/serotype prediction				Pneumococcal identification and serogroup/serotype predictiona
	Total no. of reads mapped against serotype sequences	% of reads mapped against correct serotype	% of reads mapped against second match	Serotype(s) determined by target-based NGS	Total no. of reads mapped against serotype sequences	% of reads mapped against correct serotypeb	% of reads mapped against second matchb	No. of reads mapped for typical lytA	No. of reads mapped for atypical lytA	Serotype(s) determined by target-based NGS
7B	1,667	74.8	7.7	7B/7C/40	3,700	72.4	11.1	4,830	0	7B/7C/40
7C	7,036	64.8	9.6	7B/7C/40	6,601	65.1	14.8	10,432	0	7B/7C/40
10C	14,929	78.1	6.8	10C/10F	15,278	80.5	7.8	11,564	0	10C/10F
10F	7,382	60.3	12.7	10C/10F	7,989	66.8	15.9	11,992	0	10C/10F (15A/15F)
15F	4,370	93	2.6	15A/15F	9,554	94.9	9.1	3,385	0	15A/15F
13	5,426	75.1	8	13	6,050	83.3	9.1	9,858	0	13
15A	7,897	81.9	12.1	15A/15F	18,701	95.7	1.8	7,575	0	15A/15F
16F	6,286	82.6	6.5	16F	7,819	87.3	5	4,173	0	16F
39	9,738	96.1	1.4	39	3,974	96.7	1.1	1,515	0	39
17A	2,721	75.5	4.2	17A	4,556	80.5	4	4,710	0	17A
19B/19C	8,756	85.2	3	19B/19C	6,179	80.9	4.2	5,249	0	19B/19C
21	10,255	69.5	8.6	21	9,786	80.2	5.4	14,760	0	21
23A	6,086	69.8	8.7	23A	7,970	82.1	5.4	18,645	0	23A
24F	4,247	90	2.2	24A/24B/24F	1,902	64.7	8.8	6,025	0	24A/24B/24F
23B	6,992	70.8	12.6	23B	11,687	88.9	2.9	6,502	0	23B
25A	6,677	81	6.2	25A/24F/38	5,269	62.3	9.2	10,103	0	25A/24F/38
27	2,353	88.4	2.9	27	1,914	82.3	6.3	1,805	0	27
35A	2,549	73.4	8.5	35A/35C/42	1,938	76.7	5.1	2,254	0	35A/35C/42
28F	10,356	76.1	8.4	28A/28F	13,502	80.1	4.7	4,696	0	28A/28F
29	10,893	76.2	3.8	29	11,204	76.4	5.7	10,271	0	29
31	13,968	89.5	3.8	31	6,383	80.2	5.6	4,251	0	31
33B	4,163	97.1	0.1	33B/33C/33D	1,362	88.9	3.2	1,163	0	33B/33C/33D
32A	4,796	94.8	1.7	32A/32F	6,976	93.0	2.2	847	0	32A/32F
33C	4,678	52.7	4.2	33B/33C/33D	3,470	72.3	12.9	4,833	0	33B/33C/33D
34	14,260	99	0.3	34	4,117	94.1	2	6,032	0	34
41F	6,475	85.3	3.2	41A/41F	2,114	67.6	8.8	316	0	41A/41F
35B	18,106	78	6.1	35B	12,078	81.0	6	9,415	0	35B
35F	15,598	81.2	4.2	35F/47F	13,292	85.0	3.4	3,507	0	35F/47F
36	8,061	73.9	8.4	36	5,014	68.0	11.2	12,799	0	36
41F	3,117	88.2	4.2	41A/41F	3,227	83.6	5.4	1,960	0	41A/41F
28A	8,378	85.6	4.7	28A/28F	5,028	88.6	3.1	6,341	0	28A/28F
40	4,260	75.6	10.5	7B/7C/40	3,723	74.1	11.3	4,045	0	7B/7C/40
48	10,193	96.3	0.1	48	8,084	96.6	0.7	3,247	0	48

^aWith inclusion of lytA primers for pneumococcal identification.

^bPercentage of reads against serotype sequences for the given sample.

Comparison of serotypes determined by target-based NGS versus conventional culture-based serotyping.

Of 155 respiratory samples, 22.6% of the samples (35 samples) yielded S. pneumoniae in culture. The serotypes of these isolates are listed in Table 2. DNA was prepared from sweeps of bacterial colonies from 81 of 155 samples that revealed bacterial growth in the primary cultures. Of these, 40 samples were positive after PCR enrichment and were subjected to sequencing by NGS, and the results are presented in Table 2. Of the 40 samples, 38 fulfilled the criteria for pneumococcal identification, giving a pneumococcal positivity rate of 24.5%. Thirty-seven of these samples gave predicted serotypes; 36 samples contained a single serotype with >500 reads mapped against a serotype accounting for >15% of reads, while one sample (sample 27) had two serotypes predicted based on the same criteria for serotype allocation. One sample (sample 95) was identified as containing S. pneumoniae based on the lytA criteria but did not fulfill the criteria for serotype allocation.

TABLE 2.

Detailed results for pneumococcal identification and serotype prediction in respiratory samples

Sample no.	Serotype(s) determined by culture	Sweep of primary culture					Direct detection by NGS					Pneumococcal identification and serotype prediction
		Total no. of reads mapped against serotype sequences	Serotype(s)/serogroup(s) with greatest proportion of reads mapped	% of reads mapped against correct serotypea	% of reads mapped against second matcha	% of reads mapped against pneumococcal lytA sequenceb	Total no. of reads mapped against serotype sequences	Serotype(s)/serogroup(s) with greatest proportion of reads mapped	% of reads mapped against correct serotypea	% of reads mapped against second matcha	% of reads mapped against pneumococcal lytA sequenceb	Sweep of primary culture	Direct detection by NGS
1	Oral flora	5,167	39	38.5	29.8 (9N/9L) and 27.3 (19B/19C)	3.0	2,221	9N/9L	34.8	30.1 (39) and 19 (19B/19C)	8.8	No pneumococci	No pneumococci
6	15A/15F	6,175	15A/15F	91.5	5.7	43.2	9,005	15A/15F	82.9	7.9	50.5	15A/15F	15A/15F
9	19F	6,012	19F	97.4	0.9	31.4	4,578	19F	80.9	8.8	73.9	19F	19F
22	3	3,691	3	85.2	6.1	76.2	2,789	3	69.7	16.6 (23A)	80.9	3	3
23	23A	7,049	23A	97.6	1.2	35.9	6,904	23A	87.1	2.2	26.2	23A	23A
25	23A	1,166	23A	87.3	5.2	71.4	3,535	23A	91.5	5.6	62.6	23A	23A
26	6D	3,461	Sgc 6	95.7	2.2	43.9	8,691	Sg 6	94.8	3.3	40.5	Sg 6	Sg 6
27	6A	3,370	Sg 6	74.2	19.3 (34)	80.1	3,595	Sg 6	92.0	5	54.9	Sg 6 with minor cocolonizer	Sg 6
28	15B/15C	2,000	15B/15C	93.4	2.2	64.5	2,562	15B/15C	87.6	9	63.8	15B/15C	15B/15C
29	23A	8,213	23A	97.7	1.1	34.6	7,663	23A	97.0	0.7	39.9	23A	23A
30	3	1,511	3	86.1	3.4	68.7	2,278	3	88.3	4.8	71.7	3	3
32	6A	2,292	Sg 6	95.1	2.2	59.4	2,916	Sg 6	66.9	14.7	30.9	Sg 6	Sg 6
33	23A	1,459	23A	85.4	7.1	77.6	7,900	23A	86.2	8.4	69.3	23A	23A
34	6C	6,328	Sg 6	96.5	2	59.6	12,069	Sg 6	98.1	0.7	37.4	Sg 6	Sg 6
35	6C	527	Sg 6	79.3	12.5	42.0	4,140	Sg 6	98.0	1	25.4	Sg 6	Sg 6
36	10A	758	10A/10B	76	16.5	74.5	2,654	10A/10B	84.9	8.3	72.5	10A/10B	10A/10B
37	15B/15C	1,658	15B/15C	90.9	3.7	46.3	2,801	15B/15C	87.6	7.5	73.6	15B/15C	15B/15C
44	Oral flora	2,949	3	93.6	5.2	58.9	2,949	3	93.6	5.1	58.9	3	3
71	No growth						6,062	Sg 6	98.4	0.5	36.0	No pneumococci	Sg 6
73	Oral flora	148	28F	19.6	15.5 (41F)	90.65	169	Sg 6	34.3	24.3 (3)	0.0	No pneumococci	No pneumococci
94	3	1,833	3	81.1	8.2	80.7	1,810	3	82.7	5.7	81.8	3	3
95	10A	1,002	10A/10B	39.9	28.9 (19A)	62.1	1,490	23A	24.4	16.6 (19A), 13.6 (10A/10B)	87.0	Pneumococcus identified; serotype not predicted	Pneumococcus identified; serotype not predicted
99	17F	2,056	17F	77.7	9.4	85.3	4,030	17F	78.3	7.8	81.1	17F	17F
100	Oral flora	893	19B/19C	75.6	9.6	10.3	2,194	19B/19C	65.0	15.4	18.7	19B/19C	19B/19C
102	3	1,903	3	78.6	15.5 (34)	79.2	1,587	3	86.4	4.4	76.7	3	3
103	15B/15C	1,968	15B/15C	94.6	1.2	68.8	4,798	15B/15C	90.4	4.4	66.7	15B/15C	15B/15C
104	No growth						4,569	23A	81.4	6.8	72.5	No pneumococci	23A
106	6C	2,933	Sg 6	92.9	2.8	60.7	2,868	Sg 6	92.3	4	59.4	Sg 6	Sg 6
111	15A/15F	2,029	15A/15F	78.8	10	74.8	3,330	15A/15F	92.4	5.2	55.1	15A/15F	15A/15F
112	19A	13,724	19A	96.2	2.2	56.5	9,256	19A	98.7	0.4	25.5	19A	19A
113	19A	3,958	19A	95.4	3.7	54.0	2,986	19A	79.6	11.8	66.7	19A	19A
114	19F	6,736	19F	79.8	10.6	68.9	7,337	19F	79.7	14.4	59.7	19F	19F
116	Oral flora						9,932	11B/11C	76.8	12.1	7.5	No pneumococci	No pneumococci
117	3	1,528	3	80.6	6.2	82.6	3,089	3	63.7	31.3 (15A/15F)	80.0	3	3 with minor cocolonizer
125	15A/15F	8,546	15A/15F	97.6	0.5	28.1	18,404	15A/15F	99.1	0.2	14.5	15A/15F	15A/15F
128	15A/15F	7,360	15A/15F	84.3	6.9	68.9	25,102	15A/15F	98.9	0.4	18.3	15A/15F	15A/15F
132	14	1,772	14	84	6.9	67.5	2,854	14	72.0	23.4 (15A/15F)	66.1	14	14 with minor cocolonizer
137	No growth						12	Oral	NA	NA	29.4	No pneumococci	No pneumococci
142	15A/15F	10,385	15A/15F	82.9	10.7	57.2	9,362	15A/15F	97.3	1	19.6	15A/15F	15A/15F
143	Oral flora	3,052	15B/15C	82.7	14.2	69.2	2,400	15B/15C	93.7	3.4	52.0	15B/15C	15B/15C
145	23A	5,942	23A	96	1.2	39.5	5,865	23A	93.0	4.2	50.8	23A	23A
146	23A	2,686	23A	85.5	6.5	66.7	11,800	23A	91.4	4.9	47.4	23A	23A
150	6A	8,332	Sg 6	93.3	3.5	61.6	14,680	Sg 6	96.5	1.9	27.1	Sg 6	Sg 6
152	6C	2,811	Sg 6	90.9	7.1	38.8	11,209	Sg 6	95.1	2.6	8.3	Sg 6	No pneumococci

^aPredicted serotype based on previously established criteria for assignment to a serotype (10), i.e., >500 reads mapped to a particular serotype sequence and representing >15% of all reads for the sample.

^bFor pneumococcal identification, each sample should have >10% of its mapped reads aligned against the typical pneumococcal lytA gene for further consideration for serotyping.

^cSg, serogroup.

Of all 155 samples tested with the direct NGS method, 44 had amplified products after multiplex PCR for target enrichment and were subjected to NGS sequencing; 39 samples fulfilled the criteria for pneumococcal identification, giving a positivity rate of 25.2%. Thirty-eight of 39 samples fulfilled the criteria for prediction of serotypes and were mapped to a predominant serotype. In addition, two of these samples (samples 117 and 132) had sufficient numbers and percentages of reads to map against a second serotype. One sample (sample 95) had 24.4% of reads mapped against serotype 23A, 16.6% against serotype 19A, and 13.6% against serotype 10A/10B; however, as none of the serotypes had >500 reads mapped against the given type, this sample did not fulfill the criteria for serotype allocation.

All 35 samples yielding S. pneumoniae from cultures were positive for S. pneumoniae by sweeps of primary culture DNA for NGS, and the results also corroborated the serotype predictions. However, only 34 of 35 samples were considered positive by the direct sample DNA NGS method (yielding a sensitivity of 97.4% versus sweep culture identification), as one sample (sample 152) did not meet the cutoff criterion and had only 8.3% of total mapped reads aligned to the typical lytA gene. Sample 95 did not meet the criteria for serotype allocation by either primary culture or the direct sample DNA NGS method.

Sweep culture DNA NGS identified 3 additional samples as containing S. pneumoniae, and these were also identified and confirmed by the direct sample NGS method (Table 3). Two of these three samples (samples 44 and 143) yielded pneumococcal isolation of the corresponding serotypes on repeat culture. Another two samples (samples 71 and 104) were identified as containing pneumococci by the direct sample NGS method. All five samples were confirmed to contain S. pneumoniae by repeat testing of the original DNA samples, with a second-round PCR using the CDC PCR method for prediction of serotypes. Of the three samples identified as having a second serotype (Table 3), two from the direct NGS method were confirmed to have the second serotype with reanalysis of the original DNA with additional PCR cycles using the CDC PCR method.

TABLE 3.

Discrepancies between conventional culture and NGS methods for pneumococcal identification and serotype prediction

Sample no.	Serotype identified			Repeat culture with 50 μl of NPA sample and identification of up to 200 colonies		Sweep culture DNA for serotype prediction based on CDC PCRs		Direct sample DNA for serotype prediction based on CDC PCRs
	Culture	Sweep culture DNA for NGS	Direct sample DNA for NGS	Pneumococcal identification	Serotype identified	Pneumococcal identification	Serotype identified	Pneumococcal identification	Serotype identified
Additional samples identified as having S. pneumoniae
44	Oral flora	3	3	+	3	+	3	+	3
71	No growth	No growth	Sga 6	−	NAb	−	NA	+	Sg 6
100	Oral flora	19B/19C	19B/19C	−	NA	+	19B/C	+	19B/19C
104	No growth	No growth	23A	−	NA	−	NA	+	23A
143	Oral flora	15B/15C	15B/15C	+	15B/15C	+	15B/15C	+	15B/15C
Samples identified as having second serotype
27	6A	Sg 6 plus 34	Sg 6	+	6A	+	6A	+	6A
117	3	3	3 plus 15A/15F	+	3	+	3	+	3 plus 15A/15F
132	14	14	14 plus 15A/15F	+	14	+	14	+	14 plus 15A/15F
Samples with other discrepancies
95	10A	10A/10B (400 reads) plus 19A (290 reads)	23A (364 reads) plus 19A (248 reads)	+	10A, 19A, 23A	+	10A, 19A	+	10A, 23A
73	Oral flora	1,412 (90%) reads against pneumococcal lytA with <500 reads for serotype sequences	Oral flora	−	NA	−	NA	−	NA
116	Oral flora	No pneumococci	<10% of reads for pneumococcal lytA but >500 reads for serotype 11B	−	NA	−	NA	+	11B
1	Oral flora	<10% of reads for pneumococcal lytA but >500 reads for serotypes, with 39, 9N/9L, and 19B/19C	<10% of reads for pneumococcal lytA but >500 reads for serotypes, with 39, 9N/9L, and 19B/19C	+	9N/9L	+	9N/9L, 39	+	9N/9L, 39

^aSg, serogroup.

^bNA, not applicable.

For sample 95, which was identified as having pneumococci but did not fulfill the criteria for serotype allocation, the original respiratory sample was recultured on 3 plates in 50-μl aliquots, and the progeny of 50 colonies were serotyped. Thirty-six of the colonies belonged to serotype 10A, 9 to serotype 19A, and 5 to serotype 23A (Table 3). The results of reanalysis of the samples with <10% of total reads mapped against the pneumococcal lytA gene with >500 reads mapped for sequence-specific reads are presented in Table 3.

DISCUSSION

The target-based NGS method described herein was capable of identifying all serogroups, including 32 serotypes at the serotype level. NGS offers a more versatile, high-throughput alternative to detection with previously described detection methods (11, 14).

Recently, a number of novel methods to increase the capacity to detect a broader range of pneumococcal serotypes have been described. These methods include a multiplexed PCR coupled to an automated microarray assay differentiating 22 serotypes and 24 other serotypes to the subgroup level, the sequetyping method, which relies on sequencing products of a single consensus pair of primers capable of amplifying products of 84 serotypes and differentiating 46, a 5-plex multiplex PCR followed by ionization mass spectrometry, which is capable of differentiating 45 serotypes, and a set of three multiplex PCRs with 40 pairs of previously described primers followed by fragment analysis using automated fluorescence-based capillary electrophoresis, which is capable of differentiating 39 serotype/serogroups (15–17). These methods are similar to our current method in terms of serotype resolution; however, to the best of our knowledge this is the first instance in which NGS coupled with target enrichment has been used to determine S. pneumoniae with serogroups directly from clinical specimens, although recently a similar method was used to identify other bacterial isolates (18). The potential advantages of whole-genome sequencing (WGS) over the current method include the potential ability to detect novel serotypes and the ability to distinguish more serogroups to the serotype level. In this study, however, the identification of lytA positivity in the absence of an identified serotype may suggest novel serotypes that deserve reinvestigation of the original samples. Also, the current method is much more affordable, in terms of per-isolate or per-sample costs, and enables greater throughput than WGS.

The rates of detection by direct NGS from clinical samples (25.2%) and sweep culture DNA identification (24.5%) were both higher than the rate of culture-based identification (22.6%). Culture-independent methods for pneumococcal detection have been shown to increase the detection rates in both colonization studies and assessments of sterile samples (19). It might be presumed that S. pneumoniae identified from only direct sample DNA (samples 71 and 104) and not from cultures or sweep cultures reflected remnant DNA of S. pneumoniae with serotypes predicted from the sequences. Thus, this method could potentially detect nonviable organisms or organisms from specimens from patients already receiving antibiotic treatment.

A drawback of molecular methods for detecting S. pneumoniae directly from clinical samples is the potential misidentification of nonpneumococcal isolates with similar genetic make-ups. Viridans group streptococci have been found to harbor a large number of genes originally identified as pneumococcal genes (20, 21). Our assay incorporated a specific pair of primers that amplified a signature of the streptococcal autolysin gene that differentiated S. pneumoniae from nonpneumococcal isolates. The presence of specific lytA gene sequences but not serotype-specific sequences at the given cutoff criteria could potentially be used to indicate novel serotypes. This could also be due to the presence of multiple nondominant serotypes in low abundance, as exemplified by the results of reanalyzing sample 95.

One key factor in the successful application of the current method is the determination of cutoff read numbers and proportions to consider a sample for serotype allocation. Pooling multiple samples into one index sample, while helping to reduce costs, could potentially introduce false positivity of minor serotypes due to chimera formation-related issues. Further evaluations of the cutoff values in relation to the number of samples pooled to distinguish a true minor serotype versus a sample pooling artifact are needed. The number of samples pooled together is best kept uniform for a given diagnostic assay after full validation (22).

Detection of multiple serotypes in colonization is one key area in pneumococcal research that is of increasing importance, due to changes in the capsular types with the use of vaccines (2). Multiplex PCRs, microarrays, and latex agglutination assays have been used to detect multiple serotypes in colonization, with different success rates (23–25). The use of culture for detection of multiple serotypes is limited by the number of colonies that need to be identified in order to have a realistic probability of identifying a minor serotype (26, 27). The method described offers an attractive alternative to detection of multiple serotypes in colonization from primary cultures or direct sample DNA.

This method is sufficiently versatile to be applied to pneumococcal isolates, sweep cultures, and direct clinical samples. As the enrichment requires only two multiplex PCRs optimized to the same thermocycling conditions, the method is amenable to automation, and NGS could be adapted to different sequencing platforms with modified library preparation protocols. With the emergence of competitively priced kits for rapid library preparation, this method can be carried out in a relatively short time. The bioinformatics pipeline used for this method is simple. Alternatively, on-instrument data-processing pipelines may be used for analysis.

Siira et al. (28) indicated that the best way to utilize different methodologies for serotyping is to use them in a complementary manner, whereby molecular methods are used to rapidly screen large numbers of samples in a high-throughput manner and then the more precise but costly standard methods are used for further identification where needed (28). We propose our target enrichment-based sequencing method as a versatile adaptable method that can be used in conjunction with standard methods.