Clinically significant VGS isolates (n = 181) from sputum (n = 81), blood cultures (n = 29), tracheal aspirates (n = 11), midstream urine (n = 9), and other various sterile and non-sterile sites (n = 51) from Peking Union Medical College Hospital (PUMCH; 2013–2014), were studied (Supplementary Table S1). Initial identification of these isolates was done by conventional methods (positive Gram stain, coccus morphology in chains, alpha-hemolysis, and a negative catalase test) and Vitek 2 compact system. Optochin and bile solubility tests were performed to differentiate pneumococci from non-pneumococci. The isolates were stored at −70°C in skim milk until further testing.

Template DNA was prepared as described by. Dubois et al. (2013). The16S rRNA gene was amplified for all the isolates using the universal primers 27F 5′-AGAGTTTGATCMTGGCTCAG-3′ and 1492R 5′-TACGGYTACCTTGTT ACGACTT-3′ (Seng et al., 2009). Purified PCR products and sequencing primers (the same as for amplification) were mixed and sent to Ruibiotech (Beijing, China) for sequencing. Species identification was performed by comparing the obtained sequences against those in the GenBank database using BLASTn (www.ncbi.nlm.nih.gov/blast). A sequence similarity of 99% was applied as species identification “cut-off” value for the 16S rRNA gene region. Amplification and sequencing of the gyrB gene encoding DNA Gyrase subunit B was performed for all the non-pneumococci using primers gyrB7F 5′-GAAGTDGTIAARATYACBAAY CG-3′ and gyrB5R 5′-ACATCDGCATCRGTCAT-3′ (Maeda et al., 2011). Both the gyrB nucleotide and amino acid sequences were analyzed through BLASTn and BLASTp and a nucleotide sequence similarity of 96% or a signature amino acid at a certain position was applied as species identification standard according to Galloway-Pena et al. (2014). Besides, the phylogenetic trees were generated based on 16S rRNA gene and gyrB gene. The region used for phylogenetic analyses and species identification was nucleotides 86–1336 for 16S rRNA gene and nucleotides 1113–1512, corresponding to amino acids 371–503 for gyrB gene respectively. Following alignment with Clustal W, the sequences were analyzed in MEGA version 5.2 to create radial trees using the neighbor-joining statistical method and the maximum likelihood composite model.

Two MALDI-TOF MS systems, Bruker Biotyper and Vitek MS, including both the In Vitro Diagnosis (IVD) and Research Use Only (RUO) modes were used to identify the 181 VGS isolates, according to the manufacturer's instructions. For both MALDI-TOF MS systems, direct transfer method were used for sample preparation. A small portion of a single colony after 24 or 48 h of incubation was smeared onto a target plate using a wooden cocktail stick, and covered with 1 μl matrix solution (α- cyano-4-hydroxycinnamic acid in 50 acetonitrile/2.5% trifluoroacetic acid, Bruker Daltonics, Bremen, Germany; α-cyano-4-hydroxycinnamic acid, VITEK® MS CHCA) immediately. For Bruker Biotyper, measurements were performed with the Bruker Biotyper MALDI-TOF MS system using FlexControl software with Compass Flex Series version 1.3 software and a 60-Hz nitrogen laser (337 nm wavelength). Spectra ranging from 2000 to 20,000 m/z were analyzed using the MALDI Biotyper system's automation control and the current Bruker Biotyper V.3.3.1.2 software and library [database (DB) 5989 with 5989 entries]. For Vitek MS, measurement was performed using the manufacturer's suggested settings using automated collecting spectra. Captured spectra (mass range of 2–20 kDa in the linear mode) were analyzed using the MALDI-TOF MS IVD MYLA database v2.0 and also the MALDI-TOF MS RUOsystem with the SARAMIS™ database v4.10 of bioMérieux. All identifications displaying a single result with a confidence score ≥1.7 or a confidence value of 99.9% were considered acceptable for Bruker Biotyper MS and Vitek MS, respectively (Karpanoja et al., 2014). For both MS systems, all isolates yielding a single result without acceptable confidence level or multiple results or “no identification” results were re-tested. If a single, species-level identification was obtained upon repeat analysis, this identification was considered to be the final MS result; otherwise no further analysis was performed.

Agreement and validity values were calculated with a 95% confidence interval (CI) based on an exact binomial distribution. Data were analyzed using SPSS, version 15.0 (SPSS Inc, Chicago, IL, USA).

Using a combination of 16S rRNA gene and gyrB gene sequencing, all the 181 VGS isolates were assigned to species level on the basis of a ≥99 and ≥96% gene sequence similarity with published sequences in the GenBank. The DNA sequences of the16S rRNA gene and gyrB gene of the 181 VGS isolates studied have been submitted to GenBank database (accession numbers of KX661043–KX661223 for 16S rRNA gene and KX661224–KX661319 for gyrB gene respectively (Supplementary Table S2). Figure 1 shows the phylogenetic trees generated by gyrB gene and 16S rRNA gene. As is shown in Figure 1A, gyrB nucleotide sequence successfully delineated all of the 96 non-pneumococci strains into individual species branches while in Figure 1B, all of the S. oralis isolates were branched with S. mitis isolates by 16S rRNA sequences. Besides, the bootstrap values between the mitis group were rather low compared to gyrB gene, which emphasizs the difficulties of using 16S gene sequences to correctly assign VGS strains to particular species. The identification results of the 181 VGS isolates are summarized in Supplementary Tables S3, S4, and S5, using gene sequencing as the gold standard.

A high level of similarity in the 16S rRNA gene sequences of S. pneumoniae, S. pseudopneumoniae, S. mitis, and S. oralis, which could make it difficult in distinguishing these species, has been reported (Kawamura et al., 1995; Arbique et al., 2004; Suzuki et al., 2005; Haanpera et al., 2007). Nevertheless, a recent study by Scholz et al. demonstrated that S. pneumoniae has a cytosine at position 203 of the 16S rRNA gene, while all other mitis group streptococci have adenine in that position (Scholz et al., 2012). In this study, we also checked position 203 of the 16S rRNA gene sequences, and confirmed that 85 isolates were pneumococci and the other 96 isolates non-pneumococci.

Compared to the gold standard, the Bruker Biotyper system correctly identified 88.4% (160/181) of the isolates to species level and misidentified 11.6% (21/181) of the isolates. This system performed poorly in the identification of species in the mitis group, correctly identifying 80.4% (86/107) of the isolates to species level and misidentifying 19.6% (21/107) of the isolates. Optochin tests were performed on all isolates with an identification result of S. pneumoniae in an atmosphere of 5% CO₂, among which 85 isolates were sensitive and the other 21 were resistant. Specifically, 100 (11/11), 100 (2/2), and 88.9% (8/9) of S. mitis, S. oralis, and S. pseudopneumoniae isolates were misidentified as S. pneumoniae, respectively. For all other groups, the Bruker Biotyper system performed excellently, accurately identifying all the species within each of the respective groups. Furthermore, the system did not yield an invalid or “no identification result” on any isolates (Supplementary Table S3).

Vitek MS IVD correctly identified 98.9 (179/181) of the VGS isolates to species level, 0.55 (1/181) to group level, and 0.55% (1/181) with no identification. In contrast to the Bruker MS system which performed dismally in the identification of S. pseudopneumoniae isolates, the Vitek MS IVD correctly identified the majority (8 of 9, 88.9%) of these isolates to species level and the remaining one isolate to group level (Supplementary Table S4). Furthermore, the Vitek MS IVD gave a “no identification” result on one isolate within the bovis group (S. gallolyticus).

In comparison to Bruker Biotyper and Vitek MS IVD, the Vitek MS RUO system performed very poorly in VGS identification, with an overall correct species level, group level, and genus level identification rate of 38.7 (70/181), 42.0 (76/181), and 7.2% (13/181) respectively (Supplementary Table S5). The system especially performed poorly in identifying species within the mitis and bovis groups, with only 15.9 (17/107) and 25% (2/8) of the isolates correctly identified, albeit low number of isolates. In addition, 8.8 (16/181) and 3.3% (6/181) of the isolates studied were misidentified and yielded a “no identification” result, respectively.

Sensitivity, specificity, positive predictive value (PPV) and negative predictive value (NPV) of the three databases are shown in Table 1. Overall, Bruker Biotyper gave a 100% sensitivity for the majority of the species identified, except for the mitis group species, namely S. mitis (0%), S. oralis (0%), and S. pseudopneumoniae (11.1%), most of which were misidentified as S. pneumoniae (21 of 22, 95.6%). Furthermore, PPV was incalculable for S. mitis and S. oralis as none of these isolates were correctly identified by the Bruker Biotyper. In contrast to other species in the mitis group, S. pneumoniae, had a 100% (95 CI: 95.8–100.0%) sensitivity and a 100% (95 CI: 95.5–100.0%) NPV, while the specificity and PPV were relatively low, at 78.1% (95 CI:68.5–85.9%) and 80.2% (95 CI:71.3%–87.3%), respectively.

Compared with Bruker Biotyper, Vitek MS IVD gave a better resolution for S. mitis/S. oralis identification with a sensitivity of 100% (95 CI: 75.3–100%) though the two species couldn't be distinguished from each other due to database limitation. Similarly, the overall sensitivity for S. pseudopneumoniae identification [77.8% (95 CI: 40.0–97.2%)] was higher than that of Bruker Biotyper [11.1% (95 CI: 0.3–48.3%)].

Expectedly, Vitek MS RUO showed low sensitivity for the identification of most species, except for S. anginosus, S. intermedius, S. sanguinis, S. gordonii, and S. salivarius.

The authors declare that the research was conducted in the absence of any commercial or financial relationships that could be construed as a potential conflict of interest.