Article Text
PDF
Abstract
Background Differentiation of Streptococcus pneumoniae from other viridans group streptococci is well known to be challenging in clinical laboratories. Matrix assisted laser desorption ionisation–time of flight mass spectrometry (MALDI-TOF MS) had been reported to be a good alternative for Streptococcus species level identification. However, differentiation of S. pneumoniae from other Streptococcus mitis group organisms was found to be problematic using the Bruker MALDI Biotyper system.
Methods This study used the Bruker MALDI Biotyper system in addition to a mass spectra model analysis generated by 10 reference strains of S. pneumoniae, 8 strains of S. mitis and 2 strains of S. oralis in the ClinProTools to identify 28 clinical isolates of S. pneumoniae and 47 isolates of S. mitis/oralis. The results were compared with those generated by the MALDI Biotyper system alone.
Results The percentages of correct species level identification using the MALDI Biotyper system alone and the direct transfer and extraction method were 66.7% (50/75) and 70.7% (53/75), respectively. With the additional ClinProTools mass spectra analysis, the percentages of correct identification by the direct transfer and extraction method increased to 85.3% (64/75) and 100% (75/75), respectively. This new workflow significantly improved the accuracy of S. pneumoniae and S. mitis/oralis identification.
Conclusions The additional ClinProTools mass spectra analysis with extraction method after MALDI Biotyper identification significantly improved the accuracy of identification among S. pneumoniae, S. oralis and S. mitis. The extra 15 min processing time of spectra analysis should be affordable in most clinical laboratories. We suggest that the same approach could be further explored in handling other bacterial species with high similarities.
- STREPTOCOCCUS
- MICROBIOLOGY
- DIAGNOSTICS
Statistics from Altmetric.com
Introduction
Streptococcus pneumoniae is one of the common human respiratory pathogens causing systemic infections, such as meningitis and endocarditis, in adults and children, with a high mortality rate.1 ,2 For appropriate antimicrobial therapy, it is clinically important to accurately differentiate S. pneumoniae from other viridans group streptococci (VGS). However, due to the extraordinarily high similarity in phenotypic and genotypic characteristics among VGS members, especially Streptococcus mitis group species, differentiation of these species is complicated.3 ,4
In general, laboratory tests (colony morphology examination, optochin susceptibility and bile solubility) are recommended for S. pneumoniae identification.5 However, interpretation of these tests may occasionally lead to incorrect identification.6 ,7 Genotypic methods are useful for VGS species identification. However, there is no generalised method for genotyping most Streptococcus species. Streptococcus housekeeping genes, such as rpoB, which exhibit >15% sequence divergence among species, were suggested for species differentiation.8 ,9 Unfortunately, PCR sequencing is relatively expensive and time consuming, which may not be suitable for clinical laboratory use.
Recently, matrix assisted laser desorption ionisation–time of flight mass spectrometry (MALDI-TOF MS) has been implemented in clinical laboratories for bacterial identification.10 S. pneumoniae identification by MALDI-TOF MS has been reported to be problematic due to the extraordinarily high homology of the protein mass spectra among S. pneumoniae, S. mitis, S. oralis and S. pseudopneumoniae.11 S. mitis and S. oralis were occasionally reported erroneously as S. pneumoniae by the Bruker MALDI Biotyper system with spectra library V.3.3.1.2 (4613 spectra).12 To improve the accuracy of Streptococcus species identification, the manufacturer incorporated 100 additional VGS mass spectra into the updated spectra library V.4.0.0.1 (5627 spectra) in early 2014.
In addition to the MALDI Biotyper for bacterial identification, mass spectra analysis through the use of Bruker ClinProTools software was introduced to discover biomarker mass peaks and to generate classification models for identification of certain phylogenetically related bacterial species, including VGS species.11 ,13–15 ClinProTools can generate classification models from large numbers of spectra in a rapid and user friendly manner. By using the generated models, a common signature among the spectra of each of the model generation classes can be identified, and spectra from unknown isolates can be classified simply by the model.16
In this study, we tested the performance of the most updated MALDI Biotyper spectra library, V.4.0.0.1 (5627 spectra), for S. pneumoniae and S. mitis/oralis identification. Then we added the ClinProTools mass spectra analysis after MALDI Biotyper identification. Identification results from the mass spectra analysis were compared with those generated by the MALDI Biotyper alone.
Methods and materials
Strains
A total of 75 clinical isolates, including 28 isolates of S. pneumoniae and 47 isolates of S. mitis/oralis, collected in 2013, were selected for the study. These isolates originated from different types of clinical specimens, including 30 sputum, 21 blood culture, 10 urine, 5 fine needle aspirate tissue, 7 peritoneal fluid, 1 pleural fluid and 1 biopsy. The isolates were cultured on Columbia blood agar, with incubation at 37°C with 5% CO2 for 18 h. Isolated colonies were then used for identification. Another 10 reference strains of S. pneumoniae and 10 reference strains of S. mitis groups were used to build up the mass spectra analysis model in the ClinProTools software.
Routine clinical diagnostic methods
The identities of the 75 clinical isolates were initially classified by colony morphologies, optochin susceptibility and bile solubility. Biochemical or enzymatic characteristics were further determined by the VITEK 2 system (BioMerieux, Marcy l'Etoile, France) if necessary. For isolates with an uncertain identification, S. pneumoniae specific lytA real time PCR, 16S rRNA and VGS rpoB gene sequencing were performed for species confirmation.17–19 Sequences were submitted to the NCBI Genbank and the HKU 16SpathDB V.2.0 database for species confirmation.20
Identification by Bruker MALDI Biotyper alone
All 75 isolates were prepared for MALDI-TOF MS using the manufacturer's recommended direct transfer and ethanol–formic acid (EtOH-FA) extraction protocols. A single isolated colony was inoculated directly onto the MSP96 target plate spot by the direct transfer method. The EtOH-FA extraction method was performed according to the Bruker protocol.21 Each bacterial extract was spotted onto the MSP96 target plate after centrifugation. Each spot was overlaid with α-cyano-4-hydroxycinnamic acid matrix (Sigma Aldrich, St Louis, Missouri, USA).
The target plate was analysed by the Bruker microflex LT system (Bruker Daltonics, Bremen, Germany). The protein profile of each spot with m/z values of 3000–15 000 generated was analysed by the MALDI Biotyper V.3.1 with the most updated spectra library, V.4.0.0.1 (5627 spectra). The top 10 identification matches were generated along with confidence scores, ranging from 0.0 to 3.0; a score >2.0 indicated promising species level identification.
Mass spectra analysis identification by ClinProTools
A mass spectra peak analysis model consisting of 370 S. pneumoniae/mitis/oralis mass spectra was created from 10 identity confirmed S. pneumoniae and 10 identity confirmed S. mitis group reference strains (8 strains of S. mitis and 2 strains of S. oralis) (table 1). For each strain in the model, 18–24 high quality spectra were prepared by the extraction method and were captured using flexControl V.3.4 software (Bruker Daltonics). The spectra were then imported into the ClinProTools V.3.0 software (Bruker Daltonics) for recognition of mass spectra patterns between S. pneumoniae and other S. mitis species.22 Spectra pretreatment, peak picking and peak calculation operations were performed using the preset configuration. Classification models were generated using all four available algorithms (genetic algorithm, support vector machine (SVM), supervised neural network and QuickClassifier) and compared. For the SVM algorithm, the number of peaks selected in the model was set to 10, while the default settings were left unaltered for the other modelling algorithms. For each model, the recognition capability (RC) and cross validation (CV) percentage was generated to demonstrate the reliability and accuracy of the model. RC and CV percentages were indicators of the model's performance and useful predictors of the model's ability to classify test isolates. The model with the highest RC and CV values were used in the analysis.
Characteristics of the 20 Streptococcus isolates used to build up the mass spectra analysis model in ClinProTools
For the 75 clinical isolates, their mass spectra generated from direct transfer and the EtOH-FA extraction method were imported into the ClinProTools after MALDI Biotyper identification for further analysis. Significant differences in identification accuracy among the different methods (MALDI Biotyper with direct transfer method, MALDI Biotyper with EtOH-FA extraction method, ClinProTools with direct transfer method and ClinProTools with EtOH-FA extraction method) were compared using Cochran's Q test in MedCalc software V.14.10.2 (MedCalc Software, USA).
Results
Routine biochemical and molecular identification
Among the 75 clinical isolates, 28 were confirmed as S. pneumoniae and another 47 were identified as S. mitis or S. oralis using the biochemical and molecular methods. All data are shown in supplementary table 1. All S. pneumoniae isolates were optochin sensitive, bile soluble and lytA gene PCR positive. The other 47 S. mitis/oralis isolates were all optochin resistant and bile resistant. Their 16S rRNA and rpoB gene sequencing results further confirmed them to be either S. mitis or S. oralis. However, due to the high similarity in biochemical responses and genotypic sequences between S. mitis and S. oralis, these two species could not be well differentiated. Therefore, these two species were grouped as S. mitis/oralis in further analyses.
MALDI Biotyper identification alone
Using the direct transfer method, 50 of 75 clinical isolates (66.7%) were identified correctly at the species level by the latest version of the MALDI Biotyper system, with scores >2.0 (see supplementary table 1). All 28 clinical isolates of S. pneumoniae (100%) were correctly identified (scores 2.000–2.307). In contrast, 22 of 47 (46.8%) S. mitis/oralis isolates were correctly identified as either S. oralis or S. mitis as the top choice among the top 10 identifications by the MALDI Biotyper (scores 2.019–2.249). The other 13 (27.7%) S. mitis/oralis isolates could only be identified at the genus level by the MALDI Biotyper, with scores of 1.717–1.993. There were another 11 isolates (23.4%) that were misidentified as S. pneumoniae but with scores >2.0 (scores 2.027–2.248).
Using the EtOH-FA extraction method, overall 53 of 75 clinical Streptococcus isolates (70.7%) were identified correctly by the MALDI Biotyper system alone (see supplemetary table 1). All 28 isolates of S. pneumoniae (100%) were correctly identified at the species level (scores 2.000–2.405). For the other 47 isolates of S. mitis/oralis, 25 (53.2%) were correctly identified as either S. oralis or S. mitis by the MALDI Biotyper (scores 2.049–2.383), while the other 22 (46.8%) isolates were misidentified as S. pneumoniae (scores 2.004–2.337).
MALDI Biotyper identification plus ClinProTools mass spectra analysis
In the ClinProTools, classification models using the genetic algorithm and the SVM algorithm demonstrated values of 100% for RC, CV, sensitivity and specificity through validation (table 2). After validating with other Streptococcus reference strains in the laboratory, the SVM model, analysing 10 peaks of each mass spectrum, was selected for this study (figure 1, table 3). Using the SVM classification model, spectra analysis from the direct transfer method and the EtOH-FA extraction method generated correct identification for 64 (85.3%) and all 75 (100%) clinical isolates, respectively (see supplementary table 1).
Parameters of the classification models generated on the basis of different algorithms
ClinProTools peak statistics for the 10 peaks included in the support vector machine model
(A) Comparison of the average spectra of the Streptococcus pneumoniae and Streptococcus oralis/mitis reference isolates between 3300 Da and 3500 Da. (B) Corresponding gel view representation (rainbow scale colour scheme) of the same region. S. pneumoniae (class 1) (bottom) and S. oralis/mitis (class 2) (top). Peaks with a mass of 3445, 3477 and 3485 Da were selected in the support vector machine classification model.
The additional spectra analysis with the EtOH-FA extraction method significantly improved the identification accuracy of S. pneumoniae and S. oralis/mitis compared with the MALDI Biotyper with the direct transfer method (from 46.8% to 100%), the MALDI Biotyper with the extraction method (from 53.2% to 100%) and the MALDI Biotyper plus spectra analysis with the direct transfer method (from 85.3% to 100%) (Cochran's Q test, p<0.01) (table 4; supplementary table 1).
Summary of matrix assisted laser desorption ionisation–time of flight results of 75 Streptococcus clinical isolates
Discussion
Reliable identification for Streptococcus species has been an important unresolved issue in clinical laboratories.23 Recent studies demonstrated that different MALDI-TOF MS systems could be used to differentiate most VGS at the species level.12 ,24 ,25 However, the Bruker MALDI-TOF MS system with spectra library V.3.3.1.2 (4613 spectra) was reported to have identification problems for S. pneumoniae, S. oralis and S. mitis.11 ,26 To resolve this problem, the manufacturer tried to increase the number of mass spectra of S. pneumoniae, S. mitis and S. oralis in the latest version of the spectra library. Our study is the first to evaluate the accuracy of identification of the 5627 spectra library among Streptococcus species. We evaluated the library itself as well as developing a new practical workflow which can significantly improve the accuracy of S. pneumoniae and S. mitis/oralis identification by the addition of the ClinProTools mass spectra analysis.
From our study, the use of the latest version of the MALDI Biotyper alone was found to be unreliable for S pneumoniae and S. mitis/oralis identification. Although the MALDI Biotyper identified all S. pneumoniae clinical isolates correctly, there was a high proportion of S. mitis/oralis isolates that were misidentified as S. pneumoniae. Poor identification of S. mitis/oralis by this method was similar to that reported in other studies.12 ,27 This indicates that the newly updated spectra library V.4.0.0.1 cannot resolve the misidentification problem. From our experience, the critical problem could be due to the low resolution of the MALDI Biotyper identification algorithm, which only analysed the peak patterns of the whole spectrum, instead of analysing the presence or absence of specific peaks in the spectrum.
Among those S. oralis/mitis isolates misidentified as S. pneumoniae in our study, S. oralis or S. mitis was frequently found in the top 10 identification choices. Despite the manufacturer's claim that the EtOH-FA extraction method could generate better quality spectra and lead to better performance of the MALDI Biotyper, our data showed that the extraction method did not significantly improve the performance of the MALDI Biotyper for Streptococcus species identification.
As a practical way of solving this misidentification problem, we suggest using the MALDI Biotyper with the direct transfer method, for firstline screening. For samples identified as S. pneumoniae, S. mitis or S. oralis, isolates can be re-extracted by the EtOH-FA extraction method, followed by mass spectra analysis in the ClinProTools. Spectra generated by the extraction method can be submitted to the pre-developed SVM classification model in the ClinProTools, in which 10 specific peaks in each query spectra are compared statistically. Our data showed that addition of mass spectra analysis after MALDI Biotyper identification significantly increased the correct identification rate to 100% for both S. pneumoniae and S. oralis/mitis group isolates. The working procedure for the additional mass spectra analysis was simple, and handling time was approximately 15 min for 10–12 samples.
Although our study may be limited by the small sample size of only 75 clinical isolates of Streptococcus from various specimen types, this is the first report providing a practical solution for clinical laboratories in handling S. pneumoniae misidentification problems. Also, use of the SVM classification model for other VGS species differentiation requires further evaluation.
In conclusion, addition of ClinProTools mass spectra analysis after MALDI Biotyper identification is a clinically practical approach which can significantly improve the accuracy of S. pneumoniae and S. mitis/oralis identification. The same approach could be further explored in handling other bacterial species with high similarities.
Take home messages
The most updated matrix assisted laser desorption ionisation (MALDI) Biotyper version could not solve the streptococcus misidentification problem.
Addition of ClinProTools mass spectra analysis significantly improved the correct identification rate for S. pneumoniae and S. mitis/oralis for clinical diagnostic use.
This spectra analysis approach can be used for other bacteria species with high genetic or proteinomic similarities.
References
Supplementary materials
Supplementary Data
This web only file has been produced by the BMJ Publishing Group from an electronic file supplied by the author(s) and has not been edited for content.
- Data supplement 1 - Online supplement
Footnotes
Handling editor Slade Jensen
Contributors JHKC, W-CY, VCCC, PCYW and K-YY designed and supervised the study. JHKC, KKKS and O-YW performed the study. JLLT, SKPL and PCYW provided the sample collections. JHKC wrote the manuscript.
Funding The work was supported by the Research Fund for the Control of Infectious Diseases of the Food and Health Bureau of the Hong Kong Special Administrative Region Government.
Competing interests None.
Provenance and peer review Not commissioned; externally peer reviewed.