Comparison of the Accuracy of Two Conventional Phenotypic Methods and Two MALDI-TOF MS Systems with That of DNA Sequencing Analysis for Correctly Identifying Clinically Encountered Yeasts

We assessed the accuracy of species-level identification of two commercially available matrix-assisted laser desorption ionization-time of flight mass spectrometry (MALDI-TOF MS) systems (Bruker Biotyper and Vitek MS) and two conventional phenotypic methods (Phoenix 100 YBC and Vitek 2 Yeast ID) with that of rDNA gene sequencing analysis among 200 clinical isolates of commonly encountered yeasts. The correct identification rates of the 200 yeast isolates to species or complex (Candida parapsilosis complex, C. guilliermondii complex and C. rugosa complex) levels by the Bruker Biotyper, Vitek MS (using in vitro devices [IVD] database), Phoenix 100 YBC and Vitek 2 Yeast ID (Sabouraud's dextrose agar) systems were 92.5%, 79.5%, 89%, and 74%, respectively. An additional 72 isolates of C. parapsilosis complex and 18 from the above 200 isolates (30 in each of C. parapsilosis, C. metapsilosis, and C. orthopsilosis) were also evaluated separately. Bruker Biotyper system could accurately identify all C. parapsilosis complex to species level. Using Vitek 2 MS (IVD) system, all C. parapsilosis but none of C. metapsilosis, or C. orthopsilosis could be accurately identified. Among the 89 yeasts misidentified by the Vitek 2 MS (IVD) system, 39 (43.8%), including 27 C. orthopsilosis isolates, could be correctly identified Using the Vitek MS Plus SARAMIS database for research use only. This resulted in an increase in the rate of correct identification of all yeast isolates (87.5%) by Vitek 2 MS. The two species in C. guilliermondii complex (C. guilliermondii and C. fermentati) isolates were correctly identified by cluster analysis of spectra generated by the Bruker Biotyper system. Based on the results obtained in the current study, MALDI-TOF MS systems present a promising alternative for the routine identification of yeast species, including clinically commonly and rarely encountered yeast species and several species belonging to C. parapsilosis complex, C. guilliermondii complex, and C. rugosa complex.


Introduction
Invasive candidiasis and candidemia remain leading causes of morbidity and mortality in immunocompromised hosts, particularly among patients with hematological malignancies in intensive care units [1][2][3][4][5]. Optimal and early therapy has been shown to significantly reduce the incidence of morbidity and mortality in patients with invasive yeast infections and has been shown to be associated with an overall reduction in cost of hospital care for these patients [5]. Although Candida albicans remains the predominant agent responsible for healthcare-associated infections, non-albicans Candida species, such as C. glabrata, C. parapsilosis complex, C. tropicalis, C. guilliermondii complex, C. dubliniensis, C. lusitaniae, and C. krusei have emerged as significant opportunistic pathogens [2][3][4]. Given the inherently variable antifungal susceptibility profiles of different Candida species and the emergence of infections due to previously rarely identified Candida species, correct identification to the species level is necessary for clinicians to make informed therapeutic decisions [2][3][4][5]. Recently developed molecular identification methods have shown that several members of Candida species (e.g. C. parapsilosis complex, C. guilliermondii complex, and C. rugosa complex) belong to the same complex [6][7][8].
In this study, we assessed the accuracy of species-level identification of two commercially available MALDI-TOF MS systems (Bruker Biotyper and Vitek MS) and two conventional phenotypic methods (Phoenix 100 YBC, Vitek 2 Yeast ID) for the identification of clinical yeast isolates. Accuracy was determined by comparing the results of those techniques with the results of rDNA gene sequencing analysis. Clustering analysis of the spectra of C. parapsilosis complex, C. guilliermondii complex, and C. rugosa complex by the Bruker Biotyper system were also performed.

Yeast Isolates
A total of 272 isolates of yeasts, including 90 randomly selected blood culture isolates of C. parapsilosis complex, were included in this study. These isolates included 91 challenge isolates provided by Becton-Dickinson Diagnostic Systems (Sparks, MD, USA) and 181 non-duplicate and randomly selected clinical isolates from patients who were treated at the National Taiwan University Hospital (NTUH) ( Table 1). There were no differences in species distribution of yeasts between BD challenge isolates and NTUH isolates. All the Candida isolates collected from the NTUH were recovered from blood specimens from patients with candidemia. A total of 20 isolates of Cryptococcus species were recovered from blood (n = 10) and cerebrospinal fluid (n = 10). The other yeasts were isolated from various clinical specimens from patients treated at the NTUH. Among 90 isolates of C. parapsilosis complex, 18 isolates were randomly selected and included in the initial evaluation by four systems to reflect of common population distribution of different Candida species of the hospitals [1][2][3][4]. All the yeast isolates from the hospital (NTUH) and the 91 challenge strains from BD were subcultured twice from original stock solution to get the pure cultures (no mixed cultures) and were identified simultaneously by two commercial phenotypical methods, two MALDI-TOF MS methods, and molecular methods from the same culture plates these isolates grew. When any discordant results among isolates by these four phenotypical methods and molecular methods occurred, we repeated the identification methods to all the isolates (with discordant results) simultaneously.
The study was reviewed and approved by the institutional Review Board of National Taiwan University Hospital (number 201211073RSC) with waiver of requirement for obtaining informed consent. Patient-related data associated with the yeast isolates, i.e., age and sex of the patients, are not presented for ethical considerations. No verbal or written informed consent was obtained from the patients in this study.

Phenotypic Characteristics of Isolates
Species identification was performed using the Vitek 2 yeast identification card (Vitek 2 Yeast ID; bioMérieux, Marcy l'Etoile, France) and Phoenix 100 Yeast ID panel (Becton-Dickinson Microbiology Systems, Sparks, MD, USA), according to the manufacturer's instructions [15]. Prior to testing, the isolates were subcultured onto Sabouraud dextrose agar (SDA, bioMérieux) and trypticase soy agar on 5% sheep blood agar plates (BAP, BD Microbiology System) and were incubated at 35uC in an aerobic atmosphere for 24-48 h. Extended incubation to 72 h was needed Thermus scotoductus Rhodosporidium toruloides  Candida species Cryptococcus species 20 C. neoformans 17 to ascertain adequate growth. C. albicans ATCC 14053 (Vitek 2 YST), C. albicans ATCC 24433, and C. parapsilosis ATCC 22019 (Phoenix Yeast ID panel) were used as control strains.

Performance of Bruker Biotyper
For analysis of the 272 yeast isolates by the MALDI-TOF Biotyper (Bruker Biotyper) system, samples were prepared as previously described [16]. A single colony from a BAP (BD Microbiology System) was subjected to an ethanol-formic acid extraction procedure for microorganism profiling. Briefly, two to three yeast colonies were transferred using a 5-ml inoculating loop into 300 ml of distilled water and 900 ml of ethanol. The suspension was pelleted after centrifugation at 12,000 rpm for 2 min, dried, and then reconstituted in 50 ml of 70% formic acid. After incubation for 30 seconds, 50 ml of acetonitrile (Sigma-Aldrich) was added. The suspension was then centrifuged at 12,000 rpm for 2 min. A volume of 1.0 ml of the supernatant was applied to a 96-spot polished steel target (Bruker Daltonik GmbH, Bremen, Germany) plate and dried. A saturated solution of 1.0 ml of MALDI matrix (HCCA; Bruker Daltonik) was applied to each sample and dried. Measurements were performed with Bruker Microflex LT MALDI-TOF MS (Bruker Daltonik GmbH) using FlexControl software with Compass Flex Series version 1.3 software and a 60 Hz nitrogen laser (337 nm wavelength). Spectra were collected in the linear positive mode in a mass range covering 1,960 to 20,132 m/z. Spectra ranging from the mass-to-charge ratio (m/z) 2,000 to 20,000 were analyzed using MALDI Biotyper automation control and the Bruker Biotyper 3.1 software and library (version 3.1.66, with 4,613 entries; Bruker Daltonics). Identification scores of $2.000 indicated species-level identification, scores of 1.700 to 1.999 indicated genus-level identification, and scores of ,1.700 indicated no identification.
Cluster Analysis by Bruker Biotyper for C. parapsilosis Complex, C. guilliermondii Complex, and C. rugosa

Complex Isolates
Clustering analysis of the isolates was performed using ClinProtools 3.0 (Bruker Daltonics GmbH, Bremen, Germany). C. parapsilosis complex, C. guilliermondii complex, and C. rugosa complex were analyzed for specific signals by clustering analysis [17]. In this study, there were 90 isolates of C. parapsilosis complex, including 30 isolates of C. parapsilosis, 30 isolates of C. metapsilosis, and 30 isolates of C. orthopsilosis. There were 18 isolates of C. guilliermondii complex, including 13 isolates of C. guilliermondii and five isolates of C. fermentati. In addition, four isolates of C. rugosa complex, including three C. rugosa and one C. pararugosa isolates.

Performance of the Vitek MS System
Samples of the 272 yeast isolates were prepared for analysis by the Vitek MS system as reported previously with slight modifications [16]. Briefly, a 1-ml loop (Sarstedt, Newton, NC) was used to apply a portion of a single colony to a disposable target slide (bioMérieux) composed of a polypropylene carrier. Formic acid (0.5 ml, 25% [vol/vol]; bioMérieux) was then applied directly to the isolate immediately after application on the target plate. The formic acid overlay was allowed to dry at room temperature. Then 1 ml of CHCA matrix solution (3.1% [wt/vol] a-cyano-4hydroxycinnamic acid; bioMérieux) was applied and allowed to dry at room temperature prior to mass spectrometric analysis. The isolate information was transferred to the Vitek MS acquisition station using Myla v3.2 middleware. The samples were analyzed with the Vitek MS in linear positive-ion mode across the mass-tocharge ratio range of 2,000 to 20,000 Da. The resulting quantitative value (confidence value) was then calculated and expresses the similarity between the unknown yeast and every Vitek MS (CU database) Lodderomyces elongisporus Sporobolomyces salmonicolor Thermus scotoductus No ID (1) Geotrichum klebahnii

Molecular Identification
For sequence analysis of the isolates, two fungus-specific universal primers were used to amplify the region spanning the internal transcribed spacer region of the 5.8S rRNA gene (59-TCC GTAGGTGAACCTGCGG-39) and the internal transcribed spacer region of the 18S and 28S rRNA genes (59-TCCTCCGCTTATTGATATGC-39) [11]. For species identification of Trichosporon species, sequences of the ribosomal DNA IGS1 region, which is located between the 26S and 5S rRNA genes, were also analyzed [18]. The IGS1 region (which ranges in length from 195 to 719 bp) was amplified using the primers 26SF (59-ATCCTTTGCAGACGACTTGA-39) and 5SR (59-AGCTT-GACTTCGCAGATCGG-39) [18]. The determined sequence was then compared with sequences available in GenBank by means of the BLAST program (http://blast.ncbi.nlm.nih.gov/ Blast.cgi). Results were considered acceptable if homology with other entries in the databases used for comparison was .99.5%.  C. guilliermondii complex isolates were further subjected to PCR with the primers RIBO-F (59-ACAGTTGGTCGAGGTGGTC-39) and RIBO-R (59-CCTGGGTTCCCAAGTAGTCA-39) for the riboflavin synthetase gene [7]. The amplified products were cleaved by restriction enzyme HgaI or HincII to differentiate between C. guilliermondii and C. fermentati. C. guilliermondii BCRC22093 (ATCC 46036), C. fermentati BCRC23164, and (CBS9966) were used as control strains.

Statistical Analysis
The chi-square test was used to evaluate the performance of two conventional phenotypic methods (Phoenix 100 YBC, Vitek 2 Yeast ID) and two commercially available MALDI-TOF MS systems (Bruker Biotyper and Vitek MS) for the identification of all isolates of Candida species, Cryptococcus species, and all yeast isolates. P values ,0.05 are considered statistically significant.

Yeast Isolates
Of the 272 isolates of yeast species confirmed by DNA sequencing analysis, 214 were Candida species, 20 were Cryptococcus species, 13 were Rhodotorula mucilaginosa, 8 were Trichosporon species, 5 were Saccharomyces cervevisiae, 2 were Pichia species and the remaining 10 isolates were unusual species. The sequencing results had .99.5% homology with other entries in the databases. Among the 90 isolates of C. parapsilosis complex, all were subjected to cluster analysis by the Bruker Biotyper system and 18 randomly selected isolates were used for assessment of the accuracy of two commercially available MALDI-TOF MS systems and two conventional phenotypic methods. Table 1 shows the number and proportions of the 200 yeast isolates that were correctly identified by the Phoenix YBC system and the Vitek 2 Yeast ID system on SDA and BAP. Overall, the performance of both systems on SDA and BAP was similar for Candida species (P = 0.640 for Phoenix YBC system and P = 1.000 for Vitek 2 Yeast ID system) and all yeast isolates (P = 0.445 for Phoenix YBC system and P = 0.909 for Vitek 2 Yeast ID system). The correct identification rates of the 200 isolates to the species or complex (C. parapsilosis complex, C. guilliermondii complex, and C. rugosa complex) levels on SDA by the Phoenix 100 was better than that by Vitek 2 Yeast ID systems (89.0% vs. 74.0%; P,0.001), respectively. For 142 isolates of Candida species, the Phoenix 100 YBC (SDA) and Vitek 2 Yeast ID (SDA) systems correctly identified 93.7% (n = 133) and 83.1% (n = 118) (P = 0.005), respectively, of the isolates to the species or complex levels. Neither system was able to correctly identify species among isolates of C. parapsilosis complex and C. guilliermondii complex. Using the Vitek 2 Yeast ID system, one C. albicans isolate was identified as C. albicans/C. famata on BAP. In addition, four isolates of C. krusei on BAP and one on SDA were reported as C. krusei/C. lambica/C. inconspicua ( Table 2). The majority of misidentified species among C. guilliermondii complex isolates were C. famata or C. famata/C. guilliermondii. (Table 2). Kodamaea ohmeri was identified both on the BAP and SDA by the Vitek 2 system and C. guilliermondii var. membranaefaciens was identified on BAP by Phoenix 100 YBC system. Because K. ohmeri is an ascosporogenous yeast and a teleomorph of C. guilliermondii var. membranaefaciens, both identifications were considered correct as C. guilliermondii complex.

Results of DNA Sequencing Compared with those of Vitek 2 Yeast ID and Phoenix 100 YBC
For Cryptococcus isolates, 90.0% (n = 18) were correctly identified to the species level by the Phoenix 100 YBC system and 75.0% (n = 15) were correctly identified to the species level by the Vitek 2 Yeast ID (SDA) system (P = 0.210). When using the Phoenix system for identification of Trichosporon species, we found that SDA resulted in a better identification rate than BAP (87.5% vs. 62.6%). Of the 13 isolates of Rhodotorula mucilaginosa, 76.9% could be accurately identified to the species level by the Phoenix 100 YBC system; however, none of the isolates were correctly identified using the Vitek 2 Yeast ID system.  Table 1). The low rate of correct identification of C. parapsilosis complex isolates to the species level by the Vitek MS system (IVD database) was due to the fact that 12 of the 18 C. parapsilosis complex isolates were C. metapsilosis (n = 6) and C. orthopsilosis (n = 6), which are not listed in the CU database. Among 90 isolates of C. parapsilosis complex, the Bruker Biotyper system was able to correctly identify all three species of C. parapsilosis (score value range, 1.996-2.241), C. metapsilosis (score value range, 1.731-2.271) and C. orthopsilosis (score value range, 1.701-2.086).

Results of DNA Sequencing Compared with those of the Bruker Biotyper and Vitek MS Systems
Vitek MS (CU database) failed to identify four isolates of C. neoformans and both systems failed to identify two other Cryptococcus species. For isolates of Trichosporon, the Vitek MS (IVD database) system also failed to accurately identify all of the isolates to the species level, including T. asahii. Both systems could adequately identify all S. cervevisiae isolates. The Bruker Biotyper but not the Vitek MS (IVD database) could correctly identify all isolates of P. wickerhamii, Lodderomyces elongisporus, and Sporobolomyce salmonicolor (Table 1).
Using the Vitek MS plus SARAMIS system (RUO database) to 89 initially misidentified yeast isolates, 39 isolates, including 27 of C. orthopsilosis, were correctly identified ( Table 3). All the six C. orthopsilosis isolates in the 200 initially evaluated isolates were correctly identified by Vitek MS plus SARAMIS system RUO cluster 1 (C. parapsilosis), cluster 2 (C. orthopsilosis), and cluster 3 (C. metapsilosis), which were generated by ClinProTools with the genetic algorithm, are 3380.19 m/z, 4,993.07 m/z, 5383.92 m/z, 5411.78 m/z, 5582.74 m/z, 5668.24 m/z, 7014.07 m/z, 7070.65 m/z, 7212.67 m/z, and 7326,88 m/z. Three signals 4,993.07 m/z, 5411.78 m/z, and 7070.65 m/z were observed only in the spectrum for cluster 1 (all were C. parapsilosis) spectra, three signals 3380.19 m/z, 5383.92 m/z, and 7014.07 m/z were observed only in the spectrum for cluster 2 (all were C. orthopsilosis) spectra, and four signals 5582.74 m/z, 5668.24 m/z, 7212.67 m/z, and 7326.88 m/z were observed only in the spectrum for cluster 3 (all were C. metapsilosis) spectra. The absolute intensities of the ions are shown on the y axis, and the masses (m/z) of the ions are shown on the x axis. The m/z values represent the massto-charge ratio. (C) Distribution of identification score values among three species of C. parapsilosis complex by MALDI-TOF Bruker Biotyper system. doi:10.1371/journal.pone.0109376.g001 databases. This resulted in an increase in the rate of correct identification of all Candida species (94.3%, [134/142]) and all yeast isolates (87.5% [175/200]) tested by Vitek MS plus SARAMIS system RUO databases. There was no significant difference in the correct identification of all Candida species (P = 0.053) and all yeast isolates (P = 0.096) between Bruker Biotyper system and Vitek MS plus SARAMIS system RUO databases.
Clustering Analysis of the Spectra of C. parapsilosis Complex, C. guilliermondii Complex, and C. rugosa

Complex by the Bruker Biotyper System
The 90 isolates of C. parapsilosis complex were subjected to cluster analysis ( Figures 1A and 1B). Ten peaks in the spectra of these isolates, i.e., 3380.19 m/z, 4,993.07 m/z, 5383.92 m/z, 5411.78 m/z, 5582.74 m/z, 5668.24 m/z, 7014.07 m/z, 7070.65 m/z, 7212.67 m/z, and 7326.88 m/z which were generated by ClinPro-Tools with the genetic algorithm, were used to define clusters 1, 2, and 3 ( Figures 1A and 1B). Three signals, namely 4,993.07 m/z, 5411.78 m/z, and 7070.65 m/z, were observed only in the cluster 1 spectrum (all were C. parapsilosis), three signals, namely 3380.19 m/ z, 5383.92 m/z, and 7014.07 m/z, were observed only in the spectrum for cluster 2 (all were C. orthopsilosis), and four signals, namely 5582.74 m/z, 5668.24 m/z, 7212.67 m/z, and 7326,88 m/z, were observed only in the cluster 3 spectrum (all were C. metapsilosis).
As for the 18 isolates of C. guilliermondii complex (13 C. guilliermondii and five of C. fermentati isolates), five signals, i.e., 3181.57 m/z, 3359.10 m/z, 5218.25 m/z, 6717.05 m/z, and 6732.14 m/z were used to define clusters I and II (Figures 2A  and 2B). Three signals, namely 3181.57 m/z, 5218.25 m/z, and 6732.14 m/z, were observed only in the spectrum for cluster I (all were C. guilliermondii) and two signals, 3359.10 m/z and 6717.05 m/z, were observed only in the spectrum for cluster II (all were fermentati) (Figures 2A and 2B).
As for four C. rugosa complex isolates, one C. pararugosa isolate was successfully identified by the Bruker Biotyper system but not by Vitek MS (IVD or RUO database). There were six characteristic peaks used to differentiate group I (C. rugosa) and group II (C.

Discussion
Several important findings were demonstrated in this study. First, among the four systems evaluated, only the Bruker Biotyper system could differentiate among three species of C. parapsilosis complex and none of the systems could differentiate between the two species of C. guilliermondii complex. Second, several characteristic peaks in the spectra by cluster analysis could be used to differentiate between species of the two complex isolates. Third, the performance of identification of Candida isolates or of all yeasts isolates by Vitek MS plus SARAMIS system RUO databases was similar to that by Bruker Biotyper system. Finally, the performance of the Phoenix 100 YBC system was better than that of the Vitek 2 Yeast ID system at identifying Candida species and all yeasts tested. Both systems had similar results using BAP and SDA media.
The results of the Phoenix Yeast ID system for species-level identification of yeasts (n = 250) and those of the Vitek 2 YST card system were compared with the results of ribosomal ITS sequencing analysis [15]. For all isolates, 94.4% and 89.6% of the isolates were correctly identified by BD Phoenix and Vitek 2, respectively. Considering only the species included in each system's database (n = 245), 96.3% and 91.4% of the isolates were correctly identified by BD Phoenix and Vitek 2, respectively [15]. However, only two isolates of C. guilliermondii complex were studied in their study and one was misidentified by Vitek 2 system [15]. In this study, failure of accurate identification of C. guilliermondii complex, R. mucilaginosa and C. neoformans isolates (both BAP and SDA) was the main reason of the poorer performance of Vitek 2 Yeast ID system than that of Phoenix 100 YBC system. R. mucilaginosa has also been recognized as an emerging fungal pathogen causing invasive infection [19,20].
Mancini et al recently compared the performance of the Bruker Biotyper system with that of the Vitek MS system for identification of yeasts [16]. The rate of correct identification to the species level was comparable using the commercial databases (89.8% versus 84.3%; P = 0.712). Importantly, the rate of misidentification was significantly higher for Vitek MS (1% versus 12.1%; P,0.0001). Our results demonstrated a similar performance by Vitek MS plus SARAMIS system RUO databases as compared with the identification rate reported by Mancin et al [16].
Clinically, separation of three species among C. parapsilosis complex isolates is important because of the differences in virulence and antimicrobial susceptibilities [6,21]. C. orthopsilosis had a similar behavior to C. parapsilosis, whilst C. metapsilosis seems to possess a reduced virulence potential [21]. L. elongisporus has been recognized as a yeast species closely related to C. parapsilosis. Ahmad et al. showed that currently available phenotypic identification methods (e.g. Vitek 2 system and API 20C) frequently misidentify isolates of L. elongisporus as C. parapsilosis [22]. However, recent data on small-subunit rRNA gene sequencing have shown that L. elongisporus and C. guilliermondii should not be included as members of C. parapsilosis complex and C. guilliermondii complex, respectively [22,7,23]. Chen et al demonstrated that C. guilliermondii was more commonly isolated from bloodstream infections than C. fermentati [40 of 47 (85%) versus 2 of 5 (40%), P = 0.043], although C. fermentati constituted about 10% of C. guilliermondii complex isolates [7,23]. Moreover, P. wickerhamii and S. salmonicolor have been reported to cause invasive infection in humans [24,25].
Westblade et al reported that, based on sequence analysis of the D2 region of the 26S rRNA gene, 97.8% of 852 medically important yeast isolates could be identified to the species level by Vitek MS [26]. Twenty-four isolates (2.8%) were not identified and five isolates (0.6%) were misidentified [26]. Misidentified isolates included one isolate of C. albicans identified as C. dubliniensis, one isolate of C. parapsilosis identified as C. guilliermondii complex spectra, i.e., cluster I (C. fermentati) and cluster II (C. guilliermondii). B. The five peaks used to define cluster I (C. fermentati) and cluster II (C. guilliermondii), which were generated by ClinProTools with the genetic algorithm, are five signals, i.e., 3181.57 m/z, 3359.10 m/z, 5218.25 m/z, 6717.05 m/z, and 6732.14 m/z. The signals of 3181.57 m/z, 5218.25 m/z, and 6732.14 m/z were observed in cluster I spectra but not in cluster II spectra, and those of 3359.10 m/z and 6717.05 m/z were observed in cluster II spectra but not in cluster I spectra. The absolute intensities of the ions are shown on the y axis, and the masses (m/z) of the ions are shown on the x axis. The m/z values represent the mass-to-charge ratio. doi:10.1371/journal.pone.0109376.g002 pelliculosa, and three isolates of Geotrichum klebahnii identified as Geotrichum candidum. The rate of correct identification of C. guillermondii complex (n = 36) was 97.2% [26]. Our results partly support previous findings [26][27][28].
In the present study, one C. pararugosa isolate and three C. rugosa isolates were successfully identified by Phoenix 100 YBC (SDA) and Bruker Biotyper. Both the Vitek 2 Yeast ID system (both BAP and SDA) and Vitek 2 MS (IVD and RUO) failed to accurately identify the C. pararugosa isolate, even though the Vitek MS (RUO database) contains the spectra of C. rugosa and C. pararugosa [8]. C. rugosa is a poorly known fungal species in humans [8,29,30]. Recently, C. rugosa was characterized as a species complex comprising four taxa: C. rugosa sensu stricto, C. pseudorugosa, C. neorugosa and C. mesorugosa [8,[29][30][31]. A molecular analysis of the sequences of the D1/D2 domains and the ITS region of the ribosomal genes of 24 clinical isolates phenotypically identified as C. rugosa found that 41.6% were identified as C. pararugosa [29]. The true prevalence, susceptibility profiles, and clinical roles of these species have not been well elucidated, although C. rugosa was reported to be less susceptible to caspofungin compared with other C. rugosa complex species [8,[29][30][31].
Several investigators have demonstrated that MALDI-TOF MS could accurately identify all species in C. neoformans complex and could differentiate between C. neoformans and C. gattii [26,32]. Among the 17 isolates of C. neoformans evaluated in our study, all were correctly identified by Bruker Biotyper and Phoenix, and 88.2% by Vitek 2 (SDA). Three isolates of C. neoformans complex were not identified by Vitek MS using RUO databases. Although the C. neoformans complex isolates tested in this study were not specified to the species level, a previous study showed that only 1% of C. neoformans complex isolates were C. neoformans var. gattii (C. gattii) [33].
In this study, several isolates of commonly encountered yeasts were not correctly identified by either Phoenix 100 YBC or Vitek 2 Yeast ID systems on BAP or SDA, particularly C. guilliermondii complex and C. neoformans by Vitek 2 Yeast ID system on both BAP and SDA, C. krusei by Vitek 2 Yeast ID system on BAP, and T. asahii by the Phoenix system on BAP or by the Vitek 2 Yeast ID system on SDA. These misidentifications might have a great impact on clinical management. Two isolates of C. nivariensis, a rare but emerging and multidrug-resistant pathogenic yeast [34], were accurately identified by the Bruker Biotyper system. Two C. zeylanoides isolates were not correctly identified by either of the systems but one was accurately identified by the Bruker biotyper and Vitek 2 MS (IVD database) systems. This organism has been reported to be associated with bloodstream infections [4].
In this study, using cluster analysis and external validation processes, several characteristic peaks were demonstrated to be useful for differentiating among the three species of C. parapsilosis complex. Most importantly, this study is the first to report using characteristic MALDI-TOF MS spectra to separate successfully between C. guilliermondii and C. fermentati.
In summary, we evaluated the performance of two commercial biochemical identification systems and two MALDI-TOF MS systems for identifying common and rare yeasts. Differences in performance do exist among these systems; however, the differences are not always found in other reports from other laboratories. Regional variations in species distribution of yeasts, various versions of applied database systems, and inconsistent preparation processing of yeasts might contribute to these differences. In clinical microbiology laboratories, MALDI-TOF MS systems present a promising alternative for the identification of not only clinically commonly encountered yeast species, but also several species belonging to C. parapsilosis complex, C. guilliermondii complex, and C. rugosa complex, and other clinically less encountered yeasts. Figure 3. Spectra of C. rugosa complex isolates. A.Two groups of C. rugosa complex spectra, i.e., group 1 (C. rugosa) and group II (C. pararugosa), generated by MALDI-TOF Bruker Biotyper results. B. The six characteristic peaks used to differentiate group I (C. rugosa) and group II (C. pararugosa): i.e. 3050.6 m/z, 3462.3 m/z, 5409.0 m/z, 6102.2 m/z were observed in group I spectra but not in group II spectra, and those of 3131.9 m/z and 6363.4 m/z were observed in group II spectra but not in group I spectra. The absolute intensities of the ions are shown on the y axis, and the masses (m/z) of the ions are shown on the x axis. The m/z values represent the mass-to-charge ratio. doi:10.1371/journal.pone.0109376.g003