Accuracy of matrix-assisted LASER desorption ionization–time of flight mass spectrometry for identification of Candida

Candida is a common parasitic fungus in human respiratory, gastrointestinal and urogenital systems. It can cause candidemia by invading the tissues, and even death when human immunity decreases [1]. Candidemia is a blood infection with increased morbidity and mortality rate of up to 40% [2]. The prognosis and mortality of patients are related to the distribution of Candida species. Candidemia is the fourth common cause of death among infectious diseases of the blood system [3]. At present, among all types of Candida infections, Candida albicans is predominant, but the proportion of non-C. albicans is rising [4]. Candida glabrata and Candida krusei are less susceptible to common azole anti-fungal drugs and can develop natural resistance, which may be related to the increasing incidence of candidemia [1]. Early diagnosis is the key to successful treatment of candidemia [5]. Candida can be examined in many ways, such as direct examination, Gram staining, culture etc. But these checks are time-consuming and sometimes the results may be uncertain, which delays patient’s treatment time and aggravates his/her condition [5]. A simpler, faster and higher accuracy-test method is urgently required for clinical examination.

MALDI-TOF-MS is a new type of soft ionization mass spectrometry, with its own spectral database [6], which was introduced as a rapid method for identifying bacteria and yeast [7]. It is a powerful device for proteomic analysis [8]. Its advantages are simple and intuitive [9]. It can be used to directly identify macromolecule mixtures, without the need to separate and slice [10]. The species can be identified in three steps: first, the sample collected is placed on a specially designed metal target plate. Then the instrument conducts the measurements. Finally, the model is used to infer the species by combining spectra with well-known and/or well-defined species of spectral databases [11]. Moreover, it has high sensitivity, rapid detection [12], high throughput, large-scale identification of proteins and determination of the molecular weight of biological macromolecules [13]. Although the efficiency of MALDI-TOF-MS is well established, its accuracy needs to be urgently tested, in order to facilitate its use in clinical examination. There is no meta-analysis on the diagnostic accuracy of MALDI-TOF-MS for candidemia in evidence-based medicine. The aim of the present study was to evaluate the accuracy of MALDI-TOF-MS in the identification of Candida species, in order to provide a new means and the gold standard for clinical diagnosis.

Our study date is from December 2017 to date. A systematic review of the diagnostic accuracy of MALDI-TOF-MS in candidemia was performed, followed by a meta-analysis.

We searched the keywords ‘MALDI-TOF-MS’ and ‘Candidemia’ among four databases of Pubmed, Embase, Cochrane Library and Web of Science, and collected the articles published before February 2018. Four investigators (T.-A. Xie, Y.L. Liu, Y.Y. Huang, and C. Liang) independently screened the retrieved publications according to pre-established inclusion and exclusion criteria and extracted the data from the papers. Any differences were resolved by discussing with another researcher (J.-W. Li).

Before reviewing the articles, the researchers established the criteria for inclusion and exclusion. The inclusion criteria were formulated based on the PICOS criteria. (1) Objectives: Clinical specimens were identified as Candida or standard strains by reference methods. (2) Types of study: test of diagnostic accuracy, the data of identifying Candida species level can be extracted, limited to English language. (3) Measurement indicators: accuracy. (4) Diagnostic experimental methods: the identification of Candida by MALDI-TOF-MS. The following studies were excluded: duplicate studies, abstracts, conference summaries, case reports/reviews/posters; lack of a reference method or a detailed number of isolates. According to the category of the strain and the MALDI-TOF-MS system, the number of isolates was correctly identified and total isolates were abstracted. After preliminary screening of the documents that met the inclusion criteria, we used EndNoteX8 software for document management and extracted data from the articles, including the author name, year of publication, study design, strain distribution area, detection instrument, detection system, and strain source, in an Excel spreadsheet.

The quality of the included studies was assessed using the Quality Assessment of Diagnostic Accuracy Studies (QUADAS-2) guidelines [14]. The guidelines comprise four key domains that judge bias and applicability of the studies by reviewing how the patients were selected, index test, reference standard, and the flow of patients through the study. These variables were entered in a main data Excel spreadsheet.

The main measurement index was the correct identification rate of Candida. The correct identification rate is the ratio of the number of strains identified by MALDI-TOF-MS to the number of strains identified by the reference methods in the study. Stata 12.0 software was used to merge the correct identification rates of Candida and Candida subspecies. I² measure was used to estimate heterogeneity between studies. In the case of greater heterogeneity, subgroup analysis of the included studies was conducted to explore the impact of factors causing heterogeneity on the overall effect and the degree of impact [15]. Sensitivity analysis was used to study how the effect of Candida level data could be apportioned to the total response [16]. Egger’s test was used to evaluate the publication bias of the included studies [17]. All analyses were performed with Stata statistical software package, version 12.0 (Stata Corp LP, College Station, U.S.A.).

After a comprehensive database search, we identified 16 articles in Pubmed, 79 in Embase, 1 in Cochrane Library, 30 in Web of Science and 3 from other sources. Of the 129 references, 55 were duplicate. After reviewing the title and abstract, a total of 22 articles remained for full-text screening. Of these, two articles were excluded due to inability to extract data; five articles were discarded as duplicates, and four basic research articles were excluded. Finally, ten articles [1,2,5,7,18–23] were included and their research data were extracted for meta-analysis (Figure 1).

We extracted information such as the author name, year of publication, study design, strain distribution area, detection instrument, detection system, strain source etc., from the included studies (Table 1). The identification accuracy rate of species of these studies is summarized in Table 2.

The subjects in the ten studies were identified by conventional methods or genetic analysis before validation. The quality evaluation of the included articles was conducted (Table 3).

In the ten included articles, a total of 1854 Candida isolates were assessed. The overall statistical results of the meta-analysis at the genus level identification were summarized by forest plots of the random-effects model using Stata 12.0 software (Figure 2). Identification accuracy of C. albicans was at 100% (P=0.709; I² = 0.0%), C. glabrata was at 100% (P=0.998; I² = 0.0%), C. tropicalis was at 100% (P=0.750; I² = 0.0%), and C. parapsilosis was at 100% (P=0.755; I² = 0.0%). The overall identification accuracy was at 100% (P=0.998; I² = 0.0%).

Subgroup analysis was performed on the instruments of the collected data. We combined and compared different instruments (Bruker; bioMérieux). In studies using Bruker to identify C. albicans, identification accuracy of C. albicans was at 100% (P=1.000; I² = 0.0%). In studies using bioMérieux to identify C. albicans, identification accuracy of C. albicans was at 99% (P=0.097; I² = 63.6%) (Figure 3).

In studies using Bruker to identify C. glabrata, identification accuracy of C. glabrata was at 100% (P=1.000; I² = 0.0%). In studies using bioMérieux to identify C. glabrata, identification accuracy of C. glabrata was at 100% (P=0.299; I² = 7.1%) (Figure 4).

In studies using Bruker to identify C. tropicalis, identification accuracy of C. tropicalis was at 100% (P=1.000; I² = 0.0%). In studies using bioMérieux to identify C. tropicalis, identification accuracy of C. tropicalis was at 100% (P=0.305; I² = 5.0%) (Figure 5).

In studies using Bruker to identify C. parapsilosis, identification accuracy of C. parapsilosis was at 100% (P=1.000; I² = 0.0%). In studies using bioMérieux to identify C. parapsilosis, identification accuracy of C. parapsilosis was at 100% (P=0.050; I² = 74.0%) (Figure 6).

In studies using Bruker to identify Candida, identification accuracy of Candida was at 100% (P=0.000; I² = 90.3%). In studies using bioMérieux to identify Candida, identification accuracy of Candida was at 97% (P=0.962; I² = 0.0%) (Figure 7).

The influence of the included articles on the total pooled efficacy was investigated using the sensitivity analysis. The sensitivity analysis refers to a new meta-analysis conducted every time a certain study is deleted, in which the combined effect is compared with the overall effect to detect any change in the results. The vertical solid line of 0.97 in the middle represents the overall combined effect. The left and right vertical solid lines represent the upper and lower limits of the 95% confidence interval of the total pooled effect (Figure 8).

Statistically significant results are more likely to be accepted and published in similar studies than non-statistically significant studies. The control of publication bias is difficult and influences the results of systematic evaluation. The P-values of the funnel chart and Egger’s test were used for the evaluation of publication bias in the present study. The combined results of Candida (t = −2.04; P=0.076) are shown in Figure 9.

In the present study, we focused on the importance of rapid identification of Candida by MALDI-TOF-MS. The accuracy of the MALDI-TOF MS for clinical Candida isolates identified by the gold standard method was considered to be in accordance with the conditions of the meta-analysis. The results illustrated that the overall identification accuracy was 100% (P=0.998; I² = 0.0%), indicating that MALDI-TOF-MS is very accurate in identifying Candida.

In our study, we mainly investigated four kinds of common clinical Candida species (C. albicans, C. glabrata, C. tropicalis, C. parapsilosis). Due to the more frequent use of azole anti-fungals and invasive procedures, other Candida species have been increasing [24]. In the articles we extracted, the accuracy of identification of most of the rare yeasts was very high. For example, C. guilliermondi (100%), C. krusei (100%), C. dubliniensis (100%), and so on. But there were a few exceptions, like C. dubliniensis (2/3) [23] and C. rugosa (6/7) [18]. Because of too few samples, in order to ensure scientific preciseness and avoid excessive heterogeneity, we cannot verify the accuracy of MALDI-TOF-MS in these rare Candida species.

In subgroup analysis, we compared the accuracy of two MALDI-TOF-MS systems in identifying Candida species. Eight of the ten articles were Bruker MALDI Biotyper system, and 2114 strains were studied. The accuracy of the Bruker MALDI Biotyper System in identifying Candida was 100%. Two were bioMérieuxVitek MS system, and 576 strains were studied. The accuracy of bioMérieuxVitek MS system in identifying Candida was 97%.

The results of instrument subgroup analysis of collected data indicated differences in identification capabilities of the two commercial MALDI-TOF-MS systems (Bruker MALDI Biotyper and bioMérieuxVitek MS). According to the results of bioMérieuxVitek MS detection of different subspecies of Candida, the overall heterogeneity of Candida was higher than that of Bruker MALDI Biotyper, especially for C. albicans and C. parapsilosis. The I² values of C. albicans and C. parapsilosis detected by bioMérieuxVitek MS were 63.6 and 74.0%, respectively, suggesting moderate heterogeneity. However, the I² of different subspecies detected by Bruker MALDI Biotyper was zero, indicating no heterogeneity.

Two of the ten included articles used the bioMérieux Vitek MS system. Analysis of these two articles revealed several factors that can explain the observed heterogeneity. We found that Li et al. [23] mentioned that several isolates were identified as ‘bad spectrum during acquisition’ in the discussion of the article. We think this may affect the results of mass spectrometry to some extent. Besides, the database of VITEK MS used in the present study wasin vitro devices (IVD), not the VITEK MS database for research only (RUO), which would lead to discrepant results. In addition, many factors, such as the difference between Candida spp., the reference method, were incorrectly identified, the protein profiles of Candida collected in the database were incomplete, the strains used in the study were not identified by a unified gold standard, the ease of polymicrobial bloodstream infections when directly identifying the Candida from blood culture, manmade operation level etc., may lead to heterogeneity, thus reducing the efficiency of the study.

The results of the present study illustrated that the level of Candida was distributed on both sides of the axis (0.97) and did not exceed the 95% confidence interval (0.92–1.02). No single study result affected the total combined effect. In Egger’s test, P>0.1 indicates no publication bias. The results of the present study revealed P=0.076 suggesting minor publication bias, which is permissible in a meta-analysis.

Although classical molecular identification methods continue to be used in clinical diagnosis, MALDI-TOF-MS is increasingly used in clinical microbiology laboratories and has become a gold standard [25,26]. The MALDI-TOF test procedure was completed in approximately 13 min, while the conventional identification method required 24–48 h [5]. According to the data collected in this study, the accuracy of MALDI-TOF-MS reached 100%. The most frequently isolated bloodstream Candida species include C. albicans, C. tropicalis, C. glabrata, and C. parapsilosis, which accounted for 67.9% of our total collection. The correct identification rate of the four subspecies was 100%. The identification accuracy of other subspecies did not reach 100%, and the organisms that were not identified were not in the database library. The limitations of MALDI-TOF-MS are reflected in the number of species in the database. There is a disagreement between those who want to continue adding new species and those who believe that the current database already has sufficient clinical coverage. However, some specific Candida species, such as C. auris, cannot be identified by Bruker Biotyper and can only be correctly identified when using a library containing C. auris [27].

Our study evaluated the accuracy of MALDI-TOF-MS in the identification of four most frequently isolated bloodstream Candida species, which can provide a new means and the gold standard for clinical diagnosis. In summary, MALDI-TOF-MS has proven to be a reliable and rapid method for identification of four most frequently isolated bloodstream Candida species. Further research is necessary to determine whether MALDI-TOF-MS can be used as a clinical diagnostic standard for the identification of Candida.

T.-X.J., X.-Y.L., S.O., and X.-G.G. conceived and designed the experiments. T.-A.X., Y.-L.L., C.L., and Y.-Y.H. analyzed the data and made the tables. T.-A.X., J.-W.L., Z.-W.L., S.-J.F., and J.-T.C. contributed to the production of figures by the analysis tools. All authors participated in the writing, reading, and revising of the manuscript and approved the final version of the manuscript.

The authors declare that there are no competing interests associated with the manuscript.

This work was supported by the Guangzhou Medical University [grant number 2017A046]; The Third Clinical School of Guangzhou Medical University [grant number 2018B088]; the grants from Natural Sciences Foundation of GuangDong Province [2015A030313684]; and the National Nature Science Foundation [grant numbers 81673040, 81700004 and 81803884].

PICOS

Participant Intervention Comparison Outcome and Study

QUADAS-2

Quality Assessment of Diagnostic Accuracy Studies

RUO

research only