Matrix-Assisted Laser Desorption Ionization: Time of Flight Mass Spectrometry-Identified Models for Detection of ESBL-Producing Bacterial Strains

Background

Antibiotic resistance is a critical public health problem around the world, especially the spread of multi-drug-resistant organisms (MDRO), and has led to increased infected patient mortality. Many of these increases are due to deficiency of available antibiotics. The number of reported cases of MDRO nearly quadrupled in the past decade and the World Health Organization has declared that MDRO is among the top 3 threats to human health [1–3]. Extended-spectrum beta-lactamases (ESBL)-producing gram-negative bacteria are the most common MDRO and can hydrolyze beta-lactam antibiotics such as penicillin, cephalosporin, and monobactams. ESBL enzymes are mainly mediated by the conjugative plasmids and transferred between bacteria. So far, more than 100 groups of ESBL have been identified around the world and 3 of the major groups are TEM, SHV, and CTX-M. These enzymes are most commonly produced by Klebsiella spp. and Escherichia coli. They have also been found in other Enterobacteriaceae bacteria and some non-fermenters [4,5].

The infection rate of ESBL-producing bacteria is probably underestimated because these bacteria often remain undetected by routine testing methods. Kirby-Bauer disk diffusion method is most commonly used to detect ESBL in many clinical microbiology laboratories. However, ESBL strains might show a false sensitive zone of inhibition in this method, which is time-consuming and cannot provide timely diagnostic information for anti-infective therapy [6]. Another common method for ESBL detection is a molecular test, which can provide faster results than culture, but its use is limited by high cost [7–9].

In recent years, matrix-assisted laser desorption/ionization time of flight mass spectrometry (MALDI-TOF-MS) has already been used in clinical diagnosis and medical research fields, including discovery of cancer biomarkers, diagnosis of bacterial infection, identification of mutations and genotypes of viruses, and detection of antibiotic resistance [10–20]. MALDI-TOF-MS technique is more precise, rapid, and cost-effective than the traditional methods, and at present the main application of this technology is bacterial identification in clinical microbiology laboratories [18,21]. Recently, MALDI-TOF-MS has emerged as a fast and accurate technology for the detection of antibiotic resistance. In this study, we assessed the feasibility of the use of mass spectrometry for detection of antibiotic-resistant strains, and found some specific mass peaks to set up algorithm models to discriminate ESBL-producing and non- ESBL-producing strains.

Material and Methods

BACTERIAL STRAINS:

All bacterial strains were isolated from clinical samples from 302 People’s Liberation Army (PLA) hospitals in Beijing, China. A total of 69 clinical isolates were tested by MALDI-TOF-MS, including 44 E. coli species (18 ESBL-producing and 26 non-ESBL-producing) and 25 K. pneumoniae species (15 ESBL-producing and 10 non-ESBL-producing). The distribution of the sources of the samples is shown in Table 1. Another 34 isolates (12 ESBL-producing and 22 non-ESBL-producing) were tested as blinded validation samples to confirm the clinical applicability of the methods. All bacterial strains were identified by the Vitek2 system and ESBLs were confirmed by Kirby-Bauer disk diffusion method. The bacteria were incubated on Columbia blood agar plates (BioMerieux, France) overnight at 37°C.

SAMPLE PREPARATION:

Cefotaxime (purchased from the National Institute for the Control of Pharmaceutical and Biological Products, Beijing, China) was dissolved in distilled water to final concentration of 0.5 mg/ml [11]. Five bacterial colonies were picked and resuspended in 10 μl of antibiotic solution. Subsequently, the solutions were incubated at 37°C under agitation for 3 h and then centrifuged for 2 min at 12 000 g at room temperature. The supernatant was removed for analysis by MALDI-TOF-MS. To confirm the presence of ESBLs, clavulanic acid at 0.05 mg/ml concentration was added to antibiotic solutions in another parallel test.

MALDI-TOF MS ANALYSIS:

We transferred 1 μl of supernatant from the incubated solution onto a 384 polished steel target plate and then samples were air-dried under a biosafety cabinet. Every dried sample was mixed with 1 μl of MALDI matrix (10 mg/ml of α-cyano-4-hydroxycinnamic acid (HCCA) dissolved in 50% acetonitrile and 2.5% trifluoroacetic acid (TFA); Bruker Daltonics, Bremen, Germany). Measurements were performed using an Autoflex MALDI-TOF MS (Bruker Daltonics, Germany) instrument. The parameter settings were: positive reflector mode (RP) in the mass range of 200 Da to 1500Da; ion source 1:20kV; ion source 2:17.5 kV; pulsed ion extraction: 120 ns; laser wavelength: 337 nm; laser frequency: 67–100 Hz; lens voltage: 6.5 kV; detector gain: 2650 V.

DATA ANALYSIS AND MODEL GENERATION:

Data analysis was performed with ClinPro Tools software (v3.0). Peaks with signal-to-noise ratio (S/N) >5 were picked out and performed the statistical analysis. Normally distributed data were analyzed with Student’s t tests and non-normally distributed data were analyzed by the Wilcoxon test. To distinguish ESBL and non-ESBL strains, 3 different machine-learning algorithms were used: genetic algorithm (GA), supervised neural network (SNN), and quick classifier (QC).

BLINDED VALIDATION OF CLINICAL SAMPLES:

To confirm the clinical applicability of the models established, a blinded validation test was conducted following the completion of the statistical analysis. A total of 34 clinical bacterial strains were collected and detected ESBLs by 2 methods (Mass spectrometry models and Kirby-Bauer disk diffusion method) simultaneously. K-B method results served as the criterion standard. Accuracy, sensitivity, specificity, positive and negative predictive values (PPV and NPV), and Youden’s index were calculated to assess the performances of the models.

Results

DISCREPANCY ANALYSIS OF MASS SPECTRA:

The hydrolysis of the cefotaxime by ESBL-producing and non-ESBL-producing strains was analyzed. The molecular peaks of cefotaxime (456 Da and 396Da) were found in the spectra. Incubation of cefotaxime with resistant strains resulted in the decrease of the molecular peaks at 456 Da and 396Da, but the peaks did not disappear completely. In addition, all spectra derived from resistant bacteria revealed increased peaks at 370 Da and 371 Da, corresponding to the hydrolyzed form of cefotaxime. Using ClinPro Tools software, the discrepancy analysis of the 4 peaks showed that all of them were significantly different (p<0.05) between ESBL and non-ESBL strains. The statistical results of the 4 peaks are shown in Table 2. To demonstrate the differences visually, statistical plots of the 4 peaks are shown in Figures 1–4. The spectra peaks distribution maps (Figure 1) of to the 4 peaks showed that 2 groups of spots were completely separate and we could easily discriminate the ESBL from the non-ESBL strains. From the simulated 2-dimensional gel electrophoresis map and whole mass spectra map (Figures 2–4), peak intensities at 370 Da and 371 Da increased significantly in resistant strains. In contrast, the peaks at 456 Da and 396 Da were clearly reduced in resistant strains. The area under the curve (AUCs) of the ROCs for peaks 370 Da,371 Da, 456 Da, and 396 Da were 1.00, 1.00, 0.97, and 0.93, respectively (Figure 5). To rule out the possibility of the presence of non-ESBL mechanisms, clavulanic acid was added to antibiotic solutions in another parallel test. Figure 6 demonstrates that the peaks at 370 Da and 371 Da were clearly reduced in the ESBL group in the presence of clavulanic acid, but peaks at 456 Da and 396 Da were clearly increased.

ESTABLISHMENT OF ESBL STRAINS IDENTIFIED MODELS:

We used ClinPro Tools software to analyze the data and establish 3 identified models. Four individual peaks (370 Da, 371 Da, 456 Da, and 396 Da) were combined to establish the GA, SNN, and QC models. Results of identification for ESBLs by the 3 models were: GA model with a cross-validation of 90.15% and a recognition capability of 97.5%, QC model with a cross-validation of 97.62% and a recognition capability of 92.5%, and SNN model with a cross-validation of 97.62% and a recognition capability of 92.5%. From the results in Table 3 we infer that by using the 4 peaks to construct algorithm models, we could achieve around 95% accuracy rate for detecting ESBL strains. Therefore, a blinded validation study with more clinical samples should be done with the follow-up research.

BLINDED VALIDATION OF CLINICAL SAMPLES:

A total of 34 clinical isolated bacterial strains, including 12 ESBL-producing and 22 non-ESBL-producing strains, were successfully analyzed as validation samples by MALDI-TOF MS models. Among these 34 samples, 28 (82.4%) were correctly identified by GA models with 4 false-negatives and 2 false-positives, giving a sensitivity of 71.4% and a specificity of 90.0%. We correctly identified 30 (88.2%) samples by SNN models with 2 false-negatives and 2 false-positives, giving a sensitivity of 83.3% and a specificity of 90.9%. For QC model, the number of correctly identified samples was 28 (82.4%), with 3 false-negatives and 3 false-positives, resulting in a sensitivity of 75.0% and a specificity of 86.4%. The diagnostic performances of 3 algorithm models are listed in Table 4.

Discussion

MALDI-TOF-MS is a powerful tool for the detection and identification of proteins, peptides, polysaccharides, nucleic acids, and other biological molecules. It has been used in clinical diagnosis and medical research successfully since the 1980s. In recent years MALDI-TOF-MS has become a routine method for bacterial identification in clinical microbiology laboratories because of its advantages of rapid and high throughput. However, the development of antibiotic resistance research using mass spectrometry lags far behind the identification of bacteria; therefore, many bacteriologists have focused their attention on the use of mass spectrometry for investigating antibiotic resistance. Many studies have attempted to find the differences in the spectral patterns between resistant strains and non-resistant strains, but there are still no reliable “biomarkers” for identifying resistant strains [14,22–25]. In 2012, Sparbier K. et al. [11] created a novel approach to identify resistant bacteria strains by using MALDI-TOF-MS. The principle of the method is: β-lactam antibiotics were incubated with the sensitive strains and resistant strains and hydrolysis of antibiotics was detected by using MALDI-TOF-MS after incubation. The hydrolysis of the antibiotics leads to a molecular mass change, which can be easily detected by mass spectrometry. Therefore, resistant strains can be identified according to the changes in the mass spectra profiles. However, it is difficult to determine the standardization of assay evaluation.

In this study, ClinPro Tools software was used to obtain information about the changes in the mass spectra and establish mathematical models to identify ESBL strains automatically, thereby avoiding the problem of standardization of assay evaluation. We evaluated the performance of the MALDI-TOF-MS system for the detection of ESBLs from the clinical isolated gram-negative strains, including 25 K. pneumoniae strains and 44 E. coli strains. Several beta-lactam antibiotics at different concentrations were tested with 69 clinical isolates (data not shown). Cefotaxime at a concentration of 0.5 mg/ml obtained the best effect in discriminating ESBL-producing from non-ESBL-producing strains. A total of 4 molecular peaks in cefotaxime were selected to continue our data analysis – they were molecular peaks of cefotaxime (456 Da and 396 Da) and a hydrolyzed form of cefotaxime (370 Da and 371 Da). All spectra derived from resistant bacteria revealed a clear reduction of molecular peaks of cefotaxime (456 Da and 396 Da) and an increase of the peaks of the hydrolyzed form of cefotaxime (370 Da and 371 Da).

The identified models based on the 4 peaks showed similar performance and yielded a cross-validation and recognition rate of about 95% in identifying ESBL strains and can meet the needs of clinical application after further confirmation. However, despite the previous experiments showing that the cross-validation rate and recognition rate of the models for identifying ESBL strains were very high, the results of blinded validation are still not satisfactory because these models have imperfect clinical applicability at present and their accuracy rate is still only about 85%.

Conclusions

The numbers of samples tested in our study limits the reliability of the models, and results of the blinded validation study are not yet reliable enough for use in routine clinical diagnosis. Despite this, we still found some specific peaks to discriminate ESBL-producing from non- ESBL-producing strains. We established 3 identified models of ESBL-producing strains, and the results of validation were very good. Therefore, we believe that MALDI-TOF-MS should be regarded as a promising tool for the detection of ESBL-producing bacteria and will become widely used in clinical work in the future.