3.2 FTIR-S analysis
After analyzing different spectral ranges, infrared absorbance in the lipid region (1400-1500 and 2800-3000 cm-1) showed the highest discriminatory power for the correct classification of P. aeruginosa outbreak strains. HCA using Euclidean distance and Ward linkage differentiated the outbreak strains (ST175) from control strains, which belonged to different STs such as ST227, ST253, ST381, ST557 and ST885 (Figure 3), with similar discriminatory power to PFGE classification. In this case, the cut-off dendrogram distance for outbreak and controls differentiation was 0.350. For isolate differentiation, the automated cut-off distance assigned by IR Biotyper was 0.062. The two strains belonging to the same patient were clustered together with lower distance between them. Using PCA analysis in both IR Biotyper software and Clover MS Data Analysis software, ST175 strains were clustered together, distinctly from other STs (Figure 4A and 4B). The specific differences in infrared spectra absorbances among ST175 and other STs were found in 1445 cm-1 (Figure 5A), 2925 and 2955 cm-1 (Figure 5B). The three subgroups inside the outbreak observed by WGS were not discriminated by HCA (Figure 3). In the PCA scatter plot (Figure 4C and 4D), although WGS Group 2 was partially separated from other groups, Groups 1 and 3 overlapped. Using PLS-DA and RF algorithm (Figure S4), it was also showed that Group 2 formed a distinct cluster, and the k-fold cross validation obtained for differentiation of the three groups were 88.9% and 97.8%, respectively (Table S4).
DISCUSSION
The implementation of MALDI-TOF MS and FTIR-S technology, combined with Machine Learning algorithms, allowed the correct classification of the MDR P. aeruginosa isolates causing a nosocomial outbreak in 2013-2014: 97.0% and 100% of the isolates were correctly classified by the algorithms applied to the protein peaks by MALDI-TOF MS with intensities above the established threshold (0.1) or to the specific biomarker peaks found with this technology, respectively. These results showed that MALDI-TOF MS yields a discrimination power similar to PFGE, the reference method for bacterial typing. Besides, the same protein spectra were further classified according to the information provided by WGS analysis. In this case, PLS-DA and SVM algorithms allowed a good classification of the P. aeruginosa isolates specifically correlated with the outbreak (Group 1) and showed that Group 2 was closely related with the outbreak strains, as the genomic analysis pointed out, and Group 3 and the control group were clearly unrelated to the outbreak. The validation of the classification models, carried out with 32 P. aeruginosa isolates characterized by ASO-PCR, yielded 90.0% correct classification of their protein spectra. Only three isolates were misclassified using the developed models, which indicate that the methodology described in this study may be applied as a rapid screening method when an outbreak is suspected. The implementation of FTIR-S technology showed the same discrimination power as PFGE to differentiate P. aeruginosa outbreak isolates. However, when the same level of classification provided by WGS technology was attempted with FTIR-S, the control group and outbreak Group 2 were clearly defined but Groups 1 and 3 overlapped. However, FTIR-S provided complementary information to the classification obtained with MALDI-TOF MS spectra by showing correct classification of the different STs analyzed.
Previous studies have demonstrated that specific biomarker peaks present in the protein spectra obtained by MALDI-TOF MS could be used for the monitoring of P. aeruginosa sequence types. Cabrolier et al. described a specific peak at 7359 m/z that, combined with the absence of peaks at 7329 and 12154 m/z were specific of P. aeruginosa ST175 (Cabrolier et al., 2015). These results were further confirmed by Mulet et al., who also found the peak at 7359 m/z as a biomarker for P. aeruginosa ST175 and described another peak at 6911 m/z as specific for this ST (Mulet et al., 2021). In the present study, a peak at 6915 m/z has been described as a biomarker of the P. aeruginosa outbreak strains belonging to the same sequence type. Although the difference of only 4m/z between both peaks fall within the accepted margin of error of the applied pipeline (± 4.5 m/z ) and no other marker peak has been found in the area, further analysis is needed to confirm that both studies refer the same peak. Finally, the other two biomarker peaks described for the P. aeruginosa outbreak strains (5169 and 7236m/z ) have never been described before and, therefore, they might be outbreak-specific markers.
The reproducibility of this MALDI-TOF MS based method for the rapid detection of a P. aeruginosa outbreak was evaluated. With the exception of the peak at 5740 m/z , whose inter-day CV was just below 30.0%, the peaks located between 2000 and 7000 m/z -where the most common bacterial proteins locate-, showed CV values ranging between 10.0% and 15.0% and have been considered as acceptable in previous studies (R. T. K. Pang et al., 2004). The specific biomarker peaks that allowed the differentiation of the outbreak isolates showed an average intra-day CV of 13.6% (range between 7.0 and 20.5%). Again, these values can be considered as acceptable. Only the biomarker peak at 6915 m/z showed an inter-day CV variability CV of 32.5% (Figure S3D). Although this value is above the established limit (20.0%), the peak was always detected inP. aeruginosa outbreak strains regardless its intensity. Therefore, the presence of the 6915 m/z peak can be reliably correlated with the outbreak along with the detection of the other two biomarker peaks at 5169 and 7236 m/z .
In recent years, FTIR-S has emerged as a reliable technology for outbreak analysis in clinical microbiology laboratories (Quintelas, Ferreira, Lopes, & Sousa, 2018). The simplicity and low costs of the sample preparation procedure and interpretation of results allows the follow-up of nosocomial outbreaks in real time since the turnaround time for the analysis of 30 isolates with this technology is approximately 3 hours. Although the most studied microorganism so far is Klebsiella pneumoniae(Rakovitsky et al., 2020), FTIR-S has been evaluated for typing other bacterial species such asSalmonella (Cordovana et al., 2021), or Streptococcus pneumoniae(Burckhardt et al., 2019).
In our study, FTIR-S was able to discriminate P. aeruginosaoutbreak isolates from non-outbreak isolates at the same level than PFGE, either by HCA with a cut-off score of 0.350 or PCA (Figure 3 and Figure 4). In addition, the two strains isolated from the same patient (numbers 139 and 143) were clustered together with very low distance, which indicates that this method recognizes them as the same strain and is reproducible (Figure 3). For ST differentiation, the lipid region (1400-1500 and 2800-3000 cm-1) showed the highest discriminatory power and allowed the correct classification of P. aeruginosa outbreak strains according to PFGE results. However, when this classification was compared with WGS information, the three outbreak subgroups were not clearly differentiated by FTIR-S by applying HCA. Similarly, the implementation of PCA algorithm did not cluster the three groups of isolates separately, although Group 2 isolates were almost grouped apart (Figures 4C and 4D). When using PLS-DA and RF analyses for differentiation of these groups, it was also observed that Group 2 was clearly separated from the other outbreak groups according to WGS and the biggest differences among them were found in the 800-1600 cm-1 region (Figure S5). It is important to note that WGS groups are based on SNP distance, and maybe these differences are not expressed phenotypically, and thus, FTIR-S clusters may not reflect differences detected by WGS. At the moment of writing, only one previous study has evaluated P. aeruginosa typing by FTIR-S in comparison to PFGE and MLST results (Martak et al., 2019). The authors showed the reliability of the method for differentiation of STs with an optimal cut-off distance between 0.184 and 0.374. Besides, these results can be obtained in a turnaround time of 3h, a great advantage over PFGE.
One of the limitations of this study is the small number of strains available, but it is important to acknowledge that bacterial outbreaks usually involve a limited number of patients if they are well contained, thus making larger number of samples unavailable for research purposes. The other limitation would be the limited number of isolates characterized by WGS. But, due to the costs of this technique, a more affordable approach (ASO-PCR) was carried out to classify the rest of the isolates.
However, despite these drawbacks, the results of this study showed that rapid diagnostic methods such as MALDI-TOF MS and FTIR-S may represent fast alternatives to conventional strategies -based on DNA sequencing- for real-time monitoring of nosocomial outbreaks, providing complementary information for the prompt characterization of the suspected isolates in a cost-efficient way. Although confirmation of the outbreak strains may request further analysis by WGS, the implementation and further validation of these rapid typing methods could help to reduce the number of isolates that require confirmation by expensive tests available to a limited number of microbiology laboratories.