Discrimination of clinically relevant Candida species by Fourier-transform infrared spectroscopy with attenuated total reflectance (FTIR-ATR)

Abstract

Accurate Candida species identification remains a challenge due to their phenotypic and genotypic similarity. Species belonging to the *psilosis group, are even described as phenotypically indistinguishable. Also, most of the genotypic methods commonly used to discriminate these species are laborious and very expensive. In this work we developed a Fourier-transform infrared spectroscopy with attenuated total reflectance (FTIR-ATR) based method as a reliable alternative for the discrimination of 12 Candida species. The collection comprises 82 clinical isolates obtained from distinct biological products, recovered between 2007 and 2014 in Portugal and Brazil and previously characterised by CHROMagar Candida and PCR-based sequencing techniques. Infrared spectra were analysed with principal component analysis (PCA) and partial least squares discriminant analysis (PLSDA). The results demonstrated that the 12 species could be successfully discriminated using the proposed infrared spectroscopy based method. Noteworthy, the PLSDA model led to the correct identification of 99.6% of the analysed clinical isolates. This rapid, low cost, and environmental friendly technique proved to be a reliable alternative for the identification of Candida species that share many phenotypic and genotypic characteristics and are often difficult to distinguish.

---

1. Introduction

Candida genus, Fungi kingdom and Deuteromycetes class, currently comprises more than 350 heterogeneous species.^1,2 Candida species are mostly considered commensal yeasts though they can also act as opportunist pathogens causing candidiasis.^3,4 The incidence of Candida infections has been increasing in recent years due to the widespread use of broad-spectrum antifungal agents and the growing numbers of HIV-infected and immunocompromised individuals.^3,5 Despite the predominance of Candida albicans, non-Candida albicans Candida (NCAC) species such as Candida glabrata, Candida tropicalis, Candida parapsilosis, Candida krusei, Candida guilliermondii, Candida dubliniensis, Candida lusitaniae and Candida bracarensis are emerging both as colonizers and pathogens causing superficial and/or systemic infections.^1–4 These species possess specific infectious potential, antifungal susceptibility and mortality rates.³ Additionally, due to its genotypic and phenotypic similarity identification can be a cumbersome task using the existing typing methods.⁶ Several genotypic methods have been proposed for Candida species identification being the most popular the polymerase chain reaction (PCR) and sequencing of species-specific DNA regions. However, these methods are laborious, time-consuming and expensive.^6,7 In this context, a rapid, low cost and reliable method for Candida species identification will play a fundamental role in infection management, helping to direct the therapy to the patient needs and thus decreasing mortality rates and the economic burden associated to these infections. The development of infrared spectrometers such as Fourier-transform infrared spectroscopy with attenuated total reflectance (FTIR-ATR) and the availability of sophisticated mathematical tools led to a growing interest in the application of this spectroscopy type in the microbiology field. This technique has been successfully applied for bacterial typing at different taxonomic levels^8–10 but barely explored in yeasts with only a few published studies thoroughly dedicated to Candida species discrimination.^11–13 Of note, Timmins et al.¹¹ showed the ability of this technique to discriminate isolates of. C. albicans, Candida stellatoidea and C. dubliniensis;¹¹ Tintelnot and co-workers¹² successfully discriminate C. albicans from C. dubliniensis and Essendoubi et al.¹³ discriminated a collection of Candida isolates comprising 6 different species (C. albicans, C. glabrata, C. parapsilosis, C. tropicalis, C. krusei and Candida kefir). Other studies proved that FTIR were able to discriminate isolates from different patients,^14–16 to assess structural differences in C. albicans blastospores and hyphae¹⁷ and to characterise the exopolyssacharide of biofilm-forming C. albicans isolates.¹⁸ Despite the relevance of the available results pointing to the reliability of FTIR as a potential typing method for Candida species^11–13 these studies included few species and/or isolates neither including very closely related species as those belonging to the *psilosis group or C. glabrata, Candida nivariensis and C. bracarensis complex. The main goal of this work was to develop a combined FTIR-ATR and chemometrics method to discriminate the most relevant pathogenic Candida species (C. albicans, C. glabrata, C. krusei, C. parapsilosis and C. tropicalis). Additionally, other species, as C. lusitaniae, Candida famata, C. dubliniensis, Candida orthopsilosis, Candida metapsilosis, C. bracarensis and C. guilliermondii, were considered due to their increasing relevance and their phenotypic and genotypic similarity with the previous ones. To the best of our knowledge this is the first attempt to successfully discriminate a collection of Candida isolates belonging to twelve distinct species some of them very closely related and often difficult to distinguish by the current molecular biology based typing methods.

2. Material and methods

2.1. Rationale of the study

In this work, a supervised infrared based chemometric model (PLSDA) was developed to discriminate a collection of 63 clinical isolates belonging to 5 Candida species (C. albicans, C. glabrata, C. krusei, C. parapsilosis and C. tropicalis). These species were selected due to their clinical relevance and to the number of available isolates to develop a robust supervised model (minimum of 10 isolates per species). Additionally, 19 clinical isolates belonging to 7 Candida species (C. lusitaniae, C. famata, C. dubliniensis, C. orthopsilosis, C. metapsilosis, C. bracarensis and C. guilliermondii) were also included due to their increasing relevance and their phenotypic and genotypic similarity with the previous five species allowing strengthening the hereby proposed methodology performance. Due to the reduced number of isolates of each of these second set of species, only unsupervised models (PCA) were developed. Considering the genotypic and phenotypic similarity among all species, three distinct PCA models were developed to discriminate: (1) C. parapsilosis, C. orthopsilosis and C. metapsilosis; (2) C. albicans, C. tropicalis, C. famata, C. guilliermondii, C. lusitaniae and C. dubliniensis and (3) C. bracarensis and C. glabrata.

2.2. Candida clinical isolates

A total of 82 clinical isolates (Table ESI 1†) of C. albicans (n = 13), C. glabrata (n = 13), C. krusei (n = 11), C. parapsilosis (n = 14), C. tropicalis (n = 12), C. famata (n = 2), C. dubliniensis (n = 3), C. orthopsilosis (n = 2), C. metapsilosis (n = 4), C. lusitaniae (n = 4), C. bracarensis (n = 2) and C. guilliermondii (n = 2) were included in this study. The isolates were recovered from distinct biological products (vaginal exudate, urine, blood and sputum) in one Brazilian (Regional University Hospital of Maringá) and two Portuguese hospitals (Hospital of Braga, Braga and Hospital of S. João, Porto) from 2007–2014.^19,20 All isolates were previously identified by CHROMagar Candida (CHROMagar, Paris, France) and PCR-based sequencing using specific primers (ITS1 and ITS4) targeting an intergenic spacer region of the 5.8S ribosomal RNA subunit gene.⁷ The isolates were stored at −80 °C in a glycerol/LB medium 30%/70% (vol/vol).

2.3. FTIR-ATR experiments

Candida isolates were cultured on Sabouraud's dextrose agar (37 °C, 18 h) and colonies were directly transferred from the agar plates to the ATR crystal. Infrared spectra were acquired using a PerkinElmer Spectrum BX FTIR System spectrophotometer with a PIKE Technologies Gladi ATR accessory from 4000 to 600 cm⁻¹ with a resolution of 4 cm⁻¹ and 32 scans co-additions. For each isolate, three instrumental replicates (obtained in the same day) and three biological replicates (obtained in three different days from different agar plates) were obtained corresponding to a total of nine spectra for each isolate. Between each isolate measurement, a background was taken. Spectra corresponding to the instrumental replicates were averaged prior to the analysis.

2.4. Chemometric analysis

FTIR-ATR spectra were processed with standard normal variate (SNV)²¹ followed by the application of a Savitzky-Golay filter (37 smoothing points, 3^rd order polynomial and 3^rd derivative).²² After pre-processing, spectra were modelled by PLSDA or PCA. In all cases, data were mean centred prior to each chemometric method application. The PLSDA model which is based on the PLS regression requires a previous knowledge of assigned classes (species) for all tested isolates.^23,24 The model was calibrated considering all samples and the leave-one-sample-out cross-validation procedure was adopted in order to prevent overfitting.^25,26 The cross validation (where 70% of randomly selected isolates are used to calibrate and 30% to test the model for 100 times) was performed in order to assess the proportion (%) of correct predictions for each species through the confusion matrix. More information regarding the structure of the PLSDA model and its application in a similar context is described elsewhere.²⁷

The PCA model was used when the number of isolates was too small to develop a robust supervised model. This mathematical procedure converts a set of correlated variables into a set of uncorrelated ones, the principal components (PCs). This procedure does not require any information about isolates species.²⁸

All data analyses were performed in Matlab version 7.9 (Mathworks, Natick, MA) and the PLS Toolbox version 5.5.1 for Matlab (Eigenvector Research, Manson, WA).

3. Results

3.1. Discrimination between C. albicans, C. glabrata, C. krusei, C. parapsilosis and C. tropicalis species

Fig. 1 shows the mean infrared spectrum of each of the 5 Candida species considered. Spectra present a very similar and typical shape containing the absorption bands of lipids (3000–2800 cm⁻¹), proteins/amides I and II (1700–1500 cm⁻¹), mixed region of phospholipids/DNA/RNA (1500–1185 cm⁻¹), polysaccharides (1185–900 cm⁻¹) and the fingerprint region (900–600 cm⁻¹) (Naumann et al., 1991; Maquelin et al., 2002). A vibration band at 1745 cm⁻¹ usually associated to the C=O stretching vibration of esters could also be observed among the five species (Stuart 2004). After spectra pre-processing (Fig. 2A–D) it is possible to observe some degree of spectral variability among the 5 species. Considering the polysaccharides and phospholipids/RNA/DNA wavenumber ranges (Fig. 2A and B, respectively), the main spectral differences are found to be between C. glabrata, C. krusei and C. albicans, C. tropicalis, C. parapsilosis at 1075 cm⁻¹, from 1025–950 cm⁻¹ and at 1380 cm⁻¹. Vibrations at these wavenumbers are usually attributed to CO–O–C symmetric stretching of lipids, to the ribose C–O stretching of DNA and RNA and to the CH₃ symmetric bending of leucine, respectively (Stuart 2004). These observations strongly suggest that C. albicans, C. tropicalis and C. parapsilosis present a very similar nucleic acids' profile as well as some lipids and aminoacids similarity and somehow different from C. glabrata and C. krusei. C. glabrata also seems to possess the most dissimilar proteins profile (Fig. 2C). Considering the vibration range of lipids (Fig. 2D) C. glabrata, C. krusei and C. parapsilosis appears to have a similar profile.

Fig. 1 Candida species FTIR-ATR mean spectra (4000–600 cm⁻¹). Legend: C. albicans, C. glabrata, C. krusei, C. parapsilosis, C. tropicalis.

Fig. 2 Candida species FTIR-ATR mean spectra processed with SNV and Savitzky-Golay (37 points filter size, 3^rd degree polynomial, 3^rd derivative) in the region (A) 1200–900 cm⁻¹; (B) 1500–1200 cm⁻¹; (C) 1750–1500 cm⁻¹ and (D) 3100–2800 cm⁻¹. Legend: C. albicans, C. glabrata, C. krusei, C. parapsilosis, C. tropicalis.

The developed PLSDA model (spectral range 1776–790 cm⁻¹), considering the first two latent variables (LVs), which encompasses 72.0% of the spectral variability, showed five perfectly individualized clusters, each encompassing isolates of one single Candida species (Fig. 3 and ESI 1†). Also, the clusters spatial organization perfectly corroborates the observed spectral similarity, being C. albicans, C. tropicalis and C. parapsilosis discriminated from C. glabrata and C. krusei by the first latent variable (LV). According to the confusion matrix of the developed PLSDA model, 99.6% isolates were correctly predicted (Table 1). Noteworthy, C. krusei, C. parapsilosis and C. tropicalis isolates had 100% of correct species assignment.

Fig. 3 Score plot corresponding to the first two LVs of the PLSDA regression model for discriminating between: () C. albicans, () C. glabrata, () C. krusei, () C. parapsilosis and () C. tropicalis.

Table 1 Confusion matrix obtained for the PLSDA species discrimination model, considering the region 1750–900 cm⁻¹. About 99.6% of the analysed isolates were correctly predicted (obtained by sum of the diagonals)

FTIR-ATR prediction	Species
	C. albicans	C. glabrata	C. krusei	C. parapsilosis	C. tropicalis
C. albicans	24.54	0.26	0	0	0
C. glabrata	0	19.59	0	0	0
C. krusei	0	0	15.65	0	0
C. parapsilosis	0.08	0	0	19.74	0
C. tropicalis	0.05	0	0	0	20.09

---

3.2. Discrimination between C. parapsilosis, C. metapsilosis and C. orthopsilosis species

The infrared spectra of these 3 Candida species are very similar (Fig. 4A–C). However, some differences can be noted such as the distinct sugar profile of C. metapsilosis (Fig. 4A) and the distinct lipidic profile of C. orthopsilosis (Fig. 4C). Noteworthy, vibrations in the 980–945 cm⁻¹ range are usually attributed to DNA ribose-phosphate skeletal motions and from 2900–2800 cm⁻¹ to CH₃ and CH₂ symmetric and asymmetric stretching of lipids (Stuart 2004). On the other hand, these three Candida species seem to possess a very similar proteins profile (Fig. 4B).

Fig. 4 C. parapsilosis, C. metapsilosis and C. orthopsilosis FTIR-ATR mean spectra processed with SNV and Savitzky-Golay (37 points filter size, 3^rd degree polynomial, 3^rd derivative) in the region (A) 1500–900 cm⁻¹; (B) 1750–1500 cm⁻¹; (C) 3100–2800 cm⁻¹. (D) Score plot corresponding to the first two PCs of the PCA model. Legend: () C. parapsilosis, () C. metapsilosis and () C. orthopsilosis.

Despite this similarity it was possible to discriminate the three species with a PCA model (considering the spectral range between 1500 and 1200 cm⁻¹). The model's score map exhibited three individualized clusters each containing one single species (Fig. 4D and ESI 2†). C. parapsilosis was discriminated from C. orthopsilosis and C. metapsilosis by the first PC and C. orthopsilosis from C. metapsilosis by the second PC.

3.3. Discrimination between C. albicans, C. tropicalis, C. lusitaniae, C. famata, C. guilliermondii and C. dubliniensis species

The main spectral differences between the six species were found in the carbohydrates region (Fig. 5A) and associated to the DNA and RNA ribose vibrations (1025–950 cm⁻¹). C. lusitaniae and C. guilliermondii seem to possess a different phospholipid composition being the remaining four species quite similar (Fig. 5B). Considering the proteins vibration range, the most dissimilar one is C. famata with a characteristic vibration band at 1670 cm⁻¹ usually attributed to the C=O stretching of asparagine and to the C–N stretching and N–H bending of proteins (Fig. 5C) (Stuart 2004).²⁹ C. albicans, C. tropicalis and C. dubliniensis possess a very similar lipidic profile presenting substantial differences to the other three species (Fig. 5D).

Fig. 5 Candida species FTIR-ATR mean spectra processed with SNV and Savitzky-Golay (37 points filter size, 3^rd degree polynomial, 3^rd derivative) in the region (A) 1200–900 cm⁻¹; (B) 1500–1200 cm⁻¹; (C) 1750–1500 cm⁻¹ and (D) 3100–2800 cm⁻¹. Legend: () C. albicans, () C. tropicalis, () C. lusitaniae, () C. famata, () C. guilliermondii and () C. dubliniensis.

The spectral similarity between these six species was reflected on the inability of a simple PCA model to perform their discrimination. Instead, a three-step approach was needed (spectral range 1776–790 cm⁻¹) to achieve the targeted discrimination: step 1 − discrimination of C. famata, C. guilliermondii and C. lusitaniae from each other and from C. albicans, C. tropicalis and C. dubliniensis (Fig. 6A and ESI 3A†); step 2 – discrimination of C. tropicalis from C. dubliniensis and C. albicans (Fig. 6B and ESI 3B†) and step 3 – discrimination between C. albicans and C. dubliniensis (Fig. 6C and ESI 3C†).

Fig. 6 Score plot corresponding to the first three LVs of the PCA model including: (A) () C. lusitaniae, () C. famata and () C. guilliermondii from () C. albicans, () C. tropicalis and () C. dubliniensis; (B) () C. tropicalis from () C. albicans and () C. dubliniensis and (C) () C. albicans from () C. dubliniensis.

3.4. Discrimination between C. glabrata and C. bracarensis species

Mean infrared spectrum of both C. glabrata and C. bracarensis (Fig. 7A–C) present a very similar pattern mainly among the protein and lipid vibration range (Fig. 7B and C, respectively) with only slight differences. The main dissimilarities were noted in the carbohydrates and phospholipid/RNA/DNA regions from 1020–975 cm⁻¹ usually attributed to the DNA and RNA C–O stretching. A PCA model (Fig. 7D and ESI 4†) was developed (spectral range 3100–2800 cm⁻¹ and 1776–790 cm⁻¹) and the discrimination of the two species was achieved with the first 3 PCs encompassing 80.6% of the spectral variability.

Fig. 7 C. glabrata and C. bracarensis FTIR-ATR mean spectra processed with SNV and Savitzky-Golay (37 points filter size, 3^rd degree polynomial, 3^rd derivative) in the region (A) 1500–900 cm⁻¹; (B) 1750–1500 cm⁻¹; (C) 3100–2800 cm⁻¹. (D) Score plot corresponding to the first three PCs of the PCA model. Legend: () C. glabrata and () C. bracarensis.

4. Discussion

Accurate Candida species identification remains a major issue due to their phenotypic and genotypic similarity. Several studies still report a very high rate of miss identifications among these microorganisms even using the gold standard of molecular biology techniques.^30–32 Noteworthy, Desnos-Ollivier and co-workers³³ stated about 58% of misidentified C. famata isolates. In this context, infrared based spectroscopic techniques have been explored as an alternative for Candida species identification and/or discrimination,^11–13 as well as to characterize their biofilms,¹⁸ blastopores and hyphae¹⁷ or the origin of the isolate.^13,14,16 However, the published work regarding species discrimination^11–13 included few isolates and/or species never combining together highly related species and/or present a low rate of correct species identification compared with the results herein obtained. The developed PLSDA model to discriminate the five Candida species was able to correctly predict 99.6% isolates based on their infrared spectra. Even the PCA models clearly showed well defined clusters, confirming the methodology ability to discriminate between species. It should be noted that the species discrimination achieved with the PLSDA (corroborated with PCA model results) was only possible with spectra processed with a 3^rd derivative while most similar studies involving bacteria^8,9,33 or yeasts^12,13 used 2^nd derivative spectra even for very closely related species.⁹ A filter width of 37 points is indeed somewhat higher than usually reported and may slightly attenuate some spectral bands; however, the results substantially improve when working on this range of Savitzky-Golay smoothing points. With the exception of one model, the developed models considered all major biomolecules vibration regions and part of the fingerprint region to achieve successful species discrimination (1776–790 cm⁻¹). These findings and the need of a 3^rd derivative spectrum stress the phenotypic and genotypic similarity that is undoubtedly reflected in their infrared spectra. Diezmann and co-workers³⁴ combined the DNA sequences of six nuclear genes (ACT1, EF2, RPB1, RPB2, 18S rDNA, and 26S rDNA) to establish the phylogenetic relationships among 38 strains from the Saccharomycetales order and observed several distinct clades in the phylogenetic tree comprising the different species. According to those authors, C. albicans, C. tropicalis and C. parapsilosis were in the same clade and C. glabrata and C. krusei in a distinct one. This data perfectly corroborate our findings based on the infrared spectra of the five species which points to a higher degree of similarity between C. albicans, C. tropicalis and C. parapsilosis and between C. glabrata and C. krusei (Fig. 2A–D). Also, the relative closeness of the species within each clade in the Diezmann et al.³⁴ work agrees with the results herein obtained. C. albicans and C. tropicalis, the closest ones within the first clade, present the most similar infrared spectral profile and closeness in the PLSDA scores map (Fig. 3) followed by C. parapsilosis. The relative similarity stated among C. glabrata and C. krusei is also in agreement with their relative distance within the clade of the Diezmann et al.³⁴ phylogenetic tree. Also, the developed PCA model to analyse C. albicans, C. tropicalis, C. lusitaniae, C. famata, C. guilliermondii and C. dubliniensis spectra (Fig. 6) pointed to a higher dissimilarity of C. lusitaniae, C. famata and C. guilliermondii, the first discriminated species and also the most distant in the clades of the Diezmann et al.³⁴ phylogenetic tree. On the other hand, the closer ones in the phylogenetic tree (C. albicans and C. dubliniensis) present the most similar infrared spectral profile and were the last ones to be clustered by PCA modelling. Regarding C. ortho-, meta- and parapsilosis, firstly considered C. parapsilosis group I–III due to their similarity,³⁵ the PCA model point to a high potential of discrimination based on their infrared spectra. Also, this technique proved to be able to discriminate C. bracarensis from C. glabrata, a very similar species.³⁰

As a spectroscopic technique, FTIR highly competes with other high throughput techniques mainly as mass based ones as matrix-assisted laser desorption/ionization time of flight (MALDI-TOF MS). Indeed MALDI-TOF MS as already been extensively exploited for Candida species identification with a high degree of success making this technique a first line choice.^36–38 However and despite being already available in some laboratories from the academia to industry and hospitals this mass spectrometers still are very costly and require expensive and specialized maintenance when compared with infrared spectrometers. Also, prior to mass spectra acquisition, a protein extraction step must be undertaken and in some cases the use of specific chemical reagents, as trifluoracetic acid, are need to enhance the ionization. All of these issues clearly increases the cost and time of the analysis making FTIR a true competitor with no need of any kind of sample processing nor chemical reagents.

5. Conclusions

FTIR-ATR coupled with chemometrics proved to be a reliable alternative to discriminate five of the most common Candida species (C. albicans, C. glabrata, C. krusei, C. parapsilosis and C. tropicalis) as well as very closely related and less frequent Candida species. Also, this infrared based technique also predicted species relatedness in agreement with some genotypic methods. The results obtained in this work strongly suggest that FTIR-ATR could be considered as a valuable alternative to highly demanding genotypic methods for Candida typing at species level.

Conflict of interest

The authors have no competing interests.

Acknowledgements

This work was funded by Fundação para a Ciência e a Tecnologia (FCT) under the strategic projects UID/QUI/50006/2013 and UID/BIO/04469/2013 and COMPETE 2020 (POCI-01-0145-FEDER-006684) and the project RECI/BBB-EBI/0179/2012 (FCOMP-01-0124-FEDER-027462). Sónia Silva was supported by a post-doctoral grant from Fundação para a Ciência e a Tecnologia (SFRH/BPD/111645/2015).

References

1. M. Ruhnke, V. Rickerts, O. A. Cornely, D. Buchheidt, A. Glöckner, W. Heinz, R. Höhl, R. Horré, M. Karthaus, P. Kujath, B. Willinger, E. Presterl, P. Rath, J. Ritter, A. Glasmacher, C. Lass-Flörl and A. H. Groll, Mycoses, 2011, 54, 279–310.
2. D. W. Williams, T. Kuriyama, S. Silva, S. Malic and M. A. O. Lewis, J. Periodontol., 2011, 55, 260–269.
3. S. Silva, M. Negri, M. Henriques, R. Oliveira, D. Williams and J. Azeredo, FEMS Microbiol. Rev., 2012, 36, 288–305.
4. E. Rodrigues, S. Silva and M. Henriques, Crit. Rev. Microbiol., 2104, 10, 1–13.
5. S. Silva, M. Negri, M. Henriques, R. Oliveira, D. Williams and J. Azeredo, Trends Microbiol., 2011, 19, 241–247.
6. K. H. Neppelenbroek, R. S. Seó, V. M. Urban, S. Silva, L. N. Dovigo, J. H. Jorge and N. H. Campanha, Oral Dis., 2014, 20, 329–344.
7. D. W. Williams, M. J. Wilson, M. A. O. Lewis and A. J. C. Potts, J. Clin. Microbiol., 1995, 33, 2476–2479.
8. C. Sousa, L. Silva, F. Grosso, A. Nemec, J. Lopes and L. Peixe, Eur. J. Clin. Microbiol. Infect. Dis., 2014a, 33, 1345–1353.
9. C. Sousa, L. Silva, F. Grosso, J. Lopes and L. Peixe, J. Photochem. Photobiol., B, 2014b, 133, 108–114.
10. R. Branquinho, C. Sousa, H. Osório, L. Meirinhos-Soares, J. Lopes, J. Carriço, H. Busse, A. Abdulmawjood, G. Klein, P. Kämpfer, M. Pintado and L. Peixe, Int. J. Syst. Evol. Microbiol., 2014, 64, 3867–3876.
11. É. M. Timmins, S. U. Howell, B. K. Alsberg, W. C. Noble and R. Goodacre, J. Clin. Microbiol., 1998, 36, 367–374.
12. K. Tintelnot, G. Haase, M. Seibold, F. Bergmann, M. Staemmler, T. Franz and D. Naumann, J. Clin. Microbiol., 2000, 38, 1599–1608.
13. M. Essendoubi, D. Toubas, M. Bouzaggoua, G. P. Pinon, M. Manfait and G. D. Sockalinguma, Biochim. Biophys. Acta, 2005, 1724, 239–247.
14. C. Sandt, G. D. Sockalingum, D. Aubert, H. Lepan, C. Lepouse, M. Jaussaud, A. Leon, J. M. Pinon, M. Manfait and D. Toubas, J. Clin. Microbiol., 2003, 41, 954–959.
15. M. Essendoubi, D. Toubas, C. Lepouse, A. Leon, F. Bourgeade, J. M. Pinon, M. Manfait and G. D. Sockalinguma, J. Microbiol. Methods, 2007, 71, 325–331.
16. D. Toubas, E. Mohammed, I. Adt, J. M. Pinon, M. Manfait and G. D. Sockalingum, Anal. Bioanal. Chem., 2007, 387, 1729–1737.
17. I. Adt, D. Toubas, M. Manfait and G. D. Sockalingum, Arch. Microbiol., 2006, 185, 277–285.
18. P. Lal, D. Sharmal and P. Pruthi, J. Appl. Microbiol., 2010, 109, 128–136.
19. S. Silva, M. Henriques, A. Martins, R. Oliveira, D. Williams and J. Azeredo, Med. Mycol., 2009, 47, 681–689.
20. M. Martins, M. Henriques, A. P. Ribeiro, R. M. Fernandes, V. Gonçalves, A. Seabra, J. Azeredo and R. Oliveira, Rev. Iberoam. Micol., 2010, 27, 119–124.
21. T. Næs, T. Isaksson, T. Fearn and T. Davis, A user-friendly guide to multivariate calibration and classification, NIR Publications, Chichester, UK, 2002.
22. A. Savitzky and M. J. E. Golay, Anal. Chem., 1964, 36, 1627–1639.
23. P. Geladi and B. R. Kowalsky, Anal. Chim. Acta, 1986, 185, 11–17.
24. B. K. Alsberg, D. B. Kell and R. Goodacre, Anal. Chem., 1998, 70, 4126–4133.
25. S. Wold, Pattern Recogn., 1976, 8, 127–139.
26. R. De Maesschalck, A. Candolfi, D. L. Massart and S. Heuerding, Chemom. Intell. Lab. Syst., 1999, 47, 65–77.
27. C. Sousa, F. Grosso, L. Meirinhos-Soares, L. Peixe and J. Lopes, J. Biophotonics, 2014c, 7, 287–294.
28. I. T. Jolliffe, Principal Component Analysis, Springer, Berlin, 1986.
29. B. Stuart, Biological Applications, in Infrared Spectroscopy Fundamentals and Applications, Wiley J. & Sons, New Jersey, 2004.
30. A. Correia, P. Sampaio, S. James and C. Pais, Int. J. Syst. Evol. Microbiol., 2006, 56, 313–317.
31. M. D. Ollivier, M. Ragon, V. Robert, D. Raoux, J. C. Gantier and F. Dromer, J. Clin. Microbiol., 2008, 10, 3237–3242.
32. T. R. Southern, A. Aelew and P. C. Iwen, J. Clin. Microbiol., 2015, 2015(53), 1999–2000.
33. C. Sousa, Â. Novais, A. Magalhães, J. Lopes and L. Peixe, Sci. Rep., 2013, 3270–3278.
34. S. Diezmann, C. J. Cox, G. Schonian, R. J. Vilgalys and T. G. Mitchell, J. Clin. Microbiol., 2014, 12, 5624–5635.
35. D. Lin, W. Lung-Chi, M. G. Rinaldi and P. F. Lehmann, J. Clin. Microbiol., 1995, 33, 1815–1821.
36. A. Pinto, C. Halliday, M. Zahra, S. T. Olma, K. Maszewska, J. R. Iredell, W. Meyer and S. C. Chen, PLoS One, 2011, 6(10), e25712.
37. A. K. Ghosh, S. Paul, P. Sood, S. M. Rudramurthy, A. Rajbanshi, T. J. Jillwin and A. Chakrabarti, Clin. Microbiol. Infect., 2015, 21, 272–378.
38. A. L. Roberts, A. Alelew and P. C. Iwen, Diagn. Microbiol. Infect. Dis., 2016, 85, 73–76.

---

Footnote

† Electronic supplementary information (ESI) available. See DOI: 10.1039/c6ra16769a

---