Influence of the Sample Preparation Method in Discriminating Candida spp. Using ATR-FTIR Spectroscopy.

Several studies have investigated the capacity of ATR-FTIR spectroscopy for fungal species discrimination. However, preparation methods vary among studies. This study aims to ascertain the effect of sample preparation on the discriminatory capacity of ATR-FTIR spectroscopy. Candida species were streaked to obtain colonies and spectra were collected from each preparation type, which included: (a) untreated colonies being directly transferred to the ATR crystal, (b) following washing and (c) following 24-h fixation in formalin. Spectra were pre-processed and principal component analysis (PCA) and K-means cluster analysis (KMC) were performed. Results showed that there was a clear discrimination between preparation types. Groups of spectra from untreated and washed isolates clustered separately due to intense protein, DNA and polysaccharide bands, whilst fixed spectra clustered separately due to intense polysaccharide bands. This signified that sample preparation had influenced the chemical composition of samples. Nevertheless, across preparation types, significant species discrimination was observed, and the polysaccharide (1200–900 cm−1) region was a common critical marker for species discrimination. However, different discriminatory marker bands were observed across preparation methods. Thus, sample preparation appears to influence the chemical composition of Candida samples; however, does not seem to significantly impact the species discrimination potential for ATR-FTIR spectroscopy.


Introduction
There has been a significant increase in incidence and prevalence of fungal infections in humans since the 1980s [1]. Of the fungi infecting humans, those of the genus Candida are most common and present clinically as superficial or systemic infections [2]. Candida species are commonly found as commensals in the gastrointestinal tract, oral cavity, esophageal tract, skin and genitourinary tract [1]. The rise in incidence of Candidiasis can be attributed to the increased number of immunocompromised patients, invasive procedures and inappropriate use of antibiotics [3]. Figure 1 includes the raw and pre-processed spectra of Candida glabrata for the different preparation methods. Significant bands are shown in the pre-processed data, and the corresponding assignments are outlined in Table 1.
The most significant spectral differences observed between preparation types in the average second derivative spectra occurs with bands at 1404 cm −1 , assigned to the COO-symmetric stretching (ν s COO-) from carboxylic acids and free amino acids of lipids and proteins; at 1371 cm −1 , assigned to the CH 2 wagging in lipids and β-1,3 glucans of lipids and polysaccharides; at 1343 cm −1 , assigned to CH 2 wagging vibrations of lipids; at 1022 cm −1 , assigned to β-1,4 glucans of polysaccharides; at 993 cm −1 , assigned to β-1,6 glucans of polysaccharides; and at 965 cm −1 , assigned to mannans and the C-O stretch (νC-O) of phosphodiesters and ribose of DNA and polysaccharides.
Following this, PCA was performed on spectra from Candida glabrata species with the three preparation methods. Figure 2 depicts the PCA scores plot and loadings plot corresponding to PC1 versus PC2. The PCA and PC1 loadings plot of all preparation types with all species is shown in Supplementary Materials (S1).   PCA is an unsupervised multivariate data analysis (MVDA) method that is exploratory in nature, which means that it is used to differentiate between data without a priori knowledge. These techniques are useful in sorting data and identifying the differences, heterogeneity and complexity of unknown datasets [31,32]. For PCA data analysis, scores plots are interpreted in conjunction with their corresponding loadings plots. Loadings plots display variables or spectral features that contribute to most of the clustering patterns observed in the principal components (PC) of the scores plots. As spectra were converted to their second derivatives, samples that are positively scored in the scores plot are correlated with negatively loaded bands in the corresponding PC loadings plot. Likewise, negatively scored samples in the scores plot correlate with positively loaded bands in the corresponding PC loadings plot.
Investigation of the PC1 loadings plot revealed clear discrimination between sample preparation methods. Across the PC1 versus PC2 scores plot, the untreated and washed isolates clustered together along PC1 due to loadings bands at 1634 cm −1 , 1084 cm −1 , 1047 cm −1 and 989 cm −1 , which were assigned to proteins, DNA and polysaccharides, whilst fixed isolates clustered separately due to an intense loadings band at 1018 cm −1 , which was assigned to polysaccharides.

Principal Component Analysis for Species Discrimination
PCA was performed using spectra from each preparation method to investigate the capacity for species discrimination, as illustrated in Figure 3. In some instances, the differences between spectra were relatively subtle and will not appear in the PC depicting the most variation (PC1). In datasets from all preparation methods PC2 showed the best discrimination between clusters of spectra assigned to different Candida species. PC1 seemed to show differences between species replicates and PCA scores plots, and the corresponding PC1 loadings are shown in Supplementary Materials (S2).
Investigation of the PCA scores plots of PC2 versus PC3 ( Figure 3) revealed that each preparation method demonstrated similar species discrimination. Candida krusei and Candida glabrata clustered together whilst Candida parapsilosis and Candida albicans also clustered together. However, different marker bands contributed to the discrimination seen in each preparation type (Table 2).
Nonetheless, across all preparation types, loadings plots revealed that bands 1141 cm −1 of untreated isolates, 1137 cm −1 of fixed isolates and 1145 cm −1 of washed isolates, which corresponded to the C-O stretching of carbohydrates, contributed to the separation of clusters of Candida krusei and Candida glabrata spectra in scores plots. Additionally, bands 989 cm −1 of the untreated and washed PCA is an unsupervised multivariate data analysis (MVDA) method that is exploratory in nature, which means that it is used to differentiate between data without a priori knowledge. These techniques are useful in sorting data and identifying the differences, heterogeneity and complexity of unknown datasets [31,32]. For PCA data analysis, scores plots are interpreted in conjunction with their corresponding loadings plots. Loadings plots display variables or spectral features that contribute to most of the clustering patterns observed in the principal components (PC) of the scores plots. As spectra were converted to their second derivatives, samples that are positively scored in the scores plot are correlated with negatively loaded bands in the corresponding PC loadings plot. Likewise, negatively scored samples in the scores plot correlate with positively loaded bands in the corresponding PC loadings plot.
Investigation of the PC1 loadings plot revealed clear discrimination between sample preparation methods. Across the PC1 versus PC2 scores plot, the untreated and washed isolates clustered together along PC1 due to loadings bands at 1634 cm −1 , 1084 cm −1 , 1047 cm −1 and 989 cm −1 , which were assigned to proteins, DNA and polysaccharides, whilst fixed isolates clustered separately due to an intense loadings band at 1018 cm −1 , which was assigned to polysaccharides.

Principal Component Analysis for Species Discrimination
PCA was performed using spectra from each preparation method to investigate the capacity for species discrimination, as illustrated in Figure 3. In some instances, the differences between spectra were relatively subtle and will not appear in the PC depicting the most variation (PC1). In datasets from all preparation methods PC2 showed the best discrimination between clusters of spectra assigned to different Candida species. PC1 seemed to show differences between species replicates and PCA scores plots, and the corresponding PC1 loadings are shown in Supplementary Materials (S2).
Investigation of the PCA scores plots of PC2 versus PC3 ( Figure 3) revealed that each preparation method demonstrated similar species discrimination. Candida krusei and Candida glabrata clustered together whilst Candida parapsilosis and Candida albicans also clustered together. However, different marker bands contributed to the discrimination seen in each preparation type (Table 2).
Nonetheless, across all preparation types, loadings plots revealed that bands 1141 cm −1 of untreated isolates, 1137 cm −1 of fixed isolates and 1145 cm −1 of washed isolates, which corresponded to the C-O stretching of carbohydrates, contributed to the separation of clusters of Candida krusei and Candida glabrata spectra in scores plots. Additionally, bands 989 cm −1 of the untreated and washed isolates and band 985 cm −1 of fixed isolates, which corresponded to β-1,6 glucans, also contributed to the separation of clusters in the spectra of these species.
Similarly, bands 1047 cm −1 and 1010 cm −1 , which corresponded to mannans and the C-O stretching of carbohydrates, respectively, contributed to the separate clustering of Candida albicans and Candida parapsilosis spectra.
Molecules 2020, 25, x FOR PEER REVIEW 6 of 12 isolates and band 985 cm −1 of fixed isolates, which corresponded to β-1,6 glucans, also contributed to the separation of clusters in the spectra of these species.

KMC Analysis
To visualise clustering patterns, KMC analysis was conducted. This technique is also an unsupervised MVDA method and is useful for qualitative analysis. It aids in observing the relationships between spectra rather than investigating the spectral features that contribute to separation. KMC analysis was performed for each preparation in spectral windows in the ranges of 3000-2800 cm −1 and 1800-900 cm −1 . The best separation was achieved using the polysaccharide region, 1400-900 cm −1 . The untreated isolates provided the best classification, as illustrated in Figure 4, with each species forming individual clusters. K-means cluster analysis plots for the washed and fixed datasets are presented in Supplementary Materials (S3). Similarly, bands 1047 cm −1 and 1010 cm −1 , which corresponded to mannans and the C-O stretching of carbohydrates, respectively, contributed to the separate clustering of Candida albicans and Candida parapsilosis spectra.

KMC Analysis
To visualise clustering patterns, KMC analysis was conducted. This technique is also an unsupervised MVDA method and is useful for qualitative analysis. It aids in observing the relationships between spectra rather than investigating the spectral features that contribute to separation. KMC analysis was performed for each preparation in spectral windows in the ranges of 3000-2800 cm −1 and 1800-900 cm −1 . The best separation was achieved using the polysaccharide region, 1400-900 cm −1 . The untreated isolates provided the best classification, as illustrated in Figure 4, with each species forming individual clusters. K-means cluster analysis plots for the washed and fixed datasets are presented in Supplementary Materials (S3).

Discussion
In this study, the majority of 4 Candida spp. were well discriminated using ATR-FTIR spectroscopy with three different preparation methods. There also appeared to be a clear

Discussion
In this study, the majority of 4 Candida spp. were well discriminated using ATR-FTIR spectroscopy with three different preparation methods. There also appeared to be a clear discrimination between preparation methods; however, within each sample preparation type, similar species discrimination was achieved. The polysaccharide region was the common critical discriminatory spectral region; however, each preparation method also displayed unique marker bands. To the best of our knowledge, no previous studies have attempted to investigate the effect of sample preparation on Candida spp. discrimination using ATR-FTIR spectroscopy.
There seemed to be a clear separation of preparation types signifying that sample preparation had influenced the macromolecular composition of Candida samples. In PC1, the untreated and washed samples showed intense protein, DNA and polysaccharide bands, whilst the fixed samples showed intense polysaccharide bands. In PC2 the untreated and fixed samples showed intense protein, lipid and protein and DNA and polysaccharide bands, whilst the washed isolates showed intense protein, lipid and protein and polysaccharide and DNA bands. This could be due to the washing step damaging the integrity of cell wall polysaccharides or the loss of some water-soluble polysaccharides. Fixation may also affect the integrity of proteins.
The discrimination observed in each preparation type is consistent with previous literature. Silva et al. [19] demonstrated that Candida albicans and Candida parapsilosis clustered together and Candida glabrata and Candida krusei clustered together in PCA scores plots. These findings are also consistent with phylogenetic relationships between the four species based on a study conducted by Diezmann et al. [33], which, in a combined maximum analysis, places Candida albicans and Candida parapsilosis in the same clade and Candida glabrata and Candida krusei in a separate one, based on six genes (ACT1,EF2,RPB1,RPB2,18S rDNA and 26S rDNA) [19].
There appears to be a consensus among existing literature that the best interspecies discrimination is achieved in the polysaccharide region, thereby highlighting that polysaccharide profiles are species specific [19,22]. Findings from this study are consistent with this. Across preparation types, bands corresponding to β-1,6 glucans (985 cm −1 , 989 cm −1 ), mannans (1047 cm −1 ) and the C-O stretching of carbohydrates (1010 cm −1 , 1141 cm −1 , 1137 cm −1 1145 cm −1 ) contributed to the observed discrimination. Glucans and mannans are components of the fungal cell wall and these findings highlight their role in discrimination.
Each preparation type enabled species discrimination. Across preparation types, bands corresponding to C-O stretch of carbohydrates and β-1,6 glucans contributed to the clustering of Candida krusei and Candida glabrata, whilst C-O stretching of carbohydrates and mannans contributed to the clustering of Candida parapsilosis and Candida albicans. This highlights that these bands contribute most to species discrimination, regardless of preparation type.
Furthermore, for the untreated and fixed isolates, the amide I band at 1634 cm −1 , attributed to β-pleated sheet components, was observed for the clustering of Candida glabrata and Candida krusei, whilst the C-O stretching band at 1166/1162 cm −1 and a band corresponding to the phosphodiester stretch at 1080/1084 cm −1 was observed in the clustering of Candida albicans and Candida parapsilosis.
The band corresponding to mannans and the C-O stretch of phosphodiesters and ribose was also observed at 969 cm −1 in the untreated and washed isolates for these species. Similarly, the band corresponding to β-1,4 glucans at 1026 cm −1 was observed in the washed Candida krusei and Candida glabrata isolates but was not observed in the corresponding fixed and untreated datasets. This highlights the fact that although each preparation method provides similar species discrimination, the marker bands are unique to each preparation method.
Here we demonstrated the influence of sample preparation on the spectral profile of Candida, with clear-cut discrimination between different preparation methods. Despite this, within each sample preparation, species discrimination was achieved, predominantly based on the polysaccharide region. We identified several discriminatory bands, universal to sample preparation methods (1145 cm −1 , 1141 cm −1 , 1137 cm −1 , 1047 cm −1 , 1010 cm −1 , 989 cm −1 and 985 cm −1 ), along with several discriminatory bands specific to selected sample preparation methodology. This highlights the need for standardisation of a sample preparation strategy, particularly in the context of future potential clinical application.

Fungal Collection
Four Candida species were obtained from a repository at the Monash Department of Microbiology and used for species discrimination. These isolates were stored in 25% glycerol broth at −80 • C and streaked to obtain colonies for spectral collection. Table 3 shows the strain information for each of the Candida species used for this study.

Culture Conditions
Candida species were cultured on standard yeast extract peptone dextrose (YPD) plates that were prepared in the Monash Department of Microbiology. YPD was selected as it is a complete medium for yeast growth and is most commonly used for growing yeast under non-selective conditions. All stock plates were incubated at 30 • C for 24 h.

Sample Preparation
Three sample preparation strategies were used: (1) direct transfer of untreated colonies for YPD onto the ATR crystal using a sterile loop; (2) transfer of colonies from YPD plate to an Eppendorf tube, followed by washing with 500 µL of ultrapure Milli-Q water and centrifugation (3000 g × 5 min.) conducted in triplicate; and (3) transfer of colonies from YPD plate to an Eppendorf tube followed by 24 h fixation in 1 mL of 10% neutral buffered formalin followed by washing in ultrapure Milli-Q water (as described in (2)). For each preparation type, 3 biological replicates were prepared, grown on separate YPD plates.

ATR-FTIR Spectroscopy
Samples from each respective preparation method were placed on an ATR crystal and dried using the cool setting of a blow-dryer. As yeast cells are generally around 4-6 × 6-10 µm (Candida albicans), sample thickness was estimated to be a minimum of 6 µm [3]. Spectra were acquired using a Bruker Alpha FTIR instrument (Bruker Optics, USA) with a diamond crystal attenuated total reflectance (ATR) accessory. An ATR correction was applied to all data.
For each prepared sample, 3 technical replicates were collected. Spectra were acquired over the wavenumber interval 4000 cm −1 to 600 cm −1 at a spectral resolution of 8 cm −1 . For each spectrum, 64 interferograms were co-added. Prior to acquisition of each new spectra, a spectrum of the background, from a cleaned ATR crystal, was obtained to account for experimental conditions (such as changes in atmospheric CO 2 and H 2 O). For each background spectrum, 128 interferograms were co-added.

Data Pre-Processing
Prior to analysis, all spectra were pre-processed by conversion into their second derivatives using the Savitzky-Golay method, smoothed using 11 smoothing points and then normalized using the standard normal variate (SNV). Mean spectra for each preparation type and each species with each preparation type was generated for further analysis.

PCA and KMC Analysis
Following pre-processing, data were mean-centred and analysed with PCA using MATLAB 9.1 release 2016b (MathWorks, Natick, MA, USA), using functions in the proprietary PLS_Toolbox package (Eigenvector Research Inc., Manson, WA, USA). Several spectral regions were tested for analysis: 3000-2800 cm −1 , 1800-900 cm −1 and 1400-900 cm −1 . Outliers with high leverage on the influence plot were excluded (this constituted 0.03% of the dataset) and different combinations of PCs were tested (for example, PC1 vs. PC2, PC2 vs. PC3). For this analysis, 3 PCs were chosen. The vector combination producing the most robust separation of classes was subsequently chosen for further analysis. KMC analysis was also performed on the pre-processed data in different spectral regions for each preparation type.

Conclusions
This study shows that the preparation methods of Candida spp. for ATR-FTIR spectroscopy significantly affects the spectral profile; however, this does not alter the species discrimination capacity of ATR-FTIR spectroscopy. PCA analysis shows that discrimination was achieved in each preparation type mainly in the polysaccharide regions, presumably due to species-specific cell wall structures. However, although each preparation type demonstrated a similar discrimination, the bands that contributed to discrimination varied for each preparation type. These findings outline how these preparation methods do not seem to affect drastically the discrimination of these Candida spp. for ATR-FTIR spectroscopy.
Supplementary Materials: The following are available online, Figure S1: PC1 versus PC2 scores plot of each preparation method and corresponding loadings for PC1, Figure