YEAST SMELL LIKE WHAT THEY EAT: ANALYSIS of VOLATILE ORGANIC COMPOUNDS of Malassezia Furfur in GROWTH MEDIA SUPPLEMENTED WITH

YEAST SMELL LIKE WHAT THEY EAT: ANALYSIS OF VOLATILE ORGANIC

COMPOUNDS OF Malassezia furfur IN GROWTH MEDIA SUPPLEMENTED

WITH DIFFERENT LIPIDS

THESIS PRESENTED AS PARTIAL REQUIREMENT TO OBTAIN THE TITLE AS MASTER OF SCIENCE IN CHEMISTRY

Presented by:

MABEL CRISTINA GONZÁLEZ MONTOYA

Advisor

CHIARA CARAZZONE, Ph.D.

Co-advisor

ADRIANA M. CELIS, Ph.D.

CHEMISTRY DEPARTMENT UNIVERSIDAD DE LOS ANDES DECEMBER 2018 BOGOTÁ D.C.

CONTENT 1. ABSTRACT...... 6 2. ACKNOWLEDGEMENTS ...... 7 3. INTRODUCTION TO THE GENERAL TOPIC...... 8 4. PROBLEM STATEMENT AND JUSTIFICATION OF THE RESEARCH PROJECT...... 9 5. OBJECTIVES OF THE STUDY...... 10 5.1. General objective...... 10 5.2. Specific objectives...... 10 6. THEORETICAL FRAMEWORK...... 10 6.1. Microbial and fungal VOCs...... 10 6.2. Methodologies used for analyse Microbial Volatile Organic Compounds (mVOCs)... 13 6.3. Statistical techniques used for volatile profiling...... 15 6.3.1. Univariate analysis (UVA)...... 16 6.3.2. Multivariate analysis (MVA)...... 16 7. MATERIALS AND METHODS...... 19 7.1. Chemicals...... 19 7.2. Media preparation...... 19 7.3. Yeast growth...... 23 7.4. VOCs Extraction...... 23 7.5. Chromatographic analysis and annotation...... 24 7.6. Data analysis...... 24 8. RESULTS AND DISCUSSION...... 25 8.1. Profiling of VOCs from M. furfur...... 25 8.2.VOCs profile differentiation in four growth media in exponential and stationary phase 33 8.3. VOCs profile comparison in Agilent GC-MS with preliminary analysis in Thermo GC- MS 43 9. OUTPUT OF THE STUDY...... 45 10. CONCLUSIONS...... 45 11. BIBLIOGRAPHY...... 45 12. APPENDIX...... 52

3 TABLES INDEX

Table 1. Some fungal VOCs their structures, functions and odors. Information of this Table has been reproduced from Morath et al., (2012)...... 12 Table 2. Summary of some statistical techniques used in MVA. Information of this Table has been reproduced and combined from Eliasson et al., (2011), Putri & Fukusaki (2014) and Paliy & Shankar (2016). 18 Table 3. Chemical composition of mDixon medium...... 20 Table 4. Chemical composition of oleic acid (OA), palmitic acid (PA) and OA+PA media. The components of the minimal medium (MM) are listed in Materials and Methods...... 21 Table 5. Volatiles ´ proﬁles of M. furfur in four media and two growth phases...... 28 Table 6. Signiﬁcant statistical univariate parameters found on 17 VOCs from M. furfur and compounds highlighted by PCA and PLS-DA...... 34

4 FIGURES INDEX

Figure 1. Main metabolic pathways for mVOCs production. This figure has been reproduced from Schmidt et al., (2015). Volatiles are depicted in coloured dashed rectangles indicating different chemical classes. Representative examples are given per class: alcohols (e.g. ethanol), aldehydes (e.g. benzaldehyde), alkanes (e.g. undecane), alkenes (1-undecene), aromatic compounds (e.g. 2- phenylethanol), esters (e.g. 2-phenylethyl ester), fatty acids (e.g. butyric acid), isoprene, lactic acid, lactones (e.g. gammabutyrolactone), methylketones (e.g. acetone), monoterpenes (e.g. farnesol), nitrogen compounds (e.g. benzonitrile), sesquiterpenes (e.g. pinene) and sulphur compounds (e.g., dimethyl disulphide)...... 11 Figure 2. Headspace (HS) extraction techniques used for sampling mVOCs. Information of this Fig- ure has been reproduced from Agrawal (2004)...... 13 Figure 3. A graphical representation of the different analytical approaches and informatics techniques employed in metabolomics studies. A standard experimental design in metabolomics includes sample preparation and data acquisition using for instance LC–MS, GC–MS, NMR or FT-IR. Once the data have been generated, several types of univariate statistical and multivariate algorithms such as discriminant analysis, classification trees and machine learning can be applied to reduce the dimensionality of the data and facilitate biological interpretation, this could also improve the design of subsequent or future experiments. Information of this Figure has been reproduced from Gromski et al., (2015)...... 15 Figure 4. Growth curves of M. furfur in mDixon, oleic acid (OA), palmitic acid (PA) and OA+PA medium. 23 Figure 5. Chromatograms of M. furfur in exponential and stationary phases grown in four media with different FA content. De = Dixon in Exponential phase, Ds = Dixon in Stationary phase, Oe = Oleic acid in Exponential phase, Os = Oleic acid in Stationary phase, OPe = Oleic acid + Palmitic acid in Exponential phase, OPs = Oleic acid + Palmitic acid in Stationary phase, Pe = Palmitic acid in Exponential phase, Ps = Palmitic acid in Stationary phase...... 26 Figure 6. Pirate-plots representing univariate statistical differences in the absolute peak areas of 17 VOCs in which the grouping variable Medium-Phase significantly affects the emission (P < 0.05 in ANOVA and Kruskal-Wallis test). Variation of data is represented by Bayesian High-Density Intervals after running 10000 iterations. Bars show median values for each treatment. The first line of plots is y-scaled from 0 to 16x107, in the following line y-axis were scaled from 0 to 1.2x107, while the others two lines are scaled from 0 to 1.2x106...... 35 Figure 7. Heat-map of Hellinger transformed and Pareto-scaled peak areas from 61 tentative annotated volatiles found in M. furfur analysis using HS-SPME/GC-MS. Dendrograms represent similarities between samples (objects) and VOCs (variables) using Euclidean distances as a measure. Clus- ters of Medium-phase samples were colored according to the four growth media, while the compounds were colored according to the chemical group. Compounds with * represent VOCs reported previously in Fungi in the mVOC 2.0 Database...... 37 Figure 8. 3-dimensional PCA of the volatiles’ profiles of M. furfur grown in eight different experimental treatments. Ellipses in this plot are confidence intervals of 95% using normal-distribution. 39 Figure 9. a) PLS-DA of volatiles’ profiles of M. furfur grown in eight different experimental treatments. Ellipses in this plot are confidence intervals of 95% using t-distribution. b) VIP-Plot showing the more important VOC from M. furfur differentiated between four medium and two growth phas-

5 es. The VIP score of each compound was calculated as a weighted sum of the squared correlations between the PLS-DA components and the original variables...... 42 Figure 10. 3-dimensional PCA of the volatiles’ profiles of M. furfur grown in eight different experimental treatments and analysed in the Thermo GC-MS. Ellipses in this plot are confidence intervals of 95% using normal-distribution...... 44 Figure 11. Experimental mass spectra and comparison with the tentative annotated compound or the compound with the major similarity in their fragmentation patterns in EI...... 60

1. ABSTRACT

Malassezia furfur is part of the human skin microbiota. Its VOCs possibly contribute to the characteristic scalp odour in humans, as well as to the microbiota interaction and might have important biotechnological applications for developing flavours and fragrances derived from lactones. Being a ipid dependent yeast, the aim of this study was to analyse how the lipid composition of the liquid medium influences the productions of VOCs emitted by M. furfur. The strain M. furfur CBS 1878 was grown in four different media: 1) mDixon, 2) oleic acid (OA), 3) oleic acid + palmitic acid (OA+PA), and 4) palmitic acid (PA). The profiles of the volatiles compounds were characterized using HS-SPME/ GC-MS in each media in exponential and stationary phase. A total number of 61 VOCs was found in M. furfur, among which alkanes, alcohols, ketones, and furanic compounds were the most abundant. Some compounds previously reported for Malassezia (of γ- dodecalactone, 1-butanol, 3-methyl, and 1-hexanol) were also found. Through our experiments, using univariate and multivariate unsupervised (HCA and PCA) and supervised (PLS-DA) statistical techniques, we have proven that each tested growth medium stimulates the production of a different volatiles profile in M. furfur. Carbon dioxide, 1-hexanol, the isomer 5 of methyldecane, dimethyl sulphide, 5-undecene, isomer2 of methylundecane, isomer1 of methyldecane and furan, tetrahydro-2-methyl were established as differentiating compounds for the eight experimental treatments by all the techniques. The significance of our findings deserves future research to analyse in vivo conditions of the actual nutritional lipids available in the human skin and to search possible volatile biomarkers of metabolic changes related with its beneficial or pathogenic role.

Keywords: Fungal VOCs; GC-MS; HS-SPME; Malassezia, microorganisms; lipids; exponential growth phase; stationary growth phase.

7 2. ACKNOWLEDGEMENTS

I am immensely grateful for my family's support. To my dad, for teaching me the importance of doing something I love. To my mother, for taking care of me, loving me and understanding me. To my sister, for reminding me that I can be stronger than what I think I am. To Chepito, for recharging me with happiness just by seeing it.

I also have no words to thank Chiara for her support and trust. For her teachings, for her patience and above all for being a good human being, always with the best disposition to be able to help in thousands of ways.

I thank Adriana very much for placing her trust in me. For allowing me to know another fascinating world, the world of the Malassezias; and for the late nights doing the analysis. I thank Jorge, for opening the doors of the microbiological world to me by mediating ecological interactions of animals and humans, and for his time in so many advance meetings that we had.

I am also grateful for the support of Marcela Guevara and for the late nights doing the analyses and preparing the media. Because through this research I got to know an outstanding researcher and now a friend. I also acknowledge to Diana Marcela Tabares for her contributions during the preliminary analysis of VOCs.

I thank all my colleagues at LATNAP, especially Lida, for their patience in diﬃcult times and also for the good times we shared. To my friends, who this year have been a very special motor to continue in the moments of diﬃculty.

And ﬁnally I thank the “Universidad de los Andes”, for being my home. For allowing me to train personally and professionally.

3. INTRODUCTION TO THE GENERAL TOPIC

All species in the genus Malassezia are lipid dependent, due to their inability to synthesize de novo C14 or C16 fatty acids [1] for lack of a cytosolic fatty acid synthase complex (FAS) [2]. In order to survive, these yeasts rely on lipids encountered in host skin, that are subsequently metabolised by lipases and phospholipases [3]. The 18 species of the genus are classiﬁed in the class Malasseziomycetes of phylum Basidiomycota [4]. Malassezia species could represent between 53% to 80% of human skin fungal microbiota, accumulating in diﬀerent body parts, especially abundant behind the ears [5] and on the scalp [6].

These organisms have been recognized as aetiological agents of pityriasis versicolor and could be related also to other skin diseases such as seborrhoeic dermatitis, dandruff, folliculitis, atopic/eczema dermatitis syndrome and psoriasis [7]. Depending on the stimulation, or occasionally, the downregulation of the immune response against Malassezia, this yeasts could remain as a commensal or eventually become pathogenic [8]. The causes related to this interconversion are still unknown, but there are different studies suggesting that a relationship exists between pathogenicity and modifications in the lipid compositions of the host [9,10] or changes in the skin microbiota [11,12]. The complexity of the human skin environment hinders the possibility to fully comprehend the host-microbe and microbe-microbe interactions [13], because the chemical composition of the skin determines the microbiota and, at the same time, the resident microbial communities modify this environment in a dynamic process [14]. Host demographics and genetics, human behavior, local and regional environmental characteristics and transmission events determine the human skin microbiota variability [15]. The spatial distribution of the microorganisms is also highly variable [5]. Malassezia and Propionibaterium, reside in areas with a high density of sebaceous glands (like face, chest and back), whereas Staphylococcus and Corynebacterium are mostly prevalent in areas with high temperature and humidity (like groin, axilla, and toe web) [14,16]. Any changes in these conditions may affect homeostasis and probably promote that a pathogen or a commensal, like Malassezia, can cause diseases.

Some Malassezia species could be easily differentiated by the quantity of lipids necessary to grow in certain media [17,18], therefore, the Malassezia assimilation assay is widely used to determine the lipid requirement of different strains [3]. For example, the palmitic acid (PA) medium causes a fungicidal activity for M. pachydermatis, an atypical strains of M. furfur, and M. sympodialis, but not in M. furfur CBS 1878 [19] (Fig. 4), whereas, the addition of oleic acid (OA) produces a fungistatic effect on M. pachydermatis, an atypical strain of M. furfur, and M. furfur CBS 1878 (Fig. 4). On the other hand, when the growth medium is supplemented with a mixture of PA and OA, the toxic effect of PA is relieved, at least partly, in these species/strains (Fig. 4). In M. sympodialis, OA produces a fungicidal effect. In addition, recent research has demonstrated that M. pachydermatis, an atypical strain of M. furfur, M. sympodialis and M. furfur contains the Δ9-desaturase gene, OLE1, involved in the conversion of saturated acids, such as PA and stearic acid, to palmitoleic acid and OA, respectively [3,20]. Unpublished analysis developed by Celis et al. (2017) have demonstrated that M. furfur is capable of growing in modified Dixon (mDixon,) and OA media, whereas only of surviving

9 in OA+PA and PA media (Fig. 4), as can be veriﬁed by the absence of a growing (exponential phase) in these last two media [20]. Furthermore, Dixon, mDixon and Leeming and Notham (LNA) media are the most widely used for culturing Malassezia, because their more complex composition and presence of ox bile (obtain after purifying and drying fresh bile from oxen, see Table 3) oﬀer the essential nutritional requirements for their growth [8,21].

The lipid-dependent and lipophilic life style of Malassezia spp. determine its ecological niche in the skin environment of human and animal hosts [19]. Understanding lipid metabolism of these yeasts is expected to provide new insights in their biology and ecology, and therefore, illuminate the path to understanding interconversions between commensalism and pathogenicity [20]. Fungal VOCs are one of the means to assess the metabolic profiles of fungi related to certain diseases [22–28] as they can be collected easily and relatively quickly to monitor static or dynamic metabolic changes [26,29–31]. Other important uses of fungal VOCs are biotechnological process [29,32,33], to signal ecological communication between different species [30,34], and to obtain information about fungal development [35,36]. The high diversity of VOCs released by bacteria and fungi justifies the creation of a database that summarizes the taxonomic distribution of around 2000 compounds and their possible functions. The first version of mVOC Database was created in 2014 and renewed this year [37,38]. No volatile compound of Malassezia has been reported in this database, because their data sources have been limited to PubMed publications [38]. The first and only description of Malassezia volatiles, in which the authors reported the presence of 11 different VOCs, was performed in 1979, when the genus was recognized as Pityrosporum [39].

4. PROBLEM STATEMENT AND JUSTIFICATION OF THE RESEARCH PROJECT

Having this in mind, the aim of this study is to analyse how the lipid composition of a liquid medium influences the production of VOCs released by M. furfur, the only Malassezia species known to survive in the PA medium. To accomplish this, we prepared four different media for growing the strain CBS 1878: 1) mDixon, 2) oleic acid (OA), 3) oleic + palmitic acid (OA+PA), and 4) palmitic acid (PA), and afterwards we analyse the volatiles profiles released by the yeast through HS-SPME/GC-MS. The mDixon medium was selected as the reference in vitro growing conditions, while the other media pretend to identify the VOCs released under restricted nutritional conditions when an unsaturated acid (OA), a saturated acid (PA) or a combination of both is present in the medium. In addition, for each medium we performed samplings of VOCs in exponential and stationary phase. Finally, we obtained results from eight experimental treatments: M. furfur growing in mDixon in exponential phase (De), in mDixon in stationary phase (Ds), in oleic acid in exponential phase (Oe), in oleic acid in stationary phase (Os), in oleic and palmitic acid in exponential phase (OPe), in oleic and palmitic acid in stationary phase (Ops), in palmitic acid in exponential phase (Pe), in palmitic acid in stationary phase (Ps).

Previous reports on other species, demonstrate that changes in the microbial growth conditions produce qualitative and quantitative alterations in the VOCs produced by an organism [27,40,41]. Based on these findings, we hypothesized the existence of intra- species variations in the VOCs profiles released by M. furfur due to the different lipid composition of the medium and to the phase of growth. Univariate comparisons and

10 unsupervised and supervised multivariate analysis have been applied to measure the metabolic diﬀerentiation and its signiﬁcance.

5. OBJECTIVES OF THE STUDY

5.1. General objective

To analyse how the lipid composition of the liquid medium inﬂuences the productions of VOCs emitted by Malassezia furfur strain CBS 1878 in exponential and stationary phase.

5.2. Speciﬁc objectives

➢ To use HS-SPME to perform the extraction of volatiles of M. furfur in four growth media: mDixon, oleic acid (OA), oleic acid + palmitic acid (OA+PA), and palmitic acid (PA), in both exponential and stationary phases. ➢ To separate and characterize VOCs of M. furfur produced in each growth medium and phase through GC-MS to estimate the intra-species variation of the volatiles profile. ➢ To annotate the volatile constituents emitted by M. furfur in eight experimental treatments (De, Ds, Oe, Os, O+Pe, O+Ps, Pe, Ps) to compare their qualitative and semi-quantitative differences. ➢ To use univariate (UVA) and multivariate (MVA) statistical analysis for determining if the lipid supplements in the media and the growth phases produce differences in the volatile profiles emitted by M. furfur. ➢ To compare the findings obtained in two different GC-MS from the Chemistry Department of the “Universidad de los Andes” and discuss possible causes of experimental dissimilarities.

6. THEORETICAL FRAMEWORK

6.1. Microbial and fungal VOCs

Microorganisms are ubiquitous in the biosphere and there are just a few places on earth where they do not exist. In the same way as other living things, microbes produce different volatile organic compounds, characterized by low molecular mass (100 – 500 Da), high vapour pressure, low boiling point, and usually by having no more than 20 carbons in its structure [42]. For the facility of dispersion in gaseous and aquatic media, microbial volatile organic compounds (mVOC) have important roles in different environments mediating antagonistic interactions between microorganisms occupying the same ecological niche, mediating multitrophic interactions, or functioning as indicators of biocontamination or disease [23,42–44]. In general terms, bacterial volatiles profiles are dominated by alkenes, alcohols, ketones, terpenes, benzenoids, pyrazines, acids and esters, whereas fungal volatiles profiles are dominated by alcohols, benzenoids, aldehydes, alkenes, acids, esters and ketones [44]. Most microbial volatiles are considered as side products of primary (from the synthesis of DNA and amino and fatty

11 acids) and secondary metabolism (from intermediates of the primary metabolism) [43]. The main metabolic pathways for mVOCs production are summarized in the Figure 1.

Figure 1. Main metabolic pathways for mVOCs production. This ﬁgure has been reproduced from Schmidt et al., (2015). Volatiles are depicted in coloured dashed rectangles indicating diﬀerent chemical classes. Representative examples are given per class: alcohols (e.g. ethanol), aldehydes (e.g. benzaldehyde), alkanes (e.g. undecane), alkenes (1- undecene), aromatic compounds (e.g. 2-phenylethanol), esters (e.g. 2-phenylethyl ester), fatty acids (e.g. butyric acid), isoprene, lactic acid, lactones (e.g. gammabutyrolactone), methylketones (e.g. acetone), monoterpenes (e.g. farnesol), nitrogen compounds (e.g. benzonitrile), sesquiterpenes (e.g. pinene) and sulphur compounds (e.g., dimethyl disulphide).

Many mVOCs have distinct odours, so the interest in them began with the bacteria and fungi that humans can smell. Relating specifically to fungi, there are two examples widely studied. One is the highly valued odour of mushrooms and truffles in culinary, which include mixtures of alcohols, aldehydes, terpenes, aromatics and thiols. The second example relates to the musty odor of fungal VOCs emitted from microscopic fungi, which has provided a foundation for studies that investigate the possible negative effects (irritation, sensitisation, toxicity, and inflammation) [43] of molds on human health, in what is often referred to as “sick building syndrome”[29]. This last finding has been also used to track markers of other human diseases, and exploited for detecting food spoilage, but certainly the biological roles in the ecosystems of most of the fungal VOCs remain to be investigated. The reported functions of some fungal VOCs are summarized in the Table 1.

12 Table 1. Some fungal VOCs their structures, functions and odors. Information of this Table has been reproduced from Morath et al., (2012). Molecule Structure Potential function(s); odors

Semiochemical; earthy, 1-Octen-3-ol “mushroomy” odor !

1-Butanol-3-, methyl-, acetate Antifungal; banana odor

Sabinene Unknown; peppery odor

Antibiotic; coconut odor Plant- 6-Pentyl-α-pyrone growth !

Plant-growth promoting; woody- β-Caryophyllene spicy odor Antifungal;

Isobutyric acid Antifungal; rancid cheese-like odor

Anti-microbial; almond odor Benzyl aldehyde Antifungal; !

1,8-Cineole Antifungal; camphor-like odor

Fungivore attractant; mild alcohol 2-Methyl-1-propanol ! odor

2-Heptanone Unknown; cheese odor !

3-Methyl-butanol Unknown; component of truﬄe odor !

13 The production of mVOCs depends on the species, growth phase, nutrients, pH, humidity and temperature [43], so they are useful for studying inter and intraspeciﬁc communication, where all of these variables modify VOCs emission. Around 2000 VOCs are registered from more than 1000 bacteria and fungi in the mVOC2.0 database [38], but about 1018 microbial species are expected to exist on earth, so there is a huge potential of this emerging research ﬁeld.

6.2. Methodologies used for analyse Microbial Volatile Organic Compounds (mVOCs)

Early studies were usually conducted using steam distillation coupled with liquid- liquid (L-L) extraction and subsequent concentration, applied for example to the analysis of mushroom aromas [30]. The analysis of microbial volatiles usually involved headspace (HS) extraction techniques (Figure 2) combined with GC or GC-MS. Sampling is performed using and adsorption tube (or similar), open only in the lid end, where it has contact with the cultures for a standardized time. This method directly collects the volatiles present in the gas phase above the growing fungus. A second, less frequent method requires the extraction of volatiles present in the fungal biomass using liquid- liquid (L-L) extraction assisted by microwave or ultrasound. In HS sampling an adsorbent material is needed, such as activated carbon black, a synthetic polymer like Tenax TA or a coated fiber (Solid Phase Microscale Extraction, SPME). Carbon black tubes are normally desorbed using solvents, e.g., dichloromethane or diethylether, whereas, Tenax tubes and SPME fibers are analyzed by thermal desorption. The tube, the syringe or the fiber are successively desorbed in the GC for separating the analytes.

In HS sampling, an equilibrium is achieved between the phase composed by the fungus + growth medium and the gas phase. Some physical (thermal or mechanical) assistance may be needed to help the rapid distribution of metabolites between the two phases. Like one key point in the L-L extraction is to select extraction solvents with a high affinity for metabolites of interest, in case of SPME the selection of the fiber composition is crucial [45]. Other extraction techniques used for the analysis of mVOCs are steam distillation, purge and trap, and supercritical fluid extraction [46].

Figure 2. Headspace (HS) extraction techniques used for sampling mVOCs. Information of this Figure has been reproduced from Agrawal (2004).

Gas chromatography (GC) can be used to analyze volatiles directly. The major advantage of GC is to deliver a high separation power and give structural information in one run. The most important step in GC is the injection, which, if performed poorly, can have a severe effect on the separation power. More volatile compounds require a thicker film column, whereas high separation power is best obtained by thin film columns, so an intermedium compromise between these two variables is needed to perform the analysis of volatile compounds. A mass spectrometer can be used to provide structural information from fragmentation in electron ionisation (EI) easily searched in databases and comparing the characteristic fragmentation patterns from each compound eluting from the column. The limitations of MS in the identification of unknowns can be an insufficient information content of the mass spectrum to stereo and positional isomers [45].

Other analytical techniques are proton transfer reaction-mass spectrometry (PTR- MS), selected ion flow tube-mass spectrometry (SIFT-MS) and nuclear magnetic resonance (NMR) spectroscopy [46]. Within PTR-MS, protonated water (H3O+) is used as a chemical ionisation reagent to measure the concentration of VOCs, enabling direct sampling of the gas HS and on-line measurements in real time. This has been coupled to a novel automated sampling system and shown not only clear species specific discriminations according to the microbial VOC profiles produced, but also temporal evolution of VOCs during bacterial growth. SIFT-MS utilises soft ionisation using multiple reagents (H3O+,O2+ and NO+), which are delivered via fast-flowing helium into a flow tube where the reactant sample gas is introduced at a defined flow rate. The resultant reaction products are separated and counted downstream by quadrupole mass spectrometry, as in GC-MS. SIFT-MS technology has been used previously for detecting HS VOCs from bacterial cultures [47,48]. NMR has been used less frequently, especially for to the elucidation of new chemical structures, as for the description of some terpenes emitted by Leptographium lundbergii [49].

6.3. Statistical techniques used for volatile proﬁling

Many living organisms including humans, animals, plants and even microorganisms produce large varieties of VOCs. The sum of all VOCs produced by an organism has been termed the volatilome. Gas chromatography-mass spectrometry (GC- MS)-based methods have been the tool to analyse VOC profiles. The main limitation of such methods is that they requires sample manipulation, resulting in laborious procedures, applied also to metabolomics research [31]. VOCs, as other metabolites are down-stream products of gene transcription and translation (proteins). As a result, metabolomics approaches can provide a clearer picture of the phenotype of a biological system. However, compared to the genome and the proteome, the metabolome is much more complex, for example the plant kingdom is estimated to contain 200,000 or more metabolites and phytochemicals, and there are more than 41,815 metabolites currently reported on the human metabolome database [50]. A typical metabolomics study begins with a biological question which, usually encompasses two groups or treatments, or more. This experimental design (Fig. 3) is followed by the selection and employment of appropriate analytical techniques, then a set of statistical techniques should be selected. These may include t-test, ANOVA analysis, discriminant analysis, classification trees and machine learning. However, each of these techniques has related advantages and disadvantages. Thus, a combination of different analytical technologies and statistical tools must be used to gain a comprehensive interpretation of the results.

Figure 3. A graphical representation of the diﬀerent analytical approaches and informatics techniques employed in metabolomics studies. A standard experimental design in metabolomics includes sample preparation and data acquisition using for instance LC– MS, GC–MS, NMR or FT-IR. Once the data have been generated, several types of univariate statistical and multivariate algorithms such as discriminant analysis, classiﬁcation trees and machine learning can be applied to reduce the dimensionality of the data and facilitate biological interpretation, this could also improve the design of subsequent or future experiments. Information of this Figure has been reproduced from Gromski et al., (2015).

The high amount of data obtained through GC-MS, LC-MS and NMR platforms requires different steps for data handling. First, the raw data should be processed with signal processing methods, alignment algorithms and normalization procedures are applied. Secondly, data should be analysed and interpreted. Finally, different statistical validation procedures are employed to assure the significance of the results [51].

A data matrix (sample vs. metabolites) acquired from a mass spectrometry-based metabolomics should be translated in order to extract the meaningful information from the complicated and huge data matrix. This could be accomplished by means of statistical analysis [52]. A preliminary step to statistical analysis is the data pre-treatment. These procedures are performed to ensure that the analysis lies on sample differences and not on bias or intensity mean offset. The selection of pre-treatment procedures is essential and will greatly affect which metabolites are identified to be most important. In order to avoid that the final model is dominated by one or a few high-intensity variables, a common pre-treatment is mean-centering. This procedure is done by subtracting the mean value of each variable, assuring that variables fluctuate around zero instead of their respective means. Another pre-treatment procedure is scaling, where each variable is divided by a scaling factor. Some of the more common scaling methods are unit variance scaling, pareto scaling, range scaling, and vast scaling. The final pre-treatment procedure is transformation, where data are converted, commonly by log, power, root or Hellinger arithmetic process, to deal with matrices with many zeros or to adjust data to specific assumptions of different statistical approaches [51,53].

There are two types of approaches for data interpretation. Univariate analysis (UVA) offers simple and instinctive results. Usually, the objective is to identify biomarkers related to the factor of interest, in which features are tested one by one through parametric or non-parametric tests [54]. Whereas, multivariate analysis (MVA) allows the recognition of patterns combining all the data matrix at the same time, using supervised or unsupervised methods. Nevertheless, MVA also requires algorithms more difficult to interpret compared to those generated by univariate analyses [53]. Finally, often it is not clear which MVA should be applied to a particular data set, so the researcher needs to understand advantages and caveats of different statistical techniques. For, both UVA and MVA is also important to remember that correlations between variables do not inherently imply causality (unless explanatory variables are experimental treatments).

6.3.1. Univariate analysis (UVA)

In this type of approach only one variable is analyzed at a time. These methods test individually metabolites that are increased or decreased significantly between different groups. These tests include parametric methods for data that are normally distributed, the most common being ANOVA, t-tests, and z-tests. Non-parametric methods can also be applied, like the Kruskal–Wallis test. Methods of univariate data analysis are constantly being refined. Incorporation of confidence intervals and bootstrapping besides the information of the p-value is strongly recommended [55].

6.3.2. Multivariate analysis (MVA)

17 Many classical statistical tools are developed for pairs of variables, such as correlation analysis and simple linear regression, and these methods can be extended for situations with more than two variables to handle hundreds or thousands of variables simultaneously as in the case of metabolomics data sets. When discussing multivariate methods it is important to stress that these methods make use of all variables simultaneously and deal with the simultaneous relationship among variables. Apart from using the mean, multivariate methods make use of covariances or correlations which reﬂect the extent of the relationships among the variables, in contrast to univariate methods that focus solely on the mean and the variance of a single variable [56]. Multivariate approaches could be performed in the form of unsupervised or supervised methods.

Unsupervised methods could be used for exploratory or interpretative purposes to 1) reduce the amount of dimensions that describe a profile (i.e., projecting the multivariate profile into a space of significantly smaller dimensionality whilst trying to minimize information loss), 2) clustering (identifying groups of points which are more similar to each other than to the rest of the data), or 3) for correlation analysis. Examples of techniques oriented to reduce dimensionality are Principal Component Analysis (PCA), Non-metric Multidimensional Scaling (NMDS), and Correspondence Analysis (CA) [53,55]. For clustering there are Hierarchical Cluster Analysis (HCA) and Disjoint Clustering [53]. Examples of correlation oriented to maximize the association between the explanatory variable(s) and the measured variables, include Canonical Correlation Analysis (CCorA), Co-Inertia Analysis (CIA) and Procrustes Analysis (PA) [53].

The supervised methods may be used when the classes or values of the responses that one is trying to predict (also known as the Y-data) and associated with each of the sample inputs (X-data) are known. These methods try to maximize the separation of objects among diﬀerent classes, and include Random Forest (RF), Projections to Latent Structures Discriminant Analysis (PLS-DA) and Linear Discriminant Analysis (LDA) [53,55]. In the Table 2 the general characteristics of some of these multivariate techniques are summarized.

18 Table 2. Summary of some statistical techniques used in MVA. Information of this Table has been reproduced and combined from Eliasson et al., (2011), Putri & Fukusaki (2014) and Paliy & Shankar (2016). Statistical Method Supervised/Unsupervised Characteristics

o Make a summary review of samples and metabolites in the informative axis on the PCA Unsupervised basis of the data variance. o Find the outlier samples.

o In a (small) number of ordination axes explicitly chosen prior to the analysis the multivariate NMDS Unsupervised data are ﬁtted to those dimensions. o A ‘stress’ parameter is computed to measure the lack of ﬁt.

o See the correlation relationship of the CA Unsupervised metabolites especially via time-course experiment.

o See the similarities (distances) among samples HCA Unsupervised and metabolites on the multidimensional spaces.

o K-means clustering, seeks to partition all Disjoint Clustering Unsupervised variables into predeﬁned K number of individual clusters.

o Find linear combinations of the x’s and the y’s CCorA Unsupervised that would provide maximum correlation between X and Y.

o Reveal relationships between sets of variables, CIA Unsupervised based on the covariance rather than on a correlation.

o Compares the distributions of multiple sets of corresponding objects, depicted as a cloud of PA Unsupervised objects in a multidimensional space. o This analysis is also referred to as comparison of shapes.

o Produce a complex visualization of decision trees to predict the values of response variable(s) based on the given values of predictor variables. RF Supervised o Handles thousands of input variables without variable deletion. o Produce estimates of what variables are important in the classiﬁcation.

o Find the sample diﬀerences and signiﬁcant metabolites with the binary (0, 1) values. PLS-DA Supervised o Construct the robust discriminant model by the latent variables.

o Find linear combinations of observed variables that maximize the grouping of objects into separate classes. LDA Supervised o Is more computationally eﬃcient than iterative methods because is based on the noniterative eigen- vector-based solution.

7. MATERIALS AND METHODS

7.1. Chemicals

Chemicals were purchased from Sigma (St Louis, MO, USA) unless otherwise indicated.

7.2. Media preparation Modified dixon broth medium (mDixon), OA medium, OA+PA medium and PA medium were prepared at the “Grupo de Investigación Celular de Micoorganismos Patógenos (CeMoP)” from Universidad de los Andes one day previously to experimental analysis. The mDixon was prepared containing 36 g/L malt extract [Oxoid], 10 g/L peptone [BD], 20 g/L Ox bile, 2 mL/L glycerol, 2 mL/L oleic acid, and 10 mL/L Tween 40) (Table 3). The OA, OA+PA and PA medium were prepared using a minimal medium (MM) that contained 10 mL K-buffer pH 7.0 [200 g/L of K2HPO4, 145 g/L of KH2PO4, 20 mL of M-N [30 g/L of MgSO4·7H2O, 15 g/L of NaCl, 1 mL 1% of CaCl2·2H2O [w/v], 10 mL 20% of glucose [w/v], 10 mL of 0.01% FeSO4 [w/v], 5 mL of spore elements [100 mg/L of ZnSO4·7H2O, 100 mg/L of CuSO4·5H2O, 100 mg/L of H3BO3, 100 mg/L of MnSO4·H2O, 100 mg/L of Na2MoO4·2H2O, and 2.5 mL 20% of NH4NO3 [w/v]. A mixture of 4 mM Tween 40 and 4 mM of oleic acid (Carlo Erba), or 4 mM of a mixture 50:50 of oleic (Carlo Erba) and palmitic acid (Merck, Darmstadt, Germany), or 4 mM of palmitic acid (Merck, Darmstadt, Germany), was added to the MM respectively. Mixtures containing palmitic acid were supplemented with 1% Brij-58, an emulsifier that is not metabolized [19] (Table 4).

20 Table 3. Chemical composition of mDixon medium. Componen Description or molecular structure Percentage t

malt extract

Bile composition (water 92 g/dL, bile salts 6 g/dL, bilirubin 0.3 g/dL, cholesterol 0.3 to 0.9 g/dL, FA 0.3 to 1.2 g/dL, lecithin 0.3 g/dL and 200 meq/L 3.6 inorganic salts [57] desiccated oxbile Lecithins: phospholipids, glycolipids or triglyceride. Glycerophospholipids as phosphatidylcholine, phosphatidylethanolamine, 2.0 phosphatidylinositol, phosphatidylserine, and phosphatidic acid [20]

Natural sources of amino-acids, peptides and proteins

peptone 0.6

glycerol 0.2

oleic acid 0.2

Tween 40 1

Table 4. Chemical composition of oleic acid (OA), palmitic acid (PA) and OA+PA media. The components of the minimal medium (MM) are listed in Materials and Methods.

Percentag Media Components Components found in the GC-FID analysis[19] or molecular structure e

! 78%

! 10%

3% 1% of Oleic ! acid 2%

! OA

! 1%

1% Tween 40

1% of a mixture 50:50 of Oleic ! acid and Palmitic acid

OA+PA

1% Tween 40

1% Brij-58 !

1% of Palmitic acid 98%

2% ! PA

1% Tween 40

1% Brij-58 !

7.3. Yeast growth

The strain CBS 1878 of M. furfur was purchased from the Fungal Biodiversity Center (Westerdijk institute, Utrecht, The Netherlands), and used in this study. Two loops of cells from 7-day-old mDixon agar grown colonies were suspended in 5 mL sterile water, of which 3 mL was added to 27 mL mDixon broth or 27 mL of MM; containing 4.2 mM oleic acid (Carlo Erba, Val-de-Reuil, France) and / or 4.2 mM palmitic acid (Merck, Darmstadt, Germany) supplemented with 1% Brij-58. After 3 days of growth at 180 rpm, 0.3 mL culture was used to inoculate 29.7 mL of fresh mDixon broth or MM containing 4.2 mM oleic acid and / or 4.2 mM palmitic acid supplemented with 1% Brij-58 in 150 mL Erlenmeyers with 20 mm seals with PTFE septa (JG Finneran, USA) to start the growth phase for the sampling. They were incubated at 33°C and 180 rpm. The reference for the exponential phase was 1.17 days ± 0.17, and for the stationary phase was 3.08 ± 0.19 days. These times were previously validated as the mid-point of exponential phase growth, and the initial stationary phase, respectively, in the mDixon medium (Fig 4) [20].

Figure 4. Growth curves of M. furfur in mDixon, oleic acid (OA), palmitic acid (PA) and OA+PA medium.

7.4. VOCs Extraction VOCs released by each growth media was analysed ﬁrst using three experimental replicates. These samples were used as medium control analysis, with the aim to diﬀerentiate VOCs from the medium and those emitted by M. furfur CBS 1878. Afterwards, VOCs from this strain of M. furfur in both exponential and stationary growth phase were analysed in the four liquid media: mDixon and MM supplemented with oleic acid (OA), or oleic and palmitic acid (OA+PA), or palmitic acid (PA). Five replicates were used to analyse the VOCs from M. furfur growing in each experimental treatment (De, Ds, Oe, Os, OPe, OPs, Pe, Ps).

In order to extract a wide range of VOCs a SPME fiber of Divinylbenzene/Carboxen/ Poly(dimethylsiloxane) (DVB/CAR/PDMS, 50 μm /30 μm, gray) (SUPELCO, PA, USA), with different polarities was used. The fiber was exposed for 20 min to the headspace of the erlenmeyer maintained at 33 °C, where the yeast were growing, and the agitation

24 diminished to 90 rpm during the sampling, in order to capture the VOCs released by the yeast without compromising the state of the ﬁber.

7.5. Chromatographic analysis and annotation

Subsequently, the desorption process was carried out in the GC HP 6890 Series equipped with an Agilent Mass Selective Detector 5973 (Agilent technologies, Palo Alto, CA, USA), at 250°C using splitless injection and a flow rate of 1.3 mL/min. Separation was performed on a BP-5 capillary GC column (30 m x 0.25 mm x 0.25 μm, SGE, USA) using helium as carrier gas. The temperature gradient program started at 40°C held for 0.5 min, followed by an increase to 60°C at a rate of 6°C/min, then the temperature increased until 150°C, using a rate of 3°C/min, and finally to 250°C at 10°C/min, maintaining this temperature for 6 minutes. The GC-MS filament source and the quadrupole temperature were set at 230 and 150°C, respectively. The electron ionization (EI) source was set at 70 eV and the mass spectrometer was operated in full scan mode applying a mass range from m/z 30 to 300 at a scan rate of 2.0 scan/s. All samples, including linear alkanes, were run under the same chromatographic conditions. Linear alkanes of the series C8–C20, used for the determination of retention indexes (RI), were purchased from Sigma-Aldrich and used later for the tentative assignation of compounds.

7.6. Data analysis To conduct the analysis of the GC-MS, the profiles with VOCs obtained in five biological replicates for each Medium-phase treatment were analyzed using the MSD ChemStation D. 02.00.275 (Agilent technologies) and automatic integration using a threshold of 14 units was performed. Only those peaks absent in any of the three replicates of the medium control analysis were considered as volatile released by M. furfur. Tentative annotation of compounds was done using NIST MS search 2.0 with NIST 14 database and comparing experimental RI with RI of compounds reported in the literature from the same (or equivalent) column. Exported data files of chromatograms in .csv format were used for plotting chromatograms later. Integration results were reviewed and peak areas used to construct a matrix reporting tentatively annotated volatile compounds as columns/variables and peak areas estimation in each chromatogram as rows/observations. As grouping variables, we included the four media (mDixon, OA, OA+PA and PA), both growth phases (exponential and stationary), and the combination medium-phase of the eight experimental treatments (De, Ds, Oe, Os, OPe, OPs, Pe, Ps). Each peak area was transformed using a Hellinger transformation, adequate for matrices with many zeros [58] and afterwards we applied Pareto scaling, useful for chromatographic analysis on GC-MS platform [59].

To investigate the hypothesis that supervised techniques could afford more informative models than unsupervised analysis for the differential data from each treatment, and thus unravel variables that influence class separation (differentially distributed in the treatments), each data set was analyzed by three techniques. Transformed and scaled matrix of VOCs profiles were analyzed using Hierarchical Cluster Analysis (HCA), Principal Component Analysis (PCA) and Projection to Latent Structures Discriminant Analysis (PLS-DA), scaling data to have unit variance and zero centered. HCA is a statistical tool to pool samples based on similarities, by measuring distances between all possible pairs of samples and plotting a topology for objects (in this case samples for each treatment) and variables (in this

25 case VOCs). PCA analysis was done for reducing redundancy between correlated VOCs. To support this analysis we used PERMANOVA with 999 permutations using Euclidean distances in order to discriminate which grouping variable (medium, growth phase, or both) better explained the volatiles profiles’ differentiation. The PLS-DA model intend to find components or latent variables that correlates variation from VOCs matrix to a response variable Y. The Y response variable of the model were the eight experimental medium-phase treatments (De, Ds, Oe, Os, OPe, OPs, Pe, Ps). PLS-DA was cross-validated using seven groups (Q 2 > 0.5 was considered significant [60], and the significance of the Q 2 value was tested with 1000 random permutations of the VOCs dataset. We performed all statistical tests using R studio software (http://www.rstudio.org/) and the packages stats, HybridMTest (http://bioconductor.org/packages/HybridMTest/), vegan (https://CRAN.R-project.org/ package=vegan), MetabolAnalyze (https://CRAN.R-project.org/package=MetabolAnalyze ), psych (http://CRAN.R-project.org/package=psych), ropls (http://bioconductor.org/ packages/ropls/) and ggplot2 (https://CRAN.R-project.org/package=ggplot2).

8. RESULTS AND DISCUSSION

8.1. Proﬁling of VOCs from M. furfur

HS-SPME/CG-MS analysis was applied to determine the profiles of volatile compounds from M. furfur grown in four different media: mDixon, oleic acid (OA), oleic and palmitic acid (OA+PA) and palmitic acid (PA) in both exponential and stationary phase. mDixon medium is the more complex with a supplementation of different lipids (see Table 3 for details about the composition), while the other three medium were included as restricted dietary conditions to analyze the importance of this unsaturated and saturated fatty acid regulating the VOCs production in M. furfur (see Table 4 for details about the composition). With the objective of analysing exclusively VOCs released by the organism we established the following criteria for including a compound in the following matrices: (1) the VOC was absent in any replicate of any medium control analysis (see more details in Materials and Methods section); and, (2) the VOC was found in more than one replicate of the yeast. Qualitative differences between the chromatographic profiles of the eight experimental treatments (Fig. 5) show a combination of both: the VOC released by M. furfur and the VOC released from growth media, during sampling.

Figure 5. Chromatograms of M. furfur in exponential and stationary phases grown in four media with diﬀerent FA content. De = Dixon in Exponential phase, Ds = Dixon in Stationary phase, Oe = Oleic acid in Exponential phase, Os = Oleic acid in Stationary phase, OPe = Oleic acid + Palmitic acid in Exponential phase, OPs = Oleic acid + Palmitic acid in Stationary phase, Pe = Palmitic acid in Exponential phase, Ps = Palmitic acid in Stationary phase.

HS-SPME/CG-MS analysis of M. furfur allowed the description of 61 diﬀerent VOCs (Table 5), among which alkanes, alcohols, ketones, and furanic compounds were the most prevalent. Carbon dioxide, 1-hexanol, octane, and nonane showed the higher peak areas. The lack of terpenes, a common VOC found in other species of fungi [48,61–63] is remarkable. Average peak areas, standard deviations and number of replicates in which each VOC was found, are summarized in Table 5. Tentative annotation of 27 compounds was carried out by comparison of mass spectra with databases and experimental RI deviations below 30 units from the theoretical RI. In opposition, 30 compounds were annotated to a speciﬁc structure or at least to a chemical class based only on mass spectra comparisons. From these, most of alkanes could not be exhaustively annotated because their EI fragmentation patterns are very similar. Four compounds were annotated as unknowns.

The presence of γ-dodecalactone, 1-butanol, 3-methyl, and 1-hexanol, previously described as signature VOCs of Malassezia was reported [39]. The γ-dodecalactone was detected in the mDixon and the OA medium in both growth phases, while the 1-hexanol was found in all treatments, with the exception of the PA medium. In contrast to the first analysis of Labows et al. (1979) who found many different gamma lactones in the volatile profile of Malassezia, we detected the presence of only one of these gamma lactones. These differences could be related to: genetic differences from the tested species, because the 1979 current synonyms of Pityrospoum ovale does not correspond to M. furfur; to growing conditions, because they used solid media and Tween 80 (a detergent as Tween 40, but with the difference that the lateral chain is composed by oleic acid, instead of palmitic acid); or to sampling conditions, because they absorbed the compounds in tenax, instead of SPME fibers. Although mDixon medium also contains OA, the dietary importance of this unsaturated acid in the biosynthesis of γ-dodecalactone in M. furfur is still challenging. If the presence of OA had been controlling this pathway, we will had observed also the emission of γ-dodecalactone in the OA+PA medium, unless PA was an inhibitory supplement. In other fungi OA is indeed a suitable substrate for the biosynthesis, which suffers hydroxylation, β-

27 oxidation, isomerization and lactonisation reactions [32]. Taking into account the rest of VOCs, it is clear that the DVB/CAR/PDMS -SPME ﬁber allowed the sampling of a wider range of polarities than the sampled by Labows et al using tenax (1979), evidenced in the description of more compounds not previously reported for Malassezia and even for Fungi according to the data summarized in the mVOC 2.0 Database. 16 compounds previously reported in other fungi were marked with * in the Table 5.

From all volatiles´ profiles, 25 compounds were detected in the five replicates of any media, but the absence/presence of most of the compounds was highly variable. The maximum number of VOCs (27), was reported for the stationary phase of the mDixon medium (Ds), followed by 23 compounds detected in the exponential phase of the OA medium (Oe). Subsequently 20 compounds were encountered in the exponential phase of the OA+PA medium (OPe) and finally, just 8 VOCs were revealed as emitted in the stationary phase of the PA medium (Ps). PA medium produced a drastic reduction in the number of VOCs and even in the chemiodiversity, reduced almost exclusively to alkanes. Comparisons between our analysis of VOCs from M. furfur (grown in mDixon, OA, OA+PA, and PA media) and from M. slooffiae, M. obtusa, M. sympodialis (previously known and analysed as a single specie Pityrospoum ovale), M. pachydermatis, and M. globosa (previous P. orbiculare) (grown in Littman agar and Sabouraud-dextrose agar) [39,64] demonstrates that the composition of the lipid substrate is essential for both: growing and determining the quantity of VOCs released. The interaction between these two variables: growth and diversity of volatiles in M. furfur are still unknown because growing (determined by the CFU) and volatiles sampling (through HS-SPME/GC-MS) could not be performed simultaneously on each replicate for maintaining the headspace conditions without air contamination. Still ignoring the direct effect of growing on VOCs diversity, we have indirect evidence that the lack of growing (static pattern observed for the growth curves in PA and OA+PA media in the Fig. 4) does not determine the quantity of VOCs released. While the number of VOCs detected shows that PA is a more challenging medium for the synthesis of VOCs, in the other growth-static medium (OA+PA) M. furfur emitted more than 15 volatiles.

33 8.2.VOCs proﬁle diﬀerentiation in four growth media in exponential and stationary phase

Using univariate (UVA) and multivariate (MVA) statistical analysis we proved that growth media and phases produced diﬀerences in the volatile proﬁles emitted by M. furfur.

We demonstrated that for 17 VOCs peak areas changes significantly when the Medium- growth phase is modified (ANOVA, 2.37 < F < 560.26, P < 0.05) (Kruskal- Wallis test, 14.34 < K < 34.60, P < 0.05) (Table 6). We reported VOCs with p < 0.05 in parametric and non-parametric test, because more than a half of compounds (34 of 61) had not normal distribution (Shapiro-Wilk test, 0.59 < W < 0.91, P < 0.05). Fig. 6 summarize the semi-quantitative differences in the total peak areas between each treatment for this 16 VOCs. Peak areas of carbon dioxide released by M. furfur grown in mDixon were higher than areas from other treatments, as well as growing (Fig. 4), which could evidences an increased respiration on this medium. Octane and nonane were released in the mDixon and OA+PA media having semi-quantitative differences in their areas. 1-Hexanol, 1-butanol, and the Isomer3 of methyldodecane were emitted in all media, except PA. In contrast, the rest of VOCs were released by M. furfur almost exclusively in the mDixon or the OA+PA media (Fig. 6). Differences between exponential and stationary phases were also observed, being some compounds increased in the stationary phase, as dimethyl sulphide and others decreased, as octane. The presence of dimethyl sulphide, and other volatile organic sulphur compounds (VOSC) has been previously reported in bacteria [65], ascomycota yeasts [45,66], and basidiomycota yeast [67], but to the best of our knowledge this is the first report of this compound in the genus Malassezia. Another commensal microfungi, Candida albicans share the presence of dimethyl sulfide, 1- propanol, 2-methyl-1-pentanone, and 1-butanol, 3-methyl- with M. furfur [68], but the ecological functions of these VOCs for both species remain unknown. The presence/ absence of some specific VOC in M. furfur growing in certain media highlights the importance of analysing the volatiles’ profiles (in plural) of a species, especially when the objective of a research is tracking biomarker compounds with chemotaxonomy, biotechnological purposes [30] or therapeutic purposes [48]. Intra-species variation in the emission of VOCs should be studied whenever possible. There are previous reports that using UVA show that the emission of some VOCs in other fungi as Ceratocystis sp., Lentinus lepideus, Aspergillus versicolor, Aspergillus fumigatus, Penicillium commune, Cladosporium cladosporioides, Paecilomyces variotii, Phialophora fastigiata Trichoderma spp., and Kluyveromyces marxianus depends highly on differences in the nutritional supplement in the growth medium [24,40,41,69–71].

34 Table 6. Signiﬁcant statistical univariate parameters found on 17 VOCs from M. furfur and compounds highlighted by PCA and PLS-DA. Statistics from ANOVA Statistics from Kruskal-Wallis PLS- PCA DA Compound F pval K pval

✓ ✓ 16.1 < 0.001 25.8 < 0.001 Carbon dioxide

✓ 36.8 < 0.001 33.8 < 0.001 Octane

92.5 < 0.001 34.6 < 0.001 ✓ ✓ 1-Hexanol

Nonane 13.5 < 0.001 26.7 < 0.001

2-Pentanone 13.1 < 0.001 25.7 < 0.001

1-Butanol-3-methyl 6.9 < 0.001 20.3 0.0049

✓ 7.2 < 0.001 32.8 < 0.001 1-Butanol

106.8 < 0.001 22.4 0.0022 ✓ 6-Hepten-1-ol

cis-2-Nonene 132.1 < 0.001 22.4 0.0022

✓ ✓ 56.2 < 0.001 22.0 0.0025 Isomer5 of methyldecane

✓ ✓ 29.9 < 0.001 14.3 0.0450 Dimethyl sulﬁde

✓ ✓ 371.0 < 0.001 22.4 0.0022 5-Undecene

94.5 < 0.001 22.0 0.0026 ✓ ✓ Isomer2 of methylundecane

✓ ✓ 133.7 < 0.001 22.0 0.0026 Isomer1 of methyldecane

✓ 51.0 < 0.001 19.0 0.0083 1,2-Epoxyundecane

Isomer3 of methyldodecane 4.0 0.0030 15.9 0.0258

✓ ✓ 116.2 < 0.001 22.2 0.0023 Furan, tetrahydro-2-methyl

Figure 6. Pirate-plots representing univariate statistical differences in the absolute peak areas of 17 VOCs in which the grouping variable Medium-Phase significantly affects the emission (P < 0.05 in ANOVA and Kruskal-Wallis test). Variation of data is represented by Bayesian High-Density Intervals after running 10000 iterations. Bars show median values for each treatment. The first line of plots is y-scaled from 0 to 16x107, in the following line y-axis were scaled from 0 to 1.2x107, while the others two lines are scaled from 0 to 1.2x106.

Using Hierarchical Cluster Analysis (HCA), the first unsupervised multivariate technique, we discovered that the five replicates from M. furfur grown in the mDixon and OA+PA media in exponential and stationary phase has a signature profile easily

36 differentiated from the profiles of other treatments, but the grouping is not related to a specific chemical class of compounds (Fig. 7). These aggrupation validates the reproducibility of these volatiles’ profiles, although the evident high variance of transformed and scaled peak areas of some VOCs in the heatmap. In addition, topology demonstrates that a dynamic transformation of VOCs production occurs when yeast moves from exponential to stationary phase times, even for the OA+PA medium that remain growing static (Fig. 7 and Fig 4). In the PA medium there are higher semi-quantitative similarities between samples of different growth phases, which explain the lack of clustering of the five replicates of the exponential and stationary phase. Nevertheless volatiles’ profiles from PA medium are clustered together demonstrating the VOCs differentiation. In contrast, the OA medium seems did not have a distinctive volatiles’ profile, caused by the higher variability between the VOCs found in each replicate in both growth phases. Some of replicates from OA medium in both growth phases, had just a couple of VOCs with low intensity-peak areas, and were clustered with the mDixon medium. The other group of replicates are clustered as the outgroup of the Dixon, OA and OA+PA group, having more VOCs with higher peak areas. In relation to the main biochemical pathways related with microbial VOCs, we can suggest that from the chemical groups analysed in M. furfur, furanic compounds as lactones, methylketones, esters, alkenes and maybe alcohols could be obtained through catabolism of fatty acids; this last group also could be synthesised from pyruvate pathways intermediaries; alkanes probably could be obtained from poliketides synthesised in the acetate pathway, while sulphur and nitrogen compound could have been produced from aminoacids [42,43].

Figure 7. Heat-map of Hellinger transformed and Pareto-scaled peak areas from 61 tentative annotated volatiles found in M. furfur analysis using HS-SPME/GC-MS. Dendro- grams represent similarities between samples (objects) and VOCs (variables) using Eu- clidean distances as a measure. Clusters of Medium-phase samples were colored according to the four growth media, while the compounds were colored according to the chemical group. Compounds with * represent VOCs reported previously in Fungi in the mVOC 2.0 Database.

We performed a second multivariate non-supervised approximation using PCA analysis to summarize the total variation of the volatiles’ proﬁles of M. furfur in each medium and growth phase, reducing redundancy between correlated VOCs (Fig. 8).

38 The three first PCs represented 56.35% of explained variance from 61 volatiles annotated in the original Matrix. In comparison to the analysis obtained with the HCA, all media and medium-phases are differentiated in a 3-dimensional space, including the samples obtained from OA medium, which were grouped together in the PCA space, despite having the higher variation, represented by bigger ellipses. In order to discriminate whether the medium or the growth phase or both were the variables affecting the volatiles´ profiles produced by M. furfur, we considered all three as grouping variables in the PCA analysis. Taking into consideration the total multivariate volatile profile, the differentiation is better explained through the interaction medium- phase (PERMANOVA, R = 0.6945, F = 10.393, P < 0.001). Furthermore, when considered separately as independent variables, medium (PERMANOVA, R = 0.5457, F = 14.415, P < 0.001) and phase (PERMANOVA, R = 0.0516, F = 2.068, P = 0.043), both variables were also statistically supported as volatiles’ profile differentiation factors. PC1, which discriminates samples from both mDixon treatments, is associated with an increment in dimethyl sulfide, 1-propanol, 2-methyl, isomer1 of methyldecane, isomer5 of methyldecane and the isomer2 of methylundecane. PC2, which discriminates samples from the OA+PA medium in the lower values, is associated with an increment in the isomer2 of 1-undecene, the isomer2 of methyldodecane, the isomer2 of trimethyldodecane, the isomer of methylundecanol and the isomer2 of methyltridecane. PC3, which discriminates samples from PA medium in the lower values, is associated with an increment in Furan, tetrahydro-2-methyl, 1-hexanol, Acetic acid, pentyl ester, 5-undecene and a decrease in carbon dioxide. The half of the aforementioned compounds (9) also were stood out in the UVA (Table 6 and Fig. 6), but the PCA allows the interpretation of the direction of differentiation of each compound according to the loading values on each axis. To the best of our knowledge there are many investigations who have used MVA to evaluate differentiation of the metabolites’ profiles of bacteria and fungi to discriminate species, even with chemotaxonomy purposes [47,72–74], but this is the first study where MVA analysis demonstrates intra- species fungi VOCs profiles differentiation caused by changes in both, medium or/and the phase of growth. There are some examples were the effect of one of these variables on the volatiles´ profile has been tested on bacteria [75–77], but studies on fungi are less frequent [78].

Figure 8. 3-dimensional PCA of the volatiles’ profiles of M. furfur grown in eight different experimental treatments. Ellipses in this plot are confidence intervals of 95% using normal-distribution.

We performed a third multivariate supervised approximation using PLS-DA (Fig. 9a). The quality of the model was evaluated by R2X, R2Y, and Q2 metrics. R2X and R2Y supports the goodness of fit, while Q2 > 0.5 reflects that the predictability of the model is significant (R2X = 0.686, R2Y = 0.624, Q2 = 0.505). This analysis demonstrate that the volatile´s profile from mDixon and OA+PA medium are differentiated, as also occurs for the exponential and stationary phases in these two medium. To ensure that a model based on a subset of the data can perform equally well on other subsets we used cross- validation in seven groups and 1000 permutations. Despite the differences in the number of VOCs between OA and PA media, discussed before, the Fig. 9a shows a considerable overlap of these two media in both growth phases, suggesting that some metabolic changes produced on M. furfur in both restricted dietary conditions could be similar, in comparison with the metabolic performance in OA+PA or mDixon media. Our findings using this supervised technique demonstrate that the absence of both: unsaturated oleic acid, and saturated palmitic acid, produces major changes in the fungal VOCs emitted by M. furfur. Major changes on the VOCs produced by S. cerevisae during the fermentation of different beverages occurs when linoleic acid is added or supplemented at different concentrations [79]. In the case of S. cerevisae, these changes are very important for food quality of biotechnological and industrial process, whereas for M. furfur these modifications could be very important defining the yeast-other microbes interactions or the VOCs released by the human skin where this yeast live. Surprisingly the VOC profile in the OA+PA medium is different to that from the sum of OA and PA medium individually (Table 5, Fig. 8 and Fig. 9a). This has been previously observed in Trichoderma spp., where the volatile profile found combining three different amino acids in the growth medium was dif-

40 ferent from the sum of the proﬁles obtained when each amino acid was supplied individually [40].

VIP (variable importance on projection) was used to assess the significance of the variables in the differentiation between treatments. Usually values higher than 1 are considered important. 27 VOCs from M. furfur followed this requirement (Fig. 9b). Com- pounds found exclusively in one treatment or medium has the higher VIPs values. For example, the 3-heptene, found in M. furfur growing in the OA+PA medium in exponential phase (OPe) is the compound which better differentiates the four medium and two growth phases, followed by Isomer 3 of 1-undecene, a signature compound in the OA+PA medium in stationary phase (OPs). 12 from these 27 compounds were also signalled as statistical significant using UVA (Table 6 and Fig. 6), coincidences that validates the interpretation of these two types of approximations. Nevertheless, the VIP values from PLS-DA of- fer more information, because organized the importance of each VOC discriminating the experimental treatments. This is not possible for the UVA, where each VOC is analysed separately and not correlated to other compounds of the profile. 8 of the 27 VIP compounds had the higher loadings in the PCA analysis, and also were highlighted as relevant in the UVA (carbon dioxide, 1-hexanol, the isomer 5 of methyldecane, dimethyl sulphide, 5-undecene, isomer2 of methylundecane, isomer1 of methyldecane and furan, tetrahydro-2-methyl). Whereas PCA and PLS-DA analysis shared 10 from these 27 VIP com- ponds. In addition to the last eight, 1-propanol, 2-methyl, acetic acid, pentyl ester were important differentiation factors in both analysis. This findings demonstrate the differences of the multivariate models built under supervised and unsupervised conditions, and the limitations of UVA. While the construction of the PCA model does not maximizes distances between treatments, the PLS-DA construct the multivariate components in function to this response variable (in this case, the medium and phase of growth) [50]. This could explain why the VOCs with the higher VIP (for example, 3-heptene), found exclusively in OPe, was not recognized as significant in the UVA or the PCA.

Ecological meaning of the differentiation that the absence of PA and OA produced on the volatiles’ profiles of M. furfur need to be explored further. The human skin sebum contains a complex mixture of triglycerides, fatty acids, wax esters, sterol esters, cholesterol, cholesterol esters, and squalene. Unsaturated fatty acids are more predominant, but in general, the sebum is composed of stearic acid, oleic acid, linoleic acid, palmitic acid, sapienic acid and palmitoleic acid [80]. In comparison to other animals, the human sebum has relatively high amount of palmitic acid [20], so spatial differences in the PA distribution could lead to spatial changes in the VOCs emitted by M. furfur or other species from human microbiota. Additionally, M. furfur is able not only to grow in presence of OA and PA, but also is able to hydrolyze triglycerides of human sebum and the released free fatty acids, as OA, that cause irritation in individuals susceptible to dandruff [10]. So, the human skin metabolites and the metabolites released by each species from human microbiota should maintain a balance in order to coexist together. From Malassezia already exists an example of an enzyme able to attenuate the formation of biofilms in Staphylococ- cus aureus, an opportunistic bacteria from human skin microbiota [12], but volatile metabolites emitted by the microbiota could have similar effects, however this should be corroborated. VOCs functioning as quorum sensing molecules [35,61] or antibiotics [42,81] has been reported in other microorganisms as Candida albicans, Trichoderma spp. and Muscodur albus, but the ecological function of the VOCs emitted by M. furfur in me-

41 dia with diﬀerent lipid composition in both growth phases should be studied in the future. The aforementioned studies demonstrates how any alterations on human skin could be responsible of a disequilibrium in the homeostasis of the total microbiota and maybe relates with the interconversion from commensal to pathogenic states in M. furfur, as well as in other opportunistic species.

b) Figure 9. a) PLS-DA of volatiles’ profiles of M. furfur grown in eight different experimental treatments. Ellipses in this plot are confidence intervals of 95% using t-distribution. b) VIP-Plot showing the more important VOC from M. furfur differentiated between four medium and two growth phases. The VIP score of each compound was calculated as a weighted sum of the squared correlations between the PLS-DA components and the original variables.

43 8.3. VOCs proﬁle comparison in Agilent GC-MS with preliminary analysis in Thermo GC-MS

Initially, we performed a preliminary analysis from the VOCs of M. furfur on a different GC-MS. For this analysis, we ran only three replicates for each treatment (De, Ds, Oe, Os, O+Pe, O+Ps, Pe, Ps) and one replicate for the medium control analysis. Unfortunately, this data could not be used in the final paper for a methodological difficulty caused between runs of different batches. Nevertheless, in this thesis we present and compare the results obtained with the objective to discuss the analytical implications of this difficulty.

The methodological aspects related to chemicals, media preparation, yeast growth, and extraction were equivalent to those presented previously. The analysis was performed on a Thermo Trace 1300 Gas Cromatograph, ISQLT Single Quadrupole Mass Spectrometer and a GC column ZB-5MSi (HP-Zebron 30 m x 0.25 mm x 0.25 µm). During the VOCs analysis of M. furfur, another researcher used the same instrument to run a different analysis and contaminated the column. To overcome this problem and complete our analysis, we change the column for another one of the same brand and with exactly the same characteristics. At the end of the data analysis and annotation of VOCs, we included the type of column (A or B) as a grouping variable in the Matrix, in order to analyze the significance of its effect. Peaks from each chromatogram were analyzed using the Qual Browser of Xcalibur 3.0 software (Thermo Scientific) and manual integration of each peak with a S/N ratio over 3 was performed. Annotation of compounds was done using NIST MS search 2.0 with NIST 14 database. Each peak area was transformed as described in Materials and Methods for data obtained from the GC-MS Agilent. We built a Matrix where the columns/variables were the tentatively annotated volatile compounds and rows/observations were the peak areas estimation in each chromatogram. We compared the annotated VOCs reported by each GC-MS, finding that from the total 61 compounds found in M. furfur using the GC-MS Agilent, only 13 of these compounds were also reported in the Thermo equipment. A total number of 48 compounds were reported in the preliminary analysis developed in the Thermo, but 22 of these VOCs were found in the medium control analysis replicates performed on the GC-MS Agilent, so these were released by the growth media, and not by M. furfur. The other 13 VOCs were principally aromatic compounds detected in one replicate and annotated as possible contaminants. Although the comparison between the analyses performed in the GC-MS Agilent and the GC-MS Thermo could not be performed directly, because all the experimental conditions and dates of analysis were not the same, these results suggest a broader sampling of compounds in the GC-MS Agilent. The causes for these differences could be related to the sensibility of the equipment (who has not been quantitatively compared yet), or to variances produced by random errors.

Finally, to compare the multivariate profiles we exclusively performed a PCA analysis with these data (Fig. 10). The three first PCs represented 54.8% of explained variance from 48 volatiles annotated in the original Matrix. In order to discriminate whether the column, the medium or the growth phase or both were the variables affecting the volatiles´ profiles produced by M. furfur in the GC Thermo, we considered all four as grouping variables in the PCA analysis. Taking into consideration the total multivariate volatile profile, the differentiation is better explained through the interaction medium-phase (PERMANOVA, R = 0.7219, F = 5.934, P < 0.001). Furthermore, when considered separately as independent variables, medium (PERMANOVA, R = 0.4347, F

44 = 5.126, P < 0.001) and phase (PERMANOVA, R = 0.0852, F = 2.048, P = 0.025), were also statistically supported as volatiles profile differentiation factors. In addition, when we considered the effect of the type of column we discovered that changing the column produced significant differences (PERMANOVA, R = 0.2300, F = 6.568, P < 0.001) in the volatiles profiles of M. furfur. Comparing the two PCA from the GC-MS Agilent and GC-MS Thermo (Fig 8 and Fig 10), we discovered that despite the qualitative differences between the chromatographic profiles, both analysis shared a major differentiation of the VOCs´profile of the mDixon medium in relation to the other media. The significance of changing the column on the analysis performed in the Thermo equipment was the reason that conducted to repeating all of the experimental analysis, in order to improve the reproducibility of data. From this preliminary experience, we learned that we should increase the number of replicates for experimental treatments and for medium control analysis to at least 4 replicates. In addition, we performed all of the analysis in a single workflow to avoid interferences between runs of different batches.

Figure 10. 3-dimensional PCA of the volatiles’ profiles of M. furfur grown in eight different experimental treatments and analysed in the Thermo GC-MS. Ellipses in this plot are confidence intervals of 95% using normal-distribution.

9. OUTPUT OF THE STUDY

The research of this thesis is in prep. for submission to the Journal Molecules.

10. CONCLUSIONS

Combining the results of UVA and MVA we have strong evidence to support that changing the lipid composition of growing media in M. furfur CBS 1878 produces qualitative and semi-quantitative differences in their volatiles’ profiles. Additionally, the phase of growth also affects the type and intensity of the VOCs emitted, despite growing of the yeast is not equal in the tested media. Different VOC were emphasized as meaningful differentiation factors in the UVA and MVA, being carbon dioxide, 1- hexanol, the isomer 5 of methyldecane, dimethyl sulphide, 5-undecene, isomer2 of methylundecane, isomer1 of methyldecane and furan, tetrahydro-2-methyl supported for all the techniques. The complexity of the physiological and biochemical changes produced by the stressful nutritional conditions in OA, OA+PA and PA medium should be studied further. mDixon medium, with more complex lipid composition promotes the most diverse chemical profile, whereas PA medium restricts the chemiodiversity and abundance of the VOCs produced. The significance of our findings deserves future in vivo research to analyse how variable are the lipid compositions from different human skins, and how these differences promote changes in the volatiles’ profiles produced by M. furfur. These modifications on human bouquets will determine many ecological interactions where VOCs play an important role, as microbre-microbe, microbe-host and microbe-insects interactions. Another important direction of research is the comparison of the VOCs profiles when M. furfur behaves as commensal vs. pathogenic, with the aim to understand better the virulence of this microorganism and how changes according to all the dynamic complex transformations on the microbiota.

11. BIBLIOGRAPHY

1. Shifrine, M.; Marr, A.G. The Requirement of fatty acids by Pityrosporum ovale. J. Gen. Microbiol. 1963, 32, 263–270, doi:10.1099/00221287-32-2-263. 2. Xu, J.; Saunders, C.W.; Hu, P.; Grant, R.A.; Boekhout, T.; Kuramae, E.E.; Kronstad, J.W.; DeAngelis, Y.M.; Reeder, N.L.; Johnstone, K.R.; Leland, M.; Fieno, A.M.; Beg- ley, W.M.; Sun, Y.; Lacey, M.P.; Chaudhary, T.; Keough, T.; Chu, L.; Sears, R.; Yuan, B.; Dawson, T.L. Dandruﬀ-associated Malassezia genomes reveal convergent and divergent virulence traits shared with plant and human fungal pathogens. Proc. Natl. Acad. Sci. 2007, 104, 18730–18735, doi:10.1073/pnas.0706756104. 3. Wu, G.; Zhao, H.; Li, C.; Rajapakse, M.P.; Wong, W.C.; Xu, J.; Saunders, C.W.; Reeder, N.L.; Reilman, R.A.; Scheynius, A.; Sun, S.; Billmyre, B.R.; Li, W.; Averette, A.F.; Mieczkowski, P.; Heitman, J.; Theelen, B.; Schröder, M.S.; De Sessions, P.F.; Butler, G.; Maurer-Stroh, S.; Boekhout, T.; Nagarajan, N.; Dawson, T.L. Genus-Wide comparative genomics of Malassezia delineates its phylogeny, physiology, and niche adaptation on human skin. PLoS Genet. 2015, 11, 1–26, doi:10.1371/journal.pgen.1005614. 4. Lorch, J.M.; Palmer, J.M.; Vanderwolf, K.J.; Schmidt, K.Z.; Verant, M.L.; Weller, T.J.; Blehert, D.S. Malassezia vespertilionis sp. nov.: a new cold-tolerant species of

46 yeast isolated from bats. Persoonia - Mol. Phylogeny Evol. Fungi 2018, 41, 56–70, doi:10.3767/persoonia.2018.41.04. 5. Gao, Z.; Perez-Perez, G.I.; Chen, Y.; Blaser, M.J. Quantitation of major human cu- taneous bacterial and fungal populations. J. Clin. Microbiol. 2010, 48, 3575–3581, doi:10.1128/JCM.00597-10. 6. Ashbee, H.R.; Evans, E.G. V. Immunology of diseases associated with Malassezia species. Clin. Microbiol. Rev. 2002, 15, 21–57, doi:10.1128/CMR.15.1.21-57.2002. 7. Mayser, P.A.; Lang, S.K.; W., H. Pathogenicity of Malassezia yeasts. In The Mycota. Human and Animal Relationships.; Brakhage, A., Zipfel, P., Eds.; Springer-Verlag Berlin Heidelberg: Jena, 2008; pp. 115–141 ISBN 978-3-540-87406-5. 8. Boekhout, T.; Mayser, P.; Guého-Kellermann, E.; Velegraki, A. Malassezia and the skin. Science and clinical practice; Boekhout, T., Mayser, P., Guého-Kellermann, E., Velegraki, A., Eds.; Springer Berlin Heidelberg: Berlin, Heidelberg; ISBN 978-3-642- 03615-6. 9. Kesavan, S.; Holland, K.T.; Ingham, E. The effects of lipid extraction on the im- munomodulatory activity of Malassezia species in vitro. Med. Mycol. 2000, 38, 239– 247, doi:10.1080/mmy.38.3.239.247. 10. DeAngelis, Y.M.; Gemmer, C.M.; Kaczvinsky, J.R.; Kenneally, D.C.; Schwartz, J.R.; Dawson, T.L. Three etiologic facets of dandruff and seborrheic dermatitis: Malassezia fungi, sebaceous lipids, and individual sensitivity. J. Investig. Dermatol. Symp. Proc. 2005, 10, 295–297, doi:10.1111/j.1087-0024.2005.10119.x. 11. Bjerre, R.D.; Bandier, J.; Skov, L.; Engstrand, L.; Johansen, J.D. The role of the skin microbiome in atopic dermatitis: a systematic review. Br. J. Dermatol. 2017, 177, 1272–1278, doi:10.1111/bjd.15390. 12. Li, H.; Goh, B.N.; Teh, W.K.; Jiang, Z.; Goh, J.P.Z.; Goh, A.; Wu, G.; Hoon, S.S.; Raida, M.; Camattari, A.; Yang, L.; O’Donoghue, A.J.; Dawson, T.L. Skin commensal Malassezia globosa secreted protease attenuates Staphylococcus aureus biofilm formation. J. Invest. Dermatol. 2018, 138, 1137–1145, doi:10.1016/j.jid. 2017.11.034. 13. Cogen, A.L.; Nizet, V.; Gallo, R.L. Skin microbiota: a source of disease or defence? Br. J. Dermatol. 2008, 158, 442–455, doi:10.1111/j.1365-2133.2008.08437.x. 14. Bouslimani, A.; Porto, C.; Rath, C.M.; Wang, M.; Guo, Y.; Gonzalez, A.; Berg-Lyon, D.; Ackermann, G.; Moeller Christensen, G.J.; Nakatsuji, T.; Zhang, L.; Borkowski, A.W.; Meehan, M.J.; Dorrestein, K.; Gallo, R.L.; Bandeira, N.; Knight, R.; Alexan- drov, T.; Dorrestein, P.C. Molecular cartography of the human skin surface in 3D. Proc. Natl. Acad. Sci. 2015, 112, E2120–E2129, doi:10.1073/pnas.1424409112. 15. Rosenthal, M.; Goldberg, D.; Aiello, A.; Larson, E.; Foxman, B. Skin microbiota: Mi- crobial community structure and its potential association with health and disease. Infect. Genet. Evol. 2011, 11, 839–848, doi:10.1016/j.meegid.2011.03.022. 16. Findley, K.; Oh, J.; Yang, J.; Conlan, S.; Deming, C.; Meyer, J.A.; Schoenfeld, D.; Nomicos, E.; Park, M.; Becker, J.; Benjamin, B.; Blakesley, R.; Bouffard, G.; Brooks, S.; Coleman, H.; Dekhtyar, M.; Gregory, M.; Guan, X.; Gupta, J.; Han, J.; Hargrove, A.; Ho, S.L.; Johnson, T.; Legaspi, R.; Lovett, S.; Maduro, Q.; Masiello, C.; Maskeri, B.; McDowell, J.; Montemayor, C.; Mullikin, J.; Riebow, N.; Schandler, K.; Schmidt, B.; Sison, C.; Stantripop, M.; Thomas, J.; Thomas, P.; Vemulapalli, M.; Young, A.; Kong, H.H.; Segre, J.A. Topographic diversity of fungal and bacterial communities in human skin. Nature 2013, 498, 367–370, doi:10.1038/nature12171. 17. Mayser, P.; Haze, P.; Papavassilis, C.; Pickel, M.; Gruender, K.; Gueho, E. Differenti- ation of Malassezia species: selectivity of Cremophor EL, castor oil and ricinoleic acid for M. furfur. Br. J. Dermatol. 1997, 137, 208–213, doi:10.1046/j. 1365-2133.1997.18071890.x.

47 18. Wilde, P.F.; Stewart, P.S. A study of the fatty acid metabolism of the yeast Pity- rosporum ovale. Biochem. J. 1968, 108, 225–231, doi:10.1042/bj1080225. 19. Triana, S.; de Cock, H.; Ohm, R.A.; Danies, G.; Wösten, H.A.B.; Restrepo, S.; González Barrios, A.F.; Celis, A. Lipid metabolic versatility in Malassezia spp. yeasts studied through metabolic modeling. Front. Microbiol. 2017, 8, 1–18, doi:10.3389/ fmicb.2017.01772. 20. Celis, A.M. Unraveling lipid metabolism in lipid-dependent pathogenic Malassezia yeasts. Ph.D. Thesis, Utrecht University, 2017. 21. Kaneko, T.; Makimura, K.; Onozaki, M.; Ueda, K.; Yamada, Y.; Nishiyama, Y.; Yam- aguchi, H. Vital growth factors of Malassezia species on modified CHROMagar Candida. Med. Mycol. 2005, 43, 699–704, doi:10.1080/13693780500130564. 22. Scotter, J.M.; Langford, V.S.; Wilson, P.F.; McEwan, M.J.; Chambers, S.T. Real-time detection of common microbial volatile organic compounds from medically important fungi by Selected Ion Flow Tube-Mass Spectrometry (SIFT-MS). J. Microbiol. Methods 2005, 63, 127–134, doi:10.1016/j.mimet.2005.02.022. 23. Bazemore, R.A.; Feng, J.; Cseke, L.; Podila, G.K. Biomedically important pathogenic fungi detection with volatile biomarkers. J. Breath Res. 2012, 6, doi: 10.1088/1752-7155/6/1/016002. 24. Heddergott, C.; Latgé, J.P.; Calvo, A.M. The volatome of Aspergillus fumigatus. Eu- karyot. Cell 2014, 13, 1014–1025, doi:10.1128/EC.00074-14. 25. Koo, S.; Thomas, H.R.; Daniels, S.D.; Lynch, R.C.; Fortier, S.M.; Shea, M.M.; Rear- den, P.; Comolli, J.C.; Baden, L.R.; Marty, F.M. A breath fungal secondary metabolite signature to diagnose invasive aspergillosis. Clin. Infect. Dis. 2014, 59, 1733–1740, doi:10.1093/cid/ciu725. 26. Zhang, Q.; Zhou, L.; Chen, H.; Wang, C.-Z.; Xia, Z.; Yuan, C.-S. Solid-phase mi- croextraction technology for in vitro and in vivo metabolite analysis. Trends Analyt. Chem. 2016, 80, 57–65, doi:10.1016/j.trac.2016.02.017. 27. Chung, H.; Lee, N.; Seo, J.A.; Kim, Y.S. Comparative analysis of nonvolatile and volatile metabolites in Lichtheimia ramosa cultivated in different growth media. B i o s c i . B i o t e c h n o l . B i o c h e m . 2017, 81, 5 6 5 – 5 7 2 , d o i : 10.1080/09168451.2016.1256756. 28. Rees, C.A.; Burklund, A.; Stefanuto, P.H.; Schwartzman, J.D.; Hill, J.E. Comprehen- sive volatile metabolic fingerprinting of bacterial and fungal pathogen groups. J. Breath Res. 2018, 12, doi:10.1088/1752-7163/aa8f7f. 29. Morath, S.U.; Hung, R.; Bennett, J.W. Fungal volatile organic compounds: A review with emphasis on their biotechnological potential. Fungal Biol. Rev. 2012, 26, 73– 83, doi:10.1016/j.fbr.2012.07.001. 30. Hung, R.; Lee, S.; Bennett, J.W. Fungal volatile organic compounds and their role in ecosystems. Appl. Microbiol. Biotechnol. 2015, 99, 3395–3405, doi:10.1007/ s00253-015-6494-4. 31. Tejero Rioseras, A.; Garcia Gomez, D.; Ebert, B.E.; Blank, L.M.; Ibáñez, A.J.; Sin- ues, P.M.L. Comprehensive real-time analysis of the yeast Volatilome. Sci. Rep. 2017, 7, 1–9, doi:10.1038/s41598-017-14554-y. 32. Romero-Guido, C.; Belo, I.; Ta, T.M.N.; Cao-Hoang, L.; Alchihab, M.; Gomes, N.; Thonart, P.; Teixeira, J.A.; Destain, J.; Waché, Y. Biochemistry of lactone formation in yeast and fungi and its utilisation for the production of flavour and fragrance compounds. Appl. Microbiol. Biotechnol. 2011, 89, 535–547, doi:10.1007/ s00253-010-2945-0. 33. Braga, A.; Belo, I. Biotechnological production of γ-decalactone, a peach like aro- ma, by Yarrowia lipolytica. World J. Microbiol. Biotechnol. 2016, 32, 1–8, doi:

48 10.1007/s11274-016-2116-2. 34. Wheatley, R.E. The consequences of volatile organic compound mediated bacterial and fungal interactions. Antonie van Leeuwenhoek, Int. J. Gen. Mol. Microbiol. 2002, 81, 357–364, doi:10.1023/A:1020592802234. 35. Hornby, J.M.; Jensen, E.C.; Lisec, A.D.; Tasto, J.J.; Jahnke, B.; Shoemaker, R.; Dussault, P.; Nickerson, K.W. Quorum sensing in the dimorphic fungus Candida albicans is mediated by farnesol. Appl. Environ. Microbiol. 2001, 67, 2982–2992, doi: 10.1128/AEM.67.7.2982-2992.2001. 36. Bennett, J.W.; Hung, R.; Lee, S.; Padhi, S. Fungal and bacterial Volatile Organic Compounds: An overview and their role as ecological signaling agents. In Fungal Associations; Springer Berlin Heidelberg: Berlin, Heidelberg, 2012; Vol. 9, pp. 373– 393 ISBN 9783642308260. 37. Lemfack, M.C.; Nickel, J.; Dunkel, M.; Preissner, R.; Piechulla, B. mVOC: a database of microbial volatiles. Nucleic Acids Res. 2014, 42, D744–D748, doi:10.1093/ nar/gkt1250. 38. Lemfack, M.C.; Gohlke, B.O.; Toguem, S.M.T.; Preissner, S.; Piechulla, B.; Preiss- ner, R. MVOC 2.0: A database of microbial volatiles. Nucleic Acids Res. 2018, 46, D1261–D1265, doi:10.1093/nar/gkx1016. 39. Labows, J.N.; McGinley, K.J.; Leyden, J.J.; Webster, G.F. Characteristic γ-lactone odor production of the genus Pityrosporum. Appl. Environ. Microbiol. 1979, 38, 412–415. 40. Bruce, A.; Wheatley, R.E.; Humphris, S.N.; Hackett, C.A.; Florence, M.E.J. Produc- tion of volatile organic compounds by Trichoderma in media containing different amino acids and their effect on selected wood decay fungi. Holzforschung 2000, 54, 481–486, doi:10.1515/HF.2000.081. 41. Sunesson, A.L.; Vaes, W.H.J.; Nilsson, C.A.; Blomquist, G.; Andersson, B.; Carlson, R. Identification of volatile metabolites from five fungal species cultivated on two media. Appl. Environ. Microbiol. 1995, 61, 2911–2918. 42. Schmidt, R.; Cordovez, V.; De Boer, W.; Raaijmakers, J.; Garbeva, P. Volatile affairs in microbial interactions. ISME J. 2015, 9, 2329–2335, doi:10.1038/ismej.2015.42. 43. Korpi, A.; Järnberg, J.; Pasanen, A.L. Microbial volatile organic compounds. Crit. Rev. Toxicol. 2009, 39, 139–193, doi:10.1080/10408440802291497. 44. Piechulla, B.; Degenhardt, J. The emerging importance of microbial volatile organic compounds. Plant, Cell Environ. 2014, 37, 811–812, doi:10.1111/pce.12254. 45. Agrawal, R. Flavors and aromas. In Fungal Biotechnology In Agricultural, Food and Environmental Applications; Arora, D.., Ed.; Marcel Dekker Inc.: New York, NY, 2004; pp. 281–289 ISBN 3527403736. 46. Schalchli, H.; Tortella, G.R.; Rubilar, O.; Parra, L.; Hormazabal, E.; Quiroz, A. Fungal volatiles: An environmentally friendly tool to control pathogenic microorganisms in p l a n t s . C r i t . R e v . B i o t e c h n o l . 2016, 36, 1 4 4 – 1 5 2 , d o i : 10.3109/07388551.2014.946466. 47. Thorn, R.M.S.; Reynolds, D.M.; Greenman, J. Multivariate analysis of bacterial volatile compound profiles for discrimination between selected species and strains in vitro. J. Microbiol. Methods 2011, 84, 258–264, doi:10.1016/j.mimet. 2010.12.001. 48. Thalavitiya Acharige, M.J.; Koshy, S.S.; Koo, S. The use of microbial metabolites for the diagnosis of infectious diseases. In Advanced Techniques in Diagnostic Mi- crobiology; Springer International Publishing: Cham, 2018; pp. 261–272 ISBN 9783319339009. 49. Abraham, W.-R.; Ernst, L.; Witte, L.; Hanssen, H.-P.; Sprecher, E. New trans-fused

49 africanols from Leptographium lundbergii. Tetrahedron 1986, 42, 4475–4480, doi: 10.1016/S0040-4020(01)87288-4. 50. Gromski, P.S.; Muhamadali, H.; Ellis, D.I.; Xu, Y.; Correa, E.; Turner, M.L.; Goodacre, R. A tutorial review : Metabolomics and partial least squares-discriminant analysis – a marriage of convenience or a shotgun wedding. Anal. Chim. Acta 2015, 879, 10– 23, doi:10.1016/j.aca.2015.02.012. 51. Eliasson, M.; Rännar, S.; Trygg, J. From data processing to multivariate validation-- essential steps in extracting interpretable information from metabolomics data. Curr. Pharm. Biotechnol. 2011, 12, 996–1004, doi:10.2174/138920111795909041. 52. Putri, S.P.; Fukusaki, E. Mass Spectrometry-Based Metabolomics. A Practical Guide; Boca Raton, 2014; ISBN 9781482223774. 53. Paliy, O.; Shankar, V. Application of multivariate statistical techniques in microbial ecology. Mol. Ecol. 2016, 25, 1032–1057, doi:10.1111/mec.13536. 54. Editors, B.J.A.M.; Daniels, M. Statistical Analysis of Proteomics, Metabolomics, and Lipidomics Data Using Mass Spectrometry; 2017; ISBN 978-3-319-45807-6. 55. Goodacre, R.; Broadhurst, D.; Smilde, A.K.; Kristal, B.S.; Baker, J.D.; Beger, R.; Bessant, C.; Connor, S.; Capuani, G.; Craig, A.; Ebbels, T.; Kell, D.B.; Manetti, C.; Newton, J.; Paternostro, G.; Somorjai, R.; Sjöström, M.; Trygg, J.; Wulfert, F. Pro- posed minimum reporting standards for data analysis in metabolomics. Metabolomics 2007, 3, 231–241, doi:10.1007/s11306-007-0081-3. 56. Saccenti, E.; Hoefsloot, H.C.J.; Smilde, A.K.; Westerhuis, J.A.; Hendriks, M.M.W.B. Reﬂections on univariate and multivariate analysis of metabolomics data. Metabolomics 2014, 10, 361–374, doi:10.1007/s11306-013-0598-6. 57. Chopra, T.; Gokhale, R.S. Chapter 12 Polyketide versatility in the biosynthesis of complex mycobacterial cell wall lipids; 1st ed.; Elsevier Inc., 2009; Vol. 459; ISBN 9780123745910. 58. Legendre, P.; Gallagher, E.D. Ecologically meaningful transformations for ordination of species data. Oecologia 2001, 129, 271–280, doi:10.1007/s004420100716. 59. van den Berg, R.A.; Hoefsloot, H.C.J.; Westerhuis, J.A.; Smilde, A.K.; van der Werf, M.J. Centering, scaling, and transformations: improving the biological information content of metabolomics data. BMC Genomics 2006, 7, 142, doi: 10.1186/1471-2164-7-142. 60. Triba, M.N.; Le Moyec, L.; Amathieu, R.; Goossens, C.; Bouchemal, N.; Nahon, P.; Rutledge, D.N.; Savarin, P. PLS/OPLS models in metabolomics: the impact of per- mutation of dataset rows on the K-fold cross-validation quality parameters. Mol. Biosyst. 2015, 11, 13–19, doi:10.1039/C4MB00414K. 61. Langford, M.L.; Atkin, A.L.; Nickerson, K.W. Cellular interactions of farnesol, a quorum-sensing molecule produced by Candida albicans. Future Microbiol. 2009, 4, 1353–1362, doi:10.2217/fmb.09.98. 62. Kramer, R.; Abraham, W.R. Volatile sesquiterpenes from fungi: What are they good for? Phytochem. Rev. 2012, 11, 15–37, doi:10.1007/s11101-011-9216-2. 63. Lin, H.C.; Chooi, Y.H.; Dhingra, S.; Xu, W.; Calvo, A.M.; Tang, Y. The fumagillin biosynthetic gene cluster in Aspergillus fumigatus encodes a cryptic terpene cy- clase involved in the formation of β-trans-bergamotene. J. Am. Chem. Soc. 2013, 135, 4616–4619, doi:10.1021/ja312503y. 64. Ashbee, H.R. Update on the genus Malassezia. Med. Mycol. 2007, 45, 287–303, doi:10.1080/13693780701191373. 65. Weimer, B.; Seefeldt, K.; Dias, B. Sulfur metabolism in bacteria associated with cheese. Antonie van Leeuwenhoek, Int. J. Gen. Mol. Microbiol. 1999, 76, 247–261, doi:10.1023/A:1002050625344.

50 66. Schöller, C.E.G.; Gürtler, H.; Pedersen, R.; Molin, S.; Wilkins, K. Volatile metabolites from Actinomycetes. J. Agric. Food Chem. 2002, 50, 2615–2621, doi:10.1021/ jf0116754. 67. Buzzini, P.; Romano, S.; Turchetti, B.; Vaughan, A.; Pagnoni, U.M.; Davoli, P. Pro- duction of volatile organic sulfur compounds (VOSCs) by basidiomycetous yeasts. FEMS Yeast Res. 2005, 5, 379–385, doi:10.1016/j.femsyr.2004.10.011. 68. Filipiak, W.; Sponring, A.; Filipiak, A.; Baur, M.; Ager, C.; Wiesenhofer, H.; Margesin, R.; Nagl, M.; Troppmair, J.; Amann, A. Volatile Organic Compounds (VOCs) released by pathogenic microorganisms in vitro: Potential breath biomarkers for early-stage diagnosis of disease. In Volatile Biomarkers; Elsevier, 2013; pp. 463–512 ISBN 9780444626134. 69. Sprecher E. Influence of strain specificity and culture conditions on terpene production by fungi. Planta medica. 1982, 44, 41–43, doi:10.1055/s-2007-971398. 70. Bjurman, J.; Kristensson, J. Volatile production by Aspergillus versicolor as a possible cause of odor in houses affected by fungi. Mycopathologia 1992, 118, 173– 178, doi:10.1007/BF00437151. 71. Gethins, L.; Guneser, O.; Demirkol, A.; Rea, M.C.; Stanton, C.; Ross, R.P.; Karagul Yuceer, Y.; Morrissey, J.P. Influence of Carbon and Nitrogen source on production of volatile fragrance and flavour metabolites by the yeast Kluyveromyces marxianus. Yeast 2015, 32, 67–76, doi:10.1002/yea.3047. 72. Frisvad, J.C.; Andersen, B.; Thrane, U. The use of secondary metabolite profiling in chemotaxonomy of filamentous fungi. Mycol. Res. 2008, 112, 231–240, doi: 10.1016/j.mycres.2007.08.018. 73. Bos, L.D.J.; Sterk, P.J.; Schultz, M.J. Volatile Metabolites of Pathogens: A System- atic Review. PLoS Pathog. 2013, 9, 1–8, doi:10.1371/journal.ppat.1003311. 74. Boots, A.W.; Smolinska, A.; Van Berkel, J.J.B.N.; Fijten, R.R.R.; Stobberingh, E.E.; Boumans, M.L.L.; Moonen, E.J.; Wouters, E.F.M.; Dallinga, J.W.; Van Schooten, F.J. Identification of microorganisms based on headspace analysis of volatile organic compounds by gas chromatography-mass spectrometry. J. Breath Res. 2014, 8, doi:10.1088/1752-7155/8/2/027106. 75. Rees, C.A.; Smolinska, A.; Hill, J.E. The volatile metabolome of Klebsiella pneumo- niae i n h u m a n b l o o d . J . B r e a t h R e s . 2016, 10, 2 7 1 0 1 , d o i : 10.1088/1752-7155/10/2/027101. 76. Tabares, M.; Ortiz, M.; Gonzalez, M.; Carazzone, C.; Vives Florez, M.J.; Molina, J. Behavioral responses of Rhodnius prolixus to volatile organic compounds released in vitro by bacteria isolated from human facial skin. PLoS Negl. Trop. Dis. 2018, 12, 1–16, doi:10.1371/journal.pntd.0006423. 77. Kladsomboon, S.; Thippakorn, C.; Seesaard, T. Development of organic-inorganic hybrid optical gas sensors for the non-invasive monitoring of pathogenic bacteria. Sensors 2018, 18, 3189, doi:10.3390/s18103189. 78. Sun, D.; She, J.; Gower, J.; Stokes, C.; Windham, G.; Baird, R.; Mlsna, T. Effects of growth parameters on the analysis of Aspergillus flavus volatile metabolites. Sepa- rations 2016, 3, 13, doi:10.3390/separations3020013. 79. Casu, F.; Pinu, F.R.; Fedrizzi, B.; Greenwood, D.R.; Villas-Boas, S.G. The effect of linoleic acid on the Sauvignon blanc fermentation by different wine yeast strains. FEMS Yeast Res. 2016, 16, fow050, doi:10.1093/femsyr/fow050. 80. Ro, B.I.; Dawson, T.L. The role of sebaceous gland activity and scalp microfloral metabolism in the etiology of seborrheic dermatitis and dandruff. J. Investig. Der- matology Symp. Proc. 2005, 10, 194–197, doi:10.1111/j.1087-0024.2005.10104.x. 81. Dennis, C.; Webster, J. Antagonistic properties of species-groups of Trichoderma.

51 II. Production of volatile antibiotics. Trans. Br. Mycol. Soc. 1971, 57, 41-IN4, doi: 10.1016/S0007-1536(71)80078-5.

12. APPENDIX

Figure 11. Experimental mass spectra and comparison with the tentative annotated compound or the compound with the major similarity in their fragmentation patterns in EI.