Methods Snifﬁng fungi – phenotyping of volatile chemical diversity in Trichoderma species

Volatile organic compounds (VOCs) play vital roles in the interaction of fungi with plants and other organisms. A systematic study of the global fungal VOC proﬁles is still lacking, though it is a prerequisite for elucidating the mechanisms of VOC-mediated interactions. Here we present a versatile system enabling a high-throughput screening of fungal VOCs under controlled temperature. In a proof-of-principle experiment, we characterized the volatile metabolic ﬁngerprints of four Trichoderma spp. over a 48 h growth period. The developed platform allows automated and fast detection of VOCs from up to 14 simul-taneously growing fungal cultures in real time. The comprehensive analysis of fungal odors is achieved by employing proton transfer reaction-time of ﬂight-MS and GC-MS. The data-min-ing strategy based on multivariate data analysis and machine learning allows the volatile metabolic ﬁngerprints to be uncovered. Our data revealed dynamic, development-dependent and extremely species-speciﬁc VOC proﬁles from the biocontrol genus Trichoderma . The two mass spectrometric approaches were highly complementary to each other, together revealing a novel, dynamic view to the fungal VOC release. (cid:1) This analytical system could be used for VOC-based chemotyping of diverse small organisms, or more generally, for any in vivo and in vitro real-time headspace analysis.


Introduction
Fungi are known to emit a wide range of volatile organic compounds (VOCs) with high chemical diversity, including alcohols, benzenoids, aldehydes, alkenes, acids, esters, terpenoids and ketones (Morath et al., 2012;Li et al., 2016;Lemfack et al., 2017;Misztal et al., 2018). VOCs are characterized by low molecular weight, high vapor pressure and polarity (Lee et al., 2019). VOCs evaporate easily at room temperature and are distributed into the surrounding air, enabling them to act as signal substances in intraand interorganismic communication (Insam & Seewald, 2010;Penuelas et al., 2014;Kanchiswamy et al., 2015). More than 300 fungal VOCs (fVOCs) have been identified so far  and the number of identified compounds, as well as the number of microbial species analyzed for their VOCs, continues to increase. This can be well deduced from the increasing number of entries in the database of microbial volatiles (mVOC database) (Lemfack et al., 2017). In the last decade, progress has also been made in understanding the ecological functions of fVOCs and how they might mediate interorganismic communication (Piechulla et al., 2017;Li et al., 2018). fVOCs may play crucial roles in the formation and regulation of symbiotic associations and in the distribution of saprophytic, mycorrhizal and pathogenic organisms in the ecosystem Kanchiswamy et al., 2015;Mhlongo et al., 2018). An ecological function has only been described for a few individual fVOCs and, moreover, knowing that the released fVOCs depend strongly on the developmental stage and environmental conditions (Romoli et al., 2014;Schmidt et al., 2015;Weikl et al., 2016), so far, the scientists have probably only revealed the tip of the iceberg. A systematic research on the global VOC profile of fungi is still lacking, although it is a prerequisite for elucidating the mechanisms of fVOC-mediated organismic interactions.
Over the past 10-20 yr the methods and analytical techniques for measuring VOCs have evolved considerably (Zhang & Li, 2010;Misztal et al., 2018). This development has had an impact on the analysis of volatile compounds from microbes and fungi, which have complex scent profiles and, given the high number existing on Earth, bear the potential to identify new substances (Zhang & Li, 2010;Hung et al., 2015;Lemfack et al., 2017). So far, technologies such as GC-MS have dominated the analysis of fVOCs owing to the reliable and cost-effective separation, identification and quantification of substances (Matysik et al., 2009;Morath et al., 2012;Siddiquee et al., 2012). Recently GC-MS analyses were often combined with a passive VOC collection (e.g. stir bar sorptive extraction, SBSE) (Wihlborg et al., 2008;Bicchi et al., 2009;Zhang & Li, 2010). The required pre-enrichment of VOCs before the GC-MS analysis (and the GC-MS run itself; Bicchi et al., 2009) can, however, be time-consuming and, moreover, do not enable real-time detection of volatile compounds. Thus, so far fVOC analyses have mostly been conducted at low time resolution as a result of the long collection time necessary for proper detection (e.g. 6 h or 16 h) Guo et al., 2019). Further bias might rise from common adsorption materials that preferably trap medium-to high-molecularweight compounds and from the use of thermodesorption units that favor medium-to-high thermostabile analytes (Kataoka et al., 2000;Marcillo et al., 2017). The separation ability of a GC column and sensitivity to specific compounds depends, moreover, on the specific physical properties of the stationary phase in the column. Employing polar and nonpolar stationary phases would theoretically allow detection of a wider range of compounds but also request longer sampling time (Mondello et al., 2005).
As the interest in fungal and other microbial VOCs in interorganismic interactions has increased considerably (Penuelas et al., 2014;Werner et al., 2016), the demand of a new generation of analytical technology enabling fast, real-time detection of microbial VOCs has become imperative. A promising tool to achieve the new goal is employment of proton transfer reaction-MS (PTR-MS) in combination with a time-of-flight (ToF) analyzer (Lindinger & Jordan, 1998;Graus et al., 2010). The PTR-ToF-MS allows quick VOC measurements with high sensitivity and high mass resolution (c. 4000-5000 m Dm -1 ) in a typical mass range of 15 to c. 350 amu (Graus et al., 2010). The high mass resolution makes it possible to separate isobaric compounds (e.g. pure hydrocarbons) from isobaric oxygenated molecules . The limit of detection (LOD) of a PTR-ToF-MS is currently ≤ a few parts per trillion volume compared to parts per billion of GC-MS (Thet & Woo, 2019). However, the PTR-ToF-MS is not able to separate isomers and therefore a proper compound identification may become tedious or even impossible for heavier molecules. For example, sesquiterpenes are a class of terpenoids that, having many chemical structures but usually the same chemical formula (Chadwick et al., 2013), are detected as a single mass by PTR-ToF-MS. Our setup therefore combines the strengths of both the PTR-ToF-MS, allowing the fungal emissions to be monitored in a semi-online fashion, and the GC-MS, also allowing the compounds with similar masses to be identified.
Proton transfer reaction-ToF-MS has previously been used to measure volatile emissions from plants (Farneti et al., 2015;Li et al., 2019), soils (Veres et al., 2014;Mancuso et al., 2015), yeast (Khomenko et al., 2017) and also from bacteria and fungi (Adams et al., 2017;Infantino et al., 2017;Misztal et al., 2018). These studies have employed online MS, which, however, benefits from further GC complementation of the chemical formulas with structural isomer information. To analyze complete volatilome and temporal fluctuation in emission patterns, a controlled high-throughput analysis platform is necessary. Such a platform is a first desirable step to determining development and environment-dependent fVOC profiles. It is also the first step to deciphering the VOC-based intra-and interspecies-specific chemical diversity. The large and complex fVOC datasets generated by such a platform are, however, challenging for the classic statistic approaches suitable for simpler datasets (Bzdok et al., 2018;Xu & Jackson, 2019). Instead, a machine learning approach, which has huge potential to analyze the increasingly complex biological -omics data (Camacho et al., 2018), is instrumental to uncovering the sophisticated fVOCsfactors and patterns.
In this study, we present an automated, online VOC monitoring cuvette system that allows the in vivo analysis of time series of fVOC emissions to dynamically capture changes in the fungal odor profiles.
In combination with PTR-ToF-MS, SBSE-GC-MS and a datamining approach (combining statistics and supervised and unsupervised machine learning), we investigated the chemical diversity of four Trichoderma (Hypocreales, Ascomycota) species, including T. harzianum (WM24a1), T. hamatum (QL15d1), T. reesei (QM6a) and T. velutinum (GL15b1). Trichoderma spp. are able to establish in the rhizosphere of host plants, leading to beneficial effects (Guzm an-Guzm an et al., 2018). The growth-promoting effects are based on mobilization of nutrients (Harman, 2011), induction of plant systemic resistance (Shoresh et al., 2010;Estrada-Rivera et al., 2019), and a mycotrophic lifestyle towards plant pathogens Kubicek et al., 2011). Recent studies show very species-specific volatile profiles for Trichoderma (Siddiquee et al., 2012;Lee et al., 2016;Guo et al., 2019), whose patterns can drastically change in interaction with other organisms (Guo et al., 2019), suggesting important ecological functions for fVOCs of Trichoderma. In the present study, the complement of PTR-ToF-MS and GC-MS allowed us to obtain comprehensive measurements of the odor profiles of the four Trichoderma species. The applied data-mining strategy formed a crucial part of our fVOC measurement system, facilitating the elucidation of fVOCs and hence paving the way to unearth the fVOC-mediated organismic communication. The demonstrated automated system proved to be a very time-effective strategy for characterizing the intraspecific chemical diversity of the Trichoderma volatiles. It allowed access to biologically driven changes of an individual volatile emission in real time. Within the present work, by investigating the fast-growing Trichoderma species, we demonstrate a proof-of-principle of a strategy to decipher fungalas well as other microbialvolatilomes in solitary cultures and in interactions with each other.

Fungal species and cultivation
Four Trichoderma species (Trichoderma harzianum WM24a1, Trichoderma hamatum QL15d1, Trichoderma reesei QM6a and Trichoderma velutinum GL1561, all kindly provided by Monica Schmoll, Austrian Institute of Technology, Tulln, Austria) were used for the fVOC analysis. The fungi were cultivated in glass cuvettes (7 cm diameter and 6.6 cm depth, total volume c. 254 ml) containing modified Melin-Norkrans synthetic medium (40 ml per cuvette; described previously by A. M€ uller et al., 2013) and were grown in a chamber at 23°C and continuous darkness (Guo et al., 2019). VOC analysis was started at the beginning of the exponential hyphal growth stage, because most secondary metabolites of fungi are produced in this period, which is after completion of its initial growth and right before the next developmental stage, represented by the formation of spores (Calvo et al., 2002). Before the study, growth curves of the four Trichoderma species were used to determine the exponential growth stage (Supporting Information Fig. S1). Pictures of the Trichoderma cultures were taken at the beginning and at the end of the PTR-ToF-MS measurements (Fig. S2).

Measurement system
The outline of the measurement platform is illustrated in Fig. 1. The core element of the platform was the cuvette system integrated in a growth incubator (Model 3100 Series, Thermo Scientific, Marietta, OH, USA), which allows cultivation of fungi under well controlled temperature conditions (23°C in the present study). The system consisted of a series of 14 glass cuvettes (the same ones in which the fungi were cultivated; 7 cm diameter, 6.6 cm depth) with gas-tight tin plate lids covered with a Teflon sheet (0.12 mm thick) at the bottom to minimize fVOC deposition. The lids each contained two stainless steel bulkhead unions for connection of the air supply and the sample line. A gas calibration unit (GCU; Ionicon Analytik GmbH, Innsbruck, Austria) provided VOC-free air at the desired water moisture content under ambient CO 2 concentration. This clean and humidified air (70% relative humidity at 22°C) is used to flush the cuvettes and tubes during PTR-ToF-MS analysis. The cuvette inflow was diverted by a Teflon deflector in order to avoid a direct air stream onto the fungi, which had been shown to affect fungal growth. The lines from the cuvettes to the inlet of the PTR-TOF-MS were heated to 40°C to avoid condensation and limit VOC adsorption onto the sample lines. A multiport valve (C25Z-31814D; VICI-Valco, Houston, TX, USA) served as a central control for the sequential sampling of air from the individual cuvettes. The switching of the multiport valve between the different cuvettes and the data acquisition process were controlled by PTR-MANAGER, a piece of software shipped with the PTR-ToF-MS. The online PTR-ToF-MS data were recorded using TOFDAQ software (Tofwerk AG, Thun, Switzerland). The data were stored in a hierarchical data file format (HDF5) which includes meta information on data acquisition parameters.
All tubings assembled in the cuvette system consist of polytetrafluoroethylene (PTFE) or PFA Teflon, or polyether ether ketone to minimize deposition and reactions of VOCs on the surfaces (Jud et al., 2018). In order to prevent fungal spores from entering the reaction chamber of the PTR-ToF-MS, a BOLA single-stage PFA flow filter (Bohlender GmbH, Gr€ unsfeld, Germany) containing a PTFE filtering membrane (5 lm pore size) was installed between the multiport valve and the PTR-ToF-MS. The cuvette system was operated in a slight overpressure mode by connecting a high-efficiency particulate air (HEPA) filter at the outlet of the exhaust line. This helped to prevent ambient air leaking into the cuvettes.

fVOC measurements by PTR-ToF-MS
The dynamic emissions of fVOCs emitted by the Trichoderma species were measured sequentially using a PTR-ToF-MS (Ionicon Analytik GmbH, Innsbruck, Austria). In a PTR-ToF-MS, VOCs with a proton affinity higher than that of water (691 kJ mol -1 ; Hunter & Lias, 1998) are ionized via proton transfer from primary H 3 O + ions according to the following reaction scheme: The measurement system was run at stable, well-defined conditions. The inlet flow rate was set at 450 standard cm 3 min À1 . Fig. 1 Schematic of the multiple cuvette system. Fourteen glass cuvettes (labeled 1-14) were integrated in an incubator. The transfer lines (red) from the cuvette outlets and of the multiport valve to the proton transfer reaction-time-of-flight-MS (PTR-ToF-MS) analyzer were heated up to 40°C with an encapsulated heating line to increase the vapour pressure of the substances and to keep deposition on the line surface as low as possible. The gas calibration unit (GCU) was used to generate ultraclean carrier gas stream for conduction of headspace fungal volatile organic compounds (fVOCs). The multiport valve serves as a central control for the sequential sampling of air from the individual cuvettes. Cuvette 1 (kept empty) and cuvette 2 (sole media) were used to monitor and correct background emission.

New Phytologist
The PTR-ToF-MS ion drift tube was operated at 60°C, 540 V drift voltage and 2.3 mbar drift pressure, corresponding to an E/ N of 120 Td (E being the electric field strength and N the gas number density; 1 Td = 10 À17 V cm 2 ).
The cuvette inlet flow (F in ) and the cuvette volume V define two important parameters of the cuvette system: the characteristic time constant s and its reverse, the exchange rate 1/s of the cuvette system: With a cuvette inlet flow of 450 ml min À1 and a cuvette volume of 254 ml, it takes 2.82 min (5s) until the air inside one of the cuvettes was exchanged to more than 99% (Graus, 2005). The measurement procedure consisted in continuous switching between the 14 cuvettes. Cuvette no. 1 was completely empty and served as reference. Cuvette no. 2 was containing growth medium only and was used to detract the signal background from the cuvette nos. 3 to 14 containing fungi. Every time before switching from one cuvette to the next, the system was switched for 10 s to the completely empty background cuvette in order to flush the sample lines with zero air. This helps to avoid carrying over VOCs from the preceding cuvette into the sample air of the following cuvette. From each cuvette we sampled for 5 min, during which air was drawn into the PTR-ToF-MS. When not been sampled, the air in the cuvettes was resting and fVOCs accumulated for c. 70 min.
The PTR-ToF-MS raw data were analyzed using the routines described in M. M€ uller et al. (2013) and Jud et al. (2018). The calculated signals in counts per second (cps) were normalized to 10 6 reagent ion counts (sum of the signals of H 3 O + and its cluster, H 3 O + ·H 2 O, divided by 10 6 ) to account for differences in the absolute humidity in the different cuvettes. Eventually, we got the signals in normalized cps (ncps).
After the raw data of the PTR-ToF-MS measurements were evaluated and separated cuvette-wise, the TOF DATA PLOTTER program (see https://sites.google.com/site/ptrtof/file-cabinet) was used to visualize the data and calculate the fungus-specific volatile signals. From a list of about 500 peaks present in the mass spectra we narrowed down the number of mass over charge ratios (m/z, z = 1) for further analysis by going manually through all spectra (using DATA PLOTTER) and searching for those showing differences between the background and the fungi cuvettes. Additionally, all m/z attributable to isotope signals were removed. This way we ended up with a list of 56 mass features/compounds.
The m/z signals were first background-corrected by subtracting the signal of the cuvette containing only growth medium. To this end, cubic splines were fitted through the mean signals measured from the growth medium cuvette throughout the whole measurement and the interpolated signals were then subtracted from all fungi cuvette signals measured in between.
Afterwards, the time traces of the signals were separated cuvette-wise and the cumulative signals of all m/z of interest in each cuvette sample interval were calculated. Except for analyzing the time series, the data (cumulative signal of 5 min) points over the whole measurement period were averaged in each cuvette. Eventually, the data were normalized to the respective area of fungal mycelium and the c. 70 min accumulation time for further analysis. Lacking authentic calibration standards of many of the detected compounds (identified with the GC-MS) and signal interferences of isomers we refrained from further converting the cumulative ncps in actual amounts (e.g. in nmol).
The nontargeted feature of PTR-ToF-MS measurements challenges the precise identification of the detected mass, particular for the separation of the chemical classes 'ketone' and 'aldehyde'. The lack of commercially available standards and PTR-ToF-MSbased fVOC literature, moreover, hindered the annotations of the detected masses. Nevertheless the annotation was greatly helped by the previously reported VOCs on mainly soil matrix as well as PTR-ToF-MS-based fungal and plant VOC records and the mVOC database (Table S1; Lemfack et al., 2017). To validate the annotations, we manually checked the degree of match and alignment with the isotopic patterns. Further, to improve the annotation rate, we determined the correlation (R 2 > 0.8) for the masses that are most likely a fragment of a (related) compound using the ToF DATA PLOTTER programme. Taking the abovementioned strategies together, we carefully annotated the detected masses as tentatively assigned compounds.

fVOC measurements with GC-MS
The VOCs from Trichoderma species were collected with the SBSE technique (polydimethylsiloxane (PDMS) twisters; Gerstel GmbH, M€ ulheim an der Ruhr, Germany) and analyzed as described previously (Guo et al., 2019). After the analysis with PTR-ToF-MS, twisters were fixed onto the inner side of the lid (with a magnet mounted outside the lid) and allowed to trap fVOCs over a period of 16 h. The fVOCs were analyzed by thermal desorption (TDU, Gerstel) GC-MS (GC type 7890A, MS type 5975C; Agilent Technologies, Palo Alto, CA, USA) using a 5% phenyl 95% dimethyl arylene siloxane capillary column (70 m 9 250 lm 9 25 lm DB-5MS + 10 m DG; Agilent Technologies). The TDU-GC-MS was run following the well-established procedures (Ghirardo et al., 2012. The general GC-MS parameters are given in Weikl et al. (2016), further modified by Guo et al. (2019). Peak annotation and quantification followed Guo et al. (2019) and Kreuzwieser et al. (2014). GC-MS data were normalized to the area of fungal mycelium and VOC collection time. For calculating the LOD (Table S2), three sigma (r) of the background signals were used (Shrivastava & Gupta, 2011).

Statistics
All data analysis and visualization were performed in R v.3.6.1 (R Core Team, 2018). Time series analysis were performed on the PTR-ToF-MS dataset with the 'dtwclust' package (Sarda-Espinosa et al., 2019) with 'gak' distance and 'median' centroid. Principal component analysis (PCA) was used to analyze the volatile emission patterns of the four species using merged PTR-ToF-MS and GC-MS dataset. Data were standardized to have Ó 2020 The Authors New Phytologist Ó 2020 New Phytologist Trust New Phytologist (2020) 227: 244-259 www.newphytologist.com means of 0 and variance of 1 to remove the effects of different units and scales (Maynard et al., 2019). PCA was performed using the 'prcomp' function with GGFORTIFY package (Tang et al., 2016). The variables contribution was extracted with 'fviz_contrib' function in the FACTOEXTRA package (Kassambara & Mundt, 2017). Data were z-score-standardized internally before time series analysis to obtain static features (Tang et al., 2016;Sard a-Espinosa, 2017). Random forest (RF) analysis (Breiman, 2001) was performed to elucidate the important compounds discriminating the four Trichoderma species. The number of variables that randomly sampled as candidates at each split (mtry) and the number of trees to grow (ntree) were tuned (grid search) using the caret package (Kuhn, 2008) to obtain the optimal predictive ability, stability and accuracy. RF analysis was performed separately on the PTR-ToF-MS and GC-MS datasets, to avoid the possible influence of scale difference and sparsity difference (Karlsson & Bostr€ om, 2014).

General features of combining PTR-ToF-MS, SBSE-GC-MS and data mining
The new developed VOC analysis system includes several working steps, the core part being: the online VOC analysis by PTR-ToF-MS; offline analysis employing SBSE-GC-MS; and the data-mining strategy (Fig. S3). This combination together with the automated cuvette system was built in order to allow measurement and analysis of microbial volatiles, in particular, in a novel, comprehensive and dynamic manner. PTR-ToF-MS measurements covered the detection of especially small, polar compounds, whereas SBSE-GC-MS allowed for the detection of rather nonpolar, less volatile compounds and, moreover, facilitated the identification of compounds with equal masses (Tables S1, S3). The real-time PTR-ToF-MS analysis allowed detecting development-dependent changes in the fVOCs emission from 12 fungal (and two control) samples sequentially (Figs 1, 2). When not been sampled, the fVOCs were allowed to accumulate for c. 70 min, which decreased the LOD. When a cuvette was sampled, the dilution of the fungal headspace with VOC-free air resulted in a continuous drop of the fVOC mixing ratios (Fig. 2c), which assisted in the separation of very volatile, nonpolar compounds from less volatile, polar and often oxygenated compounds. The signals of the latter were decaying more slowly during the sample period as a result of the greater 'stickiness' of these compounds onto the inner surface of the cuvette and the sample lines.
Shorter collection time allowed fast screening of a higher number of samples, whereas choosing fewer samples increased the time resolution.
The applied data-mining strategy facilitated the elucidation of the complex data reached by combining the two MS systems and the automated 14-cuvette platform. Within the present work we applied this new system combination to explore the fVOC patterns of four plant beneficial Trichoderma species.

Sesquiterpenes emission differ between Trichoderma species
The sesquiterpene detection capability exemplifies well the main features of the two applied MS systems. Our analysis with PTR-ToF-MS showed a strong sesquiterpene signal (m/z 205.195) in cultures of T. hamatum (Fig. 2a) and T. reesei (Fig. 2b), while it was very low in those of T. harzianum and T. velutinum (Fig. S4). We further detected a transient change of the sesquiterpene signal from T. hamatum and T. reesei over a measurement period of 2 d (Fig. 2). Zooming into a single measurement cycle, the differences in the transient course of the m/z 205.195 from the individual cuvettes can be observed (Fig. 2c). When the measurements started in an individual cuvette, a high signal was measured resulting from the preaccumulation. Owing to the gas exchange in the fungal headspace the signal then decreased, approximating a steady state by the end of the 5 min sampling time. The cumulative signal over the measurement period, therefore, corresponds to the overall amount of sesquiterpenes synthesized and released by the mycelia within a time period of c. 70 min.
Sesquiterpenes are a large class of isomeric terpenes having the same sum formula (C 15 H 24 ), but differing in their chemical structure. Therefore, all nonfragmented sesquiterpenes are recorded in the PTR-ToF-MS as a unique ion signal bare of any structural information (Chadwick et al., 2013;Fig. 2). Different fragmentation patterns of distinct sesquiterpenes could, in principle, aid in identifying some of them, but only as long as merely single sesquiterpenes are measured (Kim et al., 2009) and no complex mixtures as shown herein. In order to identify the individual sesquiterpene isomers, we therefore performed a complementary VOC analysis by GC-MS ( Fig. S3; Table 1). This combination allowed us to describe the dynamics of sesquiterpene emissions during the mycelial growth (by PTR-ToF-MS) and moreover to decipher their chemical structure (by GC-MS). As shown in Fig. 3, T. hamatum and T. reesei emitted a high number of sesquiterpenes, while T. harzianum and T. velutinum emitted only a few. The GC-MS measurements also showed that the emission patterns differed in different species: The sesquiterpene no. 1 (b-elemene), sesquiterpene no. 19 (a-selinene) and sesquiterpene no. 24 (trans-c-bisabolene) together (63.01%) dominated the profiles of T. hamatum (Fig. 3f); sesquiterpene no. 4 (a-cedrene), sesquiterpene no. 8 (b-cedrene) and sesquiterpene no. 6 (b-curcumene) together (52.69%) dominated the profile of T. reesei (Fig. 3g), while the profile of T. harzianum and T. velutinum had no distinct major component (Fig. 3e,h).

Species-specific temporal changes of fVOC emissions
Proton transfer reaction-ToF-MS is a powerful online monitoring technique for studying volatile emissions, but had not yet, to our knowledge, been systematically applied to elucidate development-dependent changes in VOC profiles of different fungi. Our measurements demonstrated the potential of this analytical method for the detection of the dynamic VOC emission patterns. Our online measurements by PTR-ToF-MS showed four  Fig. 4a). The temporal emission profile of cluster 1 was typical for 65.4% of T. harzianum-emitted fVOCs, while for T. hamatum about half (46.9%) of the emitted compounds fitted into the emission profile of cluster 2, as did 28.4% of T. harzianum volatiles. The emission patterns of the two other Trichoderma species, T. reesei and T. velutinum, however, could be described by clusters 3 and 4 (Fig. 4b). However, the emission profile of a single compound could differ in different species: for example, the emission of a compound with mass m/z 205.195 (total sesquiterpenes) showed different temporal patterns in different Trichoderma species (Fig. 4c).
Apparently, the emission patterns were not correlated with the moisture content inside the cuvettes. The relative humidity of the inlet air was set to 70% at 22°C at the beginning and was not further regulated throughout the experiment (and consequently through part of the developmental stages of the fungi samples). Lacking in situ air moisture sensors, we used the signal of the water cluster (H 2 O-H 3 O + ) isotope detected at m/z 39.033 with the PTR-ToF-MS as proxy for the relative air humidity and therefore for the overall moisture conditions inside the individual cuvettes. This signal was stable throughout the experiments in all cuvettes containing growth medium (i.e. cuvettes 2-14; cuvette 1 was completely empty (see Material and Methods section)). Owing to the sequential measurement of the cuvettes, resulting in cuvette flushing periods of 5 min every c. 70 min, the humidity of the growth medium seemed to be sufficient to maintain stable moisture conditions over the 2 d of experiment. The time trace of the water cluster (H 2 O-H 3 O + , m/z 39.033) therefore fits into the temporal profile of cluster 4 in Fig. 4(a).

Chemical diversity of VOCs emitted by Trichoderma species
To illustrate the chemical diversity of the individual Trichoderma species, we have clustered the volatile compounds into 11 chemical classes (monoterpenes, sesquiterpenes, sesquiterpene-alcohols, acyclic alkenes, alcohols, alkanes, aldehydes, ketones, acids, benzenoids and esters). Employing both, PTR-ToF-MS and GC-MS, a pronounced chemical diversity could be detected in the fVOC emissions (Fig. 5). However, the chemical class membership differed depending on the MS method applied. Mainly alcohol, aldehyde and ketone compounds were detected from the four Trichoderma species (accounting for 86.6% (T. harzianum), 96.8% (T. hamatum), 94.6% (T. reesei) and 94.0% (T. velutinum) of all the compounds; Fig. 5a) by PTR-ToF-MS, whereas the compounds detected by GC-MS were more diverse (Fig. 5b). A large part of the T. hamatum and T. reesei released compounds detected by GC-MS were sesquiterpenes, whereas those were only minor part of the compounds detected by PTR-ToF-MS ( Fig. 5; Tables S1, S3). Also, benzenoids were detected by GC-MS but almost not at all by PTR-ToF-MS (Fig. 5). Actually, almost none of the compounds detected by PTR-ToF-MS were found within the GC-MS chromatograms of the present study. These compounds detected by PTR-ToF-MS were also only very rarely reported in previous studies that explored fungal volatiles employing GC-MS (Tables S1, S3). The PTR-ToF-MS and GC-MS have different capabilities to detect various compounds and compound classes (Table 1). Our data show that measurement with PTR-ToF-MS was ideally suited to detect the frequently occurring compounds (76.8% of the detected compounds were commonly shared by the four Trichoderma species), while GC-MS was more likely to detect species-specific compounds (4.5% commonly shared compounds) (Fig. S5). Based on the GC-MS dataset, sesquiterpenes were the dominant VOCs of T. hamatum and T. reesei (Fig. 5b). Trichoderma reesei, in particular, emitted many specific compounds that promoted the statistical separation of the emission profile from the three other Trichoderma species (Fig. 5c,d).
We performed a PCA to investigate the structure of the emission profile of the fungal odor profiles (combination of PTR-ToF-MS and GC-MS dataset; data are shown in Table S4). The biplot (Fig. 6a) illustrates that the first two components could explain 73.1% of the variance, with component 1 (48%) mainly separating T. reesei from the other three Trichoderma species. The component 2 (25%) separated mainly T. hamatum from the other three Trichoderma species. The two species T. velutinum and T. harzianum could also be separated by component 2 (Fig. 6a). The three most important VOCs responsible for the separation by component 1 is an unknown oxygenated sesquiterpene (o-SQT-3), as well as the two sesquiterpenes, b-curcumene and a-muurolene. The most important VOCs for the separation in component 2 were the mass m/z 135.116 (C 10 H 14 + H + , pcymene or o-cymene), selina-4(15),7(11)-diene and the mass m/z 83.086 (C 6 H 10 + H + , hexanal or 2,3-dimethyl-1,3-bytadiene) (Fig. 6a).
Further, we used a supervised RF algorithm to determine the key compounds that characterized the Trichoderma species for each MS dataset. The top five predictors are illustrated in Fig. 6 (b). A heatmap (Fig. 6c) visualizes the distribution of predictors across the four species (Fig. 6c). The emission of a compound resulting in a signal at mass m/z 71.049 (C 4 H 6 O + H + , tentatively originating from either methyl vinyl ketone (MVK), methacrolein (MACR) or the dehydrated form of butyric acid), b-curcumene, a-muurolene, unknown-o-SQT-3 and b-cedrene predicted T. reesei. Trichoderma hamatum was, by contrast, discriminated by the compounds with m/z 135.116 (p-cymene or ocymene), m/z 45.034 (C 2 H 4 O + H + , acetaldehyde) and selina-4 (15),7(11)-diene. For T. harzianum and T. velutinum, no distinct predictor could be identified despite of the slight statistical separation by the PCA analysis (Fig. 6a).

Discussion
Traditionally, fVOCs have often been overlooked partly as a result of analytical limitations (Morath et al., 2012;Schmidt et al., 2015;Li et al., 2016). In the present study we demonstrated a platform to analyze fungal and other microbial VOCs efficiently and systemically. Previously Misztal et al.

Research
New Phytologist (2018) developed a cuvette system for microbial VOC measurements where, however, the VOC analysis was restricted to online MS. The present system takes advantage of the isomer speciation capability of GC-MS and the sensitivity and online measuring capability of PTR-ToF-MS. The detection sensitivity of the latter was additionally increased as fVOCs were allowed to accumulate in the headspace of the culture vessels as a result of the sequential switching between the cuvettes. Combination of the two MS systems with the cuvette system allowed us to comprehensively analyze the fungal volatilomes of the four Trichoderma species. The platform is completed by a data-mining approach enabling analysis of complex datasets. Together, the developed system allows rapid, as well as long-term, microbial/fungal VOC analyses.
To test the performance of the platform, we investigated development-dependent changes in fVOC emission of four fast-growing Trichoderma species over 48 h. The measurements revealed four completely different emission patterns for the four species. Using the data-mining approach, the emission patterns of each individual compound were visualized. The analysis procedure and the results prove that combination of all properties of the new system significantly enhances the accuracy and efficiency of VOC measurements from fungi (and for microorganisms in general). It also enables an analysis of development-dependent emission processes. The revealing of comprehensive fVOC pattern forms the necessary basis for further phenotyping and analysis of fungal chemical diversity. In the present work, we analyzed longterm changes in VOC emission, but alternatively, if the collection In the current study, the PTR-Tof-MS detected the sum signal of all sesquiterpenes (at m/z 205.195 and some other fragment m/z). By contrast, the SBSE-GC-MS analysis revealed up to 27 individual sesquiterpenes from the four Trichoderma spp. As SBSE-GC-MS is an offline method, the integral analysis time using GC-MS cannot be directly compared with the measurement intervals of the semi-online technique (i.e. PTR-ToF-MS). However, the combination of both analytical techniques facilitated the recovery of a virtually complete emission patternthe volatilomeof the investigated fungi. It turned out that the PTR-ToF-MS-detected VOCs were shared very commonly by all four Trichoderma species, whereas the GC-MS analysis enabled us to identify and quantify more species-specific compounds, discriminating, for example, the two species T. hamatum and T. reesei. The detected species-specific VOC profiles here are in accordance with previous studies describing individual Trichoderma species according to their VOCs (Guo et al., 2019). So far, an ecological function has been described only for a few of the Trichoderma VOCs; however, these few have been shown to transmit remarkable effects on the plant performance (Hung et al., 2013;Kottb et al., 2015;Lee et al., 2016Lee et al., , 2019. For example, 6-pentyl-2H-pyran-2-one (6PP) increased the defense of Arabidopsis thaliana with parallel reduced growth (Kottb et al., 2015) and, recently, Trichoderma VOCs were shown to induce  (Estrada-Rivera et al., 2019). Our GC-MS analysis revealed a large chemical diversity of the odors of the different Trichoderma species. Interestingly, previously large adjustments of the emission patterns were shown when Trichoderma was grown in the presence of other fungi (Guo et al., 2019). The high chemical diversity as well as the adjustment of emissions to the changing environment both suggest important ecological functions for these compounds (Guo et al., 2019). The present results revealed strong species dependency, especially for sesquiterpenes.
The sesquiterpene emission patterns showed high chemical diversity between the species, and we also revealed four completely differently fluctuating emission patterns over the 48 h measurement period. Many of other compounds, such as m/z 71.086 (cyclopentane or pentene) and m/z 93.091 (toluene or bicyclo[3.2.0]hepta-2,6-diene), also showed species-specific, temporally fluctuating emission patterns. Although in recent years, a remarkably growing number of studies have focused on the determination of VOCs from fungi Dickschat, 2017), only a few studies have explored temporal changes of fungal emission patterns Weikl et al., 2017;Misztal et al., 2018). Weikl et al. (2017) and Lee et al. (2015) took the first steps to analyze age-dependent changes of fungal VOC emission patterns: Weikl et al. (2017) measured the weekly emission changes from Fusarium spp. and Alternaria spp, whereas Lee et al. (2015) reported age-dependent VOC emissions from 5and 14-day-old T. atroviride. Khomenko et al. (2017), who analyzed the yeast VOCs via PTR-ToF-MS, demonstrated temporal changes in VOC emission during the yeast colony growth. More recently, using a similar approach employing PTR-ToF-MS, Misztal et al. (2018) reported on the emission patterns of selected bacterial and fungal species under various abiotic and biotic environmental conditions. Their results show, except for variation between taxa, that microbial VOC emissions are also dependent on substrate type, biotic environment, growth phase and life cycle. Sesquiterpene emission, in particular, seems to vary strongly according to the growth environment (Gonz alez-P erez et al., 2018;Guo et al., 2019;Misztal et al., 2018). Taking this previous knowledge and the present results together, it seems that several fungal species may alter especially sesquiterpene release when adjusting to the surrounding environment. However, the underlying principles behind the differently behaving emission patterns remain to be elucidated. Fungal VOCs may indeed also have some other, completely different and not yet understood ecological functions in the interorganismic interactions, as suggested in the review by Kramer & Abraham (2012). Moreover, although we talk here about fungal VOCs, we cannot completely rule out the possibility that the investigated axenic cultures involved bacterial endosymbionts (Uehling et al., 2017) which might have influenced the performance of the fungi or even emitted volatiles themselves.
Contrary to the GC-MS analysis, the compounds detected by the PTR-ToF-MS measurements were mostly less complex, suggesting an origin in primary rather than secondary metabolism. To date, an ecological function has been described for only a few of these compounds and, astonishingly, most of these compounds detected by PTR-ToF-MS have never been before reported for Trichoderma species. In addition to the aspects of developing a high-throughput platform for the VOC emission analysis from fungi or other organisms, we propose a basic methodology for data mining of VOCs from different organisms. This approach, based on multivariate statistics and machine learning (RF), significantly facilitates the decoding of the chemical diversity of volatile compounds from fungi and paves the way for the studies on fVOC-mediated fungus-fungus and fungusplant interactions. Machine learning helped us to determine the characteristic VOCs for different species. This approach revealed that Trichoderma compounds detected by PTR-ToF-MS, such as the ones detected at m/z 71.049 (tentatively MVK, MACR or a butyric acid fragment) and m/z 45.034 (acetaldehyde), were crucial discriminators of T. reesei and T. hamatum, respectively. Previous studies have shown that acetaldehyde can inhibit the growth of black mold, Aspergillus niger (Stotzky et al., 1976). Some of the other detected compounds, such as ethanol (m/z 47.049), acetone (m/z 59.050) and l-octen-3-ol/3-octanone (m/z 129.127), were also shown to harbor antifungal activity (Stotzky et al., 1976;Toffano et al., 2017;Baiyee et al., 2019;Lee et al., 2019). Both l-octen-3-ol (m/z 129.127) and 2-octenal/l-octen-3one (m/z 127.112) were also shown to have other crucial ecological functions, such as regulation of plant and seed growth (Kishimoto et al., 2007;Lee et al., 2019) and attraction of insects (Pierce et al., 1991;Chaiphongpachara et al., 2019). Nonetheless, the biological functions of most of the other compounds emitted by Trichoderma spp. and detected by PTR-ToF-MS remain so fa unknown.
Taken together, and considering the analysis of chemical diversity, PTR-ToF-MS and GC-MS (in the present setup) seem to be suitable, in different ways, to distinguish the individual Trichoderma species from each other, as they have different compound detection preferences. While the PTR-ToF is a convenient method for detecting polar, shorter-chained volatile compounds (up to c. m/z 300), the employment of the SBSE-GC-MS is better suited to the separation of nonpolar, long-chained compounds. Together the two complementary techniques allowed us to detect a much wider spectrum of VOCs than using one MS method alone. A future addition will be the integration of an infrared gas analyzer for CO 2 and H 2 O acquisition in the outlet gas flow of the cuvette system. This enables measurements of humidity and fungal respiration rates, which will allow a better correlation of VOC emission rates with the metabolic activity of the fungi.
Moreover, not only because of the limited number of fungal species that have been systematically investigated for their odor profile to date, but also based on the present knowledge that the Ó 2020 The Authors New Phytologist Ó 2020 New Phytologist Trust New Phytologist (2020) 227: 244-259 www.newphytologist.com fungal physiology, as well as biotic and abiotic environments, seems to drastically alter the pattern of fVOCs released (Schmidt et al., 2015;Nieto-Jacobo et al., 2017;Guo et al., 2019), it can be assumed that a large number of volatile compounds are still undiscovered . Also, the present study reveals 52 volatiles (34 volatiles from PTR-ToF-MS and 18 from GC-MS) that have not been previously reported from Trichoderma species.
Recently, co-cultivation systems of fungi and plants have been increasingly used to study the biological effects of fVOCs on plant growth and fitness (Ditengou et al., 2015;Kottb et al., 2015;Lahrmann et al., 2015;Ameztoy et al., 2019;Garc ıa-G omez et al., 2019;Moisan et al., 2019). None of these studies report changes in the emission pattern of a fungi-plant system over time (e.g. the over pre-establishment phase until symbiosis). This can be changed in future by applying analyses platforms such as presented here. So far, only in one study has the effect of fVOCs from two different ages of fungi on plant growth been compared . Indeed, Lee and colleagues revealed a fungal age-dependent response of plants to the fVOCs, alleviating the importance of timing in plant-fungi interactions. Our results also suggest that fungal age matters and can lead to completely different results of interactions. Fundamental technical improvements are necessary to better analyze the physiological processes of formation and release of fVOCs, for example in relation to abiotic and biotic environments. Even basic experimental conditions such as CO 2 concentration, relative air humidity and water availability can have drastic implications for the experimental results (Dannemiller et al., 2017). Future developmental stages of our platform will, therefore, incorporate the cuvettewise regulation of relative air humidity of the inlet air. The control of individual CO 2 concentrations in the inlet air of each individual cuvette is desirable, but a great technical challenge, as individual dilutions would have to be produced for each cuvette using individual mass flow controllers. Further recordings of real-time emissions from fungi, both alone and in interaction with other organisms, will enhance our understanding of bioactive VOCs and their functions. It will be particularly interesting to monitor real-time changes in fungi or other microbes growing towards other microbial or plant species. High-throughput techniques, such as those described here, will help to accelerate the research on fVOCs. This platform can be used not only for chemotyping analysis of VOC emissions from individual fungal species, but also for co-cultivation experiments with fungi, microorganisms and small plants.

Supporting Information
Additional Supporting Information may be found online in the Supporting Information section at the end of the article.      The compounds (mass to charge ratios m/z) detected by proton transfer reaction time-of-flight mass spectrometry (PTR-ToF-MS) and corresponding tentative annotations.

Table S2
Limit of detection (LOD) of individual compounds related to potential emission rate that may be detected from fungi using the VOC platform (normalized to cm -2 mycelium area).

Table S3
Chemical identification and chromatographic characteristics of detected VOCs by GC-MS analysis.
Please note: Wiley Blackwell are not responsible for the content or functionality of any Supporting Information supplied by the authors. Any queries (other than missing material) should be directed to the New Phytologist Central Office.
New Phytologist is an electronic (online-only) journal owned by the New Phytologist Trust, a not-for-profit organization dedicated to the promotion of plant science, facilitating projects from symposia to free access for our Tansley reviews and Tansley insights.
Regular papers, Letters, Research reviews, Rapid reports and both Modelling/Theory and Methods papers are encouraged. We are committed to rapid processing, from online submission through to publication 'as ready' via Early View -our average time to decision is <26 days. There are no page or colour charges and a PDF version will be provided for each article.
The journal is available online at Wiley Online Library. Visit www.newphytologist.com to search the articles and register for table of contents email alerts.
If you have any questions, do get in touch with Central Office (np-centraloffice@lancaster.ac.uk) or, if it is more convenient, our USA Office (np-usaoffice@lancaster.ac.uk) For submission instructions, subscription and all the latest information visit www.newphytologist.com Ó 2020 The Authors New Phytologist Ó 2020 New Phytologist Trust New Phytologist (2020) 227: 244-259 www.newphytologist.com