Smell the change: On the potential of gas‐chromatographic ion mobility spectrometry in ecosystem monitoring

Abstract Plant volatile organic compounds (pVOCs) are being recognized as an important factor in plant–environment interactions. Both the type and amount of the emissions appear to be heavily affected by climate change. A range of studies therefore has been directed toward understanding pVOC emissions, mostly under laboratory conditions (branch/leaf enclosure). However, there is a lack of rapid, sensitive, and selective analytical methods, and therefore, only little is known about VOC emissions under natural, outdoor conditions. An increased sensitivity and the identification of taxon‐specific patterns could turn VOC analysis into a powerful tool for the monitoring of atmospheric chemistry, ecosystems, and biodiversity, with far‐reaching relevance to the impact of climate change on pVOCs and vice versa. This study for the first time investigates the potential of ion mobility spectrometry coupled to gas‐chromatographic preseparation (GC‐IMS) to dramatically increase sensitivity and selectivity for continuous monitoring of pVOCs and to discriminate contributing plant taxa and their phenology. Leaf volatiles were analyzed for nine different common herbaceous plants from Germany. Each plant turned out to have a characteristic metabolite pattern. pVOC patterns in the field would thus reflect the composition of the vegetation, but also phenology (with herbaceous and deciduous plants contributing according to season). The technique investigated here simultaneously enables the identification and quantification of substances characteristic for environmental pollution such as industrial and traffic emissions or pesticides. GC‐IMS thus has an enormous potential to provide a broad range of data on ecosystem function. This approach with near‐continues measurements in the real plant communities could provide crucial insights on pVOC‐level emissions and their relation to climate and phenology and thus provide a sound basis for modeling climate change scenarios including pVOC emissions.

In spite of these studies, however, major knowledge gaps persist, which are mainly due to the technical limitations of the available methodologies. Most research on pVOCs is conducted with branch/ leaf enclosure experiments (Materić et al., 2015), providing accurate measurements of the quantity and quality of the pVOCs (Šimpraga et al., 2011). For calculating vegetation level emissions, the data have to be extrapolated based on known (species specific) pVOC emissions per unit dry leaf mass in combination with estimates of overall leaf mass per unit area (Lahr et al., 2015;Purves et al., 2004), introducing a range of uncertainties into the calculations. Alternatively, the ambient air can be analyzed directly (Seco et al., 2015), but here the relative and absolute contribution of individual plant species tend to remain unclear and analyses frequently have to be restricted to a subset of chemical substances for methodological reasons, typically isoprenes (Pegoraro et al., 2006;Seco et al., 2015).
However, in order to reliably model vegetation level changes in pVOC emissions, changes across the different (major) species components have to be recorded in long time series of more or less continuous measurements. However, the commonly used, very expensive, and large mass spectrometry instruments are usually stationary and can be used for individual measurements only.
Furthermore, monitoring different climatic (temperature, humidity) and ecological factors (herbivore attack, shifts in species composition) as well as the concomitant pVOC-loads in the ambient air is required, ideally in parallel with other sources of atmospheric chemistry (pollutants such as pesticides). pVOC emissions in temperate regions are expected to be strongly affected by phenology, for example, by bud break, anthesis, fruit maturation, and leaf fall (Peñuelas & Staudt, 2010). However, no data sets on real-life shifts in pVOC emissions in response to phenology have been provided for temperate vegetation-mainly because of the technical limitations currently in place. pVOC measurements presently have been carried out mostly in a laboratory setups, using preconcentration tools and subsequent thermal desorption for gas-chromatography coupled to mass spectrometry (GC/MS) (Tholl et al., 2006). GC/MS analysis is comprehensive and-by using preconcentrations steps-also sensitive.
However, the instruments are bulky and expensive, sampling, and analyses are time-consuming, and therefore, GC/MS is not suitable for continuous or even quasi-continuous monitoring. Other methods applied for a fast analysis such as electronic noses (e.g., Kunert, Biedermann, Koch, & Boland, 2002) are not sufficiently selective to provide rapid and concurrently sensitive, selective pVOC identification and quantification.
Accordingly, there is a pressing need of tools for continuous monitoring of ecosystem function and associated responses of biodiversity as well as for the relevant environmental parameters. In this context, for the first time we investigated the potential of ion mobility spectrometry (IMS) coupled to gas-chromatographic (GC) preseparation for a continuous monitoring of biodiversity. In particular, we used gas-chromatographic preseparation (GC-IMS) for the characterization of species-specific metabolite patterns, potentially opening the way to identify plant species in a natural community and quantify their phenology.

| I ON MOB ILIT Y S PEC TROME TRY COUPLED TO G C-IMS
Ion mobility spectrometry is an analytical method for the sensitive and selective identification and quantification of trace substances in the gas phase. In general, the operational gas (nitrogen or synthetic air)-the so-called drift gas-is ionized by a radioactive source which initially ionizes the nitrogen molecules. After a multistep reaction chain, protonated water clusters remain as stable reactant ions-the full theory is described in the literature (Eiceman, Karpas, & Hill, 2016). The ions are accelerated in a weak electric field (330 V/cm) toward the Faraday plate functioning as detector. The drift region is separated from the ionization region by a Bradbury-Nielsen grid. This grid is opened every 100 ms for 100 μs to allow an ion cloud to enter the drift region. During their drift toward the detector, the ions collide with the present drift gas molecules. Collision frequency depends on the shape and size of the ion and results in a characteristic resulting velocity of the ion in the drift region. Measuring the drift time of the ion and normalizing it to drift length, temperature, and pressure permits the calculation of the reduced ion mobility, which is characteristic for the ion (Vautz, Bödecker, Bader, & Perl, 2009).
When a sample containing a particular analyte with a proton affinity higher than water is introduced in the ionization region, proton transfer from the reactant ions to the analyte molecules takes place. Those ions in general have different collision cross sections and therefore reach the detector with a different velocity which is characteristic for the analyte and renders its identification possible.
Starting in the 1970s, this technology was first applied for military purpose such as the detection of chemical warfare agents and later on for security, for example, at airports for the detection of drugs and explosives (Eiceman et al., 2016). However, those instruments are optimized to detect only the targeted compound groups and therefore do not provide the full analytical performance. In the last two decades, IMS research has focused more and more on other, civilian applications in process control (Baumbach, 2006;Vautz, Mauntz, Engell, & Baumbach, 2009), biology (Perl et al., 2011;, and medicine Vautz et al., 2010).
All these applications have in common, that the samples are extremely complex mixtures, often with high humidity. Introducing such a sample directly into the ionization region leads to clustering of the different ions with each other and with water molecules, thus complicating identification or rendering it impossible. To solve this problem, IMS is coupled to rapid gas-chromatographic preseparation. Thereby, analytes ideally reach the ionization region already preseparated, thus avoiding the clustering effect on the one hand and additionally obtaining characteristic retention times as a measure for identification on the other.
The functional scheme of an IMS measurement is shown in Figure 1. The experimental setup required is quite simple-the method works at ambient pressure and temperature. Therefore, small and even mobile instruments are available today without compromising the analytical performance. This makes the method highly relevant for many applications where in situ identification of volatile compounds is required.
It needs to be emphasized, that under field conditions with fluctuating temperature and pressure, the performance of the instrument is still reliable. This is due to the fact, that the reduced ion mobility obtained from the measurements is the inverse of the measured drift time of the ions normalized to the experimental (electric field and drift length) and to the environmental conditions (temperature and pressure). Therefore, the reduced ion mobility is stable under normal environmental conditions. Furthermore, with the reactant ion peak, the instruments provide an internal standard for the determination/validation of the mobility values.
In the complex data obtained from such measurements consists of signal intensity vs. ion mobility and retention time. Particular constituents of the sample can be identified by comparing those values for the peaks detected with a reference data base (Bödeker, Vautz, & Baumbach, 2008;Perl, Bödeker, Jünger, Nolte, & Vautz, 2010). If unknown peaks are detected, parallel measurements on adsorption materials and subsequent thermal desorption GC/MS analysis provide a hypothesis on their identity, which can then be validated by GC/IMS analysis of the respective reference substances and inclusion in the reference database. For quantification, the instruments are calibrated by evaporation of the reference analyte and dilution down to the ppt V range by help of a suitable calibration gas generator (Vautz & Schmäh, 2009).

| E XPERIMENTAL
For the determination of the pattern of volatiles released from the leaves, they were placed in a 20-ml glass vial and the headspace was introduced in the GC-IMS system as summarized in the scheme in  to avoid memory effects.

| GC-IMS
Data acquisition software was also custom-made at ISAS, and the data processing and evaluation was carried out by IONysos, a custom-made software by ION-GAS GmbH, Dortmund, Germany.
In particular, the software automatically normalizes the drift time to the known ion mobility of the reactant ion peak (Vautz, Bödecker, et al., 2009) and the retention time to that of a known compound (Bödeker et al., 2008;Perl et al., 2010). After the automatic or manual location of a particular compound peak (ion mobility and rentention time), signal intensities, and peak volumes can be compared for a vast data set. Moreover, after a previous calibration, signal intensities can be converted into concentrations (Vautz & Schmäh, 2009). Finally, regions of interest can be defined and a weight can be assigned related to the relevance of the particular signal, thus enabling an automatic characterization of an unknown data set later on.

| Headspace sampling
For the analysis, 3 g of the leaves of each plant was introduced into a 20-ml glass vial at 60°C. This was performed immediately after harvesting the leaves, and the leaves were placed in the vial as a whole. The glass vial was flushed with a carrier gas flow of 100 ml/ min nitrogen (quality 6.0) and after 5 min-to enable a steady state in the vial-the carrier gas was introduced into the sample loop of the GC-IMS for 10 s to avoid memory effects.

| Exemplary plants
In order to investigate the presence of taxon-specific differences in pVOC emissions, we selected a total of nine common herbaceous plant species from Central Europe (see Table 1). Some of them have a characteristic vegetative odor detectable by the human nose (Cota tinctoria, Lamium album) and thus were expected to have a characteristic "bouquet" detectable by the analytical technique of GC-IMS here employed. However, the other plants investigated were specifically chosen because they have no distinctive scent at all (e.g., Carex muricata, Juncus effuses, Galium album) to evaluate the potential of the method also for such species.

| Data acquisition and evaluation, reproducibility
Before each particular measurement, a blank with clean and humid nitrogen was measured to avoid memory effects. We repeated the measurements for three different plants of each species and for each particular plant at least three times with different leaves. Data were acquired using qIMS software (ISAS, Dortmund, Germany), and data evaluation was carried out using IONysos (ION-GAS GmbH, Dortmund, Germany).
In general, when starting investigations on particular plants or ecosystems, respectively, some of the detected peaks will not be included in the GC-IMS substance database. In this case, reference analysis of the samples drawn on adsorption materials and later thermal desorption GC-MS will be carried out for a comprehensive analysis.

By comparison of the GC/IMS and GC-MS retention time, proposals
for the identity of the unknown substances can be obtained and will be validated subsequently by analysis of the pure substance. From this procedure, the completeness of the substance database will increase with time and will render the identification of more on more substances directly from the GC-IMS measurements possible.
This procedure was not carried out for the present study due to the high technical expenditure and the required personnel. However, from the procedure chosen here, it is assured that the compounds detected derive from the plants only. But certainly, for application F I G U R E 2 Scheme of the experimental setup for detecting the volatiles from leaves. During sampling (left), the sample loop is flushed with the headspace of the leaves. During analysis (right), the volume of the sample loop is introduced into the analytical system In the following, GC-IMS data are presented as threedimensional chromatogram with the inverse ion mobility as xaxes (which is proportional to the measured drift time), with the retention time as y-axes and the color coded signal intensity (see Figure 3 as an example).

| RE SULTS AND D ISCUSS I ON
In the first step, leaves of all particular plants were analyzed as It is obvious that the pattern of 47 peaks is much more complex than required for a reliable differentiation of the nine plant species. Therefore, we condensed the table in Figure 5 the minimum number of peaks required for identification of the specific plants using the most significant signals first. Furthermore, we used threshold values for each of the peaks. By that, we can built a decision tree as demonstrated in Figure 6, which enables the identification of each one of the plants from on nine peaks detected in the headspace as describes above.

| CON CLUS I ON S AND OUTLOOK
We here demonstrate that a unique and characteristic pattern of metabolites could be determined for each specific plant in the set of nine species investigated here-even those which have no perceptible odor to the human nose-using ion mobility spectrometry Under real-life atmospheric conditions pVOCs concentrations can be expected to be much lower than in the setup chosen here.
Therefore, further work has to be directed toward significant increase in sensitivity, for example, using in-line preconcentration tools. The sensitivity of the technique in the present setup is in the low ppb V down to the ppt V range. However, in-line preconcentration tools can be applied and, with sampling intervals of few minutes, detection limits in the ppq V range can be obtained. This work is already in progress at ISAS, Dortmund, Germany. With such a setup, the concentration range to be expected in nature can be covered in a quasi-continuous mode with a temporal resolution below 1 h.
Furthermore, a comprehensive database of pVOCs-probably focused on a particular ecosystem-needs to be developed, including anthropogenic compounds such as pesticides and pollutants from industry or traffic. For this reason, all compounds detected need to be identified to permit their classification as pVOCs, pesticides, or industrial/traffic pollutants.
In the present study, by comparison with our substance data base, several compounds could be identified such as many terpenes (e.g., 3-Carene, Limonene, γ-Terpinene), compounds usually found in essential oils and some aldehydes could be identified. However, those results have not been independently validated. Other compounds not included in the database need to be identified in the future. To do so, additional GC-MS measurements are required, followed by a validation with reference compounds. This has to be performed only once and will permit automatic identification in the following. Moreover, the pVOC patterns documented here will also need to be investigated with regard to stability, phenological variation, and to environmental and anthropogeneous influences.
With the data provided by this method, GC-IMS has the potential to give additional and in particular continuous information on short-term and long-term variations in pVOC emissions, permitting detailed insights into the physiological status of vegetation, its reaction to changing environmental conditions, and the feedback between vegetation and climate. In real-world applications, the patterns recorded are likely to be more complex and at this stage it will be only possible to determine the major components (dominant species) for a given ecosystem, plus the major contaminants. However, these data will already provide a very good proxy of ecosystem reactions to intrinsic and extrinsic factors and permit a new approach in ecosystem modeling and the quantification of ecosystem services.
The aim of the presented approach is thus not the complete documentation of biodiversity in a given habitat-which will not change in minutes or hours-but rather to trace physiological ecosystem reactions in near real-time and thus to understand ecosystem reaction to extrinsic and intrinsic factors. These physiological shifts are what precedes and often causes shifts in ecosystem function and biodiversity. It is exactly these shifts that have been impossible to detect and record with any precision in the past, and it is these shifts we need to understand in order to monitor and model ecosystem reactions to, for example, climate change. Furthermore, there is potential for the monitoring of many other compounds relevant for ambient air chemistry and thus on climate.

ACK N OWLED G M ENTS
The financial support of the Bundesministerium für Bildung und Forschung and the Ministerium für Innovation, Wissenschaft und Forschung des Landes Nordrhein-Westfalen is gratefully acknowledged.

CO N FLI C T O F I NTE R E S T
All authors disclose any potential sources of any conflict of interest.