Methodological and Ecological Caveats in Deciphering Plant Volatile Emissions: the Case Study of Tomato Exposed to Herbivory and Resource Limitation

Over recent years, major progress in experimental approaches have bring insights about the ecological functions of volatile organic compounds (VOCs) released by plants. However, deciphering volatile emissions in a methodologically robust and ecologically relevant manner remains a challenging issue. A surge in interest is required to characterize potential blind spots in volatile sampling that could result in dramatic bias in our understanding of VOCs. In parallel, ecologists need to account for various environmental factors in order to address appropriately the sources of variations of VOCs. Here we use two common porous polymers, polydimethylsiloxane (PDMS) and Porapak Q, to collect VOCs released by tomato exposed to herbivory in combination with nitrogen shortage. We dissect two key features of volatile blends, i.e., their composition and their diversity. Upon nitrogen limitation, Porapak Q stresses the up-regulation of a common defensive compound (methyl salicylate), while herbivory induces three terpenes involved in the recruitment of natural enemies of Tuta absoluta (2-carene, (cid:0) -pinene and β-phellandrene). This study suggests that the combination of resource availability and herbivory governs the differential production of generalist and specic VOCs that are active against a broad spectrum or particular herbivore species, respectively. But PDMS was found unsuitable to observe such patterns in the composition of VOC emissions. Additionally, Porapak Q was found more sensitive than PDMS to track the increase in the diversity of stress-related VOC emissions upon nitrogen limitation. This suggests that plants growing with poor resources release more information in surroundings. We discuss particular implications for tri-trophic-mediated plant defences.


Introduction
Plants release in their surroundings a large number of volatile organic compounds (VOCs) that play a key role in ecological networks. Ecological research in plant-insect interactions has shown that VOCs induce and/or prime anti-herbivore defences in adjacent plants (Li 2016) , affect host-plant location by herbivores (Conchou et al. 2019), attract natural enemies (Dicke and Baldwin 2010) and in uence hyper-parasitoids (Cusumano et al. 2019) . In other words, variations in volatile pro les in uence several trophic levels and, ultimately, determine major ecosystem processes (Hunter 2016) . With this in mind, ecologists expend considerable efforts in addressing sources of variations in VOCs. However, methodological and ecological issues make this purpose challenging. On the one hand, the restricted tools ecologists have access to explore VOCs emissions likely undermine our ability to account for the natural complexity of volatile pro les released by plants. On the other hand, while plants are rarely exposed to individual stress in nature, our understanding of VOCs is mainly based on disparate studies that address separately the in uence of different stresses on volatile emissions. Taken together, we advocate accounting for both sampling issues and realistic combinations of factors driving volatile emissions in order to unravel the roles of VOCs in a robust and relevant manner.
Over the last two decades, great progress has been made in the development of sensitive methods for headspace collection of VOCs (Tholl et al. 2020) . However, most of these methods rely on porous polymers that capture a sub-fraction of genuine VOCs pro les. We hypothesize that the use of single porous polymers, as it is generally done, results in blind spots that hamper the functional interpretation of plant volatile pro les. Because the a nity between porous polymers and individual VOCs determines the fraction of volatilomes collected, sampling methods may conduct in important pitfalls that make our understanding of plant VOCs highly versatile. Among the most common porous polymers, polydimethylsiloxane (PDMS) is used in two widespread sampling procedures, i.e., solid-phase microextraction (SPME) and stir-bar sorptive extraction (SBSE) (Bicchi and Maffei 2012) . In addition, Porapak Q (copolymers of ethylvinylbenzene and divinylbenzene) has gained popularity for its suitability with solvent extraction (Tholl et al. 2020) . In animal ecology, Noushini et al. (2020) have recently shown that sampling methods entail major biases when interpreting insect semiochemicals. In the same line, we need a surge in interest in plant ecology to compare stress-related volatile emissions across different porous polymers. This endeavour may improve the robustness of our functional interpretation of VOCs.
Another important avenue to better understand the functions of VOCs revolves around the integration of induced responses to multiple stress (Pareja and Pinto-Zevallos 2016) . Although insect herbivory (Rowen and Kaplan 2016) and plant nutrition (Islam et al. 2017;Ormeño et al. 2008) in uence VOCs emissions, there are still major gaps in knowledge about joint effects of both factors on plant volatile pro les (but see Chen 2013 in cotton; De Lange and Rodriguez-Saona 2019 in cranberry ). This line of research is required to test classic hypotheses of plant defence theory linking soil fertility with the production of antiherbivore defensive compounds (Stamp 2003) . For instance, the growth:differentiation hypothesis stands that plants growing with poor resources increase their investments in anti-herbivore compounds due to the high cost of tissue replacement upon harsh conditions (Herms and Mattson 1992) .
However, little has been done to investigate how the combination between herbivory and resource availability affects the production of generalist and speci c defensive compounds that are active against a broad spectrum or particular herbivore species, respectively. In the absence of herbivores, resource availability helps to predict the quantity of defensive metabolites (Le Bot et al. 2009). But the quality of such compounds remains the poor relative of the growth:differentiation hypothesis. We hypothesize that plants experiencing resource limitation without herbivory may invest in the production of generalist defensive compounds. In contrast, after herbivores start feeding, a large body of evidence points the elicitation of species-speci c VOCs (Rowen and Kaplan 2016). As a consequence, plant defences are enhanced according to herbivore identity (de Moraes et al. 1998;Mc Cormick et al. 2012), even if the in uence of resource availability in such patterns remains understudied. Beyond the composition of VOC pro les, we believe that their diversity, measured as Shannon entropy, may provide complementary knowledge about the roles of VOCs in plant defences. Shannon entropy is largely used in the assessment of biological diversity but only few ecologists keep in mind that such index is rooted in information theory (Shannon 1948) . In brief, Shannon entropy has been initially developed to estimate the amount of information a message contains. Thereby, studying Shannon entropy of VOC pro les may help to understand how plants exposed to multi-stress adjust information they release in surroundings and, in turn, determine the breadth of VOC-mediated ecological interactions.
Over the last decade, although 69 studies interested in plant-insect interactions have analysed tomato's VOCs emissions, none of them has integrated joint effects of herbivory and nutrient. Only one study has compared volatile pro les across different methods of collection but the authors focused on performance in VOC extraction rather than functional interpretation of VOCs (Shu et al. 2010). In this context, we hope that our approach may help to dissect (i) potential blind spots in sampling procedures of VOCs that play a key role in plant-insect interactions (ii) how nitrogen resources and herbivory affect the production of anti-herbivore VOCs, whether they are active against a broad spectrum or speci c herbivores and (iii) how nitrogen resources and herbivory affect the information amount conveyed by VOC pro les.

Methods And Materials
Sampling Procedures Commonly Used to Explore Volatiles Released by Tomato. Between 2010 and 2020, tomato has been widely used to explore plant VOCs emissions from plant-insect interaction perspectives. We performed a Web of Science query on February 9th 2021 with the following combination of terms: (tomato* OR Solanum lycopersicum) AND ('plant volatile' OR 'volatile organic compound' OR VOC) AND (insect OR herbivore OR pest). We applied additional criteria on the 251 results by removing (i) studies based on electroantennography, electronic nose, speci c families of VOCs, leaf exudates, root VOCs and plant samples that were grounded prior VOC collection and (ii) measures interested in VOCs released by essential oils, mutant plants or plants infested with nematodes, bacteria or viruses. In total, 19 studies used Porapak Q, 19 studies used Tenax, 14 studies used PDMS-based SPME or SBSE, 9 used Super Q, 6 used HayeSep Q and 2 studies used charcoal to collect VOCs (Table S1). Only three studies used a combination of two porous polymers (Tenax and Carboxen) but they did not aim at comparing the respective fractions of VOCs collected. Overall, we report 210 records of VOC pro les. The average number of VOCs collected per plant is relatively low, i.e., 17.67, and displays a high standard deviation, i.e., 13.49 (Table S1). This variability raises critical questions when comparing VOC emissions across studies.
Plant Material and Growth Conditions. Tomato seeds (Solanum lycopersicum L. cv. Better Bush VFN Hybrid, breeder: Tomato Growers, Florida, USA) were sown on humid lter paper in Petri dishes. Five days after sowing (DAS), seedlings were transferred for six days to a hydroponic system adapted to young germinating plantlets set in a growth chamber with the following conditions: 16 h hemeroperiod, 23°C/18°C (day/night) in air and 25°C in solutions, 60 ± 20 % air humidity. The full nutrient solution supplied N-NO 3 − at a low concentration of 0.3 mM. Better Bush is an early dwarf cultivar, easy to manage and well adapted to hydroponics. It is responsive to scari cation elicitation (Royer et al. 2013) and polyphenol concentration is enhanced by nitrogen limitation (Larbat et al. 2012;Larbat et al. 2014) .
Eleven days after sowing, the plantlets were transferred to a Nutrient Film Technique (NFT) system set in the same growth chamber. The entire experiment comprised 32 plants. Half of the plants were maintained at low nitrogen (0.3 mM N-NO 3 − ), which has been shown limiting for tomato growth (Adamowicz and Le Bot 2008) , while the other half received a full nutrient solution with high N-NO 3 − concentration (7 mM). Nitrogen concentration and pH were measured and adjusted manually with increasing frequency as growth proceeded, up to once a day for high nitrogen availability and twice a day for nitrogen limitation. For more details on the composition of nutrient solutions and on their maintenance, see (Royer et al. 2013) . The limited nitrogen concentration does not induce de ciency symptoms on plants during the experiment (Adamowicz and Le Bot 2008;Royer et al. 2016) . All the plants could not be grown concurrently because of the limited space inside the growth room and a limited number of VOC trapping systems. Therefore, the experiment was run over four successive periods using four herbivore-damaged and four control plants each time grown under low or high nitrogen conditions.
Tuta absoluta Larvae and Leaf Infestation. Prior to the experiment, T. absoluta adults were reared on tomato plants in an insectarium set in a climate chamber adjacent to the main experiment. A large population of larvae developed on the leaves. Infestation was made the 39th day after sowing. When plants were at the vegetative stage with 6-8 developed leaves, 12 third instar larvae selected from this population were deposited on the three terminal lea ets of the 4th true leaf for 8 plants exposed to high and low nitrogen regimes. Control plants were slightly touched with a paintbrush to mock larval deposition. Larvae were allowed to feed on leaves until plant harvest just over two days later.
Collection of plant volatiles. One day before the larval deposition, plants were set in PET-bags (polyethylene terephthalate / Albal®, Cofresco Frischhalteprodukte, Minden Germany; 31 cm diameter and 55 cm high). A metal wire, set inside the bag around each plant, prevented contacts between the bag and the leaves. For each plant studied, VOCs were sequentially collected on two adsorbent phases, polydimethylsiloxane (PDMS) tubes (Carl Roth) and Porapak Q lters (ARS, Mecanopoly, Florida, USA), after 47h and 48h of herbivory, respectively. Thereby, we aimed at collecting as many different volatile compounds as possible.
A push-pull system with two vacuum pumps (Thomas, Gardner Denver Inc, Milwaukee, WI USA) was used to pump air into and out of the bag (Tholl et al. 2020) . The inlet and outlet air ow were adjusted with air regulators at 1.5 L min − 1 and 1L min − 1 , respectively. Air was pulled in from the outside through a particle lter at the side of each plastic box and pushed into the PET bag through a charcoal lter and polytetra uoroethylene tubing. Air pulled out the system went through adsorbent phases. The bag bottoms and tops were wrapped around inlet and outlet tubing and sealed by cable binders.
PDMS tubes had an external diameter of 2 mm and a length of 5 mm (Kallenbach et al. 2014) . PDMS tubes were attached with metal strings and hung inside the bags wrapping the tomato shoots for passive trapping. VOC collections on PDMS were performed 47 hours after infestation during a 40 min period. After the collection period, PDMS tubes were stored in glass vials at -20°C until thermodesorption.
Porapak lters consist of glass tubes containing 30 mg Porapak Q (ARS, Florida, USA). VOCs were collected on Porapak 48 hours after infestation during a 2h period. After VOC collection, the volatile compounds were eluted with 200 mL of dichloromethane containing 10 ng mL − 1 of an internal standard (n-nonylacetate; Sigma-Aldrich, Seelze, Germany). Eluates were then stored in glass vials at − 20°C until analysis.
Volatile Analysis and Quanti cation. Compounds collected with PDMS were analysed with gas chromatography-mass spectrometry using a thermal desorption unit (TD20, Shimadzu) connected to a quadrupole GC-MS (QP2010Ultra, Shimadzu) with transfer line temperature: 240°C, ion source temperature: 220°C, and mass range: 33-400 m/z. Volatiles were desorbed from PDMS tubes placed in 89 mm glass TD tubes (Supelco), through a liquid nitrogen ow at a rate of 60 mL min − 1 at 200°C for 8 min. Compounds were separated with helium at a ow rate of 40 cm sec − 1 through a 30 m x 250 mm, 0.25 mm Rtx-5MS column. The oven temperature was held at 40°C for 5 minutes and then increased to 185°C at a rate of 5°C min − 1 . At the end of the run, it increased to 280°C at 30°C min − 1 for 0.83 minute to clean the column. Every ten samples, a blank tube was included. Data processing was performed using the program Shimadzu GC-MS solution software (v2.72) provided by the manufacturer and exported via its export function into an MS Excel macro le. The relative proportion of each molecule in the volatile blend was computed by dividing its peak area by the total peak area of all identi ed molecules.
Analyses of the compounds trapped with the Porapak lters were performed using gas chromatography (Agilent 6890 series, Agilent Santa Clara, CA, USA) coupled either to a quadrupole mass selective detector (Agilent 5973, interface temp. 270°C, quadrupole temp. 150°C, source temp. 230°C; electron energy 70 eV) or a ame ionization detector (FID, temp. 300°C). Compounds were separated in a 30 m x 250 mm x 0.25 mm DB5-MS column (Wilcom GmbH, Heppenheim, Germany) using He (mass detector) or H 2 (FID) carrier gases. One microliter of the sample was injected without splitting at an initial oven temperature of 40°C (2 min hold) followed by a ramp to 155°C (1°C min − 1 ), a ramp to 300°C (60°C min − 1 ), and a hold for 3 min. A semi-quanti cation of all VOCs was performed using their ame ionization detection peak area referenced to the area of the internal standard, using the same method as GC-MS. Emission rates (X) were calculated in pg.h − 1 : with RF the response factor corresponding to the compound (RF = 1, when unknown), PA(x) the compound X peak area, PA(IS) the internal standard peak area, the factor "2" corresponds to the trapping time (hours), and the bag inlet and Porapak lter outlet ow rates, respectively.
The following compounds were identi ed by comparison of retention times and mass spectra with authentic standards: (E)-β-caryophyllene, α-copaene, α-humulene, (Z)-β-ocimene, α-phellandrene, α-pinene, α-terpinene. In the absence of standards, compounds were tentatively named by comparison with mass spectra libraries of Wiley and the National Institute of Standards and Technology, as in McCormick et al. (2016) . The compounds were tentatively named only when they matched with more than 85 % identity. Putative identi cation of compounds with similar or unknown chemical structures were indexed with Arabic numerals.
Statistical Analyses. All the statistical analyses were performed with R, version 4.0.3 (R core team 2020) .
In order to identify speci c signatures in the composition of VOC pro les, we performed a discriminant analysis of principal components (Yang and Yang 2003) . For this purpose, we rst carried out a principal component analysis with the package 'vegan' (Oksanen et al. 2020; function 'rda'). Then, we used the principal component scores of each individual pro le to conduct a linear discriminant analysis (LDA; function 'lda'). This approach resolves collinearity computational issues that generally impede classical LDA. Moreover, we believe that performing ordination based on principal components, rather than speci c productions of VOC, is particularly relevant in metabolomic. Indeed, principal components can be considered as composite dimensions accounting for positive and negative correlations between single metabolites (i.e., shared pathway and trade-offs, respectively). In brief, LDA is a supervised classi cation procedure that aims to predict the probability of belonging to a given category based on predictor variables. The suitability of the LDA is assessed with the percentage of correct classi cation. In our case, correct classi cations were optimal for LDA performed on scores deriving from the rst 6 principal components (78% and 66% of correct classi cations for Porapak Q and PDMS, respectively).
We estimated the diversity of plant volatile pro les based on Shannon entropy computed with the package 'vegan' (Oksanen et al. 2020; function 'diversity'). Signi cant differences in the diversity of VOC pro les across the four treatments (i.e., nitrogen by herbivory factorial matrix) were assessed with linear models. We performed multiple comparison via the package 'multcomp' (Hothorn et al. 2008; function 'glht'; Tukey's HSD tests).

Results
The Two Porous Polymers Collect Different Fractions of VOC Pro les. Porapak Q and PDMS collect an equal number of compounds, i.e., 33 VOCs (Table 1 and Table 2). However, two third of these VOCs were speci c to each porous polymer. Only 12 compounds were common to Porapak Q and PDMS: 2 aldehydes (decanal and nonanal), 1 alkane (decane), 5 monoterpenes (p-cymene, β-phellandrene, αpinene, α-terpinene and terpinolene) and 4 sesquiterpenes ((E)-β-caryophyllene, α-humulene, copaene and δ-elemene). a -Compounds in bold were identi ed by comparison of retention times and mass spectra with authentic standards. In the absence of standards, tentative identi cation was done by comparison with mass spectra from the libraries of Wiley and the National Institute of Standards and Technology (more than 85% of identity was required to assign a tentative name) .
A semi-quanti cation of all VOCs was performed using their ame ionization detection peak area referenced to the area of the internal standard. a -Compounds in bold were identi ed by comparison of retention times and mass spectra with authentic standards. In the absence of standards, tentative identi cation was done by comparison with mass spectra from the libraries of Wiley and the National Institute of Standards and Technology (more than 85% of identity was required to assign a tentative name) .
A semi-quanti cation of all VOCs was performed using their ame ionization detection peak area referenced to the area of the internal standard. a -Compounds in bold were identi ed by comparison of retention times and mass spectra with authentic standards. In the absence of standards, tentative identi cation was done by comparison with mass spectra from the libraries of Wiley and the National Institute of Standards and Technology (more than 85% of identity was required to assign a tentative name) .
A semi-quanti cation of all VOCs was performed using their ame ionization detection peak area referenced to the area of the internal standard. a-Compounds in bold were identi ed by comparison of retention times and mass spectra with authentic standards. In the absence of standards, tentative identi cation was done by comparison with mass spectra from the libraries of Wiley and the National Institute of Standards and Technology (more than 85% of identity was required to assign a tentative name) .
The relative proportion of each compound in individual volatile blends was computed by dividing its peak area by the total peak area of all identi ed compounds. a-Compounds in bold were identi ed by comparison of retention times and mass spectra with authentic standards. In the absence of standards, tentative identi cation was done by comparison with mass spectra from the libraries of Wiley and the National Institute of Standards and Technology (more than 85% of identity was required to assign a tentative name) .
The relative proportion of each compound in individual volatile blends was computed by dividing its peak area by the total peak area of all identi ed compounds. a-Compounds in bold were identi ed by comparison of retention times and mass spectra with authentic standards. In the absence of standards, tentative identi cation was done by comparison with mass spectra from the libraries of Wiley and the National Institute of Standards and Technology (more than 85% of identity was required to assign a tentative name) .
The relative proportion of each compound in individual volatile blends was computed by dividing its peak area by the total peak area of all identi ed compounds.
Changes in the Composition of Stress-Related VOC Pro les. Linear discriminant analyses account for most of inter-treatment variability, both with Porapak Q (95%; Fig. 1a) and PDMS (98%; Fig. 1b). In the legend of Fig. 1, only the 3 VOCs harbouring the highest correlations with the 6 principal components are shown (c.f., Tables S2 and S3 for exhaustive correlations between VOC emissions and principal components).
For VOCs captured on Porapak Q, we report a signi cant discrimination between the four treatments along LD1 and LD2 (Fig. 1a). This discrimination is mainly based on PC2, PC3 and PC4. Low and high nitrogen regimes result in opposite patterns of VOC emission in herbivore-free plants. Non-infested plants exposed to limited nitrogen display high emissions of methyl salicylate and unknown #1 along with reduced production of propanoic ester #2 (i.e., negative correlations with PC4). On the contrary, uninfested plants with non-limiting nitrogen reduced the release of methyl salicylate and unknown #1 while they enhance the emission of propanoic ester #2 (i.e., positive correlations with PC4). The in uence of herbivory on VOC emissions depends on nitrogen regimes. Upon high nitrogen supply, herbivory triggers higher emissions of three terpenes, 2-carene, -pinene and β-phellandrene (i.e., positive correlations with PC2). When nitrogen supply is limited, herbivory is associated with a general relaxation of VOCs such as alkane #1, decane and ascaridole (i.e., positive correlations with PC3).
With PDMS, volatile pro les discriminate mainly based on PC6 and PC3. Only herbivore-infested plants exposed to nitrogen limitation segregate along LD2 (Fig. 1b). These plants are characterized by reduced emissions of terpinolene and 4-caranol and high releases of 4-methylmannitol (i.e., negative correlations with PC3). The three other treatments partially overlap along LD1 although herbivore-free plants exposed to high and limited nitrogen supply clearly differ in VOC emissions. First, herbivore-free plants with nonlimiting nitrogen mainly produce terpinolene and 4-caranol parallel to a decrease in the emission of 4methylmannitol (i.e., positive correlations with PC3). Second, herbivore-free plants under restricted nitrogen regime produce preferentially geraniol and reduce the emission of alloocimene and octanoic acid ester (i.e., positive correlations with PC6).
Changes in the Diversity of Stress-Related VOC Pro les. Porapak Q enables to point variations in the diversity of VOC pro les between treatments (Fig. 2a). Although herbivory does not in uence the Shannon entropy of plant volatile emissions, our results show that nitrogen limitation results in signi cantly higher diversity of VOC pro les. The Shannon entropy increases by 18% between high and low nitrogen nutrition (1.47 ± 0.04 and 1.73 ± 0.04, respectively). However, PDMS does not allow to track such variations (Fig. 2b).

Discussion
This study highlights two pitfalls in exploring the role of plant volatiles. Because porous polymers differ in their a nity with VOCs, they entail important blind spots that hamper the functional interpretation of plant volatile emissions. Additionally, we stress the importance to consider the interplay between biotic and abiotic stresses to better understand variations in the composition and the diversity of VOC pro les.
More particularly, we show that the combination of nitrogen limitation and herbivory governs the differential production of generalist and speci c anti-herbivore VOCs. Hence, our results not only endorse the growth:differentiation hypothesis, they also help to complete it by dissecting the relationships between soil fertility and the nature of defensive compounds produced by plants. Finally, we argue that accounting for Shannon entropy of VOC pro les may bring insights to estimate the information amount plants release in surroundings and, thereby, how they adjust the breadth of VOC-mediated ecological interactions in response to stress.
In our case, Porapak Q is more e cient than PDMS to collect VOCs playing key roles in plant defences against herbivores (e.g., see below for methyl salicylate and 2-carene). Moreover, our results indicate that nitrogen limitation is an important driver in the production of anti-herbivore VOCs. To our knowledge, this is the rst study reporting the effect of nitrogen limitation on the up-regulation of methyl salicylate, a common compound involved in anti-herbivore defences (Heil and Ton 2008;Rodriguez-Saona et al. 2011) . In the same line, Copolovici et al. (2014) have shown that abiotic stress can elicit plant volatile defences against herbivores. This pattern endorses the growth:differentiation hypothesis (Herms and Mattson, 1992) . So far, while resource availability has been shown to explain general patterns in the distribution of major plant defensive ecotypes (Fine et al. 2006;Woods et al. 2012) , little has been done to nely dissect the nature of anti-herbivore compounds produced by plants growing in poor and fertile soils. Here, we corroborate our initial hypothesis predicting that herbivore-free plants exposed to nitrogen limitation enhance the production of generalist VOCs, such as methyl salicylate, that prevent infestation from a large spectrum of herbivores. Indeed, methyl salicylate plays an important role in the acquisition of systemic resistance against insects and pathogens (Heil and Ton 2008) but also in the recruitment of a wide range of generalist natural enemies of herbivores (Rodriguez-Saona et al. 2011) . Once herbivory becomes effective, plants switch the production of VOCs towards more speci c anti-herbivore compounds. Our results indicate that herbivory by T. absoluta results in the overexpression of three terpenes (2-carene, -pinene and β-phellandrene) which have been shown to attract Macrolophus pygmaeus, a predator of the tomato leafminer (de Backer et al. 2017) . However, the up-regulation of these speci c terpenes involved in the recruitment of natural enemies is restricted to plants growing with high nitrogen resources. This pattern re ects how resource availability likely in uence the cost-bene t balance of synthetizing anti-herbivore specialized metabolites. The present study shows that, parallel to authentic variations in plant volatile emissions, differences in stress-related VOC pro les arise from chemical collection e ciencies. In our case, PDMS was found ine cient to collect two VOCs playing a key role in tomato's defences against T. absoluta, i.e., methyl salicylate and 2-carene. In addition, although -pinene and β-phellandrene were collected on both PDMS and Porapak Q, only the latter enables to point their induction in response to herbivory. Therefore, studies based on particular porous polymers may fail in interpreting the ecological function of VOCs. In this context, we advocate for a shift in practice in plant ecology, from single-to multi-bed adsorbent tubes containing different porous polymers. However, such sampling methods need to be nely optimized as it has been shown in atmospheric sciences (Ho et al. 2018) . Moreover, this study stresses the necessity to integrate resource availability when addressing the functions of VOCs in anti-herbivore plant defences.
Here, we show that nitrogen resources are a major source of variations in volatile emissions that helps to better understand plant strategies in the production of generalist and speci c anti-herbivore VOCs, but also in the diversity of volatile pro les. Altogether, this study points the need to improve sampling procedures and, at the same times, account for realistic combination of factors driving VOC emissions in order to explore the ecological function of VOCs.

Declarations
Funding -The Ph.D. grant of Victoire Coqueret was co-funded by INRA (Département Environnement et Agronomie) and the Regional Council of Lorraine. All our thanks to the Regional Council of Lorraine for its  Table S2

Supplementary Files
This is a list of supplementary les associated with this preprint. Click to download. Supplementarymaterial.docx