Electroantennogram and machine learning reveal a volatile blend mediating avoidance behavior by Tuta absoluta females to a wild tomato plant

Tomato cultivation is threatened by the infestation of the nocturnal invasive tomato pinworm, Tuta absoluta. This study was based on field observations that a wild tomato plant, Solanum lycopersicum var. cerasiforme, grown in the Mount Kenya region, Kenya, is less attacked by T. absoluta, unlike the cultivated tomato plants like S. lycopersicum (var. Rambo F1). We hypothesized that the wild tomato plant may be actively avoided by gravid T. absoluta females because of the emission of repellent allelochemical constituents. Therefore, we compared infestation levels by the pest in field monocrops and intercrops of the two tomato genotypes, characterized the headspace volatiles, then determined the compounds detectable by the insect through gas chromatography-linked electroantennography (GC-EAG), and finally performed bioassays using a blend of four EAG-active compounds unique to the wild tomato. We found significant reductions in infestation levels in the monocrop of the wild tomato, and intercrops of wild and cultivated tomato plants compared to the monocrop of the cultivated tomato plant. Quantitative and qualitative differences were noted between volatiles of the wild and cultivated tomato plants, and between day and night volatile collections. The most discriminating compounds between the volatile treatments varied with the variable selection or machine learning methods used. In GC-EAG recordings, 16 compounds including hexanal, (Z)-3-hexenol, α-pinene, β-myrcene, α-phellandrene, β-phellandrene, (E)-β-ocimene, terpinolene, limonene oxide, camphor, citronellal, methyl salicylate, (E)-β-caryophyllene, and others tentatively identified as 3,7,7-Trimethyl-1,3,5-cycloheptatriene, germacrene D and cis-carvenone oxide were detected by antennae of T. absoluta females. Among these EAG-active compounds, (Z)-3-hexenol, α-pinene, α-phellandrene, limonene oxide, camphor, citronellal, (E)-β-caryophyllene and β-phellandrene are in the top 5 discriminating compounds highlighted by the machine learning methods. A blend of (Z)-3-hexenol, camphor, citronellal and limonene oxide detected only in the wild tomato showed dose-dependent repellence to T. absoluta females in wind tunnel. This study provides some groundwork for exploiting the allelochemicals of the wild tomato in the development of novel integrated pest management approaches against T. absoluta.


Results
Infestation levels of Tuta absoluta in monocrops and intercrops of the two tomato genotypes. The results showed that the infestation levels were significantly different in all the weekly leaf samples between the monocrop of cultivated and wild tomato plants and the intercrop systems (Fig. 1). In general, the numbers of mines and larvae per leaf were not affected by the experimental block design (F (3,9) = 0.544; P > 0.05 and F (3,9) = 1.034; P > 0.05, respectively) but were significantly influenced by the treatment (F (3,9) = 24.32; P < 0.001 and F (3,9) (Fig. 1a,b). By the seventh week, the preference for cultivated tomato plants dropped and there were no significant differences in the mean number of larvae per leaf per week between the four treatments (P > 0.05) (Fig. 1a). However, the mean number of mines remained higher in the monocrop of cultivated tomato (CT) than in the other treatments (Fig. 1b).
When the weekly samplings were pooled per treatment, the mean numbers of mines and larvae per leaf were significantly different among the cropping systems (χ 2 = 261.2, df = 3, P < 0.001 for mines and χ 2 = 73.04, df = 3, P < 0.001 for larvae) (Fig. 2). The numbers of mines were highest in the monocrop of cultivated tomato, followed by the intercrops of the two genotypes, then the monocrop of the cultivated tomato surrounded by the wild tomato, and lowest in the monocrop of the wild tomato (Fig. 2). Likewise, the mean numbers of larvae were higher in the monocrop of the cultivated tomato compared to the other cropping systems (Fig. 2). In the intercrops, infestation by T. absoluta was lower when the wild tomato was used as a border crop than when it was used as an interline crop (Fig. 2).
Analyses of tomato headspace volatiles. The wild and cultivated tomato plants emitted distinct chemical profiles during the day and at night. A total of 74 VOCs were recorded and 61 identified, mainly comprised of monoterpenes and sesquiterpenes, and also some aldehydes, alcohols, ketones, and benzenoids ( www.nature.com/scientificreports/ were both qualitative and quantitative differences in the volatiles between the two tomato genotypes, and the collection times. The wild and cultivated tomato plants emitted 74 and 69, respectively, during the day collection, while 60 and 46 were detected, respectively, in the wild and cultivated night volatile extracts. The compounds (Z)-3-hexenol, limonene oxide, camphor, citronellal and germacrene B were uniquely detected in headspace of the wild tomato plant. The compounds tricyclene, dill ether, α-terpineol, (-)-car-3-en-2-one, α-terpinen-7-al, p-ethyl acetophenone, p-cymen-7-ol, thymol, (Z)-β-caryophyllene, α-cedrene and four unidentified compounds (no. 8, 10, 11 and 12) were detected in the day volatiles of both wild and cultivated tomato plants but not in the night volatiles (Table 1). Quantitative differences were also found in the volatile emission between the tomato plants, and between the collection times. The total emission of volatiles was higher in the day collection than in the night collection ( Table 1). The most dominant monoterpenes were δ-2-carene, α-phellandrene, p-cymene, and β-phellandrene , and the most dominant sesquiterpene was (E)-β-caryophyllene from the wild and cultivated tomato plants ( Table 1). The compounds p-cymene, allo-ocimene, decanal, 4-(1-methylethyl)-benzaldehyde, 3-(1-methylethyl)-phenol, β-bourbonene, γ-methyl ionone, 7-methyl-3-octen-2-one, dill ether, and germacrene D were highly emitted during the day compared to the night. On the contrary, the emission rates of limonene oxide, citronellal and camphor were highest in the night volatiles of the wild tomato plant compared to the volatiles of the other treatments (Table 1). There were few quantitative variations between the wild and cultivated tomato plants. The emission rates of α-thujene and δ-3-carene were higher in the wild tomato compared to the cultivated tomato, whereas those of tricyclene and caryophyllene oxide were higher in the cultivated than in the wild tomato (Table 1). www.nature.com/scientificreports/ The multivariate analytical tools [i.e., random forest (RF), the sparse partial least square discriminant analysis (sPLS-DA), and the non-metric multidimensional scaling (NMDS)] used for the selection of important variables showed different VOCs as discriminants between day and night volatiles of the wild and cultivated tomato plants (Fig. 3a-c). A total of 20 VOCs appeared in the top 10 of the discriminating VOCs highlighted by these methods, and were hence considered as the most discriminating between day and night volatiles of the wild and cultivated tomato plants. These discriminating VOCs included limonene oxide, citronellal, camphor, nonanal, tricyclene, germacrene B, (E)-β-caryophyllene, (Z)-3-hexenol, β-pinene, δ-3-carene, γ-elemene, α-terpinene, trans-isolimonene, δ-3-carene, 3,7,7-trimethyl-1,3,5-cycloheptatriene, α-pinene, β-phellandrene, α-phellandrene, α-humulene, δ-2-carene and p-cymene (Fig. 3a-c). Based on these discriminating VOCs, the multidimensional scaling (MSD), the sPLSDA and NMDS clustered the volatile treatments into four groups: (i) wild tomato plants of which volatiles were trapped during the day (WTDV), (ii) wild tomato plants of which volatiles were trapped at night (WTNV), (iii) cultivated tomato plants of which volatiles were trapped during the day (CTDV), and (iv) cultivated tomato plants of which volatiles were trapped at night (CTNV) (Fig. 3d-f). The classification accuracy of the RF analysis was very high, about 81.25%, and release rates of the most discriminating volatile compounds varied significantly across the volatile treatments (one-way ANOSIM based on Bray-Curtis dissimilarity, p < 0.0001, R = 0.54 for NMDS; and R2X = 0.74, R2Y = 0.76, Q2 = 0.59 for sPLS-DA). The sPLS-DA and NMDS biplots revealed that (Z)-3-hexenol, δ-3-carene, γ-elemene, α-pinene, 3,7,7-trimethyl-1,3,5-cycloheptatriene, germacrene B, and (E)-β-caryophyllene were associated with wild tomato plant day volatiles, while limonene oxide, citronellal, and camphor were associated with the wild tomato plant night volatiles. On the other hand, β-phellandrene, nonanal and tricyclene were associated with cultivated tomato plant day volatiles (Fig. 3e,f). The clustering heatmap and k-means plot (line graph) showed that the most discriminating VOCs are abundant in replicates of volatiles collected during the day from the wild and cultivated tomato plants than in those collected at night, except for limonene oxide, citronellal, camphor which were abundant in the night volatiles of the wild tomato plant; and β-pinene, δ-2-carene, β-phellandrene, and α-terpinene which did not show variation between day and night collections ( Fig. 4a,b).

Discussion
We investigated the infestation rates of the tomato pinworm T. absoluta on wild and cultivated tomato plants in a field experiment, and characterized the volatiles mediating the avoidance behavior of the insect. Our findings indicated that the wild tomato, S. lycopersicum var. cerasiforme, is not preferred by T. absoluta, unlike the cultivated tomato, Solanum lycopersicum L. (var. Rambo F1). This is reflected in the reduced infestation levels recorded in the monocrop of the wild tomato plant compared to intercrops of the wild and cultivated tomato plants, which also had lower infestation levels compared to the monocrop of the cultivated tomato plant. When used as a border crop in the intercrops, the wild tomato plant was effective to reduce infestation by T. absoluta to the level reported in the monocrop of the wild tomato, indicating that it is enough to plant the wild tomato plant (as many as possible) around the cultivated tomato plant. Ghaderi et al. 26 reported different field infestation levels by T. absoluta between monocrops of cultivated tomato cultivars, with the highest infestation level recorded on the cultivar 'Cal JN3' and the lowest on the cultivars 'Early Urbana Y' and 'Super Strain B' . However, by the seventh week, there was a reduction of the larval infestation in the monocrop of cultivated tomato plants. The decrease in infestation could be explained by avoidance of tomato plants previously infested by conspecifics 28,29 , and older tomato plants by the moth 34 , as well as change in the volatile emission between phenological stages of tomato plants 35 . In our study, several factors including visual cues, trichome density, and headspace volatile composition could have contributed to the observed differences in the infestation levels between the wild and cultivated tomato plants. In our experiments, the infestation level was low on the plants, and it is worth replicating this study during a season of high infestation by T. absoluta in the field. Our preliminary behavioral tests in the wind tunnel showed that the cultivated tomato plant (var. Rambo F1) was found to be very attractive to gravid Table 1. Relative mean release rates (± SE) (ng/µL/h) of volatile organic compounds detected in headspace volatiles collected during the day and at night from wild and cultivated tomato plants grown in the field (n = 4). Compounds were identified using retention times (RT), electron ionization spectrum, and retention indices calculated (RI cal) relative to relative to C 8 -C 23 n-alkanes run on an HP-5MS, and those obtained from the literature (RI lit): (A) 30 ; (B) 31 ; (C) 32 ; (D) 33 , as well as comparison of their spectra with the library data and published Kovats retention indices and mass spectra from online NIST library database. Compounds marked with a star (*) are those that were confirmed using available authentic standards run on an HP-5MS column. Significant p-values at α = 0.05 are in bold and means with different letters are significantly different based on Kruskal-Wallis ANOVA used for comparison of at least three means, while Wilcoxon paired signed-rank test was used for comparison of two means. nd = not detected. www.nature.com/scientificreports/ T. absoluta females, whereas the wild tomato plant (var. cerasiforme) was avoided by the moths. These observations are in agreement with the findings of Proffit et al. 27 who reported that volatiles of the cultivated tomato (cv. Santa clara) were very attractive to females of T. absoluta while those of a wild tomato cultivar (S. habrochaites) were avoided by the insects. The "preference performance hypothesis" stipulates that insects will preferentially oviposit on host plants that guarantee the survival of their offsprings 36 . The avoidance of the wild tomato plant by T. absoluta females suggests that the offspring of the insect may not be able to complete development on the wild tomato plant, and it would be interesting to study the development of the larval stages of the pest on the wild tomato plant, and other tomato plants treated with extracts of the wild tomato. Dias et al. 37 reported that T. absoluta females exhibited a reduced oviposition, and poor performance of larvae on the wild tomato S. pennellii, which was found to have high levels of acylsugars, unlike the cultivated tomato S. lycopersicum cv. Redenção' which was attractive and suitable for the moth. Using biochemical and transcriptomic analyses, Chen et al. 38 showed that the low level of infestation by T. absoluta on eggplant compared to tomato plant was attributed to specific signaling genes associated with emission of salicylic acid, and the high level of phenols in eggplant. Allelochemicals play a major role in attracting or deterring gravid T. absoluta females, and in mediating their oviposition on host plants [27][28][29] . Our chemical analyses of the headspace of the tomato plants revealed quantitative and qualitative differences between the tomato plants and the collection times (day and night). Five compounds, namely limonene oxide, (Z)-3-hexenol, citronellal, camphor, and germacrene B were only detected in the volatile extract of the wild tomato (var. cerasiforme) and were not present in the volatile extract of the cultivated tomato (var. Rambo F1). However, Paudel et al. 39 did not find any of these compounds in the volatiles of the wild tomato plant. The differences in the volatile profiles between our study and that of Paudel et al. 39 could be explained by  39 , volatiles were collected at the vegetative stage in the laboratory, whereas in our study volatile collection was done in the field at the onset of flowering and fruiting stage. Among compounds found to be specific to the wild tomato (var. cerasiforme), germacrene B was reported in the wild tomato S. habrochaites 27 , and (Z)-3-hexenol, citronellal and limonene oxide were reported in S. pimpinellifolium 40 which is considered to be a parental line of the wild tomato (var. cerasiforme) 41 . In this study, we found that the tomato plants emitted more volatiles during the day than at night. The monoterpenes tricyclene, α-terpineol and thymol, the ketones (-)-car-3-en-2-one and p-ethyl acetophenone, the aldehyde α-terpinen-7-al, the alcohol p-cymen-7-ol, and the sesquiterpenes (Z)-β-caryophyllene and α-cedrene were detected in the day volatiles of both wild and cultivated tomato plants but these compounds were absent in the night volatile extracts ( Table 1). The differences between day and night volatile compositions could be explained by differences in the physiological responses of the plants when exposed to environmental stresses. In our experiment, the temperature www.nature.com/scientificreports/ at night was 20 ± 3 °C, slightly lower than that of the day, 25 ± 3 °C. It has been reported that tomato plants exhibit a lower degree of transpiration and partial stomata opening during low night temperatures than at high day temperatures 42,43 , and these are among the factors that have been reported to explain higher volatile emission in the day than at night 44 . Irrespective of the collection times, β-phellandrene and δ-2-carene were the most abundant monoterpenes, while (E)-β-caryophyllene and α-humulene were the most abundant sesquiterpenes, which are in agreement with a previous study on volatiles emitted by the wild tomato (var. cerasiforme) 39 , as well as cultivated tomato cultivars like Moneymaker 45 , Semiramis 29 , and Kilele F1 33 . We found that the total emissions of volatiles did not significantly differ between the wild and cultivated tomato plants during the day or at night, but quantitative differences were noted in the emission of some compounds between the tomato genotypes. The emission level of α-thujene was higher in the wild tomato plant than in the cultivated, whereas emission levels of tricyclene and β-pinene were higher in the cultivated than in the wild tomato plant. The use of multivariate tools (machine learning) for the identification of discriminating volatiles between treatments is gaining attention in the field of chemical ecology 46,47 . Several of these variable selection methods including the random forest (RF), the sparse partial least square discriminant analysis (sPLS-DA) and the nonmetric multidimensional scaling (NMDS) have been used in the literature to narrow down important volatiles for use in behavioral assays to determine bioactive compounds 33,48,49 . We found that volatile compounds selected as important for discriminating the volatile extracts of the wild and cultivated tomato plants varied with the multivariate method used (Fig. 3a-c); hence a combination of methods could provide a better discrimination. Similarities in the top 10 selected discriminating compounds were common when using the mean decrease in accuracy of RF and the variable importance in the projection of the sPLSDA, unlike for the NMDS. In our study, the top 10 discriminating VOCs predicted by RF and sPLS-DA were either minor (e.g., (Z)-3-hexenol and β-pinene), absent in some volatile treatments (e.g., limonene oxide, citronellal, camphor and germacrene B unique to the wild tomato plant), or significantly different among the treatments irrespective of the volatile amount (e.g., δ-3-carene and α-humulene). Our observations are consistent with findings reported by Peterson  Table 1, and indicate compounds that elicited consistent antennal responses (marked with *) at least from three antennae from three insects. A total of 16 compounds were EAG-active, and listed as follows: 1 = hexanal; 2 = (Z)-3-hexenol; 9 = α-pinene, 10 = 3,7,7-Trimethyl-1,3,5-cycloheptatriene; 13 = β-myrcene; 15 = α-phellandrene, 19 = β-phellandrene, 22 = (E)-β-ocimene, 26 49 who used RF to highlight compounds that distinguished volatiles of three plant species. On the contrary, we observed that NMDS was tuned to select compounds that are abundant in the volatile treatments, even at levels not significantly different between the treatments. Using NMDS to select distinguishing volatiles between six Lygodium plant species, Wheeler et al. 50 also found that the top 6 discriminating VOCs were the major constituents emitted by the plants. The observed differences in the discriminating compounds selected by the marching learning methods could be explained by differences in assumptions and algorithms that composed each method 51 . When looking at compounds that best discriminate headspace volatile composition and those that could potentially play a role in the behavioral responses of insects, a single variable selection method is unlikely to predict most of the bioactive compounds. We found 16 compounds (i.e., hexanal, (Z)-3-hexenol, α-pinene, 3,7,7-Trimethyl-1,3,5-cycloheptatriene, β-myrcene, α-phellandrene, β-phellandrene, (E)-β-ocimene, terpinolene, methyl salicylate, (E)-β-caryophyllene, germacrene D and cis-carvenone oxide) that were detected by antennae of T. absoluta females. Among these compounds, five (i.e., (Z)-3-hexenol, limonene oxide, camphor, citronellal and (E)-β-caryophyllene) were present in the top 10 discriminating VOCs highlighted by RF or sPLSDA, while other four compounds (α-pinene, 3,7,7-Trimethyl-1,3,5-cycloheptatriene, α-phellandrene and β-phellandrene) were predicted by NMDS. Therefore, a combined use of variable selection methods could provide a better comparison of discrimination between different headspace volatiles and increase the chance of discovering more bioactive compounds, although this would imply testing several compounds, which might also be tedious. Tuta absoluta females are sensitive to small variations in headspace compositions emanating from different tomato varieties and cultivars, as well as other host plants, which led to different levels of attraction or repellency [27][28][29]52 . The availability of several overlapping bioactive compounds in the headspace volatiles of host plants has been associated with the polyphagy nature of the herbivorous pest 53 . Among the EAG-active compounds reported in this study, antennae of T. absoluta females were found to detect synthetic standards of hexanal, (Ζ)-3-hexenol, α-pinene, α-phellandrene, β-myrcene and methyl salicylate 29 and be attracted to a blend of five compounds among them β-ocimene, and (E)-β-caryophyllene 53 . To the best of our knowledge, camphor, citronellal, limonene oxide, β-phellandrene, terpinolene, (E)-β-ocimene, (E)-β-caryophyllene, and others tentatively identified as 3,7,7-Trimethyl-1,3,5-cycloheptatriene, germacrene D and cis-carvenone oxide were not reported as EAG-active compounds for T. absoluta. In future research, it is important to run a mixture of synthetic standards of these compounds in GC-EAD recordings to confirm their detection by the insect's antennae. Although the wild and cultivated tomato plants had opposite effects on the responses of T. absoluta females, most of the EAG-active compounds were commonly emitted by both plants. Hence, it is likely that the EAG-active compounds that mediated the observed avoidance behavior in T. absoluta females to the wild tomato were those which were present only in volatiles of the wild tomato. Interestingly, we found that the 4-component blend of the EAG-active compounds [(Z)-3-hexenol, camphor, citronellal and limonene oxide] which were uniquely emitted by the wild tomato plant elicited avoidance behavior in T. absoluta females in a dose-dependent way when compared to control solvent. Moreover, the blend repelled the insects when compared to volatiles of the wild or cultivated tomato plant. Tuta absoluta is a nocturnal moth, hence the insect could be more sensitive to volatiles emitted at night. Except for (Z)-3-hexenol which was emitted at a higher level during the day than at night, the other three compounds of the blend were released in higher amounts at night. Steen et al. 54 also reported that the optimal foraging activity by the nocturnal moth, the pine hawkmoth Sphinx pinastri L. (Lepidopera: Sphingidae) www.nature.com/scientificreports/ to Platanthera chlorantha plant coincided to the period of increase in the night emission of (E)-β-ocimene and (Z)-β-ocimene. Natural enemies and insect pests may detect the same VOCs from host plants, which can turn out to be of advantage or disadvantage for the control of insect pests when using allelochemicals. Among the compounds that are detectable by antennae of T. absoluta, some T. absoluta parasitoids like Trichogramma cordubense Vagas & Cabello and Trichogramma achaeae Nagaraja & Nagarkatti (Hymenoptera: Trichogrammatidae) were reported to detect β-myrcene and (Z)-3-hexenol 55 , and Trichogramma chilonis Ishii (Hymenoptera: Trichogrammatidae) was found to detect hexanal, citronellal, (Z)-3-hexenol, α-pinene, α-phellandrene and (E)-β-caryophyllene in EAG recordings 56 . Moreover, De-Backer et al. 57 reported that antennae of the predator Macrolophus pygmaeus (Rambur) (Hemiptera: Miridae) detect hexanal, α-pinene, and β-phellandrene. In olfactometer bioassays, other natural enemies of T. absoluta such as the parasitoid Dolichogenidea gelechiidivoris (March) (Hymenoptera: Braconidae) were reported to be attracted to α-pinene, α-phellandrene, β-ocimene, methyl salicylate and (E)-βcaryophyllene 33 , and the predator Nesidiocoris tenuis (Reuter) (Hemiptera: Miridae) to α-pinene, α-phellandrene, β-ocimene and δ-3-carene 58 , and to (Z)-3-hexenol and methyl salicylate 59 . Furthermore, some of the compounds attractive to the natural enemies were emitted in high amounts by the wild tomato plant, suggesting that the use of this tomato in intercropping systems could play a double beneficial role in the control of T. absoluta, through the attraction of the natural enemies and deterrence of oviposition by the moth. The possible effect of the wild tomato plant, and the 4-component blend (mixture of (Z)-3-hexenol, camphor, citronellal and limonene oxide) repellent to T. absoluta with the inclusion of other EAG-active compounds, on the behavioral responses of these natural enemies is worth evaluating in future research.
In summary, our findings show that the level of infestation by T. absoluta was lower on the wild tomato S. lycopersicum var. cerasiforme compared to that on the cultivated tomato S. lycopersicum (var. Rambo F1). Chemical analyses reveal both qualitative and quantitative differences in the volatile compositions of the wild and cultivated tomato plants. A total of 16 compounds were detected by antennae of T. absoluta females. Among these EAG-active compounds, the 4-component blend of (Z)-3-hexenol, camphor, citronellal, and limonene oxide identified only in volatiles of the wild tomato plant showed dose-dependence repellency to T. absoluta females. The present study lays down some groundwork for the development of a 'push-pull' strategy to control T. absoluta through intercrops of wild tomato with cultivated tomato plants, and deployment of traps or agronets baited with optimized controlled-release of the repellent blend.

Materials and methods
Wild and cultivated tomato plants. Seeds  This wild tomato is commonly known as the wild form of cherry tomato, and is considered to be an evolutionary intermediate between the domesticated or cultivated tomato S. lycopersicum and its wild ancestor, S. pimpinellifolium 41,60 . Seedlings of the two tomato genotypes were grown in nurseries using a mixture of soil and manure (goat and chicken dungs) in the ratio of 3:1, at Wang'uru in Mwea planes (S 00′4137.4″; E 037°22′12.3″), Kirinyaga County, Kenya. After four weeks in the nurseries, seedlings of each plant were transplanted on the mixture of soil and manure in the open field with the addition of 15 g of diammonium phosphate (DAP) at planting. These plants were used to assess the effects of monocropping and intercropping of the wild and cultivated tomato plants on infestation levels of T. absoluta. Other seedlings of each tomato were transplanted in a screened nethouse to protect them from herbivorous pests, and these plants were thereafter used for volatile collection in the field. Watering was done twice per week. After 3 weeks, a top dressing with 15 g of calcium ammonium nitrate was done for each plant.
Tuta absoluta colony. Tuta absoluta adults were obtained from the insectary at icipe (S01°13.140′; E036°53.440′), Nairobi, Kenya, which was established in early 2015 and supplemented regularly with wild adults that emerged from larvae-infested tomato plants collected from open field plantations. Adult males and females of T. absoluta were placed in Perspex cages (65 cm × 45 cm × 45 cm) and fed with a mixture of 80% honey and water. The colony was maintained under controlled laboratory conditions at a temperature range of 25 ± 2 °C and 70 ± 5% RH and a cycle photoperiod of 12:12 h L:D. Potted plants of the cultivated tomato were exposed to T. absoluta adults for 3 days for oviposition, and then the plants with eggs were transferred to other cages where the eggs hatched. The larvae which developed were offered additional plants to ensure better development through to the pupation stage. Mated T. absoluta females of 3-days old were used in the bioassays. To ensure that gravid females were used in the behavioral assays, couples that were mating in the rearing cage were aspirated and put into individual cages (10 cm × 10 cm × 6 cm). Thereafter, females were separated from males based on differences in their body morphology (i.e., females are larger with a broader abdomen tip than the males who have a slender abdomen tip).  61 . The monoterpene α-pinene and the sesquiterpene α-humulene were run at different concentrations and the linear equations were generated from the calibration curves ( Supplementary Fig. S5). Thereafter, the means of the release rates (ng/µL/h) of the compounds were computed.

Gas chromatography-linked electroantennographic detection (GC-EAD). Hewlett-Packard (HP)
5890 Series II gas chromatograph equipped with an HP-5MS column (30 m × 0.25 mm i.d. × 0.25 µm film thickness, Agilent Technologies, Palo Alto, California, USA) was used with high purity nitrogen as the carrier gas at a flow rate of 1 mL/min flow. Four microlitres (4 µL) aliquot of each volatile extract was injected into the entry port of the GC-FID at an injection temperature of 280 °C and a split valve delay of 5 min. The oven temperature was held at 35 °C for 3 min, then programmed to increase at 10 °C/min to reach 280 °C and maintained at this temperature for 10 min. The column effluent was split 1:1 for simultaneous detection by the flame ionization detector (FID) and electroantennographic detection (EAD). For EAD, silver-coated wires in drawn-out glass capillaries (1.5 mm i.d.) were filled with Ringer saline solution (7.5 g NaCl, 0.7 g KCl, 0.2 g CaCl 2, and 0.2 g MgCl 2 dissolved in 1L of distilled water). Antennal preparations were made by putting the gravid T. absoluta female in a glass vial which was placed on ice for 10 min to immobilize the insect, after which the base of the head and distal end of the antennae were cut off with a scalpel. The base of the head was then connected to the reference electrode www.nature.com/scientificreports/ while the tip of one antenna was connected to the recording electrode. The analog signal was detected through a probe (INR-II, Syntech, Hilversum, the Netherlands), captured and processed with a data acquisition controller (IDAC-4, Syntech, the Netherlands), and later analyzed with software (EAG 2000, Syntech). Each plant volatile extract (i.e., cultivated tomato plant day volatiles, cultivated tomato plant night volatiles, wild tomato plant day volatiles, and wild tomato plant night volatiles) was analyzed 6 to 10 times using a fresh female antenna at each run, making a total of between 24 and 40 antennae tested in the GC-EAD recordings. The EAG recordings were checked, and 4 replicates were selected per each headspace volatile, then every EAG response that matched with a compound was screened, and only compounds that elicited three consistent EAG responses, i.e., compounds detected by at least 3 antennae, were considered as EAG-active compounds used in the results.
Behavioural responses of gravid Tuta absoluta females to synthetics of electrophysiologically-active compounds. The responses of mated T. absoluta females to a blend of synthetic compounds were observed in a dual-choice cuboidal, Perspex glass wind-tunnel (150 cm × 21 cm × 21 cm) ( Supplementary  Fig. S6). Each end of the wind tunnel was connected to a 250 mL quick-fit glass jar (Sigma Scientific, Gainesville, FL, USA) serving as an odor source container, using flexible PTFE tubes ( Supplementary Fig. S6). Ambient air was sucked using a vacuum pump (Analytical Research system Inc., Gainesville, FL, USA), then passed through an activated charcoal column filter and a water chamber respectively for filtering and moisturization. The airflow rate was set at 350 mL/min in each arm of the wind tunnel using a flowmeter (AALBORG, Orangeburg, NY, USA), and was pulled from the bioassay set-up at a rate of 700 mL/min through the centre of the tunnel. A red fluorescent tube was suspended 1.5 m above the tunnel giving about 1000 lx incident red light under which behavioral observations were performed to prevent the insects from using visual cues when making choices. The bottom of the tunnel was marked every 5 cm from the centre to the two ends to facilitate the measurement of the insects' flight distances. Among the consistent EAG-active compounds which were unique to the wild tomato, an original blend Bo made of (Z)-3-hexenol, limonene oxide, camphor, and citronellal was composed using a tenfold of the compound release rates mixed in ratio found in the volatile emission of the wild tomato (Table 1). Specifically, the blend Bo contained 1.2 ng/µL (Z)-3-hexenol, 1.2 ng/µL limonene oxide, 1.4 ng/µL citronellal and 2 ng/µL camphor. Thereafter, the concentration of the blend Bo was halved (Bo/2) and doubled (2Bo) to make two additional blends. The blends were prepared in hexane used as the solvent. An aliquot of 200 µL of the blends (treatments) or the solvent (control) was applied onto a 1 cm long Luna dental roll (Roeko ® , Langenau, Germany). Hence, the dose tested for the original blend Bo contained 240 ng (Z)-3-hexenol, 240 ng limonene oxide, 280 ng citronellal and 400 ng camphor. The following dual choices were tested: (i) pure hexane versus pure hexane (control test), (ii) blend Bo dose versus the control (pure hexane), (iii) half of blend Bo dose (Bo/2) versus the control, (iv) double of Bo dose (2Bo) versus the control, (v) blend Bo dose versus volatiles of the cultivated tomato, and (vi) blend Bo dose versus volatiles of the wild tomato. The impregnated dental rolls were left for 20 min at room temperature to enable evaporation of the solvent after which the odor sources (i.e., the dental roll impregnated with solvent or volatile blend, or the healthy plant) were immediately introduced into the odor containers (i.e., 250 mL quick-fit jar for the volatile blend, and a cuboidal Plexiglass cage (61 cm × 35 cm × 35 cm) for the plant). Ten minutes later, a group of ten (10) mated T. absoluta females were released through a hole made at the centre of the wind tunnel. The released insects were observed for 20 min, and the numbers of responsive individuals were recorded. Tuta absoluta females that landed beyond 35 cm of either side away from the centre were considered to have responded to the corresponding odor or the control; whereas insects that remained between the release point and 35 cm of either side were considered as non-respondents ( Supplementary Fig. S6) which were thereafter not included in the data analysis. After each run of ten insects, air flow was allowed to pass through the wind tunnel set-up for 1 h to remove the previous test odor, then odor sources and containers were renewed, and locations of the containers switched between left and right of the wind tunnel. Each assay was replicated five times on 5 days, with a group of ten insects tested per day per choice test, making a total of 50 insects tested per choice test. The odor containers were cleaned with hot water and Teepol odourless detergent, rinsed with distilled water, then dried in the oven at 150 °C overnight before they were used again on another day. The bioassays were conducted at 23 ± 2 °C and 65 ± 5% RH and between 0500 to 1100 h AM as preliminary observations in the laboratory had indicated this period as the optimal activity time for the insect.

Statistical analyses.
All data analyzed in this study were collected between 2015-2016. All statistical analyses were performed using R statistical software, version 4.0.2 62 . The data on infestation levels and volatile release rates were first subjected to test for normality using Shapiro-Wilk's test, and homogeneity of variance using Bartlett's test. The data on infestation levels were auto-scaled using log 10 χ + 1 and then subjected to a oneway analysis of variance (ANOVA) to compare the mean numbers of mines and larvae per week per leaf, which was followed by Student Neuman Keuls (SNK) test to separate means when a significant difference was noted. www.nature.com/scientificreports/ Thereafter, the data were averaged per treatment level and ANOVA was performed to find the effects of the treatments on the numbers of mines and larvae throughout the experimental period. The data on volatile release rates were analyzed using a non-parametric Kruskal-Wallis ANOVA to compare amounts of VOCs between day and night volatiles of the cultivated and wild tomato plants, and when a significant difference was noted a post hoc Dunn's test with Bonferroni's adjustment was applied for mean separation 63 . Based on the release rates, the VOCs that best distinguished the wild tomato from the cultivated tomato were screened using a combination of machine learning variable selection tools, e.g., the mean decrease in accuracy (MDA) of the random forest (RF) analysis 64 , the variable importance in the projection (VIP) of the sparse partial least square discriminant analysis (sPLS-DA) 65 , and the similarity percentage (SIMPER) of the non-metric multidimensional scaling (NMDS) 66 .
Compounds that appeared in the top 10 of the VOCs highlighted by these three variable selection methods were considered as the most discriminating VOCs between day and night volatiles of the wild and cultivated tomato plants. Using the release rates of the most discriminating VOCs, biplots of sPLS-DA and NMDS were performed, respectively, in the mixOmics package and Past Software, to illustrate the correlation between volatile compounds and the tomato plants 65 . A multidimensional scaling (MDS) plot was also performed to visualize the similarity among volatiles of the plants using the function "MDSplot" of the RF package 67 . A clustering heatmap was done using the function "cim" in the mixOmics package 68 to illustrate variations in the discriminating VOCs across replicates of day and night volatile extracts of the tomato plants. The out-of-bag (OOB) error of the RF analysis was used to appreciate the classification accuracy (100%-OOB error) by RF 64 . The sPLS-DA model was validated using the function "perf " and the parameters (Q2, R2X, R2Y), as well as the "leave-one-group-out" cross-validation method in the mixOmics package 68 . One-way analysis of similarities (ANOSIM) associated with the Bray-Curtis dissimilarity index was performed to appreciate the validity of the NMDS model 66 . For the behavioral assays, the frequencies of choice made by T. absoluta females in the wind tunnel were compared between the odors using a Chi-square (χ 2 ) goodness of fit test.