Greenleaf desmodium, Desmodium intortum, silverleaf desmodium D. uncinatum, Brachiaria cv. Mulato II, and maize, Zea mays, hybrid “SC Duma 43” (Seed Co, Kenya), plants were grown individually in pots filled with fertilized soil (3.5 g Diammonium phosphate-DAP, Elgon Kenya) in an insect-proof screen house at icipe-Thomas Odhiambo Campus (ITOC), Mbita Point (0°25′S, 34°12′E; c. 1,200 m above sea level), western Kenya. All plants were grown under natural conditions (25°C; 65% RH; 12L: 12D). Maize plants used in the experiments were 2–3 weeks old.

Spodoptera frugiperda, were reared on a standard diet for multiple lepidopteran species (Southland Products Inc., Lake Village, Arkansas, United States) under controlled conditions (25 ± 2°C, 70 ± 5% RH, 12L: 12D) using methods described by Mihm (1983). The culture was initiated with field-collected insects. Field-collected insects were added on a regular basis to refresh the mass-reared culture and maintain its original behavioral characteristics (Chambers, 1977). Larval endoparasitoids Cotesia icipe and Coccygidium luteum were reared on second instar larvae of S. frugiperda. Parasitoid cultures were maintained at 24 ± 3°C, 70 ± 5% RH, 12L: 12D.

Chemical standards used for GC quantification and comparison of GC retention times were (E)-2-hexenal, 1-octen-3-ol, 3-octanone, (Z)-3-hexenyl acetate, β-ocimene, (S)-linalool, (E,E)-allo-ocimene, methyl salicylate (MeSA), indole, α-copaene, β-caryophyllene, and (E)-β-farnesene. All standards were purchased from Sigma Aldrich (Gillingham, United Kingdom), except α-copaene which was purchased from Cayman Chemical Ltd. (Cambridge, United Kingdom). The homoterpenes (E)-4,8-dimethyl-1,3,7-non-atriene (DMNT) (>98%) and (E,E)-4,8,12-trimethyl-1,3,7,11-tridecatetraene (TMTT) (>98%) were synthesized from geraniol and (E,E)-farnesol, respectively, by oxidation to their corresponding aldehydes followed by Wittig methylenation (Leopold, 1990).

Volatile compounds from the different plant species were collected by headspace sampling (Agelopoulos et al., 1999; Tamiru et al., 2011) for use in subsequent bioassays and electrophysiological and chemical analyses. Volatiles were collected for 48 h. Leaves of plants were enclosed in polyethyleneterephthalate (PET) bags (volume 3.2 L) heated to 150°C before use and fitted with Swagelock inlet and outlet ports. Charcoal-filtered air was pumped (600 ml min^–1) through the inlet port. Volatiles were collected on Porapak Q (50 mg, 50/80 mesh; Supelco, Bellefonte, PA, United States) filters inserted in the outlet port through which air was drawn at 400 ml min^–1. Porapak Q filters were conditioned before use by washing with dichloromethane. All connections were made with polytetrafluoroethylene (PTFE) tubing (Alltech Associates, Lancashire, United Kingdom) with brass ferrules and fittings and sealed with PTFE tape. After entrainment, volatiles were eluted with 0.75 ml dichloromethane (EMSURE; Merck KGaA, Darmstadt, Germany) and stored in vials (Agilent Technologies, Stockport, United Kingdom; 2 ml with solid cap, 9 mm, w/PTFE/RS liner) at −80°C until required for chemical analysis and/or bioassay.

Behavioral responses of mated female S. frugiperda to plant volatiles were studied in a wind tunnel (25 ± 2°C; RH:70 ± 5%). The wind tunnel was made from glass (7 mm thickness) on an aluminum frame, with a design, including odor source tubes, as shown in Jaramillo et al. (2013) but with larger dimensions (122 × 32 × 32 cm). Plants were used to provide volatile stimuli in bioassays but were positioned outside the wind tunnel and not at the upwind end, in order to avoid any visual cues. Plant headspace volatiles were obtained by pumping charcoal filtered air into the headspace of 3-week-old experimental plants (4–6 leaves) covered with polyethyleneterephthalate (PET) bags (volume 3.2 L; heated to 150°C before use). Plants used were greenleaf desmodium, D. intortum (GL), silverleaf desmodium, D. uncinatum (SL), and Brachiaria cv. Mulato II (BR). These were tested alone in Series 1 and in combination with maize volatiles in Series 2. Naturally emitted volatiles from experimental plants were introduced at the upwind end of the wind tunnel, through an air inlet at 300 ml min^–1. Two-day old mated female S. frugiperda moths were transferred from the insect mass rearing unit using a small Perspex cage covered with black cloth and kept in the bioassay room for 1 h before commencing the experiments for acclimatization.

Moths were carefully introduced individually through a side-panel at the downwind end of the wind tunnel, 100 cm away from the odor source tube at the upwind end. Distance flown upwind and any close approaches to the odor source were recorded during each 5 min bioassay period. It was not possible to record source contacts because the odor release point was not suitable for landing on. Bioassays were conducted at scotophase with reduced lighting provided by two bulbs (40 W, fitted with red filter) positioned at 30 cm above the wind-tunnel and angled to provide even coverage of the whole tunnel. The air velocity in the wind tunnel was 30 cm s^–1. Air from the wind tunnel was continuously blown out of the bioassay room through rear air exhaust tube aided by a suction pump (50/60 Hz, 10A). Twenty replicates of individual female moths were used for each treatment and each female was tested in the wind tunnel once only.

The above wind tunnel bioassay procedure was also used to test moth responses to synthetic standards of plant volatiles that showed electrophysiological activity with female of S. frugiperda in GC-EAG recordings (described below). Blends of these synthetic plant volatiles were formulated at the same concentration and ratio as in natural headspace samples of the companion and maize plants. Authentic standards were obtained as liquid, weighed out and diluted in hexane to obtain the required concentration. Different volumes of higher concentration stock solution (1 mg/ml or 100 ng/μl) were mixed together to get the required final amounts. Details of concentrations and ratios of blend components are provided in Supplementary Table 1. In each replicate (n = 20) of the bioassay, 10 μl of test solution was released from a piece of filter paper.

To assess oviposition behavior of S. frugiperda on plants, an oviposition bioassay was carried out under greenhouse conditions (25 ± 2°C, 65 ± 5% RH, 12L:12D) in a no-choice bioassay following modifications of methods described by Khan et al. (2007). Six different plant treatments were tested: (1) Brachiaria cv. Mulato II (BR), (2) Greenleaf Desmodium, D. intortum (GL), (3) Silverleaf Desmodium, D. uncinatum (SL), (4) Maize (cv. “SC Duma 43”) as a control. (5) Maize + GL. (6) Maize + SL. Given that Brachiaria plants are normally planted as a border crop in Push-Pull fields, no combined treatment of Brachiaria and maize was included in this assay. All plants were grown in pots (4 L) with standard ITOC agricultural soil, under the same conditions mentioned above. For the combined treatments (5 and 6), 50 Desmodium seeds were sown first, then maize (one seed per pot) was sown in the same pot 15 days later. This was because Desmodium grows more slowly than maize. Maize plants when used for the experiment were 12–15 days-old and had two fully developed leaves. In all cases, plants were grown and kept in separate screenhouse compartments before bioassay, to minimize unwanted induction of plant defense or “priming” effects by plant–plant volatile interactions between treatments (Sobhy et al., 2020).

Firstly, to investigate moth oviposition behavior under realistic conditions, with access to plants, oviposition cage experiments were carried out. Potted plants with the six different plant treatments (described above) were placed at the center of oviposition cages (80 × 80 × 100 cm) constructed of fine mesh attached to a wooden frame. Thereafter, five gravid S. frugiperda females (2–3 days old) were introduced into each cage before dusk and allowed to oviposit overnight under natural conditions of L12:D12. The numbers of S. frugiperda eggs were then counted on two oviposition sites: eggs deposited on plants and eggs deposited on the cage. For Maize + GL and Maize + SL treatments, eggs deposited on both maize and desmodium plants were counted. Ten replicates were conducted for each treatment. All cages were supplied with a 10-cm diameter wad of cotton wool moistened with water for moths to drink ad libitum.

Secondly, to investigate responses to volatiles, moth oviposition behavior was observed in compartments where access to plants was prevented but moths could still detect volatiles from the plants. This experiment was designed to test if eggs were laid further away from the odor source when exposed to companion plant volatiles. Compartments were made from polyethylene (PET) drink bottles with the bottom and the top cut off (20 cm height × 20 cm diameter), as described in Erb et al. (2011), and were inserted into the soil at the edge of each pot (Supplementary Figure 1). To ensure air circulation and prevent any direct contact between moths and plants, the upper openings of the bottles were covered with a sheet of mesh cloth which was held in place with a rubber band around the cylinder of the bottle. Another polyethylene (PET) cylinder (20 cm, height × 20 cm, diameter) was then placed on the top of the first cylinder with potted plants. Then five gravid S. frugiperda females (2–3 days old) were released for oviposition in the top cylinder. The upper opening of the oviposition unit was closed with a sheet of mesh cloth, identical to the bottom layer and held in place with a rubber band after releasing the moths (Supplementary Figure 1). The whole setup was covered with a dark cloth and moths were allowed to oviposit overnight (15 h). Subsequently, the number of S. frugiperda eggs was counted on two oviposition areas: Top = upper half sides of the bottle and the top mesh; and Bottom = lower half sides of the bottle as well as the bottom mesh, using a dissecting microscope. Fifteen replicates were conducted for each treatment.

Behavioral responses of larval endoparasitoid wasps, C. luteum and C. icipe to plant volatiles were tested in a Perspex four-arm olfactometer (Webster et al., 2010; Tamiru et al., 2011), in a controlled environment room (25 ± 2°C, 70 ± 5% RH) between 9:00 a.m. and 5:00 p.m. The olfactometer was placed on a table that was homogeneously illuminated by fluorescent tubes. Air was drawn through the four arms to exit through a hole at the center at 260 ml min^–1. Headspace samples (10 μl aliquots) were applied, using a micropipette (Drummond “microcap,” Drummond Scientific Co., Broomall, PA, United States), to a piece of filter paper (4 × 25 mm; Whatman Filter Paper, Maidstone, United Kingdom). The solvent was allowed to evaporate for 30 s then paper strips were placed in an inlet port at the entrance to each olfactometer arm. Headspace samples were in dichloromethane. Prior to assay, mated female parasitoids (2–3 day-old) were provided contact experience with maize leaves damaged by S. frugiperda larvae (Turlings et al., 1989). Subsequently, experienced females were transferred individually into the central chamber of the olfactometer using a custom-made piece of glass tubing. Time spent in each olfactometer arm was recorded with Olfa software (F. Nazzi, Udine, Italy) for 12 min and after every 3 min the position of the olfactometer was rotated clockwise by 90° to eliminate bias. The experiments were replicated at least 12 times. A choice test was carried out to compare insect responses to plant volatile headspace samples vs. solvent control. One arm had the plant volatiles, whereas the other three control arms were treated similarly with the same volume of solvent (dichloromethane) on filter paper strips.

Olfactometer bioassays were also used to test insect responses to synthetic standards of plant volatiles that showed electrophysiological activity with C. luteum and C. icipe females in GC-EAG recordings (described below). Blends of these synthetic plant volatiles were formulated at the same concentration and ratio as in the headspace samples of the companion and maize plants. Details of concentrations and ratios of blend components are provided in Supplementary Table 1. Similar to the headspace samples, 10 μl of each synthetic formulated blend were applied using a micropipette to a piece of filter paper.

Analysis of plant volatiles, to determine which compounds could be detected by insects, was carried out using a Gas Chromatography Coupled Electroantennogram (GC-EAG) system, in which the effluent from a gas chromatograph (GC) column was simultaneously directed to the antennal preparation and the GC detector, as described by Wadhams (1990). Compounds were separated with a 7,820 GC (Agilent Technologies) fitted with flame ionization detector (FID) and a non-polar HP5-MS capillary column (30 mm × 0.25 mm × 0.25 μm film thickness; Agilent Technologies), which used hydrogen carrier gas at a constant flow rate of 1.2 ml/min. Manual injections of 1 μl were made in splitless mode (285°C) with the oven temperature maintained at 35°C for 5 min, and then programmed to increase at 10°C min^–1 to 285°C, run time 45 min. The column effluent was split using a salinized glass Y-tube connector (Syntech, Kirchzarten, Germany). One arm of this connector was connected with fused silica tubing (50 cm × 0.32 mm i.d.) to the FID (250°C) and the other to an equal length of deactivated silica tubing passing through a heated (250°C) transfer line (Syntech) into a glass tube (4 mm i.d.) through which air passed (15 cm s^–1) over the EAG preparation (Meza et al., 2020).

Electroantennogram recordings were made using an IDAC-2 acquisition controller (Syntech) connected to an antennal preparation as the second detector of the GC. Glass electrodes containing electrolyte solution (0.1 M potassium chloride) were attached to silver wires held in micromanipulators (Syntech). Female adult S. frugiperda, C. icipe, and C. luteum antennae (2–4 days old) were prepared for GC-EAG analysis by excising the head after being chilled in ice for 5 min, and the tips of antennae were removed to ensure a good contact. The reference electrode was either inserted into the back of the head or the antenna was excised and proximal segments of the antenna were mounted to the reference electrode, depending on the species of insect. Then, the circuit was completed by bringing the recording electrode into contact with the distal tip of one antenna (Meza et al., 2020). Both the FID and EAG signals were monitored simultaneously after passing through a high impedance amplifier and analyzed with Syntech software (v4.6.1). For each sample, at least 5 replicate antennal preparations were used for GC-EAG recordings (Sobhy et al., 2017). Thereafter, volatile compounds that stimulated consistent responses in at least three different antennae preparations were considered bioactive and replicable.

To identify bioactive compounds that elicited responses in GC-EAG recordings, volatile samples were analyzed using an Agilent 7820A GC coupled to a mass spectrometer (5977B single quad mass selective detector, Agilent Technologies). The GC was fitted with a non-polar HP5-MS capillary column (30 mm length × 0.25 mm inner diam. × 0.25 μm film thickness) coated with (5%-phenyl)-methylpolysiloxane (Agilent Technologies). Hydrogen was used as carrier gas at a constant flow rate of 1.2 ml min^–1. Automated injections of 1 μl were made using a G4513A autosampler (Agilent Technologies) in splitless mode (285°C), with oven temperature maintained at 35°C for 5 min then programmed at 10°C min^–1 to 285°C. Ionization was by electron impact at 70 eV, 250°C (source temperature). Identifications of volatiles were made by comparison of spectra with mass spectral databases (NIST, 2017) and retention index relative to retention times of n-alkanes, and co-chromatography with authentic compounds. Subsequently, authentic standards of bioactive compounds were co-injected with the natural samples of plant volatiles to confirm their tentative identification by peak enhancement (Pickett, 1990). The amounts (in nanograms) of identified volatiles present in collected headspace samples were quantified using a single point external standard quantification method using an authentic sample of 100 ng/μl (E,E)-allo-ocimene (Skelton et al., 2010). HP Chemstation software was used for data analyses.

To obtain data on current performance of Push-Pull companion cropping, extensive sampling of S. frugiperda infestation and damage levels, in farmers’ fields, was carried out in six different counties (Migori, Homa Bay, Siaya, Kisumu, Vihiga, and Bungoma) in the region around icipe-Mbita (0°40′ to 0°58S) in Western Kenya. Nine districts were sampled in the short rains season 2019 and 12 districts were sampled in the long rains season 2019. In each district, 30 smallholder farmers were randomly selected from the list of farmers practicing Push-Pull companion cropping with an adjacent maize monocrop, for data collection. In total 9 × 30 = 270 farms were sampled in the short rains 2019 and 12 × 30 = 360 farms were sampled in the long rains 2019.

The treatments were laid out in two plots at each farm sampled: Push-Pull companion cropping and monocropped maize for comparison of S. frugiperda infestation levels in the two farming systems. The plot sizes ranged from 20 m × 20 m to 30 × 30 m for each treatment. Maize was planted, according to a standard procedure, with 75 cm between rows and a 30 cm spacing between plants in a row, thinned to one plant per planting hole. In the Push-Pull plot, maize was intercropped with greenleaf desmodium (D. intortum; one row between each maize row) and Brachiaria cv. Mulato II was planted, as a border crop, around the field with a 50 cm spacing between plants. Phosphorus fertilizer (di-ammonium phosphate) was applied at planting at the rate of 60 kg ha^–1 in each plot and top-dressed with nitrogen fertilizer (calcium ammonium nitrate) at 4 weeks after emergence at the rate of 60 kg ha^–1 in both Push-Pull and maize monocrop plots. At each farm where data were collected, plot sizes were the same for Push-Pull and maize monocrop plots. There was a separation distance of at least 10 m between Push-Pull and maize monocrop plots.

Five sampling cells measuring 3 × 3 m were randomly marked in each plot. In each cell, 20 randomly selected plants were observed individually for presence/absence of S. frugiperda damage symptoms in leaves, whorl, tassel, and cobs. Thus, a total of 100 plants was inspected in each plot for S. frugiperda infestation. Data were collected 10–11 weeks after emergence of maize. Furthermore, percentage parasitism was assessed in larvae collected from Push-Pull and monocrop maize fields in the Homa Bay (0.6221°S, 34.3310°E), Migori (0.9366°S, 34.4198°E), Siaya (0.0917°S, 34.7680°E), and Kisumu (0.0617°S, 34.2422°E) districts. Spodoptera frugiperda larvae (2nd–4th instar) were collected from the whorl of maize plants grown in Push-Pull and maize monocrop plots. Larvae (60–122 per treatment) were then kept individually in small glass tubes (10 × 2.5 cm) and provided with artificial diet (Southland Products Inc.) and reared until pupation or parasitoid emergence. The latter was monitored daily and recorded.

Prior to analysis, data were examined for normality with the Shapiro–Wilk test, and homogeneity of variances was tested by the Levene test before analysis. Wind tunnel bioassay data were analyzed using univariate analysis of variance (ANOVA) to investigate whether the flight behavior of S. frugiperda females differed when exposed to volatiles from companion plants and maize. For the first oviposition bioassay, two-way ANOVA was used to evaluate the effects of plant treatments [Brachiaria cv. Mulato II (BR); greenleaf Desmodium, D. intortum (GL); silverleaf Desmodium, D. uncinatum (SL); Maize (cv. “SC Duma 43”); Maize + GL; Maize + SL], oviposition sites (plants and cage) and their interaction of the number of S. frugiperda eggs laid in the oviposition cage experiment. Two-way ANOVA was also used to analyze the second oviposition bioassay (in compartments). The factors (plant treatments -described above) and oviposition sites (Top vs. Bottom) and their interactions were evaluated. In both wind tunnel and oviposition bioassays, comparisons among means were performed using the Holm-Sadek method. Oviposition data were Log transformed to meet assumptions of normality and homogeneity of variances. Parasitoid responses (attraction) in the olfactometer choice test were recorded using OLFA software (F. Nazzi, Udine, Italy), which provides a summary of time spent by the parasitic wasps in each of the four olfactometer arms. Bioassay data (time spent) were analyzed by one-way ANOVA using general linear model (GLM) after converting the data into proportions to address dependence of visiting time by parasitoids within the olfactometer fields, and log-ratio transformation to account for the compositional nature of the proportions (Aitchison, 1982; Tamiru et al., 2011). Comparisons of means were performed using the Student–Newman–Keuls (SNK) post hoc test (P < 0.05).

To visualize the overall differences in volatile profiles emitted from Desmodium spp., Brachiaria, and maize plants, a principal component analysis (PCA) was performed using the concentrations of the detected volatiles as dependent variables. Loading and score plots were derived after mean-centering and cube root transformation of volatile data. Average linkage hierarchical clustering based on Ward clustering algorithm of the Euclidean distance measure for the differentially emitted volatiles detected by ANOVA (P < 0.05) was used to construct a heatmap displaying the concentrations of different volatiles. Visualization together with hierarchical clustering of volatile data was done using the comprehensive online tool suite MetaboAnalyst 4.0 (Chong et al., 2018). A multivariate analysis of variance (MANOVA) was performed with plant species (i.e., Desmodium spp., Brachiaria, and maize) as a fixed factor and the concentrations of each volatile compound as dependent variables using SPSS software (IBM SPSS Statistics version 22.0, Armonk, NY, United States). Subsequently, univariate analyses of variances were performed to investigate whether the concentrations of the individual volatile compounds differed between treatments. Comparisons among means were performed using the Holm-Sidak method.

For the field survey, two-way ANOVA was used to compare the effects of sampling locations (different districts) and farming system (Push-Pull and monocrop maize) and their interaction on S. frugiperda infestation (larvae per 100 plants), damage level, and parasitism rate. All univariate analyses were performed with SIGMAPLOT 12.3 (Systat Software Inc., San Jose, CA, United States), except parasitoid responses which were analyzed using SAS version 9.2 (SAS Institute Inc, 2008).

To test the hypothesis that companion plants have repellent activity, the flight behavior of mated female S. frugiperda moths was observed when exposed to volatiles from maize and companion plants (alone and in combination). In the first series of bioassays (Figure 2A), with volatiles from single plant species, maize volatiles were significantly more attractive than volatiles of any of the three companion crops tested, greenleaf desmodium, D. intortum (GL), silverleaf desmodium, D. uncinatum (SL), and Brachiaria cv. Mulato II (BR). Females of S. frugiperda flew significantly further up the upwind from the release point and flew closer to the odor source with maize than to volatiles of GL, SL, or BR [F_{(3, 76)} = 7.916; P < 0.001].

In the second series of bioassays, when maize volatiles were presented in combination with volatiles from Push-Pull companion plants (desmodium species or Brachiaria) a significant reduction in upwind flight toward maize plants was observed [F_{(3, 73)} = 9.009; P < 0.001; Figure 2B].

In oviposition cage experiments where S. frugiperda females have the chance to contact and lay eggs on plants, there were significant differences between plant treatments in the total number of deposited eggs on the plants [two-way ANOVA, F_{(5, 108)} = 7.054, P < 0.001; Figure 3A]. Significantly more eggs were laid on the treatment containing maize alone compared to the rest of the treatments where companion plants were present. Overall, there were no significant differences between eggs laid on the plants compared to those laid on the cage [two-way ANOVA, F_{(1, 108)} = 1.642, P = 0.203; Supplementary Table 2 and Figure 3A]. However, females laid significantly more eggs on the cages inner surface with the GL, SL, and M + SL treatments while they laid more eggs on plants with the maize treatment. This underscores the repellent activity of desmodium plants, particularly SL. There was a significant interaction between both factors investigated (i.e., plant treatments and oviposition sites [two-way ANOVA, F_{(5, 108)} = 9.382, P < 0.001; Supplementary Table 2 and Figure 3A]. In the combined treatments (i.e., Maize + GL and Maize + SL), moths laid significantly more eggs on maize than on GL (P = 0.02) and SL (P = 0.01) plants (Figure 3B).

In the oviposition compartment experiment, in a no choice test where moths were confined above plants in chambers sealed at the top and bottom with mesh cloth, there was no significant difference between plant treatments in the total number of eggs deposited on the plants [two-way ANOVA, F_{(4, 140)} = 2.036, P = 0.093; Supplementary Table 2 and Supplementary Figure 2]. More eggs were laid at the top of the compartment than at the bottom [two-way ANOVAs, F_{(1, 140)} = 27.413, P < 0.001; Supplementary Table 2 and Supplementary Figure 2], when moths were exposed to odors from greenleaf desmodium, D. intortum (GL), silverleaf desmodium, D. uncinatum (SL), and Brachiaria cv. Mulato II (BR). This indicates that S. frugiperda females were repelled by the volatiles of these companion plants. No interaction was found between both investigated factors [i.e., plant treatments and oviposition sites two-way ANOVA, F_{(4, 140)} = 1.320, P = 0.265; Supplementary Table 2 and Supplementary Figure 2A]. However, moths ovipositing in the oviposition compartment above maize plants laid a similar number of eggs on the top and bottom of the compartment (Supplementary Figure 2).

To test the hypothesis that companion plant volatiles were attractive to natural enemies, the behavioral response of two larval endoparasitoids, C. icipe and C. luteum, was assessed against the volatiles of companion plants (Figure 4) in a 4-arm olfactometer. Coccygidium luteum females spent significantly more time in the presence of volatiles from Brachiaria [F_{(1, 46)} = 6.64, P = 0.013], D. intortum [F_{(1, 46)} = 7.83, P = 0.007], and D. uncinatum [F_{(1, 46)} = 9.43, P = 0.003], compared to blank controls. Similarly, C. icipe females spent significantly more time in the presence of volatiles from Brachiaria [F_{(1, 46)} = 10.18, P = 0.002], D. intortum [F_{(1, 46)} = 11.81, P = 0.001], and D. uncinatum [F_{(1, 46)} = 20.15, P < 0.001] compared with blank controls.

GC-MS analysis of headspace collections from both Desmodium spp., Brachiaria Mulato II and maize detected 34 plant volatiles in eight chemical classes (alcohols, aldehydes, benzenoids, esters, homoterpenes, ketones, monoterpenes, and sesquiterpenes; Supplementary Table 2). Overall, multivariate analysis (MANOVA) revealed significant differences in the volatile profiles emitted by the different plant species, i.e., the companion plants, Desmodium spp. and Brachiaria Mulato II, and maize [Pillai’s Trace = 2.909, F_{(9, 72)} = 3.983, P < 0.015]. Principal component analysis (PCA) of the emitted volatiles showed that the first two principal components accounted for 47.1% of the total variation in the volatile emission. Hence, these two principal components (PCs) accounted for most of the variation in the data of likely biological relevance. Furthermore, a clear separation based on the first principal component (PC1) is visible between the companion plants (i.e., Desmodium spp. and Brachiaria Mulato II) and maize (Figure 5A). Another separation based on the second principal component (PC2) is obvious between the volatile profiles of Brachiaria Mulato II and D. intortum in one cluster and D. uncinatum, suggesting that the emitted volatiles of these two groups are dissimilar.

In descending order, the greatest loadings of PC1 (Figure 5B) were for β-ocimene (0.237), camphene (0.231), D-limonene (0.215), (E)-2-hexenal (0.209), DMNT (0.199), (Z)-3-hexenyl acetate (0.196), TMTT (0.189), and methyl salicylate (0.174), whereas the major loadings of PC2 were for (E)-2-hexenal (0.176), nerolidol (0.153), 1-octen-3-ol (0.130), β-myrcene (0.120), and (Z)-2-hexen-1-ol (0.114). These volatile organic compounds (VOCs) shown to contribute to PC1 and PC2 may impact both S. frugiperda and parasitoid behavior. A heatmap showed differential magnitude of volatile emission from the companion plants (i.e., Desmodium spp. and Brachiaria Mulato II) and maize plants. Specifically, alcohols and aldehydes dominated D. intortum (GL) volatiles, but D. uncinatum (SL) profiles are characterized by the emission of sesquiterpenes. Meanwhile, Brachiaria Mulato II (BR) profiles were characterized by high emission of benzenoids and homo- and monoterpenes (Supplementary Figure 3).

Univariate analyses of variance revealed significant differences in total [F_{(3, 24)} = 6.182, P = 0.003; Supplementary Figure 4] and individual volatile emission from the different companion plants (i.e., Desmodium spp. and Brachiaria Mulato II) and maize plants, particularly due to the high differential emission of several VOCs (Supplementary Table 3, Supporting Information).

Coupled GC-electroantennography (GC-EAG) analysis using S. frugiperda females showed electrophysiological activity for 14 volatiles from the companion plants Desmodium spp. and Brachiaria (Figure 6). The bioactive volatiles were identified by GC-MS and then confirmed by GC peak enhancement as: (Z)-3-hexenyl acetate, (S)-linalool, DMNT, (E)-β-farnesene, and TMTT from Brachiaria; (E)-2-hexenal, 3-octanone, DMNT, unknown compound, and β-caryophyllene from D. intortum; (S)-linalool, (E, E)-allo-ocimene, α-copaene, (E)-β-farnesene, β-selinene, and TMTT from D. uncinatum. In addition, S. frugiperda showed a robust EAG response to β-ocimene and MeSA, which were emitted from both Desmodium spp. and Brachiaria plants.

GC-EAG recordings from the antennae of C. luteum females showed electrophysiological activity for 11 compounds from the companion plants Desmodium spp. and Brachiaria (Figure 7A). The bioactive volatiles were tentatively identified by GC-MS and then confirmed by GC peak enhancement as: (S)-linalool, DMNT, MeSA, indole, and β-caryophyllene from Brachiaria; (E)-2-hexenal, 1-octen-3-ol, (Z)-3-hexenyl acetate, and β-ocimene from D. intortum; DMNT, (E, E)-allo-ocimene, β-caryophyllene, and TMTT from D. uncinatum.

GC-EAG recordings from the antennae of C. icipe females showed electrophysiological activity for 9 volatiles from the companion plants Desmodium spp. and Brachiaria (Figure 7B). The bioactive volatiles were tentatively identified by GC-MS and then confirmed by GC peak enhancement as: (Z)-3-hexenyl acetate, β-ocimene, (S)-linalool, MeSA, and TMTT from Brachiaria; (E)-2-hexenal and 1-octen-3-ol from D. intortum; β-ocimene, β-selinene, and TMTT from D. uncinatum. In addition, C. icipe showed a robust EAG response to DMNT, which was emitted from both Desmodium spp. and Brachiaria plants.

To determine the role of companion plant volatiles in modifying S. frugiperda flight behavior, we tested blends of the synthetic plant volatiles formulated at the same concentration and ratio as in the headspace samples of the companion plants (Supplementary Table 1). Moth responses to maize volatiles presented alone and in combination with the greenleaf desmodium (GL) and silverleaf desmodium (SL) synthetic blends were compared. In the wind tunnel bioassay, S. frugiperda showed attraction to maize alone by flying further upwind to headspace volatiles of maize plants presented alone compared to maize volatile released in combination with synthetic blends of companion plant volatiles or solvent control [F_{(3, 67)} = 7.32, P < 0.001] (Figure 8). Moreover, when presented with maize volatiles in combination with companion plant volatiles, S. frugiperda moths made fewer close approaches to the odor sources and made more return flights back down the tunnel, toward the release point, compared to maize volatiles presented alone.

Parasitoid responses to synthetic blends of bioactive companion plant volatiles formulated in the same concentration and ratio as the natural headspace samples were tested. Observed results confirmed behavioral responses of the parasitic wasps to the identified companion crop volatiles. In four-way olfactometer bioassays, the synthetic blends (0.5 μg dose) of D. intortum volatiles elicited potent attraction of both C. luteum [F_{(1, 46)} = 7.37, P = 0.009] and C. icipe [F_{(1, 42)} = 25.95, P < 0.001]. Similarly, the parasitic wasps were significantly attracted to synthetic equivalents of D. uncinatum [C. luteum: F_{(1, 42)} = 10.73, P = 0.002; C. icipe: F_{(1, 46)} = 4.68, P = 0.035], and Brachiaria volatiles [C. luteum: F_{(1, 42)} = 11.04, P = 0.001; C. icipe: F_{(1, 42)} = 5.42, P = 0.024] compared to solvent controls (Figure 9). Responses to synthetic blends were similar to natural samples indicating that the correct bioactive compounds had been identified. Olfactometer response (time spent in the treated area minus time spent in the control area) was compared using an unpaired t-test. For both parasitoid species (C. luteum and C. icipe), there was no significant difference in response to natural plant headspace samples and synthetic blends for all three plant species (C. luteum: P = 0.62, P = 0.64, P = 0.36 for D. intortum, D. uncinatum, and Brachiaria, respectively; C. icipe: P = 0.95, P = 0.94, P = 0.90 for D. intortum, D. uncinatum, and Brachiaria, respectively).

Extensive field surveys confirmed earlier reports and showed that S. frugiperda infestation and damage levels were substantially reduced when Push-Pull companion cropping was used (Figure 10). Overall mean (±S.E.) infestation level (larvae per 100 plants) was 6.6 times lower in Push-Pull plots in the short rains 2019 (Push-Pull: 3.69 ± 0.34; Monocrop maize: 24.68 ± 0.60) and 7.7 times lower in the long rains 2019 (Push-Pull: 3.21 ± 0.10; Monocrop maize: 24.85 ± 0.34). Mean damage level was 3.6 times lower in the short rains 2019 (Push-Pull: 13.04 ± 0.15; Mono maize: 47.95 ± 0.16) and 4.89 times lower in the long rains 2019 (Push-Pull: 9.30 ± 0.18; Monocrop maize: 45.51 ± 0.17). Two-way ANOVA, taking districts (i.e., different sampling locations) and farming system (i.e., Push-Pull vs. monocrop maize) as factors, revealed that the overall S. frugiperda infestation level and damage level were significantly affected by both factors and their interactions (Supplementary Table 4 and Figure 10A). Furthermore, similar findings were observed during the long rains season that districts, farming system and their interactions significantly affect the overall S. frugiperda infestation level and damage level (Supplementary Table 2 and Figure 10B).

Higher percentage parasitism was found in larvae collected from companion cropped maize in all four districts sampled. The mean parasitism level of larvae sampled from Push-Pull companion cropped maize was 18.7%, whereas it was only 2.1% in monocropped maize plots. Two-way ANOVA analysis showed that “farming system” [two-way ANOVA, F_{(1, 44)} = 28.452, P < 0.001] and “districts” [two-way ANOVA, F_{(3, 44)} = 3.026, P = 0.039] had a statistically significant impact on parasitism rate (Supplementary Figure 5), but no interaction effect occurred [two-way ANOVA, F_{(3, 44)} = 2.034, P = 0.123].