Study organisms. Experiments were performed using seven insect species belonging to three different trophic levels associated with the cultivation of sweet pepper (Capsicum annuum) or raspberry (Rubus idaeus). These included two aphid species, three primary parasitoids and two hyperparasitoids. Specifically, experiments were conducted using non-winged adults of Amphorophora idaei and Myzus persicae var. nicotianae (both Hemiptera: Aphididae), adult females of their primary parasitoids Aphidius colemani, A. ervi and A. matricariae (Hymenoptera: Braconidae), and adult females of the hyperparasitoids Asaphes suspensus (Hymenoptera: Pteromalidae) and Dendrocerus aphidum (Hymenoptera: Megaspilidae). The European large raspberry aphid A. idaei is a specialist on R. idaeus plants (Blackman et al., 1977), while M. persicae var. nicotianae is a subspecies of the green peach aphid Myzus persicae, one of the most polyphagous aphid species. The subspecies M. persicae var. nicotianae is found mainly on tobacco and sweet pepper, although it can survive and reproduce on a relatively wide range of plant species (Clements et al., 2000). Aphidius spp. are generalist aphid parasitoids, commonly used in commercial biological control (Yano, 2006). Dendrocerus aphidum and A. suspensus are generalist, secondary idiobiont ectoparasitoids attacking pre-pupal and pupal stages of primary aphid parasitoids, such as Aphidius spp., inside mummified aphids (Walker and Cameron, 1981; Höller et al., 1993).
Initial aphid colonies of A. idaei and M. persicae var. nicotianae were obtained from the Research Centre for Fruit Cultivation (Sint-Truiden, Belgium) and NIOO-KNAW (Wageningen, the Netherlands), respectively. Aphids were reared and maintained under controlled conditions (22°C, 70% RH and a 16L:8D photoperiod) on R. idaeus var. “Kwanza” (Advanced Berry Breeding B.V., Hedel, Belgium) and C. annuum var. “Yolo Wonder” (Mexada DIY B.V., Sint-Agatha-Berchem, Belgium), respectively. Weekly, fresh plants were added to the colonies. All Aphidius spp. were obtained in the form of parasitized aphid mummies from Biobest (Westerlo, Belgium) (A. colemani: Aphidius-system®; A. ervi: Ervi-system®; A. matricariae: Matricariae-system®). Mummies were placed inside a nylon insect cage (17.5 cm × 17.5 cm × 17.5 cm, 96 × 26 grids per inch2 − 680 µm aperture, BugDorm, MegaView Science Co., Ltd.) and kept under controlled conditions (22°C, 70% RH and a 16L:8D photoperiod) until parasitoid emergence. Dendrocerus aphidum and A. suspensus were reared in the laboratory on fresh (1 day old) M. persicae var. nicotianae mummies parasitized by A. colemani. Hyperparasitoids were allowed to hyperparasitize the mummies for a period of 48 hours. Subsequently, the mummies were placed in a fine-mesh nylon insect cage (24.5 cm × 24.5 cm × 24.5 cm, 150 × 150 grids per inch2 − 160 µm aperture, BugDorm, MegaView Science Co., Ltd.) and kept under controlled conditions (22°C, 70% RH and a 16L:8D photoperiod) until hyperparasitoid emergence. All experiments were performed with < 24-hour-old, food- and water-deprived females.
Three bacterial strains were used in this study. Strains were isolated from the aphids Macrosiphum euphorbiae (Hemiptera: Aphididae) and M. persicae var. nicotianae, and the primary parasitoid A. ervi. Based on 16S ribosomal RNA (rRNA) gene sequencing, strains were found to be ubiquitous environmental bacteria, and assigned to Staphylococcus saprophyticus (ST18.16/160), Curtobacterium sp. (ST18.16/085) and Bacillus sp. (ST18.16/133), respectively. Sequencing of the RNA polymerase B subunit gene (rpoB) classified ST18.16/133 as Bacillus pumilus (Goelen et al., 2020a). Previous Y-tube olfactometer bioassays using a small set of insects revealed that A. colemani females responded positively to VOCs from ST18.16/133, negatively to volatiles from ST18.16/160, and showed a neutral response to ST18.16/085. Further, it was found that D. aphidum showed neutral responses towards strains ST18.16/133 and ST18.16/160, whereas ST18.16/085 was attractive (Goelen et al., 2020a). Strains were stored in tryptic soy broth (TSB; Oxoid) containing 25% glycerol at -80°C until use.
Production of VOCs. For production of VOCs, the procedure of Goelen et al. (2020a) was followed. Briefly, bacterial stock cultures were plated on tryptic soy agar (TSA; Oxoid) and incubated at 25°C for 24 hours. Subsequently, strains were re-streaked on the same medium and incubated at 25°C for another 24 hours. Next, a single colony was inoculated in 10 ml TSB and incubated overnight at 25°C at 120 rpm. After incubation, cells were washed twice and diluted in sterile physiological water (0.9% NaCl) until an optical density (OD 600 nm) of 1 was reached. Next, 1.5 ml of the cell suspension was inoculated in a 250 ml-Erlenmeyer flask containing 150 ml filter-sterilized GYP25 medium. Flasks were sealed with sterile silicone plugs and incubated for 48 hours at 25°C at 120 rpm. Fermentations were set up in triplicate, and non-inoculated, blank GYP25 medium was included as a negative control (also in triplicate). After incubation, the media were centrifuged for 15 min at 10,000 × g. Subsequently, collected supernatants were filter-sterilized to obtain cell-free test media containing the VOCs. The cell-free samples were then stored in small aliquots in sterile, amber glass vials at -20°C until further use.
Olfactometer bioassays. To investigate the olfactory response of the insects, a Y-tube olfactometer bioassay was performed as described previously (Goelen et al., 2020a). The Y-tube (stem: 20 cm; arms: 12 cm with a 60° angle at the Y-junction; inner diameter: 1.5 cm) was put on a table that was homogeneously illuminated by four 24W T5 TL-fluorescent tubes (16 × 549 mm, 1350 Lumen, 5500K, True-Light®, Naturalite Benelux) with a 96% colour representation of true day light at a height of 45 cm. Further, the Y-tube was mounted at a 20° incline to stimulate movement of the insects towards the bifurcation, and a charcoal-filtered air was led through each arm of the Y-tube at a speed of 400 ml min− 1. To eliminate any visual cues, the olfactometer was fully enclosed with white curtains. Olfactory response was investigated by loading 150 µl of the cell-free cultivation medium onto a filter paper (diameter: 37 mm; Macherey-Nagel, Düren, Germany) and subsequently placing it in one of the odour chambers, while a second filter paper loaded with 150 µl of the blank medium was placed in the second odour chamber. Insects were tested in 24 cohorts of five adult individuals and were released at the base of the stem section of the olfactometer. Olfactory response was evaluated 10 min after parasitoid or hyperparasitoid release (Goelen et al., 2020a), or 20 min after aphid release, which was found to be the optimal time point of evaluation in preliminary experiments. Individuals that had passed a set line in one of the olfactometer arms (1 cm from the Y-junction) at the time of evaluation were considered to have chosen the odour source presented by that olfactometer arm (Goelen et al., 2020a). All other insects were considered non-responding individuals and were eliminated from statistical analysis. Aphids were food- and water-deprived one hour prior to testing, while the tested freshly emerged parasitoids and hyperparasitoids did not receive any food or water before subjecting them to the bioassays. For every bioassay, new individuals were used. After every two releases the filter papers in the odour chambers were replaced with fresh filter papers with 150 µl medium. Furthermore, to compensate for unforeseen asymmetry in the setup, we swapped the odour chambers after releasing six cohorts. At the same time, the Y-tube glassware was replaced by clean glassware. At the end of the bioassay, all olfactometer parts were cleaned with tap water, distilled water, acetone and finally pentane, after which the different parts were put in an oven at 150oC until the next day. All bioassays were conducted at 23oC ± 1oC and 65 ± 5% RH between 08h00 and 17h00. As the VOC composition of all three biological replicates was highly similar, olfactory response was determined for one of the three biological replicates.
Chemical analysis of VOCs. For each biological replicate, the VOC composition was analysed by headspace solid phase micro-extraction gas chromatography followed by mass spectrometry analysis (HS-SPME-GC-MS). Volatile compounds were separated by a Thermo Trace 1300 GC system equipped with a MXT-5 column (30 m length × 0.18 mm inner diameter × 0.18 µm film thickness; Restek, Bellefonte, Pennsylvania, USA) and subsequently detected by a ISQ mass spectrometer. Five millilitres of each sample was supplemented with 1.75 g NaCl and kept at
-20oC until analysis. Samples were thawed at room temperature and subsequently incubated at 60oC under constant agitation in a TriPlus RSH SPME auto sampler (Thermo Fisher Scientific, Waltham, Massachusetts, USA). The HS-SPME volatile collection was conducted using a 50/30 µm DVB/CAR/ PDMS coating fibre (Supelco, Bellefonte, Pennsylvania, USA). After five min of equilibration, the SPME fibre was exposed to the headspace sample for 30 min. The compounds trapped on the fibre were thermally desorbed in the injection port of the chromatograph by heating the fibre for 15 min at 270°C. Further volatile collection and separation conditions were as described by Goelen et al. (2020a). Compounds were identified and quantified as in Reher et al. (2019). Chromatograms were analysed with amdis v2.71 (Stein, 1999) to deconvolute overlapping peaks. Obtained spectra were subsequently annotated using the NIST MS Search v2.0g software, using the NIST2011, FFNSC and Adams libraries. This yielded a list of 244 tentatively identified compounds across all bacterial VOC blends and the blank medium. Peak areas of these compounds were compared to the background signal, which was identified by running a GC-MS with a sample of five milliliters of demineralized water and 1.75 g of NaCl. This background signal was then subtracted from the peak areas of the corresponding tentatively identified compounds, when the difference fell below a set threshold peak area of 1,000, they were eliminated from further analysis. This yielded a list of 66 different compounds. To further extract and integrate the compound elution profiles, a file was used with the identified compounds containing the expected retention times and spectrum profiles. Extraction was performed for every compound in every chromatogram over a time-restricted window using weighted non-negative least square analysis (Lawson and Hanson, 1995), and for every compound, the peak areas were computed from the extracted profiles. Finally, relative peak areas were computed as the ratio between the compounds’ peak area and the total peak area per sample, and then multiplied by 100 (%). Relative peak areas were then used for further analysis, and summarized in a table (Table S1, Supporting Information).
Statistical analysis. Insect olfactory response was analysed using a Generalized Linear Mixed Model (GLMM) based on a binomial distribution with a logit link function (logistic regression) using bacterial strain as fixed factor (performed in R using the glmer function from the lme4 package). Each release of one cohort of five insects served as a replicate (n = 24). To prevent pseudo-replication and to adjust for overdispersion, the release of each cohort was included in the model as a random factor. The number of insects choosing either the control or the bacterial VOCs in each cohort was entered as a response variable. Insect preference was examined by testing the null hypothesis (Ho) that insects showed no preference for any olfactometer arm (i.e. 50:50 response) by testing Ho: logit = 0, which equals a 50:50 distribution. Results were presented by calculating the preference index (PI) by dividing the difference between the number of insects choosing the bacterial odours and the number of insects choosing the control odours, by the total number of responding insects. Furthermore, an analysis of variance Type III Wald chi-square test was performed on the GLMM to determine if there was an overall difference between the olfactory responses for the different VOC blends. A GLMM was also used to determine whether insect species or trophic level affected the olfactory response of the insects. Once again the number of insects choosing either the control or treatment side in each cohort was included as a dependent variable, insect species or trophic level were included as fixed factors. Finally, interaction GLMM models were created to determine the interaction effect of insect species or trophic level, and the different VOC blends on the olfactory response of the insects. The number of insects choosing either the control or treatment side in each cohort was included as a dependent variable, while the interaction between either insect species and the different VOC blends, or insect trophic level and the different VOC blends were included as fixed factors. The release of each cohort was included in the model as a random factor.
Differences in VOC composition were visualized by constructing a heatmap from the strain × relative peak area matrix of calculated Z-scores using the heatmap.2 function in the gplots package V3.1.0 in R (Warnes et al., 2016). For each compound in each sample, the Z-scores were calculated by subtracting the mean relative peak area of all samples for the respective compound from the relative peak area of that compound in the sample, and dividing that by the respective standard deviation. Furthermore, a non-metric multidimensional scaling (NMDS) plot was made from the strain × relative peak area matrix by using a Bray-Curtis distance matrix in the Vegan package in R (Oksanen et al., 2013). Additionally, a permutational multivariate analysis of variance (perMANOVA) was carried out to test for significant differences in the chemical composition of the VOC blends produced by the tested strains, based on 1,000 permutations. The analysis was performed using the adonis function from the Vegan package in R. Finally, differences in the relative peak area of different compound families were calculated by first checking for normality using a Levene’s test. When the normality assumption was violated, a Kruskal-Wallis one-way analysis of variance (ANOVA), followed by a Dunn’s test was performed. When the normality assumption was met, a univariate ANOVA followed by a Tukey’s HSD test was performed.
Pearson correlation tests were performed to investigate correlations between insect olfactory response and the VOC composition. Calculations were performed using the data obtained from the biological replicate which was used for the olfactometer experiments and the VOC analysis. For each insect species and each VOC blend tested, we computed the PI values for each of the 24 cohorts of five insects, and correlated this to the relative peak area of each of the volatile compounds in the VOC blends (66 compounds × 4 relative peak areas (for 3 bacteria and 1 blank medium) in total). To control for multiple comparisons, the P values were adjusted using the Benjamini-Hochberg procedure, for the correlations of each individual species (Benjamini and Hochberg, 1995). This was done by multiplying the ratio between the individual rank of the P value and the number of hypotheses (individual correlations per species = 66), by the false discovery rate. We defined the false discovery rate at 5% (Benjamini and Hochberg, 1995). In all analyses, a significance level of α = 0.05 was used to determine significant effects. All statistical analyses were performed in R 4.0.4 (R Core Development Team, 2019).