One of the most studied and best-known mutualistic relationships between insects is that between ants and phloem-feeding insects. Ants feed on honeydew excreted by phloem-feeding insects and, in exchange, attack the phloem feeders’ natural enemies, including parasitic wasps. However, parasitic wasps are under selection to exploit information on hazards and avoid them. Here, we tested whether parasitic wasps detect the previous presence of ants attending colonies of phloem feeders. Behavioural assays demonstrate that wasps left colonies previously attended by ants more frequently than control colonies. This behaviour has a cost for the parasitic wasp as females inserted their ovipositor in fewer hosts per colony. In a further bioassay, wasps spent less time on papers impregnated with extracts of the ant cues than on control papers. GC/MS analyses demonstrated that ants left a blend of cuticular hydrocarbons when they attended colonies of phloem feeders. These cuticular hydrocarbons are deposited passively when ants search for food. Overall, these results suggest, for the first time, that parasitic wasps of honeydew producers detect the previous presence of mutualistic ants through contact infochemicals. We anticipate such interactions to be widespread and to have implications in numerous ecosystems, as phloem feeders are usually tended by ants.

1. Mealybug colonies exposed to queenless ant nests

To obtain the ant nests, eight nest fragments were collected in a citrus orchard at Instituto Valenciano de Investigaciones Agrarias (Moncada, Spain) by digging up the nest from the ground. In the laboratory, the experimental nests were placed in eight plastic boxes (30.5 × 24.5 × 20 cm). The inner walls of the boxes were lined with Fluon at 60% (Aldrich Chemistry) to prevent ants from escaping. All fragments of nests were queenless and composed of ~200 workers. Nest fragments were kept in the laboratory (23 ± 3°C, natural daylight) and provided with a solution of water, honey and yeast at 1:4:1 on a piece of aluminium foil twice a week. Water was provided in a glass vial (15 × 1.5 cm in diameter) tapped with a piece of cotton wool in the middle of the vial. The vial was covered with a piece of aluminium foil and ants used it as a nest. Forty-eight hours prior to assays, food was removed to starve the ants and homogenize their feeding status.

To obtain mealybug colonies of similar size, potato sprouts were infested with 20-25 second instar to pre-ovipositional adult mealybugs, as A. vladimiri shows a host preference for older mealybugs [35]. Potato sprouts were individually placed into plastic boxes (16.5 × 11 × 6 high cm) with walls coated with Fluon both inside and outside the boxes. Sprouts were maintained with centrifuge tubes filled with bacteriological agar (20 g/L), with the sprout base inserted directly into the agar. Tubes were sealed with Parafilm^® to avoid ants digging. After infestation, mealybugs could settle and feed for 48 h.

To obtain mealybug colonies that had been in contact with ants (ant-exposed colonies), plastic boxes with the infested potato sprouts were connected to ant nests. Starved ants were allowed to forage in a mealybug colony by temporarily connecting the colony with the plastic box with a wire as a bridge (Fig. 1A). Twenty-four hours later the wire was carefully removed while ants were not using it. Immediately, potato sprouts were moved to an experimental arena to observe the behaviour of the female wasp. Arenas consisted of a polystyrene plastic box (10 × 14 × 14 cm) with a lateral hole (4 × 9 cm) covered with muslin. Inside the arena, the centrifuge tube with the infested potato sprout was placed vertically on a silicon base to elevate the colony and improve the accuracy of the observations. The same procedure was followed with mealybug colonies that had not been exposed to ants. Each treatment was replicated 55 times.

2. Mealybug colonies exposed to field queenright ant nests

To obtain mealybug colonies of similar size, green bean pods were used as plant substrate for the mealybugs. Prior to inoculation, one side of the bean pods was submerged in red paraffin wax to minimize the area for the mealybugs to settle and facilitate the observations (based on the methodology described by [36]). Bean pods were infested with 20-25 second instar to pre-ovipositional adult P. citri. Mealybugs were allowed to settle and feed for 24 h.

To obtain mealybug colonies that had been in contact with field ants, twelve citrus trees with high L. grandis activity were selected in a 15-year-old IVIA organic citrus orchard [Citrus sinensis (L.) Osbeck (Var. Navelate)]. A plastic box was used to expose the beans infested with mealybugs to foraging ants. Boxes were 38.5 × 32 × 25 cm and had four small holes (0.5 cm diameter) in one side to allow the entry of ants, and two big lateral holes (15 × 10 cm) covered with mesh for ventilation. Boxes were placed near the base of a citrus tree’s trunk and the four holes of the boxes were placed next to the path of L. grandis that were foraging in the tree canopy (Fig. 1B). In each box, infested beans were kept with insect pins on flower sponge fixed with silicone to the base of the box. Ants had access to the infested beans for 24 h. Then, the box was carefully removed from the trunk while ants were not present. Immediately after collection, infested beans were transported back to the laboratory in sterile boxes using disposable nitrile gloves. The same procedure was followed to obtain mealybug colonies non-exposed to ants, but the holes were blocked to exclude ant attendance. Experimental arenas for the behavioural assays of the field queenright nests were the same as in the queenless assays. Inside the arena, the beans infested with mealybugs were placed in two acrylic cylinders (5 cm diameter, 1 cm height) to elevate the colony and improve the accuracy of the observations. Each treatment was replicated 63 times.

3. Effect of previous ant attendance on parasitoid behaviour

For both types of ant colonies, parasitic wasp searching behaviour was recorded in mealybug colonies i) non-exposed to ants and ii) ant-exposed. A female wasp was released in an arena with the mealybug colony and the following behaviours were recorded: i) arrival of the wasp in the mealybug colony (i.e. spent more than 3 sec in the colony), ii) total time spent in the colony, iii) whether it left the colony (i.e. spent more than 3 sec out of the colony), iv) number of times it left the colony, and v) the total number of times she inserted her ovipositor in a host. Each observation begun 1 min after the parasitic wasp was released in the arena and lasted a minimum of 30 min and, after this period, the observation ended when the wasp rested or walked for more than 5 min without contacting hosts.

To account for potential temporal effects, equal numbers of each treatment were tested each day in both assays, randomizing the order of testing between days. Arenas were used once per day, cleaned with alcohol and left to dry for at least 24 h. All observations were carried out between 10:00 and 15:00 h.

4. Effect of ant cue infochemicals on parasitoid behaviour

The effect of L. grandis cue extracts on the behaviour of A. vladimiri was further investigated with a non-choice bioassay and extracts of the ant cues. Five extra trail extracts (ca. 960 ×5 = 4,800 ant-equivalents), obtained in the laboratory as described above from queenless ant nests, were gathered and used to treat filter paper squares for the bioassays. Papers treated with ant cues were impregnated with 50 µL of the pentane solution of trail extracts (ca. 320 ant-equivalents). Control papers were impregnated with the same volume of pentane (50 µL). Papers were used for experiments 5 min after they were impregnated to allow the solvent to evaporate. A drop of honey (75%) was provided on treated and control paper squares (1.5 × 1.5 cm). Papers were left in the middle of a glass Petri dish (5 cm diam.), into which a single wasp was then released. Petri dishes were used once per day, cleaned with alcohol and left to dry for at least 24 h.

After allowing the parasitic wasp to settle (1 min), the proportion of wasps that contacted the paper and the time spent on the paper were measured for a period of 10 min. Each treatment was repeated 15 times. To account for potential temporal effects, equal numbers of each treatment were tested each day, randomizing the order of testing between days.

5. Composition of ant infochemicals

Collection of chemical trails

The collection of chemical trails left by L. grandis was performed using Teflon coated wires [37], with slight differences between the queenless ant nests and field ant nests. For the queenless ant nests, metal wires (25 x 0.5 cm diameter) previously washed with ethanol were coated with Teflon tape and were employed as bridges to connect ant nests (total 300 ants) with boxes containing a sucrose solution feeder. To obtain control Teflon coated bridges, the same procedure was followed but boxes did not have ants.

For the field queenright ant nests, we used a similar methodology as for the behavioural observations. The same boxes were placed on the foraging path of the ants, but bean pods infested with mealybugs were placed inside a smaller plastic box (16.5 × 11 × 6 cm) coated with Tangle-Trap^® (Tanglefoot, Grand Rapids, Michigan). Metal wires (25 x 0.5 cm diameter) coated with Teflon tape were also employed as bridges to connect the inside of the small plastic boxes with the outside bigger box. Therefore, ants searching in the plastic containers had to walk over the Teflon coated bridges to reach the mealybug colonies. To obtain control Teflon coated bridges, the same procedure was followed but the entrance holes to the plastic container were blocked with clay to exclude ants.

In both experiments, ants were allowed to forage and cross the coated bridges for 24 h. Considering 8 h of effective activity of the total 24 h that the bridge was coated and one ant crossing the bridge each 30 s, each trail extract was considered to contain ca. 960 ant-equivalents. This is likely a conservative estimation as ants remain active during the night. Bridges were carefully removed when ants were not crossing the bridges and Teflon tapes were extracted with 3 mL of pentane (HPLC grade, Sigma-Aldrich, Madrid, Spain) to obtain the cues left behind by the ants, which were subsequently analysed by gas chromatography coupled to mass spectrometry (GC/MS). The control Teflon coated bridges were extracted in an identical way. Five replicates of chemical trails and four replicates of control samples were collected for the queenless ant nests, while ten replicates of chemical trails and five replicates of control samples were collected for the field ant nests.

Chemical analysis

Chemical trail and control extracts were concentrated to ~10 μL under gentle helium flow, and 2 μL were analyzed by GC/MS. All injections were performed on a Clarus 600 GC/MS apparatus (Perkin Elmer Inc., Wellesley, PA) equipped with a 30 m x 0.25 x 0.25 fused-silica capillary column (Zebron ZB-5MS, Phenomenex Inc., Torrance, CA). Extracts were injected in splitless mode with the oven programmed at 100 ºC for 1 min, raised at 10 ºC/min up to 180 ºC, maintained for 1 min, and then 5 ºC/min up to 280 ºC with 20-min hold. Injector temperature was set at 250 ºC and helium at 1 mL/min was used as carrier gas. The detection was performed in EI mode at 70 eV with ionization source and transfer line set at 180 ºC and 250 ºC, respectively. Scan mode was employed (m/z 35-500) and tentative identification was based on retention indices according alkane standards and diagnostic ions reported in literature, as there is no commercial sources for these cuticular hydrocarbons [38–42].

6. Statistical analysis

We compared the total time spent in the colony using ANOVA. The normality assumption was assessed using Shapiro’s test, and the homoscedasticity assumption was assessed with Levene’s test. The total time spent on the filtered paper was not normally distributed and was analyzed with a Wilcoxon test. Proportional and count data were analyzed with generalized linear models (GLMs). Initially, we assumed a Poisson error variance for count data (number of times the wasp left the colony, and number of stings per colony) and a binomial error variance for proportional data (colony detection; proportion of colonies with at least one host stung by the parasitic wasp; proportion of filter papers detected by the parasitic wasp). We assessed the assumed error structures by a heterogeneity factor equal to the residual deviance divided by the residual degrees of freedom. If we detected an over- or underdispersion, we re-evaluated the significance of the explanatory variables using an F test after rescaling the statistical model by a Pearson’s Chi-square divided by the residual degrees of freedom [43]. We present the means of untransformed proportion and count data (in preference to less intuitive statistics such as the back-transformed means of logit-transformed data).

Principal component analysis (PCA) was performed to visualize, through score and loading plots, differences in the chromatographic peak areas of all compounds in the four experimental cases (laboratory assay: queenless ant nests and controls; field assays: field queenright ant nests and controls). The chromatographic peak areas of all compounds were integrated for each replicate. The resulting data were arranged in a matrix of 24 rows (replicates) and 14 columns (chemical compounds as variables). In this data set, the zero value was assigned to those compounds not detected in a given experimental case. The minimum value of peak area was around 10⁴ units, the median was ∼10⁵ units, and the maximum was ∼10⁶ units. For compounds detected at trace levels, below the integration threshold, we used 10³ units as area value, which is 1 log-unit below the minimum integrated area. Then, to normalize the data distribution, area values were transformed by applying the quadratic root transformation. The prcomp function was employed to perform the PCA and the number of principal components to be considered was determined by examining their eigenvalues (l) and proportion of variances by using the get_eigenvalue function in the factoextra package. The ggplot function in the package ggplot2 was employed to visualize the scores. All data analyses were performed with the R v.3.6.3 statistical package [44].