Prior adaptation of parasitoids improves biological control of symbiont‐protected pests

Abstract There is increasing demand for sustainable pest management to reduce harmful effects of pesticides on the environment and human health. For pest aphids, biological control with parasitoid wasps provides a welcome alternative, particularly in greenhouses. However, aphids are frequently infected with the heritable bacterial endosymbiont Hamiltonella defensa, which increases resistance to parasitoids and thereby hampers biological control. Using the black bean aphid (Aphis fabae) and its main parasitoid Lysiphlebus fabarum, we tested whether prior adaptation of parasitoids can improve the control of symbiont‐protected pests. We had parasitoid lines adapted to two different strains of H. defensa by experimental evolution, as well as parasitoids evolved on H. defensa‐free aphids. We compared their ability to control caged aphid populations comprising 60% unprotected and 40% H. defensa‐protected aphids, with both H. defensa strains present in the populations. Parasitoids that were not adapted to H. defensa had virtually no effect on aphid population dynamics compared to parasitoid‐free controls, but one of the adapted lines and a mixture of both adapted lines controlled aphids successfully, strongly benefitting plant growth. Selection by parasitoids altered aphid population composition in a very specific manner. Aphid populations became dominated by H. defensa‐protected aphids in the presence of parasitoids, and each adapted parasitoid line selected for the H. defensa strain it was not adapted to. This study shows, for the first time, that prior adaptation of parasitoids improves biological control of symbiont‐protected pests, but the high specificity of parasitoid counter‐resistance may represent a challenge for its implementation.

aphids still predominates in open fields, biological control of aphids has been adopted widely in greenhouse production, where the confined space facilitates the deployment of natural enemies (van Lenteren, 2012). An important component of the biocontrol arsenal against aphids is parasitoid wasps (Boivin, Hance, & Brodeur, 2012).
Already when H. defensa-conferred resistance to parasitoids was discovered, it was hypothesized that it may be responsible for observed failures of parasitoids to limit aphid abundance on crops (Gillespie, Quiring, Foottit, Foster, & Acheampong, 2009;Oliver et al., 2005). When tested in a laboratory setting, the introduction of parasitoids indeed resulted in a rapid increase of aphids harboring H. defensa, such that the increasingly resistant populations escaped control by parasitoids (Käch, Mathé-Hubert, Dennis, & Vorburger, 2018). Defensive symbionts like H. defensa thus represent a challenge for biological control of pest aphids (Vorburger, 2018 Rouchet & Vorburger, 2014). Such specificity is also a characteristic of natural populations of L. fabarum. A large collection of field-collected lines varied widely in the ability to parasitize aphids infected with different strains of H. defensa (Vorburger & Rouchet, 2016).
These findings suggest that prior adaptation of parasitoids to aphids carrying protective symbionts could be a viable strategy to improve biological control of pest aphids in which such symbionts occur. We tested this hypothesis by deploying experimentally evolved lines of parasitoids from previous work (Dennis et al., 2017) in caged populations of black bean aphids. We found that prior adaptation to H. defensa can indeed allow parasitoids to control aphid populations that would otherwise escape control due to the rapid evolution of symbiont-conferred resistance. This result represents a proof of principle for the genetic improvement of biocontrol agents by experimental evolution.

| Insect lines
As hosts, we used one H. defensa-free and two H. defensa-infected clonal lines of A. fabae with the same genetic background. The common genetic background was a single clone of A. fabae referred to as A06-407. It was collected in July 2006 from Chenopodium album in St. Margrethen (Switzerland) and does not contain any known facultative endosymbionts of aphids (Vorburger, Sandrock, Gouskov, Castañeda, & Ferrari, 2009 (Rouchet & Vorburger, 2014 As parasitoids, we used three sexual stocks of L. fabarum that have a common origin but differ in their history of laboratory adaptation to aphid hosts. These populations are derived from an hosts in our laboratory. By the start of the present experiment, they had been reared on these aphid lines for c. 120 generations, and they had maintained the described pattern of specific adaptation (data not shown). we study (Vorburger et al., 2009). The addition of aphids marked day 0 of the experiment. Four days later, two additional plants were added to the cages and cages were assigned randomly to one of five treatments (5 replicates each): control (no parasitoids); H-(parasitoids evolved on H. defensa-free aphids); H76 (parasitoids evolved on H76-infected aphids); H402 (parasitoids evolved on H402-infected aphids); and H76 + H402 (mixture of parasitoids evolved on H76and on H402-infected aphids). Treatments were applied on day 7 of the experiment by adding 22 female and 12 male wasps of the respective lines of L. fabarum to each cage (11 and 6 each for the mixed treatment). On the same day, aphid density was estimated for the first time; thereafter, aphid and parasitoid density was estimated twice per week. On each occasion, we removed the two oldest plants from the cages and replaced them with two new plants. One of the removed plants was selected randomly for counting, the other was cut and returned to the cage so that aphids could migrate over to live plants and parasitoid mummies on the plants could hatch inside the cage. The retained plant was sealed in a plastic bag and frozen to arrest aphid reproduction, before all aphids and parasitoid mummies were counted. After day 18 of the experiment, plants started to show stunted growth because of aphid infestations; hence, we began to quantify plant size. For this, the plants were disassembled into stalks and leaves, which were spread out and photographed on a white background with size reference. The area of all plant material in the photographs was estimated with ImageJ v. 1.52 (Schneider, Rasband, & Eliceiri, 2012), from which we calculated the variable "plant surface" = 4 × total stalk area + 1 × total leaf area, since aphids feed on the underside of leaves and on the stalks, which have a quadratic cross section in V. faba. The last count took place 67 days after the addition of the aphids to the cages. At this point, we also took a haphazard sample of 24 aphids per cage to quantify their population composition at the end of the experiment, and we determined the total fresh weight of all plants in the cages (aboveground parts) as a measure of plant condition.

| Final composition of aphid populations
The DNA of aphids collected at the end of the experiment was extracted using the "salting out" protocol described in (Sunnucks and Hales, 1996). We tested each individual for infection with H. defensa by diagnostic PCR, amplifying part of the 16S ribosomal RNA gene with symbiont-specific primers (Ferrari, West, Via, & Godfray, 2012).
We also ran a diagnostic PCR for the obligate endosymbiont of aphids,

| Statistical analyses
Aphid density on plants was expressed as individuals per cm 2 , that is, #aphids/plant surface area, and parasitoid density as the number of mummies per cm 2 . To obtain comparable values from the early counts when plant growth was not visibly impacted by aphids and plant size not quantified, we assumed a plant surface of 149.8 cm 2 , which is the average of healthy plants of the same age. Densities were analyzed with mixed linear models after square root transformation for aphid densities and log-transformation for parasitoid densities (log(#mummies + 1)/plant surface area) to improve normality of residuals and homogeneity of variances. We tested for the effects of treatment, time (day of count), and the treatment × time interaction.
Cage was included as a random effect to account for the nonindependence of successive counts from the same cage. For aphid and parasitoid densities, we ran global models with all treatments as well as models for all pairwise comparisons between treatments with sequential Bonferroni correction to account for multiple testing (Rice, 1989). Analyses were carried out in R v. 3.5.0 (R Core Team, 2017), using the lme4 library (Bates, Maechler, Bolker, & Walker, 2015) with the lmerTest library for significance tests of fixed and random effects (Kuznetsova, Brockhoff, & Christensen, 2015).
Plant fresh weights at the end of the experiment were compared among treatments with a one-way ANOVA, followed by pairwise comparisons using Tukey's HSD (Tukey, 1977). The final composition of aphid populations was analyzed with permutational MANOVA on the arcsine square root-transformed proportions of H. defensa-free, H76-infected, and H402-infected aphids, using the adonis function in the vegan library (Oksanen et al., 2019), followed by pairwise post hoc comparisons with the pairwise.perm.manova function in the RVAideMemoire package (Hervé, 2020). For all treatments, we also tested whether the total proportion of H. defensa-infected aphids at the end of the experiment differed from the initial proportion of 0.4, using one-sample t tests.  (Figure 1). Specifically, when we introduced L. fabarum from a population that was evolved experimentally on H. defensa-free (H-) aphids, the parasitoids established successfully but had no detectable effect on aphid population density compared to parasitoid-free control cages (Figure 1a, b, Table 1). When we introduced parasitoids from a population that was evolved on H402-infected aphids (i.e., parasitoids adapted to aphids carrying H. defensa strain H402), parasitoids reached higher densities than the parasitoids adapted to H-aphids, but the effect on aphid populations remained weak, such that aphid densities, although somewhat lower, did not differ significantly from the control and the H-treatments (Figure 1c, Tables   1 and 2). However, parasitoids adapted to H76-infected aphids did have a significant effect on aphid population dynamics. They suppressed aphid population growth, resulting in significantly lower aphid densities at the end of the experiment (Figure 1d, Table 1).

| RE SULTS
Virtually the same result was observed when we introduced a mixture of H76-and H402-adapted parasitoids (Figure 1e, Table 1).
The broad bean plants clearly benefitted from successful aphid control in that plant fresh weights at the end of the experiment were highest in the two treatments where parasitoids managed to suppress aphid populations (Figure 2).

| D ISCUSS I ON
There is an urgent need for more sustainable pest control to reduce the harmful side effects of conventional control with pesticides (Geiger et al., 2010;Kim et al., 2017). Biological control with parasitoids is a much-used alternative to control pest aphids, particularly in greenhouse crops (Boivin et al., 2012), but its success can be hampered by the rapid evolution of symbiont-conferred resistance (Käch et al., 2018). Here, we show that parasitoids with improved infectivity through prior adaptation to defensive symbionts present in an aphid pest can reduce aphid population densities in a situation where unselected parasitoids fail. Improving biological control agents through selective breeding is not a new approach (Kruitwagen et al., 2018;Lommen et al., 2017). It has been applied to life-history traits such as development time and sex ratio, to behavioral traits like host finding, and to increased tolerance of host defenses (reviewed in Lirakis & Magalhães, 2019). We show for the first time, though, that parasitoid adaptation to a defense encoded by a microbial symbiont rather than the host itself does indeed improve control of a symbiont-protected pest. Aphid population suppression resulted in improved plant growth, which is the main goal of biological control. The specific counter-resistance of experimentally evolved parasitoids was also reflected in how the aphid populations responded to selection by parasitoids. When H402-adapted parasitoids were applied, aphid populations became dominated by H76-infected aphids, and when H76-adapted parasitoids were applied, almost only H402-infected aphids survived. In the first case, this rapid response to selection prevented an effective control of aphid populations ( Figure 1c), but in the second case, the parasitoids were still able to suppress aphid densities (Figure 1d). This difference is interesting, albeit not entirely explicable by our current knowledge of the system. We can exclude that H76-adapted wasps also parasitized H402-infected aphids effectively, because tests immediately before and after the present cage experiment confirmed that there was no cross-infectivity, as reported in Dennis et al. (2017)  supports sufficiently high parasitoid densities, wasps can also kill resistant aphids, not through parasitism but presumably by continuous disturbance that prevents aphid feeding and leads to starvation, or by excessive stabbing (Hertäg & Vorburger, 2018 in terms of reduced lifespan and lifetime reproduction (Vorburger & Gouskov, 2011). This could explain the slight decline of H. defensa-infected aphids in the wasp-free treatment. In a comparison of multiple H. defensa strains, it has been shown that H402 is more costly to the host than H76 (Cayetano et al., 2015). Thus, we could speculate that the H76-adapted wasps have exerted better control because they selected for H402-infected aphids that are generally less fit than the H76-infected aphids that were favored by the H402-adapted parasitoids. Whatever the correct explanation, using a combination of parasitoids adapted to different symbiont strains, as we did in one of our treatments, may safeguard against this issue ( Figure 1e). An alternative approach would be to expose parasitoids to selection regimes comprising more than one strain of H. defensa, either simultaneously or sequentially, in the hope to produce more general counteradaptation that is effective against multiple strains of H. defensa. To our knowledge, this has never been tried. It is thus unknown whether or not such an approach would be foiled by the Symbiont-conferred resistance to parasitoids is not restricted to aphids. Since the original discovery in pea aphids (Oliver et al., 2003), new cases keep being discovered (Hansen, Jeong, Paine, & Stouthamer, 2007;Xie, Butler, Sanchez, & Mateos, 2014 (Łukasik, Asch, Guo, Ferrari, & Godfray, 2013;Scarborough, Ferrari, & Godfray, 2005), or the Wolbachiamediated protection against viral pathogens in flies and mosquitoes (Glaser & Meola, 2010;Hedges, Brownlie, O'Neill, & Johnson, 2008;Teixeira, Ferreira, & Ashburner, 2008). Thus, we expect that future research will show that defensive symbionts can challenge the biological control of various arthropod pests. Just as in pesticides, strong selection by biological control agents will favor the evolution of resistance (Tomasetto et al., 2017). In pest populations where defensive symbionts occur, this will result in an elevated prevalence of these symbionts that may reduce biocontrol success (Käch et al., 2018;Oliver et al., 2008). Unlike pesticides, however, biological control agents possess genetic variation to evolve counter-resistance. This genetic variation can be managed and selected to improve the performance of natural enemies for biological control (Kruitwagen et al., 2018;Lommen et al., 2017).
Here, we provided a proof of principle, in a laboratory setting, that experimental evolution is an effective means to improve the biocontrol capacity of parasitoid wasps toward symbiont-protected pests. The confined space and the simplified single-crop habitat typical of greenhouse cultures certainly bear similarity to a laboratory setting, but it remains to be demonstrated whether the approach can also be applied successfully at the larger scale of real biocontrol interventions.

ACK N OWLED G EM ENTS
We thank P. Rodriguez for the diligent maintenance of insect lines used in this study and R. Stegmayer for assistance with molecular work. We also thank two reviewers for very helpful input and corrections. This study was supported by the Swiss National Science Foundation (Sinergia grant nr. CRSII3_154396 and project grant nr. 31003A_181969 to CV).

CO N FLI C T O F I NTE R E S T
None declared.  (1866)