Bottom‐up effect of host protective symbionts on parasitoid diversity: Limited evidence from two field experiments

Abstract Protective symbionts can provide effective and specific protection to their hosts. This protection can differ between different symbiont strains with each strain providing protection against certain components of the parasite and pathogen community their host faces. Protective symbionts are especially well known from aphids where, among other functions, they provide protection against different parasitoid wasps. However, most of the evidence for this protection comes from laboratory experiments. Our aim was to understand how consistent protection is across different symbiont strains under natural field conditions and whether symbiont diversity enhanced the species diversity of colonizing parasitoids, as could be expected from the specificity of their protection. We used experimental colonies of the black bean aphid Aphis fabae to investigate symbiont‐conferred protection under natural field conditions over two seasons. Colonies differed only in their symbiont composition, carrying either no symbionts, a single strain of the protective symbiont Hamiltonella defensa, or a mixture of three H. defensa strains. These aphid colonies were exposed to natural parasitoid communities in the field. Subsequently, we determined the parasitoids hatched from each aphid colony. The evidence for a protective effect of H. defensa was limited and inconsistent between years, and aphid colonies harbouring multiple symbiont strains did not support a more diverse parasitoid community. Instead, parasitoid diversity tended to be highest in the absence of H. defensa. Symbiont‐conferred protection, although a strong and repeatable effect under laboratory conditions may not always cause the predicted bottom‐up effects under natural conditions in the field.


| INTRODUC TI ON
Diversity at one trophic level can promote diversity at other trophic levels in a food web. There is comparative and experimental evidence that high species diversity in primary producers provides a more diverse resource base and thus promotes biodiversity across multiple levels of consumers (Scherber et al., 2010;Sobek et al., 2009). Much of the support for cascading diversity effects comes from biodiversity experiments manipulating the number and composition of plant species (reviewed by Siemann et al., 1998;Weisser et al., 2017).
However, such effects are not restricted to species-level diversity. Phenotypic variation resulting from genetic variation within plant species also affect diversity at higher trophic levels (Barbour et al., 2016;Crutsinger et al., 2006;Johnson et al., 2006).
Although less studied, similar bottom-up effects can be expected for animal resources, for example on parasite diversity, since parasites are often highly specialized (the 'host diversity begets parasite diversity hypothesis', Johnson et al., 2016;Thieltges et al., 2011).
Within-species diversity in host susceptibility to parasites may arise from genetic variation for defence traits (e.g. Ebert et al., 1998;Sandrock et al., 2010;Smith et al., 1999), as well as from their association with symbionts that provide protection (Flórez et al., 2015;Haine, 2008). Such defensive symbioses include conditionally mutualistic associations like honeydew-collecting ants defending sapsucking insects (Stadler & Dixon, 2005), relatively loose associations of animals with environmentally acquired 'probiotic' gut symbionts (Neish, 2009) and very tight associations with maternally transmitted microbes providing protection (Oliver & Moran, 2009). Heritable defensive symbionts can be seen as a second line of defence in addition to the host's intrinsic immune system. They provide effective and specific resistance against parasites and pathogens and thereby show some parallels to the immune systems of animals, such as the vertebrate major histocompatibility complex (MHC). Defensive symbionts should hence be subjected to similar evolutionary forces (Hafer & Vorburger, 2019). These have repeatedly resulted in the diversification of immune systems in response to diverse parasites and pathogens (Ghosh et al., 2011;Litman et al., 2007;Messier-Solek et al., 2010). Similarly, protective symbionts and the parasites and pathogens they protect against are expected to drive each other's diversity (Hafer & Vorburger, 2019). Over ecological time frames, host populations possessing more diverse communities of protective symbionts are thus expected to support more diverse communities of parasites.
Protective symbionts are especially well-studied in aphids, whom they protect against parasitoid wasps and pathogenic fungi (Guo et al., 2017;Oliver et al., 2014;Vorburger, 2014). These benefits are usually set off by fitness costs these symbionts impose in the absence of natural enemies Heyworth & Ferrari, 2015;Oliver et al., 2008;Parker et al., 2017;Vorburger & Gouskov, 2011). As a consequence, secondary symbionts tend to occur at intermediate frequencies in nature (reviewed by Guo et al., 2017;McLean et al., 2016;Oliver et al., 2014;. Correlations between parasitoids and the symbiont communities of their aphid hosts observed in natural populations suggest that these frequencies may be partially driven by interactions between symbionts and parasitoids Smith et al., 2015). However, more experimental approaches are required to establish the causal factors underlying these associations.
Likewise, symbiont-conferred protection has been shown to exert strong selection on parasitoids under laboratory condition, promoting the evolution of counteradaptations to overcome resistance mediated by H. defensa (Dennis et al., 2017;Dion et al., 2011) or even driving parasitoids to extinction (Käch et al., 2018). However, it remains unknown to which extent symbiont diversity can shape the composition and diversity of parasitoid communities.
Under natural conditions, the evidence for symbiont-conferred protection and its benefits is less clear. In a field experiment, Rothacher et al. (2016) found that a single strain of H. defensa did indeed reduce parasitism rates of the black bean aphid Aphis fabae.
Because the defence provided by H. defensa was unequally effective against different parasitoid species, the symbiont also altered parasitoid community composition and increased evenness by reducing the abundance of the dominant parasitoid species in protected aphids. Interestingly, the protection by H. defensa did not result in increased aphid population size in the presence of H. defensa, assumedly due the costs it imposes on aphid fitness in the absence of parasitoids (Rothacher et al., 2016). Similarly, Hrček et al. (2016) showed under field conditions that two symbionts, H. defensa and Regiella insecticola, protect pea aphids Acyrthosiphon pisum against parasitoids and pathogenic fungi, respectively, but again this did not result in overall benefits for the aphids since this protection was offset by other causes of mortality. By contrast, a more recent study failed to find any effect of H. defensa in protecting Aphis craccivora against the local parasitoid community (Lenhart & White, 2017).
Here, we conducted a field experiment with black bean aphids Aphis fabae, exposing H. defensa-free and H. defensa-infected populations to test for bottom-up effects of H. defensa on naturally colonizing parasitoid communities. In contrast to previous experiments, we worked with three different strains of H. defensa, creating different singly infected populations as well as populations possessing multiple strains by mixing the three singly infected lines. This allowed us to address the following questions: (1) How effective is symbiont-conferred protection in the field and is it comparable for different strains of H. defensa? (2) Do more diverse symbiont communities lead to more diverse parasitoid communities, that is does symbiont diversity influence parasitoid diversity? Surprisingly, the evidence for protection and differences between H. defensa strains was very limited in our study. We started the experiment by planting the outdoor pots with four 4-week old bean plants that had been inoculated with 12 aphids of the appropriate treatment for 10 days. After 2 weeks in the field, we collected one plant per pot (see sampling below) and replaced it with a new plant inoculated with the same initial number and composition of aphids. The same procedure was repeated weekly for another 12 weeks, always harvesting the oldest plants after all of the first four plants had been harvested. However, some adjustments had to be made during the course of the experiment. The number of adult aphids used to inoculate the plants was reduced from 12 (round 1-9) to 9 (round 10-13) because the size of colonies that had developed when we put out the plants was sometimes so large that it adversely impacted the plants. In the fifth sampling round, we harvested two plants and replaced them with one only to reduce the exposure time in the field from 4 to 3 weeks. We did not provide a replacement plant in the week prior to final sampling and harvested the two last plants on consecutive days so that one of them had only been in the field for 2 weeks. In total, this resulted in the harvesting of 13 plants per pot. Heatwaves in June resulted in the death of some plants.

| Experimental set-up
These were immediately (within 24 hr) replaced with new plants of approximately the same age but without any aphids. The continued addition of new aphids ensured that treatment differences in aphid composition persisted throughout the experiment, even though migration between pots or influx of wild aphids would have been possible. Indeed, a test towards the end of the experiment confirmed that mostly aphids remained in their appropriate treatment (see Supporting Information, test for migration).
We reduced access of snails and slugs to pots and removed any snail or slug we found, but otherwise pots were freely accessible to ants, parasitoids, predators and other animals. Plants were watered whenever necessary, albeit due to the water reservoirs within the pots this was rarely the case.
No ethical approval was required for this work.

| Sampling and measurements
Sampling (at the end of each block in 2018, weekly in 2019) took place by gently placing a cellophane bag over each plant prior to cutting the plant and immediately sealing the bag to ensure that all animals located on the plant at that particular moment were trapped.
We measured the length of each plant and counted the number of aphids either by counting aphids in groups of roughly five individuals (2018) or counting their exact number (2019). We also counted the number of mummies (dead aphids that were parasitized successfully), irrespective of whether or not the parasitoid had already hatched. Unhatched mummies (usually the large majority of mummies) were collected in insect dishes for hatching and subsequent identification. We additionally kept the plants in the cellophane bags to be able to catch and determine any parasitoids that would hatch from mummies that only formed after collection. Lastly, we counted all ants and predators. For predators, we recorded which order they belonged to and which stage they were in (i.e. egg, larvae, pupae or adult).
See Table S1 for a complete list of all parasitoids identified.

| Statistical analysis
All analysis and plot generation (package ggplot2; Wickham, 2016) took place in R, version 3.6.1 (R Core Team, 2019). We used linear mixed effect models (package lme4; Bates et al., 2015) to analyse the effect of aphid endosymbiont treatments on the number of mummies, mummification rate (the number of mummies divided by the number of mummies and live aphids within each replicate), the number of hatched parasitoids, the number of aphids, plant size and parasitoid diversity (species number and Shannon index for all parasitoids and for primary parasitoids only). To analyse mummification, we excluded samples where neither aphids nor mummies were found due to predation or aphid emigration. Parasitoid species number and Shannon diversity were calculated with vegan (Oksanen et al., 2019). In addition to diversity estimates based on raw count data, we obtained rarefied species richness and Shannon index through function estimateD in iNEXT (Hsieh et al., 2016) using only samples with at least five parasitoids and a resample size of 5. We did this for primary parasitoids and for all parasitoids combined (primary and secondary). Prior to analysis, we transformed response variables using TransformTukey from the package rcompanion (Mangiafico, 2019) to ensure that model assumptions were met. We analysed data from both years separately. For 2018, we first calculated data over all plants from each pot and block which we used in subsequent analysis and included block as random effect. For 2019, we used individual data from each round (i.e. sampling event) for aphid number, mummy number, mummification rate and plant size, and we included sampling round, plot and pot (nested within plot) as random effects. For the number of hatched parasitoids and parasitoid diversity in 2019, we calculated summed numbers over all rounds for each pot and included plot as random effect.
In each case, we built a linear mixed model including the contrasts H. defensa presence versus absence, H. defensa diversity (3 vs. 1 different haplotype) and H. defensa strain (among strains; comparison against H402) in that order as fixed effects to partition variance among them. We included aphid number as covariate and first factor in the formula when analysing mummy number, mummification rate, number of hatched parasitoids and plant size. Each model was followed with a type I analysis of variance using Satterthwaite's method to obtain p-values.
To analyse parasitoid species composition, we used parasitoids summarized over all plants for each block for 2018 and summarized all parasitoids per pot over all rounds for 2019 to obtain a single species matrix for each independent replicate. To test whether parasitoid community composition depended on treatment, we conducted a distance-based redundancy analysis (dbRDA) from the package vegan (Oksanen et al., 2019). We used rank correlations between dissimilarity indices and gradient separation to select the best dissimilarity index to use for subsequent analysis. We obtained significance values using permutation tests with 999 permutations.

| RE SULTS
In both years, the treatments with the highest total number of mum-  Table 1; Table S2). However, this was at least partially related to a somewhat higher number of aphids on the plants of this treatment (marginally non-significant, Table 1; Figure 1d), such that the proportion of mummified aphids (mummification rate) did not differ significantly between H. defensa-free and H. defensa-infected aphids ( Figure 1b; Table 1). The number of different H. defensa strains (1 vs. 3) did not have a significant effect on any of these responses, nor were there significant differences among the three treatments with single H. defensa strains (Figure 1a,b,d; Table 1; Table S2), albeit H. defensa strain showed a non-significant trend to affect mummification rate in 2018 (Figure 1b; Table 1; Table S2), which was highest with haplotype H402 and lowest with H76 in this year. None of the treatment differences observed in 2018 were significant in the 2019 experiment with the exception of the number of hatched parasitoids, which differed among the three treatments with one H. defensa strain. It was highest in the H15 treatment, but the variation among replicates was enormous (Figure 1c; Figures S1 and S2;  Table S2).
In both years, the total number of parasitoid species obtained from aphids without H. defensa was significantly higher than in all treatments with H. defensa-infected aphids (Figure 2a; Table 2;   Table S3). Similarly, Shannon diversity over all parasitoid species was significantly higher in the H-treatment than in all other treatments in 2018, but not in 2019 (Figure 2c; Table 2; Table S3). These patterns were similar when looking at primary parasitoids only (Figure 2b,d; Table 2; Table S3).
We cannot rule out that some of our findings may have been affected by differences in sample size (i.e. the number of hatched parasitoids obtained), albeit this seems unlikely considering that the treatment that showed the lowest diversity when using all parasitoids, H15, showed the highest number of hatched parasitoids. Nevertheless, any significant effects disappeared or became marginally non-significant when using rarefied diversity estimates ( Figure S3; Table 2; Table S3)  Please note that in 2018 the experiment was carried out in three consecutive rounds with five replicates of each treatment per round, whereas in 2019, the experiment was arranged in six spatial blocks with one replicate per treatment that was sampled repeatedly over time. This was accounted for in the analysis but not for plotting. Please refer to Figures S1 and S2 for more detailed plots. Mummification rate was calculated for each individual replicate prior to calculating mean and CI, giving equal weight to replicates with different numbers of individuals remaining on the plants. This explains why mean mummification rates are higher than what mean numbers of mummies and aphids would suggest. Error bars represent 95% CI, boxes represent SE.  we tested whether intraspecific differences in an aphid host, mediated not by the species' genotype but by the genotype of its heritable protective symbiont, could exert bottom-up effects on subsequent trophic levels, that is primary and secondary parasitoids . The experiment was motivated by the fact that symbiont-conferred resistance in aphids is highly specific, with different strains of H. defensa providing unequal protection against different parasitoid species and or different genotypes of the same parasitoid species Cayetano & Vorburger, 2013Leclair et al., 2016;McLean & Godfray, 2015;Rouchet & Vorburger, 2012;Schmid et al., 2012). Hence, we predicted that higher symbiont diversity should promote higher parasitoid diversity (see Hafer & Vorburger, 2019), yet we did not observe any such effect in our field experiments. There was not even any strong evidence that the species composition of emerging parasitoids was influenced significantly by the different H. defensa strains ( Figure 3). Only in 2019 did we observe an effect of H. defensa strain in that rarefied species number was lowest in strain H15, the least protective strain in our experiment. Additionally, Shannon diversity (in 2018 only) and the number of parasitoid species (both years) were higher in H. defensa-free aphids, which were expected to be more permissive hosts to begin with. The relatively low rates of parasitism have certainly limited the opportunity to detect responses in terms of species composition and diversity, and it was probably further reduced by a very high incidence of secondary parasitoids (Figure 3), which may well have erased any symbiont effects there may have been on primary parasitoids. We also suspect that potential effects of symbiont diversity on parasitoid diversity may have been obscured by very high aphid mortality that was unrelated to parasitism.
We often obtained low aphid counts and even observed complete losses of aphids on individual plants. Aphid mortality seems to have been driven especially by a high abundance of predators. In line with previous findings (Costopoulos et al., 2014;Kovacs et al., 2017),  Also our more straightforward prediction that parasitism should be reduced in H. defensa-protected aphid populations was not wellsupported, despite working with strains of H. defensa that are known to confer resistance against several parasitoids under laboratory conditions. Even though in both years we did obtain the highest number of parasitoid wasp individuals out of H. defensa-free aphids and aphids infected with H15, the least protective of the three symbiont strains Schmid et al., 2012), the rate of parasitism did not differ significantly among treatments. This is in contrast to an earlier study by Rothacher et al. (2016), who found clear evidence for protection using a similar approach, but only compared between H. defensa-free aphids and aphids infected with a single H. defensa strain, H402. We know that the symbionts have not lost their protective phenotype between these experiments. Most likely, the differences are explicable by differences in parasitoid occurrence.
During the field experiment of Rothacher et al. (2016), the parasitoid community was dominated by a single species, Lysiphlebus fabarum, normally A. fabae's most frequent parasitoid (Starý, 2006;Vorburger & Rouchet, 2016), against which H. defensa provides effective protection (Schmid et al., 2012;Vorburger et al., 2009 Figure S3 for rarefied diversity estimates Can any general lessons be learnt from our largely negative results? We were certainly surprised that differences in a trait that is clearly relevant ecologically-resistance to parasitoids-did not have a larger effect on the ensuing trophic networks, at least not in these experiments. In laboratory cage experiments, the effect of defensive symbionts on parasitism are unambiguous and strong (Käch et al., 2018;Oliver et al., 2008), and the role of interaction specificity in maintaining variation is demonstrable (Hafer-Hahmann & Vorburger, 2020;Rossbacher & Vorburger, 2020). There is also evidence from slightly more complex aphid-parasitoid-symbiont communities maintained in mesocosms that the presence of a protective symbiont can destabilize an experimental parasitoid community so strongly that local extinctions ensue (Sanders et al., 2016). However, laboratory or mesocosm experiments isolate a subset of interacting species from their surrounding community, which is at the same time an advantage and a limitation. They are better suited for a proof of principle that diversity at one trophic level can propagate to other trophic levels in a food web, but they tell us less about the actual importance of such effects in more complex natural communities. It is known that bottom-up and top-down effects are also affected by the surrounding community and its diversity (Cao et al., 2018;Crawford & Rudgers, 2013). More specifically, aphid species-level diversity can enhance parasitoid diversity (Petermann et al., 2010)

CO N FLI C T O F I NTE R E S T
The authors have no conflict of interest to report.

DATA AVA I L A B I L I T Y S TAT E M E N T
Data are available at Dryad Digital Repository https://doi. org/10.5061/dryad.1ns1r n8vr (Narayan et al., 2021).