Study plant and plant material
The woodland strawberry (Fragaria vesca L., Rosaceae) is an herbaceous perennial plant occurring throughout most of the Holarctic. It grows naturally in various half shaded and sunny habitats such as open forests and edges of farmland47,48. Across our study area (central Sweden) flowers are produced throughout the entire growing season from May until September, with a clear peak in early June48. In addition to sexual reproduction, woodland strawberry is also capable of clonal reproduction and forms a high number of runners that grow quickly into self-sustaining plants47.
The 16 plant genotypes used in this study were derived from distinct wild populations in Uppsala County, Sweden, during spring 201224,25,49 (Supplementary Material Table 1). The entire collection consists of 86 plant genotypes that have been screened for resistance against the strawberry leaf beetle25. For this study we selected eight plant genotypes with high resistance and eight genotypes with high susceptibility against strawberry leaf beetles to include the contrasting ends of the resistance spectra for this experiment25.
Following the clonal propagation for several generations, forty runners per plant genotype were transplanted into a common garden close to Uppsala (N59.741°, E17.684°) in autumn 2013. The common garden was covered with fabric mulch (Weibulls Horto) as a weed barrier, and no irrigation, herbicides or fertilizers were used. The full protocol for the common garden establishment is available in Muola et al.49.
Study insects
The insect herbivore and the parasitoid used in this study are native to Fennoscandia and sympatric in their northern distribution. The strawberry leaf beetle Galerucella tenella L. (Coleoptera: Chrysomelidae; Fig. 2) is oligophagous on several species of the Rosaceae family, including a range of Fragaria spp., and is commonly found on meadowsweet (Filipendula ulmaria (L.) Maxim.)50,51. It can reach pest status especially in organic strawberry cultivations in the Nordic countries19,52, the Baltic States53, and Russia54. The strawberry leaf beetle is a univoltine species that hibernates as adult in the topsoil before emerging during April to May when the plants start to grow after winter. By the end of June, the mated female beetles oviposit directly on the plant they feed on and the larvae hatch after approximately two weeks depending on the climatic conditions51. Like the adult beetles, the larvae feed on the leaves and flowers for two to four weeks until they pupate in the upper soil layer51.
Asecodes parviclava Thompson (Hymenoptera: Eulophidae) is the only known parasitoid species of G. tenella in Sweden. This gregarious endoparasitoid species attacks G. tenella in the early larval stage, inserting one or more eggs into the body of the beetle larvae. The parasitized beetle larva continues to feed and grow until the last stages of parasitism, when parasitoid larvae prepare for pupation and eventually consume all the tissue of their insect host from within. The remaining beetle cuticle turns into a mummified black shell (thereafter mummy, Fig. 4) around the pupating parasitoids. Most parasitoids overwinter in their mummy as pupae and emerge as adults during the next summer, when new beetle larvae are again available50,55.
Photographs of (A) a mummified larva of the strawberry leaf beetle (Galerucella tenella) inclosing pupae of the parasitoid species Asecodes parviclava (B) a pre-pupal stage before the beetle larva turns into a parasitized mummy or beetle pupa (C) a beetle pupa of the strawberry leaf beetle. Photos by Daniela Weber.
Insect rearing
Adult Asecodes parviclava were reared from naturally occurring larvae of the strawberry leaf beetle, which were collected from natural meadowsweet stands in the Skeppsvik Archipelago, Sweden (N63.779°, E20.616°; N63.762°, E20.594°) in July 2015. The mummies were placed individually in 1.5 mL micro tubes, sealed with a piece of cotton wool and stored over winter at a shaded and sheltered location outdoors at SLU Ultuna campus. In late spring, with rising temperatures, the tubes were monitored daily for emerging adult parasitoids. Upon emergence, the parasitoids were transferred to 250 ml plastic containers with a feeding station (i.e. a piece of cotton providing diluted honey) and allowed to mate freely.
To obtain beetle larvae for parasitism, we collected mating couples of adult G. tenella during May 2016 from six distinct natural Filipendula ulmaria populations in the area around Uppsala (N59.809°, E17.667°; N59.788°, E17.664°; N59.806°, E17.652°; N59.806°, E17.665°; N59.782°, E17.753°; N59.833°, E17.914°). The collected individuals were randomly mixed and placed in meshed cages containing a mix of woodland strawberry chosen randomly among the plant genotypes available in the common garden. The cages were kept in a greenhouse (15 °C, LD 16:8 h photoperiod, 80% RH) and after a 24 h oviposition period, plants with eggs were retrieved and placed in a separate cage to allow the development of the eggs into larvae. The larvae used for experiments were in the second instar and of approximately the same size.
Experimental setup
Larvae of the leaf beetles reared on resistant or susceptible plants were used to study effects of plant resistance and quality on parasitoid performance. Second instar beetle larvae were presented individually to a mated A. parviclava female in a 1.5 mL micro tube. The beetle larvae were considered parasitized when the parasitoid was observed to have inserted its ovipositor for at least 10 seconds and showed no further interest in the larva. This method to experimentally parasitize Galerucella beetles normally results in close-to 100% successful oviposition (Stenberg, pers. obs.), but we did not dissect the experimental larvae used for this study to confirm that all of them contained parasitoid eggs. Each parasitized larva was randomly assigned to feed one of the sixteen plant genotypes and transferred to a 30 ml plastic container containing one detached, middle aged leaf of the given plant genotype. Ten parasitized beetle larvae were individually fed plant material from each of the 16 plant genotypes, resulting in 160 rearing containers. The rearing containers were kept in a climate chamber (15 °C, LD 16:8 h photoperiod, 80% RH). We checked the parasitized larvae daily and replaced the leaves every third day. Upon pupation, we noted whether the parasitism was successful (i.e. the host larva turned into a mummy) or not (i.e. the host larva died or turned into a beetle pupa). For each mummy we recorded the date of occurrence and the weight three days after it reached the pupal/mummy stage. The development time of parasitoids was measured as the number of days between parasitism and mummification. The mummies were transferred to 1.5 ml Eppendorf tubes sealed with cotton wool and checked daily for adult egression. At the end of August the tubes with the mummies were transferred outdoors at the SLU campus in Ultuna to be stored over winter. In June the following year, we counted the enclosed parasitoid pupae by carefully dissecting the mummies under a microscope.
Metabolomic analysis of the tritrophic effects of food plant quality
To investigate which phytochemical compounds correlate with plant resistance to Galerucella tenella, and to verify whether the outcome of the larval development, following parasitism, is influenced by the general leaf chemistry, we conducted metabolomics profiling of leaf tissue from 14 of the 16 plant genotypes used in the experiment (two genotypes were left out by mistake). Leaf samples (of the same age, quality, and location as used in the parasitoid experiment) were sampled from strawberry leaf beetle damaged plants in the common garden. Two clonal plants from each genotype were separately sampled for metabolic profiling, from which five leaves per plant were collected and pooled. Compound concentrations representative of each genotype were then generated by taking the mean compound value of these two clonal replicates. Within 2–3 hours, samples were stored initially at −20 °C, and then at −80 °C before freeze drying. The dried leaves were grinded to a fine powder using an electric mill (IKA A 10 basic model) and stored in the dark to avoid degradation by extended exposure to sunlight.
Metabolomic profiling of the leaf tissue by GC/TOF-MS was performed at the Swedish Metabolomics Centre, Umeå, Sweden. For the extraction a combined extraction buffer (Chloroform/Water/Methanol (20/20/60, v/v)) was used and the procedure of sample preparation, derivatization, GC/TOF-MS analysis was carried out according to A et al.56. A detailed description of the procedure as well as all involved chemicals and standards can be found in the Supplementary Material. All non-processed GC/TOF-MS files were pre-treated using custom scripts (base-line correction, chromatogram alignment, data compression, Multivariate Curve Resolution) and subsequently identified by comparisons with libraries of retention time indices and mass spectra.
Statistical analysis
We used the multinomial logistic regression function ‘multinom’ in the R package ‘nnet’57 to estimate the probability for the occurrence of each potential outcome of the parasitism (host larva died; larva became a beetle pupa; larva became a mummy) in relation to the two plant resistance levels i.e. susceptible and resistant (model 1) and in relation to each of the 16 plant genotypes selected for this study (model 2). These terms were ran in separate models due to the fact that this function was unable to handle nested structures.
We used a linear model to assess variation in the total number of pupae per mummy among the 16 plant genotypes. We set the total number of pupae per mummy as the response variable and included weight of the mummy, development time from parasitism until mummy formation, and plant genotype nested in plant resistance level (susceptible or resistant) in the model. Plant genotype was treated here as a fixed factor since the used genotypes were not chosen randomly but were carefully selected based on the herbivore resistance level from a larger set of genotypes previously studied in Weber et al.25. A significant interaction term between weight and developmental time was retained in the final model, whereas all other interactions were excluded due to insignificance. For model validation, we assessed normality of the residuals by visual examination and conducted a Levene’s test to check for equality of variances.
Partial least squares generalised linear models (PLS GLMs) were used on the GC/TOF-MS data to test whether the respective outcome of the parasitism can be predicted based on individual compounds in the leaves of the plant genotypes on which the insect host had been feeding. PLS GLMs are considered a type of hybrid model that incorporates the use of both principal component analysis (PCA) and generalized linear regression. As such, these models harness the relative advantages of each58. Perhaps most critically, valid inferences can be made based on a large number of highly inter-correlated predictor variables, even when this far surpasses the number of response cases. Here, for instance, 14 parasitism response cases were regressed on 78 compound predictor variables. And secondly, binomial or Poisson error distributions can be explicitly accommodated.
We used the R package “plsRglm”58 to fit three Poisson PLS GLMs. These models were fitted with compound concentration as predictor variables, and a count of each parasitism outcome as an individual response per model (i.e., the ‘number of host larva died or not’, ‘number of larva that became a mummy or not’, or ‘number of beetles developed to pupa or not’). While currently no means exist to extend PLS GLMs for multinomial responses, our approach nonetheless could still account for dependency between parasitism responses. This was due to the fact that parasitism outcomes were mutually exclusive; i.e., a significant positive association between a compound and a response in one model (e.g., number of beetles developed), by default indicated a significant negative association in at least one other (e.g., number of larva that became a mummy). Only positive associations were deemed to be informative, as negative associations provided only redundant information. These a priori assumptions indicate that the need to adjust for multiple tests may arguably be relaxed. However, we nonetheless present significance results for both normal and adjusted tests (Supplementary Table 3). The adjustment was implemented using the ‘confints.bootpls’ function of the plsRglm package, where a Bonferroni type correction was applied to the alpha level for significance when estimating coefficient CIs.
For each of the three PLS GLMs, we used an AIC information criteria approach to select of the optimal number of components that should be fit in order to minimize AIC (i.e. three for the ‘host larva died’ and ‘larva became a mummy’ models, and four for the ‘beetle pupa’ model). For each compound in each model, 95% confidence intervals (CIs) were estimated based on 1000 parametric bootstraps. A significant association between the compound and the outcome was hence indicated where the range of the CI did not include zero. For model validation, we tested for over-dispersion using the ratio of the residual deviance and the residual’s degrees of freedom.
A PLS GLM was also run to test the link between the 78 compound predictor variables and plant resistance/susceptibly, encoded as a binary response. This model was fitted using a binomial family link function, but otherwise all analysis and validation procedures followed as for the Poisson models described above.
All statistical analyses were run using R v.3.5.1 and RStudio v. 1.1.456.
Source: Ecology - nature.com
