Rapid and Differential Evolution of the Venom Composition of a Parasitoid Wasp Depending on the Host Strain

Parasitoid wasps rely primarily on venom to suppress the immune response and regulate the physiology of their host. Intraspecific variability of venom protein composition has been documented in some species, but its evolutionary potential is poorly understood. We performed an experimental evolution initiated with the crosses of two lines of Leptopilina boulardi of different venom composition to generate variability and create new combinations of venom factors. The offspring were maintained for 10 generations on two strains of Drosophila melanogaster differing in resistance/susceptibility to the parental parasitoid lines. The venom composition of individuals was characterized by a semi-automatic analysis of 1D SDS-PAGE electrophoresis protein profiles whose accuracy was checked by Western blot analysis of well-characterized venom proteins. Results made evident a rapid and differential evolution of the venom composition on both hosts and showed that the proteins beneficial on one host can be costly on the other. Overall, we demonstrated the capacity of rapid evolution of the venom composition in parasitoid wasps, important regulators of arthropod populations, suggesting a potential for adaptation to new hosts. Our approach also proved relevant in identifying, among the diversity of venom proteins, those possibly involved in parasitism success and whose role deserves to be deepened.

Formula of the modified chi² statistics using these four expectations The normal chi² statistics is: Here, and are respectively the observed and expected headcounts of the lbspnm allele frequencies or of the LbGAP phenotype in an experimental population at a given generation (F6 and F10), using either data for populations raised on the R host strain, or on the S host strain.
Drift will create deviation from the expected values in each replicate, in one direction or the other (some replicates being above the expectation and some below). Selection, on the opposite, will create deviation from the expected value in the same direction in all replicates. To measure the effects of selection rather than those of drift, we modified the chi² statistics as follows. This modification can be considered as a unilateral test of the chi² with the alternative hypothesis being H1 < or > : Because the modified chi² statistics might not follow anymore a chi² distribution, the software simuPOP was used to simulate the null distribution of these two statistics under the hypotheses of panmixia and neutrality for the simulated loci. This approach also tackles a problem we did not yet discussed. While we computed the statistics, we summed the divergences between observed and expected headcounts for each experimental population, even when they belong to the same replicate (for instance, one from the generation F6 and one from the generation F10) and are thus not independent. That would have been a problem if we had compared our observed chi² to the usual chi² distribution, but it is not here since the same formula was used to simulate the null distribution.

Simulation of the null distribution of our statistics with the software simuPOP
To obtain the null distribution of our statistics, the software simuPOP was used to simulate 20,000 times the neutral evolution of a bi-allelic locus evolving in the same conditions as in our experiment.
For each simulation, we simulated a biallelic locus evolving for 10 generations, in eight populations (our eight replicates), following an initial cross between males and females differing for their alleles to the considered genes (between the ISm and ISy strains). For each simulation, our summary statistic (the modified chi²) was computed in the same way as above, using the same expectation, and the number of lbspnm alleles for LbSPN, or of lbgap homozygous or heterozygous individual genotypes for LbGAP. As in the experiment, the populations' size was 10 females and 5 males for all generations.
p-values were obtained for each H1 hypothesis by comparing the observed statistics (modified chi²) to the corresponding null distribution. Since we tested the two H1 hypotheses ( < and > ), the p-value was multiplied by two (Bonferroni correction).
Here is the script used to perform these simulations.  Table S1. Values of correlations of bands to discriminant axes. Provided for the first two discriminant axes before and after partial correlation analysis. Cluster numbers are from the clustering analysis. Significance level: non-significant (n.s.). The two last columns indicate the significant correlations to axis 1 or 2 at the end of the analysis. Protein bands in bold represent the evolving protein bands at the end of the analysis. To be considered as an evolving protein band, the sign (+ or -) of the correlation had to be the same before and after the partial correlations, with a significance level lower than 0.05 before and after the partial correlations.

Band Cluster
Before partial correlation analysis After partial Correlation Analysis Summary of Correlations   The only assumption is mendelian inheritance. Figure S2. Synthetic scheme of the analysis of the evolution of venom composition. The global and specific approaches are described.

Specific approach
Simple genetic determinism Comparison of observed evolution to evolution simulated under the null hypothesis of neutrality Linear mixed model testing for differential evolution of LbGAP2 on the two hosts    Density H1: Frequency of the lbspny allele is lower than expected under neutrality H1: Frequency of the lbspny allele is higher than expected under neutrality (same as "Frequency of the lbspnm allele is higher than expected under neutrality") (same as "Frequency of the lbspnm allele is lower than expected under neutrality")