Computational protocol: Partitioning of herbivore hosts across time and food plants promotes diversification in the Megastigmus dorsalis oak gall parasitoid complex

Similar protocols

Protocol publication

[…] Population structure within the eight‐locus microsatellite data from M. dorsalis sp.2 was examined using the Bayesian model‐based clustering algorithm implemented in Structure v2.3.2 (Pritchard, Stephens, & Donnelly, ). Both diploid females and haploid males were included, with males coded as missing one allele per locus. Initial exploratory analyses assuming a single population were conducted to determine the value of λ (the parameter describing the allele frequency distribution) in models with or without admixture. In both cases, λ was 0.638, and this value was fixed in all subsequent analyses. The best choice of ancestry and allele frequency models was determined by conducting two runs for each combination of models with/without admixture and with correlated/uncorrelated allele frequencies for numbers of assumed populations (K) ranging from 1 to 8. In admixture models, alpha was estimated and assumed to be equal across populations. No prior information on population assignment was used. All runs had a burn‐in of 100,000 states followed by 1,000,000 iterations. Having determined the best model (no admixture with correlated allele frequencies), three further runs of this model were conducted for K from 1 to 8, giving five runs for each value. The value ΔK, following Evanno, Regnaut, and Goudet (), was used to guide selection of the best‐supported K value. Results for the full dataset were compared with those for females only to check that results were robust to violating the assumption that all samples have the same ploidy level. No difference in the results was detected, so we present results for the full dataset below.Our analyses reveal strong associations between the oak section on which a host gall developed and the genotype of associated M. dorsalis sp.2 individuals. However, a small number of individuals had nuclear/mitochondrial genotypic combinations that were inconsistent with the oak association of their host gall. We therefore conducted additional ancestry analyses in Structure incorporating host food plant information as a prior to assess whether such mismatches could be explained by introgression between populations attacking hosts on the different oak lineages. Six of the 112 individuals genotyped for microsatellites showed host associations inconsistent with expectations based on their nuclear genotype, so this frequency was used as the value for the migration prior (ν  =  0.05; the probability that an individual is genetically derived, or has recent ancestors, from the population attacking galls on the oak section it was not collected from). Two further analyses using ν = 0.1 and ν = 0.01 were also conducted to examine the robustness of the result to variation in the migration prior. Each analysis was run twice, with K = 2 and assessing ancestors back two generations. Results were consistent over all three values of ν, so we present results for ν = 0.05.Impacts of host traits on genetic variation within M. dorsalis sp.2 were assessed using analyses of molecular variance (AMOVA) in Arlequin v2.001 (Schneider, Roessli, & Excoffier, ). As with the Structure analyses, the second allele for haploid males was coded as missing. After demonstrating a lack of significant spatial substructure among Hungarian populations (see , Table a), all samples were pooled for the host trait analyses. We examined potential population structuring by three gall traits: the timing of host gall development (spring vs. late summer/autumn); the plant organ galled by the host gall wasp (using the categories acorn, bud, catkin, leaf, and shoot, following Bailey et al., ); and the major lineage of oak being galled. Since closely related gall wasps often share similar trait values (see Bailey et al., ), we reduced potential impacts of phylogenetic nonindependence by including the gall wasp genus (or major clade within the diverse genus Andricus, following Stone et al., ) as a level nested within the main comparison of interest. The full data structure for individuals in these analyses is provided in Appendix . [...] Individual‐level data across the M. dorsalis complex were collapsed into unique haplotypes, and individuals were allocated to one or other of the known cryptic species in this complex following phylogeny reconstruction using MrBayes v3.2.6 (Ronquist et al., ). One sequence was included from each of the other three Megastigmus species attacking oak gall wasps in the Western Palaearctic, M. dumicola (GenBank GU123593), M. stigmatizans (GenBank FJ026675), and M. synophri (GenBank GU123575), with M. synophri specified as the out‐group in the analysis following the phylogeny presented in Nicholls, Preuss, et al. (). Initial assessment of the base substitutions present in our data revealed limited informative variation in first or second positions, and some types of transversions were never observed. Hence, the data were divided into two partitions (combined 1st/2nd codon positions, and 3rd codon positions) with independent HKY + I + G substitution models estimated for each partition. Comparison of Bayes factors [estimated using twice the difference in the natural log of the harmonic mean of model likelihoods of each model (2ΔlnHML), and assessed following Table  of Kass and Raftery ()] indicated that a strict clock model incorporating a birth–death speciation process provided a better fit to the data than either a no‐clock model or a model fitting a strict clock incorporating a coalescent process (2ΔlnHML = 122 and 2ΔlnHML = 328, respectively), and that a relaxed clock was no better than a strict clock (2ΔlnHML = 1.44). Our final analysis thus utilized a model incorporating a strict clock and a birth–death process of lineage diversification, with relative substitution rates allowed to vary for each data partition. We carried out two independent MCMC runs, each comprising four chains (one cold, three heated, with a temperature setting of 0.12), running for 10 million generations and sampled every 2,000 generations. Convergence between runs, stationarity of parameters, and appropriate levels of chain swapping were assessed using Tracer version 1.6 (Rambaut & Drummond, ), and a 50% majority‐rule consensus tree was generated from samples taken during the last 3 million generations of each run.As with the microsatellite data, AMOVAs were conducted on the Hungarian cytb data to test for parasitoid genetic structuring in relation to host gall traits. All cytb AMOVA analyses were conducted both across all samples (to test for ecological divergence between the two cryptic species in the M. dorsalis complex) and within M. dorsalis sp.2 (to test for the presence of intraspecific host races). Again, after first determining that there was no spatial genetic structure within Hungary (Table b), we pooled relevant samples for analyses across and within species. Subsequent analyses testing for host‐related structure included host gall clade nested within the main trait being examined.Three AMOVAs were conducted using the UK cytb haplotypic data. The first compared samples derived from native UK host galls with samples from Hungarian host galls, to establish whether underlying geographic structure was present to allow discrimination between the local recruitment and host pursuit hypotheses. Subsequent AMOVAs examined genetic differentiation between UK M. dorsalis emerging from invading hosts and either (a) native UK hosts or (b) Hungarian hosts.A parsimony network was computed for the cytb haplotypes in each cryptic M. dorsalis species using the program TCS version 1.21 (Clement, Posada, & Crandall, ). Demographic signatures of population growth were tested using Tajima's D statistic for the same cytb datasets, calculated in Arlequin (Schneider et al., ). […]

Pipeline specifications