Computational protocol: Extremophile Poeciliidae: multivariate insights into the complexity of speciation along replicated ecological gradients

Similar protocols

Protocol publication

[…] Dissections to collect male, female, and offspring-related life-history traits followed well-established protocols (e.g., [, , ]. We collected the following male and female life-history traits: standard length (SL [mm]), dry weight (g), lean weight (g), fat content (%), and reproductive investment [%; for males: testis dry weight divided by the sum of reproductive tissue dry weight and somatic dry weight (gonadosomatic index, GSI); for females: offspring dry weight divided by the sum of offspring dry weight plus somatic dry weight (reproductive allocation, RA)]. We note that GSI for males and RA for females do not capture total investment into reproduction, as it obviously ignores other costs of reproduction, such as energetic costs related to searching for mates, sneak-mating, and intrasexual competition for mates []. This is particularly true for males, where the relative behavioral costs of reproduction may exceed those captured by GSI []. Thus, patterns in total reproductive costs may not necessarily mirror those estimated via GSI and RA.For females we also collected data on fecundity (number of developing offspring), offspring dry weight (mg), offspring lean weight (mg), and offspring fat content (%). Prior to statistical analyses we log10-transformed (male and female SL, male and female lean weight, and embryo dry and lean weight), square root-transformed (fecundity), or arcsine(square root)-transformed (male and female fat content, male and female GSI, embryo fat content) all life-history variables, and conducted subsequent z-transformation to meet assumptions of statistical analyses (i.e., these transformations facilitated normality of model residuals).To quantify morphological differentiation, lateral photographs were taken using Canon DSLR cameras with a macro lens, and we then digitized 13 landmarks on each image (Additional file : Figure S1) using the software program tpsDig2 []. We performed geometric morphometric analysis as described in [, ] (see Additional file : OSM 3 for details), resulting in seven principal component axes (= relative warps) explaining 91.5 % of the morphological variance as variables for the statistical analyses. [...] We first tested for differences in adult body size (SL) between populations by conducting mixed-model analysis of variance (ANOVA) that included the following independent variables: clade (four levels: G. affinis, G. holbrooki, G. eurystoma/G. sexradiata, and G. hubbsi), sex, H2S (present or absent), and “site nested within clade-by-H2S” [random effect, hereafter: site(clade × H2S)]. In this and all subsequent statistical models we initially included all potential two-way and three-way interactions, but removed terms from the final model in a step-wise process if P > 0.1, with the exception that a term with P > 0.1 would be retained if a higher-order interaction term involving that term had a P < 0.1. We then conducted three separate mixed-model multivariate analyses of covariance (MANCOVA) examining variation in adult life history, offspring life history, and body morphology (see below for model structure). Assumptions of multivariate normal error distribution and homogeneity of variances and covariances were met for all analyses performed. Statistical significance was determined using an F-approximation from Wilks’s lambda for all model terms with the exception that we used an F-test using restricted maximum likelihood and the Kenward-Roger degrees of freedom adjustment [] for clade, H2S, and clade × H2S to appropriately test these fixed effects while treating site as a random term (i.e., effectively treating site as the unit of replication; see also [, ]). The latter significance test was conducted using the MIXED procedure in SAS (SAS Institute, Cary, NC; sample code in the appendix of []), while all other tests were conducted in JMP (SAS). To quantify the relative importance of model terms, we calculated effect size using Wilks’s partial eta squared (ηp2) and calculated the relative variance as the partial variance for a given term divided by the maximum partial variance value in that model.The first model tested for differentiation based on adult life histories and included lean weight, fat content, and GSI as dependent variables. To control for multivariate allometry, standard length (SL) was added as a covariate, and we included clade, sex, and H2S as fixed factors, and site(clade × H2S) as a random effect. The second model tested for differentiation based on offspring-related life histories and included fecundity, embryo lean weight, and embryo fat content as dependent variables. The fixed factors for this model were clade and H2S, while site(clade × H2S) was included as a random effect, and the covariates were SL and ‘embryonic stage of development’ (see [] for details on embryo stages). The third model tested for phenotypic differentiation among sites based on body shape variation, using the seven retained relative warps as dependent variables. We tested for effects of centroid size to control for multivariate allometry and included clade, sex, and H2S as fixed factors, and site(clade × H2S) as a random effect. Running the analyses of adult life histories and body shape variation for both sexes separately confirmed our results and interpretations presented here (results not shown).To assess the nature and strength of convergent life-history and morphological divergence in response to H2S among clades and sites, we performed a canonical analysis of the H2S-term of each MANCOVA to derive divergence vectors following []. Each divergence vector describes the linear combination of dependent variables that exhibits the greatest difference between habitat types in Euclidean space, while controlling for other terms in the model (see Additional file : OSM 4 for details). We visualized shape variation described by the H2S term included in the MANCOVA with thin-plate spline transformation grids using tpsRegr []. [...] We used 11 nuclear microsatellite loci to genotype N = 382 fish from 17 sites (Table ; Fig. ) in all species except G. affinis (for which no alcohol-preserved material was available, see Additional file : OSM 1) using previously established protocols [, ] (see Additional file : OSM 5 for details). We used ARLEQUIN v 3.5 [] to calculate pairwise FST-values between populations in each drainage and to calculate standard indicators of genetic variability (see Additional file : OSM 5). STRUCTURE v 2.3.4 [] was employed to identify the number of genetically distinct clusters (K) in each drainage with the method outlined by [] using the web-based tool STRUCTURE HARVESTER v 0.6.8 []. Ten iterations per K were run using the admixture model with a burn-in period of 106 generations, followed by 106 iterations for K = 1 up to twice the number of sampling sites included per area. Each simulation was performed using an ancestry model incorporating admixture, a model of correlated allele frequencies, and no prior information on locations.To test whether divergent selection between sulfidic and non-sulfidic sites was associated with population genetic structure (i.e., higher FST) while controlling for phylogenetic differences between clades, we employed a partial Mantel test with pairwise FST-values obtained from FSTAT v 2.9.3 [] as dependent variable, habitat difference as independent variable (0 = same habitat type, 1 = different habitat type), and clade as the covariate (0 = same clade, 1 = different clade). For this analysis, we compiled a global dataset including all Gambusia species for which microsatellite data were available (i.e., without G. affinis). To standardize comparisons across clades, we only included up to a maximum of two non-sulfidic sites for each sulfidic system in the first model, so that the final dataset consisted of the populations 5–12 for G. holbrooki, 13 and 16–17 for G. eurystoma/G. sexradiata, and 23–24 and 26 for G. hubbsi (see Table  for more details). Due to the lack of neutral genetic differentiation in the populations from the Florida Panhandle (see results), we conducted a second partial Mantel test that excluded those populations. […]

Pipeline specifications

Software tools TpsDIG, TpsRegr
Application Macroscope & basic digital camera imaging
Diseases Drug-Related Side Effects and Adverse Reactions
Chemicals Hydrogen Sulfide