Phylogenetic effects in chromosome number varied among examined clades, but were generally high. Chromosome number resulted poorly related to large scale climatic conditions, while a stronger relation with categorical variables was found. Specifically, open, disturbed, drought-prone habitats selected for low chromosome numbers, while perennial herbs, and especially woody plants, living in shaded, stable environments were associated with high chromosome numbers.
Altogether, our findings confirm our expectations and we argue that environmental stability favour higher recombination rates with respect to unstable environments. In addition, by comparing results of models testing for evolvability of 2n and of x, we provided insight into the ecological significance of polyploidy.
We investigated whether the evolution towards optima of diploid (2n) and basic (x) chromosome numbers is influenced by climatic variables (continuous predictors), habitat characteristics or plant traits (categorical predictors) within different angiosperm clades.
To this end, we used the phylogenetic comparative method implemented in the R program SLOUCH, conceived to study adaptive evolution of a trait along a phylogenetic tree (Hansen et al., 2008). The method assumes that the response trait evolves as if by an Ornstein–Uhlenbeck model of adaptive evolution, towards a ‘primary’ optimum θ, defined as the optimal state that species will approach in a given niche when ancestral constraints have disappeared (Hansen, 1997).
Phylogenetic trees are scaled to 1.0 total length (from the root to the tip in the ultrametric tree) to facilitate the interpretation of parameter estimates. The two main parameters returned by the model are the phylogenetic half-life (t1/2) and the stationary variance (vy). Phylogenetic half-life indicates the time it takes for half the ancestral influence on a trait to evolve towards the predicted optimal phenotype (Hansen, 1997). A half-life above zero indicates that adaptation is not instantaneous, while when t1/2 = 0 means that there is no evolutionary lag. The stationary variance is the stochastic component of the model and can be interpreted as evolutionary changes in the response trait induced by genetic drift or unmeasured selective forces.
Phylogenetic half-life in a model that only includes the intercept is a measure of the phylogenetic effect in the response trait. In such a model, a half-life = zero means that the response variable is not phylogenetically structured, while a half-life > 0 indicates that there is an influence of phylogeny on the data; when the half-life shows high values, this can attributed to an underlying continuous Brownian motion process.
The intercept-only model is contrasted with a model that also includes a predictor variable. This type of model is regarded as adaptation model because it tests whether the response traits evolve towards optima influenced by a predictor. By comparing a model with and without the inclusion of predictor variables, it is possible to determine how much of the phylogenetic signal can be attributed to phylogenetic inertia. No reduction in the t1/2 suggests that phylogenetic signal can be entirely attributed to phylogenetic inertia; on the contrary, when a trait evolves in response to a variable, a reduction in the half-life for the response trait should be observed.
The adaptation models which include continuous predictors (i.e. the climatic variables in our study) are fitted using maximum likelihood for estimation of t1/2 and vy and generalized least squares for estimation of the regression parameters in an iterative procedure (Hansen et al., 2008). In addition, the SLOUCH model assumes that the predictors have a longer phylogenetic half-life than the model residuals, and this is well supported by the variables involved in our study. The model returns parameters of an optimal regression and of a phylogenetic regression. The former is the relationship between the response and predictor variable that is predicted to evolve free of ancestral influence (absence of inertia). Therefore, the slope of this regression must be steeper than that of the phylogenetic regression.
To evaluate the effect of the categorical predictors on the evolution of chromosome number, the ANOVA and ANCOVA extensions implemented in SLOUCH were used. Categorical predictors were mapped onto the phylogeny using parsimony reconstruction.
Intercept-only models are compared to the adaptation models using the Akaike information criterion corrected for small sample-size (AICc); models with AICc scoring less than two units lower are considered substantially better while if more than two units lower than they are considered significantly better (Burnham & Anderson, 2004). Finally, model interpretations were based on comparisons of t1/2 and vy of the adaptation models with the intercept-only model, together with the amount of variation in chromosome number that the models explain. All statistical analyses have been carried out in R v3.2.3.
Although the majority of variation remains unexplained, in the light of strong phylogenetic inertia, even accounting for a small amount of the total variation, indicates that a Brownian motion process inadequately explains the evolution of chromosome number due to the action of natural selection (Felsenstein, 1985; Hansen, 1997).
The dataset includes more than half of Italian endemics covering 27 orders and 6 main angiosperms clades, which makes this interspecific study a synoptic assessment of the relation among chromosome number, plant traits and environmental factors in angiosperms.
At this stage, our study affords non-stochastic demonstrations for chromosome number variation. In addition, whilst phylogeny is a strong predictor of trait value, especially for x, a simple phylogenetic explanation is inadequate.