Two common, often coexisting grassland plant species differ in their evolutionary potential in response to experimental drought

Abstract For terrestrial plant communities, the increase in frequency and intensity of drought events is considered as one of the most severe consequences of climate change. While single‐species studies demonstrate that drought can lead to relatively rapid adaptive genetic changes, the evolutionary potential and constraints to selection need to be assessed in comparative approaches to draw more general conclusions. In a greenhouse experiment, we compare the phenotypic response and evolutionary potential of two co‐occurring grassland plant species, Bromus erectus and Trifolium pratense, in two environments differing in water availability. We quantified variation in functional traits and reproductive fitness in response to drought and compared multivariate genetic variance–covariance matrices and predicted evolutionary responses between species. Species showed different drought adaptation strategies, reflected in both their species‐specific phenotypic plasticity and predicted responses to selection indicating contrasting evolutionary potential under drought. In T. pratense we found evidence for stronger genetic constraints under drought compared to more favourable conditions, and for some traits plastic and predicted evolutionary responses to drought had opposing directions, likely limiting the potential for adaptive change. Our study contributes to a more detailed understanding of the evolutionary potential of species with different adaptive strategies in response to climate change and may help to inform future scenarios for semi‐natural grassland ecosystems.


Prior choice, model specifications and MCMC diagnostics
Literature shows manifold possibilities for prior specification (see Bolker et al. 2009, Stinchcombe et al. 2013).We investigated priors where the list of the diagonal elements (V) for the residual and random effects was 0.5 × the phenotypic trait variance (Vp).We further tested two different degrees of freedom (df) for the hypothesized inverse Wishart distribution.Depending on the number of dimensions of the matrices being estimated, i.e. 11 for Bromus and 10 for Trifolium, df were represented by 0.01 or 10.01 for Bromus and 0.009 or 9.009 for Trifolium.Additionally, we explored parameter-expanded priors, with prior means set to zero and the diagonal of the prior covariance matrix set to 2500 (Tab.S2).
For each model, we ran three independent chains using a burn-in period of 160.000 followed by 960.000MCMC steps and a thinning of 200, resulting in a final sample size for the posterior distribution of estimates of 4000.Convergence and mixing of the three chains were tested by visual inspection of trace plots, Gelman and Rubin's convergence diagnostics and autocorrelation tests (Gelman and Rubin 1992, Brooks and Gelman 1998, Hadfield 2019).As convergence was obtained in all cases, model outputs from all three chains were combined to obtain posterior distributions of all estimates.
For the results presented below we selected parameter-expanded priors, with df = 10.01 for Bromus and df = 9.009 for Trifolium, both with diagonal matrices of 2500.These prior specifications showed the best combination of a low deviance information criterion and a well mixing behaviour for both species (Tab.S3; see also all model fits in Tab.S4 -S11).In most cases, differences in priors and distributions had little effect on both, quantitative estimates and biological interpretations.
Table S6: Bromus erectus -G-Matrix of prior specification G11e2: Posterior mean MCMC estimates of G for both treatments (CON & DRY), with 95% Bayesian Credible intervals.Traits with significant treatment effect in genetic variances coloured in yellow; significant genetic covariances between trait pairs except with relative fitness coloured in green; significant genetic covariances with relative fitness in blue.

Figure S1 :
Figure S1: Response to selection Δz for Bromus erectus and Trifolium pratense.Quantitative traits are

Table S1 :
Coordinates of seed collection locations for each natural source population of B. erectus and T. pratense.

Table S2 :
Summary of linear mixed effect model and generalized linear mixed effect model (*) analyses

Table S3 :
Summary of linear mixed effect model and generalized linear mixed effect model (*) analyses evaluating the effect of treatments (Control & Drought) on functional trait means including the partitioning of genetic variances at source population level.For each model, estimates for intercept (Int) and simulated 95% Credible Intervals (CI) are shown for the fixed effects, estimated standard deviations (SD) are given for the random effects (block, seed family -SF, seed family treatment interaction -SF:treat, natural source population -pop, natural source population treatment interaction -pop:treat, residuals -resid).Traits with significant treatment effect are printed in bold, where higher and lower mean trait values are highlighted in blue and grey, respectively.Results demonstrate that for the majority of trait and species combinations the among-population component is smaller than the within-population component.

Table S4 :
Priors tested to best fit the model.We decided to use prior specification G11e2 and G10e2 for Bromus and Trifolium, respectively (coloured in blue) for genetic variance-covariance matrices.These models revealed to have the best combination of a low deviance information criterion (see Tab. S2) and well-behaved posterior distributions.

Table S7 :
Bromus erectus -G-Matrix of prior specification G11vp2: Posterior mean MCMC estimates of G for both treatments (CON & DRY), with 95% Bayesian Credible intervals.Traits with significant treatment effect in genetic variances coloured in yellow; significant genetic covariances between trait pairs except with relative fitness coloured in green; significant genetic covariances with relative fitness in blue.

Table S8 :
Bromus erectus -G-Matrix of prior specification G11vp2.2:Posterior mean MCMC estimates of G for both treatments (CON & DRY), with 95% Bayesian Credible intervals.Traits with significant treatment effect in genetic variances coloured in yellow; significant genetic covariances between trait pairs except with relative fitness coloured in green; significant genetic covariances with relative fitness in blue.

Table S9 :
Bromus erectus -G-Matrix of prior specification G11e3: Posterior mean MCMC estimates of G for both treatments (CON & DRY), with 95% Bayesian Credible intervals.Traits with significant treatment effect in genetic variances coloured in yellow; significant genetic covariances between trait pairs except with relative fitness coloured in green; significant genetic covariances with relative fitness in blue.

Table S11 :
Trifolium pratense -G-Matrix of prior Specification G10vp2: Posterior mean MCMC estimates of G for both treatments (CON & DRY), with 95% Bayesian Credible intervals in brackets.Traits with significant treatment effect in genetic variances coloured in yellow; significant genetic covariances between trait pairs except with relative fitness coloured in green; significant genetic covariances with relative fitness coloured in blue.

Table S14 :
G-Matrix comparison statistics: Descriptors of multivariate genetic variation -number of dimensions nD, maximum evolvability emax, and total genetic variance tgv for each treatment.Significant treatment effects (CI does not overlap other treatment mean) are coloured in blue.

Table S15 :
Bromus erectus -Quantitative trait heritabilities H² for all prior specifications (choice in bold):Posterior mean MCMC estimates for both treatments (CON & DRY), with 95% Bayesian Credible intervals in brackets.Trait estimates with significant treatment effect coloured in blue.