Analyses of trait variation in greenhouse and field conditions
We used data from three vegetative and two reproductive traits to analyse the drivers of intraspecific variation in greenhouse and field conditions. Vegetative traits were biomass, SLA and RSR (the latter only measured in greenhouse conditions), and reproductive traits were probability of flowering and fecundity. Biomass was estimated for all greenhouse and field individuals using leaf measurements and an equation obtained for a subset of plants (Appendix S3). Probability of flowering was modelled as a binary variable with data from the flowering vs. non-flowering plant status. Total inflorescence length was used as a proxy for fecundity, as we found a strong correlation between total inflorescence length and seed production (conditional R2 = 0.77; Appendix S3). In a preliminary analysis of field data, we found generally weak correlations among traits (Appendix S3). Thus we did not systematically consider trait covariation when analysing the sources of trait variation. However, the correlation between biomass and fecundity was moderately strong, so reproductive traits were analysed by controlling for biomass. This allowed us to assess size-independent reproductive investment (see below).
To analyse the effects of source and exposure environment on traits in the greenhouse, we applied 1) Linear Mixed Models (LMM) to plant biomass, SLA, fecundity and RSR and 2) Generalized Linear Mixed Models (GLMM) with a binomial error for probability of flowering (see details on Appendix S3). For each trait, we constructed a full model with four source environmental drivers (rotated components for Aridity, Temperature and Vegetation cover, and the binary variable Mowing), Water and Light treatments, interactions between environmental drivers and treatments, and Population as a random effect (Table S3). Full models for biomass, probability of flowering and fecundity included control biomass as a covariate. For a comparison of the role of genetic differentiation vs. plasticity, we assessed whether the effects of two source environmental drivers (Aridity and Vegetation Cover) were higher, similar to, or lower than the effects from their corresponding exposure treatments (Water and Light) and their interactions.
To test for the effects of environmental drivers on traits in field populations, we applied 1) LMMs to biomass, SLA and fecundity, and 2) GLMM with a binomial error distribution for probability of flowering (see details on Appendix S3). We constructed full models including the four source environmental drivers. To account for the possible influence of range (native vs. non-native), the models included the effect of range and its interaction with each environmental driver (Table S4). We added Population and Plot nested within Population as random effects. For probability of flowering and fecundity, we included biomass as a covariate.
Full models of the analyses with either greenhouse or field data were compared with all possible model subsets using the Akaike Information Criterion corrected for finite sample sizes (AICc) and the AICc weights (Burnham & Anderson 2002, Johnson & Omland 2004). We focused on the best AICc models, since they had high support and parameter values were overall consistent across competing models (see Appendix S3, Table S5, S6). Finally, we evaluated the utility of observational datasets to predict genetic differentiation. Genetic differentiation was considered predictable if the presence and direction of source environment effects on traits were the same in greenhouse and field conditions, and unpredictable otherwise. For probability of flowering and fecundity, we also assessed whether excluding the covariate biomass from the original analyses modified our evaluation of the predictability of genetic differentiation.