Statistical analysis
We used mixed-effect ANOVA to examine the effects of region of origin (Italy vs. Fennoscandia), maternal environment (greenhouse vs. Italy field site) and their interaction (fixed effects), and population nested within region of origin and its interaction with maternal environment (random effects) on germination proportion 12 weeks after seed maturation. The analysis was based on line means. We included in the analysis only populations with germination proportions available for both the greenhouse and the Italian maternal environments (43 populations in total). Seeds from the Swedish field site could not be included due to low seed production of Italian populations. The response variable was arcsine square-root transformed prior to analysis to obtain a normal distribution of residuals. Because of heteroscedasticity, variance was allowed to vary among regions of origin. Statistical significance of fixed explanatory variables was determined by F-tests with type III Sums of Squares and Kenward-Rogers adjustment for degrees of freedom, while significance of random factors was tested using Likelihood Ratio Tests between full and reduced models. The analysis was performed using the R package ‘nlme’.
Additionally, we tested the effect of maternal environment, population and their interaction on germination proportions of Fennoscandian populations collected at the two field sites. We included in the analysis only populations with germination proportions available for both field maternal environments (18 populations in total). The response variable was arcsine square-root transformed and statistical significance of the effect variables was determined by F-tests with type III Sums of Squares.
To assess the association between climate at the site of origin and germination proportion up to 12 weeks after seed maturation, we first conducted a scaled and centered principal component analysis (PCA) among the five climatic variables for each of the two regions separately using the R function prcomp from the ‘base’ package. The first three Principal Components (PCs) cumulatively explained 99% of the total variation in both regions (Table S3 ). We then used these PCs as independent variables in multiple regressions with mean germination proportion of populations (based on line means) as response variable. We chose to use principal components rather than the original climatic variables as predictors in the multiple regression to avoid the problem of collinearity among the chosen climatic variables. In addition, to assess spatial variation in seed dormancy within regions, we tested the relationship between population mean germination proportions, and latitude, longitude and their interaction using linear models analyzed separately by region. These analyses were restricted to estimates of seed germinability up to 12 weeks after seed maturation. Estimates of dormancy obtained soon after seed maturation are arguably the estimates most directly related to dormancy of seeds in natural environments due to post-dispersal environmental effects and their interaction with the maternal environment on seed dormancy release (Postma et al.2016; Coughlan et al. 2017; Buijs et al. 2020). Hence, the correlation between environmental conditions at the site of origin and estimates of seed dormancy is likely to decrease with time after seed maturation in our experiments.
Seeds produced in the greenhouse by plants in cohort 1 had markedly higher germination proportions than seeds produced by the other cohorts, most likely due to absence of the vernalization treatment. However, removal of cohort 1 from the dataset did not affect the statistical significance nor markedly change effect sizes, and the analyses presented below include cohort 1.
All statistical analyses were conducted in R version 3.4.0 (R Core Team, 2017).