Analysis of leaf trait variation
We first evaluated if traits were either normally or log-normally distributed using the ‘fitdist’ function in the ‘fitdistrplus’ R package (Delignette-Muller & Dutang, 2015), with the highest log-likelihood scores indicating the best-fit data distribution. For traits that were normally distributed, we calculated descriptive statistics as means and standard deviations (SD), while medians, median and SD values were calculated for log-normal traits. We also calculated and present trait ranges and coefficients of variation (CV) for each trait. In addition we tested for differences in mean trait values as a function of planting rows, using analysis of variance (ANOVA). These descriptive statistics and ANOVAs were complemented by variance partitioning analyses, which were used to identify the primary sources of trait variation in our dataset. These analyses entailed fitting linear mixed models (using the ‘lme’ function in the ‘nlme’ R package (Pinheiro, Bates, DebRoy, Sarkar, & Team, 2017)) to each trait individually (where n =45 in all cases), where an intercept was the only fixed effect, and plant identity within sample row identity were included as nested random effects. The proportion of trait variation explained by each nested random effect, as well as the proportion of variation unexplained by the factors considered here, was then estimated using the ‘varcomp’ function in the ‘ape’ R package (Paradis, Claude, & Strimmer, 2004).
We then assessed the influence of soil compaction on both individual traits and multivariate trait syndromes. First, we fit separate mixed models for each trait individually, where trait values were predicted as a function of soil bulk density (included as a fixed effect), while accounting for plant identity nested within sampling row (included as nested random effects). We then used a Principal Component Analysis (PCA) to test if multivariate leaf trait syndromes of ‘Chardonnay’ varied as a function of planting row/compaction or vine identity. Our PCA included four LES traits (i.e., A mass,R mass, leaf N, and LMA), two traits derived from light response curves (i.e., Φ and LLCP), and one trait related to light interception (i.e., leaf area). The relationship between individual traits and each PCA axis was evaluated using the ‘dimdesc’ function in the ‘FactoMineR’ R package (Le, Josse, & Husson, 2008). Then, we used a permutational multivariate analysis of variance (PerMANOVA) based on Euclidean distances and 10,000 permutations, to test if multivariate trait syndromes vary as a function of planting row (reflecting the soil bulk density gradient), plant identity, and a row-by-plant interaction term. Based on our PerMANOVA results, we also visualized our PCA biplot with 95% confidence ellipses surrounding the data points within each sample row. All multivariate analyses were implemented in the ‘vegan’ R package (Oksanen et al., 2017).
Finally, we evaluated bivariate trait relationships among ‘Chardonnay’ leaves using Pearson correlation tests. However, one of the main goals of our analysis was to test for the presence of an intragenotype LES in Chardonnay, and evaluate if these relationships differed from the global LES defined by Wright et al. (2004). For this, we used standardized major axis (SMA) regressions to evaluate the slope of the bivariate trait relationships that exist among four LES traits measured in our study vines, including A mass,R mass, LMA, and leaf N. All of these SMA regressions had a sample size of n =45, and were implemented with the ‘sma’ function in the ‘smatr’ R package (Warton, Duursma, Falster, & Taskinen, 2012). Then, we evaluated if these SMA slopes of the intragenotype ‘Chardonnay’ LES differed from the trait relationships found across a functionally and phylogenetically diverse set of plants species globally. This analysis entailed merging our ‘Chardonnay’ data with the GLOPNET dataset (i.e., the dataset used in the original LES analysis of Wright et al. (2004)), and testing for statistical differences in the SMA slope of each trait-trait relationship between the ‘Chardonnay’ vs. GLOPNET, using the ‘slope.test’ function of the ‘smatr’ R package (Warton et al., 2012).