3.2 Sources of ITV in hypocotyl
Variance partitioning showed that site was mainly contributed to
variation of PC1 (63.75%; Fig. 4A), while variation of RTD was mainly
explained neither by site nor genealogy (69.03% in individual; Fig.
4E).
The linear-mixed model indicated that the predictor variables explained
58% the variation in PC1 (quadratic mixed model). Specifically,
MAT2 had the strongest negative effect on PC1 (effect
size = −0.55 ± 0.12 s. e., P < 0.001), while MAT had a
significant positive effect on PC1 (0.38 ± 0.16, P <
0.001). SAL had a significant negative effect on PC1 (−0.21 ± 0.09,P < 0.05; Fig. 4B). Bivariate relationships between MAT
and PC1 (OLR: R 2 = 0.42; QR:R 2 = 0.54) were much stronger than SAL and PC1
(R 2 = 0.01; Figs. 4C, 4D). In contrast, all
these predictive factors together explained only 12% of the variation
in the shape index, though significant effects of MAP (−0.30 ± 0.07) and
AGB (−0.15 ± 0.07) were indicated by the model (Figs. 4F).
Bivariate analysis also showed
that MAP and AGB had significant relation with RTD, but the models had
low explanation (R 2 = 0.06 andR 2 = 0.02, respectively; Figs. 4G, 4H).