Predicting the Forest, but not the Trees: Genomecology of Populus trichocarpa

Accurate prediction of individual- and population-level adaptive phenotypic traits from genomic data can help mitigate the effects of climate change. We used phenotypes (bud flush, bud set and height growth) and whole-genome re-sequencing data for 525 P. trichocarpa trees sampled from natural stands along 16 rivers in the Pacific Northwest of North America to compare the predictive abilities (PAs, the correlations between predicted and observed phenotypic values) of models based on single-nucleotide polymorphisms (SNPs) to those of traditional genecological models. The overall ability to predict individual-tree phenotypes (i.e., incorporating variation across stands and rivers) was moderate to high (PA>0.5) using geography, climate, or SNPs as predictors. Average PAs based on SNPs were similar to those based on climate (0.708 versus 0.695), and only slightly higher than those based on geography (0.648). The relative advantage of SNP models was substantial (7%) for bud flush and marginal (1%) for bud set, whereas climate models were marginally better (1%) for height. Models combining SNP and climate information had the highest average predictive ability (0.713) and performed well across traits. Partitioning of PAs into hierarchical levels revealed that (1) none of the models were able to predict phenotypes for genotypes within stands; (2) models based on SNPs were consistently best at predicting stand-level phenotypes; and (3) all models were able to predict river-level phenotypes very well for each trait (PA>0.8). We illustrate the implications of these results by delineating “seed zones” based on genomic versus phenotypic, climatic and geographic data.