Seeing the forest for the trees: Assessing genetic offset predictions from gradient forest

Abstract

There is an urgent societal need to better predict how specific genotypes perform in different environments. For example, in order for assisted gene flow to contribute to robust populations of harvested forests (Aitken & Bemmels, 2016) or key fish habitat reef systems (Matz et al., 2020), transplanted genotypes must be good candidates to increase the overall population fitness. Similarly, restoration of degraded ecosystems depends heavily on the identification of optimally adapted source populations if restoration efforts are to be successful (Houde et al., 2015). Likewise, climate change is a growing threat to biodiversity (Nunez et al., 2019;Urban et al., 2016) and there is a need to address environmental impacts on vulnerable populations. Most efforts to assess climate change impacts

Does the paradigm of genotype-environment associations need to be re-assessed? The paradox of adaptive phenotypic clines with non-clinal patterns in causal alleles

“…Many scientific challenges remain in our ability to adequately address this need, including those associated with identifying adaptive units (65), implementing restoration and assisted gene flow (66), and forecasting the vulnerability of populations (67–69). Various methods for genomic forecasting and genomic offset have been proposed to meet these challenges, but many of them incorporate GEA methods to quantify linear relationships between population-level allele frequencies and environments (52,69,70). Results from this study suggest that GEA methods may be sufficient to forecast broad scale geographic patterns, but may be insufficient to make a forecast about specific individuals or populations.…”

“…Analogous to genomic or marker-assisted selection based on genotype and phenotype (74–77), the multivariate RDA trait prediction based on genotype and environment does not require accurate knowledge of the genomic basis of adaptation, only that linked loci are included in the data. Note that the RDA-multivariate-trait prediction is an unsealed relative value for comparison among individuals, because SNP data only carries relative information between the two alleles at the locus (70). Whether this relative RDA-multivariate-trait prediction will be accurate in more complex environments, in the presence of nuisance variables such as unselective environments, with dominance, and/or with trait plasticity remains an important direction for future research.…”

Evaluating the persistence and utility of five wildVitisspecies in the context of climate change

“…Local offsets cannot be compared quantitatively across species (Láruson et al 2022), because they are estimated with different SNPs and bioclimatic information. They do, however, help predict whether specific alleles are expected to be adaptive in future climates.…”

“…In contrast, S a focuses solely on genetic composition by counting whether an inferred adaptive allele is present at each position. Second, local offsets cannot be compared across species (Láruson, Fitzpatrick, Keller, Haller, & Lotterhos, 2022), but S a can because it reports a proportion – i.e., of adaptive alleles across candidate climate-related SNPs.…”

“…We also selected accessions as persistent if they had local offsets and genomic load values below 90%, 50% and 25% of their distribution across species. Since local offsets are not comparable between species (Láruson et al, 2022), the thresholds for local offsets were considered only within species. Finally, accessions were included when they had S a or Ne above the 10%, 50% and 75% of the distributions across species.…”