Reciprocal Evaluation of Genomic Offset Predictions of Climate Maladaptation with Independent Empirical Datasets.

Abstract

AbstractGenomic offsets are increasingly being used to forecast maladaptation expected from the decoupling of gene-environment associations caused by an abrupt shift in climate. Such gene-environment mismatches can arise temporally from rapid climate change in situ and also spatially through the introduction of nonlocal propagules to a new site. Studies have begun to evaluate genomic offsets using ground-truth observations of fitness traits measured in common gardens. However, empirical common garden evaluations of genomic offset predictive performance using independent training and testing data remains rare, and to our knowledge no studies have conducted fully reciprocal comparisons derived from replicated genomic and common garden data from independent sample sets. Here, we report an evaluation experiment of genomic offsets using red spruce (Picea rubens) based on two independently generated exome-capture datasets from different range-wide sets of populations. For each dataset, we train a gradient forest model using climate predictors and generate spatial genomic offset predictions for (1) common gardens planted with the same populations (within-set evaluations) and (2) common gardens planted with populations not used in model training (between-set evaluations). By leveraging multiple gardens planted at different times and locations, we also explore how predictive performance varies across garden environments and fitness proxies. We find the expected negative correlation between genomic offset and fitness across most comparisons, with the strongest associations for juvenile growth followed by adult survival. Our approach presents an important step forward for common garden evaluations of genomic offset and their ability to predict maladaptation under environmental change.