Redundancy analysis: A Swiss Army Knife for landscape genomics

Abstract Landscape genomics identifies how spatial and environmental factors structure the amount and distribution of genetic variation among populations. Landscape genomic analyses have been applied across diverse taxonomic groups and ecological settings, and are increasingly used to analyse datasets composed of large numbers of genomic markers and multiple environmental predictors. It is in this context that multivariate methods show their strengths. Redundancy analysis (RDA) is a constrained ordination that, in a landscape genomics framework, models linear relationships among environment predictors and genomic variation, effectively identifying covarying allele frequencies associated with the multivariate environment. RDA can be used at both individual and population levels, can include covariates to account for confounding factors and can be used to directly infer genotype–environment associations on the landscape. The modelling of both multivariate response and explanatory variables allows RDA to accommodate the genomic and environmental complexity found in nature, producing a powerful and efficient tool for landscape genomics. In this review, we outline the diverse uses of RDA in landscape genomics, including variable selection, variance partitioning, genotype–environment associations, and the calculation of adaptive indices and genomic offset. To illustrate these applications, we use a published dataset for lodgepole pine that includes genomic, phenotypic and environmental data. We provide an introduction to the statistical basis of RDA, a tutorial on its use and interpretation in landscape genomics applications, discuss limitations and provide guidelines to avoid misuse. This review and associated tutorial provide a comprehensive resource to the landscape genomics community to improve understanding of RDA as a modelling framework, and encourage the appropriate use of RDA across diverse landscape genomics applications. RDA is truly a Swiss Army Knife for landscape genomics: a multipurpose, adaptable and versatile approach to identifying, evaluating and forecasting relationships between genetic and environmental variation.