decreased to 3% or lower when including larger numbers of
accessions (>500).
47. The genetic relationships among Arabidopsis accessions
(genetic structure) can be downloaded from the admixture
map available at the 1001 genomes database. Alternatively,
the genetic structure can be determined using a subset of
genome-wide SNPs extracted from the 1001 genomes database, by using the PCA or kinship matrix options of TASSEL
(see Table 1).
48. Most environmental variables (e.g., climatic variables) are not
evenly distributed across geography, leading to significant correlations between values of each variable at neighboring locations (referred to as spatial autocorrelation), and to a lack of
independence of observations [73, 74]. Similarly, Arabidopsis
genetic and genomic diversity that is not involved in adaptation
shows strong spatial autocorrelation reflecting the demographical history (dispersion) of the species. This is shown by the
significant correlations found in Arabidopsis between geographic and genetic distances (so-called isolation by distance
pattern), and by the strong geographic structure of the multiple genetic groups detected among worldwide accessions
[22, 75]. Therefore, statistical methods testing the association
between genotypic and environmental data must take such
spatial autocorrelations into account. Otherwise, significant
associations are likely due to confounding effects derived
from the overall correlation between the genetic diversity and
environmental variation, owing to the demographic history of
natural populations and the geographic patterns of environmental factors.
49. Numerous statistical methods have been developed to test for
environment-genotype associations (reviewed in [76]). These
include simple tests, like logistic or standard regressions, as well
as general or mixed linear models, where the confounding
effects of the genetic structure are corrected by including a
covariance matrix with the admixture memberships to each
genetic group; the PCA values for the 3–10 main principal
components differentiating the major genetic groups; or a
pair-wise kinship matrix (see Note 47) [22, 23, 34, 56,
75]. However, to reduce the number of false associations, it
is convenient to use specific methods developed to test
environment-genotype associations (see Table 1) because environmental variables do not follow the same spatial patterns
than genetic variables.
50. Methods specifically developed to test environment-genotype
associations, such as LFMM or Bayenv2, require including
genome-wide SNPs that are used to determine the general
108
Carlos Alonso-Blanco et al.
accessions (>500).
47. The genetic relationships among Arabidopsis accessions
(genetic structure) can be downloaded from the admixture
map available at the 1001 genomes database. Alternatively,
the genetic structure can be determined using a subset of
genome-wide SNPs extracted from the 1001 genomes database, by using the PCA or kinship matrix options of TASSEL
(see Table 1).
48. Most environmental variables (e.g., climatic variables) are not
evenly distributed across geography, leading to significant correlations between values of each variable at neighboring locations (referred to as spatial autocorrelation), and to a lack of
independence of observations [73, 74]. Similarly, Arabidopsis
genetic and genomic diversity that is not involved in adaptation
shows strong spatial autocorrelation reflecting the demographical history (dispersion) of the species. This is shown by the
significant correlations found in Arabidopsis between geographic and genetic distances (so-called isolation by distance
pattern), and by the strong geographic structure of the multiple genetic groups detected among worldwide accessions
[22, 75]. Therefore, statistical methods testing the association
between genotypic and environmental data must take such
spatial autocorrelations into account. Otherwise, significant
associations are likely due to confounding effects derived
from the overall correlation between the genetic diversity and
environmental variation, owing to the demographic history of
natural populations and the geographic patterns of environmental factors.
49. Numerous statistical methods have been developed to test for
environment-genotype associations (reviewed in [76]). These
include simple tests, like logistic or standard regressions, as well
as general or mixed linear models, where the confounding
effects of the genetic structure are corrected by including a
covariance matrix with the admixture memberships to each
genetic group; the PCA values for the 3–10 main principal
components differentiating the major genetic groups; or a
pair-wise kinship matrix (see Note 47) [22, 23, 34, 56,
75]. However, to reduce the number of false associations, it
is convenient to use specific methods developed to test
environment-genotype associations (see Table 1) because environmental variables do not follow the same spatial patterns
than genetic variables.
50. Methods specifically developed to test environment-genotype
associations, such as LFMM or Bayenv2, require including
genome-wide SNPs that are used to determine the general
108
Carlos Alonso-Blanco et al.