2.5 | Genome scan analyses
Genomic patterns related to host selection were first investigated with
Pool-Seq data using the procedure detailed in Carlier et al, 2021b.
Briefly, two different genome scan approaches were used: the first,
named BayPass
(Gautier, 2015), is a
genotype-environment association method, while the second, named
PoolFreqDiff (Wiberg
et al., 2017), is a differentiation-based method. The population pairs
collected in 2011 or in 2013 were used in the same analysis using the
two approaches to detect convergent signals between the replicates and
to limit the detection of false positives. For BayPass, we used the
standard covariate model with a qualitative covariate (called Cov-co)
corresponding to the cultivars of origin (coded 1 for resistant and -1
for susceptible). Analysis with the samples collected in 2011 were also
run with two other quantitative covariates (called Cov-dS and Cov-dR )
corresponding to the least-squares means (LSMeans) of DLA-S or DLA-R
estimated in Dumartinet et al. 2019. Three independent runs were
conducted for all the BayPass analyses and produced very close results.
For the poolFreqDiff method, we rescaled all the allele counts to the
effective sample size (neff), as recommended by the authors
(Wiberg et al., 2017).
From the p-values estimated in the above analyses, we identified
putative genomic regions under host selection using the local score
approach (Fariello et
al., 2017) which accounts for linkage disequilibrium from Pool-Seq data.
Several values of the tuning parameter ΞΎ (1, 2 or 3) were used as
recommended in simulations
(Bonhomme et al.,
2019).