**3. The adaptive potential of wild relatives and landraces in common bean**

In common bean, ecological gradients related with drought stress are associated with divergent selection at the genetic level, after accounting for gene pool and subpopulation structure. This divergent selective pressure might be a consequence of local-level rainfall patterns. Specifically, in tropical environments near the equator with bimodal rainfall, a mid-season dry period occurs that can last 2–4 weeks. In contrast, in the sub-tropics, a dry period of 3 or more months can occur. In response to this mid-cycle drought of the sub-tropics, *P. vulgaris* enters a survival mode of slow growth and reduced physiological activity until rainfall resumes and flowering occurs [43]. Beans growing in wetter conditions on the other hand are less frequently subjected to these environmental pressures and have a fitness advantage to mature in a shorter length of time. Given these ecological differences, and consistent with genomic signatures of divergent selection, the reaction typically associated with drought tolerance, although favorable under dry conditions, seems detrimental under more humid conditions. The awareness about this trade-off may aid the breeding of new drought-tolerant varieties specifically adapted to unique microenvironments (e.g. [44]) and local regions rather than varieties eventually obsolete, originally intended for a wider range of environments.

In the next two sub-sections, we summarize the concrete evidence supporting these statements (Section 3.1) and explain how we can discard other fortuitous causes that may also explain the same pattern (Section 3.2), based on the approaches that we introduced in the previous section (Section 2).

#### **3.1. The signatures of adaptation in common bean are widespread throughout the genome**

SNP markers are good at recovering the well-described Andean and Mesoamerican gene pool structure and the five within-gene pool subpopulations observed in wild common bean [41]. Because of this, in a previous research by us with more than 22,000 SNP markers, QQ-plots, from the association analyses between those SNP markers and a bioclimatic-based drought index [12], indicated that GLM analyses likely had excessive rates of false positives whereas MLM models controlling for population structure and using a kinship matrix reduced more effectively the false positive rate.

Watterson's theta (θ) estimator [36], and Tajima's D [35] can be computed using the software TASSEL v.5 [37] and customized R scripts. Results of all windowed analyses are usually plotted against window midpoints in millions of base pairs (Mb) in the software R v.3.3.1 (R Core Team). The centromeres can be marked to visualize the extent of the centromeric repeats and

It is advisable to calculate bootstrap-based means and 95% confidence intervals around the mean for some summary statistics (i.e., SNP density, π, θ, and Tajima's D) when computed in sliding windows that contained or did not contain at least one marker that was associated with a bioclimatic-based index. For this, each summary statistic of windows containing and not containing associated SNPs should be randomly resampled with replacement (bootstrapping) across windows within grouping factors (associated vs. not associated). The overall mean is then stored for each grouping factor. This step should be iterated at least 1000 times using customized R scripts. Bootstrapping must be performed independently for each sum-

**3. The adaptive potential of wild relatives and landraces in common bean**

In common bean, ecological gradients related with drought stress are associated with divergent selection at the genetic level, after accounting for gene pool and subpopulation structure. This divergent selective pressure might be a consequence of local-level rainfall patterns. Specifically, in tropical environments near the equator with bimodal rainfall, a mid-season dry period occurs that can last 2–4 weeks. In contrast, in the sub-tropics, a dry period of 3 or more months can occur. In response to this mid-cycle drought of the sub-tropics, *P. vulgaris* enters a survival mode of slow growth and reduced physiological activity until rainfall resumes and flowering occurs [43]. Beans growing in wetter conditions on the other hand are less frequently subjected to these environmental pressures and have a fitness advantage to mature in a shorter length of time. Given these ecological differences, and consistent with genomic signatures of divergent selection, the reaction typically associated with drought tolerance, although favorable under dry conditions, seems detrimental under more humid conditions. The awareness about this trade-off may aid the breeding of new drought-tolerant varieties specifically adapted to unique microenvironments (e.g. [44]) and local regions rather than varieties eventually obsolete, origi-

In the next two sub-sections, we summarize the concrete evidence supporting these statements (Section 3.1) and explain how we can discard other fortuitous causes that may also explain the same pattern (Section 3.2), based on the approaches that we introduced in the

**3.1. The signatures of adaptation in common bean are widespread throughout the genome**

SNP markers are good at recovering the well-described Andean and Mesoamerican gene pool structure and the five within-gene pool subpopulations observed in wild common bean [41]. Because of this, in a previous research by us with more than 22,000 SNP markers, QQ-plots,

its correlation with overall patterns of diversity and divergence.

120 Rediscovery of Landraces as a Resource for the Future

mary statistic in order to eliminate correlations among these.

nally intended for a wider range of environments.

previous section (Section 2).

In that particular case, the MLM model with the first two PCoA axes scores used as covariates was the best at controlling for false positives. This model yielded a total of 115 SNP markers associated with the bioclimatic-based drought index at a Bonferroni-corrected significance threshold of 7.36 –log10(P-value). These markers explained on average 51.3% ± 0.4 of the variation in the bioclimatic-based drought index. The 115 SNPs were clustered in 90 different regions, defined as overlapping 1000 bp sections that flanked associated markers (**Figure 3**). Associated SNPs and regions were widespread in all 11 common bean chromosomes.

Following the previous example, chromosomes Pv3 and Pv8 had the highest number of associated SNPs with 21 and 32 SNPs clustered in 16 and 21 different regions, respectively. Chromosomes Pv1, Pv2, Pv4, Pv5, Pv6, and Pv9 contained an intermediate number of associated SNPs with 11, 6, 11, 7, 12, and 9 SNPs clustered in 11, 6, 8, 6, 8, and 9 different regions, respectively. Chromosomes Pv7, Pv10, and Pv11 had the fewest number of associated SNPs with 3, 2, and 1 SNPs clustered in 3, 1, and 1 different regions, respectively. Chromosome Pv8 had more regions with at least 2 associated SNPs than any other chromosome, and these regions had more associated SNPs than in any other chromosome for a total of 5 regions with an average number of associated SNPs of 3.2. The single region that contained more associated SNPs was also situated in chromosome Pv8 with 6 SNPs explaining on average 51.1% ± 0.3 of the variation in the bioclimatic-based drought index. After chromosome Pv8, Pv3 was also outstanding, having 4 regions (with at least 2 associated SNPs) with an average number of associated SNPs of 2.5. Therefore, a total of 75 regions, comprising 99 SNP markers associated with the bioclimatic-based drought index, contained at least 1 gene for a total of 77 genes. Most genes were in chromosomes Pv1, Pv3, and Pv8 with 11, 14, and 16 genes. Only two regions, at chromosomes Pv1 and Pv8 and containing a total of seven different SNPs, spanned two or more genes. The one in Pv8 was the region with more associated SNPs (six in total). One of the two genes in this region encoded an Ankyrin repeat-containing protein, which was associated with osmotic regulation via the assembly of cation channels in the membranes [45]. Among other identified candidate genes, there was a phototropic-responsive NPH3 gene [46] in Pv3.
