Figure 1.
Mean sample heterozygosity by population.
Mean sample heterozygosity for each sample was calculated as the number of heterozygote genotypes divided by the total number of non-missing genotypes for the set of 2,663 structure inference SNPs. The sample sizes are 89 ASIAN, 59 YRI, 56 NA, 60 CEU and 1301 CR, respectively.
Figure 2.
Spectrum plot of admixture coefficients.
The admixture coefficient for each sample was the average of 10 independent unsupervised STRUCTURE runs, all with K = 4, which turned out to be the optimal K for the total sample set including CR samples and four continental reference populations (see Figure S1).
Figure 3.
Principal component analysis result.
The analysis was conducted on a combined set of samples (the same as the one used for STRUCTURE analysis shown in Figure 2) with 2,663 structure inference SNPs. The top 4 principal components explain a total of 78.5% of the variance in the data, and the corresponding eigenvectors are shown in pairwise scatter plots in this figure.
Figure 4.
Triangular plot of the admixture coefficients.
The supervised STRUCTURE analysis was performed on a refined CR sample set together with three continental reference population sample sets. Coordinates on each axis indicate the admixture proportions from European (CEU), African (YRI) and Native American (Colombian and Mayan) ancestral populations. Each green circle inside the triangle corresponds to a CR sample, and the coordinates on three axes for each sample always add up to 1. The reference samples are at each vortex of the triangle (not shown) because we fixed those as the ancestral populations and inferred three proportion numbers for each CR sample in this analysis.
Figure 5.
Correlation between STRUCTURE admixture coefficients and PCA eigenvectors for CR samples.
Both STRUCTURE analysis and PCA are based on the same data set as detailed in the legend of Figure 4. Pearson correlation coefficients are −0.99 for CEU admixture proportion versus EV1 (left panel) and 0.96 for NA admixture proportion versus EV2 (right panel).
Figure 6.
Substructure mirrors geographic location.
The principal components analysis was performed on the refined CR samples (excluding 56 with greater than 10% Asian ancestry) together with three reference populations (CEU, NA and YRI). Three cantones (Nicoya, Santa Cruz and Tilaran) in Guanacaste are shown to have distinct distributions in the EV1versus EV2 scatter plot.
Figure 7.
Comparison of the average LD block size for each chromosome between the two subpopulations.
LD block size in base pairs for each chromosome was averaged over all LD blocks identified by Haploview16. The high-CEU group has average block sizes that were larger than those observed for the low-CEU group for all chromosomes except chromosomes 15 and 18.
Figure 8.
LD comparison among the CR rural population and its ancestral populations.
Number of tag SNPs needed to cover the same gene regions was used as a measurement for the strength of LD. Populations with stronger LD would need less number of SNPs tagging the regions given certain r2 threshold compared to populations with weaker LD.