Observed phenotypes across the European population samples

consisted of 145 blue and 64 brown eye color plus 207 non-blue,
non-brown phenotypes. The latter were not placed into a single
group because of the full range of tonalities observed, so this group
was classified as follows: green-hazel 95; intermediate-light 40;
and intermediate-dark 72. We observed a high frequency of light
eye colors in northern European subjects and this decreased
moving south as shown in supplementary Fig. S2, in agreement
with previous observations [43,44].
Supplementary Table S1 outlines the results of formal tests
for Hardy Weinberg equilibrium with most populationlocus combinations not indicating significant departure from
Hardy
Weinberg equilibrium and only six significant values but
distributed randomly across eye color populations.
From 2530 SNP pair comparisons significant LD was found in 40
SNP pairs listed in supplementary Table S2. All cases of significant
LD were randomly distributed across loci and eye color populations
except SNPs rs1375164rs4778232 in OCA2 (15:28291812
15:28281765 = 10,047 nucleotide separation).
The PCA analysis did not reveal discernable geographic
stratification i.e., the phenotypic classes that we assigned to
samples repeatedly cluster independently of their European
geographic distribution as indicated in Fig. 1C and D.
Handling the phenotypes observed as genetic populations and
using blue and brown classes as pre-defined populations, Structure
analysis suggested a pattern of continuity observed in intermediate
phenotypes distinct from blue and brown, represented in the
K:2 cluster plot of Fig. 1A. We have positioned iris photographs
above each sample range to illustrate the characteristics of the
phenotypes assigned in the study. The intermediate-light samples
tended to cluster close to the blue end and the intermediate-dark
samples mostly clustered with browns; green-hazels appear as a
largely equal mixture of the two most differentiated classes of blue
and brown. This suggests that the simplified light and dark eye
color phenotyping regime of Mengel-From continues to have
validity for the wider range of SNPs examined in our study.
Nevertheless, the green-hazel phenotype does not form a distinct
cluster at K:2, K:3, or K:4 (K:3 and 4 cluster plots in supplementary
Fig. S3) and appears as a group where all components have mixed
membership proportions. Furthermore, the optimum K estimation
indicated that more than two genetic populations were detectable
(probability plot in Fig. 1B). Therefore we adopted the green-hazel
class as a third reference phenotype in the training sets. In order to
ensure the clearest differentiation of samples among a continuum
of very similar iris colors, we excluded 35 green-hazel samples
whose iris images fell between the most clearly differentiated blue
and green-hazel phenotypes. Similarly, a further 13 brown eye
samples were excluded that fell between green-hazel and brown
phenotypes. So the training sets (256 samples) comprised: 145
blue subjects; 60 green-hazel from an original 95; 51 brown from
an original 64.
3.2. Association analysis
The p-values for association with light, dark and the five eye
color phenotypes originally defined in our sample set, for all 37
SNPs of SHEP 12, are given in supplementary Table S3 and this
data includes adjustment for the effect of rs12913832. Of the 14
additional skin and hair color-associated SNPs genotyped in the
SHEP assays, 6 of 98 association p-values were significant
(p < 0.05) after adjustment for rs12913832, but we did not pursue
the analysis of these SNPs further. The levels of association found in
the 23 SNPs previously identified as most closely associated with
eye color are summarized in Fig. 2. Association with a significance
value of p < 0.05 was recorded in twenty markers. The independent
additional effect on the prediction of phenotypes was
measured by adjustment with the most associated SNP
rs12913832. From adjustment analysis SNPs were detected to
give an additional effect differentiating light, dark and the five eye
color phenotypes independently of their linkage to rs12913832, all
located in the OCA2/HERC2 region in ranked order of effect:

rs1129038, rs1667394, rs916977, rs4778138, rs7495174,

rs4778241, rs4778232, rs8024968, rs11636232, rs7183877. These
were followed by associated SNPs in genes: SLC45A2 (rs26722,
rs16891982), SLC24A4 (rs12896399), ASIP (rs1015362), TYR