technical note: quantitative measures of iris color using high resolution photographs

Technical Note: Quantitative Measures of Iris ColorUsing High Resolution Photographs

Melissa Edwards,1 Agnes Gozdzik,1 Kendra Ross,1 Jon Miles,2 and Esteban J. Parra1*

1Department of Anthropology, University of Toronto Mississauga, Mississauga, Ontario L5L1C6, Canada2Miles Research, Escondido, CA 92025

KEY WORDS iris pigmentation; quantification of eye color; HERC2

ABSTRACT Our understanding of the genetic archi-tecture of iris color is still limited. This is partly relatedto difficulties associated with obtaining quantitativemeasurements of eye color. Here we introduce a newautomated method for measuring iris color using highresolution photographs. This method extracts colormeasurements in the CIE 1976 L*a*b* (CIELAB) colorspace from a 256 by 256 pixel square sampled from the9:00 meridian of the iris. Color is defined across threedimensions: L* (the lightness coordinate), a* (the red-green coordinate), and b* (the blue-yellow coordinate).We applied this method to a sample of individuals ofdiverse ancestry (East Asian, European and SouthAsian) that was genotyped for the HERC2 rs12913832polymorphism, which is strongly associated with blue

eye color. We identified substantial variation in theCIELAB color space, not only in the European sample,but also in the East Asian and South Asian samples.As expected, rs12913832 was significantly associatedwith quantitative iris color measurements in subjects ofEuropean ancestry. However, this SNP was alsostrongly associated with iris color in the South Asiansample, although there were no participants with blueirides in this sample. The usefulness of this method isnot restricted only to the study of iris pigmentation.High-resolution pictures of the iris will also make itpossible to study the genetic variation involved in iristextural patterns, which show substantial heritabilityin human populations. Am J Phys Anthropol 147:141–149, 2012. VVC 2011 Wiley Periodicals, Inc.

Eye color is determined by the type of melanin presentand the density and distribution of melanosomes locatedwithin the melanocytes of the iris stroma (Sturm andLarsson, 2009). Iris pigmentation exhibits a variableglobal distribution. In most populations, eye color is pri-marily limited to varying shades of brown. However,individuals of European, and to a lesser extent, NorthAfrican, Middle Eastern, Central Asian, and South Asianancestry, express a wide range of colors that includeshades of brown, green, and blue. In recent years, theuse of linkage analyses and genome-wide associationstudies has led to the identification of several of the keygenes associated with iris color variation (Frudakis etal., 2003; Duffy et al., 2007; Sulem et al., 2007; Eiberg etal., 2008; Kayser et al., 2008; Sturm et al., 2008; Bra-nicki et al., 2009; Mengel-From et al., 2009; Liu et al.,2010). However, most of the research efforts havefocused on European populations, and there have beenvery few studies exploring the phenotypic variation andgenetic basis of iris pigmentation in populations of non-European ancestry.One of the major challenges for unraveling the genetic

architecture of iris pigmentation is obtaining a quantita-tive measurement of eye color. Traditional methods ofmeasuring skin and hair pigmentation, which are basedon reflectometry, cannot be used on the iris. Conse-quently, the majority of studies investigating iris pig-mentation variation have used a limited number of dis-crete categories to characterize eye color (for a recentreview about iris color classification, see Mackey et al.,2011). Although such discrete classification methodshave successfully identified some of the major genesassociated with iris color, they also have a number ofweaknesses. For one, they are subjective and have lim-ited inter- and intra-observer reliability (Seddon et al.,

1990; Frudakis et al., 2003). Additionally, they areunable to account for the extensive quantitative varia-tion that is inherent in iris pigmentation.Recently, a number of research groups have developed

quantitative methods for measuring iris color. Melgosaet al. (2000) used a spectroradiometer to obtain a mea-surement of the combined pupil and iris in the Commis-sion Internationale de L’Eclairage L*a*b* (CIELAB)color space. German et al. (1998) studied drug responsein irides by manually extracting a number of color meas-urements from photographs taken of the human eye.These measurements included parameters in the XYZand CIELAB color spaces. Frudakis (2008) extractedRGB and luminosity measurements from iris photo-graphs and condensed this information into a single ‘‘irismelanin index.’’ Most recently, Liu et al. (2010) isolatedhue and saturation values from iris photographs; huewas used to represent the type of melanin, while satura-tion was used to represent the amount of melanin in the

Grant sponsors: Government of Ontario (ERA), Natural Sciencesand Engineering Research Council of Canada (NSERC), CanadaFoundation for Innovation (CFI), Ontario Innovation Trust (OIT).

*Correspondence to: Esteban Parra, Department of Anthropology,University of Toronto Mississauga, 3359 Mississauga Road North,Mississauga, Ontario L5L1C6, Canada.E-mail: [email protected]

Received 26 June 2011; accepted 11 October 2011

DOI 10.1002/ajpa.21637Published online 19 November 2011 in Wiley Online Library

(wileyonlinelibrary.com).

VVC 2011 WILEY PERIODICALS, INC.

AMERICAN JOURNAL OF PHYSICAL ANTHROPOLOGY 147:141–149 (2012)

iris. The development of such quantitative methods iscreating new opportunities to study the genetics of irispigmentation variation in a much more objective way.In this article, we introduce a new automated method

to quantify iris color. This method extracts color meas-urements in the CIELAB color space from high-resolu-tion photographs of the iris. CIELAB is an internationalstandardized color system that was designed to approxi-mate human color vision (CIE, 1986). The CIELAB colorspace provides quantitative measurements across threedifferent dimensions: a brightness dimension (L*), agreen/red dimension (a*), and a blue/yellow dimension(b*). The CIELAB space has advantages over other colorsystems because it is perceptually uniform: the distancebetween two color points closely corresponds to percep-tual differences in human vision. Additionally, the CIE-LAB is a metric space that factors in the color tempera-ture of the illuminant, also known as the white balanceof the source illumination. Color data from an RGB colorspace does not include the illuminant temperature; thisfactor is however included in the conversion from RGBto CIELAB. The output of the xenon flash as used inphotography is approximated as D55—that is, Daylightat 5,500 degrees Kelvin.We applied this new method to a sample of individuals

of diverse ancestry (East Asia, Europe, and South Asia)that was genotyped for the HERC2 rs12913832 polymor-phism, which shows a strong association with blue eyecolor (Eiberg et al., 2008; Sturm et al., 2008).

MATERIALS AND METHODS

Recruitment

Study participants were recruited at the MolecularAnthropology Laboratory of the University of TorontoMississauga (UTM) between 2007 and 2009. The major-ity of participants were UTM students and staff whoresponded to online and print advertisements that weredistributed throughout the University of Toronto commu-nity. To assess geographic ancestry, each participant wasasked to complete a questionnaire that inquired aboutthe participant’s parental and grandparental places ofbirth and native languages. For example, individualswho stated that their ancestors were from China, Japan,and Korea, were grouped as East Asian, while those whoreported ancestors from India and Pakistan weregrouped as South Asian. Individuals who reported beingof multiple ancestries were placed into a subgroup desig-nated as ‘‘Other,’’ and were not included in the analysis.In total, 205 subjects were included in the study. Ofthese, 66 were of East Asian ancestry, 72 were of Euro-pean ancestry, and 67 were of South Asian ancestry.This study was approved by the University of Toronto

Health Sciences Research and Ethics Board. All partici-pants provided written informed consent.

Genotyping

We selected the rs12913832 polymorphism in intron 86of the HERC2 gene for genotyping. The derived G alleleof this marker has been strongly associated with blueiris color in populations of European ancestry (Eiberget al., 2008; Sturm et al., 2008). A blood sample was col-lected from each participant in a 4-mL EDTA tube, andDNA was extracted from the blood samples using theE.Z.N.A. Blood DNA Midi Kit (Omega Bio-Tek, GA). The

samples were then sent to Kbioscience (http://www.kbioscience.co.uk) for genotyping. Kbioscience uses agenotyping method that combines allele-specific PCRwith detection by a Fluorescence Resonance EnergyTransfer (FRET) system. To evaluate genotyping quality,blind duplicates were included in the genotyping plates.The concordance rate for the genotype calls of the dupli-cated samples was 100%.

Acquisition and processing of iris photographs

We took a photograph of each subject’s left iris usingthe Miles Research Professional Iris camera (MilesResearch, CA). This camera consists of a Fujifilm Fine-pix S3 Pro 12-megapixel DSLR mounted on a Nikkor105-mm macro lens. A coaxial biometric illuminator wasused to deliver a constant and uniform source of light toeach iris at 5,500 K (D55 illuminant). Camera perform-ance in terms of color recording fidelity was verifiedusing the 24-patch GretagMacbeth Color Checker andImatest Master software (http://www.imatest.com). Allphotographs were taken using an aperture of f/19 andexposure sensitivity (ISO) of 200. The shutter speed wasset to 1/60 second and all pictures were recorded in 12megapixel jpeg format.Upon examination of the photographs used for the

analysis of iris color there was evidence of underexpo-sure. As a corrective measure, an exposure normaliza-tion procedure was applied to all 205 images using acolor-neutral brightness operator, the screen-mode blend-ing of each image with itself. This exposure compensa-tion filter shifts the colors away from black and white soas to improve discrimination of small differences in color.The effect is analogous to using two slide projectors illu-minating the same transparency image, aligned on ascreen, to create a brighter result. Mathematically, thescreen blend lightening filter (blending an eight-bits-per-channel image with itself) is expressed as: c0 5 255 –((255 2 c0)

2/255), where c0 represents each R, G, B chan-nel value as an eight-bit integer from 0 to 255, and c0 isthe filter result value.

Acquisition of color measurements

A single RGB measurement was extracted from eachiris image using a multistep process: 1/A macro was cre-ated in Adobe Photoshop CS5 that cropped each imageto the middle-third of the image height, centered on thepupil; 2/A second macro was developed that cropped outthe regions of the image containing the sclera; 3/A finalmacro was used to crop the sample down to a 256 by 256pixel square at the ciliary zone of the iris (mid-periph-ery) in the 9:00 meridian of each sample; and 4/A theprogram ImageJ (National Institutes of Health, MD) wasused to extract RGB data from each of the croppedimage samples. We used the ‘‘average RGB’’ of the 256by 256 pixel square to characterize the iris color.The RGB values were transformed into CIE 1976

L*a*b* (CIELAB) color space. The CIELAB color spaceis a universal standardized system that uses threedimensions to represent color. The first dimension, L*, isused to represent brightness and can have values thatrange from 0 to 100, where 0 is black and 100 is white.The a* and b* dimensions correspond to differences incolor, with negative values of a* indicating green andpositive values of a* indicating red, and negative valuesof b* indicating blue and positive values of b* indicating

142 M. EDWARDS ET AL.

American Journal of Physical Anthropology

yellow. To transform the RGB values into the CIELABcolor space, the RGB measurements were first convertedinto XYZ values using standard equations found athttp://www.easyrgb.com/index.php?X5MATH&H502#text2.The XYZ values were then transformed into L*, a*, and b*measurements using standard equations found at http://www.easyrgb.com/index.php?X5MATH&H507#text7. Asthe coaxial biometric illuminator delivered light to eachiris at a temperature of 5,500 K, the color space conver-sion was based on the D55 standard illuminant. The ob-server was set at 28.

Statistical analysis

Deviations from Hardy–Weinberg proportions in eachpopulation were evaluated using the Hardy–Weinbergexact test (see http://ihg2.helmholtz-muenchen.de/cgi-bin/hw/hwa1.pl).We used general linear model (GLM) analyses to test

the association between the rs12913832 genotypes andL*, a*, b* values in the European and South Asiansamples (in the East Asian sample, 64 out of 65 individ-uals were homozygous for the ancestral A allele, so dueto the lack of polymorphism in this sample, we did notcarry out the statistical analysis). The tests weredone with the statistical program SPSS (version 17.0,SPSS, 2008). Associations were considered significantwhen P � 0.05.

RESULTS

Figure 1a,b shows a graphical representation summa-rizing the procedure that was used to obtain measure-ments in the CIELAB color space from each iris, with arepresentative sample including two blue, one green,and three brown irides. Figure 2a,b depicts the measure-ments obtained for the full sample. There is a substan-tial dispersion of the L*, a*, and b* values in the three-dimensional space. As expected, the dispersion is morepronounced in the European sample, which includes eyecolors that ranged from light blue to dark brown. In theEuropean sample, the L* values ranged from 17.63 to67.08, the a* values from 26.56 to 25.52, and the b* val-ues from 214.34 to 37.62. There was also considerablevariation in the East Asian and South Asian samples. Inthe East Asian sample, which is comprised primarily ofdark brown irides, the L* values fluctuated from 10.96to 40.92, the a* values from 22.67 to 24.42, and the b*values from 5.73 to 11.12. In the South Asian sample,which includes a broader representation of shades ofbrown than the East Asian sample, the L* (ranging from10.18 to 44.31) and a* (ranging from 22.86 to 25.42) val-ues showed a distribution that is broadly similar to thatobserved in the East Asian sample. However, there wasconsiderably more dispersion in the b* axis (0.32–31.36).When examining the CIELAB coordinates consideringbroad categorical classifications of iris color, blue iridestended to have very high L* values and negative a* andb* values. Green irides usually mapped between brownand blue irides, and had a high L* value, a b* value thathovered around 0, and a negative a* value. In contrast,brown irides tended to have low L* values and positivea* and b* values. Irides with lighter shades of brownhad higher L*, a*, and b* values than irides with darkershades of brown (see Fig. 1b). Overall, it is evident thatclassifications that use broad categories such as ‘‘brown,’’‘‘intermediate,’’ or ‘‘blue’’ do not satisfactorily capture iris

pigmentation variation. We also observed that in manyindividuals, particularly those of European ancestry, thepigmentation of the central pupillary zone is darkerthan the pigmentation of the peripheral ciliary zone(central heterochromia).To test the usefulness of our method to quantify iris

color, we evaluated the effect of the HERC2 rs12913832polymorphism on the L*, a*, and b* values. This singlenucleotide polymorphism (SNP) has been associated withiris pigmentation in previous studies (Eiberg et al., 2008;Sturm et al., 2008; Liu et al., 2010). Table 1 shows thegenotype and allele frequencies in the East Asian, Euro-pean, and South Asian samples. There were no signifi-cant deviations from Hardy–Weinberg proportions in anyof the samples (Table 1). The frequency of the derived Gallele, which has been associated with blue iris color inprevious studies, was 71% in the European sample and13% in the South Asian sample. In the East Asian sam-ple, 64 out of 65 individuals were homozygous for the an-cestral A allele, so we eliminated this sample from thestatistical analysis. We employed an unconstrainedgenetic model, which estimates separately the effects ofthe homozygotes for the derived G allele and the hetero-zygotes, using the homozygotes for the ancestral A alleleas the reference genotype. The results of these tests forthe European and South Asian sample are depicted inTable 2. In the European sample, rs12913832 was signif-icantly associated with the L*, a*, and b* (P \ 0.001)dimensions of color space. In the lightness dimension,the estimates of the regression coefficient (beta) indi-cated that L* increased by 12.37 U in AG heterozygotesand by 23.27 U in GG homozygotes, with respect to AAhomozygotes. In the green/red dimension, a* decreasedby 7.19 U in AG heterozygotes and by 23.16 U in GGhomozygotes, with respect to AA homozygotes. In con-trast, in the blue/yellow dimension, b* decreased dra-matically (22.58 U) in the GG homozygotes, but wasvery similar for the AG heterozygotes and the AA homo-zygotes. Therefore, rs12913832 seems to fit a dominantmodel for the blue/yellow dimension (b*) of the CIELABcolor space, with the ancestral A allele dominant overthe recessive G allele, and a codominant model for thegreen/red (a*) and lightness (L*) dimensions. In theSouth Asian sample, rs12913832 was also significantlyassociated with the L*, a*, and b* dimensions of theCIELAB color space (P \ 0.001). In this sample we onlyobserved one homozygote for the derived G allele, so itwas not possible to explore in detail the effects of thethree rs12913832 genotypes on the color coordinates.However, it is interesting to note that in the South Asiansample, heterozygotes for the AG allele showed higherL*, a*, and b* values than homozygotes for the ancestralA allele, indicating that, on average, they have lightershades of brown (see Fig. 1b).To verify the results, we ran the same tests on a

smaller collection of irides that were photographed 1year later, using the same camera, but with a slightlydifferent shutter speed. This collection included photo-graphs of 27 East Asian, 37 European, and 25 SouthAsian participants and all of the major eye color pheno-types (blue, brown, green) were represented. The resultsare very consistent with those described above. In theEuropean sample, the rs12913832 marker was stronglyassociated (P \ 0.001) with the three dimensions of theCIELAB color space, and the parameters of the uncon-strained model are overly similar to those of the largersample. L* increased by 17.45 U in AG heterozygotes

143QUANTITATIVE MEASURES OF IRIS COLOR


and by 24.79 U in GG homozygotes, a* decreased by11.24 U in AG heterozygotes and by 20.80 U in GGhomozygotes, and b* decreased by 20.45 U in GG homo-zygotes with respect to AA homozygotes, but there werevery small differences between AA homozygotes and AGheterozygotes in the b* dimension. In the South Asiansample, rs12913932 was significantly associated with theL* (P \ 0.001) and b* (P 5 0.021) dimensions, and wasnot significant for the a* dimension (P 5 0.369). Again,heterozygote AG individuals showed, on average, higherL*, a*, and b* values than homozygote AA individuals.

DISCUSSION

We introduce a new method to measure iris color,using a series of crops from high-resolution photographsto obtain a quantitative estimate of color in the threecoordinates of the CIELAB color space. The CIELABcolor space is a universal color system that was designedto approximate human vision (CIE, 1986). Iris color vari-ation is reported across three different planes: a bright-ness plane (L*) and two color planes (a* and b*). Whenwe plotted our set of 205 human irides in CIELAB color

Fig. 1. (a) An illustration of the sample 256 by 256 pixel square that was isolated from the 9:00 meridian of six irides of varyingcolor (two blues, one green, three browns). The average RGB of this square was determined, and transformed into L*, a*, and b*values in the CIELAB color space. (b) The coordinates of overall iris color of six exemplar samples in the CIELAB color space. Eachiris is graphically represented by the 256 by 256 pixel square sample region that was used to evaluate iris color and its color spacecoordinates (L*, a*, b*).



Fig. 2. (a) The distribution of the full sample of irides of participants of East Asian (red circles), South Asian (green triangles),and European (blue squares) ancestry in the three coordinates of CIELAB color space: L* (the lightness coordinate), a* (the red-green coordinate), and b* (the blue-yellow coordinate). (b) The coordinates of the same irides for a* and b*, the chromatic compo-nents of the CIELAB color space.



space, most of the points were distributed throughoutthe three-dimensional space in a fairly continuous man-ner. This confirms that iris pigmentation, like skin andhair color, is a quantitative trait. This is evident in thethree population groups studied in our analysis, whichincluded two samples (East Asians and South Asians)that were comprised primarily of individuals with‘‘brown’’ irides. Treating eye color as a categorical traitin studies aimed at characterizing the genetic basis ofiris pigmentation disregards a substantial amount of theexisting variation and reduces the statistical power toidentify main effects and gene–gene interactions.To validate our method, we tested the association

between the HERC2 rs12913832 polymorphism and ourCIELAB color space measurements in the European andSouth Asian samples. This particular marker is found ina highly conserved region of intron 86 of the HERC2gene and has been strongly associated with blue/browniris color in previous studies in populations of Europeanancestry (Eiberg et al., 2008; Sturm et al., 2008).Although the exact mechanisms of action of this poly-morphism have not yet been fully determined, it is pur-ported to play a role in the regulation of OCA2. Eiberget al. (2008) suggested that rs12913832 falls within anOCA2 silencer complex that may be disturbed or stabi-lized depending on which of the two rs12913832 allelesis present (Eiberg et al., 2008). In contrast, Sturm et al.(2008) have argued that this particular polymorphismmay be a binding site for the regulatory protein HLTF(Sturm et al., 2008). Under this alternative model, whenthe ancestral A allele is present, HLTF binds to thisregion and promotes the transcription of OCA2. Whenthe derived G allele is present, however, HLTF cannotproperly bind to the DNA and OCA2 transcription isreduced. Studies in human melanocyte strains haveshown that the presence of the rs12913832 ancestral Aallele is associated with increased levels of OCA2 tran-script compared to the derived G allele (Cook et al.,2009).The frequencies for the derived G allele that we

observed in our sample (East Asians \1%, Europeans 571%, and South Asians 5 13%) are in overall agreementwith previously reported data. For example, the fre-quency of this allele in the Human Genome DiversityProject (HGDP) East Asian samples was \1%, for theEuropean samples 51%, and for Central-South Asiansamples, 15% (http://spsmart.cesga.es/).As expected, rs12913832 was significantly associated

with the L*, a*, and b* planes of the CIELAB color space

in our European sample. The use of quantitative meas-ures, instead of categorical variables, highlights someinteresting aspects of the effect of rs12913832 on iriscolor. For the b* (blue/yellow) dimension of the CIELABcolor space, rs12913832 fits a dominant model of inheri-tance quite well, with the ancestral A allele dominantover the G-derived allele. Homozygous GG individualshave lower b* values (corresponding to the blue regionsof the CIELAB space) than heterozygous AG or homozy-gous GG individuals, which show similar b* values.However, the L* and a* dimensions appear to fit acodominant model, with heterozygotes AG showing inter-mediate values between homozygotes AA and GG. Tradi-tionally, eye color has been described as a trait that fitsa dominant model of inheritance, with ‘‘brown’’ dominantover ‘‘blue,’’ although there have been many reportsdemonstrating inconsistencies with a dominant model ofinheritance (Frudakis, 2008; Sturm and Larsson, 2009).Focusing specifically on the rs12913832 polymorphism,Eiberg et al. (2008) showed that in a sample of �200Danish individuals this polymorphism was perfectlyassociated with blue/brown color, strongly supporting amodel in which blue eye color is caused by homozygosityof the rs12913832 G allele. However, Sturm et al. (2008)showed that the association of the rs12912832 GG geno-type with blue eye color is not perfect: some individualswith this genotype had brown eyes, and some heterozy-gous AG individuals blue eyes. These two studies usedcategorical classifications of iris color based on self-report (Eiberg et al., 2008), or the rating of a trained ob-server with cross-validation with the individual’s self-report (Sturm et al., 2008). In our sample, we also seeclear evidence that, although strongly associated withblue eye color, rs12913832 does not fit a simple dominantmodel of inheritance. Some homozygote GG individualsand heterozygote AG individuals occupy positions in theCIELAB space that do not overlap with the coordinatesobserved for most of the GG or AG genotypes. We alsoobserved that, contrary to the expectations of a simpledominant model, the AG heterozygotes show L* and a*values that are intermediate between those characteris-tic of the AA and GG genotypes. This suggests that AGheterozygotes have intermediate levels of OCA2 expres-

TABLE 1. Observed and expected genotype frequencies, allelefrequencies, and the Hardy–Weinberg exact test for HERC2

rs12913832 in the East Asian, European,and South Asian samples

PopulationGenotype frequenciesobserved (expected)

Allelefrequencies

Hardy–Weinberg

exact test (P)

East Asian A:A 5 64 (64.00) A 5 99% 1.000A:G 5 1 (0.99) G 5 1%G:G 5 0 (0.00)

European A:A 5 9 (6.00) A 5 29% 0.087A:G 5 23 (28.99) G 5 71%G:G 5 38 (35.00)

South Asian A:A 5 50 (50.09) A 5 87% 1.000A:G 5 15 (14.81) G 5 13%G:G 5 1 (1.09)

TABLE 2. Results of the general linear model analysis for thebrightness dimension (L*), the green/red dimension (a*), and

the blue/yellow dimension (b*) of the CIELAB color space in theEuropean and South Asian samples

PopulationColor spacedimension P-value Genotype Beta P-value

European L* 0.000 AA – –AG 12.368 0.000GG 23.267 0.000

a* 0.000 AA – –AG 27.187 0.003GG 223.158 0.000

b* 0.000 AA – –AG 1.989 0.585GG 222.584 0.000

South Asian L* 0.000 AA – –AG 8.358 0.000

a* 0.000 AA – –AG 2.701 0.037

b* 0.000 AA – –AG 6.788 0.000

Beta is the estimate of the regression coefficient.



sion, and this is reflected as perceptible differences inthe L* and a* dimensions of the color space.When we performed the same analysis on the South

Asian sample, rs12913832 was also significantly associ-ated with the L*, a*, and b* dimensions of color space.As our South Asian sample is comprised almost entirelyof irides that would have been defined as ‘‘brown’’ usinga categorical iris classification system, this shows thatrs12913832 also plays a role in modulating subtle grada-tions in brown eye color. In particular, AG heterozygotesshowed higher L*, a*, and b* values (e.g., lighter shadesof brown) than AA homozygotes. Again, this stronglysuggests that rs12913832 does not fit a dominant modelof inheritance.We repeated the analysis in an independent sample

measured one year later, and we obtained similarresults. However, the parameter values (beta coeffi-cients) obtained in this replication sample should be con-sidered with caution due to the small sample size (e.g.,

in the European sample, there were only three homozy-gotes for the ancestral allele).It is worth noting that, in our combined iris samples,

we found two individuals of South Asian ancestry thathad two copies of the rs12913832 derived G allele andbrown or heterochromatic irides. Additionally, weobserved that, whereas AG heterozygotes show signifi-cantly lower a* values than homozygotes for the ances-tral A allele in the European sample, in the South Asiansample AG heterozygotes have significantly higher a*values than AA homozygotes. These findings, as well asinformation from other studies described above, clearlyindicate that eye color is a polygenic trait with a complexgenetic architecture, where the effect of some of themajor loci may be modified by other polymorphisms(gene–gene interaction). A recent study in a large sampleof European descent has shown that interactionsbetween the genes HERC2 and OCA2, HERC2, andSLC24A5 and HERC2 and TYRP1 play a role in the

Fig. 3. One method of visualizing central heterochromia. The iris photographs of two heterochromatic and one homogenousirides were reduced to a width of 125 pixels. The program ImageJ (National Institutes of Health, MD) was used to extract a hori-zontal band (left to right) from the 9:00 meridian of each iris in RGB space and the RGB values were transformed into saturationvalues using standard equations found at http://www.easyrgb.com/index.php?X5MATH&H518#text18. Irides with substantial het-erochromia showed sharp changes in saturation across the width of the iris.



determination of eye color (Pospiech et al., 2011). Simi-larly, Liu et al. (2010) described that interactions of sev-eral polymorphisms in pigmentation genes are responsi-ble for some of the variation of hue and saturationobserved in their iris sample.In this study, we automatically extracted a 256 by 256

pixel square from the iris to estimate the average colorcoordinates in the CIELAB color space. As shown above,this method is an improvement over the categorical clas-sifications conventionally used to study the genetics ofeye color. However, this method does not fully capturethe complexity of iris color. Central heterochromia is acommon feature observed in human irides. In individu-als with central heterochromia, the pupillary zone of theiris has a darker color than the peripheral ciliary zone.Alternative methods of quantifying iris color based onhigh-resolution photographs can be employed to specifi-cally study central heterochromia. For example, a mea-sure describing color (e.g., hue angle, saturation, or justwavelength) can be obtained in a radial section of theiris, making it possible to compare the color of the pupil-lary and ciliary zones of the iris. Figure 3 illustratessuch a method using high-resolution photographs of indi-viduals with and without central heterochromia. In oursample, central heterochromia was present in the threepopulations examined. As well, it was found in irides ofall of the major color groups, although it tends to bemost noticeable in individuals with lighter colored eyes.Finally, apart from central heterochromia, there are a

number of other iridial structures that can be studiedusing high-resolution photographs, such as nevi (hyper-pigmented spots), Wolfflin nodules (pale nodules encir-cling the iris that are composed primarily of collagen tis-sue), Fuchs’ crypts (pit-like depressions near the collar-ette and in the ciliary zone of the iris), and contractionfurrows (circular and radial folds due to iris contractionor dilation), (Larsson and Pedersen, 2004; Sturm andLarsson, 2009). These traits show a substantial heritabil-ity in human populations (between 58 and 78%), but thegenetic basis of these and other iris patterns remainslargely unexplored (Larsson et al., 2003). Using a ge-nome-wide association (GWA) strategy, Larsson et al.(2011) reported that variants within the gene SEMA3Aare significantly associated with crypt frequency, poly-morphisms within the gene TRAF3IP1 with contractionfurrows, and variants near the pigmentation geneSLC24A4 with the presence of a peripupillary pigmen-tary ring. Interestingly, the genes SEMA3A and TRA-FIP1 are involved in pathways that control neurogene-sis, neural migration, and synaptogenesis, and this ledLarsson et al. to suggest that genes involved in normalneuronal pattern development may also influence irisstructures.In conclusion, we present a method to measure iris

color quantitatively using high-resolution photographsobtained with uniform illumination and show that, usingthis approach in individuals of diverse ancestry, it is pos-sible to gain interesting insights about the role of theHERC2 rs12913832 polymorphism in iris pigmentation.There are still important gaps in our understanding ofthe genetic architecture of iris pigmentation. The appli-cation of quantitative methods for measuring eye coloropens up new avenues in pigmentation research. Asthese methods are able to distinguish between subtlecolor gradations, it will be possible to increase the statis-tical power to identify main effects and gene interac-tions, and to study the genetics of iris pigmentation in

populations in which traditional categorical systems areof limited use. Additionally, such methods will provide ameans of unraveling the genetic architecture of othertraits, such as central heterochromia, and a number ofiris patterns that show substantial variation in humanpopulations. Advancing our current knowledge of thegenetic basis of iris color and structure is importantfrom multiple perspectives. For example, from a forensicperspective it will be useful to reliably predict iris colorand structure based on DNA information. From an an-thropological and evolutionary perspective, it will allowresearchers to gain new insights about the major evolu-tionary events associated with the current distribution ofeye color variation, which is quite different from the geo-graphical distribution of skin pigmentation. Of interestis to which extent natural or sexual selection, or theaction of demographic factors (e.g., genetic drift, rangeexpansions) shaped the current distribution of eye color.Finally, from the developmental and physiological per-spectives, discovering the genes associated with iris colorand structure will improve our understanding of thepathways that determine these traits.

ACKNOWLEDGMENTS

The authors thank all the individuals who participatedin the study. Jon Miles is the owner of Miles Research, acompany that designs and builds specialized iris cam-eras.

LITERATURE CITED

Branicki W, Brudnik U, Wojas-Pelc A. 2009. Interactionsbetween HERC2, OCA2 and MC1R may influence human pig-mentation phenotype. Ann Hum Genet 73:160–170.

CIE. 1986. CIE publication no. 15.2—colorimetry. Austria: Cen-tral Bureau of the Commission Internationale de L’Eclairage.

Cook AL, Chen W, Thurber AE, Smit DJ, Smith AG, BladenTG, Brown DL, Duffy DL, Pastorino L, Bianchi-Scarra G,Leonard JH, Stow JL, Sturm RA. 2009. Analysis of culturedhuman melanocytes based on polymorphisms within theSLC45A2/MATP, SLC24A5/NCKX5, and OCA2/P loci. J InvestDermatol 129:392–405.

Duffy DL, Montgomery GW, Chen W, Zhao ZZ, Le L, JamesMR, Hayward NK, Martin NG, Sturm RA. 2007. A three-sin-gle-nucleotide polymorphism haplotype in intron 1 of OCA2explains most human eye-color variation. Am J Hum Genet80:241–252.

Eiberg H, Troelsen J, Nielsen M, Mikkelsen A, Mengel-From J,Kjaer KW, Hansen L. 2008. Blue eye color in humans may becaused by a perfectly associated founder mutation in a regula-tory element located within the HERC2 gene inhibiting OCA2expression. Hum Genet 123:177–187.

Frudakis T, Thomas M, Gaskin Z, Venkateswarlu K, ChandraKS, Ginjupalli S, Gunturi S, Natrajan S, Ponnuswamy VK,Ponnuswamy KN. 2003. Sequences associated with humaniris pigmentation. Genetics 165:2071–2083.

Frudakis TN. 2008. Molecular photofitting: predicting ancestryand phenotype using DNA. Boston: Elsevier/Academic Press.

German EJ, Hurst MA, Wood D, Gilchrist J. 1998. A novel sys-tem for the objective classification of iris color and its correla-tion with response to 1% tropicamide. Ophthalmic PhysiolOpt 18:103–110.

Kayser M, Liu F, Janssens CJW, Rivadeneira F, Lao O, vanDuijn K, Vermeulen M, Arp P, Jhamai MM, van IJcken WFJ,den Dunnen JT, Heath S, Zelenika D, Despriet DDG, KlaverCCW, Vingerling JR, de Jong PTVM, Hofman A, AulchenkoYS, Uitterlinden AG, Oostra BA, van Duijin CM. 2008. Threegenome-wide association studies and a linkage analysis iden-tify HERC2 as a human iris color gene. Am J Hum Genet82:411–423.



Larsson M, Duffy DL, Zhu G, Liu JZ, Macgregor S, McRae AF,Wright MJ, Sturm RA, Mackey DA, Montgomery GW,Martin NG, Medland SE. 2011. GWAS findings for humaniris patterns: associations with variants in genes that influ-ence normal neuronal pattern development. Am J HumGenet 89:334–343.

Larsson M, Pedersen NL. 2004. Genetic correlations amongtexture characteristics in the human iris. Mol Vis 10:821–831.

Larsson M, Pedersen NL, Stattin H. 2003. Importance of geneticeffects for characteristics of the human iris. Twin Res 6:192–200.

Liu F, Wollstein A, Hysi PG, Ankra-Badu GA, Spector TD,Park D, Zhu G, Larsson M, Duffy DL, Montgomery GW,Mackey DA, Walsh S, Lao O, Hofman A, Rivadeneira F,Vingerling JR, Uitterlinden AG, Martin NG, Hammond CJ,Kayser M. 2010. Digital quantification of human eye colorhighlights genetic association of three new loci. PLoS Genet6:e1000934.

Mackey DA, Wilkinson CH, Kearns LS, Hewitt AW. 2011. Clas-sification of iris color: review and refinement of a classifica-tion schema. Clin Exp Ophthalmol 39:462–471.

Melgosa M, Rivas MJ, Gomez L, Hita E. 2000. Towards a colori-metric characterization of the human iris. Ophthalmic PhysiolOpt 20:252–260.

Mengel-From J, Wong TH, Morling N, Rees JL, Jackson IJ.2009. Genetic determinants of hair and eye colors in the Scot-tish and Danish populations. BMC Genet 10:88.

Pospiech E, Draus-Barini J, Kupiec T, Wojas-Pelc A, BranickiW. 2011. Gene-gene interactions contribute to eye color varia-tion in humans. J Hum Genet 56:447–455.

Seddon JM, Sahagian CR, Glynn RJ, Sperduto RD, GragoudasES. 1990. Evaluation of an iris color classification system.The Eye Disorders Case-Control Study Group. Invest Oph-thalmol Vis Sci 31:1592–1598.

Sturm RA, Duffy DL, Zhao ZZ, Leite FP, Stark MS, HaywardNK, Martin NG, Montgomery GW. 2008. A single SNP in anevolutionary conserved region within intron 86 of the HERC2gene determines human blue-brown eye color. Am J HumGenet 82:424–431.

Sturm RA, Larsson M. 2009. Genetics of human iris color andpatterns. Pigment Cell Melanoma Res 22:544–562.

Sulem P, Gudbjartsson DF, Stacey SN, Helgason A, Rafnar T,Magnusson KP, Manolescu A, Karason A, Palsson A, Thor-leifsson G, Jakobsdottir M, Steinberg S, Palsson S, JonassonF, Sigurgeirsson B, Thorisdottir K, Ragnarsson R, Benedikts-dottir KR, Aben KK, Kiemeney LA, Olafsson JH, Gulcher J,Kong A, Thorsteinsdottir U, Stefansson K. 2007. Geneticdeterminants of hair, eye and skin pigmentation in Euro-peans. Nat Genet 39:1443–1452.



technical note: quantitative measures of iris color using high resolution photographs

Documents