university of copenhagen
University of Copenhagen
Genetic analyses of the human eye colours using a novel objective method for eye
colour classification
Andersen, Jeppe Dyrberg; Johansen, Peter; Harder, Stine; Christoffersen, Susanne R;
Delgado, Mikaela C; Henriksen, Sarah T; Nielsen, Mette M; Sørensen, Erik; Ullum, Henrik;
Hansen, Thomas; Dahl, Anders L; Paulsen, Rasmus Reinhold; Børsting, Claus; Morling, Niels
Published in:
Forensic science international. Genetics
DOI:
10.1016/j.fsigen.2013.05.003
Publication date:
2013
Citation for published version (APA):
Andersen, J. D., Johansen, P., Harder, S., Christoffersen, S. R., Delgado, M. C., Henriksen, S. T., ... Morling, N.
(2013). Genetic analyses of the human eye colours using a novel objective method for eye colour classification.
Forensic science international. Genetics, 7(5), 508-15. DOI: 10.1016/j.fsigen.2013.05.003
Download date: 14. Jun. 2017
Forensic Science International: Genetics 7 (2013) 508–515
Contents lists available at SciVerse ScienceDirect
Forensic Science International: Genetics
journal homepage: www.elsevier.com/locate/fsig
Genetic analyses of the human eye colours using a novel objective
method for eye colour classification
Jeppe D. Andersen a,1,*, Peter Johansen a,1, Stine Harder b, Susanne R. Christoffersen b,
Mikaela C. Delgado a, Sarah T. Henriksen b, Mette M. Nielsen b, Erik Sørensen c,
Henrik Ullum c, Thomas Hansen d, Anders L. Dahl b, Rasmus R. Paulsen b, Claus Børsting a,
Niels Morling a
a
Section of Forensic Genetics, Department of Forensic Medicine, Faculty of Health and Medical Sciences, University of Copenhagen, DK-2100 Copenhagen,
Denmark
b
DTU Informatics – Informatics and Mathematical Modeling, Technical University of Denmark, DK-2800 Lyngby, Denmark
c
Department of Clinical Immunology, Copenhagen University Hospital, Rigshospitalet, Copenhagen, Denmark
d
Research Institute of Biological Psychiatry, Mental Health Centre Sct. Hans, Copenhagen University Hospital, Roskilde, Denmark
A R T I C L E I N F O
A B S T R A C T
Article history:
Received 21 January 2013
Received in revised form 25 April 2013
Accepted 9 May 2013
In this study, we present a new objective method for measuring the eye colour on a continuous scale that
allows researchers to associate genetic markers with different shades of eye colour.
With the use of the custom designed software Digital Iris Analysis Tool (DIAT), the iris was automatically
identified and extracted from high resolution digital images. DIAT was made user friendly with a graphical
user interface. The software counted the number of blue and brown pixels in the iris image and calculated a
Pixel Index of the Eye (PIE-score) that described the eye colour quantitatively. The PIE-score ranged from 1
to 1 (brown to blue). The software eliminated the need for user based interpretation and qualitative eye
colour categories. In 94% (570) of 605 analyzed eye images, the iris region was successfully extracted and a
PIE-score was calculated. A very high correlation between the PIE-score and the human perception of eye
colour was observed. The correlations between the PIE-scores and the six IrisPlex SNPs (HERC2 rs12913832,
OCA2 rs1800407, SLC24A4 rs12896399, TYR rs1393350, SLC45A2 rs16891982 and IRF4 rs12203592) were
analyzed in 570 individuals. Significant differences (p < 106) in the PIE-scores of the individuals typed as
HERC2 rs12913832 G (PIE = 0.99) and rs12913832 GA (PIE = 0.71) or A (PIE = 0.87) were observed. We
adjusted for the effect of HERC2 rs12913832 and showed that the quantitative PIE-scores were significantly
associated with SNPs with minor effects (OCA2 rs1800407, SLC24A4 rs12896399 and TYR rs1393350) on the
eye colour. We evaluated the two published prediction models for eye colour (IrisPlex [1] and Snipper [2])
and compared the predictions with the PIE-scores. We found good concordance with the prediction from
individuals typed as HERC2 rs12913832 G. However, both methods had difficulties in categorizing
individuals typed as HERC2 rs12913832 GA because of the large variation in eye colour in HERC2
rs12913832 GA individuals. With the use of the DIAT software and the PIE-score, it will be possible to
automatically compare the iris colour of large numbers of iris images obtained by different studies and to
perform large meta-studies that may reveal loci with small effects on the eye colour.
ß 2013 Elsevier Ireland Ltd. All rights reserved.
Keywords:
Forensic phenotyping
Eye colour
Pigmentation
Genetics
1. Introduction
Genetically based prediction of eye colours and other visible
physical traits (forensic phenotyping) is currently an important
research field in forensic genetics. Information about a perpetrator’s
phenotype may be a valuable tool in a crime case if no suspect or
* Corresponding author. Tel.: +45 35326916; fax: +45 35326270.
E-mail address: [email protected] (J.D. Andersen).
1
These authors contributed equally.
1872-4973/$ – see front matter ß 2013 Elsevier Ireland Ltd. All rights reserved.
http://dx.doi.org/10.1016/j.fsigen.2013.05.003
individual in the DNA database match the STR profile found at the
crime scene. With reliable forensic phenotyping assays, the police
investigators may concentrate on groups of individuals with traits
that are predicted by the genetic investigations.
Genetically based prediction of eye colours has great potentials
in forensic genetic case work. Eye colour is highly heritable
(H2 = 0.98) [3]. The genetics of blue and brown eye colour is well
understood. It is largely explained by the HERC2 SNP rs12913832
[4,5]. The colour of the human eye is determined by the production
of eumelanin (eumelanogenesis) in the melanocytes and to a
minor degree by the structure and density of the iridial stroma [6].
J.D. Andersen et al. / Forensic Science International: Genetics 7 (2013) 508–515
Brown eyes contain high amounts of eumelanin and blue eyes
contain small amounts of eumelanin. The blue eye colour is a result
of minimal pigmentation of the retina, and reflection of the shorter
blue wavelengths of visible light in the iridial stroma. In eyes
where the colour appear neither blue or brown, but appear as a socalled intermediate eye colour (e.g. green or hazel), the amount of
eumelanin in the melanocytes vary. Some areas in the iris may be
blue and some may be brown. Humans perceive eye colour as a
whole and observe the combination of blue and brown in the eye. A
certain combination of blue and brown colours may appear green
or hazel, but no such pigment exists in the iris.
Several genes, including TYR, TYRP1, SLC24A4, SLC45A2, ASIP, IRF4,
HERC2 and OCA2, influence eumelanogenesis, and SNPs in these
genes are found to be associated with eye colour [4,5,7–16]. In a
comprehensive study of 6,168 Dutch individuals, 37 eye colour
associated SNPs were investigated [17]. Six of these SNPs, HERC2
rs12913832, OCA2 1800407, SLC24A4 rs12896399, SLC45A5
rs16891982, TYR rs1393350 and IRF4 12203592 were selected
and multiplexed in a single base extension (SBE) assay termed the
IrisPlex [1].
Human eye colours have for the large majority of studies been
described qualitatively by assigning each eye to a predefined group.
Even though trained professionals were used to assign each eye to a
category, some bias must be expected. This bias makes it difficult to
reproduce the findings. Especially, in studies where genes or
markers were associated to minor variations in eye colour. By using
objective, software based analysis of the eye colours, the human bias
can be limited. Furthermore, a standardized and objective way of
analysis makes it easier to compare studies on eye colour genetics.
Quantitative measurements of iris colour have only been used in a
few studies. A combination of luminosity and RGB (Red–Green–
Blue) values from digital spectroscopy were used to make a
continuous representation of eye colours [15]. The H and S
components of the HSV (hue–saturation–value) colour spaces were
used to investigate association between genetic variants and eye
colour [18]. Furthermore, automated and quantitative measurements of iris colour was performed using an average CIELAB colour
space (from the RGB space) value extracted from a specific region in
the iris [19]. This method showed association between rs12913832
and L*, a* and b* components in a European population. However,
only a small part of the iris was used for the analysis. Also, RGB and
CIELAB measurements are greatly influenced by lighting conditions.
Therefore, standardisation of the photographic setup and normalization of the photos is necessary.
In this work, we used the HSV colour space to measure the eye
colour. The S component of the HSV colour space separated the
blue and brown areas of the iris efficiently and may be preferred in
quantitative eye colour investigations [20]. The V component of
the HSV space was used to normalize the light intensity
distribution and thus, minimize the intrinsic external effects of
image acquisition.We introduce a Pixel Index of the Eye (PIE-score)
based on the blue and brown pixels in the eye images. To test the
applicability of the PIE-score, the six IrisPlex SNPs were compared
to the PIE-scores of 570 individuals. The PIE-score was also
compared to a subjective eye colour categorization similar to the
ones used in previous studies [1,2,8,11,16,21]. We also compared
the PIE-scores with the two previously published prediction
models, the IrisPlex methods [17] and the Snipper model [2].
2. Materials and methods
2.1. Samples and DNA purification
Blood samples from 605 healthy, unrelated individuals were
collected at the Section of Forensic Genetics, Department of
Forensic Medicine, Faculty of Health and Medical Sciences,
509
University of Copenhagen, the Ringens Health Care Centre,
Stockholm and through The Danish Blood Donor Study
(www.dbds.dk) at the Blood Bank, Glostrup Hospital. DNA was
purified from 200 mL of blood using the DNA Blood Mini Kit
(Qiagen) as recommended by the manufacturer. DNA was eluted in
50 mL of AE Buffer (Qiagen). The study was approved by the Danish
Ethical Committee (H-4-2009-125 and M-20090237).
2.2. Digital photographs
Photographs were taken at a distance of approximately 10 cm
in ‘‘Raw’’ or ‘‘jpeg’’ format with a Canon EOS 5D Mark V with ISO
800, shutter 1/100 and AV 18 using a Canon EF 100 mm f/2.8 L IS
USM Macro Lens with manual focus. The white balance of ‘‘Raw’’
format photographs was changed to ‘‘Flash’’ using the Picture style
editor software (Canon).
2.3. SNP-typing
The IrisPlex SNPs, rs12913832, rs1800407, rs12896399,
rs16891982, rs1393350 and rs12203592 were typed as a part of a
PCR multiplex with 32 SNPs (Supplementary Table S1). Samples
were SNP-typed using the iPLEX1 Gold Kit (Sequenom) in a final
reaction volume of 6 ml. The PCR contained 2 ml DNA, 0.5 ml 10
Buffer, 0.8 ml 25 mM MgCl2, 0.1 ml 25 mM dNTP mix, 1.3 mL
0.5 mM primer mix (DNA Technology. Aarhus, Denmark), 0.2 mL
5 U/ml HotStarTaq and 1.1 ml H2O. The PCR was performed in a
GeneAmp1 PCR System 9700 thermal cycler (Life Technologies –
LT) with the following conditions: denaturation at 94 8C for 2 min
followed by 45 cycles of 94 8C for 20 s, 62 8C for 30 s, 72 8C for
1 min, followed by 72 8C for 3 min. The PCR products were treated
with Shrimp Alkaline Phosphatase (SAP) (Sequenom) in a
GeneAmp1 PCR system 9700 thermal cycler (LT) at 37 8C for
40 min and 85 8C for 5 min. The SBE reaction contained 8 ml SAP
treated PCR products and 2 ml iPLEX1 mix (Sequenom). The
iPLEX1 mix contained 0.2 ml 10 iPLEX1 buffer, 0.2 mL iPLEX1Termination mix, 0.94 ml primer mix (DNA Technology), 0.04 ml
iPLEX1-enzyme and 0.62 ml H2O. The SBE reaction was performed
in a GeneAmp1 PCR system 9700 thermal cycler (LT) with the
following conditions: Denaturation at 94 8C for 30 s followed by
40 cycles of 94 8C for 5 s, 52 8C for 5 s and 80 8C 5 s, 52 8C for 5 s and
80 8C for 5 s, 52 8C for 5 s and 80 8C for 5 s, 52 8C for 5 and 80 8C for
5 s, 52 8C for 5 s and 80 8C 5 s, followed by 72 8C for 3 min. A total of
40 ml of molecular grade water and ion exchange resin (Sequenom) was added to each sample. Samples were rotated for
approximately 4 h and kept in the refrigerator for up to 4 days
before spotting. Samples were spotted in duplicates using the
RS1000 Nanospotter (Sequenom) and visualized on the MassARRAY1 Analyzer 4 System (Sequenom) using the autorun settings.
Samples were analyzed with Typer Analyzer 4 (Sequenom) and
were autoclustered using a signal-to-noise ratio of 7. Clusterplots
were visually inspected, and outliers were further investigated. All
samples were run in duplicates.
SNP types were compared between spots and duplicate typings
using a custom function (PlateCompare) of the statistic software R
(R core team, version 2.11.0, URL http://www.R-project.org).
2.4. Statistical analyses
All statistical calculations were performed using R (R core team,
version 2.11.0, URL http://www.R-project.org). The PIE-score
distribution failed the Shapiro–Wilks test for normality (p < 108).
Hence, all between-groups-comparisons were performed using the
Wilcoxon rank sum test. Bonferroni multiple corrections was
applied. Correlations were investigated with Spearman’s correlation
test. Outliers were identified as observations outside 1.5 times the
interquartile range (whiskers) of the data.
510
J.D. Andersen et al. / Forensic Science International: Genetics 7 (2013) 508–515
Fig. 1. Detection of Iris and eyelid boundaries (A), used for Iris mask (B) and Iris map, (C) generation. A: Original eye image, with detected iris and eyelid boundaries. B: Iris
mask accounting for the eyelid regions and reflections from the camera flashes. C: Iris map, generated by a polar transformation of the detected iris region in (A).
2.5. Digital iris extraction and PIE-score using the DIAT software
All photographs were converted and standardized to ‘‘jpeg’’
Images 639 426 pixels. Due to varying illumination, a low
frequency intensity variation could be observed in some images.
The illumination of the entire image was corrected using a method
inspired by bias field correction employed in magnetic resonance
imaging. A set of pixel observations were sampled on the skin
surrounding the eyes and the idea was to correct the image so they
would all have the same intensity after the correction. A twodimensional thin plate spline (TPS) [22] was fitted to the intensity
of the samples to estimate the low-frequency background
illumination trend. The correction was performed by subtracting
the TPS fit from the actual image. The intensity correction was done
to facilitate the iris extraction and only affected the value
components of the HSV (hue, saturation, value) colour space.
Hence it did not affect the saturation component used in the final
PIE-score calculation.
The custom designed software DIAT for iris extraction and PIEscore calculation were written in MATLAB version 7.12.0.635 (The
MathWorks, Inc.). The iris region was automatically extracted from
the intensity corrected image using an algorithm based on [20],
with the modifications described in [23]. The DIAT-software is
available upon request.
The PIE-score is based on the number of pixels that were
labelled blue and brown. The blue and brown pixels in the iris maps
Fig. 2. PIE-scores and correlations with human eye colour classifications. A: Boxplots of the PIE-scores of 152 images that were intuitively categorized into six eye colour
groups, Light blue, Dark blue, Light int. (intermediate), Dark int. Light brown and Dark brown. B: Examples of eye colour images and the corresponding PIE-scores.
J.D. Andersen et al. / Forensic Science International: Genetics 7 (2013) 508–515
(Fig. 1) were located using a Markov Random Field (MRF) based
segmentation [24]. The segmentation was performed on the
saturation component of the HSV colour space as this component
showed the largest difference in pixel value between blue and
brown regions (data not shown). A blue–brown probability score
was assigned to each pixel and by applying a hard threshold, each
pixel was categorized as either blue or brown.
The PIE-score was calculated according to the equation:
Number of pixels labeled blue
PIE ¼
Number of pixels labeled brown
:
Number of pixels labeled blue
þ Number of pixels labeled brown
The PIE-score ranged from 1 to 1. The value of 1 was
observed when only brown pixels were found in the iris region;
and the value of 1 was observed when only blue pixels were found
in the iris region. The DIAT software allowed manual correction of
eye lid boundaries.
2.6. Evaluation of eye colours
Six untrained individuals were assigned to evaluate the iris
colour of 152 digital eye images according to six categories (blue,
dark blue, light intermediate, dark intermediate, light brown and
dark brown). The six individuals were placed in front of the same
511
screen at the same distance and asked to assess the colour category
of each eye image. Each eye image was assigned an eye colour
category by taking the median of the assessed categories.
2.7. Prediction of eye colours using IrisPlex and Snipper prediction
methods
We compared the predictions of the two eye colour prediction
methods, IrisPlex [1] and Snipper [2], with the PIE-score. The IrisPlex
eye colour prediction was based on a multinomial, logistic
regression model with three outcomes (blue, intermediate and
brown) [17] and a probability threshold of 0.7 for classification [1].
Snipper used a Bayesian likelihood based classification. A likelihood
ratio threshold of 3:1 for classification was applied as recommended
[2].
3. Results
3.1. Evaluation of the eye images
A total of 605 digital eye images were evaluated using the DIAT
software. It was possible to extract the iris and calculate the PIEscore in 570 (94%) images. The software failed extraction of 35 iris
regions, mainly in images where the eyelids covered a large part of
the iris or in brown eye images where the pupil was difficult to
differentiate from the dark iris.
Fig. 3. Boxplots of the PIE-scores of the six different IrisPlex SNP-types. Boxplots are shown for the PIE-scores and pairwise comparisons of the various observations of the
IrisPlex SNPs. Statistical significance (Sign.): ****p < 106, ***p < 103, **p < 102, *p < 0.05, -p > 0.05. SNP: SNP-type. Med.: Median.
512
J.D. Andersen et al. / Forensic Science International: Genetics 7 (2013) 508–515
Of the 570 images, we had to redefine the iris region in 88
images manually due to disturbing flash reflections and/or
inclusion of eye lids in the iris region. All analyses shown below
were performed on the 570 approved iris extractions.
The eye colours of 152 images were intuitively assigned by six
individuals to one of six colour categories (Fig. 2A). Fig. 2B shows
examples of eye images and their respective PIE-scores. A high
correlation (Spearman correlation = 0.81) between the calculated
PIE-scores and the intuitively assigned colour categories was
observed (Fig. 2A). In general, eye colours categorized intuitively
as blue or brown showed less variation in PIE-scores compared to
those categorized as intermediate. This was expected as the
intermediate eye colours are difficult to assign to a predefined
category. The differences in PIE-scores between light blue and dark
blue categories and between light brown and dark brown
categories were not significant (p > 0.05). By combining the light
blue and dark blue categories into one category and light brown
and dark brown categories into another one, an even better
correlation with the PIE-score (Spearman = 0.86) was observed.
3.2. Associations of the six IrisPlex SNPs with the PIE-scores
The association of the six IrisPlex SNPs, HERC2 rs12913832,
OCA2 rs1800407, SLC24A4 rs12896399, SLC45A2 rs16891982, TYR
rs1393350 and IRF4 12203592 with the PIE-score was investigated
(Fig. 3). As expected, the strongest association was observed for
the HERC2 rs12913832 SNP. There was a significant difference
(p < 106) in the PIE-scores of the individuals typed as HERC2
rs12913832 G (median PIE (PIEm) = 0.99) and rs12913832 GA
(PIEm = 0.71) or A (PIEm = 0.87). Furthermore, we observed
significant associations with the PIE-score of the OCA2 rs1800407,
SLC24A4 rs12896399, SLC45A2 rs16891982 and TYR rs1393350
SNPs. However, the IRF4 12203592 SNP was not significantly
associated to the PIE-score.
Among the individuals typed as HERC2 rs12913832 GA (Fig. 4),
a large effect of OCA2 rs1800407 on the PIE-score was observed.
Individuals with the combination of HERC2 rs12913832 GA/OCA2
rs1800407 C had significantly (p < 0.05) lower PIE-scores
(PIEm = 0.76) than individuals with the combination of HERC2
rs12913832 GA/OCA2 rs1800407 CT (PIEm = 0.21). We did not
observe any individuals with the combination of HERC2
rs12913832 GA/OCA2 rs1800407 T.
Among the individuals typed as HERC2 rs12913832 G (Fig. 5), a
small but significant effect on the PIE-score was found for the
SLC24A4 rs12896399 and TYR rs1393350 SNPs. Individuals with
the combination of HERC2 rs12913832 G/SLC24A4 rs12896399 T
had a significantly (p < 103) higher PIE-score (PIEm = 1) compared to individuals with the combination of HERC2 rs12913832 G/
SLC24A4 rs12896399 GT (PIEm = 0.98) or HERC2 rs12913832 G/
SLC24A4 rs12896399 G (PIE-score = 0.98). Individuals with the
combination of HERC2 rs12913832 G/TYR rs1393350 G showed a
significant (p < 102) lower PIE-score (PIEm = 0.98) than individuals with the combination of HERC2 rs12913832 G/TYR rs1393350
GA (PIEm = 0.99).
Fig. 4. Boxplots of the PIE-scores for individuals typed as HERC2 rs12913832 GA. Boxplots are shown for the PIE-scores and pairwise comparisons of the various observations
of the IrisPlex SNPs. Statistical significance (Sign.): *p < 0.05, -p > 0.05. SNP: SNP-type. Med.: Median.
J.D. Andersen et al. / Forensic Science International: Genetics 7 (2013) 508–515
513
Fig. 5. Boxplots of the PIE-scores for individuals typed as HERC2 rs12913832 GG. Boxplots are shown for the PIE-scores and pairwise comparisons of the different observations
of the IrisPlex SNPs. Statistical significance (Sign.): ***p < 103, **p < 102, *p < 0.05, -p > 0.05. SNP: SNP-type. Med.: Median.
We only observed 17 individuals with HERC2 rs12913832 A and
found no significant effect on the PIE-score when HERC2
rs12913832 A was combined with any other SNPs.
3.3. Comparison of the PIE-score with eye colour predictions with
IrisPlex and Snipper
A total of 76 different IrisPlex profiles were identified, twenty of
which were found in more than five individuals. The distribution of
PIE-scores for each of the twenty profiles is shown in Fig. 6. It was
evident that HERC2 rs12913832 was the main predictor of eye
colour with HERC2 rs12913832 G as a blue eye colour predictor and
HERC2 rs12913832 GA as a non-blue eye colour predictor. No
IrisPlex profiles with HERC2 rs12913832 A was found in more than
5 individuals and, therefore, no prediction was made for these
profiles.
The eye colours for each of the twenty IrisPlex profiles were
predicted as either blue, intermediate or brown using the IrisPlex [1]
and the Snipper [2] prediction methods (Fig. 6). For HERC2
rs12913832 G individuals, there was a high correlation between
the IrisPlex and Snipper predictions. All IrisPlex profiles with HERC2
rs12913832 G were predicted as blue by the IrisPlex prediction
methods. In comparison, Snipper predicted all HERC2 rs1291313832
G as blue, except for two that were not classified. Large PIE-score
variations were observed for the IrisPlex profiles predicted as blue. In
total, 32% (126/389) of the individuals classified as blue eyed had a
PIE-score outside the range of the dark blue and light blue boxplotwhiskers (PIE = 1–0.96; Fig. 2). If the lower boxplot-whisker of the
light intermediate category was used as the lower limit, 14% (53/
389) were outside the boxplot-whiskers range (PIE = 1–0.79). This
indicated that the eyes predicted as blue eyes by the IrisPlex and
Snipper predictors included not only blue eye colours, but also most
of the light, non-blue eye colours. The combinations with HERC2
rs12913832 GA were difficult to classify into one of the three eye
colour groups with either of the prediction models. Three out of nine
Irisplex profiles with HERC2 rs12913832 GA were predicted as
brown by the Irisplex prediction method. One of these was predicted
as intermediate by the Snipper prediction method, whereas two
profiles were not classified. Snipper furthermore predicted another
IrisPlex profile with HERC2 rs12913832 GA as intermediate. This
IrisPlex profile was not classified by the IrisPlex prediction model.
None of the remaining five HERC2 rs12913832 GA IrisPlex profiles
were classified by either of the prediction methods.
4. Discussion
We calculated the number of blue and brown pixels in high
resolution digital images of irides using the DIAT software and
defined a quantitative value for eye colour that we named the Pixel
Index of the Eye (PIE). The success rate of the automatic iris
extraction was 94%. The major causes of inaccurate iris extraction
were incorrect localisation of the pupil area in brown eyes and/or
incorrect detection of the outer boundary of the iris. The automatic
iris extraction relied mainly on light contrasts in the eye image. The
success rate may be improved by taking images while the
participant uses the fingers to keep the eyelids apart. This will
514
J.D. Andersen et al. / Forensic Science International: Genetics 7 (2013) 508–515
Fig. 6. Comparison of the PIE-scores with the eye colour classifications of the IrisPlex and Snipper prediction methods. IrisPlex profiles with five or more observations were
considered. The number of observations is shown above each plot. Genotypes are listed as HERC2 rs12913832, OCA2 rs1800407, SLC24A4 rs12896399, SLC45A2 rs16891982,
TYR rs1393350 and IRF4 rs12203592, respectively. Silent alleles were not considered by either of the prediction models. Homozygous genotypes are written as, e.g. GG and not
G, even though the presence of silent alleles cannot be excluded. Prediction from the IrisPlex and Snipper are listed below the genotypes. IrisPlex probability predictions >0.7
are shown in bold. Snipper predictions are shown as either blue (Blue), intermediate (Int.) or not classified (N.c.). Snipper predictions were classified when the likelihood ratio
>3:1.
ensure that the iris outer boundary is between the sclera and the
iris and that eyelids do not overlap with the circular iris.
Minimizing reflections in the eyes should be a priority when
designing the photographic setup because reflections are bright
and may be misinterpreted by the software as blue pixels.
We evaluated the PIE-score by comparing the PIE-scores of 152
eye images with human classifications of the same images. The
PIE-score correlated well with the human perception of eye colour.
The best fit was obtained when the eye colours were categorized
into four groups (blue, light intermediate, dark intermediate and
brown). The PIE-score was not able to differentiate between light
and dark blue or between light and dark brown eye colours. This
was expected because the PIE-score is based on the numbers of
brown and blue pixels and because each pixel in the iris region was
classified as either blue or brown. Thus, light and dark blue pixels
were classified as blue and light and dark brown pixels were
classified as brown. This limitation will be addressed in future
versions of the software.
The PIE-score may be used as a universal measurement of eye
colour. It will allow standardization of objective eye colour
interpretation and it will be possible to directly compare the
results from different studies, which is difficult today because
subjective interpretations of eye images are widely used.
Furthermore, the PIE-score is reflecting a continuous measure
that may be used to detect small differences in eye colour. This will
be especially valuable for differentiation of intermediate eye
colours and may be useful for studies of associations of genetic
markers with more subtle differences in eye colour. The PIE-score
may also be used in forensic genetic case work. Reporting the
interval of observed PIE-scores for a given IrisPlex profile (e.g. as in
Fig. 6) together with examples of eye images from individuals with
the same IrisPlex profile would be a simple and illustrative
presentation of the results. An eye colour category and a likelihood
calculation for that category may also be reported based on the
profile and the expected PIE-score.
We tested the association of the PIE-score with the six SNPs in
the IrisPlex [1]. As expected, the HERC2 rs12913832 SNP explained
the majority of the variation in the PIE-score. After correcting for
the effect of HERC2 rs12913832, three SNPs SLC24A4 rs12896399,
TYR rs1393350 and OCA2 rs1800407 influenced the PIE-score
significantly, whereas SLC45A2 rs16891982 and IRF4 rs12203592
did not. We observed a large variation in the PIE-score among the
HERC2 rs12913832 GA individuals. This observation was consistent with the hypothesis that HERC2 rs12913832 exerted incomplete penetrance. Only OCA2 rs1800407 influenced the PIE-scores
significantly in the HERC2 rs12913832 GA group. OCA2 and HERC2
are separated by 135,000 bps on chromosome 15. The haplotype of
the HERC2–OCA2 region was previously suggested to be the key to
understand the variations in human eye colour [2,16]. An epistatic
effect between OCA2 rs1800407 and HERC2 rs12913832 was
furthermore shown to increase the accuracy of intermediate eye
colour prediction [25]. The region surrounding HERC2 rs12913832
was recently identified as an enhancer of OCA2 expression [24]. The
derived allele G of HERC2 rs12913832 reduced the binding of the
transcription factor HLTF and the interaction between the HERC2
rs12913832 region and the OCA2 promoter [26]. OCA2 rs1800407 is
a missense mutation (arginine to glutamine) that may affect the
protein function. OCA2 rs1800407 was previously suggested as a
penetrance-modifier of HERC2 rs12913832 [4]. We hypothesize, in
concordance with others [27], that the OCA2 allele rs1800407 T
decreases the pigmentation level of the eye when found in cis
phase with the HERC2 allele rs12913832 A. If the OCA2 allele
J.D. Andersen et al. / Forensic Science International: Genetics 7 (2013) 508–515
rs1800407 T is found in cis phase with the HERC2 allele rs12913832
G, it will display little effect on the eye colour because the
transcription of OCA2 is reduced. A small, but significant
association of SLC24A4 rs12896399 and TYR rs1393350 with the
PIE-score was observed among HERC2 rs1291382 G individuals.
However, the difference in average PIE-scores were small and the
predictive value of other IrisPlex SNPs, besides HERC2 rs1291382
and OCA2 rs1800407, seemed very small.
The Irisplex and Snipper prediction models generally agreed on
the prediction of IrisPlex profiles with HERC2 rs12913832 G.
However, they either failed to classify or reached inconsistent
classifications for the IrisPlex profiles with HERC2 rs12913832 GA.
Furthermore, 14% of the Irisplex profiles classified as blue eye
colour by the prediction methods originated from individuals with
PIE-scores that were lower than the lower boxplot-whisker of the
light intermediate category (PIE = 0.79). Thus, the eye colour of
these individuals would most likely be categorized as dark
intermediate or brown by human inspection. This shows that it
is important to inform the recipients of the forensic case report that
eye colour predictions based on the Irisplex results may be
inaccurate and should be handled with care.
An epistatic relationship was proposed for HERC2 and OCA2
[1,28,29] and the HERC2-OCA2 region remains the best candidate
for identifying DNA sequences that may explain the large PIE-score
variation in HERC2 rs12913832 GA individuals as well as the nonblue eye colours among individuals typed as HERC2 rs12913832 G.
However, more research is needed to elucidate the effect of the
phase and the possible epistatic effects of the genetic variations in
the HERC2-OCA2 region.
Conflict of interest
None.
Acknowledgements
Statistical assistance: Torben Tvedebrink, MSc, PhD. Technical
assistance: Trine Leerhøj Hansen. Blood sampling: Pernille
Jensen, Anja Jørgensen, Rikke Hansen, staff at the Blood Bank,
Glostrup Hospital. The Danish Blood Donor Study was supported
by The Danish Council for Independent Research, Medical
Sciences (09-069412) and Innovation and The Danish Regions.
Also a part of this project was supported by the Ellen and Aage
Andersen’s foundation.
Appendix A. Supplementary data
Supplementary data associated with this article can be found, in
the online version, at http://dx.doi.org/10.1016/j.fsigen.2013.05.003.
References
[1] S. Walsh, F. Liu, K.N. Ballantyne, M. van Oven, O. Lao, M. Kayser, IrisPlex: a
sensitive DNA tool for accurate prediction of blue and brown eye colour in the
absence of ancestry information, Forensic Sci. Int. 3 (2010) 170–180.
[2] Y. Ruiz, C. Phillips, A. Gomez-Tato, J. Alvarez-Dios, C.M. Casares de, R. Cruz, O.
Maronas, J. Sochtig, M. Fondevila, M.J. Rodriguez-Cid, A. Carracedo, M.V. Lareu,
Further development of forensic eye color predictive tests, Forensic Sci. Int. Genet.
7 (1) (2013) 28–40.
[3] D. Posthuma, P.M. Visscher, G. Willemsen, G. Zhu, N.G. Martin, P.E. Slagboom, E.J.
De Geus, D.I. Boomsma, Replicated linkage for eye color on 15q using comparative
ratings of sibling pairs, Behav. Genet. 36 (1) (2006) 12–17.
[4] R.A. Sturm, D.L. Duffy, Z.Z. Zhao, F.P. Leite, M.S. Stark, N.K. Hayward, N.G. Martin,
G.W. Montgomery, A single SNP in an evolutionary conserved region within
intron 86 of the HERC2 gene determines human blue–brown eye color, Am. J.
Hum. Genet. 82 (2) (2008) 424–431.
515
[5] H. Eiberg, J. Troelsen, M. Nielsen, A. Mikkelsen, J. Mengel-From, K.W. Kjaer, L.
Hansen, Blue eye color in humans may be caused by a perfectly associated founder
mutation in a regulatory element located within the HERC2 gene inhibiting OCA2
expression, Hum. Genet. 123 (2008) 177–187.
[6] P.D. Imesch, I.H. Wallow, D.M. Albert, The color of the human eye: a review of
morphologic correlates and of some conditions that affect iridial pigmentation,
Surv. Ophthalmol. 2 (Suppl. 41) (1997) S117–S123.
[7] M. Kayser, F. Liu, A.C. Janssens, F. Rivadeneira, O. Lao, K. van Duijn, M. Vermeulen,
P. Arp, M.M. Jhamai, W.F. van Ijcken, J.T. den Dunnen, S. Heath, D. Zelenika, D.D.
Despriet, C.C. Klaver, J.R. Vingerling, P.T. de Jong, A. Hofman, Y.S. Aulchenko, A.G.
Uitterlinden, B.A. Oostra, C.M. van Duijn, Three genome-wide association studies
and a linkage analysis identify HERC2 as a human iris color gene, Am. J. Hum.
Genet. 82 (2) (2008) 411–423.
[8] P. Sulem, D.F. Gudbjartsson, S.N. Stacey, A. Helgason, T. Rafnar, K.P. Magnusson, A.
Manolescu, A. Karason, A. Palsson, G. Thorleifsson, M. Jakobsdottir, S. Steinberg, S.
Palsson, F. Jonasson, B. Sigurgeirsson, K. Thorisdottir, R. Ragnarsson, K.R. Benediktsdottir, K.K. Aben, L.A. Kiemeney, J.H. Olafsson, J. Gulcher, A. Kong, U. Thorsteinsdottir, K. Stefansson, Genetic determinants of hair, eye and skin
pigmentation in Europeans, Nat. Genet. 39 (12) (2007) 1443–1452.
[9] P. Sulem, D.F. Gudbjartsson, S.N. Stacey, A. Helgason, T. Rafnar, M. Jakobsdottir, S.
Steinberg, S.A. Gudjonsson, A. Palsson, G. Thorleifsson, S. Palsson, B. Sigurgeirsson,
K. Thorisdottir, R. Ragnarsson, K.R. Benediktsdottir, K.K. Aben, S.H. Vermeulen,
A.M. Goldstein, M.A. Tucker, L.A. Kiemeney, J.H. Olafsson, J. Gulcher, A. Kong, U.
Thorsteinsdottir, K. Stefansson, Two newly identified genetic determinants of
pigmentation in Europeans, Nat. Genet. 40 (7) (2008) 835–837.
[10] J. Han, P. Kraft, H. Nan, Q. Guo, C. Chen, A. Qureshi, S.E. Hankinson, F.B. Hu, D.L.
Duffy, Z.Z. Zhao, N.G. Martin, G.W. Montgomery, N.K. Hayward, G. Thomas, R.N.
Hoover, S. Chanock, D.J. Hunter, A genome-wide association study identifies novel
alleles associated with hair color and skin pigmentation, PLoS Genet. 4 (5) (2008)
e1000074–e1000080.
[11] T. Frudakis, M. Thomas, Z. Gaskin, K. Venkateswarlu, K.S. Chandra, S. Ginjupalli, S.
Gunturi, S. Natrajan, V.K. Ponnuswamy, K.N. Ponnuswamy, Sequences associated
with human iris pigmentation, Genetics 165 (4) (2003) 2071–2083.
[12] J. Graf, R. Hodgson, A. van Daal, Single nucleotide polymorphisms in the MATP
gene are associated with normal human pigmentation variation, Hum. Mutat. 25
(3) (2005) 278–284.
[13] P.A. Kanetsky, J. Swoyer, S. Panossian, R. Holmes, D. Guerry, T.R. Rebbeck, A
polymorphism in the agouti signaling protein gene is associated with human
pigmentation, Am. J. Hum. Genet. 70 (3) (2002) 770–775.
[14] D.L. Duffy, G.W. Montgomery, W. Chen, Z.Z. Zhao, L. Le, M.R. James, N.K. Hayward,
N.G. Martin, R.A. Sturm, A three-single-nucleotide polymorphism haplotype in
intron 1 of OCA2 explains most human eye-color variation, Am. J. Hum. Genet. 80
(2) (2007) 241–252.
[15] T. Frudakis, T. Terravainen, M. Thomas, Multilocus OCA2 genotypes specify human
iris colors, Hum. Genet. 122 (3–4) (2007) 311–326.
[16] J. Mengel-From, C. Borsting, J.J. Sanchez, H. Eiberg, N. Morling, Human eye colour
and HERC2, OCA2 and MATP, Forensic Sci. Int. Genet. 4 (2010) 323–328.
[17] F. Liu, D.K. van, J.R. Vingerling, A. Hofman, A.G. Uitterlinden, A.C. Janssens, M.
Kayser, Eye color and the prediction of complex phenotypes from genotypes, Curr.
Biol. 19 (5) (2009) R192–R193.
[18] F. Liu, A. Wollstein, P.G. Hysi, G.A. Ankra-Badu, T.D. Spector, D. Park, G. Zhu, M.
Larsson, D.L. Duffy, G.W. Montgomery, D.A. Mackey, S. Walsh, O. Lao, A. Hofman, F.
Rivadeneira, J.R. Vingerling, A.G. Uitterlinden, N.G. Martin, C.J. Hammond, M.
Kayser, Digital quantification of human eye color highlights genetic association of
three new loci, PLoS Genet. 6 (2010) e1000934–e1000940.
[19] M. Edwards, A. Gozdzik, K. Ross, J. Miles, E.J. Parra, Technical note: quantitative
measures of iris color using high resolution photographs, Am. J. Phys. Anthropol.
147 (1) (2012) 141–149.
[20] S. Harder, S.R. Christoffersen, C. Borsting, P. Johansen, N. Morling, J.D. Andersen,
A.L. Dahl, R.R. Paulsen, Medical Computer Vision. Recognition Techniques and
Applications in Medical Imaging, Lect. Notes Comput. Sci. 7766 (2013) 244–253.
[21] J. Mengel-From, T.H. Wong, N. Morling, J.L. Rees, I.J. Jackson, Genetic determinants of
hair and eye colours in the Scottish and Danish populations, BMC Genet. 10 (88)
(2009).
[22] F.L. Bookstein, Principal warps: thin-plate splines and the decomposition of
deformations, pattern analysis and machine intelligence, IEEE Transactions 11
(6) (1989) 567–585.
[23] M.M. Nielsen, S.T. Henriksen, Iris Extraction from Human Eye Images, Technical
University of Denmark, Informatics and Mathematical Modeling, Denmark, 2012.
[24] S.Z. Li, Markov Random Field Modeling in Image Analysis, Springer-Verlag New
York Inc., 2009.
[25] E. Pospiech, J. Draus-Barini, T. Kupiec, A. Wojas-Pelc, W. Branicki, Gene–gene
interactions contribute to eye colour variation in humans, J. Hum. Genet. 56 (6)
(2011) 447–455.
[26] M. Visser, M. Kayser, R.J. Palstra, HERC2 rs12913832 modulates human pigmentation by attenuating chromatin-loop formation between a long-range enhancer
and the OCA2 promoter, Genome Res. 22 (3) (2012) 446–455.
[27] D. White, M. Rabago-Smith, Genotype–phenotype associations and human eye
color, J. Hum. Genet. 56 (1) (2011) 5–7.
[28] W. Branicki, U. Brudnik, T. Kupiec, P. Wolanska-Nowak, A. Szczerbinska, A. WojasPelc, Association of polymorphic sites in the OCA2 gene with eye colour using the
tree scanning method, Ann. Hum. Genet. 72 (Pt 2) (2008) 184–192.
[29] J.D. Andersen, P. Johansen, H.S. Mogensen, C. Borsting, N. Morling, Eye colour and
SNPs in Danes, Forensic Sci. Int. Genet. Suppl. 3 (1) (2011) 151–152.