The Unz Review: An Alternative Media Selection
A Collection of Interesting, Important, and Controversial Perspectives Largely Excluded from the American Mainstream Media
Email This Page to Someone

 Remember My Information

Authors Filter?
Razib Khan
Nothing found
 TeasersGene Expression Blog
Eye color

Bookmark Toggle AllToCAdd to LibraryRemove from Library • BShow CommentNext New CommentNext New ReplyRead More
ReplyAgree/Disagree/Etc. More... This Commenter This Thread Hide Thread Display All Comments
These buttons register your public Agreement, Disagreement, Thanks, LOL, or Troll with the selected comment. They are ONLY available to recent, frequent commenters who have saved their Name+Email using the 'Remember My Information' checkbox, and may also ONLY be used three times during any eight hour period.
Ignore Commenter Follow Commenter
🔊 Listen RSS

Over at Scientific American Christie Wilcox has a post up with the provocative title, People With Brown Eyes Appear More Trustworthy, But That’s Not The Whole Story, which reports on a new PLoS ONE paper, Trustworthy-Looking Face Meets Brown Eyes. Like Christie I would enjoy illustrating this post with my own trustworthy and youthful brown eyed visage, but I worry that my mien is a bit on the sly side! In any case, what of the paper? Wilcox reviews the salient points of the results. In short, the issue here is that brown eyed men seem to have more ‘trustworthy faces’ than blue eyed men. When the eyes were digitally manipulated it turned out that color had no influence on perception. Rather, it was the correlation between eye color and facial proportion which which was driving the initial association. Christie finishes:

Given the importance of trust in human interactions, from friendships to business partnerships or even romance, these findings pose some interesting evolutionary questions. Why would certain face shapes seem more dangerous? Why would blue-eyed face shapes persist, even when they are not deemed as trustworthy? Are our behaviors linked to our bodies in ways we have yet to understand? There are no easy answers. Face shape and other morphological traits are partially based in genetics, but also partially to environmental factors like hormone levels in the womb during development. In seeking to understand how we perceive trust, we can learn more about the interplay between physiology and behavior as well as our own evolutionary history.

These findings do pose evolutionary questions, and I am interested in the correlations of behavior and eye color, and I have been so in the past. But, I have many qualms about the reliability of this literature now. When I read the post initially my eyes immediately sought out the plot you see above. Observe the intervals. Such intervals would not concern me in a simpler design, or in a model where the hypothesis already had prior support, but this is a peculiar and potentially counter-intuitive result. Additionally, if you peruse the methods section of the paper notice the attempts to control for demographic confounds in the linear regression model. There’s nothing wrong this, and due to the nature of the sample size (< 100) there was no chance that they’d get a perfectly ideal study population. But these sorts of statistical techniques are exactly the flavor of powerful tools which have been so abused in psychology and biomedical science, consciously and unconsciously. You can squeeze a correlation out of a rock.

This is an area where Jim Manzi would say we’re confronted with ‘high causal density.’ There is a literature which suggests that there are behavioral differences between blue eyed and brown eyed children. Unfortunately when Jason Malloy looked to see if there were differences in the huge NLSY data set he couldn’t find it. This doesn’t mean that there aren’t differences between individuals that differ by this phenotype, but the difference might be subtle, and one needs to tease apart various confounds. It reminds me somewhat of the confused literature on sexual attraction and MHC. There may be something there, but the papers often present contradictory results, or add a complexifying layer (e.g., you are attracted to individuals who smell different from your opposite sex parent, but not too different).

To cut to the chase on this specific paper and results, would I bet money that this will pan out? No. I think the results are probably not robust. Do I think that there are going to be biobehavioral differences between individuals with blue eyes vs. individuals with brown eyes? Here, much more cautiously, as my confidence is low, I think there actually will be found to be some phenomena of interest and difference. What one needs to do in this case I think is look at sibling pairs. Because as it happens due to the genetic architecture of eye color inheritance in Europeans you have a huge potential sample space of siblings with different eye colors, who share much genetically, and a common home environment.

Which brings me to genetics and evolution. Though I might nitpick with the methods and results of the paper which Christie reviewed above, I think they’re defensible, as far as it goes. But some of the discussion really leaves me scratching my head:

Therefore, we tentatively suggest that a combination of sex linkage and sexual selection is the most probable explanation for the reported covariance between brown eyes and trustworthy-looking faces. Also, the blue-eyed phenotype is now abundant in Northern Europe and hence should have some kind of adaptive advantage, most likely one favored by sexual selection…that compensates for the loss of perceived trustworthiness. The trade-off between a preference for colorful and visible physical features and the advantage of a trustworthy-looking face might have contributed to the high variability of European eye and hair color.

Consider this sentence: the dry earwax phenotype is now abundant in Eastern Asia and hence it should have some kind of adaptive advantage. Just because a trait is abundant does not mean that it is selectively advantageous. Rather, pleiotropy means that traits without advantage may spread, just as hitchhiking during a selective sweep can result in the spread of alleles which are not the direct targets of selection. Though the authors allude to the genetic literature (and it is cited), they do not explore it in much detail. This is a shame, because the genetics of blue eyes have been well explored in the past 5 years. I’d hazard to assert that we now understand it.

Heterozygote (my daughter)

The inheritance pattern of blue and brown eyes was one of the classic illustrations of the recessive expression of phenotypes in Mendelian genetics. In other words, two blue eyed parents could only give rise to blue eyed children, while brown eyed parents could potentially give rise to both eye colors, because the brown eyed phenotype was inclusive of homozygotes and heterozygotes (brown being dominant to blue). This is informative, but it is too simple a description of the way inheritance works in the real world. About ~75% of the blue vs. non-blue eye color variation in Europeans seems to be due to a locus which spans the genes OCA2 and HERC2. There are assorted modifier genes, but to a first approximation it is this locus which classic Mendelian models were detecting in terms of segregation within the population. But ~75% is not 100%, and there are more eye colors than blue and brown. In other words, eye color inheritance is complicated, but not too complicated.

Perhaps a more important point is that the OCA2-HERC2 region is not limited to iris pigmentation in its effect. This was originally a region where an albinism mutation was localized. There is evidence that it impacts skin color in Europeans and Asians. And, these variants in this region do seem to be targets of natural selection. One immediate thing that jumps out at you for the European variants is that they are characterized by a long block of the genome which is co-inherited together, a hallmark of recent natural selection. Second, it is important to note that the block is long for another reason. And that reason is why I’m skeptical that the reason that this region was selected initially for blue eye color. Though the original recessive expression single gene model is too simple, it is correct that most individuals with blue eyes tend to be homozygotes. This means that in the initial stage of the allele’s increase in frequency trajectory it will rise in proportion very slowly, because most variants will be in heterozygotes, and so would not be favored by natural selection. Rather than a very long homogeneous block you’d expect a narrower region, because recombination will have mixed & matched the region during the early phase. This may all sound abstruse, but evolutionary hypotheses are most persuasive when they rest on a solid genetic basis. We have a good understanding of the genetics of eye color, and a more modest one of its evolutionary history. We should leverage that.

It was my intent over the course of this post to back into a domain with lower ‘causal density.’ The genetics of eye color is not really simple, but it is intuitively tractable. In contrast the story outlined in the PLoS ONE strikes me as problematic because though the results are statistically significant in some specific conditions, the overall story is complicated, and requires some unpacking. A more general issue which goes at the heart of the problem of constructing plausible evolutionary stories for the origin of phenotypes is there are many, many, phenotypes. Probing for correlations across any pair of phenotypes, most of the time you won’t find one (at least to statistical significance). But the process will eventually yield correlations. Some will be giving you much insight, but many will be spurious.

So what is a future avenue of exploration of this topic? I’m interested in genetics, so you know where I would go. Look at the sibling pairs, and see if the correlation with face shape and eye color holds. But more importantly the genetics of facial morphology are finally starting to be elucidated. It turns out that the trait is highly polygenic, with each locus predicting only a small proportion of trait variance. To me this poses an immediate problem in attempting to posit a genetic correlation with eye color, since that trait has a genetic architecture where most of the variance is localized around one region of the genome. But the difference in face shape here may be much more subtle, and so not picked up in the GWAS analyses which have recently come out.

What I’m hoping for in the future are simple explanations of very large data sets. Here we got a somewhat complex explanation for a not so large data set.


🔊 Listen RSS

One of the most successful achievements of the “post-genomic era” has been the elucidation of the genetic architecture which undergird the variation in human pigmentation. I like to point out that in 2005 the geneticist Armand Leroi observed in his book Mutants that we didn’t know the genetics of normal variation in relation to the trait of skin color. In 2010 one couldn’t plausibly write that. We know the genes which control the vast majority of the interpopulational variation in human complexion. This is not due to human ingenuity, but the fortuitous hand that nature dealt us. Pigmentation is a very salient phenotype, evident by the classification of genetically very distinct populations in Africa, India and Oceania as “black.” But in terms of a genetic research project it has long been one of the ways to explore patterns of inheritance in model organisms such as mice, in particular in relation to coat patterns and pigment. And luckily for us, many of the genes which are implicated in pigment variation produce similar changes across diverse taxa. Additionally, the genetic architecture of human pigmentation variation is such that most of the variance is concentrated among a few loci of large effect. Concretely, it seems that well over 50% of the African-European difference in skin color as measured by reflectance of visible light is attributable to two genes, SLC24A5 and KITLG. In Europeans around 75% of the dichotomous variation between those with blue and non-blue eyes may be due to changes in the genomic region across HERC2 and OCA2 (these two genes are very near each other). These are the veritable low hanging fruit, amenable to studies with even small sample sizes and modest statistical power, so strong are the effects of the genetic variables.

And why is pigment important? Obviously there are social ramifications. But pigmentation is likely a major target of natural selection as well, as I suggested in relation to Neandertals. The results are sometimes confusing, but it does seem that pigmentation related loci are enriched in relation to those genomic regions which turn up as positive in tests of natural selection. Additionally, looking at variation around those genes which are correlated with lighter skin across Eurasia it also seems that it may be that our own lineage has become somewhat paler within the last 20,000 years, perhaps even more recently. And the same may have been true for our possible Neander-kin.

At the current rate in regards to pigmentation the age of revolutionary science may soon be over. Extraction of ancient DNA will probably resolve the rate and nature of evolutionary change, while further typing of current populations will flesh out our understanding of the variants responsible for normal human variation. To do that requires more than simply larger sample sizes or improved genomic techniques, it also requires better measurement. The utilization of reflectance indices in studies of study skin color are a step in the right direction, but a new paper in PLoS Genetics points the way toward the same in the study of eye color, Digital Quantification of Human Eye Color Highlights Genetic Association of Three New Loci:

We measured human eye color to hue and saturation values from high-resolution, digital, full-eye photographs of several thousand Dutch Europeans. This quantitative approach, which is extremely cost-effective, portable, and time efficient, revealed that human eye color varies along more dimensions than the one represented by the blue-green-brown categories studied previously. Our work represents the first genome-wide study of quantitative human eye color. We clearly identified 3 new loci, LYST, 17q25.3, TTC3/DSCR9, in contributing to the natural and subtle eye color variation along multiple dimensions, providing new leads towards a more detailed understanding of the genetic basis of human eye color. Our quantitative prediction model explained over 50% of eye color variance, representing the highest accuracy achieved so far in genomic prediction of human complex and quantitative traits, with relevance for future forensic applications.

The main improvement was decomposing the elements of pigment variation which contribute to the phenotypes which we observe and recognize in a gestalt manner through inspection. In particular, they focused on two parameters, hue (H) and saturation (S). It seems to me that H correspondences to the quality of color, and S to the quantity of color. Figure 1C shows how the eye color categories map onto the quantitative metrics (it is somewhat confusing that red = brown, but I assume brown is less useful for display purposes as a contrast with blue or green).


The r-squared here is such that ~60% of the variation of H can be explained by variation of S. In other words, the two are correlated. This is clear in the plot above, as you see a distribution which moves from brown to green to blue. In fact the researchers could take the variation in H and S, yank out the independent dimensions of variance, and find that one dimension accounted for 90% of the variation.

What they found by treating eye color as a quantitative phenotype, instead of a categorical one, is that though the primary loci already known to affect variation still showed up in their associations, there were secondary loci which now emerged. This makes sense, as the original procedure whereby what is really a continuous trait was transformed into several distinct categories removes information. Smaller effect loci which impact the change in trait value only on the margins would naturally not show up when you removed variation on the margins so as to collapse the trait into a few broad classes. In other words, if loci X had an effect large enough to shift eye color from blue to brown (as HERC2-OCA2 does) then it would show up. But if loci Y only affects trait value slightly it will be unlikely to shift the trait across categories, and so it would not be discovered.

Table 3 shows the effect of different variables, primarily SNPs on the different loci of interest. Beta simply is a measure of the effect of a variable within a linear model. The bigger the beta, the larger the effect on the variable you wish to predict (here, eye color).


SNPs within HERC2 and OCA2 are still the primary variates, as we’d expect. But other loci also affect the trait on the margins, effects only discernible with a better measurement of the trait. Additionally, there is some evidence that they have independent effects on the two dimensions of the trait in question. With more modest effects and less clear phenotypic measures (at least to human intuition) I would be cautious about overemphasizing these results, but clearly they’re in the right direction when it comes to filling in the smaller pieces to the puzzle of eye color.

And these aren’t simply academic questions. They’re of strong forensic interest. Many of the pigment related markers are very good at distinguishing populations, and, they allow us to accurately reconstruct the appearance of perpetrators for crimes where we only have genetic material. With the model they have here they can explain more than 50% of the variation in the two dimensions that they define of H and S. The effect sizes of the marginal loci are modest already, so one might be running into diminishing returns, but from what I gather this is already a significant improvement over eyewitness recollection in relation to eye color. Pigment is hopefully just the tip of the iceberg, perhaps at some point in the future we could predict the rough outlines of someone’s whole physiognomy so that computer reconstructions of appearance could rely primarily are retrieved genetic data.

Citation: Liu F, Wollstein A, Hysi PG, Ankra-Badu GA, & Spector TD (2010). Digital Quantification of Human Eye Color Highlights Genetic Association of Three New Loc PLoS Genetics : 10.1371/journal.pgen.1000934

• Category: Science • Tags: Eye color, Genetics, Genomics 
Razib Khan
About Razib Khan

"I have degrees in biology and biochemistry, a passion for genetics, history, and philosophy, and shrimp is my favorite food. If you want to know more, see the links at"