A Genomewide Association Study of Skin Pigmentation in a South Asian Population

Abstract

We have conducted a multistage genomewide association study, using 1,620,742 single-nucleotide polymorphisms to systematically investigate the genetic factors influencing intrinsic skin pigmentation in a population of South Asian descent. Polymorphisms in three genes—SLC24A5, TYR, and SLC45A2—yielded highly significant replicated associations with skin-reflectance measurements, an indirect measure of melanin content in the skin. The associations detected in these three genes, in an additive manner, collectively account for a large fraction of the natural variation of skin pigmentation in a South Asian population. Our study is the first to interrogate polymorphisms across the genome, to find genetic determinants of the natural variation of skin pigmentation within a human population.

Humans possess an impressive range of skin pigmentation, both within and between populations. This diversity is highly correlated with geographical location, indicating that environmental factors as well as genetics strongly influence skin color. The predominant environmental variable affecting skin pigmentation is sunlight, and it is certain that skin pigments play an important role in both protecting DNA from the effects of UV irradiation^1–4 and influencing the availability of UV radiation for the synthesis of necessary compounds, such as vitamin D.^5–7 Epidemiological studies in humans show that skin pigmentation is a polygenic quantitative trait with high heritability,^8–10 and, with the direct correlation between skin pigmentation and incident UV exposure, it has long been postulated that it is a trait under intense selective pressure.^11–14 Because of its striking diversity and potential evolutionary importance as a highly selected trait, human skin pigmentation is of great scientific interest. Over the past 100 years, pigmentary mutants in model organisms and human pigmentation disorders have been the main source for the discovery of genes involved in skin color. More than 100 pigmentation genes have been identified in mouse alone, most with identified human orthologs, and at least 18 genes are currently listed in OMIM as being involved in human albinism.^15,16 However, until very recently, few direct studies to identify the genes responsible for the normal range of human skin pigmentation have been reported^13,17–26; in addition, all these studies have investigated the role of only known pigmentation-disorder genes in normal skin-color variation. Therefore, many questions about the loci responsible for the natural diversity of human skin pigmentation have remained unanswered: how many genes are involved, are the different gene effects additive, are the genes involved in skin pigmentation the same across different ethnic populations, and are there still-undiscovered pigmentation genes?

With the availability of the entire human genomic sequence in 2001,^27,28 the identification of millions of SNPs across the genome,^29–31 and the development of high-throughput genotyping technologies, the tools were available for investigation of the genetic components controlling human skin pigmentation with use of a high-density genomewide association study. In the present study, we applied a three-tiered methodology of quantitative pooled genotyping followed by individual genotyping of associated SNPs in original and replicate population sets, as has been successfully used elsewhere.³²

In all, 1,620,742 SNPs across the genome were measured for allele-frequency differences between DNA pools made from the 20% tails of the skin-pigmentation distribution, as objectively measured by reflectance spectroscopy in a South Asian population. The top 30,000 SNPs (∼2%) with the largest allele-frequency differences between the pools were individually genotyped in the original population sample of 737 individuals, followed by individual genotyping of associated SNPs in an independent replicate sample of 231 individuals of South Asian ancestry. Despite confounding of phenotype-genotype associations by population stratification, we found three genes with replicated genomewide significance that together account for a large fraction of the skin-pigmentation variation in the South Asian population.

Material and Methods

Sample Selection

Approval for the study was obtained from ethics review boards in the United Kingdom. After a complete description of the study was given to the subjects, written informed consent was obtained. Recruitment of volunteers of South Asian descent was conducted at >50 different sites across the United Kingdom. For each qualifying subject, either both parents or all four grandparents were born in India, Pakistan, Bangladesh, or Sri Lanka. Reported ancestry was used to group subjects by region into eastern (mainly Bangladesh), northwestern (mainly Gujarat and Punjab), southern (mainly Sri Lanka), or mixed origin. Reported ancestry was defined for each individual from the grandparental information if known and from the parental information otherwise. If the country and state of birth of all four grandparents were the same, then the reported ancestry of the volunteer was defined similarly; this was also the case if there was missing information from one grandparent but the origin of the other three grandparents was the same. Furthermore, if information from both maternal and/or paternal grandparents was missing, then maternal and/or paternal information was used in its place. Subjects with disagreements about reported ancestry were categorized as “mixed.” Both males and females aged ⩾18 years were included. Volunteers were excluded if they reported consumption of oral dietary foods or supplements or the use of topical skin ointments on the measurement areas that are designed to change skin color. Other exclusion criteria included pigmentation disorders or any current skin disease of any type anywhere on the body.

Phenotype and DNA Collection

Skin pigmentation was measured with a Minolta chromameter with use of the Commission Internationale de l’Eclairage L*a*b* color system. The L* value, which measures skin reflectance, or lightness, ranges from 0 to 100, where 0 is the darkest and 100 is the lightest skin pigmentation. Skin reflectance was measured on three relatively hairless sites per arm—the sun-exposed lower dorsal forearm, the sun-protected inner volar forearm, and the sun-protected inner arm above the elbow—giving six measurements per volunteer. The highest of the six L* values was defined as maxL*. Skin chromameter measurements for 98 randomly selected individuals of South Asian ancestry living in the United Kingdom were used to estimate the distribution of maxL* within the South Asian population (fig. 1). Individuals were selected for the first phase of our association study by targeted recruitment in the upper and lower quintiles of maxL* values determined as described above. Targeted recruitment of individuals was conducted by prescreening volunteers for skin color by visual assessment. More than 2,800 volunteers were enrolled in the study and had their skin reflectance measured, with blood samples taken from >1,100 individuals for the isolation of DNA. Genomic DNA was extracted from whole blood with use of the Qiagen DNA isolation kit, in accordance with the manufacturer’s instructions. The sample recruited for the pooled genotyping phase of the study (cohort 1) consists of 395 individuals with maxL* values ⩽56 and 383 individuals with maxL* values ⩾63, who were classified as “L” for having low reflectance and as “H” for having high reflectance, respectively. For the replication population (cohort 2), 119 L and 116 H individuals were recruited.

Figure 1. .

Histogram of maximum L* values (maxL*) in volunteers of South Asian ancestry. The maxL* values were obtained from 98 randomly selected individuals of South Asian ancestry living in the United Kingdom. The maxL* values for the 20th and 80th percentiles of the distribution are indicated by vertical lines.

Population-Structure Analysis

The genetic ancestry of the individuals in cohort 1 was examined by individually genotyping them on a set of 312 SNPs, referred to as “genomic control” (GC) SNPs, which are spaced approximately uniformly across the human autosomes.³³ Population structure within the study population was evaluated using the structure program³⁴ with models having 1–4 ancestral clusters. For each model, subsets of the L and H individuals were selected so as to construct pools matched on mean cluster membership, as described elsewhere.³² Simple χ² tests for association with skin reflectance were performed for these GC SNPs. Any residual inflation in test statistics was assessed in comparison with the experimental error of allele-frequency determination in pooled genotyping.³²

For the association analysis of individual genotypes, both population cohorts used in this study were assessed jointly by use of principal-components analysis (PCA)³⁵ of standardized genotypes on the GC SNPs. The presence of population structure increases the fraction of genotypic variance that is accounted for by the first few principal components. The genotypes and phenotypes are subsequently adjusted for these to project out the contributions of ancestry to the genotypes. To assess the significance of the eigenvalues corresponding to the first several principal components, the PCA was repeated for 100 permutations of the genotype data in which SNP genotypes were scrambled, so as to eliminate all SNP-SNP correlations.

Pooled Genotyping on Oligonucleotide Arrays

From cohort 1, 286 L individuals and 285 H individuals were used to construct two DNA pools, L and H, respectively, for the estimation of 1,620,742 SNP allele-frequency differences between the low- and high-reflectance groups, with the use of methods described elsewhere.^32,36 For pooled genotyping, each DNA pool was amplified in quadruplicate with the use of 266,851 long-range PCRs; each PCR replicate was then pooled, labeled, hybridized to 228 high-density arrays, stained, and detected as described elsewhere.³⁶

Determination of Pooled Allele-Frequency Estimates

Estimates of the pooled allele frequency, , were computed from fluorescence intensities on the arrays, as described elsewhere.³² For each SNP, we obtained eight independent measurements of , four for each DNA pool. The allele-frequency difference between the L and H pools, Δ, was determined from the average for each pool. We also used an independently derived genomewide haplotype map³⁵ to obtain a better estimate of the allele-frequency difference for some SNPs, using a method described elsewhere.³² In brief, within each haplotype block across the genome, linear regression was used to solve for frequency differences of the underlying common haplotype patterns from the measured estimates of Δ. A good quality of fit (P<.05 for the F test) was considered to be indicative of conformance between the genetic structure of the population in our study and the independently derived haplotype map. For such haplotype-conforming SNPs, the regression provided improved estimates of allele-frequency differences, called fitted Δ, by effectively averaging over redundant information within each haplotype block. These linear regressions within haplotype blocks also allowed for the elimination of some redundancy in the SNP set selected for subsequent individual genotyping.

The selection of SNPs for further assessment with individual genotyping used a ranking by the magnitude of the estimated allele-frequency difference |Δ|. For haplotype-conforming SNPs, a lower threshold was employed on the fitted |Δ|, on the basis of the fact that estimates of allele-frequency differences are better for these SNPs. This approach to selecting SNPs from pooled genotype data for subsequent individual genotyping has been explained in detail elsewhere.³²

Individual Genotyping on Oligonucleotide Arrays

High-density oligonucleotide arrays for individual genotyping were designed as described elsewhere,³⁰ with each SNP interrogated by 40 distinct 25-mer probes. DNA samples were amplified by short-range multiplex PCR and were labeled, hybridized to the arrays, stained, and detected as described elsewhere.³³ The individual genotypes were determined by clustering measurements from multiple scans in the two-dimensional space defined by reference and alternate perfect-match trimmed mean intensities, as described elsewhere.^37,38 Quality-control filters were applied to both the DNA samples and the SNPs. DNA samples with call rates <0.75 or that showed evidence of misidentification were excluded from analysis. SNPs genotyped in <0.8 of the DNA samples were excluded from further analysis. Deviations from Hardy-Weinberg equilibrium (HWE) are sometimes indicative of problems with genotype clustering; however, such deviations could also legitimately arise because of population structure or association with the trait of interest. Therefore, SNPs that did not follow HWE at a level of P<.001 were discarded in most cases, with exceptions for SNPs that were also statistically significant at a level of P⩽.001 on a simple trend test for association with skin reflectance corrected with GC.³⁹ For these potentially associated SNPs, the clustering was visually inspected, and SNPs were excluded from further analysis if problems were detected.

Single-SNP Association Tests

Association tests were performed with logistic regression, with covariates to account for sex and population structure, with the use of likelihood-ratio tests to assess the significance of association with SNP genotypes. Population structure was represented by the first several principal components of the genotype matrix for the set of individually genotyped GC SNPs (see the “Population-Structure Analysis” section). Sex is a known confounder with skin pigmentation²⁴ that was not explicitly matched for in selecting study participants; thus, it was also included as a covariate in logistic regression. The model for our primary single-SNP association test can be written as

where P(H) is the likelihood of membership in the H group, P(L) is the likelihood of membership in the L group, γ is a factor denoting sex, π₁–π₅ are the first five principal components determined as described in the “Population-Structure Correction in the Association Analysis” section, and g is the SNP genotype of the individual, coded as g=0,1,2 on the basis of the count of an arbitrarily designated allele. This linear coding of genotypes corresponds to a multiplicative model of risk. A likelihood-ratio test was used to test for association of skin reflectance with genotype, comparing the model above with a null model without the genotype term,

We tested for residual systematic inflation in the test statistics by applying this likelihood-ratio test to all samples taken together (cohorts 1 and 2) for the GC SNPs that were polymorphic in the study population and that passed our quality filters for individual genotyping. Other features of interest in the genotype data (dominance, epistasis, and independence of associations) were modeled by adding corresponding terms to the logistic-regression models.⁴⁰

Independence of Associations

Cohort 2 samples were genotyped for 408 SNPs in the region on chromosome 15 between positions 45,524,265 and 48,470,992 (National Center for Biotechnology Information [NCBI] build 36), where many SNPs showed significant associations with the phenotype in cohort 1. To assess the independence of these SNP associations, an association analysis was performed conditioning on the genotype of the strongest associated SNP in this region, rs1426654. The conditional association for each other SNP in this region was assessed with a likelihood-ratio test that compared the logistic-regression model,

with the null model,

where g represents the genotypes of the SNP being assessed and g_top represents the genotypes of rs1426654. An analogous analysis with the roles of g and g_top reversed was also performed.

Dominance and Interactions

For the three primary SNPs that were found to be strongly associated with skin reflectance, dominance effects and epistasis were investigated by adding appropriate terms to the logistic-regression model. To test for dominance effects, genotypes of these SNPs were represented by two variables, g_add and g_dom, and the genotypes AA, Aa, and aa (where A and a represent the two SNP alleles, arbitrarily designated) were coded as g_add=(-1,0,1) and g_dom=(-0.5,0.5,-0.5).⁴⁰ Thus, g_dom represents deviations from a multiplicative risk model. A logistic-regression model was built on the cohort 2 samples, and a likelihood-ratio test was used to evaluate the significance of the association with g_dom.

was compared with

To test for interactions between pairs of SNPs, we constructed indicator variables of possible combinations of genotypes at the two loci. Denoting the alleles at SNP 1 by a and A and those at SNP 2 by b and B, the indicator variables are I_aAbB, I_aABB, I_AAbB, and I_AABB, where, for example, I_aAbB=1 for an individual whose genotype is aA on SNP 1 and bB on SNP 2. These terms were added to a base model that has additive terms g_add,1, g_add,2, and g_add,3 for the three SNPs. The interaction between each pair of SNPs was then assessed using a likelihood-ratio test, comparing the model

with

Results

Determination of the Phenotypic Groups

The population of South Asia, which, in this study, includes India, Pakistan, Bangladesh, and Sri Lanka, has a wide natural variation in skin pigmentation and consists of the most genetically diverse population outside of Africa,⁴¹ which makes it well suited for the investigation of the complex genetic trait of human skin pigmentation. Human skin color is largely determined by two biological components, melanin and hemoglobin, and one major environmental component, exposure to UV light.⁴² Melanin is the major determining factor in intrinsic skin pigmentation⁴³; therefore, we were most interested in identifying genes that affect melanin. The melanin contribution to skin pigmentation can be quantitatively measured and distinguished from redness caused by hemoglobin and/or inflammation by reflectance spectrophotometry with use of the L* value from the L*a*b* color system.⁴⁴ Therefore, the highest L* value (maxL*) from hairless, sun-protected sites on the arm was used to define intrinsic skin reflectance in this study, thereby controlling for confounding effects due to sun exposure and hair. To empirically determine the top 20% and bottom 20% of the skin-reflectance distribution in the South Asian population, 98 randomly selected individuals of South Asian ancestry living in the United Kingdom were recruited and had skin chromameter measurements taken. From the plot of the distribution of the maxL* values in this set (fig. 1), individuals with maxL* values ⩽56 were classified as belonging to the low-reflectance quintile (L) with the darkest skin pigmentation, and those with maxL* values ⩾63 were classified as belonging to the high-reflectance (H) quintile with the lightest skin pigmentation of the distribution. Targeted recruitment of individuals in these tails resulted in the collection of DNA from 395 and 383 subjects with low and high skin reflectance, respectively, for the pooled genotyping phase of the study (cohort 1).

Population Structure and Construction of Balanced Pools

The genetic diversity within South Asian populations potentially introduces significant confounding in an association study because of population structure.⁴⁵ To control for spurious associations due to systematic differences in ancestry between the L and H groups, we composed the two DNA pools to be used in the pooled genotyping with individuals who were matched for ancestry. The genetic ancestry of the 395 L and 383 H individuals was investigated by individually genotyping them for a set of 312 unlinked GC SNPs. Of the 312 GC SNPs, 294 yielded high-quality genotype data and were analyzed for population structure by use of the structure program,³⁴ with the use of models with 1–4 clusters. The results yielded good evidence of more than one distinct genetic population cluster in our set of 778 individuals but yielded little support for more than two population clusters. The inferred cluster-membership values for a two-cluster model correlate with reported geographical ancestry. Individuals of southern (mainly Sri Lankan) and eastern (mainly Bangladeshi) origin have a similar distribution of admixture, which is distinct from individuals of northwestern (mainly Punjabi) origin (fig. 2).

Figure 2. .

Histogram of admixture proportions in cohort 1 for the geographical ancestry groups. Admixture proportions were determined for 778 individuals in cohort 1 with use of the structure program for a two-cluster model, with genotypes from 294 GC SNPs. Reported parental and grandparental ancestry was used to group individuals by region into (A) eastern (mainly Bangladesh), (B) southern (mainly Sri Lanka), (C) northwestern (mainly Gujarat and Punjab), or (D) mixed origin.

Using the same genotype data set of 294 GC SNPs and a method described elsewhere,^32,33 we found significant population stratification between the tails of the skin-reflectance distribution. This result was not unexpected, since there was a clear bias in the recruitment of individuals for cohort 1 from the three geographical regions with respect to the reflectance tails (table 1). Under the null (one-cluster) model of no population structure, for which the L and H groups are assumed to be well matched, we observed a large excess of small P values in χ² tests for association with skin color (table 2), and a global test for population stratification based on the sum of χ² statistics⁴⁶ was highly statistically significant (P<10^-40). Using the inferred ancestry values obtained from the structure program with the two-cluster model, we composed the two phenotypic pools, using a subset of individuals such that the average proportions of genetic ancestry were similar between the groups and as many individuals as possible were retained. This matching required us to exclude ∼25% of the original individuals but reduced the population stratification considerably, as can be measured by a reduction in the sum of χ² statistics versus the unmatched groups (table 2). Although the residual stratification in the matched groups is statistically significant (P=3.9×10^-5), the effect of a sum statistic of 400 is comparable to additional experimental error of ∼1% in the pooled allele-frequency measurements. This error size is smaller than the actual level of experimental error. On the basis of these results, we constructed the DNA pools from 286 L and 285 H individuals from cohort 1. Recruitment of individuals for the replication population, cohort 2, was designed to reduce the population stratification bias with respect to skin reflectance, with the use of reported geographical ancestry as a marker for the genetic ancestry profiles we detected in cohort 1 (fig. 2).

Table 1. .

Cohort Distribution for Regional Ancestry and Skin Reflectance^[Note]

	No. of Individualsin Reflectance Group
Cohort and Region	L	H
1:
Northwestern	75	247
Eastern	235	100
Southern	75	2
Mixed	10	34
2:
Northwestern	61	62
Eastern	58	53
Southern	0	0
Mixed	0	1
Total	514	499

Note.— Self-reported ancestry was used to group subjects by region into eastern (Bangladesh), northwestern (Gujarat and Punjab), southern (Sri Lanka), or mixed origin.

Table 2. .

Measures of Population Stratification for Skin-Reflectance Pools of Different Compositions

		No. of SNPs with Small P Values in χ2 Tests for Associationb
Pooling Strategy	L/Ha	P<.0001	P<.001	P<.01	P<0.1	χ2 Sum	Overall P
Expectedc	395/383	0	0	3	29	294	…
All of cohort 1 included	395/383	5	9	24	86	740	2.9×10−40
Matched by inferred ancestryd	286/285	1	2	8	43	400	3.9×10−5

^aNumber of individuals included in each of the skin-reflectance phenotypic pools.

^bχ² tests for association were based on 294 GC SNPs.

^cThe expected number of SNPs with low P values, based on the null hypothesis of no population structure in cohort 1.

^dInferred ancestry values were obtained from the structure program, with the two-cluster model.

Pooled Genotyping Results

We attempted to estimate the allele-frequency difference, Δ, between the L and H pools for 1,620,742 SNPs across the human genome. Individual measurements of were excluded when we could identify specific experimental errors—saturated signal intensities, inconsistent hybridization patterns, or low signal-to-background ratios. We excluded SNPs for which more than one measurement per pool failed these quality-control criteria. SNPs for which the SE of Δ was >0.035 were also excluded, so that the set of SNPs having the largest absolute Δ were not dominated by a subset of measurements with very high experimental variance. “Fitted Δ” values, where information from Perlegen’s haplotype map is used to improve pooled allele-frequency estimates, were generated for a fraction of the SNPs, called “haplotype-conforming” SNPs. These fitted Δ values provide more-accurate estimates than do the measured Δ values of individual SNPs, by exploiting correlations among SNPs within haplotype blocks. After all the data-quality filters were applied, we had high-quality Δ or fitted Δ estimates for 1,502,205 SNPs.

Individual Genotyping for Cohort 1

SNPs were ranked on the basis of absolute Δ or fitted Δ, from largest to smallest, and 30,000 were chosen for individual genotyping on the basis of the capacity of a high-density oligonucleotide array. Selected SNPs from the pooled genotyping results had a fitted or a . A lower threshold was used for haplotype-conforming SNPs, because their consistency with the haplotype map provides additional evidence of allele-frequency differences at those positions. An additional set of 66 candidate SNPs was selected to provide denser coverage of variants in genes known to be involved in the pigmentation process (table 3), and 312 GC SNPs used to test for population structure were also included in the array. Short-range PCR assays were successfully designed for 30,045 SNPs, 98.9% of those selected for all categories, and were genotyped with the 778 individuals of cohort 1. After application of quality filters to the genotypes and the DNA samples, a total of 25,928 SNPs for 737 individuals, 363 L and 374 H, were used for the association analysis, including 292 polymorphic GC SNPs. These included individuals who had been excluded from the balanced pools constructed for cohort 1; for, unlike pooled genotyping, with individual genotypes we can explicitly model population structure in the association analysis, and, therefore, the inclusion of all individuals maximizes our power to detect associations.

Table 3. .

Candidate SNPs

					Allele
SNPa	Chrb	Positionc	Gened	Functione	1	2
rs6058017	20	32320659	ASIP	3′ UTR	A	G
rs11551042	13	93887736	DCT	Down	A	T
rs1540979	13	93888693	DCT	Down	T	A
rs9516414	13	93893332	DCT	Int	A	G
rs9524491	13	93893392	DCT	Int	T	A
rs2892680	13	93893530	DCT	Int	A	G
rs1407995	13	93894014	DCT	Int	T	C
rs2296498	13	93894112	DCT	Int	A	G
rs12876569	13	93916607	DCT	Int	C	G
rs3212379	16	88512632	MC1R	5′ UTR	C	T
rs3212359	16	88512678	MC1R	5′ UTR	C	T
rs3212360	16	88512718	MC1R	5′ UTR	C	T
rs3212361	16	88512723	MC1R	5′ UTR	G	A
rs3212362	16	88512845	MC1R	5′ UTR	G	A
rs3212363	16	88512942	MC1R	5′ UTR	A	T
rs1805005	16	88513345	MC1R	Nonsyn	G	T
rs1805006	16	88513419	MC1R	Nonsyn	C	A
rs2228479	16	88513441	MC1R	Nonsyn	G	A
rs2229617	16	88513477	MC1R	Nonsyn	G	A
rs3212364	16	88513485	MC1R	Syn	G	A
rs1805007	16	88513618	MC1R	Nonsyn	C	T
rs1110400	16	88513631	MC1R	Nonsyn	T	C
rs3212365	16	88513633	MC1R	Nonsyn	G	C
rs1805008	16	88513645	MC1R	Nonsyn	C	T
rs885479	16	88513655	MC1R	Nonsyn	G	A
rs3212366	16	88513753	MC1R	Nonsyn	T	C
rs1805009	16	88514047	MC1R	Nonsyn	G	C
rs3212367	16	88514067	MC1R	Syn	C	T
rs2228478	16	88514109	MC1R	Syn	A	G
rs3212368	16	88514133	MC1R	3′ UTR	G	A
rs3212369	16	88514261	MC1R	3′ UTR	A	G
rs3212370	16	88514278	MC1R	3′ UTR	C	A
rs3212371	16	88514702	MC1R	3′ UTR	A	G
rs12592271	15	25763517	OCA2	Int	G	A
rs12592307	15	25763768	OCA2	Syn	G	A
rs17566952	15	25870480	OCA2	Int	G	C
rs7170989	15	25874003	OCA2	Int	T	C
rs11638265	15	25876168	OCA2	Int	C	T
rs12439067	15	25876220	OCA2	Int	G	T
ss69374775	15	25876236	OCA2	Int	G	A
rs1800411	15	25885516	OCA2	Syn	C	T
rs12910433	15	25902239	OCA2	Int	C	T
rs1037208	15	25904952	OCA2	Int	C	A
rs2290100	15	25946945	OCA2	Int	A	G
rs1052165	12	54637613	SILV	Syn	C	T
ss69374774	5	33980486	SLC45A2	Down	A	C
rs250416	5	33983301	SLC45A2	Int	G	T
rs2278007	5	33987308	SLC45A2	Int	A	G
rs2287949	5	33990268	SLC45A2	Syn	T	C
rs26722	5	33999627	SLC45A2	Nonsyn	C	T
rs183671	5	33999967	SLC45A2	Int	A	C
rs4547091	11	88550469	TYR	Up	C	T
rs1799989	11	88550571	TYR	Up	C	A
rs1042602	11	88551344	TYR	Nonsyn	C	A
rs12804012	11	88600438	TYR	Int	G	A
rs3793975	11	88600836	TYR	Int	C	T
rs1827430	11	88658088	TYR	Int	A	G
rs28521275	11	88668617	TYR	Down	G	A
rs2075508	9	12688363	TYRP1	Int	T	C
rs2762462	9	12689776	TYRP1	Int	T	C
rs2733832	9	12694725	TYRP1	Int	C	T
rs2733833	9	12695095	TYRP1	Int	T	G
rs2733834	9	12698910	TYRP1	Int	C	G
rs2762464	9	12699586	TYRP1	3′ UTR	A	T
rs2382360	9	12700413	TYRP1	Int	T	C
ss69374773	9	12700473	TYRP1	Down	T	G

^ars or ss identifier in dbSNP.

^bChr = chromosome.

^cChromosome base position on NCBI build 36.2.

^dGene symbol from Entrez Gene.

^eInt = intronic; nonsyn = nonsynonymous; syn = synonymous; up = within 10 kb of the transcriptional start site; down = within 10 kb of the transcriptional stop site.

Population-Structure Correction in the Association Analysis

Since we observed significant population stratification between the tails of the skin-reflectance distribution in this South Asian population, we assessed and adjusted for population structure in the association analysis, using the genotypes of our GC panel of 312 SNPs. Population structure was modeled jointly for both population cohorts used in this study, with use of a PCA of the standardized genotypes on the 292 polymorphic GC SNPs that passed our quality filters. The significance of the eigenvalues corresponding to the first several principal components was assessed by repeating the PCA on 100 scrambled realizations of the SNP genotype data. The largest three eigenvalues for the true data set were larger than the largest eigenvalue in any of the permuted data sets (P<.01); the largest five eigenvalues in the true data set were observed in fewer than half the permuted data sets. In addition, four of the top five principal components show significant association with reported ancestry regions in an analysis of variance (PC1, P<1.0×10^-16; PC2, P=.19; PC3, P=2.9×10^-3; PC4, P=2.1×10^-3; PC5, P=5.9×10^-8), indicating that these axes of variation likely reflect true features of population structure. We also investigated the relationship between skin reflectance (classified as “L” or “H”) and the top 10 principal components, using logistic regression with sex as a covariate. The first and fifth components showed significant associations (P=4.6×10^-22 and P=1.7×10^-3, respectively). On the basis of these findings, we adjusted all association analyses for the first five principal components in addition to sex. After this adjustment, we tested for evidence of remaining population structure bias in the association analysis, using the 292 polymorphic GC SNPs that passed quality filters. One SNP (rs8041414) was located in a region on chromosome 15 showing the strongest association with skin reflectance. The results for the remaining 291 SNPs did not reveal a significant inflation in single-SNP test statistics for association with skin reflectance (mean deviance 1.0232; P=.37). Thus, no further corrections for population structure were deemed necessary in association tests for other SNPs in cohort 1 or cohort 2.

Association Results for Cohort 1

Single-SNP tests for association with skin reflectance were performed with logistic regression, including sex and five principal components of the GC SNP genotypes as covariates. Likelihood-ratio tests were used to assess the significance of associations with SNP genotypes. In total, we tested 1,502,205 SNPs for association with skin pigmentation. Although there is likely to be some linkage disequilibrium among the SNPs within this population, it could not be evaluated from our pooled genotyping allele-frequency estimates. Therefore, we employed a conservative Bonferroni threshold for significance of association, α=0.05/1,502,205=3.3×10^-8, and found 42 SNPs associated at this level of significance across the genome (table 4). Surprisingly, 39 of the 42 associated SNPs are all located within a single 2.4-Mb chromosomal region of 15q21.1-21.2, between chromosomal positions 45,774,265 and 48,220,992. The strongest SNP association for the entire genome is in this region, with an allele-frequency difference of 45% between the H and L reflectance groups, yielding a P value of 1.0×10⁻⁵⁰. This SNP, rs1834640, is located 21 kb from the closest gene, SLC24A5. Of the three SNPs with genomewide significance located outside the chromosomal 15q region, two are located within the TYR gene on chromosome 11; one is in an intron (rs12295166), and the other is a nonsynonymous polymorphism (rs1042602 [p.S192Y]) and is the only SNP in this set of 42 associations that was selected for individual genotyping as a candidate SNP (table 4). The last associated SNP, rs16891982, is also a nonsynonymous polymorphism (p.L374F) in the gene SLC45A2 on chromosome 5.

Table 4. .

Associated SNPs Showing Genomewide Significance in Cohort 1

								Frequencyf
SNPa	Categoryb	Chromosome	Positionc	Gene(s)d	Functione	Allele 1	Allele 2	L	H	P	OR (95% CI)g
rs1834640	PG	15	46179457	…	…	A	G	.47	.92	1.01×10−50	.08 (.05–.12)
rs12913316	PG	15	46275146	…	…	C	T	.33	.05	4.95×10−32	8.9 (5.68–13.97)
rs11070627	PG	15	46258816	MYEF2	Up	A	T	.33	.05	9.52×10−32	8.77 (5.6–13.75)
rs2924566	PG	15	46056053	…	…	G	A	.55	.25	4.17×10−23	3.86 (2.88–5.18)
rs4775730	PG	15	46087470	…	…	T	C	.52	.78	3.56×10−21	.26 (.2–.36)
rs11637235	PG	15	46420445	DUT	Int	C	T	.66	.35	6.09×10−21	3.44 (2.6–4.54)
rs9788730	PG	15	46098702	…	…	A	C	.65	.89	3.87×10−19	.23 (.17–.33)
rs10519170	PG	15	46473467	…	…	A	G	.57	.81	1.16×10−15	.3 (.22–.41)
rs2965317	PG	15	46049012	…	…	C	T	.33	.13	2.44×10−15	3.65 (2.58–5.17)
rs2965318	PG	15	46051787	…	…	T	G	.33	.13	4.43×10−15	3.4 (2.44–4.73)
rs16960541	PG	15	46157395	…	…	G	T	.83	.96	1.70×10−14	.16 (.1–.27)
rs1820489	PG	15	46472393	…	…	C	T	.41	.19	8.09×10−14	2.98 (2.2–4.05)
rs7164700	PG	15	46097633	…	…	G	A	.8	.95	1.04×10−13	.22 (.14–.34)
rs16960682	PG	15	46306954	SLC12A1	Int	G	C	.84	.97	1.14×10−13	.16 (.09–.27)
rs2413890	PG	15	46313654	SLC12A1	Int	G	T	.45	.23	1.07×10−11	2.55 (1.92–3.38)
rs16891982	PG	5	33987450	SLC45A2	Nonsyn	C	G	.97	.83	3.21×10−11	4.86 (2.88–8.21)
rs16960451	PG	15	46069778	…	…	C	T	.81	.95	4.96×10−11	.28 (.18–.42)
rs2924567	PG	15	46055778	…	…	G	T	.82	.67	1.74×10−10	2.54 (1.89–3.42)
ss69356377	PG	15	46157464	…	…	A	G	.89	.98	3.13×10−10	.13 (.06–.27)
rs4775727	PG	15	46067503	…	…	A	T	.84	.95	3.91×10−10	.25 (.16–.4)
rs2924572	PG	15	46039330	…	…	T	A	.22	.08	3.98×10−10	3.23 (2.18–4.77)
rs1042602	CS	11	88551344	TYR	Nonsyn	C	A	.96	.84	4.48×10−10	4.36 (2.64–7.2)
rs7170781	PG	15	48180287	ATP8B4	Int	C	G	.54	.73	8.85×10−10	.45 (.35–.59)
rs1869454	PG	15	46080741	…	…	C	T	.84	.95	1.33×10−9	.26 (.17–.42)
rs16960450	PG	15	46069520	…	…	C	T	.83	.94	1.49×10−9	.27 (.17–.42)
rs4774527	PG	15	46971973	SHC4	Int	G	A	.8	.63	1.73×10−9	2.34 (1.76–3.12)
rs16960434	PG	15	46062007	…	…	G	C	.83	.94	1.77×10−9	.27 (.17–.43)
rs1531916	PG	15	46313563	SLC12A1	Int	G	A	.35	.18	2.69×10−9	2.4 (1.78–3.25)
rs504376	PG	15	45957669	…	…	C	G	.59	.4	2.96×10−9	2.21 (1.68–2.89)
rs16960453	PG	15	46070428	…	…	C	G	.85	.95	3.07×10−9	.27 (.17–.43)
rs494230	PG	15	45903689	…	…	T	C	.42	.61	3.60×10−9	.47 (.36–.61)
rs12295166	PG	11	88615805	TYR	Int	T	C	.95	.77	4.81×10−9	2.84 (1.93–4.17)
rs1843144	PG	15	46311814	SLC12A1	Int	G	C	.34	.18	5.85×10−9	2.35 (1.74–3.17)
rs4774557	PG	15	48220992	…	…	T	C	.55	.73	6.47×10−9	.47 (.36–.61)
rs1912640	PG	15	45774265	…	…	T	C	.27	.11	7.53×10−9	2.57 (1.84–3.61)
rs4775728	PG	15	46078857	…	…	C	G	.87	.95	9.04×10−9	.24 (.15–.41)
ss69356376	PG	15	45998012	…	…	G	A	.91	.98	1.12×10−8	.19 (.1–.35)
rs1872304	PG	15	46446080	…	…	A	G	.45	.28	1.95×10−8	2.16 (1.64–2.86)
rs11854994	PG	15	46986684	SHC4, DUT	Int	G	A	.69	.51	2.32×10−8	2.07 (1.59–2.69)
rs784416	PG	15	46800217	…	…	G	C	.22	.1	2.57×10−8	2.68 (1.86–3.86)
rs16961610	PG	15	46847375	CEP152	Int	G	T	.79	.91	2.74×10−8	.36 (.25–.53)
rs16960843	PG	15	46454117	…	…	A	G	.85	.97	3.28×10−8	.27 (.17–.45)

^ars or ss identifier in dbSNP.

^bPG = SNP selection based on pooled genotyping results. CS = candidate SNP.

^cChromosome base position in NCBI build 36.2.

^dGene symbol from Entrez Gene.

^eInt = intronic; nonsyn = nonsynonymous; up = within 10 kb of the transcriptional start site.

^fFrequency of allele 1 in the reflectance groups of cohort 1.

^gOdds ratios (ORs) and 95% CIs for allele 1 are shown. The OR is the ratio of the likelihood of an individual being in the low-reflectance group to the likelihood of being in the high-reflectance group.

Replication of Associations in Cohort 2

All of the 42 genomewide-significant SNPs in cohort 1 were individually genotyped with an independent replicate population of 235 individuals (cohort 2). After quality filters were applied, genotypes of the 42 SNPs in 115 H and 116 L individuals were used in the association analysis, where single-SNP likelihood-ratio tests were performed in the same manner as were those for cohort 1. A majority of the associated SNPs, 32 (76%) of 42, yielded P values with a significance threshold of <.05 (table 5). These included the nonsynonymous polymorphisms in TYR and SLC45A2, as well as 30 SNPs in the 2.4-Mb region on chromosome 15. As was seen in cohort 1, the most significant association among these SNPs in cohort 2 was for rs1834640 on 15q21.1, with P=3.28×10^-15 and a 36% allele-frequency difference between the high- and low-reflectance groups. Treating the two cohorts as a single population of 968 individuals in the association analysis, instead of independently, yielded different statistical conclusions for only four SNPs, the intronic TYR SNP rs12295166 and three chromosome 15 SNPs, which show genomewide significance in the joint analysis but failed to reach statistical significance by the previous analysis (table 5).

Table 5. .

Results for Associated SNPs in Cohort 2

					Frequencye
SNPa	Chromosome	Positionb	Gene(s)c	Functiond	L	H	P	OR (95% CI)f	Joint Cohort Pg
rs1834640	15	46179457	…	…	.51	.87	3.28×10−15	.01 (.00–.04)	3.39×10−64
rs11070627	15	46258816	MYEF2	Up	.31	.06	1.49×10−10	54.67 (13.08–228.52)	6.63×10−41
rs12913316	15	46275146	…	…	.31	.06	2.30×10−10	46.44 (11.57–186.35)	3.18×10−41
rs16960682	15	46306954	SLC12A1	Int	.85	.97	2.32×10−6	.02 (.00–.12)	1.46×10−19
rs4775730	15	46087470	…	…	.52	.77	4.40×10−6	.14 (.06–.34)	2.10×10−25
rs2924566	15	46056053	…	…	.51	.25	6.25×10−6	6.84 (2.84–16.47)	7.65×10−27
rs9788730	15	46098702	…	…	.68	.88	7.30×10−6	.09 (.03–.27)	6.84×10−24
rs11854994	15	46986684	SHC4, DUT	Int	.72	.48	1.36×10−5	6.41 (2.67–15.39)	2.65×10−13
rs4774527	15	46971973	SHC4	Int	.84	.63	1.76×10−5	8.11 (2.96–22.2)	4.09×10−14
rs7164700	15	46097633	…	…	.8	.94	1.81×10−5	.05 (.01–.23)	3.07×10−18
rs504376	15	45957669	…	…	.58	.38	5.90×10−5	5.77 (2.36–14.09)	3.68×10−12
rs16960541	15	46157395	…	…	.87	.97	4.21×10−4	.05 (.01–.32)	1.93×10−16
rs2924567	15	46055778	…	…	.8	.64	9.62×10−4	4.7 (1.83–12.05)	1.63×10−12
rs11637235	15	46420445	DUT	Int	.59	.42	1.00×10−3	4.17 (1.72–10.06)	2.53×10−22
rs2965318	15	46051787	…	…	.31	.16	2.10×10−3	4.91 (1.73–13.9)	1.65×10−16
ss69356377	15	46157464	…	…	.9	.97	2.35×10−3	.06 (.01–.45)	1.73×10−13
rs2965317	15	46049012	…	…	.3	.15	3.50×10−3	4.79 (1.63–14.09)	1.97×10−16
rs16891982	5	33987450	SLC45A2	Nonsyn	.94	.85	4.56×10−3	7.37 (1.76–30.96)	5.02×10−13
rs784416	15	46800217	…	…	.21	.11	4.83×10−3	4.66 (1.56–13.93)	5.05×10−11
rs10519170	15	46473467	…	…	.64	.76	6.84×10−3	.28 (.11–.72)	1.03×10−16
rs1820489	15	46472393	…	…	.35	.24	6.98×10−3	3.59 (1.39–9.29)	3.71×10−15
rs2924572	15	46039330	…	…	.19	.09	1.09×10−2	5.02 (1.41–17.91)	1.85×10−11
rs1869454	15	46080741	…	…	.88	.95	1.22×10−2	.14 (.03–.68)	6.05×10−11
rs16961610	15	46847375	CEP152	Int	.81	.89	1.32×10−2	.25 (.08–.76)	2.80×10−10
rs16960450	15	46069520	…	…	.87	.95	1.98×10−2	.16 (.03–.79)	1.00×10−10
rs1042602	11	88551344	TYR	Nonsyn	.94	.87	2.05×10−2	5.05 (1.23–20.74)	6.54×10−11
rs16960453	15	46070428	…	…	.88	.95	2.38×10−2	.16 (.03–.82)	2.86×10−10
rs4775727	15	46067503	…	…	.87	.94	2.59×10−2	.18 (.04–.84)	5.38×10−11
rs16960434	15	46062007	…	…	.87	.94	2.88×10−2	.18 (.04–.87)	2.78×10−10
rs4775728	15	46078857	…	…	.88	.95	2.98×10−2	.17 (.03–.88)	1.23×10−9
rs16960451	15	46069778	…	…	.86	.94	3.04×10−2	.24 (.06–.92)	9.01×10−12
ss69356376	15	45998012	…	…	.91	.96	4.65×10−2	.17 (.03–1.01)	5.93×10−9
rs12295166	11	88615805	TYR	Int	.89	.81	8.26×10−2	2.25 (.89–5.68)	1.94×10−9
rs494230	15	45903689	…	…	.49	.6	8.44×10−2	.49 (.22–1.11)	6.24×10−10
rs2413890	15	46313654	SLC12A1	Int	.4	.31	1.85×10−1	1.77 (.76–4.14)	6.55×10−11
rs1912640	15	45774265	…	…	.17	.19	3.47×10−1	.58 (.18–1.83)	4.64×10−6
rs1843144	15	46311814	SLC12A1	Int	.3	.24	4.23×10−1	1.44 (.59–3.49)	1.26×10−7
rs4774557	15	48220992	…	…	.61	.62	4.79×10−1	.74 (.33–1.69)	6.20×10−8
rs7170781	15	48180287	ATP8B4	Int	.61	.63	5.04×10−1	.76 (.34–1.71)	1.12×10−8
rs1872304	15	46446080	…	…	.38	.36	5.60×10−1	1.28 (.55–2.99)	3.89×10−7
rs1531916	15	46313563	SLC12A1	Int	.29	.24	6.71×10−1	1.22 (.49–3.02)	1.67×10−7
rs16960843	15	46454117	…	…	.9	.91	9.46×10−1	.95 (.24–3.79)	1.55×10−6

^ars or ss identifier in dbSNP.

^bChromosome base position on NCBI build 36.2.

^cGene symbol from Entrez Gene.

^dInt = intronic; nonsyn = nonsynonymous; up = within 10 kb of the transcriptional start site.

^eFrequency of allele 1 in the reflectance groups of cohort 2.

^fOdds ratios (ORs) and 95% CIs for allele 1 are shown. The OR is the ratio of the likelihood of an individual being in the low-reflectance group to the likelihood of being in the high-reflectance group.

^gP value for the combined analysis of both cohorts.

Independent Associations on Chromosome 15 in the Vicinity of SLC24A5

To determine whether the large number of SNP associations observed within the 15q21.1-21.2 region were due to the presence of multiple independently associated SNPs or to one association having a large effect, we individually genotyped a dense set of SNPs within this region for cohort 2. The range for SNP selection was extended by 0.25 Mb in both directions from the boundaries established by significant cohort 1 associations in the region—that is, from positions 45,524,265 to 48,470,992 on chromosome 15. A total of 408 SNPs were successfully assayed, and association test results for this set of SNPs revealed several other significant associations across the region (table 6 and fig. 3). The most significant association in this region was observed for rs1426654, a nonsynonymous polymorphism (p.A111T) in SLC24A5, with P=1.06×10^-18 and a 39% allele-frequency difference between the high- and low-reflectance groups. This SNP is in strong linkage disequilibrium with the most significant SNP from the genomewide scan, rs1834640 (R²=0.93), indicating that the two probably represent a single associated locus.

Table 6. .

Association Test Results for a 2.4-Mb Region on Chromosome 15 for Cohort 2

				Frequencye
SNPa	Positionb	Genec	Functiond	L	H	P	Conditional Pf
rs1426654	46213776	SLC24A5	Nonsyn	.51	.90	1.06×10−18	…
rs12440932	47985162	ATP8B4	Int	.96	.90	3.01×10−3	2.20×10−4
rs11854994	46986684	SHC4	Int	.72	.48	1.36×10−5	8.65×10−4
rs4774527	46971973	SHC4	Int	.84	.63	1.76×10−5	3.49×10−3
rs17465874	46866258	CEP152	Int	.90	.95	1.42×10−2	4.77×10−3
rs17467239	46930524	SHC4	Int	.75	.51	2.39×10−5	5.57×10−3
rs8041414	46843962	CEP152	Int	.56	.77	4.15×10−6	8.92×10−3
rs784411	46827089	CEP152	Int	.56	.76	1.07×10−5	9.05×10−3
rs7176696	46861195	CEP152	Int	.57	.78	1.48×10−5	9.30×10−3
rs1912640	45774265	…	…	.17	.19	3.47×10−1	9.61×10−3
rs4143620	48055351	ATP8B4	Int	.95	.89	1.46×10−2	1.07×10−2
rs16961455	46799115	…	…	.88	.94	6.90×10−3	1.31×10−2
rs2289179	46861831	CEP152	Int	.78	.90	3.13×10−3	1.49×10−2
rs17463995	46791064	…	…	.61	.40	6.99×10−6	1.54×10−2
rs1797225	45556171	…	…	.14	.10	1.81×10−1	1.84×10−2
rs8023809	47984948	ATP8B4	Int	.38	.24	3.30×10−3	2.32×10−2
rs8039142	48059752	ATP8B4	Int	.47	.63	9.04×10−4	2.43×10−2
rs7182710	46892226	CEP152	Up	.74	.55	1.04×10−4	2.51×10−2
rs12914304	46800835	…	…	.66	.46	5.85×10−4	3.23×10−2
rs4775777	46918903	SHC4	Int	.73	.56	4.75×10−4	3.23×10−2
rs16960682	46306954	SLC12A1	Int	.85	.97	2.32×10−6	3.27×10−2
rs2304546	46890536	CEP152	Up	.70	.84	1.87×10−4	3.63×10−2
rs937171	47982041	ATP8B4	Int	.48	.34	2.00×10−3	4.15×10−2
rs854151	48330011	HDC	Int	.80	.72	7.05×10−2	4.21×10−2
rs10519188	46842454	CEP152	Int	.83	.90	1.82×10−2	4.22×10−2
rs16961645	46862521	CEP152	Int	.83	.90	1.82×10−2	4.22×10−2
rs934741	46994194	SHC4	Int	.23	.19	9.30×10−2	4.97×10−2
rs2413980	48008650	ATP8B4	Int	.84	.78	2.26×10−1	5.03×10−2
rs10519251	48056965	ATP8B4	Int	.91	.84	7.89×10−3	5.24×10−2
rs7175415	46777881	…	…	.72	.81	5.29×10−3	5.25×10−2
rs16961470	46804316	…	…	.83	.90	2.63×10−2	5.30×10−2
rs2413890	46313654	SLC12A1	Int	.40	.31	1.85×10−1	5.49×10−2
rs421853	45531623	…	…	.23	.20	2.58×10−1	6.00×10−2
rs7174374	46778158	…	…	.34	.19	1.12×10−3	6.37×10−2
rs12906304	48359690	GABPB2	Int	.86	.84	2.81×10−1	6.58×10−2
rs16962583	47600407	C15orf33	Int	.92	.89	1.09×10−1	7.23×10−2
rs627566	47839646	…	…	.66	.61	2.58×10−2	7.28×10−2
rs1036477	46702218	FBN1	Int	.27	.09	1.96×10−5	7.34×10−2
rs1224660	45788054	SEMA6D	Up	.20	.13	3.32×10−1	8.33×10−2
rs1496917	45529635	…	…	.79	.69	1.12×10−2	8.49×10−2
rs11070739	48043671	ATP8B4	Int	.61	.73	5.93×10−3	8.68×10−2
rs785016	45922856	…	…	.35	.53	1.10×10−4	8.81×10−2
rs12923	48357328	GABPB2	3′ UTR	.86	.85	3.63×10−1	8.82×10−2
rs17393761	47164183	…	…	.78	.83	1.13×10−1	9.28×10−2
rs16960451	46069778	…	…	.86	.94	3.04×10−2	9.29×10−2
rs5020564	46451845	…	…	.78	.72	5.21×10−1	9.39×10−2
rs1656602	45581286	…	…	.23	.18	2.26×10−1	9.49×10−2
rs7164451	47001348	SHC4	Int	.24	.36	1.57×10−2	9.93×10−2
rs1484556	47829263	…	…	.83	.83	6.16×10−1	1.05×10−1
rs8034382	48059377	ATP8B4	Int	.75	.66	1.05×10−1	1.06×10−1
rs16960434	46062007	…	…	.87	.94	2.88×10−2	1.14×10−1
rs9806310	45650255	…	…	.95	.94	4.89×10−1	1.15×10−1
rs2452524	48013605	ATP8B4	Nonsyn	.47	.36	2.98×10−2	1.16×10−1
rs1035704	45615618	…	…	.22	.17	3.23×10−1	1.19×10−1
rs12594966	46107263	…	…	.90	.97	1.50×10−3	1.19×10−1
rs8031403	48062303	ATP8B4	Int	.42	.49	1.08×10−1	1.21×10−1
rs7164700	46097633	…	…	.80	.94	1.81×10−5	1.28×10−1
rs9788730	46098702	…	…	.68	.88	7.30×10−6	1.30×10−1
rs4775727	46067503	…	…	.87	.94	2.59×10−2	1.35×10−1
rs7174453	46804645	…	…	.94	.99	4.67×10−2	1.35×10−1
rs9806753	46953709	EID1	Up	.37	.27	1.40×10−2	1.39×10−1
rs12914876	47225707	COPS2	Int	.91	.94	4.11×10−1	1.40×10−1
rs10519132	45661803	…	…	.78	.66	3.22×10−2	1.41×10−1
rs10519193	46950245	EID1	Up	.40	.30	1.89×10−2	1.44×10−1
rs7177445	46489749	FBN1	Exon	.08	.15	7.62×10−2	1.47×10−1
rs1426200	46885308	CEP152	Int	.08	.06	4.07×10−1	1.59×10−1
rs16960453	46070428	…	…	.88	.95	2.38×10−2	1.59×10−1
rs4775728	46078857	…	…	.88	.95	2.98×10−2	1.60×10−1
rs11857760	45778132	…	…	.53	.45	1.00×10−1	1.62×10−1
rs16960450	46069520	…	…	.87	.95	1.98×10−2	1.63×10−1
rs10519145	45885244	…	…	.80	.74	3.65×10−1	1.67×10−1
rs16961388	46757556	…	…	.91	.96	3.98×10−2	1.70×10−1
rs1869454	46080741	…	…	.88	.95	1.22×10−2	1.71×10−1
rs1453854	46016892	…	…	.69	.64	3.53×10−1	1.73×10−1
rs16963635	48407932	GABPB2	Int	.90	.97	1.14×10−2	1.83×10−1
rs12439630	45822483	SEMA6D	Int	.72	.62	9.47×10−2	1.87×10−1
ss69356377	46157464	…	…	.90	.97	2.35×10−3	1.91×10−1
rs1531916	46313563	SLC12A1	Int	.29	.24	6.71×10−1	1.95×10−1
rs1390870	45541604	…	…	.78	.69	4.48×10−2	1.98×10−1
rs1224662	45789839	SEMA6D	Up	.23	.14	1.06×10−1	1.99×10−1
rs16962923	47882435	…	…	.83	.86	1.21×10−1	2.00×10−1
rs2413914	46891800	CEP152	Up	.08	.07	6.93×10−1	2.07×10−1
rs16963151	48066954	ATP8B4	Nonsyn	.80	.75	6.39×10−1	2.09×10−1
rs504376	45957669	…	…	.58	.38	5.90×10−5	2.11×10−1
rs16959669	45612375	…	…	.95	.95	7.16×10−1	2.12×10−1
rs16960071	45846061	SEMA6D	Int	.92	.92	9.57×10−1	2.12×10−1
rs1820489	46472393	…	…	.35	.24	6.98×10−3	2.16×10−1
rs1974961	46938695	SHC4	Int	.09	.10	9.62×10−1	2.17×10−1
rs530734	45908604	…	…	.93	.95	9.45×10−1	2.25×10−1
rs16962243	47350024	GALK2	Int	.80	.75	3.36×10−2	2.26×10−1
rs854158	48340369	HDC	Int	.78	.71	1.36×10−1	2.29×10−1
rs2059474	45595646	…	…	.82	.72	2.77×10−2	2.30×10−1
rs955845	45776451	…	…	.79	.75	3.20×10−1	2.35×10−1
rs11634585	48443741	GABPB2	Up	.24	.25	6.12×10−1	2.37×10−1
rs4259993	47011664	SHC4	Int	.87	.83	3.84×10−1	2.39×10−1
rs16960541	46157395	…	…	.87	.97	4.21×10−4	2.40×10−1
rs2279842	46782678	…	…	.20	.16	3.01×10−1	2.44×10−1
rs1045688	45852779	SEMA6D	3′ UTR	.74	.66	1.60×10−1	2.48×10−1
rs16961560	46835997	CEP152	Nonsyn	.93	.92	1.39×10−1	2.48×10−1
rs16963682	48448856	…	…	.91	.90	5.01×10−1	2.52×10−1
rs1453857	46116200	…	…	.53	.49	8.71×10−1	2.53×10−1
rs4775783	46963495	EID1	Down	.60	.76	1.37×10−3	2.54×10−1
rs17466389	46894932	CEP152	Up	.84	.86	8.61×10−1	2.55×10−1
rs12442429	48071636	ATP8B4	Int	.82	.79	7.58×10−1	2.56×10−1
rs628501	45814620	SEMA6D	Int	.72	.63	1.47×10−1	2.67×10−1
rs10519170	46473467	…	…	.64	.76	6.84×10−3	2.68×10−1
rs12913008	47980304	ATP8B4	Int	.68	.61	1.83×10−1	2.68×10−1
rs4775785	46979618	SHC4	Int	.60	.76	2.21×10−3	2.70×10−1
rs7169897	46939299	SHC4	Int	.71	.82	4.95×10−3	2.73×10−1
rs11070711	47703708	C15orf33	Up	.06	.06	6.60×10−1	2.81×10−1
rs479062	47877548	…	…	.62	.63	2.01×10−1	2.83×10−1
rs10851472	47005151	SHC4	Int	.55	.65	1.39×10−1	2.86×10−1
rs7162626	46731907	FBN1	Up	.54	.35	3.55×10−4	2.91×10−1
rs16960139	45871972	…	…	.95	.97	9.23×10−1	2.92×10−1
rs1025760	45855645	SEMA6D	Down	.88	.94	1.28×10−1	2.98×10−1
rs2278167	48285140	SLC27A2	Int	.69	.59	6.14×10−2	3.01×10−1
rs12914000	45694010	…	…	.93	.93	9.69×10−1	3.04×10−1
rs1435752	45702304	…	…	.66	.80	3.17×10−3	3.07×10−1
rs7175546	46511988	FBN1	Int	.86	.92	3.51×10−2	3.11×10−1
rs1843144	46311814	SLC12A1	Int	.30	.24	4.23×10−1	3.19×10−1
rs2413922	47009173	SHC4	Int	.74	.61	1.63×10−2	3.19×10−1
rs17399591	47500513	C15orf33	Int	.98	.98	7.92×10−1	3.19×10−1
rs596942	45869443	…	…	.88	.94	1.50×10−1	3.20×10−1
rs925104	47703947	C15orf33	Up	.75	.81	1.67×10−1	3.21×10−1
rs10851470	46757572	…	…	.18	.13	9.41×10−2	3.22×10−1
rs4774504	45827510	SEMA6D	Int	.15	.07	7.09×10−2	3.22×10−1
rs11638981	47108821	KIAA0256	Int	.75	.65	1.61×10−1	3.30×10−1
rs1968825	46837115	CEP152	Int	.22	.13	6.16×10−3	3.30×10−1
rs11854557	47360252	GALK2	Int	.79	.83	3.27×10−1	3.31×10−1
rs1872304	46446080	…	…	.38	.36	5.60×10−1	3.34×10−1
rs7162426	46974966	SHC4	Int	.73	.84	3.20×10−3	3.34×10−1
rs12898878	47025722	SHC4	Int	.61	.47	1.69×10−2	3.39×10−1
rs17472989	47211658	COPS2	Int	.80	.83	3.56×10−1	3.46×10−1
rs8036322	48054416	ATP8B4	Int	.21	.21	7.35×10−1	3.48×10−1
rs11639262	48186938	ATP8B4	Int	.92	.88	2.54×10−1	3.53×10−1
rs2413996	48060962	ATP8B4	Int	.56	.65	1.57×10−1	3.53×10−1
rs16960843	46454117	…	…	.90	.91	9.46×10−1	3.53×1010−1
rs494230	45903689	…	…	.49	.60	8.44×10−2	3.54×10−1
rs12593937	48051824	ATP8B4	Int	.20	.21	6.98×10−1	3.55×10−1
rs647903	45877708	…	…	.33	.42	5.43×10−2	3.55×10−1
rs17383671	46986669	SHC4	Int	.80	.88	1.53×10−1	3.57×10−1
rs4491452	47015113	SHC4	Int	.17	.12	1.26×10−1	3.64×10−1
rs963031	45901235	…	…	.89	.91	8.85×10−1	3.75×10−1
rs17479589	47540824	FGF7	Int	.77	.81	2.44×10−1	3.77×10−1
rs1369637	45732119	…	…	.22	.32	4.40×10−2	3.79×10−1
rs586118	45837068	SEMA6D	Int	.40	.46	3.18×10−1	3.79×10−1
rs3743286	45853084	SEMA6D	3′ UTR	.73	.67	2.80×10−1	3.80×10−1
rs669653	45944278	…	…	.71	.86	1.67×10−5	3.80×10−1
rs17431095	48448741	…	…	.77	.74	7.90×10−1	3.80×10−1
rs2924572	46039330	…	…	.19	.09	1.09×10−2	3.84×10−1
rs1561483	45856780	SEMA6D	Down	.73	.66	2.39×10−1	3.87×10−1
rs1699400	46834784	CEP152	Int	.78	.87	8.09×10−3	3.91×10−1
rs935001	47141674	…	…	.75	.72	1.31×10−1	3.94×10−1
rs17388803	45814496	SEMA6D	Int	.93	.91	5.10×10−1	3.94×10−1
rs586799	45812214	…	…	.76	.70	2.03×10−1	3.99×10−1
rs16959595	45582548	…	…	.89	.94	1.13×10−1	4.04×10−1
rs12438387	45768317	…	…	.30	.32	7.80×10−1	4.12×10−1
rs688475	45818759	SEMA6D	Int	.76	.70	1.77×10−1	4.12×10−1
rs3784296	48356641	GABPB2	Down	.57	.49	8.74×10−2	4.15×10−1
rs8029889	48338192	HDC	Int	.75	.78	9.91×10−1	4.15×10−1
rs12898855	45801857	SEMA6D	Int	.74	.69	2.70×10−1	4.16×10−1
rs6493312	46329848	SLC12A1	Int	.88	.90	5.61×10−1	4.16×10−1
rs784416	46800217	…	…	.21	.11	4.83×10−3	4.18×10−1
rs17384124	47005799	SHC4	Int	.97	.95	8.36×10−1	4.18×10−1
rs2555470	46534732	FBN1	Int	.66	.74	1.86×10−1	4.24×10−1
rs4482220	46889123	CEP152	Int	.81	.81	7.47×10−1	4.24×10−1
rs16960351	46001269	…	…	.87	.94	6.30×10−3	4.28×10−1
rs12440824	46012263	…	…	.80	.90	1.02×10−2	4.34×10−1
rs4775854	48201197	ATP8B4	Up	.73	.78	5.14×10−1	4.35×10−1
rs16962047	47174522	…	…	.94	.95	6.40×10−1	4.37×10−1
rs10519262	48219786	…	…	.73	.76	6.73×10−1	4.41×10−1
rs76739	45829438	SEMA6D	Int	.72	.68	3.19×10−1	4.42×10−1
rs12903325	48140569	ATP8B4	Int	.75	.74	6.10×10−1	4.44×10−1
rs17374298	45581770	…	…	.77	.69	2.00×10−1	4.44×10−1
rs4775692	45599347	…	…	.89	.94	1.29×10−1	4.45×10−1
rs11634375	47536840	FGF7	Int	.46	.46	9.32×10−1	4.47×10−1
rs12591300	47492033	C15orf33	Int	.75	.79	2.62×10−1	4.47×10−1
rs12441775	46637949	FBN1	Int	.53	.34	4.93×10−4	4.52×10−1
rs1610098	45593304	…	…	.63	.62	5.54×10−1	4.52×10−1
rs7178146	47792109	…	…	.73	.78	2.07×10−1	4.55×10−1
rs11070743	48137976	ATP8B4	Int	.44	.43	9.98×10−1	4.56×10−1
rs2899440	47980269	ATP8B4	Int	.46	.49	3.23×10−1	4.59×10−1
rs11634811	45783028	…	…	.74	.61	9.56×10−2	4.61×10−1
rs537052	45844911	SEMA6D	Int	.39	.40	9.09×10−1	4.61×10−1
rs11633714	47929321	ATP8B4	Down	.86	.85	8.66×10−1	4.63×10−1
rs17486446	47928420	ATP8B4	Down	.86	.85	8.66×10−1	4.63×10−1
rs11637901	48426740	…	…	.59	.54	1.85×10−1	4.74×10−1
rs11855823	48407007	GABPB2	Int	.59	.54	1.85×10−1	4.74×10−1
rs2081628	48379835	GABPB2	Int	.59	.54	1.85×10−1	4.74×10−1
rs8041517	48055189	ATP8B4	Int	.38	.46	1.79×10−1	4.74×10−1
rs17508371	48378086	GABPB2	Int	.60	.54	2.02×10−1	4.75×10−1
rs11634947	47784508	…	…	.76	.80	2.73×10−1	4.82×10−1
rs3809485	45838748	SEMA6D	Int	.38	.40	3.60×10−1	4.83×10−1
rs1007662	47002601	SHC4	Int	.72	.75	1.84×10−1	4.86×10−1
rs4774517	46546583	FBN1	Int	.66	.74	1.76×10−1	4.86×10−1
rs662968	47795735	…	…	.46	.47	3.92×10−1	4.86×10−1
rs7171359	46544752	FBN1	Int	.66	.74	1.76×10−1	4.86×10−1
rs16961049	46586202	FBN1	Int	.92	.98	3.12×10−3	4.89×10−1
rs16960244	45941874	…	…	.71	.86	5.57×10−5	4.95×10−1
rs16960510	46109866	…	…	.96	.96	6.40×10−1	4.97×10−1
rs11632411	48224343	…	…	.85	.84	5.79×10−1	5.02×10−1
rs10519249	48055662	ATP8B4	Int	.38	.46	1.89×10−1	5.07×10−1
rs784405	46819991	CEP152	Down	.13	.12	8.92×10−1	5.08×10−1
rs8030581	47180201	…	…	.11	.03	1.52×10−2	5.08×10−1
rs16961937	47033530	SHC4	Int	.96	.96	9.84×10−1	5.09×10−1
rs559561	47803588	…	…	.60	.59	2.41×10−1	5.12×10−1
rs17487348	47951100	ATP8B4	Int	.78	.71	1.47×10−1	5.13×10−1
rs10519208	47175541	…	…	.71	.68	1.32×10−1	5.14×10−1
rs10519225	47508070	FGF7	Int	.75	.79	3.74×10−1	5.19×10−1
rs12439479	47522135	FGF7	Int	.76	.72	1.20×10−1	5.34×10−1
rs1042078	46490165	FBN1	Exon	.34	.26	1.69×10−1	5.39×10−1
rs25458	46584599	FBN1	Exon	.76	.79	3.76×10−1	5.40×10−1
rs7164052	46779873	…	…	.90	.88	9.72×10−1	5.44×10−1
rs7162180	48286546	SLC27A2	Int	.56	.59	5.34×10−1	5.45×10−1
rs16961610	46847375	CEP152	Int	.81	.89	1.32×10−2	5.49×10−1
rs1369636	45732174	…	…	.82	.89	1.41×10−1	5.51×10−1
rs3743281	45844250	SEMA6D	Syn	.73	.67	4.14×10−1	5.56×10−1
rs16961671	46887207	CEP152	Int	.81	.88	1.23×10−2	5.56×10−1
rs8036777	48208710	ATP8B4	Up	.43	.44	7.72×10−1	5.57×10−1
rs17352842	46481503	FBN1	Down	.86	.93	3.55×10−1	5.57×10−1
rs4494483	47939566	ATP8B4	3′ UTR	.82	.78	3.58×10−1	5.58×10−1
rs12594698	46362248	SLC12A1	Int	.69	.74	2.25×10−1	5.63×10−1
rs3985863	45820660	SEMA6D	Int	.89	.81	3.34×10−2	5.63×10−1
rs1224659	45787001	…	…	.60	.59	7.95×10−1	5.66×10−1
rs11539517	47704633	C15orf33	Up	.85	.82	7.64×10−1	5.68×10−1
rs2163095	47429179	C15orf33	Int	.53	.59	2.27×10−1	5.69×10−1
rs17314486	45557607	…	…	.88	.84	5.86×10−1	5.69×10−1
rs11854679	46344865	SLC12A1	Int	.69	.75	2.03×10−1	5.79×10−1
rs4143837	48193475	ATP8B4	Int	.54	.61	6.58×10−2	5.81×10−1
rs1435763	45647928	…	…	.18	.26	9.56×10−2	5.82×10−1
rs2078139	48447126	…	…	.59	.65	5.08×10−1	5.83×10−1
rs2924566	46056053	…	…	.51	.25	6.25×10−6	5.89×10−1
rs12442472	46191894	SLC24A5	Up	.91	.97	1.09×10−2	5.90×10−1
rs8029928	45821317	SEMA6D	Int	.68	.74	1.46×10−1	5.90×10−1
rs6493364	47432702	C15orf33	Int	.72	.69	1.54×10−1	5.92×10−1
rs16960195	45913521	…	…	.93	.92	8.48×10−1	5.93×10−1
rs1023683	47494739	FGF7	Up	.85	.82	9.41×10−1	5.98×10−1
rs8024406	45871559	…	…	.77	.84	1.70×10−1	6.00×10−1
rs16961125	46628336	FBN1	Int	.92	.98	1.05×10−3	6.02×10−1
rs16961144	46637064	FBN1	Int	.92	.98	1.05×10−3	6.02×10−1
rs16961172	46651769	FBN1	Int	.92	.98	1.05×10−3	6.02×10−1
rs16961186	46660907	FBN1	Int	.92	.98	1.05×10−3	6.02×10−1
rs16961202	46664308	FBN1	Int	.92	.98	1.05×10−3	6.02×10−1
rs16961232	46682983	FBN1	Int	.92	.98	1.05×10−3	6.02×10−1
rs11637235	46420445	DUT	Int	.59	.42	1.12×10−3	6.04×10−1
rs934740	46994089	SHC4	Int	.15	.12	5.46×10−1	6.06×10−1
rs10519252	48070953	ATP8B4	Int	.57	.60	5.93×10−1	6.08×10−1
rs2924567	46055778	…	…	.80	.64	9.62×10−4	6.17×10−1
rs17382306	46923318	SHC4	Int	.84	.88	4.25×10−1	6.20×10−1
rs12907752	45890486	…	…	.77	.82	2.34×10−1	6.22×10−1
rs11632303	47317613	GALK2	Int	.86	.83	9.10×10−1	6.23×10−1
rs1059852	47938555	ATP8B4	3′ UTR	.86	.84	9.91×10−1	6.25×10−1
rs2414021	48198648	ATP8B4	5′ UTR	.73	.71	8.00×10−1	6.26×10−1
rs16963010	47957733	ATP8B4	Int	.76	.81	3.65×10−1	6.30×10−1
rs17402569	47702698	DTWD1	Int	.87	.83	7.18×10−1	6.33×10−1
rs2965317	46049012	…	…	.30	.15	3.50×10−3	6.44×10−1
rs16960346	45999405	…	…	.81	.90	2.95×10−2	6.46×10−1
rs4775730	46087470	…	…	.52	.77	4.40×10−6	6.49×10−1
rs616614	45889302	…	…	.25	.31	1.89×10−1	6.51×10−1
rs2414049	48453732	…	…	.27	.30	5.62×10−1	6.52×10−1
rs4580097	48102545	ATP8B4	Int	.41	.44	3.39×10−1	6.62×10−1
rs363830	46507944	FBN1	Exon	.93	.97	1.53×10−1	6.64×10−1
rs2289181	46877467	CEP152	Nonsyn	.81	.88	1.83×10−2	6.65×10−1
rs2899453	48154202	ATP8B4	Int	.44	.43	9.20×10−1	6.66×10−1
rs17478785	47527596	FGF7	Int	.76	.79	5.75×10−1	6.67×10−1
rs1059850	47938821	ATP8B4	3′ UTR	.77	.75	8.52×10−1	6.71×10−1
rs11635140	46569855	FBN1	Int	.44	.62	1.13×10−3	6.74×10−1
rs17376095	45620983	…	…	.90	.84	3.26×10−1	6.77×10−1
rs16963076	47999573	ATP8B4	Int	.91	.88	5.79×10−1	6.78×10−1
rs4592603	47018806	SHC4	Int	.26	.18	1.59×10−1	6.79×10−1
rs2965318	46051787	…	…	.31	.16	2.10×10−3	6.80×10−1
rs16959719	45671245	…	…	.89	.94	9.34×10−2	6.83×10−1
rs10519233	47693662	DTWD1	Up	.87	.82	6.41×10−1	6.91×10−1
rs17483139	47791804	…	…	.87	.82	6.41×10−1	6.91×10−1
rs17361868	46584438	FBN1	Int	.88	.92	1.40×10−1	6.98×10−1
rs607289	45862845	SEMA6D	Down	.37	.38	8.72×10−1	7.01×10−1
rs17480434	47600635	C15orf33	Int	.87	.82	6.17×10−1	7.01×10−1
rs1993150	45854810	SEMA6D	Down	.53	.56	7.30×10−1	7.05×10−1
rs2009833	47980845	ATP8B4	Int	.48	.53	1.76×10−1	7.06×10−1
rs688709	45870675	…	…	.79	.71	9.31×10−2	7.09×10−1
rs11070627	46258816	MYEF2	Up	.31	.06	1.49×10−10	7.12×10−1
rs11631496	46863210	CEP152	Int	.88	.88	8.60×10−1	7.12×10−1
rs12913316	46275146	…	…	.31	.06	2.30×10−10	7.12×10−1
rs1320052	46282800	SLC12A1	Up	.31	.06	2.30×10−10	7.12×10−1
rs6493311	46326879	SLC12A1	Syn	.57	.67	6.92×10−2	7.23×10−1
rs1834640	46179457	…	…	.51	.87	3.28×10−15	7.25×10−1
rs16959955	45789992	SEMA6D	Up	.86	.87	5.77×10−1	7.30×10−1
rs34912622	46882494	CEP152	Int	.97	.96	3.73×10−1	7.31×10−1
rs363836	46510176	FBN1	Exon	.93	.97	3.67×10−1	7.33×10−1
rs12593270	45534632	…	…	.64	.64	9.21×10−1	7.40×10−1
rs11636795	47463394	C15orf33	Int	.24	.27	1.58×10−1	7.41×10−1
rs1912643	45777558	…	…	.83	.78	2.70×10−1	7.41×10−1
rs17394420	47208108	COPS2	Int	.85	.81	8.78×10−1	7.41×10−1
rs16963644	48425969	…	…	.89	.89	9.23×10−1	7.43×10−1
rs11070672	47048808	SHC4	Up	.38	.40	6.87×10−1	7.53×10−1
rs8039653	46359927	SLC12A1	Int	.70	.76	1.53×10−1	7.60×10−1
rs16960880	46481074	FBN1	Down	.88	.93	1.85×10−1	7.70×10−1
rs938043	45696805	…	…	.82	.74	2.03×10−1	7.77×10−1
rs532598	45845363	SEMA6D	Down	.64	.72	2.16×10−1	7.81×10−1
rs17476940	47412076	GALK2	Down	.80	.76	9.78×10−1	7.84×10−1
rs16961174	46652169	FBN1	Int	.93	.98	2.84×10−3	7.85×10−1
rs751467	46034246	…	…	.29	.41	4.74×10−2	7.89×10−1
rs585451	47795017	…	…	.78	.78	9.03×10−1	7.90×10−1
rs10519257	48176213	ATP8B4	Int	.75	.75	7.73×10−1	7.93×10−1
rs479173	45978990	…	…	.81	.89	5.82×10−3	7.93×10−1
rs8040116	47369881	GALK2	Int	.75	.76	6.77×10−1	7.93×10−1
rs17470994	47091617	KIAA0256	Int	.95	.94	8.79×10−1	7.93×10−1
rs16963141	48061447	ATP8B4	Int	.89	.87	3.43×10−1	7.99×10−1
rs10519150	45926447	…	…	.85	.89	2.76×10−1	8.03×10−1
rs17404262	47754911	…	…	.87	.82	4.95×10−1	8.04×10−1
rs4774545	47847312	…	…	.93	.93	7.21×10−1	8.05×10−1
rs17499601	48214938	…	…	.88	.85	6.29×10−1	8.05×10−1
rs10519174	46521916	FBN1	Int	.53	.37	3.09×10−3	8.08×10−1
rs16962914	47880927	…	…	.95	.96	6.01×10−1	8.08×10−1
rs16961274	46712407	FBN1	Int	.94	.98	3.37×10−2	8.10×10−1
rs671291	45925000	…	…	.68	.78	7.81×10−2	8.11×10−1
rs733220	46170229	…	…	.93	.95	8.52×10−1	8.11×10−1
rs755067	47955564	ATP8B4	Int	.52	.56	6.03×10−1	8.12×10−1
rs4375601	47959139	ATP8B4	Int	.77	.77	7.16×10−1	8.15×10−1
rs16960538	46156777	…	…	.94	.97	7.52×10−1	8.15×10−1
rs16960540	46156866	…	…	.94	.97	7.52×10−1	8.15×10−1
rs17423970	46089356	…	…	.80	.65	1.52×10−3	8.21×10−1
rs4544192	47939482	ATP8B4	3′ UTR	.83	.80	5.81×10−1	8.23×10−1
ss69356384	48416695	GABPB2	Int	.75	.71	3.99×10−1	8.24×10−1
rs6493426	48375850	GABPB2	Int	.75	.71	3.94×10−1	8.25×10−1
rs16961323	46746374	…	…	.85	.88	1.57×10−1	8.25×10−1
rs8041979	47951146	ATP8B4	Int	.82	.80	6.25×10−1	8.26×10−1
rs16962871	47838116	…	…	.75	.79	3.01×10−1	8.29×10−1
rs11636875	47040769	…	…	.67	.59	1.90×10−1	8.35×10−1
rs893160	46737569	…	…	.19	.18	8.44×10−1	8.36×10−1
rs491996	45965538	…	…	.67	.82	1.39×10−4	8.37×10−1
rs17499691	48216033	…	…	.89	.87	7.31×10−1	8.39×10−1
rs16963067	47996550	ATP8B4	Int	.91	.88	3.37×10−1	8.40×10−1
rs16963180	48084344	ATP8B4	Int	.97	.96	6.30×10−1	8.43×10−1
rs11853852	48199132	ATP8B4	Up	.55	.52	4.17×10−1	8.44×10−1
rs12912380	45734583	…	…	.94	.94	6.52×10−1	8.47×10−1
rs1365508	48288276	SLC27A2	Int	.70	.69	5.26×10−1	8.47×10−1
rs10519258	48187342	ATP8B4	Int	.95	.94	3.26×10−1	8.50×10−1
rs7170781	48180287	ATP8B4	Int	.61	.63	5.04×10−1	8.50×10−1
rs8036806	46926246	SHC4	Int	.96	.96	6.58×10−1	8.54×10−1
rs12915792	47225618	COPS2	Int	.26	.24	5.80×10−1	8.56×10−1
rs2067885	48009237	ATP8B4	Int	.83	.83	4.96×10−1	8.59×10−1
rs16960438	46062139	…	…	.87	.93	1.17×10−1	8.59×10−1
rs8039548	45888247	…	…	.84	.88	1.64×10−1	8.60×10−1
rs16960516	46116834	…	…	.90	.83	5.49×10−2	8.63×10−1
rs4494482	47939458	ATP8B4	3′ UTR	.84	.81	5.95×10−1	8.65×10−1
rs2934177	46000839	…	…	.03	.02	7.78×10−1	8.66×10−1
rs16960341	45997010	…	…	.93	.94	5.23×10−1	8.68×10−1
rs17509458	48452654	…	…	.74	.70	4.02×10−1	8.68×10−1
rs16963311	48192467	ATP8B4	Int	.88	.88	7.08×10−1	8.70×10−1
rs1656631	45623301	…	…	.70	.67	6.34×10−1	8.71×10−1
rs12324326	45935427	…	…	.82	.88	1.79×10−1	8.73×10−1
rs16959550	45549747	…	…	.91	.95	2.28×10−1	8.73×10−1
rs4426299	48000077	ATP8B4	Int	.91	.88	3.55×10−1	8.74×10−1
rs17499726	48216520	…	…	.95	.94	3.90×10−1	8.75×10−1
rs11854358	47007749	SHC4	Int	.30	.31	8.34×10−1	8.85×10−1
rs12591825	48098427	ATP8B4	Int	.25	.28	3.34×10−1	8.89×10−1
rs12909202	48400026	GABPB2	Int	.75	.71	3.70×10−1	8.89×10−1
rs16960493	46096739	…	…	.91	.94	5.24×10−1	8.92×10−1
rs9806163	46591131	FBN1	Int	.65	.74	9.19×10−2	8.98×10−1
rs8030172	45924790	…	…	.63	.56	7.81×10−2	9.00×10−1
rs2413959	47547717	FGF7	Int	.72	.70	3.81×10−1	9.02×10−1
rs11855291	45858167	SEMA6D	Down	.76	.80	4.81×10−1	9.03×10−1
rs4775708	45824941	SEMA6D	Int	.84	.84	8.31×10−1	9.05×10−1
rs2114438	46487221	FBN1	Down	.48	.34	1.27×10−2	9.07×10−1
rs4774557	48220992	…	…	.61	.62	4.79×10−1	9.07×10−1
rs1669832	45897944	…	…	.69	.63	3.18×10−1	9.08×10−1
rs12593807	46334019	SLC12A1	Int	.72	.78	1.12×10−1	9.10×10−1
rs8031680	47524988	C15orf33	Int	.06	.06	6.59×10−1	9.13×10−1
rs692339	47791940	…	…	.04	.04	5.79×10−1	9.14×10−1
rs769136	47799924	…	…	.70	.75	1.51×10−1	9.14×10−1
rs12594956	48409126	GABPB2	Int	.49	.50	9.27×10−1	9.16×10−1
rs16960904	46494063	FBN1	Int	.93	.98	4.26×10−2	9.16×10−1
rs6493427	48376049	GABPB2	Int	.47	.48	9.67×10−1	9.22×10−1
rs1559677	45525355	…	…	.44	.51	2.40×10−1	9.23×10−1
rs8025221	48148835	ATP8B4	Int	.84	.83	9.37×10−1	9.23×10−1
rs1114003	47336143	GALK2	Int	.94	.94	8.63×10−1	9.26×10−1
rs568215	45850368	SEMA6D	Syn	.21	.30	3.62×10−2	9.26×10−1
rs477077	45884409	…	…	.44	.47	5.75×10−1	9.27×10−1
rs860526	48342303	HDC	Int	.52	.48	2.25×10−1	9.27×10−1
rs1506524	46676280	FBN1	Int	.68	.77	1.04×10−1	9.30×10−1
rs16960979	46551641	FBN1	Int	.92	.97	2.17×10−2	9.30×10−1
rs17416784	48355447	GABPB2	Down	.97	.96	5.49×10−1	9.34×10−1
rs12912519	48220812	…	…	.84	.80	4.12×10−1	9.37×10−1
rs9788716	45926940	…	…	.83	.88	2.14×10−1	9.41×10−1
rs1797311	48392109	GABPB2	Int	.54	.53	9.98×10−1	9.44×10−1
rs8031017	46470746	…	…	.38	.30	6.23×10−2	9.49×10−1
rs16961806	46966198	EID1	Down	.82	.88	3.62×10−2	9.52×10−1
ss69356376	45998012	…	…	.91	.96	4.65×10−2	9.53×10−1
rs854160	48343002	HDC	Int	.47	.45	5.12×10−1	9.54×10−1
rs501916	45840521	SEMA6D	Syn	.57	.58	9.55×10−1	9.55×10−1
rs16960326	45984783	…	…	.94	.98	1.07×10−2	9.56×10−1
rs12591556	46558253	FBN1	Int	.86	.94	8.60×10−3	9.57×10−1
rs17415911	48284586	SLC27A2	Int	.84	.87	7.48×10−1	9.58×10−1
rs11070741	48098440	ATP8B4	Int	.22	.24	4.34×10−1	9.61×10−1
rs8041014	48085001	ATP8B4	Int	.22	.24	4.34×10−1	9.61×10−1
rs1224606	45933110	…	…	.86	.91	4.22×10−1	9.66×10−1
rs16959519	45528224	…	…	.81	.88	1.84×10−1	9.69×10−1
rs597543	45930092	…	…	.19	.27	2.88×10−1	9.72×10−1
rs16963207	48090156	ATP8B4	Int	.84	.85	8.58×10−1	9.83×10−1
rs7342574	48217475	…	…	.58	.59	4.44×10−1	9.90×10−1
rs652281	45826380	SEMA6D	Int	.83	.91	1.94×10−2	9.92×10−1
rs1866501	45839887	SEMA6D	Syn	.83	.91	1.76×10−2	9.93×10−1
rs3743279	45843511	SEMA6D	Nonsyn	.83	.91	1.76×10−2	9.93×10−1
rs8036851	45830750	SEMA6D	Int	.83	.91	1.76×10−2	9.93×10−1
rs1797316	48273679	SLC27A2	Int	.09	.10	5.98×10−1	9.94×10−1
rs677207	45970045	…	…	.70	.82	6.41×10−4	9.97×10−1
rs16959999	45816306	…	…	.83	.91	1.87×10−2	9.98×10−1

^ars or ss identifier in dbSNP.

^bChromosome base position on NCBI build 36.2.

^cGene symbol from Entrez Gene.

^dInt = intronic; nonsyn = nonsynonymous; syn = synonymous; up = within 10 kb of the transcriptional start site; down = within 10 kb of the transcriptional stop site.

^eFrequency of allele 1 in the reflectance groups of cohort 2.

^fP value conditional on rs1426654.

Figure 3. .

Association signals, linkage disequilibrium, and genes in the region 45,524,265–48,470,992 on chromosome 15q21. Negative log₁₀ association P values (found by logistic regression and likelihood-ratio tests) are displayed for cohorts 1 and 2 and for an analysis conditioned on rs1426654 in cohort 2. The R² value between each SNP genotyped in the region and the primary associated SNP rs1426654 is also displayed. The SNPs rs1834640 and rs1426654 show the strongest associations in cohort 1 and cohort 2, respectively. Red horizontal lines are drawn at the Bonferroni-corrected significance threshold for the genomewide scan (α=0.05/1,502,205=3.3×10^-8) for cohort 1 and 2 associations. In the association analysis conditional on rs1426654, no SNPs met a significance level that would correct for the number of SNPs in the region (α=0.05/407=1.23×10^-4). This figure was made using the Generic Genome Browser.⁴⁷ Chr = chromosome.

Furthermore, by performing an association analysis conditional on the genotypes of rs1426654 in cohort 2, we observed that none of the other SNPs rise above the Bonferroni threshold for significance of conditional association in the region (α=0.05/407=1.23×10^-4) (fig. 3 and table 6). When the conditional analysis was performed in reverse, by evaluation of the association of rs1426654 in cohort 2 conditional on each of the other 407 SNPs in this region in turn, the rs1426654 association was found to be significant at the genomewide level (P<3.3×10^-8) for all SNPs but rs1834640, with which it is in the strongest linkage disequilibrium (P=.003 for association of skin reflectance with rs1426654 conditional on rs1834640; P=.73 for association with rs1834640 conditional on rs1426654). Thus, the data are consistent with the existence of a single strong primary association at rs1426654 in the SLC24A5 gene, and the multiplicity of apparent single-SNP associations in this region can be explained by the pattern of linkage disequilibrium with this SNP (fig. 3). Because of the lack of genomewide individual genotyping data for a South Asian population, we are unable to compare the extent of linkage disequilibrium in this region with that in the rest of the genome. Of the 407 SNPs genotyped in this region, only rs1834640 is in high linkage disequilibrium with rs1426654 (R²>0.9), and 3 more SNPs are in moderate linkage disequilibrium (R²>0.5) (fig. 3). Although we cannot rule out other associations of small effect within this chromosomal region, the evidence points strongly to a single association of large effect size, located at or in strong disequilibrium with the nonsynonymous SNP rs1426654 within SLC24A5.

Dominance and Interactions among Three Nonsynonymous Polymorphisms

The single-SNP association analyses with cohorts 1 and 2 identified three nonsynonymous polymorphisms in three genes—rs1426654 (p.A111T) in SLC24A5, rs16891982 (p.L374F) in SLC45A2, and rs1042602 (p.S192Y) in TYR—that show genomewide significance for association with skin pigmentation, on the basis of a multiplicative model of risk. Under a simple additive allele-risk model, the squared correlation (R²) between the dichotomously defined skin-reflectance trait and the genotypes is directly related to the fraction of the variance of skin reflectance accounted for by each SNP. The primary associated SNP, rs1426654 in SLC24A5, has the largest effect on skin reflectance of the three SNPs, with R²=0.33, compared with R²=0.036 for rs16891982 in SLC45A2 and R²=0.025 for rs1042602 in TYR.

We examined whether there is evidence within the cohort 2 data of deviations from a multiplicative model of risk, such as a dominance effect or interactions among the loci. We have very good power to detect dominant or recessive inheritance for rs1426654, which has a high minor-allele frequency (49% in the L group and 10% in the H group) and a very large effect on skin reflectance. However, our power to detect deviations from additive inheritance for the other two SNP associations is much lower because of the smaller size of their effect. Likelihood-ratio tests were used to assess the significance of an added dominance term, but none of the tests showed significant deviations from multiplicative risk (P=.58 for rs1426654, P=.20 for rs16891982, and P=.91 for rs1042602), which provides no evidence to support dominant or recessive inheritance for any of the three SNPs.

We next explored whether the three SNPs collectively explain more of the trait than does the combination of their single-SNP associations. Interactions were modeled within the framework of logistic regressions by adding indicator variables corresponding to specific genotype combinations for pairs of SNPs. We tested each of the three pairs separately and found no evidence for pairwise interactions (rs1426654-rs16891982, P=.77; rs1426654-rs1042602, P=.38; rs16891982-rs1042602, P=.12). Therefore, our data are most consistent with a simple model for skin pigmentation comprising additive contributions from alleles at the three associated nonsynonymous polymorphisms. Because our study design used the tails of the skin-reflectance distribution, we cannot directly determine the overall fraction of the trait variance in the population that is accounted for by the associated SNPs. Nevertheless, it is evident that these three markers collectively account for a large fraction of the wide variability of skin pigmentation within South Asians.

Discussion

This is the first genomewide association study, to our knowledge, to investigate the genetic determinants of normal skin pigmentation within a human population. With the use of objective quantitative measurements of melanin content, our study clearly identifies three loci—TYR, SLC45A2, and SLC24A5—that contribute to the natural variation in skin pigmentation within a South Asian population. Within each of these genes, we found polymorphisms that met genomewide significance on association tests, with the associations replicated in a second South Asian cohort. The contributions of these polymorphisms to skin pigmentation were found to be independent and additive across genes, with no evidence of dominant or recessive effects at any of these loci. Collectively, these three genes account for a large fraction of the wide, naturally occurring variation in skin pigmentation in South Asians.

Strikingly, our primary associated SNP in SLC24A5 may explain >30% of the variance of the dichotomously defined skin reflectance in our study population. As a result of the interplay between the strength of this single-SNP association and linkage disequilibrium with SNPs in its vicinity, several other significant associations were observed in a 2.4-Mb region on chromosome 15q21. Although we cannot completely exclude the possibility that multiple real independent associations exist in this chromosomal 15 region, our conditional analysis shows unambiguously that the association signals on all SNPs in the region are far less significant than those for the primary associated SNP in SLC24A5.

This is also one of the first genomewide association studies to focus on a South Asian population. This population is known to be highly genetically diverse, and, because ancestry is known to be highly confounded with skin pigmentation, the treatment of population structure was crucial to our association analysis. Our analysis of population structure was based on the individual genotypes of 968 individuals with 292 unlinked SNPs. Notwithstanding the limitations imposed by the size of this data set on the detection of structure,⁴⁸ the results for these 292 SNPs indicated that we were able to adequately correct for stratification. It is encouraging that, despite the use of a study population that is genetically diverse and a trait that is strongly confounded with ancestry, we were able to detect three true SNP associations.

At the time of this study, mutations in two of these genes, TYR and SLC45A2, were known to be involved in different forms of oculocutaneous albinism (OCA1 [MIM 203100] and OCA4 [MIM 606574]),^49,50 yet their contribution to the normal range of skin pigmentation between and within populations was just beginning to be studied. The remaining locus, SLC24A5, which displays the largest genetic effect on pigmentation status in the South Asian population studied here, was an unknown pigmentation gene during our study. The SLC24A5 gene has since been shown to play an important role in pigmentation, because a mutation in this gene was identified as the genetic basis of the hypopigmentation phenotype in golden zebrafish.²² During the past few years, several candidate-gene admixture studies on a cohort of African Americans and African Caribbeans has shown that the same three nonsynonymous polymorphisms that we found in our South Asian cohort—rs1042602 in TYR, rs16891982 in SLC45A2, and rs1426654 in SLC24A5—are significantly associated with melanin content measured by reflectometry.^13,21,22,24 When taken together with the results of our study, the evidence is compelling that these three loci are a large component of the genetic influence of constitutive pigmentation in these diverse populations.

The importance of these three particular SNPs—rs1042602 in TYR, rs16891982 in SLC45A2, and rs1426654 in SLC24A5—to skin-pigmentation variation within other populations is likely to be highly variable across the globe. All three of the SNPs are essentially monomorphic in African and East Asian populations and thus are unlikely to contribute significantly to skin-pigmentation variation within those groups.^13,25,29,30 For populations of European ancestry, the contributions of these three SNPs to intrapopulation skin-pigmentation variation are quite distinct from each other. The SLC24A5 SNP rs1426654 is nearly fixed for one allele in all European groups.^{11,13,22,29,30,51} In comparison, the TYR SNP rs1042602 is highly polymorphic in all European populations, but, despite its variability, previous studies found that this SNP was not significantly associated with skin-reflectance measurements in a white population.^13,21,24,29 In contrast to the TYR SNP, the SLC45A2 SNP rs16891982 displays a gradient of increasing minor-allele frequency in European populations from north to south^13,26,52 and has been significantly associated with subjectively measured skin color in a population of European ancestry.¹⁹ It is important to distinguish the involvement of these three particular SNPs in skin-pigmentation variation within populations from the potential contribution of the three loci (TYR, SLC24A4, and SLC45A2). For example, two functional promoter variants and another nonsynonymous polymorphism in SLC45A2, rs26722 (p.E272K), have also been significantly associated with subjectively measured skin color in a population of European ancestry.²⁰ Whereas the promoter variants in SLC45A2 were not included in our study, rs26722 was assayed in the present study but showed weaker evidence of association with skin-reflectance measurements in our South Asian population (joint cohort association P=.02), as compared with the significantly associated rs16891982 SNP.

Other genes have been implicated either directly or indirectly with pigmentation-related traits in previous work. Several reported associations are based on the examination of markers in pigmentation candidate genes for allele-frequency differences between populations with known skin-pigmentation differences^11–13,51 or are associations with a different qualitative measure of pigmentation, such as the Fitzpatrick skin type and subjective classification of skin color by visual assessment,^53–55 all of which make direct comparisons with our results difficult. In contrast, other associations reported for skin pigmentation used quantitative measures of reflectance similar to those in our study^{17,18,21,23,24}; therefore, our results can be compared directly with those studies. A modest association has been found between skin pigmentation from reflectance measurements for SNP rs1800404 within the gene OCA2²¹ in an African American population (P=.01). Although rs1800404 was not assayed in our study, the closest OCA2 SNP that was included in our study, rs1037208, located 4,414 bp from rs1800404, yielded the most significant joint cohort association P value (P=5.6×10^-5) of several OCA2 SNPs that we assayed with individual genotyping. On the basis of International HapMap data for the CEU sample (Utah residents with ancestry from northern and western Europe),²⁹ these two SNPs are in modest linkage disequilibrium with one another (R²=0.35). Although we cannot establish beyond a doubt that these two SNPs correspond to a single locus in our population, it is quite plausible that they are highly correlated in South Asians; therefore, our results may support the prior association of OCA2 and skin pigmentation. Another study found an association with skin reflectance in a Tibetan population under an epistatic model (P=.01) for a pair of SNPs in OCA2 and MC1R¹⁷—rs12910433 and a nonsynonymous polymorphism, rs2228479 (p.V92M), respectively. Both these SNPs were individually genotyped in cohort 1 in our study, and, whereas the OCA2 SNP rs12910433 showed no evidence of association (P=.3), the MC1R SNP rs2228479 had the lowest P value (P=6.8×10^-4) of several MC1R SNPs that were individually genotyped. There is no evidence of interactions between these two SNPs in our study population (P=.91). Lastly, one SNP in the 3′ UTR of the ASIP gene, rs6058017, has been associated with darker skin color in female African Americans¹⁸ (P<.001). This SNP was included in our pooled study but showed no allele-frequency difference between the low- and high-reflectance groups (), and attempts to individually genotype this SNP in cohort 1 were unsuccessful. Since the previous association showed significant association with only female, but not male, skin pigmentation, our pooled genotyping strategy that used mixed sexes would not have the power to find such an association.

It is important to note that the application of a very strict Bonferroni correction of 1,502,205 independent tests, combined with the pooled genotyping design, greatly reduces the power of this study to detect real SNP associations with small effects on skin pigmentation; thus, it is possible that loci other than TYR, SLC45A2, and SLC24A5 may affect skin-pigmentation variation within the South Asian population. Furthermore, as was mentioned earlier in this discussion, several studies comparing allele frequencies of candidate skin-pigmentation SNPs across multiple global populations present evidence supporting an independent genetic mechanism for the lighter skin pigmentation of East Asians that is distinct from genetic factors influencing lighter pigmentation in Europeans.^12,13,25 Therefore, it is likely that loci other than the three identified in this study influence skin pigmentation in other global populations.