Skin pigmentation functions as a shield to prevent UV damage to the DNA of epidermal cells. A good number of GWASs on skin color variation or skin photosensitivity have been conducted that discovered a large number of associated loci (Ganguly et al., 2019Ganguly K. Saha T. Saha A. Dutta T. Banerjee S. Sengupta D. et al.Meta-analysis and prioritization of human skin pigmentation-associated GWAS-SNPs using ENCODE data-based web-tools.Arch Dermatol Res. 2019; 311: 163-171Google Scholar). Polymorphisms/mutations at these loci have been used to study the molecular type of skin color for individuals from different continental groups (Chen et al., 2021Chen Y. Branicki W. Walsh S. Nothnagel M. Kayser M. Liu F. et al.The impact of correlations between pigmentation phenotypes and underlying genotypes on genetic prediction of pigmentation traits.Forensic Sci Int Genet. 2021; 50: 102395Google Scholar) and have been associated with various forms of albinism (Marçon and Maia, 2019Marçon C.R. Maia M. Albinism: epidemiology, genetics, cutaneous characterization, psychosocial factors.An Bras Dermatol. 2019; 94: 503-520Google Scholar), loss of photoprotection, and increased rates of photoaging (Liu et al., 2016Liu F. Hamer M.A. Deelen J. Lall J.S. Jacobs L. van Heemst D. et al.The MC1R gene and youthful looks.Curr Biol. 2016; 26: 1213-1220Google Scholar). However, there have been surprisingly few GWASs conducted in East Asian populations, likely owing to the presumption of a low skin color variation in East Asians (Rawofi et al., 2017Rawofi L. Edwards M. Krithika S. Le P. Cha D. Yang Z. et al.Genome-wide association study of pigmentary traits (skin and iris color) in individuals of East Asian ancestry.PeerJ. 2017; 5: e3951Google Scholar). The question regarding the presence of East Asian‒specific skin color alleles remains to be answered. Individual typology angle (ITAº) is a quantitative skin color measurement based on the colorimetric parameters of the L∗a∗b∗ system (Chardon et al., 1991Chardon A. Cretois I. Hourseau C. Skin colour typology and suntanning pathways.Int J Cosmet Sci. 1991; 13: 191-208Google Scholar). The validity of ITAº in serving as an objective skin color measurement has been extensively evaluated with regard to its correlations with constitutive pigmentation, the geographical distribution of skin pigmentation, and biological markers of UV-induced erythema (Del Bino and Bernerd, 2013Del Bino S. Bernerd F. Variations in skin colour and the biological consequences of ultraviolet radiation exposure.Br J Dermatol. 2013; 169: 33-40Google Scholar). Despite the wide use of ITAº in clinical dermatology, few genetic studies had adopted it as the measurement of skin color. In this study, we report a GWAS of ITAº in Chinese, followed by replications in Chinese and Latin Americans. The discovery sample included a total of 6,964 individuals of Chinese origin from two cohorts: the Jidong cohort (n = 5,034) and the National Survey of Physical Traits cohort (n = 1,930). Studies in these cohorts were approved by the Ethics Committee of Kailuan General Hospital of Tangshan City and the Medical Ethics Committee, Staff Hospital, Jidong Oilfield Branch, China National Petroleum Corporation in July 2013 (approval number 2013 YILUNZI1) as well as the Ethics Committees of Fudan University (14117) and the Shanghai Institutes for Biological Sciences (ER-SIBS-261410). The replication sample included a total of 2,053 individuals from two cohorts: the Taizhou longitudinal cohort of Chinese origin (n = 1,787) and the Colombian cohort of Latin American origin (n = 266). The Taizhou longitudinal cohort study was conducted with the approval of the Ethics Committee of Fudan University (Ethics Research Approval 85), Shanghai, China, and the Colombian cohort study was conducted with the approval of the bioethics committee of the Odontology Faculty at the University of Antioquia (CONCEPTO 01-2013). All participants provided written informed consent. For the discovery cohorts, sample characteristics and phenotyping details are provided in Supplementary Materials and Methods (Supplementary Table S1 and Figure 1a and b). In brief, we derived ITAº from high-resolution portrait photos, which were processed using an in-house pipeline, involving a face detector, automated facial landmarking, cheek segmentation, and color analysis. In a subgroup of 50 randomly selected samples, the ITAº derived from the portrait photos was highly correlated with the ITAº measured by a spectrophotometer (r = 0.94, P = 4.42 × 10−24, Supplementary Figure S1). The Chardon skin color type (Chardon et al., 1991Chardon A. Cretois I. Hourseau C. Skin colour typology and suntanning pathways.Int J Cosmet Sci. 1991; 13: 191-208Google Scholar), as defined by ITAº cutoffs, was fairly concordant with human perception (Kappa = 0.47). Although dark skins were not seen and very light (1.51%) and brown (0.27%) skins were rarely seen in our Chinese samples, ITAº did show a substantial variance (Var = 8.03, mean = 35.95, minimum = –26.36, maximum = 68.42) across light (23.41%), intermediate (59.69%), and tan (15.12%) categories. Females had significant lighter skin than males (β = 6.18, P = 9.22 × 10−270) and aging increased coloration (β = –0.22, P = 8.08 × 10−244), which were consistent with previous observations (Tan et al., 2020Tan Y. Wang F. Fan G. Zheng Y. Li B. Li N. et al.Identification of factors associated with minimal erythema dose variations in a large-scale population study of 22 146 subjects.J Eur Acad Dermatol Venereol. 2020; 34: 1595-1600Google Scholar). Z-transformed ITAº was used in the subsequent genetic association analysis (Supplementary Figure S2). In the discovery stage of the study, we conducted a genome-wide inverse variance, a fixed-effect meta-analysis of two GWASs, which were independently conducted in the Jidong cohort and the National Survey of Physical Traits cohort, totaling 6,964 Chinese individuals. No evidence of population substratification or genome inflation was detected (λ = 0.99, Figure 1c). The meta-analysis identified a total of 19 SNPs at three genomic loci showing genome-wide significant association with z-transformed ITAº (Supplementary Table S2), including one previously unreported locus on 9p21.3 (SLC24A2) and two previously known loci on 15q12.6 (OCA2) and 15p21.1 (SLC24A5). The locus at 9p21.3, as the most significant signal over the genome, contained a total of 14 significant SNPs, where the lead SNP (rs10122939) was an intron variant of SLC24A2 (Figure 2a). The derived G allele showed a skin darkening effect reaching genome-wide significance in the Jidong cohort (β = –0.10, P = 1.29 × 10−9), nominal significance in the National Survey of Physical Traits cohort (β = –0.07, P = 4.71 × 10−3), and further enhanced significance in the meta-analysis (β = –0.09, P = 3.61 × 10−11). This association was also genome-wide significant for the original ITAº without z-transformation (β = –1.89, P = 7.28 × 10−10). No genome-wide significant association signal was observed at 9p21.3 after conditioning on the genotype of the lead SNP rs10122939 (Supplementary Figure S3). The G allele of rs10122939 was highly prevalent in East Asians (our sample: f = 0.34; in CHB of the 1000 Genomes Project: f = 0.29) but was rare in Europeans (in EUR of 1000 Genomes Project, f = 0.004, Figure 2b), which may explain the failure of previous European GWASs in detecting its effect (other SNPs in linkage disequilibrium showed the same pattern, Supplementary Table S2). A series of population genetic analyses did not reveal significant evidence for positive selection surrounding rs10122939 at 9p21.3 (Supplementary Figure S4). The other two loci have been previously associated with skin color. On chromosome 15q12.6, two well-known East Asian‒specific missense variants rs1800414 (His615Arg) and rs74653330 (Ala481Thr) of OCA2 (Edwards et al., 2010Edwards M. Bigham A. Tan J. Li S. Gozdzik A. Ross K. et al.Association of the OCA2 polymorphism His615Arg with melanin content in east Asian populations: further evidence of convergent evolution of skin pigmentation.PLoS Genet. 2010; 6e1000867Google Scholar; Yang et al., 2016Yang Z. Zhong H. Chen J. Zhang X. Zhang H. Luo X. et al.A genetic mechanism for convergent skin lightening during recent human evolution.Mol Biol Evol. 2016; 33: 1177-1187Google Scholar) showed genome-wide significant association with ITAº, where the derived C allele of rs1800414 and T allele of rs74653330 had a skin-lightening effect (P = 1.11 × 10−10 and P = 3.05 × 10−8, respectively; Supplementary Table S2 and Supplementary Figure S5a and b). Both alleles were nearly absent outside of East Asia (Supplementary Figure S5c and d). On chromosome 15p21.1, we confirmed the well-known effect of a missense variant rs1426654 of SLC24A5, of which the derived A allele had low frequency but a large skin-lightening effect (P = 4.64 × 10−10, Supplementary Table S2 and Supplementary Figure S6). We then looked up a total of 9,183 SNPs from four recently published large GWASs of skin/hair/eye pigmentation traits (Adhikari et al., 2019Adhikari K. Mendoza-Revilla J. Sohail A. Fuentes-Guajardo M. Lampert J. Chacón-Duque J.C. et al.A GWAS in Latin Americans highlights the convergent evolution of lighter skin pigmentation in Eurasia.Nat Commun. 2019; 10: 358Google Scholar; Hysi et al., 2018Hysi P.G. Valdes A.M. Liu F. Furlotte N.A. Evans D.M. Bataille V. et al.Genome-wide association meta-analysis of individuals of European ancestry identifies new loci explaining a substantial fraction of hair color variation and heritability [published correction appears in Nat Genet 2019;51:1190].Nat Genet. 2018; 50: 652-656Google Scholar; Simcoe et al., 2021Simcoe M. Valdes A. Liu F. Furlotte N.A. Evans D.M. Hemani G. et al.Genome-wide association study in almost 195,000 individuals identifies 50 previously unidentified genetic loci for eye color.Sci Adv. 2021; 7eabd1239Google Scholar; Visconti et al., 2018Visconti A. Duffy D.L. Liu F. Zhu G. Wu W. Chen Y. et al.Genome-wide association study in 176,678 Europeans reveals genetic loci for tanning response to sun exposure.Nat Commun. 2018; 9: 1684Google Scholar). A total of 151 SNPs at 13 loci showed nominally significant (P < 0.01) association, but except those in or close to OCA2 and SLC24A5 genes, none survived strict Bonferroni correction of multiple testing (P < 0.05/9,183; Supplementary Table S4). To replicate the finding at 9p21.3, we tested the association between rs10122939 and ITAº in an additional Chinese cohort from Taizhou (Taizhou longitudinal cohort, n = 1,787) and in a Latin American cohort from Colombia (n = 266). The effect of rs10122939 was successfully replicated at nominal significance, and the G allele also showed a skin-darkening effect in both replication cohorts (Taizhou longitudinal cohort: β = –0.09, P = 5.96 × 10−3, Colombian: β = –0.26, P = 3.89 × 10−2). A meta-analysis of all samples further enhanced the significance level of this association (P = 2.13 × 10−13). To further investigate the potential functions of rs10122939, we conducted a function annotation analysis using the quantitative scoring system embedding the 3DSNP database (Lu et al., 2017Lu Y. Quan C. Chen H. Bo X. Zhang C. 3DSNP: a database for linking human noncoding SNPs to their three-dimensional interacting genes.Nucleic Acids Res. 2017; 45: D643-D649Google Scholar), which revealed significant evidence for rs10122939 serving as an enhancer of SLC24A2 (Figure 2c). We further performed the luciferase report assays to experimentally validate this finding. The transcriptional activity of the enhancer containing the rs10122939 ancestral A allele was significantly higher than that of the corresponding derived G allele in the A375 cell line (t-test P = 0.012, Figure 2d). This pattern would predict less SLC24A2 production for the derived G allele carriers. Future studies may consider in vivo gene editing experiments to further investigate the pigmentation effect of SLC24A2. Unlike the well-known skin color genes SLC24A5 and SLC45A2, this SLC family gene SLC24A2 is not expressed in the skin but rather in the brain and adrenal, mainly involved in neuronal activity (Haque and Moon, 2018Haque M.N. Moon I.S. Stigmasterol upregulates immediate early genes and promotes neuronal cytoarchitecture in primary hippocampal neurons as revealed by transcriptome analysis.Phytomedicine. 2018; 46: 164-175Google Scholar; Zhou et al., 2020Zhou X.G. He H. Wang P.J. A critical role for miR-135-a5p-mediated regulation of SLC24A2 in neuropathic pain.Mol Med Rep. 2020; 22: 2115-2122Google Scholar). It has been recently reported that neuronal activity induced by acute stress can drive a rapid and permanent loss of melanocyte stem cells, which leads up to stress-induced hair greying (Zhang et al., 2020Zhang B. Ma S. Rachmin I. He M. Baral P. Choi S. et al.Hyperactivation of sympathetic nerves drives depletion of melanocyte stem cells.Nature. 2020; 577: 676-681Google Scholar). A possible hypothesis could be that SLC24A2 moderates neuronal activity that influences melanocyte stem cells, eventually causing changes in skin color. Alternatively, SLC24A2 may change the innervation of the skin and affect ITAº through altered skin thickness or other relevant characteristics. This hypothesis is particularly attractive because we did observe a positive correlation between ITAº and transepidermal water loss (P = 4.12 × 10–9), a phenotype thought to be strongly associated with skin thickness (Bargo et al., 2013Bargo P.R. Walston S.T. Chu M. Seo I. Kollias N. Non-invasive assessment of tryptophan fluorescence and confocal microscopy provide information on skin barrier repair dynamics beyond TEWL.Exp Dermatol. 2013; 22: 18-23Google Scholar). The rs10122939 SNP is also associated with transepidermal water loss (P = 0.01) (Supplementary Figure S8 and Supplementary Materials and Methods). In conclusion, this is a meaningful skin color GWAS in well-sized Chinese populations. We identified an intron variant of SLC24A2 (rs10122939) as a previously unreported East Asian‒European differentiating polymorphism involved in skin color variation. The underlying cellular mechanism is to be explored. Summary statistics for all analyzed variants for the Jidong Study and the National Survey of Physical Traits can be viewed at NODE under accession number OEP001341 or directly at http://www.biosino.org/node/project/detail/OEP001341 and accessed by submitting a request for data access. Data usage shall be in full compliance with the Regulations on Management of Human Genetic Resources in China. Individual genotype and phenotype data cannot be shared owing to Institutional Review Board restrictions on privacy concerns. All other relevant data supporting the key findings of this study are available within the letter and Supplementary Materials or from the corresponding author on reasonable request. Fudi Wang: http://orcid.org/0000-0002-0208-3343 Qi Luo: http://orcid.org/0000-0002-6988-2051 Yan Chen: http://orcid.org/0000-0002-6873-8648 Yu Liu: http://orcid.org/0000-0001-7110-127X Ke Xu: http://orcid.org/0000-0002-0408-1659 Kaustubh Adhikari: http://orcid.org/0000-0001-5825-4191 Xiyang Cai: http://orcid.org/0000-0002-3868-5426 Jialin Liu: http://orcid.org/0000-0003-4065-5397 Yi Li: http://orcid.org/0000-0002-2448-9176 Xuyang Liu: http://orcid.org/0000-0002-0871-8979 Luis-Miguel Ramirez-Aristeguieta: http://orcid.org/0000-0002-0961-5824 Ziyu Yuan: http://orcid.org/0000-0003-1434-0383 Yong Zhou: http://orcid.org/0000-0001-5221-8026 Fu-feng Li: http://orcid.org/0000-0002-0566-3589 Binghua Jiang: http://orcid.org/0000-0003-4526-2031 Li Jin: http://orcid.org/0000-0001-9201-2321 Andrés Ruiz-Linares: http://orcid.org/0000-0001-8372-1011 Zhaohui Yang: http://orcid.org/0000-0003-0958-4439 Fan Liu: http://orcid.org/0000-0001-9241-8161 Sijia Wang: http://orcid.org/0000-0001-6961-7867 The authors state no conflict of interest. This work has received funding from the Strategic Priority Research Program of the Chinese Academy of Sciences (XDB38020400 and XDB38010400), the National Key Research and Development Project (2018YFC0910403), Shanghai Municipal Science and Technology Major Project (2017SHZDZX01), the National Basic Research Program (2015FY111700), the CAS Interdisciplinary Innovation Team project, Beijing Advanced Discipline Fund, the National Natural Science Foundation of China (81930056, 91651507, 32070579, 31771393, and 31601016), the Open Project of Key Laboratory of Genomic and Precision Medicine of the CAS, State Key Laboratory of Genetic Resources and Evolution grant (GREKF20-13), the Leverhulme Trust (F/07 134/DF), BBSRC (BB/I021213/1), the Excellence Initiative of Aix-Marseille University - A∗MIDEX (a French Investissements d'Avenir programme, 2RUIZLRE/RHRE/ID18HRU201 and 20-07874), the Scientific and Technology Committee of Shanghai Municipality (18490750300), Ministry of Science and Technology of China (2020YFE0201600), and the 111 Project (B13016), Santander Research and Scholarship Award, and Bogue Fellowship from University College London. Correspondences regarding the luciferase assays and population comparative analyses should be addressed to ZY ([email protected]) and FL ([email protected]), respectively. Conceptualization: FW, SW; Data Curation: FW; Formal Analysis: FW, QL, YC, YLiu, XC, JL, YLi, ZY; Funding Acquisition: ZY, FL, SW; Investigation: FW, QL, KX, XL, ZY; Methodology: FW, QL; Project Administration: FW, FL, SW; Resources: LMRA, ZY, FFL, YZ, BJ, LJ, ARL, SW; Software: FW, QL; Supervision: FL, SW; Validation: YC, KA; Visualization: FW, FL, SW; Writing - Original Draft Preparation: FW, FL; Writing - Review and Editing: QL, YC, YLiu, KA, XC, JL, YLi, ZY, SW The Jidong cohort is a community-based, long-term observational cohort study to evaluate health-related risk factors. The baseline data were collected from 2013 to 2014 in the Staff Hospital, Jidong Oilfield Branch, China. Approval was obtained from the Ethics Committee of Kailuan General Hospital of Tangshan City and the Medical Ethics Committee, Staff Hospital, Jidong Oilfield Branch, China National Petroleum Corporation in July 2013 (approval number 2013 YILUNZI1). The approval had been renewed in 2018. To date, 9,078 individuals aged >18 years have been enrolled after excluding individuals who were unable or unwilling to participate. Written informed consent was obtained from all participants. This study included a total of 5,601 individuals (2,512 men and 3,089 women, aged 31–87 years) who paid the return visit in 2018. The facial photograph and blood samples were collected in the Staff Hospital at the same time. The National Survey of Physical Traits cohort (NSPT) is a population-based prospective cohort study to explore the environmental and genetic factors associated with physical traits and diseases. The NSPT cohort study was conducted with the official approval of the Ethics Committees of Fudan University (14117) and the Shanghai Institutes for Biological Sciences (ER-SIBS-261410). The NSPT totally collected samples of 3,565 Han Chinese individuals (1,320 men and 2,245 women, aged 17–83 years) in 2015–2018 from three sites (i.e., Taizhou, Nanning, and Zhengzhou). All individuals provided written informed consent. Portrait photos of 1,930 individuals (705 men and 1,225 women, aged 18–79 years) were collected in accordance with phenotyping standard operating procedure. Therefore, only 1,930 individuals were included in this study. The Taizhou longitudinal cohort study is a long-term observational cohort study to explore the environmental and genetic risk factors for common and noncommunicable diseases. This research program was conducted with the approval of the Ethics Committee of Fudan University (Ethics Research Approval 85), Shanghai, China. The detailed characteristics have been described before (Wang et al., 2009Wang X. Lu M. Qian J. Yang Y. Li S. Lu D. et al.Rationales, design and recruitment of the Taizhou Longitudinal Study.BMC Public Health. 2009; 9: 223Google Scholar). Our replication set included 1,787 healthy Han Chinese with portrait photos, aged 31–85 years. The Colombian cohort collected data from participants of several Latin American countries to study the genetics of normal variation in physical appearance. For the genetic analysis of individual typology angle (ITAº), 266 participants were recruited in the city of Medellin, Colombia. Ethical approval was obtained from the bioethics committee of the Odontology Faculty at the University of Antioquia (Medellin, Colombia) (CONCEPTO 01-2013). The age of the participants was between 18–50 years, with an average of 27 years. All participants were asked not to take part in vigorous exercise an hour before their study visit, not to wear make-up, and to refrain from alcohol and tobacco use 24 hours before the visit. All photographs were taken in a confined space with stabilized light-emitting diode light source. Besides, all participants wore a shawl to help give consistent light illumination. A Canon 70D digital camera (lens: Canon EF 40 mm f/2.8; Canon, Tokyo, Japan) was used for all subjects without the flash. The facial photograph for each participant consisted of a frontal facial shot with the eyes closed and no facial expression. The resolution of the photographs was 300 dpi. This study adopted ITAº as a quantitative measurement of skin color, which was derived from high-resolution portrait photos. ITAº is a quantitative variable designed for measuring skin pigmentation on the basis of colorimetric parameters. ITAº could be used to classify skin types, that is, very light (ITAº > 55), light (41 ∼ 55), intermediate (28 ∼ 41), tan (10 ∼ 28), brown (–30 ∼ 10), and dark (<–30) (Chardon et al., 1991Chardon A. Cretois I. Hourseau C. Skin colour typology and suntanning pathways.Int J Cosmet Sci. 1991; 13: 191-208Google Scholar). It has been shown that the ITAº-based skin color classification is physiologically relevant (Del Bino and Bernerd, 2013Del Bino S. Bernerd F. Variations in skin colour and the biological consequences of ultraviolet radiation exposure.Br J Dermatol. 2013; 169: 33-40Google Scholar). Portrait photos were processed using an in-house developed facial skin color quantification pipeline, which involves a face detector, automated facial landmarking, cheek segmentation, and color analysis. The face detection was achieved by the get_frontal_face_detector function from the C++ library Dlib (King, 2009King D.E. Dlib-ml: A machine learning toolkit.J Mach Learn Res. 2009; 10: 1755-1758Google Scholar). This function returns an object detector that is configured to find human faces that are looking more or less toward the camera, and the detector is composed of a linear classifier combined with classic Histogram of Oriented Gradients features (Dalal and Triggs, 2005Dalal N, Triggs B. Histograms of oriented gradients for human detection. Paper presented at: 2005 IEEE Computer Society Conference on Computer Vision and Pattern Recognition 20–25 June 2005;—San Diego, CA.Google Scholar), an image pyramid, and the sliding window detection scheme (Sullivan and Su, 2014Sullivan V. Su J. One millisecond face alignment with an ensemble of regression trees.in: IEEE Conference on Computer Vision and Pattern Recognition. 2014: 1867-1874Google Scholar). The face landmarking was achieved by the shape_predictor function from the C++ library Dlib on the basis of the face region detected by the get_frontal_face_detector. This predictor was created by training an ensemble of regression trees for face alignment on the iBUG 300-W face landmark dataset (Sagonas et al., 2016Sagonas C. Antonakos E. Tzimiropoulos G. Zafeiriou S. Pantic M. 300 faces in-the-wild challenge: database and results.Image Vis Comput. 2016; 47: 3-18Google Scholar; Sagonas et al., 2013aSagonas C. Tzimiropoulos G, Zafeiriou S, Pantic M. 300 faces in-the-wild challenge: The first facial landmark localization challenge. In: Proceedings of the 2013 IEEE International Conference on Computer Vision Workshops (ICCV-W); 2013a.Google Scholar, Sagonas et al., 2013bSagonas C. Tzimiropoulos G, Zafeiriou S, Pantic M. A semi-automatic methodology for facial landmark annotation. Proceedings of the 2013 IEEE International Conference Computer Vision and Pattern Recognition Workshops (CVPR-W), 5th Workshop on Analysis and Modeling of Faces and Gestures (AMFG 2013); 2013b.Google Scholar). It takes an image of a human face as input and identifies the locations of 68 facial landmarks including, the corners of the mouth and eyes, the tip of the nose, and edges of cheeks (Figure 1a). For each side of the face, the pipeline automatically outlines a rectangle to segment the cheek part according to preselected anatomical landmarks in such a way that the rectangle contains the cheek-surrounding landmarks with the minimal horizontal and vertical distances between the landmarks (Figure 1a). After the segmentation, we manually examined the scope of the rectangles to exclude the confounding margins that incorporate photographing background and human hair. For the right cheek, the CIELAB (ISO, 2007ISO. CIE standard colorimetric observers international organization for standardization. ISO 11664–1:2007 colorimetry – Part 1; 2007.Google Scholar) L, a and b values of all segmented pixels were obtained and converted to ITAº as ITAº =arctanL−50b×180π, and the mean ITAº was calculated to represent the facial skin color of a portrait. Because ITAº values were not subject to the normal distribution both in the Jidong cohort and the NSPT cohort (Supplementary Figure S2a and b), we applied Z-transformed ITAº in further analysis. For quality control purposes, additional measures were obtained in a subgroup of 50 randomly selected photographs. These included accessing skin colorimetric parameters on the right cheek using a spectrophotometer (Skin Colorimeter CL 400, Courage+Khazaka Electronics GmbH, Cologne, Germany) and four skin color types (very light, light, intermediate, and tan) perceived by an investigator according to the Chardon skin color type (Chardon et al., 1991Chardon A. Cretois I. Hourseau C. Skin colour typology and suntanning pathways.Int J Cosmet Sci. 1991; 13: 191-208Google Scholar). Note that dark skin color types were absent in our Chinese sample. Intermethod reliability was evaluated with the Kappa statistic. No significant difference in ITAº was detected between the left and right cheeks (t-test P = 0.13). Chardon's skin color type showed a high degree of concordance with the perceived skin color type as well as with the spectrophotometer values (Supplementary Figure S1b–d). For phenotyping in the Taizhou longitudinal cohort, the photos were taken with a Canon 70D digital camera (lens: Canon EF 40 mm f/2.8). The rest of the procedures (i.e., automated facial landmarking, cheek segmentation, and color analysis) were the same as those used in the discovery cohorts. For the Colombian cohort, ITAº was measured on the forehead. Owing to the smaller sample size, outlier values on either tail of the ITAº were excluded (values ≤ –3 or ≥ 54). Z-transformed ITAº was used for the analysis. Among 1,887 subjects in the NSPT cohort, we also collected the transepidermal water loss measurement. Transepidermal water loss was measured on the cheek using Tewameter TM 300 of Courage+Khazaka Electronics GmbH (median [interquartile range], g/h/m2 = 10.60 [8.20–13.90]). For both the Jidong and NSPT cohorts, genomic DNA was extracted from blood samples using the MagPure Blood DNA KF Kit. All samples were genotyped using the Illumina Infinium Global Screening Array (Illumina, San Diego, CA) consisting of about 710,000 SNPs. We implemented exclusion criteria for quality control using PLINK, version 1.9 (Chang et al., 2015Chang C.C. Chow C.C. Tellier L.C. Vattikuti S. Purcell S.M. Lee J.J. Second-generation PLINK: rising to the challenge of larger and richer datasets.GigaScience. 2015; 4: 7Google Scholar). In detail, we excluded subjects with >5% missing data, the duplicated subjects, and subjects that failed the X-chromosome sex concordance check or had ethnic information incompatible with their genetic information. We excluded SNPs that had >2% missing data, SNPs with a minor allele frequency <1%, and the ones that failed the Hardy‒Weinberg deviation test (P < 1 × 10−5). Imputation was performed using the 1000 Genomes Project phase 3 as the reference panel. The chip genotype data were firstly phased using SHAPEIT3 (O'Connell et al., 2016O'Connell J. Sharp K. Shrine N. Wain L. Hall I. Tobin M. et al.Haplotype estimation for biobank-scale data sets.Nat Genet. 2016; 48: 817-820Google Scholar), and IMPUTE2 was then used to impute genotypes (Howie et al., 2009Howie B.N. Donnelly P. Marchini J. A flexible and accurate genotype imputation method for the next generation of genome-wide association studies.PLoS Genet. 2009; 5e1000529Google Scholar). SNPs with an imputation quality score (INFO) <0.6, minor allele frequency <0.01, or a missing rate >0.01 were eliminated from further analyses. Finally, 8,039,700 SNPs passed quality control and were used for further analyses. For Taizhou longitudinal cohort, blood samples were collected, and DNA was extracted. All samples were genotyped using the Illumina HumanOmniZhongHua-8 chip (Illumina), which interrogates 894,517 SNPs. After quality control with PLINK, version 1.9, the genotype data were phased using SHAPEIT and were imputed using IMPUTE2 with the 1000 Genomes Project phase 3. Blood samples were collected from the participants in the Colombian cohort. DNA was extracted and genotyped on the Illumina Infinium Global Screening Array. After quality control using PLINK, version 1.9, 511,848 SNPs were retained. The genotype data were then phased using SHAPEIT2 and imputed using IMPUTE2, with the 1000 Genomes Project phase 3 as the reference panel. Ge