Genetic adaptation of skin pigmentation in highland Tibetans

Significance

The Qinghai-Tibetan Plateau represents one of the most extreme environments for human habitation. Tibetans are an ideal population for studying genetic adaptation to such environment. Strong ultraviolet (UV) radiation at high altitude imposes a serious selective pressure, which may induce skin pigmentation adaptation of Tibetans. We conducted skin pigmentation phenotyping and genome-wide analysis of Tibetans to understand the underlying mechanism of adaptation to UV radiation. We observe that Tibetans have darker baseline skin color compared with lowland Han Chinese, as well as an improved tanning ability. The adaptive variant of GNPAT promotes melanin synthesis, likely through the interactions of CAT and ACAA1 in peroxisomes with other pigmentation genes, leading to an improved tanning ability in Tibetans for UV protection.

Abstract

Strong ultraviolet (UV) radiation at high altitude imposes a serious selective pressure, which may induce skin pigmentation adaptation of indigenous populations. We conducted skin pigmentation phenotyping and genome-wide analysis of Tibetans in order to understand the underlying mechanism of adaptation to UV radiation. We observe that Tibetans have darker baseline skin color compared with lowland Han Chinese, as well as an improved tanning ability, suggesting a two-level adaptation to boost their melanin production. A genome-wide search for the responsible genes identifies GNPAT showing strong signals of positive selection in Tibetans. An enhancer mutation (rs75356281) located in GNPAT intron 2 is enriched in Tibetans (58%) but rare in other world populations (0 to 18%). The adaptive allele of rs75356281 is associated with darker skin in Tibetans and, under UVB treatment, it displays higher enhancer activities compared with the wild-type allele in in vitro luciferase assays. Transcriptome analyses of gene-edited cells clearly show that with UVB treatment, the adaptive variant of GNPAT promotes melanin synthesis, likely through the interactions of CAT and ACAA1 in peroxisomes with other pigmentation genes, and they act synergistically, leading to an improved tanning ability in Tibetans for UV protection.

The Qinghai-Tibetan Plateau represents one of the most extreme environments for human habitation (1, 2), and strong ultraviolet (UV) radiation serves as one of the most challenging factors on the plateau. According to the average annual UV irradiance (UVA 315 to 400 nm and UVB 280 to 315 nm) from NASA (3), UV irradiance at an elevation of 4,352 m is 15.37 W/m² (Rikaze, Tibetan Autonomous Region, China), much higher than the radiation intensities of lowland regions of the same latitude (e.g., Tongren, southwest China; 7.42 W/m²) and most areas in mainland East Asia (e.g., Dalian, northeastern China; 9.02 W/m²), and even higher than subtropical Southeast Asia (e.g., Ratanakiri, Cambodia; 13.45 W/m²).

There are diverse biological effects of UV radiation on humans (4, 5). The latitude gradient of UV radiation is the major cause of human skin color diversity (6, 7). Under the stimulation of UV radiation, melanin is synthesized within the melanosome in human melanocytes, and the matured melanosomes are transported to keratinocytes, leading to skin pigmentation that can protect humans from UV damage (8, 9). In general, the closer a population to the equator, the darker is their constitutive skin color (baseline color; skin areas not exposed to sunlight such as buttock and underarm) (6). Many studies on the genetic basis of human skin color diversity have been conducted, indicating a complex genetic architecture of baseline skin pigmentation due to different local evolutionary pressures, including Africans living in high–UV radiation regions and Eurasians living in low–UV radiation regions (10–24). Besides the “baseline color,” the color of the exposed skin areas (facultative skin color) is affected by tanning, namely the ability to temporarily produce melanin in response to UV radiation (25, 26). Consequently, the uneven distribution of UV radiation can serve as a strong selective force on human pigmentation genes that regulate melanin production for both constitutive and facultative skin colors.

Previous studies on European populations (27–30) and Mongolians (31) have identified several pigmentation genes involved in tanning ability for protecting the skin from UV damage. However, tanning responses are variable among populations, likely due to the difference in their genetic backgrounds. Given the strong UV radiation at the plateau, we hypothesize that Tibetans may have developed genetic adaptation to high-altitude UV radiation, which calls for a genome-wide search for genes contributing to such adaptation. In the current study, we collected skin color data and DNA samples of Tibetans covering four altitude gradients of the plateau. We observe clear differences in constitutive skin color and facultative skin color between Tibetans and lowland Han Chinese populations. We detect strong signals of positive selection on the pigmentation gene GNPAT in Tibetans. The adaptive allele of the GNPAT variant rs75356281 is significantly associated with the dark skin of Tibetans, and it affects the enhancer activity of GNPAT, leading to an increased tanning ability under UV radiation.

Results

Skin Color Measurement of Tibetans and Comparison among World Populations.

We measured both constitutive and facultative skin color (L*, level of darkness) of 1,243 indigenous Tibetans covering three altitude populations at 2,212, 4,352, and 5,021 m in the Tibetan Autonomous Region, China (SI Appendix, Fig. S1). In order to compare with the published skin color data of world populations, we converted L* to M (melanin index) (22). A higher M value denotes darker skin. Among the Tibetan samples, 410 samples were from 2,212 m (Linzhi, LYG; UV irradiance 12.30 W/m²), 595 samples were from 4,352 m (Rikaze, RKM; UV irradiance 15.37 W/m²), and 238 samples were from 5,021 m (Shan’nan, LKZ; UV irradiance 15.24 W/m²). We compared the constitutive skin color and facultative skin color among Tibetans, Cambodian aborigines (CAM, 818 samples; UV irradiance 13.45 W/m², altitude 314 m), northern Han Chinese (CHN, 359 samples; UV irradiance 9.02 W/m², altitude 19 m) (20), Africans (AFR-Gha, 237 samples; UV irradiance 11.75 W/m², altitude 270 m) (24, 32), and Europeans (EUR-Pol, 45 samples; UV irradiance 6.97 W/m², altitude 87 m) (24, 33).

We find that there is a significant skin color difference between Tibetan and other world populations (Fig. 1). The constitutive skin color (underarm) of Tibetans is darker than lowland CHN and EUR-Pol but lighter than CAM and AFR-Gha (Fig. 1A and Dataset S1, Table 1). Notably, although the annual UV radiation on the Qinghai-Tibetan Plateau is stronger than that on mainland Southeast Asia, the constitutive skin color of Tibetans is lighter than CAM living in Southeast Asia. By contrast, the facultative skin color of Tibetans is darker than CAM (Dataset S1, Table 1). This pattern suggests that besides the relatively darker baseline color, Tibetans might have developed an enhanced tanning ability to cope with strong UV radiation at high altitude. Importantly, a closer look at the Tibetan populations living at different altitudes reveals gradients of both constitutive (underarm and buttock) and facultative (back of the hand) skin colors; that is, the higher the altitude, the darker the skin (Fig. 1 B–D and Dataset S1, Table 1). This pattern suggests that the strength of UV radiation indeed serves as a strong selective force affecting skin pigmentation in Tibetans. Consistently, the changes of M values (darkness) from underarm or buttock to hand for Tibetans (ΔM underarm–hand = 22.25, ΔM buttock–hand = 17.17) are significantly larger than those for Han Chinese (ΔM underarm–hand = 10.65, ΔM buttock–hand = 9.84) (P underarm–hand = 2.53 × 10⁻²¹⁰ and P buttock–hand = 6.50 × 10⁻¹⁴⁰, t test) (Fig. 1 B–D and Dataset S1, Table 1). In particular, for the facultative skin color, the Tibetan populations living above 4,352 m are even darker than CAM (Fig. 1 B–D and Dataset S1, Table 1), an indication of a two-level UV defense, namely an enhanced tanning ability on top of the dark baseline color.

Fig. 1.

Skin pigmentation comparison among different populations. (A) Comparison of skin darkness among world populations. The y axis indicates the M value of underarm and the error bars indicate SD. The M values of Africans-Ghanaian (AFR-Gha), Europeans-Poland (EUR-Pol), Cambodians (CAM), and Han Chinese (CHN) are from previous reports (20, 24, 32, 33), and the M value of Tibetans (TBN) is collected in this study. (B–D) Comparison of skin darkness (hand, underarm, and buttock) among Tibetans from different altitudes. TBN-LYG: 2,212 m; TBN-RKM: 4,352 m; TBN-LKZ: 5,021 m. (E) Comparison of the constitutive and facultative skin color of TBN and CHN at similar altitudes (three different locations in Qinghai Province of China, 3,712 ± 64 m). ns, not significant, P > 0.05; *P < 0.05, **P < 0.01, ***P < 0.001.

To further test the tanning ability of Tibetans, we measured and compared both constitutive and facultative skin color of 461 Tibetans and 126 Han Chinese permanently living at the same altitude on the Qinghai-Tibetan Plateau (Qinghai Province of China, altitude 3,712 ± 64 m) (Fig. 1E and Dataset S1, Table 1). These volunteers are employees with a similar working schedule, and their exposure to outdoor UV radiation is presumably similar. In addition to the difference in constitutive skin color (P-underarm = 1.03 × 10⁻¹⁵ and P-buttock = 9.29 × 10⁻¹⁷, t test), we detect an increased M value difference between Tibetans and Han Chinese for the facultative skin color (P-hand = 9.53 × 10⁻²⁶, t test). The M value differences are 7.69 and 11.41% for unexposed skin areas (underarm and buttock, respectively), while it is 13.31% for the exposed skin area (hand), suggesting a stronger tanning ability of Tibetans compared with highland-residing Han Chinese. To rule out the potential influence of sunscreen use, we only included males who did not wear sunscreen according to our survey, and the pattern remained the same (SI Appendix, Fig. S2).

Collectively, the results suggest that Tibetans may have developed a two-level genetic adaptation to boost their melanin production in order to adapt to the strong UV radiation on the plateau.

Genome-Wide Detection of Selective Signals in Pigmentation Genes of Tibetans.

In order to search for selective signals in the genome for adaptation to UV radiation, we looked for pigmentation genes showing extraordinarily large allelic divergence between dark-skinned Tibetans and light-skinned Han Chinese (measured by F_ST) (34) (Dataset S1, Table 2). Published genome sequences of Tibetans and Han Chinese were used (35). To identify genes responsible for Tibetan skin color, we overlapped the genome-wide top 0.1% divergent (F_ST > 0.155) gene regions with the 171 known pigmentation genes (36), and we observed 13 pigmentation genes (GNPAT, OCA2, EDAR, VSX2, PTS, MREG, ADAMTS20, ATOX1, GGT1, MGRN1, NTRK1, PDGFC, and RB1) showing large divergence between Tibetans and Han Chinese (Fig. 2A). Among the 13 pigmentation genes, OCA2 and EDAR are previously reported genes showing a strong selection in Han Chinese (20, 37). In other words, the observed large allelic divergences of these two genes between Tibetans and Han Chinese are due to enriched adaptive variants in Han Chinese. Therefore, we excluded these two genes from further analysis.

Fig. 2.

Genome-wide detection of selective signals among pigmentation genes. (A) Genome-wide distribution of F_ST between Tibetans and Han Chinese. The pigmentation genes containing the top 0.1% F_ST sequence variants are indicated by red color. The EPAS1 gene with known strong selection in Tibetans is also indicated. The local divergence pattern of the genomic region containing GNPAT and EGLN1 is shown (Inset). The colored dots indicate SNPs in the coding regions and the 2-kb noncoding upstream and downstream regions of the two genes. The two GNPAT variants and the EGLN1 variant showing the largest F_ST values are indicated. (B) The LD map of GNPAT in Tibetans. The red color indicates strong linkage. The H3K4me1 signals (enhancer mark) of the GNPAT gene region in two skin tissues are from the Roadmap database. Sliding window analysis of the neutrality test using Fay and Wu’s H statistic is shown (Bottom). Africans (YRI) are used as an outgroup. A sliding window is generated by DnaSP, the window length is five points, and the step size is two points. (C) The EHH plots of the core haplotype covering the two variants (rs75356281 and rs75768180) of GNPAT and rs186996510 of EGLN1 in Tibetans. The figure shows haplotype decay of the core variants rs75356281, rs75768180, and rs186996510. The red lines indicate the derived allele haplotype and the black lines indicate the ancestral allele haplotype.

To further validate the observed genome-wide signals of selection, we then tested for evidence of recent positive selection in each of these 11 genes using Fay and Wu’s H test, a method for detecting recent positive selection (38). Since the detected 0.1% sequence variants are all located within the gene regions, we extracted gene region (exons and introns) sequence data from Tibetan highlanders (38 samples) (35) and other reference populations including Han Chinese (CHB, 97 samples), western Europeans (CEU, 85 samples), and Africans (YRI, 88 samples) from the 1000 Genomes Project (39). The sequence data cover GNPAT (38.4 kb, 70 single-nucleotide polymorphisms [SNPs]), VSX2 (23.27 kb, 104 SNPs), PTS (43.59 kb, 130 SNPs), MREG (71.03 kb, 307 SNPs), ADAMTS20 (197.02 kb, 757 SNPs), ATOX1 (12.69 kb, 30 SNPs), GGT1 (40.94 kb, 203 SNPs), MGRN1 (65.65 kb, 260 SNPs), NTRK1 (20.08 kb, 70 SNPs), PDGFC (208.75 kb, 468 SNPs), and RB1 (179.16 kb, 418 SNPs).

We calculated this statistic for Tibetans, CHB, and CEU populations separately using YRI as an outgroup for each. If GNPAT was subject to recent positive selection to enhance facultative skin pigmentation among Tibetans, we would expect to find an excess of high-frequency derived alleles in this population but not in the other populations considered. Three pigmentation genes show Tibetan-specific selective signals, including GNPAT, PTS, and GGT1 (Fig. 2B and SI Appendix, Fig. S3). The selective strengths of PTS and GGT1 are relatively weak based on the sliding-window patterns of Fay and Wu’s H test (gene-wide H score = −17.28, P = 0.066 for PTS, and H score = −23.01, P = 0.086 for GGT1). By contrast, we observe a significant deviation of high-frequency derived alleles from neutral expectation (H score = −26.97, P = 0.035) of GNPAT, suggesting a strong signal of positive selection in Tibetans (Fig. 2B), while no selection on GNPAT was detected in Han Chinese (H score = −7.99, P = 0.183) or Europeans (H score = −3.33, P = 0.305).

The selection intensity of GNPAT is rather strong in Tibetans; the highest F_ST (Tibetans vs. Han Chinese) is 0.439 for two GNPAT sequence variants (rs75768180 and rs75356281) (Dataset S1, Table 2), and it is only weaker than the well-documented hypoxic genes EPAS1 and EGLN1 (40–47) (Fig. 2A). It should be noted that the genomic location of GNPAT is close to the hypoxic gene EGLN1 (85.78 kb apart) with a known strong selection in Tibetans (41, 42) (SI Appendix, Fig. S4). A zoomed-in view of the local genomic region containing both GNPAT and EGLN1 indicates separate F_ST peaks for these two genes, suggesting independent signals of selection (Fig. 2A). To further rule out the possibility of genetic hitchhiking, we performed the haplotype-based extended haplotype homozygosity (EHH) test to detect extended homozygosity surrounding the variant under positive selection (48). The two intronic GNPAT variants (rs75356281 and rs75768180, 8.65 kb apart) with the largest allelic divergence between Tibetans and Han Chinese (F_ST = 0.439) are in complete linkage disequilibrium (LD) (D′ = 1.0) (Fig. 2B and Dataset S1, Table 2). Taking rs75356281 as an example, the derived allele T is highly prevalent in Tibetan populations (58%) but relatively rare in East Asians (18%), South Asians (15%), Europeans (3%), and Americans (4%) and absent in Africans, suggesting a Tibetan-specific enrichment due to positive selection. Consistently, we observe selection signals in Tibetans when putting rs75356281 or rs75768180 as the core variant in the EHH test (Fig. 2C). Although there are also selection signals for the known core variant of EGLN1 (rs186996510), its extended haplotype homozygosity does not affect the adjacent GNPAT (Fig. 2C).

GNPAT encodes an enzyme located in peroxisomes and many enzymes in peroxisomes participate in the processes of H₂O₂ decomposition, reactive oxygen species (ROS) removal, and UV protection (Gene Ontology: 0009650), which are essential to the synthesis of ether phospholipids, such as plasmalogen. Mutations in this gene are reportedly associated with cataracts, iris pigmentation, and rhizomelic chondrodysplasia punctata (49–51). Since GNPAT shows the strongest signal of selection in Tibetans and its function is closely related to pigmentation, our following analyses are focused on GNPAT.

Genetic Association of the GNPAT Variants with Dark Skin in Tibetans.

Presumably, the observed strong signal of selection on GNPAT is expected to affect skin pigmentation. Hence, we first conducted a genetic association analysis of the GNPAT variants with skin color in Tibetans, including 388 LYG individuals (2,212 m) and 593 RKM individuals (4,352 m), from whom we were able to obtain blood samples. The GNPAT variant (rs75356281) with the largest F_ST (Tibetans vs. Han Chinese) was genotyped in these Tibetans. Age, sex, and altitude were taken as covariates in the association analyses.

The GNPAT variant rs75356281 is significantly associated with hand skin color (additive model, P = 0.011, BETA = 0.68) and buttock skin color (additive model, P = 0.032, BETA = 0.59). The presumably adaptive allele (T) of rs75356281 contributes to darker skin color (Fig. 3A). The same trend is detected for underarm skin color in Tibetans though statistically not significant (P = 0.128, BETA = 0.33). To rule out the potential influence of population substructure, we carried out separate association analyses of the two populations and the results remained similar for RKM (additive model, P-hand = 0.024, P-underarm = 0.125, and P-buttock = 0.043), but became not significant for LYG (additive model, P-hand = 0.097, P-underarm = 0.410, and P-buttock = 0.085) due to the decreased sample size. Furthermore, we carried out meta-analysis using the weighted z-score method in METAL (52), which accounts for population stratification. The P values remained significant (additive model, P-hand = 0.005, P-underarm = 0.088, and P-buttock = 0.007). With the use of GEMMA software (53), we estimated the contribution of rs75356281 to skin color. In Tibetans, rs75356281 can explain 1.43% of the total M variance for hand, 0.87% for underarm, and 1.01% for buttock. For hand, it confers 4.74% of the M value difference (skin darkness difference) between Tibetans and Han Chinese.

Fig. 3.

Genetic association analysis and functional tests of the GNPAT adaptive variant. (A) Measurements of darkness (M value) of three skin areas with different genotypes at rs75356281. The y axis represents the M value, and the x axis shows three rs75356281 genotypes. The error bars indicate SD and the beta values are shown in parentheses. The additive genetic model is used for the association analysis. Age, sex, and altitude are considered as covariates. (B) The result of the dual-luciferase reporter assay for rs75356281. All assays were performed in at least three replicates. The presented activity values of the tested groups (the C allele and the T allele) were normalized by the controls (empty vectors). The P values are calculated by t test. (C) The results of EMSA. Lanes 1 and 2: negative controls without nuclear extracts; lanes 3 and 6: probes containing the C allele and the T allele, respectively, can both bind proteins from the nuclear extracts, but the binding affinity is stronger for the T allele (lane 6); lanes 4, 5, 7, and 8: relative binding strength tested by the competition assay using the unlabeled probes containing the T allele (lanes 5 and 7) or the C allele (lanes 4 and 8). (D) The sequence motif containing rs75356281 that can bind the transcription factor FOXP3. The size of the letter indicates binding strength and the asterisk indicates the site of rs75356281.

In Vitro Functional Tests of the GNPAT Adaptive Allele.

To understand the functional role of the GNPAT variants in skin pigmentation of Tibetans, we first acquired published epigenetic data of the GNPAT gene region indicating signals of regulator elements such as enhancers. We obtained the H3K4me1 data (enhancer mark) of eight skin tissues from the Roadmap database (54). The variant rs75356281 falls within the H3K4me1 signals in keratinocyte and melanocyte cells derived from foreskin tissue (Fig. 2B), suggesting that it may affect enhancer activity. Previous studies have shown that human melanoma A375 is a suitable cell line for studying melanin synthesis when melanocyte cells are unavailable. It was shown that the A375 cells can produce melanin and are responsive to UV radiation treatment (55, 56). We performed dual-luciferase reporter experiments (enhancer assay) using A375 under both normal conditions and UVB induction. Since the two variants (rs75356281 and rs75768180) with the highest F_ST value are in complete LD, we tested both variants in the enhancer assay.

Under normal conditions (without UV treatment), both alleles (T and C) of rs75356281 show significantly increased luciferase signals compared with the control, an indication of enhancer activity (SI Appendix, Fig. S5 and Dataset S1, Table 3). The key regulatory difference between the T allele (derived) and the C allele (ancestral) is their differential responses to UV radiation. The T allele showed a significant UV response while the C allele was basically not responsive to UV treatment (Fig. 3B). Consistently, after UV treatment, the increased percentage of luciferase signal of the T allele (12.92%) was much larger than that of the C allele (2.18%) (Fig. 3B and Dataset S1, Table 3), suggesting that the adaptive T allele is more responsive to UV radiation than the wild-type C allele. In contrast, with the same assay, neither a clear enhancer activity nor a UV induction pattern was observed for rs75768180 (SI Appendix, Fig. S6 and Dataset S1, Table 3), suggesting that the high F_ST value of rs75768180 is likely caused by genetic hitchhiking due to its complete LD with rs75356281.

The change of enhancer activity is usually caused by the binding-affinity alteration due to mutations located within the enhancer-binding motif of transcription factors. To test this, we carried out the transcription factor–binding assay (electrophoretic mobility-shift assay; EMSA) using nuclear extracts from A375 cells, and the labeled double-stranded oligonucleotides contain either the C allele (wild-type) or the T allele (adaptive) of rs75356281. The EMSA results show that the T allele has a stronger binding affinity compared with the C allele (Fig. 3C, lanes 3 and 6). The addition of a 100× molar excess of unlabeled probes with the same alleles could compete for protein binding, indicating that the binding pattern is specific to the motif containing rs75356281 (Fig. 3C, lanes 4 and 7).

With the use of the JASPAR database (57), we predicted that the transcription factors FOXP3, FOXL1, and SOX18 may bind to the sequence motif containing rs75356281. Fig. 3D shows that rs75356281 is located in one of the conserved sites of the FOXP3-binding motif, while no such pattern is seen for FOXL1 and SOX18 (SI Appendix, Fig. S7). It was reported that overexpression of FOXP3 could increase pigmentation and FOXP3 plays an important role in melanin synthesis (58).

Hence, the in vitro functional tests demonstrate that the Tibetan-enriched T allele (presumably the adaptive allele) of rs75356281 has a better response to UV radiation compared with the wild-type allele, and it can increase the binding affinity of the transcription factor FOXP3 and eventually enhance melanin production under UV radiation.

Transcriptome Analysis Reveals the Functional Role of GNPAT in the Pigmentation Pathway.

To further test the functional effect of rs75356281 in the pigmentation pathway, we generated three gene-edited cell lines using A375 cells (rs75356281: CC), including two homozygous cell lines containing the adaptive allele of rs75356281 (TT-1 and TT-2) and one heterozygous cell line of rs75356281 (CT) using the CRISPR-Cas9 gene-editing system (59, 60) (SI Appendix, Fig. S8). We detected one off-target site (point 4, chromosome X: 107,413,173; GRch 38) in the TT-1 cell line, which is located in the nongenic region and theoretically does not affect the pigmentation pathway (SI Appendix, Fig. S8). No off-target was detected in the other two cell lines (TT-2 and CT).

We simultaneously performed cell culture under normal conditions (37 °C and 5% CO₂) and UV radiation (37 °C and 5% CO₂ with a UVB lamp; Tanon), and generated three replicates for each of the four cell lines (CC, CT, TT-1, and TT-2) (see Materials and Methods for details). For RNA extraction, we obtained two aliquots for each cell line, and a total of 24 samples (6 samples for each cell line) were subjected to RNA sequencing (RNA-seq). No batch effect was detected (SI Appendix, Fig. S9).

Using the transcriptome data (RNA-seq), we first conducted a principal-component analysis (PCA). On the PCA map, under normal conditions, the cells with the same genotype background cluster together though the variances are quite large, and the cells containing at least one adaptive allele (CT, TT-1, and TT-2) cannot be entirely separated from the wild-type homozygotes (CC) (Fig. 4A). By contrast, under UV treatment, the cells containing at least one adaptive allele (CT, TT-1, and TT-2) clustered tightly together and separated from the wild-type homozygotes (Fig. 4A). This pattern strongly suggests that the GNPAT adaptive allele is highly responsive to UV radiation. Furthermore, under UV treatment, the number of shared differentially expressed genes (CT/TT-1/TT-2 vs. CC) is 1,399, much larger than that under normal conditions (74 genes), again indicating a UV-dependent alteration of transcriptional regulation caused by the adaptive allele (Fig. 4B and Dataset S1, Table 4). We also checked the p53 pathway given A375 is a melanoma cell line. We compared the transcriptomic differences between the gene-edited cells (rs75356281: TT or CT) and the control cells (rs75356281: CC), and we did not see enrichment of the p53 pathway genes, suggesting that it does not affect the observed patterns.

Fig. 4.

Transcriptome analysis of gene-edited melanoma cells with different GNPAT genotypes. (A) The PCA map showing expression divergence among samples with different GNPAT genotypes under either normal culture conditions or UVB induction. (B) Overlap of differentially expressed genes among the melanoma cells with different GNPAT genotypes (CC: wild-type homozygote; CT: heterozygote; TT: adaptive homozygote). (C) The expression level (TPM) of GNPAT under normal culture conditions and UVB induction. The error bars indicate SD. (D) The protein–protein interaction network among GNPAT and other differentially expressed pigmentation genes. The red-color labeled genes are up-regulated and the green-color labeled genes are down-regulated under UVB induction. The size of the circle indicates connectivity degree.

We examined the expression of GNPAT. For the wild-type homozygotes, no change was detected between normal and UV conditions (Fig. 4C), while increased expression was seen for the CT, TT-1, and TT-2 cells (although the P value of TT-2 was not significant), suggesting that the adaptive allele can significantly up-regulate GNPAT under UV radiation. This result is consistent with the data of the enhancer and EMSAs (Fig. 3 B and C), in which the T allele is more responsive to UV treatment and shows a stronger binding affinity to transcription factors than the wild-type allele.

Moreover, KEGG (Kyoto Encyclopedia of Genes and Genomes) pathway analysis demonstrated that the 1,399 differentially expressed genes (under UV treatment) are mainly concentrated in the p53 signaling pathway, the MAPK signaling pathway, and the cell-adhesion molecule pathway (SI Appendix, Fig. S10 and Dataset S1, Table 5). It is known that most of the genes in the p53 pathway are related to UV radiation, and the MAPK pathway is also closely related to melanogenesis (KEGG pathway: map04916). Overlapping these gene regions with 171 known pigmentation genes from the Color Genes database identified 68 pigmentation genes with altered expression (Dataset S1, Table 6) (36).

We next constructed protein–protein interaction networks between GNPAT and the 68 differentially expressed pigmentation genes for the adaptive allele homozygous cell lines (TT-1 and TT-2) (Fig. 4D and SI Appendix, Fig. S11) using STRING (61) (see Materials and Methods for details). Taking TT-1 as an example, the network contains 48 differentially expressed pigmentation genes including GNPAT, among which 9 genes directly interact with GNPAT (CAT, ACAA1, ACOX2, ACOX3, LPIN3, EPHX2, AMACR, DGKH, and DGKI) and 8 of them show the same direction of expression change (up-regulation) (Fig. 4D). The other 38 pigmentation genes show indirect interactions with GNPAT with either down-regulated (18 genes) or up-regulated (20 genes) expression. A similar network is also seen for the TT-2 cell line (SI Appendix, Fig. S11).

CAT and ACAA1 are bridges between GNPAT and the subnetworks containing many other pigmentation genes (Fig. 4D and SI Appendix, Fig. S11). In particular, CAT encodes catalase. Overexpression of CAT can decrease sunburned cell formation under UVB irradiation in the human reconstructed epidermis and catalase-specific messenger RNA (mRNA), protein, and enzymatic activity were all directly correlated with total cellular melanin content in human melanocytes (62, 63). Therefore, GNPAT likely serves as one of the hubs in the melanogenesis pathway and promotes the synthesis of melanin and catalase activity that are likely to act synergistically to participate in the UV protection process.

Collectively, based on the results of the functional assays and the transcriptome data, we show that the adaptive allele of GNPAT is more responsive to UV radiation with an up-regulated expression under UV treatment. This can affect the expression of many other pigmentation genes, eventually contributing to a stronger tanning ability in Tibetans.

Discussion

The average altitude of the Qinghai-Tibetan Plateau is more than 4,000 m. The indigenous highland peoples (mainly Tibetans) are successful examples of human adaptation to extreme environments, especially to high-altitude hypoxic adaptation (35, 40–47, 64–74). In addition to hypoxia, strong UV radiation also serves as one of the extreme environmental factors on the plateau. However, previous studies on human skin color diversity have been focusing on lowland populations (10–22), and how Tibetans adapt to high-altitude UV radiation remains largely unknown.

We show that the constitutive skin color of Tibetans is darker than that of lowland Han Chinese, suggesting that Tibetans have developed a protective mechanism to cope with strong UV radiation at altitude, a similar strategy for populations living near the equator such as Africans and Southeast Asian aborigines. Intriguingly, the constitutive skin color of Tibetans is lighter than the Southeast Asian aborigines (Cambodians) though the intensity of UV radiation on the Qinghai-Tibetan Plateau is stronger than that in mainland Southeast Asia. This suggests that besides the darker constitutive skin color, Tibetans might have developed a stronger tanning ability for further protection, and this is confirmed by our observation of darker facultative skin color (skin areas exposed to sunlight) of Tibetans compared with the Southeast Asian aborigines (Fig. 1). Hence, there should be enriched (adaptive) genetic variants in Tibetans contributing to both mechanisms of UV protection.

Through genome-wide comparison of sequence variants between highland Tibetans and lowland Han Chinese, we found 13 pigmentation genes showing large allelic divergence between Tibetans and Han Chinese (Fig. 2A). Among these genes, three genes (GNPAT, PTS, and GGT1) present signals of selection in Tibetans, and GNPAT has the strongest signal. The selection on GNPAT is further supported by the separate F_ST peak, the haplotype-based EHH test, and the Fay and Wu's H test (Fig. 2). The allelic divergence of the two GNPAT variants (rs75356281 and rs75768180) between Tibetans and Han Chinese reaches 0.439, an implication of strong positive selection, just slightly weaker than the known selection on EPAS1 and EGLN1 for high-altitude hypoxic adaptation (40–47). It should be noted that although we present evidence of independent signals of selection for the GNPAT variants, we cannot completely rule out the possibility of genetic hitchhiking by the adjacent gene EGLN1, and selection might favor both increased tanning response and adaptation to high-altitude hypoxia.

Both archaeological and genetic data suggest a Paleolithic settlement of the ancestors of Tibetans on the Qinghai-Tibetan Plateau (75, 76), an indication of a long duration of natural selection on Tibetan populations. Given hypoxia and strong UV radiation coexist at high altitude, the selection time of UV on the pigmentation genes (such as GNPAT) might be similar to that of hypoxia on the hypoxic genes (such as EPAS1) because all the identified adaptive variants are standing ones instead of de novo mutations. In addition, the selective intensity of UV is likely less than that of hypoxia, the condition that cannot be overtaken by traditional means.

Our genetic association analysis indicates that rs75356281 (in complete LD with rs75768180) is significantly associated with facultative skin color (back of the hand), suggesting its contribution to the skin tanning of Tibetans. It should be noted that the color of the back of the hand is just a rough reflection of tanning ability though we did observe a significant association (P = 0.011), and the contribution of rs75356281 to skin darkness difference between Tibetans and Han Chinese is relatively small. Further functional prediction based on the Roadmap data (54) shows that rs75356281 is located within a GNPAT enhancer (the H3K4me1 signal). The functional experiments demonstrate that under UVB radiation, the derived T allele (major allele, presumably the adaptive allele) of rs75356281 displays significantly higher enhancer activities compared with the wild-type C allele, likely caused by the observed stronger binding affinity of the T allele to transcription factors (such as FOXP3). Based on the evidence, we speculate that the major role of GNPAT for UV protection is to improve the tanning ability of Tibetans.

GNPAT may also play a role in constitutive pigmentation because we saw a significant association of the GNPAT adaptive allele with dark skin in Tibetans (Fig. 3A), though this observation did not get support from the results of the enhancer assay where, under normal conditions (without UV induction), we saw a reduced enhancer activity of the adaptive allele (Fig. 3B). However, this cannot rule out the possibility that there are different regulatory mechanisms for melanin production with or without UV radiation. In other words, GNPAT may play double roles, namely darkening the constitutive pigmentation and improving tanning under UV radiation, and different molecular components might be involved. It should also be noted that there are differences of constitutive skin pigmentation (underarm and buttock) among Tibetans living at different altitudes, which may reflect either adaptation to varied intensities of UV radiation or potential population substructure in Tibetans. To solve this issue, ethnographic data (e.g., the time duration of Tibetans at specific altitudes) and population genomic data need to be collected in future studies.

The data for functional validation was generated using the A375 cell line, a commonly used experimental material for testing the pigmentation pathway (55, 56, 77). Although it may not fully simulate the state of human melanocytes, the results of the dual-luciferase reporter assay and the RNA-seq data both confirm that GNPAT is a key gene in response to UV treatment. GNPAT encodes an enzyme located in the peroxisomal membrane and many enzymes in the peroxisome participate in the process of UV protection. With UV treatment, melanoma cells display dramatic expression changes between the GNPAT adaptive allele and the wild-type allele, involving 1,399 genes, and 68 of them are pigmentation genes, strong evidence for the functional effect of the GNPAT adaptive variant that promotes melanin synthesis under UV radiation, likely through the interactions of CAT and ACAA1 in peroxisomes with other pigmentation genes. Previous studies have shown that CAT participates in the process of UV protection. Under UV radiation, there is excessive ROS in cells. High levels of ROS exert a toxic effect on biomolecules such as DNA and proteins (78). CAT encodes catalase that plays a key role in protecting cells, and dismutates H₂O₂ to water and oxygen against oxidative stress. Catalase-specific mRNA, protein, and enzymatic activity were all reported to be directly correlated with total cellular melanin content in human melanocytes (63).

Finally, although GNPAT is likely the key gene for high-altitude UV adaptation in Tibetans based on the signal of selection as well as the functional data, there are other pigmentation genes (such as PTS and GGT1) with selective signals and they may work together with GNPAT to darken the constitutive skin color and enhance the tanning ability of Tibetans for UV adaptation on the Qinghai-Tibetan Plateau.

Materials and Methods

Sample Collection, Genomic DNA Preparation, and Skin Color Measurement.

We collected human blood samples and measured the skin color of Tibetans from the Tibetan Autonomous Region and Qinghai Province of China, including 410 Tibetans at 2,042 m (206 males and 204 females, Linzhi, LYG); 461 Tibetans (307 males and 154 females) and 126 Han Chinese (78 males and 48 females) from Qinghai (3,712 ± 65 m; Jiuzhi, Maqing, and Banma); 595 Tibetans at 4,352 m (245 males, 336 females, and 14 unknown sex, Rikaze, RKM), and 238 Tibetans at 5,021 m (99 males and 139 females, Shannan, LKZ). Among the 1,830 skin color–measured individuals, 971 of them (LYG and RKM) were subjected to genomic DNA extraction using the standard phenol-chloroform method (79). Written informed consent was obtained from each individual before their inclusion in the study. All protocols of this study were approved by the Institutional Review Board of the Kunming Institute of Zoology, Chinese Academy of Sciences (approval IDs: SWYX-20100110-02 and SMKX-20160311-45).

We used a CR-400 tristimulus colorimeter (Konica Minolta) to measure skin color. The colorimeter generates three numerical readings: L*, a*, and b*. We measured the L* value (level of darkness) and converted it into the M value in order to compare with the published M values of other world populations (22). Both constitutive skin color from skin areas not exposed to sunlight (underarm and buttock) and facultative skin color that is exposed to sunlight (back of the hand) were measured for comparisons. To minimize technical variations, each skin area was measured three times, and the average value was used for analysis.

Solar Radiation Data Collection.

The geographic locations of the sampling points and the average annual UV irradiance (UVA 315 to 400 nm and UVB 280 to 315 nm) were obtained from NASA (3). The insolation surface data cover 20 y (2001 to 2020).

Detection of Positive Selection on Pigmentation Genes in Tibetans.

The published genome sequence data of Tibetans contain 33 Tibetans and 5 Sherpa individuals living in six prefectures (Lhasa, Chamdo, Nagqu, Nyingchi, Shannan, and Shigatse) of the Tibetan Autonomous Region of China (35). We detected signatures of natural selection primarily based on the population genetic statistic F_ST (34). To identify the genes responsible for skin color, we looked for genes showing extraordinarily large divergence (top 0.1% F_ST), and overlapped the signal gene regions with 171 known pigmentation genes of the human genome (36).

Haploview software (D′ algorithm) (80) was used to analyze LD patterns of GNPAT in different populations. FastPHASE was used to reconstruct haplotypes (81). Using the program DnaSP v5 (82), we carried out tests of neutrality following the method developed by Fay and Wu (38). The African population (YRI, Yoruban individuals) was used as the outgroup, and a random YRI individual was selected for calculation. We also used the EHH test described by Sabeti et al. for detecting recent positive selection with an incomplete selective sweep, that is, the selected allele has not reached fixation (48).

Genotyping of rs75356281 and Genetic Association Analysis.

Based on the results of selection testing and evolutionary analysis, we picked rs75356281 for genetic association analysis with skin color in 971 Tibetan individuals (LYG and RKM). Genotyping was conducted by using an ABI 3730 sequencer (Applied Biosystems). Genetic association of the tag variant rs75356281 with M values (back of the hand, underarm, and buttock) was conducted by utilizing PLINK v1.07 (83) with age, sex, and altitude taken as covariates. We also performed a meta-analysis using the weighted z-score method in METAL that accounts for population stratification (52). GEMMA was used to calculate the percentage of the total skin color variance that can be explained by rs75356281 (53). We estimated the contribution of the GNPAT variant (rs75356281) to skin darkness difference between Han Chinese and Tibetans using the ExpDiff method (84). In this method, β represents the effect of the derived allele of rs75356281 on pigmentation, which is the β-value in the association analysis. ΔDAF is the frequency difference of the rs75356281 derived allele in Han Chinese and Tibetans. We obtained frequency data of Han Chinese from the 1000 Genomes Project and our Tibetan genotype data in this study. ΔM describes skin pigmentation differences, which is the mean M value of Tibetans minus the mean M value of Han Chinese.

Functional Test Using a Dual-Luciferase Reporter Assay.

We obtained GNPAT gene region locus H3K4me1 signals (enhancer mark) in two skin tissues from the Roadmap database (54). Graphics were generated using the Gviz package of R software.

To construct GNPAT enhancer reporters, we amplified fragments of 430 and 356 bp of GNPAT by PCR from genomic DNA of two individuals homozygous with respect to the corresponding genotypes of rs75356281 (CC and TT) and rs75768180 (AA and GG), using primers tailed with KpnI and XhoI restriction sites for rs75768180 and rs75356281, respectively, and directionally subcloned into the pGL3-promoter expression vector (Promega). The recombinant clones were verified by Sanger sequencing.

A375 (human malignant melanoma) cell lines were obtained from the Cell Bank Culture Collection of the Kunming Institute of Zoology, Chinese Academy of Sciences and cultured in Dulbecco’s modified Eagle’s medium (DMEM) (Corning). All culture media were supplemented with 10% fetal bovine serum (FBS; Gibco). Cells were grown in a humidified incubator at 37 °C under an atmosphere of 5% CO₂ and subcultured with 0.25% trypsin-ethylenediaminetetraacetate (EDTA) (Invitrogen) when ∼80% confluence was reached. The 5 × 10⁴ A375 cells cultured in 24-well plates were transfected with 0.5 μg recombinant plasmids constructed as described above using Lipofectamine 2000 and Opti-MEM (Invitrogen) according to the recommended protocols of the manufacturer. According to the manufacturer’s instructions, luciferase activity was measured at 36 h after transfection by using the Dual-Luciferase Reporter Assay System (Promega). All assays were performed with a minimum of three replicates.

In Vitro UVB Irradiation Experiment.

The wild-type A375 cells (marked as CC) and the gene-edited A375 cells (marked as CT, TT-1, and TT-2) were cultured according to the above method. The UVB irradiation was performed when the cell density reached 80 to 90%. UVB-treated cells were marked as CC-UV, CT-UV, TT-1-UV, and TT-2-UV.

A 10-cm cell dish containing 8 × 10⁶ cells was divided into two dishes about 24 h before UVB radiation. When we started the UVB radiation experiment, we first removed the remaining DMEM and washed the cells using 3 mL 1× phosphate-buffered saline (PBS) buffer, and the washing procedure was carried out two times. Then we removed the PBS to eliminate the effect of buffer on UVB radiation. The cover of one of the A375 cell dishes was removed and the UV lamp (Tanon) was positioned over the uncovered cell plate [the distance from the UV lamp to the cells was 15 cm, according to the previous protocol (85)]. For the other cell dish (the control dish), we put it in a carton to protect it from UVB radiation. Next, all lights in the room were turned off except for the 302-nm UV lamp. The cells were exposed to UVB for 30 s at room temperature with a total UVB dose of 60 mJ/cm². This UVB dosage falls within the published dose range of UVB (0 to 400 mJ/cm²) (86). Finally, the irradiated cells were washed twice with PBS, and 6 mL fresh DMEM with 10% FBS was added to the UV dish and the control dish, and then cultured at 37 °C in a humidified incubator with 5% CO₂ for 36 h.

EMSA.

Probes and competitors of 31-bp sequences that encompass the ancestral or the mutant alleles (underlined letters) of rs75356281 were designed and synthesized (General Biosystems) as follows:

rs75356281C sense, 5’-AGGTGCTTGTATGGTCATAGGCCCACAGGTG -3’; rs75356281C antisense, 5’-CACCTGTGGGCCTATGACCATACAAGCACCT -3’; rs75356281T sense, 5’-AGGTGCTTGTATGGTTATAGGCCCACAGGTG -3’; rs75356281T antisense, 5’-CACCTGTGGGCCTATAACCATACAAGCACCT -3’.

Probes were labeled with biotin at the 5′ end while the competitors were unlabeled. Sense and antisense oligonucleotides were annealed to form a double-stranded oligonucleotide probe and competitor for the transcription factor–binding assay (EMSA). The EMSA was performed using a gel shift assay system (Thermo Fisher). Three microliters (about 3 μg) A375 nuclear extraction was incubated at room temperature for 20 min with 2 μL 10× binding buffer, 1 μL poly dI-dC (1 μg/μL), 1 μL EDTA (200 mM), and 50 fmol probes, corrected with ddH₂O to a final volume of 20 μL. Competition experiments were performed by the addition of an 100-fold molar excess of unlabeled competitors to the reaction mixture before adding probes. After incubation for 10 min, probes were added to the mixture and then incubated for 20 min. After the binding reaction, the mixture was added to 5 μL 5× loading buffer, loaded onto a 6.5% polyacrylamide gel in 0.5× Tris-borate-EDTA (TBE) buffer, electrophoresed for 1 h at 150 V, and transferred to a nylon membrane in 0.5× TBE buffer for 40 min at 380 mA. The membrane was cross-linked under UV light on a benchtop at a distance of 10 cm for 10 min. Finally, the biotin-labeled probes were detected by chemiluminescence following the manufacturer’s instructions (ProteinSimple).

Construction of Plasmids Encoding sgRNA for rs75356281 Base Editing.

Plasmids encoding BE3 (pCMV-BE3) were obtained from Addgene (73021). The small-guide RNA (sgRNA) (the guide RNA of CRISPR-Cas9 protein) for rs75356281 base editing was designed with an online sgRNA design tool, CHOPCHOP (87), and synthesized by Tsingke Biological Technology. Then the DNA oligos were annealed and cloned into the BsaI site of the sgRNA-expressing vector, pGL3-U6-sgRNA-PGK-puromycin (51133, Addgene). The sequence of the sgRNA used in this study is 5′-TATGGTCATAGGCCCACAGGTGG-3′ (the underlined sites are the short protospacer adjacent motif flanking the target DNA site).

Cell Culture, Transfection, and Sorting.

The human malignant melanoma A375 cells were cultured in DMEM, supplemented with 10% (volume/volume) FBS and 1% penicillin/streptomycin (Solarbio), and maintained at 37 °C in a humidified incubator containing 5% CO₂.

A375 cells were seeded into 6-well plates 1 d prior to transfection at a density of 2 × 10⁶ cells per well. Next, cells were transfected with 1.5 μg base editor plasmid (pCMV-BE3) and 1.5 μg pGL3-U6-rs75356281 sgRNA-PGK-puromycin using Lipofectamine 2000. Twenty-four hours later, cells were transferred to a 10-cm cell-culture plate (Corning). After 48 h for base editing, 1 μg/mL puromycin was added into the cell-culture media to eliminate the cells without the two transfected plasmids. The cells were harvested after 48 h and subjected to flow cytometry (BD, FACSAria II cell sorter). A single A375 cell was seeded into 96-well plates and cultured for about 1 wk to form colonies.

Gene-Edited Cell-Line Genotyping.

Genomic DNAs of the gene-modified A375 cells were extracted using DNAiso reagent (Takara). Then the DNAs used as PCR template and the amplicons for Sanger sequencing were generated in PrimeSTAR Max premix (Takara). The primer sequences used in genotyping are as follows (F, forward; R, reverse).

$$\begin{array}{l}\text{rs75356281} \text{PCR}-\text{F}: 5\prime -\text{AGGGATTTGAGAGGTGGGGTCTA}-3\prime ;\\ \text{rs75356281} \text{PCR}-\text{R}: 5\prime -\text{GAGGGCATGGAGCAATAGGAGAC}-3\prime .\end{array}$$

RNA-Seq and Transcriptome Analysis.

The A375 cell RNAs were extracted by RNAiso Plus reagent (Takara), and the RNA purity was validated using agarose gel electrophoresis and a NanoDrop 2000 (Thermo Fisher). For RNA-seq, the libraries of each sample were constructed using a kit from Novogene. Then these libraries were sequenced on the Illumina HiSeq 2500 platform with paired-end 150-bp reads, and each sample generated 6G data.

The quality control of the raw data includes the following steps. First, low-quality regions of the end of the sequence were removed using the BWA algorithm, and the threshold was 30; then the joint sequence and the sequence containing the ambiguous base N were trimmed from the raw data. In addition, transcripts with reads length <60 were removed. Through the above criteria, we achieved the expected high-quality clean data for further analysis.

The reference human genome and annotation files were downloaded from the Ensembl database (88) and the clean data were mapped onto the reference genome using HISAT2 (v2.0.5) (89). The position information of reads on the reference genome and the characteristic information of the sequenced samples were obtained to generate bam files. Then, according to the mapped reads, the read counts and TPMs (transcripts per million) were used to indicate expression level using featureCounts software (90).

We obtained 24 sets of RNA-seq data, including 12 normal condition melanoma cells (CC, CT, TT-1, and TT-2; three replicates of each type) and 12 UVB irradiation melanoma cells (CC-UV, CT-UV, TT-1-UV, and TT-2-UV; three replicates of each type). UVB radiation treatment was the same as above. DESeq2 software (91) was used for the analysis of differentially expressed genes. Multiple test correction was conducted using the Benjamini–Hochberg approach (92) and adjusted P < 0.05 and |log2foldchange| > 1 were applied to filter significant expression genes (93). jvenn was used for Venn diagrams (94).

The clusterProfiler package (95) in R was used for KEGG pathway category analysis. Multiple test correction was performed using the Benjamini–Hochberg approach (92).

We used STRING (61) to further investigate the relationship between GNPAT and the 68 differentially expressed pigmentation genes in the adaptive allele (TT-1 and TT-2) cell lines. We first identified the genes that interact with GNPAT in the differentially expressed gene set (TT-1-UV vs. CC-UV or TT-2-UV vs. CC-UV). Next, we mixed these genes with the 68 differentially expressed pigmentation genes to form a network. These results were visualized using Cytoscape (v3.8.2) software (96).

Data, Materials, and Software Availability

The RNA-seq data of the human malignant melanoma cells reported in this article have been deposited in the National Genomics Data Center (NGDC) China National Center for Bioinformation (CNCB) Genome Sequence Archive database (https://ngdc.cncb.ac.cn/gsa/), a publicly available repository. The raw data (accession no. PRJCA007361) are available in the NGDC CNCB BioProject database (97).