We obtained a genome-wide 58,960-SNP dataset (Affymetrix GeneChip Human Mapping 50K Xba chip; Affymetrix Inc., Santa Clara, CA, USA) from the HUGO Pan-Asian SNP Consortium website (http://www4a.biotec.or.th/PASNP) [12]. The SNP data were obtained from 1,833 individual DNA samples collected from blood and buccal samples of 72 Asian and non-Asian ethnic groups. Filtering SNPs is an important step in genome-wide studies to minimize potential false findings. We filtered SNPs at 2 levels: individual and SNP. In the individual-level filtering step, we removed individual SNPs that had a <90% genotyping call rate and that were not in Hardy-Weinberg equilibrium (p < 0.001). We filtered out 89 of the initial 1,833 individuals and used SNPs from the 1,744 individuals for further study. In the SNP-level filtering step, 1,692 of the initial 58,960 SNPs were filtered out, because 389 SNPs had an SNP call rate below 90%, 451 SNPs were physically unmapped, 103 SNPs could not be mapped to dbSNP, and 749 SNPs were monomorphic. Finally, we obtained 56,025 SNP markers from 1,744 genotyped individuals, representing 72 ethnic groups in the HUGO Pan-Asian SNP Consortium.

For comparison with other ethnic variants, we obtained SNPs of 209 HapMap individuals representing 4 populations (China Han [CHB]; Japan Japanese [JYP]; USA European [CEU]; Africa Yoruba [YRI]) (http://www.hapmap.org/). The 243 SNPs in the HUGO Pan-Asian with the HapMap dataset were nonshared SNP loci and excluded.

Phasing is needed to determine which variants are inherited by an individual at each locus and to more accurately determine the relationships between unrelated individuals in large population datasets. Using phasing population data, we can identify haplotypes, which are essentially segments of DNA that are common to a particular ethnic group. The fastPHASE program version 1.1.4 [13] was used to estimate missing genotypes and reconstruct haplotypes from unphased SNP genotype data of unrelated individuals in order to identify ethnicity-specific SNPs.

SNP genotyping arrays recently have been used for CNP detection and analysis, because the arrays can serve dual roles for SNP- and CNV-based association studies. To detect CNP markers from the SNPs, we used the Affymetrix Genotyping Analysis Software and Copy Number Analysis Tool (CNAT version 3.0), which was downloaded from the Affymetrix website (http://www.affymetrix.com/products/software). We calculated the distribution of genomic smoothing copy number signal intensity (upper-boundary = 2.78, lower-boundary = 1.51) to detect the CNPs (Supplementary Fig. 1). We regarded markers with values beyond the upper- and lower-boundary values as CNVs. Then, we selected the CNPs that presented variation in >1% of the individuals as CNVs that occurred in more than 1% of the population [14]. We downloaded the detection tools from the website (http://www.affymetrix.com/products/software) and used these programs for our Pan-Asian SNP data. We regarded markers with values outside of the upper- and lower-boundary values as CNVs. Then, we selected CNPs which presented variation in >1% of the 1,196 individuals.

To investigate the distribution of ESNP and ECNP markers in Pan-Asian ethnic groups, we performed the following steps, as shown in Supplementary Fig. 2. First, we classified the ethnic groups into 3 hierarchical categories by population estimation of the phylogenic analysis (in Supplementary Fig. 2): superpopulation (clustering size = 4), subpopulation (ESNP clustering size = 12, ECNP clustering size = 11), and ethnic population (ESNP clustering size = 72, ECNP clustering size = 60). The clustering size in parentheses refers to how many distinct populations were used in the ESNP and ECNP analysis. The superpopulations consisted of African, Caucasian, American Indian, Asian, and outlier (Indian and Uyghur) populations. The subpopulations comprised population groups based on linguistic similarities and population structures and 2 additional groups (Mongoloid features [IN-NI]/Mongoloid features [IN-TB] and China Uyghur [CN-UG]), and the ethnic populations represented each ethnic group. Second, we identified ethnicity-specific variations based on SNPs and CNPs markers using the nearest shrunken centroid method (NSCM), which is not affected by the minor-allele frequency [15, 16]. The NSCM of the R package pamr carried out multiple tests for genetic heterogeneity and frequency spectrum on genes having ethnicity-specific variants. It shrinks each of the class centroids toward the overall centroid for all classes by a threshold. This shrinkage makes the classifier more accurate by eliminating the effect of noise. This method is a suitable attempt to solve the classification problem when there are a large number of features from which classes and a relatively small number of cases are predicted, and it is a significant approach to identify which features contribute most to the classification [15]. Third, we identified specific ethnic groups that correlated with each SNP or CNP. Finally, to investigate the characteristics of ethnicity-specific variations, we obtained the top 10% of ESNPs and ECNPs markers.

To investigate the characteristics of the ECNP- and ESNP-related genes, we selected the ECNP and ESNP markers ranked in the top 1% of the ethnically specific variations. We mapped ESNP and ECNP markers to gene structure, based on Entrez gene-centered information at NCBI [16]. We identified the relationships among ESNP- and ECNP-genes using the Ingenuity Systems Pathway Analysis tool (http://www.ingenuity.com), which is a web-based tool for pathway and network analysis of genes.

We obtained 58,960 Pan-Asian SNPs from 72 Pan-Asian populations listed in the supplementary documents (Supplementary Table 1). The SNP dataset shows the spectrum of genetic and phenotypic diversity in Asian populations and allows us determine the ethnic specificity of Asian populations based on ESNP and CNP makers.

We devised an approach to identify ethnic differences from the genotype profiles of the populations. Pan-Asian SNPs were filtered using several steps, as described in Materials and Methods. With the filtered SNPs, we examined the inter-SNP distances, the allele frequency, and the heterozygosity distribution to identify the genotypic features of each ethnic group. The average inter-SNP distance was 52 kb. More than 41% of SNPs fell below 10 kb, and 14% was over 100 kb (Supplementary Fig. 3). The inter-SNP distance distribution within 10 kb of one another was similar to the distribution observed in the HapMap data.

We investigated minor allele frequency and heterozygosity across the ethnic groups (Supplementary Fig. 4). We compared SNP genotypes in the ethnic groups with those identified by the HapMap [17] consortium. The proportion of Pan-Asian SNPs in each minor allele frequency and heterozygosity range was similar to those of the 4 HapMap ethnic groups. In particular, the SNP proportions for CN-UG, Proto-Austroloids (IN-DR), and Singapore Indian (SG-ID) were almost the same as those of the 4 HapMap ethnic groups. We found that 5 ethnic groups (America Melanesians [AX-ME], Thailand Mlabri [TH-MA], Indonesia Mentawai [ID-MT], CN-UG, and SG-ID) had minor allele frequency and heterozygosity distributions that were significantly different from those of other ethnic groups. Approximately 40% of the SNPs in the AX-ME, TH-MA, and ID-MT populations fell into the <0.05 range. The proportion of SNPs with heterozygosity <0.05 was lowest in the CN-UG and SG-ID groups, whereas the proportion of SNPs with heterozygosity <0.10 was highest in the TH-MA and AX-ME groups.

We confirmed the high proportion of African (YRI)-specific SNPs from 3 races: Asian, Caucasian, and African (Supplementary Fig. 5). Then, to search ESNPs from Pan-Asian ethnic groups, we identified 9,306 ESNPs from the Pan-Asian SNP markers and examined the distribution of the ESNPs at the 3 population levels: super-, sub-, and ethnic population (Fig. 1A). We found that 67.82% of ESNPs in the superpopulation level were others, including IN-NI, IN-TB, and CN-UG. It could show that isolated ethnic groups keep their original features and that these features are also presented in other population levels. In the subpopulations, Austronesian, IN-NI, and IN-TB had 95.31% of ESNPs. Additionally, we identified 28,282 CNPs based on the SNP chip profiles and then selected 5,167 ECNPs. We then examined the ECNPs in the 3 population levels (Fig. 1B). In the superpopulations, 36.57% of the ECNPs occurred in Caucasoids. Of the subpopulation ECNPs, 56.26% had Mongoloid features and 36.95% occurred in Uyghurs. Across the ethnic groups, 26.35% of the ECNPs were from Caucasoids (IN-NL). Most ECNPs occurred in Caucasoid populations (IN-NL, IN-IL, IN-DR, IN-SP, IN-EL, IN-WI, IN-WL, and SG-ID) and in ethnic groups that have Mongoloid features (IN-NI, IN-TB, and CN-UG). The ethnic group having Mongoloid features had more ECNPs than other groups, which shows that Pan-Asian ECNPs could serve as the criteria for ethnic specificity in Pan-Asian populations.

We obtained genes associated with ESNPs and ECNPs based on the NCBI Gene database [18]. We identified 156 ESNP-related genes and 52 ECNP-related genes (Supplementary Table 2). Most of the ESNP-containing genes were associated with known molecular functions encoded for cellular assembly and maintenance, cellular movement, cell death and survival, and lipid metabolism (Table 1). These ESNP-related gene sets (CEP192, GRK5, TUBGCP3, TUBGCP6, ABCA1, CHGA, ITGB3, VAV2, MYLK3, ANK2, ITGB3, ELN, LTBP4, FAM107B, DIDO1, HOXD3, PRPX2, KCNMA1, RASGRF1, NF1, and UBR2) showed association with diseases, such as cardiovascular disease, dermatological diseases and conditions, developmental disorders, and connective tissue disorders. Defects in latent transforming growth factor beta binding protein 4 (LTBP4) may be a cause of cutis laxa and severe pulmonary, gastrointestinal, and urinary abnormalities [19]. Centrosomal protein 192 kDa (CEP192) is part of a large multisubunit complex required for microtubule nucleation at the centrosome. This gene is regulated by hepatitis B virus and has roles in cells, such as duplication, replication, assembly, and nucleation. The ECNP-related genes are associated with known molecular functions encoded for cell-to-cell signaling and interaction, molecular transport, and drug metabolism (Table 1). These ECNP-containing genes (ARID1B, CAPN7, DDR2, SLC22A1, SLC22A3, ASTN2, GABRA5, GABRG3, RBFOX1, TLR4, MAG, TLR4, SNX9, and SYNJ2) are associated with cardiovascular disease, neurological disease, psychological disorders, and developmental disorders. Interestingly, the APOBEC gene, encoding the C→U editing enzyme family, such as apolipoprotein B mRNA editing enzyme and catalytic polypeptide-like 2, is associated with ECNPs. This editing gene could be affected by genetic diversity and admixtures in Pan-Asian populations. In this ethnically specific genetic variation study, we found the features of gene sets having ethnically specific variations. The AT-rich interactive domain 1B (SWI1-like) (ARID1B) gene encodes an AT-rich DNA interaction domain-containing protein. This protein is a component of the SWI/SNF chromatin remodeling complex and may play a role in cell-cycle activation. It is associated with chromatin-mediated maintenance of transcription and nervous system development. This gene is related to autosomal dominant mental retardation type 12. Discoidin domain receptor tyrosine kinase 2 (DDR2) is associated with colorectal neoplasm, metastatic colorectal cancer, hepatocellular carcinoma, gastrointestinal stromal tumor, and so on. These ethnically specific genetic variations represent relationships with cell cycling, metabolism, and the developmental and immune systems. These genetic variations can be affected by admixture and evolution. These ethnically specific variants may be forced to adapt in order to survive in new environments and play a significant role in development and functional systems associated with diseases [20].

Our analysis was able to identify ethnically variable SNPs associated with phenotypic changes. We selected 9,306 ESNPs and 5,167 ECNPs in 72 Pan-Asian populations. We found that representative ethnic groups with specific ESNPs are recently branched-out subpopulations, whereas representative ethnic groups with ethnic specific CNPs are early fixed subpopulations, as shown in Supplementary Fig. 6. This likely occurred due to accelerated and accumulated genetic drifts or selective pressure. This shows that ESNPs may participate in the ongoing creation of genetic variation through selective pressure or selective sweep. These ethnically specific variations are associated with 156 ESNP-related genes and 52 ECNP-related genes. Ethnically specific genetic variations may affect phenotype variation and disease susceptibility. Although it is not enough information to compare ECNP distribution and ESNP distribution directly, significant gene sets having ESNPs or ECNPs can be useful to study disease risk and drug susceptibility.