A set of novel SNP loci for differentiating continental populations and three Chinese populations

Reference populations

Twenty-five populations within four major geographical regions were chosen as the training set for preliminary assessments of selected SNPs. Six testing populations in three continents included Esan, Yoruba, Finnish, British, Beijing Han and Japanese populations, of which genetic data were downloaded from 1000 Genomes Phase 3 (Genomes Project Consortium et al., 2015). Genetic data of three subpopulations in China (Han, Mongolian, and Uygur) was obtained from CEPH-HGDP (Li et al., 2008). Detailed descriptions of these populations and their corresponding sample sizes were given in Table 1. Besides, genetic data of 48 SNPs in training, testing and three Chinese populations were presented in Table S1.

Criteria for SNP selection

SPSmart includes the genetic data of 1000 Genomes Phase I, HapMap release #28, Perlegen complete data set and the Stanford University and Michigan University CEPH-HGDP panels, which is developed to help researchers use and combine different datasets and do some statistical analyses of interest (Amigo et al., 2008). SNPs were chosen from SPSmart online tool when they met the following criteria: (1) SNP loci should locate in intron regions. (2) SNP loci were bi-allelic genetic markers; (3) SNP loci were located on different chromosomes or at least 10 Mb distances on the same chromosome; (4) Ancestral allele frequency differences between continental populations were at least 0.3; (5) SNP selected must conform to Hardy-Weinberg equilibrium (HWE) in all reference populations. Next, SNPs whose frequency differences among Han, Uygur and Mongolian populations were more than 0.3 were used for further analysis. Besides, we also retained some SNP loci that showed high genetic differentiations among continental populations/three Chinese populations. Finally, forty-eight SNPs were selected to differentiate continental populations and three Chinese populations. General information of 48 SNPs was given in Table 2.

Statistical analysis

HWE tests of SNP loci in 25 training populations were estimated by Genepop software v4.0 (Rousset, 2008). Allele frequencies of 48 SNP loci in 25 training populations were calculated by PowerStats software v1.2 (Promega, Madison, WI, USA). The informativeness for assignment (In) values of 48 SNP loci in four continental populations (African, American, European, and East Asian) were calculated by Infocalc program v1.1 (Rosenberg et al., 2003) based on the genetic data of 25 training populations. Ancestral allele frequency heatmap and the boxplot of In values of 48 SNPs were plotted by R software v3.3 (R Core Team, 2016). Principal component analysis (PCA) of four continental populations including 25 training populations was conducted by PLINK software v1.9 (http://www.cog-genomics.org/plink/1.9/), and then scatter plot of these populations was plotted by R software v3.3. Population genetic structure of 25 training populations at K = 2–5 and cross-validation error of each K value were performed by ADMIXTURE software v1.3 (Alexander, Novembre & Lange, 2009). Graphical results of estimated ancestry proportions were conducted with the CLUMPAK online tool (Kopelman et al., 2015).

To further evaluate discrimination efficiencies of 48 SNPs for continental populations, ancestry components of six testing populations were estimated by ADMIXTURE software v1.3 and their results were shown in the form of beeswarm by R software v3.3. Next, twenty-two training populations (excluding from American populations) were treated as reference populations and six testing populations were blind samples. PCA and Naïve Bayes analysis of these populations including reference populations and six testing populations were conducted with PLINK software v1.9 and the Snipper App suite v2.5 (http://mathgene.usc.es/snipper), respectively.

Ancestry analyses among Uygur, Han and Mongolian populations were conducted based on the 48 SNPs. First of all, allele frequencies of 48 SNPs in these three populations were calculated by PowerStats software v1.2. In values of 48 SNP loci in Uygur, Han and Mongolian populations were also calculated with Infocalc program v1.1. Secondly, PCA of these populations was also conducted by PLINK software v1.9. Ancestral allele frequency heatmap and the boxplot of In values for 48 SNPs, and scatter plot of these three populations were generated by R software v3.3 (R Core Team, 2016). Genetic structure analyses of these populations at K = 3 were conducted by ADMIXTURE software v1.3 and graphical results were plotted by CLUMPAK online tool. In the end, multinomial logistic regression analyses of these three populations were conducted by Snipper App suite v2.5.

Frequency distributions and population specific In values of 48 SNPs

After applying Bonferroni correction (the significant level = 0.05/48 = 0.00104), the selected 48 SNPs all conformed to HWE in 25 training populations (Table S2).

Ancestral allele frequencies of 48 SNP loci in 25 reference populations were shown in Fig. 1. Color contrasts of SNPs in pairwise populations reflected genetic differentiations of pairwise populations: more apparent color contrasts of pairwise populations were, larger genetic differentiations pairwise populations possessed, and vice versa. Besides, the SNPs with distinct color contrasts in pairwise continental populations contributed to differentiating these populations. For example, ancestral allele frequencies of the rs10918196 locus in African, American and East Asian populations were 0.8351, 0.3947 and 0.2429 (Table 2), implying the locus was beneficial to distinguish African populations from the other two continental populations. The phylogenetic tree above the heatmap (Fig. 1) revealed the relationships of SNPs: SNPs showed similar frequency distributions in continental populations tended to locate in the same sub-branches, and vice versa. The phylogenetic tree in the left part of the graph reflected genetic divergences of different continental populations: populations with the same biogeographical origins located in the same sub-branches.

In values of 48 SNPs in African, American, European and East Asian populations were shown in Fig. 2 and Table S3. Similar with allele frequency differences, In values were also used to evaluate the degree of population differentiations: genetic markers with high In values in the certain population contributed to differentiating the population from the other populations (Phillips, 2015). For these 48 SNPs, there were 12, 11, 10, and three SNPs with relatively high In values (≥0.1) in African, East Asian, European, and American populations, respectively. Besides, some SNPs which had similar frequency distributions among different continental populations showed low In values in these four continental populations (Fig. 2). Besides, we also calculated the cumulative In values of 48 SNPs in African, European, East Asian, and American populations, which were 3.4312, 3.0343, 2.7727 and 1.3028, respectively (Table S3). The low cumulative In values in American populations indicated these SNPs might possess lower power to differentiate American populations from the other populations.

PCA and ancestral component analysis of 25 training populations based on 48 SNPs

PCA of four continental populations comprising 25 training populations was shown in Fig. 3. Results revealed Africans and Europeans formed two distinct population clusters at quite some distances from the other populations. However, some American individuals clustered closely with East Asian individuals, which might be related to the shared ancestries before the divergences of East Asian populations and American populations (Li et al., 2008).

Next, we assessed genetic components of 25 training populations (Fig. 4A). At K = 2, the populations in Africa and Europe showed similar genetic components which could be distinguished from the populations in East Asia and America. At K = 3, African populations and European populations exhibited their distinct ancestry components, respectively. When K became 4, specific ancestry components in American populations could be observed and all training populations could be classified into four apparent clusters. No further distinctions among these populations could be discerned at K = 5. Given these results, apparent distinctions among these continental populations could be achieved by the 48 SNPs. Cross-validation error of each K value was estimated by ADMIXTURE software v1.3 to determine the optimum K value, as presented in Fig. 4B. Results revealed that the lowest cross-validation error was observed at K = 4, indicating K = 4 was the most appropriate for these data.

Ancestry analysis of six testing populations based on 48 SNPs

Since American populations collected in 1000 Genomes Phase 3 showed different degrees of admixture components of European, American and African ancestries (Genomes Project Consortium et al., 2015), some populations from East Asia, Europe and Africa were employed to evaluate the efficiency of 48 SNPs for differentiating continental populations. The testing set included two African populations (ESN and YRI), two European populations (GBR and FIN) and two East Asian populations (CHB and JPT). Firstly, we estimated ancestry components of these six populations by ADMIXTURE software v1.3, as shown in Fig S1A. Results indicated that populations within the same continents showed similar genetic component distributions and could be separated from the populations in the other continents (Fig S1A). Furthermore, cross-validation plot revealed K = 3 was the best value for these testing populations (Fig S1B). Therefore, estimated genetic components of each individual for six testing populations at K = 3 were presented in Fig. 5A. We found FIN and GBR individuals showed high genetic components of European ancestry; CHB and JPT individuals possessed high genetic components of East Asian ancestry; ESN and YRI individuals demonstrated high genetic components of African ancestry. Consequently, these six testing populations could be assigned into their corresponding continental origins by these 48 SNPs.

Next, twenty-two training populations were classified into three known populations (African, European and East Asian) according to their biogeographical origins; six testing populations were treated as unknown individuals. PCA of these populations was conducted, as shown in Fig. 5B. Result demonstrated that African, European and East Asian individuals formed three population clusters. Moreover, we found that these testing individuals were superimposed onto the correct population clusters in Fig. 5B. Results of Naïve Bayes analysis also revealed all testing samples could be assigned to their corresponding continental regions (Table S4). For example, individual HG02922 was classified into African individuals with more than one billion times than European and East Asian individuals. From the above results, these 48 SNP set performed well for ancestry origin predictions of three continental populations (African, European and East Asian).

Discrimination efficiencies of 48 SNP loci for three Chinese populations

Ancestral allele frequencies of 48 SNP loci in Uygur, Han and Mongolian populations were given in Supplementary Fig. 2. Distinct frequency differences of most SNPs could be observed among pairwise populations. As an example, ancestral frequencies of rs1857859 locus in Uygur, Han and Mongolian populations were 0.65, 0.76 and 0.25, respectively, indicating the locus was good for distinguishing Mongolians from Uygurs and Hans. Population specific In values of 48 SNPs in these three Chinese populations were presented in Fig. 6 and Table S3. Results revealed that 10, 7 and 5 SNP loci displayed relatively high In values (≥0.1) in Uygur, Han and Mongolian populations, respectively. Additionally, some SNP loci which had low In values in four continental populations showed relatively high In values in one of the three Chinese populations. For example, rs10852189 locus whose In values in four continental populations were less than 0.05 had In value with being more than 0.1 in Han Chinese population.

Next, we assessed the power of 48 SNPs for ancestry analyses of Han, Mongolian and Uygur populations. Firstly, population structure analysis of three Chinese populations at K = 3 was given in Fig. 7A. Different ancestral component distributions were seen among these populations: Han population showed high blue proportions; Mongolian group displayed high purple proportions; Uygur group exhibited high orange proportions. We also found some individuals showed admixture ancestry proportions, which might result from the recent admixtures of these populations. Nevertheless, individuals from three Chinese populations formed three distinct clusters in the PCA plot (Fig. 7B). Moreover, multinomial logistic regression analyses of three Chinese populations were conducted to further evaluate the efficiencies of 48 SNP loci (Table S5). Results indicated Uygur, Han and Mongolian individuals were correctly classified into their corresponding populations with the probability values of 1.0000, reflecting these 48 SNP loci could estimate biogeographical origins of these Chinese populations well.