Jomon: Paleolithic Contingent in Modern Japanese

Discussion

This study undertook molecular linkage analysis of major surnames among Japanese men with Y chromosomal polymorphisms. The Y-STR network of samples was consistent with recent observations of the Japanese population, in which Y-SNP haplogroups were biased on the divergent distribution of Y-STR haplotypes, as reported by Watahiki et al. [27]. The combination of Y-STR haplotype with Y-SNP haplogroup provided a high-resolution network structure of the phylogeny of bearers [6].

The most common surname of Satoh showed very little Y-STR haplotype sharing. More or less equal diversity was found for subjects carrying other surnames examined for this study. Ruling out effects related to sampling and the area, these results suggest that genetic drift has not occurred on a large-scale after the inception of hereditary surnames in Japan. The most common name in England, Smith, exhibited a similar network structure to that of control for non-namesake individuals. In contrast, the average probability of matching in Irish common surnames is as high as 8.15% [12], which probably derives from drift events occurring in their history.

Vertical transmission of Y-chromosome/surname linkage can be disrupted by non-paternity events such as adoption of male children, transmission of maternal surname, or deliberate surname change. At least, the different Y-SNP haplogroups within the same surname imply the presence of another founder. Investigations of genealogical records in England and Iceland suggest confounding effects of non-paternity events, occurring at a rate of 1.3–2% per generation [7, 10, 28].

This high heterogeneity, with the inclusion of many singletons, suggests that Japanese common surnames have undergone extensive loss of relation because of extensive non-paternity events, in addition to other factors such as mutation in meiosis. Moreover, a more plausible interpretation is that this extensive diversity derives from numerous founders through history. Many people gained a surname recently, i.e. within the last couple of centuries. Effects of such multiple independent founders were observed also in South Korea [29].

In contrast to heterogeneity among namesakes, many groups showed a cluster of close haplotypes in the same SNP haplogroup. Genetic testing is incapable of identifying the true lineage between two male individuals in the absence of an extensive genealogical tree. Therefore, the criteria were tentative. According to criteria used by King and Jobling [7], only five DCs of #1, #2, #6, #13, and #15 (in Supplementary data) were detectable from the present data. Their criteria were strict, but they caused underestimation. The requirements are expected to be suitable for a local rare surname group having total population of less than 10,000. Martinez-Cadenas et al. [21] speculated that five mutational steps would be maximal for any two bearers. We therefore examined how many steps are reasonable for the Japanese population.

To construct putative DCs for this study, we assumed three classes of fewer than five mutational steps. For analyses, 22 DCs were identified in 13 surname groups. In 12 clusters of fewer than three steps in group 1, a founder was assumed to have existed 279–1269 years ago. In 10 clusters within four steps in group 2, a founder was assumed to have existed 886–1887 years ago. In group 3 within five steps, up to 9 of 11 led to an estimate of more than millennial age. Apparently, when larger numbers of mutational steps are considered, a more ancient founder must be assumed. Furthermore, when more members are involved, even in the same step, larger TMRCA must be estimated. Overestimation might occur in a group within five step-neighbors. These results suggest that a group within four steps is optimal for selection of a DC in the huge population of a whole nation.

In Japanese history, surname use was prevalent among the noble class in the tenth century during the Heian period; the practice later to extended to the military class [15]. These major surnames are thought to have originated during this period, meaning that the surnames analyzed in this study spent roughly a millennium as the time-depths of surname establishment. Founders from the ancient noble families are known for some of the major surnames prevailing today. For instance, the most populous surname of Satoh is said to be derived from a local faction of the famous Fujiwara family.

After the public use of surnames had been prohibited among farmers for a couple of centuries, most people carried surnames late in the Edo period. In general, surnames are believed to have been inherited paternally throughout history. The heritage principle was established by law in the 19th century, during the Meiji period.

Our results highlight the fact that common Japanese surnames consist of DCs and many singletons, indicating mixed stratification of long-term bearers and short-term bearers in the population. The present molecular study revealed characteristic features of Japanese surnames. Japanese society is characterized by a wider variety of surnames than those of Korea and China. Given ~100,000 surnames in the country, regionally uneven distributions have been reported for common surnames and rare surnames [15]. For China, which has the world’s longest history of surnames, a large-scale study revealed a high relation of surnames with Y-chromosome polymorphism [30]. As in European countries, analyses of Y chromosome and surnames are progressing in Asia and elsewhere. Reportedly, several DNA registering banks for surnames have been established around the world to support historical studies [31]. Furthermore, forensic use of Y-chromosome DNA is anticipated for personal identification [32]. The investigation of Y-chromosome should be developed in Japan on a large scale.

In conclusion, the present examinations of the RST plot and network analyses revealed that divergence in collected surname samples progressed to a high degree. In contrast, many DCs composed of close haplotypes in the same SNP haplogroup were identified. Common Japanese surnames should consist of long-term bearers and short-term bearers in the population, which can reflect influences of historical events. This report is the first of extensive research conducted of the Japanese population.

Materials and methods

A total of 90 and 72 individuals from Izumo City of Shimane Prefecture and Makurazaki City of Kagoshima Prefecture, respectively (see Fig. 1), were recruited for this study.

It should be noted that individuals from Izumo were sampled in two periods and at different locations. We first collected blood samples at Department of Human Genetics, School of Medicine, the University of Tokyo in 2013 from 21 individuals who grew up in Izumo City and were at that time residents of Tokyo Area. Most of their grandparents also grew up in Izumo area. These data on 21 individuals were reported by Saitou and Jinam [13] as a preliminary study. We later collected saliva samples at the Archeological Museum of Kojindani, Izumo City in 2014 from 69 individuals. In total, only 51 DNA samples from Izumo contained enough good-quality DNA for genotyping.

As for the Makurazaki population, blood samples were collected from 72 individuals in 2015 at Makurazaki City with the help of the Makurazaki City Medical Association, and all 72 individual DNA samples were used for this study. Four grandparents of all these 72 individuals collected in Makurazaki grew up in the Southern Satsuma region. Both the Izumo and Makurazaki samples were genotyped using the Japonica array [14]. After filtering out SNPs based on genotyping call rate (<95%) and the Hardy-Weinberg equilibrium (p < 1 × 10−10), a total of 625,177 autosomal SNPs remained. Cryptic relatedness within each population was checked using KING software [15]. Six individuals from Izumo and eight from Makurazaki with up to 3rd degree relations (kinship coefficient more than 0.06) were omitted from further analysis. Therefore, 45 and 64 genome-wide SNP data for Izumo and Makurazaki, respectively, were used for subsequent data analyzes.

The Japonica-array SNP data for Izumo and Makurazaki were then merged with the following datasets: (1) BioBank Japan (BBJ) [16, 17] genome-wide SNP data (922,511 sites); (2) Affymetrix 6.0 micro-array data (641,314 SNP sites) of 20 Ainu, 35 Okinawa, 50 Yamato (mostly from Kanto region) [8], and 42 Chinese Han in Beijing (CHB) [18] individuals; (3) high-coverage whole genome data of 88 Koreans from the Korean Personal Genome Project [19]; (4) high-coverage whole genome data of Asian populations (CHB, Southern Han Chinese, Chinese Dai from China, Kinh from Vietnam) from the 1000 genomes project [20].

The BioBank Japan dataset (BBJ) [17] consisted of 182,546 individuals treated at hospitals in seven regions in Japan: Hokkaido, Tohoku, Kanto-Koshinetsu (Kanto for short), Tokai-Hokuriku (Tokai for short), Kinki, Kyushu, and Okinawa (Fig. 1). The genotyping data obtained by SNP array was downloaded from Japanese genotype-phenotype archive on the approval of NBDC Data Access Committee (JGA Accession ID: JGAD000123). The genotype data were converted into VCF format.

Whole genome sequence data were analyzed by the workflow recommended by the GATK best practice. We retrieved the mapped-read data in CRAM format for Asian population from the 1000 Genomes project (https://www.internationalgenome.org/). For Korean population, raw reads in fastq format were retrieved from The Personal Genome Project Korea (http://opengenome.net/Main_Page). The raw reads were then mapped to GRCh38 reference sequence by bwa-mem in order to consolidate with 1000 Genomes project data. Variant discovery was performed using the HaplotypeCaller algorithm implemented in GATK4.1 to perform joint calling. Joint calling was performed by GVCFtyper algorithm implemented in sentieon package [21] for computational efficiency. This program yields results compatible with GATK’s GenomicsDB and genotypeGVCFs programs. Each variant was scored using the VQSR algorithm to filter the variants. We applied VQSR by valCal and ApplyVarCal program of sentieon package. The HapMap and International 1000 Genomes Omni2.5 sites, the high-confidence SNPs of 1000-G sites, and the dbSNP151 sites are used as true datasets, training datasets, and known datasets, respectively. SNPs that passed the 99.9% sensitivity filter by VQSR were used for subsequent analyzes. The filtered genotype data in VCF were then merged with SNP array data of the BBJ, Izumo, and Makurazaki datasets by the merge function implemented in bcftools. SNPs that are missing in either dataset, minor allele frequency <1%, call rate <3%, or statistically significant deviation from Hardy-Weinberg equilibrium (p value < 0.00001) were filtered out. In total 106,237 SNPs remained. For the subsequent analyzes that require the independence of nearby SNPs, the linkage disequilibrium (LD)-based SNP pruning were performed by PLINK1.9 [22] with “--indep-pairwise 500 50 0.1”. We obtained 39,331 LD independent SNPs. After excluding duplicate samples and those with cryptic relatedness by KING, the principal component analysis (PCA) was performed using PLINK1.9 to remove outlier samples. The number of BBJ samples remaining after this step was 169,169. To reduce computation times for subsequent analyzes, we decided to further trim down the BBJ dataset. First, the mean values of the top 10 principal component values were calculated for each region, and the deviation of the principal component values from the mean were assessed for each sample by Euclidean distance. The top 1000 samples with the smallest Euclidean distances were selected from each region and used for subsequent analyses.

PCA was again performed using the trimmed down BBJ dataset using PLINK1.9, and Admixture [23] was used for admixture analysis. f4 test implemented in treemix [24] was used to test for gene flow between populations. GBR (British in England and Scotland) population from the 1000 Genomes dataset was used as outgroup for the f4 test. Phylogenetic analysis of populations was conducted using the neighbor-joining method [25] and Neighbor-Net [26] for phylogenetic tree and network construction, respectively. Pairwise Fst distances between populations calculated in smartpca [27] was used for tree and network construction.