Ancient genomics reveals tripartite origins of Japanese populations

Description

The subsequent period between the beginning of wet-rice cultivation and the emergence of the custom of burying the elite in keyhole-shaped tumuli is referred to as the Yayoi period, the first agrarian phase in Japanese history (1). Systematic rice paddy field agriculture was introduced into the Japanese archipelago during the first millennium BC. Within approximately 1,000 years, technological advancements in tool-making led not only to further advances in rice cultivation, but also to social stratification due to increasing contradiction between the hierarchization of social relations and the necessity of maintaining communal collaboration and the egalitarian ethos for food production (1).
The following Kofun period is regarded as the state-formation phase as represented by the beginning of the imperial reign. A prominent feature of this period is the construction of large burial mounds, whose size reflected the social status of its occupant during their lifetime (1). This trend is particularly evident in archaeological sites from the early Kofun period. However, the keyhole structure slowly began to be downsized over the course of time (1). Coinciding with this decline in the size of large keyhole tumuli, packed tumuli clusters, such as horizontal cave tombs, became prevalent in the Late Kofun period. The construction of keyhole tumuli ceased by the end of the Kofun period (the six century AD) which marked the completion of the process whereby the elite became the rulers and managed to control inevitable hierarchies that had persisted since the introduction of paddy field rice cultivation.
In the following sections, we provide details of archaeological contexts of the sites where the samples used in this study were collected from (see Fig. 1a for geographic locations of the sites).
1. Jomon 1.1. Kamikuroiwa rock shelter (Initial Jomon; JpKa6904) Kamikuroiwa rock shelter is located in Kumakogen, Kamiukena District, Ehime Prefecture of Shikoku. The site is sitting on a terrace along Kuma River, at an elevation of about 396 m above sea level. The excavation of this site was conducted from 1962 to 1970, identifying well-preserved deposits that were stratigraphically divided into nine layers. Most of the human burials, including JpKa6904 sequenced in this study, were recovered from Layer 4, coupled with large pieces of Jomon potteries; their roller-stamped designs (so-called "Oshigata-mon" in Japanese) associated Layer 4 with the Initial Jomon. The radiocarbon date of JpKa6904 reported in (89) was corrected with the IntCal20 curve, which gave 6,696-7,041 calBC (8,819 ± 172 calBP), consistent with the archaeological contexts of this layer. JpKa6904 had morphological characteristics of a female, which was further confirmed by our genetic data (see Table S3). An additional individual recovered from the same layer was screened for high-coverage sequencing but had little endogenous human DNA left in a tooth (<1%).

Odake shell-midden (Early Jomon; JpOd)
The Odake shell midden is located around 4 km inland from the present-day seacoast, at the junction of the Kureha hills in Toyama, a coastal city of north central Honshu. At the time of settlement at the Odake shell midden, the sea level was higher than the level at the present due to climatic warming and this site was located along the shore, emerging as the result of marine transgressions. At least 100 human skeletons have been discovered from this site, most of which were excavated between 2009 and 2010 due to the construction of new train lines. Buried skeletons had a specific burial practice in which the body was placed in a flexed position with bent legs. Some of the buried individuals folded stones in their arms across their chest. Animal remains excavated from this site included terrestrial and marine mammals, as well as fish, suggesting that the Jomon in this area utilized a variety of food resources. Four individuals from this site produced genomic data in this study, all of which are included in the early stage of the Early Jomon period as follows: • JpOd274: 4,169-4,339 calBC (6,204 ± 85 calBP) • JpOd6: 3,984-4,229 calBC (6,057 ± 123 calBP) • JpOd181: 3,801-3,967 calBC (5,834 ± 83 calBP) • JpOd282: 3,787-3,952 calBC (5,820 ± 83 calBP)

Funagura shell-midden (Early Jomon; JpFu)
The Funagura shell-midden is located in Kurashiki, a coastal city of western Honshu, sitting on the Seto Inland Sea that separates three main islands, Honshu, Shikoku, and Kyushu. Three burials were excavated from this site in 1991; two individuals were initially screened for human endogenous DNA; only JpFu1 was analyzed in this study due to the poor preservation of the other individual. JpFu1 was discovered in Layer 3 where the density of shell deposits was extremely high. Several Jomon potteries, together with lithic tools, were recovered from the same layer, the type of which had an association with the late stage of the Early Jomon period. Our radiocarbon dating showed 3,528-3,640 calBC (5,534 ± 56 calBP). This individual had morphological characters that were considered an adult female; our genetic analysis assigns female to this individual (Table S3). In addition to the human skeletons, animal remains were also excavated from this site, including fish (e.g., sharks and Japanese black porgy), birds, and mammals (e.g., deer and boars). These archaeological evidence supports that both fishing in the Inland Sea and hunting terrestrial animals could have been central to food collection for the Jomon population in this site.

Kosaku shell-midden (Middle-to-Late Jomon; JpKo)
The Kosaku shell-midden is located in Funabashi, a coastal city of central Honshu. This site was initially discovered in 1883; following excavations identified the Jomon pottery that represents the Late Jomon period, in particular those having the lid with many large bracelets made from shell inside the pottery. Around 100 human skeletons were recovered from this site, some of which were collectively buried. Two individuals sequenced in this study, JpKo2 and JpKo13, were buried individually, dated to be the Middle or Late Jomon period (see below). There were animal remains excavated together with the human skeletons, which included deers, boars, racoons, as well as a variety of fish and shells. We identify the excess accumulation of short runs of homozygosity (ROH) in our oldest Jomon genome, JpKa6904, compared to Mesolithic hunter-gatherers from Ireland (33) and Luxembourg (39), as well as the Upper-Paleolithic Northeast Siberian (19) and the other Neolithic or Pleistocene genomes (33,39,78) (see Fig. 3b). This suggests that the Jomon had a smaller population size than the West Eurasian hunter-gatherers likely due to a long-term isolation in insular East Asia.
To estimate the timing of divergence of the Jomon lineage, coupled with the population size, we employed a simulation-based modelling approach that can fit genome-wide patterns of ROH expected under given demographic conditions to those observed in the Jomon genome.
We applied diploid-genotype calling to the high-coverage ancient genomes including 7.5× JpKa6904. To minimize any potential confounding effects of postmortem deamination on the ROH analysis, we called genotypes only for single nucleotide polymorphisms (SNPs) with transversions filtered for global minor allele frequencies above 1% among the Phase 3 v5 1000 Genomes release (28). Additional filtering steps were applied to the genotype calling: bases with a quality >30, a sequence depth >10, and a genotype quality >20. This left 984,740 autosomal SNPs for ROH analysis. We then used PLINK v1.90 (76) to detect ROH in the Jomon genome with the following options: --homozyg --homozyg-density 50 --homozyg-gap 100 --homozyg-kb 500 --homozyg-snp 50 --homozyg-window-het 1 --homozyg-window-snp 50 --homozyg-window-threshold 0.05. The ROH profile was summarized into a spectrum of ROH fragments ranging from 0.5 to 100 Mb with a bin size of 1 Mb (Fig. S10).
The observed ROH spectrum is shaped jointly by the population size and timing of population split. The previous study has provided estimates on the divergence of Jomon lineage from the common ancestor of Han to be in between 18 and 38 ka ago, with a constant population size that falls within the range from 2,000 to 3,000 (14). We first performed a broad search for the parameter space that is defined with the population size (N) from 500 to 2,500 and population split from 10 to 40 ka ago (T) using the data from chromosomes 3 to 22 (Fig. S11). To test whether the Jomon demography influences the pattern of ROH in a genome, we generated 100 whole-genome simulations, using a coalescent simulator ms (81), for different combinations of N and T under a fixed scenario of the Out-of-Africa dispersal reconstructed from a previous study (80) with slight modifications (Fig. S10a). Our simulation assumed 25 years per generation and 1.25×10 -8 per generation (90) and 0.625×10 -8 per generation for a mutation rate and a recombination rate respectively.
We then estimated the likelihood of observing the ROH spectrum similar to that observed in JpKa6904 (shown as a dashed red line in Fig. S10b). We applied the method developed in (82) that measures the similarity of summary data (i.e., ROH spectrum) between the observed and simulated data based on a kernel density estimate and that calculates an approximated marginal likelihood (aML) of a given model. A normalized Gaussian kernel function with a bandwidth of 1.0 was used to compute kernel density estimates of aMLs. These estimates were then compared to identify the best-fitting model as an approximate Bayes factor (aBF); we calculated aBFs between a model with the highest aML and any other models.
Our broad search was able to confine the parameter space into 500 ≦ N ≦ 2,000 and 15 ≦ T ≦ 30 ka ago that includes likely scenarios for further testing with all autosomal chromosomes (Fig.  S11). Including chromosomes 1 and 2 increased the power in discriminating the highest likelihood model from the others (Fig. 3c); the model with N = 1,000 and T = 20 ka ago was significantly favoured with log10-scaled aBFs >2.0 against any other models, except for that with N = 1,000 and T = 15 ka ago (log10-scaled aBF: 0.1).

Note S3: Testing the presence of addition ancestry in the Kofun
We rigorously tested whether the Kofun have additional ancestry that is absent or reduced in modern Japanese using qpAdm. We modelled the Kofun and modern Japanese with a four-way admixture by adding populations identified from the f4-statistics as significantly closer to the Kofun (Fig. S21) to the three sources of Jomon, Northeast Asian, and East Asian ancestry (Table  S15). None of the populations tested support the four-way admixture models; the three-way admixture without the fourth source always better explains the genetic ancestry of Kofun than the four-way admixture (nested p-value > 0.05) (see Table S15). These results suggest that no additional ancestral component is present at any detectable level in the Kofun.    Table 1 and Table S1, and information about all other previously published data is summarized in Table S4.            (a) f4(Mbuti, X; Jomon_Initial, Jomon_Early) Fig. S13. Geographic and temporal display of differences in affinities between Jomon subperiods with ancient and modern populations using f4-statistics. We investigate potential gene flow from continental populations throughout different stages of the Jomon period using f4(Mbuti, X; sub_Jomoni, sub_Jomonj), where i and j are any pairs of the three Jomon sub-groups. The three sub-periods are defined as follows: Initial Jomon (JpKa6906), Early Jomon (JpFu1, JpOd6, JpOd181, JpOd274, and JpOd282) and a merged group for all Middle, Late and Final Jomon (F5, F23, IK002, JpHi01, JpKo2, and JpKo13). Populations genetically closer to the later sub-period when compared to the earlier one with Z > 3.0 are designated by red triangles and those symmetrically related to both are designated by gray circles. The populations tested are split into four different periods, depending on their ages: Upper-Paleolithic (>16,000 years BP), Jomon (from 16,000 to 3,000 years BP), Post-Jomon (from 3,000 years BP to the present), and Presentday. There are almost no significant results in this analysis, suggesting that the Jomon were genetically isolated from the rest of the continent from the Initial to Final Jomon period. The Igorot population from the Philippines were found to be significantly closer to the Middle-Late-Final . The results show a remarkable level of consistency in results between each pair of islands; however there appears to be a slightly greater affinity for Yayoi samples (from northern Kyushu) to Honshu and Shikoku when compared to the geographically distant Rebun Island, which is consistent with Fig. 4c.      Table S9.    Tables S10 and S17.      Note: Source population is highlighted by bold if the admixture is supported from modelling with a given set of the reference populations or by bold and red if the admixture is supported from modelling with and without a subset of Jomon included in the reference populations.     Note: Source population is highlighted by bold if the admixture is supported from modelling with a given set of the reference populations or by bold and red if the admixture is supported from modelling with and without a subset of Jomon included in the reference populations.  Table S12. Modelling on the genetic ancestry of the Kofun individuals as a mixture of two sources (Jomon and additional source) by qpWave and qpAdm. Note: Source population is highlighted by bold if the admixture is supported from modelling with a given set of the reference populations or by bold and red if the admixture is supported from modelling with and without a subset of Jomon included in the reference populations.   *These populations have an alternative model of three-way admixture shown in the parentheses that includes Jomon, East Asian ancestry, and additional source tested, instead of Jomon, Northeast Asian, and East Asian ancestry, likely due to their shared ancestry with the source population representing Northeast Asian ancestry (WLR_BA_o + HMMH_MN).  Table S17. Modelling on the genetic ancestry of present-day Japanese as a mixture of three sources by qpWave and qpAdm. Note: Source population is highlighted by bold if the admixture is supported from modelling with a given set of the reference populations or by bold and red if the admixture is supported from modelling with and without a subset of Jomon included in the reference populations.