The genetic history around the southeastern Mongolian Plateau traces Neolithic cultural diffusions in northern East Asia

Abstract

Cultural and material exchange between groups across the mountainous region separating the East Asian Steppe and the plains and river valleys of northern East Asia has been documented since the Paleolithic, yet the extent to which these interactions reflect prehistoric population dynamics is unknown. By sequencing and analyzing 35 ancient genomes from the southeastern Mongolian Plateau, spanning from 8,800 to 5,000 years ago, we found that Early Holocene populations from the southeastern Mongolian Plateau shared a common ancestry. We show this ancestry, which was predominant in steppe populations before the Holocene and may have been associated with the post-Last Glacial Maximum (LGM) microblade dispersal, to have lasted on the southeastern Mongolian Plateau until between 7,500 and 5,700 years ago, during which time it contributed to the West Liao River basin populations associated with the Hongshan culture. The continuity of the Early Holocene southeastern Mongolian Plateau ancestry was later disrupted by genetic influxes from both the northeastern Mongolian Plateau and West Liao River Hongshan populations, coinciding with cultural diffusion between 5,700 and 5,000 years ago. These revealed complex genetic and cultural interactions along the Eastern Steppe in the period between the LGM and the Middle Holocene.

Graphical abstract

Public summary

• •Population stability on the southeastern Mongolian Plateau (MP) from 8,800 to 7,500 before present (BP).
• •Admixture on the MP during the Terminal Pleistocene is associated with the spread of microblade technology.
• •Population and cultural shifts on the southeastern MP before 5,700 BP.
• •Southeastern MP ancestry contributed to the development of Middle Neolithic cultures in the West Liao River basin.
• •The southwestern dispersal of the Hongshan culture is associated with population expansion.

Introduction

After the Last Glacial Maximum (LGM) (∼26.5–19.0 ka [thousand years] BP [before present]),¹ populations increased in response to climate warming and improved habitat conditions from the Terminal Pleistocene to the Holocene.^2,3 The proliferation of Terminal Pleistocene archaeological sites in Siberia,⁴ northern China,⁵ Mongolia,⁶ and the Lake Baikal region⁷ provides evidence of intensified human occupation, marking a crucial stage in the demographic history of East Asians. Microblade technology, which spread across Northeast Asia after the LGM with adaptive advantages to the changing environment, played a pivotal role during this period^3,8,9,10,11 and has been widely used to infer human dynamics in post-LGM East Eurasia. Based on microblade chronology, some scholars have proposed human expansion into the Trans-Baikal region around 15.0 ka BP,⁷ while others have suggested two episodes of population contact between Northeast and North China before and after the LGM.¹² Additional hypotheses include the roles of the Gobi Desert¹³ and Mongolia¹⁴ as key conduits for cultural and demographic exchanges across Northeast and East Asia. In the subsequent Holocene, continuously improving environmental conditions supported the expansion of human populations and the rise of Neolithic cultures, characterized primarily by the adoption of agriculture and pottery.^13,15,16 The utilization of millet dates back as early as 7,600 years ago in the East Gobi^17,18,19,20 and to 6,700–6,100 years ago in the eastern Eurasian Steppe,^13,14,21,22 both likely introduced from northern China. Some studies have identified shared pottery styles across East Asia,^21,22,23 while others have suggested specific ceramic influences between the Baikal region and northeast China,^24,25 as well as from the Amur region to the Trans-Baikal area.²⁶ In addition, although the origin and dispersal of jade (nephrite) in Northeast Asia remain controversial, some archaeologists have found similarities among Neolithic nephrite artifacts from the Russian Far East,²⁷ the Baikal region,^24,25,28 and Northeast China.²⁹ In summary, the dispersal of microblade technology, the Neolithization process, and cultural interactions in prehistoric East Asia have been central topics of archaeological inquiry, and remain subjects of ongoing debate.^13,14

These long-term cultural interactions point to past population contacts in northern East Asia. Recent paleogenomic studies have also provided new insights into these population dynamics. With the end of the LGM, recurrent population shifts have been proposed in previous studies from the Mongolian Plateau and southern Siberia. These shifts have often been described as the replacement of populations with ANE (Ancient North Eurasian) ancestry, a lineage deeply diverged from present-day East Asians and distributed across southern Siberia from ∼24.0–14.7 ka BP,^30,31 by those with APS (Ancient Paleo-Siberian) ancestry, which spanned Siberia from the Terminal Pleistocene to the Early Holocene, contributing to present-day native Americans, and formed by the admixture between ANE and ancient East Asians.^32,33,34,35 During the Holocene, populations enriched in ANA (Ancient Northeast Asian) ancestry, which is closely related to present-day northern East Asians, became predominant in the region.^35,36,37,38 In contrast, the Amur River region exhibited long-term genetic continuity of ANA ancestry populations.^39,40 Further south of the Mongolian Plateau and the Amur River region, an 8,400-year-old individual from the southeastern Mongolian Plateau also exhibited ANA ancestry,⁴¹ suggesting that ANA ancestry was a dominant genetic component across a wide geographic area during this period. Despite these genetic insights, the population history underlying the region’s cultural interactions remains poorly understood, largely due to limited genetic research examining relationships among diverse ancestral groups. Another study found that these diverse ANA ancestry populations shared complex genetic connections.⁴² highlighting the need for fine-scale ancient genomic study in this region to clarify the genetic history and uncover the population dynamics underlying northern East Asian cultural interactions.

The southeastern Mongolian Plateau lies northwest of where the Yin Mountains meet the Great Khingan mountain range, which connects the Mongolian Plateau and the Amur River region to the north, the West Liao River valley to the east, and the Yellow River valley to the south. As a geographic crossroads among these regions, the area witnessed the emergence of diverse cultures, as represented by the Xinglong, Sitai, Zhengjiagou, and Leigongshan sites. During the Late Paleolithic period, Xinglong and Sitai yielded microblade assemblages similar to those found on the Mongolian Plateau, in the Baikal region, and in northern China.^20,43 In the Early Holocene (∼11.6–7.0 ka BP),⁴⁴ Neolithic cultures emerged at these sites. Xinglong was associated with the Yumin culture, which was also found at the Yumin site and distributed across the southeastern Mongolian Plateau from 8,800 to 7,000 BP. Sitai exhibited a distinct material assemblage that differed from Yumin-associated sites, referred to as the Sitai culture, which dates from 7,740 to 7,650 BP (supplemental information S1).^{17,19,20,43,45} In addition to the shared ceramic traditions across East Asia, some archaeological studies have suggested a link between the Yumin culture and the Baikal region based on shared pottery styles (Figure S9).^20,45 The remains of broomcorn millet (Panicum miliaceum) and foxtail millet (Setaria italica), along with farming tools unearthed at Xinglong and Sitai, document a transition from foraging to initial and, later, to more advanced farming practices, likely reflecting the expansion of agriculture from the North China Plain.^18,20,46,47 Later in the Middle Holocene (∼7.0–5.0 ka BP),⁴⁸ Sitai shared ceramic similarities with the Baikal region,^17,19,20 while the Zhengjiagou and Leigongshan sites demonstrated strong cultural affinities with the Hongshan culture of the West Liao River basin, particularly through unique jade artifacts and stone-pile tombs.⁴⁹ These extensive cultural connections highlight the southeastern Mongolian Plateau as a key region for studying the population history underlying cultural diffusion in northern East Asia. However, the lack of genetic data from this period and region hampers our understanding of the genetic consequences of these interactions on prehistoric populations. Although the Early Holocene represents a critical period of population growth and Neolithization,^50,51,52 only 19 Early Holocene genomes have been reported from northern China, including just one from the southeastern Mongolian Plateau⁴¹ and the remaining 18 from the Amur River basin.^39,40

To discover the population interactions of the southeastern Mongolian Plateau and its surrounding regions, and address the origin of the cultural material on the southeastern Mongolian Plateau, we have generated a time series of Neolithic genomes from the sphere of interaction between the southeastern Mongolian Plateau populations and those of the neighboring areas.

Materials and methods

Ethics statements

We collected ancient human bones in Hebei province, China, in cooperation with the local archaeologists who excavated them. The permission to sample their DNA was granted by local archaeological institutes that managed and cared for the human specimens. Further approval and supervision were granted by the Institutional Review Board at the Institute of Vertebrate Paleontology and Paleoanthropology of the Chinese Academy of Sciences for ancient genome sampling (202402200015).

Ancient DNA extraction, in-solution capture, and sequencing

Human bones from 35 individuals were extracted using less than 100 mg of bone powder, strictly following a previously published protocol.⁵³ All wet laboratory work and data analysis were performed with equipment from the Molecular Paleontology Laboratory, Institute of Vertebrate Paleontology and Paleoanthropology, Chinese Academy of Sciences. We constructed either double-stranded (DS) libraries with partial uracil-DNA-glycosylase (UDG) treatment or single-stranded (SS) libraries without UDG treatment or SS libraries with full UDG treatment for 34 samples, and both DS-partial-UDG and SS-no-UDG libraries for one sample. Human DNA enrichment was performed by hybridizing the libraries with oligonucleotide probes. We captured the whole mitochondrial genome with mtDNA probes⁵⁴ and 1.2 million SNPs (the “1240k” panel) for nuclear DNA.⁵⁵ One individual was shotgun sequenced. The mtDNA libraries were sequenced on Illumina Miseq with 2 × 76 base pair (bp) paired-end reads, and the nuclear DNA was sequenced on Illumina HiSeq X Ten with 2 × 150 bp paired-end reads.

Data processing and mapping

The raw data were first trimmed to remove adaptors and paired-end reads that overlapped at least 11 bp were merged into a single sequence using leeHom.⁵⁶ Merged sequences were then aligned to the reference genome using the BWA (0.5.10)⁵⁷ bam2bam command. The revised Cambridge Reference Sequence 53 (rCRS53)⁵⁸ and hg19⁵⁹ were used as references. We retained the mapped fragments with mapping quality scores of at least 30, and removed duplicate fragments that mapped to the same start and end with the same orientation.

Contamination estimation and pseudo-haploid genotyping

We first calculated the rates of C to T substitution to authenticate ancient DNA. Then, we estimated the present-day mtDNA contamination for each individual using ContamMix.⁶⁰ For male individuals, we also estimated the contamination of the X chromosome by ANGSD.⁶¹ Contamination rates of either mitochondria or X chromosome higher than 5% were considered as contaminated data. For these data, we restricted the dataset to only the damaged fragments using pmdtools⁶² to reduce the present-day DNA contamination, and then re-estimated the mtDNA contamination rates. If these were still above 5%, we excluded the data from later genotyping. For data passing contamination filters, we masked transition sites within 5 bp on both ends for SS-no-UDG library data and 2 bp for DS-partial-UDG library data. For SS-full-UDG library data, we masked transition sites at 1 bp on the 5′ end and 2 bp on the 3′ end. Then we used pileupCaller with default parameters to call the pseudo-haploid genotypes by randomly sampling one fragment per position.

Sex and uniparental haplogroups determination

The genetic sex was determined from the proportion of sequences mapping to the Y chromosome among all sequences mapped to sex chromosomes with a mapping quality of at least 37.⁶³ Each individual’s mitochondrial haplogroup was assigned to PhyloTree Build 17⁶⁴ using HaploGrep2,⁶⁵ and Y chromosome haplogroups were assigned to the ISOGG 2019 nomenclature using Yleaf⁶⁶ with the option “-q 37 -r 1 -old”.

Present-day dataset compilation

We collected 115 worldwide present-day populations from the Human Origin (HO) SNP Panel,⁶⁷ the Simons Genome Diversity Panel (SGDP),⁶⁸ the Human Genome Diversity Project (HGDP),⁶⁹ Tibetan and Han populations,⁷⁰ Southeast Asian populations,⁷¹ and southern China populations.³⁸ We integrated the present-day and ancient populations using the mergeit program from the EIGENSOFT package.⁷² Given the relatively low coverage and quality of ancient DNA data, present-day populations were used to establish a reference framework of human genetic variation, to which ancient samples were subsequently co-analyzed in principal-component analysis (PCA) and ADMIXTURE analysis, which require high-quality comparative datasets.

Relatedness estimation

To control for the effects of background relatedness within the population, we combined our new data with published ancient East Asian data and then used READv2⁷³ and BREADR⁷⁴ to estimate their kinship with default parameters. For each pair of identical, first- or second-degree kinship individuals estimated in our new data, we excluded the individual with fewer SNPs from population analyses.

PCA

The PCA was performed using the smartpca program from the EIGENSOFT package⁷² with lsqproject: YES, numoutlieriter: 0, and shrinkmode: YES. We set two present-day panels to calculate the principal components, one of which comprised 82 populations across Eurasia, the other comprised 50 Asian populations, and the ancient genomes were projected onto them.

Unsupervised ADMIXTURE analysis

We performed unsupervised individual-level clustering on 317 ancient genomes and 656 present-day genomes across Eurasia using ADMIXTURE.⁷⁵ We first pruned the dataset for high linkage disequilibrium (r² > 0.4) using PLINK (v.1.90),⁷⁶ with parameters “--indep-pairwise 200 25 0.4”. Then, we set the K values ranging from 2 to 12 to perform the ADMIXTURE analysis, which we ran 30 times for each K value to determine the optimal K value using the lowest cross-validation error (Figure S2).

Maximum likelihood phylogeny with migration events

We constructed the maximum likelihood phylogeny with migration events with TreeMix (v.1.13)⁷⁷ based on population allele frequency. We set the root of phylogenetic trees as Mbuti, set blocks of 500 SNPs by “-k 500”, set migration events of zero to three, and ran 1,000 replicates to get the bootstrap values by “-bootstrap -q” (details in supplemental information S4). The consensus tree was obtained by the “consense” program of Phylip.⁷⁸

f statistics

To measure allele sharing between populations, we calculated the f3 and f4 statistics using the qp3pop and qpDstat programs in AdmixTools,⁶⁷ respectively, using the present-day Central African Mbuti as the outgroup. We formed the f3 statistics as (Mbuti; X, Y), and the f4 statistics as (Mbuti, X; Y, Z) with the parameter “f4mode: YES”.

qpAdm

To study population admixture, infer ancestral sources, and quantify their genetic contribution to our target populations, we employed the qpAdm program in AdmixTools⁶⁷ with the option “allsnps: YES”. We designed the in silico modeling by setting a set of fixed outgroups. To identify distal and proxy sources for populations in different time periods, we designed different source population sets. Additionally, we used the “rotating scheme” in qpAdm analyses to increase the sensitivity and resolution, in which some of the source populations will be treated as the potential sources in qpAdm, and the others will be treated as outgroups. We tried one- to three-way admixture events using every possible combination of source populations (details in supplemental information S5).

Admixture time estimation

We estimated the admixture time using the LD-based method DATES,⁷⁹ in which we set the parameters “binsize: 0.001, maxdis: 1.0, qbin: 10, lovalfit: 0.45, numchrom: 23”, and we calculated the admixture time from the inferred admixture generations by assuming one generation of 28 years.⁸⁰ Since DATES worked on a two-way admixture model, we pooled two of the Neolithic northern Mongolian Plateau and Baikal population’s three source populations following the method’s guide.⁷⁹ Inferences with absolute Z scores greater than 2 and NRMSD less than 0.7 were treated as acceptable, and others were treated as failed inferences.

Results

In this study, we successfully retrieved genome-wide data from 35 ancient individuals at 4 sites located on the southeast edge of the Yin Mountains and the Mongolian Plateau, including 12 from Xinglong, 13 from Sitai, 8 from Zhengjiagou, and 2 from the Leigongshan site (Figure 1A; supplemental information S1). Carbon dates for 10 of the 35 individuals ranged from 8,821 to 4,956 calibrated years before present (cal. BP, BP means before 1950, Figure 1B; Table S1). Among these, 22 individuals date to between 8,800 and 7,500 years ago, more than doubling the number of available ancient genomes from Early Holocene northern China, and providing a temporally dense dataset for investigating the genetic history of this pivotal period on the southeastern Mongolian Plateau. Following previously reported methods, we enriched for DNA covering the 1,240k SNP panel.^55,81 To ensure the authenticity of ancient DNA, we estimated the present-day DNA contamination rates of mitochondrial DNA for each individual and chromosome X in males. To increase the reliability of the analysis, individuals with estimated contamination rates above 5% and total SNP counts below 20,000 were removed, leaving 30 individuals for the downstream analyses having 0.04- to 6.16-fold coverage (Table S1), and 41,814 to 981,727 SNPs.

Figure 1.

Chronological and geographic information about newly generated ancient genomes

(A) The geographic locations of newly sequenced ancient individuals and spatio-temporally related published groups. Populations newly sequenced in this study are shown in solid shapes.

(B) The ages of newly sampled populations. The shapes represent the mean ages of dated samples, and the bars are the 95% confidence interval (CI) for radiocarbon dating samples at each site. The numbers in the brackets represented the number of individuals successfully sequenced.

Spatially distinct population structure across the Mongolian Plateau before 7,500 years ago

The Mongolian Plateau and the adjacent Lake Baikal region, Amur River (AR) region, and northern China served as centers for the emergence of distinct Neolithic cultures in northern East Asia.^20,26,82,83 To explore their origins and connections, we first examined the genetic history across these regions. The Early Neolithic genomes from across this region, spanning the Russian Far East (DevilsCave_N³³ and Boisman_MN³⁸), the AR basin (AR14K,⁴⁰ AR13-10K,⁴⁰ and ARpost9K⁴⁰), the Mongolian Plateau (Mongolia_N_North³⁸ and Mongolia_N_East³⁸), and the Baikal region (Cis_Baikal_8980_8640_BP³⁵ and Shamanka_EN³⁷), were closely related, forming a coherent cluster in our PCA analysis (Figure 2A) and in the maximum likelihood phylogeny (bootstrap support: 746/1000, Figure 4B). This broadly shared genetic profile corresponds to the previously defined ANA ancestry.³⁶ Despite this overall similarity, we detected clear genetic substructures among populations from the southern, northern, and eastern parts of the Mongolian Plateau in PCA, ADMIXTURE, and outgroup-f3 statistics (Figures 2A, 2B, 3, 4A, and S1). In PCA and outgroup-f3 clustering (Figures 2A, 2B, and 4A), the southeastern Mongolian Plateau (represented by Yumin,⁴¹ Xinglong8.8-7.5k, and Sitai7.6k), the AR region together with the Russian Far East to the east (denoted as the AR groups and AR ancestry, represented by AR14K,⁴⁰ AR13-10K,⁴⁰ ARpost9K,⁴⁰ DevilsCave_N,³³ and Boisman_MN³⁸), and the northern Mongolian Plateau and its surrounding Baikal region (mainly represented by Cis_Baikal_8980_8640_BP,³⁵ Shamanka_EN,³⁷ Fofonovo_EN,³⁶ Mongolia_N_North,³⁸ and Mongolia_N_East³⁸) formed three genetic clusters. This is supported by the ADMIXTURE result, which shows that northern and eastern Mongolian Plateau groups shared distinct ancestral components that were maximized in each. In contrast, the southeastern Mongolian Plateau groups shared two ancestral components that were maximized among northern Mongolian Plateau populations and East Asians, respectively (Figure 3). Additional tools were then used to better resolve their genetic structure and population history.

Figure 2.

Genetic landscape of newly sequenced ancient populations

PCA of ancient populations projected onto present-day Eurasians (A) and East Asians (B). The legends apply to both (A) and (B). Present-day populations are shown in gray, while previously published and newly sequenced ancient samples are shown by colored dots.

Figure 4.

Genetic clustering of diverse ancient populations

(A) Outgroup-f3 clustering heatmap using f3(Mbuti; X, Y). New samples from this study are shown in red. Outlined boxes show clusters containing newly sequenced samples.

(B) Maximum likelihood phylogeny with 1 migration event inferred from TreeMix. The bootstrap values from 1,000 replicates are marked on the nodes.

Figure 3.

ADMIXTURE plot for ancient populations with K = 10

Ancient and present-day populations were co-analyzed. Represented ancient East Asians from the southeastern Mongolian Plateau, Amur River basin, Siberia, Baikal region, Mongolian Plateau, coastal China, Yellow River basin, West Liao River basin, southern China, and central and west Eurasia are shown here. The full ADMIXTURE results can be found in Figure S1.

To study the genetic characteristics and internal genetic relationships of the southeastern Mongolian Plateau cluster, we first applied PCA and outgroup-f3 clustering to Xinglong8.8-7.5k and Sitai7.6k. Both PCA (Figures 2A and 2B) and outgroup-f3 (Figure 4A) results suggest that Xinglong8.8-7.5k and Sitai7.6k share close genetic affinity and cluster with the published northern East Asian 8,400-year-old Yumin individual,⁴¹ which was also confirmed by maximum likelihood phylogeny of 1,000 bootstrap TreeMix (951/1,000), suggesting substantial genetic similarity and continuity between them (Figure 4B). To further test if they were sister groups with each other or contained any diverse genetic components, we next calculated f4(Mbuti, Xinglong8.8-7.5k/Sitai7.6k; other ancient Eurasians, Yumin) > 0, 4.3 ≤ Z ≤ 59.6; f4(Mbuti, Yumin; other ancient Eurasians, Xinglong8.8-7.5k/Sitai7.6k) > 0, 4.3 ≤ Z ≤ 58.8; f4(Mbuti, other ancient Eurasians; Yumin, Xinglong8.8-7.5k/Sitai7.6k) ∼ 0, −2.0 ≤ Z ≤ 2.1 (Figure 5A; Tables S2–S4), demonstrating that Xinglong8.8-7.5k and Sitai7.6k could act as sister groups with the published northern East Asian Yumin and had no extra genetic affinity with other populations. In support of this, both Xinglong8.8-7.5k and Sitai7.6k can best be modeled with Yumin as a single ancestry source in qpAdm (p = 0.12, 0.13, respectively, Table S11), showing a shared common ancestry among them. Given that millet farming had spread to the Yumin and Sitai cultures,²⁰ we reassessed the relationship between groups from these cultures and the earliest known millet farming ancestry (YR_MN³⁹) to examine the population dynamics underlying agricultural dispersal. In f4 statistics, the southeastern Mongolian Plateau groups exhibited no higher genetic affinity with the YR farmers compared with other northern East Asians, i.e., f4(Mbuti, YR Yangshao farmers; Neolithic northern East Asians, Xinglong8.8-7.5k/Sitai7.6k) ≤ 0, −10.2 ≤ Z ≤ 1.1 (Table S5). These results reveal a high genetic continuity on the southeastern Mongolian Plateau without any other ancestry contributions beginning at least 8,800 years ago (hereafter, this Yumin-related ancestry is denoted as the Early Holocene southeastern Mongolian Plateau ancestry). This ancestry formed a distinct genetic cluster from the AR groups, representing a unique ANA ancestry.

Figure 5.

Genetic continuity on the Early Holocene southeastern Mongolian Plateau and dating its post-LGM genetic contribution to the Eastern Steppe

(A) The f4 statistics of Xinglong8.8-7.5k, Sitai7.6k, and Yumin define them as sister groups. f4(Mbuti, Yumin; Xinglong8.8-7.5k/Sitai7.6k, Test) < 0, f4(Mbuti, Test; Yumin, Xinglong8.8-7.5k/Sitai7.6k) ∼ 0, f4(Mbuti, Xinglong8.8-7.5k/Sitai7.6k; Test, Yumin) > 0. Different f4-statistic formulas are represented by shape, and blue and red represent Xinglong8.8-7.5k and Sitai7.6k, respectively. Dashed vertical lines represent Z scores of −3 and 3. Different-colored labels on the y axis represent deeply diverged Asians, ancient southern East Asians, ancient Shandong groups, ancient YR groups, ancient WLR groups, ancient AR and Far Eastern groups, ancient Mongolian Plateau and Siberian groups, and ancient west and central Eurasians, from bottom to top.

(B) The DATES results of dating the admixture time of Cis_Baikal_8980_8640_BP and AR_EN. All feasible results are shown. The dots represent the estimated admixture time when using different source groups and targets in DATES analysis, and the bars represent ±1 SE. Different colors represent different target populations, and the different pooling strategies are marked on the right.

We then investigated the genetic signature of the cluster formed by northern Mongolian Plateau populations and the nearby Baikal regions. In PCA, compared with the southeastern Mongolian Plateau populations, the northern populations deviate toward the APS (as represented by Yakutia_Lena_16900_16450_BP,³⁵ UKY,³² and Kolyma³³) and the AR groups (Figures 2A and 2B). f4 statistics and qpAdm analysis of northern Mongolian Plateau and Baikal groups revealed that both Early Holocene southeastern Mongolian Plateau ancestry (29.9%) and AR ancestry (47.6%) are present in the oldest analyzed Early Holocene Neolithic individual on the northern Mongolian Plateau, Cis_Baikal_8980_8640_BP,³⁵ with an additional genetic contribution from APS (22.5%) (Figures 6A and S3; Table S11). Furthermore, the other groups from the northern Mongolian Plateau and Baikal region can all be modeled by these three genetic ancestries (Figures 6A and S6; Table S11; supplemental information S3), which was validated by the TreeMix inferring gene flow from AR groups to the northern Mongolian Plateau (Figure 4B), consistent with prior studies^42,84 (denoted as the northern Mongolian Plateau-like ancestry). Through increased sampling on the southeastern Mongolian Plateau, we identified a new genetic cline encompassing diverse ANA groups and revealing genetic differentiation among them. This cline extended from AR groups to populations on the southeastern Mongolian Plateau and contributed to the formation of northern East Asian ancestry on the northern Mongolian Plateau and Baikal region (Figures 2B, 6A, and S6; Table S11; supplemental information S3). The appearance of ancestry found on the southeastern Mongolian Plateau and the Amur region in regions to the north genetically links the Mongolian Plateau’s southern, northern, and the Baikal regions. However, the lack of genomic data across the broad intermediate region between the southeastern and northern Mongolian Plateau limits our understanding of which populations are directly associated with this process. In addition, genomic data predating 10,000 years ago in this region remain scarce, leaving the post-LGM population history and the emergence of the Early Holocene southeastern Mongolian Plateau ancestry largely unresolved. Nevertheless, with the new southeastern Mongolian Plateau genomes, we also observed that a previously described ∼7,400-year-old AR ancestry group from Wuqi and Zhalainuoer sites near Hulun Lake on the northeast Mongolian Plateau, AR_EN,³⁹ shifted toward the Early Holocene southeastern Mongolian Plateau genomes and fell along our newly defined genetic cline extending from the Early Holocene southeastern Mongolian Plateau populations to the AR populations (Figures 2B and 3). Similarly, we used f4 statistics and qpAdm to re-examine the ancestral components of AR_EN. We found that it can be modeled as 28.7% AR ancestry and 71.3% northern Mongolian Plateau-like ancestry (Figures 6A and S4; Table S12), demonstrating the role of admixture from surrounding regions in shaping the genetic structure of the northern Mongolian Plateau and Baikal region.

Figure 6.

The genetic composition of the southeastern Mongolian Plateau and surrounding regions

(A) The genetic composition of the Early Holocene Mongolian Plateau, as inferred from qpAdm, was shaped by the Early Holocene southeastern Mongolian Plateau ancestry, AR ancestry (represented by DevilsCave_N or ARpost9K), and APS-related ancestry.

(B) The genetic composition of the Middle Holocene southeastern Mongolian Plateau, as inferred from qpAdm, was shaped by genetic contributions from the northern Mongolian Plateau and the WLR basin.

We then tried to date the admixture events to better understand the population dynamics. The genetic profile of Cis_Baikal_8980_8640_BP was determined to be a three-way admixture of southeastern Mongolian Plateau, APS, and AR ancestry. Using DATES, we estimated admixture times by pooling all possible pairs among the three source ancestries (materials and methods). When pooling APS with either AR or southeastern Mongolian Plateau ancestry, although estimates varied depending on the representative populations used, the estimated admixture dates ranged from 11,767 to 14,235 years ago. In contrast, pooling AR with southeastern Mongolian Plateau ancestry yielded more recent admixture dates, ranging from 9,258 to 11,146 years ago. This temporal discrepancy suggests that AR and southeastern Mongolian Plateau ancestries may have initially admixed during the Terminal Pleistocene, followed by subsequent admixture with APS ancestry in the Early Holocene. For the eastern Mongolian Plateau AR_EN, we used its AR and northern Mongolian Plateau-like ancestry to infer the admixture time, and we found the admixture time dated back to ∼11,500 years ago (Figures 5B and S6; Table S17). The admixture times point to human interactions between different populations across the Mongolian Plateau at some points between the close of the LGM and the Early Holocene warming period, which followed the Younger Dryas cold event (12.9–11.7 ka BP).⁸⁵ Future refinement of this date will be crucial to determine whether this admixture could be linked to the expansion of populations in periods of increased temperature and humidity (i.e., Bølling-Allerød [∼14–12.9 ka BP]⁸⁶ or Early Holocene [∼11.7–8.2 ka BP] warming periods⁸⁷).

Since we estimated that the population admixture on the Mongolian Plateau and in the Baikal region dates to the Terminal Pleistocene, we further investigated the genetic profiles of AfontovaGora3 (an 18,000-year-old ANE from southern Siberia³⁰) and UKY (a 14,000-year-old APS from the Ust-Kyakhta-3 site in the Trans-Baikal region³²). Compared with Malta1, a 24,000-year-old ANE individual from southern Siberia,³⁰ AfontovaGora3 did not show significant excess allele sharing with AR or southeastern Mongolian Plateau groups, as indicated by f4(Mbuti, southeastern Mongolian Plateau/AR; Malta1, AfontovaGora3) ∼ 0 (−0.7 ≤ Z ≤ 1.4, Table S21). In contrast, UKY, when compared with the 16,000-year-old APS individual Yakutia_Lena_16900_16450_BP,³⁵ showed significant excess allele sharing with AR and southeastern Mongolian Plateau groups, i.e., f4(Mbuti, southeastern Mongolian Plateau/AR; Yakutia_Lena_16900_16450_BP, UKY) > 0 (2.6 ≤ Z ≤ 4.9, Table S22), and can be described as the two-way admixture of approximately 87.4% Yakutia_Lena_16900_16450_BP ancestry and 12.6% AR ancestry in qpAdm (Table S13). The admixture between two distinct ancestries observed in UKY, APS ancestry, represented by Yakutia_Lena_16900_16450_BP, and AR ancestry, provides evidence for genetic diffusion along the Mongolian Plateau prior to 14,000 years ago.

Dramatic population shift on the southeastern Mongolian Plateau before 5,700 years ago

The genetic structure of the Early Holocene Mongolian Plateau provided insight into its interaction with cultural connections. We then investigated how this structure evolved in the subsequent Middle Holocene. Firstly, in PCA, Xinglong5.7k fell into the cluster of northern Mongolian Plateau and Baikal groups instead of the Yumin-related cluster of Xinglong8.8-7.5k (Figure 2B). Xinglong5.7k and the ∼8,000- to 6,500-year-old Mongolia_N_East³⁸ from the Kherlen River south bank and Norovlin Uul site on the northern Mongolian Plateau formed a robust clade in the maximum likelihood phylogeny (bootstrap support: 836/1000, Figure 4B), indicating their strong genetic affinity, which was also supported by outgroup-f3 clustering and ADMIXTURE analysis (Figures 3 and 4A). Among the northern Mongolian Plateau and Baikal populations, f4 statistics also showed that Xinglong5.7k shared significantly more alleles with Mongolia_N_East than other groups, i.e., f4(Mbuti, Xinglong5.7k; other groups, Mongolia_N_East) ≥ 0 (1.6 ≤ Z ≤ 53.7), f4(Mbuti, Mongolia_N_East; other groups, Xinglong5.7k) ≥ 0 (−0.1 ≤ Z ≤ 56.0), f4(Mbuti, other groups; Mongolia_N_East, Xinglong5.7k) ∼ 0 (−2.0 ≤ Z ≤ 2.5) (Figure S5; Table S6). Additionally, Xinglong5.7k and Mongolia_N_East were shown to be a genetically similar population, and could be modeled using the same sources with similar proportions in qpAdm (Figures 6A and 7B; Table S11). To further confirm this connection, we modeled a large rotating set of northern Mongolian Plateau and Baikal groups who shared the same three ancestry sources with Xinglong5.7k and Mongolia_N_East in qpAdm, and the Mongolia_N_East outcompeted other populations as a single source of Xinglong5.7k (Figures 6B and 7B; Table S14), favoring a population turnover linking southeastern and northeastern Mongolian Plateau instead of an in situ admixture scenario on the southeast. Collectively, these observations suggest a plausible southward human migration from the northern Mongolian Plateau, resulting in both population and culture turnover on the southeastern Mongolian Plateau by 5,700 years ago. Moreover, populations associated with the Miaozigou culture, who inhabited the southeastern Mongolian Plateau around 5,200 years ago (Miaozigou_MN),³⁹ exhibited YR farmer ancestry in a previous study³⁹ as well as our PCA and ADMIXTURE analyses (Figures 2A, 2B, and 3), again suggesting that the southeastern Mongolian Plateau experienced a dramatic population shift after 5.7 ka BP.

Figure 7.

The genetic dynamics of the Middle Holocene southeastern Mongolian Plateau and the West Liao River basin

(A) f4 statistics of f4(Mbuti, HMMH_MN/WLR_MN/Zhengjiagou/Leigongshan; AR/YR-related, Early Holocene southeastern Mongolian Plateau). The dots represented f4 values. Thick bars represent ±2 SE, thin bars represent ±3 SE.

(B) The ancestry composition of Mongolia_N_East and Xinglong5.7k inferred from qpAdm. Error bars equal ±1 SE. The AR ancestry is represented by DevilsCave_N here.

(C) The ancestry composition of HMMH_MN, WLR_MN, Zhengjiagou, and Leigongshan inferred from qpAdm, showing the Early Holocene southeastern Mongolian Plateau contribution to the Middle Holocene WLR and similarity between WLR and southeastern Mongolian Plateau Hongshan groups. Error bars equal ±1 SE.

A previous genetic study described a ∼5,600-year-old group from the West Liao River (WLR) basin, HMMH_MN, as the admixture of YR farmers and AR hunter-gatherers,³⁹ while in our PCA including the new southeastern Mongolian Plateau individuals, HMMH_MN fell into the cline of southeastern Mongolian Plateau groups and AR groups (Figure 2B), indicating their possible inclusion into southeastern Mongolian Plateau ancestry. In f4 statistics, HMMH_MN shared the most alleles with Early Holocene southeastern Mongolian Plateau populations, i.e., f4(Mbuti, HMMH_MN; AR/YR_MN/other groups, Early Holocene southeastern Mongolian Plateau groups) ≥ 0 (−1.9 ≤ Z ≤ 54.3, Figure 7A; Table S7). Compared with Early Holocene southeastern Mongolian Plateau groups, HMMH_MN shared more alleles with AR and northern Mongolian Plateau-like groups, i.e., f4(Mbuti, Cis_Baikal_8980_8640_BP/DevilsCave_N; southeastern Mongolian Plateau groups, HMMH_MN) > 0 (1.5 ≤ Z ≤ 2.8, Table S7). In addition, and in contrast to the previous study,³⁹ HMMH_MN did not show apparent genetic affinity with YR farmers, i.e., f4(Mbuti, YR_MN/YR_LN; southeastern Mongolian Plateau groups, HMMH_MN) ∼ 0 (0.2 ≤ Z ≤ 1.9, Table S7). We could model HMMH_MN’s genetic composition as an admixture of 65.8% Early Holocene southeastern Mongolian Plateau and 34.2% northern Mongolian Plateau-like ancestries by qpAdm (Figures 6B and 7C; Table S15), validating the f4 statistics results and suggesting a genetic connection between the WLR basin’s west and the southeastern Mongolian Plateau.

Southeastern Mongolian Plateau’s genetic connections with the WLR Hongshan populations

We used newly sequenced ancient genomes from two Hongshan culture sites, Zhengjiagou and Leigongshan, to investigate the genetic dynamics underlying the cultural evidence on the southeastern Mongolian Plateau.^49,88

We first compared the genetic affinity between a previously reported 5,200-year-old Hongshan group from the WLR basin, WLR_MN,³⁹ and 5,000-year-old southeastern Mongolian Plateau Hongshan groups, Zhengjiagou and Leigongshan. In PCA, outgroup-f3 clustering, and TreeMix, these three Hongshan culture groups clustered together (Figures 2B, 4A, and 4B), showing an affinity also supported by the individual-level clustering ADMIXTURE analysis (Figure 3). Moreover, these three Hongshan culture groups shared the most alleles in f4 statistics, i.e., f4(Mbuti, Hongshan groups; other groups, Hongshan groups) > 0 (0.5 ≤ Z ≤ 48.1, Table S8). Statistical analysis further identified them as sister groups to each other, with f4(Mbuti, other groups; WLR_MN, Zhengjiagou/Leigongshan) ∼ 0 (−2.4 ≤ Z ≤ 3.8, Table S9). Furthermore, the earlier Hongshan (WLR_MN) group could be modeled as a single-source population for both of the younger Hongshan groups (Zhengjiagou, Leigongshan) (Figures 6B and 7C; Table S16), suggesting the population connection between the Hongshan populations from the WLR basin and the southeastern Mongolian Plateau at least during 5,200 to 5,000 BP.

The high frequency of Y chromosome sub-haplogroups of N1 associated with Hongshan culture populations might indicate that N1 was once distributed broadly across northern China.⁸⁹ Intriguingly, we found that, among the 11 males’ Y-haplogroups of our newly generated Early Holocene southeastern Mongolian Plateau individuals, 10 of whom were N1 (Table S1), which might imply a genetic origin for Hongshan culture-associated populations. Previous studies characterized a WLR Hongshan culture group, WLR_MN,³⁹ as the admixture of AR hunter-gatherers and YR farmers. However, we found that the Hongshan culture groups shared excess alleles with the Early Holocene southeastern Mongolian Plateau groups compared with AR- and YR-related groups, i.e., f4(Mbuti, Hongshan groups; YR/AR groups, Early Holocene southeastern Mongolian Plateau groups) ≥ 0 (−1.7 ≤ Z ≤ 4.9, Figure 7A; Table S10). We then tested qpAdm to model the Hongshan populations’ genetic composition, and found that they could be best modeled by the two-way admixture of Early Holocene southeastern Mongolian Plateau ancestry and the coastal China Shandong hunter-gather ancestry (represented by 9,500- to 8,000-year-old populations associated with present-day East Asians and distinct from ANA ancestry)⁴¹ (Figures 7C and S12; Table S15; supplemental information S6), suggesting that the ancestry represented by the 8,800–7,500 BP southeastern Mongolian Plateau populations contributed to the 5,200 BP WLR Hongshan population’s gene pool and was a likely origin of their Y-haplogroup N1.

Discussion

Our analysis of the ancestry trajectories of Neolithic northern East Asia, resolved fine-structure population dynamics and enabled us to understand the driving forces behind the spread of differing prehistoric cultural practices across this ecologically diverse area. We identified a distinct ancestry shared across the southeastern Mongolian Plateau that maintained population stability from ∼8.8 to 7.5 ka BP in Yumin,⁴¹ Xinglong8.8-7.5k, and Sitai7.6k. However, despite the genetic continuity over time, this ancestry was associated with various cultures practicing different intensities of millet-based agriculture.²⁰ Millet farming on the southeastern Mongolian Plateau was assumed to have been introduced from the North China Plain, where the earliest domesticated millets have been found, dating to before ∼10 kya.^18,46,47 Our genetic results showed that this Early Holocene introduction was not the result of an influx of YR farmers, indicating a non-demic diffusion model or in situ innovation. This pattern is consistent with scenarios proposed in previous studies for the independent development of farming in South Asia⁹⁰ and Anatolia,⁹¹ but contrasts with the demic diffusion observed in Japan.⁹² Furthermore, our findings challenge a previous paleogenomic study³⁹ that attributed the spread of agriculture in this region during the Middle Holocene to the expansion of farming populations, indicating that distinct mechanisms likely operated during the Early and Middle Holocene phases.

As a major ancestry across northern East Asia, the ANA ancestry predominated a broad region stretching from the Baikal region to the Russian Far East during the Early to Middle Holocene. However, fine-scale population dynamics within ANA groups remain poorly understood. Our detailed analyses resolved a substructure within ANA populations corresponding to distinct geographic areas, including the southeastern Mongolian Plateau, northern Mongolian Plateau, and the AR basin, providing new insights into the genetic diversity and admixture history of ANA groups. Populations on the northern Mongolian Plateau and the Baikal region between approximately 8.8 and 6.5 ka BP shared a genetic component with the Early Holocene southeastern Mongolian Plateau ancestry and AR ancestry, along with APS ancestry. We estimated that these admixture events occurred between ∼14.2 and 11.7 ka BP, likely in at least two distinct phases: an initial admixture between AR and southeastern Mongolian Plateau ancestries in the Terminal Pleistocene, followed by a subsequent phase of admixture with APS ancestry during the Early Holocene. This Terminal Pleistocene admixture is further supported by the genetic contribution from AR ancestry to the 14,000-year-old UKY individual from the Trans-Baikal region.³² Along with gene flow from northern Mongolian Plateau-like ancestry into an AR population (AR_EN)³⁹ by 7,300 years ago, these episodes of genetic interaction shaped the genetic landscape of the Mongolian Plateau and offer new insights into East Asian genetic prehistory. Previous paleogenomic studies have identified a shared ancestry component across Early Holocene East Asia, known as ANA ancestry.^35,36,37,38 Our analyses reveal fine-scale population structure within ANA groups and demonstrate that admixture played a central role in producing their shared genetic signatures. Prior phylogenetic studies on mtDNA⁹³ and Y chromosomes⁹⁴ have suggested post-LGM population dispersal across Northeast Asia, and our study further provides genome-wide evidence supporting this scenario. Moreover, the observed admixture between Yumin and AR ancestry with APS ancestry in the Early Holocene suggests that APS-related populations persisted across a broad area from southern Siberia to northern Mongolia during this period. To date, the youngest known APS population is represented by the 10,000-year-old Kolyma individual from northeastern Siberia.³³ Our findings further suggest that APS populations may have spanned a much wider region across Siberia during the Early Holocene than previously recognized.

The extensive admixture along the Mongolian Plateau also provides a plausible context for the widespread dispersal of microblade technology following the LGM. The lithic technology is consistent with technological trends observed among hunter-gatherers throughout Northeast Asia, primarily characterized by the appearance of this technology.¹³ This technological expansion has been linked to population expansion in archaeological studies,^9,95 and the admixture events we identified across Northeast Asia during the Terminal Pleistocene provide genomic support for a post-LGM demographic expansion concurrent with microblade dispersal. Archaeological studies have suggested a re-expansion into the desert region with the onset of post-LGM climatic amelioration, based on the chronology of microblade sites in the Gobi Desert.¹³ Our genetic results indicate that ancestry from the southeastern Mongolian Plateau, located southeast of the Gobi Desert, admixed with AR ancestry during the Terminal Pleistocene. The distribution of the genetic legacy of this admixture, which we traced in populations from northern Mongolia and the Baikal region, located north of the Gobi Desert, suggests that the admixture likely occurred in the Gobi region and extended across the Mongolian Plateau. This pattern supports the hypothesis of a post-LGM re-expansion into the broader region. Based on shared elements of microblade technology, some archaeologists have proposed population contacts between Northeast China and North China during and after the LGM.¹² The admixture between Early Holocene southeastern Mongolian Plateau ancestry and AR ancestry identified in our study provides genetic evidence supporting this post-LGM contact. The increasing occurrence of microblade technology found in the Trans-Baikal around 15,500 BP has been attributed to a Bølling-Allerød migration event,⁷ and the genetic admixture we detected of AR ancestry in the ∼14,000-year-old UKY individual from the Trans-Baikal provides direct evidence of human contact supporting this hypothesis.

Our genetic results also shed light on the Neolithic pottery interactions. Archaeologists have identified shared ceramic traditions across a broad region of East Asia, including northern China, Mongolia, the Baikal region, and the Amur region.^13,21,23 Population contacts we found across these areas may have facilitated these shared features. Specifically, based on the similarities in pottery styles between the Yumin culture of the Early Holocene southeastern Mongolian Plateau and those of the Baikal region, some have proposed that the Yumin culture originated in the Baikal valley.⁹⁶ However, the gene flow from the Yumin culture’s ancestry (Yumin, Xinglong8.5-7.5k) to the Baikal region challenges this interpretation, suggesting a more complex scenario of cultural diffusion. The pottery tradition of Yumin culture abruptly disappeared on the southeastern Mongolian Plateau after 7 ka BP.⁹⁶ Even at the same site, younger Xinglong remains (Xinglong5.7k) differed markedly from the Yumin culture Xinglong8.8-7.5k, likely adopting a highly mobile lifeway.⁴³ Genetically, Xinglong5.7k did not maintain the continuity observed in earlier Xinglong individuals, but instead showed population affinity with Mongolia_N_East from the northeastern Mongolian Plateau. The concurrence of the cultural and genetic shifts from 8.8–7.5 to 5.7 kya in this region suggests that a population turnover may have contributed to the significant cultural transformation observed here. However, further research is needed to test this hypothesis, as it is currently based on only two individuals from Xinglong5.7k. By contrast, the Yumin culture had extended eastward to the WLR basin by at least 5,000 years ago. Pottery from the 5,600-year-old Haminmangha site (HMMH_MN) exhibited cultural characteristics associated with the Yumin culture (Figure S11).⁹⁶ The genetic legacy of Yumin-related ancestry that we found in HMMH_MN suggests that populations associated with the Yumin culture may have contributed both genetically and culturally to the formation of the Haminmangha culture.

In addition, our genetic findings offer insight into the population dynamics underlying Neolithic jade cultures as well. The drilling technology at the Haminmangha site was proposed to be from the Baikal region based on its similarity.⁹⁷ We found the ancestry spanning from the Baikal region and the northern Mongolian Plateau made up ∼34% (represented by Cis_Baikal_8980_8640_BP) of the Haminmangha people’s genetic profile (HMMH_MN). Although the actual source of this ancestry in HMMH_MN may derive from a broad area across the Baikal region and northern Mongolian Plateau, this genetic composition suggests population contact between these regions and the WLR basin, supporting the cultural diffusion observation based on the jade artifacts. Moreover, the genetic origin and dispersal of the populations associated with the Hongshan culture, a predominant jade culture of China, remain unclear and debated. We show that the 5,200-year-old Hongshan culture population in the WLR basin (WLR_MN) shares the same ancestry with 5,000-year-old populations on the margin of the southeastern Mongolian Plateau (Zhengjiagou, Leigongshan). Archaeological evidence links these populations through the southward spread of Hongshan culture-associated jade and stone-pile tombs.⁴⁹ The genetic continuity between these groups that we observe supports this interpretation, which is consistent with another genomic study,⁹⁸ indicating this southward cultural diffusion might associate with population movements from the WLR basin to the southeastern Mongolian Plateau. Several hypotheses regarding the genetic origin of the WLR basin’s Hongshan culture have been proposed. The high-frequency N1 Y-haplogroup in Hongshan culture-associated populations was used to speculate a Northeast Asia or pre-Neolithic WLR region origin.⁸⁹ Genomic studies have inferred the ancestry of the Hongshan population as admixed with YR farmers and AR hunter-gatherers,³⁹ or Neolithic Mongolian hunter-gatherers,⁹⁸ based on different reference panels. Our study, based on a more reliable proxy ancestry source, finds that the Early Holocene southeastern Mongolian population played a crucial role in the formation of the Hongshan culture, which also explains the high N1 frequencies. We also found that the Early Neolithic Shandong hunter-gatherers’ ancestry contributed significantly to Hongshan-associated groups, while the genetic input from YR ancestry and other ANA ancestry to the Hongshan population remains possible (Table S15; supplemental information S6).

By reconstructing the prehistoric population dynamics of the southeastern Mongolian Plateau and illustrating the genetic connections with its surrounding regions, we reveal complex human interactions across East and Northeast Asia that were central to the exchange of crops, culture, technology, and genes from the Terminal Pleistocene to the Middle Holocene. We identified varying models of population-culture interaction among the groups inhabiting the region. In addition to ubiquitous demic diffusion, non-demic cultural transmission also occurred across the region. These interactions reveal complex human population dynamics underlying cultural practices such as microblade production, ceramic traditions, and jade craftsmanship throughout this period. Nevertheless, substantial chronological and geographic gaps in ancient genomic data remain in this region. Based on our current findings, the Early Holocene southeastern Mongolian Plateau ancestry may have extended widely across the Mongolian Plateau, and the timing of the establishment of the genetic cline between this ancestry and that of the AR region of East and Northeast Asia cannot be determined from the available data, highlighting the need for further multidisciplinary research to unravel the population prehistory and implications of these ancestral connections. In conclusion, our paleogenomic survey offers novel insights into the multifaceted interplay between population and cultural dynamics across East Asia.

Resource availability

Materials availability

This study did not generate new reagents.

Data and code availability

The mapped sequences are available at the Genome Sequence Archive⁹⁹ in the National Genomics Data Center¹⁰⁰ (https://ngdc.cncb.ac.cn/gsa-human/; accession no. PRJCA039072). The pseudo-diploid genotype calls (Eigenstrat format) are available at the OMIX, China National Center for Bioinformation/Beijing Institute of Genomics, Chinese Academy of Sciences (https://ngdc.cncb.ac.cn/omix; accession no. PRJCA039072). This study does not report original code.

Funding and acknowledgments

We appreciate the field work and archaeological study by the archaeological teams in Hebei Provincial Institute of Cultural Relics and Archeology, as well as the National Museum of China. This study was supported by the National Key R&D Program of China (2023YFF0905700) and the Chinese Academy of Sciences (YSBR-019).

Author contributions

Q.F. conceptualized and supervised this study. T.L., Z.Z., M.G., L.Z., W.Z., M.S., F.L., and Q.F. integrated the archaeological materials and conducted carbon dating. P.C., F.L., Q.D., W.P., and Q.F. performed and supervised the wet lab work. X.F. and Q.F. processed the data. T.L. and Q.F. analyzed the data. T.L., E.A.B., H.S., J.R., F.B., G.Z., T.W., and Q.F. discussed the results. T.L., E.A.B., and Q.F. wrote the manuscript. T.L., E.A.B., H.S., J.R., and Q.F. revised the manuscript. All authors reviewed, discussed, revised, and approved the final manuscript.

Declaration of interests

The authors declare no competing interests.

Published Online: November 20, 2025