The genetic history of Ice Age Europe.

Modern humans arrived in Europe ~45,000 years ago, but little is known about their genetic composition before the start of farming ~8,500 years ago. Here we analyse genome-wide data from 51 Eurasians from ~45,000-7,000 years ago. Over this time, the proportion of Neanderthal DNA decreased from 3-6% to around 2%, consistent with natural selection against Neanderthal variants in modern humans. Whereas there is no evidence of the earliest modern humans in Europe contributing to the genetic composition of present-day Europeans, all individuals between ~37,000 and ~14,000 years ago descended from a single founder population which forms part of the ancestry of present-day Europeans. An ~35,000-year-old individual from northwest Europe represents an early branch of this founder population which was then displaced across a broad region, before reappearing in southwest Europe at the height of the last Ice Age ~19,000 years ago. During the major warming period after ~14,000 years ago, a genetic component related to present-day Near Easterners became widespread in Europe. These results document how population turnover and migration have been recurring themes of European prehistory.

Modern humans arrived in Europe around 45,000 years ago and have lived there ever since, even during the Last Glacial Maximum 25,000-19,000 years ago when large parts of Europe were covered in ice[1]. A major question is how climatic fluctuations influenced the population history of Europe and to what extent changes in material cultures documented by archaeology and correlating to climatic events corresponded to movements of people. To date, it has been difficult to address this question because genome-wide ancient DNA has been retrieved from just five Upper Paleolithic individuals in Eurasia[2-4]. Here we assemble and analyze genome-wide data from 51 modern humans dating from 45,000 to 7,000 years ago (Table 1; Extended Data Table 1; Supplementary Information section 1).

Extended Data Table 1

The 51 ancient modern humans analyzed in this study

Sample Code	Data source	Country	Lat.	Long.	Cal BP 95.4%	Date type (ref.)	Culture	Remain	SNP Panel	Sex	mtDNA haplogroup	Y chrom. haplogroup	Genetic Cluster	Damage restrict	Mean coverage+	SNPs covered
UstIshim	[1]	Russia	57.43	71.10	47,480-42,560	Direct-UF ([1])	Unassigned	Femur	Shotgun	M	R	K (xLT)	Unassigned	No	42	2,137,615
Oase1	[2]	Romania	45.12	21.90	41,640-37,580	Direct-UF ([3])	Unassigned	Mandible	Shotgun	M	N	F	Unassigned	Yes	0.156	285,076
Kostenki14*	New	Russia	51.23	39.30	38,680-36,260	Direct-UF ([4])	Unassigned	Tibia	3.7M	M	U2	C1b	Unassigned	No	16.1	1,774,156
GoyetQ116-1	New	Belgium	50.26	4.28	35,160-34,430	Direct-NotUF ([5])	Aurignacian	Humerus	1240k	M	M	C1a	Unassigned	No	1.046	846,983
Muierii2	New	Romania	45.11	23.46	33,760-32,840	Direct-UF ([6])	Unassigned	Temporal	3.7M	F	U6		Unassigned	Yes	0.049	98,618
Paglicci133	New	Italy	41.65	15.61	34,580-31,210	Layer ([7])	Gravettian	Tooth	1240k	M	U8c	I	Vestonice	No	0.041	82,330
Cioclovina1	New	Romania	45.35	23.84	33,090-31,780	Direct-UF ([8])	Unassigned	Cranium	1240k	M	U	CT	Unassigned	Yes	0.006	12,784
Kostenki12	New	Russia	51.23	39.30	32,990-31,840	Layer ([9])	Unassigned	Cranium	3.7M	M	U2	CT	Unassigned	No	0.03	61,228
KremsWA3	New	Austria	48.41	15.59	31,250-30,690	Layer ([10])	Gravettian	Cranium	1240K	M	U5		Vestonice	No	0.11	203,986
Vestonice13	New	Czech	48.53	16.39	31,070-30,670	Layer ([9])	Gravettian	Femur	3.7M	M	U8c	CT(notIJK)	Vestonice	Yes	0.071	139,568
Vestonice15	New	Czech	48.53	16.39	31,070-30,670	Layer ([9])	Gravettian	Femur	3.7M	M	U5	BT	Vestonice	Yes	0.015	30,900
Vestonice14	New	Czech	48.53	16.39	31,070-30,670	Layer ([9])	Gravettian	Femur	390k	M	U		Vestonice	Yes	0.003	5,677
Pavlov1	New	Czech	48.53	16.39	31,110-29,410	Layer ([9])	Gravettian	Femur	3.7M	M	U5	C1a2	Vestonice	Yes	0.028	57,005
Vestonice43	New	Czech	48.53	16.39	30,710-29,310	Layer ([9])	Gravettian	Femur	3.7M	M	U	F	Vestonice	Yes	0.087	163,946
Vestonice16	New	Czech	48.53	16.39	30,710-29,310	Layer ([9])	Gravettian	Femur	3.7M	M	U5	IJK	Vestonice	No	1.31	945,292
Ostuni2	New	Italy	40.73	17.57	29,310-28,640	Direct-UF (New)	Gravettian	Femur	3.7M	F	U2		Vestonice	Yes	0.008	17,017
GoyetQ53-1	New	Belgium	50.26	4.28	28,230-27,720	Direct-NotUF ([5])	Gravettian	Fibula	1240k	F	U2		Vestonice	Yes	0.006	12,567
Paglicci108	New	Italy	41.65	15.61	28,430-27,070	Layer ([5])	Gravettian	Phalanx	1240k	F	U2′3′4′7′8′9		Vestonice	Yes	0.002	4,330
Ostuni1	New	Italy	40.73	17.57	27,810-27,430	Direct-UF (New)	Gravettian	Tibia	3.7M	F	M		Vestonice	Yes	0.245	369,313
GoyetQ376-19	New	Belgium	50.26	4.28	27,720-27,310	Direct-NotUF ([5])	Gravettian	Humerus	1240k	F	U2		Vestonice	Yes	0.012	25,400
GoyetQ56-16	New	Belgium	50.26	4.28	26,600-26,040	Direct-NotUF ([5])	Gravettian	Fibula	1240k	F	U2		Vestonice	Yes	0.005	9,988
Malta1	[11]	Russia	52.9	103.5	24,520-24,090	Direct-UF ([11])	Unassigned	Humerus	Shotgun	M	U	R	Mal’ta	No	1.174	1439501
ElMiron	New	Spain	43.26	−3.45	18,830-18,610	Direct-UF ([5])	Magdalenian	Toe	3.7M	F	U5b		El Mirón	Yes	1.012	797,714
AfontovaGora3	New	Russia	56.05	92.87	16,930-16,490	Layer ([5])	Unassigned	Tooth	3.7M	F	R1b		Mal’ta	Yes	0.17	286,355
AfontovaGora2	[11]	Russia	56.05	92.87	16,930-16,490	Direct-UF ([11])	Unassigned	Humerus	Shotgun	M			Mal’ta	No	0.071	143,751
Rigney1	New	France	47.23	6.10	15,690-15,240	Direct-NotUF ([12])	Magdalenian	Mandible	1240k	F	U2′3′4′7′8′9		El Mirón	Yes	0.017	35,600
HohleFels49	New	Germany	48.22	9.45	16,000-14,260	Layer ([13])	Magdalenian	Femur	390k	M	U8a	I	El Mirón	Yes	0.033	63,151
GoyetQ-2	New	Belgium	50.26	4.28	15,230-14,780	Direct-NotUF ([5])	Magdalenian	Humerus	1240k	M	U8a	HIJK	El Mirón	Yes	0.035	72,263
Brillenhohle	New	Germany	48.24	9.46	15,120-14,440	Direct-UF ([14])	Magdalenian	Cranium	390k	M	U8a		El Mirón	Yes	0.006	13,459
HohleFels79	New	Germany	48.22	9.45	15,070-14,270	Direct-UF ([5])	Magdalenian	Cranium	390k	M	U8a		El Mirón	Yes	0.005	11,211
Burkhardtshohle	New	Germany	48.32	9.35	15,080-14,150	Direct-UF ([15])	Magdalenian	Cranium	1240k	M	U8a	I	El Mirón	Yes	0.018	38,376
Villabruna	New	Italy	46.15	12.21	14,180-13,780	Direct-UF ([16])	Epigravettian	Femur	3.7M	M	U5b2b	R1b1	Villabruna	No	3.137	1,215,433
Bichon	[17]	Switzerland	47.01	6.79	13,770-13,560	Direct-UF ([17])	Azilian	Petrous	Shotgun	M	U5b1h	I2	Villabruna	No	8.119	2,116,782
Satsurblia	[17]	Georgia	42.24	42.92	13,380-13,130	Direct-UF ([17])	Epigravettian	Petrous	Shotgun	M	K3	J2	Satsurblia	No	1.195	1,460,368
Rochedane	New	France	47.21	6.45	13,090-12,830	Direct-NotUF ([5])	Epipaleolithic	Mandible	1240k	M	U5b2b	I	Villabruna	No	0.131	237,390
Iboussieres39	New	France	44.29	4.46	12,040-11,410	Direct-NotUF ([5])	Epipaleolithic	Femur	390k	M	U5b2b		Villabruna	Yes	0.005	9,659
Continenza	New	Italy	41.96	13.54	11,200-10,510	Layer (New)	Mesolithic	Cranium	3.7M	F	U5b1		Villabruna	Yes	0.006	11,717
Ranchot88	New	France	47.91	5.43	10,240-9,930	Direct-NotUF ([5])	Mesolithic	Cranium	1240k	F	U5b1		Villabruna	Yes	0.322	414,863
LesCloseaux13	New	France	48.52	2.11	10,240-9,560	Direct-NotUF ([18])	Mesolithic	Femur	1240k	F	U5a2		Villabruna	Yes	0.004	8,635
Kotias	[17]	Georgia	42.13	43.12	9,890-9,550	Direct-UF ([17])	Mesolithic	Tooth	Shotgun	M	H13c	J	Satsurblia	No	12.157	2,133,968
Falkenstein	New	Germany	48.06	9.04	9,410-8,990	Direct-UF ([19])	Mesolithic	Fibula	390k	M	U5a2c	F	Villabruna	Yes	0.033	64,428
Karelia	[20]	Russia	61.65	35.65	8,800-7,950	Layer ([21])	Mesolithic	Tooth	Shotgun	M	C1g	R1a1	Unassigned	No	1.952	1,754,410
Bockstein	New	Germany	48.33	10.09	8,370-8,160	Layer ([22])	Mesolithic	Tooth	390k	F	U5b1d1		Villabruna	Yes	0.011	21,977
Ofnet	New	Germany	48.49	10.27	8,430-8,060	Layer ([23])	Mesolithic	Tooth	390k	F	U5b1d1		Villabruna	Yes	0.003	6,263
Chaudardes1	New	France	49.24	3.46	8,360-8,050	Direct-NotUF ([5])	Mesolithic	Tibia	1240k	M	U5b1b	I	Villabruna	Yes	0.046	92,657
Loschbour	[24]	Luxembourg	49.70	6.24	8,160-7,940	Direct-UF ([24])	Mesolithic	Tooth	Shotgun	M	U5b1a	I2a1b	Villabruna	No	20	2,091,584
LaBrana1	[25]	Spain	42.93	−5.35	7,940-7,690	Direct-UF ([26])	Mesolithic	Tooth	Shotgun	M	U5b2c1	C1a2	Villabruna	No	3.338	1,884,745
Hungarian.KO1	[27]	Hungarian	47.93	21.20	7,730-7,590	Direct-UF ([27])	Neolithic	Petrous	Shotgun	M	R3	I2a	Villabruna	No	1.1	1,410,303
Motala12	[24]	Sweden	58.54	15.05	7,670-7,580	Direct-UF (New)	Mesolithic	Tooth	Shotgun	M	U2e1	I2a1b*	Unassigned	No	2.185	1,874,519
BerryAuBac	New	France	49.24	3.54	7,320-7,170	Direct-NotUF ([5])	Mesolithic	Radius	1240k	M	U5b1a	I	Villabruna	No	0.027	54,690
Stuttgart	[24]	Germany	48.78	9.18	7,260-7,020	Direct-UF (New)	Early Neolithic	Tooth	Shotgun	F	T2c1d1		Unassigned	No	19	2,078,724

Note: All dates are obtained as described in Supplementary Information section 1. When an individual has a direct date from an element from the same skeleton it is marked “Direct”, followed by a hyphen to indicate whether the date is obtained by ultrafiltration (“UF”) or without (“NotUF”). If the date is from the archaeological layers, we mark the date type as “Layer”. All the dates were calibrated using IntCal13[28] and the OxCal4.2 program[29].

We represent Kostenki14 in most analyses by our newly reported 16.1x capture data, but repeat key analyses on the previously reported 2.8x shotgun data[30].

Mean coverage is computed on the 3.7M SNP targets.

Ancient DNA data

We extracted DNA from human remains in dedicated clean rooms[5], and transformed the extracts into Illumina sequencing libraries[6-8]. A major challenge in ancient DNA research is that the vast majority of the DNA extracted from most specimens is of microbial origin, making random shotgun sequencing prohibitively expensive. We addressed this problem by enriching the libraries for between 390,000 and 3.7 million single nucleotide polymorphisms (SNPs) in the nuclear genome via hybridizing to pools of previously synthesized 52-base-pair oligonucleotide probes targeting these positions (this strategy makes it possible to generate genome-wide data from samples with high percentages of microbial DNA that are not practical to study by shotgun sequencing)[3,9]. We sequenced the isolated DNA fragments from both ends, and mapped the consensus sequences to the human genome (hg19), retaining fragments that overlapped the targeted SNPs. After removing fragments with identical start and end positions to eliminate duplicates produced during library amplification, we chose one fragment at random to represent each individual at each SNP.

Contamination from present-day human DNA is a danger in ancient DNA research. To address this we took advantage of three characteristic features of ancient DNA (Supplementary Information section 2). First, for an uncontaminated specimen, we expect only a single mitochondrial DNA sequence to be present, allowing us to detect contamination as a mixture of mitochondrial sequences. Second, because males carry a single X chromosome, we can detect contamination in male specimens as polymorphisms on chromosome X[10]. Third, cytosines at the ends of genuine ancient DNA molecules are often deaminated, resulting in apparent cytosine to thymine substitutions[11]. Thus, restricting analysis to molecules with evidence of such deamination filters out the great majority of contaminating molecules[12]. For libraries from males with evidence of mitochondrial DNA contamination or X chromosomal contamination estimates >2.5%—as well as for all libraries from females—we restricted the analyses to sequences with evidence of cytosine deamination (Supplementary Information section 2). After merging libraries from the same individual and limiting to individuals with >4,000 targeted SNPs covered at least once, 38 individuals remained, which we merged with newly generated shotgun sequencing data from the Karelia individual[9] (2.0-fold coverage), and published data from ancient[2-4,7,13-19] and present-day humans[20]. The final dataset includes 51 ancient modern humans, of which 16 had at least 790,000 SNPs covered (Figure 1; Table 1; Extended Data Table 1).

Figure 1

Location and age of 51 ancient samples

Each bar corresponds to a sample, the color code designates the genetically defined sample cluster, and the height is proportional to sample age (the background grid shows a projection of longitude against sample age). To help in visualization, we add jitter for sites with multiple samples from nearby locations. Four samples that are from Siberia are plotted at the far eastern edge of the map.

Natural selection has reduced Neanderthal ancestry over the last 45,000 years

We used two previously published statistics[3,7,21] to ask if the proportion of Neanderthal ancestry in Eurasians changed over the last 45,000 years. Whereas on the order of 2% of present-day Eurasian DNA is of Neanderthal origin (Extended Data Table 2), the ancient modern human genomes carry significantly more Neanderthal DNA (Figure 2) (P≪10−12). Using one statistic, we estimate a decline from 4.3–5.7% from a time shortly after introgression to 1.1–2.2% in Eurasians today (Figure 2). Using the other statistic, we estimate a decline from 3.2–4.2% to 1.8–2.3% (Extended Data Figure 1, Extended Data Table 3). Because all the European samples we analyzed dating to between 37,000 and 14,000 years ago are consistent with descent from a single founding population, admixture with populations with lower Neanderthal ancestry cannot explain the steady decrease in Neanderthal-derived DNA that we detect during this period, showing that natural selection against Neanderthal DNA must have driven this phenomenon (Figure 2). We also obtain an independent line of evidence for selection from our observation that the decrease in Neanderthal-derived alleles is more marked near genes than in less constrained regions of the genome (P=0.010) (Supplementary Information section 3; Extended Data Table 3)[22-25].

Extended Data Table 2

Estimated proportion of Neanderthal ancestry

		f4-ratios			Archaic Ancestry Informative SNPs
Sample Code	Age BP	SNPs	Est.	95% CI	SNPs	Est.	95% CI	Increase in Neanderthal ancestry with B	S.E.
UstIshim	45,020	2,137,615	4.4%	3.6% – 5.3%	778,774	3.0%	2.3% – 3.7%	−0.9%	1.3%
Oase1	39,610	285,076	9.9%	8.4% – 11.4%	59,854	7.5%	6.0% – 8.9%	2.5%	1.8%
Kostenki14	37,470	1,774,156	3.6%	2.7% – 4.4%	632,748	2.8%	2.3% – 3.3%	−1.0%	1.0%
GoyetQ116-1	34,795	846,983	3.4%	2.4% – 4.3%
Muierii2	33,300	98,618	5.2%	3.0% – 7.4%	22,189	3.0%	2.5% – 3.5%	0.6%	1.1%
Paglicci133	32,895	82,330	4.1%	2.1% – 6.0%
Cioclovina1	32,435	12,784	4.1%	−1.1% – 9.3%
Kostenki12	32,415	61,228	1.9%	−0.7% – 4.4%	13,385	2.6%	2.1% – 3.2%	1.7%	1.5%
KremsWA3	30,970	203,986	3.9%	2.6% – 5.2%			-
Vestonice13	30,870	139,568	4.6%	2.6% – 6.5%	35,983	3.3%	2.7% – 3.8%	0.3%	1.3%
Vestonice15	30,870	30,900	4.3%	0.6% – 7.9%	5,855	2.7%	2.1% – 3.4%	−1.5%	1.3%
Vestonice14	30,870	5,677	2.6%	−5.9% – 11.0%
Pavlov1	30,260	57,005	4.4%	1.6% – 7.1%	9,327	3.1%	2.5% – 3.8%	0.7%	1.2%
Vestonice43	30,010	163,946	6.9%	5.2% – 8.5%	38,749	2.9%	2.4% – 3.3%	0.9%	0.9%
Vestonice16	30,010	945,292	4.1%	3.1% – 5.1%	268,157	2.8%	2.3% – 3.3%	−0.1%	1.0%
Ostuni2	28,975	17,017	1.6%	−3.2% – 6.3%	2,746	2.3%	1.4% – 3.1%	1.3%	1.6%
GoyetQ53-1	27,975	12,567	4.8%	−0.7% – 10.3%
Paglicci108	27,750	4,330	3.4%	−6.0% – 12.7%
Ostuni1	27,620	369,313	4.2%	3.0% – 5.4%	88,449	2.6%	2.2% – 3.0%	0.1%	0.9%
GoyetQ376-19	27,515	25,400	6.5%	2.7% – 10.2%
GoyetQ56-16	26,320	9,988	3.6%	−1.9% – 9.1%
Malta1	24,305	1,439,501	2.9%	1.9% – 3.8%	437,187	2.5%	2.1% – 2.9%	1.0%	0.8%
ElMiron	18,720	797,714	3.6%	2.6% – 4.5%	250,071	2.8%	2.5% – 3.2%	0.6%	0.9%
AfontovaGora3	16,710	286,355	3.0%	1.8% – 4.2%	96,237	3.3%	2.9% – 3.7%	−1.5%	1.0%
AfontovaGora2	16,710	143,751	2.2%	0.4% – 4.0%	37,280	2.3%	1.9% – 2.7%	−0.3%	0.9%
Rigney1	15,465	35,600	0.8%	−2.6% – 4.2%
HohleFels49	15,130	63,151	2.3%	−0.6% – 5.2%
GoyetQ-2	15,005	72,263	1.7%	−0.6% – 4.0%
Brillenhohle	14780	13,459	2.5%	−3.0% – 8.1%
HohleFels79	14,670	11,211	1.7%	−5.1% – 8.5%
Burkhardtshohle	14,615	38,376	1.7%	−1.6% – 5.0%
Villabruna	13,980	1,215,433	2.7%	1.8% – 3.5%	425,148	3.3%	3.0% – 3.7%	1.1%	0.9%
Bichon	13,665	2,116,782	2.9%	1.9% – 3.8%	769,422	2.7%	2.2% – 3.2%	0.7%	1.3%
Satsurblia	13,255	1,460,368	1.5%	0.6% – 2.4%	542,561	2.0%	1.7% – 2.4%	0.9%	0.6%
Rochedane	12,960	237,390	1.9%	0.5% – 3.3%
Iboussieres39	11,725	9,659	6.4%	−0.8% – 13.7%
Continenza	10,855	11,717	4.1%	−1.4% – 9.6%	1,733	2.9%	1.8% – 4.0%	−10.6%	4.4%
Ranchot88	10,085	414,863	2.9%	1.8% – 4.0%
LesCloseaux13	9,900	8,635	−3.0%	−9.7% – 3.8%
Kotias	9,720	2,133,968	1.8%	1.0% – 2.7%	779,146	2.1%	1.8% – 2.4%	0.7%	0.5%
Falkenstein	9,200	64,428	4.8%	1.7% – 7.8%
Karelia	8,375	1,754,410	1.9%	1.1% – 2.7%	582,444	2.2%	1.9% – 2.6%	−0.2%	0.7%
Bockstein	8,265	21,977	5.7%	1.0% – 10.5%
Ofnet	8,245	6,263	9.8%	1.4% – 18.1%
Chaudardes1	8,205	92,657	1.9%	−0.2% – 3.9%
Loschbour	8,050	2,091,584	2.5%	1.6% – 3.3%	774,139	2.6%	2.0% – 3.1%	2.7%	1.7%
LaBrana1	7,815	1,884,745	1.9%	1.1% – 2.8%	642,231	2.7%	2.3% – 3.2%	0.4%	0.8%
Hungarian.KO1	7,660	1,410,303	2.1%	1.2% – 3.0%	439,408	2.4%	2.0% – 2.8%	−0.1%	1.2%
Motala12	7,625	1,874,519	2.5%	1.6% – 3.3%	655,685	2.3%	1.9% – 2.7%	−0.1%	0.7%
BerryAuBac	7,245	54,690	2.5%	−0.2% – 5.1%
Stuttgart	7,140	2,078,724	1.9%	1.1% – 2.7%	767,813	2.1%	1.8% – 2.5%	0.0%	0.7%
Dai	0	2,144,502	1.4%	0.7% – 2.1%	782,066	1.8%	1.5% – 2.1%	1.4%	0.4%
Han	0	2,144,502	1.8%	1.1% – 2.5%	782,164	2.1%	1.8% – 2.5%	1.9%	0.7%
English	0	2,144,502	1.5%	0.8% – 2.2%
French	0	2,144,502	1.5%	0.9% – 2.1%	782,386	1.7%	1.4% – 1.9%	1.4%	0.6%
Sardinian	0	2,144,502	1.2%	0.6% – 1.9%	782,351	1.7%	1.4% – 2.0%	0.7%	0.5%
Karitiana	0				782,037	2.1%	1.7% – 2.4%	1.5%	1.0%

Figure 2

Decrease of Neanderthal ancestry over time

Plot of radiocarbon date against Neanderthal ancestry for samples with at least >200,000 SNPs covered, along with present-day Eurasians (standard errors are from a Block Jackknife). The least squares fit (gray) excludes the data from Oase1 (an outlier with recent Neanderthal ancestry) and three present-day European populations (known to have less Neanderthal ancestry than East Asians). The slope is significantly negative for all eleven subsets of samples we analyzed (10−29

Extended Data Figure 1

A decrease in Neanderthal ancestry in the last 45,000 years

This is similar to Figure 2, except we use ancestry estimates from rates of alleles matching to Neanderthal rather than f-ratios, as described in Supplementary Information section 3). The least squares fit excludes Oase1 (as an outlier with recent Neanderthal ancestry) and Europeans (known to have reduce Neanderthal ancestry). The regression slope is significantly negative (P=0.00004, Extended Data Table 3).

Extended Data Table 3

Significant correlation of Neanderthal ancestry estimate with specimen age

Subset of samples	N	P-value for date correlation	Decrease in ancestry per 10,000 years	Estimate of Neanderthal ancestry at different time points
				0 years ago (present)	50,000 years ago	55,000 years ago	60,000 years ago
f4-ratio estimates
Core Set 1 (all ancient samples (except Oase1) + Han + Dai)	57	5 × 10−22	0.48–0.73%	1.1–2.2%	4.0–5.4%	4.3–5.7%	4.5–6.0%
Subset of Core Set 1 (<32kya)	50	2 × 10−15	0.59–0.98%	0.9–2.1%	4.5–6.4%	4.8–6.9%	5.1–7.4%
Subset of Core Set 1 (>32kya or <25kya)	44	4 × 10−18	0.44–0.69%	1.0–2.2%	3.7–5.2%	4.0–5.5%	4.2–5.8%
Subset of Core Set 1 (>25kya or <14kya)	47	5 × 10−21	0.48–0.73%	1.0–2.2%	3.9–5.3%	4.2–5.7%	4.5–6.0%
Subset of Core Set 1 (>14kya or present day)	37	2 × 10−18	0.47–0.74%	1.1–2.4%	4.1–5.5%	4.3–5.8%	4.6–6.2%
Subset of Core Set 1 (only ancient samples)	50	4 × 10−15	0.46–0.76%	1.0–2.3%	4.0–5.4%	4.3–5.8%	4.5–6.1%
Subset of Core Set 1 (individuals with >200,000 SNPs)	28	4 × 10−19	0.46–0.71%	1.1–2.3%	3.9–5.3%	4.2–5.7%	4.4–6.0%
Modification of Core Set 1 (replace East Asians with Europeans)	58	2 × 10−23	0.49–0.73%	1.1–2.3%	4.0–5.4%	4.3–5.8%	4.6–6.1%
All ancient samples including Oase1 + Han + Dai	58	8 × 10−29	0.57–0.81%	1.0–2.2%	4.3–5.7%	4.7–6.1%	5.0–6.5%
All ancient samples	51	1 × 10−20	0.57–0.86%	0.9–2.2%	4.4–5.8%	4.7–6.2%	5.0–6.6%
All ancient samples except Oase1 or UstIshim	49	8 × 10−12	0.45–0.81%	1.0–2.3%	4.0–5.6%	4.2–6.0%	4.5–6.4%
Ancestry informative SNPs
Core Set 2 (all ancient samples (except Oase1) + Han + Dai + Karitiana)	29	4 × 10−11	0.21–0.39%	1.8–2.3%	3.1–4.0%	3.2–4.2%	3.3–4.3%
Subset of Core Set 2 (no Han, Dai, Karitiana, Stuttgart)	25	1 × 10−4	0.11–0.36%	1.8–2.5%	2.9–3.8%	3.0–4.0%	3.0–4.1%
Subset of Core Set 2 (no Han, Dai, Karitiana, Stuttgart, UstIshim)	24	2 × 10−4	0.11–0.37%	1.8–2.5%	2.9–3.8%	2.9–4.0%	3.0–4.2%

Note: The “Core Set 1,” used for the f-ratio analyses, refers to 50 ancient samples (removing Oase1 as an outlier) along with 7 East Asians (Dai and Han). “Core Set 2,” used for the analyses of Neanderthal ancestry informative SNPs, refers to 26 ancient samples (removing Oase1) along with Han, Dai, and Karitiana

Y chromosomes, mitochondrial DNA and phenotypically important mutations

We used the proportion of sequences mapping to the Y chromosome to infer sex (Extended Data Table 4; Supplementary Information section 4), and determined Y chromosome haplogroups for the males. We were surprised to find haplogroup R1b in the ~14,000-year-old Villabruna individual from Italy. While the predominance of R1b in western Europe today is owes its origin to Bronze Age migrations from the eastern European steppe[9], its presence in Villabruna and in a ~7,000-year-old farmer from Iberia[9] document a deeper history of this haplotype in more western parts of Europe. Additional evidence of an early link between west and east comes from the HERC2 locus, where a derived allele that is the primary driver of light eye color in Europeans appears nearly simultaneously in specimens from Italy and the Caucasus ~14,000-13,000 years ago. Extended Data Table 5 presents results for additional alleles of known phenotypic importance. When analyzing the mitochondrial genomes we note the presence of haplogroup M in a ~27,000-year-old individual from southern Italy (Ostuni1) in agreement with the observation that this haplogroup, which today occurs in Asia and is absent in Europe, was present in pre-Last Glacial Maximum Europe and became lost during the Ice Age[26]. We also find that the ~33,000 year old Muierii2 from Romania carries a basal version of haplogroup U6, in agreement with the hypothesis that the presence of derived versions of this haplogroup in North Africans today is due to back-migration from western Eurasia[27].

Extended Data Table 4

Sex determination for newly reported samples.Y-rate is the ratio of NY/Nauto divided by the same quantity for the genome-wide target set. Female sex (F) is inferred as Y-rate<0.05 and male sex (M) as Y-rate>0.

Sample	Target	Type	Nauto	NX	NY	NX/Nauto	NY/Nauto	X-rate	Y-rate	Sex
	1240k or 2.2M*		1151240	49711	32681	0.0432	0.0284
	390k		388745	1819	2242	0.0047	0.0058
Kostenki14	2.2M	all	29633405	395534	262846	0.0133	0.0089	0.309	0.312	M
GoyetQ116-1	1240k	all	2122620	36391	22256	0.0171	0.0105	0.397	0.369	M
Cioclovina1	1240k	Damage	11521	184	125	0.0160	0.0108	0.370	0.382	M
Kostenki12	2.2M	Subset	63908	856	504	0.0134	0.0079	0.310	0.278	M
Muierii2	2.2M	Damage	81165	2177	8	0.0268	0.0001	0.621	0.003	F
Vestonice13	2.2M	Damage	119094	1578	1059	0.0133	0.0089	0.307	0.313	M
Vestonice15	2.2M	Damage	28762	338	227	0.0118	0.0079	0.272	0.278	M
Vestonice14	390k	Damage	4846	8	11	0.0017	0.0023	0.353	0.394	M
Vestonice43	2.2M	Damage	136933	1826	1204	0.0133	0.0088	0.309	0.310	M
Pavlov1	2.2M	Damage	54429	631	404	0.0116	0.0074	0.268	0.261	M
Vestonice16	2.2M	Subset	2433741	30463	20976	0.0125	0.0086	0.290	0.304	M
KremsWA3	1240k	all	235069	4119	2661	0.0175	0.0113	0.406	0.399	M
Ostuni2	2.2M	Damage	15749	138	1	0.0088	0.0001	0.203	0.002	F
Ostuni1	2.2M	Damage	427199	10868	47	0.0254	0.0001	0.589	0.004	F
Paglicci108	1240k	Damage	3883	124	2	0.0319	0.0005	0.740	0.018	F
GoyetQ53-1	1240k	Damage	10771	311	4	0.0289	0.0004	0.669	0.013	F
GoyetQ376-19	1240k	Damage	20052	680	10	0.0339	0.0005	0.785	0.018	F
GoyetQ56-16	1240k	Damage	8702	304	7	0.0349	0.0008	0.809	0.028	F
Paglicci133	1240k	Subset	81092	1641	983	0.0202	0.0121	0.469	0.427	M
ElMiron	2.2M	Damage	1765696	40647	196	0.0230	0.0001	0.533	0.004	F
HohleFels79	390k	Damage	10188	28	22	0.0027	0.0022	0.587	0.374	M
AfontovaGora3	2.2M	Damage	291798	8705	37	0.0298	0.0001	0.691	0.004	F
HohleFels49	390k	Damage	61051	113	111	0.0019	0.0018	0.396	0.315	M
Rigney1	1240k	Damage	32797	1131	9	0.0345	0.0003	0.799	0.010	F
GoyetQ-2	1240k	Damage	65563	1123	706	0.0171	0.0108	0.397	0.379	M
Brillenhohle	390k	Damage	12603	22	22	0.0017	0.0017	0.373	0.303	M
Burkhardtshohle	1240k	Damage	34207	563	407	0.0165	0.0119	0.381	0.419	M
Villabruna	2.2M	Subset	5505838	72055	52110	0.0131	0.0095	0.303	0.333	M
Rochedane	1240k	Subset	256325	4780	2830	0.0186	0.0110	0.432	0.389	M
Continenza	2.2M	Damage	10647	208	2	0.0195	0.0002	0.452	0.007	F
Iboussieres39	390k	Damage	8246	12	22	0.0015	0.0027	0.311	0.463	M
Ranchot88	1240k	Damage	594962	18520	119	0.0311	0.0002	0.721	0.007	F
LesCloseaux13	1240k	Damage	7326	275	2	0.0375	0.0003	0.869	0.010	F
Falkenstein	390k	Damage	58970	113	102	0.0019	0.0017	0.410	0.300	M
Bockstein	390k	Damage	20214	62	0	0.0031	0.0000	0.655	0.000	F
Ofnet	390k	Damage	5294	13	1	0.0025	0.0002	0.525	0.033	F
Chaudardes1	1240k	Damage	84052	1429	865	0.0170	0.0103	0.394	0.363	M
BerryAuBac	1240k	All	49670	902	554	0.0182	0.0112	0.421	0.393	M

We restrict analysis to the 1240k target set for study of the 2.2M capture datasets.

Extended Data Table 5

Allele counts at SNPs thought to be affected by selection in samples that have at least 1.0-fold coverage.rs4988235 is responsible for lactase persistence in Europe[59,60]. The SNPs at SLC24A5 and SLC45A2 are responsible for light skin pigmentation. The SNP at EDAR[61,62] affects tooth morphology and hair thickness. The SNP at HERC2[63,64] is the primary determinant of light eye color in present-day Europeans. We present the fraction of fragments overlapping each SNP that are derived; the observation of a low rate of derived alleles does not prove that the individual carried the allele, and instead may reflect sequencing error or ancient DNA damage. We highlight in light gray sites that we judge (based on the derived allele count) are likely to be heterozygous for the derived allele, and in dark gray sites that are likely to be homozygous.

		LCT	SLC45A2	SLC24A5	EDAR	HERC2
	SNP	rs4988235	rs16891982	rs1426654	rs3827760	rs12913832
	Ancestral	G	C	G	A	A
	Derived	A	G	A	G	G
UstIshim	Coverage	31	46	52	42	50
	Derived allele frequency	0%	0%	2%	0%	0%
Kostenki14	Coverage	140	113	6	45	52
	Derived allele frequency	0%	2%	17%	0%	0%
GoyetQ116-1	Coverage	8	6	0	9	1
	Derived allele frequency	0%	0%	n/a	0%	0%
Vestonice16	Coverage	13	18	0	4	5
	Derived allele frequency	0%	6%		0%	0%
Malta1	Coverage	1	0	2	2	2
	Derived allele frequency	0%		0%	0%	0%
ElMiron	Coverage	2	10	0	7	5
	Derived allele frequency	0%	0%		0%	0%
Villabruna	Coverage	17	52	5	19	10
	Derived allele frequency	0%	0%	0%	0%	100%
Bichon	Coverage	11	4	25	16	9
	Derived allele frequency	0%	0%	0%	0%	33%
Satsurblia	Coverage	1	2	4	1	4
	Derived allele frequency	0%	0%	100%	0%	50%
Kotias	Coverage	16	22	13	20	15
	Derived allele frequency	0%	0%	100%	0%	0%
Loschbour	Coverage	19	18	20	17	21
	Derived allele frequency	0%	0%	0%	0%	100%
LaBrana1	Coverage	8	6	2	11	3
	Derived allele frequency	12%	0%	0%	0%	100%
Hungarian.KO1	Coverage	1	2	2	1	2
	Derived allele frequency	0%	0%	50%	0%	100%
Motala12	Coverage	2	0	3	3	1
	Derived allele frequency	0%		0%	33%	100%
Karelia	Coverage	1	9	4	0	1
	Derived allele frequency	0%	67%	0%		0%
Stuttgart	Coverage	25	21	15	29	21
	Derived allele frequency	0%	0%	100%	0%	0%

Genetic clustering of the ancient specimens

This dataset provides an unprecedented opportunity to study the population history of Upper Paleolithic Europe over more than 30,000 years. In order to not prejudice any association between genetic and archaeological groupings among the individuals studied, we first allowed the genetic data alone to drive the groupings of the specimens and only afterward examined their associations with archaeological cultural complexes. We began by computing f3-statistics[14] of the form f3(X, Y; Mbuti), which measure shared genetic drift between a pair of ancient individuals after divergence from an outgroup (here Mbuti from sub-Saharan Africa), which allowed us to observe clear clusters of samples (Figure 3A; Extended Data Figure 2). Through Multi-Dimensional Scaling (MDS) analysis of this matrix (Figure 3B), as well as through D-statistic analyses[28] (Supplementary Information section 5), we identified five clusters of individuals with substantial shared genetic drift, which we name after the oldest individual with >1.0-fold coverage in each cluster (Supplementary Information section 5; Table 1; Extended Data Table 1). In contrast, we were not able to identify clear structure among these samples based on model-based clustering[29,30], which may reflect the fact that many of the samples are so ancient that present-day patterns of human variation are not very relevant to understanding their patterns of genetic differentiation[4,13]. The “Vestonice Cluster” is composed of 14 pre-Ice Age individuals from 34,000-26,000 years ago, who are all associated with the archaeologically defined Gravettian culture. The “Mal’ta Cluster” is composed of three individuals from the Glacial Maximum 24,000-17,000 years ago from the Lake Baikal region of Siberia. The “El Mirón Cluster” is composed of 6 Late Glacial individuals from 19,000-14,000 years ago, who are all associated with the Magdalenian culture. The “Villabruna Cluster” is composed of 13 post-Ice Age individuals from 14,000-7,000 years ago, associated with the Azilian, Epipaleolithic and Mesolithic cultures. The “Satsurblia Cluster” is composed of two individuals from 13,000-10,000 years ago from the northern Caucasus[2]. There were ten samples that we did not assign to any cluster, either because of evidence of representing distinct early lineages, (Ust’-Ishim, Oase1, Kostenki14, GoyetQ116-1, Muierii2, Cioclovina1, Kostenki12), or because they were admixed between major clusters (Karelia or Motala12), or of very different ancestry (Stuttgart). To classify the ancestry of additional low coverage samples, we built an admixture graph that fits the allele frequency correlation patterns among high coverage samples[28] (Supplementary Information section 6; Figure 4a). We fit each low coverage sample into the graph in turn, including all fragments from every individual rather than just ones with evidence of cytosine deamination, accounting for contamination bias by modeling (Supplementary Information section 7).

Figure 3

Genetic clustering

(A) Shared genetic drift measured by f3(X,Y; Mbuti) among samples with at least 30,000 SNPs covered (for AfontovaGora3, ElMiron, Falkenstein, GoyetQ-2, GoyetQ53-1, HohleFels49, HohleFels79, LesCloseaux13, Ofnet, Ranchot88 and Rigney1, we use all sequences for higher resolution). Lighter colors indicate more shared drift. (B) Multidimensional Dimensional Scaling (MDS) analysis, computed using the R software cmdscale package, highlights the main genetic groupings analyzed in this study: Vestonice Cluster (brown), Mal’ta Cluster (pink), El Mirón Cluster (yellow), Villabruna Cluster (light blue), and Satsurblia Cluster (dark purple). The affinity of GoyetQ116-1 (green) to the El Mirón Cluster is evident in both views of the data.

Extended Data Fig. 2

Heat matrix of pairwise f3(X, Y; Mbuti) for selected ancient samples

We analyze only samples with at least 30,000 SNPs covered at least once, which pass our quality control.

Figure 4

Population history inferences

(A) Admixture Graph relating selected high coverage samples. Dashed lines show inferred admixture events; the estimated mixture proportions fitted using the ADMIXTUREGRAPH software are labeled[28] (the estimated genetic drift on each branch is given in a version of this graph shown in Supplementary Information section 6). The samples are positioned vertically based on their radiocarbon date, but we caution that the population split times are not accurately known. We use color to highlight important early branches of the European founder population: the Kostenki14 lineage is modeled as the predominant contributor to the Vestonice Cluster (green); the GoyetQ116-1 lineage as the predominant contributor to the El Mirón Cluster (red); and the Villabruna lineage as broadly represented across many clusters. (B) Drawing together of European and Near Eastern populations ~14,000 years ago. Plot of affinity of each pre-Neolithic European population X to non-Africans outside Europe Y moving forward in time, comparing to Kostenki14 as a baseline; values Z<-3 standard errors below zero are indicated with filled symbols (we restricted to individuals with >50,000 SNPs). We observe an affinity to Near Easterners beginning with the Villabruna Cluster, and another to East Asians that affects a subset of the Villabruna Cluster.

A single founding population during most of the Upper Paleolithic period in Europe

Prior to this work, the most ambitious genetic analysis of early modern humans in Europe was based on the ~37,000-year-old Kostenki14[4]. That analysis suggested that the population to which Kostenki14 belonged harbored within it the three major lineages that exist in mixed form in Europe today[15]: (1) a lineage related to all later pre-Neolithic Europeans, (2) a “Basal Eurasian” lineage that split from the ancestors of Europeans and East Asians before they separated from each other; and (3) a lineage related to the ~24,000-year-old Mal’ta1 from Siberia. With our more extensive sampling of Ice Age Europe, we find no support for this model. When we test whether the ~45,000-year-old Ust’-Ishim – an early Eurasian without any evidence of Basal Eurasian ancestry – shares more alleles with one test individual or another by computing statistics of the form D(Test1, Test2; Ust’-Ishim, Mbuti), we find that the statistic is consistent with zero when the Test populations are any pre-Neolithic Europeans or present-day East Asians[3,13,31]. This would not be expected if some of the pre-Neolithic Europeans, including Kostenki14, had Basal Eurasian ancestry (Supplementary Information section 8). We also find no evidence for the suggestion that the Mal’ta1 lineage contributed to Upper Paleolithic Europeans[4], because when we compute the statistic D(Test1, Test2; Mal’ta1, Mbuti), we find that the statistic is consistent with zero when the Test populations are any pre-Neolithic Europeans beginning with Kostenki14, implying descent from a single founder population since separation from the lineage leading to Mal’ta1 (Supplementary Information section 9). A corollary of this finding is that the widespread presence of Mal’ta1-related ancestry in present-day Europeans[15] is due to migrations from the Eurasian steppe in the Neolithic and Bronze Age periods[9]; it is not due to population structure within pre-Neolithic Europe as proposed in the initial analysis of the Kostenki14 genome[4].

Resurgence of an early branching European lineage during the Last Glacial Maximum

Among the newly reported individuals, GoyetQ116-1 from present-day Belgium is the oldest at ~35,000 years ago. It is similar to the ~37,000 year old Kostenki14 and all later samples in that it shares more alleles with present-day Europeans (e.g. French) than with East Asians (e.g. Han). In contrast, Ust’-Ishim and Oase1, which predate GoyetQ116-1 and Kostenki14, do not show any distinctive affinity to later Europeans (Extended Data Table 6). Thus, from at least about 37,000 years ago, populations in Europe shared at least some ancestry with present Europeans. However, GoyetQ116-1 differs from Kostenki14 and from all individuals of the succeeding Vestonice Cluster in that both f3-statistics (Figure 3; Extended Data Figure 2) and D-statistics show that it shares more alleles with members of the El Mirón Cluster who lived 19,000-14,000 years ago than with other pre-Neolithic Europeans (Supplementary Information section 10). Thus, GoyetQ116-1 has affinity to individuals who lived more than fifteen thousand years later. While at least half of the ancestry of all El Mirón Cluster individuals comes from the GoyetQ116-1 cluster, this proportion varies, with the largest amount in individuals outside Iberia (Z=−4.8) (Supplementary Information section 10).

Extended Data Table 6

All European hunter-gatherers after Kostenki14 share genetic drift with present-day Europeans.We compute the statistic D(Han, Test; French, Mbuti). Measuring whether present-day French share more alleles with Han or with a Test population (restricting to ancient samples with at least 30,000 SNPs covered at least once). Present-day Europeans share significantly more genetic drift with European hunter-gatherers from Kostenki14 onward than they do with Han. Thus, by the date of Kostenki14, there was already West Eurasian-specific genetic drift.

Test	SNPs used	D-value	Z score
Ust’-Ishim	2,050,358	0.003	6.6
Oase1	278,785	0.005	10.6
Kostenki14	1,676,253	−0.002	−5.5
Muierii2	95,787	−0.004	−6.3
GoyetQ116-1	811,756	−0.004	−8.0
Kostenki12	59,850	−0.004	−5.1
Paglicci133	79,624	−0.004	−5.5
Vestonice13	136,598	−0.004	−7.1
Vestonice15	30,252	−0.006	−6.4
Vestonice16	914,141	−0.004	−9.1
Pavlov1	55,835	−0.005	−6.3
Vestonice43	160,463	−0.004	−6.9
KremsWA3	229,187	−0.005	−10.2
Ostuni1	360,347	−0.004	−8.6
Malta1	1,401,718	−0.005	−11.3
ElMiron	777,654	−0.007	−14.7
AfontovaGora2	141,073	−0.007	−13.6
AfontovaGora3	707,617	−0.006	−13.6
HohleFels49	62,816	−0.004	−5.2
Rigney1	34,445	−0.006	−6.1
GoyetQ-2	70,210	−0.006	−8.8
Burkhardtshohle	37,234	−0.006	−6.2
Villabruna	1,170,777	−0.010	−24.7
Bichon	2,034,069	−0.009	−23.6
Satsurblia	1,419,824	−0.005	−13.1
Rochedane	229,806	−0.011	−20.8
Ranchot88	402,274	−0.010	−21.8
Kotias	2,047,856	−0.006	−15.8
Falkenstein	64,043	−0.008	−11.6
Chaudardes1	90,047	−0.011	−16.0
Loschbour	2,037,082	−0.011	−25.4
LaBrana1	1,824,307	−0.009	−23.0
Motala12	1,816,201	−0.009	−23.8
Hungarian.KO1	1,372,801	−0.010	−26.5
Karelia	1,701,664	−0.009	−21.9
Stuttgart	2,023,939	−0.009	−23.9
BerryAuBac	53,028	−0.011	−14.0

A drawing together of the ancestry of Europe and the Near East after ~14,000 years ago

Beginning around 14,000 years ago with the Villabruna Cluster, the strong affinity to GoyetQ116-1 seen in El Mirón Cluster individuals who belong the Late Glacial Magdalenian Culture is greatly attenuated (Supplementary Information section 10). To test if this change might reflect gene flow from populations that did not descend from the >37,000 year old European founder population, we computed statistics of the form D(Early European, Later European; Y, Mbuti) where Y are various present-day non-Africans. If no gene flow from exogenous populations occurred, this statistic is expected to be zero. Figure 4b shows that it is consistent with zero (|Z|<3) for nearly all individuals dating to between about 37,000 and 14,000 years ago. However, beginning with the Villabruna Cluster, it becomes highly significantly negative in comparisons where the non-European population (Y) is Near Easterners (Figure 4b; Extended Data Figure 3; Supplementary Information section 11). This must reflect gene flow into the Villabruna Cluster from a population related to present-day Near Easterners rather than gene flow in the reverse direction, because we do not see similar patterns in earlier Europeans although they share substantial amounts of their ancestry with the Villabruna Cluster (Figure 4b). The “Satsurblia Cluster” individuals from the Caucasus dating to ~13,000-10,000 years ago[2] share more alleles with the Villabruna Cluster individuals than they do with earlier Europeans, indicating that they are related to the population that contributed new alleles to people in the Villabruna Cluster, although they cannot be the direct source of the gene flow, among other reasons because they have large amounts of Basal Eurasian ancestry while Villabruna Cluster individuals do not[2] (Supplementary Information section 12; Extended Data Figure 4). One possible explanation for the sudden drawing together of the ancestry of Europe and the Near East at this time is long-distance migrations from the Near East into Europe. However, a plausible alternative is population structure, whereby Upper Paleolithic Europe harbored multiple groups that differed in their relationship to the Near East, with the balance shifting among groups as a result of demographic changes after the Ice Age.

Extended Data Fig. 3

Studying how the relatedness of non-European populations to pairs of European hunter-gatherers changes over time

We examine statistics of the form D(W, X; Y, Mbuti), with the Z-score given on the y-axis, where W is an early European hunter-gatherer, X is another European hunter-gatherer (in chronological order on the x-axis), and Y is a non-European population (see legend). A: W=Kostenki14. B: W=GoyetQ116-1. C: W=Vestonice16. D: W=ElMiron. |Z|>3 scores are considered statistically significant (horizontal line). The similar Figure 4b gives absolute D-statistic values rather than Z-scores (for W=Kostenki14) and uses pooled regions rather than individual populations Y.

Extended Data Figure 4

An Admixture Graph model that fits the data for Satsurblia, an Upper Paleolithic sample from the Caucasus

This model uses 127,057 SNPs covered in all populations. Estimated genetic drifts are give along the solid lines in units of f-distance (parts per thousand), and estimated mixture proportions are given along the dotted lines. All three models provide an fit to the allele frequency correlation data among Mbuti, UstIshim, Kostenki14, Vestonice16, Malta1, ElMiron and Satsurblia to within the limits of our resolution, in the sense that all empirical f, f and f-statistics relating the samples are within three standard errors of the expectation of the model. Models in which Satsurblia is modeled as unadmixed cannot be fit.

The Villabruna Cluster includes the largest group of samples in this study. This allows us to study heterogeneity within this cluster (Supplementary Information section 13). First, we detect differences in the degree of allele sharing with members of the El Mirón Cluster, as revealed by significant statistics of the form D(Test1, Test2; El Mirón Cluster, Mbuti). Second, we detect an excess of allele sharing with East Asians in a subset of Villabruna Cluster individuals - beginning with a ~13,000 year old sample from Switzerland - as revealed by significant statistics of the form D(Test1, Test2; Han, Mbuti) (Figure 4b and Extended Data Figure 3). For example, Han Chinese share more alleles with two Villabruna Cluster individuals (Loschbour and LaBrana1) than they do with Kostenki14, as reflected in significantly negative statistics of the form D(Kostenki14, Loschbour/LaBrana1; Han, Mbuti)[4]. This statistic was originally interpreted as evidence of Basal Eurasian ancestry in Kostenki14. However, because this statistic is consistent with zero when Han is replaced with Ust’-Ishim, these findings cannot be driven by Basal Eurasian ancestry (as we also discuss above), and must instead be driven by gene flow between populations related to East Asians and the ancestors of some Europeans (Supplementary Information section 8).

Conclusions

We have shown that the population history of pre-Neolithic Europe was complex in several respects. First, at least some of the initial modern humans to appear in Europe, exemplified by Ust’-Ishim and Oase1, failed to contribute appreciably to the current European gene pool. Only from around 37,000 years ago do all the European individuals analyzed share ancestry with present-day Europeans[3]. Second, from the time of Kostenki14 about 37,000 years ago until the time of the Villabruna Cluster about 14,000 years ago, all individuals seem to derive from a single ancestral population with no evidence of substantial genetic influx from elsewhere. It is interesting that during this time, the Mal’ta Cluster is not represented in any of the individuals we sampled from Europe. Thus, while individuals assigned to the Gravettian cultural complex in Europe are associated with the Vestonice Cluster, there is no genetic connection between them and the Mal’ta1 individual in Siberia despite the fact that Venus figurines are associated with both. This suggests that if this similarity is not a coincidence[32], it reflects diffusion of ideas rather than movements of people. Third, we find that GoyetQ116-1 derives from a different deep branch of the European founder population than the Vestonice Cluster which became predominant in many places in Europe between 34,000 and 26,000 years ago including at Goyet Cave. GoyetQ116-1 is chronologically associated with the Aurignacian cultural complex. Thus, the subsequent spread of the Vestonice Cluster, which is associated with the Gravettian cultural complex, shows that the spread of the latter culture was mediated at least in part by population movements. Fourth, the population represented by GoyetQ116-1 did not disappear, as its descendants became widespread again after ~19,000 years ago in the El Mirón Cluster when we detect them in Iberia. The El Mirón Cluster is associated with the Magdalenian culture and may represent a post-ice age expansion from southwestern European refugia[33]. Fifth, beginning with the Villabruna Cluster at least ~14,000 years ago, all European individuals analyzed show an affinity to the Near East. This correlates in time to the Bølling-Allerød interstadial, the first significant warming period after the Ice Age[34]. Archaeologically, it correlates with cultural transitions within the Epigravettian in Southern Europe[35] and the Magdalenian-to-Azilian transition in Western Europe[36]. Thus, the appearance of the Villabruna Cluster may reflect migrations or population shifts within Europe at the end of the Ice Age, an observation that is also consistent with the evidence of turnover of mitochondrial DNA sequences at this time[26,37]. One scenario that could explain these patterns is a population expansion from southeastern European or west Asian refugia after the Ice Age, drawing together the genetic ancestry of Europe and the Near East. Sixth, within the Villabruna Cluster, some, but not all, individuals have affinity to East Asians. An important direction for future work is to generate similar ancient DNA data from southeastern Europe and the Near East to arrive at a more complete picture of the Upper Paleolithic population history of western Eurasia[38].