## Main

The first millennium ce saw major demographic, cultural and political change in Europe, including the rise and fall of the Roman Empire, migration and the emergence of medieval institutions that shaped the modern world. The post-Roman transformation of lowland Britain was particularly profound. The end of the Roman administration in fifth century Britain preceded a dramatic shift in material culture, architecture, manufacturing and agricultural practice, and was accompanied by language change1. The archaeological record and place names indicate shared cultural features across the North Sea zone, in particular, along the east and southeast coasts of present-day England, Schleswig-Holstein and Lower Saxony (Germany), Frisia (Netherlands) and the Jutland peninsula (Denmark)2,3,4. Examples include the appearance of Grubenhäuser (sunken feature buildings), large cremation cemeteries and the styles of cremation urns or objects that used animal art and chip-carved metal7,8,9,10,11. Moreover, wrist clasps, as well as cruciform and square-headed brooches, found in sixth and seventh century Britain had attested southern Scandinavian origins12,13. Despite these similarities across the North Sea zone, there was also insular material culture that had no continental equivalent14,15. Adding to this, some places and geographical features such as rivers retained names of Celtic or late Latin origin16,17.

From the Renaissance to the present day, the primary explanatory narrative for these changes has been invasion and conquest followed by resettlement from the continent18. On the basis of a small set of written sources, it was supposed that the local Romano-British population was largely replaced by migrants from the Germanic-speaking part of the continent. However, the extent to which these traditional cultural historical interpretations explain patterns of material culture or agree with the historical accounts has been questioned18. For example, historical sources going back to Bede (writing in the eight century) indicated Jutes as settlers in Kent. But, in an issue that became known as ‘the problem of the Jutes’19,20,21, this historically attested migration is difficult to determine from or reconcile with the archaeological record. Indeed, material culture elements found in Kent resemble those of contemporary Merovingian France and Alemannic (southern) Germany, rather than the rest of England or Denmark. Such discrepancies between the archaeological record and historical narratives could be argued to support a rejection of migration or invasion hypotheses, and this was the preferred theoretical position of many archaeologists from the 1960s onwards1,18,22. By that time, many scholars favoured a model of elite dominance involving small, mobile warbands and the acculturation of the local British population. However, the available isotopic and DNA evidence, even if hitherto small scale, suggests that immigrants were less wealthy and buried alongside locals23,24,25,26,27,28, which does not fit a model of elite influence that could explain the adoption of a West Germanic language with apparently minimal influence from Celtic or Latin29,30,31,32.

There is a history of addressing these questions using genetic data. After early attempts to use ancient genetic data failed33, researchers turned to studies based on present-day populations and uniparentally inherited markers, but still without reaching consensus. Work based on present-day Y chromosomes inferred 50–100% replacement of male lineages during the Early Middle Ages in eastern England34,35. More recently, the first genome-wide study of present-day British people concluded that immigrant continental northern European ancestry makes up less than 50% of the present-day southeastern English gene pool36. However, populations change over time through drift and gene flow, so present-day populations may be poor proxies for ancient groups of unknown genetic makeup. The feasibility of ancient DNA analyses to inform on population history in Britain was first demonstrated with the report of genome-wide ancient DNA (aDNA) data26,37 from 20 individuals from the Iron Age to the Early Middle Ages, two studies that have provided unambiguous evidence for continental ancestry in early and middle Anglo-Saxon contexts.

Here we investigate early medieval population dynamics in England and across the North Sea zone with the first large-scale genome-wide study of aDNA in this time period and region, increasing the archaeogenetic record in England specifically, from 8 to 285 individuals. We target a comprehensive time transect of sites in the south and east of England, spanning predominantly the time period 450–850 ce, starting with early Anglo-Saxon cemeteries including Apple Down, Dover Buckland, Eastry, Ely, Hatherdene Close, Lakenheath, Oakington, Polhill and West Heslerton. This allows us to address questions concerning the extent of continental migration to England, and its effect on the local insular gene pool. In addition, the association of artefacts with individuals allows us to study the dynamics of the migration process at the community level.

## New aDNA data

We sampled skeletal remains from 494 ancient northwestern Europeans from 37 different sites in England, Ireland, the Netherlands, Germany and Denmark, dated between approximately 200 and 1300 ce (Supplementary Note 1 and Supplementary Table 1). We prepared powder from skeletal material, extracted aDNA and converted it into double-stranded or single-stranded libraries (Methods). We selected 439 libraries for hybridization DNA capture to enrich for sequences that overlapped 1.24 million single-nucleotide polymorphisms (SNPs). For 40 libraries, we generated complete genomes without capture, with a mean coverage of 0.9×.

After quality filtering (Methods) and exclusion of duplicate individuals, genome-wide data for 460 individuals were available for analysis. These include 278 ancient individuals from England, and 182 individuals from neighbouring ancient populations in Ireland and the European continent (Fig. 1). We combined our newly reported data with published aDNA from 4,336 individuals (Supplementary Note 2), including 1,098 post-Neolithic genomes from northwestern Europe26,37,38,39,40,41,42,43,44. We also compiled a reference dataset of 10,176 present-day European individuals36,45,46,47 genotyped on an intersection of 445,171 SNPs (Supplementary Note 2). To aid interpretation of our genetic data, we also radiocarbon-dated 57 samples selected on the basis of ancestry composition, burial assemblage and preservation.

## Population shifts in post-Roman England

We performed principal component analysis (PCA) on 5,365 present-day northwestern Europeans, from Ireland to Sweden, and projected our ancient genomes onto this genetic variation (Fig. 2). For present-day variation, PC1 and PC2 broadly reflect geography, forming a V-shaped pattern from Scandinavians via individuals from northern Germany and the Netherlands towards those from Britain and Ireland. We highlight the position of individuals from present-day England (Fig. 2a), which follow a clinal distribution defined by the western British and Irish (WBI; which includes Irish, Northern Irish, Scottish and Welsh) at one extreme and overlapping present-day Dutch at the other extreme. The ancient genomes fall onto a slightly separate cline, with most of the early medieval individuals from Dutch, German and Danish sites plotting on top of present-day continental northern Europeans (CNEs; northern Germans and Danish), whereas Bronze and Iron Age individuals from Britain and Ireland cluster together with WBI (Fig. 2b). Of note, in contrast to the preceding Bronze and Iron Age individuals from Britain and Ireland, the majority of the early medieval samples from England (England EMA) plot together with the ancient individuals from the continental North Sea area along with the present-day CNEs. The divergence between prehistoric and early medieval individuals from England is also seen in the distribution of genetic distances (FST) as well as shared alleles (F4) on both the population (Extended Data Fig. 1) and the individual scale (Supplementary Fig. 3.3). We notice that the individuals from early medieval English sites are distinctly heterogeneous in the first two PCs and cover the full extent of the cline between the Bronze and Iron Age cluster and the early medieval cluster.

These genetic patterns suggest that early medieval individuals from England have variable amounts of CNE ancestry. Although most individuals from early medieval English sites cluster clearly with either present-day WBI samples or CNEs, many individuals fall between these two clusters, suggesting admixture between these ancestral groups. To quantitatively estimate these ancestry compositions, we decomposed ancestral sources using a supervised clustering approach implemented in the software ADMIXTURE48. Specifically, we assembled modern populations into two metapopulations that serve as proxies for the source ancestries in early medieval England defined above: CNE (n = 407) and WBI (n = 667). We confirmed that these two present-day metapopulations accurately represent the ancient admixture sources by testing their relationships to the ancient individuals from England using FST statistics and F4 statistics of the form F4(Yoruba, Test; WBI, CNE) (Extended Data Fig. 1 and 2). The resulting ancestry estimates for early medieval English individuals are indeed tightly congruent with both PCA PC1 position and F4 statistics (Pearson's |r| > 0.9 between PCA, F4 and ADMIXTURE ancestry assessments).

Applying our CNE–WBI ancestry decomposition to prehistoric samples, we found the genome-wide CNE ancestry in Britain and Ireland to be very low before the Early Middle Ages (Extended Data Fig. 3). In Bell Beaker and Bronze Age individuals from England, CNE ancestry does not account for more than 1% (Fig. 3a). This cannot be explained by genetic drift due to the temporal gap between our present-day CNE proxy and the Bronze Age, as shown by F4 statistics (Extended Data Figs. 1 and 2), which are robust against such drift. Similar proportions were also measured during the Iron Age (1% on average). CNE ancestry increased only during the Roman period, to 15% on average, although this estimate is based on only seven individuals. Six of those seven Roman-era individuals are from a single site, Eboracum (present-day York); which was a Colonia, the highest rank of Roman city with a legionary fortress, and as such it may have attracted a more cosmopolitan population than most of the rest of England (Fig. 3b).

In contrast to these previous periods, the majority of the early medieval individuals from England in our sample derive either all or a large fraction of their ancestry from continental northern Europe, with CNE ancestry of 76 ± 2% on average (Methods). Although CNE ancestry is predominant in central and eastern England, it is much less prevalent in the south and southwest of England, and absent in the one site that we analysed from Ireland (Fig. 3b). Moreover, we observed differences in continental ancestry not only between but also within sites. Although we estimate CNE ancestry to be prevalent across eastern English cemeteries, there was considerable variation at the individual level, ranging from 0% to 100% of CNE ancestry within a site. For example, at Hatherdene Close (n = 17) in Cambridgeshire, we estimated a mean CNE ancestry of approximately 70%, with eight individuals exhibiting exclusively CNE ancestry, but three individuals having low or zero CNE ancestry. Overall, these patterns of genetic heterogeneity, from the transregional to the family level, are consistent with continuous interaction between the Iron Age-derived Romano-British population and migrants from the continent.

We find no significant differences of CNE or WBI ancestry between male and female individuals (Supplementary Note 7), and find individuals of both ancestries within prominent and/or furnished burials. In England overall, individuals with CNE ancestry (here and in the following, CNE means more than 50% CNE, and WBI means less than 50% CNE) are more likely to be found with grave goods than individuals with WBI ancestry (Fisher’s exact test P = 0.016). Of note, this appears to be driven by female individuals with CNE ancestry who are more likely to be found with grave goods (P = 0.001), and in particular brooches (P = 0.012), than female individuals with WBI ancestry (both based on Fisher’s exact test). However, graves belonging to male individuals with CNE ancestry are just as likely to have grave goods (P = 0.57) or weapons (P = 1) as those with WBI ancestry (both based on Fisher’s exact test). This is underlined by specific examples, such as a near 100% WBI male burial in grave 37 at Updown Eastry found with a seax under a barrow marked by a ring ditch, indicating a prominent weapon burial associated with a prominent person or status (Supplementary Fig. 1.1).

This pattern is also visible in East Anglia specifically, where individuals with CNE ancestry more often have grave goods (P = 0.014). This is also significant when considering only female individuals (P = 0.025), but not when considering females with brooches, which display gender-related status (P = 0.197). At the site level, these patterns are partly significant at Hatherdene Close (P = 0.015, 0.036 and 0.1, respectively). Treating ancestry not as a binary but as a continuous variable largely agrees with the previous results (see Supplementary Note 7), with a notable exception of West Heslerton, which stands out from this overall pattern, where men with a greater proportion of CNE ancestry are more likely to be found with weapons (Wilcoxon rank sum P = 0.02, although non-significant with Fisher’s exact test P = 0.53), which is the only significant signal of this type that we found (Lakenheath also displays many CNE burials with weapons, but with limited sample size).

There are notable individual exceptions to these patterns, such as a predominantly (60%) WBI burial at grave 80 in Oakington, found with the skeleton of a cow, silvered disc brooches and a chatelaine, and interred under a barrow, which is one of the more notable or wealthy burials in this cemetery49 (Supplementary Fig. 1.4). We note that several burials with weapons that were previously identified as female and discussed in the literature23,50 have turned out to be genetically male in our analysis (see the highlighted entries in Supplementary Table 1). Of note, however, a single individual still displays a sex–gender difference: a teenage boy buried with an equal-arm brooch, beads and a knife (grave 122 in West Heslerton).

In Dover Buckland, one of the most comprehensively sampled cemeteries in our dataset, we observed the mixing of genetic and cultural identities at the family level. For example, we found a group of relatives, spanning at least three generations, who all exhibit unadmixed CNE ancestry (Extended Data Fig. 4a,c). Down the pedigree, we then see the integration of a female into this group, who herself had unadmixed WBI ancestry (grave 304), and two daughters (graves 290 and 426), consequently of mixed ancestry. WBI ancestry entered again one generation later, as visible in near 50:50 mixed-ancestry grandchildren (graves 414, 305 and 425). Grave goods, including brooches and weapons, are in fact found on both sides of this family tree, pre-mixing and post-mixing (for example, in the youngest and mixed generation, we found both weapons, beads and pin, and their mother with a brooch). Although the first mixed generation is buried in close proximity to each other, the grandchildren are elsewhere on the site, although placed together (Extended Data Fig. 4b).

A quite different pattern is observed at Apple Down, which is among the most western sites that we have analysed. Here graves can be classified into distinct burial configurations according to orientation, location and frequency of artefacts. We found that burials with CNE ancestry are more often buried in configuration A (located towards the middle of the site and with east–west burial orientation) than in configuration B (located more towards the edges and with north–south orientation)49 (Fisher’s exact test P = 0.048). This shows that there is a significant difference within the treatment of individuals according to their ancestry, a finding very similar to those at early medieval cemeteries in Hungary and Italy with respect to northern versus southern European ancestry51.

## Ancestry sources across the North Sea

Our new continental medieval data from regions bordering the North Sea provide a unique opportunity to further investigate the potential source of the CNE-related ancestry increase that we have described above (Supplementary Note 3). To this end, we first selected individuals who, according to our CNE–WBI decomposition, are of unadmixed CNE ancestry (CNE of more than 95%; from here from as England EMA CNE). For each site in the continental dataset, we then tested whether its individuals were genetically similar to the England EMA CNE group (n = 109) in terms of allele frequencies. Among the continental medieval groups analysed, sites from both northern Germany and Denmark are indeed indistinguishable from England EMA CNE individuals (Fig. 4). Consistently, England EMA CNE and medieval individuals from Lower Saxony exhibit almost identical genetic affinities and ancestry components (Extended Data Fig. 5 and Supplementary Fig. 3.2), possess the highest level of genetic similarity (based on F2, F3, F4 and FST statistics) (Extended Data Fig. 5 and Supplementary Fig. 3.8) and are symmetrically related to most ancient and modern populations (Supplementary Table 3.12). Together, this suggests that they are likely derived from the same source population. Using the LOCATOR52 software, which uses machine learning to map individuals into geographical space based on their genetic profiles, we infer a region spanning the northern Netherlands to the southernmost tip of Sweden as a putative source for the England EMA CNE ancestors, with a large proportion of individuals being assigned to Lower Saxony (see Methods) (Fig. 4). This similarity adds to previous evidence from the material culture and burial practices, especially between the Elbe-Weser region and the early Anglo-Saxon cemeteries, from which the archaeological migration discourse initially arose53. However, we also note the strong genetic homogeneity among most analysed sites in the northern Netherlands, northern Germany and Denmark (Supplementary Note 4), implying that, during the Early Middle Ages, the continental North Sea and adjacent western Baltic Sea area was a genetic continuum spanning most of the western North European plain without major geographical substructure (Supplementary Fig. 4.1,4.4). This, together with genetic backflow from the British–Irish Isles into continental Europe (Supplementary Table 4.2 and Supplementary Fig. 4.2,4.4), reflects the inferred linguistic history54 and precludes further identification of specific microregions that contributed gene flow to Britain. We note that, although our screening of plausible medieval continental sites is broad, it could overemphasize later developments of the genetic structure due to the increased replacement of cremation burials by inhumations on the continent. It also has a specific caveat in Scandinavia, where our medieval reference populations are mostly from Viking-era burials, which have diverse and mixed ancestries that may not be representative of the earlier populations there42,44.

Already during the Early Middle Ages, several individuals from multiple sites exhibit modest degrees of excess affinity (5.4%) to present-day individuals from the Scandinavian peninsula (Supplementary Fig. 6.2a), indicating additional sources. Although close cultural contacts to the Scandinavian peninsula are attested in the archaeological record3, we did not find this genetic variation to be geographically stratified within early medieval England (Supplementary Fig. 6.2b). This Scandinavian Peninsula-related ancestry increases substantially (to 30.6%) only during the Viking period (Supplementary Note 6).

Having established the close relatedness between specific continental regions and the individuals from early medieval England, we modelled the latter more directly using ancient source populations with the method qpAdm35. Specifically, we pooled ancient individuals in England by site and modelled each group as being admixed between two sources: one represented by pooled Iron Age/Roman period individuals from England, and the other represented by pooled early medieval individuals from Lower Saxony (from here known as LowerSaxony EMA). The resulting admixture proportions obtained in this way for early medieval sites in England are strongly correlated with our mean estimates from supervised ADMIXTURE above, as well as site-wise F4 statistics and mean PCA position (Pearson's |r| > 0.9 between all four ancestry assessments) (Extended Data Fig. 6).

Using this model, we detected an average of 86 ± 2% ancestry from Lower Saxony across all early medieval sites in England, only slightly higher than the 76 ± 2% estimated using present-day source populations and supervised ADMIXTURE. At a regional scale, we observed more ancestry from Lower Saxony in eastern England than in the southwest, consistent with ancestry arriving from the east, either in one event or over a continuous time period. Our estimate of genome-wide ancestry is supported by independent evidence for population turnover from uniparental markers (Supplementary Fig. 2.7). Before the Middle Ages, post-Neolithic individuals from Britain and Ireland carried overwhelmingly the major Y chromosomal haplogroup R1b-P312, especially the sub-haplogroup R-L21 (refs. 39,41), which in present day shows a cline across the region, with highest frequencies in the west55,56. By contrast, the early medieval population of England exhibits a substantial fraction of continental-derived haplotypes belonging to haplogroups R1b-U106, R1a-M420, I2a1-L460 and I1-M253, which are commonly found in northern and central Europe (and are also common among ancient continental individuals including the ones that we report). In particular, Y chromosomal haplogroups I1-M253 and R1a-M420 were absent from our Bronze, Iron and Roman Age British and Irish individuals, but were identified in more than one-third of our individuals from early medieval England. Overall, haplogroups absent in Bronze and Iron Age England represent at least 73 ± 4% of the Y chromosomes in our early medieval English sample, mirroring the turnover estimates from autosomal data. Similarly, mitochondrial genomes show evidence of female lineage population turnover from regions bordering the North Sea (Supplementary Note 2 and Supplementary Fig. 2.4).

Estimates of continental ancestry on the X chromosomes (Supplementary Fig. 2.11), as well as estimates of source origin of Y chromosomal haplogroups (Supplementary Fig. 2.16) point to no significant difference between male-specific lineages and autosomal admixture estimates (Supplementary Note 2). Although neither mitochondrial, Y chromosomal or X chromosomal data can exclude subtle levels of sex bias during the admixture (Supplementary Note 2), they are also consistent with a model of no sex bias, suggesting that the migrants included both men and women who mixed at similar levels with the local population. We note that absence of sex bias during the early medieval CNE–WBI admixture does not exclude possibilities for sex bias in the later admixture processes that caused the dilution of CNE ancestry in present-day England observed below.

## Recent population shifts in England

Although the most prominent signal of admixture in early medieval England is the rise in ancestry related to medieval and modern continental northern Europe, we found that several English sites include genomes that could not be explained as products of admixture between the two hypothesized ancestral gene pools—England IA or LowerSaxony EMA—using qpAdm57. Instead, these genomes have additional continental western and southern European ancestry (Supplementary Note 5). This ancestry is genetically very similar to Iron Age genomes from France5,6 (France IA) (Extended Data Fig. 7, Supplementary Table 5.1 and Supplementary Fig. 5.3,5.4). The majority of this French Iron Age-derived ancestry is found in early medieval southeastern England, namely, at the sites of Apple Down, Eastry, Dover Buckland and Rookery Hill, where it constitutes up to 51% of the ancestry identified (Extended Data Fig. 8a and Supplementary Table 5.4).

The appearance of France IA-related ancestry in early medieval England anticipates a pattern that we also clearly see in the present-day English population structure, in which we found that the same two-way CNE–WBI model that fits most ancient English fails for the modern population (Supplementary Fig. 5.8,5.11). Indeed, the missing component in the modern English population appears to be represented well by France IA (Supplementary Table 5.2 and Supplementary Fig. 5.2).

Using qpAdm (Methods), most present-day Scottish, Welsh and Irish genomes can be modelled as receiving most or all of their ancestry from the British Bronze or Iron Age reference groups, with little or no continental contribution. By contrast, for all present-day English samples the simple two-way admixture model (England LIA + England EMA CNE) fails. By extending our model to a three-way with added France IA as a third component, we now obtain fitting models (Supplementary Fig. 5.11,5.21). We estimate that the ancestry of the present-day English ranges between 25% and 47% England EMA CNE-like, 11% and 57% England LIA-like and 14% and 43% France IA-like. There are substantial genetic differences between English regions (Fig. 5a), with less ancient continental ancestry (England EMA CNE or France IA related) evident in southwestern and northwestern England as well as along the Welsh borders (Fig. 5c). By contrast, we saw peaks in CNE-like ancestry of up to 47% for southeastern, eastern and central England, especially Sussex, the East Midlands and East Anglia. We found substantial France IA ancestry only in England, but not in Wales, Scotland or Ireland, following an east-to-west cline in Britain (Pearson’s |r| > 0.86), accounting for as much as 43% of the ancestry in East Anglia (Fig. 5d). Very similar results were produced using LowerSaxony EMA as a source for CNE ancestry (Extended Data Fig. 8b). One potential caveat in this analysis is our relatively sparse Roman sample from England, where we particularly lack samples from the south, which might have pre-existing France IA-related ancestry. We, therefore, turned to one of our early medieval sites, the post-Roman cemetery of Worth Matravers at the southern coast of Dorset, whose individuals have nearly no CNE ancestry (less than 6% on average), and thus may serve as a more temporally close proxy for post-Roman Britain before the arrival of CNEs. When used as a source in our model, we found that the estimates of France IA-related ancestry in present-day England changed by less than 3% on average across the regions (Fig. 5b), suggesting that France IA-related ancestry entered England to a substantial amount after the Roman period. We note that a model involving southern or western European-like ancestry in England has been previously proposed36 on the basis of present-day samples, but we can now go further and delineate this third component more clearly against the CNE-like immigrant gene pool making up the majority of the early medieval individuals from England that we studied.

Our three-way population model for present-day England supports a view of post-Roman English genetic history as punctuated by gene flow processes from at least two major sources: first, the attested arrival of CNE ancestry during the Early Middle Ages from northern Germany, the Netherlands and Denmark, and second, the arrival of ancestry related to France IA. Although we cannot precisely date the order of those arrivals, at least substantial amounts of France IA-related ancestry seem to be absent in northern and eastern England during the Early Middle Ages and therefore must have arrived there subsequently. In other parts of England, however, it may have entered together with CNE ancestry or even earlier. Notably in southern England, namely, Eastry, Apple Down and Rookery Hill, several early medieval individuals already exhibit France IA-related ancestry, which probably results, at least in part, from localized mobility between the south of England and the Frankish areas of Europe during the Early Middle Ages (Extended Data Fig. 8a). Indeed, Frankish material culture is evident in these regions, particularly in Kent and Sussex58,59,60. Admixture from this second source is, therefore, unlikely to have resulted from a single discrete wave. More plausibly, it resulted from pulses of immigration or continuous gene flow between eastern England and its neighbouring regions.

## Discussion

The ‘Anglo-Saxon settlement’ is among the most intensely debated topics in British history, but much of the discussion remains anchored to the contents of Bede’s Ecclesiastical History and the Anglo-Saxon Chronicle18. These early writings defined the settlement as a single event, or a series of events, tied to the immediate aftermath of the Roman administration in the fifth to sixth century. In the archaeological and historical debate, this has been described as happening to varying degrees; as the Adventus Saxonum (a folk migration of named Germanic tribes), an invasion or the movement of a limited number of elite male migrants18,61. To this day, little agreement has been reached over the scale of migration, the mode of interaction between locals and newcomers, or how the transformation of the social, material, and linguistic or religious spheres was achieved. Here we provide strong evidence of large-scale early medieval migration across the North Sea zone and extend its temporal scope. In particular, we show that these migrations started earlier than previously assumed, as evidenced by individuals with CNE ancestry from later Roman contexts, and continued throughout the middle Anglo-Saxon period. Our results from middle Saxon sites such as Sedgeford push the estimated dates of arrival of CNE ancestry to as late as the eight century and merge these events with interpersonal mobility from Sweden and other Scandinavian regions during the later Viking invasion and settlement. Together, these migrations appear to be part of a continuous movement of people from across the North Sea to Britain from the later Roman period into the eleventh century ce.

Our results overwhelmingly support the view that the formation of early medieval society in England was not simply the result of a small elite migration18,61, but that mass migration from afar must also have had a substantial role. We identified numerous individuals with only continental ancestry, suggesting that many of them were migrants themselves or were their unadmixed descendants. Both the lack of genetic evidence for male sex bias, and the correlation between ancestry and archaeological features, point to women being an important factor in this migration. Although men with migrant and local ancestry were buried in similar ways, women with migrant ancestries were more often found with grave goods than women with local ancestry. This could point to social stratification, or plausibly might simply reflect the degree to which women of local ancestry were integrated into the emerging CNE families. It is clear, however, that these social differences are subtle, given that we did not find evidence for this pattern in male burials, and that we found significant regional and site-level differences. Previous hypotheses about the social mechanisms in this migration have included partial social segregation62, elite migration18,61, substantial population replacement34 or no migration at all1,22. Our combined genetic and archaeological analysis point to a complex, regionally contingent migration with partial integration that was probably dependent on the fortunes of specific families and their individual members.

In present-day Britain, we saw substantial northern continental ancestry, albeit at a lower level than during the early medieval period, pointing to a lasting demographic impact of the ‘Anglo-Saxon’ migrations. Specifically, in early medieval western England, Wales and Scotland, and more generally in England during the Norman period, further aDNA sampling may clarify how CNE ancestry spread and was subsequently diluted. Beyond the substantial early medieval immigration of northwestern continental European people found here, we have also identified a second major source of continental ancestry in modern Britain from sources more to the European south and west. This second ancestry component is already evident in our early medieval samples. In Southeast England specifically, individuals at several sites show ancestry whose closest match is in modern-day western Germany, Belgium and/or France, which matches the Frankish connections seen in the archaeological record for these regions. Our data and analyses indicate that this second genetic introgression continued further into the Middle Ages and potentially beyond.

## Methods

### Study design

#### Archaeological research

Provenance information for samples from all archaeological sites are given in Supplementary Information Section 1, together with short descriptions of each site, the institution owning the samples (or custodians of the samples), the responsible coauthor who obtained permission to analyse and the year of the permission granted.

#### Sampling

Sampling of 494 bone and teeth samples took place in clean-room facilities dedicated to aDNA work, for 296 samples at the Max Planck Institute for Science of Human History in Jena (MPI-SHH), for 65 at the Department of Biological and Geographical Sciences at the University of Huddersfield, for 33 at the Department of Genetics, Harvard Medical School (HMS), for 32 at the Institute for Scientific Archaeology of the Eberhard Karls University Tübingen, for 31 at the Leiden University Medical Centre in Leiden, for 15 at the Globe Institute of the University of Copenhagen, for 12 at the Australian Centre for Ancient DNA at the University of Adelaide, and for 10 at the Research Laboratory for Archaeology, University of Oxford. The sampling workflow included documenting and photographing the provided samples. For teeth processed at the MPI-SHH, we cut along the cementum–enamel junction and collected powder by drilling into the pulp chamber. The teeth processed at the Leiden University Medical Centre were sampled according to a previously published paper63. For the petrous bones, we either cut the petrous pyramid longitudinally to drill the dense part directly from either side64 or applied the cranial base drilling method as previously described65. We collected between 30 and 200 mg of bone or tooth powder per sample for DNA extractions.

#### DNA extraction

At MPI-SHH and HMS, aDNA was extracted following a modified protocol66, as described in www.protocols.io/view/ancient-dna-extraction-from-skeletal-material-baksicwe, in which we replaced the extended-MinElute-column assembly for manual extractions with columns from the Roche High Pure Viral Nucleic Acid Large Volume Kit67, and for automated extraction with a protocol that replaced spin columns with silica beads in the purification step68. Extraction of aDNA at the Leiden University Medical Centre, at the Globe Institute and at the University of Adelaide followed the protocols of Kootker et al.63, Damgaard et al.69 and Brotherton et al.70, respectively. Extraction of aDNA at the Universities of Oxford and Huddersfield followed a published protocol71.

#### Library construction

We generated 104 double-indexed72 double-stranded libraries using 25 µl of DNA extract and following established protocols73. We applied the partial UDG (UDG half)74 protocol to remove most of the aDNA damage while preserving the characteristic damage pattern in the terminal nucleotides. For 375 extracts, we generated double-indexed single-stranded libraries75 using 20 µl of DNA extract and applied no UDG treatment.

#### Shotgun screening, capture and sequencing

Libraries produced at MPI-SHH were sequenced in-house on an Illumina HiSeq 4000 platform to an average depth of 5 million reads and after demultiplexing processed through EAGER76. After an initial quality filter based on the presence of aDNA damage and endogenous DNA higher than 0.1%, we subsequently enriched 439 libraries using in-solution capture probes synthesized by Agilent Technologies for approximately 1240K SNPs along the nuclear genome77. The captured libraries were sequenced for 20–40 million reads on average using either a single end (1 × 75 bp reads) or paired-end configuration (2 × 50 bp reads). In addition, 40 genomes were shotgun sequenced for 225 million reads on average to low coverage. For the 120 samples processed at HMS, we enriched for sequences overlapping approximately 1240K SNPs77 as well as the mitochondrial genome78, and sequenced on Illumina NextSeq500 instruments for 2 × 76 cycles, or on HiseqX10 instruments for 2 × 101 cycles (reading out both indices) approximately until the point in which every additional 100 sequences generated yielded fewer than one additional SNP with data.

### aDNA data processing

#### Read processing and aDNA damage

For data produced at the MPI-SHH, after demultiplexing based on a unique pair of indexes, raw sequence data were processed using EAGER76. This included clipping sequencing adaptors from reads with AdapterRemoval (v2.3.1)79 and mapping of reads with Burrows–Wheeler Aligner (BWA)80 v0.7.12 against the human reference genome hg19, with seed length (-l) disabled, maximum number of differences (-n) of 0.01 and a quality filter (-q) of 30. We removed duplicate reads with the same orientation and start and end positions using DeDup76 v0.12.2. Terminal base deamination damage calculation was done using mapDamage81 v2.0.6, specifying a length (-l) of 100 bp. For the 107 libraries that underwent UDG half treatment, we used BamUtil v1.0.14 (https://genome.sph.umich.edu/wiki/BamUtil:_trimBam) to clip two bases at the start and end of all reads for each sample to remove residual deaminations, thus removing genotyping errors that could arise due to aDNA damage.

For data produced at the HMS, after trimming barcodes and adapters57, we merged read pairs with at least 15 bp of overlap, allowing no more than one mismatch if base quality was at least 20, or up to three mismatches if base qualities were less than 20. We chose the nucleotide of the higher quality in case of a conflict while setting the local base quality to the minimum of the two using a custom toolkit (https://github.com/DReichLab/ADNA-Tools). We aligned merged sequences to human genome hg19 using BWA80 v0.7.15 with a maximum number of differences (-n) of 0.01, a maximum number of gap opens (-o) of 2 and seed length (-l) of 16,500. PCR duplicates were identified by tagging all aligned sequences with the same start and stop positions and orientation and, in some cases, in-line barcodes using Picard MarkDuplicates (http://broadinstitute.Github.io/picard/). We only considered sequences that spanned at least 30 bp, and subsequently selected a single copy of each such sequence that had the highest base-quality score. To remove aDNA damage, we trimmed the last two bases of each sequence for UDG-treated libraries and the last five for non-UDG-treated libraries.

#### Sex determination

To determine the genetic sex of ancient individuals processed at the MPI-SHH, we calculated the coverage on the autosomes as well as on each sex chromosome and subsequently normalized the X reads and Y reads by the autosomal coverage82. For that, we used a custom script (https://github.com/TCLamnidis/Sex.DetERRmine) for the calculation of each relative coverage as well as their associated error bars83. Female individuals were expected to have an X rate of 1 and a Y rate of 0, whereas male individuals were expected to have a rate of 0.5 for both X and Y chromosomes. For individuals processed at the HMS, we calculated the ratio of sequences mapping to the Y chromosome to the sum of sequences mapping to the X and Y chromosome for the 1240K data. A ratio less than 3% is consistent with a female individual and a ratio higher than 35% is consistent with a male individual6.

#### Contamination estimation

We used the Analysis of Next Generation Sequencing Data (ANGSD) package84 (v0.923) to test for heterozygosity of polymorphic sites on the X chromosome in male individuals, applying a contamination threshold of 5% at the results of method two. For male and female samples, we estimated contamination levels on the mitochondrial DNA either using Schmutzi85 (v1.5.4) by comparing the consensus mitogenome of the ancient sample to a panel of 197 worldwide mitogenomes as a potential contamination source (MPI-SHH), or by estimating the match rate to the consensus sequence using contamMix86 v1.0-12 (HMS), applying a contamination threshold of 5%. We used PMDtools87 (v0.50) to isolate sequences from each sample that had clear evidence of contamination (over 5% on the X chromosome or mitogenome) according to the post-mortem damage score (PMD score > 3, using only bases with phred-scaled quality of at least 30 to compute the score), and performed contamination estimation again. If a sample scored below the threshold, it was included in the analysis and modelling. If the authenticity of a sample could not be verified or falsified, it was included in population genetic analyses but not used for modelling. In summary, the median mitochondrial DNA contamination is 1.0%, and the median X chromosome contamination is 1.1% (after PMD filtering).

#### Genotyping

We used the program pileupCaller (v1.4.0.2) (https://github.com/stschiff/sequenceTools.git) to genotype the trimmed BAM files of UDG half libraries. A pileup file was generated using samtools mpileup with parameters -q 30 -Q 30 -B containing only sites overlapping with our capture panel. From this file, for each individual and each SNP on the 1240K panel57,88,89, one read covering the SNP was drawn at random, and a pseudohaploid call was made, that is, the ancient individual was assumed homozygous for the allele on the randomly drawn read for the SNP in question. For libraries that underwent no UDG treatment, we used the parameter -SingleStrandMode, which causes pileupCaller to ignore reads aligning to the forward strand at C/T polymorphisms and at G/A polymorphisms to ignore reads aligning to the reverse strand, which should remove post-mortem damage in aDNA libraries prepared with the non-UDG single-stranded protocol.

#### Mitochondrial and Y chromosome haplogroup assignment

To process mitochondrial DNA data generated at the MPI-SHH, we extracted reads from 1240K data using samtools90 v1.3.1 and mapped these to the revised Cambridge reference sequence. At the HMS, we aligned merged sequences to the mitochondrial genome RSRS91. We subsequently called consensus sequences using Geneious92 R9.8.1 and used HaploGrep 2 (ref. 93) v2.4.0 (https://haplogrep.uibk.ac.at/; with PhyloTree version 17-FU1) to determine mitochondrial haplotypes. For the male individuals processed at the MPI-SHH, we used pileup from the Rsamtools package to call the Y chromosome SNPs of the 1240K SNP panel (mapping quality of 30 or more and base quality of 30 or more). We then manually assigned Y chromosome haplogroups using pileups of Y SNPs included in the 1240K panel that overlap with SNPs included on the ISOGG SNP index v.15.73 (Y-DNA Haplogroup Tree 2019–2020; 2020.07.11). For male individuals processed at the HMS, we automatically determined Y chromosome haplogroups using both targeted SNPs and off-target sequences aligning to the Y chromosome based on comparisons to the Y chromosome phylogenetic tree from Yfull version 8.09 (https://www.yfull.com/)6.

#### Kinship estimation

We calculated the PWMR94 in all pairs of individuals from our pseudo-haploid dataset to double check for potential duplicate individuals and to determine first-degree, second-degree and third-degree relatives. For this purpose, we also used READ95 to determine first-degree, second-degree and third-degree relatedness among individuals based on the proportion of non-matching alleles (P0) in nonoverlapping windows of 1 Mb and to calculate standard errors. We also used the method LcMLkin96, which uses genotype likelihoods to estimate the three k-coefficients (k0, k1 or k2), which define the probability that two individuals have zero, one or two alleles identical by descendent at a random site in the genome. We performed LcMLkin to distinguish between possible parent–offspring or sibling relationships.

### Population genetic analysis

#### Dataset

We merged our aDNA data with previously published datasets of 4,336 ancient individuals reported by the Reich laboratory in the Allen Ancient DNA Resource v.50.0 (https://reich.hms.harvard.edu/allen-ancient-dna-resource-aadr-downloadable-genotypes-present-day-and-ancient-dna-data). We assembled a dataset from mostly European populations for genome-wide analyses36,45,46,47,97,98,99,100,101,102,103. This modern set includes 10,176 individuals (Supplementary Note 2). Loci and individuals with less than 95% call rate as well as a 15-Mb region surrounding the HLA region36 were removed and loci on three previously reported long-range linkage disequilibrium regions on chromosomes 6, 8 and 11 (refs. 104,105) were pruned using PLINK106 (v1.90b3.29). aDNA data were merged to this dataset, correcting for reference allele and strand flips. We kept 445,171 autosomal SNPs after intersecting autosomal SNPs in the 1240K capture with the modern analysis set.

#### Abbreviations

We used the following abbreviations in population labels: N, Neolithic; C, Chalcolithic; EBA, Early Bronze Age; MBA, Middle Bronze Age; LBA, Late Bronze Age; Iron Age, IA; RA, Roman Age; EMA, Early Middle Ages; MA, Middle Ages. In Britain, these periods roughly correspond to the following simplified time ranges: Neolithic: 4000–2500 bce, Chalcolithic and EBA: 2500–1600 bce; MBA: 1600–1200 bce; LBA: 1200–800 bce; IA: 800 bce to 400 ce; EMA 400–1000 ce.

#### PCA

We carried out PCA using the smartpca software v16000 from the EIGENSOFT package (v6.0.1)107. We computed PCs on three different sets of modern European populations (Supplementary Note 2) and projected ancient individuals using lsqproject: YES.

#### F statistics

F3 and F4 statistics were computed with ADMIXTOOLS108 v3.0 (https://github.com/DReichLab). F3 statistics were calculated using qp3Pop (v435). For F4 statistics, we used the qpDstat (v755) and with the activated F4 mode. Significant deviation from zero can be interpreted as rejection of the tree population typology ((outgroup, X);(Pop1, Pop2)). Under the assumption that no gene flow occurred between Pop1 and Pop2 and the outgroup, a positive f-statistic suggests affinity between X and Pop2, whereas a negative value indicates affinity between X and Pop1. Standard errors were calculated with the default block jackknife 5 cM in size.

#### Fixation index

We calculated FST using smartpca software v16000 from the EIGENSOFT package (v6.0.1)107 with the fstonly, inbreed and fsthiprecision options set to YES.

#### Maximum likelihood tree

We constructed maximum likelihood trees using TreeMix (v1.12)109. For each tree, we performed a round of global rearrangements after adding all populations (-global) and calculated 100 bootstrap replicates to assess the uncertainty of the fitted model (-bootstrap). Sample size correction was disabled.

#### Inference of mixture proportions

We estimated ancestry proportions using qpWave57,110 (v410) and qpAdm57 (v810) from ADMIXTOOLS108 v3.0 with the allsnps: YES option and a basic set of 11 outgroups: YRI.SG, Poland, Finland, Sweden, Denmark, Ireland, Wales, Italy, Spain, Belgium and the Netherlands. For some analyses, we added additional outgroups to this basic set (Supplementary Notes 4–6).

#### Prediction of geographical origins

LOCATOR52 (v1.2) was run using a geolocated reference panel consisting of 670 Bronze Age, Iron Age and medieval European samples with 1X coverage higher than 50% and considering only polymorphisms covered at least in 50% of the samples, leaving a total of 920,060 SNPs. Default parameters were used, except that the width of each neural layer was 512 and -imputed-missing was set to YES. The best run was selected as the one showing the lowest validation error and the highest R2 numbers from a total of 40 independent runs.

#### ADMIXTURE analysis

We performed model-based clustering analysis using ADMIXTURE48 (v1.3). We used ADMIXTURE in supervised mode, in which we estimated admixture proportions for the ancient individuals using modern reference populations at various K values (Supplementary Notes 3–6). These analyses were run on haploid data with the parameter –haploid set to all (="*"). Standard errors for point estimates were calculated using 1,000 bootstrap replicates with the -B parameter. To obtain point estimates for populations, we averaged individual point estimates and calculated the standard error of the mean (s.e.m.): s.e.m. = $$\frac{\sigma }{\surd n}$$. We found that this better reflects the diversity within the population than a propagation of error approach, which underestimates the variance within the point estimate sample. For unsupervised ADMIXTURE analysis (Supplementary Notes 3 and 5), we carried out linkage disequilibrium pruning on the dataset using PLINK106 with the flag–indep-pairwise 200 25 0.4, leaving 306,393 SNPs. We ran ADMIXTURE with the cross-validation (–cv.) flag specifying from K = 2 to K = 10 clusters, with five replicates for each value of K. For each value of K, the replicate with highest log likelihood was kept.

### Reporting summary

Further information on research design is available in the Nature Research Reporting Summary linked to this article.