High-coverage whole-genome sequence studies have so far focused on a limited number1 of geographically restricted populations2,3,4,5, or been targeted at specific diseases, such as cancer6. Nevertheless, the availability of high-resolution genomic data has led to the development of new methodologies for inferring population history7,8,9 and refuelled the debate on the mutation rate in humans10. Here we present the Estonian Biocentre Human Genome Diversity Panel (EGDP), a dataset of 483 high-coverage human genomes from 148 populations worldwide, including 379 new genomes from 125 populations, which we group into diversity and selection sets. We analyse this dataset to refine estimates of continent-wide patterns of heterozygosity, long- and short-distance gene flow, archaic admixture, and changes in effective population size through time as well as for signals of positive or balancing selection. We find a genetic signature in present-day Papuans that suggests that at least 2% of their genome originates from an early and largely extinct expansion of anatomically modern humans (AMHs) out of Africa. Together with evidence from the western Asian fossil record11, and admixture between AMHs and Neanderthals predating the main Eurasian expansion12, our results contribute to the mounting evidence for the presence of AMHs out of Africa earlier than 75,000 years ago.
The paths taken by AMHs out of Africa (OoA) have been the subject of considerable debate over the past two decades. Fossil and archaeological evidence13,14, and craniometric studies15 of African and Asian populations, demonstrate that Homo sapiens was present outside of Africa ~120–70 thousand years ago (kya)11. However, this colonization has been viewed as a failed expansion OoA16 since genetic analyses of living populations have been consistent with a single OoA followed by serial founder events17.
Ancient DNA (aDNA) sequencing studies have found support for admixture between early Eurasians and at least two archaic human lineages18,19, and suggest modern humans reached Eurasia at around 100 kya12. In addition, aDNA from modern humans suggests population structuring and turnover, but little additional archaic admixture, in Eurasia over the last 35–45 thousand years20,21,22. Overall, these findings indicate that the majority of human genetic diversity outside Africa derives from a single dispersal event that was followed by admixture with archaic humans18,23.
We used ADMIXTURE to analyse the genetic structure in our diversity set (Extended Data Figs 1, 2; Supplementary Information 1.1–7). We further compared the individual-level haplotype similarity of our samples using fineSTRUCTURE (Extended Data Fig. 3). Despite small sample sizes, we inferred 106 genetically distinct populations forming 12 major regional clusters, corresponding well to the 148 self-identified population labels. This clustering forms the basis for the groupings used in the scans of natural selection. Similar genetic affinities are highlighted by plotting the outgroup f3 statistic9 in the form f3(X, Y; Yoruba), which here measures shared drift between a non-African population X and any modern or ancient population Y from Yoruba as an African outgroup (Supplementary Information 2.2.6, Extended Data Fig. 4).
Our sampling allowed us to consider geographic features correlated with gene flow by spatially interpolating genetic similarity measures between pairs of populations (Supplementary Information 2.2.2). We considered several measures and report gradients of allele frequencies in Fig. 1, which was compared to gene flow patterns from EEMS24 as a validation (Extended Data Fig. 5). Controlling for pairwise geographic distance, we find a correlation between these genetic gradients and geographic and climatic features such as precipitation and elevation (inset of Fig. 1, Supplementary Information 2.2.2).
We screened for evidence of selection by first focusing on loci that showed the highest allelic differentiation among groups (Supplementary Information 3). We then performed positive and purifying selection scans (Methods), and found some candidate loci that replicate previously known and functionally supported findings (Supplementary Table 1:3.3.4-I, Supplementary Information 3.1, Extended Data Fig. 6; Supplementary Table 1:3.1-IV,VI). Additionally, we infer more purifying selection in Africans in genes involved in pigmentation (bootstrapping p value (bpv) for RX/Y scores < 0.05) (Extended Data Fig. 6) and immune response against viruses (bpv < 0.05), while further purifying selection was indicated on olfactory receptor genes in Asians (bpv < 0.05) (Supplementary Table 1:3.1.1-II). Our scans for ancient balancing selection found a significant enrichment (FDR < 0.01) of antigen processing/presentation, antigen binding, and MHC and membrane component genes (Supplementary Information 3.2 and 3.3, Supplementary Table 1:3.3.2-I–III). The HLA (HLA-C)-associated gene (BTNL2) was the top highest scoring candidate in 8 of 12 geographic regions for the HKA test (Supplementary Table 1:3.3.1-I). Our positive selection scans, variant-based analyses (Supplementary Information 3.2 and 3.3) and gene enrichment studies also suggest new candidate loci (Supplementary Information 3.4 and 3.5, Supplementary Table 1:3.5-I–VI), a subset of which is highlighted in Supplementary Table 1:3-I.
Using fineSTRUCTURE, we find in the genomes of Papuans and Philippine Negritos more short haplotypes assigned as African than seen in genomes for individuals from other non-African populations (Extended Data Fig. 7). This pattern remains after correcting for potential confounders such as phasing errors and sampling bias (Supplementary Information 2.2.1). These shorter shared haplotypes would be consistent with an older population split25. Indeed, the Papuan–Yoruban median genetic split time (using multiple sequential Markovian coalescent (MSMC)) of 90 kya predates the split of all mainland Eurasian populations from Yorubans at ~75 kya (Supplementary Table 1:2.2.3-I, Extended Data Fig. 4, Fig. 2a). This result is robust to phasing artefacts (Extended Data Fig. 8, see Methods). Furthermore, the Papuan–Eurasian MSMC split time of ~40 kya is only slightly older than splits between west Eurasian and East Asian populations dated at ~30 kya (Extended Data Fig. 4). The Papuan split times from Yoruba and Eurasia are therefore incompatible with a simple bifurcating population tree model.
At least two main models could explain our estimates of older divergence dates for Sahul populations from Africa than mainland Eurasians in our sample: 1) admixture in Sahul with a potentially un-sampled archaic human population that split from modern humans either before or at the same time as did Denisova and Neanderthal; or 2) admixture in Sahul with a modern human population (extinct OoA line; xOoA) that left Africa after the split between modern humans and Neanderthals, but before the main expansion of modern humans in Eurasia (main OoA).
We consider support for these two non-mutually exclusive scenarios. Because the introgressing lineage has not been observed with aDNA, standard methods are limited in their ability to distinguish between these hypotheses. Furthermore, we show (Supplementary Information 2.2.7) that single-site statistics, such as Patterson’s D9,18 and sharing of non-African Alleles (nAAs), are inherently affected by confounding effects owing to archaic introgression in non-African populations23. Our approach therefore relies on multiple lines of evidence using haplotype-based MSMC and fineSTRUCTURE comparisons (which we show should have power at this timescale26; Supplementary Information 2.2.13).
We located and masked putatively introgressed27 Denisova haplotypes from the genomes of Papuans, and evaluated phasing errors by symmetrically phasing Papuans and Eurasians genomes (Methods). Neither modification (Fig. 2a, Supplementary Information 2.2.9, Supplementary Table 1:2.2.9-I) changed the estimated split time (based on MSMC) between Africans and Papuans (Methods, Supplementary Information 2.2.8, Extended Data Fig. 8, Supplementary Table 1.2.8-I). MSMC dates behave approximately linearly under admixture (Extended Data Fig. 8), implying that the hypothesized lineage may have split from most Africans around 120 kya (Supplementary Information 2.2.4 and 2.2.8).
We compared the effect on the MSMC split times of an xOoA or a Denisova lineage in Papuans by extensive coalescent simulations (Supplementary Information 2.2.8). We could not simulate the large Papuan–African and Papuan–Eurasian split times inferred from the data, unless assuming an implausibly large contribution from a Denisova-like population. Furthermore, while the observed shift in the African–Papuan MSMC split curve can be qualitatively reproduced when including a 4% genomic component that diverged 120 kya from the main human lineage within Papuans, a similar quantity of Denisova admixture does not produce any significant effect (Extended Data Fig. 8). This favours a small presence of xOoA lineages rather than Denisova admixture alone as the likely cause of the observed deep African–Papuan split. We also show (Methods) that such a scenario is compatible with the observed mitochondrial DNA and Y chromosome lineages in Oceania, as also previously argued13,28.
We further tested our hypothesized xOoA model by analysing haplotypes in the genomes of Papuans that show African ancestry not found in other Eurasian populations. We re-ran fineSTRUCTURE adding the Denisova, Altai Neanderthal and the Human Ancestral Genome sequences29 to a subset of the diversity set. FineSTRUCTURE infers haplotypes that have a most recent common ancestor (MRCA) with another individual. Papuan haplotypes assigned as African had, regardless, an elevated level of non-African derived alleles (that is, nAAs fixed ancestral in Africans) compared to such haplotypes in Eurasians. They therefore have an older mean coalescence time with our African samples.
Owing to the deep divergence between the sampled Denisova and the one introgressed into modern humans, it is possible that some archaic haplotypes have a MRCA with an African instead of Denisova and are assigned as ‘African’. We can resolve the coalescence time, and hence origin, of these haplotypes by their sequence similarity with modern Africans. To account for the archaic introgression we modelled these genomic segments as a mixture of haplotypes assigned a) as African or b) as Denisova in Eurasians and c) haplotypes assigned as Denisova in Papuans. These haplotypes are modelled (see Methods, Extended Data Fig. 9) in terms of the distribution of length and mutation rate measured as a density of non-African derived alleles. Since Eurasians (specifically Europeans) have not experienced Denisova admixture, this approach disentangles lineages that coalesce before the human/Denisova split from those that coalesce after.
We found that the xOoA signature (Fig. 2b–d; Supplementary Information 2.2.10) was necessary to account for the number of short haplotypes with ‘moderate’ nAAs density in the data (that is, proportion of non-African-derived sites higher than that of Eurasian haplotypes assigned as African but significantly lower than that of those assigned Denisova in either Eurasians or Papuans). Consistent with our MSMC findings (Supplementary Information 2.2.4), xOoA haplotypes have an estimated MRCA 1.5 times older than the Eurasian haplotypes in Papuan genomes, while the Denisovan haplotypes in Papuans are four times older than the Eurasian haplotypes. Adding up the contributions across the genome (Methods) leads to a genome-wide estimate of 1.9% xOoA (95% confidence interval 1.5–3.3) in Papuans, which we view as a lower bound.
Our results consistently point towards a contribution from a modern human source for derived29 alleles that are found in the genome sequence of Papuans but not in Africans. Possible confounders could involve a shorter generation time in Papuan and Philippine Negrito populations30, different recombination processes, or alternative demographic histories that have not been investigated here. We therefore strongly encourage the development of new model-based approaches that can investigate further the haplotype patterns described here.
In conclusion, our results suggest that while the genomes of modern Papuans derive primarily from the main expansion of modern humans out of Africa, we estimate that at least 2% of their genome sequence reflects an earlier, otherwise extinct, dispersal (Extended Data Fig. 10).
The inferred date of the xOoA split time (~120 kya) is consistent with fossil and archaeological evidence for an early expansion of H. sapiens from Africa13,14. Furthermore, the recently identified modern human admixture into the Altai Neanderthal before 100 kya12 is consistent with a modern human presence outside Africa well before the main OoA split time (~75 kya). Further studies will confirm whether the Papuan genetic signature reported here and the one observed in Altai Neanderthals reflect the same xOoA human group, as well as clarify the timing and route followed during such an early expansion. The high similarity between Papuans and the Altai Neanderthal reported in Extended Data Fig. 1 may indeed reflect a shared xOoA component. Further studies are needed to explore this model and suggest that understanding human evolutionary history will require the recovery of aDNA from additional fossils, and further archaeological investigations in under-explored geographical regions.
No statistical methods were used to predetermine sample size. The experiments were not randomized and the investigators were not blinded to allocation during experiments and outcome assessment.
We analyse a set of genomes sequenced by the same technology (Complete Genomics Inc.) which results in minimal platform differences between batches of samples analysed by slight modifications of CG proprietary pipeline (Extended Data Fig. 2; Supplementary Information 1.6). Informed consent forms and REC approvals were obtained for all samples newly collected for this study. We see good concordance between CG sequence and Illumina genotyping array results for the same samples with minor reference bias in the latter data (Extended Data Fig. 2; Supplementary Information 1.6). In the final dataset, we retained only one second-degree (Australians, to make use of all the available samples) and five third-degree relatives pairs (Supplementary Table 1:1.7-I). All genomes were annotated against the Ensembl GRCh37 database and compared to dbSNP Human Build 141 and Phase 1 of the 1000 Genomes Project dataset29 (Supplementary Information 1.1–1.6). We found 10,212,117 new SNPs, 401,911 of which were exonic. As expected from our sampling scheme, existing lists of variable sites have been extended mostly by the Siberian, Southeast Asian and South Asian genomes, which contribute 89,836 (22.4%), 63,964 (15.9%) and 40,758 (10.1%) of the new exonic variants detected in this study.
Compared to the genome-wide average, we see fewer heterozygous sites on chromosomes 1 and 2, and an excess on chromosomes 16, 19 and 21 (Extended Data Fig. 2). This pattern is independent of simple potential confounders, such as rough estimates of recombination activity and gene density (Supplementary Information 1.8), and mirrors the inter-chromosomal differences in divergence from chimpanzee31, suggesting large-scale differences in mutation rates among chromosomes. We confirmed this general pattern using 1000 Genomes Project data (Supplementary Information 1.8).
The ‘ancient genome diversity panel’ consisted of 106 samples from the main Diversity panel along with Altai Neanderthal, Denisova and the Modern Human reference genome. Sites that are heterozygous in archaic humans were removed.
Geographic gradient analyses
We used a Gaussian kernel smoothing (based on the shortest distance on land to each sample) to interpolate genetic patterns across space. Averaging over all markers, we obtained an expression for the mean square gradient of allele frequencies in terms of the matrix of genetic distance between pairs of samples (Supplementary Information 2.2.2). This provides a simple way to identify spatial regions that contribute strongly to genetic differences between samples, and can be used, in principle, for any measure of genetic difference (for fineSTRUCTURE data, we used negative shared haplotype length as a measure of differentiation).
To quantify the link between the magnitude of genetic gradients (from fineSTRUCTURE and allele frequency data) and geographic factors, we fitted a generalized linear model to the sum of genetic magnitude gradients on the shortest paths between samples to elevation, minimum quarterly temperature, and annual precipitation summed in the same way, controlling for path length and spatial random effects (Supplementary Information 2.2.2), and calculated partial correlations between genetic gradient magnitudes and geographic factors.
FineSTRUCTURE32 was run as described in Supplementary Information 2.2.1. Within the 106 genetically distinct genetic groups, labels were typically genetically homogeneous—113 of the 148 population labels (76%) were assigned to only one ‘genetic cluster’. Similarly, genetic clusters were typically specific to a label, with 66 of the 106 ‘genetic clusters’ (62%) containing only one population label.
Correction for phasing errors
To check whether phasing errors could produce the shorter Papuan haplotypes, we focused on regions of the genome that had an extended (>500 kb) run of homozygosity. We ran ChromoPainter for each individual on only these regions, meaning each individual was only painted where it had been perfectly phased. This did not change the qualitative features (Supplementary Information 2.2.1).
Removal of similar samples
Papuans are genetically distinct from other populations due to tens of thousands of years of isolation. We wanted to check whether the length of haplotypes assigned as African was biased by the inclusion of a large number of relatively homogeneous Eurasians with few Papuans. To do this we repeated the n = 447 painting allowing only donors from dissimilar populations, including only individuals who donated < 2% of a genome in the main painting. This did not change the qualitative haplotype length features (Supplementary Information 2.2.1).
Inclusion of ancient samples
We ran our smaller individual panel with (n = 109) and without (n = 106) ancient samples (Denisova, Neanderthal and ancestral human). This did not change the qualitative haplotype length features (Supplementary Information 2.2.1).
We investigated balancing, positive and purifying selection for a part of the dataset with larger group sizes which was defined as the Selection subset (Supplementary Table 1:3.1-I and 3.2-I) using a wide range of window-based as well as variant-based approaches. Furthermore, we investigated how these signals relate to shared demographic history. Where possible we contextualized our findings by integrating them with information from various functional databases. Detailed descriptions of all methods used are available in Supplementary Information section 3.
MSMC, Denisova masking, simulations of alternative scenarios and assessment of phasing robustness
Genetic split times were initially calculated following the standard MSMC procedure8, and subsequently modified as follows. To estimate the effect of archaic admixture, putative Denisova haplotypes were identified in Papuans using a previously published method27 and masked from all the analysed genomes. Particularly, whether a putative archaic haplotype was found in heterozygous or homozygous state within the chosen Papuan genome, the ‘affected’ locus was inserted into the MSMC mask files and, hence, removed from the analysis.
We note that a fraction of the Denisova and Neanderthal contributions to the Papuan genomes may be indistinguishable, owing to the shared evolutionary history of these two archaic populations. As a result, some of the removed ‘Denisova’ haplotypes may have actually entered the genome of Papuans through Neanderthal. Regardless of this, our exercise successfully shows that the MSMC split time estimates are not affected by the documented presence of archaic genomic component (whether coming entirely from Denisova or partially shared with Neanderthal).
We further excluded the role of Denisova admixture in explaining the deeper African–Papuan MSMC split times through coalescent simulations (using ms to generate 30 chromosomes of 5 Mbp each, and simulating each scenario 30 times). These showed that the addition of 4% Denisova lineages to the Papuan genomes does not change the MSMC results, while the addition of 4% xOoA lineages recreates the qualitative shift observed in the empirical data.
Phasing artefacts were also taken into account as putative confounders of the MSMC split time estimates. We re-ran MSMC after re-phasing one Estonian, one Papuan and 20 West African and Pygmy genomes in a single experiment. This way we ruled out potential artefacts stemming from the excess of Eurasian over Sahul samples during the phasing process. Both the archaic and phasing corrections yielded the same split time as of the standard MSMC runs.
Emulation of all pairwise MSMC split times
We confirmed that none of the other populations behaved as an outlier from those identified in the n = 22 full pairwise analysis by estimating the MSMC split times between all pairs. We chose 9 representative populations (including Papuan, Yoruba and Baka) from the 22, and compared each of the 447 diversity panel genomes to them. For each individual not in our panel, we obtain the positive mixture weights using the modelThe parameters are estimated using the observations for which we have data using a quadratic loss function. We can then predict the unobserved valuesExamination of this matrix (Supplementary Information 2.2.3, Supplementary Table 1:2.2.3-III) implies no other populations are expected to have unusual MSMC split times from Africa.
Mixture model for African haplotypes in Papuans
Obtaining haplotypes from painting. We define African or Archaic haplotypes in Eurasians or Papuans as genomic loci spanning at least 1,000 bp, and showing SNPs that were assigned by chromopainter a ≥50% chance of copying from either an African or Archaic genome, respectively. For each haplotype we then calculated the number of non-African mutations, defined as sites found in derived state in a given haplotype and in ancestral state in all of the African genomes included in the present study.
Modelling. We used a non-parametric model for the joint distribution of length and non-African derived allele mutation rate in haplotypes. We fit K = 20 components to the joint distribution. Each component has a characteristic length , variability and mutation rate . A haplotype of length with such mutations from component has the following distribution:This model for haplotype lengths is motivated by the extreme age of the split times we seek to model. Recent splits would lead to an exponential distribution of haplotype lengths. However, owing to haplotype fixation caused by finite population size, very old splits have finite (non-zero) haplotype lengths. Additionally, the data are left-censored since we cannot reliably detect haplotypes that are very short. We note that while this makes a single component a reasonable fit to the data, as K increases the specific choice becomes less important.
We then impose the prior and use the expectation-maximization algorithm to estimate the mixture proportions along with the maximum likelihood parameter estimates . We do this for the four combinations of haplotypes assigned as African (AFR) and Denisova (DEN) found in Papuans (PNG) or Europeans (EUR), in order to learn the parameters. Supplementary Information 2.2.10 describes this in more detail. We then describe the distribution of haplotypes for each class of haplotype in terms of the expected proportion of haplotypes found in each component,where is the number of haplotypes of class . is a vector of the proportions from each of the components.
Single-out-of-Africa model. We fit haplotypes assigned as African in Papuans as a mixture of the others in a second layer of mixture modelling:where sum to 1. This is straightforward to fit.
xOoA model. We jointly estimate an additional component and the mixture contributions under the mixtureThis is non-trivial to fit. We use a penalization scheme to simultaneously ensure we a) obtain a valid mixture for ; b) give a prediction that is also a valid mixture; c) leave little signal in the residuals; and d) obtain a good fit. Cross-validation is used to obtain the optimal penalization parameters ( and ) with the loss function:where are the residuals in each component, (for a valid mixture) and (for requirement c, good solutions will have similar residuals across components). The loss is minimized via standard optimization techniques. Supplementary Information 2.2.10 details how initial values are found and explores the robustness of the solution to changes in A and B—the results do not change qualitatively for reasonable choices of these parameters, and the mixtures are valid to within numerical error.
Genome-wide xOoA estimation. We used the estimated xOoA derived allele mutation rate estimate to estimate the xOoA contribution in haplotypes classed as Eurasian or Papuan by ChromoPainter. First we obtained estimates of and using the single out-of-Africa model above, additionally allowing for a EUR.EUR contribution. We then estimate using the observed mutation rate and that predicted under the mixture model by rearranging the mixture:Estimates less than 0 are set to 0. The genome-wide estimate is obtained by weighting each by the proportion of the genome that was painted with that donor. Neanderthal and Denisova haplotypes were assumed to be proxied by PNG.DEN (0% xOoA by assumption); African haplotypes by PNG.AFR; Papuan and Australian by PNG.PNG and all other haplotypes by PNG.EUR. We obtain confidence intervals by bootstrap resampling of haplotypes for each donor/recipient pair.
We estimate the proportion of xOoA in Papuan haplotypes assigned as both Eurasian (0.1%, 95% CI 0–2.6) and Papuan (4%, 95% CI 2.9–4.5) (Supplementary Information 2.2.10), by using the estimated mutation density in xOoA.
Y chromosome and mtDNA haplopgroup analysis. The presence of an extinct xOoA trace in the genome of modern Papuans may seem at odds with analyses of mtDNA and Y chromosome phylogenies, which point to a single, recent origin for all non-African lineages (mtDNA L3, which gives rise to all mtDNA lineages outside Africa has been dated at ~70,000 years old33,34). However, uniparental markers inform on a small fraction of our genetic history, and a single origin for all non-African lineages does not exclude multiple waves OoA from a shared common ancestor. We show analytically (Supplementary Information 2.2.12) that, if the xOoA signature entered the genome of Papuan individuals > 40 kya, their mtDNA and Y lineages could have been lost by genetic drift even assuming an initial xOoA mixing component of up to 35%. Similar findings have been reported recently13.
European Nucleotide Archive
The newly sequenced genomes are part of the Estonian Biocentre human Genome Diversity Panel (EGDP) and were deposited in the ENA archive under accession number PRJEB12437 and are also freely available through the Estonian Biocentre website (www.ebc.ee/free_data)
Support was provided by: Estonian Research Infrastructure Roadmap grant no 3.2.0304.11-0312; Australian Research Council Discovery grants (DP110102635 and DP140101405) (D.M.L., M.W. and E.W.); Danish National Research Foundation; the Lundbeck Foundation and KU2016 (E.W.); ERC Starting Investigator grant (FP7 - 261213) (T.K.); Estonian Research Council grant PUT766 (G.C. and M.K.); EU European Regional Development Fund through the Centre of Excellence in Genomics to Estonian Biocentre (R.V.; M.Me. and A.Me.), and Centre of Excellence for Genomics and Translational Medicine Project No. 2014-2020.4.01.15-0012 to EGC of UT (A.Me.) and EBC (M.Me.); Estonian Institutional Research grant IUT24-1 (L.S., M.J., A.K., B.Y., K.T., C.B.M., Le.S., H.Sa., S.L., D.M.B., E.M., R.V., G.H., M.K., G.C., T.K. and M.Me.) and IUT20-60 (A.Me.); French Ministry of Foreign and European Affairs and French ANR grant number ANR-14-CE31-0013-01 (F.-X.R.); Gates Cambridge Trust Funding (E.J.); ICG SB RAS (No. VI.58.1.1) (D.V.L.); Leverhulme Programme grant no. RP2011-R-045 (A.B.M., P.G. and M.G.T.); Ministry of Education and Science of Russia; Project 6.656.2014/K (S.A.F.); NEFREX grant funded by the European Union (People Marie Curie Actions; International Research Staff Exchange Scheme; call FP7-PEOPLE-2012-IRSES-number 318979) (M.Me., G.H. and M.K.); NIH grants 5DP1ES022577 05, 1R01DK104339-01, and 1R01GM113657-01 (S.Tis.); Russian Foundation for Basic Research (grant N 14-06-00180a) (M.G.); Russian Foundation for Basic Research; grant 16-04-00890 (O.B. and E.B); Russian Science Foundation grant 14-14-00827 (O.B.); The Russian Foundation for Basic Research (14-04-00725-a), The Russian Humanitarian Scientific Foundation (13-11-02014) and the Program of the Basic Research of the RAS Presidium “Biological diversity” (E.K.K.); Wellcome Trust and Royal Society grant WT104125AIA & the Bristol Advanced Computing Research Centre (http://www.bris.ac.uk/acrc/) (D.J.L.); Wellcome Trust grant 098051 (Q.A.; C.T.-S. and Y.X.); Wellcome Trust Senior Research Fellowship grant 100719/Z/12/Z (M.G.T.); Young Explorers Grant from the National Geographic Society (8900-11) (C.A.E.); ERC Consolidator Grant 647787 ‘LocalAdaptatio’ (A.Ma.); Program of the RAS Presidium “Basic research for the development of the Russian Arctic” (B.M.); Russian Foundation for Basic Research grant 16-06-00303 (E.B.); a Rutherford Fellowship (RDF-10-MAU-001) from the Royal Society of New Zealand (M.P.C.).
Extended data figures
This file contains Supplementary Tables.
About this article
Journal of Human Genetics (2018)