A major topic of interest in human prehistory is how the large-scale genetic structure of modern populations outside of Africa was established. Demographic models have been developed that capture the relationships among small numbers of populations or within particular geographical regions, but constructing a phylogenetic tree with gene flow events for a wide diversity of non-Africans remains a difficult problem. Here, we report a model that provides a good statistical fit to allele-frequency correlation patterns among East Asians, Australasians, Native Americans, and ancient western and northern Eurasians, together with archaic human groups. The model features a primary eastern/western bifurcation dating to at least 45,000 years ago, with Australasians nested inside the eastern clade, and a parsimonious set of admixture events. While our results still represent a simplified picture, they provide a useful summary of deep Eurasian population history that can serve as a null model for future studies and a baseline for further discoveries.
Overview of Best-Fitting Admixture Graph
As a starting point for our model, we used the set of populations [minus Dai] from an admixture graph formulated in Mallick et al. [2016]: Chimpanzee, Altai Neanderthal [Prüfer et al. 2014], Denisova [Meyer et al. 2012], Dinka, Kostenki 14 [K14, a ∼37 kya Upper Paleolithic individual from Russia belonging to the western Eurasian clade] [Seguin-Orlando et al. 2014], New Guinea, Australia, Onge [an indigenous population from the Andaman Islands], and Ami [aboriginal Taiwanese, representing East Asians]. The elements of the model in Mallick et al. [2016] were mostly relatively straightforward, with no admixture events aside from those involving archaic humans. The primary finding of interest was that the Australasians [plus Onge] fit best as a clade with East Asians; incorporating a deeper “southern route” ancestry component did not improve the fit.
Here, for our primary results, we used single nucleotide polymorphisms [SNPs] genotyped on the Affymetrix Human Origins array, which gave us access to larger sample sizes and additional populations beyond those that are currently available with whole-genome sequencing data. With the nine populations listed above [here New Guinea Highlanders {Reich et al. 2011} rather than the SGDP Papuan {Mallick et al. 2016}], we replicated the earlier results: The graph fit well with the same topology, correctly predicting all f-statistic relationships to within |Z|=2.14Z2.14 [standard errors estimated by block jackknife; see Materials and Methods].
To this preliminary model, we added four additional populations: MA1, Ust’-Ishim, Mamanwa [a “Negrito” population from the Philippines], and Suruí [an indigenous population from Brazil]. Mamanwa [with one component related to Australasians and the other to East Asians; Reich et al. 2011; Lipson et al. 2014] and Suruí [with one component related to MA1 and the other to East Asians; Raghavan et al. 2014] immediately had clear signatures of admixture, while Ust’-Ishim required excess Neanderthal ancestry [Fu et al. 2014, 2016]. After adding these admixtures and optimizing the topology, the resulting model had 42 statistics that differed by at least two standard errors from their fitted values [max Z = 2.68]. Many of these were highly correlated, for example with New Guinea substituted for Australia or with either Dinka or Chimp as an outgroup. We identified four residuals that were independent and reflected quartets of populations forming approximately unadmixed subtrees in the fitted model [to aid in interpretation; absolute value < 0.1 in 1000 times drift units]: f4[MA1, Ami; Denisova, Dinka] [residual Z = 2.03], f4[Suruí, Altai; New Guinea, Australia] [Z = 2.07], f4[Ust’-Ishim, Onge; Australia, Ami] [Z = 2.17], and f4[MA1, K14; Ami, Ust’-Ishim] [Z = 2.31]. The first and last of these motivated us to add additional admixture events to the model [see “Archaic humans” and “Western and northern Eurasians”]; the third is related to a residual signal we discuss in more detail below [“Replication with SGDP data”]; and the second could be connected to deeply splitting ancestry in Amazonian populations [Skoglund et al. 2015] [see “Native Americans”] but was not addressed here.
The full best-fitting admixture graph is shown in figure 1. It is based on a total of ∼123k SNPs and includes 13 leaf nodes [i.e., directly sampled groups]: Seven present-day populations, three ancient modern humans, two archaic humans, and Chimp. There are nine admixture events, of which six are from archaic humans [although these likely do not all represent separate historical events, as discussed below]. All f-statistics fit to within |Z|=2.31Z2.31 , including all pairwise f2-statistics to within |Z|=1.36Z1.36 [fig. 2]; inferred mixture proportions are indicated in figure 1 and can also be found in table 1. In what follows, we describe the features of the graph in more detail.
Fig. 1
Final best-fitting graph model. Colors of filled nodes [sampled populations and selected internal split points] and edge arrows correspond to subsets of the graph: green, Chimp and archaic; yellow, African and basal non-African; dark blue, eastern clade; light blue, Australasian sub-clade; red, western clade; purple, northern sub-clade. Tree edges [solid lines] are labeled with branch lengths in 1000 times drift units [rounded to the nearest integer value], while admixtures [dotted lines] are shown with their inferred proportions. The three drift lengths surrounding an admixture event [immediately preceding each mixing population and immediately following the admixed population] cannot be solved for individually in our framework and instead form a single compound parameter [Lipson et al. 2013]; we omit the first two and report the total drift on the edge following the admixture. The terminal drifts leading to ancient individuals are inflated as a result of a combination of single-individual populations, lower coverage, and/or haploid genotype calls.
Archaic Humans
The top portion of the graph contains Altai, Denisova, and their ancestors. We included one of two previously inferred admixtures into Denisova [Prüfer et al. 2014]: “unknown archaic” ancestry from a source splitting deeper than the common ancestor of the Neanderthal/Denisova clade with modern humans. We do not have enough constraint to solve for the precise mixture proportion [Materials and Methods] and thus prespecified it at 3%, within the range of the initial estimate. Allowing this proportion to vary freely only slightly improved the log-likelihood score of the model, whereas removing the admixture decreased the log-likelihood by about 5.9 [P < 0.005 by likelihood ratio test {LRT}; see Materials and Methods]. We also considered the previously reported Neanderthal admixture into Denisova, but our model does not provide sufficient constraint to observe this signal, so for the sake of parsimony we omitted it from the final graph.
Our model also includes several instances of gene flow from archaic to modern humans. All present-day non-African populations in the graph fit well with a single, shared Neanderthal introgression event. Consistent with previous results [Fu et al. 2014, 2016; Seguin-Orlando et al. 2014], Ust’-Ishim and K14 require extra Neanderthal ancestry, with inferred proportions of 1.5% each; we use the same mixing Neanderthal for all events. We note that while there is evidence that present-day Europeans and related groups have less Neanderthal ancestry than East Asians [Wall et al. 2013; Sankararaman et al. 2014; Vernot and Akey 2014; Lazaridis et al. 2016], no such populations are present in our model [although see “Western and northern Eurasians” below]. As an overall trend, we recapitulate the finding of increased archaic ancestry in ancient individuals [Fu et al. 2016], which could be evidence of purifying selection against introgressed segments over time [Harris and Nielsen 2016; Juric et al. 2016]. Thus, while the graph contains separate Neanderthal gene flow events for Ust’-Ishim and K14, these do not necessarily reflect additional historical episodes of admixture.
In addition to the previously documented Denisova-related introgression into Australasians [here 3.5% into the common ancestor of New Guinea, Australia, and Mamanwa], we find suggestive new evidence for Denisova-related ancestry in MA1, which we believe may explain the preliminary residual statistic f4[MA1, Ami; Denisova, Dinka] mentioned above. A consistent signal of excess allele sharing between MA1 and archaic humans can be observed when using any of Denisova, Altai Neanderthal, or the Vindija and Mezmaiskaya Neanderthals [Green et al. 2010] [table 2]. We also used an ancient ingroup in place of Ami to ensure that this pattern does not reflect an ancient DNA artifact [table 2, bottom half]. While the differences between the rows in table 2 are not statistically significant, MA1 appears to share the most drift with Denisova; the excess shared drift with Neanderthals would also be expected in a scenario of Denisova-related introgression on the basis of the sister relationship between Neanderthals and Denisova.
Motivated by these results, we tested the effects of including extra archaic ancestry in MA1 in our full graph model. Adding Denisova-related admixture resulted in a significant log-likelihood score improvement of 6.3 [P < 0.002], whereas instead allowing additional Neanderthal gene flow improved the score by 2.3 [P = 0.10]. [We note that while the inferred best-fitting source for the Denisova-related introgression was closer to the Denisova sample than for Australasians, the difference in fit quality was negligible, so for the sake of parsimony we used the same source for both events.] Given the consistent pattern of greater Neanderthal ancestry in ancient samples, however, a model with excess Neanderthal ancestry would perhaps be a more reasonable null hypothesis. Using such a model as a starting point, adding Denisova-related admixture improved the score by a marginally significant 4.0 [P < 0.02; when including Denisova-related admixture, the graph fit best without any extra Neanderthal gene flow]. In our final model, we therefore [tentatively] included Denisova-related [but not excess Neanderthal] gene flow into MA1, with an inferred mixture proportion of 1.2%, or 1.0% Denisova-related ancestry [95% confidence interval 0.4–1.6%; see Materials and Methods] in MA1 after dilution by eastern Eurasian gene flow [while we specified the Denisova-related admixture to be older, exchanging the order did not affect the quality of fit]. Placing the Denisova-related admixture in the deeper northern Eurasian lineage shared with Native Americans made the score slightly worse, so in the absence of any evidence for shared Denisova-related ancestry, we retained the mixture only into MA1. We also experimented with allowing Denisova-related ancestry in East Asians but did not find any improvement in the fit, although we would not have power to detect a very small contribution as previously inferred [Prüfer et al. 2014; Sankararaman et al. 2016].
Asian and Australasian Populations
Consistent with previous results obtained with a simpler admixture graph in Mallick et al. [2016], New Guinea and Australia fit well as sister groups, with their majority ancestry component forming a clade with East Asians [with respect to western Eurasians]. Onge fit as a near-trifurcation with the Australasian and East Asian lineages, while Mamanwa are inferred to have three ancestry components: One branching deeply [but unambiguously] from the Australasian lineage [prior to the split between New Guinea and Australia]; one East Asian-related [interpreted as Austronesian admixture]; and one from Denisova. The Denisova-related introgression in Mamanwa is shared with New Guinea and Australia and then diluted ∼3× by the Austronesian admixture [here 68.5%, when compared with 73% in Reich et al. [2011] and 50–60% in a simpler model in Lipson et al. [2014]]. In a previous study [Reich et al. 2011], Australia and New Guinea were modeled as having about half of their ancestry from each of two components: One forming a trifurcation with Onge and East Asians, and the other splitting more recently from the Onge lineage. Here, we obtain a satisfactory fit without this admixture, and while we cannot rule it out entirely, we do not have strong evidence for rejecting our simpler model. We also note that the previous model, by virtue of its different topology, included relatively more Denisova-related ancestry in Mamanwa [∼50% as much as in Australia], although both versions appear to fit the data satisfactorily.
Early Out-of-Africa Split Points
After the divergence of Dinka from non-Africans, the next split point on the modern human lineage in our model is that between the major eastern and western clades [the node labeled “Non-African”—although we note that the split point of Basal Eurasian would be deeper.] This split is soon followed on the western Eurasian branch by the split between K14 and Ust’-Ishim [i.e., their respective modern-human ancestry components]. The original Ust’-Ishim analysis [Fu et al. 2014] inferred a near-trifurcation at this point, and we wished to test whether K14 [and other western Eurasians] and Ust’-Ishim form a statistically supported clade. In fact, while the best-fitting position for Ust’-Ishim is on the western lineage [0.6 shared drift], the inferred 95% confidence interval for this point overlaps the eastern/western split [standard error 0.4 for the Ust’-Ishim split position], so that we cannot confidently resolve the branching order. We therefore continue to regard this cluster as approximately a trifurcation; while we show Ust’-Ishim at its best-fitting split point in figure 1, we color-code it as a basal non-African rather than a member of the western clade.
We also investigated another near-trifurcation, near the top of the eastern Eurasian clade, where the East Asian, Onge, and Australasian lineages are inferred to diverge in a short span. Here, the best-fitting arrangement features Onge and East Asians as a weak clade [ p∼0.02p0.02 ], but the model reaches a second, only slightly inferior local optimum with Onge and Australasians as sister groups instead, possibly suggesting admixture between two of the three lineages. An admixture event in either Onge [between the Australasian and East Asian lineages] or Australasians [between the Onge and deep eastern Eurasian lineages] is likewise weakly significant [ p∼0.02p0.02 ], but with no discriminatory power between these two scenarios. Ultimately, we chose to present the model with a trifurcation at this point because we felt it best conveyed our uncertainty: No pair of lineages clearly shares more drift, and it is likely that some admixture took place, but we cannot accurately determine which lineage or lineages were involved or constrain the exact proportions or sources.
Discussion
Our proposed admixture graph provides both an integrated summary of many population relationships among diverse non-African modern human groups and a framework for testing additional hypothesis. As an example of the model’s utility, it can help to evaluate a signal previously used to argue for deeply diverged ancestry in Aboriginal Australians [Rasmussen et al. 2011]: f4[Australia, Han; Yoruba, French] = 2.02 [Z = 8.23], or with related populations in our admixture graph, f4[Australia, Ami; Dinka, MA1] = 1.43 [Z = 3.01]. While southern route ancestry would indeed cause these statistics to be positive, the admixture events specified in our model provide two [partial] alternative explanations, namely Denisova-related introgression into Australasians and gene flow between eastern and western Eurasians. In fact, the predicted value of f4[Australia, Ami; Dinka, MA1] in our final graph is 1.72, slightly larger than the observed value, without any southern route ancestry in Australasians. Thus, this example illustrates how fitting a large number of groups simultaneously can add context to the interpretation of observed patterns in population genetic data.
In addition to providing a synthesis of previous results, we have also proposed two new admixture events in MA1, both of which seem plausible on geographical grounds but are not overwhelmingly statistically significant and would be interesting topics for further study. One event, consisting of gene flow from an eastern Eurasian population, appears to be present as well in other later western Eurasians, which makes it unlikely that the signal could be due to contamination in MA1. We note though that this event could have involved one or more [unknown] intermediate populations rather than being direct, and we also cannot rule out a small amount of admixture in the reverse direction. The second, consisting of Denisova-related gene flow, provides intriguing evidence of a novel instance of archaic introgression, but it should be subject to additional scrutiny with more sensitive methods to confirm whether the source has been accurately inferred [as opposed to excess introgression from Neanderthal or a different archaic group].
Furthermore, we show that we can obtain a good fit to the data with no further admixture events beyond those specified in our model. In other words, even though our graph is in some ways relatively simple, with only three admixtures among the 10 modern human populations, we do not find any large residuals [to the extent that we have statistical power]. This does not mean that we have identified all admixture events in the ancestry of these populations or that our graph represents the exact historical truth; rather, we propose that our model be viewed as a reasonable and relatively comprehensive starting point given currently available data, in the spirit of previous demographic null models [Schaffner et al. 2005; Gravel et al. 2011]. We also have not included certain groups with known complicated histories, including present-day European and Indian populations. We attempted to add Indians to the graph but failed to obtain a satisfactory fit, which we believe was primarily due to difficulty in modeling the western Eurasian [ANI] ancestry found in all Indian groups today [in addition to eastern, “ASI” ancestry] [Reich et al. 2009].
Overall, our model supports a rapid radiation of Eurasian populations following an out-of-Africa dispersal, in line with results from uniparental markers [Karmin et al. 2015; Posth et al. 2016]. Here, this pattern is reflected in the near-trifurcations among the main eastern and western Eurasian clades plus Ust’-Ishim [with Oase 1 splitting very close as well] and at the base of the eastern clade among Andamanese, Australasians, and East Asians. We note that archaeological evidence increasingly points to early eastward modern human dispersals, with suggestive remains from East and Southeast Asia [Mijares et al. 2010; Demeter et al. 2015; Liu et al. 2015] in addition to the relatively early sites confidently assigned to modern humans in Australasia [O’Connell and Allen 2015; Clarkson et al. 2015]. If these finds do all represent true modern human occupation, this would change our understanding of the timing of out-of-Africa migrations, but it would not necessarily be the case that present-day Asians and Australasians are related to these first inhabitants. It is also possible that Australasians possess a few percent ancestry from an early-dispersal population which could be detectable with more sensitive genetic analyses but not with our allele-frequency-based methods [Mallick et al. 2016; Pagani et al. 2016]. Finally, we caution that we are limited in our ability to infer the geographical locations and calendar dates of events in the admixture graph; our most powerful temporal constraint comes from Ust’-Ishim, whose date of ∼45 kya places the eastern/western Eurasian split no later than this time. Further analysis of ancient DNA in the context of present-day genetic variation promises to provide additional data points to refine and add detail to our understanding.
posted
"We note that archaeological evidence increasingly points to early eastward modern human dispersals, with suggestive remains from East and Southeast Asia [Mijares et al. 2010; Demeter et al. 2015; Liu et al. 2015] in addition to the relatively early sites confidently assigned to modern humans in Australasia [O’Connell and Allen 2015; Clarkson et al. 2015]. If these finds do all represent true modern human occupation, this would change our understanding of the timing of out-of-Africa migrations, but it would not necessarily be the case that present-day Asians and Australasians are related to these first inhabitants."
Its funny they are denying the obvious here - that the archaeological evidence does not support Out of Africa.
Posts: 949 | From: England | Registered: Oct 2015
| IP: Logged |
UBBFriend: Email this page to someone!
Printer-friendly view of this topic