Denisova admixture and the first modern human dispersals into Southeast Asia and Oceania

Abstract

It has recently been shown that ancestors of New Guineans and Bougainville Islanders have inherited a proportion of their ancestry from Denisovans, an archaic hominin group from Siberia. However, only a sparse sampling of populations from Southeast Asia and Oceania were analyzed. Here, we quantify Denisova admixture in 33 additional populations from Asia and Oceania. Aboriginal Australians, Near Oceanians, Polynesians, Fijians, east Indonesians, and Mamanwa (a "Negrito" group from the Philippines) have all inherited genetic material from Denisovans, but mainland East Asians, western Indonesians, Jehai (a Negrito group from Malaysia), and Onge (a Negrito group from the Andaman Islands) have not. These results indicate that Denisova gene flow occurred into the common ancestors of New Guineans, Australians, and Mamanwa but not into the ancestors of the Jehai and Onge and suggest that relatives of present-day East Asians were not in Southeast Asia when the Denisova gene flow occurred. Our finding that descendants of the earliest inhabitants of Southeast Asia do not all harbor Denisova admixture is inconsistent with a history in which the Denisova interbreeding occurred in mainland Asia and then spread over Southeast Asia, leading to all its earliest modern human inhabitants. Instead, the data can be most parsimoniously explained if the Denisova gene flow occurred in Southeast Asia itself. Thus, archaic Denisovans must have lived over an extraordinarily broad geographic and ecological range, from Siberia to tropical Asia.

Figure 1

Denisovan Genetic Material as a Fraction of that in New Guineans Populations are only shown as having Denisova ancestry if the estimates are more than two standard errors from zero (we combine estimates for populations in this study with analogous estimates from CEPH- Human Genome Diversity Panel populations reported previously¹²). No population has an estimate of Denisova ancestry that is significantly more than that in New Guineans, and hence we at most plot 100%. The sampling location of the AU2 population is unknown and hence the position of this population is not precise.

Figure 2

Denisovan and Near Oceanian Ancestry Are Proportional Except in the Philippines We plot p_D(X), the estimated percentage of Denisova ancestry as a fraction of that seen in New Guineans, against the estimated percentage of Near Oceanian ancestry p_N(X) by using the values from Table 1 (horizontal and vertical bars specify ±1 standard errors). The Mamanwa deviate significantly from the p_D(X) = p_N(X) line, indicating that their Denisova genetic material does not owe its origin to gene flow from a population related to Near Oceanians. A weaker deviation is seen in the Manobo, who live near the Mamanwa on the island of Mindanao.

Figure 3

A Model of Population Separation and Admixture that Fits the Data The admixture graph suggests Denisova-related gene flow into a common ancestral population of Mamanwa, New Guineans, and Australians, followed by admixture of New Guinean and Australian ancestors with another population that did not experience Denisova gene flow. We cannot distinguish the order of population divergence of the ancestors of Chinese, Onge/Jehai, and Mamanwa/New Guineans/Australians, and hence show a trifurcation. Admixture proportion estimates (red) are potentially affected by ascertainment bias and hence should be viewed with caution. In addition, although admixture graphs are precise about the topology of population relationships, they are not informative regarding timing. Thus, the lengths of lineages should not be interpreted in terms of population split times and admixture events.

Figure 4

Computation of the Estimate of Denisovan Ancestry p_D(X) The black lines show the model for how populations are related that is the basis for the p_D(X) ancestry estimate. Population X arose from an admixture of a proportion (1 − q_X) of ancestry from an ancestral non-African population C′ and (q_X) from archaic population B′ (C and B are their unmixed descendants). The expected value of f₄(A,B;C,X) is proportional to the correlation in the allele frequency differences A − B and C − X, and can be computed as the overlap in the drift paths separating A − B (blue arrows) and C − X (red arrows). These paths only overlap over the branches α and β, in proportion to the percentage q_X of the lineages of population X that are of archaic ancestry and so the expected value is q_X(α + β). When we compute the ratio p_D(X), (α + β) cancels from both the numerator and denominator, and we obtain q_X/q_{New Guinea}, the fraction of archaic ancestry in a population X divided by that in New Guinea. This provides unbiased estimates of the mixture proportion even if populations C and B have experienced a large amount of genetic drift since splitting from their ancestors, that is, even if we do not have good surrogates for the ancestral populations. This robustness arises because the genetic drift on the branches B→B′ and C→C′ does not contribute to the expectations.

Figure 5

Computation of the Estimate of Near Oceanian Ancestry p_N(X) The test population X is assumed to have arisen from a mixture of a proportion (1 − q_X) of ancestry from ancestral East Asians E′ and (q_X) of ancestral Near Oceanians N′. The Near Oceanians are, in turn, assumed to have received a proportion p_X of their ancestry from the Denisovans (E and New Guinea are assumed to be unmixed descendants of these two). The expected value of f₄(A,Australia; X, New Guinea) can be computed from the correlation in the allele frequency differences A − Australia (blue arrows) and X − New Guinea (red arrows). These paths only overlap along the proportion (1 − q_X) of the ancestry of population X that takes the East Asian path, where the expected shared drift is (1 − p_X)β+γ as shown in the figure. Thus, the expected value of the f₄ statistic is (1 − q_X)(1 − p_X)β+γ. Because q_X = 0 for the denominator of p_N(X) (no Near Oceanian ancestry), the ratio of f₄ statistics has an expected value of (1 − q_X) and E[p_N(X)] = q_X.