Post by Admin on May 27, 2023 18:23:32 GMT
Discussion
Any attempt to build detailed models of human history is subject to model misspecification. This is true of previous studies, which often assumed that data inconsistent with a single-origin model should be explained by archaic hominin admixture. It is also true of this study. Although it is difficult to fully explore the space of plausible models of early human population structure, we sought to capture uncertainty in the model by exploring multiple parameterizations of early history. The best-fit models presented here include reticulation and migration between early human populations, rather than archaic hominin admixture from long-isolated branches (Fig. 1c). Elements of both recent expansion and African multiregionalism (Fig. 1a,d) feature in our best-fit models, as indicated in the recent time of contemporary population divergence and the gene flow between disparate stems, respectively.
We cannot rule out the possibility that more-complex models involving additional stems, more-complex population structure, or hybrid models including both weak structure and archaic hominin admixture, may better explain the data. Because parameters related to the split time, migration rates and relative sizes of the early stems were variable across models, reflecting a degree of confounding among these parameters, we refrained from introducing additional branches associated with more parameters during that period. Rather than interpreting the two stems as representing well-defined and stable populations over hundreds of thousands of years, we interpret the weakly structured stem as consistent with a population fragmentation-and-coalescence model13. Other African populations, such as those from Central Africa, other Khoe-San groups or pre-Holocene ancient DNA samples, could further test our proposed models.
Formation of population structure in Africa
Our inferred models paint a consistent picture of the Middle to Late Pleistocene as a critical period of change, assuming that estimates from the recombination clock accurately relate to geological chronologies (Supplementary Information section 8). During the late Middle Pleistocene, the multiple-merger model indicates three major stem lineages in Africa, tentatively assigned to southern (stem 1S), eastern (stem 1E) and western/central Africa (stem 2). Geographical association was informed by the present population location with the greatest ancestry contribution from each stem. For example, stem 1S contributes 70% to the ancestral formation of the Khoe-San. The extent of the isolation 400 ka between stem 1S, stem 1E and stem 2 suggests that these stems were not proximate to each other. Although the length of isolation among the stems is variable across fits, models with a period of divergence, isolation and then a merger event (that is, a reticulation) out-performed models with bifurcating divergence and continuous gene flow.
A population reticulation involves multiple stems that contribute genetically to the formation of a group. One way in which this can happen is through the geographical expansion of one or both stems. For example, if, during MIS 5, either stem 1S (Fig. 3b) from southern Africa moved northwards and thus encountered stem 2, or stem 2 moved from central–western Africa southwards into stem 1S, then we could observe disproportionate ancestry contributions from different stems in contemporary groups. We observed two merger events. The first, between stem 1S and stem 2, resulted in the formation of an ancestral Khoe-San population around 120 ka. The second event, between stem 1E and stem 2 about 100 ka, resulted in the formation of the ancestors of eastern and western Africans, including the ancestors of people outside Africa. Reticulated models do not have a unique and well-defined basal human population divergence. We suggest conceptualizing the events at 120 ka as the time of most recent shared ancestry among sampled populations. However, interpreting population divergence times in population genetics is always difficult, owing to the co-estimation of divergence time and subsequent migration; methods assuming clean and reticulated splits can infer different split dates (Supplementary Figs. 28 and 36). Therefore, in the literature, wide variation exists in estimates of divergence time1,25.
Shifts in wet and dry conditions across the African continent between 140 ka and 100 ka may have promoted these merger events between divergent stems. Precipitation does not neatly track interglacial cycles in Africa, and heterogeneity across regions may mean that the beginning of an arid period in eastern Africa is conversely the start of a wet period in southern Africa33. The rapid rise in sea levels during the MIS 5e interglacial might have triggered migration inland away from the coasts, as has been suggested, for example, for the palaeo-Agulhas plain34. After these merger events, the stems subsequently fractured into subpopulations which persisted over the past 120 ka. These subpopulations can be linked to contemporary groups despite subsequent gene flow across the continent. For example, a genetic lineage sampled in the Gumuz has a probability of 0.7 of being inherited from the ancestral eastern subpopulation 55 ka, compared with a probability of 0.06 of being inherited from the southern subpopulation (see Table S8 for additional comparisons).
We also find that stem 2 continued to contribute to western Africans during the Last Glacial Maximum (26 ka to 20 ka), indicating that this gene flow probably occurred in western and/or central Africa (Table 1). Such an interpretation is reinforced by differential migration rates between regions; that is, the gene flow from stem 2 to western Africans is estimated to be five times that of the rate to eastern Africans during this period. We performed a variety of validation tests to explore the sensitivity of our assumptions, including relaxing fixed parameters (Supplementary Information section 6). Most of the validation tests resulted in parameters similar to the models discussed above. However, one exception was the inferred out-of-Africa and eastern–western African divergences, which were 10–15 ka younger than our fixed parameters. These younger dates are at odds with the accepted timing of the out-of-Africa expansion that contributed to later human populations at approximately 50 ka, based on archaeological, climatic and fossil information35,36,37,38. Because the inference approach is unbiased in simulations, we interpret the free estimate for eastern African versus European divergence as reflecting our inclusion of only a single out-of-Africa population in the model, the lack of a nearby source for back-to-Africa gene flow, and other regionally complex parameters, rather than a systematic bias that may affect all parameters in the model. Older pan-African features of our inferred models are minimally affected by the choice of these fixed parameters (Supplementary Information section 7.2).
Any attempt to build detailed models of human history is subject to model misspecification. This is true of previous studies, which often assumed that data inconsistent with a single-origin model should be explained by archaic hominin admixture. It is also true of this study. Although it is difficult to fully explore the space of plausible models of early human population structure, we sought to capture uncertainty in the model by exploring multiple parameterizations of early history. The best-fit models presented here include reticulation and migration between early human populations, rather than archaic hominin admixture from long-isolated branches (Fig. 1c). Elements of both recent expansion and African multiregionalism (Fig. 1a,d) feature in our best-fit models, as indicated in the recent time of contemporary population divergence and the gene flow between disparate stems, respectively.
We cannot rule out the possibility that more-complex models involving additional stems, more-complex population structure, or hybrid models including both weak structure and archaic hominin admixture, may better explain the data. Because parameters related to the split time, migration rates and relative sizes of the early stems were variable across models, reflecting a degree of confounding among these parameters, we refrained from introducing additional branches associated with more parameters during that period. Rather than interpreting the two stems as representing well-defined and stable populations over hundreds of thousands of years, we interpret the weakly structured stem as consistent with a population fragmentation-and-coalescence model13. Other African populations, such as those from Central Africa, other Khoe-San groups or pre-Holocene ancient DNA samples, could further test our proposed models.
Formation of population structure in Africa
Our inferred models paint a consistent picture of the Middle to Late Pleistocene as a critical period of change, assuming that estimates from the recombination clock accurately relate to geological chronologies (Supplementary Information section 8). During the late Middle Pleistocene, the multiple-merger model indicates three major stem lineages in Africa, tentatively assigned to southern (stem 1S), eastern (stem 1E) and western/central Africa (stem 2). Geographical association was informed by the present population location with the greatest ancestry contribution from each stem. For example, stem 1S contributes 70% to the ancestral formation of the Khoe-San. The extent of the isolation 400 ka between stem 1S, stem 1E and stem 2 suggests that these stems were not proximate to each other. Although the length of isolation among the stems is variable across fits, models with a period of divergence, isolation and then a merger event (that is, a reticulation) out-performed models with bifurcating divergence and continuous gene flow.
A population reticulation involves multiple stems that contribute genetically to the formation of a group. One way in which this can happen is through the geographical expansion of one or both stems. For example, if, during MIS 5, either stem 1S (Fig. 3b) from southern Africa moved northwards and thus encountered stem 2, or stem 2 moved from central–western Africa southwards into stem 1S, then we could observe disproportionate ancestry contributions from different stems in contemporary groups. We observed two merger events. The first, between stem 1S and stem 2, resulted in the formation of an ancestral Khoe-San population around 120 ka. The second event, between stem 1E and stem 2 about 100 ka, resulted in the formation of the ancestors of eastern and western Africans, including the ancestors of people outside Africa. Reticulated models do not have a unique and well-defined basal human population divergence. We suggest conceptualizing the events at 120 ka as the time of most recent shared ancestry among sampled populations. However, interpreting population divergence times in population genetics is always difficult, owing to the co-estimation of divergence time and subsequent migration; methods assuming clean and reticulated splits can infer different split dates (Supplementary Figs. 28 and 36). Therefore, in the literature, wide variation exists in estimates of divergence time1,25.
Shifts in wet and dry conditions across the African continent between 140 ka and 100 ka may have promoted these merger events between divergent stems. Precipitation does not neatly track interglacial cycles in Africa, and heterogeneity across regions may mean that the beginning of an arid period in eastern Africa is conversely the start of a wet period in southern Africa33. The rapid rise in sea levels during the MIS 5e interglacial might have triggered migration inland away from the coasts, as has been suggested, for example, for the palaeo-Agulhas plain34. After these merger events, the stems subsequently fractured into subpopulations which persisted over the past 120 ka. These subpopulations can be linked to contemporary groups despite subsequent gene flow across the continent. For example, a genetic lineage sampled in the Gumuz has a probability of 0.7 of being inherited from the ancestral eastern subpopulation 55 ka, compared with a probability of 0.06 of being inherited from the southern subpopulation (see Table S8 for additional comparisons).
We also find that stem 2 continued to contribute to western Africans during the Last Glacial Maximum (26 ka to 20 ka), indicating that this gene flow probably occurred in western and/or central Africa (Table 1). Such an interpretation is reinforced by differential migration rates between regions; that is, the gene flow from stem 2 to western Africans is estimated to be five times that of the rate to eastern Africans during this period. We performed a variety of validation tests to explore the sensitivity of our assumptions, including relaxing fixed parameters (Supplementary Information section 6). Most of the validation tests resulted in parameters similar to the models discussed above. However, one exception was the inferred out-of-Africa and eastern–western African divergences, which were 10–15 ka younger than our fixed parameters. These younger dates are at odds with the accepted timing of the out-of-Africa expansion that contributed to later human populations at approximately 50 ka, based on archaeological, climatic and fossil information35,36,37,38. Because the inference approach is unbiased in simulations, we interpret the free estimate for eastern African versus European divergence as reflecting our inclusion of only a single out-of-Africa population in the model, the lack of a nearby source for back-to-Africa gene flow, and other regionally complex parameters, rather than a systematic bias that may affect all parameters in the model. Older pan-African features of our inferred models are minimally affected by the choice of these fixed parameters (Supplementary Information section 7.2).
