We present evidence for three to five cases of interbreeding among four distinct hominin populations.… Clearly the real population history is likely to have been even more complex. For example, most cases of gene flow are likely to have occurred intermittently, often in both directions and across a geographic range. Thus, combinations of gene flow among different groups and substructured populations may have yielded the patterns detected rather than the discrete events considered here.44
Furthermore, while in figure 2.8 recent human groups are at least embedded in a blue bubble—which we might interpret as genetic exchange among them—in the tree in figure 2.9, which suggests an additional introgression from the modern human into the Neandertal lineage, the modern human populations again appear as unmixed among themselves.
Figure 2.8
“A possible model for gene flow events in the late Pleistocene.” Kay Prüfer et al., “The Complete Genome Sequence of a Neanderthal from the Altai Mountains,” Nature 505, no. 7481 (2014): 48. Reproduced with permission from SNCSC.
Figure 2.9
“Refined demography of archaic and modern humans.” Martin Kuhlwilm et al., “Ancient Gene Flow from Early Modern Humans into Eastern Neanderthals,” Nature 530, no. 7591 (2016): 432. Reproduced with permission from SNCSC.
Finally, some archaeologists, geneticists, and paleoanthropologists—including physical anthropologist Chris Stringer, who was a key figure in the synthesis of approaches around the out-of-Africa model—have issued criticism. They have proposed a complete replacement of archaic humans by incoming modern humans in Eurasia and reject many of the proposed admixture events. They defend the out-of-Africa scenario for modern human populations, even though it is not possible to pinpoint one geographical or temporal origin of modern human genomic ancestry in Africa, and even though this origin is not limited to Africa. As Stringer and colleagues explain in a paper from 2021:
Although a mitochondrial “Eve,” a hypothetical female ancestor of everyone alive today, will have existed and probably lived about 200 [thousand years ago], the location where she, or her Y-chromosomal “Adam” counterpart, lived is not necessarily expected to be the birthplace of all human ancestry. Furthermore, the small mitochondrial history traces just one out of a multitude of paths through the greater human genealogy. In many other parts of the genome, the most divergent branch will be found elsewhere in Africa, or sometimes outside of Africa.45
This leaves us with the conundrum that even though “trees are poor representations of genetic history,”46 as the authors put it, this history continues to be represented with a rather treelike figure (figure 2.10):
Figure 2.10
“Separation of modern human and archaic ancestries in the past one million years.” Anders Bergström et al., “Origins of Modern Human Ancestry,” Nature 590, no. 7845 (2021): 234. Reproduced with permission from SNCSC.
The observations so far are obviously connected to the development of technologies used to study human and hominin demographic history and diversity on the bases of modern and ancient DNA. Programs such as STRUCTURE and ADMIXTURE, which underlie the study of African population history discussed above, assess the genetic similarity between individuals and the extent to which populations form distinct clusters. However, the integration of aDNA presents problems, not least due to sample sizes, sample quality and chronological and geographic representativity. Significantly, these programs do not have underlying population-genetic models or hypothesis testing components. As we have seen above, the recovered genetic substructures could have been brought about by several different population histories. Evolutionary geneticist Liisa Loog therefore observes that “conclusions can be easily steered by the subjective biases of a particular researcher.”47 Even though, as we have pointed out above, researchers may assume that living people and populations are a product of admixture between a certain set of distinct ancestral groups that once existed in the past, the observed genetic patterns could be the result of other demographical histories. One cannot, for example, differentiate between one or several admixture events using these programs.
These may be some of the reasons why programs such as STRUCTURE, ADMIXTURE and FINESTRUCTURE, despite their popularity, are not sufficient for many scientists, especially those working with aDNA. Researchers often refer to the early history of human population genetics, and specifically to Cavalli-Sforza’s work, as discussed at the outset of this chapter, when accounting for the fact that they work with methods that can both model population relations and formally test admixture histories.48 In other words, it was partly due to that early history of the field that methods were developed to describe population-tree topologies that could include admixture events. These methods analyze individual genomes, or the allele frequency patterns among populations, and compare the amount of genetic drift to establish population histories.49
In the context of our chapter, the graph-building techniques that visualize data outputs are of particular interest. As geneticist Ajaj Pathak puts it, these techniques “analyse the genetic diversities of many populations and suggest an elaborate treelike topology, illustrating their mutual relationships.”50 Tools like TREEMIX, MIXMAPPER, and QPGRAPH build trees of populations that explain their evolutionary histories, including episodic migrations (or gene flow) and admixtures.
The program TREEMIX, designed to estimate the most likely population trees taking admixture into account, was applied by Kay Prüfer and colleagues to produce figure 2.8. It uses a graph representation to allow for admixture events as well as population splits. There are diverse techniques that are suitable for different purposes and different datasets, and all have their inherent assumptions, possibilities, and limitations, as well as pitfalls that may be exacerbated in the case of aDNA.51 Most significantly for our context, graph-based models such as TREEMIX infer a tree structure (only in subsequent steps “correcting” for admixture or gene flow events), which becomes evident in figure 2.11.
Figure 2.11
(Left): The inferred maximum likelihood tree of human phylogeny relating modern and archaic humans without considering gene flow between them. (Right): The same tree allowing for ten admixture events between continental groups of modern humans on the basis of TREEMIX. Joseph K. Pickrell and Jonathan K. Pritchard, “Inference of Population Splits and Mixtures from Genome-Wide Allele Frequency Data,” PLoS Genetics 8, no. 11 (2012): 8, 10.
The researchers from whose paper figure 2.11 is taken assume that human population history is treelike in order to simplify the search for the “best graph” for the data. While this technique may be computationally efficient—a standard desktop computer could provide the tree structure in five minutes and test for admixtures in a few hours—it models “migration between populations as occurring at single, instantaneous time points,” which the authors of the paper admit is “a dramatic simplification of the migration process.” The researchers express their expectation that the assumption of treeness could eventually be relaxed with an improved search algorithm.52 The appearance of a treelike structure of human history and relatedness respectively diversity may also be enhanced because admixture graphs such as QPGRAPH focus on admixture between populations of interest, hiding the admixed status of populations that are beyond the scope of a study.53
In the case of QPGRAPH, researchers are required to define the number of admixture events as well as the populations that are considered to be admixed. In the case of TREEMIX, the determination of the phylogeny is automated, but the users still decide the list of populations and the number of admixture events. This, combined with the fact that TREEMIX starts from an unadmixed tree (which is a problem especially if many populations are admixed), is regarded as the main drawback of such approaches.54 It has also been observed that tools that require a model for the histories of the populations not in question might lead to erroneous admixture results if these histories are modeled wrongly—again, this is especially a risk in aDNA studies.
Knowledge about population histories is not necessary for the statistical tool QPADM, which can identify plausible admixture histories and estimate admixture proportions. It has become a widely used tool, especially in aDNA studies, to test whether the genetics of a certain population can be explained by admixture between two or more source populations. QPADM is seen to yield accurate results even where data coverage is low, data are missing to a significant degree, or aDNA is damaged. However, scientists in the field have cautioned that ancient and present-day DNA should not be analyzed together, and that QPADM should not be used for population histories that might include extended periods of gene flow between groups.55 As with TREEMIX, the tool assumes a single pulse in a short time, even though, as evolutionary biologist Éadaoin Harney and colleagues note, “real population histories often involve continuous gene flow that occurs over a prolonged period of time.”56 Even in a case of continuous gene flow, QPADM might suggest plausible admixture proportion estimates as a single pulse.
Some tools are not only able to approximate rates of gene flow between different branches of a putative tree from sequence data, but also estimate past population sizes and the dates of population splits. One such tool is the software package G-PHOCS, which was used in the construction of figure 2.8. In most cases, tools for dating admixture events, such as ROLLOFF, ALDER, and MALDER, once again assume one single admixture pulse and cannot, therefore, capture continuous mixing of populations. As we have seen, and as also suggested by the trees in figure 2.11—where the right-hand tree allows for ten admixture events but not for continuous exchange—this is a general problem with these visualizations. As geneticists Joseph Pickrell and David Reich have observed, “One question is whether changes in populations over time are typically gradual—owing to consistent, low-level gene flow between neighboring populations—or punctate, with migration events rapidly altering the genetic composition of a region. One line of work on modeling human history explicitly assumes the latter.”57
While the assumption of such “punctate” events and their implementation in analytical tools are among the factors that seem to favor the persistence of treelike images of human population histories, geneticists have observed that most population genetic models more generally “rely on the assumption that the relationship between populations can be represented as … a phylogenetic tree, i.e. as abrupt splits between different branches of the tree, followed by independent evolution with potential for subsequent episodes of gene flow between them.”58
Conclusion
We began this chapter with a cursory look at the visualization of human history, diversity and kinship in early human population genetic research, when the tree was fundamental. From this foundation, we have identified something akin to a visual paradigm shift with the advent of programs such as STRUCTURE and ADMIXTURE that were associated with an interest in processes of mixing, and in individuals and populations as mixed. The mosaic structure of the colored bar plots produced by ADMIXTURE contradicts an essentialist understanding of the categories of “individual” and “population”—indeed, these categories seem to be in the process of dissolution.
At the same time, there is also a pull in the opposite direction, since the ADMIXTURE bar plots imply pure “ancestral populations” that current populations are thought to be mixtures of. This gains clear visual expression when the findings are converted into a tree that suggests independent evolution, and thus “pure populations.” Similarly, even though the advent of aDNA in population genomics has increased the focus on admixture and introgression, and these processes have been given a deeper history through aDNA, the amount of contact and genetic exchange between groups that researchers envisage tends to be minimized in modeling and visualizing, leaving us with trees that include relatively few connections between branches.
This is probably due to several factors, one of them being that ways of thinking and doing are handed down from one generation of researchers to the next. These ways may be disproportionately shaped by particularly influential scientists and labs.59 There is also the history of methodological and technological developments in a stricter sense—as in the necessity to build on what is already there—and the fact that statistical analyses aim to reduce the complexity of data or fit it to parametric models.
While we might think of computer statistics as automatized, neutral tools, they are in fact shaped by assumptions. We have seen how models that have been developed for aDNA analyses rely on previous patterns of knowing and thinking about kinship and human evolution, most notably the tree form. Although more complex models are appearing on the horizon, human population genomics—also in its so-called revolutionized state after the advent of aDNA studies—instantiates tree-thinking and tree-building that tends to render archaic and modern human groups as more or less discrete, homogeneous entities. Such tree-thinking and tree-building runs in danger of inadvertently conveying older notions about human diversity in terms of race which, as we have pointed out in the beginning of this chapter, also used to be visualized as trees.
Notes
1. In this context, the tree of life is a metaphor to describe the relationships between all organisms, both extant and extinct.
2. Marianne Sommer, “The Meaning of Absence: The Primate Tree that Did Not Make It into Darwin’s The Descent of Man,” BJHS Themes 6 (2021): 45–61.
3. An allele is one of two or more versions of a gene. The word is used to describe genetic variation among genes, but it can also be used to describe genetic variation in noncoding regions.
4. Luigi Luca Cavalli-Sforza and Anthony W. F. Edwards, “Analysis of Human Evolution,” in Genetics Today: Proceedings of the XI. International Congress of Genetics, The Hague, The Netherlands, September 1963, vol. 3, ed. S. J. Geerts (Oxford: Pergamon, 1965), 929.
5. Luigi Luca Cavalli-Sforza et al., “Reconstruction of Human Evolution: Bringing Together Genetic, Archaeological, and Linguistic Data,” Proceedings of the National Academy of Sciences 85, no. 16 (1988): 6003.
6. Cavalli-Sforza et al., “Reconstruction of Human Evolution,” 6003; Marianne Sommer, “Population-Genetic Trees, Maps, and Narratives of the Great Human Diasporas,” History of the Human Sciences 28, no. 5 (2015): 108–145.
7. Luigi Luca Cavalli-Sforza, “Analytic Review: Some Current Problems of Human Population Genetics,” American Journal of Human Genetics 25, no. 1 (1973): 96; Sommer, “Population-Genetic Trees,” 120–121.
8. Sommer, “Population-Genetic Trees,” 123–135; Marianne Sommer, History Within: The Science, Culture, and Politics of Bones, Organisms, and Molecules (Chicago: University of Chicago Press, 2016), part III. For histories and critical studies of human population genetics without a special emphasis on visualizations, see Jenny Reardon, Race to the Finish: Identity and Governance in an Age of Genomics (Princeton, NJ: Princeton University Press, 2005); Catherine Nash, Genetic Geographies: The Trouble with Ancestry (Minneapolis: University of Minnesota Press, 2015).
