A working model for Micronesian population history
We started with a model previously used to study southwest Pacific lineages (8, 9) and then added lineages and admixture events, testing alternative models for fit (Fig. 4C and figs. S12 to S15). With so many populations, the space of possible admixture graph topologies is vast, and the topology we show is unlikely to be a the only fit to the f-statistics. Nevertheless, identifying an admixture graph model is useful to demonstrate that all the features described in our analysis of individual f-statistics can jointly fit the data. We confirmed key inferences about admixture proportions and closest phylogenetic relatives of analyzed groups using qpWave and qpAdm (tables S22 to S25), which does not require making specific assumptions about deep phylogenetic relationships and allows us to test whether there are any groups that harbor genetic drift that is not present in the populations used as proxies for their ancestry (20). Finally, we used admixture linkage disequilibrium to estimate the ages of some detected admixture events with the software DATES (Fig. 4D and table S27).
(i) The Mariana Islands: Distinctive FRO ancestry without Papuan admixture. The Unai individuals from Guam whose radiocarbon dates range from 2800 to 2200 BP derive from the FROMarianas lineage (M1) and have homogeneous ancestry. Later Latte individuals from Guam and Saipan after 700 BP derive ~85% of their ancestry from the same source (fig. S13A), with substantial continuity also confirmed by their harboring the same mitochondrial haplogroups E1 and E2 that are seen in the Unai period. The Latte individuals also derived ~15% ancestry from a previously unidentified FROPalau lineage (M2), which we estimate mixed with FROMarianas 45 to 50 generations before the Latte individuals lived (2400 to 1700 BP, assuming 28 years per generation). The admixture date shows that this migration and mixture process cannot be invoked to explain the origin of the Latte archaeological phenomenon in the Mariana Islands, which began much later at ~1100 BP.
The modern CHamoru from Guam are admixed with European (~19%) and Native American (~9%) ancestry (Fig. 4B), plausibly associated with Spanish colonial activities from the mid-16th century onward (29). Their remaining ancestry is entirely FRO. Although our analyses of modern CHamoru did not allow us to unambiguously determine their FRO source, they show a greater genetic affinity to FROMarianas than to FROSouthwestPacific (Fig. 3B), and their mitochondrial haplogroups E1 and E2 are also found in the Unai and Latte individuals, suggesting that they derived much of their East Asian–associated ancestry from earlier groups in Guam.
(ii) Palau: Mixture of FROPalau and PapuanNewGuinea ancestry. Present-day Palauans are inferred to have ~62% FROPalau ancestry (M2) from the same lineage that admixed in a smaller proportion into the Latte individuals (fig. S13A) and ~38% PapuanNewGuinea ancestry (M4). We estimate the date of FROPalau-PapuanNewGuinea admixture to be ~2500 to 2200 BP, suggesting the possibility of Papuan migration into this region by this time.
(iii) Central Micronesia: Mixture of FROSouthwestPacific and PapuanNewGuinea. We infer genetic homogeneity in central Micronesia over space and time, with Pohnpei and Chuuk having similar proportions of ~73% FROSouthwestPacific (M3) and ~27% PapuanNewGuinea ancestry (M4) and forming a clade with the 11 individuals from prehistoric Pohnpei (Fig. 4B). FROSouthwestPacific is a better single-source proxy for the primary First Remote Oceanian ancestry in Central Micronesia than FROMarianas, but an entirely FROSouthwestPacific source fails in qpAdm when Unai and Latte are included as outgroups, suggesting that both FROSouthwestPacific and FROMarianas contributed. These findings also illuminate the origins of Nuclear Micronesian languages. Central Micronesians lack the Papuan ancestry that is predominant in the Solomon Islands, providing evidence against one of the three main candidate geographic regions (24). They also lack the PapuanNewBritain signature that was ubiquitous in Vanuatu by the time of the peopling of Central Micronesia, providing evidence against another candidate region. Instead, qpAdm shows that the people of Manus are a better proximate source for the PapuanNewGuinea ancestry than those of mainland New Guinea (table S24), increasing the likelihood of the third candidate—the Admiralty Islands—as the source for these languages and for the stream of migration that brought them. This should not be interpreted as implying that people specifically from Manus Island were the true source, but rather that the source was probably a genetically similar population from the Admiralty Islands or a coastal region along the northern fringe of mainland New Guinea.
We infer dates of FROSouthwestPacific-PapuanNewGuinea mixture in Chuuk and Pohnpei of 2100 to 1800 BP, showing that these lineages came into contact at least by the time of the peopling of Central Micronesia around 2000 BP and raising the possibility that the M3 and M4 lineage expansions into Central Micronesia came as part of an already mixed stream of people speaking early Nuclear Micronesian. An alternative, however, would accommodate a different perspective on the origins of Nuclear Micronesian languages, allowing M3 to have come from a FROSouthwestPacific group that spoke a Southeast Solomonic language (30), to be joined later by an M4 Papuan-Admiralties group that did not displace already established Nuclear Micronesian languages. Such a scenario of language continuity despite population replacement would parallel the situation posited for Vanuatu (8, 9). We do not yet have data from Yap but, given that Yapese is an earlier branching Proto-Oceanic language, we hypothesize that the indigenous Yap islanders might derive from a different mixture of source populations than other Central Micronesians.
Matrilocality in early Pacific islanders
We observed a notable degree of mitochondrial DNA differentiation between the FROMarianas and the FROSouthwestPacific lineages. All of the Unai individuals with mitochondrial haplogroup determinations and without evidence of high contamination carried haplogroups E1 and E2 (table S2), whereas all of the Lapita individuals had haplogroup B4 (7–10). All three haplogroups were found in Iron Age Taiwanese (5, 6), consistent with the finding that the Iron Age Taiwanese were relatively undrifted descendants of a population that was also ancestral to the Unai and Lapita individuals. Such a high level of mitochondrial differentiation is surprising given the intermediate degree of autosomal differentiation as measured by FST, a standard statistic measuring population genetic differentiation, which is 0.083 between the Unai and Lapita groups. This raises the possibility of greater genetic drift on the maternal than paternal line during the early divergence and radiation of FRO lineages.
We carried out simulations to determine the probability that completely different mitochondrial macrohaplogroups spread over the two populations since they diverged, under the null assumption that males and females had the same demographic behavior and given the observed genetic drift on the autosomes (fig. S16). This null hypothesis is rejected (P = 0.0014, Fisher’s exact test) (31). The P values are not sensitive to assumptions about the split time of the FROMarianas and the FROSouthwestPacific lineages (table S28). These patterns are qualitatively opposite to those in Neolithic and Bronze Age Europe, where patrilocal patterns of greater female than male mobility among households have been inferred by analyzing ancient DNA data (32, 33). Matrilocality in early Remote Oceanians has been hypothesized based on genetic and ethnographic studies of present-day communities, many of which have matrilocal practices in which women tend to raise their children in the same households in which they grew up (34, 35). Our results provide direct evidence for the practice of matrilocality among FRO populations.
These findings concerning matrilocality among the ancestors of Lapita and Unai individuals with little if any Papuan ancestry are not related to previous evidence of sex-biased admixture between Papuan and FRO ancestry in some Pacific populations (7). However, a new finding of this study does concern sex-biased mixture. Specifically, we find that the Papuan ancestry in Palau and Central Micronesia was primarily derived from male ancestors, based on significantly more Papuan ancestry on the autosomes than on the X chromosome (|Z|> 2.2 to 3.3) (Fig. 4E and table S29) (7). This is notable because each of the three cases of FRO-Papuan admixture that are now documented (Palau, Central Micronesia, and southwest Pacific and Polynesia) involved a different pair of Papuan and FRO groups. These events must have been independent, and yet all share the feature of Papuan ancestry being transmitted primarily by male ancestors.
Family structure and population size during the Latte period
We measured runs of homozygosity (ROH) that were longer than 4 centimorgans (cM) for 113 Latte individuals with high-enough-quality data to allow such analyses (table S30). Only two had single stretches of ROH longer than 50 cM, indicating that close-kin unions were avoided in Latte people. Nine individuals from Guam and nine from Saipan had at least one ROH longer than 20 cM, suggesting that mating pairs of close relatives such as second or third cousins on both islands were relatively common. Shorter ROH signals (>4 cM) were also abundant, implying a limited pool of reproductive partners in every generation. We estimated the size of the population from which the Latte individuals in Guam and Saipan were drawing their reproductive partners to be 315 to 356 individuals in Guam and 361 to 424 individuals in Saipan (table S32).
We further analyzed long shared DNA segments [identical by descent (IBD) blocks] between the X chromosomes of male individuals (one from Guam and the other from Saipan). We identified 149 pairs of individuals who shared IBD segments longer than 8 cM (table S31). This puts an upper bound on Ne, the size of the mating population in the combined Mariana Islands, of 1203 to 1712 (95% confidence interval) (table S32). If there were restricted migration between islands, or if there were temporal variation in the dates of the individuals we compared, these numbers would be overestimates. This implies a long-term small population size or strong founder event in Latte history.
We identified 122 pairs of closely related Latte individuals (up to third-degree relatives) (fig. S17 and table S33). Eighty of 125 Latte individuals that were studied had one or several close relatives.