Generation times in wild chimpanzees and gorillas suggest earlier divergence times in great ape and human evolution

By using direct observations of generation times in gorillas and chimpanzees and rates of mutation per generation from direct observation of mutations in human families, we estimate the species split times of humans and apes without relying on external fossil calibration points. At 7–13 Ma our estimate of the chimpanzee–human split time is earlier than those previously derived from molecular dating using fossil calibration points but similar to the range of 6.5–10 Ma suggested by the fossil record (30).

Whereas the earliest fossil universally accepted to belong to the lineage leading to present-day humans rather than to chimpanzees, Australopithecus anamensis, is 4.2 Ma (31) and thus reconcilable with a molecularly inferred human–chimpanzee split time as recent as 5 Ma, the attribution of late Miocene (5–7 Ma) fossils to the hominin lineage has posed a problem. Our estimates make it possible to reconcile attribution of fossils such as Ardipithecus kaddaba (5.2–5.8 Ma) (32), Orrorin tugenensis (6 Ma) (9), and Sahelanthropus tchadensis (6–7 Ma) (10) to the hominin lineage with speciation times inferred from genetic evidence (Fig. 1). However, our estimates cannot address the controversy of whether specimens such as these truly belong to the lineage leading to present-day humans or to other, closely related lineages (11).

For the deeper time period of 7–13 Ma, the fossil record is even more limited and difficult to interpret (31, 33). Fossils from between 8 and 11 Ma in Africa include mainly Gorilla-sized forms, such as Samburupithecus (34), Nakalipithecus (35), and Chororapithecus, the last of which is dated to 10–10.5 Ma and suggested to represent an early member of the gorilla clade (36). Our estimate of 8–19 Ma for the split of the gorilla lineage from the human–chimpanzee ancestor would be largely consistent with the attribution of such forms to the gorilla lineage.

Even though not quantified here, our results also significantly push back the date of the split between orangutans and African apes. Paleontological data (e.g., ref. 37) have been combined with genetic data (38) to suggest that this split occurred outside of Africa, with a later “Back to Africa” migration of the common ancestor of African apes. The purported “early great ape” Pierolapithecus catalaunicus from Spain, dating to about 12.5–13 Ma (39), and the presence of numerous derived African ape traits in Late Middle Miocene fossils from Europe such as Rudapithecus and Hispanopithecus fit well with this hypothesis. A split between African apes and orangutans that predates 15 Ma would challenge this model and would either put these fossils on the orangutan lineage or place them as unrelated to present-day great apes.

For more recent periods of hominin evolution, the more recent dates provided here for the human–chimpanzee split resolve an apparent contradiction between genetic and paleontological data. Using a chimpanzee/human split of 5.6–8.3 Ma for calibration, analyses of the Neanderthal genome indicated a population split between present-day humans and Neanderthals at 270–440 ka (40). This date appears to conflict with fossil evidence tracing the emergence of Neanderthal morphological characters over the course of the Middle Pleistocene in Europe (41). The earliest evidence for Neanderthal traits was proposed to date to 600 +∞/−66 ka at the Sima de Los Huesos (Atapuerca, Spain) (42), thus predating the genetically estimated population divergence times, but this date has been disputed on the basis of both the apparent conflict with the genetic data and on stratigraphic grounds (43). However, even if the early dates for Sima are disregarded, it is clear that fossils from oxygen isotope stage 11 (around 400 ka), such as the Swanscombe cranium, already show clear Neanderthal traits (44). Using the new human–chimpanzee split estimate and assuming generation times between 25 and 29 y would push back the human/Neanderthal split to 423,000–781,000 y, resolving this apparent conflict.

Recent attempts to model uncertainty in the fossil data used for molecular calibration also suggest earlier split times in the evolutionary history of apes with estimates of 6–10 Ma for the human–chimpanzee divergence and 7–12 Ma for the divergence of the gorilla (18). Our estimates of divergence dates have the advantage that they avoid fossil calibration points. However, it is possible that other aspects of our analysis may lead to unreliable split time inferences. First, because of the limited availability of data from the western gorilla species, we make the assumption that the average generation interval of mountain gorillas is applicable to both present-day species of the Gorilla genus. Although reliant primarily upon herbaceous vegetation, western gorillas also eat fruit much of the year, whereas fruit is nearly absent from the mountain gorilla habitat (45). More folivorous anthropoid primates are known to mature more quickly than similarly sized nonfolivorous primates (46), and indeed limited data from western gorillas suggest that females and males attain adulthood 2 and 3 y later, respectively, than the more folivorous mountain gorilla (47). This implies that the generation time in western gorillas may be on the order of 21 y, in contrast to the 19 y used here for gorilla generation time. However, because 19 is the shortest generation time observed among present-day mountain gorillas, chimpanzees, and humans, our use of this value is more conservative and simply contributes to a slightly broader range for the inferred split time for the divergence of the gorilla lineage from that leading to humans and chimpanzees, as well as to a broader range for the split time between the two gorilla species. As with western gorillas, parentage data for calculation of generation times in bonobos are lacking. However, neither extensive dietary differences between bonobos and chimpanzees nor substantial differences in developmental timing are apparent for these species and it is also relevant that we found no consistent differences in generation times between chimpanzees from western and eastern Africa. With regard to humans, highly similar estimates of generation time were obtained from demographic analysis of a large sample of less- and more-developed countries, a large sample of hunter–gatherer societies, and direct analysis of genealogies (22). In sum, except for the gorillas where marked ecological differences may contribute to a small degree of variation in generation time within the genus, the generation times used here seem reliable estimates for present-day great apes and humans.

A further notable assumption of our work is that the generation times calculated for present-day humans and great apes are valid proxies for their ancestors. It was recently suggested that a slowdown in mutation rate concomitant with an increase in body sizes and generation times has occurred in these lineages (8). However, there is an extraordinary diversity of ape body sizes in the fossil record since the Miocene (24 to 5 Ma) and it is difficult to know which ones may represent ancestors of present-day apes and humans (32). Even if fossil evidence strongly suggested an increase in the size of the ancestors of present-day apes and humans in the past, it is not clear that body mass is a good correlate of life history parameters related to generation time (48). Although our number of data points is necessarily limited, we found no correlation between mass and generation time in present-day apes and humans, and the notably short generation time for the relatively large mountain gorilla is consistent with the expectation that highly folivorous (46) as well as more terrestrial (49) species are expected to reproduce earlier than more frugivorous, arboreal primates. In accordance with the importance of diet and habitat use in influencing life-history parameters, it has been suggested that chimpanzees and orangutans represent the most appropriate living models for the potential life history variables of archaic hominins, and that the common ancestor of humans and chimpanzees exhibited a slow life history similar to that of present-day chimpanzees (50). Skeletal and dental analyses suggest that early hominins had growth patterns like those of present-day great apes, whereas Homo erectus and Neanderthals evolved slower development, but not to the extent seen in present-day humans (51, 52). Given the information available at this time, we suggest that the use of the ranges of the observed generation times in the present-day species, including the extremes represented by gorillas (with their comparatively fast life history and consequently short generation time) and humans (with their comparatively slow life history and consequently long generation time), results in conservatively broad estimates of hominid mutation rates and split times as shown (Table 2 and Fig. 1). Specifically, if we alternatively consider the human generation time of 29 y to be a recent phenomenon, and consider the chimpanzee generation time of 25 y to characterize the vast majority of evolution since the split between the gorilla and the chimpanzee/human lineages, we would infer the date of this split at 10.9–17.2 Ma, whereas the split between the lineages leading to chimpanzees and humans would be dated at 6.8–11.6 Ma.

We also note that we explicitly assume that the mutation rates estimated by sequencing members of present-day human families are also applicable to our closest great ape relatives. This assumption, which is based on our close evolutionary relationship and lack of evidence for differences in rates of evolution among the human and African great ape lineages (7, 53), can be explicitly tested in the future by sequencing of great ape family trios. As an additional point for future consideration, we note that the original publications, which provide the population split times that we recalibrate here, use various approaches for filtering the data analyzed, for example, exclusion of repetitive sequences or highly mutable sites. Refinements of our population split time estimates may involve reexamination of the data, including consideration of different parts of the genome, or different types of substitutions. For example, it will be interesting to compare inferences from substitutions at CpG sites that may accumulate in a time-dependent fashion with other classes of substitutions that may accumulate in a generation-dependent fashion. However, thus far studies have shown that the inclusion or exclusion of CpG sites has little impact on the timing of the human-chimpanzee split (3, 7).

Finally, we note that the estimation of generation times in chimpanzees and gorillas derives from the long-term efforts of researchers who have invested years in habituating the animals to human observation to collect information on their natural behavior and life histories. This study illustrates the value of such approaches in aiding interpretation of genomic data and suggests that continued behavioral study of wild apes, in addition to increased understanding of their behavior and cultures, is necessary to complement genomic studies for a fuller understanding of the evolutionary history of our closest living relatives as well as our own species.

Details regarding the analyses can be found in SI Materials and Methods. In brief, we compiled the ages of the genetically confirmed mothers and fathers of offspring born into eight chimpanzee groups and six mountain gorilla groups habituated to human observation. We did not limit our sample to individuals whose ages are exactly known because this would lead to a downward bias in the estimation of the generation length, as older individuals are more likely to have been born before the start of long-term research on a particular group. Instead, we included in our study individuals whose ages were estimated using standard morphological, behavioral, and life history criteria established from known-aged individuals and systematically incorporated estimation of ranges of minimum and maximum birthdates symmetrical about the assigned birthdate.

For the split time estimation, we first took the lowest and highest estimates of mutation rates in human families of 0.97 × 10⁻⁸ to 1.36 × 10⁻⁸/site/generation and applied the estimated generation times of 19, 25, and 29 y for gorillas, chimpanzees, and humans to arrive at low and high estimates of yearly mutation rates given each of these generation times. For example, the chimpanzee generation time of 25 y yields a rate of 0.39–0.54 × 10⁻⁹ mutations/site/year, whereas the human generation time of 29 y yields a rate of 0.33–0.46 × 10⁻⁹ mutations/site/year. For each split we then chose lower and upper bounds for the yearly mutation rates based upon the extreme values inferred for the taxa under consideration. For example, we assumed that the generation time of the common ancestor of chimpanzees and humans was between 25 and 29 y, the values for present-day chimpanzees and humans, respectively, and thus used the mutation rates of 0.33 and 0.54 mutations/site/year (Table 2). Similarly, the common ancestor of gorillas, chimpanzees, and humans is assumed to have a generation time between 19 and 29 y and we thus used a correspondingly broader set of mutation rates. We adjusted previously published split times (Table 2) by multiplying with the factor μ_old/μ_new, where μ_old corresponds to the previously used mutation rate per year and μ_new to our upper and lower bounds based on the range of per generation mutation rates and generation intervals appropriate for the split under consideration.

No explicit mutation rate was assumed for the calculation of the split times of Neanderthals and present-day humans in the original publication (41). However, the authors use a range of nuclear divergence times for orangutan–human to arrive at a human–chimpanzee divergence time of 5.6–8.3 million years. To recalibrate the Neanderthal split time, we use the published nuclear divergence of ca. 1.3% between human and chimpanzee (8, 16) to convert these values to a mutation rate per year (corresponding to 1.1–0.7 × 10⁻⁹).

We thank A. Abraham for laboratory assistance, R. Mundry for assistance with estimation of uncertainties in average parental ages, the three reviewers, W.-H. Li for helpful comments on the manuscript, and the many agencies and governments that support the field research (see SI ACKNOWLEDGMENTS). This project was funded by the Max Planck Society. K.E.L. was supported by a fellowship from the Alexander von Humboldt Foundation.