Sexual dimorphism in Australopithecus afarensis was similar to that of modern humans

These issues are potentially resolvable for A. afarensis, because its fossil record includes a large geologically simultaneous death assemblage from a single stratum in a single locality (A.L. 333) (18). Therefore, no a priori assumptions regarding the potential impact of such factors are required, and this site represents a unique opportunity to explore them in an early hominid species. The various postcranial fragments recovered at this occurrence, however, represent many different anatomical elements. Fortunately, an adult partial skeleton (A.L. 288-1, “Lucy”) is also available for this species. We therefore used ratios between a single skeletal dimension [femoral head diameter (FHD)] and other skeletal metrics preserved in A.L. 288-1 to predict FHDs for 22 of the specimens recovered at A.L. 333 (see accuracy assessment below) and seven additional specimens from other Hadar localities and Maka [Combined Afar (C.A.)] (Table 1). A.L. 288-1 also preserves a mandible, but we have not used it here for two reasons: (i) only three measurable adult specimens were available from A.L. 333, and (ii) postcrania are superior correlates of body size; indeed, the accuracy of judging body size dimorphism from mandibular data appears dubious (see below). FHD was selected because of its common use in size estimation. Our results would have been identical had we standardized to any other skeletal dimension, because all ratios and measures of dispersion [i.e., Binomial Dimorphism Index (BDI), CV, and MMR] would remain the same.

We calculated three measures of skeletal dimorphism for the A.L. 333 and C.A. samples: a MMR, CV, and BDI; formerly Technique Dimorphism (19). The last of these requires three assumptions: (i) both sexes are present in the sample; and (ii) any specimen has an equal prior probability of being male or female, but (iii) when two specimens are of different sex, the larger is male. To apply this simple algorithm, all specimens are first arrayed according to increasing size. There are then n – 1 possible sex allocations (from one female/all-others-male to one male/all-others-female), and n – 1 ratios for which a sexual dimorphism estimate can be calculated (mean of presumed males/mean of presumed females). BDI is the weighted mean of these n – 1 estimates using each ratio's probability in the binomial expansion. The other indices (CV and MMR) were calculated as described (12, 13). We used the small sample correction for the CV (20). To judge the accuracy of these three methods (BDI, CV, and MMR), we applied them to identical anatomical arrays of skeletal metrics randomly selected from extant Homo, Pan, and Gorilla samples (Table 2) (i.e., the exact same metrics calculable for A.L. 333 and the expanded C.A. sample). We did so 1,000 times for each taxon. In each iteration, a template specimen (acting in the role of A.L. 288-1) was randomly selected to provide ratios for estimating FHDs. Because the sizes of the hominoid reference samples are finite (n ≈ 50, Table 2), any single individual in the sample was selected multiple times to serve as the template. However, each randomly sampled anatomical array (n = 22 or 29 representing 12 different metrics) was almost certainly unique for each of the 1,000 iterations.

In addition to the three sexual dimorphism indices (see above), we also calculated two measures of actual skeletal dimorphism (i.e., male mean/female mean based on known sex) for each simulation: the first using FHD estimated by the template method for each specimen [template sexual dimorphism (TSD)], and the second using the actual FHDs for each individual [we will refer to the latter as direct sexual dimorphism (DSD)]. A comparison of these two methods permits an assessment of the potential error that arises from using a template specimen, i.e., comparison of TSD and DSD allows direct assessment of the estimated (template) vs. real (actual FHD) dimorphism in each hominoid simulation. Because calculation of sexual dimorphism for any of the hominid samples (i.e., BDI, CV, MMR) requires the use of the template specimen (A.L. 288-1), they can be assessed only by comparison to TSD from the simulations.

For the A.L. 333 sample, we faced the taphonomic problem that <22 adults may be represented. Relying on adult mandibular dentitions, the adult minimum number of individuals at A.L. 333 was nine (16). We therefore conducted additional simulations in which the number of hominoid individuals serving as a source of the 22 metrics in Table 1 was varied from 22 to 9. As shown in Fig. 2A, so restricting the number of adults did not significantly affect the outcome of the simulations.

As expected, our simulations (see Fig. 2 and Table 3) show body size dimorphism in Homo to be intermediate between nondimorphic Pan and highly dimorphic Gorilla. For the most part, the template method tends to overestimate both means and dispersions of actual dimorphism values (compare TSD with DSD in Table 3). The CV does not produce a direct estimate of dimorphism, but its behavior closely mimicked that of BDI, which does. MMRs consistently failed to substantially distinguish Pan from Gorilla and greatly overestimated dimorphism in all three taxa (Fig. 3 and Table 3). Their very poor performance is likely the result of instabilities in the method itself but was amplified by use of a template (as noted earlier, use of ratios increased the dispersion of estimates). However, even when no template was used and only actual femoral head values were used, the distributions of MMRs still grossly overestimated dimorphism and failed to substantially discriminate Pan from Gorilla. These methodological problems may not have been apparent until now, because MMRs have typically been applied to only very small samples. In contrast to MMR, both BDI and CV adequately quantified sample dimorphism, as evidenced by correlations with both TSD and direct sexual dimorphism (see the supporting information, which is published on the PNAS web site, www.pnas.org). Both performed almost identically in all simulations.

BDI and CV calculated for A. afarensis, whether from A.L. 333 or the entire C.A. sample, were most compatible with the Homo simulations. Regarding Pan, only the C.A. BDI and CV were significantly larger than the simulated distribution. However, both hominid samples differed significantly from Gorilla (in fact, the A.L. 333 BDI fell entirely outside the range of simulated values) (Table 4). Of crucial importance is the fact the C.A. sample yielded significantly higher dimorphism (e.g., BDI = 1.222) than did A.L. 333 (BDI = 1.167) (Fig. 2B), implying that the C.A. sample reflects not only sexual dimorphism but ecogeographic and temporal factors as well.

These results conform to a recent study that examined mandibular size in A. afarensis for changes over time (14). Its authors had previously concluded that A. afarensis might have been as dimorphic as Gorilla and Pongo (13). However, their postcranial samples were too small (femur = 5 and humerus = 3) to calculate CV, and they instead used MMRs. Since MMR is extremely sensitive to geographic and temporal variation, only by including data from all sampled individuals in the calculation of the index (e.g., CV or BDI) can we more reliably estimate the variation that is accountable by sexual dimorphism. Their larger sample for the mandibular corpus [n = 17 (geometric mean of mandible height and thickness at M₁)] had yielded a CV of 11.7 for A. afarensis [compared with 12.3 for (very dimorphic) gorillas and 7.78 for (nondimorphic) chimpanzees]. However, subsequent amplification of that sample (to n = 20) and removal of the four most recent specimens (still maintaining a temporal range of ≈320,000 years) reduced the A. afarensis CV to 8.49 (14), well below those for comparable measures in isolated contemporary modern human populations such as Zulu (10.2) and Spitalfields (10.1) (21). Moreover, as noted earlier, basing an estimate of body size dimorphism on the mandible may be unwarranted. The same metric in a second, independent, sample (21) yielded almost identical CVs for humans (10.4) and gorillas (10.5). If only corpus height was used (which presumably more directly reflects canine dimorphism), mandibles performed better (13.9 for gorilla vs. 11.6 for humans), but in chimpanzees, the female mean was larger than that of the males.§ In any case, if the four most recent A. afarensis specimens are once again excluded from the calculations (see above), the CV of mandibular corpus height (10.2) (14) still falls below that of modern humans.

We addressed one additional potential taphonomic problem. The smallest estimated FHD for A.L. 333 was 29.8 (Table 1). Although this value is just over 1 mm greater than “Lucy” (28.6), it could hypothetically imply underrepresentation of “Lucy”-sized females at the site (although an abundance of small juvenile specimens makes systematic preservation/recovery bias at A.L. 333 very unlikely). Nevertheless, we sequentially incremented our A.L. 333 sample with “Lucy”-sized specimens (i.e., FHDs of 28.6) until its BDI equaled that of gorillas. This required the addition of eight “Lucy”-sized metrics to the previous total of 22 (i.e., a new n = 30 total elements). However, the addition of these hypothetical elements also increases the minimum number of individuals at A.L. 333 to a total of between 10 (the original nine plus one represented by eight new elements) and 17 (the original nine plus eight new individuals, each represented by a single element) and increases the metric sample by more than one-third. This is obviously an artificial and nonrepresentative inflation of the actual A.L. 333 sample. Moreover, there is no a priori reason to presume that diminutive specimens like A.L. 288-1 represent the average size of A. afarensis females. Indeed, if they did, most intermediate specimens would then be male, and male A. afarensis body size would then be far more variable than in any other living hominoid. Similarly, any contention that A.L. 333 represents a single polygynous male accompanied by multiple mates and juveniles would require uniquely extreme size variation in females. Moreover, if sex ratios were like those of most primate groups (varying anywhere from an approximately equal sex ratio to only a single male), the dimorphism of A.L. 333 must have been within or below the modern human range (Fig. 4).

The evidence that A. afarensis is characterized by only slight to moderate degrees of skeletal dimorphism resolves the paradox articulated by Plavcan and van Shaik (4), who noted that the largely monomorphic canines of A. afarensis make its supposed great body mass dimorphism (relying again on ref. 6) difficult to interpret (see also ref. 24). They also observe that this is not necessarily an enigma, because body mass dimorphism often does not reflect male–male competition but can arise from other factors such as substrate preferences, predator avoidance, and phylogenetic inertia. Nevertheless, the issue is largely resolved by the results of the present study and especially by the distribution of the skeletal dimensions presented in Fig. 2 for both A.L. 333 and the entire A. afarensis sample (see also Fig. 1). These distributions are wholly inconsistent with marked bimaturism, as in gorillas and orangutans [i.e., the telic heteromorphosis of Jarman (25)], to which A. afarensis has been consistently compared in the past. They are instead comparable with the homeomorphic dimorphism that characterizes humans and chimpanzees.

Derived hominids possess a number of unique/unusual characters, including concealed ovulation (from both sexes) (26), elaborate epigamics in both sexes, relatively small testes (compared with body mass), relatively short sperm (26), permanently enlarged mammae, and a dramatically expanded cerebral capacity unparalleled in other mammals (27). It has been proposed that all but the last of these characters spring from a social complex including male provisioning driven by female choice that enabled Australopithecus to counteract restrictions of reproductive rate imposed by excessive K-selection (28). Although tool using scenarios have been argued to account for canine reduction, such “disuse” models fail to adequately explain directional selection for crown reduction in males and the myriad other unique anatomical characters of derived hominids. One obvious possibility is that small-canined males would be less effective competitors in a polygynous mating strategy and would thereby prove to be more reliable provisioners (and thereby differentially chosen by females). However, female choice does not provide an adequate explanation for the patterns of skeletal dimorphism observed across hominoids. Most notably because chimpanzees exhibit virtually no significant skeletal size dimorphism (body mass dimorphism is moderate), despite their demonstrably polygynous reproductive strategy. This near monomorphism is in stark contrast to the marked skeletal dimorphism of gorillas and orangutans and suggests that skeletal dimorphism in itself is a poor predictor of reproductive strategy in hominoids.

In polygynous species in which males must rely on substantial body mass and weaponry (especially large canines) to compete for access to mates, it is typical for male maturation to be delayed and growth thereby to be prolonged to accentuate these characteristics (telic heteromorphism) (25, 29, 30). Because hominoid male canine size is largely a product of prolonged growth (31), a greater time to maturity in gorillas and orangutans (29, 30, 32) is consistent with the greater individuation of male success in these species, as is reflected in their more marked canine, skeletal, and body mass dimorphism.

Male kin-related communities of P. troglodytes rely heavily on territorial maintenance by “patrols.” Selection may have therefore favored more rapid male skeletal maturation to accelerate their participation in cooperative territorial defense (this would also account for the species' reduced canine dimorphism compared with that in gorillas and orangutans) (33). Pan paniscus is slightly bimaturational by virtue of a later cessation of growth in males. Leigh and Shea (30) attribute this greater bimaturation to reduced feeding competition among females in P. paniscus, but an equally likely explanation is that bonobos instead represent the primitive condition, and female P. troglodytes have slightly delayed sexual maturation (thereby eliminating bimaturation in the species) because of their more intense female–female competition.

In any case, the minimal expression of bimaturation in both species increases the likelihood that australopithecine groups were not polygynous if they exhibited a female transfer system. As Hamada and Udono (33) have argued, “the social system and ecology of human ancestors, who evolved a characteristic growth pattern, must have been different from that of chimpanzees” (ref. 33, p. 283). First, their marked demographic success and capacity to invade new potentially dangerous habitats strongly suggest that they, like chimpanzees, dwelled in multimale groups (27, 28). If such groups were also (like chimpanzees) involved in significant territorial defense, largely by kin-groups, a similarly weak degree of chimpanzee-like skeletal dimorphism would be expected, as would greater canine dimorphism.

Instead, the moderate skeletal dimorphism of A. afarensis (greater than Pan and less than Gorilla) suggests a somewhat longer developmental period in males compared with females and is therefore inconsistent with a chimpanzee-like territorial strategy. At the same time, it is also markedly inconsistent with strategies like those of gorillas and orangutans, in which skeletal dimorphism is much more pronounced. Therefore, the cooccurrence of moderate skeletal dimorphism, such as that found in modern humans and A. afarensis, and a reduced male canine is fully consistent with a pair-bonded reproductive strategy in early hominids; that is, if their reproductive strategy was chimpanzee-like, hominids should show only minimal skeletal dimorphism, or if it was orangutan- or gorilla-like, they should show greater skeletal dimorphism. Canine dimorphism should be present in either case. Early hominid skeletal dimorphism is consistent with another special hominid character, the failure of male canine eruption to be delayed and thereby coincident with somatic maturation (as it is in all other hominoid species) (34, 35). Thus, observed levels of body size dimorphism in A. afarensis do not imply that monogamy is any less probable than polygyny as the fundamental social system of these early hominids.

Abbreviations: A.L., Afar Locality; BDI, Binomial Dimorphism Index; MMR, maximum/minimum ratio; TSD, template sexual dimorphism; CV, coefficient of variation; FHD, femoral head diameter; C.A., Combined Afar.

See commentary on page 9103.

We thank Gen Suwa, Richard J. Smith, Robert Eckhardt, and Tim White for detailed and critical reviews of this manuscript. Louise Humphrey, Chris Dean, and Chris Stringer kindly provided unpublished data, and Brian Richmond and Sang-Hee Lee brought key references to our attention. Kevin Kern contributed to data collection and computer programming. This research was supported by National Science Foundation Grants SBR-9729060 and BCS-9919211 (to C.O.L.).