Variables, Specimens, and Body Mass Data

We assembled a suite of 80 morphometric variables from the literature (Gingerich and Schoeninger, 1979; Radinsky, 1981a, 1981b, 1982; Gingerich et al., 1982; Anthony and Kay, 1993; Hartwig, 1993; Kobayashi, 1995). Appendix 1 lists measurement variables and their descriptions. Of these potential body mass proxies, 32 were measurements of osteological features on the cranium and mandible, while 48 were measurements of the dentition. All teeth except the third molars (which do not occur in callitrichines) were measured. Specimens of extant taxa examined were from the collections of the Mammalogy Division, Zoology Department of the Field Museum of Natural History, and the Mammalogy Collections, Division of Vertebrate Zoology of the American Museum of Natural History. In total, 157 specimens representing 17 platyrrhine species were measured. Specimen numbers are given in appendix 2. Extant taxa were chosen to ensure a sample containing at least one species from each of the 15 platyrrhine genera and the full range of body sizes (from 13.5 kg for Brachyteles arachnoides to 125 grams for Callithrix pygmaea). Body mass data were obtained from the Masses of Mammals Database (v. 3.03: Smith et al., 2003). Average body masses used in the regression analysis are listed by species in table 1. In addition, we measured preserved morphometric variables for the fossil taxon Chilecebus carrascoensis (SGOPV 3213, Museo Nacional de Historia Nacional, Santiago, Chile). Mean measurement values for extant taxa are given by species in appendix 3.

Table 1

Body mass for measured extant platyrrhine species Masses given in grams, from Smith et al. 2003.

Allometric Regression Analyses

Morphometric variables and body masses for the living platyrrhines were log-transformed (base 10) to determine allometric relationships between the variables and body mass. Species average masses were fit to species-average log-variables using ordinary (Model 1) least-squares regression (Sokal and Rohlf, 1995). Because similarity in morphometry may arise from close phylogenetic relationship (Felsenstein, 1985), we performed phylogenetically corrected regressions (Garland and Ives, 2000), which were executed in the PDAP (Midford et al., 2003) module for the Mesquite software package (Maddison et al., 2002, 2004). Phylogenetic relationships among extant taxa were obtained from molecular phylogenies of the Platyrrhini (Canavez et al., 1999; von Dornum and Ruvolo, 1999) (figure 1). Slopes and intercepts describing the relationship between the morphologic predictor and body mass were calculated for each variable.

Fig. 1

Cladogram from molecular phylogenies (Canavez et al., 1999; von Dornum and Ruvolo, 1999) of platyrrhine primates used in the phylogenetically corrected regressions of body mass on morphometric variables. This cladogram is a synthetic topology of these two phylogenetic analyses. The topologies for the two analyses were congruent for overlapping taxa, with two exceptions. First the Callicebus/Cacjao/Chiropotes/Pitehca clade was basal to all other Platyrrhini in von Dornum and Ruvolo (1999), but was allied with the Ateles/Bachyteles/Lagothrix/Allouata clade in Canavez et al. (1999). Second, Aotus was allied with Saimiri and Cebus in Canvez et al. (1999), but left in an unresolved polytomy in von Dornum and Ruvolo (1999). Therefore both nodes are conservatively left in polytomies here.

To assess a variable's efficacy in predicting body mass, we quantified the relative performance among estimators using the log-likelihood fit of each regression model to the data:

Body Mass Estimation in Chilecebus

The body mass of the fossil platyrrhine Chilecebus carrascoensis was estimated from 46 measurements preserved in SGOPV 3213, the holotype cranium, using the regression equations generated above from living taxa. We calculated a weighted average of these estimates using the proportional likelihoods for each regression model over the set of all models (Burnham and Anderson, 2002). Model averaging is appropriate when there is no clearly identifiable “correct” model. This method has been employed to a limited extent in previous work (e.g., Radinsky, 1971; Martin, 1990), where the mass estimates from several proxies were averaged to reduce errors associated with any single predictor. However, simply calculating the mean of the individual variable estimates implicitly assumes that all of the variables perform equally well in predicting body mass. Since regression model likelihoods quantify an individual variable's relative predictive ability, a more appropriate averaging technique takes advantage of this information. Using proportional likelihoods, poorly performing models influence the final body mass estimate less, whereas better fitting models more strongly influence the final estimate.

Estimating Relative Brain Size in Chilecebus

X-ray computed tomographic scans of the Chilecebus carrascoensis cranium were performed on the medical scanner at Children's Hospital of San Diego (CT HiSpeed Adv SYS#HSA1; 140 Kv, 170 mA) to estimate endocranial volume (Flynn et al., 1995). The skull and surrounding matrix were embedded in modeling clay to better mimic the density of the human body, and scanned at 1.0 mm resolution. The series of individual axial and transverse scan slices were composited into 3-D images using Cemax VIP software© (Freemont, CA). The resulting brain volume measure was used in conjunction with body mass estimates to determine the relative brain volume in Chilecebus.

To determine whether increased encephalization observed in extant anthropoids was inherited from their last common ancestor, or whether New and Old World anthropoid primates evolved this trait independently, log-encephalization quotients (logEQs) were calculated for the extant catarrhine and platyrrhine taxa considered in Martin (1990). The encephalization quotient (EQ) is the ratio of observed brain volume to expected brain volume for a given body mass (Jerison, 1970; Radinsky, 1971). The metric of interest, however, is not relative volume but deviation of observed volume from the allometric regression of brain volume on body mass, i.e., the natural logarithm of the EQ (logEQ) (Marino et al., 2004; Finarelli and Flynn, 2007; Finarelli, 2008). Positive logEQs imply larger than expected brain volumes for a given body mass, and negative values the opposite. To assess the relative brain volume for extant platyrrhines and the extinct taxon Chilecebus, we used the allometry relating body mass to brain volume for strepsirrhine primates (as an outgroup to anthropoids) given in Martin (1990) as a frame of reference.

Body Mass Estimators

We performed phylogenetically corrected regression of body mass on each of the 80 morphometric variables for the extant taxa. Slopes and intercepts for the predictive equations and the corresponding log-likelihoods for the phylogenetically corrected regressions are given for each variable in table 2; scatterplots for the regressions are given in appendix 4. The relative ability of the morphometric variables surveyed in this analysis to accurately predict body mass, as measured by the log-likelihood fits for the regressions, is highly uneven. The maximum likelihood observed among the variables is for mandibular length, indicating that this variable is the single best predictor of body mass among the 80 variables surveyed. Only one variable (pterygoid-zygomatic length) had a log-likelihood fit within 2LnL units of mandibular length, and therefore, there is no significant difference in the ability of those two variables to predict body mass among extant platyrrhine species.

Table 2

Regressions of body mass on morphometric variables for extant taxa Slopes and intercepts of predictive equations for phylogenetically corrected regressions for each measurement variable (appendix 1) in living platyrrhines sampled (table 1). Also given are R², error sum of squares (ESS) and the log-likelihood (LnL), rescaled such that the maximum likelihood observed among models (mandibular length) is 0. Measurement regressions are ordered in decreasing LnL. Teeth are upper (U) or lower (L) lengths, widths (anterior or posterior for molars), and areas.

The two optimal predictors are both craniomandibular, osteological measurements, not dental variables. Teeth are frequently employed as body mass predictors for fossil mammals (Gingerich, 1974; Legendre, 1986; Legendre and Roth, 1988) because of their higher preservation potential. However, dental measurements are generally considered to be less reliable estimators than osteological variables (including measurements of the postcranium) (Gingerich et al., 1982; Dagosto and Terranova, 1992), potentially reflecting specialized adaptations to unusual diets/food items or unusual food-processing methods (Gingerich and Smith, 1984). To test this assumption, we partitioned the data into osteological (both cranium and mandible), and dental (both upper and lower) splits.

Eight of the 10 best predictors in this analysis are osteological (table 2): mandibular length, pterygoid-zygomatic length, bizygomatic width, moment arm of the masseter, skull length, temporal fossa length, moment arm of the temporalis, and mandibular height. Among these, both mandibular length and pterygoid-zygomatic length exceed the best dental variable by at least 2 LnL units. While osteological features include the best predictors of body mass, they also include the three worst predictors: vault height, orbital width, and pterygoid length. Further, a Mann-Whitney test (Sokal and Rohlf, 1995) recovers no significant difference in the median rank log-likelihood scores across osteological and dental partitions (U  =  656, p > 0.05). Thus, while median log-likelihood for the osteological variables is slightly higher than that of dental variables (−6.45 versus −7.60), the variance of log-likelihoods for osteological variables also is higher. Thus, while the best predictors are osteological, among platyrrhine primates osteological characters do not universally outperform dental ones.

Among dental variables, the best predictor of body mass is lower M1 length (table 2), a commonly used proxy for body mass in mammals (Van Valkenburgh, 1990; Janis et al., 1998a, 1998b). Lower M1 area, another commonly used proxy (Gingerich, 1974; Legendre, 1986; Legendre and Roth, 1988; Finarelli and Flynn, 2006), was the next best estimator and was not significantly different than lower M1 length (LnL difference  =  0.298). As noted above, the distribution of log-likelihoods for dental variables was less dispersed than the distribution for osteological variables. Eight dental measures (as opposed to two osteological) were within 2 LnL units of the optimal dental score. However, among these, only upper cheek tooth row length was from the upper dentition. A Wilcoxon signed-rank test performed on 24 upper/lower pairs of dental measurements demonstrates that variables from the lower dentition significantly outperform their upper dentition counterparts as predictors of body mass (W  =  160, p < 0.05).

Thus, regression model likelihoods demonstrate: 1) that the single best predictor is mandibular length, and 2) that the lower dentition significantly outperforms the upper dentition in predicting body mass. Thus, it appears that predictors associated with the mandible may outperform those of the cranium among extant platyrrhines. To test this, log-likelihoods were separated into mandibular and cranial partitions, rather than osteological (including the mandible) and dental (including both upper and lower) partitions. Median mandibular log-likelihoods were higher than cranial log-likelihoods (−6.12 versus −7.81), a significant difference (Mann-Whitney test, U  =  519, p < 0.05).

Intraspecific variability must be a concern when evaluating evolutionary allometries through species mean values. Platyrrhines, and primates in general, are sexually dimorphic (Martin et al., 1994), a feature that leads to a range of variance that is discounted when averaging over species. When estimating body mass for fossil taxa allometries must use species averages. Sex identification for fossil mammals generally is not possible, except in rare instances (e.g., Coombs, 1975; Fleagle et al., 1980; Krishtalka et al., 1990; Van Valkenburgh and Sacco, 2002). This is due to the fact that fossil taxa are often known from a limited number of specimens and rarely represent a population sample for a species. It is therefore difficult to rigorously evaluate sexual dimorphism or other forms of intraspecific variation. However, large degrees of within-species variation can negatively impact the precision of prediction allometries.

More troublesome would be if there were a tendency for the range of variation to increase or decrease as a function of body mass (e.g., Rensch's rule: Rensch, 1950). To assess this potential impact of intraspecific variation on model estimates for extant Platyrrhini, we calculated the difference between log-transformed maximum and minimum values for the 17 species, yielding an observed proportional range (table 3). This range was not calculated for Brachyteles due to its low sample size. The observed intraspecific proportional ranges average approximately 14% of the total observed variation across all taxa. In addition, there is no association between the percentage of total range that the proportional range represents, and the major variable partitions. Mann-Whitney tests (Sokal and Rohlf, 1995) comparing osteological/dental splits (U  =  641, p > 0.05) and mandibular/cranial splits (U  =  818.5, p > 0.05) were not significant. We also regressed observed proportional range against the species average body mass for each taxon. For each measurement variable, a t-test for the slope of the regression (Sokal and Rohlf, 1995) was not significant; regression slopes ranged from −0.073 to 0.066 with a median value of 0.013. Therefore, proportional ranges do not vary as a function of species body size.

Table 3

Range of variation observed among platyrrhine species The range is calculated as Log[Max] – Log[Min] for each measurement. Brachyteles was not included here due to low sample sizes for this species. Species Abbreviations: Al_cara, Alouatta caraya; Ao_voci, Aotus vociferans; At_geof, Ateles geoffroyi; Ca_calv, Cacajao calvus; Ca_mela, Cacajao melanocephalus; Ci_molo, Callicebus moloch; Cl_goel, Callimico goeldii; Cx_jacc, Callithrix jacchus; Cx_pyg, Callithrix pygmaea; Ce_apel, Cebus apella; Ch_sata, Chiropotes satanas; La_lago, Lagothrix lagotricha; Le_rosa, Leontopithecus rosalia; Pi_pith, Pithecia pithecia; Sg_myst, Saguinas mystax; Sm_sciu, Saimiri sciureus.

Body Mass Estimates for Chilecebus

One of the main objectives in deriving morphometric proxies for body mass among extant taxa is to apply the predictive equations to fossil taxa. Here we apply the body mass proxies derived above to the fossil platyrrhine Chilecebus carrascoensis (Flynn et al., 1995). The only known specimen of this taxon (a skull, SGOPV 3213) lacks the mandible, and pterygoid-zygomatic length (the second best predictor variable) could not be measured. There are, however, 46 measurements that can be evaluated on this nearly complete skull. Of these, 24 variables are osteological and 22 dental. Body mass estimates derived from these variables range widely, from about 85 grams (temporalis length) to approximately 1.7 kg (upper P4 width), a greater than 20-fold difference in maximum and minimum estimates (table 4). The arithmetic mean of these estimates is 589 grams (table 4). Within the range of estimates, osteological variables generally produce lower estimates than do the dental variables; body mass estimates derived from osteological variables range from 85 to 1137 grams, with a mean of 494 grams, whereas dental variables yield a range from 374 to 1725 grams, with a mean of 997 grams. In contrast, the weighted average, which incorporates the proportional likelihoods of the regressions to weight individual measurement estimates, yields an estimated body mass of 319 grams (table 4), considerably lower than the simple mean. Notably the weighted average is also lower than the minimum mass estimate generated from the dental variables. Thus, while there is considerable overlap in body mass estimates between the osteological and dental partitions, the weighted-average model demonstrates that the lower estimates are derived from the variables known to be more reliable body mass indicators among extant taxa.

Table 4

Estimated body mass for Chilecebus carrascoensis Estimates based on morphometric proxies preserved in the holotype cranium. The first set of estimates are based on all of the preserved measurements. The second set relates only the 20 measurements for which Chilecebus conforms to the extant allometry (see text). Individual measurement estimates are given, and their arithmetic mean is presented at the bottom. Weighted estimates refer to a weighted average model using proportional likelihoods as model weights (see text). The sum of these represents the weighted average model and this is also given at the bottom.

This broad range of body mass estimates for Chilecebus is worrisome with regard to estimating body mass for fossil platyrrhines in general. If these variables conform to the same allometries in Chilecebus as are observed among extant taxa, then all should yield similar mass estimates. To the contrary, our results prove that at least some of these variables must scale differently in Chilecebus than in modern platyrrhines. While this might be interpreted as hindering reliable body mass estimation for Chilecebus, the allometric scalings observed among extant taxa must be assumed to hold for the fossil members of the group in any analysis of this kind. The large range of estimated body masses for Chilecebus is by no means unique among extinct primates; for example, body mass estimates for the basal catarrhine Aegyptopithecus (Kay and Simons, 1980) also vary widely depending on the morphometric proxy employed.

Flynn et al. (1995) originally estimated the body mass of Chilecebus at 1.0–1.2 kg, based on upper cheek tooth regressions across all primates. They noted, however, an unusual combination of morphological features in Chilecebus, including a relatively small braincase and the proportions of the upper dentition. Cranial osteological measures tend to produce lower body mass estimates than do dental variables for Chilecebus. However, does this imply that dental variables generally overestimate mass, or that the cranial osteological metrics consistently underestimate it? To determine which measures scale uniquely in Chilecebus relative to extant platyrrhines, we regressed all other variables preserved in Chilecebus against skull length for the set of extant platyrrhine species. This offers a way to describe allometric differences between Chilecebus and the suite of extant platyrrhines. Regression residuals were calculated for Chilecebus and extant taxa relative to each variable's allometry; variables in which the Chilecebus residual was greater than two standard deviations of that observed among the extant taxa were considered variables in which Chilecebus deviates significantly from extant platyrrhine allometry for that variable.

The results of this analysis demonstrate that a unique morphometry relative to living platyrrhines explains the much heavier body mass proposed in the initial description of Chilecebus. Flynn et al. (1995) noted that Chilecebus has proportionally wider upper premolars and molars than extant primates of the same skull length. Here, residuals for all upper cheek tooth widths in Chilecebus do exceed those in extant platyrrhines by more than two standard deviations (table 5). Thus, upper cheek tooth widths (and areas derived from them) will lead to overestimates of body mass. Mesiodistal lengths of both upper incisors of Chilecebus are significantly less than in the extant regression, meaning that these variables produce underestimates of body mass (table 5). Several osteological characters in Chilecebus also deviate significantly from the modern allometries, including: foramen magnum height and width (and the area derived from them), occipital condyle length and width, length and width of the auditory bullae, temporalis length, postorbital constriction, bizygomatic width, length of the frontal chord, orbital width, facial height, and the palatal width at M1 (table 5). All of these are significantly shorter in Chilecebus than predicted by modern allometries, and therefore would yield underestimates of body mass. Thus, no simple pattern accounts for the discrepancy between dental and cranial osteological variable-derived estimates of body mass in Chilecebus, rather various individual variables deviate from the modern allometries, and point to a fundamentally different skull form in Chilecebus relative to extant platyrrhines (Flynn et al., 1995).

Table 5

Measurements taken on the Chilecebus carrascoensis, and their deviation from the extant allometries for each measurement Two standard deviations of the residuals of extant taxa around each allometry are given. Values in bold indicate measurements where Chilecebus falls more than two standard deviations from the regression line. Sign indicates whether the measurement is anomalously large (+) or small (−) on Chilecebus, and therefore will over- or underestimate body mass, respectively.

Even so, 20 variables from both the osteological and dental partitions do conform to modern allometries, and should, therefore, provide reliable estimates of body mass. Limiting estimates to those variables greatly narrows the range of estimates: 430 to 1206 grams, with an average value of 722 grams (table 4). Excluding those measures not conforming to extant allometries eliminates both the extreme high and low body mass estimates for Chilecebus (e.g., upper P3, upper P4 width, and anterior widths of the upper M1 and upper M2, all of which predict masses > 1.5 kg; temporalis length, bizygomatic width, and width of the auditory bullae predicting masses ≤ 100 grams). The reduced set of 20 measurements produces a weighted-average estimate of 583 grams (table 4). We consider this constrained weighted-average to be the most reliable estimate of the body mass for Chilecebus, as this estimate is based on those metrics that can be shown to conform to the extant allometries and utilizes the relative reliabilities of predicting body mass in the extant taxa in weighting the impact of each variable upon the final estimate.

Relative Brain Size in Chilecebus

Flynn et al. (1995) noted the small braincase of Chilecebus (7.46 cc, based on CT data), and suggested that a low level of encephalization may characterize platyrrhines ancestrally. However, the initial body mass estimate for Chilecebus based on upper cheek teeth (1.0–1.2 kg) is approximately double the 583 gram estimate from this analysis. We have shown that upper cheek tooth widths overestimate body mass in Chilecebus, and therefore the original hypothesis of a relatively small brain in Chilecebus might simply reflect an inflated body mass estimate.

LogEQs for extant anthropoids, relative to a baseline allometry for strepsirrhine primates, show that extant catarrhines and platyrrhines have relatively large brains. Both clades have similar distributions of logEQs; catarrhine logEQs range from 1.39 to 2.25 (table 6: including hominoids, but not Homo sapiens), extant platyrrhines from 1.54 to 2.44 (table 6), and they are indistinguishable with a Mann-Whitney test (U  =  47, p > 0.05). Although modern anthropoids possess large brains relative to body mass compared to other mammals (Radinsky, 1973, 1977; Martin, 1990), it is unclear whether this condition typifies Anthropoidea ancestrally, or whether, as Kay et al. (1997) have suggested, Catarrhini and Platyrrhini each evolved relatively large brains in parallel. Martin (1990) analyzed 25 living haplorhines (Anthropoidea plus tarsiers), demonstrating that living anthropoids are large-brained compared to strepsirrhine primates relative to a “basal insectivores” reference frame.

Table 6

Mass, brain volumes, and calculated logEQ values Values relative to Martin's 1990 regression for strepsirrhine primates as a frame of reference. Fossil “tarsioids” and the fossil catarrhine Aegyptopithecus from Martin, 1990. Data for Chilecebus are from this study. Minimum, maximum, and median logEQs for the extant Platyrrhini and Catarrhini are given to the right.

Tejedor et al. (2006) noted that the ∼16.4 Ma cebine platyrrhine Killikaike blakei possesses a strongly vaulted frontal and large anterior cranial fossa, suggesting an expanded forebrain early in cebine evolutionary history. Only the anterior portion of the skull is preserved, thus a total brain volume cannot be estimated, although enlarged forebrains do correlate with relatively larger brains in cebines, suggesting that the brain of Killikaike was likely quite large. Contemporaneous European hominoids, however, lacked significantly expanded brains (Begun, 2002). Martin (1990) noted that several fossil haplorhines also had much smaller brains than their extant relatives, including early “tarsioids” and the early catarrhine Aegyptopithecus zeuxis (logEQs: 0.90–1.06, table 6). These fossil haplorhines have larger brains than expected relative to the strepsirrhine allometry, yet are below the smallest calculated logEQ value for extant anthropoids. The constrained weighted-average body mass model for Chilecebus (583 grams) yields a logEQ of 1.11 (table 6), which is comparable to encephalization values observed among contemporaneous haplorhines, but is lower than observed values in all extant anthropoids. The largest and smallest body mass estimates for Chilecebus from the reduced set of measurements, 1206 grams and 430 grams, yield logEQs of 0.62 and 1.32, respectively. Thus, only if the sole smallest reliable morphometric proxy (of 20 estimators) is correct for Chilecebus would its estimated encephalization even begin to approach the lower bound of observed logEQs among extant anthropoid species (table 6).

The logEQ estimates for Chilecebus and Martin's (1990) data for Aegyptopithecus, imply that early taxa basal to Platyrrhini and Catarrhini (but postdating the divergence between these two clades) had smaller relative brain volumes than extant members of either clade, suggesting that the elevated encephalizations of crown-clade platyrrhines and crown-clade catarrhines have been achieved convergently, pointing to a complex transformation in relative brain size within Anthropoidea. Martin's (1990) suggestion that tarsiers and their fossil allies may exhibit an independent evolutionary trend of increasing encephalization emphasizes this pattern of multiple independent acquisitions of large relative brain volume in Primates. If confirmed, this evolutionary pattern in Primates would mirror independent increases in encephalization observed across mammalian lineages (Jerison, 1970; Radinsky, 1971; Martin, 1984), and within subclades of Carnivora (Finarelli and Flynn, 2007; Finarelli, 2008) and the Cetacea (Marino et al., 2004). The convergent acquisition of higher degrees of encephalization in platyrrhines and catarrhines may yield insights into the evolutionary history of these features in other clades.