<<

Evidence for a convergent slowdown in molecular rates and its implications for the timing of early primate

Michael E. Steipera,b,c,d,1 and Erik R. Seifferte

aDepartment of Anthropology, Hunter College of the City University of New York (CUNY), New York, NY 10065; Programs in bAnthropology and cBiology, The Graduate Center, CUNY, New York, NY 10016; dNew York Consortium in Evolutionary Primatology, New York, NY; and eDepartment of Anatomical Sciences, Stony Brook University, Stony Brook, NY 11794-8081

Edited by Richard G. Klein, Stanford University, Stanford, CA, and approved February 28, 2012 (received for review November 29, 2011) A long-standing problem in primate evolution is the discord divergences—that molecular rates were exceptionally rapid in between paleontological and molecular clock estimates for the the earliest , and that these rates have convergently of crown primate origins: the earliest crown primate slowed over the course of primate evolution. Indeed, a conver- are ∼56 million y (Ma) old, whereas molecular estimates for the gent rate slowdown has been suggested as an explanation for the haplorhine-strepsirrhine split are often deep in the Late Creta- large differences between the molecular and evidence for ceous. One explanation for this phenomenon is that crown pri- the timing of placental mammalian evolution generally (18, 19). mates existed in the but that their fossil remains However, this hypothesis has not been directly tested within have not yet been found. Here we provide strong evidence that a particular mammalian group. this discordance is better-explained by a convergent molecular Here we test this “convergent rate slowdown” hypothesis in rate slowdown in early primate evolution. We show that molecu- primates using a two-step analysis. First, using a comprehensive lar rates in primates are strongly and inversely related to three paleontological and neontological primate dataset (Table S2), life- correlates: body size (BS), absolute endocranial volume we modeled the evolutionary history of three phenotypic traits: (EV), and relative endocranial volume (REV). Critically, these traits body size (BS), absolute endocranial volume (EV), and relative can be reconstructed from fossils, allowing molecular rates to be endocranial volume (REV). These traits were chosen because ANTHROPOLOGY predicted for extinct primates. To this end, we modeled the evo- they are correlated with primate life history (20–22), which is in lutionary history of BS, EV, and REV using data from both extinct turn related to molecular evolutionary rates (23). Furthermore, and extant primates. We show that the primate last common an- BS, EV, and REV are much more easily estimable from fossils cestor had a very small BS, EV, and REV. There has been a subse- than are life-history variables themselves. These analyses yielded quent convergent increase in BS, EV, and REV, indicating that there ancestral reconstructions for BS, EV, and REV from the entire has also been a convergent molecular rate slowdown over primate primate radiation, enabling assessments of phenotypic and life- evolution. We generated a unique timescale for primates by pre- history changes over primate evolution. Second, we explicitly dicting molecular rates from the reconstructed phenotypic values tested whether patterns of variation in molecular rates are corre- for a large phylogeny of living and extinct primates. This analysis lated with patterns of variation in BS, EV, and REV. Molecular – suggests that crown primates originated close to the K Pg bound- rates were estimated from four large DNA-sequence datasets to- ary and possibly in the Paleocene, largely reconciling the molecular taling over 100 Mb, and these rates were correlated with BS, EV, and fossil timescales of primate evolution. and REV using phylogenetically corrected regression techniques. Finally, we joined the results of these two analyses as a mo- hen molecular rates are constant, divergence dates can be lecular clock technique. We predicted molecular rates for pri- Westimated by using fossils to calibrate “molecular clocks” (1). mates based on the BS, EV, and REV reconstruction from our However, there is great variation in molecular rates among analysis of fossil and extant primates and our regression for- (2–7), and this phenomenon can lead to inaccurate or biased mo- mulae. In other words, we predict the molecular rates of long- lecular clock estimates. This problem has precipitated the de- extinct primates using our knowledge of their phenotypic velopment of methods that model rate variation across lineages to attributes rooted in the fossil and extant data. This is a significant date phylogenies when rates vary (8–11). These “relaxed clock” departure from the traditional method of generating molecular methods use multiple calibrations and allow rates to vary according rates using fossils as calibrations. Because our method estimates to the parameters of a model. Relaxed clock methods are especially molecular rates by using paleontological, phylogenetic, geo- appropriate for primates, because this exhibits large and logical, and neontological sources, we feel that it has strong systematic variation in molecular rates both within and among advantages over traditional calibration techniques. groups (e.g., the “hominoid slowdown”) (2, 4, 7, 12, 13, 14). Nev- ertheless, the application of these methods still results in large Results differences between paleontological and molecular estimates for First, we tested whether BS, EV, and REV have changed many primate groups (Fig. 1 and Table S1). This discordance is directionally over primate evolution by using a Bayesian method particularly striking for the origin of the primate crown group, to calculate the harmonic mean-likelihood values for a number because recent molecular studies suggest a Late Cretaceous esti- of models (Table 1). For all three traits, the nondirectional mate (an average of ∼82 Ma) for this event and yet the oldest crown primate fossils are ∼56 Ma old (15)—adifferenceof∼45%. Fur- thermore, studies that have statistically modeled sampling, speci- Author contributions: M.E.S. designed research; M.E.S. and E.R.S. performed research; ation, and preservation rates over the course of primate evolution M.E.S. and E.R.S. analyzed data; and M.E.S. and E.R.S. wrote the paper. are similarly consistent with an ancient origin for crown primates The authors declare no conflict of interest. (16, 17), showing that there are also wide gaps between these This article is a PNAS Direct Submission. methods and “direct reading” approaches to the fossil record. 1To whom correspondence should be addressed. E-mail: [email protected]. Here we investigate an alternative hypothesis for the dis- This article contains supporting information online at www.pnas.org/lookup/suppl/doi:10. crepancy between molecular and fossil estimates of early primate 1073/pnas.1119506109/-/DCSupplemental.

www.pnas.org/cgi/doi/10.1073/pnas.1119506109 PNAS Early Edition | 1of6 Downloaded by guest on September 24, 2021 Dates from the literature: Oldest fossil Avg. molecular Hominidae Great apes This study’s dates: Uncorrected EV corrected BS corrected REV corrected Old World Monkeys Marmosets, , owl & squirrel monkeys Tarsius Lemurs & lorises

80 70 60 50 40 30 20 10 0 Cretaceous PALEO. OLIGO. Pl-Pl Cenozoic

Fig. 1. Time-scaled phylogeny depicting divergence dates for the main groups of primates. Dates along the x axis are in Ma. Average molecular clock date estimates are from six recent studies (Table S1). The fossil femur cartoon indicates the earliest paleontological crown representatives of each : crown Primates and crown , Teilhardina asiatica, 55.8 Ma, or, less securely, slightly older Altiatlasius koulchii (15, 57, 58); crown Strepsirrhini, Sahar- agalago,37Ma(59–61); crown Anthropoidea, Biretia, 37 Ma (59); crown Catarrhini, Morotopithecus, 20.6 Ma (62), crown Cebidae, long- hypothesis (63), Branisella,26–27 Ma (64); crown Cebidae, successive radiation hypothesis (65), Lagonimico (66) and others, 13.3 Ma (66, 67); crown Cercopithecoidea, Microcolobus, 9.9 Ma (68); crown Hominidae, Sivapithecus, 12.5 Ma (69). The colored circles indicate the average divergence estimates from both uncorrected and corrected methods (Table 3).

Brownian-motion model was rejected in favor of a directional result alone generally supports a slowdown in molecular rates in evolution model based on an analysis of the harmonic mean- primate evolution. likelihood values using Bayes factors (SI Text and Tables S3–S6). Second, we tested for a specific relationship between BS, EV, Subsequently, we used a Bayesian method to reconstruct the and REV and molecular rates in four large DNA-sequence ancestral values for BS, EV, and REV at nodes throughout the datasets. In 10 of the 12 phylogenetically corrected regressions primate phylogeny (24) (Fig. 2 and Tables S7 and S8). This there is a significant inverse relationship between molecular rates analysis reconstructed the BS of the primate last common an- and our three phenotypic predictors: BS, EV, and REV (Fig. 3 cestor (LCA) as ∼55 g, with a small EV (2.3 cm3; cc) only slightly and Table 2). These traits explain a large proportion of the larger than the fossil plesiadapiform Ignacius (2.14 cc) (25). The variance in molecular rates. This result provides strong evidence REV of the primate LCA was reconstructed as having been that life-history correlates are related to molecular rates in pri- lower than that of any known primate, living or extinct. mates, as has been found for primates and other (6, 23, fi fi We conclude that BS, EV, and REV have evolved direction- 26). Speci cally noteworthy is the nding that molecular rates ally over primate evolution. Since the primate LCA, all major are very rapid in primates with small BS and EV and low REV. primate lineages have convergently evolved higher BS, EV, and Critically, these analyses generated regression models that can be used to predict molecular rates from BS, EV, and REV data REV. Because of the evidence that life-history correlates are fi inversely related to molecular rates in mammals (6, 23, 26), this in primates, allowing us to predict the speci c molecular rates associated with specific phenotypic values. We joined the ancestral reconstruction data to the formulae Table 1. Final harmonic mean likelihoods from the phylogenetically controlled regressions to predict the molecular rates for all of the lineages of the crown primate ra- Model Scaling parameters BS EV REV diation. For each alignment, the regression formula was used to Brownian – −125.185 −32.841 24.900 δ −127.322 −33.503 23.837 κ −107.814 −26.165 30.016 λ −119.938 −33.413 26.009 Hominoidea δλ − − 119.861 34.255 25.969 Cercopithecoidea δκ −109.111 −24.485 30.815 κλ −109.295 −27.616 26.336 Platyrrhini δλκ −108.973 −26.978 27.838 Tarsius Directional – −120.583 −27.893 32.761 δ −125.564 −31.753 26.340 Body Endocranial Lemuriformes Size κ −103.524 −22.331 38.399 volume λ − − Lorisiformes 113.443 28.336 38.413 Relative endocranial volume δλ −114.656 −29.661 35.745 δκ −107.025 −24.759 33.610 Fig. 2. Phenotypic reconstructions of ancestral BS, EV, and REV. Cladogram κλ −104.183 −24.570 40.275 depicts ancestral trait reconstructions for BS, EV, and REV at key nodes in primate evolution. For BS, the area of the outside circle includes the area δλκ −104.401 −25.350 37.554 under the colored circle. Branches are not to scale.

2of6 | www.pnas.org/cgi/doi/10.1073/pnas.1119506109 Steiper and Seiffert Downloaded by guest on September 24, 2021 Molecular Clock Rate

Body Size Endocranial Volume Relative Endocranial Volume

Fig. 3. Phenotypic data and molecular rate scatter plots. Scatter plots of molecular rate (per 108 y; y axis) and BS, EV, and REV. PGLS regression lines are shown. All regression statistics are found in Table 2. Red, Perelman et al. (7); green, the CYP7A1 region (16); blue, Jameson et al. (14); purple, the CFTR region of Prasad et al. (49).

generate molecular rates for each lineage based on the ancestral rule”), EV, and REV in all of the major primate lineages. These reconstructions for each phenotypic trait. This resulted in pri- results show that the early crown primates were not as “primate- mate phylogenies with “corrected” molecular rates that were like” as would be expected based on an analysis of exclusively scaled to absolute time and converted to ultrametric format (Fig. living taxa, strongly supporting the idea that extinct primates must S1 and SI Appendix). These corrected molecular clock estimates play a key role in models of early primate paleobiology. are younger than those based on traditional calibration methods Our results provide empirical support for an inverse re- for most major divergences within primates (Fig. 1 and Table 3). lationship between three primate life-history correlates and The effect is strongest at the earliest primate nodes, where the molecular rates in nuclear genomes. This supports the long- discrepancy between molecular and fossil evidence for di- standing idea that an organism’s phenotype is correlated with its vergence is most profound. The correction based on REV substitution rate, especially generation time (GT) (3, 6, 23, 26, produced the youngest dates for the deepest primate divergen- 33–35) and body size (36–38). Our finding that both EV and ANTHROPOLOGY ces. Our estimates for the origin of crown primates were 70 Ma REV are inversely correlated with molecular rates is key to using the BS correction, 68 Ma using the EV correction, and 63 testing these hypotheses, because the mutational mechanisms Ma using the REV correction. The uncorrected molecular clock behind these two hypotheses are different. The GT hypothesis estimate for this node was considerably older, both for these four assumes that most germ-line occur during DNA rep- datasets (76 Ma) and from other recent studies of primate mo- lication (3), whereas the BS hypothesis posits that smaller species lecular divergence dates (82 Ma). have higher mass-specific metabolic rates and therefore a higher rate of DNA damage from oxygen radicals that are a by-product Discussion of metabolic processes (36). Because large brain sizes are linked Our study bears on three key issues in primate evolution: the to the extended life of primates (20), the inverse re- phenotypes of early primates, the relationship between molecu- lationship between EV and REV and molecular rates is consis- lar clock rates and life-history correlates in primates, and the tent with the GT hypothesis. Basal metabolic rate (BMR) is molecular and fossil timescale of primate evolution. increased in primates with large brain sizes even when control- Our results support the hypothesis that the first crown primates ling for body size (39), however, so the BS hypothesis would were small (approximately the size of the smallest mouse lemurs) predict that larger-brained primates should have higher with relatively small brains (25, 27, 28), a phenotype that generally rates, which is inconsistent with our findings. Our results may persisted along the haplorhine and anthropoid stem lineages (29, support a role of body size in molecular rates, but only if the 30). This is at odds with alternative hypotheses that suggest that mechanism behind the is not BMR-related. An early primates were either smaller (31) or larger (32). Indeed, alternative body-size hypothesis is that rates are related to se- there has apparently been a convergent increase in BS (“Cope’s lection for higher-fidelity DNA replication in large-bodied

Table 2. Phylogenetic generalized least-square regression results Scaling DNA dataset parameters Predictor R2 F df P value (F) β P value (β) α P value (α)

1 κλ BS 0.589 17.23 2, 12 0.0003 *** −0.0459 0.0014 ** 0.3174 0.0000 *** 2 κ 0.495 5.88 2, 6 0.0386 * −0.0401 0.0515 · 0.267 0.0014 *** 3 κ 0.111 7.40 2, 59 0.0014 ** −0.0194 0.0086 ** 0.1458 0.0000 *** 4 κ 0.415 9.92 2, 14 0.0021 ** −0.0316 0.0071 ** 0.2407 0.0001 *** 1 κ EV 0.345 6.32 2, 12 0.0133 * −0.0716 0.0272 * 0.2806 0.0001 *** 2 κ 0.500 5.99 2, 6 0.0372 * −0.0474 0.0500 · 0.2076 0.0004 *** 3 κ 0.162 10.66 2, 55 0.0001 *** −0.0358 0.0019 ** 0.1303 0.0000 *** 4 κ 0.407 9.60 2, 14 0.0024 ** −0.0373 0.0079 ** 0.1917 0.0001 *** 1 κλ REV 0.467 10.53 2, 12 0.0023 ** −0.1898 0.0070 ** 0.241 0.0000 *** 2 κλ 0.761 19.07 2, 6 0.0025 ** −0.1484 0.0047 ** 0.2025 0.0001 *** 3 κλ 0.009 0.48 2, 55 0.6196 n.s. −0.0214 0.4901 n.s. 0.0968 0.0000 *** 4 κλ 0.117 1.85 2, 14 0.1940 n.s. −0.0373 0.1955 n.s. 0.1559 0.0014 **

Significance indicators: ·,P∼ 0.05; *, 0.05 < P < 0.01; **, 0.01 < P < 0.001; ***, 0.001 < P; n.s., not significant. DNA dataset 1, CYP7A1 (16); 2, Jameson et al. (14); 3, Perelman et al. (7); 4, CFTR region of Prasad et al. (49).

Steiper and Seiffert PNAS Early Edition | 3of6 Downloaded by guest on September 24, 2021 Table 3. Molecular clock results (all dates in Ma) Uncorrected standard molecular clock method BS correction EV correction REV correction

95% CI 95% CI Node Dataset Mean low high Estimate 95% low 95% high Estimate 95% low 95% high Estimate 95% low 95% high

Crown Primate 1 75.8 69.4 87.1 72.6 66.4 80.1 67.3 60.6 75.5 66.3 61.1 77.8 2 73.4 60.8 86.9 67.6 61.7 74.9 67.1 61.3 73.9 59.7 55.2 69.1 3 74.8 64.4 83.2 70.5 65.3 76.7 69.8 62.1 79.7 n.s. n.s. n.s. 4 79.1 70.2 89.8 69.2 64.1 75.2 68.7 63.6 74.6 n.s. n.s. n.s. Average 75.8 66.2 86.8 70 64.4 76.7 68.2 61.9 75.9 63 58.1 73.5 Crown Haplorhini 1 72 64.7 84.4 67.4 61.5 74.5 62.4 56.3 69.9 62.5 54.3 73.9 2 68.8 56.3 77.7 64.2 58.3 71.4 63 57.7 69.4 56.8 49.9 66.2 3 70.2 60.3 78.3 66.4 61.5 72.4 66.4 59.2 75.7 n.s. n.s. n.s. Average 70.3 60.4 80.1 66 60.4 72.7 63.9 57.7 71.7 59.7 52.1 70 Crown Strepsirrhini 1 50.7 40.3 61.1 59 54.2 64.7 54 48.9 60.2 53.9 52.1 61.6 2 56.3 42.2 67.3 52.9 48.8 57.8 53 48.8 57.9 48.2 46.4 54.6 3 52.9 42.3 62.9 54 50.1 58.6 52.4 46.9 59.5 n.s. n.s. n.s. 4 49.1 39.6 59.9 49.2 45.9 53 50.1 46.6 54.2 n.s. n.s. n.s. Average 52.2 41.1 62.8 53.8 49.7 58.5 52.4 47.8 57.9 51.1 49.2 58.1 Crown Anthropoidea 1 35.9 30.3 42.1 36.3 33 40.5 34.7 31.2 39.1 36 30.6 44.2 2 34.4 27.3 44.3 35.5 31.9 40 34.7 31.6 38.5 32.3 27.6 39.3 3 41.4 36 47.4 35.4 32.7 38.6 37.5 33.4 43 n.s. n.s. n.s. 4 41.9 36.9 47.5 34.2 31.6 37.3 33.3 31.1 35.9 n.s. n.s. n.s. Average 38.4 32.6 45.3 35.3 32.3 39.1 35.1 31.8 39.1 34.1 29.1 41.7 Crown Platyrrhini 3 19.5 16 22.4 19.8 18.6 21.1 21 19.1 23.2 n.s. n.s. n.s. 4 21 17.3 25 18.1 17.2 19.2 18.8 17.8 19.9 n.s. n.s. n.s. Average 20.3 16.7 23.7 19 17.9 20.2 19.9 18.5 21.6 —— — Crown Cebidae 1 16.2 13.1 19.9 18.7 17.6 19.9 17.6 16.5 18.9 22.5 19.8 26.4 3 15.9 12.8 18.5 16.7 15.8 17.8 17.1 15.8 18.6 n.s. n.s. n.s. 4 17.1 14.1 20.7 15.1 14.4 15.9 15.8 15 16.7 n.s. n.s. n.s. Average 16.4 13.3 19.7 16.8 15.9 17.8 16.8 15.8 18 22.5 19.8 26.4 Crown Catarrhini 1 24.2 20.9 27.6 25.2 22.5 28.6 24.9 22.1 28.5 24.1 20.0 30.7 2 24 19.7 28.6 25.8 23 29.4 25.3 22.9 28.3 22.8 19.1 28.4 3 29.3 25.8 32.9 23.8 21.8 26.2 27.3 23.8 32 n.s. n.s. n.s. 4 28.3 25.2 31.9 23.1 21.2 25.9 22.1 20.6 23.9 n.s. n.s. n.s. Average 26.4 22.9 30.3 24.5 22.1 27.6 24.9 22.3 28.2 23.5 19.6 29.5 Crown Cercopithecoidea 1 12.4 10.2 15.5 14.1 12.7 15.8 13.4 12.3 14.9 14.3 12.2 17.6 3 14.7 12.9 16.4 10.3 9.6 11.2 11.7 10.5 13.3 n.s. n.s. n.s. 4 14 12.7 16 11.1 10.2 12 11 10.3 11.9 n.s. n.s. n.s. Average 13.7 11.9 16 11.8 10.8 13 12.1 11 13.3 14.3 12.2 17.6 Crown Hominidae 1 15.7 13.2 19 16.6 14.8 18.8 17.2 15.2 19.7 15.9 13.1 20.7 (great ape) 2 13.5 10.4 18.3 16.3 14.5 18.6 15.5 14 17.3 13.7 11.4 17.5 3 16.3 14.2 18.7 15.4 14.1 17 22.1 18.7 26.8 n.s. n.s. n.s. 4 16.9 15.1 19.2 13.6 12.5 15 13.3 12.4 14.4 n.s. n.s. n.s. Average 15.6 13.2 18.8 15.5 14 17.3 17 15.1 19.5 14.8 12.2 19.1 1 6.8 6.0 7.9 7.4 6.7 8.3 8.3 7.5 9.3 9.3 7.5 12.4 2 6.9 6.0 8.3 5.6 5.2 6.2 5.7 5.3 6.3 5.9 4.8 7.7 3 6.7 6.0 7.9 7.2 6.8 7.6 12 10.3 14.3 n.s n.s n.s 4 6.8 6.0 7.9 5.2 5 5.6 5.4 5.1 5.8 n.s n.s n.s Average 6.8 6.0 8.0 6.4 5.9 6.9 7.9 7.1 8.9 7.6 6.2 10.1

DNA dataset 1, CYP7A1 (16); 2, Jameson et al. (14); 3, Perelman et al. (7); 4, CFTR (49). CI, confidence interval; n.s., nonsignificant; —, not determined.

organisms because they have larger numbers of cells to maintain calibrate molecular rates (41, 42). Our corrected dates indicate that (19). Our results also support the intriguing idea that mutation crown primates originated near the K–Pg boundary and perhaps as rates are not primarily a by-product of other processes but are recently as the Paleocene. These estimates are younger than those rather part of an overall life-history strategy (19, 40), a hypoth- provided by past molecular clock and modeling studies and are esis that deserves further consideration. much closer in age to the earliest undoubted crown primate, Teil- Regardless of the mechanism, we clearly show that early pri- hardina asiatica (∼56 Ma) (15) and direct reading interpretations of mates had low BS, EV, and REV and, by inference, also had fast the fossil record (43). Our findings suggest that the primate fossil molecular rates. Furthermore, we show that rates slowed down over the course of primate evolution, strongly supporting the ap- record is not necessarily poorly sampled. Rather, convergent evo- fi plication of the convergent rate slowdown hypothesis (18, 19) to lution of extended life histories within primates has led to arti - primates. Due to the strength of these relationships, we were able cially ancient molecular clock estimates in this group. Further work to predict the molecular rates of different primate lineages from is required to assess whether a convergent rate slowdown hypoth- ancestral reconstructions of BS, EV, and REV, side-stepping the esis (18, 19) can resolve the molecular and fossil timescales of other numerous concerns surrounding the use of the fossil record to placental mammalian .

4of6 | www.pnas.org/cgi/doi/10.1073/pnas.1119506109 Steiper and Seiffert Downloaded by guest on September 24, 2021 Materials and Methods example, the BS estimate used for the human lineage was the average of the Our study has three methodological steps. The first step is an analysis of BS, human BS, the reconstructed BS at the human/Australopithecus ancestor, EV, and REV evolution across primates, including estimation of ancestral and the reconstructed BS at the human/Pan ancestor. This was done because values. The second step is a test for a relationship between molecular clock the rate estimates are estimated for an entire branch and therefore the rates and BS, EV, and REV. The third step links the results of the first two steps same should hold for the phenotypic traits. We restricted the analysis to tips to generate corrected molecular clock rates that were applied to molecular to allow the phenotypic estimates to derive more directly from observable branch lengths, thereby generating corrected molecular divergence date data rather than reconstructions. Furthermore, it was sensible to exclude the estimates. deeper phylogenetic lineages from the regressions because it is within the earliest parts of primate phylogeny where the discord between molecular Phenotypic Evolution. Phenotypic data. We modeled the evolution of three rates and phenotypes is predicted to be most profound. Regressions were phenotypic traits: BS, EV, and REV (Table S2). Residuals from a phylogeneti- conducted in caper (52). fi β cally corrected BS/EV regression were used as a measure of REV by taking the In two cases the regression was signi cant, but the P values for were only difference between the observed and the predicted EV values (28) (SI Text). near significance [Jameson et al. (14) dataset; BS, β P = 0.05154 and EV, β P = Models of trait evolution. We used BayesTraits (44–46) to test different models 0.05001]. Because of the overall significance of the regression and the of trait evolution for BS, EV, and REV. We tested whether a directional closeness of the probability values to the standard α value (0.05), these re- model of trait evolution was preferred to a nondirectional Brownian-motion gression coefficients were used in subsequent analyses. model. These analyses also tested whether different phylogenetic scaling parameters (δ, λ, and κ) improved the model of evolution by modeling the Correcting Molecular Clock Date Estimates. Predicting corrected molecular rates. tempo, model, and phylogenetic association of a trait’s evolution (45). Our To predict the molecular clock rates for each lineage, we used the recon- analysis closely follows Montgomery et al. (28). structed trait values from the first step with the regression formula for that . The analyses require trait data from extant and fossil taxa trait for each DNA-sequence dataset from the second step. The average trait and a bifurcating, time-scaled tree. We used a supertree approach to obtain value for each lineage was used, as in the molecular rate regressions. For a time-scaled phylogeny of living and extinct primates, combining results example, for the CFTR alignment, the BS regression formula is molecular rate from multiple studies into a taxon-rich phylogeny of 62 extant primate (per 108 y) = −0.0316 × BS + 0.2407. For the Pongo lineage, with a BS value of genera and 123 extinct primates (SI Text and Fig. S2). 4.558, the formula predicts that the molecular rate is 0.0967 changes per Model testing. Models were compared using Bayes factors (BFs). BFs operate in 108 y. Corrected rategrams for each trait for each sequence dataset are in a manner similar to likelihood ratio testing. The test statistic is 2 × (log SI Appendix. [harmonic mean(better model)] − log[harmonic mean(worse model)]), and Estimating corrected molecular dates. Using these corrected molecular rates, values greater than 2 are evidence for the better model being preferred molecular clock estimates (branch lengths in absolute time) were generated (47). The harmonic mean of the likelihoods of a long run of postburnin for each lineage, for each phenotypic trait. To do this, the maximum-likeli- samples approximates the marginal likelihood of each model. For both the hood branch lengths for each DNA-sequence dataset were rescaled by the ANTHROPOLOGY directional and nondirectional model, BFs were used to compare the in- corrected molecular rates. For each dataset, the maximum-likelihood branch δ κ λ clusion of all permutations of the scaling parameters (none included, , , , lengths were calculated using baseml (50) under the HKY+Γ5 model (53, 54) δλ, δκ, κλ, and δκλ). (This was also used to choose the scaling factors for the (SI Appendix). In the CFTR alignment, for example, the Pongo branch length regression analysis used for estimating REV data.) Once the model was was 0.0131. Dividing this by the BS corrected molecular rate for Pongo chosen, BFs were used to test between the directional and nondirectional (0.0967 changes per 108 y) yields a branch length of 13.6 Ma. The un- models. For each model, the (MCMC) run had corrected estimate for the Pongo lineage for this dataset was longer 500,000 burn-in steps followed by at least 15 million steps sampled every (16.9 Ma). – 500th step. Acceptance rates were tuned to 15 40%. To assess convergence, This method was applied to the four DNA-sequence datasets for each trait ’ fi each model s run was conducted twice. If the nal harmonic mean like- for which there was a significant PGLS regression, resulting in 10 trees with > lihoods differed by 1, the number of iterations was increased until the each branch length scaled to absolute time. Corrected nonultrametric mo- < fi differences were 1. Runs had up to 100 million iterations. The nal har- lecular clock trees for each trait for each of the DNA-sequence datasets are in monic mean likelihoods from these two runs were averaged for use in the BF SI Appendix. Because there is variation in the time-scaled branch lengths, the – analysis (SI Text and Tables S3 S6 and S9). trees are not ultrametric. To generate ultrametric trees, we used the mean Ancestral trait reconstruction. For each trait, we used MCMC data from the best path length (MPL) method (55) implemented in APE (56). We generated model in conjunction with the phylogeny described above to estimate an- interval estimates for these divergence dates by conducting the same mo- cestral trait values in BayesTraits. For each reconstruction, model data were lecular date correction technique using the upper and lower limits from the compiled from two MCMC analyses to calculate ancestral values in a phylo- 95% Bayesian HPDs of the BS, EV, and REV reconstructions. Corrected genetic context (24). Each MCMC run had 500,000 burnin and 2–10 million ultrametric molecular clock trees for each trait for each of the DNA-se- subsequent steps sampled every 500th step. Acceptance rates were tuned to quence datasets are found in Fig. S3 and SI Appendix. 15–40% by changing the proposal parameter and by adding different sets of Because the MPL method generates ultrametric trees, when one particular nodes to each run. Convergence was assessed with effective sample sizes (48) branch has a large difference relative to close neighboring branches, negative and multiple runs. The final estimate and 95% highest posterior densities branch lengths are sometimes computed to allow all of the tips to “line up.” (HPDs) for each trait value for each node were averages of two runs (Tables S7 and S8). In our results, some short negative branch lengths are present within the New World monkeys and also within some groups of closely related taxa in the taxon-rich datasets (e.g., in the lemuriform clade). Within closely related Analysis of Molecular Rates. DNA-sequence datasets. The four datasets analyzed taxa, small differences in branch lengths can lead to negative branch lengths were: (i) a realignment of data from the CFTR region from 16 primates from because of short internodal branches. In these cases, there is also sometimes ref. 49, (ii) a taxonomically supplemented alignment of the CYP7A1 region phylogenetic uncertainty that may cause short negative branch lengths. It (16) from 14 primates, (iii) a genome-wide sample of codon third positions may also be the case that some of the variance in molecular rate variation is from 1,078 transcripts from eight primates (14), and (iv) a taxonomically not adequately modeled by the present method, another potential cause of well-sampled, genome-wide set of regions (7). In total, these four align- ments are over 6 Mb in length and total over 100 Mb of alignment data. the negative branch lengths. Additional fossil and phylogenetic evidence These datasets allowed estimates of rate variation from datasets that are would help determine whether these differences are stochastic, have a de- alternately very long, very genomically widespread, and very taxonomically terministic cause related to the present method (such as undocumented well-sampled. Methodological details for the alignment of the CFTR and changes in body size), or another cause. CYP7A1 data are in SI Text. Molecular rate regressions. We conducted weakly calibrated relaxed clock ACKNOWLEDGMENTS. We thank E. Douzery, J. Fleagle, A. Meade, D. Orme, analyses (11) to generate rates across the primate phylogeny for the four H. Pontzer, and the reviewers for comments on the manuscript and assistance with methods. This research was supported, in part, by National datasets independently using a relaxed clock technique (10, 11, 50) (SI Text). Science Foundation Grants CNS-0958379, CNS-0855217, and BCS-0819186 Using a phylogenetically corrected least-squares method (PGLS) (51), we and the City University of New York High Performance Computing Center. regressed these molecular rates from phylogenetic tips against the tip values The infrastructure of the Anthropological Laboratory at Hunter for BS, EV, and REV for each of the four datasets. For the phenotypic vari- College was supported, in part, by Grant RR003037 from the National Center able, the average trait value over the entire tip lineage was used. For for Research Resources, a component of the National Institutes of Health.

Steiper and Seiffert PNAS Early Edition | 5of6 Downloaded by guest on September 24, 2021 1. Zuckerkandl E, Pauling L (1965) Evolutionary divergence and convergence in . 34. Laird CD, McConaughy BL, McCarthy BJ (1969) Rate of fixation of sub- Evolving Genes and Proteins, eds Bryson V, Vogel HJ (Academic, New York), pp 97–166. stitutions in evolution. Nature 224(5215):149–154. 2. Li WH, Tanimura M (1987) The molecular clock runs more slowly in man than in apes 35. Li WH, Tanimura M, Sharp PM (1987) An evaluation of the molecular clock hypothesis and monkeys. Nature 326(6108):93–96. using mammalian DNA sequences. J Mol Evol 25:330–342. 3. Wu CI, Li WH (1985) Evidence for higher rates of nucleotide substitution in 36. Martin AP, Palumbi SR (1993) Body size, metabolic rate, generation time, and the than in man. Proc Natl Acad Sci USA 82:1741–1745. molecular clock. Proc Natl Acad Sci USA 90:4087–4091. 4. Steiper ME, Young NM (2006) Primate molecular divergence dates. Mol Phylogenet 37. Bromham L, Rambaut A, Harvey PH (1996) Determinants of rate variation in mam- Evol 41:384–394. malian DNA sequence evolution. J Mol Evol 43:610–621. 5. Steiper ME, Young NM, Sukarna TY (2004) Genomic data support the hominoid 38. Gillooly JF, Allen AP, West GB, Brown JH (2005) The rate of DNA evolution: Effects of slowdown and an Early estimate for the hominoid-cercopithecoid di- body size and temperature on the molecular clock. Proc Natl Acad Sci USA 102(1): vergence. Proc Natl Acad Sci USA 101:17021–17026. 140–145. 6. Hwang DG, Green P (2004) Bayesian Markov chain Monte Carlo sequence analysis 39. Isler K, van Schaik CP (2006) Metabolic costs of brain size evolution. Biol Lett 2: reveals varying neutral substitution patterns in mammalian evolution. Proc Natl Acad 557–560. Sci USA 101:13994–14001. 40. Bromham L (2009) Why do species vary in their rate of ? Biol Lett – 7. Perelman P, et al. (2011) A molecular phylogeny of living primates. PLoS Genet 7: 5:401 404. e1001342. 41. Graur D, Martin W (2004) Reading the entrails of chickens: Molecular timescales of – 8. Thorne JL, Kishino H (2002) Divergence time and evolutionary rate estimation with evolution and the illusion of precision. Trends Genet 20(2):80 86. multilocus data. Syst Biol 51:689–702. 42. Steiper ME, Young NM (2008) Timing primate evolution: Lessons from the discor- 9. Thorne JL, Kishino H, Painter IS (1998) Estimating the of the rate of dance between molecular and paleontological estimates. Evol Anthropol 17(4): – molecular evolution. Mol Biol Evol 15:1647–1657. 179 188. – 10. Rannala B, Yang Z (2007) Inferring times under an episodic molecular clock. 43. Gingerich PD, Uhen MD (1994) Time of origin of primates. J Hum Evol 27:443 445. Syst Biol 56:453–466. 44. Pagel M (1997) Inferring evolutionary processes from phylogenies. Zool Scr 26: – 11. Yang Z, Rannala B (2006) Bayesian estimation of species divergence times under 331 348. a molecular clock using multiple fossil calibrations with soft bounds. Mol Biol Evol 23: 45. Pagel M (1999) Inferring the historical patterns of biological evolution. Nature 401: – 212–226. 877 884. 12. Kim SH, Elango N, Warden C, Vigoda E, Yi SV (2006) Heterogeneous genomic mo- 46. Pagel M, Meade A (2011) BayesTraits (Univ of Reading, Reading, UK), Version 1.0. 47. Kass KE, Raferty AE (1995) Bayes factors. J Am Stat Assoc 90:773–795. lecular clocks in primates. PLoS Genet 2:e163. 48. Rambaut A, Drummond AJ (2007) Tracer (http://beast.bio.ed.ac.uk/Tracer), Version 13. Elango N, Thomas JW, Yi SV; NISC Comparative Sequencing Program (2006) Variable 1.4. molecular clocks in hominoids. Proc Natl Acad Sci USA 103:1370–1375. 49. Prasad AB, Allard MW, Green ED; NISC Comparative Sequencing Program (2008) 14. Jameson NM, et al. (2011) Genomic data reject the hypothesis of a prosimian primate Confirming the phylogeny of mammals by use of large comparative sequence data clade. J Hum Evol 61:295–305. sets. Mol Biol Evol 25:1795–1808. 15. Smith T, Rose KD, Gingerich PD (2006) Rapid Asia-Europe-North America geographic 50. Yang Z (2007) PAML 4: Phylogenetic analysis by maximum likelihood. Mol Biol Evol dispersal of earliest Eocene primate Teilhardina during the Paleocene-Eocene Ther- 24:1586–1591. mal Maximum. Proc Natl Acad Sci USA 103:11223–11227. 51. Freckleton RP, Harvey PH, Pagel M (2002) Phylogenetic analysis and comparative data: 16. Wilkinson RD, et al. (2011) Dating primate divergences through an integrated analysis A test and review of evidence. Am Nat 160:712–726. of palaeontological and molecular data. Syst Biol 60(1):16–31. 52. Orne D, et al. (2011) caper: Comparative analyses of and evolution in R 17. Tavaré S, Marshall CR, Will O, Soligo C, Martin RD (2002) Using the fossil record to (Imperial College, London, UK), Version 0.4. estimate the age of the last common ancestor of extant primates. Nature 416: 53. Hasegawa M, Kishino H, Yano T (1985) Dating of the human-ape splitting by a mo- 726–729. lecular clock of mitochondrial DNA. J Mol Evol 22(2):160–174. 18. Bromham L (2003) Molecular clocks and explosive radiations. J Mol Evol 57(Suppl 1): 54. Yang Z (1994) Maximum likelihood phylogenetic estimation from DNA sequences S13–S20. with variable rates over sites: Approximate methods. J Mol Evol 39:306–314. 19. Bromham L (2011) The genome as a life-history character: Why rate of molecular 55. Britton T, Oxelman B, Vinnersten A, Bremer K (2002) Phylogenetic dating with con- evolution varies between species. Philos Trans R Soc Lond B Biol Sci 366: fidence intervals using mean path lengths. Mol Phylogenet Evol 24(1):58–65. 2503–2513. 56. Paradis E, Claude J, Strimmer K (2004) APE: Analyses of phylogenetics and evolution in 20. Deaner RO, Barton RA, van Schaik CP (2002) Primate brains and life history: Renewing R language. Bioinformatics 20:289–290. the connection. Primate Life Histories and Socioecology, eds Kappeler PM, Pereira ME 57. Ni X, Wang Y, Hu Y, Li C (2004) A euprimate skull from the early Eocene of China. – (Univ of Chicago Press, Chicago), pp 233 265. Nature 427(6969):65–68. 21. Harvey PH, Clutton-Brock TH (1985) Life history variation in primates. Evolution 39: 58. Bowen GJ, et al. (2002) Mammalian dispersal at the Paleocene/Eocene boundary. – 559 581. Science 295:2062–2065. 22. Isler K, van Schaik CP (2009) The expensive brain: A framework for explaining evo- 59. Seiffert ER (2006) Revised age estimates for the later Paleogene mammal faunas of – lutionary changes in brain size. J Hum Evol 57:392 400. Egypt and Oman. Proc Natl Acad Sci USA 103:5000–5005. fi 23. Tsantes C, Steiper ME (2009) Age at rst reproduction explains rate variation in the 60. Seiffert ER, Simons EL, Attia Y (2003) Fossil evidence for an ancient divergence of – strepsirrhine molecular clock. Proc Natl Acad Sci USA 106:18165 18170. lorises and galagos. Nature 422:421–424. 24. Pagel M, Meade A, Barker D (2004) Bayesian estimation of ancestral character states 61. Seiffert ER, Bown TM, Clyde WC, Simons EL (2008) , paleoenvironment, and – on phylogenies. Syst Biol 53:673 684. age of Birket Qarun Locality 2 (BQ-2), Fayum Depression, Egypt. Elwyn Simons: A 25. Silcox MT, Dalmyn CK, Bloch JI (2009) Virtual endocast of Ignacius graybullianus Search for Origins, eds Fleagle JG, Gilbert CC (Springer, New York), pp 71–86. (Paromomyidae, Primates) and brain evolution in early primates. Proc Natl Acad Sci 62. Gebo DL, et al. (1997) A hominoid from the early Miocene of Uganda. Science – USA 106:10987 10992. 276:401–404. 26. Welch JJ, Bininda-Emonds OR, Bromham L (2008) Correlates of substitution rate 63. Takai M, Anaya F, Shigehara N, Setoguchi T (2000) New fossil materials of the earliest variation in mammalian -coding sequences. BMC Evol Biol 8:53. , Branisella boliviana, and the problem of platyrrhine origins. Am 27. Charles-Dominique P, Martin RD (1970) Evolution of lorises and lemurs. Nature 227: J Phys Anthropol 111:263–281. 257–260. 64. Kay RF, MacFadden BJ, Madden RH, Sandeman H, Anaya F (1998) Revised age of the 28. Montgomery SH, Capellini I, Barton RA, Mundy NI (2010) Reconstructing the ups and Salla beds, Bolivia, and its bearing on the age of the Deseadan South American land downs of primate brain evolution: Implications for adaptive hypotheses and Homo mammal “age.” J Vertebr Paleontol 18(1):189–199. floresiensis. BMC Biol 8:9. 65. Kay RF, et al. (2008) The anatomy of Dolichocebus gaimanensis, a stem platyrrhine 29. Ross CF (2000) Into the light: The origin of Anthropoidea. Annu Rev Anthropol 29: monkey from Argentina. J Hum Evol 54:323–382. 147–194. 66. Kay RF (1994) “Giant” from the Miocene of Colombia. Am J Phys Anthropol 30. Simons EL, Seiffert ER, Ryan TM, Attia Y (2007) A remarkable female cranium of the 95:333–353. early Oligocene anthropoid Aegyptopithecus zeuxis (Catarrhini, Propliopithecidae). 67. Flynn JJ, Guerrero J, Swisher CC, III (1997) of the Honda group. Ver- Proc Natl Acad Sci USA 104:8731–8736. tebrate in the Neotropics: The Miocene Fauna of , Colombia, 31. Gebo DL (2004) A shrew-sized origin for primates. Am J Phys Anthropol (Suppl 39): eds Kay RF, Madden RH, Cifelli RL, Flynn JJ (Smithsonian Inst Press, Washington, DC), 40–62. pp 44–59. 32. Soligo C, Martin RD (2006) Adaptive origins of primates revisited. J Hum Evol 50: 68. Nakatsukasa M, et al. (2010) Earliest colobine skeletons from Nakali, Kenya. Am J Phys 414–430. Anthropol 143:365–382. 33. Goodman M (1961) The role of immunochemical differences in the phyletic de- 69. Kappelman J, et al. (1991) The earliest occurrence of Sivapithecus from the middle velopment of human behavior. Hum Biol 33:131–162. Miocene Chinji Formation of Pakistan. J Hum Evol 21(1):61–73.

6of6 | www.pnas.org/cgi/doi/10.1073/pnas.1119506109 Steiper and Seiffert Downloaded by guest on September 24, 2021