Journal of Human Evolution 69 (2014) 123e128
Contents lists available at ScienceDirect
Journal of Human Evolution journal homepage: www.elsevier.com/locate/jhevol
News and views
News and views: Non-metric dental traits and hominin phylogeny Katherine Carter a, *, Steven Worthington b, Tanya M. Smith a a b
Department of Human Evolutionary Biology, Harvard University, 11 Divinity Avenue, Cambridge, MA 02138, USA Institute for Quantitative Social Science, Harvard University, 1737 Cambridge Street, CGIS Knafel Building, Cambridge, MA 02138, USA
a r t i c l e i n f o Article history: Received 8 July 2013 Accepted 10 January 2014 Available online 07 March 2014 Keywords: Hominin evolution ASUDAS Dental anthropology Phylogenetic methods Character weighting
Introduction Analyses of hominin dental remains conventionally include measurements of tooth crown sizes and descriptions of occlusal morphology such as minor accessory cusps, fissure patterns, and ridges (e.g., Wood, 1981; Aiello and Dean, 1990; Bailey, 2006). Following Dahlberg (1951), Turner et al. (1991) developed a formal system for dividing these ‘non-metric’ aspects of dental morphology into discrete categories. This system, termed the Arizona State University Dental Anthropology System (ASUDAS), is an effective tool for discriminating among modern human populations and for assessing inter-population relationships (Scott and Turner, 1997). Other researchers later used ASUDAS to examine the evolutionary relationships of various Pleistocene hominins (Irish and Guatelli-Steinberg, 2003; Martinón-Torres et al., 2007; but see Bailey et al., 2009). Most recently, Irish et al. (2013) used ASUDAS to assess the phylogenetic position of Australopithecus sediba, finding support for both an Au. sediba þ Au. africanus clade and a clade uniting South African australopiths with Homo. However, as Kimbel (2013) has argued, there are theoretical issues with applying ASUDAS to assess phylogenetic relationships from small samples of fossil hominin dental remains. Here we explore the
* Corresponding author. E-mail address: [email protected] (K. Carter). http://dx.doi.org/10.1016/j.jhevol.2014.01.003 0047-2484/Ó 2014 Elsevier Ltd. All rights reserved.
suitability of applying a method developed for partitioning among modern human populations to assess interspecies relationships among fossil hominins. We then discuss the ramifications of different choices made during phylogenetic estimation, including those pertaining to character weighting, clade support, and outgroup composition. We find that slight alteration of phylogenetic assumptions leads to numerous equally possible evolutionary reconstructions for Au. sediba. Human populations vs. hominin species The ASUDAS system allows researchers to assess intraspecific relationships using metrics such as ‘Mean Measure of Divergence’ (MMD; Scott and Turner, 1997), which allows creation of hierarchical clusters of human populations using a distance matrix of overall phenetic similarity (Irish, 2006; LeBlanc et al., 2008). It is also used to assess the affinity of unknown groups with known populations based on suites of characters (Hanihara, 1977; Irish and Turner, 1990; Aguirre et al., 2006; Pereira et al., 2012). For this technique, any individual trait is not diagnostic of a particular population; rather, it is suites of character frequencies, not the presence or absence of a single character, that determine the affinity of a population. Implicit in this method, and indeed in the samples used to create population standards, is the idea that much of the human variability in ASUDAS relates to differences in frequencies, rather than presence or absence of characters states (Scott and Turner, 1997). While Irish et al. (2013) used a geneticallyinformed threshold model to determine cutoff points for trait presence and absence, Scott and Turner (1997) made no claim that each successive change between ASUDAS expression levels was related to meaningful differences in genetic expression, a key factor in dividing characters into states for phylogenetic analysis (Hawkins et al., 1997; Wiens, 2001). While it is well established that dental morphology is under considerable genetic control, the number of character states is not necessarily related to the number of genetic changes needed to produce population variability. For example, while enamel extension UM1 is divided into 3 states of expression and Carabelli’s Cusp UM1 into 7, there is no evidence that there are more than twice the genes involved in Carabelli’s cusp expression. A phenetic method designed to assess human population relationships may or may not use characters relevant for determining
124
K. Carter et al. / Journal of Human Evolution 69 (2014) 123e128
evolutionary relationships among species. For the same reason that traits used to differentiate among breeds of dogs would not be useful to reconstruct the phylogeny of Canidae (Bininda-Emonds et al., 1999), there is no indication that the traits used for within-
species analysis capture the range of variability expressed among species. Some of the human variability that ASUDAS samples is monomorphic among non-Neandertal fossil hominins (e.g., absence of Bushman canine, the presence of a ridge on the mesial surface of the upper canine), and ASUDAS fails to capture variation in hominins known to be phylogenetically informative, such as the degree of P4 molarization, the rotation of the P3, degree of dimorphism in the canines, and relative molar and incisor sizes (Fig. 1A) (Wood, 1981; Aiello and Dean, 1990). Based on the expression of ASUDAS traits across human populations, the most parsimonious interpretation of character state polymorphism is that ASUDAS characters were not fixed in the last common ancestor of all modern humans (Fig. 1B). While there are multiple processes that could produce this pattern, it is likely that fossil hominins had the genetic potential for polymorphism in most of these traits, making estimates of interspecific relationships with small sample sizes problematic. Alternative explanations, such as independent acquisition of traits in different populations of recent hominins (e.g., independent acquisition of shoveling in Northeastern Asians and Neandertals; Denton, 2011) call into question the phylogenetic use of these characters (Fig. 1C). Methodological choices impact phylogenetic analyses Decisions related to character weighting schemes, branch support criteria, and outgroup composition can dramatically affect phylogenetic analysis (Nadal-Roberts and Collard, 2005; Bjarnason et al., 2011; Worthington, 2012). Irish et al. (2013) reported that all characters in their analysis were treated as being of equal weight. However, weighting nevertheless occurs as a consequence of the way character state changes are quantified. For characters with binary states, only a single change between states is possible. For multistate characters, a transition from one state to any other can occur either in a single change (‘unordered’) or only by traveling through all intermediate states (‘ordered’) (Slowinski, 1993). Irish et al. (2013) treated their characters as ordered, and combined with the unequal number of states among their sampled traits (Table 1), this resulted in characters with more states often being given more weight in their analysis. Several of the multistate characters sampled undergo non-sequential transformations (e.g., from state 0 to state 3) on their most parsimonious tree (MPT; the tree with an internal branching order that requires the fewest evolutionary events to explain the tip data). When this occurs, the hypothesis of state order implicitly gives more weight to these characters (Hauser and Presch, 1991: Figure 10, p. 260). When character state transitions are ordered, the parsimony algorithm counts change between any intermediate states as separate evolutionary steps. In other words, twice the weight is given to a two-step change from, for example, states ‘1’ to ‘3’ than to any transition between binary character states. Thus a character such as Protostylid LM1, which unites Au. sediba and Au. africanus with a character state change from ‘1’ (at the base of the clade
Figure 1. Theoretical problems that arise with applying ASUDAS in the fossil record. (A) ASUDAS, a method designed to partition intraspecific variation in Homo sapiens may not be useful for assessing interspecies variation in the fossil record as there are many phylogenetically informative characters that are not sampled (light region) and many characters which are sampled that are not phylogenetically informative (dark region). Only a small subset of phylogenetic variance is captured. Furthermore, because almost all variants of ASUDAS traits are found in all modern human populations, either (B) these traits were not fixed in the last common ancestor of modern humans or (C) these traits evolved independently in different lineages. Either explanation suggests that there are problems with using ASUDAS to estimate evolutionary relationships among fossil hominins.
Table 1 Number of states exhibited by characters in Irish et al.’s (2013) matrix. Number of states 1 2 3 4 6
Number of transitions
Inverse transition weight
Count of characters
Character number
0 1 2 3 5
0 1 0.5 0.333 0.2
1 7 7 6 1
18 3 6 9 11 15 21 22 2 4 5 8 10 17 19 1 7 12 13 14 20 16
Character numbers reflect the order in which characters are listed in Irish et al.’s (2013) Table 2, p. 2.
K. Carter et al. / Journal of Human Evolution 69 (2014) 123e128
grouping these two species with Homo) to ‘4’ (at the node linking these two southern African australopiths), contributes three-times the phylogenetic information of any binary character (Fig. 2A). Carabelli’s cusp UM1 also contributes twice the resolving power of
125
a binary character (Fig. 2B). This form of differential weighting leads to phylogenetic results that are driven primarily by a few characters with many states. Furthermore, of the five characters listed in Irish et al.’s (2013) analysis uniting Au. sediba and Au. africanus only two
Figure 2. Maximum parsimony cladograms illustrating character weighting schemes. (A) Protostylid LM1 and (B) Carabelli’s cusp UM1 optimized onto the MPT with states treated as ordered. States in parentheses are for terminal taxa, while those at internal nodes are maximum parsimony reconstructions. State changes uniting Au. sediba and Au. africanus exclusively are illustrated with arrows. (C) MPT where multi-state characters are treated as ordered, but are inversely weighted in proportion to the number of state transitions possible. (D) Bootstrap 50% majority-rules consensus tree (10K replicates) with states treated as ordered, with inverse transition weighting. (E) MPT where multi-state characters are treated as unordered. (F) Bootstrap 50% majority-rules consensus tree (10K replicates) with states treated as unordered. Abbreviations: TL: tree length, CI: consistency index, RI: retention index, RC: rescaled consistency index, HI: homoplasy index.
126
K. Carter et al. / Journal of Human Evolution 69 (2014) 123e128
(‘large’ Carabelli’s cusp and ‘trace-small cusp’ Protostylid LM1) are actually synapomorphies. The other three (‘weak’ labial curvature UI1, ‘faint’ shovel UI1, and ‘small’ cusp 7 LM1) are reconstructed using maximum parsimony as symplesiomorphies. As the two synapomorphic characters are also given the most weight based on their number of states, a relationship between Au. sediba and Au. africanus would be found even with a large amount of contradictory evidence. We reanalyzed Irish et al.’s (2013) data set (their Table 2, p. 2) using PAUP* v. 4.0b10 (Swofford, 2002), firstly replicating their protocols, then using two alternative character weighting schemes. All analyses reported here can be replicated using the PAUP* script in the Supplementary Online Material (SOM) File 1. Following the methods described in Irish et al.’s (2013) paper, we obtained a MPT with the same topology as their Fig. 1, but one additional step in length (77 vs. 76). Our first alternate weighting scheme treats multistate characters as ordered, but proportionally down-weights them according to their number of possible state transitions (Fig. 2C and D; Table 1). For example, a trait with 6 states has a maximum of 5 transitions when ordered, so is given a weight 0.2 times that of a binary character. The second method treats character states as unordered (Fig. 2E and F), where transitions between any two states are equally weighted. The phylogenetic position of Au. sediba is altered from that in Irish et al.’s (2013) MPT in both cases. Fiftypercent majority-rule consensus trees (plus other groups compatible with these trees) place Au. sediba as the outgroup to all other hominins (Fig. 2D and F). Irish et al. (2013) describe assessing branch support for their tree via a 50% majority-rule consensus tree of 10K bootstrap pseudo-replicate data sets. Strict adherence to the 50% criterion, however, would render all internal nodes in their Fig. 1 as one large polytomy, leaving the ingroup completely unresolved. Instead, Irish et al. (2013) allowed other groups compatible with the consensus tree to be included, facilitating assessment of branch support for individual clades. It is surprising that Homo, commonly accepted to be a monophyletic clade, is supported in only 27% of the trees inferred from resampled data sets (only 17% in our replicated analysis). The position of Au. sediba and Au. africanus as sister taxa to the Homo clade is recovered in only 28% of trees. One conservative approach to the interpretation of a bootstrap proportion (P) for a given clade is to consider 1 d P as the probability of falsely accepting a clade that does not exist (i.e., Type I error; Felsenstein and Kishino, 1993). Such low branch support for Irish et al.’s (2013) MPT may therefore indicate Type I error rates higher than 70% for key clades, markedly greater than the convention of 5% generally used in null hypothesis testing. We contend that the low bootstrap values suggest poor phylogenetic resolvability of this character matrix (Takahashi et al., 2001). Irish et al. (2013) reported a single MPT (Fig. 1, p. 3), but no mention is made whether other trees, not significantly different in length from their MPT, support conflicting evolutionary relationships. Using a two-tailed Wilcoxon signed-rank test, the number of additional steps needed to achieve a tree significantly less parsimonious than the MPT can be quantified (Templeton, 1983). The null hypothesis for this test is that the differences in the fit of characters to the two trees are no greater than expected from sampling error. The set of trees that are not significantly different from the MPT can be regarded as equally parsimonious once sampling error is taken into account (Trueman, 2010). It is therefore statistically arbitrary to favor one tree over another from this set, regardless of whether they differ in length. To calculate the extent of this region of equal parsimony for the Irish et al. (2013) character matrix, we first used an exhaustive algorithm in PAUP* to search the entire tree space and enumerate the complete distribution of tree lengths (Fig. 3). We then compared the MPT to trees of
Figure 3. Tree length distribution from exhaustive search in PAUP*. States are treated as ordered as in Irish et al. (2013). The shaded area below the dashed line indicates the region of the distribution where trees with lengths of 80 or less do not differ significantly from the MPT in the level of character support they enjoy. The MPT tree, with a length of 77, shares the same topology as the tree reported in Irish et al.’s (2013) Figure 1.
progressively longer length and found that none of the trees with length equal to or less than 80 (n ¼ 253) were significantly different from the MPT (Fig. 3). As many of these tree topologies are incongruent (Fig. 4), presenting only one topology is potentially misleading. The polarity of hominin dental non-metric character states can be affected by outgroup choice. As Kimbel (2013) has noted, the use of only one species (Gorilla gorilla) as the outgroup in Irish et al.’s (2013) study is problematic, as species-specific specializations can bias character polarity (Maddison et al., 1984). Given the availability of two taxa more closely related to the hominin ingroup, Pan troglodytes and Pan paniscus, an outgroup consisting of multiple species would mitigate this potential source of bias. The phylogenetic position of Au. sediba is an important question that may be relevant to understanding the emergence of the genus Homo, and dental traits are frequently useful in resolving phylogenetic hypotheses. However, we question the validity of Irish et al.’s (2013) results because of the theoretical and methodological issues present in their analysis. Our aim is not to argue for a particular evolutionary hypothesis, but rather to demonstrate that subtle differences in phylogenetic analysis can cause profound changes in the results. However, even with more robust testing of assumptions and evaluation of the resulting phylogenetic hypothesis, we feel that ASUDAS is an inappropriate metric for evaluating the phylogenetic place of Au. sediba.
K. Carter et al. / Journal of Human Evolution 69 (2014) 123e128
127
Figure 4. Maximum parsimony cladograms with states treated as ordered. (A) MPT with bootstrap branch support values. (BeF) Five additional trees that are one step longer than the MPT. Templeton test p-values reflect pairwise comparisons between the current tree and MPT. For each character, the absolute value of the difference in parsimony length between the MPT and a comparator tree is calculated. These differences are then ranked from smallest to largest, with ties given average ranks, and positive/negative signs are reattached. The test statistic, W, is the sum of these signed ranks. Abbreviations: TL: tree length, CI: consistency index, RI: retention index, RC: rescaled consistency index, HI: homoplasy index.
128
K. Carter et al. / Journal of Human Evolution 69 (2014) 123e128
Acknowledgments We thank David Pilbeam, Shara Bailey, and Joel Irish for discussion of this project. We also thank two anonymous reviewers for suggestions that improved the final paper. Appendix A. Supplementary data Supplementary data related to this article can be found at http:// dx.doi.org/10.1016/j.jhevol.2014.01.003. References Aguirre, L., Castillo, D., Solarte, D., Moreno, F., 2006. Frequency and variability of five non-metric dental crown traits in the primary and permanent dentitions of a racially mixed population from Cali, Colombia. Dent. Anthropol. 19, 39e47. Aiello, L., Dean, C., 1990. An Introduction to Human Evolutionary Anatomy. Elsevier Academic Press, Amsterdam. Bailey, S.E., 2006. Beyond shovel shaped incisors: Neandertal dental morphology in a comparative context. Period. Biol. 108, 253e267. Bailey, S.E., Weaver, T.D., Hublin, J.J., 2009. Who made the Aurignacian and other early Upper Paleolithic industries. J. Hum. Evol. 57, 11e26. Bininda-Emonds, O.R.P., Gittleman, J.L., Purvis, A., 1999. Building large trees by combining phylogenetic information: a complete phylogeny or the extant Carnivora (Mammalia). Biol. Rev. 74, 143e175. Bjarnason, A., Chamberlain, A.T., Lockwood, C.A., 2011. A methodological investigation of hominoid craniodental morphology nd phylogenetics. J. Hum. Evol. 60, 47e57. Dahlberg, A.A., 1951. The dentition of the American Indian. In: Laughlin, W.S. (Ed.), Papers on the Physical Anthropology of the American Indian. Viking Fund, New York, pp. 138e176. Denton, L.C., 2011. Shovel-shaped incisors and the morphology of the enameldentin junction: an analysis of human upper incisors in three dimensions. Ph.D. Dissertation, Colorado State University. Felsenstein, J., Kishino, H., 1993. Is there something wrong with the bootstrap on phylogenies? A reply to Hillis and Bull. Syst. Biol. 42, 193e200. Hanihara, K.H., 1977. Dentition of the Ainu and the Australian aborigines. In: Dahlberg, A.A., Graber, T.M. (Eds.), Orofacial Growth and Development. Walter de Gruyter Press, Berlin, pp. 195e200. Hauser, D.L., Presch, W., 1991. The effect of ordered characters on phylogenetic reconstruction. Cladistics 7, 243e265. Hawkins, J.A., Hughes, C.E., Scotland, R.W., 1997. Primary homology assessment, characters and character states. Cladistics 13, 275e283. Irish, J.D., 2006. Who were the ancient Egyptians? Dental affinities among Neolithic through postdynastic peoples. Am. J. Phys. Anthropol. 129, 529e543. Irish, J.D., Guatelli-Steinberg, D., 2003. Ancient teeth and modern human origins: an expanded comparison of African Plio-Pleistocene and recent world dental samples. J. Hum. Evol. 45, 113e144.
Irish, J.D., Turner, C.G., 1990. West African dental affinity of Late Pleistocene Nubians. II. Peopling of the Eurafrican-South Asian triangle. Homo 41, 42e53. Irish, J.D., Guatelli-Steinberg, D., Legge, S.S., Ruiter, D.J., Berger, L.R., 2013. Dental morphology and the phylogenetic “place” of Australopithecus sediba. Science 340. Kimbel, W., 2013. Palaeoanthropology: Hesitation on hominin history. Nature 497, 573e574. LeBlanc, S.A., Turner, C.G., Morgan, M.E., 2008. Genetic relationships based on discrete dental traits: Basketmaker II and Mimbres. Int. J. Osteoarchaeol. 18, 109e130. Maddison, W.P., Donohue, M.J., Maddison, D.R., 1984. Outgroup analysis and parsimony. Syst. Zool. 33, 83e103. Martinón-Torres, M., De Castro, J.B., Gómez-Robles, A., Arsuaga, J.L., Carbonell, E., Lordkipanidze, D., Manzi, G., Margvelashvili, A., 2007. Dental evidence on the hominin dispersals during the Pleistocene. Proc. Natl. Acad. Sci. 104, 13279e 13282. Nadal-Roberts, M., Collard, M., 2005. Impact of methodological choices on assessments of the reliability of fossil primate phylogenetic hypotheses. Folia Primatol. 76, 207e221. Pereira, C., Telles Antunes, M., Pestana, D., de Mendonça, M.C., Santos, J.C., 2012. An Unidentified Skeletal Assemblage from a Post-1755 Mass Grave of Lisbon: Dental Morphology and Population Affinity. Bull. Int. Assoc. Paleodont. 6, 12e 26. Scott, G.R., Turner II, C.G., 1997. The Anthropology of Modern Human Teeth. Cambridge University Press, Cambridge. Slowinski, J.B., 1993. “Unordered” versus “Ordered” characters. Syst. Biol. 42, 155e 165. Swofford, D.L., 2002. PAUP*: Phylogenetic Analysis Using Parsimony (*and other methods), version 4.0b10. Sinauer, Sunderland, MA. Takahashi, H., Takata, K., Goto, A., 2001. Phylogeography of lateral plate dimorphism in the freshwater type of ninespine sticklebacks, genus Pungitius. Ichthyol. Res. 48, 143e154. Templeton, A.R., 1983. Phylogenetic inference from restriction endonuclease cleavage site maps with particular reference to the evolution of humans and the apes. Evol. 37, 221e244. Trueman, J.W.H., 2010. A new cladistic analysis of Homo floresiensis. J. Hum. Evol. 59, 223e226. Turner II, C.G., Nichol, C.R., Scott, G.R., 1991. Scoring procedures for key morphological traits of the permanent dentition: the Arizona State University Dental Anthropology System. In: Kelley, M.A., Larsen, C.S. (Eds.), Advances in Dental Anthropology. Wiley-Liss Inc, New York, pp. 13e31. Wiens, J.J., 2001. Character analysis in morphological phylogenetics: problems and solutions. Syst. Biol. 50, 689e699. Wood, B.A., 1981. Tooth size and shape and their relevance to studies of hominid evolution. Phil. Trans. R. Soc. B 292, 65e76. Worthington, S., 2012. New Approaches to Late Miocene Hominoid Systematics: Ranking Morphological Characters by Phylogenetic Signal. Ph.D. Dissertation, New York University.
Contents lists available at ScienceDirect
Journal of Human Evolution journal homepage: www.elsevier.com/locate/jhevol
News and views
News and views: Non-metric dental traits and hominin phylogeny Katherine Carter a, *, Steven Worthington b, Tanya M. Smith a a b
Department of Human Evolutionary Biology, Harvard University, 11 Divinity Avenue, Cambridge, MA 02138, USA Institute for Quantitative Social Science, Harvard University, 1737 Cambridge Street, CGIS Knafel Building, Cambridge, MA 02138, USA
a r t i c l e i n f o Article history: Received 8 July 2013 Accepted 10 January 2014 Available online 07 March 2014 Keywords: Hominin evolution ASUDAS Dental anthropology Phylogenetic methods Character weighting
Introduction Analyses of hominin dental remains conventionally include measurements of tooth crown sizes and descriptions of occlusal morphology such as minor accessory cusps, fissure patterns, and ridges (e.g., Wood, 1981; Aiello and Dean, 1990; Bailey, 2006). Following Dahlberg (1951), Turner et al. (1991) developed a formal system for dividing these ‘non-metric’ aspects of dental morphology into discrete categories. This system, termed the Arizona State University Dental Anthropology System (ASUDAS), is an effective tool for discriminating among modern human populations and for assessing inter-population relationships (Scott and Turner, 1997). Other researchers later used ASUDAS to examine the evolutionary relationships of various Pleistocene hominins (Irish and Guatelli-Steinberg, 2003; Martinón-Torres et al., 2007; but see Bailey et al., 2009). Most recently, Irish et al. (2013) used ASUDAS to assess the phylogenetic position of Australopithecus sediba, finding support for both an Au. sediba þ Au. africanus clade and a clade uniting South African australopiths with Homo. However, as Kimbel (2013) has argued, there are theoretical issues with applying ASUDAS to assess phylogenetic relationships from small samples of fossil hominin dental remains. Here we explore the
* Corresponding author. E-mail address: [email protected] (K. Carter). http://dx.doi.org/10.1016/j.jhevol.2014.01.003 0047-2484/Ó 2014 Elsevier Ltd. All rights reserved.
suitability of applying a method developed for partitioning among modern human populations to assess interspecies relationships among fossil hominins. We then discuss the ramifications of different choices made during phylogenetic estimation, including those pertaining to character weighting, clade support, and outgroup composition. We find that slight alteration of phylogenetic assumptions leads to numerous equally possible evolutionary reconstructions for Au. sediba. Human populations vs. hominin species The ASUDAS system allows researchers to assess intraspecific relationships using metrics such as ‘Mean Measure of Divergence’ (MMD; Scott and Turner, 1997), which allows creation of hierarchical clusters of human populations using a distance matrix of overall phenetic similarity (Irish, 2006; LeBlanc et al., 2008). It is also used to assess the affinity of unknown groups with known populations based on suites of characters (Hanihara, 1977; Irish and Turner, 1990; Aguirre et al., 2006; Pereira et al., 2012). For this technique, any individual trait is not diagnostic of a particular population; rather, it is suites of character frequencies, not the presence or absence of a single character, that determine the affinity of a population. Implicit in this method, and indeed in the samples used to create population standards, is the idea that much of the human variability in ASUDAS relates to differences in frequencies, rather than presence or absence of characters states (Scott and Turner, 1997). While Irish et al. (2013) used a geneticallyinformed threshold model to determine cutoff points for trait presence and absence, Scott and Turner (1997) made no claim that each successive change between ASUDAS expression levels was related to meaningful differences in genetic expression, a key factor in dividing characters into states for phylogenetic analysis (Hawkins et al., 1997; Wiens, 2001). While it is well established that dental morphology is under considerable genetic control, the number of character states is not necessarily related to the number of genetic changes needed to produce population variability. For example, while enamel extension UM1 is divided into 3 states of expression and Carabelli’s Cusp UM1 into 7, there is no evidence that there are more than twice the genes involved in Carabelli’s cusp expression. A phenetic method designed to assess human population relationships may or may not use characters relevant for determining
124
K. Carter et al. / Journal of Human Evolution 69 (2014) 123e128
evolutionary relationships among species. For the same reason that traits used to differentiate among breeds of dogs would not be useful to reconstruct the phylogeny of Canidae (Bininda-Emonds et al., 1999), there is no indication that the traits used for within-
species analysis capture the range of variability expressed among species. Some of the human variability that ASUDAS samples is monomorphic among non-Neandertal fossil hominins (e.g., absence of Bushman canine, the presence of a ridge on the mesial surface of the upper canine), and ASUDAS fails to capture variation in hominins known to be phylogenetically informative, such as the degree of P4 molarization, the rotation of the P3, degree of dimorphism in the canines, and relative molar and incisor sizes (Fig. 1A) (Wood, 1981; Aiello and Dean, 1990). Based on the expression of ASUDAS traits across human populations, the most parsimonious interpretation of character state polymorphism is that ASUDAS characters were not fixed in the last common ancestor of all modern humans (Fig. 1B). While there are multiple processes that could produce this pattern, it is likely that fossil hominins had the genetic potential for polymorphism in most of these traits, making estimates of interspecific relationships with small sample sizes problematic. Alternative explanations, such as independent acquisition of traits in different populations of recent hominins (e.g., independent acquisition of shoveling in Northeastern Asians and Neandertals; Denton, 2011) call into question the phylogenetic use of these characters (Fig. 1C). Methodological choices impact phylogenetic analyses Decisions related to character weighting schemes, branch support criteria, and outgroup composition can dramatically affect phylogenetic analysis (Nadal-Roberts and Collard, 2005; Bjarnason et al., 2011; Worthington, 2012). Irish et al. (2013) reported that all characters in their analysis were treated as being of equal weight. However, weighting nevertheless occurs as a consequence of the way character state changes are quantified. For characters with binary states, only a single change between states is possible. For multistate characters, a transition from one state to any other can occur either in a single change (‘unordered’) or only by traveling through all intermediate states (‘ordered’) (Slowinski, 1993). Irish et al. (2013) treated their characters as ordered, and combined with the unequal number of states among their sampled traits (Table 1), this resulted in characters with more states often being given more weight in their analysis. Several of the multistate characters sampled undergo non-sequential transformations (e.g., from state 0 to state 3) on their most parsimonious tree (MPT; the tree with an internal branching order that requires the fewest evolutionary events to explain the tip data). When this occurs, the hypothesis of state order implicitly gives more weight to these characters (Hauser and Presch, 1991: Figure 10, p. 260). When character state transitions are ordered, the parsimony algorithm counts change between any intermediate states as separate evolutionary steps. In other words, twice the weight is given to a two-step change from, for example, states ‘1’ to ‘3’ than to any transition between binary character states. Thus a character such as Protostylid LM1, which unites Au. sediba and Au. africanus with a character state change from ‘1’ (at the base of the clade
Figure 1. Theoretical problems that arise with applying ASUDAS in the fossil record. (A) ASUDAS, a method designed to partition intraspecific variation in Homo sapiens may not be useful for assessing interspecies variation in the fossil record as there are many phylogenetically informative characters that are not sampled (light region) and many characters which are sampled that are not phylogenetically informative (dark region). Only a small subset of phylogenetic variance is captured. Furthermore, because almost all variants of ASUDAS traits are found in all modern human populations, either (B) these traits were not fixed in the last common ancestor of modern humans or (C) these traits evolved independently in different lineages. Either explanation suggests that there are problems with using ASUDAS to estimate evolutionary relationships among fossil hominins.
Table 1 Number of states exhibited by characters in Irish et al.’s (2013) matrix. Number of states 1 2 3 4 6
Number of transitions
Inverse transition weight
Count of characters
Character number
0 1 2 3 5
0 1 0.5 0.333 0.2
1 7 7 6 1
18 3 6 9 11 15 21 22 2 4 5 8 10 17 19 1 7 12 13 14 20 16
Character numbers reflect the order in which characters are listed in Irish et al.’s (2013) Table 2, p. 2.
K. Carter et al. / Journal of Human Evolution 69 (2014) 123e128
grouping these two species with Homo) to ‘4’ (at the node linking these two southern African australopiths), contributes three-times the phylogenetic information of any binary character (Fig. 2A). Carabelli’s cusp UM1 also contributes twice the resolving power of
125
a binary character (Fig. 2B). This form of differential weighting leads to phylogenetic results that are driven primarily by a few characters with many states. Furthermore, of the five characters listed in Irish et al.’s (2013) analysis uniting Au. sediba and Au. africanus only two
Figure 2. Maximum parsimony cladograms illustrating character weighting schemes. (A) Protostylid LM1 and (B) Carabelli’s cusp UM1 optimized onto the MPT with states treated as ordered. States in parentheses are for terminal taxa, while those at internal nodes are maximum parsimony reconstructions. State changes uniting Au. sediba and Au. africanus exclusively are illustrated with arrows. (C) MPT where multi-state characters are treated as ordered, but are inversely weighted in proportion to the number of state transitions possible. (D) Bootstrap 50% majority-rules consensus tree (10K replicates) with states treated as ordered, with inverse transition weighting. (E) MPT where multi-state characters are treated as unordered. (F) Bootstrap 50% majority-rules consensus tree (10K replicates) with states treated as unordered. Abbreviations: TL: tree length, CI: consistency index, RI: retention index, RC: rescaled consistency index, HI: homoplasy index.
126
K. Carter et al. / Journal of Human Evolution 69 (2014) 123e128
(‘large’ Carabelli’s cusp and ‘trace-small cusp’ Protostylid LM1) are actually synapomorphies. The other three (‘weak’ labial curvature UI1, ‘faint’ shovel UI1, and ‘small’ cusp 7 LM1) are reconstructed using maximum parsimony as symplesiomorphies. As the two synapomorphic characters are also given the most weight based on their number of states, a relationship between Au. sediba and Au. africanus would be found even with a large amount of contradictory evidence. We reanalyzed Irish et al.’s (2013) data set (their Table 2, p. 2) using PAUP* v. 4.0b10 (Swofford, 2002), firstly replicating their protocols, then using two alternative character weighting schemes. All analyses reported here can be replicated using the PAUP* script in the Supplementary Online Material (SOM) File 1. Following the methods described in Irish et al.’s (2013) paper, we obtained a MPT with the same topology as their Fig. 1, but one additional step in length (77 vs. 76). Our first alternate weighting scheme treats multistate characters as ordered, but proportionally down-weights them according to their number of possible state transitions (Fig. 2C and D; Table 1). For example, a trait with 6 states has a maximum of 5 transitions when ordered, so is given a weight 0.2 times that of a binary character. The second method treats character states as unordered (Fig. 2E and F), where transitions between any two states are equally weighted. The phylogenetic position of Au. sediba is altered from that in Irish et al.’s (2013) MPT in both cases. Fiftypercent majority-rule consensus trees (plus other groups compatible with these trees) place Au. sediba as the outgroup to all other hominins (Fig. 2D and F). Irish et al. (2013) describe assessing branch support for their tree via a 50% majority-rule consensus tree of 10K bootstrap pseudo-replicate data sets. Strict adherence to the 50% criterion, however, would render all internal nodes in their Fig. 1 as one large polytomy, leaving the ingroup completely unresolved. Instead, Irish et al. (2013) allowed other groups compatible with the consensus tree to be included, facilitating assessment of branch support for individual clades. It is surprising that Homo, commonly accepted to be a monophyletic clade, is supported in only 27% of the trees inferred from resampled data sets (only 17% in our replicated analysis). The position of Au. sediba and Au. africanus as sister taxa to the Homo clade is recovered in only 28% of trees. One conservative approach to the interpretation of a bootstrap proportion (P) for a given clade is to consider 1 d P as the probability of falsely accepting a clade that does not exist (i.e., Type I error; Felsenstein and Kishino, 1993). Such low branch support for Irish et al.’s (2013) MPT may therefore indicate Type I error rates higher than 70% for key clades, markedly greater than the convention of 5% generally used in null hypothesis testing. We contend that the low bootstrap values suggest poor phylogenetic resolvability of this character matrix (Takahashi et al., 2001). Irish et al. (2013) reported a single MPT (Fig. 1, p. 3), but no mention is made whether other trees, not significantly different in length from their MPT, support conflicting evolutionary relationships. Using a two-tailed Wilcoxon signed-rank test, the number of additional steps needed to achieve a tree significantly less parsimonious than the MPT can be quantified (Templeton, 1983). The null hypothesis for this test is that the differences in the fit of characters to the two trees are no greater than expected from sampling error. The set of trees that are not significantly different from the MPT can be regarded as equally parsimonious once sampling error is taken into account (Trueman, 2010). It is therefore statistically arbitrary to favor one tree over another from this set, regardless of whether they differ in length. To calculate the extent of this region of equal parsimony for the Irish et al. (2013) character matrix, we first used an exhaustive algorithm in PAUP* to search the entire tree space and enumerate the complete distribution of tree lengths (Fig. 3). We then compared the MPT to trees of
Figure 3. Tree length distribution from exhaustive search in PAUP*. States are treated as ordered as in Irish et al. (2013). The shaded area below the dashed line indicates the region of the distribution where trees with lengths of 80 or less do not differ significantly from the MPT in the level of character support they enjoy. The MPT tree, with a length of 77, shares the same topology as the tree reported in Irish et al.’s (2013) Figure 1.
progressively longer length and found that none of the trees with length equal to or less than 80 (n ¼ 253) were significantly different from the MPT (Fig. 3). As many of these tree topologies are incongruent (Fig. 4), presenting only one topology is potentially misleading. The polarity of hominin dental non-metric character states can be affected by outgroup choice. As Kimbel (2013) has noted, the use of only one species (Gorilla gorilla) as the outgroup in Irish et al.’s (2013) study is problematic, as species-specific specializations can bias character polarity (Maddison et al., 1984). Given the availability of two taxa more closely related to the hominin ingroup, Pan troglodytes and Pan paniscus, an outgroup consisting of multiple species would mitigate this potential source of bias. The phylogenetic position of Au. sediba is an important question that may be relevant to understanding the emergence of the genus Homo, and dental traits are frequently useful in resolving phylogenetic hypotheses. However, we question the validity of Irish et al.’s (2013) results because of the theoretical and methodological issues present in their analysis. Our aim is not to argue for a particular evolutionary hypothesis, but rather to demonstrate that subtle differences in phylogenetic analysis can cause profound changes in the results. However, even with more robust testing of assumptions and evaluation of the resulting phylogenetic hypothesis, we feel that ASUDAS is an inappropriate metric for evaluating the phylogenetic place of Au. sediba.
K. Carter et al. / Journal of Human Evolution 69 (2014) 123e128
127
Figure 4. Maximum parsimony cladograms with states treated as ordered. (A) MPT with bootstrap branch support values. (BeF) Five additional trees that are one step longer than the MPT. Templeton test p-values reflect pairwise comparisons between the current tree and MPT. For each character, the absolute value of the difference in parsimony length between the MPT and a comparator tree is calculated. These differences are then ranked from smallest to largest, with ties given average ranks, and positive/negative signs are reattached. The test statistic, W, is the sum of these signed ranks. Abbreviations: TL: tree length, CI: consistency index, RI: retention index, RC: rescaled consistency index, HI: homoplasy index.
128
K. Carter et al. / Journal of Human Evolution 69 (2014) 123e128
Acknowledgments We thank David Pilbeam, Shara Bailey, and Joel Irish for discussion of this project. We also thank two anonymous reviewers for suggestions that improved the final paper. Appendix A. Supplementary data Supplementary data related to this article can be found at http:// dx.doi.org/10.1016/j.jhevol.2014.01.003. References Aguirre, L., Castillo, D., Solarte, D., Moreno, F., 2006. Frequency and variability of five non-metric dental crown traits in the primary and permanent dentitions of a racially mixed population from Cali, Colombia. Dent. Anthropol. 19, 39e47. Aiello, L., Dean, C., 1990. An Introduction to Human Evolutionary Anatomy. Elsevier Academic Press, Amsterdam. Bailey, S.E., 2006. Beyond shovel shaped incisors: Neandertal dental morphology in a comparative context. Period. Biol. 108, 253e267. Bailey, S.E., Weaver, T.D., Hublin, J.J., 2009. Who made the Aurignacian and other early Upper Paleolithic industries. J. Hum. Evol. 57, 11e26. Bininda-Emonds, O.R.P., Gittleman, J.L., Purvis, A., 1999. Building large trees by combining phylogenetic information: a complete phylogeny or the extant Carnivora (Mammalia). Biol. Rev. 74, 143e175. Bjarnason, A., Chamberlain, A.T., Lockwood, C.A., 2011. A methodological investigation of hominoid craniodental morphology nd phylogenetics. J. Hum. Evol. 60, 47e57. Dahlberg, A.A., 1951. The dentition of the American Indian. In: Laughlin, W.S. (Ed.), Papers on the Physical Anthropology of the American Indian. Viking Fund, New York, pp. 138e176. Denton, L.C., 2011. Shovel-shaped incisors and the morphology of the enameldentin junction: an analysis of human upper incisors in three dimensions. Ph.D. Dissertation, Colorado State University. Felsenstein, J., Kishino, H., 1993. Is there something wrong with the bootstrap on phylogenies? A reply to Hillis and Bull. Syst. Biol. 42, 193e200. Hanihara, K.H., 1977. Dentition of the Ainu and the Australian aborigines. In: Dahlberg, A.A., Graber, T.M. (Eds.), Orofacial Growth and Development. Walter de Gruyter Press, Berlin, pp. 195e200. Hauser, D.L., Presch, W., 1991. The effect of ordered characters on phylogenetic reconstruction. Cladistics 7, 243e265. Hawkins, J.A., Hughes, C.E., Scotland, R.W., 1997. Primary homology assessment, characters and character states. Cladistics 13, 275e283. Irish, J.D., 2006. Who were the ancient Egyptians? Dental affinities among Neolithic through postdynastic peoples. Am. J. Phys. Anthropol. 129, 529e543. Irish, J.D., Guatelli-Steinberg, D., 2003. Ancient teeth and modern human origins: an expanded comparison of African Plio-Pleistocene and recent world dental samples. J. Hum. Evol. 45, 113e144.
Irish, J.D., Turner, C.G., 1990. West African dental affinity of Late Pleistocene Nubians. II. Peopling of the Eurafrican-South Asian triangle. Homo 41, 42e53. Irish, J.D., Guatelli-Steinberg, D., Legge, S.S., Ruiter, D.J., Berger, L.R., 2013. Dental morphology and the phylogenetic “place” of Australopithecus sediba. Science 340. Kimbel, W., 2013. Palaeoanthropology: Hesitation on hominin history. Nature 497, 573e574. LeBlanc, S.A., Turner, C.G., Morgan, M.E., 2008. Genetic relationships based on discrete dental traits: Basketmaker II and Mimbres. Int. J. Osteoarchaeol. 18, 109e130. Maddison, W.P., Donohue, M.J., Maddison, D.R., 1984. Outgroup analysis and parsimony. Syst. Zool. 33, 83e103. Martinón-Torres, M., De Castro, J.B., Gómez-Robles, A., Arsuaga, J.L., Carbonell, E., Lordkipanidze, D., Manzi, G., Margvelashvili, A., 2007. Dental evidence on the hominin dispersals during the Pleistocene. Proc. Natl. Acad. Sci. 104, 13279e 13282. Nadal-Roberts, M., Collard, M., 2005. Impact of methodological choices on assessments of the reliability of fossil primate phylogenetic hypotheses. Folia Primatol. 76, 207e221. Pereira, C., Telles Antunes, M., Pestana, D., de Mendonça, M.C., Santos, J.C., 2012. An Unidentified Skeletal Assemblage from a Post-1755 Mass Grave of Lisbon: Dental Morphology and Population Affinity. Bull. Int. Assoc. Paleodont. 6, 12e 26. Scott, G.R., Turner II, C.G., 1997. The Anthropology of Modern Human Teeth. Cambridge University Press, Cambridge. Slowinski, J.B., 1993. “Unordered” versus “Ordered” characters. Syst. Biol. 42, 155e 165. Swofford, D.L., 2002. PAUP*: Phylogenetic Analysis Using Parsimony (*and other methods), version 4.0b10. Sinauer, Sunderland, MA. Takahashi, H., Takata, K., Goto, A., 2001. Phylogeography of lateral plate dimorphism in the freshwater type of ninespine sticklebacks, genus Pungitius. Ichthyol. Res. 48, 143e154. Templeton, A.R., 1983. Phylogenetic inference from restriction endonuclease cleavage site maps with particular reference to the evolution of humans and the apes. Evol. 37, 221e244. Trueman, J.W.H., 2010. A new cladistic analysis of Homo floresiensis. J. Hum. Evol. 59, 223e226. Turner II, C.G., Nichol, C.R., Scott, G.R., 1991. Scoring procedures for key morphological traits of the permanent dentition: the Arizona State University Dental Anthropology System. In: Kelley, M.A., Larsen, C.S. (Eds.), Advances in Dental Anthropology. Wiley-Liss Inc, New York, pp. 13e31. Wiens, J.J., 2001. Character analysis in morphological phylogenetics: problems and solutions. Syst. Biol. 50, 689e699. Wood, B.A., 1981. Tooth size and shape and their relevance to studies of hominid evolution. Phil. Trans. R. Soc. B 292, 65e76. Worthington, S., 2012. New Approaches to Late Miocene Hominoid Systematics: Ranking Morphological Characters by Phylogenetic Signal. Ph.D. Dissertation, New York University.
