Introduction

A phylogenetic character is the fundamental unit of cladistic analysis (Freudenstein 2005; Yu et al. 2021). Nelson and Platnick (1981: p. 301) argued that a character is "a theory that two attributes which appear different in some way are nonetheless the same (homologous)." Mickevich and Weller (1990: p. 138) defined a character as “a set of nested synapomorphies of character states” based on their interpretation of Nelson and Platnick’s (1981) expanded discussion. Pimental and Riggins (1987: p. 201) regarded a character as "a feature of organisms that can be evaluated as a variable with two or more mutually exclusive and ordered states." However, many definitions of 'character' are less restrictive and do not require ordered character states. For example, Maddison and Maddison (1992: p. 45) defined a character as "a set of alternative conditions, called character states, that are considered able to evolve one to another." Sereno (2007: p. 570) reviewed existing definitions and concluded that a character is "a heritable, organismal feature (i.e., an observable condition) expressed as an independent variable." If we allow for definitions such as those of Maddison and Maddison (1992) and Sereno (2007), then individual characters can be treated as unordered or ordered. If a character is treated as unordered, each character state can transform to any other character state in a single step, i.e., state 0 can transition to state 1 or state 2 in one step (Slowinski 1993). If a character is treated as ordered, there is a hypothesized relationship between each of the synapomorphies defined within the character (Mabee 1989; Hauser and Presch 1991) and multiple steps will be incurred for some transformations, i.e., state 0 > 2 would cost two steps with linear ordering (Slowinski 1993). If a character is binary (two character states), the ordering of states is irrelevant and the two character states will always be neighbors (i.e., adjacent states that are directly connected to each other). When a character is comprised of three or more character states, it is a multistate character and the question of character ordering needs to be addressed (Mabee 1989). In either case, a model of evolution is assumed, i.e., leaving characters as unordered does not imply that there is no model of evolution (Hauser and Presch 1991). In treating characters as ordered or unordered, the researcher is positing a hypothesis about how the character states evolved (Mabee 1989; Mickevich and Weller 1990; Rineau et al. 2018).

Mabee (1989) recognized three types of ordered characters: branching, reticulate, and linear. The first two types allow for partial ordering of character states whereas linear ordering is total or complete. While there are many ways to order multistate morphological characters, one common approach is to minimize the amount of change between states, e.g., large ↔ small ↔ very small (Mickevich and Weller 1990; Slowinski 1993). This is achieved by examining the morphological similarity that exists between the character states, and the resulting tree that connects these character states together is termed a character state tree, morphocline, or transformation series (Maslin 1952; Hennig 1966; Farris et al. 1970; Mabee 1989; Pogue and Mickevich 1990; Hauser and Presch 1991). While the morphocline approach is commonly used, it is not universally accepted. Lipscomb (1992) argued that the morphocline criterion does not consider congruence with other morphological characters, one of Patterson’s (1982) homology criteria, and thus will not differentiate between homoplastic and homologous similarity. Mabee (1989) discussed concerns that the morphocline criterion will treat all evolutionary change as gradual (regardless of if the character evolved gradually) because intermediate character states will separate the two most extreme states. Thus, characters ordered using the morphocline criterion may not accurately represent evolutionary events (Pogue and Mickevich 1990). Despite these concerns, the morphocline criterion is often considered an acceptable way to order multistate morphological characters (Rineau et al. 2015, 2018). Morphocline theory has been supported by studies of embryology (e.g. Mickevich and Weller 1990), and also from paleontology where evidence from the fossil record provides many examples of gradual evolutionary change (e.g. Simpson 1949; Dzik and Trammer 1980; Murphy et al. 1981; Rose and Bown 1984; Bown and Rose 1987; Dzik 1989; Olempska 1989; Hallam 1997; Sulej and Majer 2005; Niechwedowicz and Trammer 2007; Silcox et al. 2008; Scannella et al. 2014; Monnet et al. 2015; Czepiński 2020; O’Leary 2021).

There has been significant debate over whether multistate morphological characters should be ordered or unordered when used to build phylogenies (Hauser and Presch 1991; Wilkinson 1992; Slowinski 1993; Wiens 2001; Grand et al. 2013; Rineau et al. 2015, 2018; Brocklehurst and Haridy 2021). Some authors have reported that the use of ordered multistate characters had little to no effect on the results of a phylogenetic analysis (Reeder et al. 2015; Baron et al. 2017; Lee and Yates 2018). Hauser and Presch (1991) examined 27 data sets and concluded that the practice of ordering multistate characters is unnecessary if the criterion for evaluating character evolution is congruence with other characters. By contrast, many studies support the ordering of multistate characters, especially when these characters form a morphocline (Slowinski 1993; Wiens 2001; Grand et al. 2013; Rineau et al. 2015, 2018). Slowinski (1993) examined 21 empirical matrices and found that phylogenetic resolution was typically greater when the ordered multistate characters were treated as ordered rather than unordered. Subsequent analyses of empirical data matrices for dissorophoid temnospondyls (Fröbisch and Schoch 2009), squamates (Gauthier et al. 2012; Simões et al. 2017), and tetrapods (McGowan 2002; Marjanović and Laurin, 2019) also showcase the effects of ordering a subset of multistate characters including increased or decreased phylogenetic resolution and altered phylogenetic hypotheses. Grand et al. (2013) and Rineau et al. (2015, 2018) performed analyses on simulated data sets and found that the use of ordered multistate characters increased phylogenetic resolution and decreased the support for artefactual clades. Grand et al. (2013) emphasized that their results only apply to multistate characters that form a morphocline. Thus, a reasonable approach is to treat multistate characters as ordered when the states form a hypothesized morphocline, but to leave a multistate character as unordered when such a relationship between the character states cannot be determined (Slowinski 1993; Wiens 2001; Grand et al. 2013).

It is also important to assess the general performance of morphology in cladistic analyses. Among placental mammals, analyses of morphological and molecular data support numerous clades that are incompatible with each other. Phylogenetic analyses of molecular data strongly support four superorders of placental mammals: Euarchontoglires, Laurasiatheria, Xenarthra, and Afrotheria (Madsen et al. 2001; Murphy et al. 2001a, b; Springer et al. 2003, 2005; Meredith et al. 2011; Foley et al. 2016, 2023; Tarver et al. 2016; Esselstyn et al. 2017; Upham et al. 2019). Of these, only Xenarthra has been supported by morphological data alone (Springer et al. 2007; O’Leary et al. 2013). A caveat is that Beck and Ballie (2018) recovered Euarchontoglires and Laurasiatheria in a morphology-only analysis, but they did so only after including “predicted ancestors” that were generated in a previous analysis with a molecular scaffold that constrained the monophyly of these clades. Differences between results of morphological and molecular analyses have raised concerns regarding the suitability of morphological data for accurately reconstructing higher-level relationships of extinct mammalian taxa (Springer et al. 2007, 2008, 2019).

Several previous studies have used ‘pseudoextinction analyses’ to assess the performance of morphological phylogenetics. This is accomplished by scoring soft morphological characters and/or any molecular data as missing for a subset of extant taxa (called pseudoextinct taxa) prior to analysis (Asher and Hofreiter 2006; Springer et al. 2007, 2019; Sansom and Wills 2013; Pattinson et al. 2014; Brady and Springer 2021). Two of these studies performed pseudoextinction analyses using the large matrix from O’Leary et al. (2013) and both found that morphological data were unable to accurately place most placental orders in clades that are well-supported by molecular studies when these orders were treated as pseudoextinct (Springer et al. 2019; Brady and Springer 2021). It remains to be determined if character ordering will influence pseudoextinction analyses with a large data set like that of O’Leary et al (2013).

While cladistic analyses of morphology-only data sets have failed to recover most of the higher-level clades of placental mammals that are well-supported by molecular data, it is important to note that morphological data perform more accurately at and below the ordinal level. For example, Asher and Hofreiter (2006) performed pseudoextinction analyses within the family Tenrecidae and found that morphology and molecular data reconstructed largely compatible phylogenies for tenrecs. Pattinson et al. (2014) performed similar analyses for taxa within Primates and concluded that morphological data sets are reliable for reconstructing relationships within this order. Additionally, pseudoextinction analyses with data sets for Placentalia have consistently recovered most orders with multi-taxon representation as monophyletic (Springer et al. 2007, 2019; Brady and Springer, 2021). Indeed, the only consistent exceptions are Afrosoricida, Eulipotyphla, and Cetartiodactyla.

It is also important to recognize that morphological data sets are not homogenous. The results from one data set are not necessarily applicable to other data sets. This is because different data sets are usually comprised of different characters (e.g., cranial versus postcranial) and different sets of taxa. Asher and Smith (2022) examined six morphological data sets comprised of characters from bones and teeth and concluded that these data sets exhibit different degrees of congruence with well-supported molecular phylogenies. Also, congruence can be sensitive to minor changes in taxon and character sampling within a given morphological data set.

In this study, we used Velazco et al.’s (2022) data set as a case study to further explore the consequences of using ordered versus unordered morphological characters in phylogenetic analyses. This data set is an expanded version of O’Leary et al.’s (2013) data set and is the largest morphological data set (in terms of number of characters) currently available for placental mammals. O’Leary et al.’s (2013) data set has been employed by many research groups (e.g. Carrillo and Asher 2017; Beck and Baillie, 2018; Springer et al. 2019; Brady and Springer 2021), and it is expected that future projects will continue to build upon this data set by adding more taxa. For these reasons, Velazco et al.’s (2022) data set is appropriate to employ in a case study to assess the impact of ordering morphological characters on cladistic analysis. We were also interested in exploring phylogenetic support for Tamirtheria, a newly proposed sister group to Placentalia that includes Ukhaatherium nessovi and all species more closely related to U. nessovi than to Elephas maximus (Asian elephant) (Velazco et al. 2022). Representatives of Tamirtheria in Velazco et al.’s (2022) analyses included four leptictids (Gypsonictops, Leptictis, Palaeictops, Prodiacodon), one cimolestid (Maelestes), two asioryctitherians (Ukhaatherium, Kennalestes), one zalambdalestid (Zalambdalestes), and the mid-Cretaceous eutherian Prokennalestes. To accomplish our goals, we reviewed each character description in Velazco et al.’s (2022) data set to determine which multistate characters should be treated as ordered versus unordered. We then examined how ordering these characters affected reconstructions of eutherian mammal phylogeny both with and without the use of a molecular scaffold. Finally, we examined the impact of using ordered characters in pseudoextinction analyses that sequentially treated 19 extant placental orders as pseudoextinct.

Methods

Nomenclature

We use the names Afrosoricida, Cetartiodactyla, and Eulipotyphla throughout this paper for three of the extant orders of placental mammals. These names are widely used in the literature (e.g., Meredith et al. 2011; Foley et al. 2016, 2023). At the same time, we note that Asher and Helgen (2010, 2011) prefer different names for these orders. To avoid confusion, the equivalencies are as follows: Afrosoricida = Tenrecoidea of Asher and Helgen (2010, 2011); Cetartiodactyla = Artiodactyla of Asher and Helgen (2010); Eulipotyphla = Lipotyphla of Asher and Helgen (2010).

Morphological data set

Morphological data for mammalian species were obtained from MorphoBank project 3871 (Velazco et al. 2022). Velazco et al.’s (2022) data set expands upon O’Leary et al.’s (2013) data set by adding six additional fossil taxa to the morphological matrix. Velazco et al.’s (2022) data set contains 4,541 unordered morphological characters (3,684 of which are parsimony informative) that were scored for 92 taxa. The matrix includes two extant monotremes, two extant marsupials, 42 extant placentals, and 46 taxa that are extinct. We modified Velazco et al.’s (2022) morphological matrix to incorporate updated postcranial character states for the notoungulate Thomashuxleya. Specifically, Beck and Baillie (2018) modified the character states for Thomashuxleya in O'Leary et al.'s (2013) matrix by merging these character states with new character information that was provided by Carrillo and Asher (2017) based on their description of an unusually complete specimen of Thomashuxleya. We also made changes to the scoring of character 128 for ten taxa (see below). To accommodate the reuse of the molecular scaffold of Brady and Springer (2021), the six fossil taxa added to the matrix by Velazco et al. (2022) were moved to the bottom of the matrix.

Morphological character ordering

We used Wilkinson’s (1992) method of intermediates to evaluate all multistate characters in Velazco et al.’s (2022) matrix. All multistate characters that were hypothesized to lie on a morphocline (linear, branching, reticulate, closed pathway), or that were meristic in nature, were treated as ordered. Our evaluation of multistate characters in Velazco et al.’s (2022) matrix identified 796 ordered multistate characters (~ 17.5% of the total data set). Linear, branching, and reticulate multistate characters have been discussed by previous authors (e.g., Mabee 1989), but we are unaware of formal descriptions of closed pathway characters, which have some resemblance to reticulate multistate characters (Mabee 1989: p. 153). In our usage, these are characters with character states that lie on a closed pathway rather than on a line with endpoints. For example, character 2205 pertains to the position of the metastylid relative to the metaconid on the second lower molar (m2) and the individual character states are arranged in a circle around the metaconid (Fig. 1a). In the case of Equus caballus (horse), the metastylid is distal to the metaconid (state 3). Other possible character states for this character are metastylid mesiolingual to the metaconid (state 0), metastylid buccal to the metaconid (state 1), and metastylid distolingual to the metaconid (state 2). For a closed pathway character with four character states, each of the four states is equally close to two other character states and further away from the final character state.

Fig. 1
figure 1

Example of a closed pathway character (character number 2205) from Velazco et al.'s (2022) matrix. Character 2205 is the position of the metastylid relative to the metaconid on m2. The four character states for this character are metastylid mesiolingual to the metaconid (state 0), metastylid buccal to the metaconid (state 1), metastylid distolingual to the metaconid (state 2), and metastylid distal to the metaconid (state 3). a. The m2 of Equus caballus (horse, specimen UW.GZ2010 from the Zoological Collection of the University of Warsaw), which exhibits character state 3. Other possible locations for the metastylid relative to the metaconid are depicted by arrows 0, 1, and 2. b. Sankoff stepmatrix for character 2205. This stepmatrix shows the number of steps that are required to get from character state i (rows) to character state j (columns)

Full size image

Most multistate characters in Velazco et al.’s (2022) matrix were coded by these authors so that character states with an intermediate morphocline value (e.g., small) also had an intermediate numerical value (1) between the end member character states for this character (e.g., large [0], small [1], very small [2]). For linear multistate characters that were not coded in this manner (e.g., small [0], large [1], very small [2]), as well as for branching, reticulate, and closed pathway characters, we employed stepmatrices to enforce character ordering (Fig. 1b). Of the 796 ordered characters, 99 required a stepmatrix. Eighty-seven stepmatrix characters are linear (e.g., 1↔0↔2), eight are closed pathway (e.g., 0↔2↔1↔3↔0), one is branching (e.g., 0↔1↔2 or 3), and four are reticulate (e.g., 0↔1↔2 and 1↔3 and 2↔3). All ordered multistate and stepmatrix characters are referred to as ordered below.

Alterations to the morphological matrix

While working with the matrix, errors were noted in the original scoring of character 128: infraorbital foramen, horizontal position relative to maxillary toothrow. Several taxa were scored as having the infraorbital foramen above the third upper premolar (P3). The original O’Leary et al. (2013) matrix recognizes an alternate hypothesis of homology in which P3 has been lost in many eutherian groups (as suggested by McKenna 1975). Thus, premolars that were historically called P3 or P4 in eutherians were treated as P4 and P5, respectively, in this matrix. Many taxa that lack P3 were scored as having their infraorbital foramen above P3 (due to confusing P4 with P3) for character 128. As a result, we rescored character 128 for all taxa scored as “having an infraorbital foramen above P3” even though they were also scored as lacking this tooth (character 1745). We also reviewed images, published on MorphoBank, for all taxa scored as having their infraorbital foramen above P4 to make sure their scored character states for character 128 are consistent with the hypothesis of P3 being absent in most eutherian groups (to check if the infraorbital foramen should be scored as above P4 or above P5). In total we changed the scorings for character 128 in 10 taxa. These changes are summarized in Table 1. Our matrix, including all modifications to character 128, is available in figshare (https://doi.org/10.6084/m9.figshare.21711848).

Table 1 Summary of changes to the scorings of character 128 for 10 taxa for in Velazco et al.’s (2022) matrix. The taxa are presented here in the same order as they appear in the data set
Full size table

Molecular scaffolds

We used the same molecular scaffold as Brady and Springer (2021), which includes clades that were well supported by Meredith et al. (2011). These clades are also well-supported by other molecular phylogenetic analyses (dos Reis et al. 2012; Emerling et al. 2015; Foley et al. 2016; Tarver et al. 2016; Esselstyn et al. 2017). This scaffold includes all 46 extant taxa in Velazco et al.’s (2022) matrix and is fully bifurcating with the following exceptions: 1. (Xenarthra, Afrotheria, Boreoeutheria); 2. (Hyracoidea, Proboscidea, Sirenia); 3. ((Pholidota + Carnivora), Perissodactyla, Cetartiodactyla, Chiroptera); 4. ((Dermoptera + Primates), Scandentia, Glires); 5. ((Saccopteryx + Nycteris), Myotis, Pteronotus); and 6. (Talpa, Erinaceus, Sorex). These polytomies were retained because these nodes were resolved differently in DNA versus amino acid analyses or had low bootstrap support (< 90%) or Bayesian posterior probabilities (< 0.95) (Meredith et al. 2011). This molecular scaffold (ScaffE) was employed as a backbone constraint for extant taxa only.

Additionally, we employed an expanded molecular scaffold (ScaffEF) that included two extinct South American ungulates from Velazco et al.’s (2022) matrix, Protolipterna ellipsodontoides (Order Litopterna) and Thomashuxleya externa (Order Notoungulata). Molecular data are not available for either of these species and are unlikely to become available due to their ages (early and middle Eocene, respectively; Carrillo and Asher 2017; Zanesco et al. 2019). However, molecular sequences have been obtained from other representatives of these orders: Macrauchenia (Order Litopterna) and Toxodon (Order Notoungulata) (Buckley 2015; Welker et al. 2015). Phylogenetic analyses with these sequences suggest that both Litopterna and Notoungulata belong to the superordinal clade Panperissodactyla (Perissodactyla + (Litopterna + Notoungulata)). Specifically, this clade is supported by phylogenetic analyses of collagen sequences obtained for Macrauchenia and Toxodon (Buckley 2015; Welker et al. 2015). In addition, Litopterna is supported as the sister-taxon to Perissodactyla based on a complete mitochondrial genome obtained from Macrauchenia via ancient DNA methods (Westbury et al. 2017). Mitochondrial genome data are not available for Toxodon. The Velazco et al. (2022) matrix does not include Macrauchenia or Toxodon, but if Litopterna and Notoungulata are monophyletic (see below) then Macrauchenia and Toxodon can be used to constrain the phylogenetic placement of P. ellipsodontoides and T. externa, respectively. In the absence of molecular data, there is no guarantee that Litopterna and Notoungulata are monophyletic, but morphology has done much better in reconstructing the monophyly of placental orders than in reconstructing higher-level relationships among these orders (Springer et al. 2019; Brady and Springer 2021).

Macrauchenia is a well-established member of the order Litopterna. Ameghino (1889) included Macrauchenia within this group. Since then, most authors have upheld this taxonomic classification (Simpson 1934, 1948; Scott 1937; Cifelli 1983; McKenna and Bell 1997; Schmidt and Ferrero 2014; McGrath et al. 2018). The affinities of Protolipterna are less clear, but it appears to be a member of Litopterna (Cifelli 1983; McKenna and Bell 1997) or in a paraphyletic didolodontid group that may be ancestral to litopterns (Chimento and Agnolin 2020).

Thomashuxleya and Toxodon are well established members of the order Notoungulata. Simpson (1934) classified the families Isotemnidae (containing Thomashuxleya) and Toxodontidae (containing Toxodon) in Notoungulata, albeit in different suborders. Subsequent phylogenetic analyses using morphology have placed Thomashuxleya and Toxodon within the notoungulate suborder Toxodontia (Billet 2010, 2011; Deraco and García-López 2016). Our Panperissodactyla scaffold (Equus + (Protolipterna + Thomashuxleya)) allowed us to explore whether or not the shift in position of Litopterna and Notoungulata impacted other phylogenetic relationships among placental mammals.

Phylogenetic reconstructions

Finding the shortest tree

Phylogenetic analyses were performed in PAUP* 4.0a (build 167–169) (Swofford 2003) using heuristic searches with TBR branch swapping. Morganucodon oehleri and M. watsoni (early mammaliaforms) were designated as the outgroup taxa. We used maximum parsimony with stepwise addition and 10,000 random taxon addition sequences to search for the most-parsimonious tree(s). Different searches were performed without a molecular scaffold (NoScaff), with a molecular scaffold that only included extant species (ScaffE), and with a molecular scaffold that included extant species plus two fossil species of South American ungulates (Protolipterna ellipsodontoides, Thomashuxleya externa) (ScaffEF). We also performed searches with alternate versions of our data set that treated all multistate characters as unordered versus a mix of unordered and ordered multistate characters. For simplicity, these analyses are subsequently referred to as unordered and ordered. In all, we performed six analyses using different combinations of scaffold type (NoScaff, ScaffE, ScaffEF) and treatment of multistate characters (ordered, unordered): (1) NoScaff + ordered; (2) NoScaff + unordered; (3) ScaffE + ordered; (4) ScaffE + unordered; (5) ScaffEF + ordered; and (6) ScaffEF + unordered. Searches that resulted in more than one most-parsimonious tree (MPT) were summarized with strict consensus trees. All MPTs in Newick format (unrooted) are available in Figshare (https://doi.org/10.6084/m9.figshare.21713102).

Note: We considered using TNT (Goloboff et al. 2008; Goloboff and Catalano 2016) instead of PAUP* to perform our analyses and tested the performance of each program with TNT and NEXUS versions of Velazco et al.’s (2022) matrix that were downloaded from MorphoBank. Traditional heuristic searches with PAUP* (TBR branch swapping, 1000 randomized taxon input orders) recovered four trees at 28,389 steps. We also analyzed the TNT version of the matrix with the non-traditional search algorithms that are available with TNT and obtained the same four trees of length 28,389 steps. Thus, PAUP and TNT perform equally well with the Velazco et al. (2022) matrix. Given these results, and also because PAUP* and TNT performed equally well on other data sets with comparable taxon sampling (Goloboff et al. 2022), we decided to use PAUP* in our analyses and further note that searches for the best tree(s) become even easier when constraints on relationships are employed with a molecular scaffold.

Bootstrap analyses

Bootstrap analyses were performed in PAUP 4.0a (Swofford 2003) with maximum parsimomy and 500 pseudoreplicates. We employed heuristic searches with TBR branch swapping and stepwise addition (1,000 random taxon addition sequences) for each pseudoreplicate, each of which resampled 4,541 characters (100% of the original sample size). We used the same six combinations of molecular scaffolds and unordered/ordered characters that were used in analyses to find the most parsimonious trees. The results of each bootstrap analysis were summarized with a 50% majority-rule consensus tree, and we retained groups that were compatible with the 50% majority-rule consensus. We also saved all bipartitions with bootstrap support percentages that were ≥ 5%. Bootstrap support percentages for each analysis were mapped onto the single MPT, or strict consensus tree when there were multiple MPTs.

Analyses with 796 characters

The set of 796 ordered characters comprises a relatively small percentage (~ 17.5%) of the total data set. To determine if alternate treatments of the 796 characters by themselves impacted the resulting tree(s) including the recovery of well-supported molecular clades that were incorporated into our molecular scaffolds, we performed heuristic searches (TBR with 10,000 randomized taxon input orders) with just the subset of 796 characters. These analyses were performed with 46 extant taxa.

Pseudoextinction analyses

Velazco et al.’s (2022) phenomic matrix includes osteological (1–3,660) and soft tissue/behavioral characters (3,661–4,541). For pseudoextinct taxa, non-osteological characters were scored as missing (?) since these characters would not typically be preserved in the fossil record. We treated each of 19 placental orders as pseudoextinct. Matrix alterations were performed in Mesquite (Maddison and Maddison 2018). All pseudoextinction matrices are available in Figshare (https://doi.org/10.6084/m9.figshare.21711842). Our molecular scaffold backbone constraint for extant taxa (ScaffE) was altered for each pseudoextinction analysis so that the pseudoextinct taxon was not included in the constraint.

Each pseudoextinct matrix was analyzed in PAUP 4.0a (Swofford 2003). Pseudoextinction analyses were performed with and without fossil taxa, and with both ordered and unordered characters. We used maximum parsimony with stepwise addition, 10,000 random taxon addition sequences, and TBR branch swapping to search for the shortest tree for each pseudoextinct matrix. All unrooted MPTs were exported in NEXUS format. Additionally, analyses were run in which no taxa were treated as pseudoextinct. All of the MPTs from the pseudoextinction analyses are provided in Newick format in Figshare (https://doi.org/10.6084/m9.figshare.21712850).

Tree comparisons

NEXUS trees from pseudoextinction analyses (extant taxa only) were imported back into PAUP 4.0a and compared to the tree obtained when no taxa were treated as pseudoextinct using Robinson-Foulds (RF) distances (Robinson and Foulds 1981) and normalized RF distances. The trees obtained using the ordered subset of characters were also compared to the trees obtained using all unordered characters. RF distances measure the number of internal branches that are unique to one of the two trees. An RF distance can be defined as X + Y, where X is the number of bipartitions on the first tree but not the second tree and Y is the number of bipartitions on the second tree but not the first tree. An RF distance of zero indicates that two trees are identical; higher RF distances indicate that the trees have more unique internal branches, i.e., the trees are less similar. RF distances were calculated with unrooted, fully-bifurcating trees for the same taxa. We also calculated normalized RF (nRF) distances, which were obtained by dividing RF distances by the maximum possible RF distance. Normalized RF distances can range from 0 (two trees are identical) to 1 (all internodes between two trees are in conflict). Consensus trees with polytomies were not used to calculate RF or nRF distances.

Results

Phylogenetic reconstructions

The most parsimonious trees (MPTs) based on six different analyses with Velazco et al.’s (2022) morphological data set were obtained, and a summary of the results for these analyses is provided in Table 2. Figures 2, 3, 4, 5, 6, and 7 show the single MPT or strict consensus of multiple MPTs for each of the six analyses. Bootstrap support percentages are also mapped onto these trees. The NoScaff + ordered analysis recovered a single MPT (Fig. 2) whereas the other five analyses (Figs. 3, 4, 5, 6, and 7) recovered ≥ 2 MPTs (Table 2). The ordered analysis with ScaffE (Fig. 4) produced the largest number of MPTs (13). Analyses that included two extinct species in the molecular scaffold resulted in fewer MPTs relative to analyses that only included extant taxa in the scaffold: 2 versus 13 MPTs for ordered analyses; 4 versus 6 MPTs for unordered analyses (Table 2).

Table 2 Summary of parsimony results for six different analyses. The four placental superorders are Xenarthra, Afrotheria, Laurasiatheria, and Euarchontoglires. Abbreviations: MPT, most parsimonious tree; NA, not applicable; NoScaff, no molecular scaffold; ScaffE, molecular scaffold for extant taxa only; ScaffEF, molecular scaffold for extant species and two fossil species
Full size table
Fig. 2
figure 2

Single most-parsimonious tree (30,345 steps) for the analysis of the ordered data set without a molecular scaffold (NoScaff). Bootstrap support percentages are 100% unless otherwise indicated

Full size image
Fig. 3
figure 3

Strict consensus of four most parsimonious trees (28,380 steps) for the analysis of the unordered data set without a molecular scaffold (NoScaff). Bootstrap support percentages are 100% unless otherwise indicated

Full size image
Fig. 4
figure 4

Strict consensus of 13 most parsimonious trees (31,046 steps) for the analysis of the ordered data set with a molecular scaffold for all extant taxa (ScaffE). ScaffE enforced the monophyly of four placental superorders (Xenarthra, Afrotheria, Laurasiatheria, Euarchontoglires) and most relationships within these superorders (see main text). Bootstrap support percentages are 100% unless otherwise indicated

Full size image
Fig. 5
figure 5

Strict consensus of six most parsimonious trees (29,002 steps) for the analysis of the unordered data set with a molecular scaffold for all extant taxa (ScaffE). ScaffE enforced the monophyly of four placental superorders (Xenarthra, Afrotheria, Laurasiatheria, Euarchontoglires) and most relationships within these superorders (see main text). Bootstrap support percentages are 100% unless otherwise indicated

Full size image
Fig. 6
figure 6

Strict consensus of two most parsimonious trees (31,126 steps) for the analysis of the ordered data set with a molecular scaffold for all extant taxa and two extinct taxa (ScaffEF). ScaffEF enforced the monophyly of four placental superorders (Xenarthra, Afrotheria, Laurasiatheria, Euarchontoglires) and most relationships within these superorders including the positions of the extinct taxa Thomashuxleya externa and Protolipterna ellipsodontoides (see main text). Bootstrap support percentages are 100% unless otherwise indicated

Full size image
Fig. 7
figure 7

Strict consensus of four most parsimonious trees (29,086 steps) for the analysis of the unordered data set with a molecular scaffold for all extant taxa and two extinct taxa (ScaffEF). ScaffEF enforced the monophyly of four placental superorders (Xenarthra, Afrotheria, Laurasiatheria, Euarchontoglires) and most relationships within these superorders including the positions of the extinct taxa Thomashuxleya externa and Protolipterna ellipsodontoides (see main text). Bootstrap support percentages are 100% unless otherwise indicated

Full size image

In analyses that did not employ a molecular scaffold, both NoScaff + ordered (Fig. 2) and NoScaff + unordered (Fig. 3) failed to recover three of the four placental superorders (Afrotheria, Euarchontoglires, Laurasiatheria), and only Xenarthra was recovered as monophyletic within a larger ‘edentate’group that also includes edentulous pangolins, aardvarks with enamelless teeth, and the extinct palaeanodont Metacheiromys that had reduced cheekteeth. In both analyses, Tamirtheria is diphyletic, with the Leptictida component of this wholly extinct clade nested within a large ‘insectivore’ group that also includes members of both Afrotheria (Macroscelidea, Afrosoricida) and Laurasiatheria (Eulipotyphla). Archonta (Primates + Dermoptera + Scandentia + Chiroptera) was also recovered in both of the non-scaffolded analyses. Even with these similarities between the two analyses, the treatment of characters as ordered versus unordered resulted in a different phylogenetic position for Glires (Rodentia + Lagomorpha), which are the sister-group to the mostly edentate clade in the ordered analysis and the sister-group to a large ungulate group that includes paenungulates, perissodactyls, cetartiodactyls, and various extinct species (e.g., two South American ungulates) in the unordered analysis. A second difference is that extant cetartiodactyls are all more closely related to each other than they are to other living placentals in the unordered analysis. By contrast, some extant cetartiodactyls (Sus, Bos, Lama) are more closely related to perissodactyls than they are to other living cetartiodactyls in the ordered analysis.

The analysis of the ordered data set with ScaffE resulted in 13 MPTs and the strict consensus of these trees is shown in Fig. 4. The 13 trees are distributed across five tree islands (Fig. 8). The lack of resolution on the strict consensus tree, which includes a polytomy with 20 branches at the base of Eutheria, is primarily the result of two 'rogue' taxa that occur in distinctly different positions among the tree islands. First, Leptictida either: (1) belongs to a paraphyletic Tamirtheria at the base of Placentalia (tree islands 1, 4, 5); (2) belongs to a monophyletic Tamirtheria that is the sister to Placentalia (tree island 2); or (3) is the sister to Atlantogenata (Afrotheria + Xenarthra) within crown Placentalia (tree island 3) (Fig. 8). The second rogue taxon is Thomashuxleya externa, which is nested in Afrotheria (tree islands 1 and 2) or Laurasiatheria (tree islands 3–5). By contrast with the analysis of the ordered data set with ScaffE, the analysis of the unordered data set with ScaffE yielded six MPTs that all belong to the same tree island. The strict consensus of these trees (Fig. 5) is much better resolved than is the strict consensus based on the ordered data set (Fig. 4). Still, relationships among three tamirtherian clades (Leptictida, Maelestes + Ukhaatherium + Zalambdalestes, Prokennalestes + Kennalestes) are unresolved. Finally, analyses with the ordered and unordered data sets and ScaffEF are shown in Figs. 6 and 7, respectively. Tamirtheria is paraphyletic at the base of Placentalia on the strict consensus tree based on ordered + ScaffEF (Fig. 6). Tamirtheria is monophyletic, in turn, on the strict consensus tree based on unordered + ScaffEF, albeit with weak bootstrap support (30%) (Fig. 7).

Fig. 8
figure 8

Relationships among the major groups of Eutheria on different tree islands based on an analysis of the ordered data set with a molecular scaffold for extant taxa (ScaffE). Membership of the individual MPTs to the five tree islands is as follows (see files in Figshare): tree island 1: tree 1; tree island 2: tree 2; tree island 3: trees 12–13; tree island 4: trees 9–11; tree island 5: trees 3–8). Thomashuxleya externa is nested within Afrotheria on tree islands 1 and 2 and within Laurasiatheria on tree islands 3–5

Full size image

Phylogenetic analyses with 796 characters

In analyses with 796 characters and extant taxa only, there was a single MPT (7697 steps) when the characters were treated as ordered and six MPTs (6151 steps) when the characters were treated as unordered. The mean RF distance between the single MPT based on the ordered data set and the six MPTs based on the unordered data set is ~ 33.67, and the mean nRF distance is ~ 0.391. The normalized distance indicates that ~ 39.1% of the internal branches are different on the ordered versus unordered trees. However, both analyses only recovered 11 of the internal branches that are well-supported by molecular data and are part of the molecular scaffold for extant taxa. Further, the 11 internal branches that were recovered with the ordered data set of 796 characters are the same as the 11 internal branches that were recovered with the unordered data set of 796 characters: Theria, Marsupialia, Dermoptera, Scandentia, Primates, Anthropoidea, Carnivora, Chiroptera, Cetacea, Glires, Rodentia.

Pseudoextinction analyses

The results of our pseudoextinction analyses (extant taxa only) with maximum parsimony for individual orders of placental mammals are shown in Figs. 9 and 10 for ordered and unordered characters, respectively. Table 3 summarizes Robinson-Foulds (RF) and normalized RF (nRF) distances between the fully bifurcating, maximum parsimony trees that resulted from pseudoextinction analyses with ordered versus unordered characters. Finally, RF and nRF distances between the fully bifurcating, maximum parsimony trees in pseudoextinction analyses (ordered and unordered) and fully-scaffolded analyses (ordered and unordered characters) are provided in Table 4.

Fig. 9
figure 9

Summary of pseudoextinction analyses with the ordered data set. The species tree is based on a maximum parsimony analysis that was performed with a molecular scaffold for extant taxa and without any pseudoextinct taxa. Extinct taxa were excluded from the analyses. The ordered data set treated 796 multistate characters from the Velazco et al. (2022) data set as ordered and the remainder of the multistate characters as unordered. Pseudoextinct orders and individual species that were recovered in topological positions that are inconsistent with the molecular scaffold are indicated by arrows. Seven orders (Proboscidea, Sirenia, Lagomorpha, Hyracoidea, Cingulata, Pilosa, Rodentia) remained in interordinal positions that are consistent with the molecular scaffold. The use of ordered characters altered the interordinal positions of four orders (Cetartiodactyla, Eulipotyphla, Pholidota, Tubulidentata) when compared to the unordered analysis

Full size image
Fig. 10
figure 10

Summary of pseudoextinction analyses with the unordered data set. The species tree is based on a maximum parsimony analysis that was performed with a molecular scaffold for extant taxa and without any pseudoextinct taxa. Extinct taxa were excluded from the analyses. The unordered data set treated all of the multistate characters from the Velazco et al. (2022) data set as unordered. Pseudoextinct orders and individual species that were recovered in topological positions that are inconsistent with the molecular scaffold are indicated by arrows. Seven orders (Proboscidea, Sirenia, Lagomorpha, Hyracoidea, Cingulata, Pilosa, Rodentia) remained in interordinal positions that are consistent with the molecular scaffold

Full size image
Table 3 Robinson-Foulds (RF) distances and normalized RF (nRF) distances between fully bifurcating, maximum parsimony trees that resulted from pseudoextinction analyses with ordered versus unordered characters. An RF distance of 0 indicates that the trees are identical. In cases of more than one most-parsimonious tree in a pseudoextinction analysis (Perissodactyla), all tree comparisons were performed, and all RF and nRF values are reported
Full size table
Table 4 Robinson-Foulds (RF) distances and normalized RF (nRF) distances between fully bifurcating, maximum parsimony trees in pseudoextinction analyses (ordered and unordered) and fully scaffolded, maximum parsimony trees (ordered and unordered characters). Only one analysis, pseudoextinct Perissodactyla with ordered characters, resulted in two MPTs. In this case, both RF distances are provided. An RF distance of 0 indicates that the trees are identical. Column mean and median values are provided to compare the effects of using ordered versus unordered characters
Full size table

As seen in Table 3, the trees resulting from most pseudoextinction analyses were not greatly affected by using ordered characters. Twelve of 19 analyses produced identical trees (RF = 0) when using ordered versus unordered data sets. An additional three analyses (treating Chiroptera, Pilosa, and Perissodactyla as pseudoextinct) differed only by supporting Atlantogenata (Xenarthra + Afrotheria) as monophyletic on one or all of the MPTs. Seven placental orders (Cingulata, Hyracoidea, Lagomorpha, Pilosa, Proboscidea, Rodentia, Sirenia) retained interordinal positions that are consistent with the molecular scaffold in both ordered and unordered analyses, although in the case of Rodentia the intraordinal position of Rattus shifted to a position that is incompatible with the scaffold (Figs. 9 and 10). Eight orders (Afrosoricida, Carnivora, Chiroptera, Dermoptera, Macroscelidea, Perissodactyla, Primates, Scandentia) shifted to equivalent positions in ordered and unordered analyses that are inconsistent with the molecular scaffold (Figs. 9 and 10). Only four placental orders (Cetartiodactyla, Eulipotyphla, Pholidota, Tubulidentata) were recovered in different topological positions in ordered versus unordered analyses. The largest shift occurred within Cetartiodactyla where both cetaceans (Tursiops, Caperea) were recovered as the sister group to Theria in the ordered analysis (Fig. 9) and the sister group to Perissodactyla + other cetartiodactyls in the unordered analysis (Fig. 10). Pseudoextinction analyses with ordered characters resulted in a sister-group relationship between Pholidota and Tubulidentata when either order was treated as pseudoextinct. In the unordered analyses, Pholidota was recovered as the sister group to Pilosa when pseudoextinct and Tubulidentata became the sister-group to Euungulata (Cetartiodactyla + Perissodactyla) when pseudoextinct. Eulipotyphla joined Afrosoricida when the former was treated as pseudoextinct, although the exact topological relationships within this ‘Lipotyphla’ insectivore group varied slightly in the ordered and unordered analyses (Figs. 9 and 10). The average RF distance for all pseudoextinct placental orders is 6.58 for the ordered analyses and 5.58 for the unordered analyses. The median scores are both 4 (Table 4).

We also performed pseudoextinction analyses with the full data sets (ordered and unordered) that included fossil taxa (Online Resources 1–3). In both cases, all pseudoextinct placental orders that exhibit rearrangements with extant taxa only also exhibit rearrangements with the inclusion of fossil taxa. However, the rearrangements are different for eight (unordered) or nine (ordered) placental orders when fossil taxa were included in the analyses (Online Resource 1). Pseudoextinct Cetartiodactyla, Chiroptera, Dermoptera, Pholidota, Primates, Rodentia, Scandentia, and Tubulidentata exhibit different rearrangements from analyses with extant taxa only with both the ordered and unordered data sets. For example, pseudoextinct Tubulidentata moves from a sister relationship with Pholidota (ordered) or Euungulata (unordered) to a sister relationship with Xenarthra (ordered and unordered), pseudoextinct Pholidota moves from a sister relationship with Tubulidentata (ordered) or Pilosa (unordered) to a sister relationship with Xenarthra (ordered and unordered), and pseudoextinct Scandentia moves from a sister relationship with Primates (ordered and unordered) to a sister relationship with Dermoptera (ordered and unordered). Cetartiodactyla and Chiroptera, both of which include species that move outside of Theria in pseudoextinction analyses with ordered characters and extant taxa only, remain entirely inside of Boreoeutheria in analyses with the ordered data set that includes fossil taxa. The pseudoextinction analysis with Cetartiodactyla and unordered characters also places the two cetaceans (Tursiops, Caperea) as sister to other therians in the analysis with extant taxa only, but these species are part of a paraphyletic Cetartiodactyla in analyses that also include fossils. Finally, pseudoextinct Macroscelidea exhibits a different position when fossils are included in the ordered analysis (Online Resource 1–2). Specifically, Macroscelidea is sister to Leptictida, and these two orders group with Eulipotyphla inside of crown Placentalia. Leptictida is also positioned inside of crown Placentalia on one or more trees in 13 of the pseudoextinction analyses with ordered characters and two of the pseudoextinction analyses with unordered characters (Online Resource 1).

Discussion

Impact of ordered characters on the Velazco et al. (2022) data set

Understanding phylogenetic relationships is a fundamental goal of evolutionary biology. In the case of long extinct species where molecular data are not available, morphological characters are necessary to hypothesize phylogenetic relationships (Hillis 1987; Wiens 2004; Giribet 2015; Lee and Palci 2015). The debate about how to analyze clinal and meristic multistate morphological characters, whether as ordered or unordered, remains unsettled. Previous studies have arrived at conflicting conclusions about the benefits and pitfalls of treating multistate morphological characters as ordered versus unordered. Reeder et al. (2015), Baron et al. (2017), and Lee and Yates (2018) tested the effects of ordering morphological characters in phylogenetic analyses of squamates, dinosaurs, and crocodilians, respectively. All of these authors concluded that using ordered characters had little overall impact on their results. In contrast, Fröbisch and Schoch (2009) found that using ordered characters resulted in a significantly more resolved consensus tree for dissorophoid temnospondyls despite recovering a similar number of MPTs; McGowan (2002) and Marjanović and Laurin (2008) found the opposite in another data set of early tetrapods. Moreover, Marjanović and Laurin (2008) concluded that ordering characters can increase the strength of a signal in the data, which can increase the resolution but also decrease it by revealing character conflict that remains hidden when all characters are unordered. Other studies have also reported important shifts in topologies when using ordered versus unordered characters. Simões et al. (2017) recovered gekkotans as monophyletic when using ordered characters, but this group was recovered as paraphyletic in the unordered analysis. In this study, we explored the effects of using ordered characters with Velazco et al.’s (2022) large morphological matrix for placental mammals.

Non-scaffolded phylogenetic analyses (NoScaff) with ordered versus unordered characters recovered many of the same groups when 796 multistate characters were treated as ordered rather than unordered. Among the groups that were recovered in analyses with both the ordered and unordered data sets are a large “insectivore” group (Leptictidae + Macroscelidea + diphyletic Eulipotyphla + diphyletic Afrosoricida), a ‘spiny hedgehog’ group (Erinaceus + Echinops), Archonta (Dermoptera + Scandentia + Primates + Chiroptera), an “ant & termite eating” group (Xenarthra + Pholidota + Palaeanodonta + Tubulidentata), a large “ungulate” group (Perissodactyla + paraphyletic Cetartiodactyla + Mesonyx + diphyletic Paenungulata + two South American ungulates), and a “cud-chewing” group (Bos + Lama). These polyphyletic ecomorphology groups were also recovered in O’Leary et al.’s (2013) parsimony analysis of the original “phenomic” data set (see Springer et al. 2013). In addition, Tamirtheria was recovered as a diphyletic group with both ordered and unordered characters in non-scaffolded analyses.

Although many of the clades that were recovered in the NoScaff analyses are the same, there are also differences between the results of analyses with ordered versus unordered characters. For example, the analysis with ordered characters recovered Glires and the “ant & termite eating” group as sister taxa. The “ungulate” group, in turn, is the sister-taxon to this collective group. By contrast, in the unordered analysis Glires is the sister-group to the ‘ungulate’ group. The ordered and unordered analyses also recovered different versions of paraphyletic Cetartiodactyla. In the ordered analyses perissodactyls were embedded within Cetartiodactyla whereas in the unordered analyses the mesonychid Mesonyx was found there. Notably, all of the analyses without a molecular scaffold failed to recover Afrotheria, Euarchontoglires, and Laurasiatheria, which are three of the four major groups of placental mammals. All of the NoScaff analyses also recovered multiple clades that are inconsistent with the results of molecular/phylogenomic studies (Meredith et al. 2011; dos Reis et al. 2012; Tarver et al. 2016; Foley et al. 2023) and highlight the ongoing conflict between morphological and molecular studies addressing higher level relationships among placental mammals.

By contrast with the non-scaffolded analyses, the use of ordered versus unordered characters had fewer impacts on our results when molecular scaffolds were applied. This is not surprising given the rigidity of the scaffolds for all extant taxa (ScaffE analyses) or all extant taxa plus two fossil (extinct) South American ungulates (ScaffEF analyses). The most noteworthy differences in tree topology concern the deployment of taxa within Tamirtheria. Among the 13 MPTs that resulted from the analysis of ordered characters with ScaffE, different tree islands deploy Tamirtheria as monophyletic, paraphyletic at the base of Placentalia with three monophyletic subgroups (Leptictida, Kennalestes + Prokennalestes, Maelestes + Zalambdalestes + Ukhaatherium), or diphyletic with Leptictida inside of Placentalia and other tamirtherians outside of Placentalia (Fig. 8). Analysis of the unordered data set with ScaffE also recovered one or more MPTs with either a monophyletic Tamirtheria or a paraphyletic Tamirtheria at the base of Placentalia, but a topology with Leptictida inside of Placentalia was not among the MPTs (Fig. 5). The analysis of ordered characters with ScaffEF resulted in MPTs with a paraphyletic Tamirtheria (Fig. 6) whereas the analysis of unordered characters with ScaffEF resulted in a monophyletic Tamirtheria (Fig. 7).

The contents of Tamirtheria

A main conclusion of Velazco et al. (2022) is that there is a wholly extinct sister-group to Placentalia, named Tamirtheria, that had not previously been recognized or described. As recovered by Velazco et al. (2022), Tamirtheria is a diverse clade that includes fossils from the Cretaceous and Paleogene. Velazco et al. (2022) defined Tamirtheria to include Ukhaatherium nessovi and all species more closely related to U. nessovi than to Elephas maximus. Representatives of Tamirtheria in Velazco et al.’s (2022) analyses include four leptictids, one cimolestid, two asioryctitherians, one zalambdalestid, and the mid-Cretaceous eutherian Prokennalestes. This group was recovered as monophyletic in the total-evidence parsimony analysis performed by Velazco et al. (2022) on their unmodified data set that included polymorphic characters and characters with more than ten character states. Prior to Velazco et al. (2022), the phylogenetic position of various tamirtherians has been controversial. For example, some studies recovered Leptictis as a stem eutherian (Wible et al. 2007, 2009; Halliday et al. 2019) whereas others recovered Leptictis as a crown placental (Asher et al. 2003; O’Leary et al. 2013). Given the hypothesized close relationship between the Paleogene Leptictis and the Cretaceous Gypsonictops (Lillegraven 1969; Wible et al. 2007, 2009; Halliday et al. 2019; but see Wang and Wang 2023), the phylogenetic position of Leptictida is of great interest. Gypsonictops has now been reported from the Turonian (93.9–89.8 Ma) stage of the Late Cretaceous (Cohen and Cifelli, 2015; Cohen, 2017; Halliday et al. 2019), and if Leptictida (including Gypsonictops) is a member of crown Placentalia, then paleontological objections to the Long Fuse Model of placental diversification, which places most interordinal cladogenesis of placental mammals in the Cretaceous, would largely be rebutted (Springer et al. 2019).

As noted above, there was mixed support for the monophyly of Tamirtheria (sensu Velazco et al. 2022) in our searches, and this group was also recovered as paraphyletic or polyphyletic. These results contrast with those of Velazco et al. (2022), wherein Tamirtheria was supported by a Bremer support value of 12. However, Velazco et al. (2022) also performed a maximum likelihood analysis with RAxML on a modified version of their matrix that scored both polymorphic character states and character states > 9 as “?” because of limitations of the RAxML program. This analysis recovered Tamirtheria as paraphyletic rather than monophyletic, although parsimony analysis of the modified matrix still recovered Tamirtheria, albeit with a low jackknife support score (26%). The combination of Velazco et al.’s (2022) results and the results of this study suggest that Tamirtheria monophyly is sensitive to: (1) the use of a combined matrix (DNA + morphology) versus a morphology matrix with a molecular scaffold; (2) the type of analysis that is performed (i.e., parsimony versus likelihood); and (3) the use of ordered versus unordered characters, since analyses that treated all characters as unordered were the most likely to recover Tamirtheria as monophyletic (see Table 2). Moreover, it is difficult to confidently determine whether Tamirtheria is monophyletic given the absence of molecular data for all species that comprise this hypothesized clade.

Velazco et al. (2022) provided a stem-based definition (sensu Sereno 1999) for Tamirtheria, so the definition of this clade is stable even if its composition is uncertain. Our results suggest that Tamirtheria minimally includes Ukhaatherium, Maelestes, and Zalambdalestes because Maelestes and Zalambdalestes are almost always more closely related to Ukhaatherium than to Elephas (Figs. 2, 3, 5, 6, and 7). Kennalestes and Prokennalestes, in turn, are occasional members of Tamirtheria (Figs. 2, 3, and 7) but are closer to Elephas than to Ukhaatherium in other cases (Fig. 6). Finally, the four leptictids are the least probable members of Tamirtheria based on our analyses (Fig. 7) and are often more closely related to Elephas than to Ukhaatherium (Figs. 2, 3, 6, and 8), either as the sister group to Placentalia or even inside the crown group. Evidence for the latter is not overwhelming, but the possibility that a Turonian member of Leptictida (i.e., Gypsonictops) is a crown placental should not be discounted and potentially has great ramifications for reconciling molecular and paleontological views on the timing of the placental radiation.

The performance of morphology in cladistic analyses using pseudoextinction

Understanding the phylogenetic relationships of extinct taxa is necessary to fully comprehend the Tree of Life. The vast majority of life on Earth is extinct, and accurately reconstructing the evolutionary relationships of extinct taxa remains a challenge. Since long extinct species cannot be analyzed using molecular data (Hofreiter et al. 2001) the performance of phylogenetic analyses based on morphological data sets must be critically evaluated. Pseudoextinction analyses are one approach to evaluate the performance of morphological phylogenetics. These analyses rescore soft morphological and molecular data as missing for a subset of taxa prior to running a phylogenetic analysis (Asher and Hofreiter 2006; Springer et al. 2007, 2019; Sansom and Wills 2013; Pattinson et al. 2014; Brady and Springer 2021). Previous analyses that employed O’Leary et al.’s (2013) large morphological matrix concluded that the morphological characters in this matrix were insufficient to reconstruct interordinal relationships congruent with a strongly supported molecular scaffold for most placental orders when these orders were treated as pseudoextinct (Springer et al. 2019; Brady and Springer 2021). In the present study, we performed pseudoextinction analyses on Velazco et al.’s (2022) data set, which includes the same suite of extant taxa as does O’Leary et al.’s (2013) data set. Most importantly, we expanded our analyses to include a comparison of unordered versus ordered morphological characters to determine if character ordering would improve the performance of morphological cladistics with this data set.

Our results show that character ordering has little effect on pseudoextinction analyses with this data set (Figs. 9 and 10). Unfortunately, ordered characters were just as likely as unordered characters to reconstruct the phylogenetic relationships of the majority of placental orders in positions incongruent with our molecular scaffold when they were treated as pseudoextinct. Indeed, 13 of 19 pseudoextinct placental orders showed interordinal and/or intraordinal rearrangements that are inconsistent with the molecular scaffold. Only four of the 13 wayfaring pseudoextinct orders (Cetartiodactyla, Eulipotyphla, Pholidota, Tubulidentata) were reconstructed in different topological positions when using ordered versus unordered characters (Figs. 9 and 10). The most extreme differences between the ordered and unordered pseudoextinction analyses are seen in Cetartiodactyla. In the unordered analysis (Fig. 10), the two cetaceans (Tursiops and Caperea) are positioned at the base of Cetartiodactyla whereas these taxa lie outside of Theria in the analysis with ordered characters (Fig. 9). This shift in position may be related to the extremely derived morphology of cetaceans in conjunction with sparse taxon sampling. However, a promising result is that the two extant cetaceans are restored to a position within Euungulata (Cetartiodactyla + Perissodactyla) (Online Resource 2) when the pseudoextinction analysis with ordered characters is performed with extant taxa and fossils. Similarly, pseudoextinction results with extant taxa only (ordered and unordered) both result in the movement of Chiroptera to a position outside of Theria, but bats are restored to a position inside of Placentalia when fossils are included in the analyses. These results suggest that continued additions of fossil taxa to the O’Leary et al. (2013)/Velazco et al. (2022) matrix have the potential to break up long branches and improve the performance of morphological cladistics for resolving relationships among extant and extinct mammals.