The Website
The 196 characters first described in [17] are available in web-format via the author's institutional website [26] and is archived on the BMC website [see additional file 1]. With few exceptions, images were photographed using museum collections in Berlin (ZMB), New York (AMNH), Washington DC (USNM), London (NHM), Pretoria (TM), and Cambridge (UMZC). Images and character descriptions were combined and exported as JPEG or GIF files using Adobe Photoshop and Illustrator. These were linked into HTML files using Mozilla Composer.
Morphology matrix
The current web-matrix includes corrections to Appendices 1 and 2 of [17] [see additional file 1]. Among the typographical errors listed, only one had an effect on the analysis: character 41 of Tapirus ("mastoid exposure in braincase") was inadvertently omitted from the printed Appendix 1 from [17]. It should have been listed as state "0" for Tapirus (mastoid exposed). With this correction, and using either PAUP [23] or NONA [27] under the analytical defaults of POY 2.7 [28] (e.g., polymorphisms treated as missing data), the morphological dataset published in appendix 1 of [17] yields the reported 4 trees at 1088 steps.
The terms "fenestra rotunda", "fenestra cochleae", and "round window" have been used interchangeably for the aperture in the ventrum of the petrosal pars cochlearis, leading into the cochlea, just posterior to the fenestra vestibularis (or oval window; see [29]). Asher et al. [17,24] had previously used the descriptor "rotundum" for this structure in characters 4 and 5, which should have been reserved for the distinct exit foramen for the maxillary division of the trigeminal nerve (as in primates, carnivorans, and marsupials). In order to avoid confusion between the fenestra "rotunda" (round window) and the foramen "rotundum" (exit foramen for V-2), text and images for characters 4–7 now use the term "fenestra cochleae" for this opening on the ventrum of the pars cochlearis, following [29].
Relative to the descriptions first published in [17], the text for several characters has been changed in order to better correspond to the specimens available for display on the website.
In addition to the typographical corrections summarized above, some of the coding decisions in [17] have also been changed [see additional file 1], which of course do influence the structure of the tree. Six of these were indicated in [24]; four additional improvements are identified here.
First, instead of identifying a separate character state for "glenoid poorly defined" for character #56 in Manis, this character is coded as in most other mammals: state 0, "glenoid even with petrosal." This increases consistency in how the fossil taxon Plesiorycteropus was coded, and reflects the actual position of the glenoid fossa for the mandible in a transverse plane near the petrosal bone, as opposed to the dorsally situated glenoid in, for example, chrysochlorids or caviomorph rodents.
Second, the lacrimal bone (character #71) in leporid skulls is not always well ossified to surrounding bones, and in some specimens it may fall out leaving an artefactual "fenestra" in the anterior orbit. This was incorrectly coded in [17,24] as a separate character state, "fenestra in anterior orbit." Here, this is recoded in the leporid terminal as "lacrimal foramen present."
Third, Didelphis possesses a distinct foramen rotundum (i.e., exit foramen for the maxillary [2nd] division of the trigeminal nerve, character #48), just posterior to the sphenorbital fissure [30,31]. The foramen rotundum was mistakenly coded as "confluent with sphenorbital fissure" in [2,17,24]. It is here corrected to state 1 ("distinct") to reflect the ossified, separate exit foramen for the maxillary division of the trigeminal nerve in this taxon.
Fourth, character #39 "condyloid foramina" should have been worded to specifically indicate the hypoglossal foramen, reflecting the usage of [31]. As summarized by [[32]: p. 175], the terms "condylar" or "condyloid" foramen have been used for this structure [2]. However, the descriptor "condylar" or "dorsal condylar" may also refer to small, nutrient foramina adjacent to the occipital condyle [[32]: p. 151]. Several taxa show multiple foramina that perforate the basioccipital anterior to the occipital condyle (e.g., Didelphis); others show a single, conspicuous hypoglossal foramen (e.g., Pteropus), and others lack a hypoglossal foramen (e.g., Balaenoptera). Asher et al. [17,24] had previously coded Orycteropus, Sus, and Sorex as lacking hypoglossal foramina; here, these codings are corrected to state 1 ("single") for the former two, and states 0 and 1 (polymorphic) for Sorex.
DNA sequence and indel dataset
Sequences of the tyrosinase (TYR) gene in Equus (accession AF252540) were added to the alignment of [9]. In addition, several interruptions of the reading frame and placements of several indels were adjusted (see additional file 1), amounting to 34 alterations in presumed sequence homology. In addition, 221 insertion-deletion indel characters from protein-coding genes in this DNA dataset were incorporated into a new phylogenetic analysis using MP [23] and MrBayes [33]. Each indel character is coded as 0 (for gaps) or 1 (for insertions) and consists of one or more units of three contiguous gaps. Regardless of length, such occurrences were coded as a single, binary character, shared by two or more taxa when they show overlap. Elongate gaps that overlapped with multiple, smaller gaps were coded as a single event; i.e., when an elongate gap character in taxon A overlapped with multiple, smaller gap characters in taxa B and C, the smaller gap-characters were coded as inapplicable for taxon A and treated as missing data in the analysis, based on the method of "simple indel coding" [34]. The newly-aligned sequence dataset is available linked to additional file 1. Exclusion of sites identified as "alignment ambiguous" by [9] did not have a significant effect on the topologies reported here.
Taxon sample
The choice of Recent taxa for inclusion in this dataset is based on maximizing the overlap of the morphological dataset with the 19 nuclear and 3 mitochondrial gene dataset used by [9]. This is the same sample used by [24], and is slightly smaller than that used by [17], including 41 extant and 12 extinct mammalian terminals. Not included are the sciurid, Bradypus, Tadarida, and Vampyrum sequences used by [9]; and a single terminal is used for the Caribbean lipotyphlan Solenodon (using sequence data for Solenodon paradoxus). Several terminal taxa are composites, listed here with suprageneric names, and are identified in table 1 of [24].
Phylogenetic analysis parameters
Different schemes for weighting third positions codons in MP (excluded, transitions ignored, included) were explored. Sequence data for all fossils were coded as missing; all morphological character changes were treated as nonadditive (unordered). In all MP analyses, multistate characters were treated as polymorphic, indel characters embedded in the sequence data matrix were treated as missing data (but were represented in an additional character matrix), and tree searches using PAUP [23] were heuristic using at least 200 random addition replicates and TBR branch-swapping. Bootstrap values are based on at least 100 pseudoreplicates of a 3-replicate TBR random addition sequence.
Analyses with MrBayes [33] used the AIC as applied in MrModeltest [35], based on ML scores generated by PAUP [23], to determine the model of evolution for each genetic locus independently as well as for the combined nuclear and mitochondrial genes as two discrete partitions. In most cases this identified the GTR+G+I model as optimal (Table 2). Bayesian treebuilding was computationally intensive. Partitioning the data into units of nuclear (ca. 15KB) and mitochondrial (ca. 1.5KB) DNA, plus 221 indel characters, the former two with an independent GTR+G+I model and the latter with a restriction site model (as recommended in MrBayes documentation), and combining them with the datasets for morphology including fossil taxa, took 18 days for 2 million generations on a single mac G5 processor (2.5 GHz and 2.5 GB RAM) with MrBayes 3.1. This still did not yield convergence across two independent runs. Hence, Bayesian analyses included three of the 12 sampled fossils (plus all 41 Recent taxa), using just over 1.6 million generations in two independent runs, which yielded the same consensus of post-burnin topologies (Fig. 3).
Analysis of sequence data for the 41 extant terminals only, with three unlinked evolution models defined for nucDNA, mtRNA, and indels, yielded convergence for two independent runs after ca. 3 weeks of uninterrupted computing time for one million generations on a 2Ghz P4 desktop PC with 512MB RAM. Using 21 unlinked models of sequence evolution for each gene (Table 2) in two additional runs of one million generations each yielded the same post-burnin, majority rule consensus topology as the 3-model analysis. Based on manual inspection of likelihood scores, Bayesian analyses across these analyses reached stationarity after approximately 15K generations; burn-in was conservatively defined after 50K generations.
Statistical tests of competing topologies were carried out in PAUP 4.0b10 [23]. One of the four MPTs including all data with all changes equal (Fig. 1), and one of the four MPTs resulting from the analysis excluding third coding positions (Fig. 2), were compared with several alternatives (Table 1). Because of differences in taxon sample across studies concerning the root of Placentalia [e.g., [9,15,16]], these alternatives were constructed with the present dataset, using backbone-constraints derived from each study. For example, taxa from the present dataset sampled in common with [16] were constrained in PAUP to fit figure 1 from [16], which supported erinaceid insectivorans basal followed by murid rodents. One of the resulting MPTs was then compared to an unconstrained, optimal MPT using the present morphology-DNA-indel dataset under the assumptions given in Fig. 1 (equal weighting) and Fig. 2 (third positions excluded). The same procedure was followed for hypotheses supporting basal positions of Atlantogenata [14], Xenarthra [12], Afrotheria [9], Glires [15], and Muridae (Fig. 4B).
Authors' contributions
RJA assembled the morphological and DNA sequence data matrices (the latter based on an alignment supplied by A. Roca and W. Murphy), designed the web-database, carried out the phylogenetic analyses, and wrote the manuscript. All authors read and approved the final manuscript.
Acknowledgements
I thank Al Roca and Bill Murphy for making available their DNA sequence alignment. Two anonymous reviewers and the editorial staff at BMC provided comments that helped to improve the manuscript. For financial support I thank the Deutsche Forschungsgemeinschaft (grant AS 245/2-1), which enabled photography and processing of the images used on the morphology web-database, as well as employment of my colleague Kristina Fritz who was of great help in completing both tasks. I thank in addition the European Commission's Research Infrastructure Action via the SYNTHESYS Project (GB-TAF 218), the Museum für Naturkunde Berlin, and the University of Cambridge Museum of Zoology. I am grateful to the staff at several mammalogy collections for access, particularly the Museum für Naturkunde (Berlin), American Museum of Natural History (New York), the National Museum of Natural History (Washington DC), the Natural History Museum (London), the Transvaal Museum (Pretoria), and the University Museum of Zoology Cambridge.
Answers to all your Biology Questions
