Geological Context of the Perudyptes Type Locality

Cenozoic marine sediments are exposed throughout the Pisco Basin of coastal southern Peru (fig. 1). The stratigraphy of the Pisco Basin was described in detail by Dunbar et al. (1990) and DeVries (1998). Marine deposits were laid down over the course of the Cenozoic during six major transgressions (DeVries, 1998) and preserve a wealth of fossil marine vertebrates (e.g., Muizon, 1981, 1984, 1988, 1993; Muizon et al., 2003, 2004; Stucchi, 2002, 2003, 2007; Stucchi and Urbina, 2004; Stucchi and Emslie, 2005; Stucchi et al., 2003; Clarke et al., 2007; Esperante et al., 2008; Ksepka et al., 2008; Lambert et al., 2008) and invertebrates (e.g., Rivera, 1957; de Muizon and DeVries, 1985; DeVries, 2007). Perudyptes devriesi is known from a single specimen collected from the middle Eocene Paracas Formation (Choros Formation of Dunbar et al., 1990). The Paracas Formation directly overlies crystalline basement rocks and is separated by an unconformity from the overlying middle-upper Eocene Otuma Formation (DeVries et. al., 2006; DeVries, 2007). The Paracas Formation comprises a basal transgressive sandstone member representing a comparatively nearshore paleoenvironment and grading upward into a fine-grained, tuffaceous, and diatomaceous silty sandstone member representing a more distal shelf paleoenvironment (DeVries, 2007). The Perudyptes devriesi holotype (MUSM 889) was collected from the southern wall of the Quebrada Perdida locality (14°34′S, 75°52′W) in an orange, coarse-grained, thick-bedded, siliciclastic sandstone bed that has been identified as part of the basal member of the Paracas Formation (DeVries, 1998). The beds containing the Perudyptes devriesi holotype are located a few tens of meters above the rugged platform of crystalline basement rock. Based on the presence of late middle Eocene radiolarians (Cryptocarpium ornatum, Lithocyclia aristotelis, Lithocyclia ocellus) higher in the section and the gastropod Turritella lagunillasensis in correlative sandstone beds, the age of these beds is ~42 Ma (DeVries et al., 2006).

Figure 1

Map showing the type locality of Perudyptes devriesi (star) and the extent of the Pisco Basin, with simplified stratigraphic column reflecting our current understanding of stratigraphy within the Pisco Basin (after Devries et al., 2006).

Littoral invertebrate remains support deposition of the basal Paracas sandstone at Quebrada Perdida as close to the paleoshoreline. During the Eocene, Quebrada Perdida would have been located at a paleolatitude nearly equivalent to the 14°34′S present-day latitude, as there has been essentially no latitudinal translation of the area since the late Jurassic (Jesinkey et al., 1997; Somoza and Tomlinson, 2002; Hartley et al., 2005). Cold-water upwelling along the western coast of Peru appears to have been in place by the Late Cretaceous or early Tertiary (Keller et al., 1997), and this “proto-Humboldt” current may have influenced low-latitude penguin diversity by cycling cold, nutrient-rich water into the ecosystem (Clarke et al., 2007). Sedimentary evidence indicates that land adjacent to the coast was continuously arid or semiarid from the Jurassic to present day (Hartley et al., 2005).

Additional fossils from the Paracas Formation currently awaiting description suggest that a minimum of three penguin taxa occupying a range of body sizes occurred in the middle Eocene of Peru. Based on the size of preserved limb bones, the holotype individual of Perudyptes devriesi was comparable in size to the King Penguin (Aptenodytes patagonicus), the second largest extant penguin species. A smaller penguin (~75% the size of Perudyptes devriesi) is represented by a partial tibiotarsus and a partial tarsometatarsus (MUSM 890). Three large specimens significantly surpassing comparable elements of Perudyptes devriesi in size have also been collected (MUSM 891, 892, 894), but because of incomplete preservation it remains uncertain whether all specimens belong to the same taxon or multiple species are represented. However, these materials indicate that a diverse penguin fauna was well established in Peru by the middle Eocene (Clarke et al., 2007). Only one extant species of penguin (Spheniscus humboldti, Peruvian Black-footed Penguin) occurs regularly along the coast of Peru today, and only this species plus Spheniscus mendiculus (Galapagos Penguin) occur at latitudes below 20°S (Williams, 1995). While time averaging must be taken into account, the presence of at least three distinct taxa in a relatively narrow stratigraphic interval suggests greater sphenisciform diversity in Peru, and at low latitudes in general, during the Eocene than in the present.

Taxonomic Background

Clarke et al. (2003) proposed phylogenetic definitions for higher taxa within the penguin total group. Under these definitions, Pansphenisciformes is applied to the clade including all taxa more closely related to Spheniscidae than any other extant avian lineage. Sphenisciformes is applied in a more exclusive sense to the clade including all Pansphenisciformes that share the apomorphic loss of aerial flight. These taxa currently include the same set of known species. However, volant basal members of the penguin lineage (not yet reported in the fossil record) would be placed within Pansphenisciformes but excluded from Sphenisciformes. Spheniscidae is applied to the crown clade of penguins, comprising the most recent common ancestor of all living penguin species and its descendants. We employ this recommended taxonomy throughout this paper.

Institutional Abbreviations

AMNH, American Museum of Natural History; CADIC, Centro Austral de Investigaciones Científicas, Tierra del Fuego, Argentina; CM, Canterbury Museum, Christchurch, New Zealand; IB/P/B, Prof. A. Myrcha University Museum of Nature, University of Bialystok, Bialystok, Poland; MLP, Museo de La Plata, La Plata, Argentina; MNZS, Museum of New Zealand Te Papa Tongarewa, Wellington, New Zealand; MUSM, Museum of San Marcos University, Lima, Peru; NMV, National Museum of Victoria, Melbourne, Australia; OM, Otago Museum, Dunedin, New Zealand; OU, Otago University Geology Museum, Dunedin, New Zealand; SAM, South African Museum, Cape Town, South Africa; SAMA, South Australian Museum, Adelaide, Australia; UCMP, University of California Museum of Paleontology, Berkeley, CA; USNM, National Museum of Natural History, Smithsonian Institution, Washington, DC.

Primary Analysis

In order to evaluate the phylogenetic placement of Perudyptes devriesi and recently discovered fossil taxa such as Madrynornis mirandus, Pygoscelis grandis, and Spheniscus muizoni, we employed a large combined morphological and molecular dataset for Sphenisciformes. Earlier versions of this dataset were formulated and utilized by Giannini and Bertelli (2004), Bertelli and Giannini (2005), Bertelli et al. (2006), Ksepka et al. (2006), Clarke et al. (2007), and, most recently, by Ksepka (2007). In the present study, we modified Ksepka's (2007) version of this matrix, including 25 morphological characters and 11 fossil taxa added since the most recent published version of the matrix (Clarke et al., 2007). In addition to increasing taxonomic sampling, we were also able to incorporate additional codings for Palaeospheniscus biloculata based on new material reported by Acosta Hospitaleche et al. (2007) and for Palaeospheniscus patagonicus based on new material reported by Acosta Hospitaleche et al. (2008). The current dataset contains 219 morphological characters and >6000 bp of sequence data. Fossil specimens examined are presented in appendix 1, GenBank accession numbers for sequences are presented in appendix 2, morphological character definitions are presented in appendix 3, and the morphological matrix is presented in appendix 4.

Outgroups

Recent studies of avian higher level phylogeny have generally supported a sister group relationship between Sphenisciformes and Procellariiformes (McKitrick, 1991; van Tuinen et al., 2000; Livezey and Zusi, 2006, 2007; Hackett et al., 2008), Gaviiformes (Mayr and Clarke, 2003), or a clade uniting Procellariiformes + Gaviiformes (Sibley and Ahlquist, 1990; Cooper and Penny, 1997) (see also summary hypothesis and discussion of Cracraft et al., 2004). Groth and Barrowclough (1999) recovered a Sphenisciformes-Gaviiformes clade, but they did not include representatives of Procellariiformes in their analysis. A few authors have recovered a sister group relationship between Sphenisciformes and a clade uniting Gaviiformes and Podicipediformes (Cracraft, 1985, 1988; Mayr and Clarke, 2003; Bourdon et al., 2005), although clustering of Gaviiformes and Podicipediformes has typically been attributed to convergence in these two foot-propelled diving lineages (e.g., Storer, 1971; Mayr and Clarke, 2003). Strong molecular evidence (van Tuinen et al., 2000; Chubb, 2004; Ericson et al., 2006; Hackett et al., 2008), as well as some morphological data (Mayr, 2004; Manegold, 2006), suggests that Podicipediformes are the sister taxon of Phoenicopteriformes, and that these two taxa are distantly related to Gaviiformes and Sphenisciformes.

Two alternative hypotheses regarding the higher level relationships of Sphenisciformes have recently been proposed. One hypothesis posits that Sphenisciformes are closely related to core Pelecaniformes (Fain and Houde, 2004; Mayr, 2005, 2007). This hypothesis was supported by sequence data from the beta-fibrinogen gene (Fain and Houde, 2004), although it should be noted that when additional genes are considered alongside beta-fibrinogen, Sphenisciformes are either placed in an unresolved position (Ericson et al., 2006) or united with Procellariiformes (Hackett et al., 2008). Morphological data have been presented for a sister group relationship between Sphenisciformes and the extinct flightless diving clade Plotopteridae, and for the placement of this clade within Pelecaniformes (Mayr, 2005, 2007), although some workers instead attribute similarities between penguins and plotopterids to convergence (e.g., Olson and Hasegawa, 1979, 1996). Mayr (2005) conducted a phylogenetic analysis resulting in a Sphenisciformes-Plotopteridae clade, nested within a larger clade including most traditional members of Pelecaniformes. Data from the basal penguin Waimanu, unpublished at the time of Mayr's (2005) analysis, change the distribution of some characters presented to support these proposed relationships (Ksepka, 2007; Mayr, 2007), and one morphological analysis of higher level avian relationships including data from Waimanu resulted in a sister group relationship between Sphenisciformes and Gaviiformes + Podicipediformes to the exclusion of representative Pelecaniformes (Slack et al., 2006). Unfortunately, the cranial anatomy of Plotopteridae remains essentially unknown, precluding further exploration of the Sphenisciformes-Plotopteridae hypothesis.

A second alternative hypothesis links Sphenisciformes and Ciconiiformes (Slack et al., 2003, 2006; Harrison et al., 2004; Watanabe et al., 2006). Thus far, this hypothesis has been supported exclusively by mitochondrial sequence data, and no morphological evidence has not been put forward to support a Sphenisciformes-Ciconiiformes clade.

Interestingly, several recent large-scale analyses of avian phylogeny support a general framework in which Sphenisciformes are part of a large seabird clade including Procellariiformes, Gaviiformes, Pelecaniformes, and Ciconiiformes (Ericson et al., 2006; Livezey and Zusi, 2006, 2007; Hackett et al., 2008), indicating that each alternate hypothesis may be supported by underlying synapomorphies of this larger clade. The balance of available evidence suggests Sphenisciformes and Procellariiformes are sister taxa, while Gaviiformes, Pelecaniformes, and Ciconiiformes represent more distal outgroups to penguins. In the present study we selected two species of Gaviiformes and 13 species of Procellariiformes as outgroups. Trees were rooted to Gavia immer.

Ingroup Taxonomic Sampling

We included 19 extant penguin taxa in our ingroup, reflecting ongoing refinement of penguin species limits. Recently, Banks et al. (2006) presented molecular evidence supporting species status for three taxa formerly considered subspecies of Eudyptes chrysocome: Eudyptes chrysocome (Southern Rockhopper Penguin), Eudyptes moseleyi (Northern Rockhopper Penguin), and Eudyptes filholi (Eastern Rockhopper Penguin). This conclusion is also supported by morphological differences and the allopatric distribution of the three species (Banks et al., 2006). In past studies (Giannini and Bertelli, 2005; Bertelli and Giannini, 2006; Ksepka et al., 2006; Clarke et al., 2007), Eudyptes chrysocome chrysocome and Eudyptes chrysocome moseleyi were coded as separate operational taxonomic units in recognition of these morphological differences. Here, we treat all three of the newly recognized rockhopper species as separate terminals.

We included 33 extinct penguin species as well as multiple specimens exhibiting distinct and novel combinations of characters (Ksepka, 2007). Two important sets of specimens originally referred to Palaeeudyptes antarcticus (Marples, 1952), but which differ significantly from one another and cannot be assigned to that species (see Simpson, 1971b; Fordyce and Jones, 1990; Fordyce, 1991; Ando, 2007; Ksepka, 2007), were included as two distinct terminals named for their localities: Burnside “Palaeeudyptes” (OM GL435) and Duntroon “Palaeeudyptes” (OM GL427, OM GL432). We also included a giant penguin tarsometatarsus (UCMP 321023) preserving synapomorphies of Anthropornis but differing strongly in proportions from Anthropornis nordenskjoeldi and Anthropornis grandis (Ksepka, 2007) as a separate terminal. Following Ksepka (2007), we excluded multiple fossil taxa that are taxonomic equivalents of more well-known species included in our matrix. The exclusion of taxonomic equivalents can improve resolution in consensus trees without changing the relationships of the remaining taxa (Wilkinson, 1995). A few fragmentary taxa are not taxonomic equivalents because they include infrequently recovered elements (e.g., Palaeeudyptes marplesi, which is known from a few incomplete limb elements but includes the rarely preserved patella). However, these taxa were excluded, as simultaneous inclusion of all such taxa has been demonstrated to result in a complete lack of resolution in the strict consensus of resultant trees (Ksepka, 2007). Instead, these incomplete specimens were evaluated using synapomorphies identified in the primary analysis (see below).

Treatment of Problematic Fossils

Isolated specimens can create difficulties for both taxonomy and phylogenetic analysis. The first fossil penguin species to be named was based on a single broken tarsometatarsus (Huxley, 1859), and the taxonomic difficulties associated with fragmentary holotypes and disassociated specimens have remained a constant thread in discussions of penguin evolution (e.g., Simpson, 1971b; Fordyce, 1991; Ksepka, 2007). Deciding how to treat isolated fossils is particularly difficult when multiple taxa are present at a locality. Erecting new taxa from single, nonoverlapping elements in cases where comparisons to previously named taxa cannot be made is poor taxonomic practice, and it can obscure understanding of radiation and extinction by artificial taxonomic inflation of species diversity. On the other hand, to make referrals without evidence from associated specimens risks the creation of chimeric taxa, which may confound phylogenetic analysis. We comment on the taxonomic issues surrounding two recently proposed species based on isolated elements below.

Challenges posed by isolated fossils must be addressed in dealing with penguin fossils from the Eocene La Meseta Formation of Seymour Island (Antarctic Peninsula). For largely historical reasons, the taxonomy of fossil penguins from the La Meseta Formation has been based primarily on the tarsometatarsus (Wiman, 1905a, 1905b; Marples, 1953; Myrcha et al., 1990, 2002). Although penguin fossils are abundant particularly in the upper levels of the La Meseta Formation, articulated specimens remain extremely rare, and no specimens including a tarsometatarsus and additional elements identifiable to a single individual are known. Davis and Briggs (1998) examined 1233 USNM penguin specimens from the La Meseta and found all to be completely disarticulated and 90% to be broken. Jadwiszczak (2006a: 7) indicated the IB/P/B collections include “more than a thousand almost exclusively isolated bones.” Tambussi et al. (2006) noted that of >2000 Antarctic penguin elements curated in the MLP collections, only one partial skeleton is articulated. This presumably refers to the single incomplete specimen of Palaeeudyptes gunnari described by Accosta Hospitaleche and Reguero (in press). Aside from this specimen, we are aware of only three articulated penguin specimens from Seymour Island, all of which are highly incomplete and none of which preserves the tarsometatarsus (see Ksepka and Bertelli, 2006). Thus, it currently remains impossible to directly identify non-tarsometatarsus elements from the La Meseta to the species level by association (except for the case of Palaeeudyptes gunnari).

Five distinct small penguin humerus morphotypes have been described from the La Meseta Formation: type A1, type A2, type B (sensu Jadwiszczak, 2006a; see Discussion below), a Tonniornis mesetaensis type, and a Tonniornis minimum type. These humerus morphotypes probably represent isolated humeri belonging to six previously named species known only from the tarsometatarsus: Delphinornis arctowskii, Delphinornis gracilis, Delphinornis larseni, Delphinornis wimani, Marambiornis exilis, Mesetaornis polaris. Given the lack of associated material to guide referral of isolated elements to named taxa, however, it is not possible to assign the five humerus morphotypes to previously named species with certainty. In order to provide a rigorous consideration of the character data from these fossils, we conducted a series of 14 iterative analyses alternatively combining codings from these humerus morphotypes with those from the six named species based on tarsometatarsi and also including the humeri as separate terminals (see Additional Analyses). A detailed discussion of taxonomic issues surrounding the La Meseta material is presented below.

Tonniornis mesetaensis and Tonniornis minimum are comparatively recently named species from the La Meseta Formation (Tambussi et al., 2006). The holotype specimen of each species is a single isolated humerus, and these species were diagnosed solely on features of the humerus. These diagnoses do not allow Tonniornis mesetaensis and Tonniornis minimum to be differentiated from six previously named, similarly sized taxa known from tarsometatarsi: Delphinornis arctowskii, Delphinornis gracilis, Delphinornis larseni, Delphinornis wimani, Marambiornis exilis, and Mesetaornis polaris. These six taxa and the two Tonniornis holotypes are known from the same unit of the La Meseta Formation (Telm VII of Sadler, 1988; Submeseta Allomember of Marennsi et al., 1998).

Humeri (i.e., elements directly comparable to the Tonniornis holotypes) have only been formally referred to one of the nine tarsometatarsus-based species. Nine isolated humeri were referred to Delphinornis larseni (Tambussi et al., 2006). Referral of these humeri to Delphinornis larseni is unfortunately not based on associated specimens. The humeri assigned to Delphinornis were differentiated as “conspicuously smaller and more slender than those of Tonniornis” (Tambussi et al., 2006: 155). However, the difference in length between the holotype of Tonniornis minimum and the only complete specimen referred to Delphinornis larseni is <0.5 mm (93.6 versus 93.2 mm: table 1 of Tambussi et al., 2006).

Tambussi et al. (2006) noted that the Tonniornis holotype humeri differ from humeri assigned to Delphinornis, Mesetaornis, and Marambiornis in the unpublished Master's thesis of Kandefer (1994). However, the referrals of Kandefer (1994) were also not based on articulated or associated specimens (Jadwiszczak, personal commun.) and were made before Mesetaornis and Marambiornis were formally named (Myrcha, 2002). Jadwiszczak (2006a) examined the same material studied by Kandefer (1994) and noted that genus and species referrals could not be reliably made. He therefore classified these humeri into three morphotypes (types A1, A2, and B) and concluded that “there is no justification for erecting new taxa based solely on bones other than tarsometatarsi” from the La Meseta Formation given our present state of knowledge (Jadwiszczak 2006b: 296). The A1, A2, and B humerus types differ from one another by discrete features, and all three also differ from the Tonniornis holotype humeri in possessing a weakly bipartite tricipital fossa (this fossa is single in the Tonniornis humeri).

Because none of the five small humerus morphotypes can be firmly associated with a species named from the tarsometatarsus, it is not possible to determine whether the humeri used to erect the Tonniornis species belong to previously named taxa or do in fact represent distinct species, or to determine which species the type A1, A2, or B humeri might represent. Further hindering attempts at species referrals for these isolated humeri, intraspecific size variation (as well as interspecific variation) in the relative lengths of the forelimb and hindlimb elements are significant in penguins. For example, tarsometatarsus length ranges from 34% to 53% of humerus length in extant penguins (Ksepka, personal obs.). This makes matching humeri to tarsometatarsi based on length alone risky at best. Firm referrals await associated specimens as well as further quantification of variation in the La Meseta taxa.

Molecular Data

The molecular dataset utilizes sequences from the nuclear gene RAG-1 (recombination-activating gene 1) and the mitochondrial genes 12S, 16S, COI (cytochrome oxidase I) and cytochrome b, sampled for extant penguins and some outgroup taxa by Baker et al. (2006). Additionally, because we recognize Eudyptes chrysocome, Eudyptes moseleyi, and Eudyptes filholi as distinct species, we included sequences for these taxa provided by Banks et al. (2006). For outgroup taxa unsampled by Baker et al. (2006), we collected available sequences from GenBank (accession numbers and authorship of all sequences are provided in appendix 2). This molecular dataset is largely the same as that employed by Ksepka et al. (2006) and Clarke et al. (2007). For a few outgroup taxa, recently published data were added (e.g., COI sequences for Oceanites oceanicus from Kerr et al., 2007). Sequences for each gene were aligned separately in ClustalX (Thompson et al., 1997), adjusted manually in MacClade (Maddison and Maddison, 1992), and then concatenated.

Search Strategy

Searches were conducted in TNT (Goloboff et al., 2008) and consisted of 10,000 random taxon addition replicates with tree bisection-reconnection branch swapping, saving 10 trees per replicate. All characters were equally weighted and branches with a minimum length of 0 were collapsed. Support values were calculated in PAUP*4.0b10 (Swofford, 2003). Bremer support was calculated using a decay index file constructed in MacClade and executed in PAUP*4.0b10, with the same search strategy as the primary analysis applied.

Additional Analyses: Testing New Hypotheses Regarding the Crown Penguin Radiation

One major debate in penguin evolution revolves around the timing of the crown radiation (see additional discussion in Recommended Calibration Points for Divergence Estimation). Molecular studies have placed the basal divergence in crown Spheniscidae in the Eocene (Baker et al., 2006), whereas the late appearance of crown penguins in the fossil record suggests this divergence occurred more recently, probably during the Miocene (Ksepka et al., 2006; Clarke et al., 2007). Recently, Jadwiszczak (2006b) proposed that some species of the small penguin taxa Delphinornis, Marambiornis, and Mesetaornis from the Eocene La Meseta Formation of Seymour Island (Antarctic Peninsula) could represent the last common ancestor of extant Spheniscidae. This hypothesis is intriguing, particularly because the age of these fossil taxa would be consistent with the molecular dating studies and biogeographic reconstructions of Baker et al. (2006), which estimate that the basal divergence within crown Spheniscidae occurred ~40 million years ago in Antarctica. We conducted additional iterative phylogenetic analyses designed to test this hypothesis as outlined below.

Jadwiszczak (2006b: 298) specifically hypothesized that “small penguins from the Eocene of Antarctica (Delphinornis, Marambiornis, and Mesetaornis) are promising candidates for ancestors of extant Sphenisciformes.” Jadwiszczak (2006b) considered these taxa to be possible candidates for the ancestor of Spheniscidae because their estimated body size range falls within that of living penguins, whereas he ruled out the “giant” La Meseta Formation taxa as extremely specialized lineages representing evolutionary dead ends (Jadwiszczak, 2001, 2003). Jadwiszczak (2006b) gave special mention to Delphinornis arctowskii because this species possesses a tarsometatarsus similar in proportions to extant penguins (see Myrcha et al., 2002), and he presented characters in support of his hypothesis, including the derived coalescence of the intermediate hypotarsal crests of the tarsometatarsus (primitively separated by a groove).

Jadwiszczak's (2006b) hypothesis was also based partially on observations from isolated small penguin humeri from the La Meseta Formation (humerus morphotypes A1, A2, and B). These humeri preserve a bipartite tricipital fossa, a feature that is is present in crown Spheniscidae, but also some stem taxa (i.e., Palaeospheniscus, Eretiscus, Dege, Marplesornis). Although the character data in these isolated humeri are important, it is not yet possible to assign these humeri to any particular species because all of the plausible candidate species (Delphinornis arctowskii, Delphinornis gracilis, Delphinornis larseni, Delphinornis wimani, Marambiornis exilis, and Mesetaornis polaris) are known only from tarsometatarsi (see Treatment of Problematic Fossils above). We therefore designed a set of iterative analyses exploring different ways of incorporating character codings from these humeri into the phylogenetic dataset.

Jadwiszczak's (2006a) humerus morphotypes A1, A2, and B are differentiable, but they are taxonomic equivalents when evaluated for the character matrix used here. We therefore selected the most complete humerus preserving the morphologies central to Jadwiszczak's (2006b) hypothesis (IB/P/B-0382, a type B specimen) for evaluation. In a series of six analyses, codings from this representative humerus were iteratively combined with those from the holotype and referred tarsometatarsi of six La Meseta species (i.e., Delphinornis arctowskii, Delphinornis gracilis, Delphinornis larseni, Delphinornis wimani, Marambiornis exilis, and Mesetaornis polaris). In one additional analysis, IB/P/B-0382 was scored as a separate terminal to model the possibility that the type A1, A2, and B humeri belong to distinct taxa that are as yet unknown from other elements.

As discussed above, we think that it is highly likely that the holotype humeri of Tonniornis mesetaensis and Tonniornis minimum belong to previously named species known from the tarsometatarsus (i.e., some species of Delphinornis, Marambiornis, and/or Mesetaornis). Unfortunately, we were able to assess only line drawings and descriptions (Tambussi et al 2006) of the Tonniornis mesetaensis and Tonniornis minimum holotypes for this study. These species share identical codings for our matrix. In another series of six analyses, representative codings from the holotype of Tonniornis mesetaensis (MLP 93-X-1-145) were iteratively combined with those from the holotype and referred tarsometatarsi of Delphinornis arctowskii, Delphinornis gracilis, Delphinornis larseni, Delphinornis wimani, Marambiornis exilis, and Mesetaornis polaris to create new terminals for each species. Finally, we conducted one analysis including Tonniornis mesetaensis and Tonniornis minimum as separate operational taxonomic units to test the possibility that they represent distinct taxa.

All 14 permutations of the data outlined above were evaluated using the same matrix, outgroups, and search strategies as the primary analysis.

Primary Analysis

The primary analysis resulted in 420 most parsimonious trees (MPTs) of 4492 steps. In the strict consensus of these trees (fig. 11), monophyletic Sphenisciformes and Procellariiformes are recovered. Outgroup relationships are the same as those reported by Clarke et al. (2007: see supplemental data) and agree with Ksepka (2007) except for greater resolution within Diomedeidae. Although the matrix was not designed specifically to consider fine-scale relationships within Procellariiformes, it is worth noting that two of the four traditional families within Procellariiformes are supported as monophyletic: Diomedeidae (albatrosses) and Hydrobatidae (storm petrels). Monophyly of Pelecanoididae (diving petrels) was not tested, as only a single species (Pelecanoides urinatrix) from this group was included. However, the monophyly of this monogeneric family of highly apomorphic diving birds has never been questioned. Procellariidae (shearwaters, fulmars, giant petrels, and allies) as traditionally defined is rendered nonmonophyletic because Pachyptila is recovered as the sister taxon of Pelecanoididae, and other traditional members of Procellariidae (Procellaria, Pterodroma, Puffinus, Daption, and Macronectes) are unresolved with respect to this clade. This finding in particular deserves further scrutiny, and increased sequence representation within Procellariiformes is desirable to confirm or refute the relationships recovered here.

Relationships within Sphenisciformes are well resolved. Support values are low for most branches, but this is not unexpected given the large proportion of missing data for many fossil taxa. Perudyptes devriesi is recovered as a relatively basal member of Sphenisciformes. Placement of this species and other taxa shared in common between the current analysis and that of Clarke et al. (2007) are the same, with the exception that the positions of Icadyptes salasi, Pachydyptes ponderosus, and Burnside “Palaeeudyptes” are less resolved in the strict consensus cladogram. These three taxa form a polytomy with Palaeeudyptes gunnari, Palaeeudyptes klekowskii, and more crownward penguins in the present result, whereas they were placed one node closer to the crown clade than Palaeeudyptes gunnari and Palaeeudyptes klekowskii in the results of Clarke et al. (2007). No codings were changed for these taxa, so this difference may stem from the addition of new taxa and characters, more complete codings for several fossil taxa (reflecting newly described specimens), and a single modified coding for Perudyptes devriesi (coding for character 200 changed from 2 to ?; see description of tarsometatarsus above). Given the congruence with the topology of Clarke et al. (2007), we concentrate discussion on the placements of newly sampled taxa.

The analysis supports the monophyly of Delphinornis, here proposed to also include the taxon “Archaeospheniscus” wimani (see Taxonomic Implications below). Delphinornis wimani is nested within a clade of small Eocene Antarctic penguins, including Delphinornis larseni and the newly included Delphinornis arctowskii and Delphinornis gracilis in our result. Delphinornis wimani shares no close relationships with any species of Archaeospheniscus in out results, although another newly added small penguin, Duntroonornis parvus from the Oligocene of New Zealand, is placed in a polytomy with Archaeospheniscus lowei, Archaeospheniscus lopdelli, and a clade of more crownward penguins.

The rather poorly known African fossil Dege hendeyi (represented only by the humerus and tarsometatarsus) is placed just outside of Spheniscidae, a relationship that merits further evaluation should more material become available for study in the future. At present, only a single plesiomorphic character state (153: 1), moderate separation of the m. supracoracoideus and m. latissimus dorsi insertion scars on the humerus, supports exclusion of Dege hendeyi from the crown clade. Thus, we view the phylogenetic position of this taxon as tentative. It is noteworthy that if Dege hendeyi does in fact represent a stem member of Sphenisciformes, a minimum of two invasions of Africa by penguins would be supported (one by the lineage leading to Dege hendeyi and a second by the extant Spheniscus demersus lineage). The status of the poorly known African fossils Inguza predemersus, Nucleornis insolitus, and “Palaeospheniscus” huxleyorum is summarized in table 2 below. Olson (1983) considered these four species valid, but probably congeneric. The phylogenetic positions of these taxa cannot be resolved sufficiently to provide additional insight into African penguin taxonomy or biogeography at present. Only a single penguin species, Spheniscus demersus (Jackass Penguin), occurs in coastal South Africa today. Despite early uncertainty in the precise age of the four fossil species mentioned above, all are now known to occur in the early Pliocene, supporting a significantly higher level of penguin diversity in Pliocene Africa than in the present day (Dingle, 1983; Olson, 1983).

TABLE 2

Most exclusive placement of incomplete fossils supported by synapomorphies identified in the primary phylogenetic analysis

Numbers in placement column refer to the clades labeled in figure 21. When interpreting table 2, it is important to bear in mind that the placements presented are the most exclusive that can be justified by preserved synapomorphies, not estimates of precise phylogenetic position. Region abbreviations: AF, Africa; AP, Antarctic Peninsula; AU, Australia; NZ, New Zealand; SA, South America.

Relationships among the three extant Rockhopper Penguin groups recently recognized as species agree with those reported by Banks et al. (2006) in that the Northern Rockhopper, Eudyptes moseleyi, is sister taxon to a clade uniting the Southern Rockhopper, Eudyptes chrysocome, and Eastern Rockhopper, Eudyptes filholi. The implications of this topology were discussed by Banks et al. (2006). More detailed comments on critical sectors of the tree are presented below.

Identifying Extinct Parts of the Crown Penguin Radiation: Congruence and Conflict with Previous Studies

Our analysis supports placement within the extant penguin radiation for three recently described fossil penguins: Spheniscus muizoni (Göhlich, 2007), Madrynornis mirandus (Acosta Hospitaleche et al., 2007), and Pygoscelis grandis (Walsh and Suárez, 2006). Spheniscus muizoni has not been previously included in a phylogenetic analysis, although Göhlich (2007) listed characters supporting the assignment of this fossil to the genus Spheniscus. This assignment is corroborated here, although the relationships of Spheniscus muizoni relative to other extinct species of Spheniscus (i.e., Spheniscus urbinai, Spheniscus megaramphus, and Spheniscus chilensis) remain unresolved. The estimated age for this taxon (~11–13 Ma; Göhlich, 2007) makes it the oldest confirmed record of crown penguins (see further discussion below). Spheniscus chilensis, Spheniscus urbinai, and Spheniscus megaramphus are also recovered as members of the Spheniscus clade, although all are younger than Spheniscus muizoni.

Madrynornis mirandus, one of the most complete fossil penguins yet discovered, was previously identified as the sister taxon of extant Eudyptes by Acosta Hospitaleche et al. (2007, 2008). However, these authors reported a significantly different phylogeny (fig. 12) than that recovered in the present study. Because those results supported inclusion of the early Miocene taxon Palaeospheniscus in the crown clade and the nonmonophyly of the extant genus Pygoscelis, they are directly relevant to our discussion of fossil calibration points and taxonomy. Specifically, if this alternative topology is correct, Palaeospheniscus (Gaiman Formation: early Miocene–earliest middle Miocene; Flynn and Swisher, 1995; Palazzesi et al., 2006) would represent the oldest crown spheniscid fossil.

Some differences between our results and those of Acosta Hospitaleche et al. (2007, 2008) can be accounted for by the effect of the different preferred rooting for crown penguins when molecular data are included (discussed by Bertelli and Giannini, 2005). However, reexamination of the character evidence suggests that three major differences are driven by character sampling issues and erroneous scorings. These include the placement of Megadyptes as sister taxon to a clade uniting Pygoscelis, Aptenodytes, and Palaeospheniscus (versus as sister taxon to Eudyptes + Madrynornis in the present study), the paraphyly of extant Pygoscelis (versus monophyly in the present study), and the placement of Palaeospheniscus in Spheniscidae (versus as a stem sphenisciform in the present study).

Exclusion of >100 osteological and soft tissue characters from previous studies that could not be scored in sampled fossil taxa (Paraptenodytes, Palaeospheniscus, and Madrynornis) from the analyses of Acosta Hospitaleche et al. (2007,2008) may account for some differences in topology. Rather than improving accuracy or resolution, exclusion of characters with missing data is likely to reduce accuracy (e.g., Wiens, 1998). Bertelli et al. (2006) showed that in the case of penguins, excluding particular classes of characters (e.g., integumentary, osteological, molecular) resulted in lower resolution and the failure to recover clades that are well supported by both the full morphological dataset and the combined dataset. Exclusion of synapomorphies grouping Megadyptes and Eudyptes (e.g., shelf of bone bounding the salt gland fossa, yellow crown feathers) and supporting the monophyly of extant Pygoscelis (e.g., fusion of the synsacrum and ilium) appears to account for the failure to recover these clades. We consider the monophyly of all extant Pygoscelis species and the monophyly of the clade uniting Megadyptes and Eudyptes to the exclusion of all other extant penguins to be well supported based on the strong molecular support for these clades reported by Baker et al. (2006), morphological support reported by others (Giannini and Bertelli, 2004; Bertelli and Giannini, 2005; Walsh and Suárez, 2006), and the results of the analyses presented in this paper.

Palaeospheniscus is represented by three species that occur in the Miocene of South America. In the present study, we found all Palaeospheniscus species to form a clade that also includes Eretiscus tonnii, placed outside of crown Spheniscidae (fig. 10). In contrast, Acosta Hospitaleche et al. (2007, 2008) placed Palaeospheniscus within the crown clade, as sister taxon to extant Aptenodytes. Although Acosta Hospitaleche et al. (2008) commented that Bertelli et al. (2006) proposed the same set of relationships, Bertelli et al. (2006) did not include Palaeospheniscus in their analyses or mention this taxon in their discussion.

Figure 10

Reconstruction of Perudyptes devriesi diving, with preserved elements of the holotype indicated (left humerus, pelvis and left femur not shown). Silhouette by Kristin Lamm.

Figure 11

Strict consensus of 420 most parsimonious trees of 4492 steps fom the primary analysis. Bremer support values are indicated above branches. Support values reflect the low proportion of codable informative cells for several fossil taxa and the high proportion of missing data in all fossil taxa (all 6000+ molecular characters are missing data for all fossils). For a more complete understanding of support within extant clades, see Bertelli and Giannini (2005) and Baker et al. (2006).

Figure 12

Single most parsimonious tree from analysis of 44 morphological characters by Acosta Hospitaleche et al. (2007, 2008).

Plesiomorphic character states for two unambiguous synapomorphies of the crown clade support exclusion of Palaeospheniscus from Spheniscidae: m. latissimus dorsi and m. supracoracoideus insertion scars on the humerus separated by a moderate gap (character state 153: 1) and ulnar condyle of humerus rounded and projected (character state 164: 0). In crown Spheniscidae, the m. latissimus dorsi and m. supracoracoideus insertion scars are nearly confluent (character state 153: 2) and the ulnar condyle is flattened (character state 164: 1). Additionally, Palaeospheniscus possesses a shallow sulcus between metatarsals III and IV (character state 197: 1), whereas a deep sulcus (character state 197: 2) is optimized as a synapomorphy of the clade uniting Dege hendeyi, Marplesornis novaezealandiae, and Spheniscidae (reversed in Pygoscelis).

The incorporation of 12 vertebrae into the synsacrum of Palaeospheniscus bergi (see Simpson, 1946) may also represent a primitive feature given that the most basal known penguin, Waimanu manneringi, possesses 11 synsacral vertebrae (Slack et al., 2006) and extant penguins typically possess 13 synsacral vertebrae. We observed 14 synsacral vertebrae in Pygoscelis papua and some exemplars of Eudyptes robustus, and 12 synsacral vertebrae in Eudyptula minor (these totals are optimized as apomorphic increases/decreases on our strict consensus tree). Thus, there appears to be a trend toward an increase in number of synsacral vertebrae in penguins. Unfortunately, complete synsacra remain unknown for almost all fossil taxa, and synsacral vertebral count varies significantly amongst outgroup taxa, precluding a full understanding of the phylogenetic distribution of this character.

Several characters proposed to support placing Palaeospheniscus in the crown clade (Acosta Hospitaleche et al., 2007) were scored differently in our analysis or omitted as uninformative. Absence of the medial proximal vascular foramen of the tarsometatarsus (character state 40: 1 of Acosta Hospitaleche et al., 2007) was proposed as an unambiguous synapomorphy of Pygoscelis, Palaeospheniscus, and Aptenodytes. However, this foramen is actually present in all species of Pygoscelis and Aptenodytes (character 200 of this study; see also Watson, 1883: plate VII; Bertelli and Giannini, 2005: fig. 27; Walsh and Suárez, 2006: fig. 2). Additionally, this foramen is present in all other crown taxa. Therefore, its absence in Palaeospheniscus does not support the proposed placement of this fossil taxon. Lateral proximal vascular foramen of the tarsometatarsus “present in posterior view” (character state 41: 1 of Acosta Hospitaleche et al., 2007) was proposed as a synapomorphy of Megadyptes, Pygoscelis, Aptenodytes, and Palaeospheniscus. However, the lateral proximal vascular foramen is present and visible in posterior (plantar) view in Palaeospheniscus, all penguin taxa included in the analysis of Acosta Hospitaleche et al. (2007), and all outgroup taxa (see Bertelli and Giannini, 2005: fig. 27; Ksepka et al., 2006: fig. 15), rendering it uninformative for the sampled taxa. Given these observations and the characters discussed above, we consider the stem position for Palaeospheniscus recovered here to be the more strongly supported hypothesis.

Pygoscelis grandis was placed in the crown clade in the results of our analysis. However, because this fossil is placed in a polytomy including extant Pygoscelis and a large clade of non-Pygoscelis taxa, the monophyly of Pygoscelis grandis plus extant Pygoscelis species could not be confirmed. In their original description of Pygoscelis grandis, Walsh and Suárez (2006) found this species to be the sister taxon of a clade uniting the three extant Pygoscelis species (fig. 13). This topology is present in a subset of the most parsimonious trees recovered in the present study. Two alternative placements of Pygoscelis grandis were also found to be equally parsimonious: sister taxon to the clade uniting Spheniscus, Eudyptula, Megadyptes, Madrynornis, and Eudyptes or sister taxon to a more inclusive clade uniting those taxa and extant Pygoscelis. Recovery of these alternative arrangements may be the result of differences in the morphological character sets between our study and that of Walsh and Suárez (2006), changing character optimizations due to the inclusion of stem penguin fossils in our analysis, or the influence of molecular data on crown penguin relationships. The results of the present analysis indicate that Pygoscelis grandis possesses an interesting combination of character states, and suggest that future finds, particularly of cranial material, may yield better insight into its affinities. Pygoscelis grandis lacks partial or complete fusion of the ilia to synsacrum (character states 180: 1–2), a synapomorphy shared by all extant Pygoscelis species. Hence, if Pygoscelis grandis is part of a clade including the extant Pygoscelis species, it almost certainly represents a divergence basal to any extant member of the Pygoscelis lineage, as hypothesized by Walsh and Suárez (2006).

Figure 13

Strict consensus cladogram from analysis of 37 morphological characters by Walsh and Suárez (2006), designed to test the relationships of Pygoscelis grandis.

Evaluating Potential “Direct Ancestors” of Spheniscidae from the Eocene

In the results of the primary analysis, the small Eocene Antarctic taxa Delphinornis, Mesetaornis, and Marambiornis were placed well outside the crown clade. However, because these taxa are known with certainty only from one bone (the tarsometatarsus), support for their placement is necessarily weak. Iterative analyses (details in Additional Analyses above) explore the effects of adding data from isolated humeri that may belong to these taxa. This allows investigation of Jadwiszczak's (2006b) hypothesis that these penguins represent ancestors of the Spheniscidae by evaluating their placement relative to the crown. These iterative analyses also provide a test of the different possible species assignments for the Tonniornis humeri (see Treatment of Problematic Fossils above).

Analysis of the permutation of the dataset adding the type B humerus (IB/P/B-0382) morphotype codings to the Delphinornis arctowskii terminal resulted in an increase in the number and length of most parsimonious trees recovered (2642 MPTs, 4494 steps). The strict consensus tree (fig. 14) is congruent with, but less resolved than, the tree obtained from the primary analysis. All taxa placed intermediate between Waimanu and clade 8 (see fig. 21), including the modified Delphinornis arctowskii terminal, are collapsed into a polytomy. Analyses adding codings from the type B humerus to the terminals for the other three species of Delphinornis resulted in strict consensus trees identical to the strict consensus tree recovered in the Delphinornis arctowskii permutation, although the number of MPTs varied somewhat among the Delphinornis gracilis (1920 MPTs, 4494 steps), Delphinornis larseni (2170 MPTs, 4494 steps), and Delphinornis wimani (2832 trees, 4494 steps) results.

Figure 14

Results of iterative analyses testing the effect of assigning character codings from isolated elements to species terminals. Strict consensus of 2642 MPTs of 4494 steps from the analysis applying codings from IB/P/B-0382 (type B humerus) to the Delphinornis arctowskii terminal. An identical strict consensus topology, with some variation in number of MPTs, resulted from the permutations adding the type B humerus codings to the Delphinornis gracilis (1920 MPTs, 4494 steps), Delphinornis larseni (2170 MPTs, 4494 steps), and Delphinornis wimani (2832 trees, 4494 steps) terminals.

Analysis of the permutation of the dataset adding the type B humerus codings to the Marambiornis exilis terminal resulted in 1503 MPTs of 4494 steps. The strict consensus tree (fig. 15) differs from those from the Delphinornis permutations in supporting one branch (that uniting Duntroon Palaeeudyptes + clade 8) that was recovered in the primary analysis, but collapsed in the Delphinornis permutations. Analysis of the permutation by adding B humerus codings to the Mesetaornis polaris terminal also resulted in 1503 MPTs of 4494 steps (Marambiornis exilis and Mesetaornis polaris share identical sets of character codings).

Figure 15

Results of iterative analyses testing the effect of assigning character codings from isolated elements to species terminals. Strict consensus of 1503 MPTs of 4494 steps from the analysis applying codings from IB/P/B-0382 (type B humerus) to the Marambiornis exilis terminal. An identical strict consensus topology resulted from the permutations adding the type B humerus codings to the Mesetaornis polaris terminal.

When the type B humerus (IB/P/B-0382) is treated as a separate terminal, 480 MPTs of 4494 steps are recovered. In the strict consensus tree (fig. 16), the type B humerus terminal does not cluster with Delphinornis, Mesetaornis, or Marambiornis, but is instead recovered as part of clade 6 (see fig. 21). The strict consensus is otherwise similar to that from the primary analysis, with the exception that the branch uniting Duntroon Palaeeudyptes + clade 8 is collapsed. As mentioned above, the type A1 and type A2 humerus morphotypes of Jadwiszczak (2006a) are taxonomic equivalents (sensu Wilkinson, 1995) to the type B humerus terminal when coded into our matrix from preserved material.

Figure 16

Strict consensus of 480 MPTs of 4494 steps from the analysis including the type B humerus (IB/P/B-0382) as a separate terminal.

Analysis of the permutation of the dataset adding codings from the Tonniornis mesetaensis holotype humerus (MLP 93-X-1-145) to the Delphinornis gracilis terminal also resulted in an increase in the number and length of most parsimonious trees recovered (1920 MPTs, 4493 steps) relative to the primary analysis. In the strict consensus of these trees (fig. 17), all four Delphinornis terminals shifted to a slightly more crownward position. These terminals form a polytomy with Pachydyptes ponderosus, Icadyptes salasi, Palaeeudyptes gunnari, Palaeeudyptes klekowskii, Duntroon “Palaeeudyptes,” Burnside “Palaeeudyptes,” and clade 8 (see fig. 21). Additionally, the branch supporting a more crownward position for Perudyptes devriesi than Mesetaornis polaris and Marambiornis exilis collapses. For the permutation adding codings from the Tonniornis mesetaensis humerus to the Delphinornis larseni terminal, the strict consensus tree (2090 MPTs, 4493 steps) is identical in topology to that from the Delphinornis gracilis permutation.

Figure 17

Results of iterative analyses testing the effect of assigning character codings from isolated elements to species terminals. Strict consensus of 1920 MPTs of 4493 steps from the analysis applying codings from the Tonniornis mesetaensis holotype humerus to the Delphinornis gracilis terminal. In the results of the permutation for Delphinornis larseni (6390 trees of 4493 steps), the same strict consensus tree was recovered.

Permutations adding codings from the Tonniornis mesetaensis holotype humerus (MLP 93-X-1-145) to the Delphinornis arctowskii terminal resulted in 3090 MPTs of 4493 steps. The strict consensus of these trees (fig. 18) is less resolved than the permutations for the other Delphinornis species. All taxa placed intermediate between Waimanu and clade 8 in the primary analysis are collapsed into a polytomy, with the exception that the monophyly of Anthropornis is supported. For the permutations adding codings from the Tonniornis mesetaensis humerus to the Delphinornis wimani terminal, the strict consensus tree (2850 MPTs, 4493 steps) is identical in topology to that from the Delphinornis gracilis permutation.

Figure 18

Results of iterative analyses testing the effect of assigning character codings from isolated elements to species terminals. Strict consensus of 3090 MPTs of 4493 steps from the analysis applying codings from the Tonniornis mesetaensis holotype humerus to the Delphinornis arctowskii terminal. In the results of the permutation for Delphinornis wimani (2850 trees of 4493 steps), the same strict consensus tree was recovered.

Permutations adding codings from the Tonniornis mesetaensis humerus to the Mesetaornis polaris terminal resulted in 930 MPTs of 4493 steps. In the strict consensus tree (fig. 19), all taxa placed intermediate between Delphinornis and clade 7 in the primary analysis are collapsed into a polytomy, with the exception that the monophyly of Anthropornis is supported. Results were identical for the Marambiornis exilis permutation.

Figure 19

Results of iterative analyses testing the effect of assigning character codings from isolated elements to species terminals. Strict consensus of 930 MPTs of 4493 steps from the analysis applying codings from the Tonniornis mesetaensis holotype humerus to the Mesetaornis polaris terminal. An identical strict consensus topology resulted from the permutations adding the Tonniornis mesetaensis humerus codings to the Mesetaornis polaris terminal.

Finally, the analyses including the holotypes of Tonniornis mesetaensis and Tonniornis minimum as operational taxonomic units result in 1400 most parsimonious trees of 4492 steps, placing both fossils in a large polytomy with other members of clade 6 in the strict consensus tree (fig. 20).

Figure 20

Strict consensus of 1400 MPTs of 4492 steps from the analysis including the Tonniornis mesetaensis and Tonniornis minimum holotype humeri as separate terminals.

The failure of the small humeri to cluster with any of the La Meseta penguin taxa founded on tarsometatarsi could result from two phenomena. Either (1) all described humeri from the La Meseta Formation represent taxa that are unrelated to Delphinornis, Mesetaornis, and Marambiornis, or (2) homoplasy is present, resulting in different placements for isolated humeri and isolated tarsometatarsi belonging to the same species. The first case would require that humeri of Delphinornis, Mesetaornis, and Marambiornis remain either unpreserved or undiscovered, which we consider unlikely given the large collections of tarsometatarsi from these taxa (e.g., Wiman, 1905a, 1905b; Simpson, 1971c; Myrcha et al., 2002). We consider it more likely that the effect of homoplastic characters, combined with the scarcity of informative characters codable from single elements, accounts for this situation. Given the presence of six small- to medium-sized penguin species based on tarsometatarsi, the five differentiable small- to medium-sized humerus types (A1, A2, B, Tonniornis mesetaensis, and Tonniornis minimum) are most conservatively accounted for as representatives of these species.

Despite difficulties stemming from taxonomic uncertainties, some important conclusions can be drawn. Critically, all coding strategies resulted in Delphinornis, Mesetaornis, and Marambiornis, as well as IB/P/B-0382 and the Tonniornis humeri when treated as separate terminals, being placed several nodes basal to the crown clade. Our analyses suggest that regardless of what taxonomic status one affords these isolated specimens, all of the Eocene La Meseta fossil penguins are comparatively distantly related to the crown clade. The placement of Delphinornis, Mesetaornis, and Marambiornis near the base of the sphenisciform tree in the primary analysis, despite relying on sparse morphological data, is the most parsimonious solution that relies only on definitively assignable elements and further agrees well with stratigraphic distribution. This placement is unambiguously supported by the primitive absence of a sulcus between metatarsals II and III (character state 197: 0) and primitive presence of a distal vascular foramen (character state 201: 0) for all species, and additionally by the primitively less shortened tarsometatarsus (character 194: 1) for Delphinornis gracilis, Delphinornis larseni, Marambiornis exilis, and Mesetaornis polaris. Bifurcation of the tricipital fossa of the humerus is most parsimoniously interpreted as homoplasy in the taxa represented by the type A1, A2, and B humeri and clade 11 (Palaeospheniscus, Eretiscus, Dege, Marplesornis, and Spheniscidae), regardless of whether the type A1, A2, and B humeri are considered to belong to Delphinornis, Mesetaornis, or Marambiornis or are considered to represent distinct taxa.

The situation addressed here underscores the challenges of working with isolated remains, a problem that has historically hindered studies of penguin evolution (discussed by Fordyce and Jones, 1990) but is increasingly being alleviated by discoveries and descriptions of more complete specimens (e.g., Slack et al., 2006; Clarke et al., 2007; Ando, 2007; Acosta Hospitaleche et al., 2007, 2008). Results from the iterative evaluations of isolated elements reveal potential instability in the placement of Delphinornis, Mesetaornis, and Marambiornis and suggest that caution in both taxonomic and phylogenetic interpretations of the La Meseta penguins is needed until articulated materials become available to resolve these issues. Specifically, we note that because the inclusion of data from exemplar humeri results in widened array of equally most parsimonious positions for Delphinornis, Mesetaornis, and Marambiornis, these taxa should be regarded as poorly phylogenetically constrained pending discovery of more complete remains.

A New Consideration of Fragmentary Fossils: Implications for the Timing of Penguin Evolution

Although incompleteness may preclude full resolution of the phylogenetic position of many fragmentary penguin fossils, the relationships of these fossils are relevant to understanding the timing of the radiation of Sphenisciformes. We evaluated the phylogenetic position of all valid named fossil taxa and several unnamed but important specimens that were not included in the primary analysis by identifying synapomorphies in the strict consensus tree from the primary analysis. Specimens were placed to the most exclusive level supported by the distribution of observable synapomorphies. We also evaluated whether these fossils retained character states optimized as plesiomorphies that support exclusion from the crown clade. In cases where the monophyly of a set of specimens assigned to a species was in question, we considered specimens individually. For example, we evaluated several fossils (e.g., the “Seal Rock Palaeeudyptes” specimen) originally assigned to Palaeeudyptes antarcticus, but no longer considered referable to that taxon (Simpson, 1971b), at the specimen level.

Of 27 poorly known taxa and isolated fossils evaluated, 17 were found to represent stem lineage Sphenisciformes based on the presence of plesiomorphic character states, 4 were found to represent crown Spheniscidae based on synapomorphies, and 6 preserved insufficient character evidence to determine crown/stem status (table 2).

Additional data from more fragmentary penguin fossils considered here are also consistent with a recent radiation of the crown clade. All of the 14 pre-Miocene fossils evaluated can be excluded from Spheniscidae, as they possess plesiomorphic character states unknown in crown taxa. However, one of these fossils contributes stratigraphic data relevant to deeper divergences (reflected in the dotted line in fig. 21). SAMA P14157, a partial wing from the late Eocene of Australia (formerly referred to Pachydyptes or Anthropornis, see Taxonomic Implications below), is supported as a member of clade 7 by one synapomorphy of that clade (character state 173: 1, extension of metacarpal III markedly distal to metacarpal II). All members of clade 7 identified in the primary analysis are Oligocene or younger in age. Therefore, SAMA P14157 extends the known range of clade 7 by several million years.

Figure 21

Phylogeny of Sphenisciformes calibrated to the stratigraphic record. Ghost lineages (Norell, 1992) are minimized. Dotted lines extending the range of clade 7 are based on identification of fragmentary fossils of late Eocene age to this clade. We utilize age ranges for Spheniscus urbinai and Spheniscus megaramphus provided by Stucchi (2007). Most penguins from the La Meseta Formation occur in Telm 7 (34.2–36.1 Ma), but a few species have been reported from lower units (see Myrcha et al., 2002; Jadwiszczak, 2006a), accounting for the extended ranges of Delphinornis larseni, Anthropornis grandis, and Palaeeudyptes gunnari. A specimen from the Chilcatay Formation (see fig. 1) referred to Palaeospheniscus sp. by Acosta Hospitaleche and Stucchi (2005) is provisionally referred to Palaeospheniscus patagonicus (Ksepka, 2007), accounting for the longer range of this taxon. For taxa that are poorly constrained in age, the midpoint of possible ages is used (e.g., Marplesornis is placed at the late Miocene). Geographical locality for fossil taxa and breeding range of extant taxa are provided in parentheses following taxa names. Asterisks indicate taxa with extensive breeding ranges including oceanic islands (see Bertelli and Giannini [2005] for more data on these species). Abbreviations: AF, Africa; AN, Antarctica; AU, Australia; NZ, New Zealand and surrounding islands; SA, South America.

Six fossils cannot be included in, or excluded from, Spheniscidae based on preserved character states. However, none of these fossils can be demonstrated to be older than Spheniscus muizoni. Therefore, even if future discoveries were to identify one or more of them as representing Spheniscidae, it would not extend the stratigraphic range of the crown clade. Among these six fossils are the holotypes of two New Zealand taxa (Aptenodytes ridgeni and Pygoscelis tyreei). Both of these specimens were both collected from eroded blocks found loose on a beach, and they are thus poorly constrained in age. A range of ages spanning from middle Miocene to early Pleistocene have been considered plausible for these fossils (Simpson, 1972b). Most recently Feldmann and Keyes (1992) noted that a middle-late Miocene age for these fossils was most likely, although not certain. Aptenodytes and Pygoscelis are today restricted to higher latitudes, and so if correctly identified to these clades, the fossil species would extend the geographic ranges of these taxa (Simpson, 1972b). Aptenodytes ridgeni may belong within Aptenodytes given overall resemblance (Simpson, 1972b), but we cannot identify any unambiguous synapomorphies of extant Aptenodytes in the partial hindlimb holotype. Likewise, whether Pygoscelis tyreei forms a clade with extant Pygoscelis cannot be confirmed or refuted given the available material. Simpson (1972b: 159) also considered the affinities of these two species to be “not absolutely certain.”

Four fragmentary fossils are positively supported as members of crown Spheniscidae (clade 13) based on synapomorphies. All of these taxa are Miocene or younger in age. Two of these taxa, Inguza predemersus and Tereingaornis moisleyi, were originally considered to represent new species of the extant genus Spheniscus (Simpson, 1971c; see discussion in Scarlett, 1983). Relationships of Inguza predemersus cannot currently be resolved relative to other crown penguin taxa, due to the paucity of known material. Tereingaornis moisleyi lacks one unambiguous synapomorphy of Spheniscus (character state 162: 1, posterior trochlear ridge of humerus fails to extend beyond the ventral margin of the humeral shaft). This taxon further has a distinctly shaped humeral shaft with a prominently curved ventral margin, supporting its separation from Spheniscus and also differentiating it from other extant penguin genera. The third fossil assigned here to Spheniscidae, Pygoscelis calderensis, is known from three partial skulls. These skulls preserve weakly developed temporal fossa (character state 83: 1) and a shelf of bone bounding the lateral margin of the salt gland fossa (character state 81: 1) (Acosta Hospitaleche et al., 2006). These morphologies are observed in combination only in extant Pygoscelis. However, given the current optimizations of these character states in the strict consensus tree, their presence in Pygoscelis calderensis would also be consistent with placement of Pygoscelis calderensis as the sister taxon to clade 14 or sister taxon to clade 15. Thus, the monophyly of Pygoscelis calderensis plus extant Pygoscelis cannot be supported or refuted given current data.

The last of the four extinct taxa identified to the crown clade is the late Holocene (760 ± 60 ybp) subfossil taxon Tasidyptes hunteri (van Tets and O'Conner, 1983). Whether Tasidyptes hunteri represents a diagnosable species remains debatable (Fordyce and Jones, 1990) due to the incomplete nature of the fossil and the tenuous link between holotype and referred specimens. Some of the hypodigm elements are indistinguishable from corresponding elements of extant Eudyptes, and it is possible that these represent a vagrant individual of Eudyptes sp. (Ksepka, 2007). Given the young age of Tasidyptes hunteri and successful recovery of sequence data from bones of Megadyptes waitaha dated to 600–1500 ybp (Boessenkool et al., 2009), recovery of sequence data may be one possible avenue to firmly resolving the status of this taxon.

Pre-cladistic assessments of penguin relationships (e.g., Simpson, 1946; Marples, 1952, 1962) implied that pre-Miocene fossil taxa were not closely related to extant penguins. All such taxa were placed in subfamilies separate from extant penguins. The present study confirms Simpson's (1946) hypothesis that the fossil taxa known at the time he created his classification scheme are indeed not part of the crown penguin radiation. However, it should be noted that the original subfamily classification of fossil Sphenisciformes (Simpson, 1946), although still occasionally employed, was discarded by Simpson (1971b) himself when he concluded that these subfamilies did not reflect evolutionary relationships. Phylogenetic analyses support Simpson's (1971b) decision. Of the four extinct subfamilies originally proposed, only Palaeospheniscinae has been supported as monophyletic, while Palaeeudyptinae, Anthropornithinae, and Paraptenodytinae are polyphyletic (Ksepka et al., 2006; Ksepka, 2007).

Taxonomic Implications

The results of the primary analysis support several taxonomic revisions and also highlight some taxonomically problematic areas.

Recommended Calibration Points for Divergence Estimation

Phylogenetic analysis confirms two internal calibration points for molecular sequence-based divergence estimation projects targeting Spheniscidae. Spheniscus muizoni provides a minimum estimate for the Spheniscus-Eudyptula split. The age of the deposits at the Cerro la Bruja locality where the Spheniscus muizoni holotype was collected (Göhlich, 2007) is proposed to be latest middle Miocene or earliest late Miocene (11–13 Ma) based on biostratigraphy (Muizon, 1988; Dunbar et al., 1990). Unfortunately, radiometric dates are unavailable from the relevent deposits and the estimated age range can not yet be rigorously constrained. More precise dates are unavailable at present, so ideally stratigraphic uncertainty should be considered when applying this calibration point.

Madrynornis mirandus provides a minimum estimate for the age of the Eudyptes-Megadyptes split. The single known specimen, comprising most of a skeleton, was collected from the “Entrerriense” sequence of the Puerto Madryn Formation (Acosta Hospitaleche et al., 2007). This sequence was deposited at ~10.0 ± 0.3 Ma based on ⁸⁷Sr/⁸⁶Sr dates obtained from fossil mollusks (Scasso et al., 2001).

The oldest specimens of Pygoscelis grandis were collected from the Bonebed Member of the Bahía Inglesa Formation ~7 m below an ash layer dated to 7.6 ± 1.3 Ma (Walsh and Suárez, 2006). Given the unresolved placement of Pygoscelis grandis in our analysis, we do not recommend it as a calibration point for the divergence of Pygoscelis from other penguin lineages. Indeed, even if the phylogenetic placement of Pygoscelis grandis was fully resolved, the young age of this fossil precludes its use as a calibration point. Older fossils of Spheniscus muizoni and Madrynornis mirandus both calibrate more nested divergences, and they thereby provide an older minimum date for divergence of the Pygoscelis clade.

Waimanu manneringi remains the oldest stem record of Sphenisciformes and provides a calibration for the Procellariiformes-Sphenisciformes divergence at 60.5–61.6 Ma (Slack et al., 2006). This calibration of course represents a minimum, and the true age of this divergence may well be significantly older given the patchy distribution of fossil Sphenisciformes and lack of fossil Procellariiformes in the Paleocene.

Molecular divergence estimates have placed the basal divergence within crown Spheniscidae in the Eocene (~40 Ma: Baker et al., 2006), which would require a dramatically different picture of penguin evolution than presented in this contribution. One possible explanation for the conflict between the fossil record and the molecular divergence estimates is that Paleogene fossils of Spheniscidae remain unsampled. Another potential explanation is that such fossils have been collected, but not yet recognized as belonging to crown penguins. Reevaluation of fragmentary remains using synapomorphies in the present study indicates that the apparent absence of crown clade fossils in the Paleogene is most likely not an artifact of a failure to properly identify less complete specimens. While the fossil record of many avian groups remains patchy, Sphenisciformes have one of the more complete records due to their marine habitat and dense bone structure. Clarke et al. (2007) calculated that to account for an Eocene origination of Spheniscidae, the fossil record of penguins would have to be 172%–205% more incomplete as measured in lengths of ghost lineages (Norell, 1992) than inferred from cladogram topology and stratigraphy alone (range results from taking into account different possible resolutions of polytomies). Such a large amount of inferred missing record seems unlikely, given the reasonably continuous global Cenozoic record of penguins and the strong fit of phylogeny to stratigraphy exhibited by penguins (Clarke et al., 2007; Ksepka, 2007) (fig. 21).

Alternatively, problematic aspects of divergence dating methods potentially contribute to overestimation of the age of Spheniscidae. The ~40 Ma estimates for the origin of the penguin crown clade were obtained using two externally derived calibration points: a 104 Ma estimate for the Galloanserae-Neoaves divergence and a 90 Ma estimate for the Galliformes-Anseriformes divergence (Baker et al., 2006). However, these calibration points are not based on avian fossils, but instead they represent extrapolations derived from a primary reptile-mammal calibration point based on the fossil record and a secondary primate-rodent calibration point (itself extrapolated from the primary reptile-mammal calibration) (van Tuinen and Hedges, 2001). Secondary calibrations have been widely criticized for methodological inconsistencies that may lead to erroneous estimates and artificially narrow envelopes of uncertainty (e.g., Shaul and Graur, 2002; Graur and Martin, 2004). However, these challenges can be overcome with new fossil data that sidestep the problems of secondary calibration points.

The true age of Spheniscidae most likely lies somewhere in between the extremes of the Eocene estimate obtained from divergence dating and the first appearance of the crown clade in the fossil record in the middle-late Miocene. Primary calibration points from stratigraphically and phylogenetically constrained fossils have been advocated as the most appropriate sources of calibrations for divergence studies (e.g., Reisz and Müller, 2004; Müller and Reisz, 2005). Two such well-corroborated fossil calibration points (Spheniscus muizoni and Madrynornis mirandus) have now been identified for future divergence dating studies targeting Sphenisciformes. Such primary internal calibrations may help reduce discordance between the fossil record and molecular estimations.