Seymour (Marambio) Island is located in the Weddell Sea, within the James Ross Basin, a back-arc feature that developed east of a magmatic arc of the Antarctic Peninsula (Fig. 1; e.g., Porębski 2000). The siliciclastic sedimentary rocks of the Eocene–?early Oligocene La Meseta Formation crop out in the north-eastern part of the island (Fig. 1). These uplifted nearshore, estuarine and deltaic, mostly poorly consolidated, sediments are very fossiliferous, and have yielded a diverse marine and terrestrial fauna (including numerous penguins) and flora (Porębski 2000; Myrcha et al. 2002; Marenssi 2006; Tatur et al. 2011; Reguero et al. 2013, references therein). A number of studies revealed that the sediments had come from the west–north-west, hence the source rocks must have been located within the nearby northern Antarctic Peninsula (Reguero et al. 2013, references therein). The recent micromorphological investigation of quartz sand grains (Kirshner & Anderson 2011) supports the dominant view that the La Meseta Formation is devoid of any evidence of glaciation. Hence, it represents the youngest preglacial succession on the Seymour Island (e.g., Marenssi et al. 2002).

The base of the uppermost (or the youngest) unit of the La Meseta Formation, Telm7/Submeseta Allomember, was interpreted by Porębski (2000) as a basinward facies shift (supposedly bay-head delta deposits over those interpreted as the estuary mouth) followed by “transgressive reactivation” (alternately: flooding surfaces of transgressive marine origin and shorefaces). According to Marenssi (2006), the lower part of Telm7 represents the inner estuary channels/shoals, its upper levels the shallow marine environment. For detailed stratigraphy of this unit, consult Myrcha et al. (2002, figure 2) and Reguero et al. (2013, figure 3.1).

Unit Telm7 of the La Meseta Formation has been traditionally regarded as late Eocene in age (Myrcha et al. 2002; Jadwiszczak 2006; Marenssi 2006; Acosta Hospitaleche & Reguero 2010). Some recent works, however, have adopted the view that its lower levels can be dated to the middle Eocene (Reguero et al. 2013, references therein). The youngest documented ages (34.78–33.57 Mya) from the formation are based on strontium isotope stratigraphy and suggest that up to several of the uppermost metres overlaying the unit Telm7 may be as young as the earliest Oligocene (Ivany et al. 2006). Unfortunately, the GPS coordinates for the sampling locality reported by Ivany et al. (2006) are very inaccurate and do not allow locating the described section (see Marenssi et al. 2010).

The new material reported here is a short fragment of the spinal column of an early penguin from the Eocene–?early Oligocene La Meseta Formation (Seymour Island, Antarctic Peninsula). It was found within the uppermost part of the formation, which corresponds to the upper level of the Telm7 unit or Submeseta Allomember (late Eocene), at the so-called “shelves” locality. The locality is situated 300 m west–south-west of the end of the airstrip, ca. 20 m below the top of the plateau in terms of altitude (Fig. 1). The specimen was collected by Andrzej Gaździcki and Andrzej Tatur (Polish Academy of Sciences) in 1992. The “shelves” site is known from numerous fossil penguin bones, Cucullaea and Eurhomalea bivalves, Hiatella gastropods and Bourchardia brachiopods as well as rare echinoderms (asteroids and echinoids) and whale remains (Tatur, pers. comm.). This specimen (IB/P/B-0981) is permanently deposited at the Institute of Biology, University of Bialystok, Poland, in the collection of the Andrzej Myrcha University Centre of Nature (abbreviated IB/P/B).

Another short segment of the penguin spinal column discussed here was collected in 2008 by Marcelo Reguero (Museo de La Plata [MLP], Argentina) within the lower part of the Telm7 unit/Submeseta Allomember of the La Meseta Formation (Fig. 1). These “thoracic vertebrae still joined” come from the locality DPV 10/84, situated ca. 100 m below the top of the formation, and were accompanied by a number of other bones, including a partial tarsometatarsus and almost complete humerus, belonging in the same skeleton (for more details, including stratigraphy and depositional setting, see Acosta Hospitaleche & Reguero [2010]). The specimen is late Eocene (Acosta Hospitaleche & Reguero 2010) or late middle Eocene (figure 4.5 in Reguero et al. 2013) in age. A single catalogue number (MLP 96-I-6-13) was assigned to the entire partial skeleton, which was attributed by Acosta Hospitaleche & Reguero (2010) to Palaeeudyptes gunnari. It is housed at the MLP, La Plata, Argentina.

The calculation of the Bayes factor (see Rouder et al. 2009 for a readable introduction), used to validate the taxonomic position of the latter specimen, was based on posterior probabilities obtained by Jadwiszczak & Acosta Hospitaleche (2013, appendix I) and the prior odds for the hypothesis H₀: P. gunnari to H₁: P. klekowskii Myrcha, Tatur & Del Valle, 1990 equal one (not favouring either hypothesis). The factor quantified the evidence of the data for H₀ (not necessarily a “null hypothesis”) vs. H₁. The strength-of-evidence scale by Kass & Raftery (1995), categorizing the relative success of the hypotheses/models in predicting the data, was utilized. Bayes factors can be also, probably more correctly, interpreted as measures of change in support for H₀ (relative to H₁) (Lavine & Schervish 1999).

The comparative material studied directly is represented mainly by bones of both fossil and extant penguins from collection of the Andrzej Myrcha University Centre of Nature. Anatomical nomenclature follows that of Baumel & Witmer (1993). Measurements were taken with digital callipers and rounded to the nearest 0.1 mm.

The specimens described earlier represent two series made of three articulated vertebrae, hence each of them obviously represent a contiguous segment of the spinal column. According to Baumel & Witmer (1993) and Müller et al. (2010, supporting information), the cervical and thoracic vertebrae in birds can be separated by elements transitional in configuration (cervicodorsal vertebrae; Fig. 2b, d, f). Both penguin fossils discussed here possess such combination of features. These characters are discussed below and allow interpreting each set of bones in terms of two cervicodorsal vertebrae followed by a single thoracic element.

In specimen IB/P/B-0981 (new material), the presence of the heterocoelous (saddle-shaped) condition within two joints preserved is evident. The large and almost complete second cervicodorsal vertebra IB/P/B-0310 (vertebra cervicodorsale in Jadwiszczak 2006) is also conspicuously heterocoelous, and this is observed in a number of badly preserved cervicodorsal vertebrae of large stem penguins as well (e.g., IB/P/B-0220). However, the clearly narrow centrum of the broken caudalmost vertebra of IB/P/B-0981 strongly suggests the change in the character of the caudal articular surface and this condition is also evident in a quite well-preserved isolated vertebra IB/P/B-0982 (unpublished material, here identified by me as the first “true” thoracic element). According to Baumel & Witmer (1993), in penguins (also auks and some other birds), the elements of the thoracic series are characterized by the concave caudal articular surfaces (opisthocoelous condition). It is perfectly in line with the observation first made by Owen and confirmed by Watson (see Watson 1882) that the third thoracic bone (which corresponds to the 16th or first “truly” thoracic vertebra in Fig. 2) “shows the opisthocoelous character,” although its cranial surface is still saddle-shaped (Watson 1882; author's observation). Stephan (1979) was incorrect in stating that the saddle-shaped joint is still present between the third and fourth thoracic vertebrae (i.e., the first and second “truly” thoracic element).

The condition of the ventral processes and adjacent portion of the bone in the analysed specimen provides further evidence of the transitory nature of this segment. The abrupt change in width (and morphological details) between the two cranialmost vertebrae from the wide (typical of the neck) ventral aspect into slender (typical of the thorax) one is conspicuous (see also Watson 1882; Stephan 1979). Such a change in width is also conspicuous in dorsal view. Moreover, the presence of the cranioventral foveae in both aforementioned elements, and its absence in the caudalmost vertebra is consistent with Stephan's (1979) observation, that some movement of the spine in the median plane between the two cranialmost post-cervical vertebrae is still possible, but more caudally it becomes “virtually impossible.” Baumel & Witmer (1993) noted that, during the ventral flexion of the neck, this fovea accommodates the ventral margin of the articular surface of the preceding vertebra. The large area for attachment of the elastic ligaments that prevent instability of the joint, located caudally to the base of the spinous process of the cranialmost vertebra, is indicative of the relative importance of these two vertebrae for the aforementioned mechanism. Obviously, both cranialmost elements of specimen IB/P/B-0981 were active during such a flexion, assisting the caudalmost cervical vertebrae in forming the curvature within the terminal fragment of the neck (see figure 5 in Guinard et al. 2010).

Stephan (1979) determined that vertebrae 14 and 15 in penguins (which correspond to the first and second transitional vertebrae; Fig. 2) are connected with articulated ribs that terminate free. According to the character matrix by Ksepka & Clarke (2010), this is decidedly the most common condition, but in some species transition to free cervicodorsal ribs begins at the 13th (two species) and, in a single taxon, even the 15th vertebra (Bertelli & Giannini 2005). The articular facets for rib attachment in specimen IB/P/B-0981 are preserved in the middle and caudalmost bones, the posterior one being noticeably larger. This presumably reflects the transition between the cervicodorsal and thoracic series (slender and short vs. more robust and much longer ribs).

Baumel & Raikow (1993) noted the presence of the exceptionally large intervertebral foramina in the cervicodorsal segment in birds, and the one between the two cranialmost bones of IB/P/B-0981 is huge indeed. Such a condition is also evident in extant penguins—e.g., king penguin (Aptenodytes patagonicus Miller, 1778), Magellanic penguin (Spheniscus magellanicus [Forster, 1781]), gentoo penguin (Pygoscelis papua [Forster, 1781]); (author's observation)—and is associated with the brachial plexus (see Baumel & Raikow 1993). This explains the relatively large diameter of the cranial portion of the vertebral canal in IB/P/B-0981 in terms of the room for the caudalmost portion of the so-called “cervical enlargement,” a swelling of the spinal cord at the level of the wings. Interestingly, a decrease in size of the foramen between the middle and caudalmost vertebrae, relative to its predecessor, appears to be more well pronounced in the fossil specimen than in its recent counterparts.

The remnants of the transverse process in the anterior vertebra of IB/P/B-0981 appear to indicate that it was much broader dorsoventrally than its counterparts in consecutive bones, which is typical of the relevant region in extant penguins (author's observation). The abrupt change in depth and character of the lateral concavities among the two cranialmost elements is also, to a lesser degree, present in the transitional series of some extant penguins (e.g., from the genera Eudyptes and Spheniscus). Interestingly, the only fossil specimen identifiable as the first transitional vertebra of a “giant” penguin (IB/P/B-0922; vertebra cervicodorsale in Jadwiszczak 2006; here I clarify its location) and having the satisfactorily preserved lateral concavity is generally similar to the anterior bone. However, its concavity, though well-developed, is relatively shallower and not so sharply delineated.

In present-day birds, diverticula of the cervical air sacs can extend caudally as far as anterior thoracic series (e.g., Wedel 2005). The conspicuous lateral concavity in IB/P/B-0981 can represent a kind of “bony recess,” a blind fossa, that may be occupied by part of such a structure. Some support for this hypothesis is provided by the matte surface of the excavation (assuming this is an actual feature), for this texture may be an external osteological imprint of a pneumatic diverticulum (see also Wedel 2005).

In modern Sphenisciformes, the vertebrae located at the base of the neck become clearly shorter than those from the thoracic region, which is especially well marked in dorsal view (author's observation; see also Shufeldt 1901). This trait together with the increasing oblique elevation of the cranial zygapophyses can be traced in specimen IB/P/B-0981.

Specimen MLP 96-I-6-13 (Palaeeudyptes gunnari), although less well exposed from the matrix than IB/P/B-0981, also possesses a number of features typical of two craniodorsal and the first thoracic vertebrae in extant Sphenisciformes. The most obvious of such traits are those related to conspicuous differences in width of the arches, caudal zygapophyses and transverse processes, dorsoventral thickness of the last-mentioned structures, and the oblique elevation of the cranial zygapophyses. The presence of the outline of a large intervertebral foramen just caudal to the dorsoventrally thick transverse process together with the remnants of the moderately robust rib located posteriorly also strongly suggests, similarly to the condition in IB/P/B-0981, the actual location of the segment at the root of the neck. The morphology of the caudal fragments of the two cranialmost vertebrae in ventral view suggests probably similar joint mobility relative to that in the series discussed earlier.

Two specimens discussed here undoubtedly represent the cervicodorsal region of the spinal column of Eocene penguins. In each case, two transitional vertebrae are followed by the first thoracic element, and they are the only such fossils reported from the Palaeogene of Antarctica. Although both of them resemble, in general terms, their counterparts in recent Sphenisciformes (as far as it can be assessed), some features appear to be unique, most notably the sharp delineation of the deep lateral concavity (maybe a pneumatization-related state) and the surprisingly small intervertebral foramen between the two caudalmost vertebrae in specimen IB/P/B-0981. The former feature distinguishes it from at least some fossil “giant” penguins. The condition of the articular surfaces, cranioventral fovea and attachment surfaces for the elastic ligaments located on the dorsal side, suggest that the joint mobility in the median plane appears to be roughly comparable to that in recent species. The early sphenisciform represented by specimen IB/P/B-0981 was probably larger than the extant gentoo penguin, but smaller than the present-day king penguin, and its taxonomic position remains unknown. Another specimen discussed earlier, MLP 96-I-6-13, is clearly larger than IB/P/B-0981, and its assignment to Palaeeudyptes gunnari (see Acosta Hospitaleche & Reguero 2010) has been strongly supported here. Both fossils provided valuable insights into form and function in early penguins.