Five individual penguin specimens were examined in this study, including three fossil specimens from the Eocene of Seymour Island and two extant comparative specimens from the Antarctic region. From each individual, two bone elements were selected for µ-XRF analysis based on preservation quality and exposed surface area. Long bones were selected when available, but the choice of fossil elements was constrained by preservation and completeness.

The fossil material (specimens CUGB P2003, P2004, and P2005) was collected during the 2020 and 2024 Chinese–Chilean Antarctic Research Expeditions on Seymour Island, Antarctic Peninsula. These fossils represent three stratigraphically distinct intervals across the Eocene succession of Seymour Island (Fig. 1C), ranging from 55.3 to 37.7 Ma. Specimen CUGB P2004, recovered from the Telm 2 subunit (Acantilados I; Marenssi et al. 1998; Montes et al. 2013), is dated to 55.3–54.1 Ma (Ivany et al. 2008; Beamud et al. 2015; Montes et al. 2019), corresponding to the lower Ypresian. The analyzed elements include an incomplete furcula (Fig. 2a) and a radius (Fig. 2d). Although CUGB P2004 is well preserved, its taxonomic study has not yet been completed. Specimen CUGB P2005, collected from Telm 5 (Cucullaea II; Marenssi et al. 1998; Montes et al. 2013), comprises the distal trochlea of a tarsometatarsus (Fig. 2b) and a pedal phalanx (Fig. 2e). Its stratigraphic age of 49.1–45.8 Ma (Ivany et al. 2008; Beamud et al. 2015; Montes et al. 2019) spans the upper Ypresian to lower Lutetian. Because the specimen is highly fragmentary and has not been taxonomically assessed in detail, it is treated as Sphenisciformes indet. Specimen CUGB P2003, assigned to Palaeeudyptes sp. (Xia et al. 2024), was collected from the Submeseta Formation (Submeseta II; Montes et al. 2013). The analyzed elements include the proximal portion of a femur (Fig. 2c) and a tarsometatarsus (Fig. 2f). This horizon is dated to 41.2–37.7 Ma (Ivany et al. 2008; Beamud et al. 2015; Montes et al. 2019), corresponding to the upper Lutetian–Bartonian interval. Given the limited taxonomic information available for CUGB P2004 and CUGB P2005, these fossil specimens were used only for stratigraphically constrained geochemical analysis.

The extant comparative penguin material was obtained from the osteological collections of the China University of Geosciences (Beijing) and the Polar Research Institute of China. Both specimens were collected in Antarctica, with detailed localities shown in Fig. 1A, B. Specimen CUGB E2001 (Fig. 2g, h), identified as Pygoscelis antarcticus (chinstrap penguin), was collected near Great Wall Station on King George Island. Specimen CUGB E2002 (Fig. 2i, j), identified as Aptenodytes forsteri (emperor penguin), was collected near Zhongshan Station on Penguin Island. For each extant specimen, the same skeletal elements (one femur and one tibiotarsus) were selected.

The specimens were analyzed using a Bruker M6 Jetstream mobile XRF scanner (Bruker Nano GmbH, Berlin, Germany) housed at the Micro-X-ray Fluorescence Imaging Laboratory, Institute of Earth Sciences, China University of Geosciences (Beijing, China). The system was equipped with a rhodium (Rh) target X-ray source operated at 50 kV and 600 µA and polycapillary optics capable of generating a nominal spot size smaller than 50 µm for Mo Kα excitation (Alfeld et al. 2013; Liu et al. 2024; Colombo et al. 2025). The X-ray beam was perpendicular to the sample surface. Fluorescence signals were collected using a silicon drift detector (SDD) with an active area of 50 mm² and an energy resolution of 145 eV at Mn Kα (275 kcps) (Liu et al. 2024).

All specimens were scanned under ambient atmospheric conditions. This approach eliminated the need for a vacuum chamber, preventing pressure-induced stress on large and fragile fossil bones while enabling fully non-destructive analysis with reliable elemental detection. Specimens were placed as horizontally as possible on a table beneath the scanner head, with small pieces of modeling clay placed beneath them to level the main scanned surfaces and keep their surface heights approximately uniform. This setup was used to minimize variation in the working distance during scanning. Because of its large size and pronounced curvature, the furcula of CUGB P2004 (Fig. 2a) could not be accommodated within a single focal plane. To preserve spatial resolution, only the lateral portion of the left furcular ramus—where a consistent working distance could be maintained—was selected for µ-XRF mapping (as shown in Fig. 2). Nevertheless, minor surface-height differences related to the curved or uneven surfaces of individual bones could not be completely avoided; therefore, the resulting µ-XRF values are interpreted as relative signal intensities rather than absolute concentrations. The scanner head was moved across each specimen using a motorized X–Y stage to acquire maps of the selected scanning surfaces, with Z-axis adjustment used to set the working distance during scanning. A step size of 50 µm and a dwell time of 10 ms/pixel were chosen to balance spatial resolution and acquisition efficiency. For all scanned samples, the combined mapping field covered 2,651 × 1,968 pixels, with a total collection time of 18 h and 46 min.

Data processing and analysis were performed using ESPRIT Spectrum (Bruker Nano GmbH, Berlin, Germany). Characteristic peaks in the XRF spectra were identified to determine the elemental composition of each sample, primarily based on Kα peaks. To improve peak resolution, a software-assisted peak-fitting algorithm was applied to accurately discriminate between overlapping fluorescence peaks, particularly where Kα peaks overlapped with Kβ or L peaks. Potentially interfering peaks were manually verified and excluded when necessary. The Y-axis of the XRF spectra was set to a base-10 logarithmic scale (Fig. 3), allowing fluorescence peaks of different intensities to be visualized within the same spectrum.

For spatial analysis, single-element distribution maps and heat maps were generated for the selected elements (Figs 4, 5, Suppl. material 1: figs S1, S2). Single-element distribution maps are presented as monochromatic false-color images to show the spatial distribution of each element. Heat maps were generated to display relative variations in µ-XRF signal intensity within each element map, thereby helping to identify areas of comparatively higher or lower signal intensity across the scanned bones. Quantitative comparisons among specimens were based on area-based region-of-interest (ROI) analyses and the mean µ-XRF values reported in Table 1.

All 10 bone elements were scanned together within a single mapping field. For each skeletal element, an ROI was defined to include its complete visible scanned surface. Mean elemental values were calculated by averaging all pixels within each skeletal ROI for each selected element using the area-scan function in ESPRIT Spectrum. The background value was calculated from the non-sample area of the same mapping field after excluding all skeletal ROIs. This non-sample background was used to estimate the mean background signal level under the same scanning conditions. These data are reported as relative µ-XRF counts or signal intensities.

In the extant bones, Ca and P, the principal constituents of bioapatite, exhibited strong and nearly continuous signals across all samples. Mean Ca and P µ-XRF values varied among the extant specimens and partly overlapped with those of the fossil bones, although CUGB E2001 showed particularly high Ca values (Table 1). The Ca and P maps (Fig. 4B, E) displayed broadly homogeneous distributions, whereas the corresponding heat maps (Fig. 5B, E) revealed slightly higher intensities in the midshaft regions than at the articular ends in most specimens, consistent with the thicker cortex and more compact bone in the diaphysis.

S and K were present at uniformly low signal intensities throughout the extant bones (Fig. 4C, D). Si and Mn were also detectable but were present at low signal intensities, forming thin rims or patchy, low-intensity enrichments along the bone margins (Fig. 4A, G). In contrast, Al, Ti, Zn, Fe, and Sr generally yielded values at or below background levels and lacked consistent spatial patterns (Figs 4, Suppl. material 1: figs S1, S2). Their mean counts (Table 1) indicate that these measurements predominantly reflect background-level signals or analytical noise rather than consistent enrichment within the bone tissue.

Fossil bones retained definite Ca and P signals, although their mean µ-XRF values varied among specimens and partly overlapped with those of the extant bones. In contrast to the broadly continuous distributions observed in the extant specimens, Ca and P showed more heterogeneous spatial patterns in several fossil elements (Figs 4B, 4E, 5B, 5E). Among the fossil specimens, CUGB P2003 exhibited the highest mean Ca and P values, whereas CUGB P2004 showed the lowest values (Table 1).

In contrast to the extant bones, several non-apatitic elements showed elevated µ-XRF signal intensities in the fossils. The early Eocene specimen CUGB P2004 was particularly distinctive: it showed the highest mean µ-XRF values for Si, K, and Ti among all samples, exceeding those in both younger fossils and extant bones (Table 1). These elements formed continuous, high-intensity enrichments over most bone surfaces (Figs 4A, 4D, 4F, 5A, 5D, 5F), and their spatial patterns were well correlated. Fossils from younger strata (CUGB P2003 and P2005) displayed markedly lower and more patchy Si, K, and Ti signals, with only localized hotspots.

Sulfur also showed stratigraphically variable behavior. It was almost absent in CUGB P2004 but exhibited pronounced localized enrichment in CUGB P2005 and more moderate enrichment along the bone margins of CUGB P2003. Mn and Fe produced elevated signals in all three fossil specimens relative to the extant bones, and mean Feµ-XRF values were markedly higher than the background values and those in the extant bones (Table 1). Their heat maps displayed irregular, patchy signal distributions across the bone surfaces, with heterogeneous high- and low-intensity regions in all fossils (Figs 4G, 4H, 5G, 5H).

Sr exhibited distinct spatial patterns among the fossils. In the Sr maps, CUGB P2005 and CUGB P2003 showed localized hotspots, particularly in joint regions, whereas CUGB P2004 remained uniformly low in intensity (Figs 4I, 5I). Mean Srµ-XRF values in all fossil specimens were generally below background levels, except for CUGB P2005, which exhibited a slight elevation above the background value (Table 1). Thus, despite localized hotspots, Sr generally showed low relative signal intensities and modest differences among fossils. Al and Zn in the fossil bones yielded values below background levels and lacked organized spatial distributions (Suppl. material 1: figs S1, S2, Table 1).

Si, K, and Ti in sedimentary records are predominantly derived from the breakdown and alteration of silicate minerals. As their relative abundances reflect the intensity and products of chemical weathering, they are closely linked to continental weathering processes (Nesbitt and Young 1982; Taylor and McLennan 1985). Among these elements, Ti is particularly valuable as a geochemical tracer because of its marked immobility and extremely low solubility under typical surface weathering conditions (Orians et al. 1990; Young and Nesbitt 1998). During weathering, Ti is preferentially incorporated into clay and silt fractions (Garcia et al. 1994; Taboada et al. 2006), thereby facilitating its preservation in sediments. Ti remains largely unaffected by subsequent chemical alteration, preserving its original geochemical signal throughout sediment transport and early diagenesis (Greber et al. 2017). Furthermore, the stable oxidation state of Ti (Ti⁴⁺) and its biological inertness enhance its reliability as a proxy for terrestrial input (Greber et al. 2017). Collectively, these properties make Ti an effective proxy for tracing historical terrestrial weathering fluxes and sedimentary dynamics.

Consistent with this, the early Eocene fossil specimen CUGB P2004 exhibits marked enrichment in Si, K, and particularly Ti compared with both younger fossils and extant bones (Figs 4, 5; Table 1). The mean Tiµ-XRF value is approximately twice those in the two younger fossil specimens, whereas all three elements show substantially lower values in the extant specimens than in CUGB P2004 (Table 1). When considered together with the sedimentological and paleoclimatic context, this pattern is consistent with increased input of terrestrially derived weathering material into the depositional basin during the early Eocene (Fig. 6). Such material likely originated predominantly from regional igneous and continental sources (Marenssi et al. 2002; Montes et al. 2019). Volcanic arc-derived detritus could, in principle, contribute to elemental enrichment. However, evidence indicates that volcanic activity on the Antarctic Peninsula was relatively subdued during the early Eocene (Smellie 2021). This interpretation is supported by petrographic analyses of the La Meseta Formation, which reveal a mixed lithic assemblage without a dominant volcanic component (Marenssi et al. 2002).

These geochemical observations align with paleoclimatic reconstructions that indicate warm and humid conditions in Antarctica during the early Eocene (Zachos et al. 2001; Pross et al. 2012). Such a climate would have enhanced the intensity of chemical weathering on adjacent landmasses, thereby increasing the supply of weathered material to nearshore marine environments (Fig. 6). Under the relatively low-energy, fine-grained depositional conditions characteristic of the lower La Meseta Formation, prolonged sediment–water interaction would have facilitated the accumulation of these weathering products within and on skeletal remains, helping to explain the elevated Si, K, and Ti signals observed in CUGB P2004. As the Antarctic Peninsula experienced climatic cooling from the middle to late Eocene, weathering intensity likely decreased, leading to reduced terrestrial input to the basin and resulting in lower, more stable Tiµ-XRF values in younger fossils.

Although Si and K were also released in large quantities during intensified weathering in the early Eocene, these elements differ from Ti in being more easily transported in dissolved form (Gaillardet et al. 1999). However, under stable depositional conditions, Si and K could still accumulate in fossil bones through processes such as silicification or infiltration by clay-rich porewater. Intense weathering typically mobilizes Al through the formation of aluminosilicates; however, Alµ-XRF signals in the analyzed bones remain at near-background levels (Table 1). This absence of Al enrichment may partly result from analytical attenuation of the low-energy Al Kα signal under non-vacuum µ-XRF conditions (Mantler and Schreiner 2000). Additionally, because Al tends to remain bound within detrital or authigenic aluminosilicates in the surrounding matrix rather than being incorporated into the bone apatite lattice (Hughes and Rakovan 2015), this geochemical behavior likely contributes to the subdued Al signal in the µ-XRF maps.

Sedimentary energy conditions appear to have exerted strong control over elemental distribution patterns among the Seymour Island penguin fossils. The early Eocene specimen CUGB P2004, from the lower La Meseta Formation (Telm 2), is associated with fine-grained, low-energy shelf deposits (Marenssi 2006; Montes et al. 2019) and shows the highest relative µ-XRF values for Si, K, and Ti among all analyzed bones (Figs 4, 5, Table 1). By contrast, the younger specimens CUGB P2005 from Telm 5 and CUGB P2003 from the Submeseta Formation were deposited in higher-energy environments characterized by coarser sediments and stronger hydrodynamic activity, and both exhibit lower overall relative µ-XRF values for Si, K, and Ti, accompanied by more heterogeneous distribution patterns. Extant penguin bones show only low, near-background signals for these elements, indicating that the enrichments observed in fossils cannot be explained by primary bone composition alone.

Under the relatively low-energy, fine-grained depositional conditions of the lower La Meseta Formation, fossil bones would have remained in prolonged contact with porewater circulating through Si- and K-bearing detrital and authigenic phases. These conditions would have favored the accumulation of silicate- and clay-derived material within and on skeletal remains, providing a plausible explanation for the continuous, high-intensity Si, K, and Ti signals observed in the µ-XRF maps of specimen CUGB P2004. Similar patterns, in which elevated Si and Al in fossil bones and teeth reflect the infilling of bone porosity by fine-grained clay minerals, have been documented in other diagenetic settings (Kohn et al. 1999). The µ-XRF data do not directly resolve mineralogy or pore-scale textures, but the high relative Si, K, and Ti signals and their spatial correlation in CUGB P2004 are consistent with clay and silt enrichment under low-energy, mud-rich conditions. In contrast, the higher-energy environments represented by Telm 5 and the Submeseta Formation were dominated by coarser-grained, more mobile sediments, with stronger currents and wave activity driven by sea-level fluctuations and glacial input (Marenssi 2006; Montes et al. 2019). In such settings, bones likely underwent episodic reworking and more rapid burial, shortening the duration of bone–porewater interaction and resulting in more patchy and overall lower Si, K, and Ti signals, as observed in CUGB P2005 and CUGB P2003.

Changes in redox conditions during burial strongly influence the behavior of Fe, Mn, and S because of the redox-dependent solubility of these elements. Under reducing conditions, Fe²⁺ and Mn²⁺ are relatively mobile in porewater; under more oxidizing conditions, they tend to precipitate as oxyhydroxides or, in the presence of sulfide, as sulfide minerals (Slomp et al. 1997). In the penguin fossils analyzed here, Fe and Mn are marked elevated relative to their levels in extant bones and background levels, with heat maps revealing irregular, patchy high- and low-intensity regions across bone surfaces (Figs 4G–H, 5G–H, 5Table 1). These heterogeneous patterns are consistent with spatially variable redox conditions during burial. Repeated cycles of local reduction and oxidation would lead to the dissolution and reprecipitation of Fe- and Mn-bearing phases within bone pores and on bone surfaces. Similar secondary Fe and Mn oxide coatings, formed under unstable redox conditions, have been documented on fossil bones and other authigenic substrates across various diagenetic settings (Van Cappellen and Wang 1996; Vasilatos and Economou-Eliopoulos 2018).

The spatial association of Fe and S in CUGB P2005 further supports the involvement of redox-sensitive mineralization. In this specimen, zones of elevated S intensity partially overlap with Fe-enriched areas (Figs 4C, 5C, 5H), suggesting the precipitation of Fe sulfides in localized reducing microenvironments (Canfield 1988). Sulfate-reducing bacteria in such settings may convert sulfate (SO₄^2–) to sulfide (H₂S), which then reacts with dissolved Fe²⁺ to form Fe sulfide minerals (Fortin and Beveridge 1997). Although the µ-XRF data cannot directly identify specific mineral phases, the observed Fe–Mn enrichment and Fe–S colocalization are consistent with models in which microbial and abiotic redox processes jointly control the development of secondary Fe-, Mn-, and S-bearing minerals in fossil bones during diagenesis (Machel 2001).