The IRR results demonstrate that the VHI evaluation of virtual sections is congruent with the OHI rating of histological thin sections. As noted, certain structures taken into consideration by the OHI, such as individual concentric lamellae, cannot be differentiated, nor can structures smaller than 1 µm (e.g., osteocytes or their lacunae) be visualized without synchrotron-based microCT (Andronowski et al., 2017a,b), but undifferentiated concentric lamellae, and features such as shallow pits and tunnels can be seen with lab-based microCT. Moreover, our inability to visualize osteocytes and their lacunae may result from a contrast-to-resolution issue during the scan. Thus, the VHI is able to describe the level of bioerosion that a sample has undergone, specifically that which results from bacterial attack. It may be possible that non-MFD bioerosion can be detected and differentiated with microCT images; however, the samples included in this study almost exclusively exhibited MFD, and not Type 1 Wedl-, nor cynobacterial tunneling. Importantly, in other respects the virtual images are also consistent with what has been qualitatively described in conventional thin sections.
Jans et al. (2002) specify five categories of diagenetic alterations as visualized via histology (1) presence and type of MFD; (2) presence of inclusions (e.g., sand, fungi, framboids); (3) presence of infiltrations, such as stains exogenous to the bone matrix; (4) presence of microfissures, and (5) birefringence intensity (see Fig. 1b). We found that MFD can be clearly visualized both in virtual sections and in 3D volumes, and can, moreover, be virtually measured (as can anatomical features) to refine their identification by size (Fig. 10). Such measurements are possible due to the pixel-size calibration inherent in microCT scans. In badly affected samples it is difficult to ascertain which type of tunnelling is present, as the bioerosion presents as large dark grey patches. However, this problem is also encountered with thin section micrographs of badly preserved samples, and, moreover, though morphologically distinct, different MFDs are very likely, “all aspects of the same type of bacterial attack, differing only in tissue microarchitecture and local hydrology” (Turner-Walker, 2019, p. 35). We were also unable to visualize the Type 2 Wedl tunnels diagnosed via histology. This is likely due to the small size of the tunnels, which although they may be long, are not thick enough to be visualized at the resolution we used.
Fig. 10 Virtual measurements of MFD and Haversian canal in midline of a single virtual transverse cross-section (1 voxel [8.0 µm] thick) of femoral sample GÖ01
We agree with Booth et al. (2016) that although inclusions can be observed based on differences in density (grey levels) and texture, no further identification can be made. Infiltrations such as stains are more difficult to distinguish in microCT images. While it may be possible to detect a stain after its having been recognized through histology, it is not immediately apparent using virtual images alone that a change in grey values results from, for example, the impregnation of organic or inorganic substances as manifested macro- or microscopically as a stain, or if changes to the grey values result from demineralization caused by bacterial attack. Furthermore, while dissolution and recrystallization of HAp can be visualized as changes in grey values, it is not possible to determine the uptake or exchange of specific materials like uranium or fluorine, or nitrogen content as identified via energy dispersive X-ray spectroscopy, nor can microCT visualize the hypermineralized cuffs surrounding tunnels that result from mineral reprecipitation as is specifically achievable with SEM (Kendall et al., 2018; Turner-Walker & Syversen, 2002). Microfissures recognized in histological sections can be clearly detected in virtual sections; it is also possible to determine if the microfissures are post-depositional, embedding artefacts, or result from excavation damage. Birefringence intensity cannot be assessed, though collagen content can be visualized with staining procedures not employed in this study (Handschuh et al., 2017). Lastly, virtual evaluation of sub-micron spongiform porosity (0.1–1 µm in diameter), a common form of tunnelling in archaeological bone (Turner-Walker, 2012; Turner-Walker et al., 2002), requires SR-microCT (Caruso et al., 2020).
As concerns qualitative descriptions of how bacteria invade bone, we once again find congruence between microscopy and virtual imaging. Yoshino et al. (1991) described regions heavily affected by bacteria and fungi as potentially corresponding to low X-ray density. As noted, areas of low density (dark grey values) that result from mineral dissolution and redistribution (Hackett, 1981) correspond to MFD in the virtual sections. Yoshino et al. (1991) also reported 5–10 µm in diameter vacuoles surrounding circumferential and Haversian lamellae, while Jackes et al. (2001) describe the orientation of MFDs along circumferential lamellae and the axis of Haversian systems. These vacuoles, or MFDs, were also visible in virtual sections within and surrounding the lamellae of Haversian systems and in circumferential lamellae. It has also been suggested that bacterial attack and chemical demineralization are impeded by cement lines (Kendall et al., 2018; Turner-Walker, 2012; Turner-Walker et al., 2002). We may be visualizing this in samples such as GE08 (Supplementary Materials ESM 8) where bioerosion appears to specifically attack the undifferentiated concentric lamellae, while in some regions of S01 (see Fig. 5) bioerosion seems to affect the interstitial lamellae, surrounding but not penetrating the concentric lamellae. However, there were no observable cement lines in sample S01. Moreover, after a certain point, when bioerosion has become severe, even if only within patches as with SB02 (Supplementary Materials ESM 6), it is not possible to distinguish where bacterial attack begins or ends. Further research is thus required to confirm that bacterial attack is curbed by cement lines.
The volume rendering depictions of bacterial attack are worth particular mention as we were able, using the “glow” colormap, to visualize in 3D what has previously only been reported in 2D. It is hypothesized that bacteria disseminate through and enlarge the canalicular network, accessing collagen via Haversian and Volkmann canals (Hackett, 1981; Kendall et al., 2018; Turner-Walker, 2019). Turner-Walker (2012) and Yoshino et al. (1991) also describe the occupation of Haversian canals by bacteria. While it is not possible to visualize bacterial attack on the canalicular network due to restrictions imposed by scan resolution, we saw evidence of bacterial attack within the canals. Sample GÖ01 (Fig. 11) provided a particularly interesting volume in which fine red trails follow the interior length of well-preserved yellow canals, the colors reflecting differences in X-ray density. These are arguably the demineralized traces of bacteria that have entered the Haversian and Volkmann canals, consumed their soft tissues (i.e., blood and lymphatic vessels, and nerve fibers), and then attacked the more mineralized concentric and interstitial lamellae. Indeed, we see further evidence of bioerosion in the form of miniscule holes that penetrate some canals (Fig. 11), and as endocasts of bacteria which take the form of narrow, almost translucent red tunnels through the bone matrix. Jackes et al. (2001) also described what they believe to be the endocasts of bacterial chains as varied trajectories of circumscribed tunnels, visible in both longitudinal- and cross-sections. These tunnels present in volume renderings (e.g., Supplementary Materials ESM 5) as thin red channels that migrate through the bone, often at first in the direction of the Haversian canals but also in numerous directions. Depending on the severity of the erosion, they may present as discrete spiral-like red tunnels distinct from the anatomical canals to an almost cobweb-like film that veils degraded canals (e.g., Supplementary Materials ESM 6), or even as an amorphous spongy texture (e.g., Supplementary Materials ESM 8) in the most poorly preserved samples.
Fig. 11 3D volume rendering showing MFD tunneling (a), red trails of bioerosion (b) and miniscule holes in Haversian canals (c) in midline of femoral sample GÖ01
Jackes et al. (2001) noted that in previously published figures (e.g., Bell, 1990; Garland, 1987; Grupe & Dreses-Werringloer, 1992; Hackett, 1981; Piepenbrink, 1986; Schultz, 1986) that illustrate MFD, the magnification was not high enough to clearly visualize the bioerosion and that the cross-sections these authors used resulted in a view that resembles grape clusters, which do not match what can be visualized using longitudinal sections such as those published by Bell (1990). These grape-like clusters are also visible in the 3D volume of GÖ01 (see Fig. 4). When the volume is viewed from different angles it can be seen that these clusters are actually tunnels that run longitudinally and at angles through the bone (Fig. 12). This visualization is similar to, but clearer than what is seen at very high magnification (>600×) with microscopy. MicroCT is a valuable tool in this respect, as it permits the selection and visualization of any location within a sample in any orientation. Samples scanned, for example, in the coronal plane can be virtually resliced to the transverse plane. The virtual dataset can also be rotated to align with any anatomical plane using landmarks. We chose to reslice our samples for assessment in the orientation of traditional thin sections for two reasons (1) to facilitate comparison with traditional micrographs, and (2) 3D visualization of the canals and erosion in planes that do not follow the orientation of the Haversian systems made it difficult to assess bioerosion patterns in a three-dimensional sample. Furthermore, samples can be virtually cropped to visualize smaller volumes. The capacity to reslice, rotate, and virtually crop or dissect samples, repeatedly if desired, is specifically achievable with virtual but not conventional histologic methods.
Fig. 12 3D volume rendering showing MFD tunneling in femoral sample GÖ01
According to both Dal Sasso et al. (2014) and Booth et al. (2016) individual slices within a stack are representative of the sample, or entire femoral midshaft, and thus the OHI applied to a single transverse cross-section may be “considered representative of the sample itself” (Dal Sasso et al., 2014, p. 37). We assessed the complete stack of virtual sections from microCT images and confirmed that within a given stack that individual slices are homogeneously affected by bioerosion. While it appears probable that this holds true for the entire midshaft, further research should assess different regions of the femoral diaphysis to confirm homogeneity within the entire shaft, particularly when staining is present in patches.
Effects of skeletal sample region: Using microCT Dal Sasso et al. (2014) found that skull, rib, and femur fragments present with different levels of bioerosion with femoral samples being the best preserved, and further suggest that the poorer preservation of rib and skull bone microstructure results from the higher volume of trabecular bone and porosity associated with the latter, which can be more heavily affected by inclusions and MFD. In the current study we had both femoral and rib samples from only two individuals (GÖ01/GÖ24 and GÖ04/GÖ25), none of which were badly affected by bioerosion; thus, we can neither confirm nor refute their findings. Nevertheless, our research suggests that, although the femoral diaphysis, which has a thick cortical layer, is preferred for visualizing canal structure and Haversian systems, the VHI can also be applied to mandibular, cranial, and rib samples. However, the volume renderings can be more difficult to interpret. For example, mandibular sample SB01 (Supplementary Materials ESM 4), an OHI/VHI 5, has numerous anatomical canals that can be clearly visualized within the void, while subadult cranial sample SB04 (Supplementary Materials ESM 12), also an OHI/VHI 5, which has fewer canals, is not so clearly visualized. Sample R02 (Supplementary Materials ESM 13) (OHI/VHI 2), also a cranial fragment, but this time of an adult, provides a clearer image; however, there is higher density recrystallization around the canals and trabeculae of this sample, which facilitates visualization. Humeral fragment SB02 (Supplementary Materials ESM 6) (OHI/VHI 2.5) of a juvenile with numerous canals also provides a clear visualization. Rib samples GÖ24 (Supplementary Materials ESM 14) (OHI/VHI 4.5) and GÖ25 (Supplementary Materials ESM 15) (OHI 4/VHI 5), which have numerous canals (particularly GÖ24) can be clearly visualized but are not as easy to interpret as femoral samples.
Effects of cremation: We suggest that some cremated samples can be evaluated with the VHI. However, the degree of burning may affect how easily this can be done. As noted in the assessment of IN01 (see Fig. 7), cremation results in the homogenization of grey values, which obscures anatomical features like concentric lamellae. Samples that have been cremated at presumably higher temperatures, such as IN04 (Supplementary Materials ESM 16) are more difficult to evaluate. Lemmers et al. (2020) conducted experiments to ascertain the visibility of bioerosion in histological sections of bone that have been cremated, concluding that it can be identified in burnt remains. Thus, the classical histological analysis provides good results; however, further experiments are required assessing the effects of burning on bioeroded versus uneroded remains in microCT images to better understand how bioerosion may be visualized in cremated remains. However, given the high concordance between the VHIs and OHIs of cremated samples, as demonstrated by the IRR, we included them in the dataset. Interestingly, we found that although microanatomy could still be visualized in virtual sections, even if poorly, the volume renderings of cremated samples were often, but not always extremely poor, even in instances of an OHI 5. Sample R01 (Supplementary Materials ESM 17) provided a significantly clearer 3D visualization than IN01 (see Fig. 7), IN02 (Supplementary Materials ESM 18) or IN03 (Supplementary Materials ESM 19). Specific calibration of the microCT scans to measure absolute mineral density, or more particularly calcium content, may help to better visualize cremated remains (Handschuh et al., 2017). Lastly, previous research suggests that cremation results in bone shrinkage, cracking and crystallization (Boschin et al., 2015; Ellingham & Sandholzer, 2020; Hanson & Cain, 2007). While we are able to visualize dense (bright white) infilling in the scans, which may be recrystallization, as well as cracking, morphometric analysis is required to confirm shrinkage.
Effects of pathological alteration: With any histological evaluation familiarity with microanatomical structures is required to identify their presence and destruction by bioerosion. It is also important to have an understanding of, for example, hematologic and metabolic bone disease, and how these features are visualized in microCT images. Pathological alterations affect the visualization and interpretation of skeletal remains, including, we found, in volume renderings using the “glow” colormap. While it was still possible to apply the VHI to the virtual sections of sample S04 (see Fig. 6), a late-mature female, it was immediately clear that this individual was severely osteoporotic. The volume rendering, using the “glow” colormap, of heavily demineralized samples such as S04 visualizes very little. Thus, for heavily demineralized, or samples characterized by pathological changes, it is worthwhile to try a variety of colormaps to test which can be used to best visualize canal structure and bioerosion in 3D.
Faunal samples: Again, although applying the VHI to non-human samples yielded similar ratings to the OHI, they too were less clearly visualized in 3D. As noted in the Results, little was visualized in faunal sample R04 (Supplementary Materials ESM 10) (OHI 3.5/VHI 3), tentatively identified as belonging to Artiodactyla, which is potentially due to the paucity of Haversian canals in bone of this mammalian order. Sample GÖ14 (Supplementary Materials ESM 20), an ovicaprid humerus (OHI 3.5/VHI 4.25), also yielded a poor volume rendering in which very little was visualized, consistent with both the SEM and virtual sections.