Virtual histology of archaeological human deciduous prenatal enamel through synchrotron X-ray computed microtomography images

The implementation of synchrotron radiation X-ray microtomography for non-destructive visualization of the dental prenatal enamel microstructure of intact archaeological human deciduous teeth is presented and compared with the physical thin-sectioning of the same archaeological dental series. Prenatal daily enamel secretion rates, i.e. the distance between two adjacent daily enamel microstructures, have been measured with the two approaches. Results show a good agreement between physical and virtual histology.


Introduction
Human teeth hold a wealth of information about an individual's growth, development, diet, chemistry and non-specific illnesses (Hillson, 2014;Guatelli-Steinberg, 2016). Because of this, histologists routinely draw on the cell mechanisms underlying tooth morphology to reconstruct enamel growth in order to contribute insights into the evolution of hominins, their phylogenies and life history (Dean, 2016). Relative to permanent teeth, human deciduous teeth have been less studied for this purpose but they retain unique information about prenatal ontogeny Mahoney, 2015), and aspects of maternal diet, health and mobility during pregnancy Nava et al., 2020;Li et al., 2020).
Enamel forms incrementally with a 24 h circadian rhythm and retains a record of this growth in the form of incremental markings known as cross striations (Lacruz et al., 2012;Zheng et al., 2013). A fundamental dental-growth parameter is the amount of enamel deposited by cells over this 24 h period (i.e. daily secretion rate: DSR). Measurements of prism lengths ISSN 1600-5775 between two adjacent cross striations is the amount of enamel deposited over 24 h. There are many applications of DSRs in studies of biological anthropology and human evolution. DSRs calculated from physical thin sections have been used to reconstruct the speed at which enamel cells form a human tooth crown in archaeological and contemporary series (Birch & Dean, 2014;Mahoney, 2012;, and have provided insight into the evolution of hominin dental development and life history through the analysis of fossil dental specimens Macchiarelli et al., 2006;Dean et al., 1993Dean et al., , 2001Dean et al., , 2010Dean et al., , 2016Smith et al., 2018;Austin et al., 2013;Rosas et al., 2017;Lacruz et al., 2008;Dean & Smith, 2009;Joannes-Boyau et al., 2019). Prenatal enamel usually shows less distinct microstructures with respect to the postnatal one, possibly as a consequence of the protected and buffered intrauterine environment in which the tissue develops Noré n, 1983;Rythé n et al., 2008). For this reason, data on prenatal enamel DSRs are scarce and mostly available from modern reference collections of exfoliated/extracted deciduous teeth (Mahoney, 2012(Mahoney, , 2015Birch & Dean, 2014;Dean et al., 2020).
Most studies of enamel histology involve destructive thin sectioning of teeth (e.g. Mahoney, 2015;Nava et al., 2019). Increasingly, anthropological studies have turned to nondestructive virtual histology of rare and fragile fossil specimens of mammalian teeth. Virtual histology is an advanced imaging technique for phenotyping, depiction and discrimination of various biological tissues (Albers et al., 2018;Dullin et al., 2017;Sanchez et al., 2012;Tafforeau & Smith, 2008). This typically requires phase-contrast synchrotron radiation computed microtomography (SRmCT) (Le Cabec et al., , 2017Smith et al., 2007Smith et al., , 2015Tafforeau et al., 2006Tafforeau et al., , 2007Tafforeau et al., , 2012Xing et al., 2019;Tafforeau & Smith, 2008). Virtual histology can be considered nondestructive, as long as the experimental parameters are set to reduce the total dose delivered and therefore minimize the potential deleterious effects of X-rays on ancient DNA that could be preserved in teeth and made available for possible future analysis (Immel et al., 2016).
The comparability of growth parameter estimates for the enamel fine microstructures (i.e. those whose dimensions are on the order of a few micrometres) derived from these two approaches is largely unexplored. One study conducted on permanent teeth using phase-contrast SRmCT -with a submicrometric pixel size -reported that enamel-growth parameters calculated from physical thin sections and virtual histology were comparable with a maximum difference of less than 1% (Tafforeau & Smith, 2008).
This study aims to analyze prenatal enamel from the Roman Imperial necropolis of Velia (I-II centuries CE, Salerno, Italy) (Fiammenghi, 2003) using 3D virtual histology of intact deciduous teeth. The goal is to extend the validation of virtual histology to deciduous teeth and particularly to the less readable prenatal enamel (Hillson, 2014). Here, we use the SRmCT setup of the SYRMEP beamline (Tromba et al., 2010) of the Elettra synchrotron facility in Basovizza (Trieste, Italy).
Virtual histology is conducted in propagation-based phasecontrast mode, which allows us to consistently image and quantify the fine growth markers in prenatal enamel. For the first time, we compare virtual data acquired using different effective pixel sizes to DSRs for the same tooth type and archaeological skeletal series  calculated from physical thin sections.

Materials and methods
A sample of four intact deciduous central upper and five lower incisors (Table 1) were selected from an archaeological sample of infants from the large Imperial Roman necropolis at the ancient port of Velia (I-II centuries CE, Campania, Southern Italy) (Fiammenghi, 2003;Craig et al., 2009;Bondioli et al., 2016). This skeletal collection was chosen because previous histological analysis of the prenatal enamel of upper and lower central deciduous incisors revealed cross striations and other incremental markings were well preserved .
The age-at-death of the individuals selected for the present analysis ranges between 2 and 12 months based on skeletal and dental development (AlQahtani et al., 2010;Cunningham et al., 2016;Ubelaker, 1989;Buikstra & Ubelaker, 1994). The skeletal collection is curated in the Museo delle Civiltà , Museo Nazionale Preistorico Etnografico Luigi Pigorini of Rome, to which one of the authors (LB) was affiliated and in charge of the collection at the time of the analysis.

Acquisition of the SRmCT dataset
SRmCT measurements were conducted at the SYRMEP beamline of Elettra using a multi-scale approach. At SYRMEP the X-ray beam is delivered by a bending magnet  source and experiments were performed in a filtered whitebeam configuration (filters: 1.5 mm Si, 2.0 mm Al) with a mean energy of $27 keV (estimated entrance dose of the order of kiloGray). At the time of measurement, the only microtomographic setup available for pink beam experiments was installed $17 m from the source. The mechanical limitation of this setup did not allow a tiling scanning procedure (Du et al., 2018), so only a single scan was acquired for each tooth.
Projections were recorded with a macroscope camera based on a 16-bit, water-cooled sCMOS detector (2048 Â 2048 pixels) coupled to a 15 mm-thick LSO:Tb scintillator screen. Using the variable optical zoom of the detecting system, effective pixel sizes of 2.0, 1.3 and 0.9 mm (see Table 1) were set corresponding to fields of view of about 4.1 Â 4.1, 2.6 Â 2.6 and 1.8 Â 1.8 mm, respectively.
For each scan, we acquired 2000 projections over a 180 rotation with an exposure time/projection of 2.0 s (effective pixel size = 0.9 mm) or 1800 projections over a 180 rotation with an exposure time/projection of 1.5 s (effective pixel sizes of 2.0 and 1.3 mm).
The experiment was conducted in propagation-based phase-contrast mode with the sample-to-detector distance set to 150 mm for all measurements, independent of the pixel-size selection.
The slice reconstruction was carried out using the customdeveloped SYRMEP Tomo Project (STP) software (Brun et al., 2015).
Before reconstruction, a ring-removal algorithm (Rivers, 1998) was applied. Then, projections were processed by a single-distance phase-retrieval algorithm (Paganin et al., 2002) with a / parameter (ratio between the real and imaginary parts of the refraction index of the material n = 1 À À i) chosen on an ad hoc basis for a good visualization of fine details in the tooth enamel balancing between contrast and spatial resolution. Since acquisition conditions were not ideal (that is to say homogeneous objects and a narrow X-ray energy bandwidth) a slight blurring can be observed in the reconstructed images. Paganin's algorithm can be employed, with some caution, even on dense and multi-materials (Langer et al., 2014;Arzilli et al., 2015) and with polychromatic light (e.g. Myers et al., 2007). In this case the / ratio was set to 10.

Virtual histology extraction
The final reconstructed images for histological analyses were generated by importing the volumes in the ImageJ software (Schneider et al., 2012), using the Reslice tool to generate a stack of images following the proper histological sectioning plane (buccolingual plane, passing through the tip of the dentine horn). The image stack was processed through the ZProject average intensity function varying the range of the stack to obtain different virtual thicknesses. In ImageJ, the brightness/contrast tool and the unsharp mask filter were used to enhance visualization of the features of interest, taking care to avoid the creation of artefacts.

Thin sectioning
The physical sectioning of the deciduous central incisors from the same collection was obtained using standard histological methods (Mahoney, 2015;Nava et al., 2019). The teeth were first cleaned and embedded in epoxy resin (Epo-Thin, Buehler Ltd). A longitudinal buccolingual thin section passing through the centre of the cusp, $300 mm thick, was obtained using a diamond blade microtome (Leica 1600, Leica AG). The sectioned tooth was attached to a microscope slide, previously treated with liquid silane (3M RelyX ceramic primer), using a light-cured adhesive (3M Scotchbond multipurpose adhesive) placed under UV light for 60 s. The sections were ground to a final thickness of $100 mm with waterresistant abrasive papers of different grits (Carbimet, Buehler Ltd), and then polished with a micro-tissue (Buehler Ltd) and alumina powder (MicroPolish II alumina 1.0 mm, Buehler). The enamel surface of the sections was etched using a phosphoric acid gel (Etchant Delivery System, 3M Scotchbond) for 10 s, cleaned and finally mounted with a coverslip using Eukitt (mounting medium for microscopic preparations, O. Kindler GmbH & Co.).
Overlapping images of the entire crown were obtained with a microscope camera (Leica DFC 295, 6.55 Â 4.92 mm CMOS sensor, 2048 Â1536 pixel imaging, 30 bits colour depth, with an effective pixel size of 0.52 mm when coupled with a 10Â objective and 0.63Â video optic adapter) paired with a transmitted light microscope (Laborlux S, Leica AG) using polarized light. The images were assembled into a mosaic using Microsoft Image Composite Editor [ICE 2.0 (https://www.microsoft.com/en-us/research/product/computa tional-photography-applications/image-composite-editor/)].

Daily enamel secretion rates
DSRs were collected from virtual and physical histological sections. Rates were calculated by measuring the length of prisms between five or six cross striations (representing four or five days of enamel secretion, respectively). This was repeated several times within the enamel regions where cross striations were visible so that a grand mean DSR could be calculated. Overall, a total of 112 DSRs were collected in the virtual thin sections and 51 in the physical thin sections. The Shapiro-Wilk normality test was used to check the normality of the DSR distributions. Non-parametric statistical tests were used to compare the DSR sample distributions because the Shapiro-Wilk test showed that the DSR distribution deviates from normality. The statistical package R (version 4.1.1) (R Core Team, 2021) was used for the statistical analysis and generation of graphs.

Virtual histology: visualization of the enamel microstructures
A total of 18 synchrotron radiation microtomographic scans were acquired on 9 individual teeth from Velia (Table 1). Overall, the visibility of enamel microstructures in the reconstructed images was good and comparable to that of the physical sections (Fig. 1). Daily cross striations, a neonatal line and other accentuated markings are visible. The anatomical morphology of these features make them recognizable despite of the presence of linear artefacts (even more visible in Fig. 2 as horizontal stripes) related to a non-optimal ring removal procedure.
Retzius lines and the neonatal line were more visible on slices that are 20-130 mm thick (Fig. 2). Cross striations were more easily discernable in sections where the thickness of the reformatted image is equivalent to the voxel size (Fig. 3). The neonatal line was clearly visible in all the reconstructed and physical thin sections (Fig. 4).

Comparison of measurements at different voxel sizes of daily secretion rates
We were able to record DSRs in all images reconstructed with isotropic voxel sizes of 2.0, 1.3 and 0.9 mm (Fig. 3). DSRs were determined through multiple estimates for ten virtual histological sections across the entire prenatal portion of the crown. Fig. 5 shows the boxplots comparing the distribution of DSR values in the virtual (isotropic voxel size of 0.9, 1.3 and 2.0 mm) and physical thin sections from deciduous central incisors of the same skeletal series ( Table 2).
The Fligner-Killeen test for homogeneity of variance between real histology and different voxel size SRmCT measures shows a statistically significant difference between the DSR variances (chi-squared = 15.658, DOFs [degrees of freedom] = 3, p < 0.05). The Kruskal-Wallis test shows a statistically significant difference between the DSR distributions (chi-squared = 60.327, DOFs = 3, p < 0.01). The pairwise Wilcoxon rank-sum test with continuity correction and Bonferroni correction of probabilities shows no significant difference between the physical sections and SRmCT measurements at 0.9 or 1.3 mm voxel sizes. Significant differences were only detected between physical sections and SRmCT measurements at 2.0 mm voxel size (mean difference = 1.3 mm, p < 0.01). Measurements also    differed significantly between the SRmCT measurements at 1.3 and 2.0 mm voxel sizes (mean difference = 0.97 mm, p < 0. 01) and slightly but significantly between 0.9 and 1.3 mm voxel sizes (mean difference = 0.66 mm, p < 0.05).
Overall, the virtual histological assessment of DSRs produces values comparable to those from the physical sections when the smallest voxel size is used. In fact, by increasing the voxel size, the partial volume effect (values of the greyscale levels associated with a single material changing in different parts of the image) is more evident. This artefact can partially mask features of interest if their size is within 2-3 times the voxel resolution and can cause an erroneous measurement of the parameters of interest (Bull et al., 2013;Withers et al., 2021), resulting in larger uncertainty of the DRS values (Table 2).
It is reasonable to compare the mean values between different individuals and between different measurement techniques. In fact, the DSR values obtained from the present virtual histological analysis are in agreement with those already estimated with classical histology for the same population and same tooth class, always in the prenatal enamel [Wilcoxon rank-sum test with continuity correction W = 3159.5, p > 0.05, reference data from the work by ]. Velia 387 lower deciduous central incisor. Vestibular-lingual reformatted images from SRmCT, isotropic voxel size = 1.3 mm, slice thickness = 1.3 mm. The fine enamel microstructures are discernible. On the right, the magnification of a portion of the left image shows enamel prisms (yellow line) and cross striations (white arrows). Each dark-and-bright couple highlighted by the white arrows represents a single cross striation, i.e. 24 h of enamel deposition.

Conclusions
The results presented in this study show the comparison between the quantitative assessments of growth parameters (DSR) of dental deciduous enamel in histological thin sections through optical microscopy and SRmCT images acquired at the SYRMEP beamline of Elettra.
Several factors can affect spatial and contrast resolution in CT images, including X-ray beam energy and quality (geometry, monochromaticity, coherence), sample size and composition, the mechanical stability of the setup, detector performances (scintillator screen, optics, camera), image preprocessing (artefact removal, phase-retrieval), and reconstruction algorithm.
When using 1.3 and 0.9 mm pixel sizes for the phase-contrast SRmCT measurements, it was possible to produce data that varied only slightly from that obtained from physical thin sections. Indeed, the range of deciduous DSRs from the microtomographic setup lay within the range of DSRs known for modern human incisors (Mahoney, 2012), and those published previously for the same archaeological series analyzed here .
Despite the above-mentioned factors affecting the reconstructed volumes and taking into account the quite heavy postprocessing applied to these datasets, the deciduous enamel microstructures are well visible in the final images. Retzius lines and the neonatal line were easily identified with the microtomographic setup used. Even if images from physical thin sections tend to have higher contrast, virtual imaging has the obvious advantage that it allows extraction of a very high number of slices of different thicknesses In summary, the results demonstrate that virtual histology can be applied to obtain reliable and quantitative measurements of prenatal DSRs in deciduous dental crowns.

Figure 5
Box and whiskers plot of the DSR distributions in classical and virtual histology at different voxel sizes. Virtual 0.9: virtual histology at 0.9 mm pixel size. Virtual 1.3: virtual histology at 1.3 mm pixel size. Virtual 2.0: virtual histology at 2.0 mm pixel size. Bold line -median. N -number of samples.