Mesolithic and Chalcolithic Mandibular Morphology: Using Geometric Morphometrics to Reconstruct Incomplete Specimens and Analyse Morphology

: Human skeletal remains are routinely used to examine cultural and biological aspects of past populations. Yet, archaeological specimens are frequently fragmented/incomplete and so excluded from analyses. This leads to decreased sample sizes and to potentially biased results. Digital methods are now frequently used to restore/estimate the original morphology of fragmented/incomplete specimens. Such methods include 3D digitisation and Geometric Morphometrics ( GM ) . The latter is also a solidly established method now to examine morphology. In this study, we use GM - based methods to estimate the original morphology of incomplete Mesolithic and Chalcolithic mandibles originating from present Portugal and perform ensuing morphological analyses. Because mandibular morphology is known to relate to population history and diet, we hypothesised the two samples would di ﬀ er. Thirty - seven specimens ( 12 complete and 25 incomplete ) were CT - scanned and landmarked. Originally complete specimens were used as reference to estimate the location of absent anatomical landmarks in incomplete specimens. As predicted, our results show shape di ﬀ erences between the two samples which are likely due to the compounded e ﬀ ect of con -trasting population histories and diets.


Materials and Methods
This study is based on a total of 37 Mesolithic and Chalcolithic specimens originating from several sites located in the present Portugal (Table 1 and Figure 1).
All specimens were digitised using a Toshiba Astelion CT scanner (120 kV, voxel size 0.348 × 0.348 × 0.3, revolution time 0.75 s, spiral pitch factor 0.94) at the Faculty of Veterinary Medicine of the University of Lisbon. Segmentation ensued in 3D Slicer (Fedorov et al., 2012) using standard protocols described by O'Higgins (2017, 2018), Godinho, Spikins, and O'Higgins (2018) and . Fragmented specimens were virtually pieced together . After this procedure, coordinates were extracted from a total of 21 anatomical landmarks (LMs ; Table 2) from the most complete hemi-mandible of each specimen to capture mandibular morphology ( Figure 2). The use of left hemimandibles was favoured. When specimens were incomplete, the location of the missing LMs was estimated using the thin plate spline (TPS) function of the Geomorph R package following the recommendations of Godinho, O'Higgins, and Gonçalves (2020). Specifically, excessively incomplete specimens were not reconstructed because reconstruction error may be larger than inter-individual differences and hence may lead to biased results. This led to the exclusion of specimens missing more than 5 LMs. Only 1 specimen with 5 missing LMs was included and incomplete specimens most often lacked 2 LMs (Table A1). Mesolithic specimens were used as reference to geometrically estimate the location of the missing landmarks in the incomplete Mesolithic specimens. The same procedure was applied to the Chalcolithic sample using complete Chalcolithic specimens. Thus, chronological specific references were used. This is because the use of inappropriate references (i.e., specimens with meaningful morphological differences due to, e.g., contrasting population history) leads to larger errors in the estimation of the location of missing anatomical regions (Gunz et al., 2009;Neeser, Ackermann, & Gain, 2009;Senck, Bookstein, Benazzi, Kastner, & Weber, 2015). Nevertheless, we tested if this population-specific reconstruction approach could be driving the hypothetical inter-population differences. To that end, a non-population specific reference was created using all complete specimens from both chronologies for ensuing reconstruction of incomplete specimens. Results from both reconstruction approaches were then compared using Principal Component Analysis (PCA) and Discriminant Function Analysis (DFA) (given below).
After estimation of the missing LMs, standard GM analysis ensued. The landmark coordinate datasets of all specimens were superimposed using Generalised Procrustes Analysis (GPA). GPA removes the effects of size, location, and orientation and produces shape variables that are used in shape analysis. Shape differences between samples were examined via the PCA and visualised using Thin Plate Splines (TPS) that depict shape differences along the selected PCs. The impact of reconstruction approach was examined by Condyle, anterior A point on the antero-superior aspect of the mandibular notch (on the condyle) 20 Condyle, posterior The centre of the condyle from a posterior view 21 Ramus, posterior Posteriormost point of the ramus that is in line with the ramus root performing a PCA in which all specimens from the population-specific and non-population-specific reconstructions were included. DFA with 10,000 permutations and cross-validation scores was also used to examine the inter-population differences and was implemented using MorphoJ (Klingenberg, 2011). DFA was also used to examine the impact of population-specific vs non-population-specific reconstruction approaches (given above).
To examine if hypothetical morphological differences between Mesolithic and Chalcolithic samples are most likely related to population history or masticatory mechanics (given above), dental wear was also examined because it is related to the latter (Chattah & Smith, 2006;Smith, 1984). Wear magnitude was scored according to Smith (1984), averaged per individual and compared between the two samples using a boxplot and the Wilcoxon non-parametric statistical test.

Results
Digitisation and the use of GM-based reconstruction allowed estimating the original morphology of 25 originally incomplete specimens, thereby, increasing the sample size from 12 to 37 specimens.
Ensuing morphological analysis including all specimens shows limited overlap between the two samples (when plotting PC1 and 2, which explain ∼34% of the total variance; Figure 3). This is because the Mesolithic sample clusters mostly along the positive values of PC2 and the Chalcolithic sample mostly clusters along the negative values. Morphologically, this corresponds to Chalcolithic specimens having, e.g., generally wider rami, taller coronoid processes, shorter mandibular symphyses, and more alveolar prognathism. Although there is overlap between the two samples in PC1, the most extreme positive specimens are Chalcolithic. Such specimens have, e.g., shorter mandibular symphyses, more flexure of the posterior border of the ramus and more anteriorly positioned coronoid processes. PCA comparison of full samples including population-and non-population-specific reconstruction approaches show meaningful overlap between specimens and very little impact on the overlap between groups ( Figure A1). DFA is unable to reliably discriminate between population-and non-population-specific reconstructions of the same populations, and discriminates similarly between Mesolithic and Chalcolithic specimens based on the two different reconstruction approaches (Tables A2-A7). This shows inter-population differences are not due to reconstruction approach. DFA using 10,000 permutations shows significant inter-group differences (T-Square: p < 0.0001). Nevertheless, cross-validation results show misclassification of 4/11 Chalcolithic and 2/26 Mesolithic specimens (Table 3).
Dental wear magnitude is significantly heavier in the Mesolithic sample (Figure 4).

Discussion
The use of digital methods enabled the objective and reproducible reconstruction of 25 specimens that were originally incomplete. Thus, sample size was increased to a total of 37 specimens, which enabled further GM-based morphological analysis, a better representation of morphological variance, and hence more reliable results than if only the 12 originally complete specimens were included. As expected, morphological analyses show shape differences between the Mesolithic and Chalcolithic samples. Our results also show negligible differences between population-specific and non-populationspecific reconstructions. Thus, contrasting shapes between the two populations are not related to the reconstruction approach. Because mandibular morphology is known to relate to both population history (Buck & Vidarsdottir, 2004;Katz et al., 2017;Mounier et al., 2018) and masticatory mechanics (Galland et al., 2016;Katz et al., 2017;May, Sella-Tunis, Pokhojaev, Peled, & Sarig, 2018;Pokhojaev et al., 2019;von Cramon-Taubadel, 2011), these shape differences may relate to either of these two underlying factors.
Specifically, Iberian Mesolithic populations derived from previously existing Post-Glacial Upper Palaeolithic populations (Brewster et al., 2014;López-Onaindia, Gibaja, & Subirà, 2019). By no later than ∼5500 cal. BC, populations originating in the Middle East reached the Iberian Peninsula and introduced  agriculture (Martins et al., 2015;Zilhão, 2000Zilhão, , 2001. Ancient DNA studies show marked genetic discontinuity between Mesolithic hunter-gatherers and Neolithic agro-pastoralists, thus suggesting population replacement mainly in most European regions. However, such studies also show the presence of Mesolithic DNA in post-Mesolithic individuals and so at least some level of admixture exists between the local Mesolithic and the incoming Neolithic populations (Haak et al., 2015;Olalde et al., 2015Olalde et al., , 2019Villalba-Mouco et al., 2019). Because mandibular morphology is known to relate to population history (Buck & Vidarsdottir, 2004;Katz et al., 2017;Mounier et al., 2018), our results showing shape differences between the two samples are to be expected and likely also related to population history. Despite contrasts in mandibular shape being likely related to differences in population history in the samples, masticatory mechanics has also probably impacted mandibular morphology to some extent. The Mesolithic hunter-gatherer diet has been consistently said to be mechanically more demanding than the post-Mesolithic agro-pastoralist diet (Cohen, 1989;Larsen, 1997Larsen, , 2006Stock & Pinhasi, 2011). This is because the latter included more processed food items that made the overall diet softer and so less demanding (Cohen, 1989;Larsen, 1997Larsen, , 2006Stock & Pinhasi, 2011). Previous experimental studies using non-human mammal models have shown that differences in the material properties of diet impact skull morphology (Beecher & Corruccini, 1981;Bouvier & Hylander, 1984;He & Kiliaridis, 2003;Kiliaridis, Engström, & Thilander, 1985;Menegaz & Ravosa, 2017;Menegaz, Sublett, Figueroa, Hoffman, & Ravosa, 2009;Ravosa, Kunwar, Stock, & Stack, 2007;Ravosa et al., 2008a,b), and so differences in skull form between hunter-gatherers and agro-pastoralists are frequently linked to differences in the masticatory demands due to dietary differences (Galland et al., 2016;Katz et al., 2017;May et al., 2018;Pokhojaev et al., 2019;von Cramon-Taubadel, 2011). Our results showing significantly heavier wear in Mesolithic specimens are consistent with previous studies (Larsen, 1997;Lukacs, 1989) and support the hypothesis that mandibular shape differences between the two samples are also related to differences in diet and therefore in masticatory demands. This is because dental wear is known to relate to the material properties of food, and so it is frequently used to examine differences in diet and food pre-processing (Chattah & Smith, 2006;Smith, 1984).
In summary, our results confirm our prediction that mandibular morphology differs between Mesolithic hunter-gatherers and Chalcolithic agro-pastoralists. This is probably due to the compounded effect of population history and masticatory mechanics. Although we are unable to discern which of these factors impacted morphology the most, previous research about limb skeletal morphology showed that differences in mechanical loading fail to erase the impact of population history in bone form (Agostini, Holt Brigitte, & Relethford John, 2018). This is consistent with previous studies showing that mandibular morphology is impacted more by population history than by masticatory mechanics (Katz et al., 2017), and so the mandibular morphological differences detected in this study are most likely related to population history and, possibly, enhanced by contrasting masticatory demands.    Chalcolithic (population specific) 2 9 11 From cross-validation Chalcolithic (non-population specific) 5 6 1 1 Chalcolithic (population specific) 6 5 1 1 Mesolithic (non-population specific) 0 2 6 2 6 From cross-validation Chalcolithic (non-population specific) 8 3 1 1 Mesolithic (non-population specific) 2 2 4 2 6 From discriminant function Chalcolithic (non-population specific) 11 0 11 Mesolithic (population-specific) 0 2 6 2 6 From cross-validation Chalcolithic (non-population specific) 8 3 1 1 Mesolithic (population-specific) 1 2 5 2 6 Table A5: DFA with cross-validation of the population-specific reconstructed Chalcolithic sample vs the non-populationspecific reconstructed Mesolithic sample (see details about reconstruction parameters in Materials and Methods)

Allocated to
True group Mesolithic (non-population specific) Mesolithic (population specific) Total From discriminant function Mesolithic (non-population specific) 22 4 26 Mesolithic (population specific) 6 2 0 2 6 From cross-validation Mesolithic (non-population specific) 7 1 9 2 6 Mesolithic (population specific) 17 9 26 Figure A1: Shape PCA comparing population-specific and non-population-specific reconstruction of incomplete specimens. Note there is complete or almost complete overlap between specimens despite differences in reconstruction method.