The relationship between dental complexity and mandibular shape: implications for dietary inference in stem mammals By Wade Harris (1646023) ORCID number: 0000-0002-6981-954 A dissertation submitted to the Faculty of Science, University of the Witwatersrand, Johannesburg, in fulfilment of the requirements for the degree of Master of Science in Palaeontology. Supervisor: Prof. Jonah N. Choiniere Co-supervisors: Prof. Roger Benson, Dr. Kathleen Dollman ii 1. DECLARATION I declare that this dissertation is my own, unaided work. It is being submitted for the Degree of Master of Science at the University of the Witwatersrand, Johannesburg. It has not been submitted before for any degree or examination at any other University. _____________________________________ Wade Harris Signed on the 10th day of June 2024 at the Evolutionary Studies Institute, University of the Witwatersrand, Johannesburg iii 2. ABSTRACT Non-mammalian cynodonts exhibit some of the first major morphological innovations that contributed to the success of their descendent lineage, mammals. This includes features that are hypothesized to enable adaptation to a highly varied suite of diets such as specialized tooth crowns with complex occlusal surfaces and a jaw muscle configuration where two major muscles contribute to jaw closing. Surprisingly, inferences on cynodont diets so far have been based on qualitative evidence, and the quantification of these dietary adaptations could assist in testing these inferences. Here, I evaluate the relationships of mandibular shape, dental complexity and the combined data on body mass and relative mandible size, to known diets in living mammals, to assess the utility of these ecomorphological proxies for inferring the diets of extinct non-mammalian cynodonts. To assess relationships between diet and jaw shape, I collected 12 fixed landmarks (type 2) and four sliding landmarks (type 3) for six non- mammalian cynodonts, 51 marsupial mammals and 211 placental mammals. Dental complexity (OPCR) values were collected using the R package molaR, for a sample of 19 non-mammalian cynodonts, 47 marsupial mammals and 193 placental mammals. Procrustes-aligned shape coordinates, OPCR values, body mass estimates and relative mandible size data were then subjected to phylogenetic Procrustes ANOVA regressions and phylogenetic regressions. Neither mandible shape nor OPCR are strongly correlated to diet (carnivory, frugivory, granivory, herbivory, nectivory, invertivory), but the combination of these variables are a somewhat reliable predictor of diet, particularly mandible shape. Extinct non-mammalian cynodonts have mandible shapes that are comparable to those of mammals, however they occupy a narrow morphospace and their OPCR scores are generally much lower than those of mammals. Consequently, when these variables were entered into a predictive framework for diet, I observed limited inferential power since non-mammalian cynodonts do not strongly resemble mammals. iv 3. ACKNOWLEDGEMENTS I’d like to thank everyone who played a role making this research successful. This endeavour would not have been possible without the mentorship of Prof. Jonah Choiniere, I have learned so much while working alongside you, thank you for all the guidance and encouragement. I’m deeply indebted to my supervisors: Prof. Jonah Choiniere, Prof. Roger Benson and Dr. Kathleeen Dollman. Thank you all for being incredibly patient, and generously providing your expertise. Particularly Prof. Roger Benson, I’m amazed at the invaluable set of skills I’ve learnt from our interactions, thank you. I’d like to thank Gideon Chinamatria for taking the time to CT-scan my specimens. To my parents, my brother, my partner, and friends thank you for believing in me and encouraging me to keep moving forward. To my lab-members, thank you for all your support and advice, especially to Frederick Tolchard, for the many insightful conversations, and to Chandelé Montgomery whose contagious enthusiasm has never failed to brighten up my days. This research would not have been possible without the support from Genus: DSI- NRF Centre of Excellence in Palaeoscience under Grant No. 86073. v 4. CONTENTS 1. DECLARATION ................................................................................................ ii 2. ABSTRACT ..................................................................................................... iii 3. ACKNOWLEDGEMENTS ............................................................................... iv 4. CONTENTS ...................................................................................................... v 5. LIST OF INSTITUTIONAL ABBREVIATIONS ................................................. ix 6. LIST OF FIGURES .......................................................................................... xi 7. LIST OF TABLES ......................................................................................... xvii 8. LIST OF SYMBOLS ....................................................................................... xx 9. INTRODUCTION .............................................................................................. 1 9.1 The Karoo Basin ............................................................................................ 1 9.2 Cynodonts and their dietary adaptations ....................................................... 2 9.3 Dietary inferences in the fossil record............................................................ 6 9.4 Dietary evaluation using craniodental anatomy ............................................. 7 9.4.1 Dietary inferences using mandible shape ............................................... 7 9.4.2 Dietary inferences using teeth................................................................ 11 9.4.3 Size and its association with diet .......................................................... 14 10. RESEARCH RATIONALE .............................................................................. 16 11. MATERIALS ................................................................................................... 18 11.1 Data samples ........................................................................................... 18 11.2 Phylogenetic data ..................................................................................... 18 11.3 Functional data ......................................................................................... 19 11.3.1 Diet .................................................................................................... 19 11.3.2 Body mass ......................................................................................... 19 12. METHODS ...................................................................................................... 21 12.1 Specimen imaging .................................................................................... 21 12.2 Tooth model development ........................................................................ 21 vi 12.3 Phylogeny ................................................................................................ 23 12.4 Diet categorization ................................................................................... 24 12.5 Landmarking protocol ............................................................................... 25 12.6 OPCR ....................................................................................................... 28 12.7 Statistical analyses ................................................................................... 29 12.7.1 Mandible shape variation ................................................................... 29 12.7.1.1 Phylogenetically-aligned principal component analysis (PCA) .......... 29 12.7.1.2 Phylogenetic ANOVA regressions (procDpgls) ................................. 30 12.7.1.3 Phylogenetically informed discriminant analysis (pDFA) ................... 31 12.7.2 Mandible size and its association with diet ........................................ 32 12.7.2.1 Phylogenetic regressions (pgls) ........................................................ 32 12.7.2.2 Phylogenetically informed discriminant analysis (pDFA) ................... 35 12.7.3 Dental complexity .............................................................................. 36 12.7.3.1 Generation of box and whisker plots ................................................. 36 12.7.3.2 Phylogenetic regressions (pgls) ........................................................ 36 12.7.3.3 Phylogenetically informed discriminant analysis (pDFA) ................... 37 12.7.4 Dietary inferences using mandible shape and dental complexity ...... 38 12.7.4.1 Phylogenetic ANOVA regressions ..................................................... 39 12.7.4.2 Phylogenetically informed discriminant function analysis (pDFA) ..... 39 13. RESULTS ....................................................................................................... 40 13.1 Mandible shape and its association to diet ............................................... 40 13.1.1 PCA ................................................................................................... 40 13.1.2 Phylogenetic ANOVA regressions (procDgls) .................................... 47 13.1.3 pDFA: mandible shape ...................................................................... 58 13.2 Mandible size and its association with diet ............................................... 64 13.2.1 Phylogenetic regressions (pgls) ......................................................... 64 13.2.2 pDFA: mandible size .......................................................................... 71 vii 13.2.2.1 pDFA: considering only mandible size .............................................. 71 13.2.2.2 pDFA: considering mandible size and shape .................................... 75 13.3 Dental complexity (OPCR) and its association with diet ........................... 79 13.3.1 Average dental complexity (OPCRavg): distribution and trends ........ 79 13.3.1.1 OPCRavg: box and whisker plots ..................................................... 79 13.3.1.2 OPCRavg: phylogenetic regressions (pgls) ...................................... 84 13.3.2 Summed dental complexity (OPCRsum): distribution and trends ...... 88 13.3.2.1 OPCRsum: box and whisker plots .................................................... 88 13.3.2.2 OPCRsum: phylogenetic regressions (pgls) ..................................... 93 13.3.3 Dental complexity of the first molar (OPCRm1): distribution and trends 97 13.3.3.1 OPCRm1: box and whisker plots ...................................................... 97 13.3.3.2 OPCRm1: phylogenetic regressions (pgls) ..................................... 102 13.3.4 Dental complexity of the second molar (OPCRm2): distribution and trends 106 13.3.4.1 OPCRm2: box and whisker plots .................................................... 106 13.3.4.2 OPCRm2: phylogenetic regressions (pgls) ...................................... 111 13.3.5 Dental complexity of the third molar (OPCRm3): distribution and trends 115 13.3.5.1 OPCRm3: box and whisker plots ..................................................... 115 13.3.5.2 OPCRm3: phylogenetic regressions (pgls) ..................................... 120 13.3.6 pDFA: OPCR ................................................................................... 125 13.3.6.1 pDFA: OPCRavg and OPCRsum .................................................... 125 13.3.6.2 pDFA: OPCRm1-m3 ....................................................................... 133 13.3.6.3 pDFA: OPCRm1-m3 and log10 body mass ...................................... 141 13.4 Dietary inferences using mandible shape and OPCR variables ............. 146 13.4.1 Phylogenetic ANOVA regressions .................................................... 146 13.4.2 pDFA: mandible shape and OPCR .................................................. 150 viii 14. DISCUSSION ............................................................................................... 157 14.1 Mandibular form and its correlation to dietary disparity in mammals ...... 157 14.2 Dental complexity’s correlation to dietary disparity in mammals ............ 165 14.3 Implications for dietary inference in cynodonts ....................................... 169 14.4 Implications for macroevolutionary evolution .......................................... 176 15. CONCLUSION .............................................................................................. 181 16. APPENDICES .............................................................................................. 183 16.1 Appendix A ............................................................................................. 183 16.2 Appendix B ............................................................................................. 204 16.3 Appendix C ............................................................................................. 207 16.4 Appendix D ............................................................................................. 262 16.5 Appendix E ............................................................................................. 270 16.6 Appendix F ............................................................................................. 320 17. REFERENCES ............................................................................................. 345 ix 5. LIST OF INSTITUTIONAL ABBREVIATIONS AMNH American Museum of Natural History, New York, New York, USA BP Evolutionary Studies Institute (formerly Bernard Price Institute for Palaeontological Research), University of the Witwatersrand, Johannesburg, South Africa Du Baa Duke University, Durham, North Carolina, USA FLMNH Field Museum of Natural History, Chicago, Illinois, USA GWU Center for Advanced Study of Human Palaeobiology, George Washington University, Washington, D.C, USA JSM University of Chicago, Chicago, Illinois, USA MCP Science Technology Museum, Porto Alegre, Brazil MCN Museu de Ciências Naturais da Fundação Zoobotânica do Rio Grande do Sul, Porto Alegre, Brazil MCZ Museum of Comparative Zoology, Harvard University, Cambridge, Massachusetts, USA NHM-PV Natural History Museum, London,England, UK NHMUK Natural History Museum, London, England, UK NMQR National Museum, Bloemfontein, South Africa NMS National Museum Scotland, Edinburgh, Scotland, UK OUMNH Oxford University Museum of Natural History, Oxford, England, UK PVL Instituto Miguel Lillo, San Miguel de Tucumán, Argentina SAM-PK Iziko, the South African Museum, Cape Town, South Africa UF Florida Museum of Natural History, Gainesville, Florida, USA UFRGS Universidade Federal do Rio Grande do Sul, Farroupilha, Brazil x UMMZ University of Michigan Museum of Zoology, Ann Arbor, Michigan,USA UMZC University Museum of Zoology, Cambridge, UK USNM Smithsonian National Museum of Natural History, Washington, D.C, USA UWBM UW Burke Museum of Natural History and Culture, Seattle, Washington, USA WLM Wits Life Science Museum, University of the Witwatersrand, Johannesburg, South Africa YPM Yale Peabody Museum of Vertebrate Zoology, New Haven, Connecticut, USA xi 6. LIST OF FIGURES Figure 1: Lithostratigraphy and vertebrate biozonation ranges of the Beaufort and Stormberg groups of South Africa's Karoo Supergroup showing the stratigraphic ranges of South Africa's cynodonts. ............................................................................ 2 Figure 2: Cynodont phylogeny alongside stratigraphic scale showing relationships and occurrences of members. .................................................................................... 5 Figure 4: Postcanine tooth form variation in a carnivorous, herbivorous, and invertivorous mammal. ............................................................................................. 12 Figure 5: Differences in tooth complexity (OPCR) in the first molar of a carnivorous and herbivorous mammal. ........................................................................................ 14 Figure 6: Tooth model generation for an exemplar specimen (Acinonyx jubatus). .... 22 Figure 7: Landmarks placed on examplar specimen. ............................................... 26 Figure 8: Scatterplot of log10 mandible centroid size and log10 body mass, illustrating an example of a significant interaction effect. ........................................................... 34 Figure 9: PC 1 and PC 2 axes associated with major mandible shape variation in fully landmarked mammal species. .................................................................................. 40 Figure 10: Landmark configurations corresponding to the PC 1 and PC 2 extreme values of fully landmarked mammal species. ........................................................... 41 Figure 11: PC 1 and PC 2 axes associated with major mandible shape variation in partially landmarked mammal and non-mammalian cynodont species. .................... 43 Figure 12: Landmark configurations corresponding to the PC 1 and PC 2 extreme values of partially landmarked mammal and cynodont species. ............................... 44 Figure 13: PC 1 and PC 2 axes of shape variation amongst partially landmarked mammal and non-mammalian cynodont species, partitioned according to major taxonomic groups. .................................................................................................... 46 Figure 14: Scatterplot of shape regression scores and log10 body mass, alongside the mandibular shape differences associated with the effect of increasing log10 body mass. ........................................................................................................................ 48 xii Figure 15: Scatterplot of shape regression scores and log10 mandible centroid size, alongside the mandibular shape differences associated with increasing log10 centroid size ........................................................................................................................... 49 Figure 16: Scatterplot of shape and the arcsine transformed carnivory values, alongside the mandibular shape differences associated with the partial effect of carnivory. .................................................................................................................. 50 Figure 17: Landmark configurations in a small-bodied and large-bodied carnivore. . 51 Figure 18: Scatterplot of shape and the arcsine transformed nectivory values, alongside the mandible shape differences associated with the partial effect of nectivory. .................................................................................................................. 52 Figure 19: Landmark configurations in a small-bodied and large-bodied nectivore. . 53 Figure 20: Scatterplot of shape and the arcsine transformed granivory values, alongside the mandible shape differences associated with the partial effect of granivory. .................................................................................................................. 53 Figure 21: Landmark configurations in a small-bodied and large-bodied granivore.. 54 Figure 23: Landmark configurations in a small-bodied and large-bodied invertivore. 55 Figure 22: Scatterplot of shape and the arcsine transformed invertivory values, alongside the mandible shape differences associated with the partial effect of invertivory. ................................................................................................................ 55 Figure 24: Scatterplot of shape and the arcsine transformed herbivory values, alongside the mandible shape differences associated with the main effect of herbivory. .................................................................................................................. 56 Figure 25: Landmark configurations in a small-bodied and large-bodied herbivore.. 57 Figure 26: Landmark configurations in a small-bodied and large-bodied frugivore... 57 Figure 27: Landmark configurations representing mandible shape differences in a correctly classified herbivorous mammal and herbivorous mammals misclassified as a carnivore or invertivore. ......................................................................................... 59 Figure 28: Landmark configurations representing mandible shape differences in a correctly classified carnivorous mammal and carnivorous mammals misclassified as an invertivore or carnivore. ....................................................................................... 60 xiii Figure 29: Landmark configurations representing mandible shape differences in a correctly classified invertivorous mammal and invertivorous mammals misclassified as a carnivore or herbivore. ...................................................................................... 62 Figure 30: Landmark configurations representing mandible shape of non-mammalian cynodonts estimated as carnivorous, herbivorous and invertivorous. ....................... 64 Figure 31: Scatterplot of log10 mandible centroid size and log10 body mass. ............ 67 Figure 32: Scatterplot of log10 mandible centroid size and log10 body mass, illustrating the interaction that log10 body mass with frugivory, relative to non-frugivorous mammals. ................................................................................................................. 68 Figure 33: Scatterplot of log10 mandible centroid size and log10 body mass, illustrating the interaction that log10 body mass has with granivory, relative to non-granivorous mammals. ................................................................................................................. 69 Figure 34: Scatterplot of relative mandible size and log10 body mass amongst mammals and non-mammalian cynodonts. .............................................................. 70 Figure 35: Box and whisker plots showing the distribution of average OPCR scores (OPCRavg) amongst non-mammalian cynodonts within qualitatively inferred diet categories. ................................................................................................................ 79 Figure 36: Box and whisker plots showing the distribution of average OPCR scores (OPCRavg) amongst major mammalian orders relative to non-mammalian cynodonts. ................................................................................................................ 80 Figure 37: Box and whisker plots showing the distribution of average OPCR scores (OPCRavg) amongst mammal species within major diet categories. ....................... 83 Figure 38: Box and whisker plots showing the distribution of average OPCR scores (OPCRavg) amongst mammal species within finer diet categories. ......................... 84 Figure 39: Scatterplot of average OPCR scores (OPCRavg) and log10 body mass, illustrating the interaction that log10 body mass has with granivory, relative to non- granivorous mammals. ............................................................................................. 87 Figure 40: Box and whisker plots showing the distribution of summed OPCR scores (OPCRsum) amongst non-mammalian cynodonts within qualitatively inferred dietary categories. ................................................................................................................ 89 xiv Figure 41: Box and whisker plots showing the distribution of summed OPCR scores (OPCRsum) amongst major mammalian orders relative to non-mammalian cynodonts. ................................................................................................................ 90 Figure 42: Box and whisker plots showing the distribution of summed OPCR scores (OPCRsum) amongst mammal species within major diet categories. ...................... 92 Figure 43: Box and whisker plots showing the distribution of summed OPCR scores (OPCRsum) amongst mammal species within finer diet categories. ........................ 93 Figure 44: Scatterplot of summed OPCR scores (OPCRsum) and log10 body mass, illustrating the interaction that log10 body mass has with granivory, relative to non- granivorous mammals. ............................................................................................. 96 Figure 45: Distribution of OPCR scores for the third-last postcanine tooth, or functionally analogous ‘first molar’ (OPCrm1), among non-mammalian cynodonts within qualitatively inferred dietary categories. ......................................................... 98 Figure 46: Box and whisker plots showing the distribution of OPCR scores for the first molar (OPCRm1) amongst major mammalian orders relative to non-mammalian cynodonts. ................................................................................................................ 99 Figure 47: Box and whisker plots showing the distribution of OPCR scores for the first molar (OPCRm1) amongst mammal species within major diet categories. ............ 101 Figure 48: Box and whisker plots showing the distribution of OPCR scores for the first molar (OPCRm1) scores amongst mammal species within finer diet categories. ... 102 Figure 49: Scatterplot of OPCR scores for the first molar (OPCRm1) and log10 body mass, illustrating the interaction that log10 body mass has with granivory, relative to non-granivorous mammals. .................................................................................... 105 Figure 50: Distribution of OPCR scores for the second-last postcanine tooth, or functionally analogous ‘second molar’ (OPCrm2), among non-mammalian cynodonts within qualitatively inferred dietary categories. ....................................................... 106 Figure 51: Box and whisker plots showing the distribution of OPCR scores for the second molar (OPCRm2) amongst major mammalian orders relative to non- mammalian cynodonts. ........................................................................................... 107 Figure 52: Box and whisker plots showing the distribution of OPCR scores for the second molar (OPCRm2) amongst mammal species within major diet categories. . 110 xv Figure 53: Box and whisker plots showing the distribution of OPCR scores for the second molar (OPCRm2) amongst mammal species within finer diet categories. ... 111 Figure 55: Distribution of OPCR scored for the posteriormost postcanine tooth, or functionally analogous ‘third molar’ (OPCRm3), among non-mammalian cynodonts within qualitatively inferred dietary categories. ........................................................ 116 Figure 56: Box and whisker plots showing the distribution of OPCR scores for the third molar (OPCRm3) amongst major mammalian orders relative to non-mammalian cynodonts. ............................................................................................................... 117 Figure 57: Box and whisker plots showing the distribution of OPCR scores for the third molar (OPCRm3) amongst mammal species within major diet categories. ..... 119 Figure 58: Box and whisker plots showing the distribution of OPCR scores for the third molar (OPCRm3) amongst mammal species within finer diet categories. ...... 120 Figure 59: Scatterplot of OPCR scores for the third molar (OPCRm3) and log10 body mass, illustrating the interaction that log10 body mass has with granivory, relative to non-granivorous mammals. .................................................................................... 123 Figure 60: Scatterplot of OPCR scores for the third molar (OPCRm3) and log10 body mass, illustrating the interaction that log10 body mass has with carnivory, relative to non-carnivorous mammals. .................................................................................... 124 Figure 61: Occlusal views of OPCR 3D colours maps and associated OPCR scores (OPCRavg and OPCRsum) amongst a correctly classified invertivorous mammal and invertivorous mammals misclassified as a herbivore or carnivore. ......................... 126 Figure 62: Occlusal views of OPCR 3D colours maps and associated OPCR scores (OPCRavg and OPCRsum) amongst a correctly classified carnivorous mammal and carnivorous mammals misclassified as an invertivore or herbivore ........................ 128 Figure 63: Occlusal views of OPCR 3D colours maps and associated OPCR scores (OPCRavg and OPCRsum) amongst a correctly classified herbivore and herbivorous mammals misclassified as an invertivore or carnivore. ........................................... 129 Figure 64: Occlusal views of OPCR 3D colour maps and associated OPCR scores (OPCRavg and OPCRsum) in cynodont species that pDFA iterations estimated as a carnivore, invertivore and herbivore. ...................................................................... 131 xvi Figure 65: Occlusal views of OPCR 3D colours maps and associated OPCR scores (OPCRm1-m3) amongst a correctly classified carnivorous mammal and carnivorous mammals misclassified as a herbivore or invertivore. ............................................ 135 Figure 66: Occlusal views of OPCR 3D colours maps and associated OPCR scores (OPCRm1-m3) amongst a correctly classified herbivorous mammal and herbivorous mammals misclassified as an invertivore or carnivore. ........................................... 136 Figure 67: Occlusal views of OPCR 3D colours maps and associated OPCR scores (OPCRm1-m3) amongst a correctly classified invertivorous mammal and invertivorous mammals misclassified as a herbivore or carnivore. ......................... 137 Figure 68: Occlusal views of OPCR 3D colour maps and associated OPCR scores in cynodont species estimated as a carnivore, invertivore and herbivore. .................. 140 Figure 69: Scatterplot of shape regression scores and OPCRm1 scores, alongisde the mandibular shape differences associated with the effect of increasing OPCRm1 scores. .................................................................................................................... 147 Figure 70: Scatterplot of shape regression scores and OPCRm3 scores, alongisde the mandibular shape differences associated with the effect of increasing OPCRm3 scores. .................................................................................................................... 149 Figure 71: Landmark configurations alongside associated OPCR 3D colour maps and OPCRm1-m3 scores amongst a correctly classified herbivorous mammal and herbivorous mammals misclassified as a carnivore or invertivore. ......................... 151 Figure 72: Landmark configurations alongside associated OPCR 3D colour maps and OPCRm1-m3 scores amongst a correctly classified invertivorous mammal and invertivorous mammals misclassified as a carnivore or herbivore. ......................... 153 Figure 73: Landmark configurations alongside associated OPCR 3D colour maps and OPCRm1-m3 scores amongst a correctly classified carnivorous mammal and carnivorous mammals misclassified as an invertivore or carnivore. ....................... 154 Figure B1: Time-scaled phylogenetic tree of mammal species. .............................. 204 Figure B2: Time-scaled phylogenetic tree used for non-mammalian cynodont species. .................................................................................................................. 205 Figure B3: Merged super-tree containing mammal and non-mammalian cynodont species. .................................................................................................................. 206 xvii LIST OF TABLES Table 1: Majority consumed diet categories and descriptions. .................................. 25 Table 2: Traditional, or ‘major’ diet categories and descriptions................................ 25 Table 3: Landmark number, count and description associated with the dermacation of the placement of each landmark. .............................................................................. 27 Table 4: An example of a pgls model which assesses whether herbivores have a different pattern of mandible allometry than non-herbivores. .................................... 33 Table 5: Phylogenetic ANOVA regressions (procDgls) assessing the association between mandible shape (Shape) and log10 body mass (BM), log10 centroid size (Size) and dietary variables amongst mammal species. ........................................... 47 Table 6: Confusion table for pDFAs which estimated diet using mandible shape. .... 58 Table 7: The estimated diets of non-mammalian cynodont species associated with pDFAs which made use of mandible shape. ............................................................. 63 Table 8: Phylogenetic regressions assessing log10 mandible centroid size's association with log10 body mass and dietary variables amongst mammal species. 65 Table 9: Confusion table for pDFAs which estimated diet using log10 mandible centroid size residuals and log10 body mass. ............................................................ 71 Table 10: The estimated diets of non-mammalian cynodont species associated with pDFAs that used log10 mandible centroid size residuals and log10 body mass. ........ 74 Table 11: Confusion table for pDFAs which estimated diet using mandible shape, log10 mandible centroid size residuals and log10 body mass. .................................... 75 Table 12: The estimated diets of non-mammalian cynodont species associated with the pDFAs that used mandible shape, log10 mandible centroid size residuals and log10 body mass. ....................................................................................................... 78 Table 13: Phylogenetic regressions assessing the association that OPCRavg has with log10 body mass, log10 mandible centroid size and dietary variables amongst mammal species. ...................................................................................................... 85 Table 14: Phylogenetic regressions assessing the association that OPCRavg has with log10 body mass, log10 mandible centroid size and dietary variables amongst living taxa. ................................................................................................................. 94 xviii Table 15: Phylogenetic regressions assessing the association that OPCRm1 has with log10 body mass, log10 mandible centroid size and dietary variables amongst living taxa. ........................................................................................................................ 103 Table 16: Phylogenetic regressions assessing the association that OPCRm2 has with log10 body mass, log10 mandible centroid size and dietary variables amongst living taxa. ......................................................................................................................... 112 Table 17: Phylogenetic regressions assessing the association that OPCRm3 has with log10 body mass, log10 mandible centroid size and dietary variables amongst living taxa. ........................................................................................................................ 121 Table 18: Confusion table for pDFAs which estimated the diet of mammal species using OPCRavg and OPCRsum scores. ................................................................ 126 Table 19: The estimated diets of cynodont species associated with the pDFAs which that used using OPCRavg and OPCRsum scores as training data. ....................... 132 Table 20: Confusion table for pDFAs which estimated diet using OPCRm1-m3 scores. .................................................................................................................... 133 Table 21: The estimated diets of cynodont species associated with the pDFAs which that used using OPCRm1-m3 scores as training data. ........................................... 139 Table 22: Confusion table for pDFAs which estimated diet using OPCRm1-m3 scores and log10 body mass. .............................................................................................. 141 Table 23: The estimated diets of cynodont species associated with the pDFAs which that used using OPCRm1-m3 scores and log10 body mass. ................................... 145 Table 24: Phylogenetic ANOVA regressions assessing mandibular shape's association with OPCR variables. ........................................................................... 146 Table 25: Confusion table for pDFAs which estimated diet using mandible shape and OPCRm1-m3 scores. ............................................................................................. 151 Table 26: The estimated diets of cynodont species associated with pDFAs which made use of mandible shape (first seven PCA axes) and OPCRm1-m3 scores. ... 155 Table A1: List of mammal species sampled, together with their specimen numbers and taxonomic affiliations. ...................................................................................... 183 Table A2: List of cynodont species sampled, together with their specimen numbers and taxonomic affiliations. ...................................................................................... 201 xix Table A3: CT scanning parameters of specimens that were scanned at the Evolutionary Studies Institute (ESI). ....................................................................... 203 Table C1: Diet data used for mammal species and the associated metadata. ....... 207 Table C2: Expected diets of non-mammalian cynodont species inferred based off qualitative observations and the associated metadata. .......................................... 261 Table D1: Body mass data for mammal species and associated metadata. ........... 262 Table D2: Body mass approximations used for non-mammalian cynodont species and associated metadata........................................................................................ 268 Table D3: Skull length (SL) measurements and the respective body mass estimates. ............................................................................................................................... 269 Table E1: OPCR scores used for the lower premolars (pm1-3) of mammal species. ............................................................................................................................... 270 Table E2: OPCR scores used for the lower premolars and molars (pm4; m1-m2) of mammal species. .................................................................................................... 283 Table E3: OPCR scores used for the lower molars (m2-m4) of mammal species. . 295 Table E4: Postcanine teeth counts reported amongst non-mammalian cynodont species. .................................................................................................................. 303 Table E5: OPCR scores used for the lower molars (m3-5) in non-mammalian cynodont species. ................................................................................................... 304 Table E6: OPCR scores used for the lower postcanines (pc1-3) of non-mammalian cynodont species. ................................................................................................... 306 Table E7: OPCR scores used for the lower postcanines (pc4-pc6) in non-mammalian cynodont species. ................................................................................................... 308 Table E8: OPCR scores used for the lower postcanines (pc7-9) in non-mammalian cynodont species. ................................................................................................... 310 Table E9: OPCR scores used for the lower postcanines (pc10-13) in non-mammalian cynodont species. ................................................................................................... 312 Table E10: OPCRsum and OPCRavg scores used for mammal and non-mammalian cynodont species. ................................................................................................... 314 xx 7. LIST OF SYMBOLS P5 5th percentile P95 95th percentile PP Posterior probability 1 8. INTRODUCTION 8.1 The Karoo Basin The sedimentary rock units of South Africa’s Karoo Basin contain an abundant, almost-continuous plant and animal fossil record from the Late Carboniferous (300 Ma) to the Early Jurassic (190 Ma) (Smith et al., 2020). These fossils (particularly from the Beaufort and Stormberg groups) document the early evolutionary history of major extant vertebrate groups including fish, amphibians, reptiles, birds (via dinosaurs), and most importantly mammals via their cynodont ancestors (Smith, 1990, Smith et al., 2020). Cynodonts are a key group of synapsids that includes all living mammals and their closest extinct relatives among Therapsida (Hopson and Kitching, 2001, Liu and Olsen, 2010). Within Cynodontia, several reciprocally monophyletic groups form a nested grade of successive sister-taxon relationships with Mammaliaformes, including tritylodontids, brasilidontids, trithelodontids and gomphodonts (Kemp, 1983, Rowe, 1988, Smith, 1990, Rowe, 1993, Liu and Olsen, 2010). These groups, which fall outside mammaliaforms, form a paraphyletic grade often referred to as “non-mammalian cynodonts”. Fossils of non-mammalian cynodonts are well represented in the Main Karoo Basin (MKB), with 24 recognized taxa (Figure 1) (Botha and Smith, 2020, Hancox et al., 2020, Rubidge and Day, 2020, Smith et al., 2020, Smith, 2020, Viglietti, 2020, Viglietti et al., 2020a, Viglietti et al., 2020b). Their first appearance in the MKB is documented in the Endothiodon Assemblage Zone of the Beaufort Group, which dates back the Early Late Permian (Wuchiapingian age) (Botha et al., 2007, Rubidge and Day, 2020), and their last appearance is in the Early Jurassic (Rhaetian age), as recorded in the Massospondylus Assemblage Zone (Crompton, 1964). Their temporal range, spanning approximately 58 Ma, documents the acquisition of mammalian characteristics (Luo, 2007, Lautenschlager et al., 2017). The cynodonts that emerged during this period often lived contemporaneously (Figure 1), and exhibited a wide range of size classes and feeding apparatuses (i.e., tooth forms and mandible forms) (Abdala et al., 2006, Kemp, 2016, Abdala and Gaetano, 2018, Hendrickx et al., 2019, Hendrickx et al., 2020). 2 8.2 Cynodonts and their dietary adaptations Living mammal species are characterized by their primary jaw joint being between the dentary and squamosal, and the lower jaw consists of a single bone, the dentary (Crompton, 1963, Crompton et al., 1972). Non-mammalian cynodonts have postdentary bones (angular, articular, surangular and splenial), and in early members, the jaw joint is between the quadrate and the angular. During the course of cynodontian evolution, these postdentary bones eventually become reduced and the quadrate and articular bones migrate to form part of the middle ear (Luo, 2007). This transition coincided with changes in muscle orientation, mandible shape, and tooth form and occlusion (Gow, 1978, Lautenschlager et al., Figure 1: Lithostratigraphy and vertebrate biozonation ranges of the Beaufort and Stormberg groups of South Africa's Karoo Supergroup showing the stratigraphic ranges of South Africa's cynodonts. Solid lines indicate known ranges, dotted lines indicated suspected ranges that have not been confirmed. Figure modified from Smith et al. (2020). 3 2017), which are often regarded as adaptations associated with the evolutionary success and ultimate radiation of mammals (Evans and Pineda-Munoz, 2018, Berkovitz and Shellis, 2018). Early members of Cynodontia, such as Procynosuchus, had a prominent dentary bone with a high coronoid process, present but highly reduced post-dentary bones and heterodont dentition. The occlusal surfaces of the posterior postcanine tooth crowns are more “complex” than the anterior teeth because they bear cuspules and a fully expressed cingulum. The teeth did not precisely occlude at this stage. According to Kemp (1979), these features implied that Procynosuchus ate insects and was somewhat adapted to chewing them instead of swallowing them whole. Cynodonts branching later than Procynosuchus and its close relatives like Thrinaxodon form the Eucynodontia (Figure 2). Even the earliest eucynodonts are distinguished from basal members by an even larger dentary, with postdentary bones that are reduced to a small rod. The coronoid process is tall, and there is a large angular process ventrally (Kemp, 2005). These features, together with a prominent fossa on the lateral surface of the coronoid process, indicate that these eucynodonts had already developed masseteric musculature (Crompton and Hotton III, 1967, Barghusen, 1968, Bramble, 1978, Lautenschlager et al., 2017). There was no mammal-like jaw joint, but the group possessed a secondary contact between the surangular and squamosal bones (Jenkins, 1971). Gomphodontia are an early-branching diverse group of eucynodonts that lived from the Middle Late Triassic (Seeley, 1895, Crompton, 1972, Crompton et al., 1972, Hopson et al., 1991, Kemp, 2005). Gomphodonts are characterized by the presence of buccolingually expanded molariform postcanine teeth that occluded in a manner similar to mammals, although it was unlikely to have been precise (Crompton et al., 1972, Kemp, 1982, Sues and Hopson, 2010, Hendrickx et al., 2016). Gomphodont postcanines are broadened and contain accessory cusps, ridges and basins that increase the complexity of their occlusal surface (Seeley, 1895, Sues and Hopson, 2010, Abdala and Gaetano, 2018, Hendrickx et al., 2019). The diets of gomphodonts are broadly inferred as herbivorous or omnivorous based off the presence of these gomphodont postcanines (Reisz and 4 Sues, 2000, Abdala and Ribeiro, 2003, Abdala and Sa-Teixeira, 2004, Hopson and Sues, 2006, Sues and Hopson, 2010, Hendrickx et al., 2020). Gomphodontia contains three main groups: Diademodontidae, Trirachodontidae and Traversodontidae (Figure 2) (Seeley, 1895, Liu and Abdala, 2014). Diademodontidae are the earliest-branching clade of gomphodonts and have a combination of gomphodont and sectorial teeth (Gow, 1978). Trirachodontidae branch after diademodontids, and are currently hypothesized to be a sister group to Traversodontids (Sidor and Hopson, 2017). Trirachodontids also have a combination of gomphodont and sectorial teeth, with fewer sectorial teeth than in diademodontids (Sidor and Hopson, 2017, Hendrickx et al., 2019). Traversodontidae are the most taxonomically diverse gomphodont group, and they formed a major component of the cynodont fauna in the Triassic of Gondwana. Traversodontid postcanines consist of mainly gomphodont teeth (Abdala and Ribeiro, 2003, Abdala et al., 2006, Sues and Hopson, 2010). They are the latest-surviving group amongst gomphodonts, with their last appearance dating back to the Late Triassic (Hopson, 1984, Sues and Olsen, 1990, Sues et al., 1992, Gow et al., 1993, Sues et al., 1999). The ~30 Ma that their lineage persisted records an increase in body sizes, concomitant with an increase in tooth complexity (i.e., qualitatively observed) (Martinelli and Soares, 2016), and these features are assumed to correspond to a change in food resources (Abdala and Ribeiro, 2010, Martinelli and Soares, 2016, Abdala and Gaetano, 2018). 5 Figure 2: Cynodont phylogeny alongside stratigraphic scale showing relationships and occurrences of members. The relationships within families of Gomphodontia are annotated. Figure modified from Ruta et al. (2013). 6 8.3 Dietary inferences in the fossil record Dietary inferences in the fossil record are made using a wide range of methods such as dental microwear, stable isotopes, and qualitative comparative studies (e.g., Goswami et al., 2005, Clementz, 2012, Pineda‐Munoz et al., 2016, Evans and Pineda-Munoz, 2018, Melstrom and Irmis, 2019, Morales-García et al., 2021). However, inferring diet from the fossil record is difficult because direct evidence of paleoecology (i.e., fossilized stomach contents, coprolites) is relatively rare or understudied (Martill et al., 1994, Kriwet et al., 2008, Zipfel et al., 2023). Nevertheless, reconstructing the diets of fossils is important because diet is a fundamental aspect of biology which can influence a species’ physiology and morphology (Carbone et al., 1999, Price and Hopkins, 2015). Additionally, reconstructing the diets of fossils is crucial for understanding past ecosystems. The importance of studying diet in deep time and its evolutionary history therefore calls for continued efforts to find osteomorphological traits that strongly correlate with dietary preferences (Grossnickle et al., 2019). Ecomorphological studies that are targeted at understanding the link between form and function are highly relevant to palaeobiology because our understanding of an extinct animal’s ecology hinges on the relationship that morphological adaptations have with behaviour in living organisms (Price and Hopkins, 2015, Benevento et al., 2019, Morales-García et al., 2021). Living mammal species have an incredibly diverse diet, and they encompass a wide range of dietary niches that evolved repeatedly among phylogenetically independent groups (Jones and Safi, 2011, Wang et al., 2020). This dietary variation is coupled with conspicuous differences in osteological components of the feeding apparatus, particularly the lower jaw and teeth (Hoshi, 1971, Ungar, 2010), making them an ideal natural testing ground for dietary ecomorphology. Non-mammalian cynodonts document the evolutionary origins of some of the major adaptations to which the adaptive success of Mammalia has been attributed. These adaptations include precise tooth occlusion and the squamosal- dentary jaw joint (Kardong, 2012, Bonnan, 2016). These changes enabled mammals to become the “champion chewers” that they are today, and the disparity of tooth forms, mandible shapes and sizes are widely regarded as morphological adaptations to the types of food eaten (Hoshi, 1971, Liem et al., 7 2001, Bonnan, 2016). By comparing the breadth of dietary adaptations in extant mammal species to non-mammalian cynodonts we can learn more about the evolutionary sequence that led to the first true mammals (Botha et al., 2005). Comparative ecomorphological studies often focus on craniodental features, particularly the mandible and teeth, because of the association that these features have with dietary adaptations (Frederich et al., 2008, Cassini, 2013, Hedrick and Dumont, 2018). Additionally, these elements are thought to be evolutionarily decoupled from adaptations of the sensory systems (Ross et al., 2012, Prevosti et al., 2012). Historically, studies of non-mammalian cynodont diets have used qualitative inference through anatomical comparisons of mandibular and dental morphology (e.g., Gow, 1978, Reisz and Sues, 2000, Fastovsky, 2001). A growing number of recent studies have attempted to establish quantitative means of dietary inference (Morales-García et al., 2021, Huttenlocker et al., 2021). This allows more explicit testing of non-mammalian cynodont diets. However, these inferences are limited if they are not transparently benchmarked against the known diets of living taxa. This thesis aims to evaluate the relationships between diet and the predictor variables of mandible shape, size (the combined information from body mass and relative mandible size), and dental complexity in extant mammals. My aims are: 1) to assess the strength of these associations in living mammal species, where diet is known; and 2) to assess the utility of these variables for inferring the diets of cynodont species. 8.4 Dietary evaluation using craniodental anatomy 8.4.1 Dietary inferences using mandible shape Mandible shape provides key information about the biomechanics of the feeding apparatus (Marcé-Nogué et al., 2017). The well-developed theory of lever mechanics provides testable links between the shape changes of the mandible, orientation of the muscle action and aspects of feeding such as gape and bite (Fabre et al., 2018, Sella-Tunis et al., 2018). Clear osteomorphological features have been hypothesized to have relevance to the functional adaptations of carnivorous and herbivorous diets (Radinsky, 1981a, 8 Liem et al., 2001, Cox et al., 2012, Barbero et al., 2023). For example, carnivorous mammals have a large temporalis muscle, which inserts on a large triangular shaped coronoid process (Figure 3) (Radinsky, 1981a, Hartstone‐Rose et al., 2012). The orientation of the temporalis increases the mechanical advantage and maximizes the muscle’s line of action, enabling a powerful bite force (Liem et al., 2001, Hartstone‐Rose et al., 2012, Hartstone‐Rose et al., 2019). Relative to carnivores, herbivorous animals have much large masseter and mandibular pterygoid muscles which inserts on an enlarged posterior mandible body (Figure 3) (Pérez-Barbería and Gordon, 1999, Crompton et al., 2010). The mandibular condyles of herbivores are elevated (Figure 3), and this together with the orientation of the masseter and pterygoid increases the mechanical advantage. This provides herbivores with the necessary force and freedom of movement (around the mandibular condyle) that is required for grinding plant material (Liem et al., 2001). Morphometrics is the study of shape variation and its covariation with other variables. The association, causes and effects that variables have with shapes are assessed by applying multivariate statistical analyses to sets of morphological variables (Bookstein, 1991). Early research in morphometrics entailed quantifying shape through linear distance measurements, angles or ratios (Bouvier, 1986b, Takahashi and Pan, 1994), but using these metrics became challenging as there was no consensus amongst researchers on suitable size correction methods. Additionally, linear distances were not always defined by homologous points and these measurements did not capture the general shape of the object well. These challenges prompted researchers to explore alternative methods of quantifying shape, which eventually gave rise to modern geometric morphometrics (Kendall, 1984, Rohlf and Slice, 1990, Bookstein, 1991, Rohlf and Marcus, 1993, Adams et al., 2004). In geometric morphometrics, shape is quantified as a set of landmark- based coordinates or outlines that represent biologically defined points. This enables the retention of the general shape or geometry of an object, particularly the relative spatial positions of points (Slice, 2007). 9 There are three main categories of landmarks. Type 1 landmarks are discrete juxtapositions which are distinct objectively locatable points in space, these include foramina and the points at which three bones meet. Type 2 landmarks are maxima of curvatures or points of correspondence indicated by geometry, such as the tip of a process or the notch of a bone. Type 3 landmarks are extremal endpoints which are defined in relation to the positions of other structures Figure 3: Representative variation in mandibular morphology in A: carnivorous mammal (a cheetah, Acinonyx jubatus; scale bar equals 39mm); B: herbivorous mammal (a vicuña, Vicugna vicugna; scale bar equals 54 mm). Mandibles shown in lateral view. Blue shading represents the temporal group’s muscle attachment sites, and the red shading represents the masseter muscle group’s attachment sites. Musculature based off Pérez-Barbería and Gordon (1999) and Hartstone‐Rose et al. (2012). 10 elsewhere on the bone, such as the anterior-most point of a bone and inter- landmark segments (Bookstein, 1991, Bookstein, 1985, Bookstein and Cutting, 1988, Cooke and Terhune, 2015, Palci and Lee, 2019). Generalized Procrustes Analysis (GPA) translates and rotates each specimen to minimize the squared summed distance between corresponding landmarks, and by doing this specimens are scaled to the same unit (centroid size). This brings the landmarks to a common coordinate system which enables differences in configuration (not orientation, scale or rotation) of shapes to be expressed (Bookstein, 1991, Rohlf, 1999, Slice, 2007). Previous ecomorphological studies that made use of mandible shape (i.e., using geometric morphometric frameworks) generally found that diet explains little of the variation in mandible shape (Hautier et al., 2008, Prevosti et al., 2012, Wang et al., 2021). However, there is often no standardization in dietary categorization and the proxies (e.g., Procrustes coordinates, PCA coordinates, partial warps) used to represent mandible shape which makes it difficult to compare their results and understand which factors cause studies to arrive at different conclusions. Procrustes coordinates directly represent shape variation, and studies that made use of these found that diet explains approximately 8% of the mandible shape variation in rodents (MANOVA R2=0.0082) and bovids (procDgls R2=0.08). Hautier et al. (2008) partitioned diet as hypercarnivory, animal-dominated omnivory, plant-dominated omnivory, and herbivory whereas Wang et al. (2021) partitioned diet according to browsers, mixed-feeders and grazers. Diet explained 42% (pgls using PCA coordinates of shape R2=0.42) of the variation in the major aspects of mandible shape as described by principal axes (PC) in bats (Nogueira et al., 2009, Hedrick and Dumont, 2018). Nogueira et al. (2009) partitioned the diet of bats into insectivory, frugivory, nectivory, and carnivory. Hedrick and Dumont (2018) instead categorized diet by assigning one of five character states, informed by the mechanical properties of food, whereby one was indicative of a liquid diet and five was indicative of a very hard diet. Prevosti et al. (2012) found that diet plays a key factor in mandibular shape of marsupial and placental carnivorous taxa using partial warps. Although no direct correlation statistic is reported, Prevosti et al. (2012) reports that there was still a 11 substantial amount of variation in mandible shape that phylogenetic patterns or diet could not explain. Diet was partitioned into hypercarnivores, meso-carnivores, insectivores, and herbivores. Outside Mammalia, the disparity of mandible shape has been associated with the diet of aquatic snakes, birds, and some groups of fish (Burress et al., 2016, Fabre et al., 2016, Navalón et al., 2019, Orkney et al., 2021). Although dietary hypotheses have been tested using mandibular morphometrics in extinct mammals (Morales-García et al., 2021), the mandible shapes of cynodonts have not been studied using a morphometric framework. 8.4.2 Dietary inferences using teeth Teeth operate as the main mechanical interface between food and an animal’s body, and teeth are directly involved in both food acquisition and food processing in mammals. Dental form and function have been the subject of many studies due to this close relationship (e.g., Crompton and Hiiemae, 1969, Sheine and Kay, 1977, Popowics, 2003, Evans and Sanson, 2006). Postcanine teeth in mammals are made up of premolars and molars. The premolars puncture and crush food, and the molars grind, crush, or shred food. The postcanine teeth are mainly adapted for mechanical processing of food within the oral cavity, and have stronger dietary signals relative to incisors and canines (Evans and Pineda- Munoz, 2018). For example, members of the Carnivora have specialized postcanine molars known as “carnassials” which bear a sharpened surface used to cut flesh (Figure 4) (Butler, 1946, Popowics, 2003). Insectivorous mammals generally have tribosphenic or dilambdodont molars, characterized by shearing crests and a deep crushing basin (Figure 4) (Strait, 1993, Butler, 1996). Herbivorous mammals have developed a suite of modifications enabling their plant-rich diet that aid in cellulose cell wall breakdown in many taxa, including hardened enamel lophs and crests and increased occlusal surface areas through the molarization of premolars and larger tooth size (Figure 4) (Janis and Fortelius, 1988). 12 Due to this qualitative association between tooth shape and diet, gross tooth morphology has historically been used to infer the diet of fossil species. The disadvantage of these qualitative observations is that they often rely on the presence of specific features, and the observations may be influenced by subjective interpretations, potentially leading to bias (Pineda‐Munoz et al., 2016). The first attempts to quantify tooth morphology entailed using geographic information systems (GIS) approaches in a novel way to quantify tooth shape without relying on landmarks (Zuccotti et al., 1998, Jernvall and Selänne, 1999, Ungar and Williamson, 2000). Further development of GIS methods gave rise to dental topography analysis- which is broadly defined as making use of a method that quantifies 2D/3D tooth shapes using a single metric without the use of landmarks (Dennis et al., 2004, Winchester et al., 2014). Figure 4: Postcanine tooth form variation in A, B; a carnivorous mammal (a cheetah, Acinonyx jubatus); C, D an herbivorous mammal (a sheep, Ovis aries); E, F an invertivorous mammal (moonrat, Echinosorex gymnura). Note differences in number of teeth and shape. Tooth rows shown in labial (A,C,E) and occlusal views (B,D,F). A,B,C scale bar equals 14mm. D scale bar equals 11mm. E,F scale bar equals 8mm. 13 There are many metrics that quantify various attributes of the tooth, such as relief index, Dirichlet normal energy and orientation patch count, or orientation patch count rotated (OPCR) (Berthaume et al., 2018). Amongst these metrics, OPCR is generally preferred for its broad applicability as a methodology, which enables inferences of diet in mammals (Evans et al., 2007, Santana et al., 2011) and reptiles (Melstrom and Wistort, 2021, Shipps et al., 2023). Orientation patch count (OPC) is a univariate metric that quantifies the surface complexity of a 2D/3D mesh model of a tooth surface. Dental complexity is a measure of the number of additional features, or ‘tools’, on a tooth surface (Evans et al., 2007). Conceptually, this analysis is a proxy for the mechanical processing capacity of a tooth. Processing capability is hypothesized to improve when more ‘tools’, or features, are added to the tooth surface, as it increases the tooth’s capability to divide food per mastication event. In other words, by adding more tools to the tooth, the tooth becomes more efficient at breaking food (Evans et al., 2007, Evans and Jernvall, 2009, Bunn et al., 2011). Hence, OPC can quantitatively reflect the differences in the mechanical processing capabilities between the teeth of grazers and carnivores. A grazer’s tooth often has more “tools” to break down plant material because plant cell walls contain cellulose which is difficult to digest and requires breaking down physically through extensive mastication, and this results in higher OPC scores. This contrasts a carnivore’s tooth, which generally have less “tools” as meat is readily digestible and requires minimal processing (mainly requires shearing), resulting in lower OPC scores (Figure 5) (Evans et al., 2007, Santana et al., 2011). OPC summarizes the slope orientation and topographic elevation of the occlusal region of a tooth into distinct patches (Figure 5). Contiguous regions of the tooth that have the same slope orientation and topographic elevation are binned together (forming one patch) according to eight categories based on the intercardinal direction that the patch is orientated in. The dental complexity is the count of this topographic patch count (Evans et al., 2007). OPC is particularly sensitive to the initial orientation of the tooth. To reduce this sensitivity, Evans and Jernvall (2009) modified OPC to be more robust to orientation differences in the input surface. This was done by rotating an occlusally aligned tooth clockwise and calculating OPC at each new orientation for a specified number of iterations and 14 averaging all OPC values together. This modified method is called orientation patch count rotated (OPCR). Dental complexity has proved to be a useful tool, largely because it enables dietary inferences without relying on specific homologous features. Dental complexity has significant correlations with diet in rodents, mammals, marsupials, bats, and lemurs (Evans et al., 2007, Santana et al., 2011, Godfrey et al., 2012, Smits and Evans, 2012).The correlation between dental complexity and diet in such diverse groups is important, as it potentially allows inferences of diet in extinct groups that don’t have close living relatives with similar ecological diversity. For example, this correlation has been used to infer diet in stem crocodylians and large-bodied diprotodontids (Melstrom and Irmis, 2019, White et al., 2021). It also facilitates evaluations of the dietary breadth of groups which have no modern analogues, such as multituberculates (Wilson et al., 2012). The dental complexity of cynodonts has only recently been evaluated, however these dental complexity values have never been compared to those of extant taxa with known dietary habits (Hendrickx et al., 2024). 8.4.3 Size and its association with diet Allometry is broadly defined as the study of how biological variables (e.g., shape, length, physiology), change with size (Huxley, 1924, Huxley and Teissier, 1936, Thompson, 1942, Kleiber, 1947, Huxley, 1950). A fundamental observation is the Figure 5: Differences in tooth complexity (OPCR) in the first molar of A, B: a carnivorous mammal (a cheetah, Acinonyx jubatus) and D, E: an herbivorous mammal (Vicuña, Vicugna vicugna). 15 square-cube law (or surface law), whereby an organism’s surface area and length should scale proportionally with mass (Huxley and Teissier, 1936, Thompson, 1942, Kleiber, 1947). The scaling relationships between mass other variables are powerfully influenced by this fundamental scaling principal. This can be explained arithmetically, as a cube with a length of 1cm, will have a surface area of 1cm2 and a volume of 1cm3. A cube that has double this length (2cm), will have a surface area of 4cm2 and a volume of 8cm3. Similarly, when an organism’s volume increases proportionally to mass, or when its length increases with mass to the power 1/3 (2=cubed root of eight), we say that it scales isometrically (West and West, 2012). Deviations from this are referred to as allometric scaling (West et al., 1999). Considering the effects of body mass is important because an animal’s ecology and evolution are considerably influenced by diet and body mass (McNab, 1986, Price and Hopkins, 2015, Cooke et al., 2022, Reuter et al., 2023). For instance, terrestrial mammalian invertivores have a maximum mass of 52,350g (in myrmecophages, Orycteropus afer) or 900g (in other invertivores such as Tenrec ecaudatus), because the distribution, abundance and energy content associated with feeding on exclusively small protein rich invertebrates becomes energetically infeasible at larger body sizes (Carbone et al., 1999, Cooke et al., 2022). Extant mammalian herbivores occupy a wide range of body sizes (13g— 2,915,040g), and include species with large body sizes, unique to herbivores (not attained by mammalian carnivores and invertivores). This distribution has been linked to large-bodied mammalian herbivores strategy to prioritize food resource abundance over food resource quality (Clauss et al., 2013, Pineda-Munoz et al., 2016). Mammalian carnivores occupy medium to large size classes (12,000g—30,000g), which has been linked to the energy constraints associated with predation as heavier carnivores are those that hunt large-bodied prey (Gittleman, 1985). Although size, and body mass have not generally been used as a quantitative proxy for dietary inferences in fossil species (but see Clark et al., 2023), these measures have often been used to justify qualitative dietary inferences in older publications (e.g., Gow, 1978). 16 9. RESEARCH RATIONALE Understanding diet in deep time is important for reconstructing past ecosystems, and exploring how dietary adaptations possibly played a role in the diversification and adaptive success of members from major amniote lineages, but inferring the diets of extinct species can be difficult. Traditionally, the ecology (i.e., diet, lifestyle) of extinct species would be inferred using qualitative approaches (Crompton and Jenkins, 1968, Jenkins and Parrington, 1976, Gow, 1978). However, in recent years there have been more studies that make use of quantitative approaches (White et al., 2021, Morales-García et al., 2021). This is largely because quantitative inferences are less subjective to human error and bias. However, this advancement in palaeobiology has been coupled with the development of more techniques that quantify aspects of morphology (Fortelius and Solounias, 2000, Evans et al., 2007, Boyer, 2008, Pineda‐Munoz et al., 2017, Shan et al., 2019, Fulwood et al., 2021), and it’s important to evaluate the strength of the correlations that these variables have with diet before drawing inferences about extinct species. Understanding cynodonts’ adaptations to different diets and the possible interplay that this had with body mass is important because derived cynodonts document many of the first major innovations to which the success of mammals has been attributed. These innovations include features related to improved masticatory ability such as tooth occlusion and the mammalian muscle configuration. Given that existing knowledge relied on qualitative approaches to infer the diet most cynodont species, predominantly though craniodental observations, this study will evaluate the applicability of geometric morphometrics and OPCR in inferring dietary preferences in cynodonts. While these methods are established for the crown group of mammals, cynodonts are located on the stem lineage and may not have developed the typical mammalian associations between diet, dental complexity, and mandibular shape. These relationships may still be useful in understanding cynodont’s dietary adaptations. This research has two primary aims: firstly, to understand the correlation that diet has with mandible shape, mandible size, body mass and dental complexity 17 (OPCR) in extant mammalian species; and secondly, to assess the utility of these variables for inferring the diets of extinct cynodonts. 18 10. MATERIALS 10.1 Data samples To evaluate the relationships of mandible shape and dental complexity to diet and size, I compiled a database of three-dimensional (3D) models of the mandibles of 262 taxa across the major mammalian clades (Afrotheria, Euarchontoglires, Laurasiatheria, Marsupialia, Monotremes and Xenarthra). I selected mammals from these clades to capture as much variance in evolutionary history, size and dietary habits as possible (Table A1). These data were mostly obtained from previous scans available from the online 3D data repository, Morphosource (www.morphosource.org). Two of the mammal species were scanned at the ESI using a Nikon Metrology XTH 22/320 LC dual source micro-CT system (Table A3, Castor canadensis_WLM unnumbered, Dugong dugon_WLM unnumbered). To determine the degree of confidence in dietary inferences made using mandible shape, size and OPCR for non-mammalian cynodonts, I broadly sampled members across fossils of extinct members of Cynodontia. This palaeontological dataset included early-diverging cynodonts such as Procynosuchus and Cynognathus, as well as more derived members such as gomphodont cynodonts and the mammaliaform, Megazostrodon (Table A2). Four of these specimens were CT scanned at the ESI (Table A3), and 3D models of Procynosuchus and Trirachodon were obtained from MorphoSource (Table A2). Photogrammetry models of the postcanine dentition from 14 gomphodont cynodonts were also included in the database (Table A2). 10.2 Phylogenetic data A time-scaled phylogenetic tree of mammals (Figure B1) was obtained from VertLife (www.vertlife.org), a multi-institutional project that aims to study the biodiversity of terrestrial vertebrates (Gauthier, 2019). The phylogenetic tree was assembled from input phylogenies inferred from DNA data (Upham et al., 2019a, Upham et al., 2019b). The topology of the phylogenetic tree that I constructed for extinct cynodonts (Figure B2) was based on Ruta et al. (2013) and Hendrickx et al. (2020). Ruta et al. (2013) made use of 150 discrete skeletal characters, and Hendrickx et al. http://www.morphosource.org/ http://www.vertlife.org/ 19 (2020) made use of mostly dental characters, but also included skeletal characters. The occurrences of cynodont species that were used to time scale the phylogenetic trees were obtained from the Palaeobiology Database (www.paleobiob.org). 10.3 Functional data 10.3.1 Diet The diets of mammal species were collected from the Elton Traits Ecological Archives (E095-178-D1), which describes diet quantitatively using data sourced from existing literature (Wilman et al., 2014). The authors of the dataset partitioned diet into nine diet categories: invertivory, endothermic vertebrates, ectothermic vertebrates, unknown vertebrates, piscivory, scavenger, frugivory, nectivory, granivory and herbivorous. Under this classification scheme, the diet of an obligate grazer would be scored as “100 % herbivorous” and a taxon that has more omnivorous dietary habits could be scored, for example, as “20% herbivorous”, “50% endothermic vertebrates” and “30% nectivorous” (Table C1). This approximated the dietary breadth of each taxon. For comparability, I additionally collected verbal descriptions regarding the diet of living specimens from Nowak (1999). The diet assigned to each extinct cynodont species was ascertained from the literature (Table C2). I refer to this as the “expected diet” throughout. These prior dietary inferences were often made using qualitative interpretations of tooth morphology, tooth wear and some authors considered body mass and the lever mechanics of the mandible (Jenkins and Parrington, 1976, Gow, 1978). The diets of many gomphodont cynodonts are broadly categorized as herbivorous based on their buccolingually expanded postcanine dentition, and amongst these the traversodontids are hypothesized to be the most specialized for herbivory (Kemp, 2005, Hendrickx et al., 2019). 10.3.2 Body mass The body mass of living species was collected from the Elton Traits Ecological Archives (E095-178-D1) (Wilman et al., 2014) (Table D1). http://www.paleobiob.org/ 20 When possible, body mass estimates for extinct cynodont species were collected from the literature (Table D2). However, there were some instances where body mass had not been reported, and I approximated body mass for five extinct taxa using skull length measurements reported in the literature (Table D3). The skull is not a major weight-bearing bone and does not have the strongest relationship with body mass (Campione and Evans, 2012). However, the postcranial skeletons of cynodonts are generally understudied, and skull lengths were more frequently reported in the literature. I modified the methodology used to infer body mass from Pavanatto et al. (2019), and used two equations to estimate the body mass. The first equation (BMa= 2.7 × (𝑆𝐿 ÷ 10) 3) used was developed by Quiroga, based on the relationship between skull length and body mass in therapsids (Quiroga, 1980, Quiroga, 1984). The second equation (BMb=(103.13×log(𝑆𝐿)−5.59 × 1000)) was developed by Van Valkenburgh, based on the relationship between skull length and body mass in extant mammals (Van Valkenburgh, 1990). The average of these computations was used as the body mass for five extinct taxa (Table D3). 21 11. METHODS 11.1 Specimen imaging 180 mammal species and two cynodont specimens were already rendered as 3D models available for download from MorphoSource. I made 3D models of the mandibles of an additional 88 mammal species and four cynodonts in VG Studio Max version 3.2.3. These reconstructions and existing 3D models were subsequently used to landmark specimens and create 3D tooth models. Photogrammetry images compiled by Christophe Hendrickx using a Dino-lite digital microscope were used to build photogrammetry models using Agisoft Photoscan Professional. These were exported as STL models and were subsequently used to create 3D tooth models. 11.2 Tooth model development Postcanine tooth models were created in Avizo 3D version 2021.1. Teeth were cropped by approximately following the enamel cap of the tooth. Cropping often resulted in the mesial and distal portions of the tooth being removed, as these contacted the teeth in front and behind it. However, this had a negligible effect since these portions are not occlusal and do not play an active role in chewing. Each tooth model was smoothed using the “Smooth surface” tool, with the number of iterations set to 100 and lambda set to 0.6. The resolution varied after smoothing, ranging from 7,000 to 12,000 faces. To standardize the data to comparable resolutions, I used the “Simplification editor” tool to reduce the surface’s complexity to 1,000 faces as the literature has established that 1,000 faces is sufficient in retaining the occlusal topography without obscuring the model with artefacts of scanning (i.e., cracks) (Melstrom and Wistort, 2021). The resultant number of faces and vertices present in each tooth model was recorded for screening purposes (Appendix E, Tables E1-E9). The quality of the tooth model was assessed based on the number of faces, number of vertices and whether the occlusal topography was retained in the model. The postcanine tooth models of 40 mammals had low-resolutions or had clearly lost a considerable amount of their occlusal topography after standardization and were disregarded in 22 further analyses (refer to specimens labelled “F” for OPCR calculated in Table A1). The number of tooth models generated for each specimen were dependent on the dental formula. Tooth models were generated for the three posteriormost teeth. Amongst eutherian mammal species, these were often molars or a combination of molars and premolars. Metatherian mammal species have four molars, and tooth models were generated for all four molars. The dental formula amongst the cynodonts sampled is highly variable and they could have up to 13 postcanine teeth (Martinelli et al., 2009). I generated tooth Figure 6: Tooth model generation for an exemplar specimen (Acinonyx jubatus). A: illustration of how tooth models were cropped from the mandible, shown in lateral view. Dark blue regions indicate areas to be cropped, while the lighter blue region highlights the resultant tooth model. Scale bar equals 40.61 mm. B: tooth model in occlusal view before smoothing and simplification. Scale bar equals 6.74 mm. C: tooth model in occlusal view after smoothing and simplification. Scale bar equals 6.88 mm. 23 models for the three most posterior postcanines along each specimen’s tooth row. Extinct cynodonts would have used these most-posterior postcanines to do most of the food processing (Gow, 1978), making them functionally analogous to the posteriormost teeth that were sampled for most mammals. Non-mammalian cynodonts exhibited a wide variation of tooth forms along their tooth row, however the three posteriormost postcanine teeth were often a combination of sectorial and gomphodont teeth. Some exhibited only gomphodont teeth in these positions, this included Cricodon, Luangwa, Massetognathus, and Santacruzodon. The posteriormost postcanines of Trirachodon sp. are only sectorial teeth. 11.3 Phylogeny The phylogenetic tree for extinct species (Figure B2) was made using Mesquite version 3.70 (build 490) (Maddison, 2007). I scaled the branch lengths in the statistical computing language and environment R, version 2.3.4 (R Core Team, 2013). This was done using the DatePhylo command from the strap package, version 1.6-0 (Bell et al., 2022), and made use of occurrences from Palaeobiology Database (www.paleobiob.org). The extant and extinct trees were merged to create a super-tree using the bind.tree command from the ape package, version 5.7-1 (Paradis and Schliep, 2019). The resultant composite tree was used during some statistical analyses (Figure B3). The analyses that only contained mammal species made use of the scaled phylogenetic tree for mammal species sourced from VertLife (Figure B1). These analyses included: phylogenetically-aligned principal component analysis (Collyer and Adams, 2021) (see results section 13.1.1); phylogenetic Procrustes ANOVA regressions (Adams, 2014) (see results sections 13.1.2 and 13.4.1); and phylogenetic regressions (Grafen, 1989, Orme et al., 2023) (see results sections 13.2.1, 13.3.1.2, 13.3.2.2, 13.3.3.2, 13.3.4.2, and 13.3.5.2). The analyses that contained a combination of cynodont and mammal species made use of the super-tree (Figure B3), this included: phylogenetically-aligned principal component analysis (Collyer and Adams, 2021) (see result section http://www.paleobiob.org/ 24 13.1.1); and phylogenetically informed discriminant function analyses (Motani and Schmitz, 2011) (see result sections 13.1.3, 13.2.2, 13.3.6, and 13.4.2). 11.4 Diet categorization The quantitative dietary data for mammal species was modified by creating an additional column that contains the sum of the values from the columns containing data on the consumption of vertebrates (endothermic vertebrates, ectothermic vertebrates, unknown vertebrates, piscivory, scavenger). This was done because ecomorphological studies that are relevant to palaeobiology do not make use of many subdivisions of carnivory (e.g., Melstrom and Irmis, 2019, Morales-García et al., 2021, White et al., 2021). Modifying this data resulted in a total of six dietary categories (invertivore, carnivore, frugivore, nectivore, granivore and herbivore), which were quantitatively encoded, representing the proportions at which mammal species consumed different food items. I converted the percentages to decimal form and arcsine transformed the quantitative dietary data to normalize the data for analyses. The individual effects of dietary variables were assessed separately in analyses to avoid multicollinearity. The quantitative dietary data was converted into two sets of categorical variables, but this was used for visualization purposes only. Two dietary categories were assigned to each species using the quantitative data and the descriptions from Nowak (1999). The first dietary category, majority consumed (Table 1), is based on the food type that each specimen ate most frequently and includes the same six categories as the quantitative dietary data. 25 Table 1: Majority consumed diet categories and descriptions. Diet category Description Invertivore Shrimp, krill, squid, crustaceans, molluscs, cephalopods, polychaetes, gastropods, orthoptera, ground insects, insect larvae, worms, orthopterans, flying insects, and other general invertebrates Carnivore Mammals, birds, reptiles, snakes, amphibians, salamanders, fish, scavenge, garbage, offal, carcasses, carrion, general unknown vertebrates Frugivore Fruit, drupes Nectivore Nectar, pollen, plant exudates, gums Granivore Seed, maize, nuts, spores, wheat, grains Herbivore Grass, ground vegetation, seedlings, weeds, lichen, moss, small plants, reeds, cultivated crops, forbs, vegetables, fungi, roots, tubers, legumes, bulbs, leaves, above-ground vegetation, twigs, bark, shrubs, herbs, shoots, aquatic vegetation, aquatic plants The second dietary category (Table 2), which I often refer to as traditional, or major dietary categories, contains a total of three categories and partitions the diet into the three more broad categories that are commonly used. Table 2: Traditional, or "major" diet categories and descriptions. Dietary category Description Herbivore Plant materials Carnivore Vertebrates Invertivore Hard and soft-bodied arthropods 11.5 Landmarking protocol 26 A geometric morphometric landmark dataset was created that characterized the shape of the mandible. This was informed by the location of teeth, biomechanics, Figure 7: Landmarks placed on examplar specimen (a cheetah, Acinonyx jubatus), in A: right lateral; B: ventral; C: anterolateral and D: posterior views. The yellow circles represent fixed landmarks; green circles represent fixed landmarks which delineate stop/start points of curve semi-landmarks, and the red circles represent the curve semi-landmarks bound between the stop/start points. 27 and the overall shape of the mandible. The landmark dataset contained a total of 16 landmarks: 12 type 2 fixed homologous landmarks and four type 3 series of curve semi-landmarks that were treated as sliding semi-landmarks in my analysis (Figure 7, Table 3). The landmarks were placed on the lateral surface of all specimens using Avizo’s landmarking tool (Figure 7). This recorded the landmarks as coordinates that encoded geometric information, such as distances and angles between landmarks (Slice, 2007). In this way, the coordinate system reflects the unique orientation and shape of each mandible. Table 3: Landmark number, count and description associated with the dermacation of the placement of each landmark. Landmark Type Count Description L1 Type 2 fixed homologous 1 Middle of incisors L2 Type 2 fixed homologous 1 Most lateral point of incisors L3 Type 2 fixed homologous 1 Most anterior point of tooth row after canine (If canine is present) L4 Type 2 fixed homologous 1 Most posterior point of tooth row L5 Type 2 fixed homologous 1 Beginning of anterior edge of coronoid process which marks the beginning of SL1 L6 Type 2 fixed homologous 1 Tip of coronoid process which marks the end of SL1 and start of SL2 L7 Type 2 fixed homologous 1 Middle anterior point of the condyle which marks the end of SL2 L8 Type 2 fixed homologous 1 Most lateral point of condyle L9 Type 2 fixed homologous 1 Most medial point of the condyle, marks the end of SL4 L10 Type 2 fixed homologous 1 Tip of the angular process, marks the start of SL3 and SL4 L11 Type 2 fixed homologous 1 Most anterior point of masseteric fossa L12 Type 2 fixed homologous 1 Ventral end of the mandibular symphysis, marks the end of SL3 SL1 Type 3 sliding 30 Anterior edge of the coronoid process SL2 Type 3 sliding 22 Posterior edge of the coronoid process SL3 Type 3 sliding 46 Ventral edge of the mandible SL4 Type 3 sliding 32 Posterior edge of the mandible 28 A total of 268 specimens were landmarked, comprising a dataset of 262 mammal species and six non-mammalian cynodont species. Two extinct cynodont species were only partially landmarked and did not contain the anterior-most landmarks L1 and L2 (Table 1) due to the fragmentary nature of the specimens. However, the rest of the sample was fully landmarked. Of the landmarked specimens, 127 mammal species were previously landmarked by Alberto Martin-Serra and Catherine Johnson, and I landmarked the remaining 135 mammal species and the six cynodont species. The landmarks were read into R using a script provided by Roger Benson (see Appendix F, R script for mandible shape analyses). The script reflected left-sided landmarks to ensure that the compiled landmark list only consisted of landmarks for the right mandible (or reflected left mandible). The script resampled the points along each curve semi-landmark series to equal counts. I created two functional landmark lists, one called the fully-landmarked set, containing only mammal species that were fully landmarked. And another, called the partially-landmarked set, containing mammal and cynodont species which were partially landmarked and did not contain the anterior-most landmarks, L1 and L2 (Table 3). I used the earlier mentioned script provided by Roger Benson to drop L1 and L2 in specimens that were fully landmarked. Following reading the landmarks into R, both sets (full-landmarked and partially- landmarked) of the landmark data were scaled, rotated and aligned to the mean shape by computing generalized Procrustes analysis (GPA). This was done in R by making using of the gpagen function from the geomorph package, version 4.0.5 (Adams and Otárola‐Castillo, 2013). The outputs of GPA include (i) Procrustes coordinates, which are used as a shape variable in my analyses, and (ii) the centroid size of the mandible, which is used as a size variable in my analyses. 11.6 OPCR Tooth complexity was calculated in R using the OPCr command from the molaR package, version 5.3 (Pampush et al., 2016b). The minimum face count threshold was set to three triangles per patch, this threshold is established to capture important morphological patterns (Melstrom and Wistort, 2021). Visualizations of 29 these scores were generated using the OPC3d function from the molaR package. This renders the mesh faces and assigns them to patches, these colour maps were inspected to identify and correct possible errors. This procedure was done for a total of 259 specimens, comprising 240 mammal species and 19 non- mammalian cynodont species (Appendix E, Tables E1-E9). Average and summed OPCR scores were computed in R using the rowSums and rowMeans functions from native R. In metatherians, the first molar was excluded from the average and sum calculation. The posteriormost three teeth were selected because this is hypothesized to be analogous to the last three teeth in eutherian mammals (O'Leary et al., 2013, Williamson et al., 2014). 11.7 Statistical analyses All statistical analyses were computed in R (Appendix F) and made use of the geomorph version 4.0.5 (Adams and Otárola‐Castillo, 2013), caper version 1.0.2 (Orme et al., 2023), and the custom code available from Motani and Schmitz (2011). 11.7.1 Mandible shape variation 12.7.1.1 Phylogenetically-aligned principal component analysis (PCA) I assessed the major axes of evolutionary variation in mandible shape amongst mammal species using a phylogenetically-aligned principal component analysis on the Procrustes shape coordinates from the fully-landmarked dataset of mammal species (n= 262). A phylogenetically-aligned PCA aligns the data to the axis of greatest phylogenetic signal, not the axis of greatest dispersion (Collyer and Adams, 2021). This was done using the gm.prcomp command from the geomorph package version 4.0.5 (Adams and Otárola‐Castillo, 2013). A separate phylogenetically aligned principal component analysis was computed to assess how the morphospace of cynodonts (n=6) compares to that of mammal species (n=262). This analysis made use of the use of the Procrustes shape coordinates associated with the partially-landmarked dataset. I generated shape-difference plots using the plotRefToTarget command from the geomorph package. The mean shape of the Procrustes coordinates was calculated using the mshape command from the geomorph package (Adams and 30 Otárola‐Castillo, 2013). The shape associated with the extreme ends (minimum and maximum) of the PC axes were compared to the mean shape. I used the adjectives “long” and “short” to describe anteroposterior variation and the adjectives “tall” and “low” to describe dorsoventral variation. 12.7.1.2 Phylogenetic ANOVA regressions (procDpgls) I assessed the association that mandible shape variation had with dietary variables and body mass by computing Procrustes ANOVAs (procDpgls) using the procD.pgls function from the geomorph package. These tested statistical hypotheses and described patterns of shape variation and covariation within a set of Procrustes shape coordinates. These hypotheses were evaluated in a phylogenetic framework that made use of permutation processes (Adams and Otárola‐Castillo, 2013, Adams, 2014, Adams and Collyer, 2015, Adams and Collyer, 2016, Adams and Collyer, 2018). Considering the shared evolutionary history amongst species is important because assuming independence of species traits causes an inflation of p values, potentially leading to false-positive inferences (Felsenstein, 1985, Grafen, 1989). I evaluated the association between individual predictor variables and mandible shape (e.g., models of the form Shape ~ Herbivory). Since many studies have shown that skeletal elements show shape allometry, the partial (e.g., models of the form Shape ~ Body Mass + Herbivory) and interaction effects (e.g., models of the form Shape ~ Body Mass * Herbivory) of dietary variables were assessed. The model “Shape ~ BM + Herbivory” assesses whether herbivory influences mandible shape after accounting for the effects of body mass on shape variation. If the partial effect of herbivory is significant in this model, the answer to this question is yes (Grafen and Hails, 2002). If the model “Shape ~ BM * Herbivory” is significant it suggests that herbivores have a different pattern of shape allometry to non-herbivores (Grafen and Hails, 2002, Franzese and Kam, 2009). I explored how predictor variables affected mandible morphology in models that were significant in explaining mandible shape variation by plotting the shape variation implied from the procDpgls models. The coefficients of a variable are interpreted as the effect that the predictor (e.g., diet) has on the response variable (i.e., shape) for every unit change. The effects of predictor variables were 31 visualized by generating shape deformation plots, where the mean shape was represented by model’s intercept coefficients and “effect” of the predictor variable was characterized by scaling the coefficients of the respective predictor variable with an exaggeration factor. The model “Shape ~ BM * Herbivory” assesses whether herbivores have different patterns of shape allometry than non-herbivores. There is no existing technique/method available that evaluates the shape deformations associated with significant interaction effects in a geometric morphometric framework. To understand the effects of body mass on shape within dietary variables I manually approximated the shape disparity associated with significant interaction effects. This was done by plotting the landmark configurations of the associated small- bodied and large-bodied species within the diet category. However, this only provides a broad approximation. There is also no current method available to evaluate the Akaike information criterion (AIC) in a geometric morphometric framework, instead the R2 values from procDpgls ANOVA tables were used to compare the goodness of fit amongst models. 12.7.1.3 Phylogenetically informed discriminant analysis (pDFA) I evaluated how informative mandible shape is for estimating diet by computing a phylogenetically flexible discriminant function analysis (pDFA) using the custom code available from Motani and Schmitz (2011). I followed the methodology used in Morales-García et al. (2021) and used the coordinates of the first seven PCA axes as the input predictor variables, these coordinates summarized most of the total mandible shape variation (99.99%, see section 13.1.1). The pDFA analyzes the association that mandible shape (PC axes) has with dietary categories (herbivore, carnivore and invertivore) in mammal species and used the trends found from this association to estimate the diet of mammal and non-mammalian cynodont species based on the input predictor variables. Each taxon is assigned a posterior probability (PP) that coincides with the chances/probability that the taxon belongs to the assigned diet based on the associations found between mandible shape and the dietary categories. The posterior probability of an assigned diet being high is indicative of a high 32 confidence that this taxon belongs in the diet; inversely a low probability is indicative of a lower confidence level. Posterior probabilities above 0.75 are considered “high”, and posterior probabilities less than 0.50 are considered “low” throughout this study. My sample of mammal species contained an unequal number of herbivores (n=147), carnivores (n=43) and invertivores (n=72). Biased samples such as these, with unequal representation among classes can strongly affect posterior estimates and to correct for this I followed the methodology of Chapelle et al. (2020) and Hermanson et al. (2022) using iterative random sampling of an equal number of species from each dietary category. I computed 400 iterations of a pDFA, and each pDFA iteration randomly sampled 30 carnivores, 30 herbivores and 30 invertivores from my larger sample of mammal species. This meant that each iteration had 90 mammal species, and under these parameters the random guess success rate is 33.33% (the probability of randomly guessing the correct class= 1/3; error rate of 66.67%). I used the median, the 5th percentile (P5) and the 95th percentile (P95) of the percentage of correct classifications per iteration to evaluate the overall success. Mammals that were frequently misclassified, particular