15The extent to which evolutionary transitions are shaped by developmental bias remains 16 poorly understood. Classically, morphological variation is assumed to be abundant and 17 continuous, but if morphogenesis biases how traits vary than evolutionary transitions might 18 follow somewhat predictable steps. Compared to other anatomical structures, teeth have an 19 exceptional fossil record which documents striking evolutionary trajectories toward 20 complexity. Using computer simulations of tooth morphogenesis, we examined how varying 21 developmental parameters influenced transitions from morphologically simple to complex 22 teeth. We find that as tooth complexity increases, development tends to generate 23 progressively more discontinuous variation which could make the fine-tuning of dietary 24 adaptation difficult. Transitions from simple to complex teeth required an early shift from 25 mesiodistal to lateral cusp patterning which is congruent with patterns of dental 26 complexification in early mammals. We infer that the contributions of primary enamel knot 27 cells to secondary enamel knots which are responsible for patterning lateral cusps may have 28 been an important developmental innovation in tribosphenic mammals. Our results provide 29 evidence that development can bias evolutionary transitions and highlights how 30 morphogenetic modelling can play an important role in building more realistic models of 31 morphological character evolution.32 33 34 35Despite more than two centuries of investigation the extent to which 'laws of variation'(1) 36 govern the tempo and mode of evolution remains poorly understood (2-5). Both as a 37 simplifying assumption and because of ignorance about generating processes, morphological 38 variation has traditionally been assumed to be abundant and continuous (1, 6). However, 39 several decades of evo-devo studies have highlighted how interactions between genes, cells, 40 tissues, and the external environment, favour the generation of some types of variation, whilst 41 conspiring against others (2, 4, 7, 8). This 'anisotropic' model of morphological variation 42 implies that evolutionary transitions are more likely to follow some evolutionary pathways 43 than others, and suggests that knowledge of morphogenesis can be used to predict some 44 aspects of morphological evolution (9-11). But, the extent to which these evo-devo insights 45 have penetrated other areas of evolutionary biology, such as phylogenetics, is limited (12-14), 46 at least in part because empirical models of developmental bias are sorely lacking. As an 47 example, the most widely used model of character substitution in morphological 48 phylogenetics, the Markov k (Mk) model (15), assumes equal likelihoods of transition 49 between character states as well as character independence; both assumptions which can be 50 violated by developmental correlations (11, 12, 16, 17). Despite accumulating evidence for 51 wide-spread developmental non-independence and bias in how morphological traits vary (4) 52 t...