Within artiodactyls there is consistency between hypotheses—the spatiotemporal hypotheses are better supported than the buccolingual hypotheses (Fig. 5a; Table 1), but not among perissodactyls, where the Lingual and Spatial hypotheses best explained the observed modularity patterns with limited support for the Temporal hypothesis (Fig. 5b; Table 2). (Artiodactyl ES = 4.91; Perissodactyl ES = 1.82; Fig. 5; Tables 1, 2). Our modularity results lead us to suggest that the developmental timing of the premolar trigonid is a contributor to the evolution of artiodactyl molarized premolars. Perissodactyl modularity results, however, show little support for an interregional mode of molarization, implying perissodactyl molarization is primarily a within-premolar process. Our modularity results demonstrate relative divergence in modes of molarization across the p4-m1 boundary in artiodactyls and perissodactyls. However, both clades share support for some level of modularity at the p4-m1 boundary, likely resulting from conserved developmental programs across all mammalian teeth (Butler 1939, 1952, 1972; Osborn 1978; Kavanagh et al. 2007; Couzens et al. 2021; Guo et al. 2022). Differences among clades are evident in the relative performance of modularity hypotheses. While we discuss these modularity hypotheses as independent, dental development is complex and difficult to capture with morphometrics (Jernvall and Thesleff 2000; Hallgrímsson et al. 2009). We acknowledge that while some of our hypotheses produce stronger signals than others, our results also show that these modularity hypotheses are difficult to tease apart statistically (Fig. 5; Tables 1, 2). Teeth are a complex evolutionary system and their evolution is confounded by a long history of convergence with sporadic changes in evolvability (Luo et al. 2001, 2007; Salazar-Ciudad and Jernvall 2010; Harjunmaa et al. 2014; Couzens et al. 2021). An inability to statistically differentiate contributing factors to modular patterning is predicted empirically and theoretically in this framework (Mitteroecker and Bookstein 2007). None of the proposed modularity hypotheses can be regarded as mutually exclusive given the “palimpsest” like developmental mechanisms and extrinsic mechanical factors which regulate dental development (Klingenberg 2008; Hallgrímsson et al. 2009; Conith et al. 2021; Sadier et al. 2023). However, the differences in modular patterning revealed in our analysis may provide a useful heuristic in dissecting the complexities of cheek tooth evolution.
Artiodactyls and temporal influence on the evolution of molarization
We hypothesized that if the timing of p4 development influenced premolar molarization, there would be evidence of modularity between the molar morphology and the premolar cusps that had been developing the longest. This idea is represented by the Temporal hypothesis which was the best supported modularity hypothesis for artiodactyls (Table 1). Our results emphasize the importance of developmental timing of the premolar trigonid for the evolution of molarized premolars in artiodactyls.
The trigonid usually develops before the talonid, but the development of the trigonid happens at different times for the p4 and m1 (Luckett 1993; Jernvall and Jung 2000; Harjunmaa et al. 2014; Sadier et al. 2021). This implies that it is unlikely that the morphogens driving cusp development in the m1 would have a direct impact on p4 cusp development. The poor performance of the Spatial hypothesis in artiodactyls also directly supports this idea – the talonid is the most proximal portion of the p4 to the m1 and therefore should have shown stronger modularity if a spatial signal was vital for inter-regional morphogenesis. If an artiodactyl molarization signal exists in our data, it would not exist in the molar or the premolar enamel knots, but rather the developmental precursor of both the p4 and m1—the dental lamina comprised of cranial neural crest ectomesenchyme and ectodermal epithelium. We hypothesize that the molarization mechanism for artiodactyls exists prior to the formation of individual teeth in dental morphogenesis. To evaluate the potential of this hypothesis, we require additional developmental context for the deciduous fourth premolar (dp4), p4, and m1 in artiodactyls.
The primary dental lamina gives rise to both the dp4 and the m1, while the p4 is derived from a secondary dental lamina branching lingually from the dp4 (Štembírek et al. 2010; Dosedělová et al. 2015; Guo et al. 2022). The dp4, m1, and p4 have a common ancestry and therefore potential for co-option of developmental mechanisms across these teeth. Co-option of developmental mechanisms has been suggested to produce “novel” morphologies in the vertebrate head – an example is the migratory sequences of neural crest cells following embryonic circulatory and innervation pathways (Gammill et al. 2006; McLennan et al. 2010; Horie et al. 2018). Additional support for a conserved mechanism exists in the timing of development of these three teeth. In extant mammalian models, the m1 develops closest in time to the p4 rather than the dp4 (Moustakas et al. 2011; Wang et al. 2014; Wakamatsu et al. 2019). The dp4 achieves its immutable form (completed mineralization) by the time the m1 enters the late cap stage and the p4 starts to form from the secondary lamina – which is a notable transition point in dental development where inductive potential shifts from the epithelium to the mesenchyme (Santagati and Rijli 2003; Cho et al. 2007; Du et al. 2017; Malik et al. 2020). The start of development of the p4 is marked by an inductive shift and the transition of the M1 from cap to bell stage – posing a potential window for co-option of “molariform” determining mechanisms as the inductive potential shifts. From the developmental context, it is apparent that the p4 develops at a critical point in dental development for the m1 in artiodactyls, a potential reason why the Temporal hypothesis was the best supported of our a priori hypotheses.
Trigonid development in the cheek teeth is dictated by the onset of enamel knot development (Cho et al. 2007; Du et al. 2017; Malik et al. 2020). The onset of p4 development at the same time as the m1 inductive shift provides an opportunity for developmental co-option. Previous work has found more variation in trigonid than talonid morphology in mammalian cheek teeth, reflecting the increased amount of time and cellular resources available to those initial cusps (Janis and Lister 1985; Jernvall and Jung 2000; Jernvall et al. 2000; Kavanagh et al. 2007; Harjunmaa et al. 2014). These lines of evidence and our results can be synthesized to support the regulation of the p4 trigonid development relative to molar development as a major contributor in the evolution of molarized premolars in artiodactyls.
Perissodactyls and independence across the p4-m1 boundary
In contrast to artiodactyls, perissodactyl shape data favour independence of the p4 and m1. The most common modes of perissodactyl molarization do not result from an inter-regional mechanism. Perissodactyl taxa have convergently molarized premolars in many ways (Butler 1952; Holbrook 2015; Bai et al. 2019). Most of these modes of molarization involve the modification of cingula to join or separate cusps to make the premolars less or more molariform. Modification of cingulum morphology is an intrinsically morphogenetic process relying on the interactions of enamel knots within a tooth (Cai et al. 2007; Wang et al. 2014; Dasgupta et al. 2021; Sadier et al. 2021). Morphogenetic models have demonstrated that small perturbations to the placement, timing, or resources available to enamel knots can drastically alter tooth form – single parameter changes in crown complexity evolution simulation software have shown the ability to gain or lose cingula in mammalian teeth (Salazar-Ciudad and Jernvall 2002, 2004, 2010; Polly 2008). If cingula modifications were the primary mode of molarization in perissodactyls, small disruptions could have changed non-molariform premolars to molariform premolars and vice versa. We hypothesize that each occurrence of evolving molarized premolars in perissodactyls was likely followed by a rapid canalization event to prevent developmental perturbations from reversing molarization of the premolars. Increasing the modularity and reducing the integration of the premolars and molar shape are candidate evolutionary-developmental mechanisms for this phenomenon within the context of our results.
We suggest that the molarization potential for perissodactyls shifted from before enamel knot development to after enamel knot development relative to their most recent common ancestor with artiodactyls. Another piece of evidence from our results can be used to support this hypothesis—the performance of the Temporal hypothesis in perissodactyls. While the Temporal hypothesis was the best supported in artiodactyls, it was the poorest performing in perissodactyls. If perissodactyls increased modularity of cusp development in the premolars, it would favour the removal of any interactions between the premolars and molars to ensure effective premolar molarization.