Convergence and constraint in the cranial evolution of mosasaurid reptiles and early cetaceans

Abstract. The repeated return of tetrapods to aquatic life provides some of the best-known examples of convergent evolution. One comparison that has received relatively little focus is that of mosasaurids (a group of Late Cretaceous squamates) and archaic cetaceans (the ancestors of modern whales and dolphins), both of which show high levels of craniodental disparity, similar initial trends in locomotory evolution, and global distributions. Here we investigate convergence in skull ecomorphology during the initial aquatic radiations of these groups. A series of functionally informative ratios were calculated from 38 species, with ordination techniques used to reconstruct patterns of functional ecomorphospace occupation. The earliest fully aquatic members of each clade occupied different regions of ecomorphospace, with basilosaurids and early russellosaurines exhibiting marked differences in cranial functional morphology. Subsequent ecomorphological trajectories notably diverge: mosasaurids radiated across ecomorphospace with no clear pattern and numerous reversals, whereas cetaceans notably evolved toward shallower, more elongated snouts, perhaps as an adaptation for capturing smaller prey. Incomplete convergence between the two groups is present among megapredatory and longirostrine forms, suggesting stronger selection on cranial function in these two ecomorphologies. Our study highlights both the similarities and divergences in craniodental evolutionary trajectories between archaic cetaceans and mosasaurids, with convergences transcending their deeply divergent phylogenetic affinities.


Introduction
Over their 390 Myr history, more than 60 lineages of tetrapods have independently reinvaded aquatic ecosystems (Vermeij and Motani 2018). The shared constraints they faced as part of this transition led to many textbook examples of evolutionary convergence in terms of feeding ecology, sensory biology, and locomotion, among others (Kelley and Pyenson 2015). One comparison that has received relatively little attention is that between archaic cetaceans (the Eocene and Oligocene ancestors of modern whales, dolphins, and porpoises) and mosasaurids (a clade of Late Cretaceous marine squamates). Both groups show similar raptorial ecomorphotypes, such as putative megapredators (Gallagher 2014;Voss et al. 2019); changes in postcranial anatomy, such as loss of sacral attachment and changes in limb morphology (Uhen 2010;Lindgren et al. 2011); and a shift from axial propulsion in the form of undulation to more efficient locomotion based on caudal oscillation (Buchholtz 2001;Lindgren et al. 2010Lindgren et al. , 2011. The earliest fully aquatic cetaceans (the basilosaurid lineage of archaeocetes) existed for more than 7 Myr before giving rise to the two neocete groups that survive to the present day: the odontocetes (toothed whales) and the mysticetes (baleen whales) Coombs et al. 2022). Mosasaurids, on the other hand, diversified into three parallel lineages early on in their evolutionary history (Russellosaurina, Mosasaurinae, and the more basal Halisaurinae), and radiated in a series of different waves until their extinction at the Cretaceous/Paleogene (K/Pg) boundary (Everhart 2005). Both mosasaurids and early cetaceans ultimately achieved near global distributions (Polcyn et al. 2014;Buono et al. 2019), high levels of taxonomic diversity (Polcyn et al. 2014;Marx and Fordyce 2015), and notable craniodental disparity (Fitzgerald 2010;Boessenecker et al. 2017b;Coombs et al. 2022;Cross et al. 2022;MacLaren et al. 2022).
Despite being frequently cited as a classic example of evolutionary convergence (Kelley and Pyenson 2015), similarities in ecomorphology between extinct marine tetrapods have only recently begun to be investigated using rigorous quantitative methods. Much of the focus of this research has been on cranial and dental morphology due to the wealth of fossilized remains and the ecological information that can be extracted (Kelley and Motani 2015;Motani et al. 2015;Stubbs and Benton 2016;Fischer et al. 2017;Reeves et al. 2021;Cross et al. 2022). In this paper, we quantitatively analyze cranial evolution in mosasaurids and early (fully aquatic) cetaceans during the first ca. 20 Myr of their evolutionary histories. We explicitly test for possible instances of ecomorphological convergence in the skulls and teeth between the groups; based on previous qualitative comparisons (Gallagher 2014;Kelley and Pyenson 2015), we predict a high level of convergence between megapredatory mosasaurids (e.g., Mosasaurus, Tylosaurus) and basilosaurid archaeocetes. In addition, we anticipate high convergence scores between putative small-prey specialists with an elongate snout. Finally, previous studies have suggested that the shift from axial-based to caudal-based locomotion resulted in increased efficiency and more effective colonization of open-ocean niches (Fish 2001;Lindgren et al. 2011), with ramifications for feeding and sensory ecology; we therefore hypothesize that skull ecomorphology will exhibit trajectory shifts with the acquisition of new locomotor techniques. Institutional

Methods
Data Sampling.-Our analyses focus on the skulls of 21 mosasaurid species (1 halisaurine, 11 russellosaurines, and 9 mosasaurines) and 17 cetacean species (4 archaeocetes, 5 toothed mysticetes, and 8 odontocetes) ( Table 1). The Oligocene cetacean Kekenodon has uncertain phylogenetic affinities; we follow previous work in placing the taxon as sister to the Neocete node (Clementz et al. 2014) and group it in our analyses with the other archaeocetes belonging to the family Basilosauridae. We took 12 linear measurements of each skull and jaw, either directly on the specimen or from high-precision 3D models (Fig. 1). Where neither option was available, measurements were taken from figured specimens using ImageJ (v. 1.53), and these data were cross-checked using other photographs and information from associated papers. We used our measurements to calculate 10 morphofunctional ratios with well-established functional and biomechanical outcomes, for example, mechanical advantage for jaw adduction calculated from mandibular lever arms (Anderson et al. 2011; Table 2). Functional ratios were adapted from previous studies (Anderson et al. 2011;Stubbs and Benton 2016;MacLaren et al. 2017;Fischer et al. 2020) and selected specifically to enable viable comparisons between mosasaurids and cetaceans (Supplementary Material).
To place our results into ecological context, we searched the literature for observations and evidence regarding the feeding and locomotor ecology of our study species (Supplementary Tables 1, 2). Given the breadth of taxa in this dataset, as well as inevitable uncertainties on paleoecology and life history for fossil species, we used relatively broad categories of diet (apex, fish/squid, benthic) and locomotion (anguilliform, sub-carangiform, and carangiform) following published studies (Pauly et al. 1998;Kelley and Motani 2015;Gutarra and Rahman 2022). Inferences about feeding ecology were not based on morphological information considered in our dataset so as to avoid circular reasoning; rather, we considered preserved stomach contents and craniodental features such as tooth wear, which fell outside the scope of our measurements.
Ecomorphospace Occupation, Phylogeny, and Disparity.-All analyses were carried out in the statistical software R v. 4.1.0 (R Core Team 2021). The ecomorphological dataset was passed through a completeness threshold of 45% per taxon, then z-transformed and converted to a Euclidian distance matrix. Pairwise biplots and correlations between all traits were computed using the psych v. 2.1.3 package. We employed two different types of ordination: (1) principal coordinates analysis (PCoA) using the ape v. 5.5 package, applying the Caillez correction for negative eigenvalues (Paradis et al. 2004); and (2) nonmetric multidimensional scaling (NMDS), using the vegan v. 2.5-7 package (Oksanen et al. 2007), with two dimensions predefined. NMDS is better for visualizing morphospace (as it can account for all variation in two axes); however, as it is nonmetric and cannot be used for statistical analyses, PCoA was also computed for quantitative use.
Skull size is an important factor in marine vertebrate ecology (McCurry et al. 2017b), and here we visualize it via two proxy measurements: skull length (commonly used for marine reptiles) and bizygomatic width, here defined as the maximum distance between the outer edges of the squamosals (commonly used in cetaceans) (Pyenson and Sponberg 2011). In an attempt to assimilate marine reptile and cetacean datasets, we chose to employ both measurements. Size was used here for scaling data points in ordination analyses; it was not used as an independent ecomorphological trait in itself. The natural logarithm of size metrics was used to explore frequency of differentsized taxa within and between the two groups.
We created a phylomorphospace to visualize ecomorphological trends across the evolution of both groups and test for convergence. Our composite tree combines recently published topologies for cetaceans and mosasaurids Strong et al. 2020). Taxa not included in these studies were grafted on the phylogeny using the phytools v. 0.7-80 and paleotree v. 3.3.25 packages (Bapst 2012;Revell 2012), based on their  Fig. 2A). We then dropped all tips for which we have insufficient (i.e., not passing the completeness threshold) or no data, using the ape v. 5.6-2 package (Paradis et al. 2004). The resulting tree was time-scaled using the minimum branch length algorithm (minimum = 3 Myr), using the paleotree v. 3.3.25 package (Bapst 2012 and Plioplatecarpus) had outlying data points based on isolated teeth, which we chose to remove, as they extended their respective temporal ranges by more than 10 Myr. The oldest age of the range was used to calibrate the tree (dateTreatment="firstLast" argument in the timePaleoPhy function). The root was then manually increased to ensure a mid-Carboniferous (318 Ma) split between Reptilia and Synapsida (Brocklehurst et al. 2022). Disparity (both sum of ranges and sum of variances) was calculated via the dispRity v. 1.6.1 package (Guillerme 2018) for both cetaceans and mosasaurids, as well as for major subclades (Russellosaurina, Mosasaurinae, Halisaurinae, Basilosauridae, Odontoceti, and Mysticeti), without rarefaction. All PCoA axes were used to calculate disparity, as the loadings on each axis are low.
Convergence Analyses.-Upon reviewing the results of ordination analyses, we chose a number of taxon pairs to be tested for inter-and intraclade ecomorphological convergence ( Table 3). We applied the C1 metric of Stayton (2015), which compares the morphological distance of two taxa with the morphological distance between their respective ancestral nodes, and thus quantifies how much of this difference has been lost through putative evolutionary convergence. A C1 value closer to 1 indicates greater convergence (Grossnickle et al. 2020). We used the first two PCoA axes (17.9% of variation) and all axes (100% of variation), with significance tested using the convevol v. 1.3 package (Stayton 2014) using 1000 Brownian simulations of character evolution for each pair. To test for the influence of long branches, we also ran these tests with the node determined by the minimum branch length (i.e., divergence in the Aptian).

Results
Morphological Evolution.-The earliest mosasaurids in our sample, the Turonian russellosaurines Russellosaurus and Tethysaurus, had small, relatively gracile skulls (Fig. 2B), likely limiting their diet to small prey items. Later russellosaurines and all mosasaurines radiate throughout the ecomorphospace, with no clear trajectory (Fig. 2B). Several back-and-forth occupations of novel and more ancestral phenotypes are observed; for example, the early mosasaurine Clidastes and the later (and larger) mosasaurine Plotosaurus both exhibit more longirostrine skulls with elongate teeth and a relatively small area of temporal musculature (Fig. 2B). Ecomorphological variation is present within genera with multiple species, such as Mosasaurus and Prognathodon (Fig. 2B).
Mosasaurid apex predators like Prognathodon, Mosasaurus, and Tylosaurus are split into two distinct regions of ecomorphospace, with Prognathodon exhibiting more robust snout and mandibles, indicating higher stress-resistance during biting in this genus (Fig. 2B). The durophagous mosasaurid Globidens dakotensis is not separate from other mosasaurids, rather plotting among the large apex predators (Fig. 2B). In both the sum of variances and sum of ranges disparity analyses, mosasaurids demonstrate higher mean disparity than cetaceans (Fig. 3A, Supplementary Fig. 1). Furthermore, in the sum of ranges analysis, both individual mosasaurid subclades (Russellosaurina and Mosasaurinae) exhibit higher disparity than the cetacean subclades (Fig. 3B). However, when sum of variance is used, the unusual mysticete Janjucetus hunderi drives a higher disparity result in this subclade (Fig. 3A, Supplementary Fig. 2).
Unlike their mosasaurid counterparts, the earliest fully aquatic cetaceans (basilosaurid "archaeocetes") had large skulls with extensive areas of temporal musculature and robust teeth (low crown aspect ratio) and plot close to the megapredatory mosasaurids Mosasaurus and Tylosaurus (Fig. 2B, Supplementary Fig. 3). All "archaeocetes" plot in a similar region of ecomorphospace (Fig. 2B) and show low ecomorphological disparity when compared with both mosasaurid clades and more derived cetaceans (Fig. 3B). Variation in this group is spread along an axis describing postorbital robusticity, with the basilosaurid Basilosaurus exhibiting a deep postorbital skull and large, robust anterior dentition compared with the more flattened, Snout width just before orbit:snout length Anterior mechanical advantage Distance between fulcrum and coronoid process:distance between fulcrum and anterior tip of dentary tooth row (mandible length) Opening mechanical advantage Distance between fulcrum and retroarticular process (mosasaurs)/ angular process (cetaceans):distance between fulcrum and anterior tip of dentary tooth row (mandible length) Relative temporal musculature Temporal fenestra length:skull length shallow-snouted cranium of Kekenodon ( Supplementary Fig. 4). Oligocene toothed mysticetes are more disparate than both odontocetes and basilosaurids (Fig. 3B), which is reflected in both the results of the disparity analysis as well as their spread across the   Fig. 2). Archaic mysticetes (Mystacodon selenensis and Coronodon havensteini) plot close to basilosaurids, whereas more crownward forms (aetiocetids) occupy a region of ecomorphospace characterized by long, thin snouts and smaller, narrower teeth. The mammalodontid J. hunderi consistently stands apart from the other cetaceans; it has a large relative temporal fenestra size, with high ratios for snout width, snout depth, and orbit diameter ( Supplementary Fig. 5) and plots as an outlier to both cetacean and mosasaurid ecomorphospace occupation (Fig. 2B). Early odontocetes evolved a suite of features associated with more longirostrine snouts and smaller teeth, indicating a somewhat more constricted ecomorphospace occupation by early odontocetes and aetiocetes. The basal odontocete Simocetus rayi plots within the center of mosasaurid ecomorphospace, despite notable morphological differences between Simocetus and the majority of mosasaurids.
With the exception of the toothed mysticete J. hunderi (see below), cetaceans in general have shallower snouts, smaller orbits, and more variably sized-and often larger-temporal fenestrae (Supplementary Figs. 5, 6) than mosasaurids. Furthermore, cetaceans show a wider range of adductor mechanical advantage, including the species with the lowest (Aetiocetus weltoni) and highest (J. hunderi) values ( Supplementary Figs. 5, 6). The size range for mosasaurids and cetaceans (using both metrics) is similar, with the cetacean distribution indicating larger skulls overall than mosasaurids (Fig. 3C,D). Only two pairs of functional traits are significantly correlated across both mosasaurs and cetaceans: snout depth to temporal fenestra length (mosasaur R = 0.67; cetacean R = 0.84) and snout depth to snout width (mosasaur R = 0.80; cetacean R = 0.75) (Supplementary Figs. 5, 6). Some of the other correlated traits in cetaceans are likely due to one outlying taxon (J. hunderi; Supplementary Fig. 5). Neither the NMDS nor the PCoA plots clearly distinguish long-from robust-snouted species (Fig. 2,  Supplementary Fig. 7). Putative apex predators, inferred as active hunters of large vertebrate prey, plot at higher values on both axes irrespective of the ordination method (e.g., the mosasaurids Tylosaurus bernardi and Prognathodon overtoni, and the cetacean Basilosaurus isis). There is no obvious association between locomotion guild and either skull ecomorphology or dietary class ( Supplementary Figs. 3, 8). However, it should be noted that both dietary class and locomotion guild have a high percentage (around 65%) of missing data (Supplementary Material).
Convergence Tests.-Statistical analyses identify a number of different taxa as convergent in their skull ecomorphology, albeit at different levels (Table 3). Three mosasaurid-cetacean pairs were statistically convergent for both sets of PCoA axes tested: Gavialimimus almaghribensis versus Waipatia maerewhenua; T. bernardi versus  Table 3).

Discussion
Differences in Evolutionary Trajectory.-Early mosasaurids and archaeocete cetaceans occupy clearly distinct regions of the ecomorphospace, possibly reflecting their different terrestrial ancestries. Little is known about the ecology of semiaquatic mosasauroids ("aigialosaurs"), with most research instead focusing on their phylogenetic relationships to other squamates (Carroll and Debraga 1992). However, these were small reptiles with skulls and teeth comparable to those of early mosasaurids and a probable diet of small prey (Carroll and Debraga 1992;Bardet et al. 2008;Cross et al. 2022). By contrast, semiaquatic archaeocetes not included in this study (e.g., Protocetus) were relatively large and powerful, with a diet that likely consisted of a wide variety of prey types and sizes (Fahlke et al. 2013). The difference in ecomorphospace occupation between basilosaurids and early mosasaurids could also reflect available niche space. In the aftermath of the K/Pg mass extinction, early cetaceans likely faced relatively little competition other than selachians (Lindberg and Pyenson 2006). This was not the case during the Cenomanian-Turonian radiation of mosasauroids, which had to navigate coexistence with other marine reptiles, including large platypterygiine ichthyosaurians, both long-and short-necked plesiosaurians, sharks, and large teleosts (Bardet 1994;Fischer 2016;Reeves et al. 2021;Cross et al. 2022). It is possible that marine ecosystem turnover and the demise of ichthyosaurs at the end of the Cenomanian allowed mosasaurids to diversify and occupy higher trophic levels (Bardet et al. 2008;Cross et al. 2022).
Mosasaurids and cetaceans both radiated during times of high marine productivity (Polcyn et al. 2014;Pyenson et al. 2014), but did so at different stages in their respective evolutionary histories. Our results suggest that mosasaurids radiated in skull ecomorphology soon after becoming fully aquatic; however, it should be noted that the fossil record of these early forms is poor (Cross et al. 2022). This is in contrast to basilosaurid cetaceans, whose skulls remained comparatively ecomorphologically conserved until the origin of neocetes during the latest Eocene (Boessenecker et al. 2017b;Coombs et al. 2022 ; Fig. 2B). The ecomorphological evolution of mosasaurids lacks an obvious pattern. In several cases, later and highly derived forms plot in proximity to less-derived predecessors; one striking example of this is the mosasaurine Plotosaurus bennisoni, which occupies a region of cranial ecomorphospace similar to that of more basal mosasaurids (Fig. 2) such as Clidastes and Tethysaurus (Fig. 2B, Supplementary Fig. 4), despite its highly derived akinetic skull and postcranial anatomy (Lindgren et al. 2007;LeBlanc et al. 2013). In fact, our analysis recovered Plotosaurus as statistically convergent with one of the oldest mosasaurid species in the analysis, the russellosaurine Tethysaurus nopcsai (Table 3), demonstrating that for the craniodental characteristics investigated in this study, both early (basal) and late (derived) mosasaurids from a wide range of phylogenetic clades may have adopted similar functional roles, albeit likely at different scales given the discrepancy in body sizes between the taxa (Fig. 2). At least six mosasaurid taxa occupy a region of low NMDS 2 values, with many of these species exhibiting elongate rostra and narrow dentition (e.g., Ectenosaurus, Plotosaurus, Tethysaurus) often associated with rapid jaw adduction and fast-prey capture. Although specimen selection may have influenced overall values and placement in ecomorphospace (e.g., P. bennisoni UCMP 32778 is possibly a juvenile; LeBlanc et al. 2013), the size-independent nature of most of the functional characteristics used precludes large differences to be expected from ontogeny, and we find no evidence for severe ontogenetic allometry in mosasaurids in the literature. Rather than radiating out from one common source and not reverting, mosasaurids from different times and phylogenetic clades are recovered in similar regions of ecomorphospace, indicating a recurring longirostrine ecomorphology transcendent of phylogenetic relatedness (Fig. 2B). Overall, mosasaurids do not follow a clear trajectory within ecomorphospace occupation in relation to phylogeny or temporal occurrence, but rather occupy a range of ecomorphospace indicative of widespread niche partitioning within and between clades (Schulp et al. 2013).
The trajectory of early cetacean skull ecomorphological evolution is much easier to discern than for mosasaurids. Basilosaurids all occupy a similar region of ecomorphospace (Fig. 2B, high NMDS 2). The earliest mysticetes in the study (Mystacodon and Coronodon) plot near basilosaurid ecomorphospace; subsequently, both mysticetes and odontocetes evolved along similar (but nonidentical) trajectories toward shallower, longer snouts with smaller teeth. We interpret these changes as adaptations to feeding on smaller prey (reducing the necessity for high bending resistance in the snout and mandible), as also reflected in the gradual emergence of simplified teeth and the attendant need to swallow prey whole (Peredo et al. 2018). Innovations in feeding ecology may have been a major driver of neocete (mysticetes + odontocetes) diversification (Marx and Fordyce 2015;Boessenecker et al. 2017b). Some highly distinctive taxa like the odontocete Inermorostrum and the toothed mysticete Mammalodon were too incomplete to include in our analysis, and early toothless mysticetes were not considered ecologically comparable to mosasaurids; the inclusion of these taxa would likely have increased the ecomorphological disparity of Oligocene neocetes even more so than is recovered with the taxa sampled here (Figs. 2B, 3).
Some cetaceans diverge significantly from the trajectories seen in the rest of the clade. For example, the earliest odontocete Simocetus rayi plots among the mosasaurids and is at least slightly convergent with the small plioplatecarpine mosasaurid Selmasaurus johnsoni (Table 3). This result may be spurious, as Simocetus has a highly unusual skull shapeincluding edentulous premaxillae and a ventrally deflected rostrum (Fordyce 2002)-not captured by our functional trait measurements. These and other features have led to Simocetus being interpreted as a benthic suction feeder, a lifestyle seemingly never adopted by mosasaurids. Any similarities between the two, such as a relatively short and deep snout, are thus likely superficial rather than functional in nature.
The same may be true for the unusual Oligocene toothed mysticete Janjucetus hunderi, whose deep, blunt snout and large eyes may reflect a route to a megapredatory ecomorphology quite removed from that of mosasaurids (e.g., Mosasaurus spp. and Prognathodon currii); these morphological features may also be directly involved in suction feeding (Fitzgerald 2010(Fitzgerald , 2012Young et al. 2012). This taxon is an outlier in our analyses, and despite plotting closest to the mosasaurid Prognathodon solvayi, the two are not significantly convergent. Janjucetus hunderi has the largest relative temporal fenestra size and anterior mechanical advantage of all the species in this study, suggesting a slow and powerful bite, whereas P. solvayi has values that are comparatively unremarkable compared with other taxa in the analysis. Janjucetus has been compared with various other secondarily aquatic tetrapods, including the plesiosaur Rhomaleosaurus (Fitzgerald 2006), the metriorhynchid Dakosaurus (Young et al. 2012), and the pinniped Hydrurga (Fitzgerald 2006). The deep snouts of all these taxa are well adapted for resisting torsional stress during grip and tear feeding (Taylor 1992;Fitzgerald 2006;Young et al. 2012), and their size range suggests that this feeding style was possible for both apex and lower-level megapredators. The stark contrast between the stout, brevirostrine snouts of mammalodontid odontocetes (such as Janjucetus) compared with the elongate, latirostrine crania of both aetiocetid mysticetes (e.g., Aetiocetus) and many Oligocene odontocetes (e.g., Xenorophus) indicates a clear divergence in ecomorphological trajectory in Oligocene cetaceans. Interestingly, occupation of ecomorphospace by all three of these groups is not shared with any group of mosasaurids.
One of the more surprising results of our analysis was the apparent decoupling of swimming ability and cranial ecomorphology, with sub-carangiform and carangiform species plotting broadly with their anguilliform ancestors. Despite the limitations of our postcranial dataset, this result is consistent with other studies indicating distinct evolutionary pressures on craniodental and postcranial regions-for example, in short-necked plesiosaurs (Fischer et al. 2020) and ichthyosaurs (Gutarra et al. 2019).
Convergence, Heritage, and Context.-Despite being statistically significant, the convergence recovered between three mosasaurid-cetacean pairs (Basilosaurus isis vs. Tylosaurus bernardi, Cynthiacetus peruvianus vs. Mosasaurus sp. (IRSNB R 12), and Waipatia maerewhenua vs. Gavialimimus almaghribensis) is not reflected in a complete overlap of ecomorphospace occupation between the two clades. They are thus examples of incomplete convergence, wherein taxa are ecomorphologically similar-for example, the presence of a robust skull and elongate snout in B. isis/T. bernardi and C. peruvianus/Mosasaurus sp.-yet exhibit unique morphological traits (Grossnickle et al. 2020;Watanabe et al. 2021). The latter may be unique adaptations, such as the predental rostrum of T. bernardi (Jiménez-Huidobro and Caldwell 2016) and the distinctive prismatic cutting edges on the teeth of Mosasaurus (Lingham-Soliar 1995), or ancestral constraints, such as pterygoid teeth and cranial kinesis in mosasaurids (LeBlanc et al. 2013) and heterodont teeth and the mammalian jaw joint in early cetaceans (Uhen 2018).
One previous study interpreted Basilosaurus as an "Elvis taxon" that filled a niche vacated by megapredatory mosasaurids at the K/Pg boundary (Gallagher 2014). The term "Elvis taxon" was erected to describe a phenomenon seen during postextinction recovery among invertebrate communities whereby a morphology reappears long after it was thought to have become extinct (Erwin and Droser 1993). The original extinct species is replaced by a new, unrelated form that is morphologically indistinguishable from its predecessor (Erwin and Droser 1993). This situation clearly does not apply to B. isis and T. bernardi, which are not morphologically identical. In addition to these ecomorphological differences, substantial changes took place in oceanic ecosystems at the K/Pg boundary and in the 15 Myr of subsequent recovery before the first semiaquatic archaeocetes evolved (Thewissen et al. 2009). Late Cretaceous oceans were particularly hot and deep and sometimes poorly oxygenated (Skelton et al. 2003), whereas the late Eocene oceans were cooler and punctuated at the Eocene-Oligocene boundary by the onset of Antarctic glaciation and the precursor of the Antarctic Circumpolar Current (Marx and Fordyce 2015). Rather than filling a specifically mosasaurid-shaped hole, B. isis emerged in the context of an ecosystem that had recovered from the bottom up and lacked any large secondarily aquatic tetrapod predators. Instead of one impersonating the other, B. isis and T. bernardi might be better understood as large open-ocean megapredators with similar craniodental proportions, reflecting the constraints of their shared ecological niche.
Convergence between the longirostrine mosasaurid G. almaghribensis and the odontocete W. maerewhenua (Table 3) also highlights the repeated evolution of longirostrine (putatively piscivorous) ecomorphologies in marine amniotes. Longirostry is thought to increase hydrodynamic efficiency during sweep feeding on small, fast prey (McCurry et al. 2017a;Strong et al. 2020), and longirostrine species have jaws that are biomechanically adapted to open swiftly and capture prey (Anderson et al. 2011). Our results suggest that early odontocetes were able to explore longirostry to a greater extent than mosasaurids. Whereas G. almaghribensis is one of the most longirostrine mosasaurids in this study, W. maerewhenua has a shorter rostrum than several other early odontocetes and may have been a ram feeder (Tanaka and Fordyce 2017). Longirostry in early odontocetes perhaps evolved in tandem with both cranial telescoping and echolocation (Geisler et al. 2014;Boessenecker et al. 2017b). While we cannot know how mosasaurids would have evolved had they survived the K/Pg mass extinction, it is clear that later odontocetes evolved extremely elongate rostra (Lambert and Goolaerts 2021).
Our results add to increasing evidence that incomplete convergence is more common in the natural world than previously realized (Meloro et al. 2015;Grossnickle et al. 2020;Watanabe et al. 2021;Alfieri et al. 2022) and suggest that the textbook convergence in marine tetrapods may be superficial and likely restricted to general body shape (Motani 2002). When one focuses in on the details of craniodental architecture, strong ecomorphological convergence appears rare, especially when analyzing distant clades that colonized marine niches in widely distinct biosphere contexts, such as mosasaurids and early cetaceans. Strong ecomorphological convergence has been theorized to occur when considering a specific, restricted niche with a single optimal morphology (Alfieri et al. 2022); this has been observed in short-necked plesiosaurs, which have an adaptive landscape defined by "peaks" of optimal morphology (Fischer et al. 2020). However, incomplete convergence may result in an adaptive landscape better described as a "slope," where groups show similar or parallel trajectories in ecomorphological evolution that are offset by their ancestral heritage (Grossnickle et al. 2020;Alfieri et al. 2022). Our results fit this pattern, with no clear optimal peaks of ecomorphology and trajectories that appear to be strongly influenced by intrinsic phylogenetic constraints. We posit that extrinsic environmental influences, such as differences in ocean temperature and oxygenation between the Late Cretaceous and Eocene-Oligocene (Skelton et al. 2003;Marx and Fordyce 2015), as well as available ecological niche space in the pelagic ecosystem, were of importance. These distinct contexts likely combined with phyletic heritage and historical contingency to limit marine tetrapod convergence. While we did not consider convergence in postcranial anatomy during this study, it is likely that this would show a similar trend-broad similarities in axial or appendicular morphology that are limited by constraint inherited from mammalian or reptilian ancestors (e.g., dorsoventral vs. lateral axial movement).
Incomplete convergence can also occur as a result of multiple morphologies performing the same purpose, a "many-to-one" relationship between form and function (Zelditch et al. 2017). This has been seen in a number of terrestrial groups, including pack-hunting carnivorans (Meloro et al. 2015), gliding mammals (Grossnickle et al. 2020), and slow arboreal xenarthrans (Alfieri et al. 2022). Modern (and presumably also extinct) marine raptorial predators are usually opportunistic and often overlap in prey choice and capture methods (Hocking et al. 2017). Our study highlights the complicated relationship between form and function in these animals, which may best be described as "many-to-many" (Zelditch et al. 2017).

Conclusions
Despite their superficial similarities, mosasaurids and early cetaceans show different evolutionary trajectories in skull ecomorphology. The earliest mosasaurids were small, low-level predators that rapidly radiated into a range of different ecomorphologies and inferred niches. By contrast, early fully aquatic cetaceans were likely all megapredatory and show a general evolutionary trend toward adaptations for smaller prey. The evolutionary pathways in these groups are strongly influenced by intrinsic phylogenetic constraints as well as extrinsic environmental influences.
We found several examples of convergence in skull ecomorphology between megapredatory forms, but also between longirostrine forms. Despite numerous shared features, convergence is generally incomplete and neither overrides ancestral constraints nor precludes the presence of unique adaptations in certain species. Our results suggest that qualitative assessment of marine tetrapod convergence is too superficial and can overlook the more nuanced differences in ecomorphology between different secondarily aquatic tetrapod radiations. awarded to V.F. (Project SEASCAPE;grant MIS F.4511.19). E.J.C. was funded by the London Natural Environment Research Council Doctoral Training Partnership (London NERC DTP) training grant NE/ L002485/1. We would like to thank A. LeBlanc and one anonymous reviewer for their comments, which significantly improved the article. The authors declare no conflicts of interest.