Morphological evolution in therocephalians breaks the hypercarnivore ratchet

Abstract

Large carnivorous mammals have been suggested to show a ratchet-like mode of morphological evolution. A limited number of specializations for hypercarnivory evolve repeatedly in multiple clades, with those lineages evolving such specialities being unable to retreat back along their evolutionary trajectory or jump between adaptive peaks. While it has been hypothesized that such mechanisms should have applied to the evolution of other terrestrial carnivores, the non-mammalian synapsid clade Therocephalia appears to defy this expectation. The earliest, basalmost members of this clade are large macropredators, and it is later that small carnivores appear, seemingly evolving from top-predator ancestors. In order to test this reading of therocephalian evolution, variation in rates of body size evolution were tested for and incorporated into an ancestral reconstruction. Similar studies were made of the evolution of discrete characters related to carnivory. All analyses indicate the ancestral therocephalian was a large macro-predator, with serrated teeth, elongated canines and robust lower jaws. Small sizes apparently evolve later. It is therefore suggested that the hypercarnivore ratchet is a feature of mammalian evolution.

1. Introduction

Ratchet-like mechanisms of evolution (where the scope of change is limited to a single direction [1,2]) have been a point of great interest in the last few decades. Discussion began in the realm of molecular evolution with Muller's ratchet: because mutation reversals are rare, asexual clonal organisms will accumulate deleterious mutations, leading to reduced resistance of populations to environmental upheavals and increased likelihood of their extirpation [3–5]. Ratchet-like mechanisms have also been posited for morphological evolution, for example, Dollo's Law [6], which posits that a complex trait, once lost, cannot be regained. It has also been suggested that certain characteristics may reduce the ability to attain other states [7–9], with the ultimate result that extreme specialization will not only limit a clade's ‘evolvability' (potential morphological diversity) but also their ability to adapt to environmental disturbances [10,11].

Evolutionary ratchets have been invoked to explain patterns of evolution in mammalian carnivores. In Carnivoromorpha, certain specializations evolve repeatedly in different lineages: cat-like morphology (short snout and long canines, retaining the ability to pronate wrists), dog-like morphology (limbs adapted for cursorial hunting, long robust jaws and teeth to deliver snapping bites) and hyena-like morphology (skull and teeth adapted for powerful bite to crush bone) [12–16]. Increased specialization along these lines is accompanied by a trend towards large body size (Cope's Rule) [14]. The clades evolving these specializations have reduced disparity relative to their near relatives [15] and are unable to retreat back along their evolutionary trajectory or jump from one peak in the adaptive landscape to another [16]. These specializations also appear to increase extinction risk, with hypercarnivorous lineages having shorter durations [12–16]. The large size evolved by hypercarnivores has been suggested as reducing their resilience to extinction; large predators have lower population densities and are therefore more vulnerable [14]. In short, the top-predator niche is an evolutionary dead-end, with the short-term selective pressures driving lineages into a rut from which they cannot escape when environmental upheavals occur.

These evolutionary trends in hypercarnivores have been less extensively studied outside mammals. They have been suggested to apply to non-mammalian synapsids and dinosaurs [13,17], but without any quantitative analysis comparing the morphological trends observed. In fact, there is one non-mammalian synapsid clade where the ‘conventional wisdom' regarding its evolutionary history seems to directly oppose the hypercarnivore ratchet.

Therocephalia is a therapsid lineage that first appears in the fossil record in the middle Permian sediments of the Karoo Supergroup of South Africa [18]. The lineage includes a diverse range of morphologies, including large predators, small generalist carnivores and insectivores, and small herbivores [19–21]. In recent publications [21,22], it was commented that, unlike other therapsid lineages, therocephalians appear to show multiple distinct waves of niche occupation. The earliest therocephalians known are scylacosaurids [18], an assemblage of questionable monophyly [21–24] containing large-bodied predatory species, which represented the top predators across the middle-late Permian transition [21,25,26]. Towards the latter stages of the Permian, however, these were replaced by the predominantly small-bodied insectivorous Eutherocephalia, and later in the Triassic by herbivorous Bauriidae.

Discussion of therocephalian evolution has thus far assumed they originated as large-bodied predators. Characters present in the predatory scylacosaurids such as large canines and serrated teeth have generally been discussed as representing the primitive condition for therocephalians [22,23,27]. Large body size has also been considered primitive for the group. An analysis of body size of therocephalians, although focussing on patterns across the end-Permian extinction, did indicate the size estimated at the root of Therocephalia, which was within the range observed in the scylacosaurids [28]. The small-bodied eutherocephalians were found to represent a later miniaturization. However, none of this literature has commented on how unusual this trend in body size evolution is, and how this Palaeozoic synapsid lineage is showing an evolutionary trajectory not seen in their Cenozoic relatives; that they do not appear to be constrained by the same macroevolutionary ratchet that has restricted mammalian carnivores.

Here, I carry out a thorough assessment of morphological evolution in therocephalians, with an emphasis on their earliest evolution. Evolution of size is re-examined, along with evolution of characters associated with a predatory lifestyle, this time incorporating heterogeneity of rate and mode into the analyses.

2. Material and methods

(a). Phylogeny

The phylogenetic framework for this study is a recent and comprehensive phylogenetic analysis of therocephalians [21]. The character-taxon matrix from this study was expanded by the addition of a non-therocephalian outgroup: the recently described cynodont Abdalodon [29]. As it has been hypothesized that cynodonts originated at small sizes [28,29], and this clade is the immediate outgroup to Therocephalia [23], sampling of the earliest members of this clade is necessary for accurate reconstruction of ancestral therocephalian morphology. The final matrix therefore contains 59 taxa: 51 therocephalians, and eight non-therocephalian outgroups (four Permian cynodonts, composite codings of Gorgonopsia and Anomodontia, Titanophoneus potens and Biarmosuchus tener).

A time-calibrated phylogeny was produced by subjecting this matrix to a Bayesian tip-dating analysis using the fossilized birth–death (FBD) model [30], whereby relationships and branch length are simultaneously estimated using both the character data and the stratigraphic ages of taxa. This method was implemented in MrBayes 3.2.6 [31]. To account for uncertainty in time of the first appearances, the ages of taxa were represented by a uniform probability distribution covering the full possible age of the formation in which they first appear. An offset exponential root prior was applied, with a minimum age 0.2 Myr before the first appearance of therocephalians and a mean age at the Cisuralian/Guadalupian boundary. This choice of prior is based on studies suggesting a major faunal turnover at this time, along with the principal diversification of therapsids [32–36]. However, the impact of prior choice was examined (see the electronic supplementary material). The analysis was run four times for 10 000 000 generations, sampling every 1000. Twenty-five per cent of trees sampled were discarded as burn-in. The maximum clade credibility tree (figure 1a) was used as the basis of subsequent analyses. In all these, the outgroup taxa Anomodontia and Gorgonopsia were dropped (as they are composites rather than representing the primitive condition of these clades), as was Titanophoneus potens, which is a highly derived rather than plesiomorphic dinocephalian [37].

Figure 1.

(a) Rate scalar applied to each branch of the therocephalian phylogeny. Colour of legend indicates rate, length acts as scale bar for branch length of the tree. (b) Phylogram showing body size evolution of therocephalians.

(b). Size evolution

As in Huttenlocker [28], skull length was used as a proxy for size. Values were taken from this study, and from the literature for those taxa added to the dataset. These were subjected to analysis using the separate slopes model [38]. This starts with the Brownian motion model of trait evolution as its base, where the amount of trait variation along a branch is directly proportional to its duration, and there is no selection in any direction or rate variation. Such assumptions underlay the ancestral state reconstructions of Huttenlocker [28] but can bias reconstructions and mask evolutionary trends or positive phenotypic selection [39,40]. The separate-slopes model incorporates rate variation by scaling branches to allow more or less trait variation for a given amount of time. A Markov chain Monte Carlo method is used to identify the pattern of rate variation which best fits the observed trait data and phylogeny.

This analysis was run for 10 000 000 generations, sampling every 1000, with 25% discarded as burn-in. The mean rate scalars (the amount the inferred change exceeds what would be expected from Brownian motion) were applied to the branch lengths, and a likelihood ancestral state reconstruction of skull length was carried out using the rescaled phylogeny, using the ace function in the package ape [41].

(c). Discrete morphological traits

Three characters from the character-taxon matrix, thought to be associated with the macro-predatory niche, were subjected to further analysis. These were: the presence or absence of serrated incisors (character 99), size of canine relative to maxilla (character 94) and robustness of the lower jaw (character 73; relevant owing to the need to bite down on and resist struggling prey [13,17]).

Ancestral character state reconstructions of these characters were carried out, again attempting to account for rate variation. Models of discrete character evolution were fitted to the phylogeny and compared using the sample-size corrected Akaike information criterion. For the presence or absence of serrations (a binary character), two models were fitted: an equal rates (ER) model, where the transition rate was the same in both directions, and an all-rates-different (ARD) model, where the transition rates differed from each other. The other two characters were three-state characters, and so a third model was also tested: a meristic, or ordered, model where transition between states 0 and 2 was not permitted. These model-fitting analyses were carried out using the fitDiscrete function in the R package geiger. This function also calculates the best-fitting q-matrix of transition rates for each model. The q-matrices of the best fitting model of each character was used in the ancestral state reconstructions.

3. Results

Despite the addition of new data (including a comparatively small outgroup cynodont) and allowing rate variation, the ancestral size of therocephalians is still found to be large, within the ranges observed by the macropredatory scylacosaurids. In fact, the new data and methods showed the ancestral therocephalian skull length to be slightly larger than previously found [28]: 193 mm, roughly that of wolf-sized therocephalian Gorynychus [21].

The large size of scylacosaurids does not result from unusually high rates of evolution (figure 1a). While Pristerognathus and Glanosuchus do exhibit exceptionally rapid evolution towards larger sizes than their close relatives (a mean rate scalar of more than three), the lineages leading to the clade Scylacosauridae, as well as Lycosuchus and Gorynychus (two other basally diverging large predatory therocephalians) show evolution little different from Brownian motion. High rates of evolution (mean rate scalar of more than two) are found along the branch leading to the clade containing Bauriodea and Whaitsiodea, containing predominantly small taxa. We must therefore infer later selection towards small sizes from large ancestral therocephalians (see the electronic supplementary material for discussion of potential size-related preservation bias).

The three discrete characters studied all showed different modes of evolution (electronic supplementary material, table S1). The presence or absence of serrated incisors was found to best fit an ARD model where loss of serrations occurred more frequently than their gain. The size of the canine relative to the maxilla best fitted the meristic (ordered) model, although support was not vastly greater than for the ER model. The evolution of jaw robusticity fitted the ER model best.

Ancestral state reconstructions using the q-matrices inferred for the relevant models found little uncertainty regarding the ancestral morphology of therocephalians (figure 2). There was nearly 100% support for the primitive therocephalian having serrated teeth, large canines and a robust lower jaw, results entirely consistent with a hypercarnivorous niche.

Figure 2.

Ancestral state reconstructions. (a) The presence/absence of tooth serrations; (b) canine size; and (c) robustness of the lower jaw. Colours at tips indicate character state observed. Pie charts at nodes indicate probabilities of ancestral states.

4. Discussion

If these results are reliable, they support the traditional paradigm that therocephalians originated as large predators, and only later evolved small body sizes. The patterns observed in mammals do not appear to apply to therocephalians. Mammalian carnivores, once they have reached large size and a specialized bauplan, are apparently unable to leave this adaptive peak. Therocephalians, on the other hand, retreated from the hypercarnivore niche and evolved small sizes later in the Permian. The high rates of evolution along the lineage leading to the small baurioids and whaitsioids indicates that this miniaturization was fairly rapid, effectively a ‘jump' from one adaptive peak to another.

If we infer that therocephalians are not restricted by the same hypercarnivore ratchet as mammals, the next question is whether it is the pattern in therocephalians that is unusual, or that of mammals. While research into this issue is limited outside mammals, there is a wealth of research into body size in other terrestrial tetrapods, the results of which may shed light on this question.

The clade which occupied the top-predator role prior to the therapsids, the pelycosaur family Sphenacodontidae, do appear to follow a pattern similar to mammalian carnivores: early selection towards large size and further gradual increase [42,43]. The later Permian predatory gorgonospians have not been subjected to quantitative study, but the observed trend again appears to be small basal members and a later transition to large size [44,45]. However, the Mesozoic predatory dinosaurs provide a different perspective. Numerous lineages within Theropoda evolved into large macro-predators. In a series of studies identifying shifts in body size evolution over a dinosaur supertree, results were found which suggest theropod macro-predators were able to retreat from their adaptive peak [46,47]. A shift towards larger body size was found at the base of Tetanurae, hypothesized to have been owing to increased macropredation [46]. Within Tetanurae, however, numerous shifts towards smaller sizes are observed, e.g. Coelurosauria, Compsognathidae and Paraves [46,47].

While analyses over a wider range of clades would be useful, it does seem that the hypercarnivore ratchet is not a universal rule. Why, then it is so prevalent in mammals, which might be thought to be more flexible in their dental evolution? Recent studies of tooth development in mammals have shown that morphology and arrangement of cusps are under control of a small number of signalling pathways, and so a wide range of morphologies are possible from small changes early in crown development [48–51]. However, this developmental flexibility does not universally translate to evolutionary flexibility. In hypercarnivorous mammals, for example, the loss of the talonid (the crushing basin of the molars) is a relatively inflexible loss compared to other mammalian dental modifications, potentially constraining hypercarnivores strongly to their adaptive peak [52].

The syclacosaurid therocephalian macro-predators retain teeth that are highly plesiomorphic; the overall structure both of the pre- and postcanines is the simple conical morphology primitive to therapsids [53]. This might allow them greater flexibility to explore other regions of ecospace, even from a hypercarnivore starting point. The family Bauriidae evolve multiple cusps and occlusion of molariform teeth [23,53], but this evolution of multicusped teeth is possibly paedomorphic; juvenile therocephalians possess tricuspid teeth which are primitively lost in adulthood [54]. Possibly it is the combination of plesiomorphic tooth morphology in adulthood, and multicusped teeth in juveniles, which gave therocephalians greater flexibility in their ecological evolution, and ensured that they didn't dig themselves too deep into an evolutionary rut.

Data accessibility

All data is included as the electronic supplementary material.

Competing Interests

I declare I have no competing interests.

Funding

The study was funded by Deutsche Forschungsgemeinschaft grant number BR 5724/1-1.