The evolution and ecology of gigantism in terror birds (Aves, Phorusrhacidae)

Abstract

Terror birds (Aves, Phorusrhacidae) were large flightless apex predators in South America during the Cenozoic. Here, we estimate a new phylogeny for phorusrhacids using Bayesian inference. We demonstrate phylogenetic evidence for a monophyletic Patagornithinae and find significant support for a distinct crown group associated with the quintessential ‘terror bird’ characteristics. We use this phylogeny to analyse the evolution of body size and cursoriality. Our results reveal that size overlap was rare between co-occurring subfamilies, supporting the hypothesis that these traits were important for niche partitioning. We observe that gigantism evolved in a single clade, containing Phorusrhacinae and Physornithinae. The members of this lineage were consistently larger than all other phorusrhacids. Phorusrhacinae emerged following the extinction of Physornithinae, suggesting the ecological succession of the apex predator niche. The first known phorusrhacine, Phorusrhacos longissimus, was gigantic but significantly smaller and more cursorial than any physornithine. These traits likely evolved in response to the expansion of open environments. Following the Santacrucian SALMA, phorusrhacines increased in size, further converging on the morphology of Physornithinae. These findings suggest that the evolution and displacement of body size drove terror bird niche partitioning and competitive exclusion controlled phorusrhacid diversity.

1. Introduction

Phorusrhacids (Cariamiformes, Phorusrhacidae), commonly known as ‘terror birds’, are a group of extinct Cenozoic birds with specialized predatory adaptations, several of which have no modern analogues. These large terrestrial predators exploited new ecological niches and evolved to dominate their respective environments. Prominent predatory characteristics, including large body size [1], long powerful legs [2] and akinetic skulls reinforced for striking prey [3], facilitated their rise to apex predator status.

Placental carnivores were absent in South America, where most phorusrhacids are found, during the continent's roughly 52-million-year isolation before the Great American Biotic Interchange (GABI) [4]. These pre-GABI environments allowed avian and marsupial species to diversify into large predators [4,5].

Throughout their evolutionary history, phorusrhacids varied in their morphology and ecology and occupied diverse niches [5–7]. Some species are thought to have been pursuit predators, while others were more robust and unlikely to sustain prolonged chases [8,9]. While all species had reinforced skulls, only some were fully akinetic [3,10,11]. Gigantism (anomalous large size) appears to have played a prominent role in facilitating the ecological dominance of phorusrhacids [5–7]. Other phorusrhacid species, however, were smaller, some of which were comparable in size to the modern Cariamiformes, the seriemas [1,7,12].

(a) . Phorusrhacid palaeoecology

Phorusrhacids are believed to have occupied two main ecological categories. The smaller, more delicate species may have filled a niche similar to that of modern seriemas [5,7,11]. They probably fed on smaller prey in terrestrial environments while possibly retaining limited flight capabilities [7,11]. Larger species were completely flightless and hunted large mammals [5–7,11]. This larger size class can then be further divided, with some clades being similar in size to modern ratites, while others exhibited true gigantism, reaching sizes as large or larger than Struthio camelus [1]. It is presumed that gigantic species, such as Phorusrhacos longissimus, consumed different prey compared to the more moderately sized Patagornis marshi and the small, seriema-like Psilopterus bachmanni [6,7].

Here, we aim to quantify character displacement in Phorusrhacidae through geologic time. The displacement of ecologically correlated characters has been extensively documented as a mechanism for niche partitioning in both modern and fossil contexts [13–15]. Niche partitioning refers to the selective differentiation of ecological niches to minimize competitive effects [16–18]. Body size is often associated with different dietary niches among predators [19,20]. Today and throughout prehistory, carnivore guilds vary in body size. These size differences correspond to observable differences in prey selection [19]. Formations with high phorusrhacid diversity show body size variation among species [5–7,12]. This suggests that body size evolved to facilitate intrafamilial niche partitioning. Other traits, such as cursoriality, also differed among phorusrhacid species, perhaps presenting an additional mode of niche partitionment [8,9]. Species with disparate means of locomotion are capable of coexisting through either differences in resource gathering behaviours or landscape use [21]. The hypothesis that body size and cursoriality facilitated niche partitioning in phorusrhacids has not been tested with phylogenetic comparative methods. Such tools are necessary for evaluating evolutionary hypotheses to account for the non-independence of shared ancestry on trait evolution [22].

(b) . Phorusrhacid phylogenetics

Previous analyses regarding the interrelationships of phorusrhacids have been inconclusive because of their sparse fossil record. Most species lack significant amounts of skeletal material and therefore lack diagnostic osteological traits [10,23–25]. Alvarenga & Höfling [1] attempted to clarify their systematics, but they did not place individual species within a phylogenetic context. Subsequent work by Alvarenga et al. [24] produced a phylogeny but did not resolve many major relationships. Agnolín's phylogenetic work was better resolved but lacked several important taxa [23,25]. The analysis of Degrange et al. [10] was more taxonomically complete but still did not resolve the crown group relationships.

(i) . Classification of Brontornis burmeisteri

The placement of Brontornis within Phorusrhacidae remains highly contentious. Although traditionally identified as a robust phorusrhacid [1,23], representing the crown of the derived subfamily Brontornithinae, numerous authors have contended that this taxon's extreme autapomorphies warrant its exclusion from Phorusrhacidae [10,26,27]. For example, Agnolín [26] inferred Brontornis as a basal anseriform, closely associated with dromornithids and Gastornis (Diatryma). This finding aligned with those later presented by Degrange et al. [10]. Thus, Brontornis has been excluded from more recent phylogenetic analyses, and other brontornithines are often assigned to the subfamily Physornithinae [10,11,23,25–28]. Other authors have continued to include Brontornis as a member of Phorusrhacidae [24,29,30]. Worthy et al. [31] found evidence for including Brontornis within Cariamiformes, loosely conferring Brontornis to Phorusrhacidae. Most recently, Degrange [11] reclassified Brontornis as a cariamiform of uncertain affiliation, whereas Agnolín [27] argued that Brontornis should remain within Anseriformes. The variability of results across these studies highlights the uncertainty in Phorusrhacidae systematics, which is due, in part, to the poor fossil record of Brontornis and phorusrhacids in general [1,11,31].

(ii) . Reanalysis of the evolution of Phorusrhacidae

Tests for ecological and evolutionary hypotheses require an understanding of phylogenetic uncertainty. In this study, we infer a new Bayesian phylogeny for Phorusrhacidae and assess the statistical support for major subfamilies. Subsequently, we use a selection of equally probable trees to assess body size and cursoriality as potential mechanisms for niche partitioning among phorusrhacids. Our goal is to determine how these traits evolved and establish a framework for subsequent analyses on the ecology and evolution of terror birds.

2. Methods

(a) . Phylogenetics

(i) . Character matrix

We developed a character matrix by independently evaluating the character matrices reported by previous authors [10,23,25,26]. We added Falco peregrinus, Llallawavis scagliai and Titanis walleri to the character matrix using our observations and previous descriptions of the fossil material [1,10,24,29,32,33]. Due to their poor specimen record, we excluded the taxa Psilopterus affinis and Psilopterus colzecus [1]. We also excluded the putative phorusrhacids Paleopsilopterus itaboraiensis, Lavocatavis africana, Eleutherornis cotei and Patagorhacos terrificus as their assignment to Phorusrhacidae is highly questionable and based on very fragmentary material [6,10,11,34,35]. Non-phorusrhacid Cariamiformes were represented by the extant seriemas (Aves, Cariamidae), Chunga burmeisteri and Cariama cristata, as well as the extinct taxa Strigogyps, Idiornis, Paracrax and Bathornis. Our final reformatted matrix consisted of 22 taxa, including 15 phorusrhacids, six other cariamiforms and one falconiform, described with 134 characters (electronic supplementary material, S1 Nexus and S1 Text).

(ii) . Phylogenetic inference

We conducted our phylogenetic analysis using MrBayes V3.2.7 [36] (for phylogenetic parameters see electronic supplementary material, S1 Nexus and S1 Text). We generated a Bayesian posterior distribution of 5000 trees using a 5 000 000 iteration Markov chain Monte Carlo (MCMC) simulation. The MCMC was sampled every 1000 iterations and the first 1 250 000 (25%) iterations were discarded as burn-in. We calculated an extended majority-rule consensus time tree (chronogram) and subsampled the final 1000 chronograms (iterations 4 001 000–5 000 000) to account for phylogenetic uncertainty in our comparative analysis.

(iii) . Phylogenetic position of Brontornis burmeisteri

To control for the uncertainty in the phylogenetic position of Brontornis, we repeated our phylogenetic analysis under the same parameters but including the putative phorusrhacid B. burmeisteri. All constraints were otherwise equivalent, and a new consensus tree was inferred for comparison (electronic supplementary material, S1 Text).

(b) . Ecomorphology

We evaluated body size and cursoriality, traits known to be ecologically correlated [9]. Body size is expected to reflect the relative prey size of each species [5–7], while cursoriality is a measure of locomotor style [8,9]. We obtained measurements of phorusrhacid fossils from the literature or from photographs (electronic supplementary material, S1 Table and S2 Table) [1,6,10,30,37–40]. Due to the limited number of specimens, we used averaged values in our analyses (electronic supplementary material, S1 Text). When direct measurements were available, we calculated the averages solely based on those values, disregarding measurements from photographs.

Phylogenetic comparative analysis was completed using BayesTraits [41]. Posterior distributions were compared using the highest posterior density interval (HPDI), as calculated using the R package HDInterval [42,43]. To account for phylogenetic uncertainty, we conducted our comparative analyses using a subsample of 1000 trees from the posterior distribution of our Bayesian phylogenetic inference. The likelihood convergence of all phylogenetic comparative analyses was assessed using the program Tracer 1.7 [44]. We compared the likelihood support of competing models using the Bayes factor test:

$$\mathrm{BF}=2\ast \hspace{0.17em}|Lh(M_1)\hspace{0.17em}-Lh(M_2)|,$$

where Lh represents the log marginal likelihoods estimated from models 1 (M₁) and 2 (M₂), respectively. We estimated Pagel's λ to assess the phylogenetic structure in the data for each test [45]. We used Bayes factors to evaluate the presence of phylogenetic signal, comparing the marginal likelihood of models estimating λ to equivalent tests assuming no phylogenetic signal (λ = 0.0001). If models estimating λ were found to be significant (BF > 2), we included the parameter in our final trait analysis. If no significant difference was found or if Bayes factors supported the model that constrained λ to zero, we used the latter model in our final trait analysis. We ran MCMC simulations for 110 000 000 iterations, sampling every 1000, with a burn-in of 10 000 000 iterations. To estimate the log marginal likelihood, we employed a stepping-stones algorithm, using 1000 stones, each calculated with 10 000 iterations [46].

(i) . Body size

It has previously been suggested that mass played a crucial role in the division of niches among phorusrhacid species [5–7,12]. When multiple species coexist, each species appears to have a distinct body mass, which likely corresponds to size of prey [5–7]. We used proxies to measure mass for the species included in our analysis. Femur length (FL) serves as a reliable indicator of avian body mass [31,47]. However, due to the incomplete fossil record of Phorusrhacidae, femur length is unknown for several species. To address this, we considered three alternative proxies for avian body mass: femur distal condyle width (FDW), tarsometatarsus proximal width (TMTPW) and tarsometatarsus trochlea III width (TMTIIIW). We independently assessed each metric using phylogenetic generalized least squares (PGLS) regression in BayesTraits V4.0 to determine if they correlated with femur length and, consequently, body size [41] (electronic supplementary material, S1 Text). The statistical significance of regression coefficients was determined by calculating the proportion of parameter estimates that crossed a value of 0 (p_MCMC). A regression coefficient is considered significant if less than 5% of posterior parameter estimates cross 0 (p_MCMC < 0.05). All three proxies were found to be reliable linear correlates to femur length in phorusrhacids ($R_{\mathrm{Avg}}^2>0.975$, $R_{\sigma }^2<0.0003$, p_MCMC < 0.0002) (electronic supplementary material, S1 Text). To determine a single metric for comparison, we performed regressions of TMTPW and TMTIIIW, FDW and TMTIIIW, and FDW and TMTPW ($R_{\mathrm{Avg}}^2>0.955$, $R_{\sigma }^2<0.0004$, p_MCMC < 0.0003) (electronic supplementary material, S1 Text). From these models, we estimated a set of phylogenetically corrected retrodictions (past predictions) of TMTIIIW for all species where measurements were unavailable (TMTIIIWR) (electronic supplementary material, S1 Text) [48].

Hutchinson [16] proposed a minimum offset of sizes between sympatric competitors which would allow for coexistence. According to this ratio, if one taxon is more than 1.3 times larger than the other, body size likely evolved to reduce competition and facilitate coexistence [16,19,49]. This principle applies well in some contexts, where body size represents the only means of resource partitioning, but it is less applicable when habitat, behaviour, and other morphologic differences allow for the reduction of interspecific competition [18,19,49,50]. Moreover, this concept has been criticized extensively regarding its statistical power [49,51,52]. However, as a tool for identifying size displacement, this measure can provide an initial indication of the presence of character displacement prior to conducting more statistically robust and phylogenetically controlled analyses. Using our average TMTIIIW values, we calculated whether the size ratio between each pair of cohabitating taxa was greater than or less than 1.3:

$${\displaystyle \frac{\mathrm{TMTIIIW}\hspace{0.17em}\mathrm{Larger}\hspace{0.17em}\mathrm{Species}}{\mathrm{TMTIIIW}\hspace{0.17em}\mathrm{Smaller}\hspace{0.17em}\hspace{0.17em}\mathrm{Species}}.}$$

We evaluated differences in size among phorusrhacid lineages by performing a phylogenetic analysis of covariance (PANCOVA). TMTIIIW was our response variable, and time and clade identity were our explanatory variables along with the interaction between TMTIIIW and clade identity. We used binary (dummy-coded) indicator variables as covariates in our models to designate clade identity (electronic supplementary material, S1 Text). This tests for the significance of the size differences, over time, between the clades that exhibit gigantism and those that do not. Subfamilies were not acceptable covariates as each contains too few species to produce a statistically sound result. Instead, we used three clades: Mesembriornithinae + Psilopterinae, Patagornithinae and Physornithinae + Phorusrhacinae. These represent the major diverging lineages of Phorusrhacidae, comprising two unrelated clades of seriema to ratite-sized phorusrhacids and the gigantic phorusrhacids, respectively [1]. Patagornithinae, although monophyletic with Phorusrhacinae and Physornithinae, was treated as a separate group, as this clade does not exhibit gigantism. Composed of only three measurements, two of which are retrodictions, the trend for this clade lacks statistical power, but this separation helps constrain the trend calculated for gigantism to only those species of gigantic size. We conducted the PANCOVA in BayesTraits V4.0 using a PGLS regression model, sampling a posterior distribution of slopes and intercepts [41]. Bayes factor tests found no support for λ differing from 0 (BF < 2), and thus λ was constrained to 0 during our simulations (λ = 0.0001). Given the extinction of Phorusrhacidae before the Holocene, it is not possible to observe a 0 Ma intercept. Therefore, instead of using a 0 Ma intercept, we calculated slopes and intercepts with age centred around the last appearance datum (LAD) of the youngest phorusrhacid included in our tree, T. walleri (LAD = 1.8 Ma). This analysis will help identify any size disparity near the extinction of gigantic Phorusrhacidae. Similarly, we centred our dataset around the age of the oldest taxon within our tree, in terms of LAD, Paraphysornis brasiliensis (LAD = 23.03 Ma). By doing so, we calculated size trends at the earliest stage of phorusrhacid evolution that can be observed in our tree. To determine significance, we calculated a p_MCMC for the slope and intercept of each regression line (electronic supplementary material, S1 Text). We evaluated differences in size by observing the overlap of the 95% HPDIs of the intercepts (electronic supplementary material, S1 Text).

We reconstructed body size, based on TMTIIIW, for the most recent common ancestor (MRCA) of each subfamily, to compare body size evolutionary dynamics. We divided Phorusrhacinae into two clades, following our phylogenetic results: we labelled Kelenken guillermoi and Devincenzia pozzi Robust Phorusrhacinae, and we referred to P. longissimus and T. walleri as Gracile Phorusrhacinae. A Bayes factor test revealed statistical support for phylogenetic signal (BF > 2), thus λ was estimated in our ancestral state reconstructions. Using a continuous random walk model, we generated a posterior distribution of 100 000 possible ancestral states for each clade. We then calculated the 95% HPDIs of the MRCAs for Physornithinae, Robust Phorusrhacinae, and Gracile Phorusrhacinae to compare size across the clades relevant to our evaluation of gigantism (electronic supplementary material, S1 Text). We likewise calculated the 95% HPDIs of the MRCAs for Mesembriornithinae, Patagornithinae, and Psilopterinae to compare size overlap among small- and medium-sized species (electronic supplementary material, S1 Text). We compared the overlap of HPDIs between each clade (electronic supplementary material, S1 Text) and against the measured values for each species (electronic supplementary material, S1 Table).

(ii) . Cursoriality

We calculated cursoriality using two equations: the proportion of the tarsometatarsus to the total length of the hindlimb [53]; and the robusticity of the tarsometatarsus [8]. According to the Storer [53] method,

$${\displaystyle \frac{\mathrm{TMTL}}{\mathrm{FL}+\mathrm{TBL}+\mathrm{TMTL}},}$$

the ratio of the tarsometatarsus length to the sum length of the femur (FL), tibiotarsus (TBL), and tarsometatarsus (TMTL) is greater than 0.30 in cursorial species and less than 0.30 in non-cursorial, or graviportal, species. Using the method proposed by Angst and colleagues [8],

$${\displaystyle \frac{\mathrm{TMTL}}{\mathrm{TMTMW}},}$$

we calculated the ratio between the length of the tarsometatarsus (TMTL) and the minimum transverse width of the tarsometatarsus (TMTMW). Species are considered cursorial if they have a ratio greater than 12 and graviportal if the ratio is less than 12. It is important to note that in general analysing ratios should be avoided [54]. However, in this case, we cautiously proceed to ensure comparability with previous studies [8,9,53]. Averaged values were used for each variable (electronic supplementary material, S1 Text). Both equations were used to classify species as either cursorial or graviportal.

(c) . Phorusrhacid palaeoecology

To characterize the ecology of phorusrhacids at different points in time, we evaluated the conditions present in each South American Land Mammal Age (SALMA) we observe phorusrhacids [55]. SALMAs are designated in terms of change to the gross faunal assemblage of South America, as opposed to strict geological periods, which allows us to evaluate how phorusrhacid lineages responded to large scale ecological turnover [55]. We focused on environments in Argentine Patagonia, as this region has produced the highest density and diversity of phorusrhacid fossils [6]. Ecological inferences are largely based on ecomorphology, palaeoenvironmental reconstructions, and co-occurrence with potential prey and competitor species. We summarized previous interpretations and integrated our observations of evolutionary dynamics to determine long-term trends in phorusrhacid ecology.

3. Results

(a) . Phylogenetics

We present and evaluate a new phylogeny of Phorusrhacidae based on our Bayesian posterior distribution of trees (figure 1). We report here our phylogenetic results and the accompanying posterior probabilities (PP).

Figure 1.

Majority-rule consensus Bayesian phylogeny of Phorusrhacidae with posterior probabilities reported. Silhouettes of Falconiformes (Falco peregrinus peregrinus by Ferran Sayol: CC0 1.0), Cariamidae (Cariama cristata by George Edward Lodge, vectorized by T. Michael Keesey: Public Domain Mark 1.0) and Phorusrhacidae (Psilopterus lemoinei, Mesembriornis milneedwardsi, Patagornis marshi, Phorusrhacos longissimus, and Paraphysornis brasiliensis by Thomas W. LaBarge: CC BY 4.0) sourced from PhyloPic (phylopic.org).

We find that Psilopterinae (Psilopterus and Procariama) and Mesembriornithinae (Mesembriornis and Llallawavis) are retained as subfamilies within Phorusrhacidae, although Psilopterinae is weakly supported (PP = 0.62). As originally reported by Patterson and Kraglievich [56], both subfamilies form a distinct lineage separate from other phorusrhacids, albeit with low support (PP = 0.64).

All remaining phorusrhacids belong to a single lineage with high node support (PP ≈ 1). No previous time-calibrated analysis has resolved the relationships of derived phorusrhacids – previous work produced a polytomy of various species assigned to Phorusrhacinae, Patagornithinae and Brontornithinae/Physornithinae [10,24].

We demonstrate a clear separation between larger derived phorusrhacids and Patagornithinae (PP = 0.89). Patagornithinae (Andrewsornis, Patagornis and Andalgalornis) is resolved as a distinct, monophyletic subfamily (PP = 0.86).

Phorusrhacinae, as traditionally defined (Phorusrhacos, Devincenzia, Kelenken and Titanis) is paraphyletic, a result previously obtained by Agnolín [23,25]. A distinct clade comprising P. longissimus and T. walleri is observed, but K. guillermoi and D. pozzi are instead grouped with Physornithinae (likewise a result previously reported by Agnolín [23,25]). However, this grouping has low support (PP = 0.50), and the node linking P. longissimus and T. walleri also presents high uncertainty (PP = 65). Provided their ecological and biostratigraphic context (figure 2) and numerous synapomorphies with P. longissimus and T. walleri (electronic supplementary material, S1 Nexus and S1 Text), additional study may support the monophyletic affinity of K. guillermoi and D. pozzi within Phorusrhacinae [1,11]. However, we currently lack sufficient phylogenetic evidence to support or refute an affinity with Physornithinae. For this reason, we continue to separate Phorusrhacinae into Robust Phorusrhacinae and Gracile Phorusrhacinae for the purpose of comparison.

Figure 2.

Size disparity, biostratigraphy, and diversity of Phorusrhacidae. Size represented by normalized average TMTIIIW. Photograph measurement used for: L. scagliai. TMTIIIWR values used for: A. abbotti, A. steulleti and P. fortis. Values normalized to the standard deviation of measured TMTIIIW, excluding photograph measurements. Silhouettes (Psilopterus lemoinei, Mesembriornis milneedwardsi, Patagornis marshi, Phorusrhacos longissimus and Paraphysornis brasiliensis by Thomas W. LaBarge: CC BY 4.0) sourced from PhyloPic.

We find strong support (PP ≈ 1) for the subfamily Physornithinae (Physornis and Paraphysornis). When B. burmeisteri is added to the dataset, we observe no significant change to either tree topology or posterior probabilities (electronic supplementary material, S1 Text). Thus, our interpretations should remain robust if, with additional study, Brontornis is conferred to Phorusrhacidae (in which case Physornithinae would be suppressed and Brontornithinae would be included within Phorusrhacidae). The inclusion of B. burmeisteri results in a slight decrease in the affinity of Robust Phorusrhacinae with Physornithinae (PP = 0.43) (electronic supplementary material, S1 Text).

(b) . Ecomorphology

(i) . Body size

We plotted the occurrence and size of each species along a geologic time scale (figure 2) to observe changes in body size, size overlap, diversity and subfamily co-occurrence over time. We standardized the measured (TMTIIIW) and estimated (TMTIIIWR) body size values to the standard deviation of the measured (TMTIIIW excluding photograph measurements) values to compare the sizes of phorusrhacid species (electronic supplementary material, S3 Table). In a given SALMA, we observe no exact size overlap among individual species present (figure 2).

We determined that among co-occurring taxa three dyads possessed size ratios less than Hutchinson's ratio (1.3). These exceptions were between two psilopterines during the Santacrucian (P. bachmanni and Psilopterus lemoinei: 1.04), two physornithines during the Deseadan (P. brasiliensis and P. fortis: 1.09), and between the first known mesembriornithine and the last known patagornithine, during the Huayquerian (M. incertus and A. steulleti: 1.14). All other species combinations were offset in average TMTIIIW values by 1.3 times or more.

The PANCOVA revealed an early divergence of two different size trends (figure 3). At 1.8 Ma, we estimate an average intercept of 3.841 cm (TMTIIIW) for Physornithinae + Phorusrhacinae, as opposed to 2.269 cm for Mesembriornithinae + Psilopterinae. These intercepts are significantly different (p_MCMC = 0.023), showing less than 4% overlap of posterior distributions at the 95% HPDI. Likewise, significantly different intercepts were estimated at 23.03 Ma (p_MCMC = 0.018), with no overlap observed at the 95% HPDI. At 23.03 Ma, an average intercept of 4.363 cm is estimated for Physornithinae + Phorusrhacinae and 0.441 cm for Mesembriornithinae + Psilopterinae. The regression of Physornithinae + Phorusrhacinae TMTIIIW has an average slope of −0.025 (cm/Ma) across the entire phylogenetic history of this clade, with no significant trend (p_MCMC > 0.185) (figure 3). Thus, the average size for this clade largely did not shift, either positively or negatively, over time. Mesembriornithinae + Psilopterinae show an average directional slope of +0.086 (cm/Ma) at both 1.8 Ma and 23.03 Ma (p_MCMC < 0.028). The intercept for Mesembriornithinae + Psilopterinae increased over time, shifting from an average intercept of 0.441 cm to 2.269 cm. Small- to medium-sized phorusrhacids therefore appear to have increased in size over time while gigantic species remained fairly consistent. The gigantic and non-gigantic lineages exhibited sizes which were continuously offset, even as the latter group increased in size. While the magnitude of this offset varied, sizes were always significantly different.

Figure 3.

PANCOVA of phorusrhacid body size. Regression analysis completed using BayesTraits V4.0 [41]. Average slopes (β) and intercepts (α) reported at 1.8 Ma (LAD of T. walleri) and 23.03 Ma (LAD of P. brasiliensis) for Physornithinae + Phorusrhacinae (a) and Mesembriornithinae + Psilopterinae (b). No trend is reported for Patagornithinae due to inadequate sample size. Species used in the analysis are labelled and superimposed onto the PANCOVA graph. Illustrations represent gigantic (a) and non-gigantic lineages (b), depicting P. longissimus (a) and P. bachmanni (b), respectively. Silhouettes (Psilopterus lemoinei, Mesembriornis milneedwardsi, Patagornis marshi, Phorusrhacos longissimus and Paraphysornis brasiliensis by Thomas W. LaBarge: CC BY 4.0) sourced from PhyloPic. Illustrations work of corresponding author.

The posterior distributions of body size ancestral state reconstructions for Robust Phorusrhacinae and Physornithinae overlap significantly (greater than 90%) with each other. Gracile Phorusrhacine overlaps partially (38%) with Physornithinae, and Physornithinae overlaps moderately (54%) with Gracile Phorusrhacinae. Physornithinae and Robust Phorusrhacinae do not overlap at all with Patagornithinae, Mesembriornithinae and Psilopterinae. Gracile Phorusrhacinae overlaps marginally with Mesembriornithinae and Patagornithinae and vice versa (21% or less). Psilopterinae presents no overlap with Gracile Phorusrhacinae. Psilopterinae overlaps significantly (greater than 75%) with Mesembriornithinae and moderately (58%) with Patagornithinae. Patagornithinae overlaps significantly (greater than 95%) with Mesembriornithinae and significantly (greater than 75%) with Psilopterinae. Mesembriornithinae likewise overlaps significantly (greater than 90%) with Patagornithinae and significantly (greater than 80%) with Psilopterinae.

(ii) . Cursoriality

The majority of Phorusrhacidae were cursorial, according Storer's [53] and Angst and colleagues' [8] equations (8 of 11 species analysed) (electronic supplementary material, S1 Text). Utilizing the Storer method [53], the following species were determined to be cursorial: P. bachmanni, P. lemoinei, Procariama simplex, M. milneedwardsi, L. scagliai, P. marshi and P. longissimus; and one species was graviportal: P. brasiliensis. Using Angst and coauthors’ [8] equation, we find the following species to be cursorial: P. bachmanni, P. lemoinei, P. simplex, M. incertus, M. milneedwardsi, L. scagliai and P. marshi; and the following species to be graviportal: P. longissimus, K. guillermoi, D. pozzi and P. brasiliensis.

4. Discussion

(a) . Phylogenetics

The consensus of our Bayesian posterior distribution of trees is generally well resolved and has high clade credibility values associated with most major nodes (figure 1). We find strong support for many previously hypothesized relationships within Phorusrhacidae [1]. All five subfamilies of Phorusrhacidae are represented in our phylogeny, with varying levels of statistical support (figure 1). We use this tree to justify classifying 17 distinct phorusrhacid species following Alvarenga and colleagues' [24] systematics. Overall, this phylogeny is generally congruent with previous analyses using the parsimony criterion [10,24] but significantly clarifies crown-group relationships.

(i) . Interrelationships of Phorusrhacidae

In our interpretation, Psilopterinae remains a unique clade, despite low posterior probability support, given their overall similarities in size and inferred ecologies [1,7,12,38,40]. Given their similar size and morphology [1], and the phylogenetic results of previous authors [10,24], we assign P. affinis and P. colzecus to Psilopterinae. Mesembriornithinae is moderately supported as a distinct subfamily, containing L. scagliai, M. incertus and M. milneedwardsi, but excluding P. simplex.

With the addition of new cranial characters, we present here phylogenetic evidence for the purported subfamilies of crown Phorusrhacidae. Patagornithinae and Physornithinae are well-supported within our tree, with high posterior probabilities (PP > 0.85). We recover the subfamily Phorusrhacinae, but currently composed of only two taxa (P. longissimus and T. walleri) and with limited support (PP = 0.65). K. guillermoi and D. pozzi fall outside their hypothesized relationships, instead showing an affinity with Physornithinae, but their placement is poorly resolved (PP = 0.50). However, K. guillermoi and D. pozzi are poorly represented in the fossil record. We note their strong similarity in cranial morphology to P. longissimus (and to the limited cranial material available from T. walleri), characters largely unknown in Physornithinae (electronic supplementary material, S1 Text). The massive stature of K. guillermoi and D. pozzi compared to the relatively gracile and cursorial construction of P. longissimus, may be the result of parallel evolution to Physornithinae, thus resulting in apparent synapomorphies. We therefore continue to interpret these taxa as members of Phorusrhacinae. Regardless of their affinity, P. longissimus, K. guillermoi, and D. pozzi appear to succeed Physornithinae, occupying a similar macropredatory niche.

(ii) . Taxonomic affinity of Brontornis burmeisteri

We find that the phorusrhacid tree is generally congruent, including or excluding Brontornis (electronic supplementary material, S1 Text). There is a strong affinity between B. burmeisteri and P. brasiliensis and P. fortis (PP ≈ 1). However, as our phylogenetic analysis is restricted to Australaves, we cannot determine at this time whether a position outside this clade would be supported. A definitive systematic assessment will require a phylogenetic reappraisal of Brontornis including many more modern and fossil neognathan birds, in particular Gastornithidae and Dromornithidae [31].

(b) . Ecomorphology

In comparison with co-occurring taxa and modern avifauna, we interpret the ecologies of individual phorusrhacid species. Using relative body size, we categorize them into the following predator size classes. Generalist predators were species of small body size with inferred diets and behaviours similar to modern Cariamiformes, which are able to use a wide variety of prey resources. Mesopredators are defined as species of intermediate body size, similar in size to modern ratites. Macropredators, species exhibiting gigantism, represented the largest phorusrhacid species in the ecosystem and were likely apex predators. We use these ecological states as size class descriptors to evaluate the niches occupied by various phorusrhacid species. These are relative terms meant to clarify our ecological interpretations based on potential prey size.

In addition to considering body size, we also categorize predatory niches based on cursoriality. Species with cursorial proportions are considered pursuit predators, relying on speed to chase and kill prey. Species with graviportal proportions are labelled as ambush predators, relying on stealth to approach and capture prey. Currently, there is no direct evidence of predation by phorusrhacids, so the species categorized here as ambush predators could have potentially employed other carnivorous methods such as scavenging. This ecological interpretation is solely based on the predatory behaviours observed in extant Cariamiformes [57]. Our final interpretation combines both locomotion and body size to describe the niche occupied by each terror bird species.

(i) . Body size

Using our body size classifications, we observe that psilopterines, having sizes most similar to modern and contemporary Cariamidae, were likely generalist predators of small prey [1,5–7]. Patagornithinae and Mesembriornithinae, composed of medium-sized species, were mesopredators [1,5–7]. Phorusrhacinae and Physornithinae, which demonstrate gigantism, are classified as macropredators [1,5–7]. Our analysis of body size revealed several trends indicating the central role it played in phorusrhacid ecology. Small, intermediate, and large phorusrhacids all coexisted within the same ecosystems (figure 2). No more than two species of the same size class are ever observed in the same ecosystem, and such cases are associated with either additional modes of resource partitioning (such as locomotory differences, dietary generalism or geographical separation) or resulted in extinction. A consistent size class appears to have been selected within each subfamily that is maintained over geologic time (figure 2).

PANCOVA revealed significant differences in both sizes and the direction of size evolution between the gigantic and non-gigantic lineages (figure 3). We also find that body size divergence occurred early in the history of Phorusrhacidae (figures 2 and 3). While sizes may have increased or decreased over time in certain clades, sizes never converged enough to generate a shift in size class. Body size adaptations appear to have been the result of the initial radiation of this group to fill ecological niches rather than a prolonged evolutionary trend (figure 2).

The findings of our ancestral state reconstruction support the hypothesis of size-driven niche partitioning, competitive exclusion, and subfamily to subfamily succession within size classes. The ancestor to Physornithinae was functionally equivalent in size to the ancestor of Robust Phorusrhacinae. These two lineages never overlapped in time or space but when comparing physornithines to the MRCA of Robust Phorusrhacinae, we find that all TMTIIIW and TMTIIIWR values for Physornithinae fall within the 95% HPDI of Robust Phorusrhacinae (electronic supplementary material, S1 Table and S1 Text). This indicates that no significant size difference existed between the last surviving physornithine and the first robust phorusrhacine.

A similar trend is observable in medium-sized species as well. The patagornithine Andalgalornis steulleti co-occurred with the first known member of Mesembriornithinae, M. incertus, but Patagornithinae went extinct soon after. We find no significant difference between the body size of Mesembriornithinae and Patagornithinae, with all TMTIIIW and TMTIIIWR values falling within the overlapping range of 95% HPDIs (1.443 cm–3.074 cm) (electronic supplementary material, S1 Table and S1 Text). The Hutchinson's ratio for these species is less than 1.3, suggesting that they were not partitioned by size. We therefore hypothesize that the extinction of Patagornithinae was due to competition with, and eventual replacement by, Mesembriornithinae.

The MRCA for Psilopterinae overlaps significantly with Mesembriornithinae and Patagornithinae. However, of the taxa used in this analysis, only P. simplex possessed specimens within the 95% HPDI range of Mesembriornithinae and Patagornithinae (electronic supplementary material, S1 Table and S1 Text). The MRCA calculations do not include several psilopterine taxa (P. affinis and P. colzecus) which were smaller than P. simplex (electronic supplementary material, S1 Table). Pleistocene psilopterines possessed smaller sizes, more similar to Psilopterus [38]. Thus, the calculated ancestral state reconstruction is likely biased towards larger taxa. Given the survival of Psilopterinae into the late Pleistocene [38,58], we can likewise conclude that Psilopterinae was not outcompeted by another known phorusrhacid.

(ii) . Cursoriality

Both cursoriality equations produced similar results, except for P. longissimus, which was found to be cursorial according to the Storer equation [53] and graviportal according to Angst and colleagues' equation [8]. However, given that the value calculated using the method developed by Angst et al. [8] is close to the boundary between cursorial and graviportal (11.3) (electronic supplementary material, S1 Text) and that the Storer method [53] favours cursoriality, we conclude that P. longissimus, was predominately cursorial.

Psilopterinae, Mesembriornithinae and Patagornithinae were all found to be composed of cursorial species. As all psilopterines were found to be cursorial, and their proportions and sizes remain well within the range observed for other psilopterines, we hypothesize that P. colzecus and P. affinis were cursorial [1]. It should be considered, however, that in comparison with other phorusrhacids the pelvises of Psilopterinae are proportionally wide suggesting less efficient running abilities [5]. Morphometrically, psilopterine hindlimbs align closer to walking and wading birds than modern cursorial taxa [6,40]. Thus, while cursorial in hindlimb proportions, psilopterines may not have been as efficient runners as patagornithines or mesembriornithines [5,6,40]. Both patagornithines and mesembriornithines were found here to be cursorial and were previously supported as cursorial taxa through cursoriality analyses [8,9], geometric morphometrics [6,40] and biomechanical modelling [2]. Among Psilopterinae, less pronounced cursoriality [5,6,40] and a more generalized, plesiomorphic skeletal morphology (i.e. short height, slender frame, large wings, thin hindlimbs) [1,7,12,40] suggests less specialization for pursuing and dispatching large prey. Large and small psilopterine species probably maintained a more generalist niche, partitioned from the more specialized, pursuit predator niche occupied by Patagornithinae and Mesembriornithinae [2,5,40].

As was previously observed by Angst et al. [8], P. brasiliensis was graviportal. Based on the similarly robust skeletal construction of P. fortis [1], Physornithinae was likely entirely composed of graviportal taxa. P. longissimus was cursorial, while the Robust Phorusrhacines, D. pozzi and K. guillermoi, were both graviportal in their proportions. Robust Phorusrhacines were notably more robust than P. longissimus, which we attribute to convergence with Physornithinae. It should be noted, however, that the tarsometatarsi of these taxa are not as robust as P. brasiliensis (electronic supplementary material, S1 Text). Chiappe & Bertelli [59] observed that neither D. pozzi nor K. guillermoi were as robust as B. burmeisteri and were likely more cursorial in their habits. The tarsometatarsus of P. brasiliensis was approximately as robust as B. burmeisteri (electronic supplementary material, S1 Text), therefore we must likewise conclude that although graviportal in construction, Robust Phorusrhacinae likely possessed superior running abilities to Physornithinae. T. walleri lacks the material to designate a state of cursoriality.

(c) . Phorusrhacid palaeoecology

We interpret all psilopterines as generalist predators, similar to modern and extinct seriemas. Likewise, we interpret all patagornithines and mesembriornithines as pursuit mesopredators. We suggest that P. longissimus was a pursuit macropredator and K. guillermoi and D. pozzi were ambush macropredators. All physornithines were likely ambush macropredators, and T. walleri was a macropredator of unknown locomotory style/hunting habit. All macropredators are estimated to have been over 100 kg [1], however P. fortis, P. brasiliensis, K. guillermoi and D. pozzi were all significantly larger than either P. longissimus or T. walleri (figure 2). If B. burmeisteri was indeed a phorusrhacid, its graviportal proportions and large size would demarcate it as an ambush macropredator according to our scheme. In such a case, given their differences in cursoriality, there probably would not have been complete niche overlap between B. burmeisteri and P. longissimus, which would have allowed for their coexistence.

(i) . Community ecology

During the early Deseadan SALMA (Early Oligocene), Patagonia was wet and humid with a temperate to cold-temperate climate [60]. In the Late Oligocene, temperatures increased and tropical plants started to spread southward, resulting in continued dominance of forests in Patagonia [60]. Towards the end of the Deseadan and through the Colhuehuapian SALMA (Early Miocene), the climate became more arid, particularly in lowland areas, instigating the transition from closed, forested environments to open and xeric habitats [60,61]. The Santacrucian SALMA (Early Miocene) was characterized by a mosaic of open shrubland interspersed with patches of closed woodland, marshes, and seasonally flooded water basins [5,60–62]. Grasses were present but not yet widespread during this time [60].

These dry and open conditions were most suitable for pursuit predators, such as P. longissimus [5,8,21]. Physornithines were likely ambush predators, which are generally better suited to closed environments (i.e. Deseadan forests), especially when lacking cursorial adaptations [8,21]. Open environments continued to expand during the Middle-Late Miocene (Colloncuran to Montehermosan) [60]. All subsequent phorusrhacids that evolved in this region had strong adaptations for living in open habitats. Later phorusrhacines such as K. guillermoi were larger and more robust, but overall maintained superior running abilities compared to physornithines, and could probably have competed as ambush predators in open environments [21,59]. Grasses started to dominate the floral community in the Late Miocene (Mayoan to Montehermosan), shaping the landscape that characterizes the modern Patagonian steppe [60].

Plausible prey for phorusrhacids would have varied depending on size. In pre-interchange environments prey species would have been restricted to those lineages endemic to South America, many of which went extinct prior to or during the GABI [4]. Psilopterines likely possessed a broad and variable (generalist) diet, similar to modern seriemas [5–7]. They probably preyed on invertebrates and small vertebrates, including birds, snakes, other small reptiles, and mammals such as rodents, small marsupials, and the very small members of Notoungulata [5–7,63]. Mesembriornithines and patagornithines probably preyed upon larger taxa, including rheid birds, medium-sized notoungulates, and possibly smaller members of Litopterna in the case of Mesembriornis [4–6,62,63]. The largest phorusrhacids (physornithines and phorusrhacines) likely consumed large (greater than 100 kg) notoungulates and litopterns [4–7,62,63]. The largest mammals in phorusrhacid ecosystems, such as Astrapotheria and the extremely large notoungulate families (e.g. Toxodontidae, Homalodotheriidae) are generally thought to have been too large to have been hunted by phorusrhacids [4–6,63]. However, juvenile individuals of these large mammal species would likely have been vulnerable to predation [62]. Glyptodonts and other large members of Cingulata were probably invulnerable to predation from phorusrhacids due to their extensive dermal armour [5,6]. However, smaller, armadillo-sized Cingulata and other unarmoured members of Xenarthra could have been prey to phorusrhacids [4,62].

Phorusrhacids would have competed for prey with other native South American carnivores. Psilopterinae probably would have experienced some competitive pressure from early Cariamidae, which have persisted in South America from the Santacrucian to the present day [5,12,64]. However, just like modern seriemas, psilopterines probably consumed a wide variety of prey items, reducing the need for resource partitioning [5–7]. The major competitors to medium-sized and gigantic Phorusrhacidae would have been the Sparassodonta [62,65]. Sparassodonts remained the top mammalian carnivores in South America until the Late Pliocene (Chapadmalalan) [62,65]. Most sparassodont carnivores were scansorial, probably inhabiting closed, forested environments in contrast with the open environments inferred to have been preferred by cursorial phorusrhacids [5,21,65]. Scansorial sparassodonts may have competed with phorusrhacids in the forested habitats that typified the Deseadan, but following the expansion of xeric shrublands and grasslands, many sparassodonts probably occupied different habitats than phorusrhacids [65]. Borhyaenids, however, were large ground-dwelling carnivores and probably had some competitive effects on Phorusrhacidae [62,65]. Borhyaenids persisted from before the Deseadan to the Huayquerian [65]. Borhyaena was one of the few cursorial sparassodonts and co-occurred with phorusrhacids in the Santa Cruz formation (Santacrucian) [5,62,65]. Borhyaena is therefore the most likely of the sparassodonts to have competed with phorusrhacids.

(ii) . Evolutionary ecology

Based on our results, we hypothesize that competitive exclusion was a significant selective factor, preventing species of comparable niches from coexisting [66]. Co-occurring species were typically of disparate morphologies and differing ecologies. High phorusrhacid diversity within a single environment appears to have been maintained by niche partitioning facilitated by strong displacement in body size. Typically, each hypothesized niche is occupied only by one member of a single subfamily in a given environment, and co-occurrence does not occur within a single niche over extended periods of geologic time (figure 2).

Examples of apparent niche overlap within a subfamily occurred between the ambush macropredators, P. brasiliensis and P. fortis, and between the generalist predators, P. bachmanni and P. lemoinei. P. lemoinei and P. bachmanni were roughly equivalent in size (figure 2), but probably had variable diets similar to modern seriemas [5–7] as well as contemporary Cariamidae [64]. Able to coexist with other seriema-sized Cariamiformes, the smaller generalist phorusrhacids may have been exempt from the exclusionary trend observed in larger lineages. Although P. brasiliensis and P. fortis were both extremely large and graviportal, these species likely had different spatial niches. P. brasiliensis has only been observed in Brazil, while P. fortis was native to southern Argentina [1]. P. brasiliensis and P. fortis would not have directly competed under this scenario, thus their temporal co-occurrence did not violate the trend of competitive exclusion.

We conclude that no two species of the same body size could persist for very long at the same time unless there were mitigating circumstances which reduced niche overlap such as geographical separation or the ability to partition resources via generalist ecology. Where two subfamilies are observed occupying the same niche, we find that extinction in one lineage is always close at hand. We observe what appear to be two cases of replacement, with two unrelated lineages occupying the same niche in rapid succession.

We find evidence that Patagornithinae was replaced by Mesembriornithinae. The emergence of M. incertus during the Huayquerian, a competitor of equivalent niche to the patagornithine A. steulleti, immediately precedes the extinction of Patagornithinae with the transition to the Montehermosan SALMA (figure 2). This appears to be a unique instance of intraguild competition within Phorusrhacidae, as all other cases of niche overlap mentioned above can be explained by ecological evidence. Both subfamilies appear to have occupied the same niche in the same habitat. This competition, we hypothesize, drove the extinction of Patagornithinae. This short interval of co-occurrence, followed by the reaffirmation of niche partitioning, suggests extinction by competitive exclusion [66].

The more conclusive case of faunal succession occurred between gigantic phorusrhacids. Phorusrhacines, in particular the robust phorusrhacines, were morphologically most similar to Physornithinae, suggesting similar niche occupation. Unlike the above case, no temporal overlap is observed; Phorusrhacinae appears in the fossil record after the last appearance of Physornithinae (figure 2). Based on size, it appears that the phorusrhacines evolved to occupy the macropredator niche space left behind by Physornithinae. As evident by our PANCOVA, this niche represents a unique strategy employed by Phorusrhacinae and Physornithinae (figure 3). The initial successor, P. longissimus, was more cursorial and slightly smaller than P. fortis, but nonetheless was gigantic and apparently inherited the macropredator niche previously occupied by P. fortis. Given our interpretation that K. guillermoi and D. pozzi are indeed phorusrhacines, we see an increase in average phorusrhacine size over time (figure 2). These sizes are indistinguishable from those of Physornithinae, further indicating ecological succession.

If Brontornis is ultimately conferred Phorusrhacidae, this trend is still robust. The differences we observe in size and cursoriality between B. burmeisteri and P. longissimus would probably have provided the necessary ecological disparity to allow for their co-occurrence. As a phorusrhacid, B. burmeisteri would have occupied an ambush macropredator niche distinct from the pursuit macropredator niche occupied by P. longissimus. Robust Phorusrhacinae would then have filled the ambush macropredator niche left behind by the extinction of Brontornis.

Our findings suggest that competitive exclusion based on body size shaped terror bird diversity. Niche partitioning evolved early in Phorusrhacidae, dividing phorusrhacid ecologies by subfamilies. Niche overlap between and within lineages was rare, and each case is associated with either extinction, allopatry, or generalism. Gigantism emerged as part of the initial radiation of this group and was maintained throughout the evolutionary history of Phorusrhacidae. Gigantism in Physornithinae and Phorusrhacinae provided the means to exploit the apex predator niche and coexist with other predators on the landscape, including other subfamilies of Phorusrhacidae. Size based niche partitioning prevented competition between clades and was responsible for the continual high diversity and speciosity of terror birds in pre-interchange South American ecosystems.

Ethics

This work did not require ethical approval from a human subject or animal welfare committee.

Data accessibility

The datasets supporting this article are available in the electronic supplementary material [67]. Executable nexus file with matrix and MrBayes commands is available in ‘S1 Nexus’. Phylogenetic parameters and character list are available in ‘S1 Text’. Fossil measurements and literature sources are available in ‘S1 Table and S2 Table’.

Declaration of AI use

We have not used AI-assisted technologies in creating this article.

Authors' contributions

T.W.L.: conceptualization, data curation, formal analysis, funding acquisition, investigation, visualization, writing—original draft, writing—review and editing; J.D.G.: conceptualization, formal analysis, methodology, supervision, validation, writing—review and editing; C.L.O.: methodology, project administration, supervision, validation, writing—review and editing.

All authors gave final approval for publication and agreed to be held accountable for the work performed therein.

Conflict of interest declaration

We declare we have no competing interests.

Funding

This work was supported by the Montana State University Undergraduate Scholars Program (Undergraduate Research Grant, 2019).