Differential locomotor and predatory strategies of Gondwanan and derived Laurasian dromaeosaurids (Dinosauria, Theropoda, Paraves): Inferences from morphometric and comparative anatomical studies

Tetrapod limbs morphology is a reliable proxy of locomotor capacities. Beyond this, other aspects of life habits, such as predation abilities, can also be relevant to determine main morphofunctional appendicular properties, which ultimately reflect a compromise between different factors of the biological role. Dromaeosauridae is a dinosaur clade belonging to Theropoda, a group of bipedal predators. Dromaeosaurids represent an interesting study case, in which the hindlimbs have been proposed to be involved in both locomotion and predation activity. A peculiar feature characterizing all dromaeosaurids is a modified second pedal digit, which is typically related to predation. This theropod group is closely related to birds and diversified during the Cretaceous Period, mainly in the Northern Hemisphere (Laurasia). However, a subclade of dromaeosaurids, the Unenlagiinae, was recently recognized for Gondwana. Nevertheless, there are morphological differences between derived Laurasian dromaeosaurids (eudromaeosaurs) and unenlagiines. Such differences are observed in the proportions between hindlimb bones and in the presence of a subarctometatarsalian condition in unenlagiines, which is mainly characterized by a proximally constricted metatarsal III. To evaluate the function of these divergent morphologies, we conducted morphometric analyses and comparisons of qualitative morphological aspects, encompassing unenlagiines, other dromaeosaurids, as well as taxa from other theropod groups, including extant birds. The former approach consisted of two phylogenetic principal component analyses, one based on the main measurements of the hindlimb, and the other focused on the lengths of the pedal phalanges. The first analysis drew the unenlagiines close to taxa with long tibiae, as well as long and slender metatarsi. Instead, eudromaeosaurs are closer to taxa with shorter tibiae, and shorter and wider metatarsi. The second analysis showed that eudromaeosaurs and unenlagiines have similar phalangeal proportions, including the elongation of distal phalanges. However, the shorter second phalanx of the pedal digit II of eudromaeosaurs could have increased the force generated by this digit, which was the main predatory tool of the autopodium. This, together with a shorter and wider metatarsus, and a marked hinge‐like morphology of the articular surfaces of metatarsals and phalanges, possibly allowed eudromaeosaurs to exert a great gripping strength and hunt large prey. Conversely, the longer and slender subarctometatarsus, and less well‐marked hinge joints of unenlagiines possibly gave them greater cursorial capacities. Additionally, the longer second phalanx of digit II allowed unenlagiines fast movements of this digit to hunt smaller and elusive prey. Thus, the distinctive morphological evolutionary pathways of these two dromaeosaurid clades seem to have been influenced by the particular locomotor and predatory specializations that characterized each of these lineages.


| INTRODUC TI ON
Unenlagiinae is a clade of Gondwanan paravians, first recognized by Bonaparte (1999), which has been generally considered a subfamily of dromaeosaurids since the phylogenetic analysis of Makovicky et al. (2005). However, recent studies have challenged the dromaeosaurian affinities of unenlagiines and proposed instead an alternative phylogenetic hypothesis locating these theropods within the stem of Avialae Novas, 2011, 2013). Despite and beyond the ongoing controversy about the relationships of unenlagiines, there are many shared morphological features between unenlagiines and other dromaeosaurids. One of these shared traits is the presence of a modified pedal digit II, with a hyperextensible phalanx II-2 and a hypertrophied sickle-shaped claw. The peculiar form of this digit has led many researchers to propose multiple interpretations about its possible function (e.g. Colbert and Russell, 1969;Ostrom, 1969;Manning et al. 2006;Senter, 2009;Fowler et al. 2011). However, they all agree that this digit was involved in food procurement, as the main structure implied in the submission and/or the killing of the prey. Nevertheless, these functional interpretations are based mainly on the anatomy of derived Laurasian taxa (Eudromaeosauria, following Longrich and Currie, 2009;Cau et al. 2017; among others), such as Deinonychus, Velociraptor, Saurornitholestes, Achillobator and Dromaeosaurus. In these taxa the phalanges are markedly modified with respect to the plesiomorphic theropod morphology. The digit II of unenlagiines is similarly modified, although there are some anatomical differences from the digit II of eudromaeosaurs.
Moreover, the anatomical differences between unenlagiines and eudromaeosaurs are not limited to those found in this pedal digit, but also in other parts of the hindlimb. The metatarsus also differs between the two groups, since unenlagiines generally have a subarctometatarsalian condition (Novas and Pol, 2005;Currie and Paulina Carabajal, 2012;Brissón Egli et al. 2017;Gianechini et al. 2018), as also reported in microraptorine dromaeosaurids and some basal troodontids (e.g. Xu et al. , 2002Xu et al. , 2008Xu, 2002;Xu and Norell, 2004;Zheng et al. 2010;Xu and Qin, 2017).
By contrast, in eudromaeosaurs, the metatarsus has a structure more similar to the plesiomorphic theropod condition (e.g. Ostrom, 1969;Barsbold, 1983;Norell and Makovicky, 1999;Xu et al. 2010). In the subarctometatarsalian condition, the metapodium has a partially similar morphology to the arctometatarsus, a type of metatarsal morphology observed in some theropod groups, such as tyrannosaurids, ornithomimids and alvarezsaurids. White (2009) pointed out how these morphologies differ, indicating that in the subarctometatarsus the proximal end of the metatarsal III, although constrained, is equally visible in anterior and posterior views (whereas it is completely constrained proximally in the arctometatarsus and not visible). Additionally, in posterior view, the third metatarsal is visible through the entire length of the metatarsus, excluding metatarsals II and IV from buttressing. Several functional hypotheses have been raised regarding the arctometatarsus, most of them linking it with an increase of mechanical efficiency during locomotion (Coombs, 1978;Wilson and Currie, 1985;Holtz, 1994;Snively and Russell, 2003;Snively et al. 2004;White, 2009). The subarctometatarsalian condition could also have been related to enhancement of locomotor efficiency, and some authors consider it as transitional between the plesiomorphic morphology and the arctometatarsalian condition (White, 2009).
In unenlagiines and eudromaeosaurs, the hindlimb, especially the autopodium, is implied in both locomotor and predatory functions.
Accordingly, the morphological differences possibly reflect different locomotor and predatory habits. Based on the previous ideas about the functional implications of the subarctometatarsalian and the arctometatarsalian conditions, unenlagiines would have had locomotor capabilities not present in eudromaeosaurs. Such hypotheses have already been mentioned by previous authors (e.g. Fowler et al. 2011), although they have not been evaluated in a quantitative form, at least not for the case of unenlagiines. The goal of the present study was to perform an analysis including both unenlagiines and eudromaeosaurs, and to approach, in a quantitative manner, the morphological differences between these groups. Additionally, exhaustive morphological comparisons between the autopodia of unenlagiines and eudromaeosaurs and taxa of other theropod groups are performed in order to arrive at a conclusion about the possible functions of the autopodium, and the hindlimb in general, in these two dromaeosaurid clades. and phalanges, possibly allowed eudromaeosaurs to exert a great gripping strength and hunt large prey. Conversely, the longer and slender subarctometatarsus, and less well-marked hinge joints of unenlagiines possibly gave them greater cursorial capacities. Additionally, the longer second phalanx of digit II allowed unenlagiines fast movements of this digit to hunt smaller and elusive prey. Thus, the distinctive morphological evolutionary pathways of these two dromaeosaurid clades seem to have been influenced by the particular locomotor and predatory specializations that characterized each of these lineages.

K E Y W O R D S
Dromaeosauridae, functional morphology, life habit, Paraves, Unenlagiinae

| MATERIAL S AND ME THODS
In order to evaluate quantitatively how unenlagiines and eudromaeosaurs differ morphologically, a morphometric analysis was performed employing a set of linear measurements of the hindlimb bones of several theropod taxa. A diverse sample of theropod clades was considered, with the aim of covering a wide spectrum of morphologies, proportions and sizes of the hindlimb elements.
The sample includes measurements of Herrerasaurus, non-Tetanurae neotheropods, basal tetanurans, and representatives of most coelurosaur clades, including Mesozoic avialans. Data from more recent, albeit extinct groups of birds, i.e. Dinornithiformes, were also considered, as were data from extant taxa representatives of diverse ecomorphs, in which the locomotor habit, mode of feeding, and particular abilities of the foot such as 'grasping', are known. Extant bird taxa considered in the study include mainly ground-dwellers with high cursorial capacities as well as less cursorial or even non-cursorial, raptorial birds with different hunting modes and 'grasping' capacities, as well as perching birds with more arboreal habits, such as passeriforms, which also have 'grasping' abilities (Appendix S1). The measurements considered include the proximodistal lengths of the femur (FL), tibiotarsus (TL), metatarsus (MtL), and non-ungual pedal phalanges, and the mediolateral width of the metatarsus at midshaft (ML). The MtL measurements were taken for the longest element, typically the metatarsal III. For modern birds, the length of the tarsometatarsus was considered as MtL, due to the complete fusion of the distal tarsals and the metatarsals. The dimension ML refers to the mediolateral diameter of the articulated metatarsals (MTs) II, III and IV at midshaft of these bones.
Most of the measurements were obtained from published datasets (e.g. Holtz, 1994;Karhu and Rautian, 1996;Xu, 2002;Mayr et al. 2007;Zhou et al. 2010;Turner et al. 2011;Lü and Brusatte, 2015;Tsogtbaatar et al. 2017). Other values were measured directly on materials deposited in different collections (Appendix S1). For many taxa with published measurements, ML was not provided by the authors, so in these cases it was estimated from the published photographs of the specimens. For each taxon, we specified from which specimen the measurements were taken, except in some cases where such identification was not provided by the authors who originally published the data. In the case of taxa with published measurements of several specimens, we decided to consider the data of only one of the specimens, specifically the largest one, in order to avoid data from potentially juvenile individuals. We included mainly complete specimens, i.e. those with all the bones of the hindlimb completely preserved, in order to gather all the required measurement data. In some cases, measurements were estimated for bones with only small portions missing, so that our estimates should approximate reliably the real dimensions of the element. Additionally, measurements of some taxa were obtained directly from materials housed in Argentinean repositories, including one specimen of the alvarezsauroid Alnashetri cerropoliciensis (MPCA), specimens of 17 taxa of extant birds (MACN, one specimen by taxon), and one specimen of Struthio camelus (CFA-OR), as specified in Appendix S1.
The length of the pedal ungual phalanges is measured by some authors in a straight line from the proximal end to the distal end of the phalanx, while others measure only the external curvature. In consequence, published lengths of pedal unguals of theropods are not taken according to the same criteria.
Therefore, in the absence of a consensus, we did not consider the lengths of unguals in our analyses. The length of the phalanges of digit I is also not included, because in certain taxa of the clades included in the analysis, i.e. ornithomimids, this digit is atrophied or absent.
We performed phylogenetic principal component analysis (phylogenetic PCA or (PPCA); see Revell, 2009Revell, , 2012 using these measurements instead of the traditional PCA. The phylogenetic PCA allows the reduction of the original variables to principal components, while taking into account the non-independence among the former due to the phylogenetic relationships between species.
In this way, in a phylogenetic PCA the samples are not considered as independent datapoints, an assumption of the traditional PCA and frequently violated due to the phylogenetic relationships between samples (Revell, 2009).
Given that the purpose of these analyses was the study of shape changes between species that cover a wide range of sizes, the phylogenetic PCAs were constructed from size-standardized, Mosimann variables (Mosimann and James, 1979) instead of the original ones.
Each Mosimann variable was obtained as the ratio between the original variable and the geometric mean of all variables considered for the corresponding phylogenetic PCA.
From the complete dataset, two phylogenetic PCAs were performed. One of them includes the measurements of the long bones of the hindlimb, i.e. FL, TL, MtL and ML, and the other, the lengths of the non-ungual pedal phalanges. Due to data availability (Appendix S1), the first PPCA included 74 taxa, whereas the second included 32 taxa (Fig. 1). This methodological design resulted in a different taxonomic coverage in each PPCA (in relation to the available data and due to the inability to perform these analyses with missing data), allowing the maximization of the number of morphologies and taxa considered in each analysis.
After the phylogenetic PCA was computed, the phylogenetic relationships between species were projected into bivariate plots of morphospaces, constructing phylomorphospaces (Revell, 2012). The evaluation of the phylogenetic signal on each phylogenetic principal component was made with the K statistic proposed by Blomberg et al. (2003) calculated for each axis. The K statistic provides a measure of the strength of the phylogenetic signal within the data. Values smaller than one indicate a lack of phylogenetic signal, or strong adaptative processes. Values near one are expected if the character evolved following the phylogenetic relationships, under a Brownian motion model. Values greater than one show that phylogenetically close taxa are more similar than expected and they eventually indicate stasis (Blomberg et al. 2003;Losos, 2008).
Additionally, the size effect on each axis of the morphospaces was calculated using phylogenetic generalized least squares (PGLS) regressions (Martins and Hansen, 1997), considering the geometric means as the independent variables. A PGLS regression allows the incorporation of the phylogenetic structure of samples as the error term of the regression equations, and then consideration of the biases caused by phylogeny in the calculation of the relationship between the analysed variables.
The curvature angles of unguals of unenlagiines and Laurasian dromaeosaurids were measured using the methodology applied by Fowler et al. (2009), which in turn is based on that of Pike and Maitland (2004). Both the external and inner curvature angles of the unguals are measured with this methodology, obtaining the angle between the base and the tip of the claw. However, as this methodology was used to measure ungual curvatures of extant taxa of birds with soft tissue on digits, some modifications were made. For extant birds, the base of the claw is considered at the point where the keratinous sheath emerges from the skin of the digit. However, in fossil unguals lacking the sheath and soft tissue, the same methodology cannot be applied for the measurement of the curvature angles.
Accordingly, we take the proximodorsal tip of the ungual bone as the dorsal base to measure the external curvature angle, and the tip of the flexor tubercle as the ventral base (Fig. S1). However, the flexor tubercle shows two ventral tips in the unguals of the analysed theropods, separated by an extension of the side groove of the claw. In such cases, the distal tip was taken as the base to measure the angle of the inner curvature. The angles were taken from photographs of the ungual phalanges using the measure tool in Adobe Photoshop.
The incomplete unguals which have not preserved the distal or the proximoventral ends were reconstructed, although in these cases it was indicated that the angle values are estimated. and minor contributions from the metatarsus length and ML (both variables negatively correlated). High positive PPC2 scores identify mainly taxa with elongated femora, and somewhat short and slightly slender metatarsi, whereas negative scores usually characterize taxa with shorter femora, and slightly longer and wider metatarsi (see also Appendix S2).

| Description of the phylogenetic PCA based on measurements of the hindlimb long bones
In general, extant birds and Dinornithiformes are partially segregated from the MzTer, toward negative scores of PPC1 and PPC2 ( Fig. 2). This is mainly because these groups have longer and more slender metatarsi, longer tibiae, and shorter femora in comparison with the MzTer. F I G U R E 1 Composited phylogeny indicating the relationships between taxa included in the study, based on cladograms published by previous authors (cited in the text). Taxa in red correspond to those included in the phylogenetic principal component analysis (PPCA) based on the measurements of the hindlimb long bones, taxa in green correspond to those included in the PPCA based on the lengths of the pedal phalanges, and taxa in blue are those included in both analyses The MzTer with high negative PPC1 scores include the alvarezsauroids, derived ornithomimids, some oviraptorosaurs, basal avialans, troodontids, microraptorines and unenlagiines.   (Fig. 2). These taxa show a long metatarsus, although slightly shorter and wider than other MzTer such as the alvarezsaurids.
Thus, they are located on fewer negative PPC1 scores and higher positive PPC2 scores. Rahonavis is closer to the oviraptorosaur Wulatelong than to Buitreraptor, presenting fewer negative PPC1 scores and slightly lower positive PPC2 scores. This separation appears because Rahonavis has a slightly shorter and wider metatarsus than Buitreraptor and the other taxa clustered with this Argentinian unenlagiine.
The eudromaeosaurs Deinonychus and Velociraptor segregate and locate on lower negative PPC1 scores than other dromaeosaurids, including Buitreraptor, since they have markedly shorter and wider metatarsi and shorter tibiae. In fact, Deinonychus lies closer to tyrannosaurids than to other dromaeosaurids. Velociraptor is located on remarkably higher positive PPC2 scores because it has an even shorter metatarsus and tibia, relative to the femur. Bambiraptor is markedly separated from other derived Laurasian dromaeosaurids, mainly because it has a comparatively long metatarsus.

| Influence of phylogeny in the distribution of taxa across the morphospace
The Blomberg K values indicate that the taxa distribution along the PPC1 is strongly influenced by the phylogenetic relationships of major clades (K = 2.714). PPC2 is less influenced by deep phylogenetic relationships, and is instead related to the influence of the phylogenetic structure of terminals and of more inclusive clades (K = 0.262; Table   S1). Thus, the segregation and relatively scarcely overlapping distribution of the major clades along the PPC1 can be related to the high K

| Description of the phylogenetic PCA based on lengths of the phalanges
In the phylogenetic PCA made using the lengths of the phalanges, the contributions of the variables to PPC1 represent 39.0%, and to PPC2, 29.1% of the total variation (Fig. 4). Because these two axes explain a small percentage of the variation, we also analysed the third component (PPC3; 10.8% of the total variation).
In the graphic of PPC1 vs PPC2 (  Additionally, the dispersion of taxa is linked to the large number of morphological convergences between distantly located taxa (Table   S2). The PPC3 shows a K value closer to 1 (K = 0.811), and thus it fits more closely with a stochastic model (i.e. the distribution of taxa follows their phylogenetic relationships, but is not particularly strongly influenced either by deep nodes or by terminal relationships).
For instance, basal taxa included in the analysis, such as basal tetanurans and the basal coelurosaur Gorgosaurus, are almost overlapping on similar values of PPC1, although they are separated along PPC2 and PPC3 (Figs 6 and 7).
Dromaeosaurids show a certain amount of convergence between basal and derived taxa, since Buitreraptor is located near the derived dromaeosaurids Deinonychus and Bambiraptor (Figs 6 and 7).
These three taxa have a comparatively elongated digit IV relative to Microraptor and Sinornithosaurus, which are more derived than Buitreraptor, although more basal with respect to Deinonychus and Bambiraptor (Fig. 1).

| Influence of size in the distribution of taxa across the morphospace
The results of the PGLS regressions indicate that the axes that compose the morphospace analysed for the phalanx measurements (i.e.  (Gambaryan, 1974;Coombs, 1978;Hildebrand, 1982Hildebrand, , 1985Hildebrand, , 1988Garland and Janis, 1993;Carrano, 1999). From the perspective of locomotor performance, animals known as cursorials have the capacity to move at greater velocities or for long distances with a low energetic cost (Gregory, 1912;Garland and Janis, 1993;Stein and Casinos, 1997;Carrano, 1999). However, Carrano (1999) considered that a discrete categorization of the locomotor habits might not be appropriate. Instead, these habits should be evaluated along a multivariate continuum between two locomotor extremes, i.e. strictly graviportal and cursorial. Theropods can be generally considered as cursorial animals (or 'subcursorial', according to Coombs, 1978), as they were bipeds, digitigrades and with long and parasagittally oriented hindlimbs (Farlow et al. 2000). However, different theropod taxa would be dispersed along a continuum that includes different grades of cursoriality. The distribution in the morphospace obtained in the multivariate analyses performed could indeed reflect such ecomorphological diversity. Taxa with more elongated distal segments of the hindlimbs (i.e. tibia and metatarsus), a more slender and compressed metapodium, and reduced lateral pedal digits likely had a greater cursorial capacity (Hildebrand, 1988;Carrano, 1999).
These taxa would locate closer to the 'cursorial extreme' of the multivariate continuum than taxa with shorter segments of the hindlimb, with a more robust metapodium, and lateral digits less reduced.
The elongation of the distal elements of the hindlimb (tibia and metatarsus) allows increasing of the stride length and speed of movements, which are related to a greater cursorial capacity (Carrano, 1999;Fowler et al. 2011). Garland and Janis (1993) explained that the ratio between the lengths of metatarsus and femur (MT/F) was repeatedly considered by some authors as a predictor of locomotor performance in fossil forms. However, Garland and Janis (1993) and other authors (Gatesy and Middleton, 1997;Zeffer et al. 2003;Habib and Ruff, 2008) warned that ratios between hindlimb bones are not good predictors of the type of locomotion, so limb proportions must be considered with caution. Thus, it is important to also take into account qualitative aspects, such as the morphology of the metapodium, to make inferences about locomotor abilities of specific taxa.
The arctometatarsalian and subarctometatarsalian conditions could confer significant cursorial capabilities to their bearers. Some authors (Holtz, 1994;White, 2009) have noted that theropod taxa presenting these conditions have distal elements of the hindlimb significantly more elongated than taxa with a plesiomorphic metapodium. Moreover, many authors have postulated biomechanic hypotheses about the performance of the arctometatarsalian and subarctometatarsalian foot (Coombs, 1978;Wilson and Currie, 1985;Holtz, 1994;Russell, 2002, 2003;Snively et al. 2004;White, 2009). These authors indicated how the interaction between metatarsals and the transference of forces along the metatarsus provide advantages during locomotion, and how this could represent benefits for the cursorial habit.
By contrast, extant birds with a grasping foot are characterized by an elongation of the distal pre-ungual phalanges of the digits, especially the penultimate phalanx (Fisher, 1946;Hopson, 2001;Zhou and Farlow, 2001;Kambic, 2008;Fowler et al. 2011;Kavanagh et al. 2013). This feature can be observed both in perching and in raptorial extant birds. The elongation of the distal phalanges is convergently observed even in arboreal mammals which have grasping autopodia, such as the sloths (Kavanagh et al. 2013, and references therein).

| Interpretation of the phylogenetic PCAs: their significance for the locomotor habits of theropods
Taking into account the diverse factors and how they affect the hindlimb elements differentially, it is important to consider both analyses together (i.e. proportions of the long bones and the phalanges) between the two taxa (Fig. 2). However, the second phylogenetic PCA based on the lengths of the phalanges reveals clear dissimilarities between these taxa (Figs 4 and 5). The latter analysis indicates that the cursorial capacities of Avimimus are greater than those of Sinornithoides, whose phalangeal proportions are possibly more related to a grasping function.
Based on the results of the second phylogenetic PCA made using lengths of the phalanges, taxa such as Avimimus, Cariama and Rhea are considered to have greater cursorial abilities (Gonzaga, 1996;Picasso, 2010;Degrange, 2017 I   II   III   IV   I   II   III   IV   I   II   III   IV   I   II   III   IV   I   II   III   IV  II   III   IV   II   III   IV   I   II   III   IV   I   II   III robust metatarsi. These taxa can be considered to have less advanced cursorial abilities than those taxa located at low positive to negative values of PPC2 (Dilong, Archaeornithomimus, Elaphrosaurus and Herrerasaurus), taxa which have longer and slender metatarsi.
Modern birds also show the same general trend.
Along the PPC1, taxa having positive to low negative scores can be considered as having less cursorial capacities than those located at higher negative scores (Fig. 9). Accordingly, taxa such as Linhenykus and Parvicursor are interpreted as highly cursorial, an interpretation that is consistent with their highly derived, markedly elongated and slender arctometatarsus (Karhu and Rautian, 1996;Xu et al. 2011Xu et al. , 2013. Unfortunately, these two taxa have not preserved all the pedal phalanges and so they cannot be included in the second analysis based on the lengths of the phalanges. However, in Sinovenator the phalanges of digit III appear to shorten distally and phalanx IV-4 is slightly longer than IV-3 (Xu, 2002).

| Functional implications related with the proportions of the long bones
The main differences between the hindlimbs of unenlagiines and eudromaeosaurs are related to the relative length and form of the metatarsus, as well as to the morphology of the phalanges of digit II Porfiri et al. 2011;Agnolín and Novas, 2013;Gianechini et al. 2018). The results of the phylogenetic PCAs indicate that in unenlagiines (except in Rahonavis), the metatarsus is significantly elongated when compared with the femur and tibia. Additionally, the lateromedial width of the metatarsus (ML) is significantly reduced related to its total length (MtL; except in Rahonavis) (Figs 8 and 9). By contrast, in eudromaeosaurs, the metatarsus is definitely shorter, and the ML/MtL ratio is higher.
These comparisons indicate that the metatarsus of eudromaeosaurs is more robust overall than that of the unenlagiines. Bambiraptor is not a eudromaeosaur, although it has morphological characters of the autopodium more similar to those of eudromaeosaurs and also it is phylogenetically closer to these than to other dromaeosaurids ( Fig. 1).
The metatarsi of Neuquenraptor (MCF PVPH 77) and Austroraptor (MML 195) are incomplete, although their approximate length can be estimated, indicating that they were very elongated with respect to the tibia and femur. Thus, these taxa possibly had length proportions of the hindlimb bones much similar to those of Buitreraptor.
The proportions of the hindlimb long bones of Buitreraptor are remarkably different with respect to those of Bambiraptor and the eudromaeosaurs analysed here, i.e. Velociraptor and Deinonychus (Fig. 9). Instead, Buitreraptor is more similar in this respect to other taxa with a relatively elongated metatarsus, such as Mahakala, Alnashetri, Zhongjianornis, Zhenyuanlong, Sinovenator and Mei. These taxa are similar in size or smaller than Buitreraptor Xu and Norell, 2004;Turner et al. 2007Turner et al. , 2011Zhou et al. 2010;Gao et al. 2012;Makovicky et al. 2012Makovicky et al. , 2016Lü and Brusatte, 2015).

According to previous authors, similar size and hindlimb proportions
would presumably indicate a similar locomotor mode (Holtz, 1994;Gatesy and Middleton, 1997;White, 2009).
Rahonavis departs from the general morphology of other unenlagiines by its shorter tibia, as well as shorter and wider metatarsus (Forster et al. 1998;Figs 8 and 9). Nevertheless, Rahonavis has hindlimb proportions more similar to those of unenlagiines than those of eudromaeosaurs, especially because it has a comparatively short femur and long tibia. Thus, Rahonavis can be considered as the least cursorial unenlagiine analysed, although clearly more cursorial than the eudromaeosaurs.

| Functional implications inferred from proportions of the pedal phalanges
The only unenlagiine to date with all the pedal phalanges preserved is Buitreraptor. Our results indicate that it has proportions of the phalanges similar to those of Bambiraptor and of the eudromaeosaur Deinonychus. The three taxa share their markedly elongated digit IV, which has a total length greater than that of digit III (Fig. 8). We can estimate that Neuquenraptor and Rahonavis have a digit IV shorter than the digit III, as do Sinornithosaurus and Microraptor, because the sum of lengths of the other pre-ungual phalanges of digit IV is significantly lower than the total length of digit III. Thus, the complete digit IV would have been slightly shorter than digit III, even if Ph.
IV-4 had the same length or was even slightly longer than Ph. IV-3.
By contrast, in other MzTer included in the analysis such as derived troodontids, non-paravian coelurosaurs, and basal tetanurans, the digit III is clearly the longest and digit IV is significantly shorter. These inter-digit proportions are linked with the cursorial abilities of the taxa involved (Abourachid and Renous, 2000;Moreno et al. 2007;Fowler et al. 2011). Accordingly, the length proportions of dromaeosaurids digits, including unenlagiines and especially Buitreraptor, seem to indicate a restriction to their cursorial habit.
Furthermore, dromaeosaurids show a significant elongation of the distal pre-ungual phalanges, a feature related to grasping capacities (see literature cited above). In extant birds with a grasping foot, such as Turdus and Bubo, the distal phalanges are significantly elongated (Appendix S1 and Fig. 8), independently of their ecological habits and the type of objects they usually hold with their feet. Generally, in unenlagiines, the length proportions of the distal phalanges of digit III are similar to those of the eudromaeosaur Deinonychus. However, in unenlagiines the second phalanx of digit II is shorter than the first one (Appendix S1), indicating  Ostrom (1976); (d) based on Russell and Dong (1993); (f) modified from Gatesy and Middleton (1997); (h) based on Karhu and Rautian (1996) However, the available data and the apparently long distal phalanges of digit IV in Neuquenraptor indicate that it probably had more accentuated grasping capacities than other unenlagiines, resembling those of Bambiraptor and the eudromaeosaur Deinonychus (Appendix S1).

| Functional implications related to qualitative aspects of the metatarsus and motion range of digits
Regarding its qualitative features, the subarctometatarsalian condition of unenlagiines also supports the presence of high cursorial abilities within the group. This condition is observed in Buitreraptor, Neuquenraptor, and possibly Austroraptor (based on the specimen MML 220). However, Rahonavis has a non-subarctometatarsalian metatarsus, indicating less cursorial abilities than other unenlagiines.
Additionally, previous authors noted differences in the distal articular surfaces of metatarsals between unenlagiines and eudromaeosaurs (e.g. Agnolín and Novas, 2011;Fowler et al. 2011;Gianechini et al. 2018). In eudromaeosaurs the MT I, II and III have a well-developed ginglymoid distal articular surface (Colbert and Russell, 1969;Ostrom, 1969;Makovicky, 1997, 1999;Fowler et al. 2011). This could indicate that the first phalanges flexed and extended predominantly in a single plane (Fowler et al. 2011). Instead, in unenlagiines the ging-

| Functional implications related to qualitative aspects of the pedal phalanges
From the point of view of qualitative aspects, the digit II of unenlagiines is modified in a similar way to that of eudromaeosaurs, although important differences can be observed. First, in unenlagiines such as Buitreraptor, Neuquenraptor and Unenlagia paynemili (MUCPv 1066), the distal articular surface of phalanx II-2 is less proximally extended.
This feature restricts the extension of the ungual phalanx, as can be observed in an isolated articulated digit II of Buitreraptor (MPCA 478, Gianechini et al. 2018), in which the ungual seems to be totally extended (Fig. 10). which probably was inserted onto the proximoventral zone of the phalanx as in extant birds; Hudson, 1937;Vanden Berge and Zweers, 1993). Thus, the shortness of Ph. II-1 could maximize the mechanical advantage of the flexor muscle and the grasping strength of the digit II. Another difference is the more proximally extended proximoventral heel of phalanx II-2 of eudromaeosaurs, which possibly was an insertion point of flexor muscles (Ostrom, 1969).
The mentioned characters of the digit II of the eudromaeosaurs seem to indicate the capacity to exert stronger predatory efforts, which could be an advantageous feature for subduing large prey.
Conversely, the phalangeal morphology of unenlagiines would indicate weak predatory efforts. Moreover, the longer Ph. II-1 of unenlagiines also suggests faster movements of digit II, which could be eventually useful for hunting small prey.
Differences in the degree of development and curvature of the claw of digit II between eudromaeosaurs and unenlagiines are difficult to evaluate, mainly because most unenlagiines have not preserved a complete ungual. The available data (Table S3)

| Morphological and functional correlates in extant raptorial birds and possible resemblances with dromaeosaurids
An interesting convergence is observed in the morphospace of the long bone measurements between extant raptorial birds and some eudromaeosaurs. Both groups tend to show positive PPC2 values (Fig. 2), as they have relatively long femora and consequently shorter metatarsi. Moreover, raptorial birds converge specifically with Deinonychus and Velociraptor in the presence of wider metatarsi, as reflected by their fewer negative values in the PPC1. In general, in extant raptorial birds a shorter and robust metatarsus is related to the ability of the foot to exert a greater grip force.
By contrast, a longer metatarsus is correlated with a minor grip force although it supports the capacity for rapid movement (Ward et al. 2002;Zeffer et al. 2003;Einoder and Richardson, 2007;Habib and Ruff, 2008;Fowler et al. 2009Fowler et al. , 2011. In a general way, owls (Strigiformes) have the shortest and most robust metatarsus, whereas falconiforms and especially accipitrids have a longer and more slender metatarsus (Ward et al. 2002;Einoder and Richardson, 2007;Fowler et al. 2009). Thus, owls have a greater grip capacity and strength, although these abilities are also related to other characters of the foot, such as the presence of sesamoids, a specialized tendon-locking mechanism, and a facultative zygodactyl condition (Ward et al. 2002;Einoder and Richardson, 2007;Fowler et al. 2009 (Mosto et al. 2013).
Similarly, the short and robust metatarsus of eudromaeosaurs, such as Velociraptor and Deinonychus, could have allowed a great grip force (Ostrom, 1969;Fowler et al. 2011). By contrast, the elongated subarctometatarsus of unenlagiines could have allowed a greater capacity of rapid movement, like in falconiforms and accipitrids, although it could have reduced grip strength (Fowler et al. 2011).
Although certain morphological and even functional features can be compared among these dromaeosaurids and extant raptorial birds, it must be also considered that these birds are predominantly aerial with a generally limited terrestrial locomotion (but see Mosto et al. 2013). Raptorial birds share many features of the autopodium, e.g. elongation of the distal non-ungual phalanges, independently of their specific type of prey and the hunting method they employ. This can be interpreted as the result of a predominant influence of hunting and grasping specializations, instead of terrestrial locomotion (Eyton, 1867;Fisher, 1946;Hopson, 2001;Kambic, 2008;Kavanagh et al. 2013). Conversely, dromaeosaurids had a terrestrial locomotion, although a partially or primarily arboreal habitat has been suggested for some taxa (Chatterjee, 1997;Manning et al. 2006Manning et al. , 2009. So, it is to be expected that both factors of selective pressures, i.e. predation and terrestrial locomotion, had a great influence on their hindlimb and autopodium. This is a main reason behind the segregation between extant birds and dromaeosaurids in the morphospace. Also it might

| Locomotor and predatory habits of Buitreraptor and other unenlagiines
Based on the previous discussion, unenlagiines possibly had a better cursorial locomotor performance and the capacity to reach greater running velocities than eudromaeosaurs. Of course, this does not mean that eudromaeosaurs did not have an effective locomotion and the ability to run fast. Possibly, eudromaeosaurs may have made sudden bursts of runs at high speed, but for shorter periods of time and/or for short distances, whereas unenlagiines could have maintained an accelerated pace for longer time and/or distance. The metatarsus of eudromaeosaurs has a structure with functional adaptations possibly more useful to effective predation than to cursorial locomotion. The morphological differences of the pedal phalanges between the two groups, especially those of digit II, could be more directly related to different predatory habits.
It is remarkable that the metapodium had a great morphological plasticity along the evolution of dromaeosaurids. Its structure differs drastically between unenlagiines and eudromaeosaurs (and microraptorines, which also have a subarctometatarsalian condition), depending on the relative and differential importance of the mechanical benefits associated both with predatory and locomotor functions. By contrast, as the results of the phylogenetic PCA indicate, the length proportions of the phalanges are not meaningfully dissimilar between these groups. A probable explanation for this is that the phalanges are the main elements implied in predatory functions. Consequently, the predatory habit exerted a greater selective pressure on the morphology of the phalanges, regardless of the particular feeding strategy (i.e. the way of hunting the prey) and locomotor habit of the taxa concerned. Nevertheless, some specific differences observed in unenlagiines are remarkable, such as the longer and slender phalanx II-1 (as the phylogenetic PCA indicates), and the greater freedom of movement of the remaining digits (which can be inferred from the interphalangeal articular morphology).
These features could have allowed unenlagiines a fast and secure grip of small and agile/elusive prey that did not demand great efforts to be subdued.
Unenlagiines and microraptorines have similar modifications of the metapodium, and thus they probably had a similar mode of moving on the ground, except the capacity of gliding postulated for some microraptorines (Xu et al. 2003;Chatterjee and Templin, 2007;Alexander et al. 2010). Probably, these two groups of dromaeosaurids used digit II for predation, although their predatory habits, i.e. the way of hunting and the type of prey, were not necessarily the same. Notably, some microraptorines (at least Microraptor and Sinornithosaurus) have a phalanx II-1 shorter than II-2 (Appendix S1), as in eudromaeosaurs. Additionally, some specimens of Microraptor gui indicate it fed on mammals, enantiornithine birds and fishes, which is evidence for diversified feeding habits and for their ability to exploit different ecological niches on ground, trees and water (Larsson et al. 2010;O'Connor et al. 2011;Xing et al. 2013).
It is likely that unenlagiines preyed on rapid and elusive animals, although it is difficult to know more specifically the type of prey that they hunted. In addition, there is no direct evidence of their feeding habits, such as the gut content of the Microraptor specimens. Nevertheless, it is possible to achieve an approximation of the feeding habits of unenlagiines, especially for the better represented taxa such as Buitreraptor. Regarding other unenlagiines the available information is more scarce, so it is more difficult to infer whether they differed in their predatory strategies or the type of prey they hunted.
Based on the small size, slender proportions (especially those of metapodium), and the inferred cursorial capacities of Buitreraptor, it probably foraged on the ground searching small prey, such as invertebrates, reptiles or mammals, over large distances and probably employing high-speed pursuits in some cases.
The fauna recorded from the fossiliferous area of La Buitrera, where Buitreraptor was discovered, also includes remains of small tetrapods such as snakes, sphenodonts, crocodyliforms and mammals (Carignano et al. 2002;Apesteguía and Novas, 2003 and Tsaagan (Colbert and Russell, 1969;Ostrom, 1969;Currie, 1995;Barsbold and Osmólska, 1999;Norell et al. 2006;Turner et al. 2012), although many taxa have denticles only on the distal carina. Such a dentition would have allowed ingestion of larger prey or tearing and cutting the flesh from them into smaller pieces. Feeding models have been proposed for some taxa, such as Deinonychus (Fowler et al. 2011), although they are difficult to apply to Buitreraptor because the size of the teeth and their lack of denticles. Buitreraptor did not have other flesh-tearing structures (e.g. the tomial tooth of extant raptorial birds), so it is very likely that it consumed whole small animals and that the teeth were mainly employed as a tool to hold them. Also, it is possible that these teeth have been used to tear apart small prey, in order to consume them in more than one swallow. In previous works, it has been postulated that the dentition of Buitreraptor would indicate a piscivorous feeding mode ). Certainly, Microraptor also had small non-serrated teeth and there is evidence that it fed on fish. However, this unique feature is not a reliable indicator of a strict piscivorous diet, since other morphological evidence must also be taken into account.
Moreover, Microraptor included in its diet other animals in addition to fish, as mentioned above. Buitreraptor is also characterized by having long forelimbs and hands (Agnolín and Novas, 2013;Novas et al. 2018), which could also have used to handle the prey once it was captured and subjugated with the feet.
Extant long-legged and predominantly terrestrial birds that forage on the ground and hunt small prey include the seriemas (Cariamiformes) and the Secretary Bird (Falconiformes). The Secretary Bird kicks and stamps on the prey until it is wounded or incapacitated, and then takes it with its beak (Kemp and Kemp, 1978;Kemp, 1995;Portugal et al. 2016). By contrast, the Red-legged Seriema (Cariama cristata) takes the prey with its beak and hits it on the ground with sudden movements of the head until it is injured (Boyle, 1917). An interesting trait of this seriema species is that it has a markedly curved ungual phalanx on the second digit (Burmeister, 1937;Jones, 2010; F.A.G., personal observation of MACN 23873).
Some authors proposed that this bird uses its enlarged claw to hold the prey against the ground, although others do not agree (Gonzaga, 1996, and references therein). The extinct phorusrhacids were terrestrial, generally flightless carnivorous birds, which are also characterized by having a markedly developed and curved ungual of the second digit (Sinclair and Farr, 1932;Alvarenga and Höfling, 2003;Jones, 2010). Some authors have proposed that this claw could be used as a means of apprehending the prey on the substrate, before using the beak to tear it apart (Jones, 2010). Buitreraptor could have used its pedal claw in a way similar to that proposed for seriemas and phorusrhacids, although there is no direct evidence for this.
Other unenlagiines, such as Austroraptor, probably used a strategy of hunting and subjection of the prey similar to that of Buitreraptor. Although Austroraptor is significantly larger (estimated total length: 5 m), it has numerous, non-serrated, and small teeth in comparison with the size of the skull (Novas et al. 2009;Gianechini et al. , 2017Gianechini, 2014). However, the teeth of Austroraptor are conical, so they probably were more resistant and could have been employed to seize and dismember large prey.
Austroraptor probably had length proportions of the hindlimb bones similar to those of Buitreraptor. Also it had a subarctometatarsalian condition, suggesting potentially good cursorial abilities.
However, Austroraptor had strikingly shorter arms compared to other unenlagiines, so it would not have used them to manipulate the prey, or at least not in the same way that Buitreraptor.
Rahonavis was probably a less cursorial taxon due to its hindlimb morphology. However, it had a relatively long tibia, so fast chases of prey cannot be ruled out as a hunting strategy used by this taxon. Moreover, Rahonavis has a digit II similar to that of other unenlagiines, so it probably had similar functional capacities. Nevertheless, the distal phalanges are shorter than in other unenlagiines, so it probably had slightly lower gripping abilities.
Unfortunately, cranial remains and teeth of Rahonavis are unknown, so it is more difficult to speculate about the type and size of animals that it could have been preyed upon. Surely it fed on small prey, although is not possible to know if it was able to tear flesh from larger prey.
For Rahonavis an arboreal habit can be proposed, based on its small size and the, albeit weak, grasping ability of its autopodium.
Many extant birds with grasping abilities of the foot are 'perchers' and have this habit, i.e. they are predominantly arboreal foragers (Glen and Bennet, 2007). Furthermore, this type of lifestyle is correlated in paravians with aerial locomotor capacities.
Rahonavis shows evidence of feathered forelimbs (Forster et al. 1998) and many osteological traits that suggest the capacity for flapping flight (Agnolín and Novas, 2013). Also, the claw of pedal digit II has been considered as a potential tool for climbing trees (Chatterjee, 1997;Manning et al. 2006Manning et al. , 2009 (Turner et al. 2007).
The specimens of Neuquenraptor, U. comahuensis and U. paynemili are much more fragmentary than those of other unenlagiines, as for example none of them have cranial remains, and thus attempts to infer the habits of these taxa represents a greater challenge. However, in the case of Neuquenraptor the features of its hindlimb indicate that velocity was probably important to obtain its prey. Only scarce hindlimb remains are known for U. comahuensis and U. paynemili, including the phalanges of digit II, which are very similar to those of the other unenlagiines (Novas and Puerta, 1997;Calvo et al. 2004;Porfiri et al. 2011). Accordingly, mainly due to the lack of skull and metatarsus remains, as well as of most of the pedal phalanges, it is more difficult to infer locomotor and predatory habits of U. comahuensis and U. paynemili.

| CON CLUS ION
In conclusion, morphological differences between the hindlimbs of unenlagiines and eudromaeosaurs reflect differences both in locomotor and in predatory habits. In unenlagiines the presence of a long tibia and of a long, slender and subarctometatarsalian metatarsus suggests greater cursorial capacities with respect to eudromaeosaurs. Conversely, eudromaeosaurs have a shorter, wider and non-subarctometatarsalian metatarsus. The two groups of dromaeosaurids have similar length proportions of the pedal digits, and share the elongation of the distal pedal phalanges, the latter feature probably allowing them the ability to grasp. However, certain morphological traits of eudromaeosaurs, such as a more robust metatarsus, markedly ginglymoid distal articular surfaces of metatarsals I, II and III, as well as interphalangeal articular surfaces, and a shorter phalanx II-1, indicate that these dromaeosaurids possibly exerted higher grip strength than unenlagiines. By contrast, foot proportions and slenderness of unenlagiines would not have allowed them to perform high-force grasping. Instead, unenlagiines may have been able to make faster movements with both the metatarsus and the digit II. Comparable morphofunctional difference is analogously observed in extant raptorial birds. Those taxa with the shortest metatarsi, such as owls, have the ability to produce the greatest grip force. By contrast, taxa with longer metatarsi, such as polyborine falconiforms, generate a lesser grip force but can effect faster movements with the pes.
The morphological differences between the pedal phalanges of unenlagiines and eudromaeosaurs are not as drastic as those observed between their metatarsi. This fact, together with the similar length proportions of the pedal phalanges, seems to indicate that the morphology of these pedal elements varied to a small extent along dromaeosaurid evolution. Since the digits were the main parts of the autopodium involved in the predatory function it is probable that the predatory habit exerted a greater selective pressure on the morphology of the phalanges.
Buitreraptor gonzalezorum, with its small size, high cursorial capacities, a long metatarsus and phalanx II-1, more mobile phalanges, and tiny teeth, was probably a terrestrial predator that preyed on small elusive animals, such as arthropods, lizards and mammals, using rapid gripping movements of its pes. Rahonavis ostromi was also a small-sized unenlagiine, although its morphology seems to indicate that it had less well-developed cursorial abilities. On the other hand, its small body size and potential climbing and aerial locomotion capacities could be related to an arboreal habit. Other unenlagiines, such as the large-sized Austroraptor cabazai and the medium-sized Neuquenraptor argentinus probably preyed on larger animals, also making use of their well-developed cursorial faculties. U. comahuensis and U. paynemili are more fragmentary and so it is more difficult to infer a locomotor and predatory habit for them.
During dromaeosaurid evolution, the different lineages seem to have diverged acquiring varied lifestyles, as documented by unenlagiines, microraptorines, eudromaeosaurs, and recently by halszkaraptorines (Cau et al. 2017, although also see Brownstein, 2019). Future studies, such as reconstructions of the muscular system, will be necessary to analyse the hindlimb as an osteo-muscular integrated complex, and the ways it would have been involved both in locomotion and predation in dromaeosaurids. These paleobiological aspects will contribute to a better comprehension of the dromaeosaurid evolutionary story, as well as about the role of these theropods within the ecosystems in which they lived.

INS TITUTIONAL AB B RE VIATIONS
Below is a list of the institutional abbreviations used in the present paper.