Vertical support use and primate origins

Adaptive scenarios of crown primate origins remain contentious due to uncertain order of acquisition and functional significance of the clade’s diagnostic traits. A feature of the talus bone in the ankle, known as the posterior trochlear shelf (PTS), is well-regarded as a derived crown primate trait, but its adaptive significance has been obscured by poorly understood function. Here we propose a novel biomechanical function for the PTS and model the talus as a cam mechanism. By surveying a large sample of primates and their closest relatives, we demonstrate that the PTS is most strongly developed in extant taxa that habitually grasp vertical supports with strongly dorsiflexed feet. Tali of the earliest fossils likely to represent crown primates exhibit more strongly developed PTS cam mechanisms than extant primates. As a cam, the PTS may increase grasping efficiency in dorsiflexed foot postures by increasing the path length of the flexor fibularis tendon, and thus improve the muscle’s ability to maintain flexed digits without increasing energetic demands. Comparisons are made to other passive digital flexion mechanisms suggested to exist in other vertebrates. These results provide robust anatomical evidence that the habitual vertical support use exerted a strong selective pressure during crown primate origins.

. Species mean PTS measurements in extant taxa …………………………… 42  (2), which provided the topology for all living and extinct crown strepsirrhines, we chose to use the allcompat tree derived from analysis of their full morphological dataset with the fossilized birth-death prior on branch lengths. For the strepsirrhine nodes that overlap with those in the tree derived from analysis of the Gunnell et al. (1) matrix, those in the latter tree are, on average, 77.8% the age of those in the Herrera and Dávalos tree, consistent with Herrera and Dávalos' (recovery of a much more ancient "euprimate" node (68.04 Ma versus 60.11 Ma). To maintain consistency with the Gunnell et al. tree, we reduced the depth of all of the lemuriform branches in Herrera and Dávalos' tree by multiplying them by 0.778.
For our Bayesian tip-dating analysis of the platyrrhine matrix of Kay (3), we first converted polymorphisms into their own discrete states and characters with <6 states were ordered if they were ordered in Kay's analyses (characters with more states cannot be ordered due to software limitations). The value for the clockratepr parameter was derived from an R script written by S. Heritage, which uses the pathlengths from the allcompat tree output by a non-clock Bayesian analysis, and the ages of living and extinct taxa, to calculate a mean and standard deviation for the rate of character evolution (see Gunnell et al. [1] for additional details). The non-clock analysis that was run for this purpose imposed partial constraints that enforced a "molecular scaffold"; this analysis was run for 50 million generations, and the first 25% of sampled trees were discarded as burn-in. The resulting allcompat consensus was used to determine the values for the clockratepr parameter. In order to keep the platyrrhine divergences broadly consistent with Gunnell et al. (1) (which employed the molecular supermatrix of Springer et al. [6]), divergences within crown Atelinae, crown Callithricidae, crown Pitheciinae, and crown Catarrhini were fixed to fit the means of the four different divergence estimates for each node provided by Springer et al. (6), while all other nodes were free to vary. These nodes were selected because no fossil species in the Kay matrix (3) have been placed within these clades in recent parsimony or Bayesian phylogenetic analyses of morphological data. In addition, a truncated normal calibration was used for the age of the entire platyrrhine clade, based on the age of Branisella (27.2 Ma), the oldest platyrrhine in the Kay matrix (3).
We also ran a Bayesian tip-dating analysis of Gunnell's matrix (4) of Eocene notharctines. We assigned polymorphisms in the matrix intermediate states, maintained the same ordering scheme that was used by Gunnell (4), and ran a non-clock Bayesian analysis to obtain values for the clockratepr parameter (20 million generations, first 25% of sampled trees discarded as burn-in). The runs from the subsequent Bayesian tip-dating analysis of the matrix failed to converge, however, and did not produce a resolved allcompat tree, so we used the topology from the nonclock Bayesian analysis and used the '1-Ma rule' to assign divergence dates.
Other modifications to the resulting supertree include: 1) the divergence between Ptilocercus and Tupaia was based on the morphological clock-based divergence of Ptilocercus kylin from Tupaia (48.24 Ma); 2) Ignacius was placed into a polytomy with non-purgatoriid plesiadapiforms; 3) Plesiadapis rex and Plesiadapis cookei were grafted onto the Plesiadapis branch following Yapuncich et al. (7); 4) Teilhardina brandti was grafted mid-way along the branch between Teilhardina belgica and Steinius, following previous papers (7-9); 5) the omomyiform Ourayia was grafted onto the tree as the sister taxon of Hemiacodon following Tornow (10); 6) the species of Eosimias in the Gunnell et al. matrix (1) is Eosimias centennicus, but the tali measured for this study are from older E. sinensis, so the tip of the terminal branch for Eosimias was placed at 45 Ma (11); 7) the hominoid-cercopithecoid divergence, placed at 21.7 Ma in the Gunnell et al. matrix (1), was placed at 25.2 Ma following Stevens et al. (12), and all internal branches separating extant crown catarrhines were lengthened by 1.162% to reflect this change; 8) Australopithecus afarensis, which was not sampled in any of the morphological matrices that were used to create the composite supertree, was given a 1-Ma terminal branch, placing it along the hominin stem lineage at 4.3 Ma; the same arrangement was used for the 1.7 Ma East Turkana hominins sampled here, which joined the hominin stem lineage at 2.7 Ma.
The resulting supertree is presented below as Tree S2. Because this tree recovers polyphyly among Adapiformes, we wanted to evaluate alternative topology that maintained monophyly of this group. To construct this alternative tree (Tree S3), we added Donrussellia provincialis to the tree used in Yapuncich et al. (7) following protocols detailed in that paper.

Supplementary Results
Comparative plates showing variation of PTS index among extant and extinct taxa are shown in Figs. S2-S4.
Scaling relationships of posterior trochlear shelf and body size. There were no significant relationships observed between body mass and the PTS index and significant autocorrelation was detected within all groups except lemuriforms (Table S3). Among the components of the PTS index, significant positive allometry was recovered between lnRadius and lnBM, which indicates that the base circle of the cam (i.e., the curvature of the lateral tibial facet) increases more quickly than body mass increases (Table S4). Essentially, the joint surface becomes flatter as body mass increases; this finding is consistent with previous assessments of allometric scaling of articular surfaces in primates (13,14).
When trochlear width is used as a body size proxy, the PTS index exhibits a significant negative relationship, indicating that the cam effect decreases as body size increases. This finding is consistent with observed postural differences among euarchontans: larger-bodied taxa are more likely to employ above-branch quadrupedal or suspensory behaviors (15,16), rather than postures that require habitually dorsiflexed feet. There were also significant relationships recovered between each of the component distances and trochlear width, but the slope of these regressions did not deviate from the expectations of isometry (Table S5).
Correlations between PTS cam size and other talar metrics. All regressions exhibited significant phylogenetic autocorrelation, as detected by Pagel's lambda. Only FFG Ellipse showed a significant (positive) correlation with the PTS index (Table S6). FFG Ellipse models the groove for the flexor fibularis as an ellipse by calculating the ratio of the semi-major (mediolateral) axis to the semi-minor (anteroposterior) axis (7). The significant positive correlation between these two variables indicates that the groove becomes mediolaterally wider and anteroposteriorly shallower as the PTS is more strongly developed. Given that the measurement protocol used to quantify the development of the PTS involved placing a landmark at the saddle point of the groove for the flexor fibularis, the significant positive correlation observed between the PTS index and FFG Ellipse is highly intuitive.
ANOVA and one-way t-tests. The average intraspecific range of the PTS index (0.17) is 14% of range observed across the entire sample (1.20). As assessed by ANOVAs and post hoc comparisons, there are significant differences between several groups (Table S7) at the level of species means. Among extant strepsirrhines, lorisids exhibit PTS indices that are significantly lower than other strepsirrhine groups (Figs. S5-S6, Table S7). Additionally, indriids exhibit PTS indices that are significantly higher than galagids. There is broad overlap among anthropoid groups, largely as a result of high variation within hominoids (Pongo as a very low PTS index, while Homo exhibits a high PTS index). Our sample of cercopithecoids, which exhibit high PTS indices, are the only examined anthropoid group that is statistically differentiated (from hominoids, atelids, and callitrichines) (Figs. S5-S6, Table S7).
In comparisons of all major euarchontan groups (Figs. S5-S6, Table S7), we consolidated all strepsirrhine and anthropoid OTUs that were not significantly different from all other OTUs in the initial comparisons. Strepsirrhines were represented by lorisids and non-lorisids, while all anthropoid groups were combined. In the euarchontan-wide analysis, the mean PTS index of non-lorisid strepsirrhines was significantly higher than that of all other groups except tarsiers. The mean PTS index of lorisids was significantly lower than that of all other groups except nonprimates. Finally, post hoc differences reveal significant differences between tarsiers and nonprimates.
Results of the one-way t-tests are shown in Table S8. The mean PTS indices for all strepsirrhine clades except lorisids are significantly greater than 1, indicating that the distance from the joint axis to the FHL groove is greater than the circle ascribed by the curvature of the lateral tibial facet (i.e., the PTS will function as a cam during dorsiflexion). The mean PTS index of lorisids is significantly less than 1, indicating that the distance from the joint axis to the FHL groove is less than the circle ascribed by the curvature of the lateral tibial facet (i.e., no cam effect). Tarsiers also exhibit a mean PTS index that is significantly greater than 1. Among anthropoids, atelids and callitrichines have mean PTS indices that are significantly less than 1, but there are no significant differences among any other anthropoids.
Principal component analyses of talar metrics. We used principal components analysis (PCA) to visualize the morphospace created within Euarchonta by the PTS index and several previously published metrics of talar morphology. Analyzed metrics include PTS index, fibular facet angle (FFA) (8), the first principal component of three metrics of medial tibial facet morphology (MTF PC1) (9), and the natural logs of the position (FFG Position) and depth (FFG Ellipse) of the flexor fibular groove (7). PCAs were conducted in PAST (17) with species mean values for 123 taxa; values can be found in Data S2. The ancestral crown primate is also represented in these analyses using the ASR values of the talar variables reported by Boyer and Seiffert (8), Boyer et al. (9), and Yapuncich et al. (7). Two PCAs were generated using the correlation matrices: the first PCA does not include FFG Ellipse, while the second does.
The four-variable PCA is shown in Fig. S7 and eigenvalues and explained variance are reported in Table S9. Together, the first two principal components explain 78.5% of the total variance. Loadings of talar metrics onto each principal component are reported in In both PCA plots, taxonomic groups are well differentiated from each other, continuing a trend of increasing discriminatory power as euarchontan talar morphology has been placed into a more comprehensive and quantitative framework (7)(8)(9).
Two important points are revealed by these principal component analyses. First, these talar metrics partition euarchontans in the manner expected from previous discussion of these features (18)(19)(20). The first principal components of both PCAs are strongly correlated with three talar metrics (fibular facet angle, medial tibial facet morphology, position of the flexor fibularis groove) and neatly divide strepsirrhine and haplorhine primates. Thus, these PCAs confirm the initial description by Gebo (20) that these three features contribute to the primary axis of talar morphological variation separating primate suborders. This is particularly evident in the loadings of the first PCA (Table S12). Subsequent research has used these features to determine the taxonomic affinities of enigmatic fossil crown primates including eosimiids (21,22) and amphipithecids (23,24). While the three features with the strongest loadings on the first PCs do not necessarily discriminate crown primates from other euarchontans, taxonomic separation along the second principal components suggest that development of the posterior trochlear shelf is useful for distinguishing crown primates from other euarchontans. Again, these results confirm early descriptions of the posterior trochlear shelf as a crown primate feature (25).
Second, the PCAs reveal the strong morphological similarities in the tali of the earliest crown primates. In both analyses, omomyiforms have positive loadings on PC1 (representing a more sloping fibular facet, an expanded medial tibial facet, and a more laterally positioned flexor fibularis groove) and strong negative loadings on PC2 (representing a strongly developed PTS cam) and plot closer to notharctids than tarsiers within these morphospaces. Though omomyiforms and notharctids are thought to represent the initial radiations of haplorhines and strepsirrhines respectively, they are not as distinctly partitioned as extant taxa by the set of talar features proposed by Gebo (20) to distinguish these groups. Indeed, reduced discriminatory ability should be expected close to the base of the order. This expectation has provided justification for several recent studies of talar morphology (7)(8)(9) to quantify these talar features in comparative samples that include other euarchontans. Fossil evidence suggests that the Haplorhini-Strepsirrhini suborder split occurred very closely in time with the origin of crown primates, so extending the framework established by Gebo (20) has the potential to reveal morphological changes in the ankle associated with crown primate origins.
These PCAs also confirm previous analyses or clarify outstanding issues for other fossil taxa. Dermopterans and plesiadapiforms (except Carpolestes) are well separated from crown primates along the second principal component in both analyses. Lorisids and the subfossil lemurs Palaeopropithecus, Babakotia, and Megaladapis exhibit PC2 values like those of non-crown primate taxa; this morphological similarity is also evidenced in the SURFACE analyses below. Tupaia and Carpolestes both exhibit low PC1 values and modest PC2 values, similar to the nailbearing callitrichines. This similarity is most evident in Fig. S8 and is detected in the SURFACE analyses below. Ptilocercus, the pen-tailed tree shrew, has been suggested as a model for the positional behaviors of the ancestral crown primate (26)(27)(28)(29); in both PCAs, Ptilocercus is more similar to the earliest crown primates than any other non-crown primate taxon.
When Yapuncich et al. (7) performed a similar PCA analysis without the PTS index, ASR values for ancestral crown primate plotted within the polygon containing other euarchontans (their Figure 12). In the talar metric PCA of Boyer et al. (30), Donrussellia, the earliest known adapiform, plotted in very similar position within the non-crown primate euarchontan polygon (their Figure 5a). Here, when the PTS index is included, both Donrussellia and the ancestral crown primate plot much more closely to other crown primates (and fall within the lemuriform polygon in Fig. S8). Compared to earlier analyses, the position of Eosimias, the earliest known anthropoid (21,22), also changes substantially when the PTS index is included. In the PCA of Yapuncich et al. (7), Eosimias plotted within the non-crown primate euarchontan polygon; here, the taxon occupies a much more central position in both PCAs. These results reinforce the uniqueness of the PTS cam mechanism among crown primates, as the tali of all three taxa (the ancestral crown primate, Donrussellia, and Eosimias) become more crown primate-like when the PTS index is included.
Among other crown primate fossil taxa, the talar morphology captured by these metrics suggest that later occurring adapiforms (e.g., adapines and caenopithecines) are more similar to slowclimbing lorisids, confirming previous assessments of the positional behaviors of these taxa (31)(32)(33). Asiadapines, adapiforms from the early Eocene of India, are well-represented by postcrania and have been interpreted as generalized arboreal quadrupeds (34,35). However, Boyer et al. (36) and Yapuncich et al. (7) noted some similarities in the talus to slow-climbing taxa such as lorisids. Here, particularly in Fig. S8, the asiadapines Asiadapis and Marcgodinotius are more similar to extant lemuriforms than lorisids (adapines plot between asiadapines and lorisids). This suggests that asiadapine tali are not particularly specialized for slow-climbing (at least in a manner homologous to extant lorisids). The results from the SURFACE analyses below complicate this interpretation but underscore that asiadapines exhibit a mosaic of postcranial adaptations and remain an important group for further study.

Ancestral state reconstruction.
Ancestral states values are summarized in Table S13 for the  delta model and Table S14 for the kappa model. Estimated marginal likelihoods for all models are presented in Table S15, and contrasts between values reconstructed with Tree S2 and Tree S3 with delta model are presented in Table S16. Text versions of all phylogenetic trees used in this study are provided on the following pages. Node numbers for Tree S2 are shown in Fig. S9.
PTS hypertrophy within the hominin lineage. The estimated node values of the PTS index increase substantially within the hominin lineage ( Figure 4; Table S13; It is possible that PTS hypertrophy has some functional advantages for bipedalism as well. Namely, the PTS cam mechanism could increase tension in the flexor fibularis tendon to concentrically flex (and thus shorten) the toes. Shorter effective digits would be better able to resist high bending moments induced by ground reaction forces during bipedal locomotion (37,38). Electromyography of the flexor fibularis shows the muscle is most active during the middle of the support phase as the foot transitions from heel strike to toe-off (39). Furthermore, the lengths of the pedal intermediate and proximal phalanges are reduced in humans and fossil hominins relative to those of chimpanzees and gorillas (40), which may indicate that resisting bending moments was a strong adaptive pressure during the evolution of obligate bipedalism. It is notable that digits 4 and 5 are particularly reduced in humans (41), as the flexor fibularis (=flexor hallucis longus in human anatomy) typically does not insert on the lateral digits (42,43), and thus could not be recruited to resist bending moments.
Tree S1. Phylogenetic tree of extant taxa used for body mass PGLS regressions (      Table S17. The range of adaptive regimes for select clades are shown in Fig. S10. Adaptive regimes are detailed in Tables S18-21 and plotted on the phylogenies in Figs. S11-S14.
Version 1 -all taxa included: Adaptive optimum and convergent clades are detailed in Table  S18 and adaptive regimes are shown for both topologies in Fig. S11.
With the Gunnell et al. (1) Table S21 and adaptive regimes are shown in Fig. S14. Portions of these results are presented in Fig. 2a but are also presented here to facilitate comparison.
When considering extant taxa, there are no differences between the Gunnell et al. (1) and "traditional" topologies. There are four adaptive regimes identified with SURFACE. The ancestral node, scandentians, and dermopterans share an adaptive optimum of 0.74. A regime shift to a higher adaptive optimum (1.59) occurs at the ancestral primate node, which is maintained across most extant primates. Pongo has a unique (though problematic) adaptive regime (Ɵ = -1.58), while lorisids and Callithrix converge on an optimum of -0.08.

Summary of SURFACE analyses:
In the main text, we presented results from the SURFACE analysis of extant taxa for simplicity (Fig. 2a). The regime shifts identified in the extant only analysis align very well with the ancestral state reconstruction (ASR) results (Fig. 3, Table S13), as the SURFACE analysis recovers a substantial increase in the PTS index adaptive optimum at the base of Primates (Fig. S14, Table S21). However, the inclusion of fossil taxa in SURFACE analyses adds complexity to the evolution of the feature within Euarchonta. When modeling both extant and extinct taxa, two factors suggest that Version 3 with the "traditional" topology should be the preferred model. First, this version does not estimate any optima that are outside the PTS values observed in our sample (Table S1-S2, Table S20), so all regimes seem reasonably estimated. Second, this version has the lowest AICc of all examined versions, indicating that it is the best-fit model (Table S17) (although we note that sample sizes are not equivalent across all versions).
The key difference between ASR and SURFACE results from Version 3 concerns the direction of change deep in the tree. ASR results estimate that the PTS index at the ancestral euarchontan node was quite low and subsequently increased (Fig. 3, Table S13). In contrast, the best-fit SURFACE model recovers a high adaptive optimum at the base of Euarchonta (1.61), with a shift to a reduced optimum at the Primatomorpha node (1.11). Optima are reduced independently in the lineages of Scandentia (0.78) and Dermoptera (0.61). Thus, the best-fit SURFACE model suggests the selective pressure to develop a pronounced posterior trochlear shelf decreased from the ancestral euarchontan node to the Primatomorpha node. Given the absence of PTS development in any other mammalian group, we do not consider this scenario likely. Future analyses should include non-euarchontan outgroups to address issue.
Overall, SURFACE analyses detect regime shifts that decrease adaptive optima more frequently than shifts that increase optima, which may be due to widespread homoplasy of reduced PTS indices across Euarchonta. In all analyses, dermopterans, plesiadapiforms, lorisids, adapines, and caenopithecines exhibit (and often converge toward) adaptive optima substantially less than 1. Several taxa within this set have been compared favorably with one another by previous authors. Beard (44)(45)(46) and Bloch and Boyer (47) suggest some similarities in the postural behaviors of claw-bearing plesiadapiforms and dermopterans, though these similarities do not necessarily extend to locomotor behaviors such as mitten-gliding (29,(48)(49)(50)(51)(52). Based on comparisons of multiple postcranial elements, Dagosto (31,53) and Gebo (54) have favorably compared the postural behaviors of lorisids and adapines. Similarities in tarsal morphology led Boyer et al. (33) to suggest that the caenopithecine Afradapis had a locomotor profile like those of extant lorises. Seiffert et al. (32) found phenetic similarities between the caenopithecine Caenopithecus and extant lorisids for both the talus and calcaneus.
In every analysis, palaeopropithecids and Pongo converge toward a very low (occasionally negative) optimum for the PTS index. Morphological similarities between these taxa have long been recognized, ranging from strongly curved proximal phalanges (55), lumbar vertebrae morphology (56), and long bone cross-sectional dimensions (57). SURFACE analyses conducted here lend further support to the hypothesis that palaeopropithecids were large-bodied suspensory taxa, with an overall positional behavior similar to Pongo.
Across all analyses, few regimes exhibit increased adaptive optima (Fig. S10). Among extant taxa, lemuriforms (except Daubentonia) consistently exhibit increased optima ranging from 1.24 to 1.43, while galagids often (but not always) have optima greater than 1 (0.93 to 1.2). Notharctids and omomyiforms share adaptive optima that are generally greater than 1 (0.93 to 1.43), similar to those of galagids. The only other taxa that consistently have optima greater than 1 are the adapiforms Anchomomys and Djebelemur (0.93 to 1.35).
Comparisons of optima estimated across SURFACE analyses reveal that the differences in the included sample and topology do not have a strong impact on the estimated adaptive optima for most groups (Fig. S10). Notable exceptions include asiadapines, which exhibit the most variation in their estimated adaptive optima, and plesiadapiforms, which were excluded from Version 3 in order to resolve the polytomy at the base of Euarchonta. It is likely that short branch lengths and shifting phylogenetic position explain the high variance in adaptive optima for both taxa.
Based on analyses of multiple postcranial elements, asiadapines have been reconstructed as generalized arboreal quadrupeds (34,35), though some tarsal features are similar to those observed in slow-climbing taxa (7,8). The four specimens examined for this study have modest PTS indices (ranging from 0.91 to 1.25; Table S2; Fig. S6) but the clade has adaptive optima that range from -0.39 to 2.30. It is possible that short branch lengths and shifting phylogenetic position may explain the variability in estimated adaptive optima for asiadapines. In the Gunnell et al. (1) topology, asiadapines are the most basal strepsirrhines, a position which is occupied by notharctids in the "traditional" topology. It is worth noting that this region of the phylogeny also generates large contrasts when comparing estimated node values of the two versions of ancestral state reconstruction (Table S16). Using the Gunnell et al. (1) topology, SURFACE tends to estimate multiple regimes within the strepsirrhine lineage, and asiadapines are one of several regimes with high Ɵ values. In contrast, when asiadapines are bracketed by the taxa with high PTS indices (notharctids and Djebelemur + Anchomomys) in the "traditional" topology, their modest PTS indices necessitate very low optima (Versions 1 and 2). In the best-fit model (Version 3 with the "traditional" topology), asiadapines share a low adaptive optimum with palaeopropithecids and Pongo, suggesting strong selective pressures to reduce the PTS index with the group. Given the observed PTS index values for the group, asiadapines are quite distant from their adaptive optimum. Overall, the estimated Ɵ values of Version 3 are consistent with interpretations of asiadapines as generalized quadrupeds (34,35), while remaining concordant with studies that have noted tarsal features that are shared with slow-climbing taxa (7,8).

Fig. S1.
Schematic of cam mechanism (a) and measurement protocol (b) to quantify development of the posterior trochlear shelf in two primate taxa. 1) The articular surface of the lateral tibial facet is highlighted in Geomagic Studio. 2) As an approximation of the base circle of the cam, a best-fit cylinder is fit to the selected area.
3) The radius of the best-fit cylinder is calculated. 4) A landmark is placed at the saddle point of the groove for the tendon of the flexor fibularis muscle. The groove to cylinder distance is measured with the "measure distance to feature" function of Geomagic Studio. PTS index is calculated as the ratio of the distance between the groove and the axis of the base circle (=Radius + Groove to Cylinder) relative to the radius of the base circle (=Radius).

Fig. S2. (previous page)
Comparative plates of posterior trochlear shelf morphology for scandentians, dermopterans, plesiadapiforms, tarsiiforms, and notharctids. These plates provide a guideline for development of PTS; actual measurements were taken following protocols detailed in the methods section and shown in Fig. S1. Circle approximates curvature of the lateral tibial facet (i.e., the base circle of the cam); dotted lines indicate anterior and posterior extent of the base circle. Ellipses approximate tendon of the flexor fibularis muscle and are color-coded by species mean PTS index. Blue: PTS index < 0.75; light blue: 0.75-1.00; light red: 1.00-1.25; red: >1.25. Views for each specimen are medial (left) and dorsal (right). * indicates chirality has been reversed for consistency. Scale bars equal 3 mm.      Table S7). Abbreviations for taxonomic groups are defined in the caption of Fig. S5.     Adaptive regimes identified with SURFACE using two different topologies. All taxa included. Nodes with regime shifts are numbered and detailed in Table S18. Branch colors indicate regime shifts (red: Ɵ > 1.0, blue Ɵ < 1.0); color intensity indicates rank order of Ɵ values (darker colors indicate more extreme optima).

Fig. S12.
Adaptive regimes identified with SURFACE with two different topologies. Those taxa with taxon-specific regimes in Fig. S11 have been removed. Nodes with regime shifts are numbered and detailed in Table S19. Branch colors indicate regime shifts (red: Ɵ > 1.0, blue Ɵ < 1.0); color intensity indicates rank order of Ɵ values (darker colors indicate more extreme optima).

Fig. S13.
Adaptive regimes identified with SURFACE with two different topologies. Plesiadapiforms and taxa with taxon-specific regimes in Fig. S11 have been removed. Nodes with regime shifts are numbered and detailed in Table S20. Branch colors indicate regime shifts (red: Ɵ > 1.0, blue Ɵ < 1.0); color intensity indicates rank order of Ɵ values (darker colors indicate more extreme optima).   Table S3.
Clade-specific phylogenetic generalized least squares (PGLS) and ordinary least squares (OLS) regressions between PTS index and ln(body mass). Regressions conducted with Tree S1.