O R I G I NA L A RT I C L E

doi:10.1111/evo.13815

Head-turning morphologies: Evolution of

shape diversity in the mammalian
atlas–axis complex
Abby Vander Linden,1 Kristin M. Campbell,2 Erin K. Bryar,2 and Sharlene E. Santana2,3,4
1
Graduate Program in Organismic and Evolutionary Biology, University of Massachusetts Amherst, Amherst,
Massachusetts
2
Department of Biology, University of Washington, Seattle, Washington
3
Burke Museum of Natural History and Culture, Seattle, Washington
4
E-mail: ssantana@uw.edu

Received February 13, 2019

Accepted July 1, 2019

Mammals flex, extend, and rotate their spines as they perform behaviors critical for survival, such as foraging, consuming prey,
locomoting, and interacting with conspecifics or predators. The atlas–axis complex is a mammalian innovation that allows precise
head movements during these behaviors. Although morphological variation in other vertebral regions has been linked to ecological
differences in mammals, less is known about morphological specialization in the cervical vertebrae, which are developmentally
constrained in number but highly variable in size and shape. Here, we present the first phylogenetic comparative study of the
atlas–axis complex across mammals. We used spherical harmonics to quantify 3D shape variation of the atlas and axis across a
diverse sample of species, and performed phylogenetic analyses to investigate if vertebral shape is associated with body size,
locomotion, and diet. We found that differences in atlas and axis shape are partly explained by phylogeny, and that mammalian
subclades differ in morphological disparity. Atlas and axis shape diversity is associated with differences in body size and locomotion;
large terrestrial mammals have craniocaudally elongated vertebrae, whereas smaller mammals and aquatic mammals have more
compressed vertebrae. These results provide a foundation for investigating functional hypotheses underlying the evolution of
neck morphologies across mammals.

KEY WORDS: Cervical vertebrae, mammals, morphology, locomotion, spherical harmonics analysis.

Clade-wide evolutionary analyses of anatomical structures have size in mammals has also created structural and mechanical con-
proven fruitful in understanding the ecological pressures that straints that influence the morphospace within which the mam-
shape morphological diversity (e.g., Felice and Goswami 2018; malian skeleton can evolve (Gould 1966; Jungers 1984; Chris-
Slater and Friscia 2019). The radiation of mammals and its accom- tiansen 2002; Cardini and Polly 2013; Arnold et al. 2017).
panying morphological and ecological diversity provides a rich Relative to the skull and limbs, the macroevolution of the
opportunity for exploring macroevolutionary patterns in form and spinal skeleton has been greatly understudied in mammals; most
function and their potential adaptive significance. For example, comparative work has focused on the morphology of thoracic and
the morphology of mammal skulls, teeth, jaws, and limbs have lumbar vertebrae only within individual families (e.g., Johnson
been shaped by demands for efficient food processing, locomo- and Shapiro 1998; Shapiro and Simons 2002; Pierce et al. 2011;
tion, and combat (e.g., Coombs 1983; Van Valkenburgh 1985; Granatosky et al. 2014). The evolution of functional and morpho-
Popowics and Fortelius 1997; Pérez-Barberı́a and Gordon 1999; logical diversity in the mammalian cervical spine generally, and
Caro et al. 2003; Clauss et al. 2008; Polly 2008; Santana et al. the atlas–axis complex in particular, has received less attention,
2010; Goswami et al. 2011). Conversely, the evolution of body even though the atlas–axis complex is a major skeletal innovation


C 2019 The Author(s). Evolution 
C 2019 The Society for the Study of Evolution.

2060 Evolution 73-10: 2060–2071

 AT L A S - A X I S S H A P E E VO L U T I O N

in mammalian evolution (Evans 1939). These first two cervical in body size, diet, and ecology across Mammalia. First, we hy-
vertebrae represent a simplification of the vertebral morpholo- pothesize that the shape of these two vertebrae exhibits an allomet-
gies found in nonmammalian synapsids (Sidor 2001), and allow ric relationship with body mass, such that larger-bodied mammals
for a greater variety and range of head movements and postures have relatively more robust vertebrae that would support increased
than those available to any mammalian ancestor (Evans 1939; skull weight while providing larger attachment surfaces for larger
Kemp 1969; Crompton and Jenkins 1973). The articulation of the neck muscles (Manfreda et al. 2006). Second, we hypothesize
mammalian atlas with two occipital condyles in the skull allows that behaviors associated with consuming animal prey (killing,
flexion and extension of the head with less strain of the spinal shaking, tearing, etc.) as opposed to plant material (browsing
cord than when a single occipital condyle is present, as in rep- and grazing) impose different mechanical demands on the neck
tiles (Crompton and Jenkins 1973). Although the atlanto-occipital of carnivores versus herbivores and omnivores, resulting in the
joint enables flexion and extension, the peg-like dens of the axis evolution of different atlas and axis morphologies across species
acts as a pivot about which the atlas and head rotate laterally with different diets. We expect to find shorter, more compressed
(Evans 1939). As a consequence, the atlas–axis complex allows vertebrae with larger transverse and spinous processes in carni-
mammals to decouple these motions, and to more precisely per- vores, which would allow for a greater range of neck movements
form behaviors critical for survival, such as subduing, killing, and and provide larger attachment areas for powerful neck muscles
biting prey (Van Valkenburgh and Ruff 1987; Ydesen et al. 2014); (Pierce et al. 2011). Third, we hypothesize that mechanical de-
elevating and depressing the head for grazing, gnawing, or dig- mands of different locomotor modes require different degrees of
ging (Pellis and Officer 1987; Du Toit 1990; Reichman and Smith head support, neck rotation, and flexibility. Therefore, we expect
1990; Lessa et al. 2008; Zsoldos and Licka 2015); and combating to find shorter vertebrae with more surface for rotation of the axis
conspecifics (Kitchener 1988; Stankowich 2012). Consistent with and flexion/extension of the head in climbing/gliding taxa when
their functional roles, the atlas and axis appear to exhibit broad compared to terrestrial taxa. As these three hypotheses are not mu-
morphological diversity across mammals. tually exclusive, we explore whether models with single factors
Little is known about the factors underlying the morpho- or combinations of body size, diet, and locomotion better explain
logical diversity and evolutionary patterns of the atlas and axis the observed diversity in atlas and axis morphology. Alternatively,
vertebrae across extant mammals; previous studies have focused evolutionary history might explain most of the diversity in atlas
on examining morphological transitions in extinct taxa (Jenkins and axis shape (Randau et al. 2016a). We use phylogenetic com-
1960; Kemp 1969; Kikuchi et al. 2012), or on comparing atlas parative methods and spherical harmonics analyses to quantify
and/or axis morphologies within specialized groups of mammals and compare the three-dimensional shape of vertebrae, and test
(e.g., primates, tree shrews, felids; Sargis 2001; Manfreda et al. whether evolutionary changes in their morphology are associated
2006; Randau et al. 2016a, 2016b; Nalley and Grider-Potter 2017). with differences in body size, diet, and locomotion across the
Comparative studies of atlas shape have found contrasting trends mammal phylogeny.
across the mammal groups studied to date. Atlas shape has been
linked to changes in body size and the evolution of bipedal pos-
ture in primates (Manfreda et al. 2006), but a study in felids Materials and Methods
found that atlas shape was related to body size but not to diet SAMPLING AND ECOLOGICAL DATA
or locomotor ecology (Randau et al. 2016a, 2016b). Within Eu- All specimens included in this study were obtained from the Burke
archontogliran mammals, including primates and rodents, atlas Museum of Natural History and Culture (University of Wash-
shape is associated with body size, but not with head size or ington, Seattle, WA; Table S1). We sampled one atlas and one
locomotion (Vander Linden et al. 2019). Furthermore, because axis vertebra from 80 species of mammals, with each species
almost all mammals, from giraffes to humans, possess seven cer- belonging to a different family across all orders. This sample
vical vertebrae (Narita and Kuratani 2005), it has been proposed included representative species from families within the orders
that the extensive variation in neck length is mainly due to scal- Monotremata and Marsupialia and the superoders Laurasiathe-
ing and shape changes within individual vertebrae (Arnold et al. ria, Euarchontoglires, Afrotheria, and Xenarthra (Meredith et al.
2017). However, no quantitative studies have tested this idea nor 2011). To minimize any effect of sexual dimorphism, we used
the relative importance of different behaviors (e.g., feeding or adult female specimens except in a few cases where only adult
locomotion) on the evolution of atlas and axis diversity across males were available. We incorporated phylogenetic relation-
mammals. ships and divergence times between taxa using a time-calibrated
Given the functional and evolutionary importance of the family-level molecular tree of Mammalia published by Mered-
mammalian atlas–axis complex, we aim to investigate its mor- ith et al. (2011) and pruned to our study species (Fig. S1 and
phological diversity and evolution within the context of variation Table S2).

EVOLUTION OCTOBER 2019 2061

A . V. L I N D E N E T A L .

We sourced body mass data for each species from the Pan- microCT), and automatically segmented pixels representing bone
THERIA dataset (Jones et al. 2009) using the Wilson and Reeder in Mimics 17.0 (Materialise, Belgium). We then rendered these
(2005) taxonomy, and log10 -transformed these data to improve as 3D objects, and exported them as ∗.STL files.
normality for statistical analyses. Using the Kissling et al. (2014) We used Geomagic Studio 14.0 (3D Systems, Inc., Rock Hill,
mammal diet dataset, we assigned each species to one of the CA) to process all STLs prior to spherical harmonics (SPHARM)
following diet categories: carnivore (consuming predominantly analyses. We removed floating polygons, automatically filled
invertebrate or vertebrate animals), herbivore (consuming pre- small holes, and deleted polygons representing internal trabec-
dominantly plant material, including leaves, stems, roots, seeds, ulae in models generated from µCT scans to create a watertight
fruits, nectar, or other plant parts), or omnivore (consuming both surface mesh (Fig. 1A). Because SPHARM requires united ob-
plants and animals). We assigned species to one of three locomotor jects with no holes (Shen et al. 2009), we filled the transverse
categories based on literature accounts (Table S1 and Supporting foramina and vertebral foramina of all vertebrae using Geomagic
Information References): (1) aquatic—fully aquatic species such algorithms that automatically fill holes following the curvature of
as whales and dolphins, as well as species that primarily locomote the surface, thus preserving the shape of the mesh. All meshes
and forage in water such as seals and walruses; (2) ground—any were then automatically resurfaced to improve triangle quality
species with primarily terrestrial, semi-fossorial, or fossorial lo- and reduced to 20,000 triangles (Fig. 1B).
comotion, including semi-aquatic species that also spend a sub-
stantial amount of time foraging and traveling on land, such as SPHERICAL HARMONICS ANALYSES
the capybara; and (3) above ground—species that use scansorial We performed separate SPHARM analyses of the atlas and axis
or primarily arboreal locomotion and species that glide or fly. We datasets using the SPHARM 2.0 (Shen et al. 2009) software in
opted to group fossorial and terrestrial species, as well as arbo- Matlab R2014a (The Mathworks, Inc., Natick, MA). Six land-
real and gliding/volant species, to achieve sufficient sample sizes marks on each vertebra were used to reorient, rescale, and repo-
within groups for meaningful statistical analyses while capturing sition (i.e., register) the models within each dataset (Fig. 1B; see
broad similarities in locomotor mode. Table S3 for landmark descriptions). Following McPeek et al.
Some species included in our study have fused portions of (2008), we resized all meshes to a centroid size of 1, and reg-
the cervical vertebral column, resulting in a syncervical anatomy istered them to a template model using the 3D landmarks. We
(Vanburen and Evans 2016). In species with a fused atlas and axis selected for the SPHARM algorithm to model the vertebrae using
(C1-C2; the whale Berardius bardii and dolphin Lagenorhynchus 15° of smoothing, resulting in an output of 768 spherical harmon-
obliquidens), we opted to treat the syncervical as analogous to the ics coefficients for each model (McPeek et al. 2008) (Fig. 1C).
atlas, considering it a first functional unit of the cervical spine.
In species with fused vertebrae caudal to the atlas (the porcupine STATISTICAL ANALYSES
Erethizon dorsatum C2-C3, kangaroo rat Dipodomys microps C2- To reduce the dimensionality of the spherical harmonics coef-
C3, and armadillo Dasypus novemcinctus C2-C4), we treated the ficients dataset and determine the major axes of morphological
fused vertebrae as analogous to the axis or a second functional variation, we performed a principal component analysis (PCA)
unit. of the spherical harmonics coefficients for each species in both
the atlas and axis datasets in R 3.3.3, using the prcomp function
3D IMAGING in the base R library (R Core Team 2018). Because the spherical
We used both laser scanning and micro-computed tomography harmonics coefficients are complex numbers, the PC scores for
(µCT) scanning to generate 3D models of vertebrae. We scanned each species were also complex numbers. However, the complex
specimens >5 cm in diameter using a NextEngine HD laser scan- portions of the scores were either 0 or very close to 0, so we
ner (NextEngine, Inc., Santa Monica, CA). We scanned each ver- discarded the complex portions of the numbers before using the
tebra in two orientations, resulting in 12 partial scans spanning PC scores in subsequent analyses (following McPeek et al. 2009).
360°. We then aligned and merged the partial scans using the Nex- The first two PCs in both the atlas and axis datasets accounted
tEngine Scan Studio software, and exported ∗.STL (surface) files for over 50% of the total variance, with all other PCs account-
for postprocessing. For specimens too small to be imaged by the ing for less than 7% each (Table S4). To visualize shape change
laser scanner, we used a Skyscan 1174 µCT scanner (Bruker Mi- across the mammalian atlas and axis morphospaces, we used the
croCT, Belgium). We used a 0.25 mm Al filter, a voltage of 50 kV, SPHARM software to generate “eigenshapes,” or spherical har-
and a current of 800 uA for all µCT scans. Voxel sizes ranged from monics models representing the average vertebral shape and the
9 to 30 µm, depending on specimen size. We reconstructed µCT vertebral shapes at ±2 standard deviations from the average along
scans as image stacks (slices) in NRecon version 1.6.9.18 (Bruker each PC axis (McPeek et al. 2009) (Fig. 2).

2062 EVOLUTION OCTOBER 2019

 AT L A S - A X I S S H A P E E VO L U T I O N

ATLAS AXIS
cranial lateral dorsal
cranial lateral dorsal

A Original 3D
mesh

3 1
3
1 1 4 5
B Mesh filled 5 5 2
2 2
4 1
and land-
marked for 5 6
5 1
SPHARM 2
3 6 3 6 2 3 1 6 3 5
4

C New mesh
generated
from
SPHARM
coefficients

Figure 1. Illustration of atlas and axis morphology of an example specimen (Pteropus poliocephalus, the gray-headed flying fox) and
the SPHARM modeling process. (A) A 3D mesh is generated from µCT or laser scan data. (B) The transverse foramina and vertebral
foramina are filled using automatic hole-filling algorithms and the mesh is reduced to exactly 20,000 triangles. 3D coordinates for six
landmarks are recorded for each specimen; numbers refer to landmark descriptions in Table S3. (C) Once all meshes have been prepared
and landmarked, a random mesh is chosen as a template in the SPHARM software and all other models are registered to the template
using the landmarks. The SPHARM algorithm then calculates the spherical harmonics coefficients for each specimen and generates a new
mesh that visually represents those coefficients. The SPHARM coefficients are then used for further analyses.

To estimate the effect of shared evolutionary history, we used lambda to scale the branch lengths of the underlying phyloge-
the phytools package in R (Revell 2012) to calculate Blomberg’s netic variance–covariance matrix in each PGLS regression (Pagel
K statistic for all 80 PCs of the atlas and axis datasets, respec- 1999). Using the MuMIn package in R (Bartoń 2016), we com-
tively. For all phylogenetic analyses, we incorporated tree topol- pared Akaike information criterion (AICc) values corrected for
ogy and branch length information from the Meredith et al. (2011) small sample sizes of models including all possible combinations
phylogeny. We tested for significant differences between the es- of predictor variables. We considered models with AICc > 2
timated K value and 0 (no phylogenetic signal) using 1000 ran- to be best supported by the data (Burnham et al. 2011). Because
dom permutations of the data (Revell 2012). We used the mor- the caper PGLS function requires a univariate response variable,
phol.disparity function in the geomorph R package (Adams and and because each of our PC axes represents distinct and inter-
Otárola-Castillo 2013) to estimate multivariate shape disparity pretable morphological trends, we analyzed each PC separately.
across major mammal clades. We estimated shape disparity as We initially conducted AIC model comparisons for the effects of
Procrustes variance in PC scores from all 80 PC axes between our predictor variables on PC 1–5 (>70% of shape variation) for
the six mammalian clades for both the atlas and axis datasets. We both the atlas and axis datasets. However, we found the model
tested for pairwise differences in disparity between groups using comparison results for PC3, PC4, and PC5 to be uninformative,
1000 random permutations of the residuals. as all models either had very similar AIC scores, or were simply
We examined the relationship among atlas and axis shape supported in order of increasing model complexity, with the best
and body size, diet, and locomotor ecology across mammals via model being just the response variable. Therefore, we report the
phylogenetic generalized least squares (PGLS) regressions con- PGLS analyses and phylogenetic signal for PC1 and PC2 (but see
ducted with the caper package in R (Orme et al. 2013). The full Table S5 for results of PCs 3–5).
model for both atlas and axis analyses included log10 -transformed PGLS models with a combination of continuous and cat-
body mass, diet, and locomotion categories for each species as egorical predictor variables are difficult to represent graphi-
predictors of the PC scores for each principal components axis an- cally. To visualize trends in vertebra shape between diet and
alyzed. We used a maximum-likelihood (ML) estimate of Pagel’s locomotor categories, we generated boxplots of PC1 and PC2

EVOLUTION OCTOBER 2019 2063

A . V. L I N D E N E T A L .

Giraffe

A ATLAS

0.010
Townsend’s mole

Lama
Colugo

0.005
PC2 - 15.7% Baird’s beaked whale

Kangaroo Jumping mouse Pronghorn

rat
Asiatic elephant

Gorilla
0.000

Echidna Lesser anteater

Tree hyrax
Monotremata
Crested porcupine
Marsupialia
Two-toed sloth Puma Afrotheria
Xenarthra
-0.005

Virgina opossum Euarchontoglires

Tasmanian devil Bearcat Laurasiatheria

-0.005 0.000 0.005 0.010 0.015

PC 1 - 39.7%

Crested porcupine
B AXIS
Kangaroo rat

Monotremata
Marsupialia
0.005

Afrotheria
Townsend’s
mole Two-toed sloth Xenarthra
Euarchontoglires
PC2 - 12.9%

Puma
Laurasiatheria

Bearcat
0.000

Tasmanian
Asiatic devil Pronghorn
elephant Giraffe

Lesser anteater Llama

Jumping
mouse
-0.005

Colugo
Tree hyrax
Virgina opossum
Squirrel monkey Echidna

-0.010 0.000 0.010 0.020

PC 1 - 44.3%

Figure 2. Morphospace plots showing the first two PC axes summarizing SPHARM coefficients for (A) atlas and (B) axis shape across
mammals. Convex hulls illustrate morphospace occupation by different subclades. 3D mesh eigenshapes representing ±2 standard
deviations of each PC axis are shown in cranial and lateral view.

2064 EVOLUTION OCTOBER 2019

 AT L A S - A X I S S H A P E E VO L U T I O N

scores for the atlas and axis across different diet and locomotor LOCOMOTION DIET
groups, including plots of the original PC scores as well as the
Size-corrected
ATLAS PC scores
residuals of PC scores regressed on log10 -body mass (Fig. 3). PC scores

These plots are illustrative and do not contain information on A B

significance of shape differences between groups, because we

PC1
opted to use an information-theoretic approach rather than null-
hypothesis significance testing (Burnham et al. 2011). Linear re-
gression coefficients of PC scores on log10 -body mass and scatter-
plots of PC scores versus body mass are presented in Table S6 and
Fig. S2. above- ground aquatic carnivore omnivore herbivore
ground

C D
Results
ATLAS AND AXIS MORPHOSPACES

PC2
PC1 of atlas shape primarily describes changes in the cranio-
caudal length of the vertebra and the orientation of the trans-
verse processes (Fig. 2A). Species with low PC1 scores tend
to have short, craniocaudally compressed vertebral arches and
above- ground aquatic carnivore omnivore herbivore
proportionally narrow, horizontally oriented transverse processes, ground

whereas species with high PC1 scores have craniocaudally longer

Size-corrected
atlases and somewhat caudally rotated transverse processes. Most AXIS PC scores
PC scores
species in our analyses have low PC1 scores, but a few euar-
E F
chontoglirans (e.g., the gliding colugo) and many laurasiatheres
(particularly ungulates) exhibit positive PC1 scores. PC2 of at-
las shape mainly describes changes in the dorsoventral height
PC1

of the vertebral arch and the lateral extension of the transverse

processes. Species with negative PC2 scores have laterally ex-
tended transverse processes and more dorsoventrally compressed
above- ground aquatic carnivore omnivore herbivore
vertebral arches, whereas species with positive PC2 scores have ground

reduced transverse processes and relatively taller vertebral arches G H

(Fig. 2A).
The axis morphospace reveals patterns of shape variation
PC2

similar to those observed in the atlas. PC1 of axis shape is strongly

influenced by the craniocaudal length of the dens, centrum, verte-
bral arches, and spinous process, all of which increase along PC1
(Fig. 2B). Axis PC2 describes changes in the dorsoventral height, above- ground aquatic carnivore omnivore herbivore
ground
craniocaudal length, and angle of the spinous process, as well as
changes in the dorsoventral height and mediolateral width of the
vertebral arches. Species with negative PC2 scores have wide, Figure 3. Box plots showing the relationship between locomo-
tion and shape (A, C, E, and G) and diet category and shape (B,
dorsoventrally compressed vertebral arches and short, anteriorly
D, F, and H) for atlas and axis vertebrae. Light gray boxes repre-
projecting or nonangled spinous processes; species with positive
sent PC scores, whereas dark gray boxes represent size-corrected
PC2 scores have narrower vertebral arches and taller, more steeply residuals obtained from a linear regression of PC scores on log10
angled spinous processes (Fig. 2B). body mass (see Table S4 for regression coefficients). Eigenshapes
Although species positions within the atlas and axis mor- representing the extreme vertebra shapes of each PC are shown
phospaces are roughly equivalent along PC1 (corresponding to in lateral view. These plots are intended to illustrate trends from
craniocaudal length in both vertebrae), species are more evenly the PGLS analyses and do not contain information on statistical
scattered across PC2 in the axis morphospace than in the atlas significance. Dots above and below boxes represent outliers.

morphospace. In the latter, a few laurasiatherian species (e.g., gi-

raffe, llama, and Baird’s beaked whale) have high PC2 scores,
whereas most other species are concentrated near the origin or
toward low PC2 scores.

EVOLUTION OCTOBER 2019 2065

A . V. L I N D E N E T A L .

Table 1.Multivariate disparity of atlas and axis shape between mammalian subclades, measured as the Procrustes variance in all 80 PCs
between groups using the morphol.disparity function in the geomorph R package.

No. of Estimated Atlas

No. of families Percentage divergence time Procrustes Axis Procrustes
Subclade families sampled sampled (million years ago) variance variance

Laurasiatheria 69 31 45 84.6 8.668 × 10–5 1.445–4

Euarchontoglires 52 27 52 83.3 3.854 × 10–5 7.642 × 10–5
Marsupialia 24 12 50 81.8 3.861 × 10–5 6.999 × 10–5
Afrotheria 12 6 50 80.9 3.425 × 10–5 6.663 × 10–5
Xenarthra 5 3 60 65.4 4.339 × 10–5 1.178 × 10–4
Monotremata 2 1 50 36.7 N/A N/A

Atlas pairwise disparity was significantly different between Laurasiatheria and Euarchontoglires (P = 0.001) and Laurasiatheria and Marsupialia (P = 0.03);
axis pairwise disparity was significantly different between Laurasiatheria and Euarchontoglires (P = 0.009) and Laurasiatheria and Marsupialia (P = 0.038);
no other between-group comparisons were significant (P > 0.05). Sampling coverage of families within each superorder is included, as well as estimated
divergence time (from Meredith et al. 2011).

In both the atlas and axis datasets, PC1 and PC2 exhibit D). Atlas morphology (PC1 and PC2) did not exhibit marked
K statistics significantly different from 0 (and see Table S7 trends across diet categories (Fig. 3B, D).
for the rest of the PCs), indicating phylogenetic signal in the
shape data (Blomberg et al. 2003). Inspection of the atlas and
axis morphospaces reveals that laurasiatherian mammals occupy
PREDICTORS OF AXIS SHAPE DIVERSITY
larger portions of morphospace than other groups (Fig. 2). The
Axis shape variation along PC1 is best explained by a model that
multivariate shape disparity, measured as the Procrustes vari-
includes both body mass and locomotion as predictors, but two
ance in all PCs among groups, was highest in Laurasiatheria
other models (one including body mass, locomotion, and diet;
for both atlas and axis vertebrae (Table 1). Among the six sub-
one including only body mass) had similar explanatory power
clades, atlas and axis pairwise disparity was significantly differ-
(Table 2C). Together, these three models comprised 90% of the
ent between Laurasiatheria and Euarchontoglires, and between
AICc weight among all models. Size-corrected PC1 scores of
Laurasiatheria and Marsupialia (Table 1). Although some sub-
aquatic taxa were relatively lower than those of above-ground
clades were represented by fewer species, our sample repre-
and ground taxa, indicating more craniocaudally compressed
sented 45–60% of the family-level diversity within each group
vertebral arches, centra, and spinous processes in these species
regardless of taxonomic diversity (as determined by Meredith
(Fig. 3E).
et al. 2011).
The best supported model explaining axis shape variation
along PC2 includes locomotion as the only predictor, but is
PREDICTORS OF ATLAS SHAPE DIVERSITY closely followed by models including both locomotion and diet,
The best-supported model predicting atlas shape variation along and body mass and locomotion (Table 2D). Thus, we found no
PC1 included both body mass and locomotion. However, a slightly clear trend in variation in axis PC2 with respect to specific pre-
more complex model including body mass, locomotion, and diet dictors. Conversely, raw and size-corrected values of axis PC2 are
was similarly well supported (AICc = 2.61; Table 2A). Taken slightly larger in ground taxa than above-ground and aquatic taxa
together, these two models comprise 90% of the AICc weight. (albeit exhibiting large variation). This suggests that some species
Plots of size-corrected atlas PC1 show relatively lower resid- that locomote on the ground have mediolaterally wider vertebral
uals in aquatic taxa than species that locomote on the ground bodies, and dorsoventrally taller and craniocaudally lengthened
and above ground, indicating relatively more craniocaudally com- spinous processes (Fig. 3G). As in the atlas, both size-corrected
pressed vertebral bodies and transverse processes in aquatic lin- and raw axis PC1 and PC2 scores do not seem to differ among
eages (Fig. 3A). For atlas PC2 regressions, the model with the dietary categories (Fig. 3F, H).
lowest AICc score included only the intercept and no other In general, both the atlas and axis are more craniocaudally
variables, but three additional models were similarly supported compressed in smaller mammals than in larger mammals, and
(Table 2B). Thus, changes in the mediolateral width of the verte- larger mammals exhibit greater variation in shape (e.g., very high
bral arches and transverse processes, represented by PC2, were not PC1 scores in the giraffe, and low PC1 scores in the elephant and
clearly associated with particular ecological categories (Fig. 3C, whale; Fig. 2; see Fig. S2 for PC vs. body size plots).

2066 EVOLUTION OCTOBER 2019

 AT L A S - A X I S S H A P E E VO L U T I O N

Table 2. Aikaike information criterion (AICc) scores for phylogenetic generalized least squares (PGLS) regressions of ecomorphological
variables (body mass, locomotion, and diet) on (A) atlas PC1, (B) atlas PC2, (C) axis PC1, and (D) axis PC2.

(A) Atlas PC1

Model df AICc AICc Akaike weight ML lambda Lambda 95% CI

Log mass + locomotion 7 −683.3 0 0.71 0.97 0.80, NA

Log mass + locomotion + diet 9 −680.7 2.61 0.19 0.97 0.80, NA
Log mass + diet 4 −677.5 5.75 0.04 0.99 0.85, NA
Log mass 2 −676.7 6.61 0.03 1.0 0.88, NA
Diet 3 −676.2 7.09 0.02 1.0 0.86, NA
Locomotion + diet 8 −672.9 10.43 0.00 1.0 0.89, NA
Intercept 1 −671.4 11.90 0.00 1.0 0.93, NA
Locomotion 6 −669.9 13.39 0.00 1.0 0.92, NA
(B) Atlas PC2

Intercept 1 −710.9 0 0.43 0.74 0.38, 0.93

Log mass 2 −709.8 1.11 0.25 0.75 0.41, 0.93
Log mass + locomotion 7 −707.8 3.12 0.09 0.76 0.48, 0.93
Locomotion 6 −707.7 3.25 0.09 0.76 0.46, 0.93
Diet 3 −707.5 3.37 0.08 0.77 0.42, 0.94
Log mass + diet 4 −706.1 4.82 0.04 0.77 0.43, 0.94
Locomotion + diet 8 −703.3 7.61 0.01 0.78 0.47, 0.94
Log mass + locomotion + diet 9 −703.2 7.71 0.01 0.76 0.46, 0.93
(C) Axis PC1

Log mass + locomotion 7 −576.7 0 0.52 0.73 0.14, 0.98

Log mass + locomotion + diet 9 −575.5 1.16 0.29 0.73 NA, 0.98
Log mass 2 −573.1 3.51 0.09 0.75 0.18, 1.0
Log mass + diet 4 −570.9 5.70 0.03 0.77 0.22, NA
Diet 3 −570.3 6.31 0.02 0.84 0.49, NA
Locomotion + diet 8 −570.3 6.39 0.02 0.84 0.51, NA
Intercept 1 −569.0 7.69 0.01 0.86 0.50, NA
Locomotion 6 −568.2 8.48 0.01 0.86 0.51, NA
(D) Axis PC2

Locomotion 6 −655.2 0 0.34 0.52 NA, 0.95

Intercept 1 −653.5 1.65 0.15 0.62 NA, 0.97
Locomotion + diet 8 −653.5 1.71 0.14 0.62 NA, 0.98
Diet 3 −653.1 2.11 0.12 0.68 NA, NA
Log mass + locomotion 7 −652.8 2.41 0.10 0.522 NA, 0.95
Log mass 2 −652.1 3.04 0.07 0.62 NA, 0.97
Log mass + locomotion + diet 9 −651.1 4.09 0.04 0.62 NA, 0.99
Log mass + diet 4 −650.8 4.33 0.04 0.68 NA, NA

Column headings: model = predictor variables; df = degrees of freedom; AICc = AIC score corrected for small sample sizes; AICc = difference in AICc score
between a particular model and the model with lowest AICc value; Akaike weight = relative likelihood of model; ML lambda = maximum-likelihood estimate
of Pagel’s lambda for the PGLS model; Lambda 95% CI = 95% confidence interval of lambda value estimate.

Discussion describe trends in morphological diversity across a taxonomi-

We present the first study examining the morphological diversity cally broad sample of extant mammals and examine the potential
of the atlas–axis complex, an anatomical innovation that enabled ecological drivers of this diversity. We found that shape disparity
mammals to achieve and specialize on ecologically important be- of both atlas and axis vertebrae varies across major clades and is
haviors (Evans 1939; Kemp 1969; Crompton and Jenkins 1973). greater in laurasiatherian mammals (Table 1). This is consistent
Using 3D imaging and high-dimensional analytical tools, we with adaptive scenarios explaining the evolution of these skeletal

EVOLUTION OCTOBER 2019 2067

A . V. L I N D E N E T A L .

elements, as Laurasiatheria comprises an eclectic array of species, Smaller species tend to have short, compressed cervical vertebrae,
from bats to whales to horses, which are highly diverse in many but some large species, such as the Baird’s beaked whale and Asi-
ecological and physical aspects. Moreover, the lower disparity atic elephant, also exhibit relatively compressed vertebrae even at
in vertebral shape in the most speciose clade, Euarchontoglires, massive body sizes. Conversely, other large species, such as the
highlights the possibility that morphological diversity in the cer- giraffe, llama, and elk, have a relatively long atlas and axis for
vical spine can be decoupled from species diversity in mammals their size (Fig. S2). Therefore, atlas and axis length appears to be
(i.e., as previously described in mammalian body size, Venditti more variable among large-bodied taxa than small-bodied taxa.
et al. 2011; squirrel jaw shape, Zelditch et al. 2015; and baleen These results are consistent with a study of neck length modifi-
whale morphology, Marx and Fordyce 2015). cation in mammals, which found greater variation in the overall
We found some support for our hypothesis that atlas and length of the cervical spine in larger species (Arnold et al. 2017).
axis shape diversity can be explained by ecomorphological fac- Cervical vertebra count is remarkably consistent and devel-
tors such as body size, locomotion, and possibly diet. We also opmentally constrained in mammals (Narita and Kuratani 2005;
found evidence for the influence of past evolutionary history on Asher et al. 2011), totaling seven vertebrae except in two sloth
the shape variation of these vertebrae, which is consistent with genera (Bradypus and Choleopus) (Hautier et al. 2010) and man-
previous comparative studies of mammalian cranial (Marroig and atees (Trichechus) (Buchholtz et al. 2014). The tight constraint on
Cheverud 2001; Raia et al. 2010; Goswami et al. 2011), appendic- cervical vertebra number may be due to the involvement of Hox
ular (Polly 2008; Martı́n-Serra et al. 2015), and axial (Randau et al. genes responsible for cervical patterning in other developmen-
2016a) structures at large phylogenetic scales. Previous quanti- tal functions, as homeotic transformations of these elements are
tative analyses indicated that primate atlas shape is associated associated with major congenital abnormalities and cancer inci-
with locomotion when phylogenetic history is ignored (Manfreda dence in humans (Galis 1999; Galis et al. 2006). Further evidence
et al. 2006), but a more recent study in felids found that phylogeny has linked constraints in cervical count to the development and
influences cervical vertebra shape more than diet or locomotion muscularization of the diaphragm (Buchholtz et al. 2012). With
(Randau et al. 2016a). The atlas articulates tightly with the base these meristic and homeotic constraints in the number of cervi-
of the cranium, and the shape of this region exhibits strong phy- cal vertebrae, the diversity of neck morphologies in mammals is
logenetic signal in several groups of primates (Cardini and Elton achieved largely via homologous variation in the size and shape of
2008; Gilbert 2011) and carnivorans (Goswami 2006; Figueirido the vertebrae, likely involving regionally expressed growth factors
et al. 2010). If this is the case for other mammals, atlas shape may (Buchholtz 2012). It remains unclear what developmental mecha-
be generally constrained by this articulation, which is important nisms determine the length and width of individual vertebrae, but
for head flexion and extension, and could subsequently influence our results illustrate that, while very small mammals tend to have
axis shape (Evans 1939). compressed atlases and axes, large mammals have evolved both
Our analyses revealed an association between some aspects long and short neck vertebrae in spite of their constrained cervical
of atlas and axis shape and body size, such that smaller species count. Therefore, the study of vertebrae developmental patterns
tend to have craniocaudally compressed atlases and axes, whereas and mechanisms in large mammals could prove more fruitful for
larger species show more variation in atlas and axis craniocaudal understanding the intrinsic sources of vertebrae morphological
length. Therefore, scaling may be one of the mechanisms un- diversity.
derlying variation in atlas and axis morphology across lineages. In addition to body size, our analyses revealed associations
Previous studies of the axial skeleton within mammal orders and between atlas and axis shape and locomotor modes, although
families support this idea. For example, the shape of the atlas in these relationships differed between the two vertebrae. This is not
primates scales allometrically with body size, and larger primate unexpected given the divergent functions of the atlas (supporting,
species possess disproportionally thicker and more robust verte- flexing, and extending the head) versus the axis (rotation of the
bral arches than smaller species (Manfreda et al. 2006). In felids, head about the neck) (Evans 1939). Importantly, our models sup-
the atlases of larger bodied species display relatively longer ven- ported an association between atlas and axis PC1 and locomotion
tral arches and taller centra (positively allometric), but relatively when body mass was included as a covariate, but not when body
narrower prezygophyseal distance (negatively allometric) (Ran- mass was excluded (Table 2A and C). Smaller and larger species
dau et al. 2016b). In bovids, larger species have proportionally differ in their shape diversity, and our PGLS results indicate that
(isometrically) longer centra of the lumbar vertebrae, but dispro- the relationship between morphology and locomotor ecology may
portionally (positively allometric) wide centra that may restrict be confounded by the effect of body size. Comparisons of size-
flexion and extension at the lumbosacral joint (Halpert et al. 1987). corrected atlas and axis PC1 scores across locomotor categories
Although body mass is a predictor in our best supported models suggest that aquatic taxa have craniocaudally shorter, more com-
for PC1 of both the atlas and axis, this relationship is nuanced. pressed vertebrae than above-ground and ground-dwelling taxa

2068 EVOLUTION OCTOBER 2019

 AT L A S - A X I S S H A P E E VO L U T I O N

(Fig. 3A, E). This shortening of the cervical vertebrae has been results highlight that the evolution of body size and locomotor
considered an adaptation to aquatic life in cetaceans and man- behavior may have imposed different mechanical demands on the
atees, in which a shorter neck reduces flexibility and decreases necks of mammals, influencing the morphological evolution of
drag during swimming (Reidenberg 2007). their atlas–axis complex.
We found little support for the hypothesis that atlas and axis
shape is associated with diet in mammals. PGLS models that AUTHOR CONTRIBUTIONS
included only diet, or diet in addition to body mass, were not A.V.L. performed CT and laser scans, conducted the SPHARM analy-
among the best supported models predicting any set of PC scores ses and data analysis, and made the figures. K.C. cleaned 3D meshes,
compiled ecological data from literature, and assisted with SPHARM
for either the atlas or axis (Table 2). Boxplots of atlas and axis
analyses. E.B. performed laser scans and cleaned 3D meshes. S.E.S. de-
PC scores showed no discernable trend in shape variation across signed the study and assisted with data analysis. All authors contributed
diet categories (Fig. 3B, D, F, and H). When highly supported to writing the manuscript and gave final approval for publication.
models included diet as a predictor, it was always in addition to
locomotion (Table 2A, C, and D). These results suggest that the ACKNOWLEDGMENTS
effects of feeding behaviors on atlas and axis shape evolution, if We thank J. Bradley and volunteers at the Burke Museum of Natural His-
any, may be surpassed by locomotor specializations. tory and Culture, Seattle, WA, for assisting with specimen use; M. McPeek
for kind assistance with the spherical harmonics software SPHARM; J.
Because this study explores shape variation in the atlas and
Campbell for Matlab assistance and troubleshooting; members of the
axis at a broad taxonomic scale, our analyses of the relationship Santana Lab for helpful input during data collection and analysis; and
between morphology and ecological traits were similarly broad D. Irschick for insightful feedback on the manuscript. This study was
in scope. Realistically, mammals exhibit much greater ecological supported by the University of Washington’s Department of Biology via
startup funds to S.E.S.; A.V.L. received support from NSF GRFP award
and functional diversity than what is captured by the categories
1451512 while conducting this research.
employed here. However, dividing our sampled taxa into finer-
scale categories, such as gliding or fossorial locomotion groups,
DATA ARCHIVING
would have led to insufficient sampling of each category to per-
Supporting Information includes seven tables, three figures, supplemental
form robust statistical analyses. Further, morphospace plots coded references, and can be found with this article online. R code for analyses,
by locomotor mode illustrate that locomotor specialists such as raw data, and 3D models of vertebrae can be found in Dryad repository
gliders are not outliers in atlas and axis shapes (Fig. S3). We https://doi.org/10.5061/dryad.1nq8md7.
anticipate that the broad-scale trends presented here will provide
a useful starting point to discern the complicated interplay of LITERATURE CITED
factors that may influence the evolution of vertebral shape in spe- Adams, D. C., and E. Otárola-Castillo. 2013. Geomorph: an R package for the
collection and analysis of geometric morphometric shape data. Methods
cific mammal orders. For example, future studies could focus on
Ecol. Evol. 4:393–399.
differences between carnivores that consume invertebrate versus Arnold, P., E. Amson, and M. S. Fischer. 2017. Differential scaling patterns
vertebrate prey (Van Valkenburgh 1985), or species that dig with of vertebrae and the evolution of neck length in mammals. Evolution
their forelimbs versus teeth (Lessa et al. 2008)—each of which 71:1587–1599.
may have different functional implications for the shape of the Asher, R. J., K. H. Lin, N. Kardjilov, and L. Hautier. 2011. Variability and
constraint in the mammalian vertebral column. J. Evol. Biol. 24:1080–
cervical vertebrae. Likewise, in species with horns or antlers, at- 1090.
las and axis morphology may be under selection to accommodate Bartoń, K. 2016. MuMIn: multi-model inference. Available at http://r-forge.r-
muscles and ligaments involved in supporting the weight of these project.org/projects/mumin/.
cranial structures and deploying them as weapons in intraspecific Blomberg, S. P., T. Garland, and A. R. Ives. 2003. Testing for phylogenetic
signal in comparative data: behavioral traits are more labile. Evolution
combat or predator defense (Stankowich 2012).
57:717–745.
Buchholtz, E. A. 2012. Flexibility and constraint: patterning the axial skeleton
in mammals. Pp. 230–256 in R. J. Asher and J. Muller, eds. From clone to
bone: the synergy of morphological and molecular tools in paleobiology.
Conclusions Cambridge Univ. Press, New York.
Buchholtz, E. A., H. G. Bailin, S. A. Laves, J. T. Yang, M.-Y. Chan, and L.
This study is the first to quantify the morphological diversity of E. Drozd. 2012. Fixed cervical count and the origin of the mammalian
the atlas–axis complex across mammals. Despite developmental diaphragm. Evol. Dev. 14:399–411.
constraints on cervical number, atlas and axis shapes are remark- Buchholtz, E. A., K. L. Wayrynen, and I. W. Lin. 2014. Breaking constraint:
ably diverse across the mammals in our sample. In explaining this axial patterning in Trichechus (Mammalia: Sirenia). Evol. Dev. 16:382–
393.
diversity, we found strong support for phylogenetic and scaling
Burnham, K. P., D. R. Anderson, and K. P. Huyvaert. 2011. AIC model selec-
effects, moderate support for specialization for locomotion styles, tion and multimodel inference in behavioral ecology: some background,
and no apparent effect of broadly defined dietary categories. Our observations, and comparisons. Behav. Ecol. Sociobiol. 65:23–35.

EVOLUTION OCTOBER 2019 2069

A . V. L I N D E N E T A L .

Cardini, A., and S. Elton. 2008. Does the skull carry a phylogenetic signal? species-level database of life history, ecology, and geography of extant
Evolution and modularity in the guenons. Biol. J. Linn. Soc. 93:813–834. and recently extinct mammals. Ecology 90:2648.
Cardini, A., and P. D. Polly. 2013. Larger mammals have longer faces because Jungers, W. L. 1984. Aspects of size and scaling in primate biology with
of size-related constraints on skull form. Nat. Commun. 4:1–7. special reference to the locomotor skeleton. Am. J. Phys. Anthropol.
Caro, T. M., C. M. Graham, C. J. Stoner, and M. M. Flores. 2003. Correlates 27:73–97.
of horn and antler shape in bovids and cervids. Behav. Ecol. Sociobiol. Kemp, T. S. 1969. The atlas-axis complex of the mammal-like reptiles. J.
55:32–41. Zool. London 159:223–248.
Christiansen, P. 2002. Mass allometry of the appendicular skeleton in terres- Kikuchi, Y., Y. Nakano, M. Nakatsukasa, Y. Kunimatsu, D. Shimizu, N.
trial mammals. J. Morphol. 251:195–209. Ogihara, H. Tsujikawa, T. Takano, and H. Ishida. 2012. Functional
Clauss, M., T. Kaiser, and J. Hummel. 2008. The morphophysiological adap- morphology and anatomy of cervical vertebrae in Nacholapithecus ke-
tations of browsing and grazing mammals. Pp. 47–88 in I. J. Gordon rioi, a middle Miocene hominoid from Kenya. J. Hum. Evol. 62:677–
and H. H. T. Prins, eds. The ecology of browsing and grazing. Springer, 695.
Berlin, Germany. Kissling, W. D., L. Dalby, C. Flojgaard, J. Lenoir, B. Sandel, C. Sandom,
Coombs, M. C. 1983. Large mammalian clawed herbivores: a comparative K. Trojelsgaard, and J.-C. Svenning. 2014. Establishing macroecolog-
study. Trans. Am. Philos. Soc. 73:1–96. ical trait datasets: Digitalization, extrapolation, and validation of diet
Crompton, A. W., and F. A. Jenkins. 1973. Mammals from reptiles: a review preferences in terrestrial mammals worldwide. Ecol. Evol. 4:2913–
of mammalian origins. Annu. Rev. Earth Planet. Sci. 1:131–155. 2930.
Du Toit, J. T. 1990. Feeding-height stratification among African browsing Kitchener, A. 1988. An analysis of the forces of fighting of the blackbuck
ruminants. Afr. J. Ecol. 28:55–61. (Antilope cervicapra) and the bighorn sheep (Ovis canadensis) and the
Evans, F. G. 1939. The morphology and functional evolution of the atlas-axis mechanical design of the horn of bovids. J. Zool. 214:1–20.
complex from fish to mammals. Ann. N. Y. Acad. Sci. 39:29–104. Lessa, E. P., A. I. Vassallo, D. H. Verzi, and M. S. Mora. 2008. Evolution of
Figueirido, B., F. J. Serrano-Alarcón, G. J. Slater, and P. Palmqvist. 2010. morphological adaptations for digging in living and extinct ctenomyid
Shape at the cross-roads: homoplasy and history in the evolution of the and octodontid rodents. Biol. J. Linn. Soc. 95:267–283.
carnivoran skull towards herbivory. J. Evol. Biol. 23:2579–2594. Manfreda, E., P. Mitteroecker, F. L. Bookstein, and K. Schaefer. 2006. Func-
Felice, R. N., and A. Goswami. 2018. Developmental origins of mosaic evo- tional morphology of the first cervical vertebra in humans and nonhuman
lution in the avian cranium. PNAS 115:555–560. primates. Anat. Rec. - Part B New Anat. 289:184–194.
Galis, F. 1999. Why do almost all mammals have seven cervical vertebrae? Marroig, G., and J. M. Cheverud. 2001. A comparison of phenotypic vari-
Developmental constraints, Hox genes, and cancer. J. Exp. Zool. 285: ation and covariation patterns and the role of phylogeny, ecology, and
19–26. ontogeny during cranial evolution of new world monkeys. Evolution
Galis, F., T. J. M. Van Dooren, J. D. Feuth, J. A. J. Metz, A. Witkam, S. Ru- 55:2576–2600.
inard, M. J. Steigenga, and L. C. D. Wijnaendts. 2006. Extreme selection Martı́n-Serra, A., B. Figueirido, J. A. Pérez-Claros, and P. Palmqvist. 2015.
in humans against homeotic transformations of cervical vertebrae. Evo- Patterns of morphological integration in the appendicular skeleton of
lution 60:2643–2654. mammalian carnivores. Evolution 69:321–340.
Gilbert, C. C. 2011. Phylogenetic analysis of the African papionin basicranium Marx, F. G., and R. E. Fordyce. 2015. Baleen boom and bust: a synthe-
using 3-D geometric morphometrics: the need for improved methods to sis of mysticete phylogeny, diversity and disparity. R. Soc. Open Sci.
account for allometric effects. Am. J. Phys. Anthropol. 144:60–71. 2:140434.
Goswami, A. 2006. Morphological integration in the carnivoran skull. Evolu- McPeek, M. A., L. Shen, J. Z. Torrey, and H. Farid. 2008. The tempo and
tion 60:169–183. mode of three-dimensional morphological evolution in male reproduc-
Goswami, A., N. Milne, and S. Wroe. 2011. Biting through constraints: cranial tive structures. Am. Nat. 171:E158–E178.
morphology, disparity and convergence across living and fossil carniv- McPeek, M. A., L. Shen, and H. Farid. 2009. The correlated evolution of
orous mammals. Proc. R. Soc. B Biol. Sci. 278:1831–1839. three-dimensional reproductive structures between male and female
Gould, S. J. 1966. Allometry and size in ontogeny and phylogeny. Biol. Rev. damselflies. Evolution 63:73–83.
41:587–640. Meredith, R. W., J. E. Janečka, J. Gatesy, O. A. Ryder, C. A. Fisher, E. C.
Granatosky, M. C., P. Lemelin, S. G. B. Chester, J. D. Pampush, and D. Teeling, A. Goodbla, E. Eizirik, T. L. L. Simão, T. Stadler, et al. 2011.
Schmitt. 2014. Functional and evolutionary aspects of axial stability in Impacts of the Cretaceous Terrestrial Revolution and KPg extinction on
euarchontans and other mammals. J. Morphol. 275:313–327. mammal diversification. Science. 334:521–524.
Halpert, A. P., F. A. Jenkins, and H. Franks. 1987. Structure and scaling of the Nalley, T. K., and N. Grider-Potter. 2017. Functional analyses of the primate
lumbar vertebrae in African bovids (Mammalia: Artiodactyla). J. Zool. upper cervical vertebral column. J. Hum. Evol. 107:19–35.
London 211:239–258. Narita, Y., and S. Kuratani. 2005. Evolution of the vertebral formulae in
Hautier, L., V. Weisbecker, M. R. Sanchez-Villagra, A. Goswami, and R. mammals: a perspective on developmental constraints. J. Exp. Zool.
J. Asher. 2010. Skeletal development in sloths and the evolution of Part B Mol. Dev. Evol. 304:91–106.
mammalian vertebral patterning. Proc. Natl. Acad. Sci. USA 107:18903– Orme, D., R. Freckleton, G. Thomas, T. Petzoldt, S. Fritz, N. Isaac, and W.
18908. Pearse. 2013. caper: comparative analyses of phylogenetics and evolu-
Jenkins, F. A. 1960. The evolution and development of the dens of the mam- tion in R.
malian axis. Anat. Rec. 164:173–184. Pagel, M. 1999. Inferring the historical patterns of biological evolution. Nature
Johnson, S. E., and L. J. Shapiro. 1998. Positional behavior and vertebral 401:877–884.
morphology in atelines and cebines. Am. J. Phys. Anthropol. 105:333– Pellis, S. M., and R. C. E. Officer. 1987. An analysis of some predatory
354. behaviour patterns in four species of carnivorous marsupials (Dasyuri-
Jones, K. E., J. Bielby, M. Cardillo, S. A. Fritz, J. O’Dell, D. L. Orme, K. dae), with comparative notes on the eutherian cat Felis catus. Ethology
Safi, W. Sechrest, E. H. Boakes, C. Carbone, et al. 2009. PanTHERIA: a 75:177–196.

2070 EVOLUTION OCTOBER 2019

 AT L A S - A X I S S H A P E E VO L U T I O N

Pérez-Barberı́a, F. J., and I. J. Gordon. 1999. The functional relationship Sargis, E. J. 2001. A preliminary qualitative analysis of the axial skeletal of
between feeding type and jaw and cranial morphology in ungulates. tupaiids (Mammalia, Scandentia): functional morphology and phyloge-
Oecologia 118:157–165. netic implications. J. Zool. 253:473–483.
Pierce, S. E., J. A. Clack, and J. R. Hutchinson. 2011. Comparative axial Shen, L., H. Farid, and M. A. McPeek. 2009. Modeling three-dimensional mor-
morphology in pinnipeds and its correlation with aquatic locomotory phological structures using spherical harmonics. Evolution 63:1003–
behaviour. J. Anat. 219:502–514. 1016.
Polly, P. D. 2008. Adaptive zones and the pinniped ankle: a three-dimensional Sidor, C. A. 2001. Simplification as a trend in synapsid cranial evolution.
quantitative analysis of carnivoran tarsal evolution. Pp. 167–196 in E. Evolution 55:1419–1442.
J. Sargis and M. Dagosto, eds. Mammalian evolutionary morphology. Stankowich, T. 2012. Armed and dangerous: predicting the presence and
Springer, Dordrecht, The Netherlands. function of defensive weaponry in mammals. Adapt. Behav. 20:32–43.
Popowics, T. E., and M. Fortelius. 1997. On the cutting edge: tooth blade Van Valkenburgh, B. 1985. Locomotor diversity within past and present guilds
sharpness in herbivorous and faunivorous mammals. Ann. Zool. Fennici of large predatory mammals. Paleobiology 11:406–428.
34:73–88. Van Valkenburgh, B., and C. B. Ruff. 1987. Canine tooth strength and killing
R Core Team. 2018. R: a language and environment for statistical computing. behaviour in large carnivores. J. Zool. London 212:379–397.
R Foundation for Statistical Computing, Vienna, Austria. VanBuren, C. S., and D. C. Evans. 2016. Evolution and function of anterior
Raia, P., F. Carotenuto, C. Meloro, P. Piras, and D. Pushkina. 2010. The cervical vertebral fusion in tetrapods. Biol. Rev. Camb. Philos. Soc.
shape of contention: adaptation, history, and contingency in ungulate 92:608–626.
mandibles. Evolution 64:1489–1503. Vander Linden, A., B. P. Hedrick, J. M. Kamilar, and E. R. Dumont. 2019. Atlas
Randau, M., A. R. Cuff, J. R. Hutchinson, S. E. Pierce, and A. morphology, scaling and locomotor behaviour in primates, rodents and
Goswami. 2016a. Regional differentiation of felid vertebral column relatives (Mammalia: Euarchontoglires). Zool. J. Linn. Soc. 185:283–
evolution: a study of 3D shape trajectories. Org. Divers. Evol., 299.
https://doi.org/10.1007/s13127-016-0304-4. Venditti, C., A. Meade, and M. Pagel. 2011. Multiple routes to mammalian
Randau, M., A. Goswami, J. R. Hutchinson, A. R. Cuff, and S. E. Pierce. diversity. Nature 479:393–396.
2016b. Cryptic complexity in felid vertebral evolution: shape differen- Wilson, D. E., and D. M. Reeder, eds. 2005. Mammal species of the
tiation and allometry of the axial skeleton. Zool. J. Linn. Soc. 178:183– world: a taxonomic and geographic reference (Vol. 1). JHU Press,
202. Baltimore.
Reichman, O. J., and S. C. Smith. 1990. Burrows and burrowing behavior by Ydesen, K. S., D. M. Wisniewska, J. D. Hansen, K. Beedholm, M. Johnson,
mammals. Pp. 197–244 in H. H. Genoways, ed. Current mammalology. and P. T. Madsen. 2014. What a jerk: prey engulfment revealed by high-
Plenum Press, New York. rate, super-cranial accelerometry on a harbour seal (Phoca vitulina). J.
Reidenberg, J. S. 2007. Anatomical adaptations of aquatic mammals. Anat. Exp. Biol. 217:2239–2243.
Rec. 290:507–513. Zelditch, M. L., J. Li, L. A. P. Tran, and D. L. Swiderski. 2015. Relationships of
Revell, L. J. 2012. phytools: an R package for phylogenetic comparative diversity, disparity, and their evolutionary rates in squirrels (Sciuridae).
biology (and other things). Methods Ecol. Evol. 3:217–223. Evolution 69:1284–1300.
Santana, S. E., E. R. Dumont, and J. L. Davis. 2010. Mechanics of bite force Zsoldos, R. R., and T. F. Licka. 2015. The equine neck and its function during
production and its relationship to diet in bats. Funct. Ecol. 24:776–784. movement and locomotion. Zoology 118:364–376.
Shapiro, L. J., and C. V. Simons. 2002. Functional aspects of strepsirrhine
lumbar vertebral bodies and spinous processes. J. Hum. Evol. 42:753–
783.
Slater, G. J., and A. R. Friscia. 2019. Hierarchy in adaptive radiation: a case Associate Editor: A. Goswami
study using the Carnivora (Mammalia). Evolution 73:524–539. Handling Editor: Mohamed A. F. Noor

Supporting Information
Additional supporting information may be found online in the Supporting Information section at the end of the article.

Figure S1. Phylogenetic relationships of taxa in this study, pruned from the 2011 time-calibrated molecular family-level tree of mammals published by
Meredith et al. [41].
Figure S2. Scatterplots illustrating the relationship between body size and shape for the atlas (A and B) and axis (C and D) vertebrae.
Figure S3. Morphospace plots showing PC1 and PC2 scores coded by primary locomotor behavior (see SI Table 1) for (A) atlas and (B) axis.
Supporting Information
Table S1. Catalogue information, taxonomic information, and ecological data for all study taxa.
Table S2. List of pruned tip labels from the family-level phylogeny published by Meredith et al. (see main text Ref. 41) with corresponding representatives
from each family included in this study.
Table S3. Positions and descriptions of landmarks used for SHARM model registration (see main text Fig. 1B).
Table S4. Standard deviation and proportion of variance explained by each principal component for the atlas and axis vertebrae.
Table S5. Aikaike information criterion (AICc) scores for phylogenetic generalized least squares (PGLS) regressions of ecomorphological variables (body
mass, locomotion, and diet) on Atlas PC1–PC5 (A–E) and Axis PC1–PC5 (F–J).
Table S6. Coefficients from linear regression of atlas and axis PC scores on log-transformed body mass.
Table S7. Phylogenetic signal in atlas and axis shape estimated using Blomberg’s K statistic for the first two PC scores for each species (phylosig function
in the phytools R package).

EVOLUTION OCTOBER 2019 2071