The metabolic cost of walking in humans, chimpanzees, and early hominins
Herman Pontzera,*, David A. Raichlenb, Michael D. Sockolc
aWashington University, Department of Anthropology, St. Louis MO 63130, USA
bUniversity of Arizona, Department of Anthropology, Tucson AZ 85721, USA
cUniversity of California at Davis, Department of Anthropology, Davis CA 95616, USA
a r t i c l e i n f o
Received 3 October 2007
Accepted 28 July 2008
a b s t r a c t
Bipedalism is a defining feature of the hominin lineage, but the nature and efficiency of early hominin
walking remains the focus of much debate. Here, we investigate walking cost in early hominins using
experimental data from humans and chimpanzees. We use gait and energetics data from humans, and
from chimpanzees walking bipedally and quadrupedally, to test a new model linking locomotor anatomy
and posture to walking cost. We then use this model to reconstruct locomotor cost for early, ape-like
hominins and for the A.L. 288 Australopithecus afarensis specimen. Results of the model indicate that hind
limb length, posture (effective mechanical advantage), and muscle fascicle length contribute nearly
equally to differences in walking cost between humans and chimpanzees. Further, relatively small
changes in these variables would decrease the cost of bipedalism in an early chimpanzee-like biped
below that of quadrupedal apes. Estimates of walking cost in A.L. 288, over a range of hypothetical
postures from crouched to fully extended, are below those of quadrupedal apes, but above those of
modern humans. These results indicate that walking cost in early hominins was likely similar to or below
that of their quadrupedal ape-like forebears, and that by the mid-Pliocene, hominin walking was less
costly than that of other apes. This supports the hypothesis that locomotor energy economy was an
important evolutionary pressure on hominin bipedalism.
? 2008 Elsevier Ltd. All rights reserved.
The adoption of habitual terrestrial bipedalism is evident in the
earliest fossil hominins (White et al.,1994; Ward et al., 2001; Galik
et al., 2004; Zollikofer et al., 2005) and has long been considered
the distinctive evolutionary event marking our divergence from the
other African apes (Darwin, 1871; Dart, 1925; Washburn, 1967;
Richmond et al., 2001; Ward, 2002). This critical transition has
received nearly continuous attention for well over a century (see
Richmond et al., 2001), and yet, robust debate regarding the origin
of our bipedalism persists. Since energy efficiency is often thought
tobe an importantevolutionary pressure,and since humanwalking
is efficient when compared to other species (Rubenson et al., 2007),
several studies over the past four decades have focused on the
energetic cost of walking in early hominins and the longstanding
hypothesis that selection for increased locomotor efficiency drove
the adoption and persistence of hominin bipedalism (e.g., Rodman
and McHenry, 1980; Stern and Susman, 1983; Susman et al., 1984;
Leonard and Robertson, 1997; Sockol et al., 2007).
In this paper, we examine locomotor energetics in early homi-
nins by applying new biomechanical and energetic data from
chimpanzees and humans to several competing hypotheses for
early hominin anatomy and gait. We use chimpanzees as a model
for the last common ancestor (LCA) of humans and chimpanzees,
primarilyas a matterof parsimony. Ourplacewithin theAfrican ape
clade as a sister-taxon to the panines (chimpanzees and bonobos;
Ruvolo, 1997), and the postcranial and locomotor similarities
shared by gorillas, bonobos, and chimpanzees, suggest that
quadrupedal knucklewalking was the ancestral condition for the
hominin lineage (Washburn, 1967; Pilbeam, 1996). Further,
morphological evidence supports the hypothesis that hominins
evolved from knucklewalking apes (Richmond and Strait, 2000;
Richmond et al., 2001). However, while a chimpanzee-like knuck-
lewalking ape is our working model for the human-chimpanzee
LCA, the methods developed here can be extended to alternative
reconstructions, such as recent work suggesting that the extended
bipedal gait used by orangutans in the canopy may be the ancestral
hominin condition (Thorpe et al., 2007).
Reconstructing hominin locomotor costs
Previous efforts to reconstruct locomotor energetics in early
hominins have ranged from comparative allometric approaches to
* Corresponding author.
E-mail address: email@example.com (H. Pontzer).
Contents lists available at ScienceDirect
Journal of Human Evolution
journal homepage: www.elsevier.com/locate/jhevol
0047-2484/$ – see front matter ? 2008 Elsevier Ltd. All rights reserved.
Journal of Human Evolution 56 (2009) 43–54
sophisticated computer modeling. In an early study investigating
energetics and the origin of hominin bipedalism, Rodman and
McHenry (1980), using a comparative allometric approach, pre-
sented data indicating that human walking was less costly and
chimpanzee walking more costly, than expected for a quadruped of
similar body mass. Noting that previous work on chimpanzees and
capuchins indicated that bipedalism was no more costly than
quadrupedalism for these quadrupedally-adapted primates (Taylor
and Rowntree, 1973), Rodman and McHenry (1980) went on to
suggest that the bipedalism of early hominins likely provided an
energetic advantage over their quadrupedal forebears, since any
adaptations for habitual bipedalism would likely have improved
locomotor efficiency. A similar approach was taken by Leonard and
Robertson (1997), who argued that the energetic advantage of early
hominin bipedalism was likely even greater for females. Such
comparative approaches have the advantage of being straightfor-
ward and drawing on robust data sets, but present challenges in
incorporating details of early hominin locomotor anatomy other
than body mass and do not consider differences in posture.
Other reconstructions of hominin locomotor performance have
focused on morphology preserved in fossil material. Much of this
work has centered on Australopithecus afarensis (Stern and Susman,
1983; Susman et al.,1984; Latimer,1991; Stern, 2000; Ward, 2002),
since postcrania from earlier hominins are poorly known (Ward,
2002). These analyses have resulted in competing reconstructions
of early hominin locomotor performance, with some (Stern and
Susman,1983; Susman et al.,1984; Stern, 2000) suggesting that the
hind limb anatomy of A. afarensis indicates an inefficient, ‘‘bent-hip,
bent-knee’’ (BHBK) gait for this species and others (Latimer, 1991),
arguing that walking in A. afarensis was similar in gait and effi-
ciency to that of modern humans (see Ward, 2002 for a review of
this debate). While these morphological analyses present a detailed
assessment of locomotor form and function in early hominins,
distinguishing between their competing conclusions is hampered
by a lack of experimental studies linking the implicated aspects of
morphology, such as pelvic architecture (see Stern, 2000), directly
to walking kinematics or cost.
Various modeling approaches have also been used to assess
locomotor performance in early hominins, and have the advantage
of examining gaits and anatomy not present in extant taxa. For
example, Crompton et al. (1998), using an inverse dynamics
approach, suggested that a BHBK gait is implausible for A. afarensis
due to high muscle power requirements and associated heat load
(but see Stern, 1999). Recent inverse (Wang et al., 2004) and
forward (Nagano et al., 2005; Sellers et al., 2005) dynamics models
have indicated that, even using a human-like upright gait, A. afar-
ensis would have used more energy per kilogram of body mass than
do modern humans. This result is consistent with a recent
numerical model (Pontzer, 2005, 2007a,b) indicating that walking
costs are greater for individuals and species with shorter hind
limbs, a condition present in early bipeds (Jungers, 1982). In
contrast, Kramer (1999; Kramer and Eck, 2000) using an inverse
dynamics model, has argued that walking in A. afarensis was more
efficient than in modern humans due to reduced mechanical work
associated with swinging shorter hind limbs.
These mathematical simulations provide useful, quantitative
comparisons for different hypotheses regarding early hominin gait
and posture, but like any model are constrained by the assumptions
and data used to construct and validate them. Inverse dynamics
models (Crompton et al., 1998; Kramer, 1999; Kramer and Eck,
2000; Wang et al., 2004) require movement patterns and limb
segment properties as inputs, details typically derived from
humans. Forward dynamics approaches, in which locomotion for
a given species is modeled through iterative computer simulation
(Sellers et al., 2003, 2005; Nagano et al., 2005), generate movement
profiles de novo and are flexible enough to incorporate a range of
anatomical designs and optimization criteria. Previous forward
dynamics studies of fossil hominin gait have used human-like
muscle lengths and limb segment inertias as inputs (Nagano et al.,
2005; Sellers et al., 2005), and a narrow range of optimization
criteria (maximizing energetic efficiency: Sellers et al., 2005;
maintaining a human-like posture: Nagano et al., 2005). While this
powerful approach can be validated against experimental data (e.g.,
Sellers et al., 2003, 2005), the lack of experimental data on loco-
motor cost in non-human apes has prevented the validation of
forward dynamics models for these species, important points of
comparison for early hominins.
Experimental approaches to reconstructing early hominin
walking cost have been limited by the inherent difficulties of
measuring oxygen consumption during locomotion in captive
primates. Until recently (Sockol et al., 2007), the only study of
locomotor energy cost in apes examined two juvenile chimpanzees
(Taylor and Rowntree,1973) and notably found no difference in cost
between bipedal and quadrupedal running. While this study has
been central to the debate regarding early hominin energetics (e.g.,
Rodman and McHenry, 1980), the use of juveniles and the lack of
biomechanical analyses makes it difficult to assess the reliability of
the metabolic data in this study (see Steudel-Numbers, 2003), or to
link the high cost of locomotion reported for chimpanzees to any
aspect of anatomy or gait. Sockol and colleagues (2007) addressed
these issues by examining energetics and mechanics in a sample of
adult chimpanzees. Results of this study supported previous work
indicating that chimpanzee locomotion is energetically costly
relative to humans and other mammals (Taylor and Rowntree,
1973; Taylor et al., 1982). However, the relative costs of bipedalism
and quadrupedalism in chimpanzees varied, with three of five
chimpanzees using more energy to walk bipedally, and two having
equivalentoreven lower bipedal costs (Sockol et al., 2007). Notably,
differences in cost corresponded to differences in the volume of
muscle activated to support bodyweight at each step, indicating
a strong causal link between locomotor anatomy, posture, and cost
(Sockol et al., 2007).
Other experimental studies have used modern humans or
highly trained macaques as models for investigating walking costs
in early hominins. Carey and Crompton (2005), in a study of
humans, found BHBK walking to be approximately 50% more
expensive than upright walking, consistent with earlier modeling
work (Crompton et al., 1998). A comparison of quadrupedal and
bipedal walking in trained macaques (Nakatsukasa et al., 2004,
2006) indicated that bipedalism was approximately 25% more
expensive, in contrast to the earlier data from chimpanzees and
capuchins (Taylor and Rowntree, 1973). The use of tractable model
species provides a greater degree of experimental control and
depth of analysis for these studies, but as with computer models,
anatomical and kinematic differences between early hominins and
these species constrain their application.
An integrated experimental-numerical approach
In this study, we investigate early homininwalking cost using an
integrated approach that combines experimental measures of cost
and kinematics in chimpanzees and humans with a numerical
model linking locomotor anatomy and gait to cost. Previous work
has indicated a predictable, causal relationship between locomotor
anatomy and gait, and the metabolic cost of locomotion in terres-
trial animals (Kram and Taylor, 1990; Roberts et al., 1998a,b;
H. Pontzer et al. / Journal of Human Evolution 56 (2009) 43–5444
Pontzer, 2007a,b). We apply this framework here, using recent
kinetic, kinematic, and metabolic data from chimpanzees and
humans (Sockol et al., 2007), as well as published data for a range of
other species to construct and test a predictive numerical model for
locomotor cost based on work by Roberts and colleagues (1998a,b).
Our model reliably predicts the mass-specific cost of transport (VO2
kg?1m?1) using three parameters, each closely tied to anatomy or
gait: 1) effective mechanical advantage, which is a function of
posture and muscle moment arm length (Biewener, 1989), 2) step
length, which is a function of limb length (Kram and Taylor, 1990;
Hoyt et al., 2000; Pontzer, 2007a,b), and 3) muscle fascicle length.
These links between anatomy, gait, and cost allow us to estimate
the walking costs of early hominins for a range of hypothetical limb
lengths, postures, and muscle lengths, using bipedal chimpanzees
and modern humans as boundary conditions for anatomy and gait.
We then use the validated model toexamine the hypothesis that
locomotor efficiency was a critical selective pressure for the origin
of hominin bipedalism. Comparing estimates of walking cost from
the model against observed quadrupedal walking costs in chim-
panzees, we test the prediction that bipedalism in early hominins
was less costly than the quadrupedalism of their ape-like forebears.
Specifically, we investigate the degree to which limb length, muscle
length, and posture would need to change in order for early ape-
like hominins to reap an energetic benefit from adopting a bipedal
gait. Further, we compare the independent effects of limb length,
muscle length, and posture on locomotor cost. Finally, we reex-
amine previous reconstructions of A. afarensis within the frame-
work of the present model in order to establish a range of plausible
locomotor costs for this species.
Modeling locomotor cost
While it is generally accepted that walking and running cost
derive from the muscle activity required to support and propel the
body and move the limbs (Biewener, 2003), several approaches
have been used for predicting locomotor cost. Some workers have
used multivariate statistics to link locomotor cost to anatomical
variables (e.g., Steudel-Numbers and Tilkens, 2004). This inductive
approach is useful for explicating the statistical relationship
between anatomy and cost, but is inherently sample-specific and is
therefore difficult to apply to broader comparative contexts. Others
have focused on the mechanical work done to move the center of
mass and limbs (Cavagna and Kaneko, 1977; Heglund et al., 1982;
Willems et al., 1995; Minetti et al., 1999), or more recently, to
redirect the centerof mass during the heel-strike collision (Donelan
et al., 2002; Collins et al., 2005). However, while the mechanical
work performed must be reflected in the metabolic energy
consumed during locomotion, mechanical work has proven to be
a relatively poor predictor of metabolic cost in empirical studies
because the apparent efficiency with which work is performed
changes with speed and between species (Cavagna and Kaneko,
1977; Heglund et al.,1982; Willems et al.,1995; Minetti et al.,1999).
An alternative approach is to estimate locomotor cost from the rate
at which muscle force is generated to support bodyweight (Kram
and Taylor,1990; Taylor,1994; Roberts et al.,1998a,b; Pontzer, 2005,
2007a). This force-production approach is related to analyses of
work, since greater displacements of the center of mass will
generally require higher forces. However, the approaches are
mathematically distinct (see Kram and Taylor,1990; Pontzer, 2005)
and by focusing on muscle force rather than muscle work, the
force-production approach better incorporates isometric contrac-
tions, which constitute a large portion of muscle activity during
terrestrial locomotion (e.g., Roberts et al.,1997), but which perform
no mechanical work because they produce no displacement.
Finally, some forward-dynamics approaches (e.g., Sellers et al.,
2003, 2005) have used complex, empirically validated muscle
models (e.g., Umberger et al., 2003) to convert muscle activation
patterns to metabolic cost. This method shares similarities with
both force-production and work approaches, since increases in
work and muscle force will increase predicted cost. We use the
force-production approach here, because it does not require muscle
activation patterns and because it has proven reliable in predicting
locomotor energy cost (Kram and Taylor, 1990; Taylor, 1994; Rob-
erts et al., 1998a,b; Pontzer, 2005, 2007a).
During walking and running on level ground the body’s center
of mass rises and falls with each step. Gravity accelerates the body
downward, while muscles act toacceleratethe body upward. When
averaged over a stride cycle, these accelerations must be equal in
magnitude, effectively canceling each other. However, while the
magnitude of gravity is constant, the magnitude of upward accel-
eration, measured as the vertical ground force exerted by the limbs,
changes over the course of the stride cycle. This is most apparent
during running: vertical ground force is zero during the aerial
portion of the stride and two- to three-times bodyweight during
stance phase. As contact time (i.e., stance phase duration)
decreases, the magnitude of ground forces must increase so that
the mean vertical ground force magnitude, averaged over the entire
stride, is equal to bodyweight (Kram and Taylor, 1990; Biewener,
2003; Pontzer, 2005).
approach for explaining locomotor cost (Kram and Taylor, 1990;
Taylor,1994), has indicated that the metabolic cost of locomotion is
a product of the volume of muscle activated each step to produce
these ground forces (Taylor et al., 1980; Kram and Taylor, 1990;
Taylor, 1994; Roberts et al., 1998a,b; Pontzer, 2005, 2007a). This
relationship explains both the increase in energy use with speed
and the scaling of transport cost, two consistent trends in studies of
running energetics. First, since contact time decreases as an animal
runs faster, ground force impulses increase with speed, so that the
rate of energy use (VO2s?1) increases with speed (Kram and Taylor,
1990). Second, animals with shorter legs take shorter, more
frequent steps, and therefore, must generate greater ground forces
at higher rates than longer-legged animals running at the same
speed. Consequently, the mass-specific cost per meter (VO2kg?1
m?1) is primarily a function of effective limb length, or hip height,
with longer-legged animals using less energy per meter (Kram and
Taylor, 1990; Pontzer, 2007a, b)1. While walking mechanics differ
markedly from running mechanics, these relationships between
contact time, limb length, and cost hold; with shorter contact times
leading to larger, more frequent ground forces, and thus, higher
metabolic costs (Pontzer, 2005, 2007a).
Calculating the volume of muscle activated during walking or
running in order toestimate metabolic cost requiresinformation on
the magnitude of the ground force impulse, the length of the
muscle fascicles (lfasc), and the ratio of the anatomical moment arm
of the muscles (r) to the load arm of the ground force vector (R). The
product of the force generated by the limb muscles and their
anatomical moment arm (Fmusc?r) must balance the torque
generated by the groundforce (Fground?R). This relationship can be
rearranged as Fmusc¼Fground/EMA, where EMA is the effective
mechanical advantage of the limb (r/R; Biewener, 1989). Since the
tension generated by a muscle is a product of the cross-sectional
area of active muscle, and the magnitude and frequency of ground
1In contrast, the work needed to swing the limbs should theoretically increase
with limb length. However, the proportion of total locomotor cost spent on leg
swing is relatively small (w20%; Marsh et al., 2004; Pontzer, 2007a) and does not
appear to scale strongly with body mass (Hildebrand, 1985; Pontzer, 2007a). Thus,
we make the simplifying assumption of ignoring swing cost here.
H. Pontzer et al. / Journal of Human Evolution 56 (2009) 43–5445
force impulses is inversely proportional to contact time (tc;Kram
and Taylor, 1990; Pontzer, 2005, 2007a), the mass-specific volume
of muscle activated per second (_Vmusc) at a given joint, or for
a simplified idealized one-joint limb, can be calculated as:
_Vmusc ¼ g ?1
where g is gravitational acceleration and s is a constant relating
tension to muscle cross-sectional area (usually 20 N/cm2; Biewener
et al., 2004). Note that in this generalized equation, lfascand EMA
are mean values for the entire limb, weighted by muscle physio-
logical cross-sectional area (PCSA; see Roberts et al., 1998b; Biew-
eneret al., 2004). Also, note that the volume of muscle calculatedby
Equation 1 is inherently mass-specific (i.e., the cm3of muscle per kg
of body mass), because the term 1/tcgives the mass-specific rate of
ground force production (Kram and Taylor, 1990; Pontzer, 2005,
Because both lfascand EMA scale as (body mass)0.26(Biewener,
1989; Kram and Taylor, 1990), these variables generally cancel in
interspecific comparisons, leaving only 1/tcas the critical variable
determining the rate of muscle activation. This was the key insight
of the Kram and Taylor (1990) study demonstrating that muscle
activation, and thus locomotor cost, was a function of 1/tc for
terrestrial animals. Subsequent work demonstrated the importance
of considering lfascand EMA for species that do not conform to
mammalian scaling patterns (Roberts et al., 1998a,b). Given the
variation in lfasc and EMA among the species in our data set,
particularly the long muscle fibers and crouched posture of chim-
panzees compared to humans, we calculated muscle volume for all
species following Equation 1 as:
where lfasc/EMA is calculated for each limb joint (a, b, .i) and
summed, and this sum multiplied by 1/tc. For each joint, lfascand r
are calculated as the mean fascicle length and anatomical moment
arm for all muscles thatextend the joint, weighted byeach muscle’s
PCSA (see Roberts et al., 1998b; Biewener et al., 2004). Equation 2
gives the rate of muscle activation (cm3kg?1s?1) to calculate the
mass-specific volume of muscle activated per distance traveled
(cm3kg?1m?1); Equation 2 is divided by walking or running speed
where Lstepis step length, the horizontal distance covered by the
center of mass during stance phase. This was the equation used to
predict the metabolic cost per distance traveled.
Testing the model
To test the locomotor cost model, we assembled data on meta-
bolic cost, kinematics, and kinetics during walking and running for
a broad comparative sample of birds and mammals, including
humans and chimpanzees (Table 1). Data for lfasc, EMA, and Lstep
were taken or calculated from the literature (Roberts et al., 1998b;
Thorpe et al., 1999; Biewener et al., 2004; Sockol et al., 2007). Due
to the limited number of species for which published force-plate
data is available, the ratio of lfasc/EMA for turkeys was used for
guinea fowl, bobwhite quail, rhea, and emu (Roberts et al., 1998b).
This approach implicitly assumes that the ratio of lfasc/EMA is
similar for these birds, which is consistent with previous work on
the metabolic cost of running in these species (Roberts et al.,
1998a). Metabolic cost data, specifically, the mass-specific net cost
of transport (COT; ml O2kg?1m?1), for dogs and all bird species;
were taken from studies combining metabolic and force-produc-
tion measurements (Roberts et al., 1998a, b). Cost data for chim-
panzees were taken from Sockol et al. (2007) for the three
chimpanzees (see Sockol et al., 2007, subjects C1–3 in their Table 1)
for whom force-plate and cost data were both collected.
COT data for humans were collected separately for this study,
using sixhealthyhuman subjects
mass¼69.0 kg, mean hip height¼89.6 cm) with no apparent gait
abnormalities, size-matched to the samples used in force-plate
studies of human walking (Sockol et al., 2007) and running
(Biewener et al., 2004). Washington University approval for human
data collection was obtained prior to the study and subjects gave
informed consent prior to their participation; institutional guide-
lines were followed throughout. Walking and running costs were
measured while walking and running on a treadmill (Sole Fitness
F85) at a range of speeds, using standard open-flow methods
described previously (Fedak et al., 1981; Pontzer, 2007a). Briefly,
subjects wore a loose mask and air was drawn past their face,
through the mask, at a high rate (mass-flow rate: 200–300 lpm) in
order to capture all expired air. Collected air was continuously
sampled and monitored for oxygen concentration (Sable Systems
PA-1B) after having water vapor and CO2 removed; oxygen
concentration was recorded at a rate of 10 Hz using AxoScope data
acquisition software (Molecular Devices?). Mean oxygen concen-
tration was calculated for the last minute of each trial, after
a minimum of three minutes of steady walking. Only trials inwhich
oxygen consumption visibly plateaued, indicating steady-state
aerobic energy use, were included. Note that while early energetics
studies often used much longer trials (ca. 20 minutes, see for
example Taylor and Rowntree, 1973; Taylor et al., 1982), the
continuous monitoring of oxygen concentration afforded by
improved equipment has enabled shorter trial lengths to be used
(e.g., Roberts et al., 1998a; Wickler et al., 2000; Griffin et al., 2004;
Pontzer, 2007a). For example, recent energetics work in horses
(Wickler et al., 2000; Griffin et al., 2004) has used three-minute
trials, since these shorter exercise bouts produced equivalent
results to much longer (w15 minute) trials (Wickler et al., 2000)
and are less taxing for the subjects.
To calculate the net rate of oxygen consumption (i.e., with
resting oxygen use subtracted), we used the intercept-subtraction
method as in Taylor and Rowntree (1973) and many other loco-
motor studies (see Taylor et al., 1982; Rubenson et al., 2007): for
each subject, we fit a trendline to a plot of oxygen consumption per
second versus speed and subtracted the intercept value (i.e., the
expected value at 0 ms?1; see Rubenson et al., 2007). This approach
was used for the chimpanzee sample (Sockol et al., 2007), and is
employed here for humans to maintain similar methods. Note that
mean intercept cost for chimpanzees (4.63 ml O2s?1), which was
not reported in Sockol et al. (2007), was near the predicted value for
their body mass (4.53; see Taylor et al., 1982), while the human
intercept cost for the sample in this study (6.56) was marginally
higher than expected for their mass (5.73). The net rate of oxygen
consumption was divided by body mass, and the mass-specific rate
of oxygen consumption was calculated for speeds matched to the
walking and running force-plate studies (Table 1). This rate of
oxygen consumption was then divided by speed to give the mass-
specific COT (ml O2kg?1m?1), the cost to travel a meter.
Estimating lfascand EMA
Mean fascicle length for each joint (lfasc) and EMA for all species
in this data set were calculated using similar methods (Roberts
et al.,1998b; Bieweneret al., 2004; Sockol et al., 2007). First, fascicle
length for each muscle in a given muscle group (plantar flexors,
knee extensors, hip extensors, shoulderextensors, elbow extensors,
H. Pontzer et al. / Journal of Human Evolution 56 (2009) 43–54 46
and wrist flexors; see Roberts et al.,1998b; Thorpe et al.,1999; and
Biewener et al., 2004 for group definitions) was weighted by that
muscle’s PCSA to determine mean lfascfor that group, following
Roberts et al. (1998a,b). Thus, lfascfor an extensor group consisting
of a set of muscles (d, e, .i) was calculated as:
These cadaver-based values for the lfascof each extensor groupwere
then scaled to each subject, assuming the relevant cadaver-based
ratio of lfascto limb segment length (e.g., knee extensor lfasc:thigh
length) was similar forall subjects within a species. Weighted mean
values of lfascfor each joint for humans and chimpanzees are given
in Table 1.
Meananatomicalmoment arm(r) foreachmusclegroup(Table1)
was taken from published values for humans (Biewener et al., 2004,
their Table 2) or chimpanzees (Thorpe et al., 1999; their Tables 9
and 15). Shoulder values for chimpanzees were estimated from the
plot of coracobrachialis and biceps moment arms about the
shoulder in Thorpe et al. (1999), since the forelimb ground reaction
force (GRF) vector generally passes posterior tothe shoulder during
stance (see Sockol et al., 2007). These mean values of r were then
scaled to each subject assuming r scales geometrically, as (body
mass)0.33. Using a point-estimate for r is a simplification, since in
fact r changes with joint angle (Thorpe et al.,1999). However, since
these estimates are for mid-stance, when GRF magnitude is
greatest, and since using a point-estimate of r greatly simplifies
calculation of Vmusc, we considered this approach appropriate for
The moment arm of the ground reaction force vector (R) was
calculated from kinematic and force-plate data. In the simplest
case, R could be calculated solely as the perpendicular distance
from the GRF vector to the center of joint rotation (e.g., Biewener,
1989; Roberts et al.,1998b; Hutchinson, 2004; Fig.1). Here, we used
a modified approach in which the GRF vector, limb segment
accelerations, and the flexor moments generated by two-joint
muscles were combined to calculate net joint moments, following
Winter (2005) and Biewener and colleagues (2004). First, kine-
matic data were smoothed using a fourth-order zero-lag Butter-
worth filter with a 12 Hz low-pass filter, following Biewener and
colleagues (2004), and segment accelerations were calculated
using the finite differences method (Winter, 2005). Net inertial,
gravitational, and GRF moments (M) were then calculated for each
kinematic frame at each joint using the free-body method
described in Winter (2005: p.91)2. Extensor muscle forces (Fankle,
Fknee.Fshoulder) needed to generate these moments were then
calculated by solving the system of equations given in Biewener
and colleagues (2004: Eqs. 1–3):
Hind limb system:
Mknee¼ Fkneerknee? FG;kneerG;knee? FH;kneerH;knee
Mhip¼ Fhiprhip? FRF;hiprRF;hip
Melbow¼ Felbowrelbow? FWF;elbowrWF;elbow
Mshoulder¼ Fshoulderrshoulder? FT;shoulderrT;hip
Flexor moments generated by two-joint muscles (G: gastrocne-
mius, H: hamstrings, RF: rectus femoris, WF: wrist flexors, T:
triceps, long head) were calculated assuming the force produced by
each muscle in an extensor group is proportional to its PCSA
(Biewener et al., 2004) and using scaled anatomical moment arms
for these muscles (r) as described above. For each system of
equations (5–7 for the hind limb and 8–10 for the forelimb of
quadrupedal chimpanzees), the muscle force at the most distal
joint was calculated first (i.e., Eqs. 5 and 8) and the other two
equations for each limb were then solved simultaneously.
Using a Matlab?routine, extensor muscle force at each joint was
calculated foreachvideo frame of a stepand mean net moment was
calculated for the step, disregarding any frames for which the joint
Anatomical and biomechanical variables and locomotor cost for humans and chimpanzees
Mean muscle fascicle length and moment arms (cm)Vmusc/mcSourced
HipKnee Ankle ShoulderElbow Wrist
lfasc rRlfasc rRlfasc rRlfasc rRlfasc rRlfasc rR
0.90 15.1 5.2 26.3
0.64 15.1 5.2 18.3
7.9 2.5 1.3 7.1 3.7 10.6 10.6 1.7 1.7 6.2 2.1 5.8 8.0 2.0 1.7
7.9 2.5 3.7 7.1 3.76.0
3.0 10.0 1.4 0.7 3.9 3.9 10.6
3.0 10.0 1.4 5.6 3.9 3.9 10.6
Mean EMA (r/R)e
aSubjects C1, C2, and C3 in Table 1 of Sockol et al. (2007).
bMean speed from force-plate trials.
cCalculated using Eq. (3).
dSources: 1) Sockol et al., 2007, 2) new metabolic data, 3) Biewener et al., 2004, 4) Roberts et al., 1998b, 5) Roberts et al., 1998a.
eMean for all joints.
fEstimated from ratio for turkeys.
2Segment inertial properties were calculated from segment lengths following
Winter (2005) for humans or from segment lengths and circumferences following
Raichlen (2004) for the chimpanzees.
H. Pontzer et al. / Journal of Human Evolution 56 (2009) 43–54 47
moment was <25% the maximum for that step, following Biewener
and colleagues (2004). Mean extensor moment (Fr) was then
divided by mean GRF magnitude to calculate R. This was performed
for 2 to 5 successful force plate strides per subject and the mean R
for these strides used for analysis; values are given in Table 1.
Calculating R in this manner is analogous to calculating R from
FgroundR¼Fmuscler, but essentially modifies Fmuscleto account for
segment inertia and gravity as well as the action of two-joint
muscles. This approach is computationally more intensive, but has
the advantage of providing a more accurate measure of Vmusc. By
comparison,R calculatedfromthe GRF vectorandjoint centers asin
Fig. 1, neglecting limb segment inertia and two-joint muscles,
provides similar measures. For the four humans and three chim-
panzees for whom kinetic measurements were taken, the differ-
ence in R calculated using these two approaches did not achieve
statistical significance (p¼0.07, student’s two-tailed paired t-test),
although a larger sample would likely reveal statistically significant
differences. We discuss the merits and effects of these different
Muscle volume versus locomotor cost
Muscle volume activated per meter was calculated following
Equation 3, using values given in Table 1, assuming s¼20 N/cm2.
For bipedal chimpanzees, humans, and birds, lfasc/EMA was sum-
med for the ankle, knee, and hip joints. EMA for running humans
was adapted fromwalking values, keeping EMA the same at the hip
and ankle, but decreasing EMA at the knee by 75%, following
Biewener et al., (2004). For quadrupedal walking in chimpanzees
and trotting in dogs, lfasc/EMAwas summed for the ankle, knee, hip,
wrist, elbow, and shoulder, and these values were weighted by the
proportion of bodyweight supported by the forelimb or hind limb.
Values of lfasc/EMA for the wrist, elbow, and shoulder were multi-
plied by 30.5%, the percentage of bodyweight borne by the fore-
limb, while lfasc/EMA values for hind limb joints were multiplied by
69.5%. Separate joint values of lfasc/EMA for dogs are not given in
Roberts et al. (1998b), and thus, the mean value of lfasc/EMA was
used for all joints, with forelimb and hind limb values weighted
evenly (i.e., each multiplied by 50%).
The mass-specific active muscle volume per meter (Eq. 3) was
plotted against COT for each species to determine the reliability of
this approach for predicting locomotor cost. Ordinary least squares
(OLS) was used to determine the relationship between muscle
activation and cost. Other commonly used predictors of COT,
including step length, body mass, and limb length, were also
plotted against COT to determine the performance of these
predictors relative to active muscle volume.
Modeling walking cost in early hominins
Upon validating our approach for predicting locomotor cost
(see below), we used our model to estimate walking cost for
early hominins. Following Equation 3, we estimated active
muscle volumes for a range of morphologies and postures,
manipulating lfasc, EMA, and Lstep independently. We then
calculated a predicted locomotor cost using the OLS equation
from the validation study.
Boundary values for lfasc, EMA, and Lstep were adapted from
means for bipedal chimpanzees and humans, assuming that plau-
sible values for early hominins fall between these extremes. In
order to control for differences in morphology due purely to
differences in body size, values of lfasc, EMA, and Lstepwere calcu-
lated for a 30 kg human and 30 kg chimpanzee, assuming
geometric similarity within species. Species means for lfasc, r, and R
were calculated for each species and estimated for a 30 kg indi-
vidual of that species assuming that these lengths scale with (body
mass)0.33. To estimate a scaled value of Lstep, the ratio of Lstepto hip
height was first calculated, since step length is a function of hip
height (Kram and Taylor,1990; Hoyt et al., 2000; Pontzer, 2007a, b).
Hip height was calculated for a 30 kg human and chimpanzee
assuming that this length scales with (body mass)0.33. Lstepwas
then calculated from hip height, using the ratio of Lstep/(hip height)
for chimpanzees or humans. For purposes of comparison, these
values and an estimated COT were also calculated for a 30 kg
Predicted dimensions for 30 kg hominoids
Model Mass (kg) Hip height (cm) Lstep(m) HipKneeAnkleShoulder ElbowWristVmusc/maEstimated COTb
rRlfasc rRlfasc rRlfasc rRlfasc rRlfasc rR
Quadrupedal chimpanzee 30
12.8 4.4 22.2 6.6 2.1 1.1 6.0 3.1 9.0 9.0 1.4 1.5 5.3 1.8 4.9 6.8 1.7 1.4 42.6
12.8 4.4 15.5 6.6 2.1 3.1 6.0 3.1 5.1
4.9 4.02.3 7.6 1.1 0.5 3.0 3.0 8.0
aVolume of muscle activated per meter traveled (cm3/m), calculated following Equation 3.
bCalculated from Vmuscusing the OLS regression equation given in Fig. 2d.
Fig. 1. Schematic of parameters used to estimate active muscle volume adapted from
a video frame of a bipedal chimpanzee force plate trial. Fground¼the ground force
trajectory (indicated by block arrow); r ¼muscle moment arm; R¼ ground force
moment arm; lfasc¼muscle fascicle length; Lstep¼ step length. Circles (5) indicate
centers of rotation for the hip, knee, and ankle. EMA is shown here for the hip; lfascis
shown for the ankle extensors.
H. Pontzer et al. / Journal of Human Evolution 56 (2009) 43–54 48
quadrupedal chimpanzee. Values of lfasc, EMA, Lstep, and COT for
30 kg chimpanzees and humans are given in Table 2.
We then estimated walking cost for a 30 kg early hominin over
a range of hypothetical morphologies and gaits, varying lfasc, EMA,
and Lstepindependently between the chimpanzee-like and human-
like conditions. These costs were compared to those for a 30 kg
quadrupedal chimpanzee to determine what degree of change in
lfasc, EMA, or Lstepwould lower walking costs for early hominins
below those of quadrupedal apes. Parameter values werecalculated
by the general formula:
Parameter ¼ ð%Chimpanzee ? ChimpanzeeValueÞ
þ ð%Human ? HumanValueÞ
where %Chimpanzee and %Human are the degrees to which the
parameter is chimpanzee-like or human-like, respectively, and
Chimpanzee and Human Values are those estimated for a 30 kg
chimpanzeeorhuman,given inTable 2. Values wereassumed tofall
along a continuum between boundary conditions such that
%Chimpanzee¼100 – %Human. Thus, 100%Chimpanzee is the
chimpanzee-like boundary condition for a parameter, 100%Human
is the human-like boundarycondition (Table 2). Hip height, R, r, and
lfascwere calculated directly using Equation 11. For Lstep, the ratio of
Lstep/(hip height) was calculated using Equation 11 assuming that
the %Chimpanzee for this ratio was equal to the %Chimpanzee for
EMA. This ratio and estimated hip height were then used to
calculate Lstep; thus, Lstepvalues change both as a function of hip
height and EMA. Varying Lstep in this way acknowledges its
dependence on posture, as well as hip height, with more crouched
postures and greater limb excursions resulting in longer Lstep
relative to hip height (see Schmitt, 1999; see Table 2).
In order to compare our approach to previous modeling efforts,
we estimated walking cost for the A.L. 288 A. afarensis specimen.
Body mass of 28 kg for A.L. 288 was taken from McHenry (1992).
Vmusc, and thus COT, was estimated over a range of EMA and lfasc,
from a chimpanzee-like boundary condition to a human-like
boundary condition. Boundary condition parameters were esti-
mated assuming geometric scaling as with the 30 kg human
and chimpanzee models and parameter values between these
boundary conditions were calculated using Equation 11. Hip height
was estimated from published femur and tibia length (McHenry,
1992), assuming knee flexion while standing was 155 degrees for
the chimpanzee-like boundary condition (similar to chimpanzees
in our sample) and 180 degrees for the human-like boundary
condition. In all cases, 3 cm was added to femurþtibia length to
account for the height of the foot, based on the ratio of lateral
malleolus height to hip height for a sample of 10 humans (Pontzer,
unpublished data). Vmuscwas estimated using Equation 3 over the
full range of hip height, EMA, and lfascvalues, from 100% chim-
panzee-like to 100% human-like, and estimated COT was compared
to that estimated for 30 kg chimpanzees and humans. This enabled
us to determine the degree to which lfascor EMA for A.L. 288 would
need to change from a primitive, chimpanzee-like condition to
a derived human-like condition in order to decrease walking cost
below that of similarly sized quadrupedal chimpanzees. Results
were also compared to previous cost estimates for this specimen.
Finally, we examined the sensitivity of our results to changes in
anatomical and postural variables.
Active muscle volume and locomotor cost
As predicted, the mass-specific COT (ml O2kg?1m?1) in our
comparative sample was strongly correlated with the estimated
volume of muscle activated per meter traveled (_Vmusc/m; Fig. 2).
Further, active muscle volume was a better predictor of cost than
other commonly used parameters. With bobwhite quail removed
from the regression to eliminate the disproportionate influence of
this smallest species, and the degrees of freedom limited to reflect
the number of separate species in the analysis,_Vmusc/m explains
91% of the variance in COT (r2¼0.91, df¼5 species, p<0.001),
outperforming body mass (r2¼0.68, df¼5, p¼0.012), Lstep
(r2¼0.59, df¼5, p¼0.021), and hip height (r2¼0.81, df¼5,
p¼0.006; Fig. 2). This holds when bobwhite quail are included in
the analysis, although r2values increase for all comparisons: body
mass¼0.85; Lstep¼0.94; hip height¼0.94;_Vmusc/m¼0.98 (Fig. 2).
Walking COT for humans (mean¼0.08 ml O2kg?1m?1), which
was equivalent to the minimum COT values measured in these
subjects, was lower than mean minimum COT values in some
previous studies (see Rubenson et al., 2007). This lower estimate is
partly a function of the method used to calculate COT. Here, we
subtracted the intercept value of the speed/cost relationship to
determine COT in order to keep methods for humans and chim-
panzees similar, whereas in other human studies the resting rate of
oxygen consumption is typically subtracted to calculate net cost.
Our approach lowered mean COT for humans by 0.015 ml
O2kg?1m?1compared to the COTcalculated by subtracting resting
Importantly, results of the model test are not sensitive to
between-study variation in COT or Vmusc. Increasing cost to match
the mean COT published in a recent meta-analysis of human
walking cost (COT¼0.10; Rubenson et al., 2007), or decreasing it to
match the human COT from our previous study (COT¼0.05; Sockol
et al., 2007), has a negligible effect on the overall fit of the model (r2
of 0.90 to 0.91 for all comparisons), and the_Vmusc/m remains the
best predictor (highest r2value) for COT. Similarly, increasing active
(_Vmusc/m¼17.90; Biewener et al., 2004) does not affect the fit or
relative performance of our model. Finally, the fit of the model is
unchanged (r2¼0.91) when bird species with estimated ratios of
lfasc/EMA are removed.
values fromprevious work
Effects of lfasc, EMA, and hip height
Given the strong correlation between_Vmusc/m and COT, we can
use the model to parse the independent contributions of lfasc, EMA,
and hip height to overall differences in cost between species and
gaits. When size is accounted for by comparing expected COT for
geometrically scaled 30 kg subjects, differences in locomotor cost
between humans and bipedal chimpanzees are nearly evenly
distributed among EMA, lfasc, and hip height (Fig. 3). That is, the
increased economy of humanwalking is nearlyequallya function of
posture, limb length, and muscle length. Further, the difference in
COT between bipedal and quadrupedal chimpanzees was equally
a function of the greater contact times afforded by quadrupedal
walking and the decreased EMA in these chimpanzees during
bipedal trials (Table 2).
Estimating walking cost for A.L. 288
Estimated walking costs for A.L. 288 range considerably
depending on the assumptions made regarding posture and muscle
fiber length. Our maximum estimate, using a chimpanzee-like EMA
and musclefascicle length,
3.78 Jkg?1m?1. In contrast, the lowest estimated cost, assuming
human-like EMA andlfasc,
1.94 Jkg?1m?1. Intermediate values produced an estimate of
0.14 ml O2kg?1m?1or 2.85 J kg?1m?1(Fig. 4). Generally, estimates
of COT fall below the estimated walking cost for a quadrupedal
chimpanzee of similar mass (Fig. 4). A modest change of only
6% from the chimpanzee-like boundary condition toward the
is 0.19 mlO2kg?1m?1
H. Pontzer et al. / Journal of Human Evolution 56 (2009) 43–54 49
human-like condition in EMA and lfasclowers estimated COTfor A.L.
288 below that of quadrupedal chimpanzees. When EMA and lfasc
are considered separately, an 18% change in EMA toward the
human-like, fully extended condition, or a 10% change in lfasc,
results in lower COT than expected for similarly sized quadrupedal
chimpanzees (Fig. 4). However, because hip heightis shorterfor A.L.
288, estimated COT was always greater than expected for a 30 kg
human, even at the extreme human-like boundary conditions of
EMA and lfasc.
Our estimates of COT for A.L. 288 accord well with previous
forward dynamics models of this specimen, which have produced
COT estimates of 0.14 ml O2kg?1m?1(Sellers et al., 2005) and
0.13 ml O2kg?1m?1(Nagano et al., 2005). However, while COT
estimates from these studies were most similar to our intermediate
estimates, the gaits produced by these forward-dynamics studies
were most similar to our human-like boundary condition. Thus,
COT estimates following our approach are generally lower than
those of forward-dynamics models; for a given gait, the Vmusc
COT(mlO2 kg-1 m-1)
Fig. 3. Changes in estimated cost of transport (COT; ml O2kg?1m?1) for a 30 kg
bipedal hominin with changes in EMA (black line, circles), hind limb length (black line,
squares), muscle fascicle length (gray line, triangles), and all parameters combined
(black line) between chimpanzee-like and human-like boundary conditions. Estimated
cost for 30 kg bipedal chimpanzees, 30 kg quadrupedal chimpanzees, and 30 kg
humans are shown for comparison (dotted lines).
EMA & Fascicle Length
COT (mlO2 kg-1 m-1)
Fig. 4. Estimated cost of transport (COT; ml O2kg?1m?1) for A.L. 288 over a range of
EMA (black line, circles), muscle fascicle length (gray line, triangles), and both
parameters combined (black line), between chimpanzee-like and human-like
boundary conditions. Estimated cost for 30 kg bipedal chimpanzees, 30 kg quadru-
pedal chimpanzees, and 30 kg humans are shown for comparison (dotted lines).
Body mass (kg)
1/Hip Height (m)
1/Step Length (m)
y = 0.4858x-0.28
r2 = 0.68
(with Bobwhite quail: r2=0.85)
y = 0.0993x + 0.0606
r2 = 0.59
(with Bobwhite quail: r2=0.94)
y = 0.0637x + 0.0856
r2 = 0.81 2
(with Bobwhite quail: r2=0.94)
y = 0.0030x + 0.0547
r2 = 0.91
(with Bobwhite quail: r2=0.98)
COT(mlO2 kg-1 m-1)
COT (mlO2 kg-1 m-1)
COT (mlO2 kg-1 m-1)
COT (mlO2 kg-1 m-1)
Fig. 2. Cost of transport (COT) versus a) Body mass, b) 1/Step length, c) 1/Hip height, and d) Vmusc. Black circles¼comparative data (Table 1); gray squares¼human walking; gray
triangles ¼bipedal chimpanzees; white triangles ¼quadrupedal chimpanzees.
H. Pontzer et al. / Journal of Human Evolution 56 (2009) 43–5450
approach used here produced a lower COT estimate. Mechanical-
work approaches for estimating cost for A. afarensis (0.03 ml
O2kg?1m?1: Fig. 9 in Kramer 1999; 0.06 ml O2kg?1m?1: Wang
et al., 2004) fall below the lowest estimates from our model.
The sensitivity of our models of early hominin locomotor cost to
differences in anatomical or postural variables can be calculated
directly using Equation 3. For both a hypothetical 30 kg hominin
and for A.L. 288, estimated active muscle volume increases directly
with lfascand R. Thus, a 20% increase in either of these variables for
all joints will lead to a corresponding 20% increase in_Vmusc/m.
Similarly, since_Vmusc/m is inversely proportional to Lstepand r,
changes in these variables are inversely proportional to their effect
on COT; a 20% increase in either of these variables leads to a 25%
decrease in COT. When joints are considered separately, the
sensitivity of our models of hominin cost are dependent on the
overall reconstruction of morphology and posture. For the ‘‘100%
Chimpanzee-like’’ boundary condition, changes at the hip have the
greatest effect on COT; a 20% change in lfascor R at the hip leads to
a 14% change in COT, while similar changes in these variables in the
knee or ankle lead to a 3% increase in COT. In contrast, for the ‘‘100%
Human-like’’ boundary condition, changes at the ankle have the
largest effect, with a 20% increase in lfascor R leading to a 14%
increase in COT, while a similar change at the knee or hip increases
COT by 5% or 4%, respectively. This context dependency reflects
postural differences between humans and chimpanzees and is
consistent with previous work showing the ankle and hip engaging
the largest muscle volumes in humans (Biewener et al., 2004) and
chimpanzees (Sockol et al., 2007) respectively. The crouched
posture of chimpanzees leads to large moments and muscle
volumes at the hip, while the extended posture of humans leads to
the highest moments and volumes at the ankle. Changes in these
joints, thus, have large effects on_Vmusc/m, and hence COT, in
chimpanzee- versus human-like reconstructions.
Gait, anatomy, and walking cost in humans and chimpanzees
Our results support the hypothesis that the rate of muscle
activation provides a reliable measure of locomotor cost. The
volume of muscle activated per meter traveled (_Vmusc/m) explained
a greater proportion of the variance in COT both between species
and between gaits, than any other predictor used here, including
body mass. When bobwhite quail are included in the analysis,
_Vmusc/m explained a remarkable 98% of the variation in COT (Fig. 2).
The model has such broad applicability because, by explicitly
incorporating step length, EMA, and fascicle length, it accounts for
effects of body size, speed, posture, and gait. This is consistent with
work indicating that these biomechanical variables underlie the
scaling of locomotor cost for terrestrial animals (Kram and Taylor,
1990; Roberts et al., 1998a,b; Pontzer, 2007a,b). The success of the
model suggests that Vmuscmay be useful for comparing locomotor
cost and efficiency when oxygen consumption cannot be measured
directly, as in fossil species or untrained primates. This approach
provides a more reliable estimate of locomotor cost than allometric
estimates, and also provides a means of linking metabolic cost
directly to muscular and skeletal anatomy.
The link between_Vmusc/m and COT is consistent with variation
in gait and cost in adult chimpanzees noted previously (Sockol
et al., 2007). Subject C4 in Sockol et al., (2007) used 28% greater hip
flexion and 17% greater knee flexion during quadrupedal walking
compared to the other chimpanzees. Bipedal joint angles were
similar to other chimpanzees (see Sockol et al., 2007, their Fig. 4),
but contact times for this individual were 5% longer during bipedal
trials, while all other chimpanzees used longer steps during
quadrupedal walking. While force plate data were not collected for
this individual, when percentage differences in minimum hip and
knee angles are converted into proportional differences in R and
active muscle volumes are calculated following Equation 3, esti-
mated_Vmusc/m during quadrupedal walking (84.0 cm3kg?1m?1) is
27% greater than for bipedal walking (61.8 cm3kg?1m?1). Given
the necessary estimation of R, this difference in muscle activation
corresponds reasonably well to the 44% greater COT during
quadrupedal walking for this individual.
The COT value for walking humans in this study (0.08 ml
O2kg?1m?1) is within the range (0.08–0.13), but below the mean
(0.10) of previous estimates of walking COT reported in a recent
meta-study of human locomotor cost (Rubenson et al., 2007). The
COT value in our previous report (Sockol et al., 2007) was lower still
(0.05), further highlighting the potential effect of individual and
between-study variation in comparisons of locomotor perfor-
mance. These differences are likely due in part to the relatively
small sample sizes used, and the intercept-subtraction approach
used hereand in our previous study (Sockol et al., 2007) tocalculate
net cost. Importantly, lower COT for humans do not affect the
overall fit of the model; r2values remained above 0.94 when esti-
mates of COT or_Vmusc/m from previous human studies were used,
and active muscle volume remains the best predictor of differences
in COT between species and gaits for humans and chimpanzees.
Still, given the variation in COT observed across human studies
(Rubenson et al., 2007) and within a sample of adult chimpanzees
(Sockol et al., 2007), broad comparisons across species should be
tempered with an understanding of the underlying variation in
locomotor performance. For example, depending on the individuals
or samples used for comparison (Rubenson et al., 2007; Sockol
et al., 2007), humans have COTs that are between w40%–80% lower
Differences in walking cost between gaits and species were
attributable nearly equally to differences in posture, limb length,
and muscle fascicle length. In particular, the importance of lfascin
determining locomotor cost is notable. Most previous consider-
ations of differences in human, ape, and fossil hominin locomotor
performance have focused on gait (Jenkins, 1972; Stern and Sus-
man, 1983; Crompton et al., 1998) or limb length differences
(Jungers, 1982; Kramer, 1999). While these aspects of locomotor
anatomy and gait are clearly important, our results indicate that
muscle morphology must be considered as well, in order todevelop
a complete reconstruction of locomotor performance. Recent work
on muscle morphology of apes (Thorpe et al., 1999; Payne et al.,
2006a,b) may prove critical in moving locomotor comparisons of
living and extinct hominoids beyond skeletal comparisons. Further,
future work might improve on the accuracy of estimating_Vmusc/m
and COT by taking non-invasive measures of these parameters for
individual subjects, rather than relying on scaled measurements
taken from cadaveric specimens.
Limitations of the model
As noted above, a number of approaches have been used to
estimate locomotor costs for extinct hominins. The novel method
we propose here has the advantage of explicitly linking locomotor
anatomy to cost using a model that can be verified across a range of
gaits and species. However, as with any modeling study, the results
are only as reliable as the underlying assumptions. Perhaps most
critically, our approach uses estimated muscle force production and
active muscle volume to predict metabolic cost. While this
approach is strongly supported by previous empirical work (Kram
and Taylor,1990; Taylor,1994; Roberts et al.,1998a,b; Pontzer, 2005,
2007a) and explains over 90% of the variance in cost in this data set
H. Pontzer et al. / Journal of Human Evolution 56 (2009) 43–54 51
(Fig. 2), it does not directly consider mechanical work or collisional
energy losses, which must also contribute to cost (see Minetti et al.,
1999; Donelan et al., 2002). Incorporating collisional losses and
mechanical work may improve the fit of the model, and could
potentially affect our results.
One important assumption we used in estimating cost for A.L.
288 and 30 kg chimpanzees and humans is that EMA and lfascscale
isometrically. This assumption has the effect of decreasing esti-
mated cost, since the ratio of r to R remains constant if both scale
with (body mass)0.33. In contrast, empirical data for the scaling of r
and R indicate that EMA scales with (body mass)0.26(Biewener,
1989). As a result, our estimated costs for 30 kg chimpanzees and
humans are near the observed COT for our larger-bodied subjects,
in contrast to the expectation of increasing COT with smaller body
size. This departure from the empirically based expectation for
EMA is suitable for this investigation, since we seek to model
explicitly chimpanzee-like and human-like boundary conditions.
However, future work might explore the effect of different scaling
regimes on_Vmusc/m and COT.
Another important simplification in our model is the calculation
and use of a single measure of lfasc, r, and R for each joint and
extensor group. This approach has been used before (e.g., Roberts
et al., 1998b) and produces estimates of active muscle volume that
correspond well with locomotor cost (Fig. 2), but it condenses
a great deal of variation in musculoskeletal anatomy (e.g., Thorpe
et al., 1999). Future work might improve upon our model by
accounting for the relationship between joint angle and r, or by
testing the assumption that each muscle’s contribution in extensor
force is proportional to its PCSA. Further, while the approach we
used in estimating R has the advantage of incorporating the
segmental inertia and the effects of two-joint muscles into esti-
mates of active muscle volume, it is a departure from the simplest
means of calculating this variable (e.g., Fig. 1). The more involved
approach is useful here, as it enables the model to capture changes
in segmental inertia and two-joint muscles between the chim-
panzee-like and human-like boundary conditions. Still, since R
was not significantly different when calculated using either
method (p ¼0.07, Student’s two-tailed paired t-test), and since
calculating R using the simpler approach (Fig. 1) does not affect
the fit of the model (r2remains 0.95), the computationally simpler
method may be preferred in some studies. Regardless, the effects
of different approaches to estimating R warrant future investiga-
tion, since sensitivity analyses indicate that changes in R are
directly proportional to changes in estimated muscle activation
While long-term bipedal training may affect locomotion in
habitual quadrupeds (Preuschoft et al., 1988; Nakatsukasa et al.,
1995), the effect of bipedal training on chimpanzee walking kine-
matics and cost in this data set are likely small. As facultative
bipeds, the chimpanzees in this study did not need to be trained to
walk bipedally per se. Rather, the training regimen was directed
toward acclimating the chimpanzees to the treadmill and to
walking for adequately long duration to measure steady-state
oxygen use (Sockol et al., 2007). The percentage of each day spent
walking or standing bipedally for these chimpanzees was not
measured, but is likelyquitelow.Evenfor the duration of the Sockol
et al. (2007) study, training sessions rarely exceeded an hour, and
only a small portion of each session was spent engaged in bipedal
walking, with multiple rest breaks interspersed between treadmill
trials; chimpanzees spent the great majority of their day walking as
they chose, usually quadrupedally. Still, since the chimpanzees in
this data set perform occasionally for commercial purposes, they
likely spend more time bipedally over the course of their lives than
do wild chimpanzees (Hunt, 1992). Therefore, while joint angles
and ground force patterns appear broadly similar to those reported
for untrained chimpanzees (e.g., Kimura,1991), the chimpanzees in
this study may be more accustomed to walking bipedally than their
The use of chimpanzees as a model for the human-chimpanzee
LCA affects our analysis by setting a chimpanzee-like boundary
condition for our reconstructions of hominin anatomy, posture, and
cost. However, our approach is flexible, and can be adapted to other
models of the human-chimpanzee LCA by using anatomical data
from other apes (e.g., Payne et al., 2006a,b) or otherwise modifying
these variables. Forexample, Payne and colleagues (2006a,b) report
shorter fascicle length and longer moment arms for the hip flexors
of orangutans as compared to chimpanzees. This, in addition to the
more extended hind limb postures reported fororangutans (Thorpe
et al., 2007), would presumably lead to lower estimates of cost for
an orangutan-like model of a proto-hominin, although differences
in step length and other extensor groups would need to be
considered. More kinetic and kinematic data from other apes, as
well as more anatomical data for late Miocene apes, may help
improve and refine models of the human-chimpanzee LCA.
Walking cost in early hominins
Our approach produces estimates of COT for A.L. 288 that
correspond reasonably well with previous forward-dynamics
models of this specimen. Given the different sets of assumptions
that go into these different modeling approaches, the similarities in
estimated cost are notable. Interestingly, COT estimates produced
by computer simulations (Nagano et al., 2005; Sellers et al., 2005)
are somewhat higher than those produced by our human-like
boundary condition, the condition that most closely resembles the
gait of these forward-dynamics models. This difference in COT
between computer-modeling and experimentally-based estimates
has been noted previously (Sellers et al., 2005). As discussed by
Sellers and colleagues (2005), the increased COT from computer
simulations may be due to the absence of common energy-saving
mechanisms in these computer models, such as elastic energy
storage in the tendons. While these energy saving mechanisms are
not explicitly modeled using_Vmusc/m to predict COT, using the ratio
of_Vmusc/m to COT borne from experimental data may effectively
account for them by assuming, implicitly, that the same energy
saving mechanisms used by extant species are available to the
modeled fossil species.
Mechanical work estimates of walking cost (Kramer, 1999;
Wang et al., 2004) in A. afarensis are well belowour estimates based
on_Vmusc/m. This is perhaps unsurprising, as mechanical work is
typically a poor predictor of metabolic cost (Cavagna and Kaneko,
1977; Heglund et al., 1982). Still, while some have previously
cautioned against using mechanical work to estimate the metabolic
cost of walking (Wang et al., 2004), estimates of mechanical work
are often used as measures of efficiency and energy savings,
particularly inwalking gaits. These estimates of work may be useful
for comparisons of gait, but they cannot be translated directly to
metabolic cost because of the inherent inefficiency of muscle and
the complex ways in which metabolic power is related to
mechanical power in musculoskeletal systems.
Walking cost and the evolution of hominin bipedalism
Estimates of walking cost for the A.L. 288 specimen are generally
lower than those expected for a similarly sized quadrupedal
chimpanzee. Modest changes of less than 10% of the morphospace
between chimpanzee- and human-like boundary conditions were
sufficient to bring estimated COT below that for quadrupedal
chimpanzees (Fig. 4). Therefore most, if not all, reconstructions of
gait and posture in this species (Stern and Susman, 1983; Latimer,
1991; Stern, 2000) would result in lower walking costs relative
to quadrupedal apes. While the degree to which the posture of
H. Pontzer et al. / Journal of Human Evolution 56 (2009) 43–54 52
A. afarensis was crouched is not typically specified in studies
proposing a bent-hip, bent-knee gait for this species (Stern and
Susman, 1983; Stern, 1999, 2000), the more dorsally projecting
ischial attachment for the hamstrings (Robinson, 1972; Stern and
Susman, 1983) suggests a less crouched posture than that seen in
chimpanzees. Any change toward a more extended posture, in
conjunction with the longer hind limbs of A.L. 288 relative to
chimpanzees, would result in a lower cost of walking than seen in
quadrupedal apes. In contrast, because the hind limb of A.L. 288 is
short relative to modern humans (Jungers, 1982), estimated
walking cost remains above the human-like boundary condition
even when human-like values of lfascand EMA are used (Fig. 4).
These results, in addition to similar estimates of walking cost for
A.L. 288 from forward dynamics models (Nagano et al., 2005;
Sellers et al., 2005), suggest that, at least by 3 million years ago
hominin bipedalism was more economical than that of our ape-like
quadrupedal forebears, but not as economical as that of modern
A lack of adequate postcranial evidence makes it more difficult
to test hypotheses regarding walking cost in the earliest hominins.
The substantial, independent effects of lfasc, EMA, and hip height on
differences in COT between bipedal and quadrupedal chimpanzees
(Fig. 3) indicate that there are a number of morphological and
postural modifications that would have lowered the walking cost of
an early, ape-like bipedal hominin below that of a quadrupedal ape.
Further, these changes would not need to be dramatic to lower the
cost of hominin bipedalism below that of quadrupedal apes.
Interestingly, changes in EMA are expected to have a smaller initial
effect on COT than changes in lfascor hind limb length (Fig. 3),
suggesting that any selection for lower walking cost might be
expected to act on muscle morphology and limb length ahead of
posture. However, the relative labilityand heritabilityof these traits
Overall, our analyses indicate that no adaptive hurdle would
need to be overcome during the transition to bipedalism. While the
three chimpanzees studied here had greater bipedal walking costs,
our modeling of Vmusc and COT indicates that relatively minor
changes in gait, muscle morphology, and hind limb length, equiv-
alent to approximately 25% of the morphospace between human-
and chimpanzee-like boundaryconditions, wouldhavelowered the
cost of bipedal walking below that of quadrupedalism (Fig. 3).
Further, the observed variation in gait and cost in adult chimpan-
zees (Sockol et al., 2007; see above) suggests that bipedalism in
even the earliest, most primitive hominins could have been more
economical than quadrupedalism. Our results also suggest that
locomotor economy was an important selective force shaping
anatomy in early hominins. By 3 millionyears ago, with A. afarensis,
hominin walking cost was likely well below that of quadrupedal
apes (Figs. 3 and 4) due to increased hind limb length and pelvic
changes allowing greater hip and knee extension. Postcranial
evidence for earlier species is needed to test whether these same
adaptations are apparent in the earliest hominins.
At least 30 distinct skeletal features have been used to argue for
the efficiency or inefficiency of walking in A. afarensis (see Stern,
2000; their Tables 1 and 2), although most have yet to be linked
empirically to locomotor speed or cost. The model employed here
identifies limb length, posture and muscle moment arm length
(EMA), and muscle length (lfasc) as the primary variables deter-
mining locomotor cost. Despite the broad range of gaits and loco-
motor anatomies represented in our comparative sample, these
four variables captured 98% of the variation in locomotor cost
suggesting that while other features undoubtedly affect walking
efficiency, they likely play a relatively smaller role. As such, func-
tional analyses of hominin locomotor anatomy may do well to take
an explicitly hierarchical approach, placing greatest weight on the
anatomical variables known to account for the largest amount of
variation in performance. For example, quantitatively tying
anatomical traits such as ischial orientation to posture would
greatly improve our ability to model early hominin costs. Such an
approach might advance the current debate regarding terrestrial
and arboreal adaptations in A. afarensis and other early hominins.
Biomechanical analyses and reconstructions can only demon-
strate how the performance of a structure evolved, and perhaps
provide clues to its evolutionary success. But even if bipedalism in
early hominins was more economical, this need not be why it
evolved initially. Hominin bipedalism may have originated in
a different behavioral context in which energetic advantages were
not critical (see Thorpe et al., 2007). Indeed, morphological solu-
tions for decreasing locomotor cost, including increased hind limb
length, improved EMA, and shorter muscle fibers, are presumably
available to quadrupedal apes and would not require a transition to
bipedalism. The maintenance of long muscle fibers, short hind
limbs, and crouched postures in chimpanzees may suggest that
competing selection pressures, such as safety in the canopy
(Pontzer and Wrangham, 2004), outweigh selection for locomotor
economy in this species, and perhaps in the other non-human apes.
In contrast, the absence of an energetic hurdle to bipedalism,
combined with evidence that the earliest hominins may well have
reaped energy benefits (Fig. 3) and that hominins had undergone
selection for walking economy beyond that of quadrupedal apes by
the mid-Pliocene (Fig. 4), suggests that locomotor energy economy
was important in the success and persistence of the hominin
A.A. Biewener, D.E. Lieberman, J. Jones, and P. Rodman gener-
ously provided necessary equipment for this study. Three anony-
mous reviewers provided useful comments. This project was
supported by grants from the National Science Foundation BCS-
0424092 to M.D.S. and the L.S.B Leakey Foundation.
Biewener, A.A., 1989. Scaling body support in mammals: limb posture and muscle
mechanics. Science 245, 45–48.
Biewener, A.A., 2003. Animal Locomotion. Oxford University Press.
Biewener, A.A., Farley, C.T., Roberts, T.J., Temaner, M., 2004. Muscle mechanical
advantage of human walking and running: implications of energy cost. J. Appl.
Physiol. 97, 2266–2274.
Carey, T.S., Crompton, R.H., 2005. The metabolic costs of ‘‘bent-hip, bent-knee’’
walking in humans. J. Hum. Evol. 48, 25–44.
Cavagna, G.A., Kaneko, M., 1977. Mechanical work and efficiency in level walking
and running. J. Physiol. 268, 467–481.
Collins, S., Ruina, A., Tedrake, R., Wisse, M., 2005. Efficient bipedal robots based on
passive dynamic walkers. Science 307, 1082–1085.
Crompton, R.H., Li, Y., Wang, W.J., Gunther, M.M., Savage, R., 1998. The mechanical
effectiveness of erect and ‘‘bent-hip, bent-knee’’ bipedal walking in Austral-
opithecus afarensis. J. Hum. Evol. 35, 55–74.
Dart, R.A.,1925. Australopithecus africanus: the man-ape of South Africa. Nature 115,
Darwin, C., 1871. The Decent of Man and Selection in Relation to Sex. John Murray,
London. reprinted by Princeton University, 1981.
Donelan, J.M., Kram, R., Kuo, A.D., 2002. Mechanical work for step-to-step transi-
tions is a major determinant of the metabolic cost of human walking. J. Exp.
Biol. 205, 3717–3727.
Fedak, M.A., Rome, L., Seeherman, H.J., 1981. One-step N2-dilution technique for
calibrating open-circuit VO2measuring systems. J. Appl. Physiol. 51, 772–776.
Galik, K., Senut, B., Pickford, M., Gommery, D., Treil, J., Kuperavage, A.J.,
Eckhardt, R.B., 2004. External and internal morphology of the BAR 1002 ’ 00
Orrorin tugenensis femur. Science 305, 1450–1453.
Griffin, T.M., Kram, R., Wickler, S.J., Hoyt, D.F., 2004. Biomechanical and energetic
determinants of the walk–trot transition in horses. J. Exp. Biol. 207, 4215–4223.
Heglund, N.C., Fedak, M.A., Taylor, C.R., Cavagna, G.A., 1982. Energetics and
mechanics of terrestrial locomotion IV: total mechanical energy changes as
a function of speed and body size in birds and mammals. J. Exp. Biol. 79, 57–66.
Hildebrand, M., 1985. Walking and running. In: Hildebrand, M., Bramble, D.M.,
Liem, K.F., Wake, D.B. (Eds.), Functional Vertebrate Morphology. Harvard
University, pp. 38–57.
H. Pontzer et al. / Journal of Human Evolution 56 (2009) 43–5453
Hoyt, D.F., Wickler, S.J., Cogger, E.A., 2000. Time of contact and step length: the
effect of limb length, running speed, load carrying, and incline. J. Exp. Biol. 203,
Hunt, K.D., 1992. Positional behavior of Pan troglodytes in the Mahale Mountains
and Gombe Stream National Parks, Tanzania. Am. J. Phys. Anthropol. 87, 83–107.
Hutchinson, J.R., 2004. Biomechanical modeling and sensitivity analysis of bipedal
running ability. I. Extant taxa. J. Morphol. 262, 421–440.
Jenkins Jr., F.A., 1972. Chimpanzee bipedalism: cineradiographic analysis and
implications for the evolution of gait. Science 178, 877–879.
Jungers, W.L., 1982. Lucy’s limbs: skeletal allometry and locomotion in Austral-
opithecus afarensis (A.L. 288-1). Nature 297, 676–678.
Kimura, T., 1991. Voluntary bipedal walking in infant chimpanzees. Hum. Evol. 6,
Kram, R., Taylor, C.R., 1990. Energetics of running: a new perspective. Nature 346,
Kramer, P.A., 1999. Modelling the locomotor energetics of extinct hominids. J. Exp.
Biol. 202, 2807–2818.
Kramer, P.A., Eck, G.G., 2000. Locomotor energetics and leg length in hominid
bipedality. J. Hum. Evol. 38, 651–666.
Latimer, B., 1991. Locomotor adaptations in Australopithecus afarensis: the issue of
arboreality. In: Coppens, Y., Senut, B. (Eds.), Origine(s) de la bipedie chez les
Hominides. CRNS Editions, Paris, pp. 169–176.
Leonard, W.R., Robertson, M.L., 1997. Rethinking the energetics of bipedality. Curr.
Anthropol. 38, 304–309.
Marsh, R.L., Ellerby, D.J., Carr, J.A., Henry, H.T., Buchanan, C.I., 2004. Partitioning the
energetics of walking and running: swinging the limbs is expensive. Science
McHenry, H.M., 1992. Body size and proportions in early hominids. Am. J. Phys.
Anthropol. 87, 407–431.
Minetti, A.E., Ardigo, L.P., Reinach, E., Saibene, F., 1999. The relationship between
mechanical work and energy of locomotion in horses. J. Exp. Biol. 202, 2329–2338.
computer modeling and simulation of upright, straight-legged, bipedal locomo-
tion of Australopithecus afarensis (A.L. 288-1). Am. J. Phys. Anthropol.126, 2–13.
Nakatsukasa, M., Hayama, S., Preuschoft, H.,1995. Postcranial skeleton of a macaque
trained for bipedal standing and walking and implications for functional
adaptation. Folia Primatol. 64, 1–29.
Nakatsukasa, M., Hirasaki, E., Ogihara, N., 2006. Energy expenditure of bipedal
walking is higher than that of quadrupedal walking in Japanese macaques. Am.
J. Phys. Anthropol. 131, 33–37.
Nakatsukasa, M., Ogihara, N., Hamada, Y., Goto, Y., Yamada, M., Hirakawa, T.,
Hirakasi, E., 2004. Energetic costs of bipedal and quadrupedal walking in
Japanese macaques. Am. J. Phys. Anthropol. 124, 248–256.
Payne, R.C., Crompton, R.H., Isler, K., Savage, R., Vereecke, E.E., Gunther, M.M.,
Thorpe, S.K., D’Aout, K., 2006a. Morphological analysis of the hind limb in apes
and humans. I. Muscle architecture. J. Anat. 208, 709–724.
Payne, R.C., Crompton, R.H., Isler, K., Savage, R., Vereecke, E.E., Gunther, M.M.,
Thorpe, S.K., D’Aout, K., 2006b. Morphological analysis of the hind limb in apes
and humans. II. Moment arms. J. Anat. 208, 725–742.
Pilbeam, D.R., 1996. Genetic and morphological records of the Hominoidea and
hominid origins: a synthesis. Mol. Phylogenet. Evol. 5, 155–168.
Pontzer, H., 2005. A new model predicting locomotor cost from limb length via
force production. J. Exp. Biol. 208, 1513–1524.
Pontzer, H., 2007a. Predicting the cost of locomotion in terrestrial animals: a test of
the LiMb model in humans and quadrupeds. J. Exp. Biol. 210, 484–494.
Pontzer, H., 2007b. Limb length and the scaling of locomotor cost in terrestrial
animals. J. Exp. Biol. 210, 1752–1761.
Pontzer, H., Wrangham, R.W., 2004. Climbing and the daily energy cost of loco-
motion in wild chimpanzees: implications for hominoid locomotor evolution. J.
Hum. Evol. 46, 315–333.
Preuschoft, H., Hayama, S., Gu ¨nther, M.M., 1988. Curvature of the lumbar spine as
a consequence of mechanical necessities in Japanese macaques trained for
bipedalism. Folia Primatol. 50, 42–58.
Raichlen, D.A., 2004. Convergence of forelimb and hind limb natural pendular
periods in baboons (Papio cynocephalus) and its implication for the evolution of
primate quadrupedalism. J. Hum. Evol. 46, 719–738.
Richmond, B.G., Begun, D.R., Strait, D.S., 2001. Origin of human bipedalism: the
knuckle-walking hypothesis revisited. Am. J. Phys. Anthropol. S33, 70–105.
Richmond, B.G., Strait, D.S., 2000. Evidence that humans evolved from a knuckle-
walking ancestor. Nature 404, 382–385.
Roberts, T.J., Marsh, R.L., Weyand, P.G., Taylor, C.R., 1997. Muscular force in running
turkeys: the economy of minimizing work. Science 275, 1113–1115.
Roberts, T.J., Kram, R., Weyand, P.G., Taylor, C.R., 1998a. Energetics of bipedal
running: I. metabolic cost of generating force. J. Exp. Biol. 201, 2745–2751.
Roberts, T.J., Chen, M.S., Taylor, C.R., 1998b. Energetics of bipedal running: II. limb
design and running mechanics. J. Exp. Biol. 201, 2753–2762.
Robinson, J.T., 1972. Early Hominid Posture and Locomotion. Chicago University
Rodman, P.S., McHenry, H.M., 1980. Bioenergetics and the origin of hominid
bipedalism. Am. J. Phys. Anthropol. 52, 103–106.
Rubenson, J., Heliams, D.B., Maloney, S.K., Withers, P.C., Lloyd, D.G., Fournier, P.A.,
2007. Reappraisal of the comparative cost of human locomotion using gait-
specific allometric analyses. J. Exp. Biol. 210, 3513–3524.
Ruvolo, M., 1997. Molecular phylogeny of the hominoids: inferences from multiple
independent DNA sequence data sets. Mol. Biol. Evol. 14, 248–265.
Schmitt, D., 1999. Compliant walking in primates. J. Zool. Lond. 248, 149–160.
Sellers, W.I., Cain, G.M., Wang, W., Crompton, R.H., 2005. Stride lengths, speed, and
energy costs in walking of Australopithecus afarensis: using evolutionary
robotics to predict locomotion of early human ancestors. J. R. Soc. Interface 22,
Sellers, W.I., Dennis, L.A., Crompton, R.H., 2003. Predicting the metabolic
energy costs of bipedalism using evolutionary robotics. J. Exp. Biol. 206,
Sockol, M.D., Raichlen, D.A., Pontzer, H., 2007. Chimpanzee locomotor energetics
and the origin of human bipedalism. Proc. Natl. Acad. Sci. 30, 12265–12269.
Stern, J.T.,1999. The cost of bent-knee, bent-hip bipedal gait. a reply to Crompton, et
al. J. Hum. Evol. 36, 567–570.
Stern, J.T., 2000. Climbing to the top: a personal memoir of Australopithecus afar-
ensis. Evol. Anthropol. 9, 113–133.
Stern, J.T., Susman, R.L., 1983. Locomotor anatomy of Australopithecus afarensis. Am.
J. Phys. Anthropol. 60, 279–317.
Steudel-Numbers, K., 2003. The energetic cost of locomotion: humans and primates
compared to generalized endotherms. J. Hum. Evol. 44, 255–262.
Steudel-Numbers, K.L., Tilkens, M.J., 2004. The effect of lower limb length on the
energetic cost of locomotion: implications for fossil hominins. J. Hum. Evol. 47,
Susman, R.L., Stern, J.T., Jungers, W.L., 1984. Aboreality and bipedality in the hadar
Hominids. Folia Primatol. 43, 113–156.
Taylor, C.R., 1994. Relating mechanics and energetics during exercise. Comparative
vertebrate exercise physiology: unifying physiological principles. Adv. Vet. Sci.
Comp. Med. 38, 181–215.
Taylor, C.R., Heglund, N.C., McMahon, T.A., Looney, T.R., 1980. Energetic cost of
generating muscular force during running: a comparison of large and small
animals. J. Exp. Biol. 86, 9–18.
Taylor, C.R., Heglund, N.C., Maloiy, G.M.O., 1982. Energetics and mechanics of
terrestrial locomotion: I. metabolic energy consumption as a function of speed
and body size in birds and mammals. J. Exp. Biol. 97, 1–21.
Taylor, C.R., Rowntree, V.J., 1973. Running on two or on four legs: which consumes
more energy? Science 179, 186–187.
Thorpe, S.K.S., Crompton, R.H., Gunther, M.M., Ker, R.F., Alexander, R.M., 1999.
Dimensions and moment arms of the hind- and forelimb muscles of common
chimpanzees (Pan troglodytes). Am. J. Phys. Anthropol. 110, 179–199.
Thorpe, S.K., Holder, R.L., Crompton, R.H., 2007. Origin of human bipedalism as an
adaptation for locomotion on flexible branches. Science 316, 1292–1294.
Umberger, B.R., Gerritsen, K.G.M., Martin, P.E., 2003. A model of human muscle
energy expenditure. Comput. Methods. Biomech. Biomed. Engin. 6, 99–111.
Wang, W., Crompton, R.H., Carey, T.S., Gunther, M.M., Li, Y., Savage, R., Sellers, W.I.,
2004. Comparison of inverse-dynamics musculo-skeletal models of A.L. 288-1
Australopithecus afarensis and KNM-WT 15000 Homo ergaster to modern
humans, with implications for the evolution of bipedalism. J. Hum. Evol. 47,
Ward, C.V., 2002. Interpreting the posture and locomotion of Australopithecus
afarensis: where do we stand? Yearb. Phys. Anthpol. 45, 185–215.
Ward, C.V., Leakey, M.D., Walker, A., 2001. Morphology of Australopithecus ana-
mensis from Kanapoi and Allia Bay, Kenya. J. Hum. Evol. 41, 255–368.
Washburn, S.L., 1967. Behaviour and the origin of man. The Huxley Memorial
Lecture. Proc. R. Anthropol. Inst. Gr. Br. Ireland. 3, 21–27.
White, T.D., Suwa, G., Asfaw, B., 1994. Australopithecus ramidus, a new species of
early hominid from Aramis, Ethiopa. Nature 371, 306–312.
Wickler, S.J., Hoyt, D.F., Cogger, E.A., Hirschbein, M.H., 2000. Preferred speed and
cost of transport: the effect of incline. J. Exp. Biol. 203, 2195–2200.
Willems, P.A., Cavagna, G.A., Heglund, N.C., 1995. External, internal, and total work
in human locomotion. J. Exp. Biol. 198, 379–393.
Winter, D.A., 2005. Biomechanics and Motor Control of Human Movement, third ed.
Wiley, New York.
Zollikofer, C.P.E., Ponce de Leon, M.S., Lieberman, D.E., Guy, F., Pilbeam, D., Likius, A.,
Mackaye, H.T., Vignaud, P., Brunet, M., 2005. Virtual cranial reconstruction of
Sahelanthropus tchadensis. Nature 434, 755–759.
H. Pontzer et al. / Journal of Human Evolution 56 (2009) 43–5454