Available via license: CC BY 4.0
Content may be subject to copyright.
Article
Elephants evolved strategies reducing the
biomechanical complexity of their trunk
Highlights
dElephants evolved strategies reducing the biomechanical
complexity of their trunk
dObject transport is achieved by propagating inward curvature
from the trunk tip
dThe trunk can form rigid segments connected by pseudo-
joints for point-to-point motion
dThe trunk tip tangential velocity obeys a power law with its
path curvature
Authors
Paule Dagenais, Sean Hensman,
Val
erie Haechler, Michel C. Milinkovitch
Correspondence
michel.milinkovitch@unige.ch
In brief
Dagenais et al. uncover the biomechanics
of the elephant trunk, showing that
complex behaviors emerge from a finite
set of basic movements. Curvature,
torsion, and strain provide an efficient
kinematic representation to decompose
trunk trajectories into motion primitives.
The trunk forms pseudo-joints and its tip
follows a speed-curvature power law.
Dagenais et al., 2021, Current Biology 31, 1–11
November 8, 2021 ª2021 The Author(s). Published by Elsevier Inc.
https://doi.org/10.1016/j.cub.2021.08.029 ll
Article
Elephants evolved strategies reducing
the biomechanical complexity of their trunk
Paule Dagenais,
1,2
Sean Hensman,
3
Val
erie Haechler,
1
and Michel C. Milinkovitch
1,2,4,5,
*
1
Laboratory of Artificial and Natural Evolution (LANE), Department of Genetics and Evolution, University of Geneva, 30, Quai Ernest-Ansermet,
1211 Geneva, Switzerland
2
SIB Swiss Institute of Bioinformatics, 30, Quai Ernest-Ansermet, 1211 Geneva, Switzerland
3
Adventure with Elephants, Bela Bela, South Africa
4
Twitter: @LANEVOL
5
Lead contact
*Correspondence: michel.milinkovitch@unige.ch
https://doi.org/10.1016/j.cub.2021.08.029
SUMMARY
The elephant proboscis (trunk), which functions as a muscular hydrostat with a virtually infinite number of
degrees of freedom, is a spectacular organ for delicate to heavy object manipulation as well as social and
sensory functions. Using high-resolution motion capture and functional morphology analyses, we show
here that elephants evolved strategies that reduce the biomechanical complexity of their trunk. Indeed,
our behavioral experiments with objects of various shapes, sizes, and weights indicate that (1) complex
behaviors emerge from the combination of a finite set of basic movements; (2) curvature, torsion, and strain
provide an appropriate kinematic representation, allowing us to extract motion primitives from the trunk
trajectories; (3) transport of objects involves the proximal propagation of an inward curvature front initiated
at the tip; (4) the trunk can also form pseudo-joints for point-to-point motion; and (5) the trunk tip velocity
obeys a power law with its path curvature, similar to human hand drawing movements. We also reveal
with unprecedented precision the functional anatomy of the African and Asian elephant trunks using medical
imaging and macro-scale serial sectioning, thus drawing strong connections between motion primitives and
muscular synergies. Our study is the first combined quantitative analysis of the mechanical performance,
kinematic strategies, and functional morphology of the largest animal muscular hydrostat on Earth. It
provides data for developing innovative ‘‘soft-robotic’’ manipulators devoid of articulations, replicating the
high compliance, flexibility, and strength of the elephant trunk.
INTRODUCTION
The elephantproboscis (trunk) functions as a muscular hydrostat:
1
the coordinated contractions of antagonist muscles are translated
into torsion, bending,elongation, shortening, and stiffening, notvia
the support ofan articulated skeleton, but through shape changes
that rely almost entirely on the near-incompressibility (constant
volume) of the self-supporting trunk tissues. Remarkably, the
elephant trunk can perform very delicate tasks such as manipu-
lating a singleblade of grass, but it is alsocapable of carrying heavy
loads up to 270 kg.
2,3
The proboscis serves elephants in multiple
additional functions: breathing, olfaction, mechanosensation,
vocalization, posture-based communication, siphoning/spraying
water, sprinkling dust, and tool handling.
2–12
In principle, a
muscular hydrostat could exhibit a virtually infinite number of de-
grees of freedom. Hence, simplified control for task-specific tra-
jectories is needed to escape the so-called curse of dimension-
ality.
13
The composition of trajectories from building blocks
(motion primitives) provides a solution to reduce the behavioral
landscape complexity in muscular hydrostats.
14–19
Here, we combine behavioral, kinematic, and functional
morphology analyses of the elephant trunk and demonstrate
the existence of simplification mechanisms in its integrated
form and function. First, we provide evidence of compositionality
at the behavioral level by showing that the elephant, when facing
the task of grabbing and transporting with its trunk objects of
various shapes, sizes, and weights (Figure 1A), resorts to a finite
repertoire of basic strategies. Our results indicate that the short-
est ballistic trajectories (extension, bending, and twisting) are
concatenated into 17 primary behaviors and that changing the
size or mass of target objects induces robust transitions in
grasping strategies (Figures 1B–1E). Second, we use high-reso-
lution motion capture to track the trajectories of the continuously
flexible trunk and we construct an effective kinematic represen-
tation of prototypical sequences via spatiotemporal maps of cur-
vature, torsion, and longitudinal strain. Reproducible kinematic
features extracted from these trajectory maps are interpreted
as motor primitives, which can be linked directly to the underly-
ing synergetic muscle groups responsible for bending, twisting,
and elongating the trunk. We find that these kinematic primitives
are specific to tasks and/or phases of motion. For example,
when reaching for an object in front, longitudinal strain is the pri-
mary motor ingredient as the trunk elongates and retracts in a
modular fashion, whereas during the transport phase, the
Current Biology 31, 1–11, November 8, 2021 ª2021 The Author(s). Published by Elsevier Inc. 1
This is an open access article under the CC BY license (http://creativecommons.org/licenses/by/4.0/).
ll
OPEN ACCESS
Please cite this article in press as: Dagenais et al., Elephants evolved strategies reducing the biomechanical complexity of their trunk, Current Biology
(2021), https://doi.org/10.1016/j.cub.2021.08.029
dominant feature is the propagation of an inward curvature front.
Conversely, when reaching for a target to the side, the trunk can
form rigid segments connected by pseudo-joints, analogous to
an elbow and wrist in an articulated skeleton.
A fundamental problem in analyzing neural control (and in
robotics) is to identify how a specific movement is chosen,
among all possibilities, to achieve a given point-to-point
displacement. A common answer is that motion primitives
might be selected as a compromise between cost func-
tions,
17,20
such as speed versus accuracy,
21
precision versus
efficiency,
22
or velocity versus path curvature.
22–27
In this
perspective, we make a remarkable observation: in all analyzed
trajectories, regardless of the specificities of the task, the 3D
tangential velocity of the trunk tip obeys a power law with its
path curvature, similar to human hand drawing movements in
2D,
24
but with a different scaling exponent. Hence, we propose
that the geometry of the path governs the kinematics in
elephant trunk trajectories.
Moreover, we suggest that the passive activity of cross-helical
connective fibers embedded in different planes, described here
for the first time in the trunk, might reduce the trunk energetic
expenditure when working through antagonist muscle groups.
Finally, a non-trivial relation between longitudinal and cross-
sectional strains suggests the contribution of nostril dilation
and contraction during trunk elongation and shortening.
A
B
D
C
E
Figure 1. Transitions in prehension strate-
gies of an African elephant trunk with intact
or severed tip fingers (target objects placed
on the ground)
(A) Schematic representation of the experiments.
(B–E) Frequency of occurrence of each behavior
for target objects of increasing dimension, (B and
D) d
1
= 3.75 cm, d
2
= 7.5 cm, d
3
= 15 cm, and d
4
=
30 cm, or increasing mass density, (C and E) m
1
=
500 kg/m
3
(wood), m
2
= 3,000 kg/m
3
(aluminum),
and m
3
= 8,000 kg/m
3
(steel), for an elephant with
intact (B and C) or missing (D and E) trunk tip fin-
gers. The color scale is adjusted to highlight be-
haviors for which frequencies are superior to 20%.
The full repertoire of behaviors is also illustrated in
Video S1. Note that the ‘‘distal twist’’ behavior had
a non-zero frequency of occurrence when objects
were handed from above the ground (Data S1B).
See also Tables S1 and S3,Video S1, and Data
S1A.
In brief, the elephant trunk illustrates
how simplified strategies can be adopted
by muscular hydrostats to negotiate with
their intrinsic complexity. A striking paral-
lel can be drawn with the octopus arm,
which has independently evolved the for-
mation of virtual joints during point-to-
point motion
28–30
and the propagation of
a bend when reaching for a target.
31–34
As these two lineages (elephants and oc-
topi) are separated by nearly a billion
years of evolution and rely on completely
different nervous systems (a centralized
cerebral motor cortex in the elephant versus a neural control
largely de-localized in the arms in the octopus
35
), it is tempting
to conjecture that these strategies (making pseudo-joints and
propagating a curvature front) are exceptionally efficient at the
level of biomechanics and/or neuronal control.
We discuss our results in the context of so-called soft robotics,
an emerging new paradigm with the ambition of developing
robotic manipulators replicating the high compliance, flexibility,
and strength of natural hydrostats.
36–40
Web applications are
publicly available at https://www.lanevol.org/projects/proboscis.
RESULTS
Behavioral strategies
The behavioral experiments (Figure 1A) were performed with two
adult male African elephants: one with an intact trunk and the
second missing its prehensile tip fingers. The animals were moti-
vated to pick and execute a point-to-point transport of various
objects of different shapes, sizes, and densities (Table S1).
Marker-based motion capture allowed us to track 3D trajectories
of the trunk with high accuracy (positional uncertainty < 2 mm).
Basic trunk behaviors and adaptation
The elephant trunk exhibits an impressive versatility in its pre-
hension capabilities, which is built upon a finite set of motor stra-
tegies, combined to compose complex trajectories. Here, we
ll
OPEN ACCESS
2Current Biology 31, 1–11, November 8, 2021
Please cite this article in press as: Dagenais et al., Elephants evolved strategies reducing the biomechanical complexity of their trunk, Current Biology
(2021), https://doi.org/10.1016/j.cub.2021.08.029
Article
identify 17 basic behaviors manifested by the trunks of African
elephants during the prehension phase of motion (Figure 1;
Video S1;Data S1A). Using this ethogram across all 550 suc-
cessful prehension sequences recorded for both animals, we
estimated the probability of executing a specific basic behavior
when presented with a specific shape (Figures 1B–1E).
We discovered that varying the size or the mass of the objects
induces transitions in behaviors, and that such transitions are
robust across different geometrical shapes. Figure 1B shows a
size-induced behavioral transition: for the smallest objects (d
1
),
the animal tends to either pinch them between the two tip fingers
or grasp them with a full tip-grip from the side. The finger pinch is
shape-specific and most probable for the thin cylinder. Less
probable for d
1
objects is a tip-grip from the top, which becomes
the single most preferred strategy for slightly larger objects (d
2
),
and one of two preferred prehension strategies (with suction,
equally probable) for yet bigger (d
3
) objects. A distal flip is very
frequently used to secure the transport phase after a d
1
to d
3
ob-
ject has been picked from the ground. The animal transitions to a
distal wrap of the trunk around d
4
objects to pick them up from
the ground. Bowing the head down in the direction of the object,
pressing it against the ground, or sweeping it closer were some-
times used for largest objects. More rarely, the animal used one
foot to push the big cylinder against its trunk or wedge it against
the tusks before releasing it to the ground in a putative playful
behavior.
Second, when varying the mass of the manipulated object, we
also observed a transition in prehension strategies (Figure 1C),
although not as sharp as when changing the size, suggesting
that the tested range of weights was less of a constraint than
the tested range of sizes. The combination of tip-grip from the
top and distal flip is the preferred strategy for the spheres and
cylinders of all tested masses, except for the heaviest cylinder,
for which the distal wrap becomes equally probable. Note also
that the heaviest objects (m
3
) were most often ‘‘kicked’’ and/or
swept closer by the trunk before grasping and lifting. During
the manipulation of disks, the animal systematically replaced
tip-grip by suction, a remarkable solution to the challenge of
grasping such a particular object placed flat on the ground.
Note that this suction behavior tends to be followed by the usual
distal flip. The animal also tends, with a probability proportional
to mass density, to sweep the disk closer to the body and to
use ground relief as a lever to grab it. Other parameters, such
as the initial position of the target and the texture of the object,
were also tested (Data S1B), but had less influence on the
prehension strategies.
The second adult male, whose trunk tip fingers are missing,
provided an exceptional opportunity to analyze behavioral adap-
tations. As this animal could not perform suction, tip-gripping, or
finger pinching, it displayed a more restrained repertoire of be-
haviors (Figures 1D and 1E). The core strategy for this animal
was the distal wrap, which was even used for grabbing very small
objects. The foot was also used more often, with greater coordi-
nation, and in situations unobserved for the uninjured elephant,
e.g., pushing the disk with the foot inside the loop of the distally
wrapped trunk.
Learning
The ethological graphs (Figures 1B–1E) do not capture the
chronology of the sequences or the strategies that failed.
Some interesting transitions occurred over time as the same
tasks were repeated: the animals tended to first try the strate-
gies that had worked previously with similar objects, and if
they encountered a failure (i.e., if the object was dropped),
they switched to other tactics until achieving a successful
transport. This phenomenon illustrates the capacity of ele-
phants for learning to execute new tasks in only a few trials
(Data S1C). The largest cone (d
4
) offered a great example of
this learning process: after three attempts, the elephant with
uninjured trunk tip figured out a robust solution to lift and carry
this unusual shape (Video S2). Finally, despite the fact that the
release phase is more difficult to categorize, due to accidental
drops and spontaneous (possibly playful) movements, it also
involves interesting behavioral transitions, adaptation, and
learning (Data S1C).
These results suggest that an elephant goes down an inter-
nally defined list of grabbing strategies or, for more complex
shapes such as the large cone, eventually explores creative
strategies until successful. The apparent complexity of
grasping behaviors can thus be reduced to a smaller set of
basic modules, accounting for the ability to quickly learn new
strategies by recombining a pre-defined set of primitive
movements. It is likely that such successive choices involve
compromises among gain/cost functions, e.g., the probability
of a successful prehension and transport, energetic input,
number of independent muscle fascicles to activate, or
duration of the sequence.
Motion primitives and kinematic invariant
We decompose trajectories into five phases of motion: (1) reach-
ing, (2) prehension, (3) transport, (4) release, and (5) reaching
back. We use ellipses-and-backbone reconstructions to
quantify the trunk conformations in 3D (Figures 2A and 2B) and
extract the dominant motion primitives in each trajectory phase.
All sequences analyzed here and virtual animations of the African
elephant trunk with projected heatmaps of kinematic variables
are available in the online supplemental material (Video S2;
online application 1). We first examine a prototypical sequence,
in which the trunk uses suction to lift and transport a wooden disk
(Video S2).
Multi-axis deformations
Cross-sectional and longitudinal strains compensate (Figure 2C)
as expected for a muscular hydrostat. Longitudinal strain moves
the trunk end effector front and back without moving the head,
and some distal shortening is associated with lifting the trunk
up. The profile of longitudinal strain is symmetrically reproduced
between the right-to-left and left-to-right sequences (Figure 2D):
the whole trunk elongates during the reaching phase up to a
maximal strain (>30%) close to the tip (at the trunk index
ix0.85, i.e., 85% of the total distance between the first and
last rows of markers). The elongation between ix0.25 and
ix0.55 is maintained for most of the sequence. On the other
hand, the 40% most distal part of the trunk (0.6 (i(1) retracts
during the transport phase but elongates again during the
release phase. While reaching back, most of the trunk progres-
sively shortens. Compared to the analogous octopus arm, which
uses a distally propagating bend when reaching for a target,
31–34
the elephant adopts a simpler strategy: modular elongation in the
direction of the object.
ll
OPEN ACCESS
Current Biology 31, 1–11, November 8, 2021 3
Please cite this article in press as: Dagenais et al., Elephants evolved strategies reducing the biomechanical complexity of their trunk, Current Biology
(2021), https://doi.org/10.1016/j.cub.2021.08.029
Article
Curvature and torsion
External variables describing the trunk position, orientation,
and velocity are shown in Figure S1. This extracorporeal repre-
sentation is useful to disentangle the different phases of motion
and to distinguish between planned and correctional actions
(Data S1D). However, it is tedious to read, and we show below
that curvature and torsion constitute a more appropriate param-
eter space (Figure S2) as they greatly facilitate interpretation of
the trunk movements (Figure 3;Data S1E).
In the majority of point-to-point transport tasks, the trunk
displacement mainly occurs in the sagittal plane; hence, the
dominant ingredients of motion are found in the up-down curva-
ture knb, as shown in Figure 3 for prehension tasks with five ob-
jects (cylinders with increasing diameter, d
2
,d
3
, and d
4
; with
increasing mass, m
1
and m
2
; and a sphere with diameter d
4
;
Video S2). The rate of the sagittal curvature front displacement
offers a simple framework to distinguish among categories of
movements. Figures 3A–3C illustrate how the trunk adapts to
cylinders of increasing size (diameter d
2
,d
3
, and d
4
). The bend
typically occurs in two steps. First, the most distal part of the
trunk is bent inward during the prehension phase followed by
proximal propagation of this curvature front with similar rates
(z0.3 s
1
) for all sizes. However, when grasping the largest cyl-
inder (d
4
), an intermediate step of much slower curvature propa-
gation (z0.06 s
1
) is observed. Note that the inward bend prop-
agates more proximally for larger objects (30%, 40%, and 60%
for d
2
,d
3
, and d
4
, respectively). For d
4
, the inward curvature is
delimited by a sharp inflection point on the trunk, i.e., positive
curvature (upward bend) is observed in the remaining proximal
part, resulting in a S-shaped trunk. Moreover, the maximally
curved posture is maintained for longer during the transport
phase. Finally, relaxation from the curved state is quite
A
CD
BFigure 2. African elephant trunk using suc-
tion to lift and transport a wooden disk
(A) Marker 3D positions superimposed on the
corresponding 2D image.
(B) Virtual backbone and ellipses fitted to the 3D
markers positions, for six equally distributed time
points.
(C) Multi-axis deformations of the trunk: cross-
sectional strain (DD=Dowith Dthe diameter aver-
aged between the large and small axes of the el-
lipses) and longitudinal strain (DL=Lo) compensate
due to the constant volume constraint inherent to
a muscular hydrostat.
(D) Spatiotemporal map of longitudinal strain
along the trunk (horizontal black curves indicate
markers positions); the five phases of motion (i,
reaching; ii, prehension; iii, transport; iv, release; v,
reaching back) are indicated. The color scale bar
indicates the amplitude of the positive or negative
longitudinal strain.
See also Figures S1,S2, and S4 and Video S2.
symmetric for d
2
and d
3
, whereas it hap-
pens in two steps for d
4
, the first one with
moderate rate (0.11 s
1
), followed by an
abrupt release. The weight of the object
also influences the curvature front propa-
gation: in comparison to the light cylinder
(Figure 3A), manipulation of a heavier cyl-
inder (Figure 3D) causes a faster and more proximal inward bend
propagation, and the trunk stays in its maximally curved state for
a longer time.
Figure 3E represents an impressive behavior executed while
maneuvering the small wooden cylinder (d
2
,m
1
): after having
picked up the object with a tip grip, the animal transferred it to
the ventral part of its trunk and performed a full-trunk inward
bend (bringing the tip fingers up to the mouth), while the cylinder
rolled in the middle portion of the curled trunk, nonchalantly in-
termingling two complex activities (point-to-point trunk tip
displacement and point-to-point object transport).
The kinematic profiles of Figures 3C and 3F illustrate the differ-
ence of strategy for grabbing a cylinder and a sphere (both with
the same large diameter d
4
) using a vertical and oblique wrap,
respectively. The more challenging sphere involved a much
longer prehension phase with the curvature propagating at a
particularly low rate, immediately followed by a two-step
relaxation.
These curvature maps exemplify how intrinsic kinematic vari-
ables can encode the unfolding of trunk conformations in serial
phases of motion. The propagation of a curvature front initiated
at the tip suggests the existence of a propagating front of muscle
activation, analogous to what has been discovered in the
octopus arm during the reaching phase.
31–34
Joint-like twist
A ‘‘joint-like twist’’ movement was often recorded while the
trunk reached behind to grab a treat (Figure 4A; Video S2): a
sharp bend about halfway through the trunk length (with the
ventral side of the trunk facing up distally to the joint), followed
by a second bend before the tip of the trunk (with fingers wide
open), creating the illusion of an arm with an elbow and a wrist.
Three independent kinematic ingredients are necessary to
ll
OPEN ACCESS
4Current Biology 31, 1–11, November 8, 2021
Please cite this article in press as: Dagenais et al., Elephants evolved strategies reducing the biomechanical complexity of their trunk, Current Biology
(2021), https://doi.org/10.1016/j.cub.2021.08.029
Article
properly encode this posture (Figure 4A, right panel): the abrupt
development of a sideway curvature front ðknaÞis combined
with a gradual accumulation of torsion ðtbaÞ, whereas the in-
ward curvature (knb < 0) exhibits two maxima (one just before
the first virtual joint and a second one at the very tip, contrib-
uting to the second virtual joint). The longitudinal strain com-
pletes the picture: the trunk is contracted by 10%–15% every-
where except at the very tip, where it is close to zero. Elephants
were previously described to form a joint at the distal part of the
trunk forming a stiff pillar to apply vertical force and pick up
granular food.
41
Once more, a parallel can be drawn with the
octopus arm, whose solution for point-to-point movements
(food to mouth) is a combination of virtual joints and stiffened
segments.
28,30
Sharp self-wipe twist
Another behavior involving a spectacular twist was observed
when the elephant attempted in early experiments to remove
markers by using the trunk to wipe itself (Figure 4B; Video S2).
Note that the capacity of a flexible arm to clean itself is of prime
interest from a robotics point of view. In the first half of the trajec-
tory, the trunk shortens homogeneously across its length (by
10%). Halfway through, the distal part elongates by 20% and
the proximal part (until ix0.6) relaxes back to the reference
length. Concomitantly, (1) a front of sagittal curvature (knb<0)
is initiated at the tip and propagates very fast up to half the trunk
length, (2) torsion (tba) is generated clockwise in a large interval
0.5 < i< 0.9 and anti-clockwise more locally near the tip, and
(3) a slight curvature toward the left (kna< 0) adds the final touch
to the posture (Figure 4B, right panel).
Velocity and path curvature relationship
One key result of our analyses is that trunk tip trajectories exhibit
a power law between 3D tangential velocity and the path curva-
ture (i.e., vtip kp
tip with px2/3; Figure 5). Other values of the
scaling exponent have been measured in various biological sys-
tems such as human hand drawing movements (p=1/3),
24
hu-
man locomotion paths (pwithin [0.26, 0.13]),
25
human tongue
movements (pwithin [0.55, 0.33]),
42
and Drosophila larval
locomotion (pwithin [0.24, 0.22]).
26
Different arguments
have been proposed to link these kinematic invariants with con-
trol aspects of motion such as minimization of trajectory vari-
ance, of jerk, or of mechanical power.
22,24,27
Escaping the high-dimensionality problem
In brief, complex trunk trajectories are decomposed into repro-
ducible piece-wise primitives expressed in the profiles of curva-
ture, torsion, and longitudinal strain. These motion primitives are
highly relevant from a motor control point of view as they can be
mapped directly to the activity of specific muscle groups respon-
sible for bending, twisting, and elongating the trunk. Certain
A
D
B
E
C
F
Figure 3. Trunk curvature maps (knb ) for six trajectories of the African elephant trunk maneuvering different target objects
(A) Grip and flip (cylinder d
2
,m
1
).
(B) Distal wrap in the vertical plane (cylinder d
3
).
(C) Distal wrap in the vertical plane (cylinder d
4
).
(D) Grip and flip (cylinder d
2
,m
2
).
(E) Grip and flip + full bend (cylinder d
2
,m
1
).
(F) Distal wrap in the oblique plane (sphere d
4
). Horizontal black curves indicate marker positions. The five phases of motion are as in Figure 2D. Propagation rates
of the curvature front are indicated by dashed lines with corresponding values (in s
1
). The color scale bars indicate the amplitude of the positive or negative
curvature. See STAR Methods (Kinematic calculations) for the definition of knb .
See also Table S1,Video S2, and online application 1.
ll
OPEN ACCESS
Current Biology 31, 1–11, November 8, 2021 5
Please cite this article in press as: Dagenais et al., Elephants evolved strategies reducing the biomechanical complexity of their trunk, Current Biology
(2021), https://doi.org/10.1016/j.cub.2021.08.029
Article
mechanical and kinematic parameters uncovered here (Table
S2;Data S1F) can be directly used in a soft robotics biomimetic
approach.
Muscular synergies
Using state-of-the-art computer tomographic (CT) scan, mag-
netic resonance imaging (MRI), an in-house developed
macro-scale serial-sectioning method (STAR Methods), and his-
tology, we characterized the anatomy of the African and Asian
elephant trunks in unprecedented detail. A combination of longi-
tudinal, radial, transversal, and oblique muscle groups, traversed
and clothed by nervous and connective tissues
3,43,44
and wrap-
ped in a hyperkeratinized skin,
45
have been previously identified
in the elephant trunk. Here, we produced high-resolution images
(30 microns/pixel) of transversal (Figure 6), sagittal, and coronal
(Figure 7) sections of the trunks as well as MRI virtual sections.
MRI series are also presented in Videos S3 and S4. In addition,
we generated high-resolution polygon meshes of (1) the skin sur-
face (Figure S3A) and (2) the six muscle groups (Figure S3B). In
the online supplemental applications 2–9, the complete series
of anatomical sections can be examined at full resolution and
meshes can be manipulated in 3D. Anatomists from the previous
century have reported up to 150,000 muscles in the Asian
elephant trunk,
3
but our images indicate that they were in fact
counting muscle fascicles. Below, we explore the anatomy of
the trunk in relation to each specific mechanical function of
muscular hydrostats,
1
which present a one-to-one correspon-
dence with the intrinsic kinematic variables described above
(strain, curvature, and torsion).
Elongation and shortening
The basic function of the radial dorsal, radial ventral, and trans-
versal (rectus nasi
43
) muscle groups (in red, dark blue, and green,
respectively, in Figures 6 and S3B) is to reduce the diameter of
the trunk, inducing its passive elongation. Muscle groups work
in antagonist pairs: contraction of the dorsal longitudinal group
(maxillo-labialis,
43
orange in Figures 6 and S3B) or the superficial
and deep oblique groups on the ventral side (pars rimana and
pars supralabialis,
43
in pink and pale blue, respectively, in Fig-
ures 6 and S3B) causes shortening of the trunk, inducing a
passive increase in diameter. We measured an exceptional
elongation capacity of the trunk of 135% (length ratio between
the longest and shortest trunk states).
The muscles’ relative volumes are very well conserved be-
tween the two species, with 28% and 24% of the total trunk
A
B
Figure 4. Special movements of the African elephant trunk
(A) Joint-like twist to reach behind for food and (B) self-wipe sharp twist. For (A), the second (distal) bend occurs close to the last row of markers, such that it is
much more visible in recorded videos (left panel) than in motion capture profiles (central panel). Left: 2D video images with overlaid marker 3D positions. Center:
virtual backbones at six equally distributed time points. Right: signature maps (color bars indicate amplitude) of torsion and curvature (tba ,knb , and kna as defined
in the Kinematic calculations section of the STAR Methods) encoding the main features of motion (instantaneous distributions, A; spatiotemporal maps, B). See
also Video S2 and online application 1.
ll
OPEN ACCESS
6Current Biology 31, 1–11, November 8, 2021
Please cite this article in press as: Dagenais et al., Elephants evolved strategies reducing the biomechanical complexity of their trunk, Current Biology
(2021), https://doi.org/10.1016/j.cub.2021.08.029
Article
volume occupied by the longitudinal muscle group for the African
and Asian elephants, respectively; 10% and 15% by the super-
ficial oblique; 14% and 15% by the deep oblique; 28% and
21% by the radial dorsal; 13% and 15% by the radial ventral;
and 7% and 10% by the transversal muscle group. Note that
the proximal end of the longitudinal muscle group attaches ante-
riorly on the cranium
43
(Video S3), similarly to the protruding
tongues of certain animals (lizard, armadillo, opossum, and
pangolin) that originate outside the oral cavity.
1,46
The relation between longitudinal and cross-sectional strain is
captured by four components of the strain-dependent Poisson’s
ratio (n), which were estimated from the graphs of Da=aoversus
DL=Loand Db=boversus DL=Lo(where aand bare, respectively,
the radii of the large and small axes of the ellipse cross-section):
naL =0:33, nbL =0:28, nLa =0:38, and nLb =0:35 (Figures S4D–
S4F; Data S1G). These values reflect approximate isotropy be-
tween the cross-sectional and longitudinal directions, as well
as between the two antagonist processes of elongation and
shortening. It is also consistent with the maximal change in vol-
ume of DV=Vo±5% for DL=Lo±15%, which we recorded
(Data S1G). The apparent compressibility of the trunk (nLa +
nLbs1) might be due to the presence of the nostrils, two air-filled
cavities with variable cross-sectional areas.
47
For example,
dilating the nostrils while contracting the outer diameter would
contribute to the elongation of the trunk without being accounted
for in the overall diameter change (DD=Do), thus causing
Poisson’s ratios to be underestimated.
Embedded crossed-fiber connective tissues play a major role
in shape control (limiting deformations) and elastic recoil during
antagonist movements.
46,48,49
Here, we show that intramuscular
connective fibers in the transversal plane (Figures 6 and S5B;
Data S1H) might store potential energy when the cross-sectional
area of the trunk is passively increased, allowing the trunk to
elongate again with less active force. Note that histological sec-
tions of the large nerve bundles running longitudinally into the
trunk (Figures 6,7,S5C, S6D, and S6E) reveal an undulated
arrangement (Figure S6E) that allows the nerves to sustain large
trunk elongation without undergoing strain-induced damage
(Data S1H).
Torsion
The fiber angle of the oblique muscle groups (measured from the
longitudinal axis) is determinant for the mechanical behavior:
below or above the critical angle (qc=54440), their contraction
induces, in addition to torsion of the trunk, its overall shortening
or elongation, respectively.
1
The superficial and deep oblique
muscles show angles q=20–30(Figure 7A). Hence, these
are also trunk shorteners. The concomitant 10%–15% nega-
tive strain that we observed during the joint-like twist and the
sharp self-wipe twist (where torsion plays an important role) is
compatible with this suggestion.
Oblique muscles run with opposite angles on the left and right
sides of the trunk, forming a V-shape with the open side facing
the tip of the trunk for the superficial oblique group and with
the pointy side facing the tip for the deep oblique group. Given
A
D
B
E
C
F
Figure 5. Power law between tangential velocity and path curvature at the trunk tip
Log-log graphs of vtip (in m/s) versus ktip (in m
1
) measured in the African elephant trunk during six motion sequences, while maneuvering (A) a wooden disk, (B) a
wooden cylinder of medium size (d
3
), (C) a wooden cylinder of large size (d
4
), (D) a wooden cone of large size (d
4
), and (E and F) during two different instances of the
‘‘joint-like twist’’ behavior. Fitted affine functions (black lines) indicate a power law vtip kp
tip with px2/3. The average scaling exponent was obtained from 20
different trajectories. Note that the same relation was observed regardless of the object’s attributes and also for trajectories not involving the transport of an
object.
ll
OPEN ACCESS
Current Biology 31, 1–11, November 8, 2021 7
Please cite this article in press as: Dagenais et al., Elephants evolved strategies reducing the biomechanical complexity of their trunk, Current Biology
(2021), https://doi.org/10.1016/j.cub.2021.08.029
Article
that these fibers follow the curvature of the trunk surface, the
right and left superficial oblique muscles form a left-handed
and right-handed helix fiber arrangement, respectively, and
inversely for the deep oblique muscles. Therefore, torsion of
the trunk can be amplified by the activation of the fibers with
the same helix-handedness, i.e., in different muscle groups on
opposite sides of the trunk. Note that, because of its more
external location from the central axis, the superficial oblique
group generates larger torque ( r
!3F
!) than the deep oblique
muscles, for an equal force.
The superficial and deep oblique groups are tightly entangled
at their junction, displaying a braided design of alternating mus-
cle strands (Figure 7A). The confinement of the muscle groups
responsible for torsion to the ventral side might be a morpholog-
ical adaptation granting greater dexterity to that side of the trunk,
specialized in wrapping, squeezing, holding, transporting, and
pressing items against surfaces. The stress generated by the
ventral oblique muscles can be transmitted to the dorsal side
by the connective tissue wrapped around the muscular core,
with a cross-helical arrangement visible in the most lateral
parasagittal sections of the trunk (Figure 7B), at the junction of
the obliques and longitudinal muscles (with helix angle jx
42). Moreover, the ventral side presents a concave surface
(akin to the palm of a hand), covered with large transversal wrin-
kles (Figure S3A), which aid in gripping and holding objects.
Bending
There are two possible mechanisms of bending: either via longi-
tudinal contraction on one side (the dorsal longitudinal or the
ventral obliques for an upward or inward bend, respectively, or
an appropriate lateral combination of those three groups for
sideways curvature) while actively keeping a constant diameter
by contraction of cross-sectional muscle groups (transversal
and/or radial), or via a reduction of the diameter while actively
keeping a constant length on one side, resulting in a passive
elongation on the convex side of the bend.
1
Accordingly, we
have shown that upward curvature (knb > 0) is accompanied by
a dorsal longitudinal shortening (DL=Lo< 0), and vice versa. For
trajectories involving the propagation of an inward bend from
the tip, we typically observe a negative peak in the longitudinal
strain of about 15% at the location of the inflexion point (where
knb changes sign or goes to zero), indicating that the dorsal lon-
gitudinal muscle group is locally shortened at the curvature wave
front.
Activating both sides (left and right) of the same oblique group
(deep or superficial) or both oblique groups (superficial and
deep) on the same side (left or right) with equal strength would
suppress any undesired torsional effect due to their opposite
handedness. The external location of the longitudinal/oblique
groups allows them to generate greater moment arm for
bending. Note that intramuscular connective tissue fibers
oriented in the transverse plane (Figures 6 and S5B) might help
the trunk to bend with more energetic efficiency, by resisting
the passive increase in diameter imposed by the unilateral longi-
tudinal contraction, and thus reducing the amount of active force
needed from the cross-sectional muscle groups. The muscle ar-
rangements in the dorsal and ventral tip fingers (Figure S7) reflect
a greater ability for bending (Data S1I), compared to a reduced
capacity for torsion.
Stiffening
One additional muscular synergy consists in multi-axial contrac-
tions allowing the elephant to produce localized stiffening of the
trunk. During our prehension experiments, we observed that the
ventral side of the trunk sometimes applied downward forces on
objects, prior to a distal wrap. In this case, stiffening allows
resisting diametrical compression. It is likely that the animal is
finely adjusting the local stiffness of its trunk to control the
surface contact with the manipulated object.
DISCUSSION
In the present study, we combined kinematic analyses with
anatomical measurements to uncover the fundamental
Figure 6. Muscular anatomy of the African
and Asian elephant trunks
Transversal sections from the tip (top panels),
center (central panels), and proximal base (lower
panels) of the trunks of an Asian (E. maximus, left
halves) and African bush (L. africana, right halves)
elephant. Left column: serial sections of frozen
trunks. Center: MRI virtual sections. Right: MRI
virtual sections with the six muscle groups su-
perimposed: longitudinal (L, orange), radial dorsal
(rd, red), radial ventral (rv, dark blue), transversal (t,
green), deep oblique (do, pale blue), and superfi-
cial oblique (so, pink). Intramuscular connective
tissue fibers are visible as white filaments in
photographed sections (left column) and are
schematized as white lines in the central column.
Arrows: nerve bundles. See also Figures S3 and
S5–S7,Data S1H and S1I, Videos S3 and S4, and
online applications 2 and 5.
ll
OPEN ACCESS
8Current Biology 31, 1–11, November 8, 2021
Please cite this article in press as: Dagenais et al., Elephants evolved strategies reducing the biomechanical complexity of their trunk, Current Biology
(2021), https://doi.org/10.1016/j.cub.2021.08.029
Article
biomechanics of the elephant trunk. First, we quantified the
prevalence of 17 basic prehension strategies of African elephant
trunks in point-to-point transport tasks, reflecting the composi-
tionality of motion as a crucial simplification mechanism for con-
trolling this muscular hydrostat. Our results highlight robust
behavioral transitions as a function of the size or mass of the
target objects. The catalog of actions reported here only encom-
passes a small subset of all possible trunk behaviors and many
other movements would be worth investigating, especially
focusing on the trunk tip region. Second, we reconstructed the
dynamics of 3D conformations of the African elephant trunk
from high-resolution motion capture data, and we showed that
specific actions are translated as signature patterns in the
spatiotemporal maps of strain, torsion, and curvature. Out of a
virtually infinite conformational space, we identified simple kine-
matic solutions accounting for distinct trajectory phases. This
suggests a potential mechanistic solution for a nervous system
to control the complexity of muscular hydrostats. In particular,
the elephant adopts basic strategies such as successively elon-
gating and retracting specific portions of the trunk when reach-
ing for a target in front, or propagating an inward curvature front
(initiated at the tip) during the transport of objects. We also re-
vealed that the elephant trunk can form functional joints when
achieving point-to-point displacements, reminiscent of a strat-
egy of the octopus arm and reinforcing the idea that joint forma-
tion in soft manipulators is an efficient solution (both in terms of
mechanics and control) for point-to-point motion.
29,30
Addition-
ally, we extracted invariance principles from the kinematic
analyses, demonstrating that a power law exists between the
tangential velocity and path curvature at the trunk tip
(vtip k2=3
tip ), which differs from the relation found in human
hand movements (vk1=3).
24
Finally, we investigated the internal morphology of the
elephant proboscis with unparalleled precision, revealing the
orientation of muscles and connective fibers and their micro-
scopic structure. By analyzing the anatomical data in the light
of our kinematic results, we draw a strong connection between
the muscular system of the trunk and its biomechanical func-
tions. Electromyographic investigations of the proboscis and
recording of the elephant brain activity would clearly deepen
our understanding of the underlying neuromotor control,
although such experiments would be particularly difficult to
implement in elephants compared to smaller species.
In conclusion, our study offers unprecedented insights into the
anatomy and biomechanics of the largest muscular hydrostat on
Earth: we characterized the elephant trunk motion as a partition
of building blocks, both at the biomechanical level (expressed as
muscular synergies) and at the kinematic level (translating into
trajectories encoded via intrinsic trunk variables). The descrip-
tion of motion based on a finite set of primitives and the identifi-
cation of kinematic invariants can greatly reduce the complexity
of control in animal motion and robotics. Solutions derived from
natural systems can inspire engineers to select proper ap-
proaches, and our study has the potential to guide the design
of an innovative biomimetic soft manipulator (https://www.
proboscis.eu).
A
B
Figure 7. Muscular anatomy in the proximal part of an Asian elephant trunk
(A) Coronal sections (MRI virtual sections in left halves and physical sections in right halves) in ventral, central (below the nostrils), and dorsal (above the nostrils)
portions. The white dashed lines highlight the braided arrangement of the two oblique muscle groups.
(B) Sagittal sections (MRI in proximal halves and physical sections in distal halves) in lateral, mid-lateral, and central (intersecting the nostril) portions. Black lines
highlight the cross-helical connective fibers at the junction of the oblique and longitudinal muscles. Muscle groups are indicated as in Figure 6. Star symbol,
proboscideal nerve bundles; arrow, blood vessel.
See also Figure S7,Videos S3 and S4, and online applications 6 and 7.
ll
OPEN ACCESS
Current Biology 31, 1–11, November 8, 2021 9
Please cite this article in press as: Dagenais et al., Elephants evolved strategies reducing the biomechanical complexity of their trunk, Current Biology
(2021), https://doi.org/10.1016/j.cub.2021.08.029
Article
STAR+METHODS
Detailed methods are provided in the online version of this paper
and include the following:
dKEY RESOURCES TABLE
dRESOURCE AVAILABILITY
BLead contact
BMaterials availability
BData and code availability
dEXPERIMENTAL MODEL AND SUBJECT DETAILS
dMETHOD DETAILS
BMotion capture experiments
BTrunk imaging procedures
dQUANTIFICATION AND STATISTICAL ANALYSIS
BCatalogue of prehension behaviors
BKinematic calculations
SUPPLEMENTAL INFORMATION
Supplemental information can be found online at https://doi.org/10.1016/j.
cub.2021.08.029.
A video abstract is available at https://doi.org/10.1016/j.cub.2021.08.
029#mmc8.
ACKNOWLEDGMENTS
Llewellyn Lloyd, Joseph Kamupambe, David Mupupu, Thabo Ramashia, Na-
than Gwangwa, Edward Mabotha, and Oscar Severino assisted with handling
the animals. Roland Pellet (mechanical workshop, physics department, Uni-
versity of Geneva) and Richard Rohart fabricated the metallic and wooden ob-
jects, respectively. We thank the Zoo Zurich (Switzerland) and Antoine Joris
from the R
eserve Africaine de Sigean (France) for providing the trunks of an
Asian and an African elephant, respectively. We thank Ruben Soto and Chris-
tine Bruguier at the CMU (Lausanne, Switzerland); Henning Richter,
Maya Kummrow, and Jean-Michel Hatt at the Tierspital (Zurich, Switzerland);
as well as Franc¸ ois Lazeyras and S
ebastien Courvoisier at the HUG (Geneva,
Switzerland) for CT and MRI scans. We thank Athanasia Tzika and Ebrahim Ja-
hanbakhsh for advice, Adrien Debry and Florent Montange for assisting in
physical sectioning, and Szabolcs Zakany for the development of the xy
motorized imaging stage. We thank Giovanni Landi, Kimmy Costa, and Fabrice
Berger for the implementation of the web-based applications, including the
trunk animations. We thank Jeffrey Thingvold (Qualisys) for helping with the
trunk animation workflow, Gregory Loichot for assembling the webpage with
supplemental material, and Fabrice Berger for the drawing of Figure 1A. We
thank Mathias Bankay (Qualisys) and Nicolas Long (Trinoma) for technica l
assistance with the motion capture set up. We thank Lucia Beccai and Barbara
Mazzolai (Istituto Italiano di Tecnologia, Pisa, Italy), Egidio Falotico (Scuola
Superiore Sant’Anna, Pisa, Italy), Shlomo Magdassi (The Hebrew University
of Jerusalem, Israel), and Sarah Karmel (Photocentric Ltd, Peterborough,
UK) for multidisciplinary discussions. This work was supported by a Future
and Emerging Technologies grant (FET-Open n.863212 – ‘‘PROBOSCIS,’’
https://www.proboscis.eu, coordinated by the Istituto Italiano di Tecnologia,
Pisa, Italy), under the European Union’s Horizon 2020 research and innovation
program. The funding bodies played no role in the design of the study, collec-
tion, analysis, and interpretation of data, and in writing the manuscript.
AUTHOR CONTRIBUTIONS
M.C.M. supervised all aspects of the study. P.D., M.C.M., and S.H. designed
and conducted the behavioral study with assistance of V.H. P.D., V.H., and
M.C.M. performed the anatomical study. Histology was performed by V.H.
P.D. and M.C.M. analyzed all data and wrote the manuscript. All authors
read and approved the final manuscript.
DECLARATION OF INTERESTS
The authors declare that they have no competing interests.
Received: May 5, 2021
Revised: June 25, 2021
Accepted: August 9, 2021
Published: August 23, 2021
REFERENCES
1. Kier, W.M., and Smith, K.K. (1985). Tongues, tentacles and trunks: the
biomechanics of movement in muscular-hydrostats. Zool. J. Linn. Soc.
83, 307–324.
2. Dehnhardt, G.F.C., and Sachser, N. (1997). Sensitivity of the trunk of Asian
elephants for texture differences of actively touched objects. Proceedings
of the International Symposium on Physiology and Ecology of Wild and
Zoo Animals 2, 37–39.
3. Shoshani, J. (1997). It’s a nose! It’s a hand! It’s an elephants’ trunk! Nat.
Hist. 106, 37–43.
4. Chevalier-Skolnikoff, S., and Liska, J. (1993). Tool use by wild and captive
elephants. Anim. Behav. 46, 209–219.
5. Rasmussen, L.E., and Munger, B.L. (1996). The sensorineural specializa-
tions of the trunk tip (finger) of the Asian elephant, Elephas maximus.
Anat. Rec. 246, 127–134.
6. Shoshani, J. (1998). Understanding proboscidean evolution: a formidable
task. Trends Ecol. Evol. 13, 480–487.
7. Hart, B.L., Hart, L.A., McCoy, M., and Sarath, C.R. (2001). Cognitive
behaviour in Asian elephants: use and modification of branches for fly
switching. Anim. Behav. 6262, 839–847.
8. Garstang, M. (2004). Long-distance, low-frequency elephant communica-
tion. J. Comp. Physiol. A Neuroethol. Sens. Neural Behav. Physiol. 190,
791–805.
9. Martin, F., and Niemitz, C. (2003). ‘‘Right-trunkers’’ and ‘‘left-trunkers’’:
side preferences of trunk movements in wild Asian elephants (Elephas
maximus). J. Comp. Psychol. 117, 371–379.
10. Shoshani, J., Kupsky, W.J., and Marchant, G.H. (2006). Elephant brain.
Part I: gross morphology, functions, comparative anatomy, and evolution.
Brain Res. Bull. 70, 124–157.
11. Niimura, Y., Matsui, A., and Touhara, K. (2014). Extreme expansion of the
olfactory receptor gene repertoire in African elephants and evolutionary
dynamics of orthologous gene groups in 13 placental mammals.
Genome Res. 24, 1485–1496.
12. Plotnik, J.M., Brubaker, D.L., Dale, R., Tiller, L.N., Mumby, H.S., and
Clayton, N.S. (2019). Elephants have a nose for quantity. Proc. Natl.
Acad. Sci. USA 116, 12566–12571.
13. Bellman, R.E. (2015). Adaptive Control Processes (Princeton University
Press).
14. Mussa-Ivaldi, F.A. (1999). Modular features of motor control and learning.
Curr. Opin. Neurobiol. 9, 713–717.
15. Mussa-Ivaldi, F.A., and Bizzi, E. (2000). Motor learning through the combi-
nation of primitives. Philos. Trans. R. Soc. Lond. B Biol. Sci. 355, 1755–
1769.
16. Wolpert, D.M., and Ghahramani, Z. (2000). Computational principles of
movement neuroscience. Nat. Neurosci. 3(Suppl ), 1212–1217.
17. Flash, T., and Hochner, B. (2005). Motor primitives in vertebrates and in-
vertebrates. Curr. Opin. Neurobiol. 15, 660–666.
18. Anderson, D.J., and Perona, P. (2014). Toward a science of computational
ethology. Neuron 84, 18–31.
19. Wiltschko, A.B., Johnson, M.J., Iurilli, G., Peterson, R.E., Katon, J.M.,
Pashkovski, S.L., Abraira, V.E., Adams, R.P., and Datta, S.R. (2015).
Mapping sub-second structure in mouse behavior. Neuron 88, 1121–
1135.
ll
OPEN ACCESS
10 Current Biology 31, 1–11, November 8, 2021
Please cite this article in press as: Dagenais et al., Elephants evolved strategies reducing the biomechanical complexity of their trunk, Current Biology
(2021), https://doi.org/10.1016/j.cub.2021.08.029
Article
20. Todorov, E. (2004). Optimality principles in sensorimotor control. Nat.
Neurosci. 7, 907–915.
21. Fitts, P.M. (1954). The information capacity of the human motor system in
controlling the amplitude of movement. J. Exp. Psychol. 47, 381–391.
22. Harris, C.M., and Wolpert, D.M. (1998). Signal-dependent noise deter-
mines motor planning. Nature 394, 780–784.
23. Viviani, P., and Terzuolo, C. (1982). Trajectory determines movement dy-
namics. Neuroscience 7, 431–437.
24. Lacquaniti, F., Terzuolo, C., and Viviani, P. (1983). The law relating the ki-
nematic and figural aspects of drawing movements. Acta Psychol. (Amst.)
54, 115–130.
25. Hicheur, H., Vieilledent, S., Richardson, M.J.E., Flash, T., and Berthoz, A.
(2005). Velocity and curvature in human locomotion along complex curved
paths: a comparison with hand movements. Exp. Brain Res. 162, 145–154.
26. Zago, M., Matic, A., Flash, T., Gomez-Marin, A., and Lacquaniti, F. (2018).
The speed-curvature power law of movements: a reappraisal. Exp. Brain
Res. 236, 69–82.
27. Matic, A., and Gomez-Marin, A. (2020). Geometric purity, kinematic
scaling and dynamic optimality in drawing movements beyond ellipses.
J. Math. Psychol. 99, 102453.
28. Sumbre, G., Fiorito, G., Flash, T., and Hochner, B. (2005). Neurobiology:
motor control of flexible octopus arms. Nature 433, 595–596.
29. Hooper, S.L. (2006). Motor control: the importance of stiffness. Curr. Biol.
16, R283–R285.
30. Sumbre, G., Fiorito, G., Flash, T., and Hochner, B. (2006). Octopuses use a
human-like strategy to control precise point-to-point arm movements.
Curr. Biol. 16, 767–772.
31. Gutfreund, Y., Flash, T., Yarom, Y., Fiorito, G., Segev, I., and Hochner, B.
(1996). Organization of octopus arm movements: a model system for
studying the control of flexible arms. J. Neurosci. 16, 7297–7307.
32. Sumbre, G., Gutfreund, Y., Fiorito, G., Flash, T., and Hochner, B. (2001).
Control of octopus arm extension by a peripheral motor program.
Science 293, 1845–1848.
33. Zelman, I., Titon, M., Yekutieli, Y., Hanassy, S., Hochner, B., and Flash, T.
(2013). Kinematic decomposition and classification of octopus arm move-
ments. Front. Comput. Neurosci. 7, 60.
34. Hanassy, S., Botvinnik, A., Flash, T., and Hochner, B. (2015). Stereotypical
reaching movements of the octopus involve both bend propagation and
arm elongation. Bioinspir. Biomim. 10, 035001.
35. Zullo, L., Eichenstein, H., Maiole, F., and Hochner, B. (2019). Motor control
pathways in the nervous system of Octopus vulgaris arm. J. Comp.
Physiol. A Neuroethol. Sens. Neural Behav. Physiol. 205, 271–279.
36. Tsukagoshi, H., Kitagawa, A., and Segawa, M. (2001). Active hose: an arti-
ficial elephant’s nose with maneuverability for rescue operation. In
Proceedings 2001 ICRA IEEE International Conference on Robotics and
Automation, pp. 2454–2459.
37. McMahan, W., Jones, B.A., and Walker, I.D. (2005). Design and implemen-
tation of a multi-section continuum robot: Air-Octor. In 2005 IEEE/RSJ
International Conference on Intelligent Robots and Systems, pp. 2578–
2585.
38. Walker, I., Dawsona, D., Flash, T., Grasso, F., Hanlond, R., Hochnere, B.,
Kier, W., Paganog, C., Rahn, C., and Zhangi, Q. (2005). Continuum robot
arms inspired by cephalopods. Proceedings of SPIE, 5804.
39. Mazzolai, B., Margheri, L., Cianchetti, M., Dario, P., and Laschi, C. (2012).
Soft-robotic arm inspired by the octopus: II. From artificial requirements to
innovative technological solutions. Bioinspir. Biomim. 7, 025005.
40. Yang, Y., and Zhang, W. (2014). ET arm: highly compliant elephant-trunk
continuum manipulator. International Conference on Intelligent Robotics
and Applications, 288–299.
41. Wu, J., Zhao, Y., Zhang, Y., Shumate, D., Braccini Slade, S., Franklin, S.V.,
and Hu, D.L. (2018). Elephant trunks form joints to squeeze together small
objects. J. R. Soc. Interface 15, 20180377.
42. Kuberski, S.R., and Gafos, A.I. (2019). The speed-curvature power law in
tongue movements of repetitive speech. PLoS ONE 14, e0213851.
43. Boas, J.E.V., and Paulli, S. (1925). The Elephant’s Head: Studies in the
Comparative Anatomy of the Organs of the Head of the Indian Elephant
and Other Mammals (Carlsberg-Fund).
44. Wilson, J.F., Mahajan, U., Wainwright, S.A., and Croner, L.J. (1991). A con-
tinuum model of elephant trunks. J. Biomech. Eng. 113, 79–84.
45. Martins, A.F., Bennett, N.C., Clavel, S., Groenewald, H., Hensman, S.,
Hoby, S., Joris, A., Manger, P.R., and Milinkovitch, M.C. (2018). Locally-
curved geometry generates bending cracks in the African elephant skin.
Nat. Commun. 9, 3865.
46. Kier, W.M. (1992). Hydrostatic skeletons and muscular hydrostats. In
Biomechanics (Structures and Systems): A Practical Approach, A.A.
Biewener, ed. (IRL Press at Oxford University Press).
47. Schulz, A.K., Ning Wu, J., Ha, S.Y.S., Kim, G., Braccini Slade, S., Rivera,
S., Reidenberg, J.S., and Hu, D.L. (2021). Suction feeding by elephants.
J. R. Soc. Interface 18, 20210215.
48. Kier, W.M., and Stella, M.P. (2007). The arrangement and function of
octopus arm musculature and connective tissue. J. Morphol. 268,
831–843.
49. Kier, W.M. (2012). The diversity of hydrostatic skeletons. J. Exp. Biol. 215,
1247–1257.
50. Kreyszig, E. (1991). Differential Geometry (Dover Publications).
51. Miller, J. (2009). Shape Curve Analysis Using Curvature (University of
Glasgow).
52. Wagner, P.H., Luo, X., and Stelson, K.A. (1995). Smoothing curvature and
torsion with spring splines. Comput. Aided Des. 27, 615–626.
ll
OPEN ACCESS
Current Biology 31, 1–11, November 8, 2021 11
Please cite this article in press as: Dagenais et al., Elephants evolved strategies reducing the biomechanical complexity of their trunk, Current Biology
(2021), https://doi.org/10.1016/j.cub.2021.08.029
Article
STAR+METHODS
KEY RESOURCES TABLE
RESOURCE AVAILABILITY
Lead contact
Further information and requests for resources and reagents should be directed to and will be fulfilled by the lead contact, Michel C.
Milinkovitch (Michel.Milinkovitch@unige.ch).
Materials availability
This study did not generate new materials.
Data and code availability
All data needed to evaluate the conclusions in the paper are present in the paper and/or the Supplemental Information. The online
web applications are available at https://www.lanevol.org/projects/proboscis.
EXPERIMENTAL MODEL AND SUBJECT DETAILS
Two adult male African bush elephants (Loxodonta africana) were involved in the motion capture experiments, one with an intact trunk
and the second with severed fingers in the prehensile tip as the result of an accident that occurred several years prior to the exper-
iments. All experiments have been approved by the University of Geneva ethical regulation authority and performed according to
South-African law. Two trunks from adult males of African bush and Asian elephants (Loxodonta africana and Elephas maximus)
were collected from deceased zoo animals and used for the anatomical study.
METHOD DETAILS
Motion capture experiments
Motion capture was performed at Adventures with Elephants (https://adventureswithelephants.com/) in Bela Bela, South Africa. The
animals were trained to grasp objects (of various shapes and sizes) with the trunk and transport them from one point to another on the
ground (or from/to the hands of a person). During the experiments, one animal handler was standing on each side of the elephant to
distribute food items as rewards for executing tasks.
Multiple objects were designed for the prehension experiments (Table S1). Four shapes (sphere, cube, cylinder and cone), each in
four different sizes (diameter or side length d
1
= 3.75 cm, d
2
= 7.5 cm, d
3
= 15 cm, d
4
= 30 cm) were made of wood (plywood and
spruce; fabricated by https://www.richard-rohart.com). The cylinders have a length/diameter ratio of 3, the cones have a height/
diameter ratio of 1. To limit the change in mass in that series of objects, the larger shapes (d
3
and d
4
) were made hollow with a
REAGENT or RESOURCE SOURCE IDENTIFIER
Biological samples
trunk of a deceased adult elephant L. africana R
eserve Africaine de Sigean, France CITES certificate
1470
trunk of a deceased adult elephant E. maximus Zurich zoo, Switzerland N/A
Experimental models: Organisms/strains
2 adult male elephants L. africana https://adventureswithelephants.com/ N/A
Software and algorithms
Qualisys Track Manager https://www.qualisys.com/software/qualisys-track-manager/?
gclid=EAIaIQobChMIyLTbosOl8AIVg6Z3Ch3stAn4EAAYAiAAEgLq5fD
_BwE
N/A
MATLAB https://www.mathworks.com/products/matlab.html N/A
Amira https://www.thermofisher.com/ch/en/home/industrial/electron-
microscopy/electron-microscopy-instruments-workflow-solutions/
3d-visualization-analysis-software/amira-life-sciences-biomedical.html
N/A
Hugin http://hugin.sourceforge.net/ N/A
ll
OPEN ACCESS
e1 Current Biology 31, 1–11.e1–e4, November 8, 2021
Please cite this article in press as: Dagenais et al., Elephants evolved strategies reducing the biomechanical complexity of their trunk, Current Biology
(2021), https://doi.org/10.1016/j.cub.2021.08.029
Article
wood thickness of 1 cm. Three shapes (sphere and cylinder of diameter d
2
, and disk with a 30 cm diameter and 1 cm length) were also
fabricated, each with three different mass densities (m
1
= 500 kg/m
3
, wood; m
2
= 3000 kg/m
3
, aluminum; m
3
= 8000 kg/m
3
, steel). The
aluminum sphere was not tested. The steel sphere (m
3
) was fabricated in two versions: one with a smooth surface and one with a
carved (rough) surface. The smooth steel sphere was presented to the animal either with a clean surface, or with its surface covered
with lubricant (glycerin and water-based) in order to make it slippery. The metallic objects were fabricated in the mechanical work-
shop of the physics department at the University of Geneva, Switzerland. Granular material (rice) was used for additional prehension
experiments. We also tested the ability of elephants to pick up uncooked spaghetti sticks (Data S1A).
We used a motion capture system (Qualisys, Go
¨teborg, Sweden), combining 10 infrared cameras (Oqus 7+, 100 Hz, 12 megapix-
els) with sun filters and two video cameras recording in the visible range (Miqus, 30 Hz, 2 megapixels) positioned in a semi-circle
around the scene, placed on tripods with alternating heights of 1 m and 2 m. Retroreflective markers (1.6 cm in diameter) were
screwed in flexible bands of bio-compatible adhesive kinesiology tape. Eight to 10 rows of markers were placed on the dorsal
side of the trunk with 3 (near the tip) to 6 (at the proximal base) markers per row. Distances in between markers were 5 cm near
the tip and 10 cm at the base. An example of the markers configuration on the trunk can be seen in Figure 2A. The animals accepted
the presence of the kinesiology bands on the dorsal side of their trunk, but not on the ventral side or on the tip fingers. Before placing
the kinesiology bands, the trunk was thoroughly brushed to remove excessive dust. Calibration was performed inside the measure-
ment volume by placing an L-shaped frame with 4 markers on the floor (to set the coordinate system) and by sweeping a wand (with
one marker at each end of its T-shaped end) across the full volume. We obtained very high spatial resolution (uncertainty z2 mm) for
the tracking of the 3D markers positions.
Trunk imaging procedures
Following the approach of the Visible Human Project (https://www.nlm.nih.gov/research/visible/visible_human.html), we explored
the internal anatomy of the trunk from adult males of African bush and Asian elephants using a combination of CT scans, MRI
and physical sectioning. The time between death and freezing the trunk was longer for the African elephant sample, resulting in a
more advanced degradation of the tissues which is reflected in the MRI scans but less so in physical serial sections.
The MRI and CT scanning sequences were optimized to reveal the soft tissues (dermis, epidermis, muscles, connective tissues).
CT scan yields a better contrast for dermis and epidermis, and was thus used to extract precise skin morphology. MRI offers better
contrast for muscular and connective tissues than for the skin but also takes a much longer time for comparable resolution (4 h for the
MRI of a full trunk versus 30 min for a CT scan). The trunks were unfrozen a few days before imaging so that they would be at room
temperature during the scans. For the Asian elephant trunk, CT scanning and MRI were performed at the Tierspital (Zurich,
Switzerland). For the African elephant trunk, CT scanning was performed at the University Medical Center in Geneva (Switzerland)
and MRI was performed at Geneva University Hospital (HUG, Geneva, Switzerland). The resolution of the CT scan (pixel spacing)
is 0.75 mm. MRI sequences were acquired with the so-called ‘T2’ protocol, which offers better contrast but can only be acquired
as 2D series, and the ‘Proton Density (PD)’ method which yields slightly reduced contrast but produces a 3D image with isotropic
resolution. The T2 scans were acquired in 3 orientations (coronal, sagittal and transversal) with a slice thickness (i.e., out-of-plane
resolution) of roughly 2 mm and in-plane resolution dependent on the field of view (FOV): from 0.6-0.7 mm (smallest FOV at the
tip) to 0.9-1.1 mm (largest FOV at the head base). The isotropic resolution for the PD sequence is equivalent to the in-plane resolution
of the T2 sequences. Six successive and overlapping portions of the trunk were scanned using different receptor antenna so as not to
displace the antennas nor the trunk during the whole procedure.
The image series (DICOM files) were imported in Amira (Thermo Fisher Scientific, Massachusetts, USA) where the piecewise scans
were stitched and used to segment and export surface meshes for the skin and the six main muscle groups. Segmentation was done
automatically (using an intensity threshold) for the skin, whereas segmentation of muscle groups was manually performed in about 40
selected planes and then interpolated over the full trunk volume.
After CT and MRI scanning, the trunks were frozen again in a straight position before being sliced on a vertical bandsaw in regular
1 cm-thick sections. The African elephant trunk was entirely sliced in about 160 serial transversal sections. The Asian elephant trunk
was sliced in about 130 serial transversal cuts, and the remaining 45 cm most proximal part of the trunk was sagittally divided in two
halves: the left half was sliced in serial sagittal cuts (20 slices) and the right half was sliced in serial coronal cuts (25 slices). Each
trunk slice was washed in tap water, placed over a green background in a plastic container, fully immersed in 2 L of tap water (to avoid
reflections) and covered with a glass plate (to prevent the slice from floating). Physical sections were imaged with the following high-
resolution photography system. A 45 mega-pixel camera (Nikon D850 with 60 mm objective, f16, shutter speed=1/50 s, iso 200) was
fixed facing down over the slice at a distance of 74.5 cm. The plastic container with the slice was positioned on an in-house developed
motorized stage for controlled horizontal (in xand y) movements. Nine overlapping pictures were taken to cover the largest slices
(about 45 cm x 35 cm) whereas more distal sections of the trunk could be imaged with six, four, and one picture. Images were stitched
using the Hugin software (http://hugin.sourceforge.net/). Alignment and cropping of the green background was done in Photoshop.
This procedure allowed us to generate the same very high resolution (z0.03 mm/pixel) for all images. Note that this resolution is much
higher than for the corresponding virtual section images obtained with CT and MRI scans. This also proved to be crucial for delim-
itating the muscle groups in the tip of the trunk. All serial sections are available (for both species) in high resolution in the form of a
virtual scroller of the internal structures of the trunk (online applications 2 and 5–7).
Finally, small blocks from different regions of the trunk (muscles and skin) were fixed in formalin, and histological slides were pre-
pared with two standard staining procedures: (i) Mallory’s trichrome and (ii) Methyl blue and Acid fuchsin.
ll
OPEN ACCESS
Current Biology 31, 1–11.e1–e4, November 8, 2021 e2
Please cite this article in press as: Dagenais et al., Elephants evolved strategies reducing the biomechanical complexity of their trunk, Current Biology
(2021), https://doi.org/10.1016/j.cub.2021.08.029
Article
QUANTIFICATION AND STATISTICAL ANALYSIS
Catalogue of prehension behaviors
Analyzing the 550 recorded prehension sequences (Table S3) for both animals, we constructed an ethogram with 17 basic trunk
behaviors (Video S1). We counted the number of occurrences for each behavior over all the prehension sequences, which can be
used to estimate the probability of executing a behavior when presented with a specific shape.
Kinematic calculations
After post-processing of the motion capture data (cleaning the trajectories with Qualisys software QTM), the 3D positions of all
markers in each time point were exported for analyses of specific trajectories. We fitted an ellipse to each row of markers. Each ellipse
has 8 parameters to be fitted (3 coordinates of its center, 3 angles for its orientation around the 3 cartesian axes and 2 parameters for
the lengths of its axes). The initial guess for the optimization process is based on the real anatomy of the trunk, using the eccentricity
(ratio of large over small diameter) of the trunk at the position of each ellipse. We imposed that the large axis of the ellipses decreases
in length toward the tip. Due to the fact that markers are only present on the dorsal perimeter of the trunk, some deviations are
expected between the fitted values for the two principal axes aand band the real cross-sectional geometry of the trunk.
Nevertheless, despite not having imposed incompressibility (constant volume), this feature is captured by the reconstruction, with
a maximal total change in volume of ± 5%. We then fit a cubic spline through the ellipse centers and extract 50 points from this spline
as an approximate virtual backbone of the trunk. The resulting model can be visualized in Figures 2B and S2A.
The velocity and acceleration are calculated using the virtual backbone spline positions and finite centered differences between
subsequent time frames. A reference sequence was selected in which the trunk was in a relaxed posture for 3 s. Longitudinal and
cross-sectional strains (DL=Lo,Da=aoand Db=bo) are defined using the arc length between subsequent ellipses and the radii
(for the large and small axes) of each ellipse, relative to the reference frame, yielding a local description of the strains.
The curvature and torsion are typically defined in differential geometry using the Frenet-Serret framework.
50
Preliminary analyses
revealed some pitfalls of this approach while evaluating those quantities across the trunk. Indeed, torsion is prone to artifacts and
difficult to interpret, as very sharp peaks of torsion can appear along the curve without any physical intuitive meaning, as observed
by Miller
51
in his work on human face shape analysis. Curvature thus defined is also prone to artifacts, especially when the number of
spline segments is small and at the extremities of the curve.
52
Therefore, we used an alternative definition for the torsion and the two
curvature components, which is much more relevant from a biomechanical point of view. This definition is based on local orthogonal
frames constructed from the ellipses: unit vectors b
a,b
band b
n=b
b3b
apointing respectively along the ellipse large axis (toward the right
of the animal), the ellipse small axis (toward the dorsal side of the trunk) and the axis of the trunk (the cross product being in the
direction of the trunk tip). This allows to define the torsion tba as the tendency of the trunk to twist in a clockwise direction, knb as
the upward/downward curvature component and kna as the left/right curvature component:
vib
n=knb b
b+kna b
a
vib
b=knb b
n+tba b
a
vib
a=kna b
ntba b
b
with videnoting the partial derivative with respect to i, the normalized position index on the trunk, starting at 0 at the most proximal
ellipse and ending at 1 at the most distal ellipse.
Or, in matrix form:
vi
2
6
6
6
6
4b
n
b
b
b
a
3
7
7
7
7
5
=
2
6
6
4
0knb kna
knb 0tba
kna tba 0
3
7
7
5
2
6
6
6
6
4b
n
b
b
b
a
3
7
7
7
7
5
These quantities are dimensionless; to obtain the usual units (m
-1
) for torsion and curvature, we must divide by the (time-depen-
dent) trunk length. In the relaxed posture, the dimensionless torsion has maximal values of t0;max =±2.5, or 1.4 m
-1
when normalized
by the trunk length. To avoid including biased torsion coming from an asymmetrical placement of the markers, tba is always
expressed as a relative difference with the reference pose: ðtt0Þ=t0;max . The torsion can be intuitively interpreted as the tendency
of the ellipse cross-section to rotate around the trunk central axis, since ðtt0Þxviq, where qis the rotation angle (in radians) of the
ellipse around its center. For example, ðtt0Þ=2pwould mean that the trunk effectuates a full twist of 2pradians around its central
axis over its full length (from i=0toi=1).
ll
OPEN ACCESS
e3 Current Biology 31, 1–11.e1–e4, November 8, 2021
Please cite this article in press as: Dagenais et al., Elephants evolved strategies reducing the biomechanical complexity of their trunk, Current Biology
(2021), https://doi.org/10.1016/j.cub.2021.08.029
Article
In addition to the instantaneous curvature of the trunk posture (kna and knb ), we also calculate the path curvature of the trunk tip
(defined in the Frenet-Serret frame), using the time-dependent position r
!ðtÞof the trunk tip (position of the center of the most distal
ellipse):
ktip =
kr
!0ðtÞ3r
!00ðtÞk
kr
!0ðtÞk
3=
kv
!ðtÞ3a
!ðtÞk
kv
!ðtÞk3
where v
!and a
!are the velocity and acceleration of the trunk tip.
Additional information is provided in Data S1.
ll
OPEN ACCESS
Current Biology 31, 1–11.e1–e4, November 8, 2021 e4
Please cite this article in press as: Dagenais et al., Elephants evolved strategies reducing the biomechanical complexity of their trunk, Current Biology
(2021), https://doi.org/10.1016/j.cub.2021.08.029
Article
