AquaSynth/research/vocal-source-grounding/sources/elemans-2015-universal-mechanisms.txt

University of Southern Denmark

Universal mechanisms of sound production and control in birds and mammals
Elemans, Coen; Rasmussen, Jeppe Have; Herbst, Christian T.; Düring, Daniel Normen;
Zollinger, Sue Anne; Brumm, Henrik ; Srivastava, K.; Svane, Niels; Ding, Ming; Larsen, Ole
Næsbye; Sober, Sam J. ; Svec, Jan G.
Published in:
Nature Communications
DOI:
10.1038/ncomms9978
Publication date:
2015
Document version:
Final published version
Document license:
CC BY
Citation for pulished version (APA):
Elemans, C., Rasmussen, J. H., Herbst, C. T., Düring, D. N., Zollinger, S. A., Brumm, H., Srivastava, K., Svane,
N., Ding, M., Larsen, O. N., Sober, S. J., & Svec, J. G. (2015). Universal mechanisms of sound production and
control in birds and mammals. Nature Communications, 6, Article 8978. https://doi.org/10.1038/ncomms9978

Go to publication entry in University of Southern Denmark's Research Portal

Terms of use
This work is brought to you by the University of Southern Denmark.
Unless otherwise specified it has been shared according to the terms for self-archiving.
If no other license is stated, these terms apply:
• You may download this work for personal use only.
• You may not further distribute the material or use it for any profit-making activity or commercial gain
• You may freely distribute the URL identifying this open access version
If you believe that this document breaches copyright please contact us providing details and we will investigate your claim.
Please direct all enquiries to puresupport@bib.sdu.dk

Download date: 28. May. 2026

ARTICLE
Received 24 Mar 2015 | Accepted 22 Oct 2015 | Published 27 Nov 2015

DOI: 10.1038/ncomms9978

OPEN

Universal mechanisms of sound production
and control in birds and mammals
C.P.H Elemans1,2,3, J.H. Rasmussen1,*, C.T. Herbst4,*, D.N. Düring1,*, S.A. Zollinger5, H. Brumm5, K. Srivastava6,7,
N. Svane1, M. Ding8, O.N. Larsen1, S.J. Sober6 & J.G. Švec4

As animals vocalize, their vocal organ transforms motor commands into vocalizations for
social communication. In birds, the physical mechanisms by which vocalizations are produced
and controlled remain unresolved because of the extreme difficulty in obtaining in vivo
measurements. Here, we introduce an ex vivo preparation of the avian vocal organ that allows
simultaneous high-speed imaging, muscle stimulation and kinematic and acoustic analyses to
reveal the mechanisms of vocal production in birds across a wide range of taxa. Remarkably,
we show that all species tested employ the myoelastic-aerodynamic (MEAD) mechanism,
the same mechanism used to produce human speech. Furthermore, we show substantial
redundancy in the control of key vocal parameters ex vivo, suggesting that in vivo vocalizations
may also not be specified by unique motor commands. We propose that such motor
redundancy can aid vocal learning and is common to MEAD sound production across birds
and mammals, including humans.

1 Department of Biology, University of Southern Denmark, Campusvej 55, 5230 Odense, Denmark. 2 QuanTM program, Emory University, Atlanta, Georgia
30322, USA. 3 Université de Saint-Etienne/Lyon, ENES/CNPS CNRS UMR8195, Saint-Etienne 42023, France. 4 Faculty of Science, Department of Biophysics,
Voice Research Lab, Palacky University, Olomouc 77146, Czech Republic. 5 Communication and Social Behaviour Group, Max Planck Institute for Ornithology,
Seewiesen 82319, Germany. 6 Department of Biology, Emory University, Atlanta, Georgia 30332, USA. 7 Coulter Department of Biomedical Engineering,
Georgia Institute of Technology, Atlanta, Georgia 30332, USA. 8 Department of Orthopaedic Surgery & Traumatology, Odense University Hospital, University
of Southern Denmark, Odense 5230, Denmark. * These authors contributed equally to this work. Correspondence and requests for materials should be
addressed to C.P.H.E. (email: coen@biology.sdu.dk).

NATURE COMMUNICATIONS | 6:8978 | DOI: 10.1038/ncomms9978 | www.nature.com/naturecommunications

1

ARTICLE

NATURE COMMUNICATIONS | DOI: 10.1038/ncomms9978

n contrast to laryngeally vocalizing mammals, B10,000
species of extant birds vocalize with a uniquely avian vocal
organ, the syrinx, located at the tracheobronchial junction and
suspended in an air sac of the respiratory system1. The syrinx is
structurally highly diverse across species1, but how morphological
diversity reflects functional diversity remains unexplored. In
addition, while songbirds are a widely used experimental animal
model for neural mechanisms underlying vocal imitation
learning2,3, we lack the empirical evidence to precisely map
motor function onto neural circuitry4. Addressing these questions
requires empirical quantification of syringeal dynamics as a
function of control parameters in different species under
physiologically realistic, controlled conditions. However,
imaging the syrinx in vivo remains a challenge5,6, and we thus
still lack this quantification of syringeal dynamics and control
parameters.
Earlier endoscopic imaging identified syringeal vibratory
tissues in songbirds and non-songbirds5,6, arguing against
purely aerodynamical whistle mechanisms in which sound is
produced without periodic movement of the vocal apparatus.
Many mathematical models for birdsong7–11 assume that
syringeal sound production is based on a myoelasticaerodynamic (MEAD) system12–15. However, conclusive
empirical evidence for MEAD is lacking4,16. The MEAD
framework explains the physical mechanism underlying
laryngeal sound production in mammals12–15,17. In brief,
self-sustaining laryngeal vocal fold oscillations are maintained
through fluid-tissue interactions and (myo)elastic restoring
forces generated within the tissues12–15, preventing the
need for muscle contractions at the rate of tissue vibration or
other periodic input18. Expiratory airflow is mechanically
converted by vocal folds into pulse-like airflow, which
causes air pressure disturbances constituting the acoustic
excitation of the system13. The mechanical properties and
recruitment of different layers of vibrating tissues affect
their resonance properties, which in combination with
aerodynamic driving forces determine the frequency and mode
of oscillation19–22.
According to MEAD theory, the medio-lateral vibration of the
inner vocal fold surface (vibrational component 1 (VC1)) that
gates airflow23 can only be self-sustaining if moving around a
stable equilibrium position and if no net energy loss occurs per
oscillation cycle13,24. The latter requires the presence of an
aerodynamic force that changes magnitude with direction of
vocal fold motion and thus is asymmetric over the oscillation
cycle13. One possibility is that during self-sustained vocal fold
vibration the required asymmetric aerodynamic force is produced
by time-varying supraglottal pressure due to inertia of air in
the vocal tract23. Such a mechanism would result in uniform
medio-lateral vibration of the vocal folds24. However, models
suggest that this mechanism limits the range of fundamental
frequencies (F0) produced14. Another, more robust possibility is
that the asymmetric aerodynamic force during self-sustained
vocal fold vibration is produced by out-of-phase motion of the
superior and inferior edge of the vibrating tissue (vibrational
component 2 (VC2))13,25. VC1 and VC2 are the respective
medio-lateral and caudo-cranial components of a tissue surface
wave, or mucosal wave, that travels on the inner vocal fold surface
along the expiratory air stream and facilitates aerodynamic energy
transfer into tissue24. The wave phase changes cause the vocal
folds to change shape from convergent during opening to
divergent during closing parts of the cycle24. Because the
intraglottal pressure is higher for the convergent shape than for
the divergent shape26,27, the vocal folds are pushed apart during
opening and pulled together during closing14,24,25. VC2 presence
thus indicates that intraglottal pressure forms the asymmetric

I

2

forcing function over opening and closing phases of vibration25
essential to self-sustained oscillation. Sound excitation events in
mammals occur mainly at glottal closure and/or opening, when
airflow abruptly stops or starts28–30.
Although isolated aspects of syringeal dynamics have been
studied in birds5,6,31–35, the asymmetric forcing function essential
to maintain self-sustained oscillation has not yet been identified,
and the caudo-cranial tissue-wave component VC2—a crucial
underlying assumption in modelling studies7,9,36—has not been
demonstrated experimentally in the intact syrinx under
appropriate physiological conditions16. Furthermore, it is
unknown how syringeal dynamics relate to sound generating
events within a single oscillatory cycle. These essential features of
syringeal dynamics required to confirm MEAD have yet to be
established16.
The translation of vocal motor commands into acoustic output
depends on neural activity, musculature, morphology and
physical mechanism of sound production4,16,37. These variables
define a multi-dimensional parameter space that the brain needs
to explore and navigate to control vocal output. In birds, these
control variables have been studied in vivo by correlating
acoustical parameters, for example, F0, with physiological
parameters, for example, lung pressure or muscle activity,
taking advantage of the highly stereotyped vocal patterns
employed by adult birds38. This approach can inform us about
a particular solution an individual uses to control its vocal output.
However, if the control space is redundant (that is, there is more
than one possible solution to achieve a specific vocal target, for
example, F0), as commonly observed in motor control systems39,
studying stereotyped in vivo behaviour provides limited insight in
the behaviour of the entire system as we only observe the final
solution the individual uses that may not be unique. Furthermore
if vocal control parameters covary it may be difficult to establish
causal relationships. To understand how the brain controls vocal
behaviour therefore requires systematic quantification of the
system’s behaviour across its multi-dimensional parameter space.
However, we currently lack an experimental paradigm to
systematically study physiological control of the vocal system
in birds.
Here, we present a novel ex vivo paradigm of the syrinx, which
allows unprecedented experimental control and high-resolution
imaging during sound production. First, to investigate if the
MEAD physical mechanism of self-sustained oscillations as
observed in mammals is applicable to birds, we test the
hypotheses that a caudo-cranial travelling tissue surface wave is
present and that this tissue wave is associated with sound
production events. Second, to determine whether vocalizations
are encoded by unique motor commands we test the hypothesis
that the physiological control space of the syrinx ex vivo is
redundant for key acoustic parameters. We show that birds
employ the MEAD mechanism for sound production and that
key vocal parameters exhibit redundant control ex vivo. We
propose that motor redundancy may accelerate vocal learning
and is common to MEAD sound production across birds and
mammals.
Results
Physical mechanism of sound production. We developed an
experimental paradigm that allows imaging of syringeal dynamics
under controlled conditions ex vivo (Methods section;
Supplementary Fig. 1). To test our first hypothesis, if the
MEAD physical mechanism of self-sustained oscillations is
applicable to birds (Fig. 1a), we took advantage of the diversity in
syringeal morphology across species and first studied the
domestic pigeon syrinx (Fig. 1b) because of its relatively simple

NATURE COMMUNICATIONS | 6:8978 | DOI: 10.1038/ncomms9978 | www.nature.com/naturecommunications

ARTICLE

NATURE COMMUNICATIONS | DOI: 10.1038/ncomms9978

VC1

VC2

Trachea
ICAS membrane

psupra

Muscle

pICAS

T4
Lateral vibratory
mass (LVM)

T2
psub
Larynx

Closing

Syrinx

Right
LVM

T1
Opening

Bronchus
Cranial

Left
LVM

T4
Ch B

DKG1
T2

Δy

DKG2

T1
Ch A

1 mm
Closing/divergent

Caudal

Collision

Opening/convergent

DKG1

Syringeal opening

Left

ΔtDKG

Right

DKG2

Left

Right

μEGG

Ch A
Ch B

Position

2

1 mm

Cranial
Min

VC2 mucosal wave

2
2 ms

4
0

6

LVM displacement
(mm)

ΔtEGG

Glottovibrogram
GVG

3
Opening / closing / opening cranial edge

2

Cranial edge

1

Caudal edge

0

Syringeal opening
(mm)

Caudal

Closing/opening caudal edge
Right

Sound (Pa)

Tracheal
VKG

Closing caudal edge
Mediolateral mucosal wave

Left

Opening/closing cranial edge

0

5

0
Excitation
10

15

20

25

30

35

Time (ms)

morphology. We found self-sustained syringeal oscillations when
both bronchial and air sac pressures were 40.5 kPa (N ¼ 12).
The lateral vibratory masses (LVMs)7 were visualized using
transillumination of the syrinx (Fig. 1c). Within an oscillation

cycle the LVM inner wall changed shape from divergent
during closing, rectangular during full collision and to
convergent during opening (Fig. 1c, Supplementary Movie 1).
A consistent phase shift in LVM position, confirmed by

NATURE COMMUNICATIONS | 6:8978 | DOI: 10.1038/ncomms9978 | www.nature.com/naturecommunications

3

ARTICLE

NATURE COMMUNICATIONS | DOI: 10.1038/ncomms9978

simultaneous micro-electroglottography (mEGG) recordings
(Methods section), was present along the caudo-cranial
axis (Fig. 1d). Spatiotemporal analysis of the LVM inner wall
displacement and syringeal opening identified both components
VC1 and VC2 of a tissue travelling wave along the caudo-cranial
axis (Fig. 1e). Furthermore, we observed medio-lateral, that is,
laterally travelling, tissue waves on the cranial surface of the
LVMs (Fig. 1f), which are continuations of the caudo-cranial
waves accompanying the LVM opening, and commonly observed
on mammalian vocal folds (Supplementary Fig. 2). Sound
pressure excitation occurred both during opening and closing
events (Fig. 1g). The presence of both VC1 and VC2 components
confirms our hypotheses that a caudo-cranial travelling tissue
surface wave is present and that this wave is associated with
sound producing events in pigeons.
To test whether the presence of a tissue surface wave is a shared
trait across birds, we studied syringeal oscillatory behaviour in six
additional bird species from five orders. We selected species
ranging in size (B15 g–200 kg) and vocal complexity, each with
highly divergent syrinx morphologies, containing one or two
paired oscillators and controlled by zero to seven pairs of intrinsic
syringeal muscles (Figs 2 and 3). We reconstructed threedimensional geometries including bone, soft tissue and muscle
based on micro-computed tomography (CT) scans (Figs 2a
and 3a). Our study species included the largest extant bird
(ostrich), a phylogenetically basal paleognathid bird (elegantcrested tinamou) and several neognathids: a parrot (cockatiel),
dove (Barbary dove) and two songbirds (zebra finch and
Bengalese finch). Medio-lateral vibration components (VC1)
were observed in each species using tracheal endoscopy (Figs 2c
and 3c). These VC1 vibrations resulted in complete syringeal
closure within an oscillatory cycle in each species except for the
ostrich, where full closure was never observed, more closely
resembling human breathy phonation40.
In each species, regardless of syringeal morphology, we
confirmed the presence of the caudo-cranial component (VC2)
of a travelling tissue wave using spatiotemporal analysis of
syringeal inner wall displacement, syringeal opening and/or
mEGG (Figs 2d and 3d, Supplementary Movies 1–6). To further
test if the tissue wave was present over a range of fundamental
frequencies, we subjected the syrinx to bronchial pressure ramps
and measured the direction and speed of the tissue wave when
present (Methods section). During self-sustained oscillations over
a range of F0 values, the tissue wave was always present (Fig. 4)
and running from caudal to cranial (indicated by all velocities
being positive in Fig. 4). The wave speed did either not vary
(pigeon and tinamou; linear regression, P ¼ 0.56 (n ¼ 10) and
P ¼ 0.26 (n ¼ 10), respectively) or increased significantly with
F0 (zebra finch and cockatiel; linear regression, Po0.001
(n ¼ 11) and Poo0.001 (n ¼ 118) respectively) within a range of
0.5–4.0 m s  1.

Furthermore, to investigate the relationship between tissue
motion dynamics and sound generation events within a single
oscillatory cycle, we quantified the delay between sound
excitation events and first opening and closing events of the
syringeal passage over a range of F0 values (Fig. 5). In the pigeon,
both opening and closing events were precisely accompanied by
an acoustic excitation at very short delays of 170 and 90 ms,
respectively (Fig. 5a), which are both below the 250 ms temporal
accuracy of the opening and closing event timing (that is, one
frame duration of high-speed video). The tinamou showed a
strong acoustic excitation on syringeal opening, with a delay of
100 ms, also below the 250 ms temporal accuracy of the opening
and closing event timing (Fig. 5b). A second, often weaker,
excitation occurred 1.63 ms after closing. The zebra finch showed
a very precisely timed strong acoustic excitation on syringeal
closing at a delay of 40 ms (at a temporal accuracy of 33 ms;
Fig. 5c).
In conclusion, these data confirm our first hypothesis that a
caudo-cranial travelling tissue surface wave is present across a
range of syringeal morphologies and sizes. Furthermore, our data
confirm a close association between opening/closing event timing
and sound generation events within single oscillatory cycles
across a range of species. Collectively, our data thus provide the
essential lacking demonstrations of syringeal dynamics and sound
generation events required to conclusively show that MEAD
theory is applicable to sound production in birds.
Vocal control redundancy ex vivo. To test our second hypothesis
that the physiological control space of the vocal organ is
redundant for key acoustic parameters, we systematically
investigated the relationship between syringeal control parameters and acoustic output ex vivo. In mammals, the brain can
control bronchial pressure and laryngeal muscle activity to
modulate aerodynamic forces and vocal fold tissue properties (for
example, geometry and elasticity) and attain a target fundamental
frequency (F0)12–15. In birds, in addition to bronchial pressure,
pressure in the interclavicular air sac (ICAS) enclosing the syrinx
also affects F0, as suggested by models7,9,10 and experimental
manipulations10. We first quantified syringeal oscillatory
behaviour in different species as a function of bronchial and air
sac pressure (Figs 6 and 7a–c). For each species the F0 ranges
produced ex vivo corresponded well to the lower-end distribution
of spontaneous vocalizations (Figs 6 and 7b,c). We further found
that in each investigated species, multiple different combinations
of bronchial and ICAS pressures could achieve the same target
frequency (iso-F0 contours in Figs 6 and 7c), indicating that F0
control is redundant within the pressure control space. To
investigate how much two other important vocal parameters, that
is, sound pressure level (SPL) and sound quality, changed with F0,
we quantified the SPL and Wiener entropy (WE) along evenly

Figure 1 | The MEAD theory explains sound production in the domestic pigeon (Columba livia) syrinx. (a) Schematic illustration of vocal organs in
mammals (larynx) and birds (syrinx) showing the two vibrational components essential to MEAD theory: medio-lateral oscillation (VC1), and caudo-cranial
phase differences (VC2). The latter are related to a tissue surface wave moving cranially and cause the inner wall of the vibratory tissue to be divergent during
closing (solid line, black arrows) and convergent during opening motion (dashed line, white arrows). The syrinx is suspended in the ICAS. Three pressures act
on the syrinx: subsyringeal (psub), suprasyringeal (psupra) and ICAS pressure (pICAS). (b) Annotated micro-CT scan of the pigeon syrinx showing LVM (yellow)
and their inner edges (right, green; left, red), muscle (pink) and tracheal rings (T1–4). (c) Transilluminated syrinx filmed at 4 kHz with traced outlines of right
(green) and left (red) LVM during closing, collision and opening phases of oscillation. The region shown is indicated by the dotted box in b. White
arrows indicate movement. Circles indicate mEGG electrode pairs channel A (cyan fill) and B (black fill). (d) Two DKG (compiled at dotted lines in c) and
simultaneous mEGG recordings show consistent caudo-cranial time delay of syringeal closure between the caudal line DKG2 and cranial line DKG1 (DtDKG) as
well as between the caudal and cranial mEGG signals (DtmEGG). (e) Spatiotemporal displacement analysis of the left LVM (top) shows a caudo-cranial travelling
tissue wave (dashed line), which is also present in the glottovibrogram of syringeal opening (bottom). (f) Simultaneous tracheal endoscopic VKG confirms
opening and closing event timing of the cranial LVM edge, and reveals tissue waves (arrowheads) travelling mediolaterally over the epithelia. (g) Abrupt
positive and negative excitation of the acoustic waveform aligns with opening and closing of the LVM cranial edge.
4

NATURE COMMUNICATIONS | 6:8978 | DOI: 10.1038/ncomms9978 | www.nature.com/naturecommunications

ARTICLE

NATURE COMMUNICATIONS | DOI: 10.1038/ncomms9978

Ostrich

Tinamou
Muscle
ICAS

1m

Tracheal view

LVM

10 mm

ventral

Muscle

LVM
100 mm

1 mm

Ventral

1

Tracheal VKG

Right

Left

Pa

Left

Opening cranial edge

GVG

DKG

VC2 wave

Caudal (pos 1)

Sound

Closing

3

Cranial

Cranial (pos 2)

VC2

Right

VC2 wave

2 ms

Caudal

F0 = 150 Hz

0.4

1 mm

Tracheal VKG

VC1

DKG

0.2

0

Syringeal opening
(mm)

2
Dorsal

Left
LVM

Right
LVM
VKG

F0 = 310 Hz

0

0

2 ms

5 ms

–0.4

–0.2
Time

Time

Figure 2 | MEAD theory explains syringeal sound production in paleognaths. (a) Three-dimensional syrinx geometry of the ostrich (Struthio camelus) and
elegant-crested tinamou (Eudromia elegans) showing bone (grey), LVM (yellow) and muscles (pink). Both views are ventral. Muscle is rendered transparent
in tinamou syrinx to show vibratory tissue. (b) High-speed imaging stills of endoscopic tracheal (ostrich) and transilluminated frontal views (tinamou) of
the syrinx at two different phases during the oscillatory cycle. The region and location of these stills is indicated with a dotted box in a. For the ostrich the
position of VKG analysis (dotted horizontal line) and DKG analysis (solid white line from point 1–2) presented in panel c is indicated. Transillumination
provides sufficient detail to identify inner edges of right (green) and left (red) LVM in tinamou. (c) Evidence for medio-lateral vibrations (VC1) using
tracheal endoscopic VKG. In the ostrich predominantly the right side vibrated. Full opening and closure of the cranial edge can be observed in tinamou.
(d) Evidence for the caudo-cranial component (VC2) of the tissue wave as demonstrated by DKG for ostrich, and glottovibrogram for tinamou.
(e) Synchronised acoustic waveforms.

spaced iso-F0 contours. A point on the iso-F0 contour was
considered to be redundant in all three acoustic parameters if
neither SPL nor WE changed along the iso-F0 contour (Methods
section). The percentage of the explored control space that
demonstrated redundancy for the three vocal parameters
measured 79.8±21.4% (mean±s.d.) for pigeon (N ¼ 7),
83.0±7.7% for Barbary dove (N ¼ 3), 67.8±4.7% for tinamou
(N ¼ 3), 71.4±18.9% for cockatiel (N ¼ 2), 44.3% for zebra finch
right hemisyrinx, 62.4±13.3% for Bengalese finch left hemisyrinx
(N ¼ 6) and 71.6±18.6% for Bengalese finch left hemisyrinx
(N ¼ 6). These results demonstrate that F0 control by pressure
ex vivo is redundant in subspaces within the pressure control
space, and that this redundancy is conserved across species
exhibiting a wide range of syringeal morphologies. Therefore, in
the ex vivo preparation redundant acoustic output can be
achieved by modulating air pressure, without any active control
of syringeal muscles.
We furthermore examined whether muscular control
introduced an additional source of redundancy and quantified
the effect of syringeal muscle recruitment on F0 using local
micro-stimulation (Fig. 7d). Muscle stimulation caused an F0
increase of 300–750 Hz and 75–200 Hz for tinamou and zebra
finch, respectively (Fig. 7e). The F0 range achieved by pressure

and muscle recruitment overlapped within each species (tinamou
(N ¼ 1) and zebra finch (N ¼ 5)) and multiple different
combinations of muscle recruitment and bronchial pressure
resulted in the same F0. These results thus demonstrate that
muscular control of F0 is redundant in the ex vivo preparation.
Taken together, the above data confirm our second hypothesis
that the physiological control space of the vocal organ ex vivo is
redundant for key acoustic parameters across a range of syringeal
morphologies and sizes.
Discussion
Our data establish that birds use MEAD as the primary physical
mechanism for sound production with strong similarities to
mammalian MEAD systems: First, we demonstrate the presence
of a tissue wave that travels from the caudal to cranial end of the
syringeal vibratory tissue (Figs 1–4). This tissue wave thus causes
the syringeal vibratory tissue shape to be convergent when
opening, and divergent when closing during expiratory sound
production, and can be considered analogous to the caudo-cranial
mucosal wave observed in mammalian vocal folds. Our data
therefore strongly suggest that in birds, just as in mammals, the
dominant asymmetric forcing function over the opening and

NATURE COMMUNICATIONS | 6:8978 | DOI: 10.1038/ncomms9978 | www.nature.com/naturecommunications

5

ARTICLE

NATURE COMMUNICATIONS | DOI: 10.1038/ncomms9978

Zebra finch

Cockatiel

Muscle
Muscle

Medial
Labia
Lateral
Labia

B3
LVM
100 mm

10 mm

1 mm

1 mm

B3
Ch B

Ch B

Right
LL

Ch A

Right

Left

Closing

Opening cranial edge

Right

Closing
Midline
Opening cranial edge

Left

μEGG

μEGG

ΔtEGG

ΔtEGG
1

F0 = 1450 Hz

Sound
Pa

0.5

GVG

Ch A
Ch B

VC2

0.4

VC2 wave

0

Syringeal opening
(mm)

VC1

Tracheal VKG

Tracheal VKG

Ch A

6
F0 = 420 Hz

0

0

2 ms

1 ms
–0.4

–1
Time

Time

Figure 3 | MEAD theory explains syringeal sound production in neognaths. (a) Three-dimensional syrinx geometry of the cockatiel (Nymphicus hollandicus,
ventral view) and zebra finch (Taeniopygia guttata, dorsal view) showing bone (grey), LVM (yellow) and muscles (pink). Muscle is rendered transparent in
zebra finch right hemisyrinx to show vibratory tissue. (b) High-speed imaging stills of frontal transilluminated frontal views of the syrinx at two different phases
during the oscillatory cycle at view and location indicated with a dotted box in a. While transillumination does not provide detail on motion in the cockatiel
syrinx, it provides sufficient detail to identify inner edges of vibratory tissue in female zebra finches (green). Circles indicate location of mEGG electrodes (cyan,
channel A; black, channel B). (c) Tracheal endoscopic VKG shows opening and closing events and thus evidence for medio-lateral vibrations (VC1). In the
zebra finch sound production was induced in the right hemisyrinx. (d) Evidence for caudo-cranial component (VC2) of the tissue wave as demonstrated by
mEGG phase delays for cockatiel and zebra finch and glottovibrogram analysis for zebra finch. (e) Synchronised acoustic waveforms.

closing phases of vibration essential to maintain self-sustained
oscillation is not formed by the mass inertance of the air column
in the vocal tract23,24, but by the tissue-wave-induced intraglottal
pressure changes. This mechanism reduces the dependency of the
self-sustained syringeal oscillations on acoustic resonances of the
vocal tract and allows for an expanded F0 range of vocalization14,
which in addition to labial morphology41 could aid birds in
extending their F0 range. Second, the magnitude range of the
tissue-wave speed measured (0.5–3.0 m s  1) is in excellent
agreement with values reported for the mammalian larynx30,41–
46 (human41,43–45, 0.5–2.0 m s  1; dog42,46,47, 0.5–2 m s  1; calf45,
0.4 m s  1; and an African elephant30, 1.2 m s  1) suggesting that
no scaling effects occur. Third, our data show that in some species
(zebra finch, cockatiel) the tissue-wave speed increased with F0,
similar to what has been observed in humans47, whereas in other
species (pigeon, tinamou) the tissue-wave remained constant with
respect to F0 (Fig. 4). The different relationships are possibly
related to biomechanical properties of the tissue layers involved in
the vibrations or varying aerodynamic forces acting on the
vibrating structures. Future work will be required to establish the
respective causality. Last, our findings demonstrate that acoustic
excitation occurs at opening and/or closing of the syringeal
passage (Fig. 5), which strongly suggests a causal relation between
6

tissue vibration and generation of acoustic energy by modulating
the glottal airflow as found in the mammalian larynx28–30. While
in humans the maximum acoustic excitation has been shown to
occur mainly at the instant of glottal closure28,29, a recent study
reported that in an excised elephant larynx the maximum
acoustic excitation occurred at the instant of glottal opening30.
The mammalian larynx thus exhibits more diversity than
previously thought in what movements cause the predominant
acoustic excitation. Our work demonstrates that this diversity is
also present in birds and that more comparative studies are
needed to explain the causal link between airflow, tissue vibration
and acoustic excitation.
In conclusion, we find that despite the large diversity present in
syringeal morphology of the bird species included in this study,
all use MEAD as the primary physical mechanism of sound
production, supporting MEAD-based approaches to modelling
avian sound production7–11. Moreover, our findings suggest that
despite their different evolutionary origins1, laryngeally vocalizing
mammals and syringeally vocalizing birds have converged on the
same physical mechanism for vocalization.
Furthermore, our data show that key acoustic parameters, such
as particular F0 values, can be achieved by multiple distinct
combinations of respiratory pressure and syringeal muscle

NATURE COMMUNICATIONS | 6:8978 | DOI: 10.1038/ncomms9978 | www.nature.com/naturecommunications

ARTICLE

NATURE COMMUNICATIONS | DOI: 10.1038/ncomms9978

vVC2 (m s–1)

1.0

Pigeon

0.8
0.6
0.4
0.2
0

vVC2 (m s–1)

4

190

200

210

220

230

240

250

260

270

Tinamou

3
2
1
0

2
vVC2 (m s–1)

P = 0.56
180

P = 0.26
230

240

250

260

270

280

290

300

310

320

330

Zebra finch

1

P < 0.001
0

240

260

280

300

320

340

360

380

400

Cockatiel
vVC2 (m s–1)

3
2
1
P << 0.001
0

1,650

1,700 1,750 1,800 1,850 1,900
Fundamental frequency (Hz)

1,950

2,000

Figure 4 | The caudo-cranial tissue wave is present across a range of
fundamental frequencies. Magnitude of the caudo-cranial component
(vVC2) of the travelling tissue-wave’s velocity in (a) pigeon and (b) tinamou,
(c) zebra finch and (d) cockatiel. The positive velocity values indicate that
the wave travelled from caudal to cranial. Wave speed remained constant
with F0 in pigeon and tinamou (Linear regression, P ¼ 0.56 (n ¼ 10) and
P ¼ 0.26 (n ¼ 10), respectively), but increased significantly with F0 in zebra
finch and cockatiel (linear regression, Po0.001 (n ¼ 11) and Poo0.001
(n ¼ 118), respectively).

recruitment ex vivo (Figs 6 and 7). This finding thus supports the
hypothesis that subspaces within the entire physiological control
space of the vocal organ ex vivo are redundant for acoustic
parameters.
Our work represents the first systematic exploration of
syringeal behaviour in a controlled environment using pressure
differentials up to 2 kPa. Although it remains unresolved to what
degree birds can independently control bronchial pressure and
ICAS air sac pressure in vivo, ventilation flow across the lungs
requires the existence of pressure differences between different
parts of the respiratory system48–50. Especially in dynamic
situations such as vocal behaviour, the passive geometry of the
air sac system can lead to substantial pressure differences between
air sacs51,52 and indeed transient pressure differences up to
1.0 kPa occur during vocalizations in Barbary doves7,53. These
pressure fluctuations can result in local transmural pressure
differentials over the syringeal walls and thus in transmural forces
on the vibrating tissues7,37. The individual bird can either avoid
or exploit such regimes during vocalization. However, regardless

of pressure differentials, only bronchial pressure in combination
with different levels of recruitment in a single vocal muscle leads
to redundancy in F0 (Fig. 7e).
How acoustic redundancy of the vocal organ ex vivo reflects the
functional motor redundancy in vivo remains to be explored. The
speed at which consecutive motor commands can be executed
in vivo (either on the same side of the syrinx or when vocal
production switches rapidly from one side to the other within a
complex syllable) may cause a significant reduction of the
available functional redundancy, especially because at least some
songbirds possess superfast syringeal muscles that produce peak
force in o5 ms (refs 54,55). Here, we did not investigate the rapid
time-varying pressure and muscle recruitment patterns that occur
in vivo, and indeed for most species more detailed experimental
in vivo data are required to meaningfully explore the highdimensional control space ex vivo. On the other hand, we show
that redundancy emerges when only varying the activity level of a
single vocal muscle. Redundancy can thus be expected to
increase when the brain can produce a given acoustic output by
choosing from a large set of redundant motor commands when
coordinating the B16 syringeal muscles56 with several thousand
motor units57.
A recent study used zebra finch song acoustics to infer lowdimensional motor commands11, which were then used to
generate synthetic songs that closely resembled the acoustic
output of natural vocal behaviour, and evoked auditory responses
that closely resembled the pattern of premotor activity. These
results led the authors to suggest that premotor area HVC
encodes a low-dimensional forward model of gesture dynamics11.
In contrast, our data indicate that particular acoustic outputs can
be specified by multiple motor states, suggesting that motor
commands cannot be uniquely inferred from acoustics. It will
therefore be important for future studies to investigate whether
the method used in ref. 11 to infer motor commands from
acoustic data yields the actual combination of control parameters
employed by the bird to generate the sound, rather than a
different set of control parameters that redundantly produces the
same acoustic output. Similarly, future ex vivo studies employing
time-varying stimulus patterns across multiple muscles will allow
us to explore the limits of vocal redundancy during sophisticated
motor trajectories similar to those in behaving animals. Despite
the apparent conflict between our findings and elements of
ref. 11, it is important to note that the instantiation of a lowdimensional forward model within HVC—which is multiple
synapses upstream from the vocal and respiratory muscles—is
not incompatible with a redundant vocal organ. Indeed, one
strength of low-dimensional encoding in HVC is that it would
allow HVC to represent control parameters (for example, vocal
fold tension) without regard to the details of the redundant
combinations of muscle contraction states required to achieve
that tension.
In many fine motor control systems such as arm reaching, the
brain must negotiate the so-called motor redundancy problem39,
in which a particular behavioural target can be achieved by vast
numbers of motor commands. However, one potential advantage
to a redundant control space is that it allows the brain to find
subspaces of possible motor commands (for example, along the
iso-F0 contours in Figs 6 and 7c in our ex vivo paradigm) rather
than searching for unique motor commands to achieve the target.
Crucially, redundant control spaces allow variability in taskirrelevant directions58,59. In vocal production, the brain may
select from within this motor command space to meet other
demands, such as acoustic targets in future and/or past syllables.
Consequently, we speculate that vocal motor redundancy may
simplify trial-and-error learning during song acquisition by
allowing the brain to rapidly discover one of many motor

NATURE COMMUNICATIONS | 6:8978 | DOI: 10.1038/ncomms9978 | www.nature.com/naturecommunications

7

ARTICLE

NATURE COMMUNICATIONS | DOI: 10.1038/ncomms9978

Acoustic excitation with syringeal
Rock pigeon

Opening

350

150

50
0
180

200

220

240

Opening

100

No. observations per bin

200

Closing

Closed

250

Open

Phase ϕ (deg)

–90 μs

10

300

8
8
6

6
4

4

2

2

0
–1

260

Tinamou

0

0
–1

1

14

350

0

10

–100 μs

1

1.63 ms

250
200
Opening Closing

150
100
50
0
220

240

260

280

300

No. observations per bin

12

300
Phase ϕ (deg)

Closing
10

–170 μs

8

6

6

4

4
2

2
0
–1

320

8

10

0

0

1

0

1

2

3

Time after opening (ms)
Zebra finch

250
200
150
100
Opening

Phase ϕ (deg)

300

50
0
240 260 280 300 320 340 360 380 400
Fundamental frequency (Hz)

18
No. observations per bin

Closing

350

–40 μs

16
14
12
10
8
6
4
2
0
–1
0
1
Time after closing (ms)

Figure 5 | Sound excitation is associated with syringeal opening and closing events. The relative duration of the syrinx in open (white bar starting at
phase j ¼ 0°) and closed (grey bar) state within a single oscillatory cycle (full oscillation equals 360°) over a range of fundamental frequencies in
(a) pigeon, (b) tinamou and (c) zebra finch (left) based on transillumination data (Methods section). Sound excitation events are associated with opening
(red circles: mean±s.d.) and/or closing (blue circles: mean±s.d.) of the syrinx. The thick vertical black line at the base of each column indicates the
duration of a single movie frame relative to the cycle duration and thus the timing accuracy for determining a syringeal closing or opening event. Note that
in zebra finch the lines are short due to high frame-rates used (25 kHz). The histograms (right) show that timing of sound excitation is closely associated
with syringeal opening (red) and closing (blue) events.

solutions60 before learning how to further improve performance
by exploiting motor redundancy39. Vocal redundancy may
therefore have aided, or even been necessary for, the
evolutionary development of vocal learning. Future studies
might evaluate this speculation by quantifying whether learned
vocalizations exploit vocal redundancy. Furthermore, because
many species do not learn their vocalizations and many vocal
learners produce innate vocalizations in addition to learned ones,
future work might also determine whether innate vocal motor
programs similarly exploit redundancy to improve performance.
8

Our ex vivo results suggest that the redundant control of key
vocal parameters represents a significant aspect of avian vocal
control, and one that merits further investigation. Because
redundancy in F0 control by pressure and muscle recruitment
is also observed in various simplified computational models of the
human larynx13, and songbird syrinx9 as well as in ex vivo dog
larynx preparations61,62, we propose that vocal control
redundancy is a typical feature of MEAD sound production
systems and hence a common feature of vocal production and
control in mammals and birds.

NATURE COMMUNICATIONS | 6:8978 | DOI: 10.1038/ncomms9978 | www.nature.com/naturecommunications

ARTICLE

NATURE COMMUNICATIONS | DOI: 10.1038/ncomms9978

Airsac pressure (kPa)

Frequency (kHz)

Rock pigeon
0.8

0.4

0

Airsac pressure (kPa)

Frequency (kHz)

Barbary dove

0.5

0
500 ms

Airsac pressure (kPa)

Frequency (kHz)

Cockatiel
4

2

0
500 ms

2
R
1
0

Airsac pressure (kPa)

Frequency (kHz)

L

0.2

180

1

0.1

0
0

1

3

2

500

0.4

450

2

400

0.2

1
0

0

F0 (kHz)

3

0

1

2

3

0

F0 (kHz)
2

3

1,100
2
1

1,500
1,800

1
0
0

Left hemisyrinx

Bengalese finch
3

230
220

2

500 ms

1

F0 (kHz)
0.3

3

3

1

2

3

0

Right hemisyrinx
F0 (kHz)
900

2,350
1,900
1,500

2
1
0
0

1

2

800
1

700

1
2
2
30
Bronchial pressure (kPa)

3

0

200 ms

Frequency (kHz)

0.2
0.1
0

500 ms

0.3
0.2
0.1

pb (kPa)

Frequency (kHz)

Ostrich
0.3

0

2

Sound
500 ms

0

Figure 6 | Pressure control of F0 in sound production is redundant across avian taxa. Spectrograms and oscillograms of spontaneous vocalisation (left)
and representative pressure control spaces ex vivo (right) of (a) pigeon (Columba livia), (b) Barbary dove (Streptopelia risoria), (c) cockatiel and
(d) Bengalese finch (Lonchura striata domestica). The F0 ranges of sounds produced ex vivo (grey vertical bar right of spectrogram) correspond well to the
lower range of spontaneous vocalizations. Pressure control may allow higher F0 values, as indicated by the grey bar fading out. Iso-F0 contours (values in
Hz) are overlaid on the pressure control spaces and shown in white when redundant for all three acoustics parameters (F0, SPL and WE), or black when
not. In the Bengalese finch, the left hemisyrinx produces higher F0 than the right hemisyrinx, corroborating earlier in vivo studies70. (e) Spectrogram and
oscillogram (left) of ostrich mating call, and ex vivo sound production including bronchial pressure (pb) patterns underlying phonation (right). A complete
pressure control space was not obtained for this species.

NATURE COMMUNICATIONS | 6:8978 | DOI: 10.1038/ncomms9978 | www.nature.com/naturecommunications

9

ARTICLE

NATURE COMMUNICATIONS | DOI: 10.1038/ncomms9978

Tinamou

Zebra finch
ICAS
membrane

M. syringealis

Right
M. syringealis
ventralis

pICAS

Frequency (kHz)

psub
2

2

1

1

0

0
100 ms

200 ms

Time (s)
F0 (kHz)
Airsac pressure (kPa)

F0 (kHz)
0.6

2

3

450

0.4
2

450
350

1

0.4

400
1
0.2

0.2

300
300

Frequency (kHz)

0

0

1

2

0

0

3
0
Bronchial pressure (kPa)

2

2

1

1

0

0

1

2

0

F0stim

F0pre
100 ms

100 ms
Time (s)

1,200
F0stim

800
F0 (Hz)

1,000
700

800
600

600
F0Pre

400
1.6

2.0

500
1.0
2.4
Bronchial pressure (kPa)

2.0

Figure 7 | Motor control of F0 in sound production is redundant. (a) Three-dimensional geometry of the tinamou and zebra finch syrinx with pressure
control parameters; bronchial or subsyringeal (psub), and ICAS pressure (pICAS). Stimulated muscles are indicated. (b) Sound spectrogram (top) and
oscillogram (bottom) of a tinamou honksqueal call and zebra finch song motif. (c) Representative pressure control spaces ex vivo of one individual. Iso-F0
contour as in previous figure (values in Hz). (d) Sound spectrogram (top) and muscle stimulus (bottom) during syringeal muscle stimulation. Black dots
indicate measurements points. (e) Fundamental frequency (F0) is affected by muscle stimulation (blue dots, no stimulation; red dots, maximal stimulation)
and pressure. The grey bar indicates the individuals’ frequency range achieved by varying only pressure. Pressure control may allow higher F0 values, as
indicated by the grey bar fading out. Multiple different combinations result in the same frequency, for example, 500 and 650 Hz in tinamou and zebra finch,
respectively, (green horizontal line).

10

NATURE COMMUNICATIONS | 6:8978 | DOI: 10.1038/ncomms9978 | www.nature.com/naturecommunications

ARTICLE

NATURE COMMUNICATIONS | DOI: 10.1038/ncomms9978

Methods
Subjects. To study sound production mechanisms and the effect of muscle stimulation on sound production, we used 12 adult domestic pigeons (Columba livia;
order Columbiformes; nine males, three females), four domestic Barbary doves
(Streptopelia risoria; order Columbiformes; three males, one female), three elegantcrested tinamous (Eudromia elegans; order Tinamiformes; two males, one female),
two ostriches (Struthio camelus; order Struthioniformes; sex unknown), three
cockatiels (Nymphicus hollandicus; order Psittaciformes; two males, one female),
six Bengalese finches (Lonchura striata domestica; order Passeriformes; six males)
and 19 zebra finches (Taenopygia guttata, order Passeriformes; five adult males;
two juvenile males (B40 dph); 12 females). Ostrich syrinxes were obtained from a
local breeding farm. They were transferred to ice immediately after extraction, flash
frozen in liquid nitrogen and stored at  80 °C. Bengalese finches were kept in
indoor aviaries on a 14-h light:10-h dark light cycle with food and water ad libitum
(Emory University, Atlanta, GA, USA). Zebra finches were kept in indoor aviaries
on a 12-h light:dark photoperiod with food and water ad libitum (University of
Southern Denmark (SDU), Odense, Denmark). All other animals were kept
in 3  6  2 m outdoor aviary with food and water ad libitum (SDU, Odense,
Denmark). Pigeons, doves, cockatiels and tinamous were obtained from local
breeders. Bengalese finch experiments were carried out at Emory University, and
all other experiments at SDU, Denmark. All experiments were conducted in
accordance with the Institutional Animal Care and Use Committee of Emory
University and of SDU.
Surgical and mounting procedures. Animals were euthanized with isoflurane and
cooled on ice or icepacks. The syrinx and associated blood vessels were dissected
out using a stereoscope (M165-FC, Leica Microsystems) through a ventral incision
to the sternum while regularly flushing with oxygenated Ringers solution (5 °C,
recipe cf. refs 54–56), and transferred to a Sylgard-covered petri dish on ice
containing oxygenated Ringer solution. The syrinx was cleaned of fat and
connected to species–specific tubing assemblies of non-reactive polyethylene tubing
(Instech Salomon, PA, USA) in the experimental chamber with 10/0 nylon suture
(S&T, Neuhausen, Switzerland). Special care was taken to mount the syrinx of each
species in its natural position by leaving structural elements, such as bronchidesmus and collagen tissue strands, intact. Two pairs of mEGG electrodes (see
below) were inserted in tissue and fixed with 10/0 suture at locations indicated in
Figs 1 and 2. For muscle stimulation experiments ex vivo, perfusion pathways were
established in the zebra finch that allowed perfusion through the original syringeal
vasculature. All arteriole capillaries were localized and closed using 10/0 suture
immediately after exposure of the syrinx by a ventral incision to the sternum. The
syrinx was then transferred to a Sylgard-covered petri dish on ice and connected to
micro-perfusion lines, gravity fed with oxygenated Ringers. Perfusion pressure was
kept at 2–4 kPa until patent perfusion was observed and immediately reduced to
1–2 kPa. The syrinx was typically mounted ventral side up in the experimental
chamber to two bronchial and one tracheal connector (Supplementary Fig. 1).
Experimental chamber design. To study syringeal sound production, we
developed an experimental chamber that allowed for study of the syrinx ex vivo
(that is, intact perfused organ) under controlled conditions. This experimental
chamber was milled out of aluminium (temperature controlled) or PVC and
covered by an airtight glass lid that allowed pressurization and a clear view
(Supplementary Fig. 1). The glass was either coated with nano-particles (Percenta
AG, Germany) or heated to 10 °C above chamber temperature by a 18-mm
diameter Formvar coated Nichrome wire assembly (A-M Systems, Sequim, WA,
USA), to prevent droplet forming fogging up the glass. In the floor centre of the
chamber, a 1-mm raised edge with glass bottom covered in Sylgard allowed
transillumination of the syrinx. We separately controlled pressure in the bronchial
connectors (pb) and chamber (pICAS) with dual valve differential pressure PID
controllers (model PCD, 0–10 kPa, Alicat Scientific, AZ, USA), referenced to
atmospheric pressure, and a response time of 1 ms. Bronchial mass flow was
measured with micro-electro-mechanical system (MEMS) flow sensors of various
ranges (PMF series, Posifa Microsystems, San Jose, USA) and a response time of
1 ms. The supply of pressurized air to the controllers was heated and humidified
with deionised water in a pressure cooker. Excess vapour condensed in a downstream second pressure cooker. The pressure controllers were temperature controlled at 37 °C (HD4034 with PT100 sensor, Hotek Technologies, WA, USA)
ensuring pre-heated, fully humidified air to the syrinx. The temperature of the
aluminium experimental chamber was controlled with a recirculating water bath
(Julebo F12-ED, Seelbach, Germany) and logged at 1 s intervals with a 1-mm
diameter J-type miniature temperature probe (USBTC01, National Instruments,
TX, USA) 1–2 mm away from the syrinx. Perfusion fluids attained the same
temperature as the chamber due to physical contact with a platform in the chamber
(labelled ‘Perfusate platform’ in Supplementary Fig. 1). During the mounting
process the temperature of the chamber was kept at 7 °C.
Image acquisition and analysis. The syrinx was imaged through the chamber lid
with a high-speed camera (MotionPro-X4, 12 bit CMOS sensor, Integrated Design
Tools, Inc.; 4,000–25,000 frames s–1) mounted on a stereomicroscope (M165-FC,
Leica Microsystems). For the zebra finch, we also used a more light-sensitive 16 bit

high-speed camera (Fastcam SA1, Photron, San Diego, CA, USA; 20,000–35,000
frames s  1). We used a 1.2-mm (Schölly Fiberoptics, Denzlingen, Germany) and
300-mm (Advanced Inspection Technologies, Melbourne, FL, USA) diameter
flexible fiberscope for endoscopic imaging. The low-light images obtained with the
endoscopes were captured with a videokymographic (VKG) system (model 2156,
Cymo B.V., Groningen, The Netherlands), which combines a high-speed linescan
camera (7,200 line images s  1) with a full frame CMOS camera (25 frames s  1).
The analogue video output from the VKG was captured together with the
microphone and a synchronization signal using a capturing device (Intensity
Extreme, Black Magic Design, Australia). Light for transillumination was provided
by a 1,700 lumen LED (Luxeon S, Philips, The Netherlands) powered by a stable
power source (PS23023, HQ Power, Belgium) or plasma light source (HPLS200,
Thorlabs, Germany) through liquid light guides and reflected of a 45° angled silver
coated prism (MRA series, Thorlabs) to absorb heat.
Kymography. A kymogram displays pixel intensity over time along a set line in an
image. The VKG system used here generates a kymogram at the fixed position of
the linescan camera63. Digital kymograms (DKG) are kymograms along any
arbitrary line in a digital image64 and were extracted from high-speed image
recordings using Matlab (The Mathworks). Two kymograms in Supplementary
Fig. 2 were previously published: human from ref. 40, and elephant from ref. 30.
Micro-electroglottography. As an independent proxy for kinematics of the
syringeal vibratory tissues, we used EGG. This non-invasive technique measures
electrical impedance between electrodes and is commonly used to quantify vocal
fold contact area in the larynx65. mEGG electrodes consisting of single stranded
25 mm diameter Formvar coated Nichrome wires (A-M Systems) were inserted as
described above and connected to a modified two-channel electroglottograph
(model EG2, Glottal Enterprises Inc. NY, USA).
Data acquisition and synchronisation. Sound was recorded with a ½ inch
pressure microphone-pre-amplifier assembly (model 46AD with preamplifier type
26AH, G.R.A.S., Denmark), amplified and high-pass filtered (10 Hz, 3-pole
Butterworth filter, model 12AQ, G.R.A.S.). The microphone sensitivity was
measured before each experiment (sound calibrator model 42AB, G.R.A.S.).
The microphone was placed at 2–3 cm from the tracheal connector outlet in the
acoustic near field, and on a 45° angle to avoid the air jet from the tracheal outlet.
The sound signal was time shifted for the travelling distance from vibratory
membranes to microphone. Microphone, mEGG, pressure and flow signals were
low-pass filtered at 10 kHz (custom-built filter). These signals together with
synchronization signals from camera systems and muscle stimulators were
digitized at 50 kHz (USB 6259, 16 bit, National Instruments, Austin, TX, USA).
These signals were synchronized with all imaging systems with an accuracy of
o21 ms before each experiment. All control and analysis software was written in
Labview (National Instruments) or Matlab.
Tissue-wave imaging protocol and analysis. Transillumination successfully
visualized the inside outlines of vibratory tissues in the domestic pigeon, Barbary
dove, elegant-crested tinamou and zebra finch. High optical density of the ostrich
and cockatiel syrinx did not allow transillumination. In the tinamou, we dissected
apart M. syringealis to allow LVMs imaging using transillumination. In zebra finch
transillumination was successful in adult females and juvenile males, but attempts
in adult males were unsuccessful due to the high optical density of M. syringealis
ventralis (VS) muscles. In the zebra finch sound production was induced in the
right hemisyrinx. Substantially different requirements in lighting conditions did
not allow for simultaneous transillumination and tracheal endoscopy in zebra
finches.
To image syringeal oscillatory behaviour we subjected the syrinx to a single
bronchial pressure ramps at constant air sac pressure, while filming from 5,000 to
35,000 frames s  1. Our system could acquire and save a maximum of 2–10 s of
data depending on frame rates, which comprised (tens of) thousands of frames.
Due to the image complexity of the data, analysis could not be automatized, and
required frame-by-frame manual analysis. On selected sequences of high-speed
images, the left and right LVM or labia were traced manually in Amira (Visage
Imaging GmbH, Berlin, Germany) and processed in Matlab. We calculated
displacement (for example, Fig. 1e top panel) as current position minus the most
lateral position during oscillation (indicated by the green line ‘min’ on the top left
in Fig. 1e). With both left and right LVM shapes quantified we could then compute
the syringeal opening as a function of position and time, a graph also known as the
glottovibrogram30 as can be seen in Figs 1e,2d and 3d.
To quantify syringeal dynamics over a range of F0 values, full glottovibrogram
reconstructions over several cycles were not suitable due to the labour intensive
manual tracing required. Therefore we used an alternative approach to determine
the presence of the VC2 component and speed of the tissue wave across a range of
F0 values. For pigeon, tinamou and zebra finch we used transilluminated frontal
views of the syrinx and calculated one DKG at the caudal edge and one DKG at the
cranial edge of the LVM or labia (top two panels in Fig. 1d) at distance Dy
(indicated in middle panel Fig. 1c). DKG’s contain many periodic traces as many
structures move with each oscillation. By tracing selected LVM and labial inner

NATURE COMMUNICATIONS | 6:8978 | DOI: 10.1038/ncomms9978 | www.nature.com/naturecommunications

11

ARTICLE

NATURE COMMUNICATIONS | DOI: 10.1038/ncomms9978

edges in video stills over five oscillation cycles, we could identify the wave
representing the motion of the LVM or labial inner edge in the DKG’s and as such
measure the time difference between maximal lateral position of LVM or labium on
the two DKG’s (DtDKG in Fig. 1d). The wave speed equalled vVC2 ¼ Dy/DtDKG and
was averaged over five consecutive cycles. Because in the cockatiel syrinx
transillumination was not successful, we used cross-correlation between two
mEGG electrode pairs located at distance Dy to calculate time delay (DtmEGG) and
wave speed vVC2. The ostrich was not included in this analysis because both
transillumination was not successful and mEGG did not resolve vibrations due to
lack of syringeal closure. We did not estimate wave speed from the tracheal
endoscopic view because the VC2 velocity component was almost normal to the
imaging plane, leading to large inaccuracies.
In these five consecutive oscillations, we quantified the timing of syringeal
opening and closing from the transilluminated frontal views as the first frame in
which the entire syringeal passage was open or closed, respectively. The minimal
precision for these measurements equals the duration of one frame as indicated by
the black vertical line at the base of each column in Fig. 5. In addition, we
quantified the timing of acoustic excitations in the sound pressure signal. The
timing of the sound signals was corrected for the time needed for the sound to
propagate from the oscillators through trachea and tubing to the microphone,
assuming a sound speed of 340 m s  1. With these parameters we quantified the
timing of acoustic excitation with respect to syringeal opening and closing events
within single oscillatory cycles.
Pressure control space protocol. To explore syringeal oscillatory behaviour in
the pressure control space we subjected the syrinx to a set of bronchial pressure
ramps at randomized and variable air sac pressure, and one additional ramp where
pb equalled pICAS. Ramp speed was 1 kPa s  1. Depending on species, we used
pressures o2.0–4.0 kPa and |pb  pICAS|o2.0 kPa to avoid regimes of high flow
(that is, combinations of high pb and low pICAS) and/or mechanical failure of
syringeal oscillatory structures. These sets were performed without transillumination imaging. SPL and WE were calculated on 100 ms segments by applying a
sliding window with 50 ms steps. If the SPL of these segments were above the 60 dB
re. 20 mPa threshold, segments were zero-padded to the next power of two, and a
4,096 point power spectral density estimate was computed (periodogram method
in Matlab). Additional filtering was used to extract the fundamental frequency (F0)
reliably from the spectral density estimate. WE was calculated as the geometric
mean of the power spectrum over the arithmetic mean of the power spectrum.
We extracted iso-F0 contours in pb, pICAS control space. SPL and WE were
evaluated along the iso-F0 contours at a resolution of 10 Pa (Supplementary Fig. 3).
To evaluate whether points on these iso-F0 contour lines were redundant for all
three quantified acoustic parameters (F0, SPL and WE) we computed acoustic
parameters at a fixed distance of 100 Pa both before and after each point along the
iso-F0 contour line locations in pb, pICAS control space (Supplementary Fig. 3). For
the 100 Pa start and end sections of the contour, only forward and backward values
could be computed. A point along the iso-F0 contour was considered redundant for
all the three parameters if the WE and SPL variation within the 100 Pa interval
was below 0.2 units of WE and 1 dB SPL (ref. 66). This resulted in a logical (1/0)
redundancy array for each iso-F0 contour. The iso-F0 contour is plotted
white when redundant and black when not in the pb, pICAS control spaces in
Figs 6a–d and 7c.
We computed iso-F0 contours at fixed intervals (Columba, 10 Hz; Streptopelia,
50 Hz Eudromia, 50 Hz; Nymphicus, 100 Hz; Lonchura, 100 Hz; Taenopygia, 50 Hz)
within the F0 range observed in the preparation. The percentage redundancy
of a pressure control space was expressed as the ratio of redundant points to
the total amount of points on all iso-F0 contours computed in the pb, pICAS
control space.
Muscle stimulation protocol. Muscles were stimulated with 50 mm diameter
Teflon-coated twisted silver wire electrodes (A-M Systems) that were fixed with
10/0 suture. We used a stimulus isolator (model A395, WPI, Sarasota, FL, USA) to
apply variable currents (0–10 mA) when limited compliance voltage was sufficient
to contract muscles (tinamou). For zebra finches and Bengalese finches, we used
two stimulators (model 14E11, DISA, Herlev, Denmark and model 2100, A-M
Systems, respectively). To ensure muscle specificity, we placed carbon microspheres
(20–40 mm diameter) at regular intervals on the muscles and surrounding tissues
and filmed contractions at 1 kHz. We stimulated the single syringeal muscle
present in tinamous, M. syringealis. In songbirds, activity of the M. syringealis
ventralis correlates to F0 (ref. 67). Therefore we focussed on this muscle to study F0
control in zebra finches (N ¼ 5 adult males). F0 of the sound was determined
30 ms prior (control) and 30 ms after stimulation. We computed 512-point fast
Fourier transforms and used parabolic interpolation around the peak power to
determine F0.
Statistics. Linear regressions were calculated to test the significance of wave speed
as a function of F0 in Matlab. All values are presented as mean±1 s.d.
Morphome reconstruction. We constructed three-dimensional annotated syrinx
morphologies, that is, morphomes, based on IKI contrasted CT scans56. All
12

specimens were scanned at Odense University Hospital with a mCT scanner
(VivaCT 40, Scanco Medical AG, Switzerland) with 10 mm resolution, except the
ostrich syrinx which was scanned with 80 mm resolution on a XtremeCT (Scanco
Medical AG). These scans were annotated manually in Amira for bone, soft tissue
and muscle using protocols cf. ref. 56.
Natural vocalizations. Sound recordings of natural vocalizations were obtained in
previous studies for domestic Barbary dove7, elegant-crested tinamou68 and zebra
finch69. A domestic pigeon call recording was obtained from www.xeno-canto.com
(catalogue nr: XC92264). We recorded the mating calls of two male ostriches at an
open-range ostrich-breeding farm (Langeskov, Denmark) from April to May 2013
using a shotgun microphone (model KMR82i, Neumann, Berlin, Germany) and a
24-bit digital recorder (Olympus LS-100). Cockatiel recordings were made with the
same shotgun microphone and solid-state recorder (model PMD-670, Marantz,
Mahwah, NJ, USA).

References
1. King, A. S. Functional anatomy of the syrinx in Form and Function in Birds
(eds King, A. S. & McLelland, J.) 4, 105–192 (Academic Press, 1989).
2. Brainard, M. S. & Doupe, A. J. Translating birdsong: songbirds as a model for
basic and applied medical research. Annu. Rev. Neurosci. 36, 489–517 (2013).
3. Fee, M. S. & Scharff, C. The songbird as a model for the generation and
learning of complex sequential behaviors. ILAR J. 51, 362–377 (2010).
4. Elemans, C. P. H. The singer and the song: The neuromechanics of avian sound
production. Curr. Opin. Neurobiol. 28, 172–178 (2014).
5. Goller, F. & Larsen, O. N. A new mechanism of sound generation in songbirds.
Proc. Natl Acad. Sci. USA 94, 14787–14791 (1997).
6. Jensen, K. K., Cooper, B. G., Larsen, O. N. & Goller, F. Songbirds use pulse tone
register in two voices to generate low-frequency sound. Proc. Biol. Sci. 274,
2703–2710 (2007).
7. Elemans, C. P. H., Zaccarelli, R. & Herzel, H. Biomechanics and control of
vocalization in a non-songbird. J. R. Soc. Interface 5, 691–703 (2008).
8. Zaccarelli, R., Elemans, C. P. H., Fitch, W. T. & Herzel, H. Modelling bird
songs: voice onset, overtones and registers. Acta Acust. United Acust. 92,
741–748 (2006).
9. Mindlin, G. B. & Laje, R. The Physics of Birdsong (Springer, 2005).
10. Amador, A. & Margoliash, D. A mechanism for frequency modulation in
songbirds shared with humans. J. Neurosci. 33, 11136–11144 (2013).
11. Amador, A., Perl, Y. S., Mindlin, G. B. & Margoliash, D. Elemental gesture
dynamics are encoded by song premotor cortical neurons. Nature 495, 59–64
(2013).
12. van den Berg, J. Myoelastic-aerodynamic theory of voice production. J. Speech
Hear. Res. 1, 227–244 (1958).
13. Titze, I. R. Principles of Voice Production (National Center for Voice and
Speech, 2000).
14. Titze, I. R. Comments on the myoelastic—aerodynamic theory of phonation.
J. Speech Hear. Res. 23, 495–510 (1980).
15. Titze, I. R. & Alipour, F. The Myoelastic Aerodynamic Theory of Phonation
(National Center for Voice and Speech, 2006).
16. Riede, T. & Goller, F. Peripheral mechanisms for vocal production in birds—
differences and similarities to human speech and singing. Brain Lang. 115,
69–80 (2010).
17. Fitch, T. in Encyclopedia of language and linguistics 2nd ed., Vol. 10 (ed Brown,
K) 115–121 (Oxford, Elsevier 2006).
18. Herbst, C. T. et al. How low can you go? Physical production mechanism of
elephant infrasonic vocalizations. Science 337, 595–599 (2012).
19. Titze, I. R. & Strong, W. J. Normal modes in vocal cord tissues. J. Acoust. Soc.
Am. 57, 736–749 (1975).
20. Titze, I. R. On the relation between subglottal pressure and fundamental
frequency in phonation. J. Acoust. Soc. Am. 85, 901–906 (1989).
21. Svec, J. G., Horácek, J., Sram, F. & Veselý, J. Resonance properties of the vocal
folds: in vivo laryngoscopic investigation of the externally excited laryngeal
vibrations. J. Acoust. Soc. Am. 108, 1397–1407 (2000).
22. Berry, D. A. Mechanisms of modal and nonmodal phonation. J. Phon. 29,
431–450 (2001).
23. Flanagan, J. L. & Landgraf, L. L. Self-oscillating source for vocal-tract
synthesizers. IEEE Trans. Audio Electroacoust. AU16, 57 (1968).
24. Titze, I. R. The physics of small-amplitude oscillation of the vocal folds.
J. Acoust. Soc. Am. 83, 1536–1552 (1988).
25. Ishizaka, K. & Flanagan, J. L. Synthesis of voiced sounds from a two-mass
model of the vocal cords. Bell Syst. Tech. J. 51, 1233–1268 (1972).
26. Scherer, R. C., Torkaman, S., Kucinschi, B. R. & Afjeh, A. A. Intraglottal
pressures in a three-dimensional model with a non-rectangular glottal shape.
J. Acoust. Soc. Am. 128, 828–838 (2010).
27. Fulcher, L. P. et al. Pressure distributions in a static physical model of the
hemilarynx: measurements and computations. J. Voice 24, 2–20 (2010).
28. Miller, D. G. & Schutte, H. K. Characteristic patterns of sub-and supraglottal
pressure variations within the glottal cycle. in Transcr. XIIIth Symp. Care Prof.

NATURE COMMUNICATIONS | 6:8978 | DOI: 10.1038/ncomms9978 | www.nature.com/naturecommunications

ARTICLE

NATURE COMMUNICATIONS | DOI: 10.1038/ncomms9978

Voice (ed. Lawrence, V.) 70–75 (The Voice Foundation, New York, NY,
USA, 1985).
29. Schutte, H. K. & Miller, D. G. Resonanzspiele der Gesangsstimme in ihren
Beziehungen zu supra- und subglottalen Druckverläufen: Konsequenzen für die
Stimmbildungstheorie. Folia. Phoniatr. (Basel) 40, 65–73 (1988).
30. Herbst, C. T. et al. Complex vibratory patterns in an elephant larynx. J. Exp.
Biol. 216, 4054–4064 (2013).
31. Larsen, O. N. & Goller, F. Role of syringeal vibrations in bird vocalizations.
Proc. R. Soc. Biol. Sci. 266, 1609–1615 (1999).
32. Fee, M. S., Shraiman, B., Pesaran, B. & Mitra, P. P. The role of nonlinear dynamics
of the syrinx in the vocalizations of a songbird. Nature 395, 67–71 (1998).
33. Fee, M. S. Measurement of the linear and nonlinear mechanical properties of
the oscine syrinx: implications for function. J. Comp. Physiol. A Neuroethol.
Sens. Neural. Behav. Physiol. 188, 829–839 (2002).
34. Dürrwang, R. Funktionelle Biologie, Anatomie und Physiologie der
Vogelstimme PhD thesis, University of Basel (Aku-Fotodruck, Basel, 1974).
35. Paulsen, K. Das Prinzip der Stimmbildung in der Wirbeltierrehe und beim
Menschen (Akademische Verlagsgesellschaft, 1967).
36. Amador, A. & Mindlin, G. B. Beyond harmonic sounds in a simple model for
birdsong production. Chaos 18, 043123 (2008).
37. Goller, F. & Riede, T. Integrative physiology of fundamental frequency control
in birds. J. Physiol. Paris 107, 230–242 (2013).
38. Suthers, R. A. & Zollinger, S. A. in Neuroscience of birdsong. (eds Zeigler, H.P.
& Marler P) 78–98 (Cambridge University Press, Cambridge, UK, 2008).
39. Bernstein, N. The Co-ordination and Regulation of Movements (Pergamon
Press, 1967).
40. Herbst, C. T., Ternström, S. & Švec, J. G. Investigation of four distinct glottal
configurations in classical singing—a pilot study. J. Acoust. Soc. Am. 125,
EL104–EL109 (2009).
41. Riede, T. & Goller, F. Morphological basis for the evolution of acoustic diversity
in oscine songbirds. Proc. Biol. Sci. 281, 20132306 (2014).
42. Kusuyama, T., Fukuda, H., Shiotani, A., Nakagawa, H. & Kanzaki, J. Analysis of
vocal fold vibration by X-ray stroboscopy with multiple markers. Otolaryngol.
Head Neck Surg. 124, 317–322 (2001).
43. Döllinger, M., Tayama, N. & Berry, D. A. Empirical eigenfunctions and medial
surface dynamics of a human vocal fold. Methods Inf. Med. 44, 384–391 (2005).
44. Boessenecker, A., Berry, D. A., Lohscheller, J., Eysholdt, U. & Doellinger, M.
Mucosal wave properties of a human vocal fold. Acta Acust. United Acust. 93,
815–823 (2007).
45. Kobler, J. B., Chang, E. W., Zeitels, S. M. & Yun, S.-H. Dynamic imaging of
vocal fold oscillation with four-dimensional optical coherence tomography.
Laryngoscope 120, 1354–1362 (2010).
46. Döllinger, M., Berry, D. A. & Berke, G. S. Medial surface dynamics of an in vivo
canine vocal fold during phonation. J. Acoust. Soc. Am. 117, 3174–3183 (2005).
47. Titze, I. R., Jiang, J. J. & Hsiao, T. Y. Measurement of mucosal wave
propagation and vertical phase difference in vocal fold vibration. Ann. Otol.
Rhinol. Laryngol. 102, 58–63 (1993).
48. Maina, J. N. The Lung-Air Sac System of Birds (Springer Science & Business
Media, 2006).
49. Maina, J. N., Singh, P. & Moss, E. A. Inspiratory aerodynamic valving occurs in
the ostrich, Struthio camelus lung: a computational fluid dynamics study under
resting unsteady state inhalation. Respir. Physiol. Neurobiol. 169, 262–270 (2009).
50. Mackelprang, R. & Goller, F. Ventilation patterns of the songbird lung/air sac
system during different behaviors. J. Exp. Biol. 216, 3611–3619 (2013).
51. Brackenbury, J. H. Lung-air-sac anatomy and respiratory pressures in the bird.
J. Exp. Biol. 57, 543–550 (1972).
52. Gaunt, A. S., Stein, R. C. & Gaunt, S. L. Pressure and air flow during distress
calls of the starling, Sturnus vulgaris (Aves; Passeriformes). J. Exp. Zool. 183,
241–261 (1973).
53. Beckers, G. J. L. Mechanisms of frequency and amplitude modulation in ring
dove song. J. Exp. Biol. 206, 1833–1843 (2003).
54. Elemans, C. P. H., Spierts, I. L. Y., Müller, U. K., van Leeuwen, J. L. & Goller, F.
Bird song: superfast muscles control dove’s trill. Nature 431, 146 (2004).
55. Elemans, C. P. H., Mead, A. F., Rome, L. C. & Goller, F. Superfast vocal muscles
control song production in songbirds. PLoS ONE 3, e2581 (2008).
56. Düring, D. N. et al. The songbird syrinx morphome: a three-dimensional, highresolution, interactive morphological map of the zebra finch vocal organ. BMC
Biol. 11, 1 (2013).
57. Sober, S. J., Wohlgemuth, M. J. & Brainard, M. S. Central contributions to
acoustic variation in birdsong. J. Neurosci. 28, 10370–10379 (2008).
58. Scholz, J. P. & Schoner, G. The uncontrolled manifold concept: identifying
control variables for a functional task. Exp. Brain Res. 126, 289–306 (1999).

59. Todorov, E. & Jordan, M. I. Optimal feedback control as a theory of motor
coordination. Nat. Neurosci. 5, 1226–1235 (2002).
60. Latash, M. L. The bliss (not the problem) of motor abundance (not
redundancy). Exp. Brain Res. 217, 1–5 (2012).
61. Chhetri, D. K. et al. Effects of asymmetric superior laryngeal nerve
stimulation on glottic posture, acoustics, vibration. Laryngoscope 123,
3110–3116 (2013).
62. Chhetri, D. K., Neubauer, J. & Berry, D. A. Neuromuscular control of
fundamental frequency and glottal posture at phonation onset. J. Acoust. Soc.
Am. 131, 1401–1412 (2012).
63. Qiu, Q. & Schutte, H. K. Real-time kymographic imaging for visualizing human
vocal-fold vibratory function. Rev. Sci. Instrum. 78, 024302 (2007).
64. Švec, J. G. & Schutte, H. K. Kymographic imaging of laryngeal vibrations. Curr.
Opin. Otolaryngol. Head Neck Surg. 20, 458–465 (2012).
65. Baken, R. J. Electroglottography. J. Voice 6, 98–110 (1992).
66. Dooling, R. J., Lohr, B. & Dent, M. L. Comparative Hearing: Birds and Reptiles
13, 308–359 (Springer, 2000).
67. Goller, F. & Suthers, R. A. Role of syringeal muscles in controlling the
phonology of bird song. J. Neurophysiol. 76, 287–300 (1996).
68. Schuster, S., Zollinger, S. A., Lesku, J. A. & Brumm, H. On the evolution of
noise-dependent vocal plasticity in birds. Biol. Lett. 8, 913–916 (2012).
69. Elemans, C. P. H., Laje, R., Mindlin, G. B. & Goller, F. Smooth operator:
avoidance of subharmonic bifurcations through mechanical mechanisms
simplifies song motor control in adult zebra finches. J. Neurosci. 30,
13246–13253 (2010).
70. Secora, K. R. et al. Syringeal specialization of frequency control during song
production in the Bengalese finch (Lonchura striata domestica). PLoS ONE 7,
e34135 (2012).

Acknowledgements
We thank T. Christensen, P. Martensen, F. Mortensen, F. Andreasen, A. Chandler, V.
Shanbogue and N. Ditzel for technical assistance, J. Brewer for optics advice, L. Jakobsen
for use of equipment, S. Pedersen for ostrich access, A. Qiu for videokymography advice
and M. Rothenberg for EGG advice. H. Mehaffey, J.M. Ratcliffe and A. Surlykke commented on the manuscript. This study was funded by grants from the Danish Research
Council (FNU), Carlsberg Foundation, visiting professorship (Université de Saint-Etienne) and QuanTM fellowship (Emory University) to C.P.H.E., a Heisenberg Fellowship
of the German Research Foundation (Br 2309/8-1) to H.B., US National Institutes of
Health grants F31DC013753 to K.S., P30NS069250 and R01NS084844 to S.J.S., European
Social Fund and the state budget of the Czech Republic, project numbers OPVK CZ.1.07/
2.3.00/30.0004 ‘POST-UP’ to C.T.H. and J.G.Š., the institutional fund of Palacký University Olomouc to C.T.H. and OPVK CZ.1.07/2.3.00/20.0057 to J.G.Š.

Author contributions
C.P.H.E. conceived the study and built the set-up. All authors contributed to data
acquisition: pigeon (C.P.H.E., J.H.R., C.T.H., D.N.D., J.G.Š.), Barbary dove (C.P.H.E.),
ostrich (N.S., C.P.H.E.), tinamou (S.A.Z., H.B., C.P.H.E.), cockatiel (O.N.L., C.P.H.E.),
Bengalese finch (K.S., S.J.S., C.P.H.E.) and zebra finch (D.N.D., C.P.H.E.). CT scans were
performed by D.N.D. and M.D., data annotation by D.N.D. All authors contributed to
data analysis and the manuscript. All authors read and approved the final manuscript.

Additional information
Supplementary Information accompanies this paper at http://www.nature.com/
naturecommunications
Competing financial interests: The authors declare no competing financial interests.
Reprints and permission information is available online at http://npg.nature.com/
reprintsandpermissions/
How to cite this article: Elemans, C. P. H. et al. Universal mechanisms of
sound production and control in birds and mammals. Nat. Commun. 6:8978
doi: 10.1038/ncomms9978 (2015).
This work is licensed under a Creative Commons Attribution 4.0
International License. The images or other third party material in this
article are included in the article’s Creative Commons license, unless indicated otherwise
in the credit line; if the material is not included under the Creative Commons license,
users will need to obtain permission from the license holder to reproduce the material.
To view a copy of this license, visit http://creativecommons.org/licenses/by/4.0/

NATURE COMMUNICATIONS | 6:8978 | DOI: 10.1038/ncomms9978 | www.nature.com/naturecommunications

13