RESEARCH ARTICLE

Auditory motion-specific mechanisms in the

primate brain
Colline Poirier*, Simon Baumann, Pradeep Dheerendra, Olivier Joly, David Hunter,
Fabien Balezeau, Li Sun, Adrian Rees, Christopher I. Petkov, Alexander Thiele, Timothy
D. Griffiths*
Institute of Neuroscience, Newcastle University, Newcastle upon Tyne, Tyne and Wear, United Kingdom

* [Link]@[Link] (CP); [Link]@[Link] (TDG)

a1111111111
a1111111111
a1111111111 Abstract
a1111111111
a1111111111 This work examined the mechanisms underlying auditory motion processing in the auditory
cortex of awake monkeys using functional magnetic resonance imaging (fMRI). We tested
to what extent auditory motion analysis can be explained by the linear combination of static
spatial mechanisms, spectrotemporal processes, and their interaction. We found that the
OPEN ACCESS
posterior auditory cortex, including A1 and the surrounding caudal belt and parabelt, is
involved in auditory motion analysis. Static spatial and spectrotemporal processes were
Citation: Poirier C, Baumann S, Dheerendra P, Joly
O, Hunter D, Balezeau F, et al. (2017) Auditory able to fully explain motion-induced activation in most parts of the auditory cortex, including
motion-specific mechanisms in the primate brain. A1, but not in circumscribed regions of the posterior belt and parabelt cortex. We show that
PLoS Biol 15(5): e2001379. [Link] in these regions motion-specific processes contribute to the activation, providing the first
10.1371/[Link].2001379
demonstration that auditory motion is not simply deduced from changes in static spatial
Academic Editor: Anthony Zador, Cold Spring location. These results demonstrate that parallel mechanisms for motion and static spatial
Harbor Laboratory, United States of America
analysis coexist within the auditory dorsal stream.
Received: October 19, 2016

Accepted: April 7, 2017

Published: May 4, 2017

Copyright: © 2017 Poirier et al. This is an open

Author summary
access article distributed under the terms of the Motion is a fundamental dimension of acoustic and visual stimuli that is critical for ani-
Creative Commons Attribution License, which mals to interact with their environment. Yet, surprisingly, we still do not understand the
permits unrestricted use, distribution, and
basic mechanisms in the brain that underlie perception of auditory motion. For the last 30
reproduction in any medium, provided the original
author and source are credited. y, this research field has been hampered by unsuccessful attempts to answer a simple but
fundamental question: is auditory motion perception deduced from processing individual
Data Availability Statement: Data are available
static sounds, or are there mechanisms in the auditory domain dedicated to detecting
from the Open Science Framework at [Link]
ut5pa/. motion? Here we report the discovery of specific motion detectors located in the auditory
cortex of primates. We demonstrate that these auditory motion detectors are close to the
Funding: Wellcome Trust [Link]
well-known visual motion detectors. Both types of detectors are likely to be crucial for the
(grant number WT091681MA, WT085002MA, and
WT092606AIA).TDG, AR, LS, AT, CIP. The funder planning of limb and eye movement. This study addresses a fundamental issue in neuro-
had no role in study design, data collection and science and sheds new light on the brain mechanisms underlying the essential aspects of
analysis, decision to publish, or preparation of the our ability to navigate the world.
manuscript. Biotechnology and Biological Sciences
Research Council [Link] (grant
number BB/J009849/1).CIP. The funder had no
role in study design, data collection and analysis,

PLOS Biology | [Link] May 4, 2017 1 / 24

 Auditory motion detectors

decision to publish, or preparation of the Introduction

manuscript. Fonds Wetenschappelijk Onderzoek -
Vlandereen [Link] (grant number Motion is a fundamental dimension of acoustic and visual stimuli that is critical for animals
25208 and 23629).CP. The funder had no role in to interact with their environment. Human psychoacoustic studies have addressed whether
study design, data collection and analysis, decision auditory motion analysis depends on sequential perception of stationary sources or whether
to publish, or preparation of the manuscript. specific motion detection mechanisms exist, but the results so far have remained inconclusive
Competing interests: The authors have declared [1–9].
that no competing interests exist. Studies of neuronal activity in various mammalian species, including macaques, have
Abbreviations: BOLD, blood-oxygen-level shown that cues supporting auditory motion perception can induce pronounced asymmetry
dependent; CL, caudolateral area; CM, caudomedial of neuronal responses to opposite motion directions in the inferior colliculus and primary
area; df, degrees of freedom; fMRI, functional auditory cortex A1 [10,11]. However, it has been argued that this apparent direction sensitivity
magnetic resonance imaging; FWE, family-wise does not represent genuine motion selectivity but results from “adaptation of excitation,”
error; Hf, high frequency; Lf, low frequency; M1,
defined as the reduced capacity of a neuron to respond to excitatory stimuli following the pre-
monkey 1; M2, monkey 2; MAA, minimum audible
angle; MAMA, minimum audible movement angle; sentation of a prior excitatory stimulus, a mechanism also called “spatial masking” [12–14].
ML, middle lateral area; STG, superior temporal Motion processing per se has not been investigated beyond A1 in animal models. However,
gyrus. sensitivity to static spatial information has been shown to increase from A1 to the caudomedial
(CM) and caudolateral (CL) belt areas in macaques [15–19], opening the possibility that these
regions might be more sensitive to dynamic spatial information than A1.
Human functional magnetic resonance imaging (fMRI) studies indicate that the planum
temporale, the region of the auditory cortex that contains areas homologous to monkey areas
CM and CL, is involved in auditory motion processing [20,21]. However, these studies did not
directly address the question of the underlying mechanisms and did not allow for any conclu-
sion about the existence of specific motion-detection processes for sound movement analysis
as opposed to sequential processing of stationary sources. Similarly, lesion studies did not
allow for distinguishing between the two hypotheses [22–25].
In this study, we measured the fMRI blood-oxygen-level-dependent (BOLD) response to
auditory motion in the whole auditory cortex of awake macaques. We performed a series of
experiments designed to elucidate the mechanisms supporting auditory motion processing.
Data revealed that auditory motion perception relies on specific computational mechanisms
beyond the simple representation of successive snapshots of location.

Results
We employed virtual auditory motion stimuli smoothly moving in azimuth at a speed of 100˚/s
and different control stimuli (Fig 1A). The choice of the movement plane and the speed of the
motion stimuli were chosen to be ecologically relevant (consistent with everyday listening expe-
rience) and to take advantage of the fact that human and macaque listeners are more accurate
in perceiving angular sound position change in the azimuthal plane than in the vertical one
[17,26,27].

Experiments 1 and 2: Motion versus stationary stimuli

The two first experiments involved motion and stationary stimuli. Four moving and five sta-
tionary stimuli were presented in a random order to subjects involved in a visual fixation task,
and a sparse-sampling paradigm was used to measure the BOLD response induced by each
stimulus (Fig 1B). Auditory stimuli were based on individual intra-auricular recordings of an
amplitude-modulated broadband noise. Virtual moving stimuli were perceived as moving
back and forth between the positions 0˚ and 80˚ within one hemispace, with half of the stimuli
starting from the midline position (0˚), while the other half started from the most lateral posi-
tion (80˚), resulting in four motion stimulus exemplars per subject. The five stationary stimuli
corresponded to sounds perceived as coming from the locations −80˚, −40˚, 0, +40˚, and +80˚,

PLOS Biology | [Link] May 4, 2017 2 / 24

 Auditory motion detectors

Fig 1. Stimuli and experimental paradigm. (A) Representation of different exemplars of the auditory stimuli used in experiments 1 to 4.
The plots represent the waveform of one moving stimulus, one stationary control stimulus, and one spectrotemporal stimulus. The
pictograms represent the location of each stimulus relative to the subject’s head, as perceived during the experiments. L: left, R: right. (B)
Schematic representation of the main experimental paradigm. We used a sparse-sampling paradigm where 4.9 s-long auditory stimuli
were presented between each scan acquisition (every 10 s). The presentation order of the stimuli was pseudorandomized. Stimuli
illustrated: L40: stationary stimulus coming from the −40˚ position; Mov R1: moving stimulus moving within the right hemispace, starting
from the 0˚ position; R80: stationary stimulus coming from the +80˚ position; Mov L2: moving stimulus moving within the left hemispace,
starting from the −80˚ position; and Mov R2: moving stimulus moving within the right hemispace, starting from the +80˚ position.
[Link]

respectively (the minus sign referring to the left hemispace and the plus sign to the right one).
Because moving stimuli starting from the central position induced similar BOLD responses as
those starting to move from the periphery, data were pooled, resulting in two motion condi-
tions corresponding to sounds moving within the left hemispace (Motion Left) and those mov-
ing within the right hemispace (Motion Right).
The goal of this experiment was to identify the neural substrates of motion perception that
were not explained by the processing of stationary sounds. We thus compared the BOLD
response induced by moving sounds within each hemispace with the average BOLD response
induced by the stationary stimuli corresponding to the spatial positions through which the
moving sounds passed: the BOLD responses induced by stationary stimuli −80˚, −40˚, and 0˚
were thus averaged and compared to those induced by sounds moving in the left hemispace

PLOS Biology | [Link] May 4, 2017 3 / 24

 Auditory motion detectors

(contrast Motion Left minus Stationary Left), while the BOLD responses induced by stationary
stimuli +80˚, +40˚, and 0˚ were averaged and compared to those induced by sounds moving in
the right hemispace (contrast Motion Right minus Stationary Right). These comparisons
allowed us to control for the laterality of the motion stimuli (nondynamic spatial information,
corresponding to the encoding of space within which the motion stimulus was moving), as
well as for their intrinsic spectrotemporal content (amplitude-modulated broadband noise fil-
tered through the pinna).
In each subject (monkey 1, M1; monkey 2, M2), the contrasts Motion Left minus Stationary
Left and Motion Right minus Stationary Right revealed widespread activation of the posterior
part of the auditory cortex contralateral to the stimuli, on the superior temporal gyrus (STG)
(Fig 2A and 2B). The activation included the three stages of the hierarchically organized audi-
tory cortex, namely the core (in A1), the belt (in the middle lateral [ML] and CL areas, sur-
rounding A1), and the parabelt (the caudal part of the parabelt, lateral to CL and ML, on the
STG convexity), and extended into the inferior bank of the STG (also known as the superior
bank of the superior temporal sulcus). Activation in the ipsilateral hemisphere was much more
limited (M1) or absent altogether (M2). These results indicate that motion-induced activity in
the contralateral posterior auditory cortex cannot simply be explained by a spatial laterality
process or by encoding of the intrinsic spectrotemporal content of the moving stimuli since
these processes were controlled for by the stationary stimuli.
While in Experiment 1 the BOLD response induced by each stimulus was measured 5 s
after the stimulus onset (targeting the peak of the hemodynamic response function [28]), in a
second experiment we measured the time course of the hemodynamic response induced by
each stimulus, recording the BOLD response 2, 3, 4, and 5 s after the stimulus onset (Fig 2C).
This was to eliminate the possibility that the greater activation triggered by motion stimuli in
Experiment 1 was due to adaptation of the BOLD response to the stationary stimulus while the
motion-induced response was still sustained at its maximal level. This experiment revealed no
interaction between the two stimulus conditions and the four time points tested, except in a
small cluster in the right inferior bank of the STG, where stationary sounds did not induce any
significant activation. This confirms that the contrast between moving and stationary stimuli
revealed in Experiment 1 was not due to a different time course of the BOLD response to the
two types of stimuli.

Experiment 3: Motion versus spectrotemporal control

The change of spatial location inherent to moving stimuli induces some temporal variations
of the sound spectral envelope due to filtering of the sounds through the pinna. This spectro-
temporal effect of motion could theoretically be the source of auditory motion selectivity as
described so far. To control for this, we generated a new stimulus by averaging at each time
point the signal coming from each channel and presenting the stimulus diotically (i.e., the
same signal was sent to each ear). This stimulus had a spectrotemporal structure similar to the
motion stimulus, allowing us to control for the combined effect of the intrinsic spectrotem-
poral content of the stimulus and the spectrotemporal effect of motion, but did not contain
any static spatial cue (Fig 1A). In our third experiment, we compared the BOLD response
induced by motion with the one induced by the spectrotemporal control stimulus. The con-
trasts Motion Left minus Spectrotemporal control and Motion Right minus Spectrotemporal con-
trol revealed activation in ML, CL, the caudal parabelt, and the inferior bank of the STG in the
hemisphere contralateral to the stimuli (Fig 3) and no activation in the ipsilateral hemisphere.
The results thus indicate that motion-induced activation in these contralateral regions cannot
be explained by spectrotemporal processes.

PLOS Biology | [Link] May 4, 2017 4 / 24

 Auditory motion detectors

Fig 2. Motion minus Stationary contrast. (A) Location of the superior temporal gyrus (STG) surface and of the auditory areas in context. The dashed line
corresponds to a cut through the operculum to expose the superior bank of the STG (in black). The relative positions of the primary core area, A1, and the
secondary belts areas, middle lateral (ML) and caudolateral (CL), are indicated on the STG surface. The two straight lines represent the low-frequency (Lf)
and high-frequency (Hf) reversals of tonotopy gradients. L; lateral; C: caudal. See also Materials and Methods and S1 Fig. (B) Motion minus Stationary
contrasts (Experiment 1). Blood-oxygen-level-dependent (BOLD) responses were measured 5 s after the stimulus onset. Significant statistical parametric t
maps are shown on a surface rendering of the STG of each individual (number of scans [n] = 2,466 [monkey 1, M1] and 3,429 [monkey 2, M2]; degrees of
freedom [df] = 2,210 [M1] and 3,071 [M2]; p-values range after correction with the family-wise error (FWE) method [PFWE] from 0.05 to below 1x 10−10 [M1
and M2]). Since activation in the hemisphere ipsilateral to the stimuli was weak or absent, only contralateral activation is shown: Motion Left minus Stationary

PLOS Biology | [Link] May 4, 2017 5 / 24

 Auditory motion detectors

Left is projected on the right STG, while Motion Right minus Stationary Right is projected on the left STG. Three different views of each STG are represented:
a top, a lateral, and a bottom view allowing the visualization of the superior bank (Sup.), the convexity (Conv.), and the inferior bank (Inf.) of the STG,
respectively. The transparent region within the black boundary corresponds to motion-responsive regions (defined by the contrasts Motion Left minus
Silence and Motion Right minus Silence, PFWE < 0.05) where the BOLD responses induced by motion and stationary stimuli were not significantly different.
The numbers 1, 2, 3 and 4 indicate the locations where the plots illustrated in panel C are extracted from. (C) Hemodynamic response functions for motion
and stationary stimuli in M1 (Experiment 2). The BOLD response (expressed by the response estimate coefficient, betas) was measured 2, 3, 4, and 5 s after
the stimulus onset. The plots illustrate data from four representative individual voxels whose locations are indicated in panel B. ANOVA tests: n = 5,794,
df = 4,905, F-values below 2.1, and all PFWE = 1. Error bars correspond to standard errors. Data are available from the Open Science Framework ([Link]
io/ut5pa/).
[Link]

Experiment 4: Motion = spectrotemporal + static spatial processes?

We then tested the hypothesis that a linear combination of spectrotemporal and stationary
spatial processes could fully explain motion-related activity. To do so, we gathered in a single
experiment the three different types of stimuli: motion, stationary, and spectrotemporal con-
trol stimuli. As a control, we first computed the same contrasts as in Experiments 1 and 3: the
contrasts Motion minus Stationary and Motion minus Spectrotemporal controls revealed pat-
terns of activation similar to those observed in Experiments 1 and 3 (Fig 4). Brain regions com-
monly activated by both contrasts were ML, CL, the caudal parabelt, and parts of the inferior
bank of the STG (green region in Fig 4). This result indicates that in these regions, motion-
related activation cannot be explained by static spatial processing (controlled in the contrast
Motion minus Stationary), spectrotemporal effect of motion (controlled in the contrast Motion
minus Spectrotemporal control), or the intrinsic spectrotemporal content of the stimulus (con-
trolled in both contrasts) alone.
We then tested the hypothesis that the addition of these processes could explain motion-
induced activation. More specifically, we tested the following model: Motion = Static central
sound + Spatial laterality + Spectrotemporal effect of motion, where spatial laterality was
defined with the contrast Stationary sounds minus Stationary central sound and the spectro-
temporal effect of motion with the contrast Spectrotemporal controls minus Stationary central
sound (S2 Fig). Motion-induced activation was not found to be significantly different from
the sum of the three components in most parts of brain regions activated by motion stimuli,
including A1 (Fig 5, transparent regions within the black boundary), indicating that the simple
additive model provides a good estimation of the activation induced by motion stimuli in
these regions. However, in parts of ML, CL, the caudal parabelt, and the inferior bank of the
STG, the BOLD response induced by motion stimuli was significantly greater than the sum of
the BOLD responses induced by stationary central sound, spatial laterality, and spectrotem-
poral effect of motion (Fig 5, green cluster).
The excess signal (i.e., the part not explained by the above components) that was present
in these areas could arise if the three processes made different contributions (i.e., different
weighting coefficients attached to each process), if the processes interacted, or if an additional,
independent (and thus genuine motion-specific) process would trigger the response. To distin-
guish between these explanations, we performed a multiple linear regression analysis across
those voxels where the BOLD response to auditory motion was not fully explained by the sim-
ple additive model (green cluster in Fig 5). This analysis revealed that the motion-induced
BOLD signal was best estimated by a model including differential weighting factors (with a
slightly smaller contribution of spectrotemporal effect of motion, compared to the two other
components in both subjects; see Table 1), weak or no interactions between the components
(R2 change between models with and without interactions: M1: 0.004, F = 0.13, p = 0.969; M2:
0.019, F = 4.09, p = 0.004; R2 for best model: M1: 0.82; M2: 0.86; p-values below 1 x 10−6 for
each subject), and, importantly, a term that assumes the presence of genuine motion-selective

PLOS Biology | [Link] May 4, 2017 6 / 24

 Auditory motion detectors

Fig 3. Motion minus Spectrotemporal control contrast (Experiment 3). Significant statistical parametric t maps are shown on a surface rendering of
the superior temporal gyrus (STG) contralateral to the stimuli (no activation in the ipsilateral hemisphere): Motion Left minus Spectrotemporal control is
projected on the right STG, while Motion Right minus Spectrotemporal control is projected on the left STG (n = 3,028 [monkey 1, M1] and 2,664 [monkey
2, M2]; degrees of freedom [df] = 2,722 [M1] and 2,382 [M2]; PFWE: from 0.05 to 4 x 10−9 [M1] and from 0.05 to below 1 x 10−10 [M2]). The transparent
region within the black boundary corresponds to motion-responsive regions (defined by the contrasts Motion Left minus Silence and Motion Right minus
Silence, PFWE < 0.05) where the blood-oxygen-level-dependent (BOLD) responses induced by motion and stationary stimuli were not significantly
different. Conv., convexity; Hf, high frequency; Inf., inferior bank; Lf, low frequency; Sup., superior bank. For more details, see Fig 2 legend. Data are
available from the Open Science Framework ([Link]
[Link]

PLOS Biology | [Link] May 4, 2017 7 / 24

 Auditory motion detectors

Fig 4. Motion minus Stationary contrasts (in cyan) and Motion minus Spectrotemporal control contrasts (in yellow) (Experiment 4). Voxels where
each contrast is statistically significant (PFWE < 0.05) are respectively colored in cyan and yellow (t-tests: n = 9,927 [monkey 1, M1] and 6,664 [monkey 2,
M2]; degrees of freedom [df] = 8,902 [M1] and 5,922 [M2]). The green area corresponds to voxels where both contrasts are significant. Statistical results are
shown on a surface rendering of the superior temporal gyrus (STG) contralateral to the stimuli: Motion Left minus Stationary Left/Spectrotemporal control is
projected on the right STG, while Motion Right minus Stationary Right/Spectrotemporal control is projected on the left STG. The transparent region within
the black boundary corresponds to motion-responsive regions (defined by the contrasts Motion Left minus Silence and Motion Right minus Silence, PFWE <
0.05) where the blood-oxygen-level-dependent (BOLD) responses induced by motion and stationary stimuli were not significantly different. Conv.,
convexity; Hf, high frequency; Inf., inferior bank; Lf, low frequency; Sup., superior bank. For more details, see Fig 2 legend. Data are available from the
Open Science Framework ([Link]
[Link]

responses to occur in these areas (the added constant term in the linear regression, Table 1,
p-values below 1 x 10−5 for each subject). This additional component accounted for about 42%
of the mean signal intensity induced by motion stimuli (M1: 42.5%; M2: 42.2%). A control

PLOS Biology | [Link] May 4, 2017 8 / 24

 Auditory motion detectors

Fig 5. Motion minus (Stationary central sound + Spatial laterality + Spectrotemporal effect of motion) contrast (Experiment 4). Voxels where the
contrast is statistically significant (PFWE < 0.05) are colored in green (t-tests: n = 9,927 [monkey 1, M1] and 6,664 [monkey 2, M2]; degrees of freedom [df] =
8,902 [M1] and 5,922 [M2]). Statistical results are shown on a surface rendering of the superior temporal gyrus (STG) contralateral to the stimuli. The
transparent region within the black boundary corresponds to motion-responsive regions (defined by the contrasts Motion Left minus Silence and Motion
Right minus Silence, PFWE < 0.05) where the blood-oxygen-level-dependent (BOLD) response induced my motion was not significantly different than the
additive model. Conv., convexity; Hf, high frequency; Inf., inferior bank; Lf, low frequency; Sup., superior bank. For more details, see Fig 2 legend. Data are
available from the Open Science Framework ([Link]
[Link]

analysis across voxels taken within the region where the full signal could be explained by the
simple additive model (transparent patch in Fig 5) revealed similar regression coefficients but
no significant constant term (Table 1). S3 Fig illustrates the goodness of fit of the best-adjusted
model with experimental data in each subject.

PLOS Biology | [Link] May 4, 2017 9 / 24

 Auditory motion detectors

Table 1. Multiple linear regression coefficients and associated statistics.

Coefficient SE t/F p
M1
Inside green cluster
Motion constant 9.0 1.51 6.0 3 x 10−6
Stationary 1.2 0.14 8.4 <1 x 10−6
Laterality 1.3 0.41 3.1 5 x 10−3
ST 0.7 0.29 2.4 2 x 10−2
Outside green cluster
Motion constant -0.2 0.83 −0.2 0.84
Stationary 1.1 0.07 15.3 <1 x 10−6
Laterality 1.0 0.24 4.1 4 x10-4
ST 0.6 0.07 7.6 <1x 10−6
M2
Inside green cluster
Motion constant 7.5 0.58 12.9 <1 x 10−6
Stationary 1.4 0.07 19.2 <1 x 10−6
Laterality 1.3 0.10 13.1 <1 x 10−6
ST 0.8 0.07 11.5 <1 x 10−6
Stationary x Laterality 0.03 0.02 1.6 0.12
Stationary x ST -0.05 0.01 −3.4 9 x 10−4
Laterality x ST -0.02 0.02 −1.4 0.18
Stationary x Laterality x ST 0.001 0.004 0.4 0.72
Outside green cluster
Motion constant 0.0 0.51 0.0 0.98
Stationary 1.2 0.07 19.1 <1 x 10−6
Laterality 1.0 0.10 9.9 <1 x 10−6
ST 0.6 0.04 12.8 <1 x 10−6

“Inside green cluster” refers to the brain region depicted in green in Fig 5, where the contrasts Motion minus Silence and Motion minus (Stationary central
sound + Spatial laterality + Spectrotemporal effect of motion) were significant. “Outside green cluster” refers to transparent voxels within the black boundary
in Fig 5 and corresponds to voxels where the contrast Motion minus Silence was significant but where the contrast Motion minus (Stationary central sound
+ Spatial laterality + Spectrotemporal effect of motion) was not significant. Interaction coefficients and associated statistics are displayed only in the case
where adding interaction terms significantly improved the model (statistically significant R2 change). Monkey 1, M1; monkey 2, M2; Motion constant,
constant term of the linear regression; SE, standard error; ST, spectrotemporal effect of motion; stationary, stationary central sound process. See also S3
Fig.

[Link]

Experiment 5: Sensitivity of auditory motion areas to visual motion

To further characterize these auditory motion-selective areas, we tested whether they were also
selective for visual motion. Visual-motion areas were identified by contrasting the BOLD sig-
nal induced by slowly moving horizontally oriented gratings with that induced by stationary
gratings. The auditory and visual motion regions were found to partially overlap on the infe-
rior bank of the STG (Fig 6).

Discussion
Existence of motion-specific mechanisms
Movement selectivity has been classically investigated by comparing moving and stationary
stimuli. In humans, the contrast Motion minus Stationary has consistently revealed activation

PLOS Biology | [Link] May 4, 2017 10 / 24

 Auditory motion detectors

Fig 6. Auditory (in green) and visual (in red) motion-specific areas in monkey 2 (M2). Visual motion-specific areas were identified by the contrast
Moving gratings minus Stationary gratings (Experiment 5, t-test: n = 1,241; degrees of freedom [df] = 1,067). Auditory motion data came from Experiment 4,
and auditory motion-specific areas were identified by the contrast (Motion minus (Stationary central sound + Spatial laterality + Spectrotemporal effect of
motion)) illustrated in Fig 5. The black area corresponds to voxels where both contrasts are significant. Significant statistical results (PFWE < 0.05) are shown
on a surface rendering of the superior temporal gyrus (STG) contralateral to the stimuli. Conv., convexity; Hf, high frequency; Inf., inferior bank; Lf, low
frequency; Sup., superior bank. See Fig 2 legend for more details. Data are available from the Open Science Framework ([Link]
[Link]

of the planum temporale but not Heschl’s gyrus [20,21,29–32]. While human A1 has tradition-
ally been considered to be located on the Heschl’s gyrus, recent tonotopy data suggest that A1
is rather found on the posterior half of the Heschl’s gyrus and slightly extends posteriorly into
a small part of the planum temporale [33,34]. The planum temporale could thus encompass
human homologues of macaque auditory areas CM, CL, and ML but also of the posterolateral
part of A1. According to this model, our results fit well with human data since the contrast
Motion minus Stationary induced the recruitment of the caudolateral parts of A1 and extended
into CL and ML (Experiments 1 and 4).
The contrast Motion minus Stationary controls for the intrinsic spectrotemporal content of
the motion stimuli and, when the stationary control sounds cover the whole spatial range
spanned by the motion stimulus, its nondynamic spatial component as well. However, it does
not control for the spectrotemporal effects of motion: filtering of sound through the pinna dif-
fers for each spatial position, inducing slow modulations of the sound spectral envelope. This
dynamic nonspatial component aspect has only been controlled for in one previous study [21].
In this human fMRI study, spectrotemporal effects of motion were found to be processed in
parts of the planum temporale overlapping with those involved in motion processing (revealed
by the contrast Motion minus Stationary). Our study provides similar results in macaques
(Fig 4).
Compared to the previous human study [21], our study in macaques went several steps
further. First, we tested for the first time the possibility that the addition of the different
components of auditory stimuli could explain the activation that they induced. Our results
demonstrate that the additive combination of (1) the intrinsic spectrotemporal component
(“Stationary central sound” processing), (2) the nondynamic spatial component (spatial

PLOS Biology | [Link] May 4, 2017 11 / 24

 Auditory motion detectors

laterality processing), and (3) the dynamic nonspatial component (spectrotemporal effect of
motion) allows a full characterization of motion processing in large parts of the STG, including
A1. However, in a circumscribed region overlapping parts of ML, CL, the caudal parabelt, and
the inferior bank of the STG, the additive model did not explain a significant fraction of the
signal induced by motion stimuli.
Second, we controlled for potential interactions between components. This was done by
introducing in the linear models all possible interactions and testing whether the percentage of
the variance explained by the model including the interactions was significantly higher than in
the model without the interaction terms. This analysis revealed weak or no interactions
between the components.
Third, we controlled for mechanisms that can influence the gain of the three different com-
ponents (i.e., different forms of adaptation or amplification). It has been argued that greater
activation induced by moving sounds compared to stationary sounds could represent adapta-
tion of responses induced by stationary sounds [35–38]. By measuring the time course of the
hemodynamic response induced by motion and stationary sounds, here we demonstrate that
the differential activation cannot be explained by (slow) adaptation of the BOLD response trig-
gered by stationary sounds (Experiment 2). Rapid adaptation of neuronal responses could
also, in principle, explain greater activation induced by moving sounds. In the present study,
the use of an amplitude-modulated noise as a stimulus should have limited such an effect.
Moreover, adaptation or amplification mechanisms were controlled in the final linear regres-
sion analysis by the coefficients weighting the different components and their interactions.
The regression analysis revealed that the magnitude of these mechanisms was moderate (coef-
ficients close to 1) and similar between the region where part of the motion processing signal
was left unexplained and the region where the signal could be fully explained by the simple
additive model. Thus, adaptation or amplification fails to explain the unaccounted signal.
The linear regression analysis revealed that after controlling for the different processes not
specific to motion, their potential interactions, and their potential adaptation or amplification,
on average 42% of the signal variance remained unexplained. Since processes directly or indi-
rectly linked to processing of several stationary sounds were controlled for in this analysis, we
conclude that this remaining part of the signal comes from a motion-specific process.
The nature of the mechanisms underlying auditory motion perception has been debated for
more than 30 y. The “snapshot hypothesis” postulates that motion is inferred from snapshots
of object successive positions, without direct appreciation of motion. According to this
hypothesis, auditory motion perception is based on the same mechanisms as those involved in
the localization of static sound sources. The alternative hypothesis, usually referred as the
“motion detector hypothesis” or “velocity detector hypothesis”, considers that motion percep-
tion is based on specific mechanisms. On the one hand, the fact that in humans, the minimum
audible movement angle (MAMA; defined as the smallest movement angle allowing a subject
to determine whether a sound is moving or not) differs from the minimum audible angle
(MAA; defined as the smallest location difference between two static sources that subjects
could discriminate) and was found to increase with speed has been interpreted as suggesting
that motion detectors exist [3,4]. It has also been argued that if moving sounds are processed
via a snapshot process, comparing the location of the starting and ending points should be suf-
ficient to perceive movement, and information about intermediary locations should be redun-
dant. However, the MAMA for moving sounds was found to be smaller than the MAA for
tone bursts marking the starting and ending positions of the moving sound [5,39], and subjects
could discriminate between accelerating and decelerating 90 ms-long stimuli starting and end-
ing at identical spatial locations [6], indicating that the human brain extracts other informa-
tion than the location of the starting and end points of a moving sound. On the other hand,

PLOS Biology | [Link] May 4, 2017 12 / 24

 Auditory motion detectors

movement detection and discrimination performances of human subjects have been explained
by estimation of the distance traversed by the source rather than appreciation of the motion
per se [1]. The MAMA and the MAA were also found to show similar dependency on sound
frequency, spectral bandwidth, and source azimuth, suggesting that static spatial cue percep-
tion and dynamic spatial cue perception are dependent on the same underlying mechanisms
[2,40]. Altogether, these data have failed to provide clear evidence about the mechanisms
underlying motion perception. Our results indicate that the parts of the auditory cortex,
including A1, analyze auditory sources in movement by processing their spatial location and
the consequence of location change (spectrotemporal effect of motion), consistent with a
“snapshot” strategy. However, we demonstrate that caudal belt and parabelt regions of audi-
tory cortex extract the motion component of moving stimuli (motion-specific process) in
addition to the non-motion specific components. The coexistence of the two mechanisms
might explain why it has been so difficult to distinguish between the snapshot and the motion
detectors hypotheses in the past. It is also relevant to psychophysical data illustrating facilita-
tion of motion perception by static spatial information [7,8].
The BOLD response measured by fMRI only provides an indirect measure of neuronal
activity. The choice of macaques as subjects of this study paves the way for a detailed investiga-
tion of the motion-specific mechanism at the cellular level. Our study indicates that motion-
specific and snapshot processes coexist in the caudal belt and parabelt regions. Electrophysio-
logical studies will be useful to determine whether the two types of processes are encoded by
different populations of neurons or not.
This study investigated the mechanisms underlying auditory motion along the azimuth axis
restricted to each hemispace. It is possible that the relative contribution of motion-specific and
snapshot mechanisms depends on the nature of the movement (in elevation, in depth, or
across hemispaces). Primates are particularly accurate at discriminating the spatial location of
sounds coming from the regions near the midline [41], and based on the main opponent-
channel hypothesis [42], the firing rate of auditory neurons contains more information about
the precise location of a sound source when it is near the midline, by opposition to the periph-
eral space. It is thus possible that the contribution of motion-specific mechanism for sounds
moving across the midline is less important than when sounds move within one hemispace.
Looming sounds are particularly relevant from a behavioral point of view, often indicating a
threat. In such sounds, information about the successive static position of the sound is limited
to monaural cues and the distance or time to arrival is systematically underestimated [43].
Thus, one might expect the relative contribution of snapshot processes to be less important.
These will be interesting hypotheses to test in the future.

A further auditory stream?

Auditory information has been proposed to be processed along two main streams: a ventral
stream connecting the rostral belt and parabelt areas to the ventral prefrontal cortex, involved
in the identification of sounds, and a dorsal stream connecting the caudal belt and parabelt to
the posterior parietal cortex and the dorsolateral prefrontal cortex, involved in spatial process-
ing [18,44]. The exact number and the respective role of each stream are still a matter of
debate. For instance, several authors have suggested that the auditory dorsal stream could be
divided into distinct substreams [15,45–47]. Anatomical tract-tracing studies suggest at least
two substreams originating from the caudal belt areas: a “dorsodorsal substream” involving
CM, projecting to Tpt posterior to CM and thence to the parietal and prefrontal cortex; and a
“dorsocaudal substream” connecting ML and CL to the caudal parabelt and the inferior bank
of the STG, which itself projects to the parietal cortex [48].

PLOS Biology | [Link] May 4, 2017 13 / 24

 Auditory motion detectors

Evidence for a role of the dorsal auditory pathway in spatial processing in nonhuman pri-
mates comes from electrophysiological data indicating that sensitivity to static spatial informa-
tion is higher in CL compared to A1 and the rostral belt areas [15,18,19]. Our study revealed a
similar refinement of static spatial processing between A1 and CL (see S4 Fig). Some single-
unit electrophysiological studies also highlighted the role of CM in spatial processing [17,19].
However, spatial selectivity seems to be weaker compared to CL [19], and modelling of neural
data suggests that the firing rates of CL neurons, but not of CM neurons, carry enough infor-
mation to account for sound localization performance in azimuth [16]. Using fMRI, we did
find strong and consistent BOLD activation induced by static and moving sounds in CL but
not in CM. Since the BOLD response indirectly reflects the activity of very large populations
of neurons, it is possible that at this scale, the spatial sensitivity of CM neurons cannot be
detected. Together, these results suggest that spatial sensitivity differs to some extent between
CL and CM and that the specialization for static auditory spatial processing mainly occurs
along the dorsocaudal substream.
In addition to this static spatial processing, our data demonstrate a particular specialization
for motion analysis within the dorsal-caudal substream, indicating that this stream carries out
higher-level spatial computation rather than just representing fixed space. This result indicates
the existence of parallel pathways for fixed and dynamic auditory spatial analysis within the
dorsocaudal stream that likely feed into distinct downstream mechanisms as in the visual sys-
tem. The exact number of substreams within the visual dorsal pathway and their respective
roles are still debated [49–51]. Subdivision of the human visual dorsal pathway into at least
two substreams has been proposed, with the dorsodorsal pathway involving the superior parie-
tal lobule, while the ventral-dorsal pathway involves the visual motion areas in the temporal sul-
cus (including MT) and the inferior parietal region [52,53]. Auditory motion-specific areas
described in the present study have been found to extend on the inferior bank of the STG
(equivalent of the superior bank of the temporal sulcus), in the vicinity of visual motion areas,
and our visual experiment revealed a small overlap between auditory and visual motion-selective
areas in this region. These data suggest that the auditory and visual motion substreams share
some neural substrates in the inferior bank of the STG, which might potentially support the
numerous behavioral interactions that have been reported between auditory and visual motion
perception [54,55]. It will be interesting to determine whether this region represents the point
where the auditory and visual motion substreams merge by testing whether the same population
of neurons respond to auditory and visual motion stimuli and where these neurons project.
Because the posterior auditory cortex in humans is involved not only in spatial analysis but
also in speech and music perception, recent models of the dorsal auditory stream incorporate
the idea that there may be a transformation of auditory information into a motor signal coding
for the action necessary to produce the sound [47,56,57]. While this auditory-motor function
could coexist with perceptual spatial processing in distinct substreams [47], it has also been
argued that spatial processing could be interpreted as a preparation for eye movement or grasp-
ing [58,59]. In this case, the whole dorsal pathway could be characterized as a mechanism for
auditory-motor integration, with different substreams supporting different auditory-motor pro-
cesses. We suggest that the ability to compute motion allows a substream of the auditory dorsal
pathway to predict the trajectory of sources in a way that helps visual tracking and grasping.

Materials and methods

Ethics statement
All procedures were approved by the Animal Welfare and Ethical Review Body at Newcastle
University and by the United Kingdom Home Office (PPL 60/4095, 60/4037 and 70/7976).

PLOS Biology | [Link] May 4, 2017 14 / 24

 Auditory motion detectors

Experiments complied with the Animal Scientific Procedures Act (1986), the European Direc-
tive on the protection of animals used for scientific purposes (2010/63/EU), and the United
States National Institutes of Health Guidelines for the Care and Use of Animals for Experi-
mental Procedures and were performed with great care to ensure the well-being of the
animals.

Subjects
Two awake male rhesus monkeys (Macaca mulatta) M1 and M2, respectively 7 and 11 y old
(weighing 7 and 17 kg), participated in the experiments. The monkeys were initially implanted
with a head holder. All surgical procedures were performed under general anesthesia and ster-
ile conditions. Details regarding surgical procedures, postoperative care, and the cleaning of
the implant are published elsewhere [60]. The animals were first habituated to the scanner
environment over the course of several days and then enrolled in the experiments.

Auditory motion and control stimuli

Sound stimuli were created in MATLAB 7.1 (MathWorks, Natick, Massachusetts, US) with a
sample rate of 44.1 kHz and 16-bit resolution. All stimuli for the auditory motion experiments
(Experiments 1 to 4) were based on a random-phase noise carrier (1–20 kHz). The noise was
amplitude modulated by a sinusoidal envelope of 80% depth at 80 Hz in order to produce an
additional localization cue [61] and to prevent adaptation. Prior to the scanning experiments,
the amplitude-modulated noise was delivered in free field in an echo-suppressed room from
17 different positions separated by 10˚, along the azimuthal axis (from −80˚ to +80˚) and
recorded with a omnidirectional miniature electret microphone (Knowles Corporation, Itasca,
Illinois, US) placed within each ear canal of the subject, resulting in the recording of the sound
convolved by the head-related transfer function of each monkey. The microphone output was
amplified and recorded digitally at a sampling rate of 44.1 kHz. In addition to interaural level
and time differences, this whole procedure preserves spectral cues specific to each individual
and has been shown in humans to induce stimuli to be perceived as localized in the external
space when delivered through headphones [62]. While providing stimuli tailored to each sub-
ject, this approach is time consuming, and when applied to macaques, it is limited by the fact
that the subject’s pinnae are mobile such that movements can distort spectral cues. By gently
holding the subject’s pinnae from the back and keeping the stimulus recording sessions short
(~1 h), we could prevent any movement. To accommodate this time limitation, we did not
attempt to record sounds separated by less than the MAA (around 3˚ in macaques; see [17,26,
63]). Instead, we took advantage of the fact that a motion percept can be induced by sequen-
tially presenting sounds coming from spatial positions separated by more than the MAA as
long as these spatial positions are not too far apart and as long as the duration of each sound is
short enough (in other words, as long as the apparent speed is high enough). Motion stimuli
moving three times back and forth between the central space (0˚ position) and the most lateral
position (+ or −80˚ position) within each hemispace were created by concatenating 100 ms-
long segments of recordings from adjacent locations (Fig 1); half of the stimuli started from
the central position and the other half started from the most lateral position. Any abrupt
change of power between segments was avoided by concatenating on-phase segments, starting
and finishing when the power of the amplitude-modulated signal was minimum. Since these
spatial locations were separated by 10˚, it resulted in stimuli virtually moving at a speed of
100˚/s. A similar approach was used to create stationary stimuli: 100 ms-long recorded seg-
ments coming from the same location were concatenated to form long examplars of stationary
sounds. Five different stationary stimuli were created, corresponding to spatial positions −80˚,

PLOS Biology | [Link] May 4, 2017 15 / 24

 Auditory motion detectors

−40˚, 0˚, +40˚, and +80˚ (Fig 1). Finally, spectrotemporal controls of each motion stimulus
were created by averaging at each time point the signal coming from each channel and pre-
senting the stimulus diotically (Fig 1). The resulting stimuli were sounds with intensity and
spectral content varying with time like the motion stimuli but with no spatial information (sti-
muli sounding as coming from inside the head, without spatial laterality). All auditory motion
and control stimuli were 4.9 s long.

Validation of stimulus motion

To validate the percept induced by the virtual motion stimuli in our monkey subjects, we char-
acterized our stimuli psychophysically in humans. This approach takes advantage of the simi-
larity of spatial perception between humans and monkeys and of the fact that the MAA is
actually smaller in humans [17,26,27,63], allowing us to establish a more exacting test in which
differences between concatenated and moving stimuli are more likely to be detected. Briefly,
we built an apparatus capable of delivering static or moving sound stimuli in free field in our
soundproof chamber. This used an electric motor with adjustable speed (controlled by a
potentiometer) with an attached rotor arm to which a small speaker was attached, to achieve
sound-source rotatory movement in the azimuthal plane through the subject’s ear canal. We
have replicated in three human participants (two males and one female with no hearing disor-
der, age range: 20–35 y, having given their informed consent) the intra-auricular recording
approach used in macaques, using exactly the same sounds and the same recording equipment.
For each human participant, we recorded from the ear canal when static sounds were delivered
from azimuthal positions recorded in 10˚ intervals from zero (midline, front). Additionally,
we recorded motion stimuli from the ear canal when a speaker moved around the head with
an angular motion of 100˚/s or 50˚/s, clockwise and anticlockwise. The two types of recordings
were based on the same amplitude-modulated noise stimulus used in the macaque work. The
recording session lasted between 2 and 3 h, requiring the human participant to remain still
during this period. Static recordings from adjacent positions were then concatenated to create
stimuli virtually moving at speeds of 100˚/s or 50˚/s, the duration of each recorded segment
being 100 and 200 ms, respectively (concatenated stimuli). This procedure was used to create
four concatenated stimuli per participant, while recordings of real moving sounds were used
to create moving stimuli matching the travelled path and the direction of the concatenated sti-
muli: two stimuli of each type moved clockwise (one moving from +90˚ to +180˚ and the
other one moving from −90˚ to 0˚), and two stimuli moved anticlockwise (one moving from
−90˚ to +180˚ and the other one moving from 90˚ to 0˚). We tested each participant’s percep-
tion of these stimuli using criterion-free psychophysics. We used an AXB psychophysical para-
digm, where X was always a moving stimulus and A and B were either a moving stimulus or a
concatenated stimulus (whether A or B was the moving stimulus was randomized across trials)
and each stimulus moved at 100˚/s along the same path. Participants were asked to identify
which of stimuli A or B was different from X. The results confirmed that no participant was
able to distinguish concatenated stimuli from motion stimuli at 100˚/s (performance was at
chance level in each participant: Chi-square tests, n = 240, degrees of freedom [df] = 1, X2/
p = 0.6/0.44 [participant 1], 0.42/0.52 [participant 2], 0.6/0.44 [participant 3]; S5 Fig). Because
spatial acuity in azimuth is better in humans than macaques [17,26,27,63], this result supports
our claim that concatenated stimuli used in the scanner were perceived by macaques as
smoothly moving. In a control experiment, we replicated the AXB psychophysical paradigm
using the concatenated and motion stimuli moving at a speed of 50˚/s. This second experiment
confirmed that the percept is speed dependent, as two participants were then able to discrimi-
nate concatenated stimuli from motion stimuli (Chi-square tests, n = 240, df = 1, X2/p = 138/

PLOS Biology | [Link] May 4, 2017 16 / 24

 Auditory motion detectors

<1 x 10−19 [participant 1], 86.4/<1 x 10−19 [participant 3]), while the third participant was still
at chance level (Chi-square test, n = 240, df = 1, X2/p = 0.42/0.52, S5 Fig).

Other stimuli
Stimuli for the tonotopy experiment were based on a random-phase noise carrier with three
different passbands, 0.5–1 kHz, 2–4 kHz, and 8–16 kHz, resulting in three different stimuli
that encompassed different spectral ranges. The carriers were amplitude modulated with a
sinusoidal envelope of 90% depth at 10 Hz to achieve robust responses.
Stimuli for the visual motion localizer experiment (Experiment 5) were horizontally ori-
ented gratings (spatial frequency: 0.5 cycles/˚) of 6˚ diameter, displayed at 7˚ to the right or to
the left of the vertical meridian. Half of the stimuli were moving at a frequency of 8 Hz, while
the remaining stimuli were stationary.

Auditory stimuli delivery

Auditory stimuli were delivered in the scanner at an RMS sound pressure level of 74 dB using
custom adapted electrostatic headphones based on a Nordic NeuroLab system (Nordic Neuro-
Lab, Bergen, Norway). These headphones feature a flat frequency transfer function up to 16
kHz and are free from harmonic distortion at the applied sound pressure level. We recorded
the spontaneous eye movements of one monkey when exposed to the stationary lateralized sti-
muli (−80˚, −40˚, +40˚, and +80˚) through the headphones. Perception of the stimuli induced
systematic eye movements in the direction of the sound (S6 Fig), indicating that the spatial
information of the stimuli was preserved through the headphones and that the monkey could
perceive it.

Experimental design
Subjects were scanned in a sitting position, head-fixed, while engaged in a visual fixation task
(fixation window: 2˚). Eye position was monitored at 60 Hz with a camera-based system (Sen-
soriMotoric Instruments, Teltow, Germany), and correct fixation was rewarded by drops of
fruit juice.
To avoid any contamination of the stimulus-induced BOLD responses by the response
evoked by the acoustic noise of the scanner, a sparse-sampling paradigm was used for all audi-
tory experiments. Images were acquired every 10 s (acquisition time: 1.6 s), stimuli being pre-
sented during the 8.4 s silent gap. Based on a previous time course characterization of the
BOLD response in the auditory system of macaques [28], the plateau phase of the BOLD
response was targeted in experiments 1, 3, and 4 by starting acquisition of the images 5 s after
the stimulus onset. In Experiment 2, image acquisition started 2, 3, 4, and 5 s after the stimulus
onset. To obtain baseline data, stimuli were omitted in 25% of the trials prior to image acquisi-
tion. The visual localizer experiment was acquired with a continuous paradigm. Stimuli were
delivered in a pseudorandomized way, ensuring that each stimulus was presented the same
number of times within each daily session. Because we aimed to only analyze trials in which
the monkey was fixating, we interrupted the session when the monkey stopped fixation for
more than 5 min. The number of trials per stimulus therefore varied from one session to the
next, according to the monkey’s willingness to participate in the visual fixation task. Between
28 and 50 images per stimulus were acquired in each daily session. For the tonotopy experi-
ment, 3 sessions were acquired in each subject. For the first, third, and fourth auditory motion
experiments, 5, 6, and 17 sessions were acquired in M1, and 7, 6, and 14 sessions were acquired
in M2. For the hemodynamic response function experiment (Experiment 2), 11 sessions were

PLOS Biology | [Link] May 4, 2017 17 / 24

 Auditory motion detectors

acquired in M1. For the visual localizer experiment (Experiment 5), 7 sessions were acquired
in M2.

Imaging data acquisition

Data were recorded in a 4.7 T actively shielded vertical MRI scanner (Bruker Biospec 47/60
VAS) equipped with an actively shielded gradient system (Bruker GA-38S) of 38 cm innerbore
diameter (Bruker BioSpin, Ettlingen, Germany). A transmit/receive volume RF coil with an
active decoupler (Bruker) was used to acquire functional and nonisotropic structural data. The
volume coil in the transmit-only mode and an 8-channel receiving surface phased-array coil
(H. Kolster, Windmiller Kolster Scientific, Fresno, California, US) were used to acquire isotro-
pic structural data in order to generate three-dimensional surfaces.
Nonisotropic structural T1-weighted images (resolution: 0.5 mm x 0.5 mm x 2 mm) were
acquired at the end of each session using a modified driven equilibrium Fourier transform
(MDEFT) sequence with the same slice geometry as the functional scans to simplify coregistra-
tion. The imaging parameters were as follows: FOV: 12.8 cm x 9.6 cm; FA: 30˚, TI: 800 ms; TE:
6 ms, TR: 2 s. Isotropic structural T1- and T2-weighted images (FOV: 10 cm x 10 cm; resolu-
tion: 0.6 mm x 0.6 mm x 0.6 mm) were acquired during a separate session using a magnetiza-
tion-prepared rapid gradient-echo (MP-RAGE) sequence (FA: 27˚, TI: 800 ms, TE: 7 ms, TR:
2.1 s), and a rapid acquisition with relaxation enhancement (RARE) sequence (TE: 14 ms,
RARE factor: 8, TR: 5.5 s), respectively. No parallel acceleration was used.
Functional data covering the whole STG were acquired with a single-shot gradient-echo
echo-planar imaging (EPI) sequence optimized for each monkey. Typical parameters were as
follows: FOV: 12.8 cm x 9.6 cm; FA: 90˚, TE: 21 ms, TA: 1.6 s, axial orientation, slice thickness:
2 mm, interleaved slice acquisition. The inplane resolution was 1 mm x 1 mm for M1 and 1
mm x 1.5 mm for M2.

Data analysis
Functional MRI data were analyzed with SPM8 ([Link] Data
acquired from each animal were processed separately in their native space. Images from each
session were first realigned to the mean EPI image. No attempt was made to coregister EPI
and structural scans. Instead, a pair consisting of a mean EPI image and a nonisotropic struc-
tural scan acquired during the same session was chosen as a reference, based on the quality of
their alignment to each other. All functional images were coregistered to this reference EPI
image, and all the nonisotropic structural scans were coregistered to the corresponding struc-
tural scan. The isotropic structural scans were coregistered to the mean of all nonisotropic
structural scans. Functional data were smoothed with a kernel of 2 mm fullwidth at half maxi-
mum, high-pass filtered with a cut-off of 300 s to account for slow signal drifts, and adjusted
for global signal fluctuations (global scaling).
In a general linear model analysis for the combined sessions of each experiment, the voxel-
wise response estimate coefficients (beta-values) and t-values (one sided t-test) for the different
contrasts of interest were calculated (head movement parameters were regressed out). Associ-
ated p-values were corrected for multiple comparisons using the FWE correction on the bilat-
eral STG (Experiments 1, 2, and 3) and for the STG contralateral to the stimuli (Experiments 4
and 5). For auditory and visual motion experiments, data acquired while subjects did not fixate
were discarded. For tonotopy experiments, all data were used.
In Experiment 4, a multiple linear regression analysis was performed across voxels in each
subject using SPSS (IBM SPSS Statistics 21.0). The analysis was first performed across the set
of voxels where the contrasts Motion Left minus (Left Stationary central sound + Left Spatial

PLOS Biology | [Link] May 4, 2017 18 / 24

 Auditory motion detectors

laterality + Left Spectrotemporal effect of motion) and Motion Right minus (Right Stationary cen-
tral sound + Right Spatial laterality + Right Spectrotemporal effect of motion) performed at the
voxel level were significant (t-values > 4.1, corrected p < 0.05). In order to determine the
source of the signal that was not explained by the linear addition of components, we tested the
following model across those voxels:

Motion = b1 x (Stationary central sound) + b2 x (Spatial laterality) + b3 x (Spectrotemporal

effect of motion) + b4 x (Stationary central sound x Spatial laterality) + b5 x (Stationary
central sound x Spectrotemporal effect of motion) + b6 x (Stationary central sound x Spatial
laterality x Spectrotemporal effect of motion) + Motion constant term.

First- and second-order interactions between the three explanatory factors were incorpo-
rated only if this more complex model significantly increased the percentage of variance
explained by the model. As a control, we performed the same analysis across a subset of voxels
taken from the region where the voxel-based contrasts did not induce any significant differ-
ence (t-values < 4.1, corrected p > 0.05). To match the statistical power of both analyses, the
size of this second voxel set was matched to the first one by selecting voxels with the smallest t-
values. Since the probability of false-negative results decreases with the t value, this procedure
reduced the risk of selecting false-negative voxels. Performing these analyses on smoothed and
unsmoothed data provided similar results. Only results based on unsmoothed data are
described in the Results section of the manuscript.

Result displayed on 3D surfaces

Isotropic structural images were used to generate the rendered surfaces. The ratio between
T1-weighted and T2-weighted images was computed, and the resulting image was used to
manually segment the gray matter of the STG. The binary image was used to generate a tri-
dimensional triangulated mesh using BrainVisa suite ([Link] The functional
results (contrast maps and t maps) were then projected with BrainVisa onto the rendered
surface.

Tonotopy maps and localization of auditory areas

Tonotopy maps (S1 Fig) were calculated by subtracting the response estimate coefficient (beta-
values) of the low-frequency condition (0.5–1 kHz) from the high-frequency condition (8–16
kHz). The contrast High frequency minus Low frequency was inclusively masked by the contrast
High frequency minus Silence, while the contrast Low frequency minus High frequency was
masked by the contrast Low frequency minus Silence (p < 0.05, uncorrected for multiple com-
parisons for both masks). The low- and high-frequency reversals of tonotopic gradients were
identified on the surfaces and used to define the position of the core and belt areas on the supe-
rior bank of the STG: the low-frequency gradient reversal defined the rostral border of A1 and
ML, and the high-frequency reversal defined the caudal border of A1 and ML and the rostral
border of CL (see Fig 2A). The caudal parabelt was defined as the region lateral to ML and CL
on the STG convexity.

Supporting information
S1 Fig. Tonotopy maps. Contrast maps, representing the degree of preference for high (cyan-
blue colors) and low frequencies (red–yellow colors), are shown on a surface rendering of the
superior bank of the STG.
(TIF)

PLOS Biology | [Link] May 4, 2017 19 / 24

 Auditory motion detectors

S2 Fig. Spectro-temporal effect of motion (in cyan) and spatial laterality (in yellow)
(Experiment 4). Voxels where each contrast is statistically significant (PFWE < 0.05) are
respectively colored in cyan and yellow (t-tests: N = 9927 (M1) and 6664 (M2); df = 8902 (M1)
and 5922 (M2)). The green area corresponds to voxels where both contrasts are significant 9.
Statistical results are shown on a surface rendering of the superior temporal gyrus (STG) con-
tralateral to the stimuli. Spectro-temporal effect of motion was assessed with the contrast Spec-
tro-temporal control minus Stationary central sound and spatial laterality with the contrast
Stationary Left/Right minus Stationary central sound. For more details, see Fig 2 legend.
(TIF)
S3 Fig. Multiple linear regression plots inside and outside the motion-specific brain region
(green cluster in Fig 5). The plots represent the relationship between the data estimated by
the best adjusted model (Motion = b1 x (Stationary central sound) + b2 x (Spectro-temporal
processes) + b3 x (Spatial laterality) + interactions + constant term) and the experimental data
in each subject (M1 and M2). Data from both hemispheres have been merged for each subject.
Inside the motion specific region. Number of voxels (n) = 30, F(3,26) = 40.1, p < 0.001 (M1),
n = 130, F(3, 126) = 221.9, P < 0.001 (M2); outside the motion-specific region: n = 30, F(3,26)
= 154, p < 0.001 (M1), n = 135, F(3.131) = 194.2, p < 0.001 (M2). For more details, see Table 1
legend.
(TIF)
S4 Fig. Encoding of static spatial location. Significant statistical results (PFWE < 0.05) are
shown on a surface rendering of the superior temporal gyrus (STG) contralateral to the stimuli
(N = 2466 (M1) and 3429 (M2); df = 2210 (M1) and 3071 (M2). Contra-hemisphere preference
(in blue) was assessed by the contrast Stationary Left minus Stationary Right, projected on the
right STG, and Stationary Right minus Stationary Left, projected on the left STG. These con-
trasts did not reveal any significant ipsilateral preference. The contrasts Stationary central
sound minus Stationary Right and Stationary central sound minus Stationary Left did not reveal
any preference for central sounds. Laterality preference (in red) was assessed by the contrasts
Stationary Left minus Stationary central sounds and Stationary Right minus Stationary central
sound. The black area corresponds to voxels where both contrasts were significant. This maps
illustrate the refinement of static spatial processing between A1 and the dorso-caudal regions
of the auditory cortex: while the broad location of static stimuli (left versus right hemispace)
was encoded in large parts of the auditory cortex, including A1, the more precise location of
the sounds (lateral versus central positions) was only processed in the most dorso-caudal
regions (CL, caudal parabelt, inferior bank of the STG).
(TIF)
S5 Fig. Human psychophysical data. Stimuli were moving stimuli and concatenated-static sti-
muli moving along one of four paths: from +90 to +180˚, from -90 to 0˚, from -90 to +180˚
and from 90 to 0˚. Plotted data represents the percentage of correct responses over 240 trials
that three human subjects made in the AXB psychophysical experiments. The dashed line rep-
resents chance level (50% of correct answer).  : significantly different from chance level using a
chi-square test (fore detailed statistics, see text).
(TIF)
S6 Fig. Monkey eye movement recordings. Eye movements when the subject was asked to
not move his eyes (Fix, fixation spot on) and when subject heard the auditory stimulus while
free to move his eyes (stimulus presentation, fixation spot off). Auditory stimuli were left (-80
or -40˚) stationary stimuli (left) or right (40 or 80˚) stationary stimuli (pink).
(TIF)

PLOS Biology | [Link] May 4, 2017 20 / 24

 Auditory motion detectors

Author Contributions
Conceptualization: Colline Poirier, Timothy D. Griffiths.
Formal analysis: Colline Poirier.
Funding acquisition: Colline Poirier, Li Sun, Adrian Rees, Alexander Thiele, Timothy D.
Griffiths.
Investigation: Colline Poirier, Adrian Rees.
Methodology: Colline Poirier, Simon Baumann, Pradeep Dheerendra, Adrian Rees, Alexander
Thiele, Timothy D. Griffiths.
Project administration: Colline Poirier, Timothy D. Griffiths.
Resources: David Hunter, Fabien Balezeau, Li Sun, Christopher I. Petkov, Alexander Thiele.
Software: Pradeep Dheerendra, Olivier Joly.
Supervision: Adrian Rees, Alexander Thiele, Timothy D. Griffiths.
Visualization: Colline Poirier.
Writing – original draft: Colline Poirier.
Writing – review & editing: Simon Baumann, Pradeep Dheerendra, Olivier Joly, David
Hunter, Fabien Balezeau, Li Sun, Adrian Rees, Christopher I. Petkov, Alexander Thiele,
Timothy D. Griffiths.

References
1. Grantham DW. Detection and discrimination of simulated motion of auditory targets in the horizontal
plane. J Acoust Soc Am. 1986; 79(6):1939–49. PMID: 3722604
2. Chandler DW, Grantham DW. Minimum audible movement angle in the horizontal plane as a function of
stimulus frequency and bandwidth, source azimuth, and velocity. J Acoust Soc Am. 1992; 91(3):1624–
36. PMID: 1564199
3. Perrott DR, Musicant AD. Minimum auditory movement angle: binaural localization of moving sound
sources. J Acoust Soc Am. 1977; 62(6):1463–6. PMID: 591679
4. Perrott DR, Musicant AD. Dynamic minimum audible angle: binaural spatial acuity with moving sound
sources. J Audiol Res. 1981; 21:287–96.
5. Perrott DR, Malborough K. Minimum audible movement angle: Marking the end points of the path trav-
eled by a moving sound source. J Acoust Soc Am. 1989; 85(4):1773–5. PMID: 2708691
6. Perrott DR, Costantino B, Ball J. Discrimination of moving events which accelerate or decelerate over
the listening interval. J Acoust Soc Am. 1993; 93(2):1053–7. PMID: 8445116
7. Carlile S, Best V. Discrimination of sound source velocity in human listeners. J Acoust Soc Am. 2002;
111(2):1026–35. PMID: 11863159
8. Getzmann S. Effects of velocity and motion-onset delay on detection and discrimination of sound
motion. Hear Res. 2008; 246(1–2):44–51. [Link] PMID:
18848613
9. Locke SM, Leung J, Carlile S. Sensitivity to Auditory Velocity Contrast. Sci Rep. 2016; 6:27725. https://
[Link]/10.1038/srep27725 PMID: 27291488
10. Ahissar M, Ahissar E, Bergman H, Vaadia E. Encoding of sound-source location and movement: activity
of single neurons and interactions between adjacent neurons in the monkey auditory cortex. J Neuro-
physiol. 1992; 67(1):203–15. PMID: 1552320
11. Spitzer MW, Semple MN. Responses of inferior colliculus neurons to time-varying interaural phase dis-
parity: effects of shifting the locus of virtual motion. J Neurophysiol. 1993; 69:1245–63. PMID: 8492161
12. Wilson WW, O’Neill WE. Auditory motion induces directionally dependent receptive field shifts in inferior
colliculus neurons. J Neurophysiol. 1998; 79(4):2040–62. PMID: 9535967

PLOS Biology | [Link] May 4, 2017 21 / 24

 Auditory motion detectors

13. McAlpine D, Jiang ZD, Shackleton TM, Palmer AR. Responses of neurons in the inferior colliculus to
dynamic interaural phase cues: evidence for a mechanism of binaural adaptation. J Neurophysiol.
2000; 83(3):1356–65. PMID: 10712463
14. Ingham NJ, Hart HC, McAlpine D. Spatial receptive fields of inferior colliculus neurons to auditory appar-
ent motion in free field. J Neurophysiol. 2001; 85(1):23–33. PMID: 11152702
15. Kusmierek P, Rauschecker JP. Selectivity for space and time in early areas of the auditory dorsal
stream in the rhesus monkey. J Neurophysiol. 2014; 111(8):1671–85. [Link]
2013 PMID: 24501260
16. Miller LM, Recanzone GH. Populations of auditory cortical neurons can accurately encode acoustic
space across stimulus intensity. Proc Natl Acad Sci U S A. 2009; 106(14):5931–5. [Link]
1073/pnas.0901023106 PMID: 19321750
17. Recanzone GH, Guard DC, Phan ML, Su TK. Correlation between the activity of single auditory cortical
neurons and sound-localization behavior in the macaque monkey. J Neurophysiol. 2000; 83:2723–39.
PMID: 10805672
18. Tian B, Reser D, Durham A, Kustov A, Rauschecker JP. Functional Specialization in Rhesus Monkey
Auditory Cortex. Science (80-). 2001; 292(5515):290–3.
19. Woods TM, Lopez SE, Long JH, Rahman JE, Recanzone GH. Effects of Stimulus Azimuth and Intensity
on the Single-Neuron Activity in the Auditory Cortex of the Alert Macaque Monkey. J Neurophysiol.
2006; 96:3323–37. [Link] PMID: 16943318
20. Baumgart F, Gaschler-markefski B, Woldorff MG, Hans-Jochen H, Henning S. A movement-sensitive
area in auditory cortex. Nature. 1999; 400:725–6.
21. Warren JD, Zielinski BA, Green GGR, Rauschecker JP, Griffiths TD. Perception of sound-source
motion by the human brain. Neuron. 2002; 34(1):139–48. PMID: 11931748
22. Griffiths TD, Rees A, Witton C, Shakir RA, Henning GB, Green GG. Evidence for a sound movement
area in the human cerebral cortex. Nature. 1996; 383(6599):425–7. [Link]
PMID: 8837772
23. Ducommun CY, Michel CM, Clarke S, Adriani M, Seeck M, Landis T, et al. Cortical motion deafness.
Neuron. 2004; 43:765–77. [Link] PMID: 15363389
24. Lewald J, Peters S, Corballis MC, Hausmann M. Perception of stationary and moving sound following
unilateral cortectomy. Neuropsychologia. 2009; 47(4):962–71. [Link]
neuropsychologia.2008.10.016 PMID: 19022269
25. Thaler L, Paciocco J, Daley M, Lesniak GD, Purcell DW, Fraser JA, et al. A selective impairment of per-
ception of sound motion direction in peripheral space: A case study. Neuropsychologia. 2016; 80:79–
89. [Link] PMID: 26586155
26. Brown CH, Schessler T, Moody MD, Stebbins WC. Vertical and horizontal sound localization in pri-
mates. J Acoust Soc Am. 1982; 72:1804–11. PMID: 7153427
27. Perrott DR, Saberi K. Minimum audible angle thresholds for sources varying in both elevation and azi-
muth. J Acoust Soc Am. 1990; 87(4):1728–31. PMID: 2341677
28. Baumann S, Griffiths TD, Rees A, Hunter D, Sun L, Thiele A. Characterisation of the BOLD response
time course at different levels of the auditory pathway in non-human primates. Neuroimage. 2010; 50
(3):1099–108. [Link] PMID: 20053384
29. Pavani F, Macaluso E, Warren JD, Driver J, Griffiths TD. A common cortical substrate activated by hori-
zontal and vertical sound movement in the human brain. Curr Biol. 2002; 12(18):1584–90. PMID:
12372250
30. Krumbholz K, Schönwiesner M, Von Cramon DY, Rubsamen R, Shah NJ, Zilles K, et al. Representation
of interaural temporal information from left and right auditory space in the human planum temporale and
inferior parietal lobe. Cereb Cortex. 2005; 15(3):317–24. [Link] PMID:
15297367
31. Poirier C, Collignon O, DeVolder AG, Renier L, Vanlierde A, Tranduy D, et al. Specific activation of the
V5 brain area by auditory motion processing: An fMRI study. Cogn Brain Res. 2005; 25(3):650–8.
32. Meng J-A, Saberi K, Hsieh I-H. Velocity Selective Networks in Human Cortex Reveal Two Functionally
Distinct Auditory Motion Systems. PLoS ONE [Internet]. 2016; 11(6):e0157131. Available from: http://
[Link]/10.1371/[Link].0157131 [Link] PMID:
27294673
33. Baumann S, Petkov CI, Griffiths TD. A unified framework for the organization of the primate auditory
cortex. Front Syst Neurosci. 2013; 7:11. [Link] PMID: 23641203
34. Saenz M, Langers DRM. Tonotopic mapping of human auditory cortex. Hear Res. 2014; 307:42–52.
[Link] PMID: 23916753

PLOS Biology | [Link] May 4, 2017 22 / 24

 Auditory motion detectors

35. Smith KR, Okada K, Saberi K, Hickok G. Human cortical auditory motion areas are not motion selective.
Neuroreport. 2004; 15(9):1523–6. PMID: 15194888
36. Smith KR, Saberi K, Hickok G. An event-related fMRI study of auditory motion perception: No evidence
for a specialized cortical system. Brain Res. 2007; 1150(1):94–9.
37. Krumbholz K, Schönwiesner M, Rubsamen R, Zilles K, Fink GR, Von Cramon DY. Hierarchical process-
ing of sound location and motion in the human brainstem and planum temporale. Eur J Neurosci. 2005;
21(1):230–8. [Link] PMID: 15654860
38. Krumbholz K, Hewson-Stoate N, Schönwiesner M. Cortical response to auditory motion suggests an
asymmetry in the reliance on inter-hemispheric connections between the left and right auditory cortices.
J Neurophysiol. 2007; 97(2):1649–55. [Link] PMID: 17108095
39. Grantham DW. Auditory motion perception: Snapshots revisited. In: Gilkey R.H. and Anderson TR, edi-
tor. Binaural and Spatial Hearing in Real and Virtual Environments. Mahwah, NJ: Lawrence Erlbaum
Associates; 1997. p. 295–313.
40. Perrott DR, Tucker J. Minimum audible movement angle as a function of the azimuth and elevation of
the source. J Acoust Soc Am. 1988; 83(4):1522–7. PMID: 3372867
41. Mills AW, Movements H, Sound D. On the minimum audible angle. J Acoust Soc Am. 1985; 30(4):237–
46.
42. Stecker GC, Harrington IA, Middlebrooks JC. Location Coding by Opponent Neural Populations in the
Auditory Cortex. PLoS Biol. 2005; 3(3):520–8.
43. Bach DR, Neuhoff JG, Perrig W, Seifritz E. Looming sounds as warning signals : The function of motion
cues. Int J Psychophysiol. 2009; 74(1):28–33. [Link] PMID:
19615414
44. Romanski LM, Averbeck BB. The primate cortical auditory system and neural representation of conspe-
cific vocalizations. Annu Rev Neurosci. 2009; 32:315–46. [Link]
051508.135431 PMID: 19400713
45. Belin P, Zatorre RJ. “What”, “where” and “how” in auditory cortex. Nat Neurosci. 2000; 3(10):965–6.
[Link] PMID: 11017161
46. Kaas JH, Hackett TA. Subdivisions of auditory cortex and processing streams in primates. Proc Natl
Acad Sci U S A. 2000; 97(22):11793–9. [Link] PMID: 11050211
47. Hickok G, Poeppel D. The cortical organization of speech processing. Nat Rev Neurosci. 2007; 8
(5):393–402. [Link] PMID: 17431404
48. Hackett TA, de la Mothe LA, Camalier CR, Falchier A, Lakatos P, Kajikawa Y, et al. Feedforward and
feedback projections of caudal belt and parabelt areas of auditory cortex: Refining the hierarchical
model. Front Neurosci. 2014; 8:1–21.
49. Kravitz DJ, Saleem KS, Baker CI, Mishkin M. A new neural framework for visuospatial processing. Nat
Rev Neurosci. 2011; 12(4):217–30. [Link] PMID: 21415848
50. Chafee M V, Crowe D a. Thinking in spatial terms: decoupling spatial representation from sensorimotor
control in monkey posterior parietal areas 7a and LIP. Front Integr Neurosci. 2012; 6:112. [Link]
org/10.3389/fnint.2012.00112 PMID: 23355813
51. Binkofski F, Buxbaum LJ. Two action systems in the human brain. Brain Lang. 2013; 127(2):222–9.
[Link] PMID: 22889467
52. Rizzolatti G, Matelli M. Two different streams form the dorsal visual system: Anatomy and functions.
Exp Brain Res. 2003; 153(2):146–57. [Link] PMID: 14610633
53. Pisella L, Binkofski F, Lasek K, Toni I, Rossetti Y. No double-dissociation between optic ataxia and
visual agnosia: Multiple sub-streams for multiple visuo-manual integrations. Neuropsychologia. 2006;
44(13):2734–48. [Link] PMID: 16753188
54. Mateeff S, Hohnsbein J, Noack T. Dynamic visual capture: apparent auditory motion induced by a mov-
ing visual target. Perception. 1985; 14:721–7. [Link] PMID: 3837873
55. Lakatos S. The influence of visual cues on the localisation of circular auditory motion. Perception. 1995;
24:457–65. [Link] PMID: 7675624
56. Warren JE, Wise RJS, Warren JD. Sounds do-able: Auditory-motor transformations and the posterior
temporal plane. Trends Neurosci. 2005; 28(12):636–43. [Link]
PMID: 16216346
57. Zatorre RJ, Chen JL, Penhune VB. When the brain plays music: auditory-motor interactions in music
perception and production. Nat Rev Neurosci. 2007; 8(7):547–58. [Link]
PMID: 17585307
58. Recanzone GH, Cohen YE. Serial and parallel processing in the primate auditory cortex revisited.
Behav Brain Res. 2010; 206:1–7. [Link] PMID: 19686779

PLOS Biology | [Link] May 4, 2017 23 / 24

 Auditory motion detectors

59. Rauschecker JP, Scott SK. Maps and streams in the auditory cortex: nonhuman primates illuminate
human speech processing. Nat Neurosci. 2009; 12(6):718–24. [Link] PMID:
19471271
60. Thiele A, Delicato LS, Roberts MJ, Gieselmann MA. A novel electrode-pipette design for simultaneous
recording of extracellular spikes and iontophoretic drug application in awake behaving monkeys. J Neu-
rosci Methods. 2006; 158(2):207–11. [Link] PMID: 16843532
61. Henning GB. Detectability of interaural delay in high-frequency complex waveforms. J Acoust Soc Am.
1973; 55:84–90.
62. Wightman FL, Kistler DJ. Headphone simulation of free-field listening. I: Stimulus synthesis. J Acoust
Soc Am. 1989; 85:858–67. PMID: 2926000
63. Recanzone GH, Beckerman NS. Effects of intensity and location on sound location discrimination in
macaque monkeys. Hear Res. 2004; 198:116–24. [Link] PMID:
15567608

PLOS Biology | [Link] May 4, 2017 24 / 24