Content uploaded by Arran T Reader
Author content
All content in this area was uploaded by Arran T Reader on Jan 13, 2016
Content may be subject to copyright.
Journal Club
Editor’s Note: These short, critical reviews of recent papers in the Journal, written exclusively by graduate students or postdoctoral
fellows, are intended to summarize the important findings of the paper and provide additional insight and commentary. For more
information on the format and purpose of the Journal Club, please see http://www.jneurosci.org/misc/ifa_features.shtml.
Semantic Organization of Body Part Representations in the
Occipitotemporal Cortex
XArran T. Reader
School of Psychology and Clinical Language Sciences, University of Reading, Reading, RG6 6AL, United Kingdom
Review of Bracci et al.
Experiments using functional magnetic
resonance imaging (fMRI) have revealed
that the ventral and lateral occipitot-
emporal cortices (VOTC and LOTC,
respectively), areas often associated with
high-level visual processing (Grill-
Spector and Malach, 2004), show prefer-
ential activation during the observation
of body parts. Early results examining
preferential brain activation during ob-
servation of body parts highlighted the
importance of a region in the LOTC. This
was termed the extrastriate body area
(EBA) by Downing et al. (2001), who
found that this region responded more
strongly to images of body parts than to
various control stimuli. Supporting these
results, Urgesi et al. (2004) applied trans-
cranial magnetic stimulation over EBA,
which resulted in a reduced ability to dis-
criminate between body parts. More re-
cently, Bracci et al. (2010) found an area of
the LOTC that responded preferentially to
the observation of hands, suggesting that
body part representation in the OTC is
likely to be heterogeneous.
While the organization of body part
representations (i.e., somatotopy) in the
somatosensory and motor cortices is well
established, little evidence existed for sim-
ilar organization in the OTC until rela-
tively recently. Orlov et al. (2010) found
that distinct areas of the OTC represented
different visually presented body parts.
They also showed that the organization of
these representations was not accounted
for simply by the shape of the body parts.
Such findings suggested that body part
representations in the LOTC and VOTC
might be organized according to one or
more criteria, for example shape or func-
tion. In a recent issue of The Journal of
Neuroscience,Bracci et al. (2015) further
examined how body-part-related activity
in the OTC is organized. Using fMRI and
a technique called representational simi-
larity analysis (RSA; Kriegeskorte et al.,
2008), they tested five different models of
body part organization in the OTC. Their
results revealed that a model based on se-
mantic similarity best accounted for the
organization of body part representations
in the OTC.
Bracci and colleagues (2015) showed
participants images of whole bodies,
hands, feet, arms, legs, chests, waists, up-
per faces, lower faces, and chairs, while
they performed a one-back repetition
detection task and fMRI data were col-
lected. Regions of interest (ROIs) in the
LOTC and VOTC were defined in each
participant by contrasting the blood-
oxygenation level dependent (BOLD) re-
sponse to whole bodies versus chairs.
ROIs within the OTC were found in both
hemispheres for all participants. Control
regions in the occipital cortex (OC) were
defined by the opposite contrast: chairs
versus whole bodies. Multivoxel pattern
analysis, a tool for comparing patterns of
activation across multiple voxels, was
then used to compare the similarities of
activation among the eight remaining
body part activations in the LOTC and
VOTC. These patterns were correlated
across the different body parts to create an
eight-by-eight neural dissimilarity matrix
(Bracci et al., 2015, their Fig. 3A).
The authors used RSA, which involves
comparing dissimilarity matrices between
a model and measured brain activation
(Fig. 1). Using RSA, different theoretical
models can be compared to discover the
degree to which each can explain the ob-
served brain activity. To use RSA, the au-
thors created dissimilarity matrices for
five possible models of body part organi-
zation: physical shape similarity (com-
puted using the shape context algorithm;
Belongie et al., 2002), perceived shape
similarity (based on six independent par-
ticipants’ arrangement of body parts
based on shape similarity), physical prox-
imity in the body, organization following
the cortical sensory and motor homun-
culi, and semantic similarity (Bracci et al.,
2015, their Fig. 2A). These dissimilarity
matrices were then compared statistically
with the neural dissimilarity matrices of
each of the four ROIs (left and right LOTC
and VOTC).
In the primary analyses, the authors
replicated the body-part-specific activa-
tion of OTC reported previously. More-
over, they found that only the perceived
Received Oct. 14, 2015; revised Nov. 19, 2015; accepted Nov. 23, 2015.
Many thanks to Dr. Nicholas P. Holmes for his comments on the manu-
script, and for his continued supervision and support.
The author declares no competing financial interests.
Correspondence should be addressed to Arran T. Reader, School of
Psychology and Clinical Language Sciences, University of Reading, Ear-
ley Gate, Whiteknights Road, Reading, RG6 6AL, UK. E-mail:
a.reader@pgr.reading.ac.uk.
DOI:10.1523/JNEUROSCI.3766-15.2016
Copyright © 2016 the authors 0270-6474/16/360265-03$15.00/0
The Journal of Neuroscience, January 13, 2016 •36(2):265–267 • 265
shape, physical proximity, and semantic
similarity models were significantly and
positively related to neural similarity in
the LOTC and VOTC. The physical shape
and cortical homunculus models failed to
account for any additional variance in the
neural similarity. When these models
were directly compared, however, the
semantic similarity model was signifi-
cantly more related to neural similarity
than any of the other models, suggesting
that the organization of body part repre-
sentations in the LOTC and VOTC is at
least partly semantic. In examining the
neural dissimilarity matrices, the authors
found three distinctly organized clusters:
effectors (hands, feet, legs, arms), nonef-
fectors (chests, waists), and faces (upper
faces, lower faces). To check if nearby re-
gions showed complementary organiza-
tion, the analysis was performed again on
the left and right OC (control ROIs). Per-
haps unsurprisingly, considering the
importance of the occipital cortex in pro-
cessing lower-level visual stimuli, only the
shape similarity models were significantly
related to patterns of BOLD response in
these areas.
Bracci et al. (2015) provide evidence
that the organization of body part repre-
sentations in the OTC is at least partially
semantic, though such a model alone
could not fully explain the neural similar-
ity. As the authors note, this may have
been due to “imperfections in the models
and/or the existence of additional organi-
zational principles that we did not con-
sider” (p. 12983). The former may be
most likely, considering that the method
by which the semantic similarity model
was defined (frequency of body part word
co-occurrence in written text) may not
adequately reflect the way in which such
semantic similarity is represented in the
OTC. It may be the case that semantic or-
ganization in the OTC is quite different
than that which can be found in the distri-
bution of words in text. For example,
while the combination of body parts in
text may reflect a wide variety of possible
functions (i.e., both feet and hands being
used in climbing, or both hand and
mouth being used in eating), their repre-
sentation in the OTC may be related to
more specific general functions such as
expressing emotion, or moving aversive/
attractive stimuli away from/towards the
body. This may also explain why the phys-
ical proximity model was also positively
associated with neural similarity because,
for example, hands are almost always used
in combination with the arms.
Regardless of how the semantic model
was generated, the clusters of organiza-
tion within the LOTC and VOTC (effec-
tors, noneffectors, faces) did indeed
suggest an organization based on func-
tion. Tools (objects used as effectors) are
represented close to body effectors in the
OTC, and these regions are functionally
connected to the left intraparietal sulcus
(IPS) and left premotor cortex (Bracci et
al., 2012). Interestingly, a whole-brain
analysis run by Bracci et al. (2015) re-
vealed additional clusters of frontal and
parietal activation that were significantly
and positively associated with the seman-
tic similarity model once it was contrasted
with the other four models. These areas
were the left and right superior parietal
lobule, left inferior parietal lobule, left
ventral premotor cortex, and left dorsal
premotor cortex. This may provide
greater insight into the role of these areas
in action perception and execution. Cer-
tainly, it supports the claim that the OTC
is functionally connected to these regions.
This association could be a reflection of
the sensorimotor and functional similar-
ity of body part representations seen in
these areas (Rizzolatti and Craighero,
2004), but it may also highlight overlap-
ping pathways relevant for semantic
meaning and action processing. For ex-
ample, the left IPS has been implicated
both in the processing of words and ob-
jects (Devereux et al., 2013) and in the
representation of observed action goals
(Hamilton and Grafton, 2006).
Bracci et al. (2015) posited that seman-
tic organization in the OTC could be a
function of connectivity demands. Body
part information related to action would
feedforward to downstream motor areas,
while areas representing faces would be
strongly connected to regions involved in
social cognition. These claims may be
supported by new findings in studies on
the movement disorder limb apraxia,
which indicate that damage to the OTC
often underlies defective pantomimed
tool use, defective imitation of panto-
mimes, and defective imitation of mean-
ingless gestures (Buxbaum et al., 2014;
Hoeren et al., 2014). In addition, lesions
to the OTC have been implicated in im-
paired (tool-related) action recognition
(Tarhan et al., 2015). Such results confirm
Figure 1. Simplified overview of RSA as used by Bracci et al. (2015). Correlation-based multivoxel pattern analysis can be used
to determine voxelwisedissimilarity between neural activity related to the stimuli a– d.This is used to create a neural dissimilarity
matrix. A dissimilarity matrix is a square, symmetrical matrix that compares the similarity between pairs of elements. Diagonal
pairs (showing the dissimilarity between an element and itself) are defined as zero. The neural dissimilarity matrix can be
comparedtothedissimilaritymatricesposited by a number of models (in this example, model 1 and model 2). InthecaseofBracci
et al. (2015), the authors used a multiple regression (for each individual participant) using the model dissimilarity matrices as
dependent variables and the neural dissimilarity matrices as independent variables. Differences between the computed regres-
sion coefficients were compared using pairwise ttests to reveal which models best explained the neural dissimilarity. This
resulted in bar graphs similar to that seen in this figure, which emphasize the degree to which each model can be related
to the observed brain activity shown in the neural dissimilarity matrix, with beta estimating the explanatory contribution
of each regressor (in this case, the two models).
266 •J. Neurosci., January 13, 2016 •36(2):265–267 Reader •Journal Club
the idea that regions in the OTC could act
as an upstream hub for multiple tasks
reliant on body part recognition. Net-
works that include the OTC could have a
semantic structure built in from the very
beginning. Take action recognition as an
example. Processing of the semantic as-
pects of action recognition is sometimes
considered to rely on top-down cognition
(Wurm et al., 2014;Davey et al., 2015).
The results of Bracci et al. (2015) suggest
that the way in which semantic and visual
information are integrated for the pur-
pose of action recognition may not rely
primarily on top-down processing. If re-
gions with semantic organization in the
OTC and other areas (i.e., IPS, premotor
cortex) are functionally linked, then the
semantic nature of the observed stimu-
lus could be derived from the pathway
through which the stimulus information
travels. Therefore, broad semantic infor-
mation regarding the action context may
in fact be integrated much earlier, follow-
ing the observation of body parts. A recent
study by Wurm and Lingnau (2015) ap-
pears to support this claim, but more
work is needed to confirm the hypothesis.
Bracci et al. (2015) have increased our
understanding of how body parts are rep-
resented in the OTC. They indicate that
semantic information regarding the pos-
sible context in which body parts are ob-
served may be available surprisingly
early, during visual processing, due in
part to the organization of visual pro-
cessing regions. More importantly, their
results have implications for how we
discuss tasks reliant on body part recog-
nition. Further study of the semantic or-
ganization in OTC could shed new light
on processes as varied as action recogni-
tion and imitation, or even social cogni-
tion as a whole.
References
Belongie S, Malik J, Puzicha J (2002) Shape
matching and object recognition using shape
contexts. IEEE Trans Pattern Anal Mach Intell
24:509–522. CrossRef
Bracci S, Ietswaart M, Peelen MV, Cavina-Pratesi
C (2010) Dissociable neural responses to
hands and non-hand body parts in human left
extrastriate visual cortex. J Neurophysiol 103:
3389–3397. CrossRef Medline
Bracci S, Cavina-Pratesi C, Ietswaart M, Cara-
mazza A, Peelen MV (2012) Closely overlap-
ping responses to tools and hands in left lateral
occipitotemporal cortex. J Neurophysiol 107:
1443–1456. CrossRef Medline
Bracci S, Caramazza A, Peelen MV (2015) Rep-
resentational similarity of body parts in hu-
man occipitotemporal cortex. J Neurosci 35:
12977–12985. CrossRef Medline
Buxbaum LJ, Shapiro AD, Coslett HB (2014)
Critical brain regions for tool related and im-
itative actions: A componential analysis. Brain
137:1971–1985. CrossRef Medline
Davey J, Rueschemeyer SA, Costigan A, Murphy
N, Krieger-Redwood K, Hallam G, Jefferies E
(2015) Shared neural processes support se-
mantic control and action understanding.
Brain Lang 142:24–35. CrossRef Medline
Devereux BJ, Clarke A, Marouchos A, Tyler LK (2013)
Representational similarity analysis reveals com-
monalities and differences in the semantic process-
ing of words and objects. J Neurosci 33:
18906 –18916. CrossRef Medline
Downing PE, Jiang Y, Shuman M, Kanwisher N
(2001) A cortical area selective for visual pro-
cessing of the human body. Science 293:2470 –
2473. CrossRef Medline
Grill-Spector K, Malach R (2004) The human
visual cortex. Annu Rev Neurosci 27:
649–677. CrossRef Medline
Hamilton AF, Grafton ST (2006) Goal represen-
tation in human anterior intraparietal sulcus.
J Neurosci 26:1133–1137. CrossRef Medline
Hoeren M, Ku¨mmerer D, Bormann T, Beume L,
Ludwig VM, Vry MS, Mader I, Rijntjes M,
Kaller CP, Weiller C (2014) Neural bases of
imitation and pantomime in acute stroke pa-
tients: distinct streams for praxis. Brain 137:
2796–2810. CrossRef Medline
Kriegeskorte N, Mur M, Bandettini P (2008)
Representational similarity analysis—con-
necting the branches of systems neuroscience.
Front Syst Neurosci 2:4. CrossRef Medline
Orlov T, Makin TR, Zohary E (2010) Topo-
graphic representation of the human body in
the occipitotemporal cortex. Neuron 68:586 –
600. CrossRef Medline
Rizzolatti G, Craighero L (2004) The mirror-
neuron system. Annu Rev Neurosci 27:169–
192. CrossRef Medline
Tarhan LY, Watson CE, Buxbaum LJ (2015)
Shared and distinct neuroanatomic regions
critical for tool-related action production
and recognition: evidence from 131 left-
hemisphere stroke patients. J Cogn Neurosci
27:2491–2511. CrossRef Medline
Urgesi C, Berlucchi G, Aglioti SM (2004) Mag-
netic stimulation of extrastriate body area
impairs visual processing of nonfacial body
parts. Curr Biol 14:2130–2134. CrossRef
Medline
Wurm MF, Lingnau A (2015) Decoding actions
at different levels of abstraction. J Neurosci
35:7727–7735. CrossRef Medline
Wurm MF, Hrkac´ M, Morikawa Y, Schubotz RI
(2014) Predicting goals in action episodes at-
tenuates BOLD response in inferior frontal
and occipitotemporal cortex. Behav Brain Res
274:108–117. CrossRef Medline
Reader •Journal Club J. Neurosci., January 13, 2016 •36(2):265–267 • 267
