Effective Connectivity within the
Distributed Cortical Network for Face
Scott L. Fairhall and Alumit Ishai
Institute of Neuroradiology, University of Zurich, 8057
Face perception elicits activation within a distributed cortical
network in the human brain. The network includes visual (‘‘core’’)
regions, as well as limbic and prefrontal (‘‘extended’’) regions,
which process invariant facial features and changeable aspects of
faces, respectively. We used functional Magnetic Resonance
Imaging and Dynamic Causal Modeling to investigate effective
connectivity and functional organization between and within the
core and the extended systems. We predicted a ventral rather than
dorsal connection between the core and the extended systems
during face viewing and tested whether valence and fame would
alter functional coupling within the network. We found that the
core system is hierarchically organized in a predominantly feed-
forward fashion, and that the fusiform gyrus (FG) exerts the
dominant influence on the extended system. Moreover, emotional
faces increased the coupling between the FG and the amygdala,
whereas famous faces increased the coupling between the FG and
the orbitofrontal cortex. Our results demonstrate content-specific
dynamic alterations in the functional coupling between visual-
limbic and visual-prefrontal face-responsive pathways.
Keywords: cortical network, dynamic causal modeling, faces, fMRI,
Face perception, a highly developed visual skill in humans, is
mediated by activation in a distributed neural system that
encompasses visual, limbic, and prefrontal regions (Haxby
et al. 2000; Ishai et al. 2005). The cortical network that mediates
face perception includes the fusiform gyrus (FG), an extra-
striate region that processes the identification of individuals
the superior temporal sulcus (STS), where gaze direction and
speech-related movements are processed (Hoffman and Haxby
2000; Puce et al. 2003); the amygdala (AMG) and insula, where
facial expressions are processed (Breiter et al. 1996; Morris et al.
1996; Phillips et al. 1997; Ishai et al. 2004); the inferior frontal
gyrus (IFG), where semantic aspects are processed (Leveroni
et al. 2000; Ishai et al. 2002); and regions of the reward circuitry,
including the nucleus accumbens and orbitofrontal cortex
(OFC), where facial beauty and sexual relevance are assessed
(Aharon et al. 2001; O’Doherty et al. 2003; Ishai 2007; Kranz and
Ishai 2006). It has been proposed that the cortical network
for face perception can be divided into a ‘‘core’’ system that
includes the inferior occipital gyrus (IOG), FG, and STS, and an
extended system that includes the AMG, insula, IFG, and OFC
(Haxby et al. 2000, 2002). It is currently unknown whether the
core and the extended systems indeed comprise a cortical
network and how these regions are functionally connected.
To investigate effective connectivity within the distributed
cortical network for face perception, we combined conventional
Statistical Parametric Mapping (SPM) with Dynamic Causal Mod-
eling (DCM), a new analytic approach that allows the assessment
of effective connectivity within cortical networks (Friston et al.
2003). DCM yields a measure of cortical connectivity that is di-
independent of coincidental stimulus-locked coupling (Friston
et al. 2003; Penny et al. 2004a). DCM has been previously em-
ployed to investigate category-selective effects (Mechelli et al.
2003; Noppeney et al. 2006); bottom-up and top-down coupling
during perception and imagery (Mechelli et al. 2004); stimulus
visibility (Haynes et al. 2005); effective connectivity during
spelling and rhyming (Bitan et al. 2005); and AMG--hippocampal
coupling during memory retrieval (Smith et al. 2006).
We used DCM and Bayesian model selection to investigate the
pattern of interactions within a network of face-selective
regions during the perception of various face stimuli. We
addressed the following issues: 1) What is the architecture of
the core system? Is it predominantly a feed-forward architec-
ture, a recurrent architecture, or is it organized in parallel? 2)
How does the core system interact with the extended system?
We hypothesized that during attentive viewing, due to the piv-
otal role of the FG in face perception, the ventral, but not the
dorsal regions of the core system would exert influence on the
extended system. 3) Are there any differences in the patterns of
effective connectivity between the left and the right hemi-
sphere? 4) Can functional specialization be explained in terms
of selective enabling of coupling between regions? We hypoth-
esized that valence and fame would differentially alter effective
connectivity between the core and the extended systems,
namely that emotional faces would increase effective connec-
tivity between the FG and the AMG, whereas famous faces
would increase the connectivity between the FG and the OFC.
Ten healthy subjects (5 males, mean age 25 years) with normal vision
participated in the study. All subjects gave written informed consent in
accordance with protocols approved by the University Hospital of
Stimuli and Task
Subjects were presented with 4 types of face stimuli: black and white
line drawings of unfamiliar faces, and gray scale photographs of
unfamiliar, famous, and emotional (fearful and happy) faces. Phase
scrambled versions of these faces were used as visual baseline (see Ishai
et al. 2005). Each stimulus was presented for 3 s. Each run included 3
alternating epochs of scrambled faces (24 s) and faces (36 s). Five runs
Cerebral Cortex October 2007;17:2400--2406
Advance Access publication December 26, 2006
? 2006 The Authors
This is an Open Access article distributed under the terms of the Creative Commons Attribution Non-Commercial License (http://creativecommons.org/licenses/by-nc/2.0/uk/) which
permits unrestricted non-commercial use, distribution, and reproduction in any medium, provided the original work is properly cited.
(line drawings, famous, emotional, and 2 runs with unfamiliar faces)
were collected for each subject and the order of stimulus formats was
randomized. Stimuli were generated using Presentation (www.neurobs.
com, version 9.13) and were projected with a magnetically shielded
LCD video projector onto a translucent screen. The subject was
instructed to attentively view the faces.
Data were collected using a 3-T Philips Intera whole body MR scanner.
Functional data were obtained from 39 transverse slices covering the
whole brain with a spatial resolution of 2.3 mm 3 2.3 mm 3 3 mm
(acquisition matrix 96 3 96), using a sensitivity-encoded singleshot
gradient-echo planar sequence. Images were acquired with fields of
view = 220 mm, time repetition (TR) = 3000 ms, time echo (TE) = 35 ms,
h = 82?, and with a SENSE acceleration factor of 2.0 (Pruessmann et al.
1999). High-resolution spoiled gradient recalled echo structural images
wereobtainedwith 1mm 3 1mm 31 mmspatialresolution(acquisition
matrix 224 3 224), TE = 2.30 ms, TR = 20 ms, h = 20?. These T1-weighted
images provided detailed anatomical information for the region of
interest (ROI) analysis.
Data were analyzed using SPM5 software (www.fil.ion.ucl.ac.uk/spm/).
All volumes were slice time corrected, realigned to the first volume,
corrected for motion artifacts, mean-adjusted by proportional scaling,
normalized into standard stereotactic space (template provided by the
Montreal Neurological Institute), and smoothed using a 5-mm full-width
at half-maximum Gaussian kernel. The time series were high-pass
filtered to eliminate low-frequency components (filter width 128 s)
and adjusted for systematic differences across trials.
Statistical Parametric Mapping
The analysis was based on a conventional general linear model (Friston,
Holmes, Poline, et al. 1995; Friston, Holmes, Worsley, et al. 1995) using 3
regressors, representing the 3 experimental effects: all faces versus
scrambled faces; emotional faces versus scrambled faces; and famous
faces versus scrambled faces. The main effect of faces was used for ROI
selection and time series extraction: A set of ROIs was independently
defined for each subject, which included the IOG, FG, STS, AMG, IFG,
and the OFC. The anatomical locations of these clusters were de-
termined by superimposing the statistical maps on the coplanar high-
resolution structural images. Regional responses were defined as the
average of face-responsive voxels within a 6-mm radius centered on the
maximum voxel. The significance threshold was set to P < 0.001.
Dynamic Causal Modeling
The hemodynamic model used by DCM has been biophysically validated
(Friston et al. 2003; Stephan et al. 2004), however, the neural model is
experiment dependent and requires specific hypotheses. Thus, the
brain regions included in the model, the anatomical connectivity
between them and the modulation by experimental conditions have
to be specified (Penny et al. 2004b). The first regressor, the main effect
of faces, was used as a direct or driving input that entered the IOG, such
that any regional connections, present in the tested models, mediate the
propagation of face-selective responses in the IOG around the face-
system. The second and third effects (valence and fame) were used as
bilinear modulators of connectivity to assess the selective enabling of
pathways by these factors. Model fitting was achieved by adjusting the
connection parameters, such that the activity predicted in each region
most closely matched that observed in the actual data (Friston et al.
2003). The parameters optimize a variational free-energy bound on
the models evidence. This ensures that the model fit uses the free
parametersin a parsimonious way. After model inversion the free energy
can be used as an approximation to the model evidence, namely the
probability of the data given the model. This quantity is used below to
compare different models using Bayes factors.
Our analysis was based on Bayesian model comparison using Bayes
factors to select among competing models. The models we explored
were based upon the theoretical division into core and extended
systems (Haxby et al. 2000, 2002). Models were contrasted based on
their ability to explain the observed data, in terms of their evidence. This
is probability of the data from the ith subjects data given the jth model:
pðyijmjÞ: Using a Bayesian framework, successive pairwise comparisons
were made between models. Two criteria were used to assess the
evidence in favor of one model versus another, namely the Bayesian and
Akaike’s Information Criteria (Penny et al. 2004b). The former is biased
toward simple models, whereas the latter is biased toward complex
models (Kass and Raftery 1993). Both are approximations to
data, given the jth model, relative to the kth model. To ensure
a consistent inference about models, the Bayes factors reported here
(and used in any statistical tests) reflect the smaller of these 2 values and
are therefore conservative estimates. We report the Bayes factors
pooled over all subjects in terms of each subject-specific Bayes factor
subjects (y1, ... ,yn). To ensure this model comparison was not unduly
affected by one subject, we also tested the consistency of the Bayes
factor for each subject using a v2-test on the proportion of subjects
favoring a particular model (i.e., BðiÞ
jk=pðyijmjÞ=pðyijmkÞ; which is the probability of the ith subjects
jk=pðy1;...;ynjmjÞ=pðy1;...;ynjmkÞ for n subjects. This is
effectively the relative evidence for each model using the data from all
Testing Bilinear Effects of Valence and Fame
The enabling effects of emotion and fame on specific connections were
assessed using classical parametric tests (t-tests) on the conditional
expectation of the bilinear effect from each subject. Coupling strengths
in dynamic models play the same role as rate constants in kinetic
models. In other words, they are measured in units of per second. Thus,
a modulatoryor enabling effect of0.24, given a regionalcouplingof 0.48,
represents a 50% increase in connectivity. We therefore report in the
Tables both the increase in coupling (rate) and its percent increase,
relative to the underlying regional coupling. As can be seen, the
selective enabling of specific connections can be quite profound, in
The main effect of faces (as compared with scrambled faces)
revealed activation within a distributed cortical network that
included visual, limbic, and prefrontal regions (Fig. 1 and Table
1). The DCM analysis included 3 regions representing the core
system (the IOG, FG, and STS) and 3 regions (the AMG, IFG, and
OFC) representing the extended system.
The Core System
Directional Connectivity within the Right Hemisphere
Bayesian model selection was used to determine the best model
of effective connectivity within the core system. To determine
the most probable pattern of effective connectivity, models
were divided into 6 prototypes (Fig. 2A). Within each prototype,
the backwards connectivity was varied systematically, creating
24 models from these 6 prototypes. We found greater evidence
in favor of a direct and separate influence of the IOG on the FG
and STS, than any other model (Fig. 2A, Prototype 2). The values
in Supplementary Table 1 reflect evidence at the level compa-
rable to classic fixed effects analysis, namely the relative
probability of data from all subjects given a particular model.
A Bayes Factor (B) greater than the natural exponent e (B >
2.718) provides strong evidence in favor of one model (Penny
et al. 2004b). Thus, the feed-forward variant of Prototype 2
(Supplementary Table 1: shaded row) was favored (B = 3.1) over
the next best 3-tier model (Prototype 1). Greater evidence for
this model (B > 1) was observed in 9 out of 10 subjects, when
considered separately (v2= 6.4, P < 0.011). Evidence against
Cerebral Cortex October 2007, V 17 N 10 2401
the role of interconnectivity between FG and STS in terms of
Prototypes 4 and 5 was present but weak when data from all
subjects were considered. However, when these data were
considered for each subject separately (in terms of the most
likely model), there was reliable evidence in favor of feed-
forward Prototype 2 in comparison to either Prototype 4 (v2=
6.4, P <0.011) or Prototype 5 (v2= 6.4, P <0.011). The weight
of evidence in favor of a feed-forward system was not increased
by including backwards connections (see Supplementary Table
1). In other words, the extra complexity induced by adding
backward connections was not offset by increased model fit
(Penny et al. 2004b). Taken together, these results suggest that
the IOG directly influences both the FG and STS with little
evidence of feedback or collateral influences on the observed
blood oxygen level--dependent response.
Functionally Selective Coupling within the Right
Hemisphere Core System
The effects of valence and fame were introduced to the DCM as
bilinear terms (see Methods). Group statistics for the bilinear
A network of face-selective regions
RegionN Mean coordinates
Note: N indicates number of subjects who showed significant activation in each region.
Coordinates are in the normalized space of the brain atlas (Talairach and Tournoux 1998).
Standard error of the means are indicated in parentheses. L 5 left, R 5 right, M 5 medial.
Figure 1. Face perception elicits activation within a distributed cortical network. Axial sections, taken from a representative subject, illustrate activation within the core (IOG, FG,
STS) and extended (AMG, IFG, OFC) systems. Coordinates are in the Talaraich space. L 5 left, R 5 right.
Figure 2. (A) Prototype models of the core system. Shown are the 6 feed-forward exemplars. Patterns of reciprocal connectivity were investigated by modifying the collateral and
feedback connections. (B) Alterations in effective connectivity within the core system produced by all face stimuli, emotional faces, and famous faces. Black arrows indicate
significant regional effects, red lines indicate significant bilinear effects, and dotted lines indicate nonsignificant effects.
Cortical Connectivity within the Face Network
Fairhall and Ishai
effects are shown in Table 2. Faces per se had a strong and
significant influence on the effective connectivity between the
IOG and both the FG and STS (Fig. 2B). Although emotional and
famous faces significantly increased the influence that the IOG
exerted on the ventral (FG) pathway, this enabling effect was
not observed along the dorsal (STS) pathway.
Effective Connectivity within the Left Hemisphere
Functional brain imaging studies have consistently shown that
the right hemisphere exhibits stronger response to face stimuli,
in terms of both the number of subjects who show significant
activation in face-responsive regions, and the spatial extent of
the activation, that is, the cluster size (Sergent et al. 1992;
Kanwisher et al. 1997; Haxby et al. 1999; Rossion et al. 2000;
Ishai, Ungerleider, Martin, et al. 2000; Ishai et al. 2005; Kranz and
Ishai 2006). To test whether the left hemisphere exhibits
differential patterns of effective connectivity, we analyzed the
regional and modulatory coupling in the 7 subjects who showed
activation in the left IOG, FG, and STS (see Table 1). Consistent
with the patterns of connectivity in the right hemisphere, we
found greater evidence (B = 2.9) in favor of a direct and separate
influence of the IOG on the FG and STS (Prototype 2, Fig. 2A) in
5 of 7 subjects. Moreover, we found that emotional and famous
faces significantly increased the coupling between the IOG and
the FG (Table 3).
The Extended System
Connectivity between the Core and the Extended Systems
To test our hypothesis that ventral rather than dorsal regions of
the core system are functionally coupled with the extended
system, the AMG, IFG, and OFC were connected in all varying
combinations to either the FG or the STS in purely feed-forward
models (Supplementary Fig. 1). This analysis included the 5
subjects who showed activation in all regions of the extended
system in the right hemisphere (see Table 1). Consistent with
our hypothesis, model comparisons clearly favored a single FG
connection to the extended system, with this model having
much more evidence than the next best model (B = 21.04).
Interestingly, the extent of evidence in favor of a model was
seen to decrease on including the STS influence (Table 4).
Collateral and Feedback Connectivity
To assess the degree of lateral and reciprocal connections
within the extended system, and the degree of feedback
connectivity between the extended and core systems, a simpli-
fied set of models were constructed (Supplementary Fig. 1) and
compared with a purely feed-forward model. Models which
included lateral connections between regions of the extended
system did not have greater evidence, nor did models that
included both lateral and feedback connections (Table 5:
bottom 2 rows). Bayesian selection could not distinguish
between the purely feed-forward model and a model with
backward regional connections from the extended to the core
system (i.e., between A.1 and B.1 in Supplementary Fig. 1).
Functionally Selective Coupling within the Extended System
The more parsimonious feed-forward model (A.1 in Supple-
mentary Fig. 1) was selected to investigate whether the effects
of valence and fame can be expressed in terms of alterations in
effective connectivity in the extended system (note that all
patterns reported here were also present in the feedback
model, see B.1 in Supplementary Fig. 1). In this model, valence
and fame were allowed to influence all forward connections as
bilinear terms. As in the core system, faces were observed to
Effective connectivity within the right hemisphere core system
Note: Means, standard deviations (SDs), standard error of the means (SEMs), and P values
showing the alteration in effective connectivity within regions of the right hemisphere core
system induced by all faces, emotional faces, and famous faces.
Effective connectivity within the left hemisphere core system
All faces FG
Note: Means, standard deviations (SDs), standard error of the means (SEMs), and P values
showing the alteration in effective connectivity within regions of the left hemisphere core system
induced by all faces, emotional faces, and famous faces.
Ventral and dorsal connections between the core and the extended systems
ModelConnections from the FG to:
Note: Each model in the rows is compared with each model in the columns. Models were
arranged from high FG connectivity (left) to high STS connectivity (right). For clarity, only FG
connections are indicated. The remaining region/s are connected to the STS.
Patterns of reciprocal connectivity within the extended system
ForwardFeedback CollateralCollateral and feedback
Collateral and feedback
Note: Each model (rows) was compared with the other models (columns). Models are
depicted in Supplementary Figure 1.
Cerebral Cortex October 2007, V 17 N 10 2403
increase the effective connectivity between the IOG and the FG
and STS (Fig. 3). Faces were also observed to have a strong effect
on the influence that the FG exerted on the AMG. Increases in
effective connectivity between the FG, IFG, and OFC were
apparent at a fixed effects level but were not significant across
subjects (Table 6). Critically, valence and fame were observed to
selectively enable dissociable pathways within the network for
face perception: Viewing emotional faces produced an increase
in effective connectivity along the IOG--FG--AMG pathway,
whereas viewing famous faces increased coupling along the
IOG--FG--OFC pathway (Fig. 3 and Table 6).
In this study we employed conventional SPM with DCM to
investigate the functional connections between regions of the
cortical network that mediates face perception. This is, to our
knowledge, the first use of Bayesian model selection to explore
functional organization in a principled way. We found that in
both hemispheres the core system is functionally organized in
a hierarchical, feed-forward architecture, with the IOG exerting
influences on both the FG and STS. Moreover, the FG, but not
the STS, exerted a strong causal influence on the AMG, IFG, and
OFC. Finally, within this network, we observed content-specific
alterations in the functional coupling between visual-limbic
regions and visual-prefrontal regions in response to emotional
and famous faces, respectively.
Faces perception elicits activation within a distributed corti-
cal network (Ishai et al. 2005) that includes all regions of the
proposed core and extended systems (Haxby et al. 2000, 2002).
Within the core system, the lateral FG plays a dominant role, as
indicated by consistent and replicable patterns of activation
within this region, irrespective of face formats, tasks, and
experimental conditions (Kanwisher et al. 1997; Haxby et al.
1999; Ishai, Ungerleider, Haxby, 2000; Ishai, Ungerleider, Martin,
et al. 2000; Grill-Spector et al. 2004; Kranz and Ishai 2006).
Intriguingly, prosopagnosic patients, despite their profound
inability to recognize faces, exhibit normal patterns of activa-
tion in the FG (Rossion et al. 2003; Avidan et al. 2005),
suggesting that activation in the FG is insufficient for face
recognition, which likely depends on integration across cortical
regions. PS, a patient with bilateral and asymmetrical lesions, is
prosopagnosic despite her intact left IOG and right FG,
suggesting that activation in these 2 ventral regions of the
core system in both hemispheres is necessary for face recog-
nition (Rossion et al. 2003).
The STS, which mediates the processing of changeable
aspects of faces (Hoffman and Haxby 2000; Haxby et al. 2002;
Puce et al. 2003), is less reliably activated across subjects and
tasks (e.g., Kanwisher et al. 1997; Haxby et al. 1999; Ishai et al.
2005). We therefore predicted that during attentive viewing the
FG, and not the STS, would enable the dynamic coupling
between visual and limbic/prefrontal face-selective regions.
Our results indicate that the FG provides the major causal input
into the extended system, which processes emotional and social
aspects of face stimuli. Although the IOG was observed to
separately and directly influence both the FG and the STS, the
ventral and dorsal regions of the core system likely comprise 2
distinct pathways. As the FG exerted influences on all regions of
the extended system (AMG, IFG, and OFC), it seems that the
extraction of motile, changeable aspects of face stimuli within
limbic and prefrontal regions are enabled via the ventral visual
It has been proposed that the STS plays a major role within
a putative network for social cognition, which includes the
AMG and the OFC (Brothers 1997; Allison et al. 2000; Adolphs
2003). According to this model, the STS, and not the FG, would
exert the dominant influence on regions of the extended sys-
tem implicated in social cognition. It should be noted that the
static pictures used in our study provide an impoverished ren-
dition of the social cues present in motile faces. During the per-
ception of static faces, STS activation is thought to result from
implied, rather than overt, biological motion (Haxby et al. 2002).
We therefore predict that the STS would exert a greater effec-
tive influence on the extended system during the perception of
animated faces. Future studies would determine whether such
social cues are mediated through the FG, as suggested by our
current study, or directly from the STS to the extended system.
When compared with neutral faces, emotional and famous
faces elicit stronger activation within regions of the core and
the extended systems (e.g., Vuilleumier et al. 2001; Pessoa
et al. 2002; Ishai et al. 2004, 2005). It is believed that these
manifestations of ‘‘valence enhancement’’ reflect top-down
modulation or feedback processes. We hypothesized that
attentive viewing of emotional faces would increase the
effective connectivity between the FG and the AMG, which
mediates the perception of emotional facial expressions, in
particular fear, anger, and disgust (Breiter et al. 1996; Morris
et al. 1996; Phillips et al. 1997; Ishai et al. 2004). Moreover, we
predicted that viewing famous faces (taken from our database of
contemporary Hollywood celebrities, see Ishai et al. 2002)
would differentially increase the functional coupling between
Figure 3. Alterations in effective connectivity within the core and the extended systems induced by all faces, emotional faces, and famous faces. Black connections indicate
significant regional effects, red connections indicate significant bilinear effects, and dotted lines indicate nonsignificant effects.
Cortical Connectivity within the Face Network
Fairhall and Ishai
the FG and the OFC, a region of the reward circuitry implicated
in the processing of beautiful, attractive, and sexually relevant
faces (O’Doherty et al. 2003; Kranz and Ishai 2006; Ishai 2007).
Consistent with our hypothesis, the DCM analysis revealed
content-specific alteration in effective connectivity between
the core and the extended systems. Thus, emotional faces
increased the effective connectivity between the IOG, FG, and
the AMG, whereas famous faces increased the effective con-
nectivity between the IOG, FG, and the OFC. These results
demonstrate dynamic alterations in AMG and OFC activity,
which depend on valence and fame-related aspects of face
stimuli. Enhanced activation within the AMG and the OFC is
caused (at least in part) by a differential increase in the
influence that the FG exerts on these regions. Due to the large
regional variability in hemodynamic response latencies, axonal
conduction delays, which are typically in the order of 10--20 ms,
cannot be estimated from fMRI data (Friston et al. 2003; Stephan
et al. 2007). Future ERP or MEG studies will determine the
temporal dynamics of coupling between these regions.
The Bayesian model selection consistently revealed greater
evidence in favor of the simple, feed-forward models. The lack
of evidence in favor of models with feedback and/or lateral
connections does not reflect the absence of such anatomical
connections, but rather, implies a unidirectional modulatory
influence (see Friston et al. 2003; Penny et al. 2004a; Stephan
et al. 2004). Thus, our results do not exclude the presence of
per se but suggest that these connections play a minor role in
the observed hemodynamic response during face viewing. Our
findings are also in accord with an intracranial recordings study,
in which coupling (in the form of transient phase synchrony)
between the FG and other cortical regions was observed during
a delayed face recognition task, however, the other regions did
not exhibit coupling with each other (Klopp et al. 2000).
We used attentive viewing to model the ‘‘default’’ network of
face perception. Although face viewing elicits activation within
multiple regions of the face network, not all subjects exhibit
activation in all ROIs (e.g., Ishai et al. 2005). Previous studies
have suggested that more engaging tasks are likely to evoke
significant activation within all regions of the extended system
(Ishai, Ungerleider, Martin, et al. 2000; Ishai et al. 2004; Kranz
and Ishai 2006). The necessity to identify each and every region
in model construction renders DCM particularly susceptible to
these variations in activation, with mandatory exclusion of
subjects who did not show activation in all 6 ROIs of the face
network. These differences in cortical activation likely reflect
subject-specific variations in signal-to-noise ratio than variations
in functional architecture. Future DCM studies may overcome
these limitations by employing more cognitively demanding
tasks that would enable the identification of all ROIs in all
In sum, our DCM analysis revealed that activation in limbic
and prefrontal face-selective regions is modulated via the
ventral visual stream, where the functional coupling between
the FG and the AMG or the OFC is dynamically altered in
response to distinct facial characteristics.
We thank Karl Friston for his guidance in preparation of this
manuscript and Klaas Enno Stephan for his helpful suggestions. This
study was supported by the Swiss National Science Foundation grant
3200B0-105278 and by the Swiss National Center for Competence in
Research: Neural Plasticity and Repair.
Funding to pay the Open Access publication charges for this article
was provided by Swiss National Science Foundation grant 3200B0-
Address for correspondence Alumit Ishai, PhD, University of Zurich,
Winterthurerstrasse 190, 8057 Zurich, Switzerland. Email address:
AdolphsR. 2003. Cognitiveneuroscienceof humansocialbehaviour. Nat
Rev Neurosci. 4:165--178.
Aharon I, Etcoff N, Ariely D, Chabris CF, O’Connor E, Breiter HC. 2001.
Beautiful faces have variable reward value: fMRI and behavioral
evidence. Neuron. 32:537--551.
Allison T, Puce A, McCarthy G. 2000. Social perception from visual cues:
role of the STS region. Trends Cogn Sci. 4:267--278.
Avidan G, Hasson U, Malach R, Behrmann M. 2005. Detailed exploration
of face-related processing in congenital prosopagnosia: 2. Functional
neuroimaging findings. J Cogn Neurosci. 17:1150--1167.
Bitan T, Booth JR, Choy J, Burman DD, Gitelman DR, Mesulam MM. 2005.
Shifts of effective connectivity within a language network during
rhyming and spelling. J Neurosci. 25:5397--5403.
Breiter HC, Etcoff NL, Whalen PJ, Kennedy WA, Rauch SL, Buckner RL,
Strauss MM, Hyman SE, Rosen BR. 1996. Response and habituation of
the human amygdala during visual processing of facial expression.
Brothers L. 1997. Friday’s footprint: how society shapes the human
mind. New York: Oxford University Press.
Friston KJ, Harrison L, Penny W. 2003. Dynamic causal modelling.
Friston KJ, Holmes AP, Poline JB, Grasby PJ, Williams SC, Frackowiak RS,
Turner R. 1995. Analysis of fMRI time-series revisited. Neuroimage.
Friston KJ, Holmes AP, Worsley KJ, Poline J-P, Frith CD, Frackowiak RSJ.
1995. Statistical parametric maps in functional imaging: a general
linear approach. Hum Brain Mapp. 2:189--210.
Grill-Spector K, Knouf N, Kanwisher N. 2004. The fusiform face area
subserves face perception, not generic within-category identifica-
tion. Nat Neurosci. 7:555--562.
Haxby JV, Hoffman EA, Gobbini MI. 2000. The distributed human neural
system for face perception. Trends Cogn Sci. 4:223--233.
Haxby JV, Hoffman EA, Gobbini MI. 2002. Human neural systems for face
recognition and social communication. Biol Psychiatry. 51:59--67.
Content-specific coupling within the core and the extended systems
All faces IOGFG
Famous faces IOG
Note: Across-subjects means, standard deviations (SDs), standard error of the means (SEMs),
and P values indicate the alteration in effective connectivity induced by all faces, emotional, and
famous faces within all regions.
Cerebral Cortex October 2007, V 17 N 10 2405
Haxby JV, Ungerleider LG, Clark VP, Schouten JL, Hoffman EA, Martin A.
1999. The effect of face inversion on activity in human neural
systems for face and object perception. Neuron. 22:189--199.
Haynes JD, Driver J, Rees G. 2005. Visibility reflects dynamic changes of
effective connectivity between V1 and fusiform cortex. Neuron.
Hoffman EA, Haxby JV. 2000. Distinct representations of eye gaze and
identity in the distributed human neural system for face perception.
Nat Neurosci. 3:80--84.
Ishai A. Forthcoming 2007. Sex, beautyand the orbitofrontal cortex. Int J
Ishai A, Haxby JV, Ungerleider LG. 2002. Visual imagery of famous faces:
effects of memory and attention revealed by fMRI. Neuroimage.
Ishai A, Pessoa L, Bikle PC, Ungerleider LG. 2004. Repetition suppression
of faces is modulated by emotion. Proc Natl Acad Sci USA.
Ishai A, Schmidt CF, Boesiger P. 2005. Face perception is mediated by
a distributed cortical network. Brain Res Bull. 67:87--93.
Ishai A, Ungerleider LG, Haxby JV. 2000. Distributed neural systems for
the generation of visual images. Neuron. 28:979--990.
Ishai A, Ungerleider LG, Martin A, Haxby JV. 2000. The representation of
objects in the human occipital and temporal cortex. J Cogn Neuro-
Ishai A, Ungerleider LG, Martin A, Schouten JL, Haxby JV. 1999.
Distributed representation of objects in the human ventral visual
pathway. Proc Natl Acad Sci USA. 96:9379--9384.
Kanwisher N, McDermott J, Chun MM. 1997. The fusiform face area:
a module in human extrastriate cortex specialized for face percep-
tion. J Neurosci. 17:4302--4311.
Kass RE, Raftery AE. 1993. Bayes factors and model uncertainty.
Technical Report, University of Washington.
Klopp J, Marinkovic K, Chauvel P, Nenov V, Halgren E. 2000. Early
widespread cortical distribution of coherent fusiform face selective
activity. Hum Brain Mapp. 11:286--293.
Kranz F, Ishai A. 2006. Face perception is modulated by sexual
preference. Curr Biol. 16:63--68.
Leveroni CL, Seidenberg M, Mayer AR, Mead LA, Binder JR, Rao SM. 2000.
Neural systems underlying the recognition of familiar and newly
learned faces. J Neurosci. 20:878--886.
Mechelli A, Price CJ, Friston KJ, Ishai A. 2004. Where bottom-up meets
top-down: neuronal interactions during perception and imagery.
Cereb Cortex. 14:1256--1265.
Mechelli A, Price CJ, Noppeney U, Friston KJ. 2003. A dynamic causal
modeling study on category effects: bottom-up or top-down medi-
ation? J Cogn Neurosci. 15:925--934.
Morris JS, Frith CD, Perrett DI, Rowland D, Young AW, Calder AJ,
Dolan RJ. 1996. A differential neural response in the human
amygdala to fearful and happy facial expressions. Nature. 383:
Noppeney U, Price CJ, Penny WD, Friston KJ. 2006. Two distinct neural
O’Doherty J, Winston J, Critchley H, Perrett D, Burt DM, Dolan RJ. 2003.
Beauty in a smile: the role of medial orbitofrontal cortex in facial
attractiveness. Neuropsychologia. 41:147--155.
Penny WD, Stephan KE, Mechelli A, Friston KJ. 2004a. Modelling
functional integration: a comparison of structural equation and
dynamic causal models. Neuroimage. 23:264--274.
Penny WD, Stephan KE, Mechelli A, Friston KJ. 2004b. Comparing
dynamic causal models. Neuroimage. 22:1157--1172.
Pessoa L, McKenna M, Gutierrez E, Ungerleider LG. 2002. Neural
processing of emotional faces requires attention. Proc Natl Acad
Sci USA. 99:11458--11463.
Phillips ML, Young AW, Senior C, Brammer M, Andrew C, Calder AJ,
Bullmore ET, Perrett DI, Rowland D, Williams SC, et al. 1997. A
specific neural substrate for perceiving facial expressions of disgust.
Pruessmann KP, Weiger M, Scheidegger MB, Boesiger P. 1999. SENSE:
sensitivity encoding for fast MRI. Magn Reson Med. 42:952--962.
Puce A, Syngeniotis A, Thompson JC, Abbott DF, Wheaton KJ, Castiello
U. 2003. The human temporal lobe integrates facial form and motion:
evidence from fMRI and ERP studies. Neuroimage. 19:861--869.
Rossion B, Caldara R, Seghier M, Schuller AM, Lazeyras F, Mayer E. 2003.
A network of occipito-temporal face-sensitive areas besides the right
middle fusiform gyrus is necessary for normal face processing. Brain.
Rossion B, Dricot L, Devolder A, Bodart J-M, Crommelinck M, de Gelder
B, Zoontjes R. 2000. Hemispheric asymmetries for whole-based and
parts-based face processing in the human fusiform gyrus. J Cogn
Sergent J, Ohta S, MacDonald B. 1992. Functional neuroanatomy of face
and object processing. Brain. 115:15--36.
Smith AP, Stephan KE, Rugg MD, Dolan RJ. 2006. Task and content
modulate amygdala-hippocampal connectivity in emotional retrieval.
Stephan KE, Harrison LM, Kiebel SJ, David O, Penny WD, Friston KJ.
Forthcoming 2007. Dynamic causal models of neural system dynam-
ics: current state and future extensions. J Biosci. 31.
Stephan KE, Harrison LM, Penny WD, Friston KJ. 2004. Biophysical
models of fMRI responses. Curr Opin Neurobiol. 14:629--635.
Talairach J, Tournoux P. 1998. Co-planar stereotaxic atlas of the human
brain. New York: Thieme Medical.
Vuilleumier P, Armony JL, Driver J, Dolan RJ. 2001. Effects of attention
and emotion on face processing in the human brain: an event-related
fMRI study. Neuron. 30:829--841.
Cortical Connectivity within the Face Network
Fairhall and Ishai