a r t ic l e s

A fast pathway for fear in human amygdala

Constantino Méndez-Bértolo1,2,10, Stephan Moratti1,3,4,10, Rafael Toledano5, Fernando Lopez-Sosa1,
Roberto Martínez-Alvarez6, Yee H Mah7, Patrik Vuilleumier8, Antonio Gil-Nagel5 & Bryan A Strange1,9
A fast, subcortical pathway to the amygdala is thought to have evolved to enable rapid detection of threat. This pathway’s
existence is fundamental for understanding nonconscious emotional responses, but has been challenged as a result of a lack
of evidence for short-latency fear-related responses in primate amygdala, including humans. We recorded human intracranial
electrophysiological data and found fast amygdala responses, beginning 74-ms post-stimulus onset, to fearful, but not neutral
or happy, facial expressions. These responses had considerably shorter latency than fear responses that we observed in visual
cortex. Notably, fast amygdala responses were limited to low spatial frequency components of fearful faces, as predicted by
magnocellular inputs to amygdala. Furthermore, fast amygdala responses were not evoked by photographs of arousing scenes,
© 2016 Nature America, Inc. All rights reserved.

which is indicative of selective early reactivity to socially relevant visual information conveyed by fearful faces. These data
therefore support the existence of a phylogenetically old subcortical pathway providing fast, but coarse, threat-related signals
to human amygdala.

A classical model of emotional responses in the brain 1 holds that
the amygdala receives direct subcortical inputs through the supe-
rior colliculus and pulvinar2, which enables crude, but rapidly proc-
essed, information about fear-related cues to bypass detailed cortical
processing in visual pathways3. This ‘low-road’ model for fear process-
ing is based primarily on rodent data1. Evidence for a fast pathway
in humans has only been inferred indirectly from neuroimaging
studies in healthy individuals4,5 that used subconscious emotional
stimulus presentation during functional magnetic resonance imaging
(fMRI)6,7 and in cortically blind patients who showed preserved
processing of unseen visual fear-related stimuli8,9, possibly medi-
ated by intact fiber connections between pulvinar, superior colliculus
and amygdala10, after damage to visual occipital areas. However,
given a lack of direct electrophysiological evidence for short-latency
npg
were also implanted in visual cortical areas, thereby enabling direct
comparison of latencies to emotional versus neutral face responses
in amygdala and ventral visual cortex.
In experiment 1, intracranial event-related potentials (iERPs) were
recorded while patients judged the gender of BSF, LSF and HSF faces
displayed on-screen for 500 ms. Patients viewed 135 faces, each of
which was ‘identity unique’, that is, faces pertained to 135 different
actors, balanced for gender. The procedure was repeated after a 10-min
break (the same stimuli were presented in different random order). We
expected that fast amygdala responses would occur for phylogenetically
‘prepared’ stimuli22, such as faces, and not for more complex emotional
stimuli, such as arousing scenes. Thus, in a second experiment, we
recorded iERPs while patients viewed complex neutral and unpleasant
arousing scenes and performed indoor-outdoor judgments. Electrode

fear-related responses in human amygdala11–14, an alternative to the
low-road model suggests that some cortical regions may be equally
fast at processing fear as the amygdala15,16.
We addressed this controversy by presenting emotional (fearful,
happy) and neutral faces to patients with medication-resistant epi-
lepsy in whom stereotactic electrodes had been implanted in the amy-
gdala for pre-surgical evaluation. The faces were presented either as
normal photographs (broad spatial frequency, BSF) or were spatially
filtered such that only their low (LSF) or high (HSF) spatial frequency
components were displayed (Fig. 1). Because subcortical pathways
are thought to carry only crude (LSF) visual input to the amygdala
via magnocellular neurons17–21, we hypothesized that rapid amygdala
responses to fearful faces would be restricted to those containing LSF
information (that is, LSF and BSF faces). In some patients, electrodes
contacts in amygdala and visual areas were localized by co-register-
ing each patient’s pre-operative structural MRI with their post-opera-
tive computed tomography (CT) scan (Fig. 2a and Supplementary
Figs. 1 and 2). By recording from 19 patients with amygdala electrodes,
we found very early latency iERPs in human amygdala to fearful, but
not happy or neutral, faces that preceded the earliest iERPs observed in
face-sensitive fusiform gyrus. As predicted, the fast amygdala response
was selectively observed to the low SF component of fearful faces,
and was not evoked by complex unpleasant scenes.
RESULTS
Fast amygdala responses to fearful faces
In experiment 1, we tested 16 patients with amygdala electrodes,
of whom eight satisfied our inclusion criteria (two right, four left

1Laboratory for Clinical Neuroscience, Centre for Biomedical Technology, Technical University of Madrid, Madrid, Spain. 2CEI Campus Moncloa, UCM-UPM, Madrid,

Spain. 3Department of Basic Psychology I, Complutense University of Madrid, Madrid, Spain. 4Laboratory for Cognitive and Computational Neuroscience, Centre
for Biomedical Technology, Technical University of Madrid, Madrid, Spain. 5Epilepsy Unit, Department of Neurology, Hospital Ruber Internacional, Madrid, Spain.
6Department of Neurosurgery, Hospital Ruber Internacional, Madrid, Spain. 7Neuroscience Research Centre, Cardiovascular and Cell Sciences Institute, St. George’s,

University of London, London, UK. 8Laboratory for Neurology and Imaging of Cognition, Department of Neuroscience and Neurology, University Hospital and Medical
School, University of Geneva, Geneva, Switzerland. 9Department of Neuroimaging, Reina Sofia Centre for Alzheimer’s Research, Madrid, Spain. 10These authors
contributed equally to this work. Correspondence should be addressed to B.A.S. (bryan.strange@upm.es).

Received 3 March; accepted 12 May; published online 13 June 2016; doi:10.1038/nn.4324

nature NEUROSCIENCE  VOLUME 19 | NUMBER 8 | AUGUST 2016 1041

 a r t ic l e s
Fearful Happy Neutral
BSF
HSF
LSF
Figure 1  Experimental face stimuli. Examples of BSF, LSF and HSF faces
with fearful, happy and neutral expressions presented in experiment 1.
Note that each stimulus was identity unique (that is, a different actor for
each of the 135 faces presented).
© 2016 Nature America, Inc. All rights reserved.
and two bilateral; Table 1, Fig. 2a,b and Supplementary Table 1).
Providing direct support for the low-road model of fear detection, our
amygdala recordings revealed a prominent fast deflection for fearful
faces that was different from that for both happy and neutral faces
as early as 74 ms post-stimulus (Fig. 2c). Cluster-based permutation
testing23 using multivariate analysis of variance (MANOVA) with
within-subject factors of emotion (fearful, happy, neutral) and spatial
frequency (SF) (broadband, HSF, LSF) was applied to all post-stimulus
time points. A time cluster expressing a significant emotion by SF
interaction was observed between 76–110 ms after face presentation
(at cluster threshold P < 0.01; summed F value = 455.89, P = 0.001),
which extended between 74–122 ms at a more liberal cluster threshold
of P < 0.05 (Fig. 2c and Supplementary Table 2).
Given our prior hypothesis that fast, fear-specific responses in
human amygdala arise via a magnocellular pathway, planned com-
parisons on iERP amplitudes across the interaction time cluster were
performed to directly test for fear-specific early responses restricted
npg
to BSF and LSF faces. Thus, a set of four linear contrasts were com-
puted, and ensuing P values were corrected for false discovery rate
(Pfdr) for these four tests. Testing responses to fearful versus happy
and neutral faces (fearful ≠ (happy = neutral)) at each SF revealed a
significant effect for BSF and LSF, but not HSF, faces for the 76–110-
ms window (BSF: F1,9 = 12.22, P = 0.007, Pfdr = 0.016; LSF: F1,9 = 9.55,
P = 0.013, Pfdr = 0.017; HSF: F1,9 = 2.72, P = 0.133, Pfdr = 0.133).
Responses to BSF and LSF fearful faces were therefore significantly
different from responses to neutral and happy faces in this time win-
dow (Fig. 2d), with this negative deflection to BSF and LSF fearful
faces being observed in eight of the ten amygdalae recorded (Fig. 2e
and Supplementary Table 3). Furthermore, short-latency amygdala
responses to BSF and LSF fearful faces were significantly different to
those elicited by HSF fearful faces (fearful HSF ≠ (fearful BSF = fearful
LSF)) (F1,9 = 11.51, P = 0.008, Pfdr = 0.016). Thus, amygdala responses
to BSF and LSF fearful faces did not differ, but were both significantly
different from HSF fearful face responses (post hoc t tests compar-
ing iERP amplitudes across the interaction time cluster; Fig. 2f and
Supplementary Table 4). Altogether, these data indicate a selective
fast amygdala response to the LSF components of fearful faces.
To determine the onset of deflections to each face stimulus type
individually, we applied cluster-based permutation statistics to test for
1042 VOLUME 19 | NUMBER 8 | AUGUST 2016  nature NEUROSCIENCE
the onset of iERPs that were significantly different from zero for each
stimulus type separately for the entire post-stimulus period. Notably,
significant fast-latency effects, beginning ~70 ms post-stimulus presen­
tation, were limited to BSF and LSF fearful face responses (for BSF
fearful faces, from 72–118 ms at cluster threshold P < 0.01, summed
t value = −73.20; P = 0.023, and from 68–194 ms at cluster threshold
P < 0.05; for LSF fearful faces from 72–146 ms at cluster threshold
P < 0.01, summed t value = −124.82; P = 0.013, and from 66–192 ms
at cluster threshold P < 0.05; Fig. 2g and Supplementary Table 5).
Next, given that we recorded from four right and six left hemi-
sphere amygdalae, we separated iERPs as a function of hemisphere
(Supplementary Fig. 3). A fast deflection is seen for fearful faces in
both left and right amygdala. To test for a laterality effect on the time
window expressing an emotion by SF interaction (76–110 ms), we
entered mean amplitude differences across this time window for each
emotion contrast, collapsing across BSFs and LSFs, into a Kruskal-
Wallis test comparing right versus left amygdalae. Although the mag-
nitude of the fast fear response appeared larger for right amygdala
(Supplementary Fig. 3), consistent with right-lateralized amygdala
responses to fearful faces in cortically blind patients8, we observed no
significant laterality effects (right versus left amygdalae tests for fearful
minus neutral χ2(1) = 2.23, P = 0.136; fearful minus happy χ2(1) = 0.73,
P = 0.394; happy minus neutral χ2(1) = 0.05, P = 0.831).
Later latency amygdala responses to emotional faces
The early effects that we observed contrast with previous reports
of rather late (200 ms) intracranially recorded human amygdala
responses to fearful faces12 or complex emotional scenes11. Here, the
emotion (fearful, happy, neutral) by SF (BSF, HSF and LSF) cluster-
based permutation MANOVA test with a less conservative cluster
threshold of P < 0.05 also revealed a relatively early main effect of
emotion in two time clusters: between 106–144 and 148–204 ms, and
282–302 ms. At cluster threshold P < 0.01, only two clusters (between
158–162 and 184–200 ms) showed a main effect of emotion (Fig. 2c
and Supplementary Table 2). Post hoc t tests (Supplementary Table 6)
revealed that, irrespective of frequency, in clusters prior to 200 ms,
responses to fearful faces were greater than those to both neutral and
happy faces, but, in the late time window (288–302 ms), fearful face
responses were only greater than those to neutral faces.
Effect of repetition on amygdala responses
Previous iERP recordings have shown habituation of amygdala
responses with multiple repetitions of emotional face presenta-
tion12, which may, in those recordings, have obscured a fast amy-
gdala response. To safeguard against habituation effects, we presented
each patient with 135 identity-unique faces that were repeated only
once in a second block. Nonetheless, to specifically verify repetition
effects on amygdala iERPs in our procedure, we performed an addi-
tional MANOVA for the mean amplitude across the early interaction
time window (76–110 ms), including a within-subject factor ‘block’.
We observed no emotion by SF by block interaction (F4,6 = 0.71,
P = 0.615) for responses in the early latency interaction window,
or a block by emotion (F2,8 = 0.16, P = 0.853) or frequency by block
(F2,8 = 1.90, P = 0.211) interaction. Only an emotion by frequency
interaction (F4,6 = 19.15, P = 0.001) was significant (equivalent
tests for later time windows expressing main effects of emotion and
frequency are given in Supplementary Table 7).
Fearful face responses are localized to lateral amygdala
The amygdala comprises several subnuclei, and the lateral component
is considered to be the sensory gateway to the amygdala1. Our iERP
 a r t ic l e s

Figure 2  Fast-latency human amygdala a L R

b L R
responses to BSF and LSF fearful faces. Patient 03 (R)
Anterior
(a,b) Schematic summary of electrode contact Patient 04 (L)
localizations in left and right amygdala for all Patient 05 (L)
Patient 06 (B)
patients included in experiment 1 are displayed Patient 10 (R)
(color coded) on serial coronal sections of the Patient 15 (L) Posterior
MNI canonical T1 image (anterior-posterior y=0
Patient 16 (B)
Patient 25 (L) Lateral Medial Medial Lateral
position indicated by y coordinates; a) and
semi-transparent segmented amygdalae c 40
(b) as viewed from above. (c) Amygdala Fearful
Happy
iERPs from ten amygdalae of eight patients y = –2 20 Neutral
(total of 26 contacts) to fearful, happy and

Amplitude (µV)
neutral faces collapsed over spatial frequencies
0
and averaged. Horizontal bars below depict
y = –4
time clusters expressing a significant main
effect of emotion (top) or SF (middle), or a –20 Emotion
significant interaction between both (bottom) Frequency
Interaction
using a cluster threshold of P < 0.05 (gray) y = –6 –40
–100 0 100 200 300 400 500 600
or P < 0.01 (black) on cluster-based non-
parametric permutation statistic, based on d 40
Fearful
e 60
MANOVA F values (Supplementary Table 2). Happy
An emotion by spatial frequency interaction 20 Neutral
Amplitude (µV)

was observed from 74–122 ms post-stimulus

Amplitude (µV)
onset (at cluster threshold P < 0.05; from 0
© 2016 Nature America, Inc. All rights reserved.

76–110 ms at cluster threshold P < 0.01).

0
A significant main effect of emotion was
–20
observed in time clusters from 106–144 ms, Time (ms)
148–204 ms and 282–302 ms, and main effect
–40
of spatial frequency from 308–334 ms (all at –100 0 100 200 300 400 500 600 –30 76–110 ms 282–302 ms
cluster threshold P < 0.05). At cluster
threshold P < 0.01, a significant main effect f 40
BSF fearful g 0 100 200 300 400 500 600
HSF fearful Time (ms)
of emotion is observed during two clusters 20 LSF fearful BSF fearful
at 158–162 ms and 184–200 ms. Shaded
Amplitude (µV)

HSF fearful
area indicates s.e.m. (d) Amygdala iERPs to LSF fearful
0 BSF happy
emotional and neutral faces, collapsed over
HSF happy
BSF and LSF only. (e) Dot plots of responses to
LSF happy
fearful faces, collapsed over BSF and LSF, of –20
BSF neutral
the ten individual amygdalae during the early HSF neutral
interaction (left) and the late main effect of –40 LSF neutral
–100 0 100 200 300 400 500 600
emotion (right) time windows (Supplementary
Table 3). Each amygdala is represented by a triangle, with orientation (left or right pointing) indicating left or right amygdala, respectively. Color coding
is depicted as in a; note that patients 6 and 16 were implanted bilaterally. Dot plots are superimposed on mean activity across all amygdalae (error bars
indicate s.e.m.). (f) Amygdala iERPs to fearful BSF, LSF and HSF faces. (g) Horizontal bars depict time windows in which responses to each stimulus
type are significantly different from zero with a cluster threshold of P < 0.05 (gray) or P < 0.01 (black) on cluster-based non-parametric permutation
npg

statistic, based on one-sample t test (Supplementary Table 5).

analyses collapsed over contacts that were anatomically limited to
the amygdala, but spanned its medial-lateral extent (Fig. 2a,b). To
refine the anatomical origin of the amygdala iERPs that we observed,
we next performed a source localization analysis. For each patient,
the minimum norm was applied to average responses to BSF and
LSF fearful faces from all contacts (in and outside the amygdala) of
each amygdala electrode to obtain mean current source densities for
the time windows of the early interaction (76–110 ms) and late main
effect of emotion (282–302 ms) (Fig. 3a). For patient 10, source locali-
zation was not possible, as seven of ten channels displayed a flat signal.
An overlap of thresholded source localizations (50% of peak activity)
of the remaining patients revealed consistent localization of responses
to lateral amygdala during the early interaction time window (eight
of nine amygdala responses; Fig. 3b). Source estimation was less
consistent for the late positive-going main effect of emotion (six of
nine amygdala responses; Fig. 3b). However, both source localiza-
tions of time intervals for early negative-going and late positive-going
waveforms overlapped substantially in lateral amygdala (Fig. 3c). The
latter observation is interesting, as the polarity inversion between
early and late responses may indicate recruitment of different
neuronal populations. However, spatially overlapping source localized
early and late fear responses in lateral amygdala could instead be
consistent with different afferent inputs in the same neural popula-
tion. Also note that including responses from all amygdala electrode
contacts (both in and outside the amygdala), and demonstrating
current source in lateral amygdala, eschews the possibility that iERPs
are conducted from brain areas outside of the amygdala.
Emotional face responses in ventral visual cortex
The demonstration of SF-selective, fast-latency amygdala responses
to fearful faces does not exclude the possibility that, for the same
stimuli, similar latency responses are present in visual cortical areas
that provide input to the amygdala. To address this, we analyzed corti-
cal responses in patients who also had electrode contacts in ventral
visual areas. Seven of the patients that completed experiment 1 had
undergone electrode implantation in both amygdala and ventral visual
cortex, and met all inclusion criteria. The primate amygdala receives
extensive input from uni- and multi-modal areas of the temporal
lobe24–26, but there is no evidence for amygdala afferents from the
occipital lobe24,25. We therefore focused our analyses on five patients
with electrodes with contacts in fusiform gyrus (Fig. 4a), a face-
sensitive region of the inferior temporal lobe27. Two of these patients

nature NEUROSCIENCE  VOLUME 19 | NUMBER 8 | AUGUST 2016 1043

 npg © 2016 Nature America, Inc. All rights reserved.

Table 1  Patient demographic and clinical data

1044
Age at Percentage of Percentage of
onset Seizure trials with epileptic trials with epileptic
Experi- Hand- Age of epilepsy type (frequency Drugs and Education spikes in amygdala spikes in amygdala
Patient ment Sex edness (years) (years) Aetiology Lesion location per months) dose (mg) VIQ PIQ completed (experiment 1) (experiment 2)
02 2 F R 38 21 Hippocampal sclerosis Right hippocampal sclerosis plus porencephalic Weekly CPS LCS 400 103 91 Tertiary NA 0.3%
a r t ic l e s

plus focal dysplasia cyst over the parieto-occipital junction LEV 3,000
03 1 M R 55 21 Focal dysplasia Right temporal neocortex Monthly CPS PHT 200 132 125 Tertiary 10.6% NA
04 1 and 2 F R 21 12 Focal dysplasia Extensive lesion over the left frontal region Daily CPS CBZ 600 86 80 Tertiary 0.7% 21.7%
involving dorsolateral and orbitofrontal cortex Daily SPS CNZ 3.5
and anterior order of the cingulum LEV 1,500
TOP 400
05 1 and 2 M R 29 16 Periventricular Bilateral occipital horn heterotopia plus focal Weekly CPS CBZ 1,000 64 73 Secondary 2.6% 0.3%
heterotopia plus focal dysplasia over the left occipital cortex Daily SPS TOP 400
cortical dysplasia
06 1 and 2 M R 49 14 Hippocampal sclerosis Left hippocampal sclerosis Weekly CPS LGT 200 115 86 Tertiary Left 0% Left 0%
Monthly SG TCS CLB 30 Right 0% Right 0%
LCS 300
08 2 F R 29 8 months Focal dysplasia Extensive right posterior dysplasia involving Monthly CPS LEV 2,000 67 71 Secondary NA 10.6%
the convexity and medial aspect of parietal, LTG 500
occipital and posterior temporal lobes
10 1 M R 59 26 Encephalocele plus Right anterior temporal pole and basal area Weekly CPS PGL 375 105 102 Tertiary 3% NA
focal cortical dysplasia Encephalocele confirmed at surgery Monthly SG TCS ECZ 1,600
13 2 F R 42 16 Focal dysplasia Right basal temporal cortex Monthly CPS PGL 450 87 88 Secondary NA 10.6%
LCS 400
15 1 and 2 F R 35 16 Focal dysplasia Left temporal pole Daily CPS OXC 1,200 97 98 Tertiary 0.4% 0%
Weekly SG TCS PHT 150
16 1 and 2 M R 30 14 Reactive gliosis, Medial wall of the left parietal region Daily CPS LCS 400 102 93 Tertiary Left 1.5% Left 0.8%
diffuse microglia (precuneus and posterior cingulum) 4 SG TCS yearly ECZ 800 Right 1.1% Right 0%
activation and small LTG 300
vessel vasculopathy VAL 300
CLB 20
ESC 20
25 1 and 2 M R 29 13 Focal dysplasia Right posterior temporobasal region Weekly CPS CBZ 1,200 116 97 Tertiary 6.7% 5.8%
Monthly SG TCS LEV 1,000
LTG 200
ESC 10
CBZ, carbamazepine; CLB, clobazam; CNZ, clonazepam; CPS, complex partial seizure; ECZ, eslicarbazepine; ESC, escitalopram; LCS, lacosamide; LEV, levetiracetam; LTG, lamotrigine; NA, Not applicable; OXC, oxcarbazepine;
PGL, pregabalin; PHT, phenytoin; PIQ/VIQ, Procedural/Verbal Intelligence Quotient of the Wechsler Adult Intelligence Scale, version III (WAIS-III); SG TCS, secondary generalized tonic clonic seizure; SPS, simple partial seizure;
TOP, topiramate; VAL, valproate.

VOLUME 19 | NUMBER 8 | AUGUST 2016  nature NEUROSCIENCE

 a r t ic l e s

a b c
Patient 16 76–110 ms 282–302 ms
y = –24 y = –25 y = –25

L R

pAm N 76–110 ms 282–302 ms

50 100 150 200 5 6 7 8

Figure 3  Amygdala iERPs to fearful faces originate from lateral amygdala. (a) Source localization of the response to BSF and LSF fearful faces in
the early time window (76–110 ms) from a representative amygdala (left amygdala of patient 16, 11 contacts). The current source map (thresholded
at 50% of maximum activity) is overlaid on a coronal section of the pre-operative T1 MRI from patient 16. The amygdala electrode is illustrated
schematically (white stripes represent amygdala contacts used for source localization; the black stripe is a bad channel with no signal). The color
bar indicates estimated current source strength in picoamperes (pA). (b,c) Group source localization maps for seven patients with nine amygdala
electrodes (total of 56 contacts in and outside the amygdala). (b) Co-incidence maps of voxels that show the number of source localization cases (N),
normalized to MNI space, that reached 50% of maximum activity, overlaid on two coronal sections of the canonical MNI (anterior-posterior locus
of each section is given by the y coordinate above). The color bar indicates the number of overlapping cases for the early interaction (left section,
MNI coordinates of: x = –26; z = –22; y = –4) and the later (282–302) main effect of emotion (right section, peak overlap: x = –26; z = –26;
y = –4) time windows. Note that right amygdalae source solutions were flipped to be overlaid onto the left amygdala for representation purposes.
© 2016 Nature America, Inc. All rights reserved.

(c) The co-incidence maps for both time windows in b have been thresholded at N ≥ 6 and superimposed to show that they overlap in lateral inferior
amygdala. The inset depicts a magnified view of the amygdala region.

had two different electrodes in fusiform gyrus, yielding seven groups
of fusiform contacts. In a first analysis, we determined an index of
face-selectivity for these fusiform contacts by comparing responses
to broadband neutral faces relative to neutral scenes (presented in
experiment 2, described below). One fusiform contact group was
excluded from this analysis because of a signal artifact in experiment 2
(Fig. 4a); thus, iERPs from six contact groups were compared at the
group level using cluster-based permutation analysis. This analysis,
comparing all post-stimulus time points, revealed a significant cluster
between 164–176 ms (summed t values = −32.07, P = 0.03), suggest-
ing a selective fusiform response to faces versus scenes at a latency
consistent with the face-sensitive N170 potential28 (Fig. 4b).
We next examined the interaction between emotional facial expres-
sion and SF in the iERPs recorded from the seven groups of fusiform
contacts during experiment 1. Group-averaged fusiform responses to
npg
P < 0.05), earlier than previously reported iERP latencies to fearful
versus neutral faces recorded in ventral visual cortex (~300 ms) 12,
but, notably, 100 ms later than the effects observed in the amygdala.
No earlier emotion or interaction effects were observed at either clus-
ter threshold (Supplementary Table 8). We observed main effects
of SF (two clusters spanning 202–284 ms at both cluster thresholds,
and a separate cluster from 170–186 at cluster threshold P < 0.05)
followed by an emotion by SF interaction around 300 ms at both
cluster thresholds (Supplementary Tables 8 and 10).
Recording from both amygdala and face-sensitive fusiform cor-
tex enabled a comparison of fear-evoked responses between these
areas. Thus, in the four patients with electrodes in both amygdala
and fusiform cortex, we compared emotional face responses between
these two areas in the early emotion by SF interaction time window.
The mean amplitude difference between emotions (fear versus happy,

face stimuli (Fig. 4c) were characterized by a positive peak at ~120 ms
post-stimulus onset, followed by a negative deflection peaking at 170 ms
and then a slower positive component. Cluster-based permuta-
tion testing employing an emotion (fearful, happy, neutral) by SF
(BSF, HSF, LSF) MANOVA showed only a main effect of frequency
from 206–240 ms at cluster threshold P < 0.01 (Supplementary
Table 8). Post hoc tests in this time window revealed that, although
responses to broad and LSF faces were greater than to HSF faces,
broad and LSF responses did not differ (Supplementary Table 9),
reflecting later latency negative deflections to HSF faces relative to
BSF and LSF (Fig. 4d). In contrast with effects observed in the amy-
gdala, we did not observe a significant interaction at early latencies,
even after relaxing the cluster threshold to P < 0.05 (Supplementary
Table 8). At this lower threshold, the earliest observed effects were
a main effect of frequency from 204–258 ms followed by an emo-
tion by SF interaction from 290–334 ms (Supplementary Table 9).
To our surprise, no significant main effect of emotion was obtained.
However, given that enhanced fusiform responses were observed less
frequently to happy than to fearful faces29, we repeated this analysis,
comparing only fearful and neutral stimuli (Fig. 4e). This analysis
yielded a significant main effect of emotion in a 172–218 ms time
cluster (summed F values, 229.44; P < 0.001, at cluster threshold
fear versus neutral, and happy versus neutral) across the 76–110-ms
window, collapsing over BSF and LSF, was calculated for both the
amygdala and fusiform gyrus, separately. These six difference val-
ues (three emotion differences for amygdala and fusiform contacts,
respectively) for each of the four patients were entered into a Friedman
test. This test showed a significant interaction between emotion and
brain region (χ2(5) =12.71, P = 0.026). Further Friedman tests for each
region separately revealed a significant emotion effect in amygdala
(χ2(2) = 8, P = 0.018) but not in fusiform (χ2(2) = 2, P = 0.368).
It remains possible, however, that fast afferent input from fusiform
cortex arrives at the amygdala before a differential response to fear-
ful versus neutral or happy faces is observed in the fusiform, that
is, emotion selectivity arises in the amygdala, but still depends on
up-stream fusiform activity. Given that fast amygdala responses are
observed to BSF and LSF fearful faces, we next tested for the earliest
onset of any upward or downward deflection in fusiform contacts to
BSF and LSF fearful faces, adopting a liberal statistical approach: we
employed a one-tailed t test against zero for each time point of the
entire post-stimulus period, applying an uncorrected alpha level of
0.05 (Fig. 4f). The first significant deflection for BSF fearful faces
was negative and spanned 174–192 ms (174 ms: t6 = −2.53, P = 0.044
uncorrected); for LSF fearful faces there was a significant positive

nature NEUROSCIENCE  VOLUME 19 | NUMBER 8 | AUGUST 2016 1045

 a r t ic l e s

a Fus1 Patient 13 b 80
Faces
Scenes
c 80
Fearful
Happy
Fus2 Patient 15 Neutral

Amplitude (µV)
40

Amplitude (µV)
40
Fus3 Patient 10
Fus4 Patient 03 0 0

Fus5 Patient 13 –40 –40

Fus6 Patient 05* Emotion
–80 Time (ms) –80 Frequency
R L Fus7 Patient 03 Interaction
–100 0 100 200 300 400 500 600 –100 0 100 200 300 400 500 600

d 80
BSF e 80 Fearful f 80
BSF fearful
HSF fearful
HSF Neutral
LSF LSF fearful
Amplitude (µV)

Amplitude (µV)

Amplitude (µV)
40 40 40

0 0 0

–40 –40 –40

Emotion
–80 –80 Frequency –80
Interaction
–100 0 100 200 300 400 500 600 –100 0 100 200 300 400 500 600 –100 0 100 200 300 400 500 600

g
Figure 4  Fast responses to fear are not observed in fusiform gyrus. (a) Summary illustration of contact sites in the fusiform 80
gyrus, depicted as colored shapes, and superimposed on a semi-transparent MNI template brain. Different groups of contacts

Amplitude (µV)
40
from the same patient are displayed in the same color. The asterisk indicates that the contact group was excluded from the
© 2016 Nature America, Inc. All rights reserved.

comparison of broadband neutral faces versus neutral scenes as a result of a signal artifact in experiment 2. (b) Averaged 0
iERPs to BSF neutral faces (solid line) and neutral complex scenes (dashed line) from six fusiform contact groups (four
–40
patients, total of 11 contacts). Responses to faces were significantly greater than those to scenes between 164–176 ms
(cluster threshold P < 0.01) on cluster-based non-parametric permutation statistic, based on two-sample t test. Horizontal –80
bars indicate time clusters expressing significant effects, as described in Figure 2. Shaded area indicates s.e.m. 72–108 ms
(c–f) Fusiform gyrus iERPs from five patients and seven contact groups (total of 13 contacts) to fearful, happy and neutral
faces, collapsed across frequencies (experiment 1; c). No significant main effect of emotion was observed on cluster-based non-parametric
permutation statistic, based on MANOVA F values (Supplementary Table 8). There was a main effect of spatial frequency from 206–240 ms
(cluster threshold P < 0.01) and an emotion by spatial frequency interaction from 290–334 ms (cluster threshold P < 0.05). (d) Fusiform iERPs
to BSF, HSF and LSF faces, collapsed across emotion. (e) Fusiform gyrus iERPs to only fearful and neutral faces, collapsed across spatial frequencies.
A main effect of emotion was significant from 172–218 ms, as well as a main effect of frequency from 170–186 ms (both at cluster threshold
P < 0.05). Significant main effects of frequency between 200 and 300 ms, as well as an emotion (fear, neutral) by spatial frequency (BSF, HSF and
LSF) interaction around 300 ms, were also observed (Supplementary Table 8). (f) Fusiform iERPs to BSF, HSF and LSF fearful faces only. (g) Dot plot
of responses observed in each of the seven fusiform contact sites to fearful faces, collapsed over BSF and LSF, during the early interaction time window
derived from the amygdala iERP statistical analysis (Supplementary Table 11). Colored markers are depicted as in a. Dot plots are superimposed on the
mean (±s.e.m.) activity across all fusiform contact sites.

deflection from 104–124 ms (104 ms: t6 = 3.68, P = 0.010 uncor- right, four left, two bilateral; Table 1, Fig. 5a,b and Supplementary
rected) and a negative deflection from 178–204 (178 ms: t6 = −2.57, Table 12). Six of these nine patients also completed experiments 1.
P = 0.042 uncorrected). Thus, even applying liberal statistical criteria, Notably, we found that the earliest latency at which iERPs evoked
npg

the earliest deflection in fusiform cortex to BSF and LSF fearful faces
(104 ms) occurred more than 30 ms after the onset of SF-dependent
amygdala responses to fearful faces observed following cluster-based
correction. Lastly, to control for the fact that the polarity of evoked
potentials is unpredictable and can vary across individual contact
sites, we tested absolute values of each post-stimulus time point of
fusiform responses against zero in a one-tailed t test, again at uncor-
rected alpha of 0.05. The first significant effect for LSF fearful faces is
again at 104 ms and for BSF fearful faces is at 108 ms. We also plotted
the average response amplitude to fearful faces, collapsed over BSF
and LSF, of all seven fusiform contact sites during the early amygdala
interaction time window (Fig. 4g and Supplementary Table 11),
which show minimal deviations from zero. The absence of fast
responses in ventral stream also safeguards against the possibility that
short-latency effects in the amygdala are a result of activity derived
from the common reference electrodes.
Emotional scenes do not evoke fast amygdala responses
In experiment 2, amygdala iERPs were recorded during presentation of
neutral and unpleasant complex pictures, such as household scenes and
mutilated bodies, respectively. The task was completed by 13 patients
with amygdala electrodes, of whom 9 satisfied inclusion criteria (three
by unpleasant and neutral pictures differed significantly was from
~190 ms post-stimulus onset. Cluster-based permutations compar-
ing emotional and neutral pictures revealed two significant clusters
(Fig. 5c and Supplementary Table 13), which, on relaxing the cluster
threshold to P < 0.05, collapsed into one cluster spanning 192–280 ms
(Fig. 5c and Supplementary Table 13). To calculate latencies at
which iERPs to each picture type differed significantly from zero,
we employed the same procedure as in experiment 1. With a cluster
threshold of P < 0.01, we observed significant differences from zero
baseline from 186 ms and 226 ms for unpleasant and neutral pictures,
respectively (Fig. 5d and Supplementary Table 14), which, on relax-
ing the cluster threshold to P < 0.05, began at 162 ms and 202 ms.
Thus, no differential response to emotional versus neutral scenes
was observed in the early time window in which fast responses to BSF
and LSF fearful faces occurred in experiment 1. To formally test for a
difference between fast amygdala responses to fearful faces and emo-
tional scenes, we compared mean amplitudes of iERPs from experi-
ment 1 versus 2 in the time cluster exhibiting an early facial emotion
by frequency interaction (76–110 ms; Supplementary Table 15). That
is, we tested for a difference between early responses to fearful relative
to neutral faces versus the response in the same early time window for
emotional relative to neutral scenes (for these analyses we collapsed

1046 VOLUME 19 | NUMBER 8 | AUGUST 2016  nature NEUROSCIENCE


Figure 5  Fast-latency amygdala a L
responses are not evoked by complex
emotional pictures. (a,b) Schematic
summary of electrode contact localizations
in left and right amygdala for all patients
included in experiment 2, as described
in Figure 2. (c) Human amygdala iERPs y = –2
from nine patients (11 amygdalae with
a total of 32 contacts) to unpleasant
pictures (experiment 2) were significantly
different from those evoked by neutral y = –4
pictures only from 192 ms to 280 ms
(at cluster threshold P < 0.05), encompassing
two smaller clusters from 224 ms to 252 ms
y = –6
and from 258 ms to 278 ms, resulting from
a cluster threshold P < 0.01, on cluster-based
non-parametric permutation statistic, based
on paired-sample t test (Supplementary y = –8
Table 13). Shaded area indicates s.e.m.
(d) Horizontal bars depict time windows in
which responses to unpleasant (top) and
y = –10
neutral (bottom) pictures were significantly
different from zero, on cluster-based non-parametric permutation statistic, based on one-sample t test (Supplementary Table 14). Different cluster
thresholds (P < 0.05 and P < 0.01) are coded as in Figure 2.
© 2016 Nature America, Inc. All rights reserved.
BSF and LSF face trials). In a first analysis, we entered normalized
mean amplitudes from all amygdalae in both experiments (10 from
experiment 1 and 11 experiment 2) into a repeated-measures ANOVA
with within-subject factor emotion (negative, neutral) and between-
subject factor experiment (experiments 1 and 2). This revealed a sig-
nificant emotion by experiment interaction (F1,19 = 23.54, P = 0.001).
Next, we restricted our sample to the six patients (eight amygdalae)
completing both tasks, and performed an ANOVA with within-
subject factors emotion and task, which revealed a significant emotion
by task interaction (F1,7 = 16.96, P = 0.004).
DISCUSSION
The existence of a subcortical route to the amygdala for rapid process-
ing of ecologically important stimuli has markedly influenced basic
and clinical research on emotional processing in the brain. However,
one important limitation of this low-road model1 has been an absence
of support from direct electrophysiological recordings in primates. An
npg
alternative account suggests that rapid visual processing of emotional
stimuli can be mediated by other visual pathways involving visual
cortex15. Our results provide direct empirical support for the low-road
model, as we found that human amygdala intracranial responses to
fearful, but not neutral or happy, faces at very fast latency (~70 ms)
were SF dependent. This effect was localized to lateral amygdala,
where sensory inputs from thalamo-amygdala and cortico-amygdala
pathways converge1. We found that this selective amygdala response
preceded any evoked activity in face-sensitive ventral visual cortex to
the same stimuli by more than 30 ms, and preceded the onset of a dif-
ferential response to fearful faces in ventral visual cortex by ~100 ms.
Our findings are therefore consistent with a bottom-up amygdala
response originating via a more direct subcortical magnocellular route
rather than top-down influences from higher level visual processing
stages. In contrast, the later latency main effect of emotion in the
amygdala (beginning from ~120 ms) was more consistent with emo-
tional information that arrives at the amygdala having been processed
in visual cortex. In support of this interpretation, this later response
was evoked by HSF fearful faces (Fig. 2f,g), which also modulated
fusiform cortex activity (Fig. 4f), indicating engagement of slower
parvocellular pathways along visual cortical areas30,31. Note that more
pronounced later latency (>200 ms) effects have been reported with
nature NEUROSCIENCE  VOLUME 19 | NUMBER 8 | AUGUST 2016 1047
a r t ic l e s
R b L R
Patient 02 (L) Anterior
Patient 04 (L)
Patient 05 (L)
Patient 06 (B)
Patient 08 (R)
Patient 13 (R)
Patient 15 (L) Posterior
Patient 16 (B)
Patient 25 (L) Lateral Medial Medial Lateral
c Unpleasant
60 Neutral
Amplitude (µV)
40
20
0
Time (ms)
–20
–100 0 100 200 300 400 500 600
d 0 100 200 300 400 500 600
Unpleasant
Neutral
tasks requiring attention to facial emotional expression12,32, but not
when patients attended to gender12, as we required in our task.
Evidence for a fast, magnocellular pathway in processing faces in non-
human primates is provided by monkey pulvinar recordings showing
50-ms latency responses to face-like stimuli, including cartoon faces33.
By contrast, relatively few intracranial recording studies in non-human
primates have compared amygdala responses to static images of threat-
related faces to non-threatening expressions. One study34 reported
increased neuronal firing to threatening faces at 120–250 ms post-
stimulus presentation, which shows homology with the main effect
of emotion that we observed in two time clusters between 118–204
ms. We could not, however, evaluate fast-latency responses to threat
because the first 100 ms after stimulus-image onset were excluded from
analyses34. Similarly, in a study reporting human single-unit amygdala
recordings32, emotion-selective units were selected in a time window
250–1,750 ms post-stimulus-onset, rendering analyses agnostic to
fearful face responses at fast latencies (before 250 ms).
The few prior studies reporting field potentials12,14 or oscillatory
responses11,35 from human amygdala depth recordings failed to find
the fast-latency amygdala response that we observed. Although two
previous studies35,36 also presented fearful faces, they found responses
at ~130 ms (approximating the latency of the main effect of emotion
that we observed in the amygdala) and used only standard broadband
photographs. In another study reporting late (200 ms) iERPs to
fearful faces12, a limited number of identities (eight) were each pre-
sented 30 times in each of two tasks. Human electrocorticographic
(ECoG) recordings demonstrate item-specific repetition suppres-
sion of electrophysiological responses37; thus, multiple repetitions
of the same stimuli may have resulted in habituation of an amygdala
fast response in this previous study12. To eschew this possibility, we
presented patients with 135 identity-unique facial expressions, and
did not observe habituation following a single repetition. A further
factor likely to have improved our ability to detect fast amygdala
responses is that, in seven of eight patients included in experiment 1,
pathology was outside of the medial temporal lobe entered into our
analyses (Table 1), consistent with preserved amygdala function.
Lastly, given that we localized fast responses to lateral amygdala, stud-
ies with recording sites more medial than those described here32 may
be unable to detect fast-latency responses.
 a r t ic l e s
The failure of previous human intracranial studies to find a rapid
amygdala response to complex emotional pictures11,14 may reflect a
fundamentally faster processing time for fearful faces, which have
important motivational and communicative value among primates38,
relative to emotional scenes with multiple objects. Using complex
emotional scenes, we also found that only late-latency human
amygdala iERP amplitudes were modulated by emotion. The fact
that fast responses were limited to fearful faces provides support
for evolutionary theories of amygdala automaticity to social threat
signals39,40. From an evolutionary perspective, stimuli associated
with recurrent survival threats, such as fearful faces, require mini-
mal neural processing for identification, a notion referred to as
‘preparedness’22. Moreover, the emergence of social communities and
social signals of emotions during evolution presumably contributed to
making amygdala-centered circuits particularly responsive to threat
cues communicated by other conspecifics, including facial expres-
sions40,41. Although the fast-latency response that we observed is
selective for fearful versus happy and neutral faces, fast responses
may also occur to other negative facial expressions such as anger.
It remains to be tested whether fast human amygdala responses occur
to simple, biologically relevant stimuli, such as snakes, that provided
© 2016 Nature America, Inc. All rights reserved.
survival threat during evolution22,42.
Accounts of automaticity in fear processing also suggest that amy-
gdala responses to emotional stimuli occur regardless of attentional
resources available or competition between concurrent inputs16,39,43,
and are not necessarily under voluntary control44. We did not explicitly
manipulate attention in either experiments 1 or 2; amygdala responses
were observed during incidental tasks of gender or indoor/outdoor
judgments, respectively. Nonetheless, we note that automatic amygdala
responses independent of attention would be predicted by the exist-
ence of fast, subcortical inputs16,39, for which our data provide, to the
best of our knowledge, the first direct electrophysiological support
in humans. However, it remains possible that amygdala responses to
coarse inputs without attention may involve other cortical or subcorti-
cal pathways receiving privileged early access to coarse (LSF) visual
information16,45. That is, a fast amygdala response could be driven by
inputs from other cortical regions, such as emotion-sensitive ventral or
orbitofrontal cortex, which also receive magnocellular pulvinar input46.
This is unlikely, as the ~70-ms latency response that we observed is
npg
faster than increased neuronal firing rates previously reported47 in
human ventral prefrontal cortex to emotional scenes (120–160-ms
latency), and faster than late latency (~500 ms) responses to fearful
faces observed in human orbitofrontal cortex12. Without simultane-
ously recording from these cortical areas, our data cannot exclude the
possibility that short-latency modulations of neural activity by SF and
fear relevance also occur in cortical regions receiving magnocellular
thalamic input15. Whether these brain regions indeed show equiva-
lent fast, LSF-dependent responses to fear-relevant stimuli in humans
remains unknown. However, our findings clearly demonstrate that
short-latency responses can be observed in human amygdala.
Our results demonstrate for the first time, to the best of our knowl-
edge, using direct electrophysiological recordings in a homogenous
sample of human patients, a fast (74 ms) and selective amygdala
response to emotional information. The early amygdala response was
specific to fearful, but not happy, facial expressions. Furthermore,
the effect was selective to socially relevant, fearful facial information
and was not evoked by unpleasant complex pictures. Fast responses
were oly observed to LSF, but not HSF, components of fearful faces,
consistent with coarse visual input providing limited, but rapid,
information via the magnocellular pathway. The latency of amygdala
responses to fearful faces was significantly faster than that observed
1048 VOLUME 19 | NUMBER 8 | AUGUST 2016  nature NEUROSCIENCE
in fusiform cortex. Thus, our data provide support for a low-road
circuit for fear detection in the human amygdala. This pathway, and its
functional properties, may also constitute an important neural substrate
for models of non-conscious processing in anxiety disorders42,48,49
and the generalization of fear responses to coarsely defined cues in
pathological conditions39,50.
Methods
Methods and any associated references are available in the online
version of the paper.
Note: Any Supplementary Information and Source Data files are available in the
online version of the paper.
Acknowledgments
We thank the electroencephalography technicians at the Hospital Ruber
Internacional. This work was supported by Project grant SAF2011-27766
from the Spanish Ministry of Science and Education and Marie Curie Career
Integration Fellowship (FP7-PEOPLE-2011-CIG 304248) to B.A.S., a PICATA
fellowship of CEI Moncloa (UCM-UPM) to C.M.-B., and a Ramón y Cajal
fellowship (RYC-2009-04974) to S.M. This work was supported by Project grant
SAF2011-27766 from the Spanish Ministry of Science and Education, Marie Curie
Career Integration Fellowship (FP7-PEOPLE-2011-CIG 304248), and BIAL
Foundation Grant 119/12 to B.A.S.
AUTHORS CONTRIBUTIONS
C.M.-B., S.M., P.V., A.G.-N. and B.A.S. designed the experiments. C.M.-B., F.L.-S.
and R.T. collected data and C.M.-B., S.M. and B.A.S. performed analyses. R.T. and
A.G.-N. monitored patients and performed clinical evaluation. R.M.-A. performed
surgical electrode implantation. Y.H.M. designed and performed electrode contact
localization. B.A.S., C.M.-B., S.M. and P.V. wrote the paper with input from all of
the other authors.
COMPETING FINANCIAL INTERESTS
The authors declare no competing financial interests.
Reprints and permissions information is available online at http://www.nature.com/
reprints/index.html.
1. LeDoux, J.E. The Emotional Brain (Simon & Schuster, New York, 1996).
2. Day-Brown, J.D., Wei, H., Chomsung, R.D., Petry, H.M. & Bickford, M.E. Pulvinar
projections to the striatum and amygdala in the tree shrew. Front. Neuroanat. 4,
143 (2010).
3. Tamietto, M. & de Gelder, B. Neural bases of the non-conscious perception of
emotional signals. Nat. Rev. Neurosci. 11, 697–709 (2010).
4. Johnson, M.H. Subcortical face processing. Nat. Rev. Neurosci. 6, 766–774 (2005).
5. Garrido, M.I., Barnes, G.R., Sahani, M. & Dolan, R.J. Functional evidence for a
dual route to amygdala. Curr. Biol. 22, 129–134 (2012).
6. Morris, J.S., Ohman, A. & Dolan, R.J. Conscious and unconscious emotional learning
in the human amygdala. Nature 393, 467–470 (1998).
7. Whalen, P.J. et al. Masked presentations of emotional facial expressions modulate
amygdala activity without explicit knowledge. J. Neurosci. 18, 411–418 (1998).
8. Pegna, A.J., Khateb, A., Lazeyras, F. & Seghier, M.L. Discriminating emotional faces
without primary visual cortices involves the right amygdala. Nat. Neurosci. 8,
24–25 (2004).
9. Morris, J.S., DeGelder, B., Weiskrantz, L. & Dolan, R.J. Differential extrageniculostriate
and amygdala responses to presentation of emotional faces in a cortically blind
field. Brain 124, 1241–1252 (2001).
10. Tamietto, M., Pullens, P., de Gelder, B., Weiskrantz, L. & Goebel, R. Subcortical
connections to human amygdala and changes following destruction of the visual
cortex. Curr. Biol. 22, 1449–1455 (2012).
11. Oya, H., Kawasaki, H., Howard, M.A. 3rd & Adolphs, R. Electrophysiological
responses in the human amygdala discriminate emotion categories of complex visual
stimuli. J. Neurosci. 22, 9502–9512 (2002).
12. Krolak-Salmon, P., Henaff, M.A., Vighetto, A., Bertrand, O. & Mauguiere, F. Early
amygdala reaction to fear spreading in occipital, temporal, and frontal cortex: a
depth electrode ERP study in human. Neuron 42, 665–676 (2004).
13. Naccache, L. et al. A direct intracranial record of emotions evoked by subliminal
words. Proc. Natl. Acad. Sci. USA 102, 7713–7717 (2005).
14. Brazdil, M. et al. Neural correlates of affective picture processing—a depth ERP
study. NeuroImage 47, 376–383 (2009).
15. Pessoa, L. & Adolphs, R. Emotion processing and the amygdala: from a ‘low road’
to ‘many roads’ of evaluating biological significance. Nat. Rev. Neurosci. 11,
773–783 (2010).
16. Vuilleumier, P. How brains beware: neural mechanisms of emotional attention.
Trends Cogn. Sci. 9, 585 (2005).

17. Schiller, P.H., Malpeli, J.G. & Schein, S.J. Composition of geniculostriate input
to superior colliculus of the rhesus monkey. J. Neurophysiol. 42, 1124–1133
(1979).
18. Berson, D.M. Retinal and cortical inputs to cat superior colliculus: composition,
convergence and laminar specificity. Prog. Brain Res. 75, 17–26 (1988).
19. Vuilleumier, P., Armony, J.L., Driver, J. & Dolan, R.J. Distinct spatial frequency
sensitivities for processing faces and emotional expressions. Nat. Neurosci. 6,
624–631 (2003).
20. Carretié, L., Hinojosa, J.A., López-Martín, S. & Tapia, M. An electrophysiological
study on the interaction between emotional content and spatial frequency of visual
stimuli. Neuropsychologia 45, 1187–1195 (2007).
21. Inagaki, M. & Fujita, I. Reference frames for spatial frequency in face representation
differ in the temporal visual cortex and amygdala. J. Neurosci. 31, 10371–10379
(2011).
22. Seligman, M.E.P. Phobias and preparedness. Behav. Ther. 2, 307–320 (1971).
23. Maris, E. & Oostenveld, R. Nonparametric statistical testing of EEG and MEG data.
J. Neurosci. Methods 164, 177–190 (2007).
24. Aggleton, J., Burton, M. & Passingham, R. Cortical and subcortical afferents to
the amygdala of the rhesus monkey (Macaca mulatta). Brain Res. 190, 347–368
(1980).
25. Stefanacci, L. & Amaral, D.G. Topographic organization of cortical inputs to the
lateral nucleus of the macaque monkey amygdala: a retrograde tracing study.
J. Comp. Neurol. 421, 52–79 (2000).
26. Amaral, D.G. & Insausti, R. Retrograde transport of D-[3H]-aspartate injected into
the monkey amygdaloid complex. Exp. Brain Res. 88, 375–388 (1992).
27. Kanwisher, N., McDermott, J. & Chun, M.M. The fusiform face area: a module
in human extrastriate cortex specialized for face perception. J. Neurosci. 17,
4302–4311 (1997).
© 2016 Nature America, Inc. All rights reserved.
28. Eimer, M. The face-sensitive N170 component of the event-related brain potential.
in The Oxford Handbook of Face Perception (eds. Calder, A., Rhodes, G., Johnson M.
& Haxby, J.) 329–344 (Oxford University Press, 2011).
29. Vuilleumier, P. & Pourtois, G. Distributed and interactive brain mechanisms during
emotion face perception: evidence from functional neuroimaging. Neuropsychologia
45, 174–194 (2007).
30. Livingstone, M. & Hubel, D. Segregation of form, color, movement, and depth:
anatomy, physiology and perception. Science 240, 740–749 (1988).
31. Merigan, W.H. & Maunsell, J.H.R. How parallel are the primate visual pathways?
Annu. Rev. Neurosci. 16, 369–402 (1993).
32. Wang, S. et al. Neurons in the human amygdala selective for perceived emotion.
Proc. Natl. Acad. Sci. USA 111, E3110–E3119 (2014).
33. Nguyen, M.N. et al. Neuronal responses to face-like stimuli in the monkey pulvinar.
Eur. J. Neurosci. 37, 35–51 (2013).
npg
nature NEUROSCIENCE  VOLUME 19 | NUMBER 8 | AUGUST 2016 1049
a r t ic l e s
34. Gothard, K.M., Battaglia, F.P., Erickson, C.A., Spitler, K.M. & Amaral, D.G. Neural
responses to facial expression and face identity in the monkey amygdala.
J. Neurophysiol. 97, 1671–1683 (2007).
35. Sato, W. et al. Rapid amygdala gamma oscillations in response to fearful facial
expressions. Neuropsychologia 49, 612–617 (2011).
36. Pourtois, G., Spinelli, L., Seeck, M. & Vuilleumier, P. Temporal precedence of
emotion over attention modulations in the lateral amygdala: Intracranial ERP
evidence from a patient with temporal lobe epilepsy. Cogn. Affect. Behav. Neurosci.
10, 83–93 (2010).
37. Rodriguez Merzagora, A. et al. Repeated stimuli elicit diminished high-gamma
electrocorticographic responses. NeuroImage 85, 844–852 (2014).
38. Dimberg, U. & Öhman, A. Behold the wrath: Psychophysiological responses to facial
stimuli. Motiv. Emotion 20, 149–182 (1996).
39. Anderson, A.K., Christoff, K., Panitz, D., De Rosa, E. & Gabrieli, J.D.E. Neural
correlates of the automatic processing of threat facial signals. J. Neurosci. 23,
5627–5633 (2003).
40. Öhman, A. Automaticity and the amygdala: nonconscious responses to emotional
faces. Curr. Dir. Psychol. Sci. 11, 62–66 (2002).
41. Kling, A.S. & Brothers, L.A. The amygdala and social behavior. in The Amygdala:
Neurobiological Aspects of Emotion, Memory, and Mental Dysfunction (ed. Aggleton,
J.P.) 353–377 (Wiley-Liss, 1992).
42. Ohman, A. & Mineka, S. Fears, phobias and preparedness: toward an evolved module
of fear and fear learning. Psychol. Rev. 108, 483–522 (2001).
43. Pessoa, L., McKenna, M., Gutierrez, E. & Ungerleider, L. Neural processing of emotional
faces requires attention. Proc. Natl. Acad. Sci. USA 99, 11458–11463 (2002).
44. Moors, A. & De Houwer, J. Automaticity: a theoretical and conceptual analysis.
Psychol. Bull. 132, 297 (2006).
45. Kveraga, K., Boshyan, J. & Bar, M. Magnocellular projections as the trigger of
top-down facilitation in recognition. J. Neurosci. 27, 13232–13240 (2007).
46. Barbas, H. Connections underlying the synthesis of cognition, memory, and emotion
in primate prefrontal cortices. Brain Res. Bull. 52, 319–330 (2000).
47. Kawasaki, H. et al. Single-neuron responses to emotional visual stimuli recorded
in human ventral prefrontal cortex. Nat. Neurosci. 4, 15–16 (2001).
48. Etkin, A. et al. Individual differences in trait anxiety predict the response of the
basolateral amygdala to unconsciously processed fearful faces. Neuron 44,
1043–1055 (2004).
49. Rauch, S.L. et al. Exaggerated amygdala response to masked facial stimuli in
posttraumatic stress disorder: a functional MRI study. Biol. Psychiatry 47, 769–776
(2000).
50. Sheline, Y.I. et al. Increased amygdala response to masked emotional faces in
depressed subjects resolves with antidepressant treatment: an fMRI study. Biol.
Psychiatry 50, 651–658 (2001).
 ONLINE METHODS
Participants. Participants were medication-resistant epilepsy patients with
depth electrodes surgically implanted to aid seizure focus localization (Table 1).
Implantation sites were chosen solely on the basis of clinical criteria. Patients
had normal or corrected-to-normal vision and had no history of head trauma
or encephalitis. Their amygdalae were radiologically normal on pre-operative
MRI (Supplementary Fig. 1). No statistical methods were used to pre-determine
sample sizes but our sample sizes are larger to those reported in previous publica-
tions11–14,35. All patients signed informed consent. The study had full approval
from the Hospital Ruber Internacional Ethics Committee.
Stereotactic electrode implantation. A contrast enhanced MRI was performed
pre-operatively under stereotactic conditions to map vascular structures prior to
electrode implantation and to calculate stereotactic coordinates for trajectories
using the Neuroplan system (Integra Radionics). DIXI Medical Microdeep depth
electrodes (multicontact, semi rigid, diameter of 0.8 mm, contact length of 2 mm,
inter-contact isolator length of 1.5 mm) were implanted based on the stereotactic
Leksell method.
Electrode contact localization. To take advantage of the visibility of individual
electrode contacts on CT images, for each patient we co-registered the pre-
electrode placement T1-weighted magnetic resonance images (pre-MRI) to
post-electrode placement CT (post-CT) whole-brain volumes. MRIs were
© 2016 Nature America, Inc. All rights reserved.
acquired on a 3 T Signa HDx GE scanner (GE Healthcare). To optimize this co-
registration, both brain images were first skull-stripped. For CTs this was done
by filtering out all voxels with signal intensities between 100 and 1,300 HU. Skull
stripping of the pre-MRI proceeded by first spatially normalizing the image to
MNI space employing the New Segment algorithm in SPM8 The resultant inverse
normalization parameters were then applied to the brain mask supplied in SPM8
to transform the brain mask into the native space of the pre-MRI. All voxels in
pre-MRI lying outside the brain mask and possessing a signal value in the high-
est 15th percentile were filtered out. The skull-stripped pre-MRI was then co-
registered and re-sliced to the skull-stripped post-CT. Next, the pre-MRI was
affine normalized to the post-CT, thus transforming the pre-MRI image into
native post-CT space. The two images were then overlaid, with the post-CT
thresholded such that only electrode contacts were visible.
Electrode contact visualization. To construct schematic summary illustra-
tions of the localizations of all amygdala contacts entered into our analyses, the
coordinates—in native post-CT space—were identified for the center of each
contact in each amygdala. For each patient, the pre-MRI (now in native post-CT
space) was then spatially normalized to MNI space, the ensuing normalization
npg
parameters applied to each of the amygdala contact coordinates for that patient,
and the resultant MNI coordinates indicated on the MNI template brain
(Figs. 2a and 5a). This approach was adopted instead of spatially normalizing
the fused pre-MRI and post-CT image to MNI space, because the latter approach
distorts contact topography. To provide a three-dimensional view of contact
localizations, the cytoarchitectonically defined amygdala51, provided in the
SPM Anatomy toolbox, was obtained by thresholding the separate laterobasal,
centromedial, and superficial group probability maps at ≥ 0.3, and combining
them into a single amygdala volume. The surface contour of the amygdala volume
(in standard MNI space) was then rendered in Paraview52 (Figs. 2b and 5b).
Experiment 1: emotional faces. Stimuli. We compiled faces of 139 different
actors (69 female) posing fearful, happy, and neutral expressions from three
databases: Karolinska Directed Emotional Faces (http://www.emotionlab.
se/resources/kdef)53, Warsaw Set of Emotional Facial Expression Pictures (http://
www.emotional-face.org/)54 and Radboud Faces Database (http://www.socsci.
ru.nl:8180/RaFD2/RaFD)55. Eye gaze of all face stimuli was directed forward.
Images were gray-scaled and enclosed in a rectangular frame (198 × 251 pixels)
excluding most hair and background (Fig. 1). Spatial frequency content in the
original stimuli (BSF) was filtered using a high-pass cut-off of >24 cycles per
image for HSF stimuli, and a low-pass cut-off of <6 cycles per image for LSF
stimuli (using Matlab, The Mathworks). Presented faces subtended a visual angle
of 7.4°, resulting in spatial frequency cut-offs of 3.24 and 0.81 cycles per degree for
HSF and LSF, respectively. Lastly, brightness was equalized across all nine condi-
tions (mean gray scale pixel values 115). Further, we ensured that there were no
nature NEUROSCIENCE
significant differences in the pixel value distributions between emotions within
each spatial frequency (Anderson-Darling k-sample tests: for BSF AD = 1.06,
tAD = −0.88, asymptotic P = 0.843; for HSF AD = 0.06, tAD = −1.81, asymptotic
P = 1; for LSF AD = 1.10, tAD = −0.84, asymptotic P = 0.824). For each patient, 135
different identities (out of the 139 that composed the whole set) were randomly
selected for presentation. Each of the 135 identities was pseudorandomly assigned
to one of the nine possible conditions: BSF fearful, HSF fearful, LSF fearful, BSF
happy, HSF happy, LSF happy, BSF neutral, HSF neutral and LSF neutral faces.
Thus, each condition was composed of 15 different identities, unique to that con-
dition. Pseudo-randomization proceeded such that either eight or seven identities
shared gender in each condition. Once the 135 faces were selected, their order of
presentation was randomized. The task was repeated twice. In the second block,
performed 10 min after the first, the same stimuli were presented as the first
block, but in a different pseudo-random order.
Procedure. Experiments were conducted during the second post-operative day.
All patients were seizure free for the previous 12h. In each of 2 experimental
blocks, faces were centrally displayed on an LCD computer screen for 500 ms
followed by a fixation cross for 3500 ms. Patients were required to make a gender
judgment, via button press, for each face (this task was employed because gender
judgments rely equally on HSF and LSF information, with no dominance by either
spatial frequency range56). Patients remained as still as possible attending the
centre of the screen while avoiding verbalizations and minimizing eye-blinks.
Data acquisition. Ongoing intracranial electroencephalogram (iEEG) activity
was acquired using an XLTEK EMU128FS amplifier (XLTEK). iEEG data were
recorded at each electrode contact site at a 500-Hz sampling rate (online bandpass
filter 0.1–150 Hz) and referenced to linked mastoid electrodes. The accuracy
of stimulus onset latencies was first measured with a photo-diode and a light-
to-voltage converter (TKK Brain Research Unit), and latency uncertainty found
to be in the range of 2 ms.
Data analysis. Patient inclusion. Of 16 patients with amygdala electrodes who
completed the task, four patients did not meet our criteria for spike-free trials
(75%) and were thus excluded. A further two patients were excluded due to poor
task engagement (12% and 38% trials in which responses were omitted, respec-
tively, compared to an average 0.67% across all other patients). One patient was
excluded due to the presence of large-amplitude slow oscillations during record-
ing. Lastly, electrophysiological responses from one patient did not demonstrate
any discernible stimulus-evoked components during the 640-ms post-onset inter-
val. Data from this patient were also excluded. Thus, we analyzed iERPs from ten
amygdalae from eight patients (four left-sided, two right-sided and two bilater-
ally implanted). Patient demographics and clinical details are given in Table 1.
Although patient 5 had procedural and verbal IQ in the borderline and extremely
low range, respectively (Table 1), this patient’s gender judgment performance was
comparable to that of other patients (Supplementary Table 1).
Pre-processing. For each amygdala contact, experimental condition, and
patient, epochs from −200 to 640-ms peri-stimulus time were extracted from
continuous iEEG data. Epochs containing epileptiform activity or artifacts
(large-amplitude slow-wave drifts or high-frequency activity) were rejected
by trial-by-trial visual inspection, as were epochs corresponding to absent or
multiple behavioral responses. Epochs were then detrended, baseline corrected
(100-ms pre-stimulus baseline) and no filter was applied. No further off-line fil-
tering of the data was performed to avoid filter effects that may distort waveforms
and hence introduce latency artifacts57. For each experimental condition, data
were then averaged across the two blocks. In the case that there was more than
one contact within the amygdala, data from all contacts were averaged within
trial for that amygdala.
Statistics. We applied a cluster-based non-parametric permutation statistic,
based on MANOVA F values with within-subject factors of emotion (fear, happy,
neutral) and spatial frequency (BSF, LSF, HSF), to determine the time points of
significant interaction between emotion and spatial frequency with respect to
iERP amplitude. This approach effectively corrects the family-wise error rate in
the context of multiple comparisons of latency bins23. Under the null hypothesis
of no differences between levels of each test (main effects of emotion or spatial
frequency, and their interaction), the amplitude values at each time point can
be permuted between conditions within each factor. After a permutation step,
a MANOVA was calculated at each time point for each test separately, with a
cluster threshold of P < 0.01 for the main effects and interaction. Thus, significant
doi:10.1038/nn.4324
 time clusters were formed by temporal adjacency of supra-threshold effects
(a cluster contained at least two significant neighbors along the time dimension).
For each cluster, the MANOVA F values (Wilk’s lambda) of the corresponding
test were summed and the greatest sum among all clusters entered into the per-
mutation distribution. Note that as the permutation distribution is a data-driven
non-parametric distribution, no degrees of freedom are given. Permutation steps
were repeated 1,000 times and permutation distributions for main effects and
interactions created. Initially, empirical cluster sums of MANOVA F values that
were greater than the 99th centile within the permutation distribution were con-
sidered as significant temporal clusters of emotion/spatial frequency main effects
or interaction. We next applied a less conservative cluster threshold of P < 0.05
and repeated the permutation testing. The more conservative (P < 0.01) and
relaxed (P < 0.05) cluster thresholds were applied in order to balance risk of false
positives (that is, that weak, neighboring single-time-point effects combine into
a significant cluster) and risk of false negatives, respectively. A MANOVA was
chosen over univariate ANOVA to eschew violation of the sphericity assump-
tion. Next, to explore the differences in each significant time cluster (for main
effects and interaction), the mean amplitude values for each subject and each
condition across the significant clusters were computed for each effect and tested
with specific contrasts and post hoc t tests. When applying parametric tests, data
distribution was assumed to be normal, but this was not formally tested. Finally,
we applied cluster-based permutation statistics on the iERP to each of the nine
face stimulus types separately, to test a null hypothesis of deflections being equal
© 2016 Nature America, Inc. All rights reserved.
to zero for the entire post-stimulus period.
Of 14 patients with amygdala electrodes who met inclusion criteria on the basis
of performance on the gender judgment task, seven also had electrodes in visual
areas. Identical data pre-processing steps were employed for these iEEG data as
for amygdala contacts, with one patient rejected for not meeting criterion for
spike-free trial number. A further patient did not have contacts in the fusiform
gyrus. Thus, from five patients, seven groups of fusiform contacts were entered
into the same cluster-based permutations statistics as applied to amygdala contact
data (again with cluster threshold of P < 0.01 followed by P < 0.05). To test for
latency effects of BSF and LSF fearful faces employing non-corrected statistics, we
compared iERPs for each stimulus type relative to zero in two-tailed one-sample
t tests. Only time clusters with more than four adjacent data points (that is, at
least 10 ms) are considered significant.
For those patients for whom responses were successfully recorded simul-
taneously from both amygdala and fusiform (four patients), we computed
the average values for fearful, happy, and neutral faces across all frequencies.
To test the interaction between electrode site (amygdala, fusiform) and emotion
(fearful, happy and neutral) we calculated difference values with respect to the
three different combinations of the factor levels for emotion at each electrode site.
npg
For each emotion, the amplitude differences between electrode sites were also
determined. The difference values were submitted to non-parametric Friedman
tests. Significant interactions between electrode sites and emotion were further
analyzed by Wilcoxon signed ranked tests applied to the factors emotion and
electrode site. Data collection and analysis were not performed blind to the con-
ditions of experiments 1 or 2. A Supplementary Methods Checklist is available
and includes statistics derived from experiments 1 and 2.
Source localization of amygdala responses. Electrodes containing amygdala
contacts were selected for each patient, and the average responses to BSF and
LSF fearful faces from all contacts of each selected electrode (that is, within
and outside the amygdala) were analyzed. Noisy channels were removed. Thus
data from 5–12 contacts per electrode were submitted to source localization at
the individual patient level, implemented in Brainstorm58 software. After co-
registration of the individual pre-operative MRIs with post-operative CT scans,
a dipole grid (20,000 dipoles) restricted to the neocortex, amygdala and hip-
pocampus was used as a brain model to estimate the current source distribution.
This dipole grid was used to calculate a forward model employing the bound-
ary element method59. An l2-weighted minimum norm algorithm (wMNE)60
was then applied to obtain mean current source densities for the time windows
corresponding to the observed early (76–110 ms) response to fear. To measure
the consistency of this localization over all amygdalae, each wMNE solution
was normalized to MNI space61 using SPM8. The wMNE solutions for right-
sided amygdalae electrodes were flipped on the left-right axis. Lastly, for each of
the 20,000 dipoles comprising the neocortex, amygdala and hippocampus, the
doi:10.1038/nn.4324
number of subjects showing at least 50% of the maximum activation at that dipole
was counted. This process was then repeated for the later main effect of emotion
time window (282–302 ms). Note that for patient 10, source localization was not
possible due to two channels within, and five channels lateral to, the amygdala
(seven of ten channels) displaying a flat signal, prohibiting the calculation of a
potential gradient along the electrode contacts.
Eye movements. Although eye movements were not recorded in the current
experiment, it is unlikely that task-related saccadic eye movements can account
for the fast amygdala responses we observe. A previous fMRI study comparing
responses to fearful versus neutral BSF, HSF and LSF faces reported no signifi-
cant effects or interaction due to SF or emotion in mean eye-position data over
a 200-ms period after stimulus onset19. Furthermore, iERP recordings are
considered relatively less susceptible to eye movement artifacts compared to
scalp-recorded ERPs62.
Experiment 2: emotional pictures. Experiment 2 proceeded in an identical man-
ner to experiment 1, except for the following:
Stimuli. Patients were presented with 40 emotional and 80 neutral color
pictures. These were drawn at random from a pool of 80 high-arousing unpleasant
(mutilations and attack) scenes selected from the International Affective Picture
System63 (IAPS), and 160 low-arousing neutral pictures: 149 taken from the IAPS
(household scenes and neutral persons) and eleven neutral landscape pictures
taken from the world-wide web. Mean normative IAPS picture ratings (SEM)
on a nine-point scale for valence were 5.05 (0.05) for neutral, and 2.04 (0.05) for
unpleasant pictures, and for arousal were 3.29 (0.06), and 6.3 (0.07) for neutral
and unpleasant pictures, respectively. Note that in both experiments 1 and 2, the
ratio of negative emotional to non-negative stimuli was 1:2.
Procedure. This experiment was conducted during the third post-operative day.
Emotional and neutral pictures were presented pseudo-randomly (presentation
time 500 ms; interstimulus interval 3,500 ms) with a constraint that emotional
pictures were separated by at least one neutral picture. Patients were required
to make an indoor-outdoor judgment to each via button-press. Prior to signing
informed consent, patients were shown one example of an unpleasant IAPS
picture and instructed that they would see similar pictures both on that day and
the next (patients saw the same pictures again the next day during a recognition
memory test).
Data acquisition. This was identical to experiment 1, but on post-operative day 3.
Data analysis. Patient inclusion. Of 14 patients who completed the task, nine
met all inclusion criteria. Two patients were excluded due to poor push-but-
ton response rate (36% and 28% trials in which responses were omitted com-
pared to 1.22% mean omissions for the nine patients included in the analysis).
A further three patients did not meet our criteria for spike-free trials (75%).
Patient demographics and clinical details are given in Table 1. Despite bor-
derline and extremely low range IQs of patients 5 and 8, these patients’ reac-
tion times on the indoor/outdoor task were comparable to other patients’
(Supplementary Table 12).
Preprocessing. After spike and artifact rejection, data preprocessing was as for
experiment 1; epochs of unfiltered data were detrended and baseline corrected
(100-ms pre-stimulus baseline).
Statistics. For statistical comparison of evoked responses, we applied the same
cluster-based permutation approach as for experiment 1, but a paired t test was
used as the initial cluster statistic to compare emotional versus neutral pictures.
To calculate the points in time at which iERPs started to differ significantly from
zero, we used the same procedure as for experiment 1 iERPs.
For the four patients with electrodes in the fusiform gyrus who completed
both experiments 1 and 2, evoked responses to BSF neutral faces recorded in
experiment 1 were compared to those evoked by neutral scenes in experiment 2,
again employing cluster-based permutations statistics with a cluster threshold of
P < 0.01 (n = 6 groups of contacts). Note that some neutral IAPS pictures (18%)
contained human faces embedded within the scenes, which were removed from
this analysis.
Data availability. The data that support the findings of this study are available
from the corresponding author upon request.
nature NEUROSCIENCE
 Code availability. Stimulus presentation, electrode contact localization, and all
iERP analyses were done using open source software packages. Visual stimuli
were presented using Cogent2000 (http://www.vislab.ucl.ac.uk/cogent_2000.
php). Electrode contacts were visualized using SPM8 (http://www.fil.ion.ucl.
ac.uk/spm). Signal processing and permutation statistics were implemented in
the Fieldtrip toolbox (http://www.fieldtriptoolbox.org/). The 3-by-3 MANOVA
for each permutation was calculated within the R-environment for statistical
computing (https://www.r-project.org/; the R-code can be obtained from the
authors upon request) and subsequent cluster forming and statistical inference
done using the fieldtrip-toolbox. Forward head modeling and cortical source
analyses were calculated using the Brainstorm toolbox (http://neuroimage.usc.
edu/brainstorm/). These open source toolboxes, with the exception of R, were
run on Matlab.
51. Amunts, K. et al. Cytoarchitectonic mapping of the human amygdala, hippocampal
region and entorhinal cortex: intersubject variability and probability maps.
Anat. Embryol. (Berl) 210, 343–352 (2005).
52. Ahrens, J., Geveci, B. & Law, C. ParaView: An end-user tool for large-data
visualization. in The Visualization Handbook (eds. Hansen, C.D. & Johnson, C.R.)
717 (Citeseer, 2005).
53. Lundqvist, D., Flykt, A. & Öhman, A. The Karolinska Directed Emotional Faces–
KDEF (Department of Clinical Neuroscience, Psychology Section, Karolinska
Institutet, 1998).
© 2016 Nature America, Inc. All rights reserved.
npg
nature NEUROSCIENCE
54. Olszanowski, M., Pochwatko, G., Kukliń ski, K., Ś cibor-Rylski, M. & Ohme, R. Warsaw
set of emotional facial expression pictures—validation study of facial display
photographs. Front. Psychol. 5, 1516 (2015).
55. Langner, O., Dotsch, R., Bijlstra, G., Wigboldus, D.H.J., Hawk, S.T. & van
Knippenberg, A. Presentation and validation of the Radboud Faces Database.
Cogn. Emot. 24, 1377–1388 (2010).
56. Schyns, P.G. & Oliva, A. Dr. Angry and Mr. Smile: When categorization flexibly
modifies the perception of faces in rapid visual presentations. Cognition 69,
243–265 (1999).
57. Tanner, D., Morgan-Short, K. & Luck, S.J. How inappropriate high-pass filters can
produce artifactual effects and incorrect conclusions in ERP studies of language
and cognition. Psychophysiology 52, 997–1009 (2015).
58. Tadel, F., Baillet, S., Mosher, J.C., Pantazis, D. & Leahy, R.M. Brainstorm: a user-
friendly application for MEG/EEG analysis. Comput. Intell. Neurosci. 2011, 879716
(2011).
59. Gramfort, A., Papadopoulo, T., Olivi, E. & Clerc, M. OpenMEEG: opensource software
for quasistatic bioelectromagnetics. Biomed. Eng. Online 9, 45 (2010).
60. Hauk, O. Keep it simple: a case for using classical minimum norm estimation in
the analysis of EEG and MEG data. Neuroimage 21, 1612–1621 (2004).
61. Collins, D.L. et al. Design and construction of a realistic digital brain phantom.
IEEE Trans. Med. Imaging 17, 463–468 (1998).
62. Lachaux, J.P., Rudrauf, D. & Kahane, P. Intracranial EEG and human brain mapping.
J. Physiol. Paris 97, 613–628 (2003).
63. Lang, P.J., Bradley, M.M. & Cuthbert, B.N. International affective picture system
(IAPS): affective ratings of pictures and instruction manual (Technical Report A-6)
(University of Florida, 2005).
doi:10.1038/nn.4324