doi:10.1093/brain/awv336
BRAIN 2016: 139; 54–61
| 54
REPORT
Early detection of intentional harm in the
human amygdala
*These authors contributed equally to this work.
A decisive element of moral cognition is the detection of harm and its assessment as intentional or unintentional. Moral cognition
engages brain networks supporting mentalizing, intentionality, empathic concern and evaluation. These networks rely on the amygdala
as a critical hub, likely through frontotemporal connections indexing stimulus salience. We assessed inferences about perceived harm
using a paradigm validated through functional magnetic resonance imaging, eye-tracking and electroencephalogram recordings. During
the task, we measured local field potentials in three patients with depth electrodes (n = 115) placed in the amygdala and in several
frontal, temporal, and parietal locations. Direct electrophysiological recordings demonstrate that intentional harm induces early activity in
the amygdala (5200 ms), which—in turn—predicts intention attribution. The amygdala was the only site that systematically discriminated between critical conditions and predicted their classification of events as intentional. Moreover, connectivity analysis showed that
intentional harm induced stronger frontotemporal information sharing at early stages. Results support the ‘many roads’ view of the
amygdala and highlight its role in the rapid encoding of intention and salience—critical components of mentalizing and moral evaluation.
1 Laboratory of Experimental Psychology and Neuroscience, Institute of Cognitive Neurology, Favaloro University, Buenos Aires,
Argentina
2 UDP-INECO Foundation Core on Neuroscience (UIFCoN), Faculty of Psychology, Universidad Diego Portales, Santiago, Chile
3 Instituto de Ingenierı́a Biomédica, Facultad de Ingenierı́a, Universidad de Buenos Aires, Argentina
4 National Research Council (CONICET), Buenos Aires, Argentina
5 Department of Psychology, and Department of Psychiatry and Behavioural Neuroscience, The University of Chicago, Chicago, IL, USA
6 Laboratory of Neuroscience, Universidad Torcuato Di Tella, Buenos Aires, Argentina
7 Departamento de Fı́sica, FCEN, UBA and IFIBA, Conicet, Buenos Aires, Argentina
8 Programa de Cirugı́a de Epilepsia, Hospital Italiano de Buenos Aires, Buenos Aires, Argentina
9 Escuela de Psicologı́a, Pontificia Universidad Católica de Chile, Santiago, Chile
10 Centre of Excellence in Cognition and its Disorders, Australian Research Council (ARC), New South Wales, Australia
11 Department of Psychology, University of Cambridge, Cambridge, UK
12 Universidad Autónoma del Caribe, Barranquilla, Colombia
Correspondence to: Agustin Ibanez,
Laboratory of Experimental Psychology and Neuroscience,
Institute of Cognitive Neurology,
Favaloro University, Buenos Aires, Argentina
E-mail: aibanez@ineco.org.ar
Keywords: amygdala; intentional harm; moral cognition; intracranial recordings
Abbreviation: wSMI = weighted Symbolic Mutual Information
Received June 16, 2015. Revised September 22, 2015. Accepted September 25, 2015. Advance Access publication November 25, 2015
ß The Author (2015). Published by Oxford University Press on behalf of the Guarantors of Brain. All rights reserved.
For Permissions, please email: journals.permissions@oup.com
Downloaded from https://academic.oup.com/brain/article/139/1/54/2468818 by guest on 22 June 2022
Eugenia Hesse,1,2,3,* Ezequiel Mikulan,1,2,4,* Jean Decety,5 Mariano Sigman,6,7
Marı́a del Carmen Garcia,8 Walter Silva,8 Carlos Ciraolo,8 Esteban Vaucheret,8
Fabricio Baglivo,1,2,3 David Huepe,2 Vladimir Lopez,9 Facundo Manes,1,2,4,10
Tristan A. Bekinschtein11 and Agustin Ibanez1,2,4,10,12
Early detection of intentional harm in the human amygdala
Introduction
Materials and methods
Participants
Three patients with intractable epilepsy who were offered surgical intervention to alleviate their condition took part in the
study. Subject 1 was a 19-year-old, right-handed female who
had completed 1 year of tertiary education and suffered from
drug-resistant epilepsy since the age of 16 years. Subject 2 was
a 57-year-old, left-handed male with an undergraduate degree
who suffered from drug-resistant epilepsy since the age of 42
years. Subject 3 was a 29-year-old, left-handed female with an
undergraduate degree who suffered from epilepsy since the age
of 18 years. All of the subjects gave written informed consent
in accordance with the Declaration of Helsinki, and the study
was approved by the Institutional Ethics Committee of the
Hospital Italiano de Buenos Aires, Argentina. They were attentive and cooperative while performing the task. Their
cognitive performance under
(Supplementary Table 1).
the
task
was
| 55
accurate
Patients’ recordings
Direct cortical recordings were obtained from semi-rigid,
multi-lead electrodes that were implanted in each patient.
The electrodes used have a diameter of 0.8 mm and consist
of 5, 10 or 15 contact leads 2-mm wide and 1.5-mm apart
(DIXI Medical Instruments). The video-SEEG monitoring
system (Micromed) records as many as 128 depth-EEG electrode sites simultaneously. Our three subjects had electrodes in
the left amygdala, although two of them were left-handed. No
data were collected from the right amygdala. All patients were
carefully selected such that the amygdalae from which the
recordings were obtained were distal to the epileptogenic
foci, and no single recording site presented epileptogenic activity (see below). Subject 1 had 128 sites recorded. Subject 2 had
90 sites recorded. Subject 3 had 105 sites recorded. The
recordings were sampled at 1024 Hz.
Post-implantation MRI and CT scans were obtained from
each patient. Both volumetric images were affine registered
and normalized using the SPM8 MATLAB toolbox. Using
MRIcron, the MNI coordinates of each contact site and their
respective Brodmann areas were obtained and are listed in
Supplementary Table 2. We used the normalized position of
the electrodes’ contact sites to an MNI coordinate space because this procedure allowed us to define the patient’s results
in a common space (Foster et al., 2015).
Experimental design: task and stimuli
We used an adaptation of the Intention Inference Task (Decety
et al., 2012; Decety and Cacioppo, 2012; Escobar et al., 2014)
to assess the detection of intentional harmful actions. The task
evaluates intentional detection in the context of intentional/
unintentional/neutral harms and consists of three scenarios:
(i) intentional harm in which one person is in a painful situation intentionally caused by another (e.g. pushing someone
off a bench); or (ii) unintentional harms in which one person
is in a painful situation unintentionally caused by another; and
(iii) neutral or control situations (e.g. one person receiving a
flower given by another). The Intention Inference Task evaluates the comprehension of the unintentional or deliberate
nature of the action and the intention of the perpetrator to
hurt. It consists of 25 animated scenarios (11 intentional, 11
unintentional, three neutral), and one practice trial for each
category (before the task). The patients were asked to perform
the task a few times. Subjects performed the task twice
(Subject 2 three times). Although these are a small number
of trials, our results were robust and were consistent with
the single trial responses observed in intracranial recordings
and their enhanced signal-to-noise ratio (Jacobs and Kahana,
2010). Each scenario consists of three digital colour pictures
presented in a successive manner to imply motion. The durations of the first, second and third pictures in each animation
were 500 (T1), 200 (T2), and 1000 (T3) ms, respectively. See
the Supplementary material for stimuli validation with behavioural and eye-tracking measures. For additional stimulus examples and validation information, see Fig. 1A, Supplementary
Fig. 1 and Supplementary material. For a video illustration of
the clips, see Supplementary Video 1.
Downloaded from https://academic.oup.com/brain/article/139/1/54/2468818 by guest on 22 June 2022
Perceiving and reacting to harm is crucial for survival and
social interaction. Indeed, the assessment of deliberately
harmful actions moulds human morality (Decety et al.,
2012; Treadway et al., 2014; Ames and Fiske, 2015).
Moral evaluation engages neurocognitive mechanisms supporting theory of mind, intentionality, empathic concern
and evaluation (Moll et al., 2005; Young et al., 2007;
Moll and Schulkin, 2009; Decety and Cowell, 2014).
Neuroimaging studies show that these cognitive domains
involve widely distributed networks (Moll and Schulkin,
2009; Decety et al., 2012; Ibanez and Manes, 2012).
This network relies on the amygdala as a critical hub
(Treadway et al., 2014), likely through frontotemporal connections indexing stimulus salience (Pessoa and Adolphs,
2010).
However, available evidence presents various limitations.
First, functional MRI studies of morality are blind to early
differences among relevant mechanisms (Huebner et al.,
2009). Second, amygdala activation is confounded by
stimulus-related signal fluctuation in nearby veins draining
distant brain regions (Boubela et al., 2015). Third, EEG/
MEG studies of subcortical source space are inaccurate.
Thus, no evidence exists of a direct and early involvement
of the amygdala in the detection of intentional harm.
To bridge these gaps, we assessed inferences about perceived harm using a paradigm previously validated through
functional MRI and eye-tracking (Decety et al., 2012) as
well as EEG recordings (Decety and Cacioppo, 2012;
Escobar et al., 2014). Participants viewed short videos depicting interactive situations that involved intentional harm,
unintentional harm or no harm at all. Their task was to
evaluate whether the actions were intentional or unintentional. All stimuli were presented in a three-frame sequence
(T1: 500 ms, T2: 200 ms, T3: 1000 ms; see ‘Materials and
methods’ section).
BRAIN 2016: 139; 54–61
56
| BRAIN 2016: 139; 54–61
E. Hesse et al.
Downloaded from https://academic.oup.com/brain/article/139/1/54/2468818 by guest on 22 June 2022
Figure 1 Amygdala responses to intentional harm. (A) Examples of stimuli used for intentional, unintentional, and neutral conditions,
together with the temporal duration of each image (see examples in Supplementary Video 1). (B) Electrode contact sites in the amygdala.
(C) Time-frequency charts. T1, T2 and T3 represent the times at which each digital picture is presented to the subject. A zero value was assigned
to points that were not significant (P 4 0.05) relative to the baseline. Left: Subtraction between the amygdala time-frequency charts obtained from
the intentional and unintentional conditions. Right: Subtraction between the amygdala time-frequency charts obtained from the intentional and
neutral conditions. (D) Averaged power spectrum of the intentional and unintentional time-frequency charts using different frequency ranges. The
green marks identify significant differences between conditions (bootstrapping, P 5 0.01). Left: Frequency range 1–40 Hz. Right: 60–150 Hz. (E)
Binary logistic regression between conditions (intentional = 1, unintentional = 0) and mean value of the power spectrum on broadband and
0–1000 ms window (B = 0.580, R = 0.44, P = 0.0000001, correct categorization = 70.54%).
Early detection of intentional harm in the human amygdala
The faces of the protagonists were not visible and thus there
were no facial emotional reactions visible to the patients.
However, body expressions and postures provided sufficient
information about the intention of the agent. Patients were
asked to respond to intentionality (was the harmful action
done on purpose?). The question was answered by selecting
‘Yes’ or ‘No’ with two different buttons.
Data analysis
Signal preprocessing
| 57
Logistic regression of the trial-by-trial
analysis
Logistic regression analysis was performed to evaluate whether
the power activity across each trial could predict the subject
categorization as intentional or unintentional. The dependent
variable was the unintentional (0) or intentional (1) condition.
The independent variable of interest was the averaged value of
power spectrum over time (0–1000 ms since stimuli onset) for
the 60–150 Hz frequency range in each trial. Statistical significance was considered to be P 5 0.01. Outliers were detected
using the Tukey two-sided method (Tukey hinge distance
factor = 1.5) (Tukey, 1977). Three outlier values were detected and left out of the analysis. This procedure was done
for the amygdala power spectrum values and for the other regions, grouped by regions of interest within subjects (see below).
Comparison between the amygdala
and the other regions
To assess the amygdala’s power activation and ability to distinguish between conditions (logistic regression) relative to the
other regions, we performed a three-step analysis: (i) to evaluate whether the region discriminates the intentional condition,
and the intentional from unintentional conditions; (ii) to compare the amygdala’s power activation with that of the regions
that did discriminate the intentional conditions and the intentional from unintentional conditions; and (iii) to perform a
logistic regression of single trial data as predictors of subject
classification. See Supplementary material for detailed information on this three-step analysis.
Time–frequency analysis
The time–frequency charts were obtained by analysing the
digitized signals using a windowed Fourier transform
(window length: 250 ms, step 8 ms, window overlap 97%)
(Gross, 2014). Our scripts were based on the newtimef.m
script. As the frequency analysis is window-centred, we consider that the earliest unbiased significant temporal value is
125 ms, when the window centre is at 0 ms (T1 stimulus
presentation, see Supplementary material). The time–frequency
charts were normalized to the baseline before the stimulus
onset. The normalization involved subtracting the baseline
average and dividing by the baseline standard deviation on a
frequency-by-frequency basis using a window from 500 to 0
relative to the stimuli onset.
We obtained the time–frequency chart for each condition
(intentional, unintentional and neutral harmful actions) and
performed subtractions between them (intentional unintentional, intentional neutral and unintentional neutral).
Significant power increases and decreases across time against
baseline values were analysed with Monte Carlo permutation
tests (5000) combined with bootstrapping, as reported in other
intracranial studies (Naccache et al., 2005; Ibanez et al.,
2013). This simple method offers a straightforward solution
for multiple comparison problems and for data distribution
assumptions. Frequency band ranges of 1 to 40 Hz and broadband (60 to 150 Hz) of the time-frequency charts were averaged for the signals obtained from the intentional and
unintentional conditions.
Amygdala connectivity analysis using
the weighted Symbolic Mutual
Information measure
To analyse the amygdala’s connectivity with the other regions
within each subject, we used the weighted Symbolic Mutual
Information (wSMI) measure (King et al., 2013). This method
calculates a non-linear index of information sharing between
two signals. The signals are transformed into symbols. By
defining a value of k, the number of samples that represent
a symbol, and , the temporal separation between them, a
frequency range is defined for which wSMI will be sensitized.
The joint probability between the signals was then calculated
for each pair of channels, for each trial, with a fixed value of
k = 3 and = 32 ms—hence establishing the frequency range to
1–10 Hz. The signals were low-pass filtered at 10 Hz to avoid
aliasing effects (see Supplementary material for more details).
A seed analysis based on the wSMI was calculated for
the amygdala’s signal with the other regions within each subject for the intentional and unintentional conditions (signals
without the baseline time window). The statistical comparisons
between the connectivity values obtained for each condition
were performed using a Wilcoxon Signed Rank test. The null
hypothesis was rejected if a t-value was greater than the most
extreme 5% of the distribution (i.e. P 5 0.05). The BrainNet
Viewer toolbox was used for visualization of wSMI.
Downloaded from https://academic.oup.com/brain/article/139/1/54/2468818 by guest on 22 June 2022
The data were bandpass filtered from 1 to 200 Hz using a zero
phase shift finite impulse filter. Then, they were notch filtered
at 50 Hz and its harmonic frequencies (100 Hz, 150 Hz) to
eliminate the line artefacts. The contact sites recorded from
each patient who presented artefacts and pathological waveforms were discarded. This was achieved by visually inspecting
the recordings and by the following criteria: (i) signal values
do not exceed five times the signal mean; and/or (ii) consecutive signal samples do not exceed 5 standard deviations (SD)
from the gradient mean. A total number of 115 contact sites
remained after applying these criteria (35 contact sites for
Subject 1, 44 for Subject 2 and 36 for Subject 3;
Supplementary Fig. 2).
Once the sites that complied with the criteria were selected,
they were referenced to the mean value (the averages of the
sites per subject were subtracted from each recording). Finally,
the data were segmented into 2000 ms epochs, including a
500 to 0 ms prestimulus baseline period. The epochs were
baseline corrected.
BRAIN 2016: 139; 54–61
58
| BRAIN 2016: 139; 54–61
Brief time span functional
connectivity
A functional connectivity analysis (Omigie et al., 2015) of the
early window (0 to 500 ms) was implemented to study the
correlation between different regions. The signals were first
bandpass filtered in two frequency band ranges, 1 to 40 Hz
and 60 to 150 Hz, for the intentional and unintentional signals. See Supplementary material for more details.
Results
unintentional harm, intentional harm induced stronger
frontotemporal connectivity in all patients (bootstrapping,
for
1–40 Hz:
Subject
1:
P = 6.3139 10 35,
t = 42.54; Subject 2: P = 7.989 10 33, t = 35.24;
Subject 3: P = 2.8841 10 10, t = 21.18; for 60–150 Hz:
Subject 1: P = 5.2694 10 48, t = 42.35; Subject 2:
P = 1.9606 10 11, t = 32.21; Subject 3: P = 2.7941
10 58, t = 43.95, see Fig. 2B), even when controlling for
the neutral condition (Supplementary Fig. 6). In addition,
we found that intentional harm elicited increased frontotemporal connectivity at medium and long range distances
(Supplementary Fig. 7). Thus, detection of intentional harm
was associated with greater fronto-amygdala information
sharing during T1–T3 and with fronto-temporal coupling
at early stages (T1).
Discussion
Previous reports pointed to the amygdala as a critical hub to
appraise intentional harmful actions and stimulus salience
(Treadway et al., 2014). Our results provide unprecedented
spatiotemporal evidence for its role in the early encoding of
intention, the subsequent categorization of harmful events,
and the automatic modulation of corticolimbic connections.
These findings support the view that the amygdala indexes
the biological significance of salient stimuli through multipathway networks (Pessoa and Adolphs, 2010).
The concept of intentionality has been variously defined
in the literature. Here, we propose that harm is intentional
insofar as it reflects the perpetrator’s motivation to deliberately hurt another person, leading to mostly negative
moral judgements (Supplementary material). We have selected a well-validated and replicated set of stimuli allowing
an early and unambiguous categorization of intentional
harm versus unintentional harm, while controlling for
basic variables (such as familiarity) (Supplementary material). Future studies should assess early modulations of
amygdala activity by intentionality manipulating both
basic variables and other potentially relevant cognitive
dimensions (Supplementary material).
The amygdala has been implicated in the processing and
transmission of sensory salient stimuli to guide behaviours
and decision-making (Janak and Tye, 2015). Consistent
with this claim, amygdala activity at broadband (a band
that captures spike firing neurons) (Manning et al., 2009)
predicted single-trial behavioural performance, as previously
reported for aversive learning (Lim et al., 2009). The rapid
(125 ms) involvement of this structure replicates previous
findings with scalp EEG (Decety and Cacioppo, 2012;
Escobar et al., 2014), highlighting its role in other automatic
processes (Pessoa and Adolphs, 2010), such as emotional salience and face/object recognition (see Supplementary material
for a deeper assessment of this finding). Note that relatively
slow (Janak and Tye, 2015) and high (Oya et al., 2002)
amygdala frequencies are sensitive to stimulus salience modulations. The absence of similar discrimination/prediction in
Downloaded from https://academic.oup.com/brain/article/139/1/54/2468818 by guest on 22 June 2022
During the task, we measured local field potentials (at T1,
T2, and T3) in three patients (Subjects 1, 2 and 3) with
depth electrodes (n = 115) placed in the amygdala (Fig. 1B,
n = 6) and in several frontal, temporal and parietal locations (n = 109; see Supplementary Fig. 2 and
Supplementary Table 2 for spatial locations).
Intentionality and content (harm versus neutral) were discriminated by an activity boost in the amygdala (all sites).
This was observed during the first 200 ms after stimulus
onset (T1) at 1–40 Hz and throughout T1–T2–T3 at broadband (60–150 Hz) (Fig. 1C and Supplementary Fig. 3).
Bootstrapped permutations of single trial analysis revealed
greater activity for intentional than unintentional harm at an
early time window (80–200 ms, 1–40 Hz) and throughout
the T1–T3 time points at broadband (Fig. 1D). This
occurred separately in each patient (Supplementary Table 3).
Moreover, a trial-by-trial analysis of amygdala responses
(averaged during T1–T3) at broadband predicted the subjects’ subsequent categorization as intentional or unintentional (Fig. 1E). Such a classification was not predicted by
the activity of any other region (Supplementary Fig. 4). The
amygdala was the only site that systematically discriminated
between critical conditions in all subjects (at both low and
high bands) and predicted their classification of events.
To examine whether such modulation in the amygdala
resonated in other regions, we analysed both (i) amygdala
connections during the full stimulus set presentation; and
(ii) connectivity at 1–40 Hz and broadband among all recording sites at early stages. First, via a wSMI analysis
(King et al., 2013), we explored the integration and
global broadcasting of information across non-linear amygdala connections. At relatively low frequencies (1–10 Hz),
enhanced fronto-amygdalar connectivity (each subject separately: Subject 1: mesial/lateral supplementary motor area;
Subject 2: orbitofrontal cortex; Subject 3: inferior frontal
gyrus, pars orbitalis, lateral and posterior medial frontal
gyrus; Wilcoxon, threshold of P 5 0.05) was observed for
intentional relative to unintentional harm (Fig. 2A). The
evoked responses of these prefrontal regions, which presented early connectivity with the amygdala, featured late
(but
not
early)
stimulus-related
modulations
(Supplementary Fig. 5). We also assessed functional
connectivity among all recording sites at an early
window (T1: 0–500 ms). Again, compared with
E. Hesse et al.
Early detection of intentional harm in the human amygdala
BRAIN 2016: 139; 54–61
| 59
frontal regions for the intentional conditions (Wilcoxon Signed Rank Test, threshold of P 5 0.05 for each subject). Each colour represents a
different subject. (B) Brief Time Span Functional Connectivity. Significant correlations of intentional and unintentional conditions for each subject
in a 0–500 ms window. Lines and line-widths indicate t-values and their absolute magnitudes, respectively. Left: A significantly larger number of
connections (1–40 Hz) was found for the intentional conditions [P 5 0.01 bootstrapping: Subject 1 (S1): P = 6.3139 10 35, t = 42.54; Subject 2
(S2): P = 7.989 10 33, t = 35.24; Subject 3 (S3): P = 2.8841 10 10,t = 21.18]. Right: A significantly larger number of connections (broadband)
were found for the intentional conditions [Subject 1 (S1): P = 5.2694 ‘10 48, t = 42.35; Subject 2 (S2): P = 1.9606 10 11, t = 32.21; Subject 3
(S3): P = 2.7941 10 58, t = 43.95].
other associated regions corroborates the specificity of amygdala activity in intentionality attribution (Treadway et al.,
2014).
We also observed an early coupling of corticolimbic networks previously implicated in intentional harm processing
(Treadway et al., 2014) (Supplementary material). As detailed in the Supplementary material, this aligns with
extant models of social cognition (Moll and Schulkin,
2009; Fumagalli and Priori, 2012; Ibanez and Manes,
2012) and fast fronto-amygdala networks (Pessoa and
Adolphs, 2010). Using a novel method (wSMI) (King
et al., 2013), we showed enhanced information sharing at
relatively slow (1–10 Hz) frequencies during observation of
intentional actions. Because theta fronto-amygdaline oscillation is enhanced during aversive stimuli processing (Janak
and Tye, 2015), this may indicate a general role for lowfrequency coupling between these regions in the face of
aversion-inducing events. Moreover, the early increases in
frontotemporal connectivity suggest rapid spreading of
amygdala boosts to other regions. Such a claim is consistent
Downloaded from https://academic.oup.com/brain/article/139/1/54/2468818 by guest on 22 June 2022
Figure 2 Corticolimbic and frontotemporal tuning of intentional actions. (A) WSMI. Significant connections of the amygdala with
60
| BRAIN 2016: 139; 54–61
Funding
This work was partially supported by grants from
CONICET, CONICYT/FONDECYT Regular (1130920),
FONCyT-PICT 2012-0412, 2012-1309, and the INECO
Foundation. V.L. is sponsored by Fondecyt 1150241,
M.S. is sponsored by CONICET and the James
McDonnell Foundation 21st Century ScienceInitiative in
Understanding Human Cognition Scholar Award.
Supplementary material
Supplementary material is available at Brain online.
References
Ames DL, Fiske ST. Perceived intent motivates people to magnify
observed harms. Proc Natl Acad Sci USA 2015; 112: 3599–605.
Boubela RN, Kalcher K, Huf W, Seidel EM, Derntl B, Pezawas L,
et al. fMRI measurements of amygdala activation are confounded
by stimulus correlated signal fluctuation in nearby veins draining
distant brain regions. Sci Rep 2015; 5: 10499.
Brazdil M, Roman R, Urbanek T, Chladek J, Spok D, Marecek R,
et al. Neural correlates of affective picture processing–a depth ERP
study. Neuroimage 2009; 47: 376–83.
Catenoix H, Magnin M, Mauguiere F, Ryvlin P. Evoked potential
study of hippocampal efferent projections in the human brain.
Clin Neurophysiol 2011; 122: 2488–97.
Decety J, Cacioppo S. The speed of morality: a high-density electrical
neuroimaging study. J Neurophysiol 2012; 108: 3068–72.
Decety J, Cowell JM. The complex relation between morality and
empathy. Trends Cogn Sci 2014; 18: 337–9.
Decety J, Michalska KJ, Kinzler KD. The contribution of emotion and
cognition to moral sensitivity: a neurodevelopmental study. Cerebral
Cortex. 2012; 22: 209–20.
Escobar MJ, Huepe D, Decety J, Sedeno L, Messow MK, Baez S, et al.
Brain signatures of moral sensitivity in adolescents with early social
deprivation. Sci Rep 2014; 4: 5354.
Foster BL, Rangarajan V, Shirer WR, Parvizi J. Intrinsic and taskdependent coupling of neuronal population activity in human parietal cortex. Neuron 2015; 86: 578–90.
Fumagalli M, Priori A. Functional and clinical neuroanatomy of morality. Brain 2012; 135(Pt 7): 2006–21.
Gross J. Analytical methods and experimental approaches for electrophysiological studies of brain oscillations. J Neurosci Methods 2014;
228: 57–66.
Huebner B, Dwyer S, Hauser M. The role of emotion in moral psychology. Trends Cogn Sci 2009; 13: 1–6.
Ibanez A, Cardona JF, Dos Santos YV, Blenkmann A, Aravena P,
Roca M, et al. Motor-language coupling: direct evidence from
early Parkinson’s disease and intracranial cortical recordings.
Cortex 2013; 49: 968–84.
Ibanez A, Manes F. Contextual social cognition and the behavioral
variant of frontotemporal dementia. Neurology 2012; 78: 1354–62.
Jacobs J, Kahana MJ. Direct brain recordings fuel advances in cognitive electrophysiology. Trends Cogn Sci 2010; 14: 162–71.
Janak PH, Tye KM. From circuits to behaviour in the amygdala.
Nature 2015; 517: 284–92.
Kawasaki H, Kaufman O, Damasio H, Damasio AR, Granner M,
Bakken H, et al. Single-neuron responses to emotional visual stimuli
recorded in human ventral prefrontal cortex. Nat Neurosci 2001; 4:
15–6.
King JR, Sitt JD, Faugeras F, Rohaut B, El Karoui I, Cohen L, et al.
Information sharing in the brain indexes consciousness in noncommunicative patients. Curr Biol 2013; 23: 1914–9.
Krolak-Salmon P, Henaff MA, 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 2004; 42: 665–76.
Lim SL, Padmala S, Pessoa L. Segregating the significant from the
mundane on a moment-to-moment basis via direct and indirect
amygdala contributions. Proc Natl Acad Sci USA 2009; 106:
16841–6.
Manning JR, Jacobs J, Fried I, Kahana MJ. Broadband shifts in local
field potential power spectra are correlated with single-neuron spiking in humans. J Neurosci 2009; 29: 13613–20.
Moll J, Schulkin J. Social attachment and aversion in human moral
cognition. Neurosci Biobehav Rev 2009; 33: 456–65.
Moll J, Zahn R, de Oliveira-Souza R, Krueger F, Grafman J. Opinion:
the neural basis of human moral cognition. Nat Rev Neurosci 2005;
6: 799–809.
Naccache L, Gaillard R, Adam C, Hasboun D, Clemenceau S,
Baulac M, et al. A direct intracranial record of emotions
evoked by subliminal words. Proc Natl Acad Sci USA 2005; 102:
7713–7.
Omigie D, Dellacherie D, Hasboun D, George N, Clement S,
Baulac M, et al. An intracranial EEG study of the neural dynamics of musical valence processing. Cereb Cortex 2015; 25:
4038–47.
Oya H, Kawasaki H, Howard MA, 3rd, Adolphs R. Electrophysiological
responses in the human amygdala discriminate emotion categories of
complex visual stimuli. J Neurosci 2002; 22: 9502–12.
Downloaded from https://academic.oup.com/brain/article/139/1/54/2468818 by guest on 22 June 2022
with the findings that (i) activity flows from limbic to frontal
structures occur between 190–347 ms (Catenoix et al.,
2011); (ii) salient stimuli modulate cortical activity in early
epochs (100 ms) (Kawasaki et al., 2001); causing (iii) parallel distributed frontotemporal processing (Krolak-Salmon
et al., 2004; Brazdil et al., 2009) (Supplementary material).
Intracranial EEG recordings can provide novel and robust
results (Foster et al., 2015). Our results provide unprecedented spatiotemporal evidence for the role of amygdala in
the early encoding of intention, the subsequent categorization
of harmful events, and the automatic modulation of corticolimbic connections, all systematically observed in each subject.
Although intracranial measures provide a unique source of
information that cannot be obtained through non-invasive
methods, they also feature important limitations. We have
carefully dealt with the well-known caveats of intracranial
EEG research by adopting several precautions to minimize
the effects of pathological tissue in our signals
(Supplementary material). Finally, we could not presently
examine laterality differences in amygdala activations, an
issue that calls for further research (Supplementary material).
By overcoming the spatiotemporal limitations of previous
neuroimaging studies of the amygdala, the present results
help to clarify the ‘many roads’ view. Consistent with this
perspective, we observed early amygdala responses guided
by stimulus salience and rapid parallel coupling with other
regions. Nevertheless, our results also highlight the amygdala’s critical role in automatically encoding and classifying
intentional harm during moral evaluation—functions that
no other region seems to subserve.
E. Hesse et al.
Early detection of intentional harm in the human amygdala
Pessoa L, Adolphs R. Emotion processing and the amygdala: from a
‘low road’ to ‘many roads’ of evaluating biological significance. Nat
Rev Neurosci 2010; 11: 773–83.
Treadway MT, Buckholtz JW, Martin JW, Jan K, Asplund CL,
Ginther MR, et al. Corticolimbic gating of emotion-driven punishment. Nat Neurosci 2014; 17: 1270–5.
BRAIN 2016: 139; 54–61
| 61
Tukey JW. Exploratory data analysis. Reading, MA: Addison-Wesley
Pub. Co.; 1977.
Young L, Cushman F, Hauser M, Saxe R. The neural basis of the
interaction between theory of mind and moral judgment. Proc Natl
Acad Sci USA 2007; 104: 8235–40.
Downloaded from https://academic.oup.com/brain/article/139/1/54/2468818 by guest on 22 June 2022