Graph theory reveals amygdala modules consistent with its anatomical subdivisions

Abstract

Similarities on the cellular and neurochemical composition of the amygdaloid subnuclei suggests their clustering into subunits that exhibit unique functional organization. The topological principle of community structure has been used to identify functional subnetworks in neuroimaging data that reflect the brain effective organization. Here we used modularity to investigate the organization of the amygdala using resting state functional magnetic resonance imaging (rsfMRI) data. Our goal was to determine whether such topological organization would reliably reflect the known neurobiology of individual amygdaloid nuclei, allowing for human imaging studies to accurately reflect the underlying neurobiology. Modularity analysis identified amygdaloid elements consistent with the main anatomical subdivisions of the amygdala that embody distinct functional and structural properties. Additionally, functional connectivity pathways of these subunits and their correlation with task-induced amygdala activation revealed distinct functional profiles consistent with the neurobiology of the amygdala nuclei. These modularity findings corroborate the structure-function relationship between amygdala anatomical substructures, supporting the use of network analysis techniques to generate biologically meaningful partitions of brain structures.

Full text

1
SCIENTIfIC REPORTS | 7: 14392 | DOI:10.1038/s41598-017-14613-4
www.nature.com/scientificreports
Graph theory reveals amygdala
modules consistent with its
anatomical subdivisions
Elisabeth C. Caparelli1, Thomas J. Ross
1, Hong Gu1, Xia Liang1,2, Elliot A. Stein1 &
Yihong Yang1
Similarities on the cellular and neurochemical composition of the amygdaloid subnuclei suggests
their clustering into subunits that exhibit unique functional organization. The topological principle
of community structure has been used to identify functional subnetworks in neuroimaging data that
reect the brain eective organization. Here we used modularity to investigate the organization of
the amygdala using resting state functional magnetic resonance imaging (rsfMRI) data. Our goal was
to determine whether such topological organization would reliably reect the known neurobiology of
individual amygdaloid nuclei, allowing for human imaging studies to accurately reect the underlying
neurobiology. Modularity analysis identied amygdaloid elements consistent with the main anatomical
subdivisions of the amygdala that embody distinct functional and structural properties. Additionally,
functional connectivity pathways of these subunits and their correlation with task-induced amygdala
activation revealed distinct functional proles consistent with the neurobiology of the amygdala
nuclei. These modularity ndings corroborate the structure–function relationship between amygdala
anatomical substructures, supporting the use of network analysis techniques to generate biologically
meaningful partitions of brain structures.
e amygdala is involved in a series of emotional processes and cognitive functions including learning, memory,
attention and perception1. Although small in total size, its complex composition of structurally and functionally
heterogeneous subnuclei supports these multiple functions. Histological studies suggest that the amygdala may
be composed of at least twenty subnuclei2 that share cytoarchitecture, myeloarchitecture and chemoarchitecture
features. Based on these similarities, schemas have been proposed to segregate these nuclei into subunits based on
their functional and anatomical characteristic3. Although the boundaries and even the names of these subunits
remain unsettled, one of the most widely accepted classication schemas classies them as the supercial (SF)
(corticoid) amygdaloid nucleus, the centromedial (CM) group and the laterobasal (LB) complex4–6. A stereotaxic
probabilistic map of the human amygdala, based on the superposition of cytoarchitectonic mapping of cell-body
stained histological sections, has been developed by Amunts and colleagues5 and classies the SF group to include
the anterior amygdaloid area, the amygdalopyriform transition area, the amygdaloid-hippocampal area and cor-
tical nuclei (intermediate, dorsal, ventral and posterior), the CM group combines the central and medial nuclei,
while the LB group incorporates the lateral, basolateral, basomedial and paralaminar nuclei. is mapping is quite
consistent with the cellular and neurochemical composition of the amygdaloid nuclei, since the CM, which is
continuous with the bed nucleus of stria terminals, is comprised of cells that are morphologically similar to those
in the striatum, while the LB and SF resemble cortical area7–10. Additionally, the LB is the main amygdala division
that receives eerent projections from basal forebrain cholinergic neurons projecting mostly to the basolateral
nucleus4,11.
e neurobiological composition of these main subdivisions indicates a potentially exclusive functional/
anatomical neural interaction that may be captured either in synchronized uctuation of the fMRI signal, as
in rsfMRI, or through anatomical connections, such as, diusion tensor imaging (DTI). For this reason, dier-
ent functional and structural MRI based approaches have been proposed to parse the amygdala architecture.
As a result, while some studies found the amygdala subdivided in two main regions, which were identied as
superior12/supercial-cluster13 and inferior12/deep-cluster13, representing the LB and smaller nuclei (inferior/
1Neuroimaging Research Branch, National Institute on Drug Abuse, National Institutes of Health, Baltimore,
Maryland, USA. 2Research Center of Basic Space Science, Harbin Institute of Technology, Harbin, China.
Correspondence and requests for materials should be addressed to E.C.C. (email: elisabeth.caparelli@nih.gov)
Received: 5 April 2017
Accepted: 4 September 2017
Published: xx xx xxxx
OPEN
Content courtesy of Springer Nature, terms of use apply. Rights reserved

www.nature.com/scientificreports/
2
SCIENTIfIC REPORTS | 7: 14392 | DOI:10.1038/s41598-017-14613-4
deep-cluster) and the CM and cortical nuclei (superior/supercial-cluster)14, others found the amygdala subdi-
vided into four main subregions, dividing LB into lateral and basal regions and the CM into central and medial,
leaving the cortical nuclei undierentiated15. erefore, these ndings are not fully in agreement with the amyg-
dala subdivisions shown in the stereotaxic probabilistic map5.
Two recent studies16,17, using imaging approaches, found three subdivisions of the amygdala forecasted by
the probabilistic atlas5. e rst study proposed a method that started from a hypothetical topographic model,
based on the functional-anatomic organization of brain networks subserving social cognition, by dening three
social networks: perception, which would be associated with the ventrolateral sector of the amygdala; aliation,
related to the medial section of the amygdala; and aversion, linked with the dorsal sector of the amygdala16.
As a result, three amygdala regions were pre-dened in those locations (spherical ROIs) and using an iterative
seed-target-seed methodology, the border of these subdivisions were established. e second study proposed
an imaging-parcellation method based on meta-analysis, where three amygdala subdivisions were identied by
computing whole-brain co-activation patterns for each amygdala seed voxel and grouping these seeds based on
similarities between their co-activation patterns17. is method required a systematic analysis of a broad range
of associated experiments (6,500 fMRI and PET studies) to increase the robustness of parcellation results and to
avoid dependence on any particular user-specied parameter. erefore, while the rst approach16 predene the
initial number of subdivisions, the second approach was a meta-analysis requiring a signicant amount of data
from dierent imaging sources.
In this work, we propose to use resting state BOLD data and the topological principle of community structure
to identify the main subdivisions of the amygdala. Our approach is based on global topological characteristics
that has been shown to produce stable brain structural and functional networks, suggesting that the observed
hierarchical modular organization coincide with the underlying anatomical communities at dierent scales18.
Furthermore, by using hierarchical modularity analysis to explore the community structure of brain networks19,20,
previous studies have shown that functionally and anatomically related brain regions are more densely inter-
connected, with relatively few connections between these clusters, indicating that these networks are intimately
related and share common topological features, such as modules and hubs21. erefore, based on these concepts,
we employed the principle of modularity to identify subnetworks within the amygdala. We hypothesized that the
modular organization of the amygdala will reect its internal functional segregation and integration consistent
with the neurobiological properties of its nuclei. For this purpose, we applied the community structure algorithm
to a high spatial and temporal resolution rsfMRI data from the Human Connectome Project (HCP)22,23. We fur-
ther assessed the functional consequence of each identied subnetwork by determining their functional connec-
tivity and then by correlating these circuits with amygdala activation during an emotional processing task. Our
ndings validate the association between the modular structure of the amygdala and the biological characteristics
of amygdala subdivisions.
Results
Modularity. Modular analysis of the amygdala revealed three distinct modules (Fig.1A and B) that are con-
sistent with the probabilistic anatomical subregion maps5, displayed in the Juelich histological atlas (50% proba-
bilistic mask, FSL - Fig.1C). However, the borders of the atlas, as implemented in FSL, do not completely coincide
with macro-anatomical landmarks of the amygdala5 (Fig.1D; Juelich atlas volume, le: 2968 mm3, right: 2904
mm3 6, making direct comparison of the modularity results and the anatomical subdivisions dicult, since the
amygdala template is smaller (le: 1608 mm3, right: 1512 mm3), but better represents the average amygdala size in
healthy adults24. Despite size discrepancies, the synergy between the relative size of the subdivisions found in the
Juelich atlas (LB > SF > CM) and the modules obtained from our modularity approach (lateral > medial > dor-
sal), together with the closeness between the subdivision location, suggest labeling the dorsal module as CM, the
medial as SF and the lateral as LB.
FC maps. Patterns of signicant correlation for each amygdala subdivision reveal unique connectivity circuits.
MRI signal at the CM was mainly correlated with the signal in the middle and anterior cingulate cortices, frontal
cortex, striatum, insula, cerebellum and precuneus, while a negative correlation was observed at the occipital
gyrus (Fig.2, Table1). However, an overall reduction in the connectivity strength was observed for the right when
compared with the le CM, which was not seen for other amygdala subdivisions (Figs2 and 2S). In spite of this
reduction in strength, the connectivity pattern is preserved between the two sides of this amygdala subdivision.
Direct comparison with correlation results from the other seed regions highlight this unique path of connections
when contrasting the functional connectivity results of the CM with either the combined results from LB and SF
(Fig.3, Table2) or the connectivity results from each of the others subdivisions (Fig.1S, Table1S).
e LB nuclei was positively correlated with the superior, medial and inferior frontal gyri, precentral gyrus,
paracentral lobule, middle temporal gyrus and cerebellum, and negatively correlated with the parietal lobule,
precuneus and cingulate gyrus (Fig.2, Table1). Further, when compared with the circuit strength from the other
two amygdala subdivision, the LB seems to be uniquely associated with inferior and middle temporal gyrus and
middle occipital gyrus (Fig.3, Table2). Pairwise comparisons also show unique correlations of LB with the infe-
rior temporal gyrus and the middle frontal gyrus when contrasted with CM and with cerebellum, and the middle
frontal gyrus and the inferior parietal lobule when compared with SF (Fig.1S, Table1S).
Activity in the SF correlated with signal uctuations in the orbitofrontal cortex, posterior insula, olfactory cor-
tex, precentral gyrus, paracentral lobule and posterior cingulate, and was anti-correlated in the anterior cingulate
cortex, superior and middle frontal gyrus, cerebellum, parietal lobule, inferior temporal gyrus and precuneus
(Fig.2, Table1). Unique connectivity path of the SF subdivision is observed at the paracentral lobule, posterior
cingulate cortex (PCC) and orbitofrontal cortex (OFC) when contrasting with the combined correlation results
for the other two subdivisions (Fig.3, Table2).
Content courtesy of Springer Nature, terms of use apply. Rights reserved

www.nature.com/scientificreports/
3
SCIENTIfIC REPORTS | 7: 14392 | DOI:10.1038/s41598-017-14613-4
Figure 1. Modularity results for the amygdala subdivisions. (A) Correlation matrix for the amygdala template
with modules overlaid in semi-transparent colors, greyscale values indicate r-values, (B) modularity results: LBL
(696 mm3), LBR (640 mm3), SFL (472 mm3), SFR (464 mm3), CML (440 mm3), CMR (408 mm3), are displayed
on coronal slices located at y-axis values: -2, -4, -6, -8 (L/R sux = le/right). e maximum modularity factors
obtained for this parcellation were QL = 0.26, QR = 0.25; (C) Juelich atlas overlaid on a coronal anatomical image
(D) Juelich atlas superimposed on the amygdala template (shown in white, under the Juelich atlas); radiological
convention.
Figure 2. FC maps for the amygdala subdivisions, CM, LB and SF, obtained from modularity results. MNI
standard space; radiological convention; signicance: p < 0.05 FWE corrected. Le/Right indicate seed locations.
Content courtesy of Springer Nature, terms of use apply. Rights reserved

www.nature.com/scientificreports/
4
SCIENTIfIC REPORTS | 7: 14392 | DOI:10.1038/s41598-017-14613-4
Subdivision x y z Brain region cluster size T-score
CML
−24 −6−14 le Amygdala 71314 >13
22 −80 28 right Superior occipital gyrus 477 −5.9
−14 −92 24 le Superior occipital gyrus 397 −5.9
CMR
24 −2−14 right Amygdala 24511 >15
−8 90 36 le cuneus 2277 −6.6
−16 −68 −46 le cerebellum (VIIIa) 2244 6.8
2 58 28 le superior frontal gyrus 1488 7.7
0−4 48 cingulate Gyrus 1394 7
16 −72 −52 right cerebellum (VIIb) 533 7.6
4−62 32 right precuneus 355 6.5
50 −46 −24 right inferior temporal gyrus 293 5.7
LBL
−28 −8−20 le amygdala 58058 >13
0 58 36 superior frontal gyrus 1383 8.8
50 −50 56 right inferior parietal lobule 856 −7.6
4 34 38 right medial frontal gyrus 658 −7
2 52 −16 right Mid. Orbital gyrus 629 8.1
−8−70 34 le precuneus 525 −7.6
−38 −56 46 le inferior parietal lobule 473 −7.6
2−22 30 right cingulate gyrus 454 −8.7
40 10 24 right inferior frontal gyrus 398 6.1
58 28 2 right inferior frontal gyrus 307 6
LBR
30 −6−22 right amygdala 42326 >13
0 58 −10 medial frontal gyrus 2564 9.2
32 −82 −36 right cerebellum (Crus I) 931 8.7
50 −46 42 right inferior parietal lobule 870 −6.6
−26 −82 −36 le cerebellum (Crus I) 665 9.2
2 32 40 right medial frontal gyrus 591 −7.6
−6−72 38 le precuneus 387 −6.6
0−26 28 cingulate gyrus 310 −6.6
SFL
−16 −6−18 le amygdala 68153 >13
38 −52 −32 right cerebellum (Crus I) 3118 −11.5
30 56 18 right superior frontal gyrus 2793 −8.8
−32 58 16 le middle frontal gyrus 1434 −9.1
2 28 40 right medial frontal gyrus 1286 −9.1
54 −44 42 right inferior parietal lobule 1087 −10.2
−40 −30 −10 le inferior temporal gyrus 793 −7
−8−68 38 le precuneus 733 −8.4
−50 −52 40 le inferior parietal lobule 641 −7.3
36 −46 0 right middle temporal gyrus 623 −7.4
4−24 28 right cingulate gyrus 451 −8.5
14 −12 6 right t halamus 308 −7.5
SFR
20 −6−20 right amygdala 65553 >13
−40 −60 −30 le cerebellum (Crus I) 1579 −9.7
2 38 22 right anterior cingulate 1541 −8.4
−30 54 24 le middle frontal gyrus 1335 −8
30 56 28 right superior frontal gyrus 1166 −7.7
26 −70 −26 right cerebellum (VI) 1050 −8.8
56 −44 36 right supramarginal gyrus 916 −8.7
−54 −48 36 le inferior parietal lobule 895 −9.4
2−22 28 right cingulate gyrus 639 −10.5
−8−72 40 le precuneus 303 −8.5
46 18 0 right inferior frontal gyrus 302 −6.6
12 −70 42 right precuneus 297 −10
−14 −12 10 le thalamus 290 −7.2
Table 1. Cluster locations for the functional connectivity maps calculated for each amygdala subdivision.
T-test results from partial correlation for each subset are corrected for multiple comparisons, pcorrected < 0.01;
cluster’s peak locations are in Montreal Neurological Institute (MNI) coordinates, locations for the cerebellum
subdivisions are from MNI-SUIT space (AFNI-SUIT atlas92), cluster size in number of voxels, XL(R) = X
le(right), X = CM, LB, SF.
Content courtesy of Springer Nature, terms of use apply. Rights reserved

www.nature.com/scientificreports/
5
SCIENTIfIC REPORTS | 7: 14392 | DOI:10.1038/s41598-017-14613-4
Pairwise comparisons show that the connectivity strength of the SF subdivision is higher when compared
with CM based circuits in the occipital lobe, PCC, precentral and postcentral gyrus, paracentral lobule, superior
frontal gyrus and OFC and higher than the connectivity path of the LB at medial and inferior frontal gyrus,
precuneus, PCC, precentral and postcentral gyrus, middle temporal gyrus, precentral gyrus and OFC (Fig.1S,
Table1S).
Connectivity across sessions. A two-way ANOVA (session x seed location), comparing the connectivity
pattern obtained for each amygdala subunit and for each session (subseries), revealed a main eect of the seed
location in the striatum, insula, OFC, ACC, superior medial gyrus, occipital gyrus, medial, middle and inferior
frontal gyrus, cingulate gyrus, inferior parietal lobule, cerebellum, superior and middle temporal gyrus (Fig.3SA,
Table3). ere was no main eect of session and no interaction between seed location and session. ICC test-retest
results of brain connectivity data also show fair to moderate reliability for the functional connectivity across ses-
sions (ICC > 0.4). Intersession consistency was highest from the FC maps obtained for the SF, followed by those
from CM and LB (Fig.3SB).
Altogether, these results indicate a stable pattern of brain connectivity for each modularity based amygdala
substructure across sessions (Fig.S2) that generally reproduced those obtained with the entire time series (Fig.2),
highlighting the consistency of the connectivity over time.
Amygdala activation and rsfMRI. As expected, the entire amygdala was signicantly activated when sub-
jects performed the face/shape matching task (Fig.4A1). However, raising the threshold (T > 13) indicated that
the most signicant activated voxels were located within the CM and the SF subdivisions (Fig.4A2). Average
BOLD signal in each amygdala subdivision shows the LB as the least activated during the fear/anger faces task,
while the supercial subdivision was the most active during the task (Fig.4B); there were no signicant dier-
ences in activation between subdivisions in the le and right hemisphere. Notably, only the CM showed sig-
nicant correlation between the average BOLD signal in this amygdala subdivision and its mean connectivity
strength with the entire brain (Figs4C and 5S), while no signicant correlation between BOLD and correspond-
ing connectivity maps was observed for the others two amygdala subdivisions.
Discussion
By using the community structure algorithm to evaluate the topological organization of the amygdala, we iden-
tied three subdivisions, which is congruent with previous ndings16,17, but with a method that did not require
an extended meta-analysis. Moreover, modularity has shown to be very reliable, even with small datasets; the
results from the HCP dataset (98 subjects) reproduced our previous pilot ndings from a group of just 18 subjects,
acquired in our center with dierent EPI sequences and dierent set of parameters (see Supplemental Material
Fig.4S). Additionally, our method is free of initial hypotheses related to the number of modules, which is auto-
matically established by the algorithm based on the principle of modularity. Finally, our method was able to
identify the CM and SF based only on information about node connections, despite the poor separation of these
modules as illustrated in Fig.1A, which may explain why previous data driven work12,13 were not able to separate
the CM from SF.
Our work further supports that topological properties may have an impact on the study of structural-function
relations in brain networks. Whilst it helps to understand the fundamental architecture of connections within
and between brain regions, it also provides a way to elucidate how this architecture supports neurophysiological
dynamics. Modularity approaches can be either top-down, such as the Newman’s method25, where a network
is repeatedly subdivides into smaller portions until there is nothing to be gained in modularity value (Q), or
bottom-up, as in Louvain’s method26,27, which is a hierarchical clustering approach, where nodes are progressively
merged to others nodes becoming larger and larger sets of nodes, a process that is repeated iteratively until no
increase of modularity is possible28. By using the Louvain’s method, we were able to identify amygdala modules
that resemble the most accepted classication schemes for its anatomical subdivision5, suggesting a degree of
functional segregation of these subunits that is consistent with the biological characteristics of the amygdala
substructures.
Dierent methods have been proposed to nd small communities in large networks, such as the Surprise
method29 and Optimal Compression method30, each having particular strengths and weaknesses. For example
the Surprise method29 is best suited for the study of binarized networks. However, it is vulnerable to noise errors
that aect small modules, and in the case of the amygdala, located in a brain area with low SNR, it may not
necessarily improve the quality of the partitions. e Optimal Compression method30, on the other hand, is not
recommended when “community structure are considered as statistical deviation from the null model in which
the degree sequence is held constant but links are otherwise equiprobable among all nodes, wherein the modularity
optimization methods by denition provides the optimal partitioning”30. In contrast, by construction, the Louvain
method26 is able to unfold a complete hierarchical community structure for the network, with each level of the
hierarchy being given by the intermediate partitions found at each pass. In our work, we considered only the top
level of this hierarchy, namely the nal partition found by the algorithm, since no further meaningful sub-module
was found in the amygdala (additional sub-modules mostly consisted of single voxels, as tested on dataset from
Fig.4S) when lower levels of modular hierarchy were evaluated.
Regional connectivity. Temporally correlated patterns of low-frequency uctuations during rest revealed
regions of distinct functional networks associated with each of the three amygdala subdivisions, which were
maintained across sessions, demonstrating the robustness and stability of our ndings. Notably, their combi-
nation reproduces the previously reported16,31 connectivity pattern for the entire amygdala, as areas from the
Content courtesy of Springer Nature, terms of use apply. Rights reserved

www.nature.com/scientificreports/
6
SCIENTIfIC REPORTS | 7: 14392 | DOI:10.1038/s41598-017-14613-4
ventro-temporal-limbic network (described by Glerean and colleagues31) are identied in the connectivity path-
way of the amygdala subdivisions.
Signal variation at the LB subdivision was positively correlated with activity in the superior temporal gyrus,
middle frontal and precentral gyri, supporting the involvement of similar circuits in associative learning and
emotion regulation32,33. In particular, the connection between the LB and the temporal lobes was previously
observed using rsFC6,12,16 and probabilistic DTI34, and is supported by axonal tracing studies in non-human pri-
mates, where a connection between the superior temporal gyrus and the ventrolateral nucleus of the amygdala
was identied35.
Signicant correlation was found between the SF region and posterior insula, OFC, subgenual ACC, inferior
parietal lobule, olfactory cortex, middle temporal gyrus and PCC. is correlation pattern suggests an involve-
ment of SF in olfaction-related aective processes, consistent with a meta-analysis of human imaging data17 and
rat anatomy36,37. Similarities of the connectivity pattern of the LB and SF mainly with the temporal lobe38, suggests
that these two amygdaloid nuclei may share a common functional organization. Comparative architectonic stud-
ies have also suggested a separation of the SF from the CM group, assigning it to the LB group39, which is justied
by the fact that these two nuclei also contain similar cortical –like neuronal composition7,9.
Despite the biological resemblance between the SF and LB, we also observed unique connectivity patterns
between SF and areas of the default mode network (DMN), although anatomical projections from the amygdala
to the PCC has not been veried40, suggesting a possible indirect connection between these regions through the
sACC, as previously reported in monkeys41. Nevertheless, a functional connection between the amygdala and
the PCC has also been reported42, suggesting an involvement of SF in self-awareness related emotional processes,
autobiographical memory, past self-relevant stimuli and future prospection43. e SF was also anti-correlated
with the more dorsal part of the precuneus, which is consistent with the pattern of positive correlations, pre-
viously observed, between the DMN and the ventral, but not the dorsal precuneus44. In addition, a decrease in
regional cerebral blood ow in the PCC, medial frontal cortex and ventral but not dorsal precuneus during a
working memory task45, suggests that dorsal precuneus (BA 7) may not be part of the DMN46. Furthermore, the
connection between the SF and the precuneus may corroborate an amygdala-precuneus pathway that has been
anatomically observed in tract tracing studies in monkeys47,48. Lastly, our ndings showing the SF amygdala as
the most activated in response to an emotional task that elicited either fear or anger, corroborates its involvement
in directing attention towards aective stimuli49.
e signicant correlation between resting activity of the CM nuclei and signal uctuation from areas attrib-
uted to the salience network (SN), e.g., anterior insula, ACC and middle cingulate cortex (MCC)50, is consistent
with the central nucleus involvement in facilitating attention to salient stimuli51. In addition, the CM was the
only amygdala subdivision in which the rsFC pattern predicted activation in response to fear and angry faces,
thus indicating its association with adverse feelings. Even though the SF had the highest average activation for
this task, its BOLD signal did not correlate with the mean functional connectivity pattern of this subdivision.
Moreover, the most activated voxels occurred not only in the SF, but also in the CM subdivision. erefore, our
Figure 3. FC-dierential maps contrasting the connectivity map of each amygdala subdivision, CM, LB and
SF, against the average connectivity pattern of the other two subdivisions, LB + SF, CM + SF, CM + LB. MNI
standard space; radiological convention; signicance: p < 0.05 FWE corrected.
Content courtesy of Springer Nature, terms of use apply. Rights reserved

www.nature.com/scientificreports/
7
SCIENTIfIC REPORTS | 7: 14392 | DOI:10.1038/s41598-017-14613-4
ndings suggest an involvement of CM and its connected areas (SN and striatum) with negative emotion and is
consistent with previous ndings. For example, the strongest amygdala functional connectivity within the SN is
seen in those with the greatest amygdala activation to and aroused by negative pictures52, which is sustained by
the role of the amygdala, SN and ventral striatum on the identication of the emotional signicance of environ-
mental stimuli and the production of aective states53. Multiple fMRI studies have also reported activation in the
ACC/MCC in anger and fear emotional tasks54, while monkey studies show that this part of the cingulate cortex
receives input from the amygdala41,55 and has been implicated in fear56. e anterior insula has also been linked
with such emotional experiences as fear and anger57, supported by a projection from the central nucleus of the
amygdala35. e striatum has been associated with anger58, such that functional connections between the CM
and the striatum are enhanced during stress59. e CM participates not only in the expression of conditioned
fear10 but also in the learning and consolidation of fear conditioning60. Experiments in rats show that lesions of
the central, medial and cortical amygdala nuclei markedly increase the number of contacts a rat will make with a
sedated cat, demonstrating a decrease of fear in these lesioned animals61. e involvement of CM in anger has also
Subdivision x y z Brain region cluster size Z-score
CML > LBL + SFL
−18 −8−10 le amygdala 33154 >13
10 −28 74 right paracentral lobule 20900 −10.0
0 26 32 le middle cingulate cortex 4865 10.6
−36 54 20 le middle frontal gyrus 1691 8.0
−68 −22 24 le postcentral gyrus 1656 8.7
30 40 28 right middle frontal gyrus 1398 6.5
68 −32 36 right supramarginal gyrus 1195 6.8
28 −6−20 right hippocampus 817 −11.2
CMR > LBR + SFR
48 −74 2 right middle temporal gyrus 23328 −8.2
30 −4−24 right amygdala 6105 <−13
−52 −48 8 le middle temporal gyrus 2288 −7.3
60 −4−10 right superior temporal gyrus 2129 −7.2
−36 −46 −34 le cerebellum (VI) 2122 7.4
2 42 20 le anterior cingulate cortex 1498 7.1
LBL > CML + SFL
−34 −4−26 le amygdala 12381 >13
0 4 30 le anterior cingulate cortex 1870 −7.0
30 −84 20 right middle occipital gyrus 1487 6.0
−18 −92 16 le middle occipital gyrus 1079 5.4
LBR > CMR + SFR 34 −2−24 right amygdala 2731 >13
−30 −8−20 le hippocampus 1007 12.3
SFL > CML + LBL
−14 −4−18 le amygdala 13575 >13
−30 −70 −24 le cerebellum (VI) 9287 −10.1
0−20 62 le paracentral lobule 8517 10.1
−30 54 16 le middle frontal gyrus 7197 −9.2
0 40 −22 le rectal gyrus 2460 9.9
56 −66 22 right middle temporal gyrus 1549 7.1
−24 −80 34 le superior occipital gyrus 1470 6.9
16 −104 −4right calcarine gyrus 1390 −6.1
0−54 18 le precuneus 1186 7.1
58 −44 36 right supramarginal gyrus 922 −7.6
−62 −42 30 le supramarginal gyrus 832 −6.7
SFR > CMR + LBR
12 −20 76 right paracentral lobule 13240 10.1
14 −2−16 right amygdala 11217 >13
−10 −54 6 le calcarine gyrus 3912 8.3
−20 −82 −30 le cerebellum (Crus I) 1793 −8.8
26 −70 −26 right cerebellum (VI) 1517 −8.1
−10 30 26 le anterior cingulate cortex 1428 −6.7
−12 −16 10 le thalamus 885 −7.1
−56 −48 36 le inferior parietal lobule 828 −7.3
40 24 10 right inferior front al gyrus 803 −6.4
Table 2. Locations of the signicant clusters obtained from the direct comparison of the functional
connectivity pattern of each amygdala subdivision with the average connectivity pattern of the other two
subdivisions. Results are corrected for multiple comparisons, pcorrected < 0.01; cluster’s peak locations are in MNI
coordinates, cluster size in number of voxels, XL(R) = X le (right), X = CM, LB, SF.
Content courtesy of Springer Nature, terms of use apply. Rights reserved

www.nature.com/scientificreports/
8
SCIENTIfIC REPORTS | 7: 14392 | DOI:10.1038/s41598-017-14613-4
been demonstrated in cats, with stimulation of the central nucleus suppressing defensive rage, while stimulation
of the medial nucleus enhances aggressive behavior62. Overall, our ndings reinforce the previously observed
participation of the CM and connected areas (SN and striatum) in fear and anger.
Validation and reproducibility. e functional connectivity circuits for the modularity dened amygdala
subdivisions reproduce the major ndings obtained by Bickart et al.16 and by Roy et al.6, which used the Juelich
atlas as a template. Both found that the CM nucleus correlated with the striatum and insula, that the LB complex
correlated with the temporal lobe, and activity of the SF mainly correlated with activity of the OFC. Additionally,
Roy and colleagues6 also reported laterality dierences of the CM connections, as observed here, that is consistent
with those previously reported63. For example, right amygdala lesions in rats have been shown to generate greater
decits in contextual fear than le sided lesions64. Previous electrophysiological studies also showed hemispheric
lateralization of pain processing by CM neurons. Specically, neurons in the le latero-capsular division of the
central nucleus of the amygdala (CeLC) did not develop increased responsiveness in a rodent model of arthritis
pain, while the right CeLC played a major role in the processing of prolonged nociceptive inputs and develops
sensitization65. is suggests a tonic inhibitory mechanism from the prefrontal cortical areas over the le CeLC,
which exert a top-down inhibitory inuence on the amygdala66. In our work, the lower connectivity strength
observed for the CM, besides being consistent with previous work, may also indicate a diminished inhibitory
eect from the prefrontal cortical areas over the right CM, since this was the amygdala subregion most connected
to these cortical areas. is may explain the unique correlation of its connectivity pattern and the activity on this
area during the faces task, once more being consistent with a predominant role of the right amygdala in negative
emotions. Finally, while the anatomical and functional basis for this lateralization remains unclear, taken together,
these ndings highlight the role of the CM on amygdala lateralization.
e functional connectivity between the LB and SF with the SMA dier from the results obtained by Roy and
colleagues6, although they are consistent with another resting state study that identied connectivity between
the LB and SF with motor areas38. Structural connections between the amygdala and motor areas has also been
previously observed in a DTI study34. Studies in cats67 and monkeys68 have shown that the projections to the
motor system arise from the magnocellular division of the basal nucleus of the amygdaloid complex. In addition,
disruption of the amygdala-motor pathway has been suggested to be responsible for the inability of those with
Autism to react to social stimuli69, which is consistent with data from both monkeys70 and humans71 reporting
that the LB and SF subregions are especially sensitive to social stimuli. erefore, together with previous ndings,
our work indicate that a direct amygdala-motor pathway might provide a mechanism by which the amygdala can
inuence more complex motor behaviors34.
Finally our results beneted from using high spatial and temporal resolution data from the HCP22. e high
spatial resolution improves the assessment of small brain regions, such as the amygdala, by limiting partial vol-
ume eects, reducing dephasing artifacts and improving in-plane MRI signal uniformity72, while the high scan
rate has the further advantage of minimizing the aliasing of physiological artifacts over the low frequencies of
interest73. Even though smaller voxels always carry the disadvantage of lower SNR, it was compensated by the
large sample size and multiple long rsfMRI runs.
Conclusions
In conclusion, we have applied a community structure approach to characterize the modular organization of the
amygdala using resting state fMRI data from the HCP. e anatomical and functional specicity of the results
provide compelling evidence of the structure–function association between the subnuclei identied herein and
x y z Brain region cluster
size F-score
−30 −8−24 right amygdala 10817 >20
2 30 32 le superior medial g yrus 6615 19
−50 −72 0 le midd le occipital gyrus 2498 13
16 −92 22 right cuneus 2333 12
2 38 −18 r ight medial frontal gyrus 1449 19
28 −84 −32 right cerebellum (Crus I) 1438 15
−40 −54 −30 Le cerebellum (Crus I) 1362 20
−28 52 16 le middle frontal gyrus 1233 13
30 54 18 right middle frontal gyrus 902 14
−58 −34 40 le inferior parietal lobule 827 9.7
−52 0 −20 le midd le temporal g yrus 780 12
62 −36 36 Right inferior parietal lobule 450 10
62 −4−10 right superior temporal gyrus 371 17
40 32 −14 Right inferior frontal gyrus 278 19
−2−28 26 Le cingulate gyrus 270 7.9
Table 3. Signicant clusters for the main eect of seed location obtained from the two-way ANOVA. Results are
corrected for multiple comparisons, pcorrected < 0.01; cluster peak locations are in MNI coordinates, locations for
the cerebellum subdivisions are from MNI-SUIT space (AFNI-SUIT atlas92), cluster size in number of voxels.
Content courtesy of Springer Nature, terms of use apply. Rights reserved

www.nature.com/scientificreports/
9
SCIENTIfIC REPORTS | 7: 14392 | DOI:10.1038/s41598-017-14613-4
histological anatomical substructures, demonstrating that graph theory based network analysis techniques can
generate biologically meaningful partitions of brain structures. Finally, our ndings suggest that characterization
of regional modular organization may be useful to evaluate disease- or age-induced abnormalities.
Methods
Subjects. Analyses were performed on 100 adult healthy volunteer (46 males and 54 females; age range 22 to
36 years; mean age 29.4 ± 3.6 years), randomly selected through the 100-unrelated option from the WU-Minn
Consortium HCP (WU-Minn HCP 500 Subjects + MEG2 Data Release), obtained through the Open Access
agreement. Data is publicly available at the HCP online database (http://www.humanconnectome.org/documen-
tation/S500/). All experiments were performed in accordance with the relevant guidelines and regulations of the
Human Connectome Project74. Detailed description on the standard operating procedures of the protocol is avail-
able at the WU-Minn HCP 500 Subjects + MEG2 Data Release: Reference Manual Appendix IV. Subject recruit-
ment procedures and informed consent forms, including consent to share de-identied data, were approved by
the Washington University (WU) institutional review board75. e use of the HCP data was approved by the
Oce of Human Subjects Research Protections at the NIH. All data presented in this paper is not identiable,
only group results are presented overlaid on a MNI template.
Data acquisition. MRI. Images were acquired on a 3 Tesla Skyra Siemens system using a 32-channel head
coil, a customized SC72 gradient insert (100 mT/m) and a customized body transmit coil. Resting state fMRI
Figure 4. (A) Amygdala activation for the threshold of T ≥ 6 (A1), location of the most signicant activated
voxels (T > 13, peak value T = 17) (A2) overlaid over the amygdala subdivisions, CM, LB and SF (color pattern
from Fig.1), radiological convention; (B) Average BOLD signal for each amygdala subdivision (*p < 0.0001);
(C) ROI analysis correlating the mean BOLD signal at the CMR with the mean positive and signicant
(p < 0.05) Fisher transformed correlation values of the CMR connectivity pathway.
Content courtesy of Springer Nature, terms of use apply. Rights reserved

www.nature.com/scientificreports/
10
SCIENTIfIC REPORTS | 7: 14392 | DOI:10.1038/s41598-017-14613-4
data were acquired in four runs on two dierent days (two runs per day); task-evoked fMRI was acquired in two
runs on the same day aer completion of the rsfMRI. Both used a multi-band gradient-echo EPI sequence22,76
(TE/TR 33.1/720 ms, resolution 2 mm isotropic, 72 oblique-axial slices, 1200 images/rsfMRI run, 176 images/
task-evoked fMRI run, MB acceleration factor = 8, BW = 2290 Hz/Pix). Within each day, acquisitions alter-
nated between phase encoding in the right-to-le (RL) direction in one run and the le-to-right (LR) direc-
tion in the other run. Anatomical images were acquired with a high-resolution (0.7 mm isotropic) T1-weighted
magnetization-prepared rapid gradient echo (3D-MPRAGE) sequence. Anatomical and functional imaging
sequences covered the whole brain. Subjects were given a simple instruction to rest and keep their eyes open with
relaxed xation on a projected bright cross-hair on a dark background (presented in a darkened room)77.
fMRI task – Emotion Processing. Participants performed a block design modied face-shape matching
task33 wherein they were asked to match which of two faces/shapes presented on the bottom of the screen with
the face/shape at the top of the screen; faces exhibited either an angry or fearful expression. Each of the two
runs includes 6 blocks (3 face blocks and 3 shape blocks), each block was composed of 6 trials of the same task
type (face or shape), with trials presented for 2000 ms and an inter-trial interval of 1000 ms. Shape blocks were
presented interleaved with face blocks and each block was preceded by a 3000 ms task cue (“shape” or “face”);
8 seconds of xation was presented at the end of each run77.
Data analysis. Preprocessing - 1st level fMRI analysis. e datasets underwent an initial preprocessing by the
HCP consortium. Anatomical images were distortion corrected, co-registered and averaged across runs, AC-PC
registered, brain extracted, B1 bias eld corrected and normalized to MNI152 space. FMRI runs (rsfMRI and
task-evoked) underwent gradient distortion correction, motion correction, registration to the T1w image, spatial
normalization to the MNI standard space, 4D global mean-based intensity normalization, B1 bias eld correction;
rsfMRI were also subject to independent component analysis (ICA)-based artifact removal. Finally, both task
and resting data were whole brain masked23,77. Local data processing was performed in AFNI78 and MATLAB
(e MathWorks Inc, Natick, Massachusetts). e pre-processed task-evoked fMRI time series were smoothed
(FWHM = 4 mm) to improve signal-to-noise ratio (SNR)79, censored for time points that exceed a Euclidean dis-
tance of 0.3 mm and then modeled in a xed eect analysis, two runs per subject, with canonical hemodynamic
responses time-locked to the shape-face epochs, contrasting face with shape condition; head motion parameters
(3 translations and 3 rotations) were also entered into the model as regressors of no interest. Time-series datasets
were scaled so that beta weights could be interpreted as percent signal-change. Each pre-processed rsfMRI run
had the rst four EPI volumes removed to ensure signal equilibrium, following by band-pass ltering (0.01–
0.10 Hz) to minimize instrument induced dris80 and physiological noise81. Multi-linear regression with the 6
time-varying realignment parameters was performed to minimize motion related uctuations in the MRI signals.
e rst three principal components (using PCA) of white matter (WM) and cerebrospinal uid (CSF) signals
were also regress out to preclude non-neuronal induced signal uctuations82,83. e WM and CSF masks were
generated by segmenting the preprocessed high-resolution anatomical images using SPM884 and down-gridding
the obtained masks to the same resolution as the pre-processed functional data. Individual subject rsfMRI time
series were smoothed (FWHM = 4 mm) to improve SNR79, and censored for time points that exceed a framewise
displacement (FD) threshold >0.5 mm and the root mean square variance across voxels (DVARS) > 0.5%85. Time
series with excessive number of time points censored (more than 30% for rsfMRI, 20% for task-evoked fMRI)
were discarded; two subjects were excluded from resting BOLD and another two from the task BOLD data.
Modularity. Modularity analysis of the amygdala was performed using a weighted-connectivity conserving null
model in fully connected, undirected networks with positive and negative weights27 from the Brain Connectivity
Toolbox (https://sites.google.com/site/bctnet/) on the remaining 98 subjects. More specically, Pearson correla-
tion coecient was computed for each voxel (node) inside of the anatomically pre-dened le and right amygdala
templates in MNI space (from Jerne Volumes of Interest database86,87), which was resampled to 2 mm isotropic
to reproduce the image resolution of the processed functional data. Individuals’ correlation matrices were then
averaged across subjects and runs to obtain the nal connectivity matrix, M. Following, aer removing the
auto-correlation values from M, a Louvain algorithm26, was used to nd the community aliation vector (C) cor-
responding to the undirected and weighted correlation matrix (M)27, accounting for the contribution of positive
and negative edges. e algorithm nds the optimal number of clusters by aggregating the nodes in the network
into groups of modules. More specically, Louvain method starts from a set of nodes that, through subsequent
passes of the algorithm, are clustered to others nodes becoming larger and larger sets of nodes (it is optimized by
allowing only local changes of communities); next the found communities are aggregated in order to build a new
network of communities, nally it is repeated iteratively until it reaches a number of modules that have a maximal
possible number of within group connections, and a minimal possible number of between-group connections88,
i.e., until the modularity factor, Q, is maximized26,27. erefore, no initial information regarding the number of
modules is needed. However, the algorithm is non-deterministic, and may produce dierent solutions each run,
so to address this limitation, the calculation was repeated 100 times (the calculation of Q is independent in each
repetition) and the maximum modularity factor, QM, was registered (by considering precision of 4 decimal places
only one Q value was found), then it was repeated another 100 times and another modularity factor was selected,
Qm; these two values were compared and if they are dierent the latest is kept to compare with the results from
the next iteration. is process was repeated until no change to the maximum Q was observed (QM = Qm). Note
that for every step (every QM and Qm) the number of modules remained constant varying only their size. Next the
nal community aliation, vector (C) was entered into the ne-tuning27 modularity algorithm, in order to rene
the modules borders. Here, again the process was repeated iteratively, as before, until no change in the maximum
modularity factor, Q, was observed; this allowed for the number of optimal partitions (modules) to be determined
Content courtesy of Springer Nature, terms of use apply. Rights reserved

www.nature.com/scientificreports/
11
SCIENTIfIC REPORTS | 7: 14392 | DOI:10.1038/s41598-017-14613-4
automatically. e modularity factor, Q, ranges between 0 and 1, and represents the goodness in which a network
is optimally partitioned into functional modules27.
Functional Connectivity (FC) maps. Seed-voxel partial correlation was used to calculate the FC maps. e nal
subdivisions of le and right amygdala were used as seeds and correlated with the entire brain for each subject,
regressing out the signal contribution from the neighbor seeds. e cross-correlation coecient maps were then
converted to z-score maps using the Fisher’s r-to-z transformation. Final FC maps for each subject were generated
by averaging the results across runs. In order to evaluate temporal variations in the connectivity pattern generated
for each amygdala seed, the original time-series were divided into four equal subseries (sessions) and FC maps
were calculated for each session and averaged across runs for each subject.
One sample t-tests were performed independently on the FC maps generated for each amygdala subset.
Dierential assessments of the FC maps were conducted to evaluate dierences in functional connectivity across
amygdala subdivisions by contrasting the connectivity path of one seed against the combination of the others
and by pairwise comparison (see results in the supplemental materials) using a linear mixed-eects modeling
approach (3dLME)89. e FC pathways for the dierent sessions (subseries) and each amygdala subdivision were
calculated with t-test, and the dierence between sessions across seed location (amygdala subdivisions) was eval-
uated with two-way analysis of variance (ANOVA) using 3dLME89. Results were corrected for multiple compari-
sons using 3dClustSim, based on Monte Carlo simulations90, including voxels of the entire brain and considering
a conservative imaging smoothing (FWHM) of 10 mm; statistical signicance was set for pcorrected < 0.05 (see
Supplemental material).
e reliability of the connectivity maps obtained for the four sessions (subseries) was also evaluated using a
two-way mixed single measures intra-class correlation (ICC(3,1))91; specically the between subject and residuals
mean square values were computed for each voxel using the function 3dICC_REML89 on the FC maps generated
for each amygdala subdivision, considering all sessions (see Supplemental material).
fMRI group analysis. One sample t-tests were performed on the beta weights generated for each subject pro-
viding a group activation map for the angry/fear faces contrasted against shapes. In addition, region-of-interest
(ROI) analyses were carried out on the beta weights using the modularity results as a mask to extract the average
BOLD signal in each amygdala subdivision for each subject. Results from the ROI analyses were compared with
the average of the positive and signicant (p < 0.05) Fisher transformed correlations obtained from the FC maps
generated from seed-voxel partial correlations for each amygdala subdivision. Both analyses were carried out for
the 96 subjects that survived head motion criteria for both resting and task-evoked datasets.
Availability of materials and data. All data analyzed in this manuscript were obtained from the
WU-Minn Consortium HCP (WU-Minn HCP 500 Subjects + MEG2 Data Release), which is publicly available at
the HCP online database (http://www.humanconnectome.org/). Detailed description of the analysis is included
in the methods section and in the supplemental material.
References
1. Baxter, M. G. & Murray, E. A. e amygdala and reward. Nat ev Neurosci 3, 563–573, https://doi.org/10.1038/nrn875 (2002).
2. Freese, J. L. & Amaral, D. G. In e human amygdala (eds P J Whalen & s E A Phelp) 3–42 (2009).
3. Swanson, L. W. & Petrovich, G. D. What is the amygdala? Trends Neurosci 21, 323–331 (1998).
4. Heimer, L. et al. e human basal forebrain Part II. e primate nervous system Part III. (1999).
5. Amunts, . et al. Cytoarchitectonic mapping of the human amygdala, hippocampal region and entorhinal cortex: intersubject
variability and probability maps. Anatomy and embryology 210, 343–352, https://doi.org/10.1007/s00429-005-0025-5 (2005).
6. oy, A. . et al. Functional connectivity of the human amygdala using resting state fMI. Neuroimage 45, 614–626 (2009).
7. Sah, P., Faber, E. S., Lopez De Armentia, M. & Power, J. e amygdaloid complex: anatomy and physiology. Physiol ev 83, 803–834,
https://doi.org/10.1152/physrev.00002.2003 (2003).
8. Puelles, L. oughts on the development, structure and evolution of the mammalian and avian telencephalic pallium. Philos Trans
 Soc Lond B Biol Sci 356, 1583–1598, https://doi.org/10.1098/rstb.2001.0973 (2001).
9. Heimer, L., Van Hoesen, G. W., Trinmble, M. & Zahm, D. S. Anatomy of Neuropsychiatry. 1–171 (2008).
10. LeDoux, J. e amygdala. Curr Biol 17, 868–874, https://doi.org/10.1016/j.cub.2007.08.005 (2007).
11. Unal, C. T., Pare, D. & Zaborszy, L. Impact of basal forebrain cholinergic inputs on basolateral amygdala neurons. J Neurosci 35,
853–863, https://doi.org/10.1523/jneurosci.2706-14.2015 (2015).
12. Mishra, A., ogers, B. P., Chen, L. M. & Gore, J. C. Functional connectivity-based parcellation of amygdala using self-organized
mapping: a data driven approach. Hum. Brain Mapping 35, 1247–1260, https://doi.org/10.1002/hbm.22249 (2014).
13. Bach, D. ., Behrens, T. E., Garrido, L., Weisopf, N. & Dolan, . J. Deep and supercial amygdala nuclei projections revealed in vivo
by probabilistic tractography. J Neurosci 31, 618–623, https://doi.org/10.1523/jneurosci.2744-10.2011 (2011).
14. Pitänen, A. Connectivity of the rat amygdaloid complex. 31–116 (Oxford University Press, 2000).
15. Saygin, Z. M., Osher, D. E., Augustinac, J., Fischl, B. & Gabrieli, J. D. Connectivity-based segmentation of human amygdala nuclei
using probabilistic tractography. Neuroimage 56, 1353–1361, https://doi.org/10.1016/j.neuroimage.2011.03.006 (2011).
16. Bicart, . C., Hollenbec, M. C., Barrett, L. F. & Dicerson, B. C. Intrinsic Amygdala-Cortical Functional Connectivity Predicts
Social Networ Size in Humans. J Neurosci 32, 14729–14741 (2012).
17. Bzdo, D., Laird, A. ., Zilles, ., Fox, P. T. & Eicho, S. B. An investigation of the structural, connectional, and functional
subspecialization in the human amygdala. Hum Brain Mapp 34, 3247–3266, https://doi.org/10.1002/hbm.22138 (2013).
18. Zhou, C., Zemanova, L., Zamora, G., Hilgetag, C. C. & urths, J. Hierarchical organization unveiled by functional connectivity in
complex brain networs. Physical review letters 97, 238103, https://doi.org/10.1103/PhysevLett.97.238103 (2006).
19. Ferrarini, L. et al. Hierarchical functional modularity in the resting-state human brain. Hum. Brain Mapping 30, 2220–2231, https://
doi.org/10.1002/hbm.20663 (2009).
20. Meunier, D., Achard, S., Morcom, A. & Bullmore, E. Age-related changes in modular organization of human brain functional
networs. Neuroimage 44, 715–723, https://doi.org/10.1016/j.neuroimage.2008.09.062 (2009).
21. Bullmore, E. & Sporns, O. Complex brain networs: graph theoretical analysis of structural and functional systems. Nature reviews.
Neuroscience 10, 186–198, https://doi.org/10.1038/nrn2575 (2009).
Content courtesy of Springer Nature, terms of use apply. Rights reserved

www.nature.com/scientificreports/
12
SCIENTIfIC REPORTS | 7: 14392 | DOI:10.1038/s41598-017-14613-4
22. Uğurbil, . et al. Pushing spatial and temporal resolution for functional and diusion MI in the Human Connectome Project.
Neuroimage 80, 80–104, https://doi.org/10.1016/j.neuroimage.2013.05.012 (2013).
23. Glasser, M. F. et al. e minimal preprocessing pipelines for the Human Connectome Proje ct. Neuroimage 80, 105–124, https://doi.
org/10.1016/j.neuroimage.2013.04.127 (2013).
24. Brierley, B., Shaw, P. & David, A. S. e human amygdala: a systematic review and meta-analysis of volumetric magnetic resonance
imaging. Brain es Brain es e v 39, 84–105 (2002).
25. Newman, M. E. Modularity and community structure in networs. Proc Natl Acad Sci USA 103, 8577–8582, https://doi.org/10.1073/
pnas.0601602103 (2006).
26. Blondel, V., Guillaume, J., Lambiotte, . & Lefebvre, E. Fast unfolding of communities in large networs. J Stat Mech, P10008, https://
doi.org/10.1088/1742-5468/2008/10/P10008 (2008).
27. ubinov, M. & Sporns, O. Weight-conserving characterization of complex functional brain networs. Neuroimage 56, 2068–2079,
https://doi.org/10.1016/j.neuroimage.2011.03.069 (2011).
28. Eicho, S. B., irion, B., Varoquaux, G. & Bzdo, D. Connectivity-based parcellation: Critique and implications. Hum Brain Mapp
36, 4771–4792, https://doi.org/10.1002/hbm.22933 (2015).
29. Nicolini, C. & Bifone, A. Modular structure of brain functional networs: breaing the resolution limit by Surprise. Scie ntic epor ts
6, 19250, https://doi.org/10.1038/srep19250 (2016).
30. osvall, M. & Bergstrom, C. T. An information-theoretic framewor for resolving community structure in complex networs. Proc
Natl Acad Sci USA 104, 7327–7331, https://doi.org/10.1073/pnas.0611034104 (2007).
31. Glerean, E. et al. eorganization of functionally connected brain subnetwors in high-functioning autism. Hum Brain Mapp 37,
1066–1079, https://doi.org/10.1002/hbm.23084 (2016).
32. Phelps, E. A. & LeDoux, J. E. Contributions of the amygdala to emotion processing: from animal models to human behavior. Neuron
48, 175–187 (2005).
33. Hariri, A. ., Booheimer, S. Y. & Mazziotta, J. C. Modulating emotional responses: eects of a neocortical networ on the limbic
system. Neuroreport 11, 43–48 (2000).
34. Grezes, J., Valabregue, ., Gholipour, B. & Chevallier, C. A direct amygdala-motor pathway for emotional displays to inuence
action: A diusion tensor imaging study. Hum Brain Mapp 35, 5974–5983, https://doi.org/10.1002/hbm.22598 (2014).
35. Stefanacci, L. & Amaral, D. G. Some observations on cortical inputs to the macaque money amygdala: an anterograde t racing study.
J Comp Neurol 451, 301–323, https://doi.org/10.1002/cne.10339 (2002).
36. emppainen, S., Jolonen, E. & Pitänen, A. Projections from the posterior cortical nucleus of the amygdala to the hippocampal
formation and parahippocampal region in rat. Hippocampus 12, 735–755 (2002).
37. Moreno, N. & Gonzalez, A. Evolution of the amygdaloid complex in vertebrates, with special reference to the anamnio-amniotic
transition. J Anat 211, 151–163, https://doi.org/10.1111/j.1469-7580.2007.00780.x (2007).
38. Li, Y., Qin, W., Jiang, T., Zhang, Y. & Yu, C. Sex-dependent correlations between the personality dimension of harm avoidance and
the resting-state functional connectivity of amygdala subregions. PLoS One 7, e35925, https://doi.org/10.1371/journal.pone.0035925
(2012).
39. Stephan, H. & Andy, O. J. Quantitative comparison of the amygdala in insectivores and primates. Acta Anat (Basel) 98, 130–153
(1977).
40. Vogt, B. A. Cingulate Neurobiology and Disease. (Oxford University Press Inc., 2009).
41. Vogt, B. A. & Pandya, D. N. Cingulate cortex of the rhesus money: II. Cortical aerents. J Comp Neurol 262, 271–289, https://doi.
org/10.1002/cne.902620208 (1987).
42. Chase, H. W., Moses-olo, E. L., Zevallos, C., Wisner, . L. & Phillips, M. L. Disrupted posterior cingulate-amygdala connectivity
in postpartum depressed women as measured with resting BOLD fMI. Soc Cogn Affect Neurosci 9, 1069–1075, https://doi.
org/10.1093/scan/nst083 (2014).
43. Marchetti, I., oster, E. H., Sonuga-B are, E. J. & De aedt, . e default mode networ and recurrent depression: a neurobiological
model of cognitive ris factors. Neuropsychol ev 22, 229–251, https://doi.org/10.1007/s11065-012-9199-9 (2012).
44. Cauda, F. et al. Functional connectivity of the posteromedial cortex. PLoS One 5, https://doi.org/10.1371/journal.pone.0013107
(2010).
45. Pfeerbaum, A. et al. Cerebral blood ow in posterior cortical nodes of the default mode networ decreases with tas engagement
but remains higher than in most brain reg ions. Cerebral cortex (New Yor, N. Y.: 1991) 21, 233–244, https://doi.org/10.1093/cercor/
bhq090 (2011).
46. Zhang, S. & Li, C. S. Functional connectivity mapping of the human precuneus by resting state fMI. Neuroimage 59, 3548–3562,
https://doi.org/10.1016/j.neuroimage.2011.11.023 (2012).
47. Parvizi, J., Van Hoesen, G. W., Bucwalter, J. & Damasio, A. Neural connections of the posteromedial cortex in the macaque. Proc
Natl Acad Sci USA 103, 1563–1568, https://doi.org/10.1073/pnas.0507729103 (2006).
48. Leichnetz, G. . Connections of the medial posterior parietal cortex (area 7m) in the money. Anat ec 263, 215–236 (2001).
49. Van Schuerbee, P., Baeen, C., Luypaert, ., De aedt, . & De Mey, J. Does the amygdala response correlate with the personality
trait ‘harm avoidance’ while evaluating emotional stimuli explicitly? Behav Brain Funct 10, 18, https://doi.org/10.1186/1744-9081-
10-18 (2014).
50. Seeley, W. W. et al. Dissociable intrinsic connectivity networs for salience processing and executive control. J Neurosci 27,
2349–2356, https://doi.org/10.1523/jneurosci.5587-06.2007 (2007).
51. app, B. S., Supple, W. F. J. & Whalen, P. J. Eects of electrical stimulation of the amygdaloid central nucleus on neocortical arousal
in the rabbit. Behav Neurosci 108, 81–93 (1994).
52. Touroutoglou, A., Bicart, . C., Barrett, L. F. & Dicerson, B. C. Amygdala tas-evoed activity and tas-free connectivity
independently contribute to feelings of arousal. Hum Brain Mapp 35, 5316–5327, https://doi.org/10.1002/hbm.22552 (2014).
53. Phillips, M. L., Drevets, W. C., auch, S. L. & Lane, . Neurobiology of emotion perception I: e neural basis of normal emotion
perception. Biol Psychiatr y 54, 504–514 (2003).
54. Vogt, B. A. Pain and emotion interactions in subregions of the cingulate gyrus. Nat ev Neurosci 6, 533–544 (2005).
55. Aggleton, J. P. e contribution of the amygdala to normal and abnormal emotional states. Trends Neurosci 16, 328–333 (1993).
56. Whalen, P. J. et al. Mased presentations of emotional facial expressions modulate amygdala activity without explicit nowledge. J
Neurosci 18, 411–418 (1998).
57. Craig, A. D. How do you feel-now? e anterior insula and human awareness. Nat ev Neurosci 10, 59–70, https://doi.org/10.1038/
nrn2555 (2009).
58. Muhlberger, A. et al. Stop looing angry and smile, please: start and stop of the very same facial expression dierentially activate
threat- and reward-related brain networs. Soc Cogn Aect Neurosci 6, 321–329, https://doi.org/10.1093/scan/nsq039 (2011).
59. Vogel, S. et al. Blocing the mineralocorticoid receptor in humans prevents the stress-induced enhancement of centromedial
amygdala connectivity with the dorsal striatum. Neuropsychopharmacology 40, 947–956, https://doi.org/10.1038/npp.2014.271
(2015).
60. Wilensy, A. E., Schafe, G. E., ristensen, M. P. & LeDoux, J. E. ethining the fear circuit: the central nucleus of the amygdala is
required for the acquisition, consolidation, and expression of Pavlovian fear conditioning. J Neurosci 26, 12387–12396, https://doi.
org/10.1523/jneurosci.4316-06.2006 (2006).
Content courtesy of Springer Nature, terms of use apply. Rights reserved

www.nature.com/scientificreports/
13
SCIENTIfIC REPORTS | 7: 14392 | DOI:10.1038/s41598-017-14613-4
61. Blanchard, D. C. & Blanchard, . J. Innate and conditioned reactions to threat in rats with amygdaloid lesions. J Comp Physiol
Psychol 81, 281–290 (1972).
62. Gregg, T. . & Siegel, A. Brain structures and neurotransmitters regulating aggression in cats: implications for human aggression.
Prog Neuropsychopharmacol Biol Psychiatry 25, 91–140 (2001).
63. Ba l l, T. et al. esponse properties of human amygdala subregions: evidence based on functional MI combined with probabilistic
anatomical maps. PLoS One 2, e307, https://doi.org/10.1371/journal.pone.0000307 (2007).
64. Baer, . B. & im, J. J. Amygdalar lateralization in fear conditioning: evidence for greater involvement of the right amygdala. Behav
Neurosci 118, 15–23, https://doi.org/10.1037/0735-7044.118.1.15 (2004).
65. Ji, G. & Neugebauer, V. Hemispheric Lateralization of Pain Processing by Amygdala Neurons. J Neurophysiol 102, 2253–2264 (2009).
66. Quir, G. J. & Beer, J. S. Prefrontal involvement in the regulation of emotion: convergence of rat and human studies. Curr Opin
Neurobiol 16, 723–727, https://doi.org/10.1016/j.conb.2006.07.004 (2006).
67. Llamas, A., Avendano, C. & einoso-Suarez, F. Amygdaloid projections to prefrontal and motor cortex. Science 195, 794–796 (1977).
68. Amaral, D. G. & Price, J. L. Amygdalo-cortical projections in the money (Macaca fascicularis). J Comp Neurol 230, 465–496,
https://doi.org/10.1002/cne.902300402 (1984).
69. Gotts, S. J. et al. Fractionation of social brain circuits in autism spectrum disorders. Brain 135, 2711–2725, https://doi.org/10.1093/
brain/aws160 (2012).
70. Homan, . L., Gothard, . M., Schmid, M. C. & Logothetis, N. . Facial-expression and gaze-selective responses in the money
amygdala. Curr Biol 17, 766–772, https://doi.org/10.1016/j.cub.2007.03.040 (2007).
71. B a l l, T. et al. Anatomical specicity of functional amygdala imaging of responses to stimuli with positive and negative emotional
valence. J Neurosci Methods 180, 57–70, https://doi.org/10.1016/j.jneumeth.2009.02.022 (2009).
72. eichenbach, J. . et al. eory and application of static eld inhomogeneity eects in gradient-echo imaging. J Magn eson Imaging
7, 266–279 (1997).
73. Cordes, D., Nandy, . ., Schafer, S. & Wager, T. D. Characterization and reduction of cardiac- and respiratory-induced noise as a
function of the sampling rate (T) in fMI. Neuroimage 89, 314–330, https://doi.org/10.1016/j.neuroimage.2013.12.013 (2014).
74. Jacob, Y. et al. Dependency Networ Analysis (DEPNA) eveals Context elated Inuence of Brain Networ Nodes. Sci ep 6,
27444, https://doi.org/10.1038/srep27444 (2016).
75. Glasser, M. F. et al . e Human Connectome Project’s neuroimaging approach. Nat Neurosci 19, 1175–1187, https://doi.org/10.1038/
nn.4361 (2016).
76. Moeller, S. et al. Multiband multislice GE-EPI at 7 tesla, with 16-fold acceleration using partial parallel imaging with application to
high spatial and temporal whole-brain fMI. Magn eson Med 63, 1144–1153, https://doi.org/10.1002/mrm.22361 (2010).
77. WU-Minn HCP 500 Subjects + MEG2 Data elease: eference Manual http://www.humanconnectome.org/documentation/S500/
78. Cox, . W. AFNI: soware for analysis and visualization of functional magnetic resonance neuroimages. Comput Biomed es. 29,
162–173 (1996).
79. Smith, S. M. In Functional MI: an introduction to methods (eds P. P. M. Jezzard, Matthews, & S. Smith, M.) (Oxford University Press,
2003).
80. Foerster, B., Tomasi, D. & Caparelli, E. C. Magnetic eld shi due to mechanical vibration in functional magnetic resonance
imaging. Magn. eson. Med. 54, 1261–1267 (2005).
81. Cordes, D. et al. Frequencies Contributing to Functional Connectivity in the Cerebral Cortex in “esting-state” Data. AJN 22,
1326–1333 (2001).
82. Lund, T. E., Madsen, . H., Sidaros, ., Luo, W. L. & Nichols, T. E. Non-white noise in fMI: does modelling have an impact?
Neuroimage 29, 54–66, https://doi.org/10.1016/j.neuroimage.2005.07.005 (2006).
83. Chai, X. J., Castanon, A. N., Ongur, D. & Whiteld-Gabrieli, S. Anticorrelations in resting state networs without global signal
regression. Neuroimage 59, 1420–1428, https://doi.org/10.1016/j.neuroimage.2011.08.048 (2012).
84. Ashburner, J. & Friston, . J. Unied segmentation. Neuroimage 26, 839–851, https://doi.org/10.1016/j.neuroimage.2005.02.018
(2005).
85. Power, J. D., Barnes, . A., Snyder, A. Z., Schlaggar, B. L. & Petersen, S. E. Spurious but systematic correlations in functional
connectivity MI networs arise from subject motion. Neuroimage 59, 2142–2154, https://doi.org/10.1016/j.neuroimage.2011.10.018
(2012).
86. Nielsen, F. A. & Hansen, L. . Automatic anatomical labelling of Talairach coordinates and generation of volumes of interest via the
BrainMap database, http://neuro.imm.dtu.d/services/jerne/ninf/voi (2002).
87. Fox, P. T., Miiten, S., Davis, G. & Lancaster, J. L. In Functional Neuroimaging: Technical Foundations (eds . W. atcher et al.) Ch.
9, 95–105 (Academic Press, 1994).
88. ubinov, M. & Sporns, O. Complex networ measures of brain connectivity: uses and interpretations. Neuroimage 52, 1059–1069,
https://doi.org/10.1016/j.neuroimage.2009.10.003 (2010).
89. Chen, G., Saad, Z. S., Britton, J. C., Pine, D. S. & Cox, . W. Linear mixed-eects modeling approach to FMI group analysis.
Neuroimage 73, 176–190, https://doi.org/10.1016/j.neuroimage.2013.01.047 (2013).
90. Ward, B. D. Simultaneous inference for FMI data. http://afni.nimh.nih.gov/pub/dist/doc/manual/AlphaSim.pdf (2000).
91. Shrout, P. E. & Fleiss, J. L. Intraclass correlations: uses in assessing rater reliability. Psychol Bull 86, 420–428 (1979).
92. Diedrichsen, J., Balsters, J. H., Flavell, J., Cussans, E. & amnani, N. A probabilistic M atlas of the human cerebellum. Neuroimage
46, 39–46, https://doi.org/10.1016/j.neuroimage.2009.01.045 (2009).
Acknowledgements
is work was supported by the National Institute on Drug Abuse Intramural Research Program. Data were
provided by the Human Connectome Project, WU-Minn Consortium (Principal Investigators: David Van
Essen and Kamil Ugurbil; 1U54MH091657) funded by the 16 NIH Institutes and Centers that support the NIH
Blueprint for Neuroscience Research; and by the McDonnell Center for Systems Neuroscience at Washington
University.
Author Contributions
Elisabeth C. Caparelli carried all the data analysis and wrote the manuscript. omas J. Ross and Yihong Yang
provided some guidelines for the data analysis and reviewed the manuscript. Hong Gu also provided some
guidelines for the data analysis and helped with script writing. Xia Liang provided some guide on using the Brain
Connectivity Toolbox. Elliot A. Stein provided important comments on the scientic signicance of the work and
reviewed the manuscript.
Additional Information
Supplementary information accompanies this paper at https://doi.org/10.1038/s41598-017-14613-4.
Competing Interests: e authors declare that they have no competing interests.
Content courtesy of Springer Nature, terms of use apply. Rights reserved

www.nature.com/scientificreports/
14
SCIENTIfIC REPORTS | 7: 14392 | DOI:10.1038/s41598-017-14613-4
Publisher's note: Springer Nature remains neutral with regard to jurisdictional claims in published maps and
institutional aliations.
Open Access This article is licensed under a Creative Commons Attribution 4.0 International
License, which permits use, sharing, adaptation, distribution and reproduction in any medium or
format, as long as you give appropriate credit to the original author(s) and the source, provide a link to the Cre-
ative Commons license, and indicate if changes were made. e images or other third party material in this
article are included in the article’s Creative Commons license, unless indicated otherwise in a credit line to the
material. If material is not included in the article’s Creative Commons license and your intended use is not per-
mitted by statutory regulation or exceeds the permitted use, you will need to obtain permission directly from the
copyright holder. To view a copy of this license, visit http://creativecommons.org/licenses/by/4.0/.
© e Author(s) 2017
Content courtesy of Springer Nature, terms of use apply. Rights reserved

1.
2.
3.
4.
5.
6.
Terms and Conditions
Springer Nature journal content, brought to you courtesy of Springer Nature Customer Service Center GmbH (“Springer Nature”).
Springer Nature supports a reasonable amount of sharing of research papers by authors, subscribers and authorised users (“Users”), for small-
scale personal, non-commercial use provided that all copyright, trade and service marks and other proprietary notices are maintained. By
accessing, sharing, receiving or otherwise using the Springer Nature journal content you agree to these terms of use (“Terms”). For these
purposes, Springer Nature considers academic use (by researchers and students) to be non-commercial.
These Terms are supplementary and will apply in addition to any applicable website terms and conditions, a relevant site licence or a personal
subscription. These Terms will prevail over any conflict or ambiguity with regards to the relevant terms, a site licence or a personal subscription
(to the extent of the conflict or ambiguity only). For Creative Commons-licensed articles, the terms of the Creative Commons license used will
apply.
We collect and use personal data to provide access to the Springer Nature journal content. We may also use these personal data internally within
ResearchGate and Springer Nature and as agreed share it, in an anonymised way, for purposes of tracking, analysis and reporting. We will not
otherwise disclose your personal data outside the ResearchGate or the Springer Nature group of companies unless we have your permission as
detailed in the Privacy Policy.
While Users may use the Springer Nature journal content for small scale, personal non-commercial use, it is important to note that Users may
not:
use such content for the purpose of providing other users with access on a regular or large scale basis or as a means to circumvent access
control;
use such content where to do so would be considered a criminal or statutory offence in any jurisdiction, or gives rise to civil liability, or is
otherwise unlawful;
falsely or misleadingly imply or suggest endorsement, approval , sponsorship, or association unless explicitly agreed to by Springer Nature in
writing;
use bots or other automated methods to access the content or redirect messages
override any security feature or exclusionary protocol; or
share the content in order to create substitute for Springer Nature products or services or a systematic database of Springer Nature journal
content.
In line with the restriction against commercial use, Springer Nature does not permit the creation of a product or service that creates revenue,
royalties, rent or income from our content or its inclusion as part of a paid for service or for other commercial gain. Springer Nature journal
content cannot be used for inter-library loans and librarians may not upload Springer Nature journal content on a large scale into their, or any
other, institutional repository.
These terms of use are reviewed regularly and may be amended at any time. Springer Nature is not obligated to publish any information or
content on this website and may remove it or features or functionality at our sole discretion, at any time with or without notice. Springer Nature
may revoke this licence to you at any time and remove access to any copies of the Springer Nature journal content which have been saved.
To the fullest extent permitted by law, Springer Nature makes no warranties, representations or guarantees to Users, either express or implied
with respect to the Springer nature journal content and all parties disclaim and waive any implied warranties or warranties imposed by law,
including merchantability or fitness for any particular purpose.
Please note that these rights do not automatically extend to content, data or other material published by Springer Nature that may be licensed
from third parties.
If you would like to use or distribute our Springer Nature journal content to a wider audience or on a regular basis or in any other manner not
expressly permitted by these Terms, please contact Springer Nature at
onlineservice@springernature.com