Intrinsic functional connectivity of the central extended amygdala

Abstract

The central extended amygdala (EAc)—including the bed nucleus of the stria terminalis (BST) and central nucleus of the amygdala (Ce)—plays a critical role in triggering fear and anxiety and is implicated in the development of a range of debilitating neuropsychiatric disorders. Although it is widely believed that these disorders reflect the coordinated activity of distributed neural circuits, the functional architecture of the EAc network and the degree to which the BST and the Ce show distinct patterns of functional connectivity is unclear. Here, we used a novel combination of imaging approaches to trace the connectivity of the BST and the Ce in 130 healthy, racially diverse, community‐dwelling adults. Multiband imaging, high‐precision registration techniques, and spatially unsmoothed data maximized anatomical specificity. Using newly developed seed regions, whole‐brain regression analyses revealed robust functional connectivity between the BST and Ce via the sublenticular extended amygdala, the ribbon of subcortical gray matter encompassing the ventral amygdalofugal pathway. Both regions displayed coupling with the ventromedial prefrontal cortex (vmPFC), midcingulate cortex (MCC), insula, and anterior hippocampus. The BST showed stronger connectivity with the thalamus, striatum, periaqueductal gray, and several prefrontal territories. The only regions showing stronger functional connectivity with the Ce were neighboring regions of the dorsal amygdala, amygdalohippocampal area, and anterior hippocampus. These observations provide a baseline against which to compare a range of special populations, inform our understanding of the role of the EAc in normal and pathological fear and anxiety, and showcase image registration techniques that are likely to be useful for researchers working with “deidentified” neuroimaging data.

1. INTRODUCTION

When extreme, fear and anxiety can become debilitating (Grupe & Nitschke, 2013; Salomon et al., 2015). Anxiety disorders are common and challenging to treat, imposing a staggering burden on public health, and underscoring the need to develop a more complete understanding of the distributed neural circuits governing the expression of fear and anxiety in humans (Bystritsky, 2006; Craske et al., 2017; DiLuca & Olesen, 2014; Global Burden of Disease Collaborators, 2016; Griebel & Holmes, 2013).

Converging lines of anatomical, mechanistic, and physiological evidence make it clear that the central extended amygdala (EAc) is a key hub in this circuitry (Figure 1a,b) (Avery, Clauss, & Blackford, 2016; Davis, Walker, Miles, & Grillon, 2010; Fox & Shackman, in press; Goode & Maren, 2017; Gungor & Paré, 2016; Shackman & Fox, 2016; Tovote, Fadok, & Luthi, 2015). The EAc encompasses a collection of subcortical regions with similar cellular compositions, neurochemistry, gene expression, and structural connectivity and it encompasses the bed nucleus of the stria terminalis (BST), the central nucleus of the amygdala (Ce), the sublenticular extended amygdala (SLEA), and portions of the accumbens shell (Alheid & Heimer, 1988; Fox, Oler, Tromp, Fudge, & Kalin, 2015a; Oler et al., 2017; Yilmazer‐Hanke, 2012). It has long been recognized that the amygdala is connected to the BST via two major fiber bundles—the ventral amygdalofugal pathway (VA) and the stria terminalis (ST) (Avery et al., 2014; Kamali et al., 2016, 2015; Nauta, 1961) (Figure 1c)—and more recent tracing studies have identified a third, indirect pathway centered on the SLEA (Ce ↔ SLEA ↔ BSTL) (deCampo & Fudge, 2013; Fudge et al., 2017; Oler et al., 2017). Anatomically, the Ce and the BST are both poised to trigger or orchestrate key signs of fear and anxiety—including alterations in arousal, behavioral inhibition, and neuroendocrine activity—via dense mono‐ and polysynaptic projections to brainstem and subcortical effector regions (Fox et al., 2015a; Freese & Amaral, 2009; Fudge et al., 2017).

Figure 1.

The EAc. (a) Simplified schematic of key EAc inputs and outputs in humans and other primates. The EAc (magenta) encompasses the BST, which encircles the anterior commissure, and the Ce. As shown by the translucent white arrow at the center of the figure, much of the sensory (yellow), contextual (blue), and regulatory (green) inputs to the EAc are indirect (i.e., polysynaptic), and first pass through adjacent amygdala nuclei before arriving at the Ce or the BST. Both regions are poised to orchestrate momentary states of fear and anxiety via dense projections to downstream effector regions (orange). Portions of this figure were adapted from the atlas of (Mai, Paxinos, & Voss, 2007; see also Yilmazer‐Hanke, 2012). (b) BST and Ce seeds. Figure depicts the location of the BST and Ce seeds used in the present study. See Supporting Information, Figure S5 for bilateral views and a more detailed description of seed derivation. (c) Structural connections of the EAc. In humans and other primates, the BST (dorsorostral magenta region) and the Ce (ventrocaudal magenta region) are structurally connected via two major fiber bundles (gold), the ventral amygdalofugal pathway and the stria terminalis (Johnston, 1923; Nauta, 1961; Yilmazer‐Hanke, 2012). From the Ce, the ventral amygdalofugal pathway courses forward and medially, passing through the SLEA, a bridge of neurons harbored within the substantia innominata. The stria terminalis, which arches dorsally over the thalamus, provides a second, less direct connection between the two major divisions of the central extended amygdala. Figure depicts deterministic tractography (gold) of these two fiber bundles. Image kindly provided by Do Tromp. Abbreviations: BL = basolateral nucleus of the amygdala; BM = basomedial nucleus of the amygdala; BST = bed nucleus of the stria terminalis; Ce = central nucleus of the amygdala; EAc = central division of the extended amygdala; La = lateral nucleus of the amygdala; Me = medial nucleus of the amygdala; SLEA = sublenticular extended amygdala [Color figure can be viewed at http://wileyonlinelibrary.com]

Consistent with this neuroanatomy, mechanistic studies in rodents indicate that microcircuits within and between the BST and the Ce play a critical role in organizing defensive responses to a range of potentially threat‐relevant cues and contexts (Calhoon & Tye, 2015; Davis et al., 2010; Fox & Shackman, in press; Goode & Maren, 2017; Gungor & Paré, 2016; Lange et al., 2017; Tovote et al., 2015) (Figure 1c). Although the BST and the Ce are often viewed as passive output relays for amygdala‐mediated emotional learning (e.g., La → Ce/BST → effector regions; LeDoux, 2000, 2007; Pare & Duvarci, 2012), more recent work in rodents has expanded this role to include relaying information about pain and aversive reinforcers (Yu et al., 2017), guiding attention to motivationally salient stimuli (Davis & Whalen, 2001; Roesch, Esber, Li, Daw, & Schoenbaum, 2012; Shackman et al., 2016a), learning aversive associations (Ciocchi et al., 2010; Han, Soleiman, Soden, Zweifel, & Palmiter, 2015; Li et al., 2013; Penzo, Robert, & Li, 2014; Penzo et al., 2015; Sato et al., 2015; Yu et al., 2017), and actively gating and regulating defensive responses (Ehrlich et al., 2009; Fadok et al., 2017; Gungor & Paré, 2016; Pare & Duvarci, 2012).

Although the causal contribution of the BST has yet to be explored in primates, the Ce has been shown to control defensive responses to potential threat in monkeys (Kalin, 2017; Kalin et al., 2016; Kalin, Shelton, & Davidson, 2004). Similarly, rodents, monkeys, and humans with amygdala damage exhibit a profound lack of fear and anxiety in response to a broad spectrum of learned and innate dangers (Antoniadis, Winslow, Davis, & Amaral, 2007; Bechara et al., 1995; Choi & Kim, 2010; Davis & Whalen, 2001; Feinstein, Adolphs, Damasio, & Tranel, 2011; Feinstein, Adolphs, & Tranel, 2016; Izquierdo, Suda, & Murray, 2005; Kalin et al., 2004; Korn et al., 2017; Mason, Capitanio, Machado, Mendoza, & Amaral, 2006; Oler, Fox, Shackman, & Kalin, 2016).

Neuroimaging research indicates that heightened activity in the EAc is associated with elevated signs of fear and anxiety in both monkeys and humans (Alvarez et al., 2015; Banihashemi, Sheu, Midei, & Gianaros, 2015; Cheng, Knight, Smith, & Helmstetter, 2006; Cheng, Richards, & Helmstetter, 2007; Fox et al., 2015b; Fox, Shelton, Oakes, Davidson, & Kalin, 2008; Kalin, Shelton, Fox, Oakes, & Davidson, 2005; Knight, Nguyen, & Bandettini, 2005; Kragel & LaBar, 2015; LaBar, Gatenby, Gore, LeDoux, & Phelps, 1998; Shackman et al., 2013; Somerville et al., 2013; van Well, Visser, Scholte, & Kindt, 2012; Wood, Ver Hoef, & Knight, 2014). Among humans, the amygdala responds to a variety of threat‐related cues (Costafreda, Brammer, David, & Fu, 2008; Fusar‐Poli et al., 2009; Lindquist, Satpute, Wager, Weber, & Barrett, 2016; Sabatinelli et al., 2011; Sergerie, Chochol, & Armony, 2008) and work using high‐resolution fMRI indicates that the dorsal amygdala in the region of the Ce is particularly sensitive to aversive visual stimuli (Hrybouski et al., 2016).

Although less intensively studied than the Ce, the BST is sensitive to emotional faces (Sladky et al., 2017), aversive images (Brinkmann et al., 2018), and a variety of threat‐related cues (Alvarez, Chen, Bodurka, Kaplan, & Grillon, 2011; Brinkmann et al., 2017b; Choi, Padmala, & Pessoa, 2012; Grupe, Oathes, & Nitschke, 2013; Herrmann et al., 2016; Klumpers et al., 2015; McMenamin, Langeslag, Sirbu, Padmala, & Pessoa, 2014; Mobbs et al., 2010; Pedersen et al., 2017; Somerville, Whalen, & Kelley, 2010; Somerville et al., 2013). While imaging research hints at potential functional differences between the two regions (Alvarez et al., 2011; Fox et al., 2015b; Meyer, Padmala, & Pessoa, 2017; Shackman et al., 2017; Somerville et al., 2013), methodological limitations preclude decisive inferences (Fox & Shackman, in press; Shackman & Fox, 2016). Importantly, other work suggests that alterations in EAc function likely plays a key role in the development, maintenance, and recurrence of anxiety disorders, depression, and substance abuse (Avery et al., 2016; Brinkmann et al., 2017a, 2017b, 2018; Buff et al., 2017; Fox & Kalin, 2014; Kaczkurkin et al., 2016; Münsterkötter et al., 2015; Shackman et al., 2016a, 2016b; Stevens et al., 2017; Williams et al., 2015; Wise & Koob, 2014).

Although this vast literature leaves little doubt that the EAc plays a crucial role in evaluating and responding to a variety of potential threats, it does not act in isolation. Fear and anxiety reflect functional circuits that extend well beyond the borders of the EAc (Chang, Gianaros, Manuck, Krishnan, & Wager, 2015; Fox & Shackman, in press; Kragel, Knodt, Hariri, & LaBar, 2016; Nummenmaa & Saarimaki, in press; Pessoa, 2017; Shackman & Fox, 2018; Shackman, Fox, & Seminowicz, 2015; Wager et al., 2015). Anatomically, the BST and the Ce are embedded within a complex web of mono‐ and polysynaptically connected brain regions (Figure 1a) (Carrive & Morgan, 2012; Fox et al., 2015a; Freese & Amaral, 2009; Fudge et al., 2017; Oler et al., 2017; Ongur & Price, 2000). This structural backbone includes subcortical regions, such as the periaqueductal gray (PAG), that are responsible for triggering specific signs of fear and anxiety (Amano et al., 1982; Assareh, Sarrami, Carrive, & McNally, 2016; Bandler, Price, & Keay, 2000; Chen et al., 2015; Fadok et al., 2017; Faull & Pattinson, 2017; Motta, Carobrez, & Canteras, 2017; Nashold, Wilson, & Slaughter, 1969; Richardson & Akil, 1977; Satpute et al., 2013; Tovote et al., 2016). It also encompasses a number of cortical regions implicated in the expression and regulation of fear and anxiety, including the anterior insula, dorsolateral prefrontal cortex, mid‐cingulate cortex (MCC), and OFC (Birn et al., 2014; Buhle et al., 2014; Cavanagh & Shackman, 2015; de la Vega, Chang, Banich, Wager, & Yarkoni, 2016; Fox et al., 2015, 2010b; Grupe & Nitschke, 2013; Mobbs et al., 2009, 2007, 2010; Shackman, McMenamin, Maxwell, Greischar, & Davidson, 2009; Shackman et al., 2011; Stout, Shackman, Pedersen, Miskovich, & Larson, 2017; Uddin, Kinnison, Pessoa, & Anderson, 2014). While it is widely believed that the synchronized flow of information across this network underlies the human capacity for flexibly regulating fear and anxiety, the functional architecture of the EAc network and the degree to which the BST and the Ce are characterized by distinct patterns of functional connectivity remains incompletely understood.

Building on prior work (Table 1), we used a combination of imaging approaches to trace and compare the intrinsic functional connectivity of the BST and the Ce. Whole‐brain “resting‐state” functional MRI (fMRI) data were acquired from a relatively large (n = 130) sample of psychiatrically healthy, racially diverse, community‐dwelling adults, providing increased statistical power and generalizability. Given the challenges of imaging the EAc (Fox et al., 2015a; Shackman & Fox, 2016; Fox & Shackman, in press), several techniques were used to maximize effective spatial resolution, including a multiband imaging sequence with 2‐mm³ nominal resolution, boundary‐based co‐registration (Greve & Fischl, 2009), a novel brain‐extraction (“skull‐stripping”) approach, and diffeomorphic normalization (Avants, Epstein, Grossman, & Gee, 2008; Avants et al., 2011, 2010; Klein et al., 2009). To further enhance anatomical specificity, analyses were conducted using spatially unsmoothed data and newly developed extended amygdala seeds. Collectively, these techniques enabled us to compare the intrinsic functional connectivity of the BST and the Ce with enhanced statistical sensitivity and anatomical precision (Table 1). Understanding these functional networks is important: it would provide a baseline against which to compare a range of special populations—including individuals at risk for developing mental illness and patients suffering from psychiatric disorders—and it would inform our understanding of the EAc's role in normal and pathological fear and anxiety.

Table 1.

Intrinsic functional connectivity of the human central extended amygdala

Citation	Population	N	Coverage	Native EPI resolution	Smoothing	Normalization	Ce seed	BST seed
Present study	Adults	130	Whole brain	2 × 2 × 2 mm	N/A	FSL‐BBR, ANTS/SyN	Prescribed by an experienced neuroanatomist using a specially processed, ultra‐high‐resolution, multi‐modal probabilistic template (CITI168)	Prescribed by 2 raters using T2 images acquired from 10 young adults and normalized using ANTS/SyN; thresholded at 25% (Theiss, Ridgewell, McHugo, Heckers, & Blackford, 2017)
Avery et al., 2014	Midlife adults	99	Whole brain	3 × 3 × 4 mm	3 mm	SPM8	Prescribed using a single ultra‐high‐resolution T2 image acquired from a 42‐year‐old male	N/A
Gorka et al., 2017	Young adults	27	Partial	1.3 × 1.3 × 1.3 mm	2.6 mm	3dAllineate, 3dQWarp	Prescribed by 2 raters for the left hemisphere using 8 study‐specific, ultra‐high‐resolution, multi‐modal probabilistic templates; thresholded at 20% (Tyszka & Pauli, 2016)	Prescribed by 3 raters using each subject's T1 image; thresholded at 66.67% (Torrisi et al., 2015)
Motzkin et al., 2015	Older adults	17	Whole‐brain	3.5 × 3.5 × 3 mm	4 mm	ANTS/SyN	N/A	Prescribed by an experienced neuroanatomist using the 1‐mm MNI152 T1 template
Oler et al., 2012	Adolescents	105	Whole‐brain	3 × 3 × 3 or 3.75 × 3.75 × 5 mm	6 mm	Affine	Prescribed by an experienced neuroanatomist using the 1‐mm MNI152 T1 template	N/A
Torrisi et al., 2015	Young adults	27	Partial	1.3 × 1.3 × 1.3 mm	2.6 mm	3dAllineate, 3dQWarp	N/A	Prescribed by 3 raters using each subject's T1 image; thresholded at 66.67% (Torrisi et al., 2015)

2. MATERIALS AND METHODS

2.1. Subjects

Data were extracted from the publicly available Nathan Kline Institute‐Rockland Sample (NKI‐RS) (http://fcon_1000.projects.nitrc.org/indi/enhanced; Nooner et al., 2012) for 185 adults (18–40 years old). Exclusionary criteria included: positive drug urine screen (n = 12); self‐reported lifetime bipolar disorder, neurological disorder, pervasive developmental disorder, or psychosis/schizophrenia (n = 14); incomplete MRI data (n = 15); and incomplete demographic data (n = 5). Using procedures detailed below, 18 additional subjects were excluded due to excessive motion artifact (n = 8), susceptibility artifact (n = 9), or unusable T1 scans (n = 1). The final sample consisted of 130 subjects (59 males, M = 25.3 years, SD = 6.1). Additional demographic details can be found in the Supporting Information.

2.2. Data acquisition

MRI data were acquired using a Siemens Magnetom Trio Tim 3 T scanner and 32‐channel head‐coil (http://fcon_1000.projects.nitrc.org/indi/enhanced/mri_protocol.html). T1‐weighted anatomical images were acquired using a magnetization‐prepared, rapid‐acquisition, gradient‐echo sequence (inversion time: 900 ms; repetition time: 1,900 ms; echo time: 2.52 ms; flip angle: 9°; field‐of‐view: 250 $\times$ 250; matrix: 256 $\times$ 256; number of slices: 176 sagittal; slice thickness: 1 mm). Building on prior work with partial‐brain coverage (Gorka, Torrisi, Shackman, Grillon, & Ernst, 2017; Torrisi et al., 2015), functional scans were obtained using a T₂*‐weighted echo‐planar image (EPI) sequence (multiband acceleration: 4; repetition time: 1,400 ms; echo time: 30 ms; flip angle: 65°; number of excitations: 1; field‐of‐view: 224 $\times$ 224 mm; number of slices: 64 oblique‐axial; matrix: 112 $\times$ 112; slice thickness: 2 mm; gap: ∼0 mm; volumes: 404), enabling us to survey the entire brain.

2.3. Data processing pipeline

2.3.1. Brain extraction and normalization

Given our focus on the BST and the Ce, methods were optimized to minimize spatial normalization error and incidental spatial blurring. Consistent with other work (Acosta‐Cabronero, Williams, Pereira, Pengas, & Nestor, 2008; Fein et al., 2006; Fischmeister et al., 2013), unpublished observations by our group demonstrate that the quality of spatial normalization is enhanced by using a brain‐extracted (i.e., “skull‐stripped” or “de‐skulled”) template and brain‐extracted T1 images. This advantage is particularly evident for publicly available datasets, such as the NKI‐RS, where portions of the skull and tissue in the region of the face have been manually removed (“de‐faced”) by the curators to mitigate risks to subject confidentiality (i.e., “anonymized” or “de‐identified”). However, this benefit is only realized when the quality of the extraction is sufficiently high and consistent, as with images that have been manually extracted by an experienced neuroanatomist. To ensure consistently high‐quality extractions, we implemented a multi‐tool strategy (for a similar approach, see Meyer et al., 2017; Najafi, Kinnison, & Pessoa, 2017). For each inhomogeneity‐corrected (using N4; Tustison et al., 2014) T1 image, six extraction masks were generated. Five masks were generated using BET (Smith, 2002), BSE (Shattuck, Sandor‐Leahy, Schaper, Rottenberg, & Leahy, 2001), 3dSkullstrip (Cox, 1996), ROBEX (Iglesias, Liu, Thompson, & Tu, 2011), and SPM unified segmentation (Ashburner & Friston, 2005), respectively. The sixth mask was generated by applying the inverse spatial transformation (see below) to the MNI152 brain mask distributed with FSL. Specifically, for each subject: (a) the defaced T1 image was spatially normalized to the MNI152 template using the unified segmentation approach implemented in SPM12; (b) the 1‐mm MNI152 template was defaced to match the idiosyncratic defacing of the T1 image; (c) the original T1 image was normalized to the individually defaced 1‐mm template using SyN; and (d) the inverse transformation was used to “reverse‐normalize” the MNI152 brain mask distributed with FSL to native space. Next, a best‐estimate extraction mask was determined by consensus, requiring agreement across four or more extraction techniques. Using this mask, each T1 image was extracted and spatially normalized to the 1‐mm MNI152 template using the high‐precision diffeomorphic approach implemented in SyN (mutual information cost function; Avants et al., 2008, 2011, 2010; Klein et al., 2009). The average of the 130 normalized T1 images is depicted in Supporting Information, Figure S1.

2.3.2. EPI data

The first 3 volumes of each EPI scan were removed and the remaining volumes were de‐spiked and slice‐time corrected using default settings in AFNI (Cox, 1996). Recent methodological work indicates that de‐spiking is more effective than “scrubbing” (Jo et al., 2013; Power, Schlaggar, & Petersen, 2015; Siegel et al., 2014) for attenuating motion‐related artifacts in intrinsic functional connectivity. Spike‐ and slice‐time‐corrected EPI data were co‐registered to the corresponding brain‐extracted, native‐space T1 image using the boundary‐based registration technique implemented in FSL (Greve & Fischl, 2009) and converted to a compatible file format using Convert3d (https://sourceforge.net/p/c3d). Motion correction was then performed using ANTS (https://stnava.github.io/ANTs). The maximum value of the frame‐to‐frame displacement was calculated for each subject and z‐transformed. Subjects with a z‐score >1.96 (p = .05) were excluded (n = 8). Residual displacement in final dataset was negligible (median = 0.11 mm, SD = 0.07 mm, maximum = 0.43 mm). To minimize incidental spatial blurring, the transformation matrices for motion correction, co‐registration, and spatial normalization were concatenated and applied to the EPI data in a single step. Normalized EPI data were resampled to 2‐mm³ voxels using fifth‐order splines. To maximize spatial resolution, no additional spatial filters were applied, consistent with recent recommendations (Stelzer, Lohmann, Mueller, Buschmann, & Turner, 2014; Turner & Geyer, 2014). Each EPI and T1 dataset was visually inspected before and after processing for quality assurance. To quantify susceptibility artifact in the medial temporal lobe (MTL), we computed the ratio of mean signal in the amygdala relative to the caudate and putamen separately for each hemisphere and subject and then standardized across subjects (i.e., z‐transformed). Preliminary visual inspection indicated that values >∼2.50 were associated with substantial signal loss (“drop‐out”) in the MTL. Accordingly, subjects with z‐scores <−2.50 were excluded (n = 9) (for a similar approach, see Birn et al., 2014). To attenuate physiological noise, white matter (WM) and cerebrospinal fluid (CSF) time‐series were identified by thresholding the tissue prior images distributed with FSL, as in prior work by our group (Birn et al., 2014) and others (Coulombe, Erpelding, Kucyi, & Davis, 2016). The EPI time‐series was orthogonalized with respect to the first 3 right eigenvectors of the data covariance matrix from the WM and CSF compartments (Behzadi, Restom, Liau, & Liu, 2007), a Legendre polynomial series (first‐ to fifth‐order), and motion estimates (6 parameters lagged by 0, 1, and 2 volumes), consistent with recent recommendations (Hallquist, Hwang, & Luna, 2013). Orthogonalized time‐series were bandpass filtered (0.009–0.10 Hz) using AFNI and rescaled to zero‐mean unit variance in MATLAB. Using 3dFWHMx, the mean spatial smoothness of the orthogonalized data was estimated to be ∼2.28 mm³.

2.3.3. Seed regions

The BST seed was implemented using a previously published probabilistic region of interest thresholded at 25% (Theiss, Ridgewell, McHugo, Heckers, & Blackford, 2017). Building on prior work by our group (Birn et al., 2014; Nacewicz, Alexander, Kalin, & Davidson, 2014; Najafi et al., 2017; Oler et al., 2012, 2017), the Ce was manually prescribed by an experienced neuroanatomist (B.M.N.) using a specially processed version of the CITI168 high‐resolution (0.7 mm), multimodal (T1/T2) probabilistic template (http://evendim.caltech.edu/amygdala-atlas; Tyszka & Pauli, 2016) and guided by the atlas of Mai et al. (2007). The methods used for processing the template and prescribing the Ce seed are detailed in the Supporting Information, Figures S2–S5. Consistent with prior reports (Birn et al., 2014; Entis, Doerga, Barrett, & Dickerson, 2012; Hrybouski et al., 2016), visual inspection indicated that this approach provides enhanced anatomical sensitivity and selectivity compared to the more widely used centromedial amygdala region‐of‐interest distributed with FSL (Amunts et al., 2005) (Supporting Information, Figure S5). The BST and Ce seeds are depicted in Figure 1b and Supporting Information, Figure S6. To minimize partial volume artifacts, seeds were decimated to the 2‐mm MNI template using an iterative procedure that maintained a consistent seed volume across templates. Specifically, each seed was minimally smoothed (2.24 mm FWHM Gaussian) and the voxel size was dilated by 0.1 mm and resliced (linear interpolation), enabling us to identify a threshold that approximated the original seed volume and better preserved anatomical boundaries (Left BST: 96 mm³; Right BST: 96 mm³; Left Ce: 152 mm³; Right Ce: 152 mm³).

2.4. Analytic plan

We adopted a standard a priori seed‐based approach to quantifying intrinsic functional connectivity (Biswal, Yetkin, Haughton, & Hyde, 1995; Fox et al., 2005). For each subject, SPM12 (http://www.fil.ion.ucl.ac.uk/spm/software/spm12) and in‐house MATLAB code was used to perform a voxelwise regression between the artifact‐attenuated, average seed time series and voxel times series throughout the brain. Single‐subject regression analyses were performed using the Cochrane–Orcutt procedure for estimating autoregressive error, which is more efficient and potentially less biased than ordinary least‐squares (Stocker, 2007). In order to identify regions showing consistent functional connectivity with the BST or Ce seeds across subjects, we tested the intercept in regression models, equivalent to a single‐sample t test (t > 5.47, p < .05, whole‐brain Šidák corrected for 228,483 voxels) (Birn et al., 2014; Oler et al., 2010; Šidák, 1967). At this threshold, clusters of negative connectivity were only identified in regions of deep white matter and gray matter adjacent to ventricles and, so, are neither reported nor interpreted. A minimum conjunction (Boolean “AND”) was used to identify regions showing significant coupling with both seeds (Nichols, Brett, Andersson, Wager, & Poline, 2005) and a paired t test was used to assess differential functional connectivity. For ease of interpretation, differential connectivity was only examined in the subset of 12,004 voxels where functional connectivity was significant for one or both seeds (t > 4.80, p < .05, Šidák corrected for the 12,004 voxel region‐of‐interest). This approach circumvents the need to interpret significant differences (e.g., BST > Ce) in regions where neither seed shows significant functional connectivity. For both analyses, we imposed an arbitrary 80 mm³ (i.e., 10 native EPI voxels) minimum‐extent criterion—in addition to the intensity‐based thresholds (p < .05, Šidák corrected)—to suppress noise. Exploratory analyses yielded no reliable sex differences in Ce or BST functional connectivity. As an additional check on the integrity of the data and our approach, we confirmed our ability to identify the default mode network (Supporting Information, Figure S7). Clusters were labeled using a combination of the Mai and Harvard–Oxford atlases (Desikan et al., 2006; Frazier et al., 2005; Goldstein et al., 2007; Mai, Majtanik, & Paxinos, 2015; Makris et al., 2006). Some figures were created using MRIcron (http://people.cas.sc.edu/rorden/mricron).

3. RESULTS

3.1. Subcortical regions

As shown in Figure 2 and Supporting Information, Figure S8, whole‐brain regression analyses revealed robust coupling between the BST and the Ce regions (p < .05, whole‐brain Šidák corrected; Tables 2, 3, 4). Analyses seeded in the BST showed significant functional connectivity with neighboring regions of the basal forebrain and basal ganglia and distal voxels in the region of the Ce. The complementary pattern was observed for the Ce seed—significant functional connectivity with neighboring regions of the dorsal amygdala and with distal voxels located in the region of the BST. Consistent with invasive tracing studies (Oler et al., 2017), the BST and Ce also showed robust coupling with anatomically intermediate voxels located in the SLEA, the ribbon of subcortical gray matter (substantia innominata) encompassing the ventral amygdalofugal pathway (Figure 3). Finally, both seeds showed significant functional connectivity with the amygdalohippocampal area and anterior hippocampus (Figure 2).

Figure 2.

Intrinsic functional connectivity of the EAc. Left and center columns depict the results of whole‐brain regression analyses for the BST and the Ce seed regions, respectively, conservatively thresholded at p < .05 whole‐brain Šidák corrected. The right column depicts the intersection or conjunction (Boolean “AND”) of the two thresholded maps (Nichols et al., 2005). The BST seed showed significant functional connectivity with neighboring voxels in the basal forebrain (cyan arrowheads) and voxels in the region of the Ce (green arrowheads), while the Ce seed showed significant coupling with neighboring voxels in the dorsal amygdala and distal voxels in the region of the BST. Analyses also demonstrated that the BST and Ce exhibit robust functional connectivity with intermediate voxels located along the path of the ventral amygdalofugal pathway in the sublenticular extended amygdala (magenta arrowheads). Finally, both regions showed significant coupling with the amygdalohippocampal area and anterior hippocampus (white arrowheads), posterior insula (brown arrowheads), and superior temporal sulcus (black arrowheads). Note: Results are depicted here and reported in the accompanying tables for clusters of at least 80 mm³. See Figures 3 and 5 for additional views of these contrasts. Abbreviations: BST = bed nucleus of the stria terminalis; Ce = central nucleus of the amygdala; EAc = central division of the extended amygdala; L = left hemisphere; R = right hemisphere [Color figure can be viewed at http://wileyonlinelibrary.com]

Table 2.

Regions showing significant functional connectivity with the BSTa

x	y	z	t	mm3	Hemisphere	Region(s)/subregions
11	45	1	7.65	176	B	Cingulate sulcus, pregenual
−21	41	29	8.55	352	L	Superior frontal sulcus, anterior
−25	33	49	10.03	896	L	Superior frontal sulcus, anterior
27	32	35	8.75	888	B	Superior frontal sulcus, anterior
−42	23	−5	7.86	272	L	Orbitofrontal cortex, basal operculum
−5	3	0	21.04	49,072	B	Midlineb
−6	4	−1	21.04	9,128	B	Basal forebrain: caudate, putamen, globus pallidus, nucleus accumbens, rostrodorsal hypothalamus, piriform cortex, sublenticular extended amygdala (ventral amygdalofugal pathway), dorsal amygdala (central and medial nuclei), amygdalohippocampal area and anterior hippocampus, thalamus, brainstem
−6	−43	5	12.96	7,648	B	Posterior cingulate/Precuneus
1	19	37	11.7	3,072	L	Cingulate: cingulate sulcus, midcingulate; cingulate sulcus, posterior; juxtapositional lobule
11	18	33	10.27	480	R	Cingulate: Cingulate sulcus, pregenual; Cingulate sulcus, midcingulate
1	53	−5	9.67	328	B	Ventromedial prefrontal cortex: OP10r/mc; inferior frontopolar gyrus; rostral gyrus; anterior cingulate cortex, pregenual
−3	−25	−3	9.48	80	L	Periaqueductal gray, dorsolateral
−53	2	−1	7.97	136	L	Superior temporal gyrus, planum polare
−39	1	59	7.12	136	L	Precentral sulcus
1	−13	−23	8.81	88	R	Cerebellum
−37	−15	17	10.79	1,648	L	Posterior insula: central operculum, parietal operculum, posterior insula (dorsal portion of the long gyri), Heschl's gyrus
53	−16	5	9.57	2,224	R	Posterior insula: central operculum, parietal operculum, posterior insula (dorsal portion of the long gyri), Heschl's gyrus
31	−17	3	7.43	184	R	Putamen
13	−17	39	8.01	160	B	Cingulate sulcus, posterior
−27	−19	5	7.91	112	L	Putamen
7	−21	−1	10.85	152	R	Thalamus
69	−22	−3	8.19	544	R	Superior temporal sulcus
−20	−29	57	11.7	3,144	L	Central sulcus
21	−29	57	13.12	3,024	R	Central sulcus
26	−37	57	8.98	360	B	Postcentral sulcus
−19	−37	65	8.04	272	L	Postcentral gyrus
57	−57	21	7.53	176	R	Angular gyrus
54	−62	31	7.04	176	R	Lateral occipital cortex
−9	−69	5	8.29	256	L	Calcarine sulcus
31	−72	−37	8.41	344	B	Cerebellum
−31	−80	−37	8.17	504	L	Cerebellum
−7	−81	1	7.94	384	L	Calcarine sulcus
−35	−83	−19	7.03	96	L	Lateral occipital cortex/fusiform, occipital
25	−85	−19	8.08	328	R	Fusiform, occipital
15	−93	1	7.9	80	R	Occipital pole

Note. Abbreviations: B, bilateral; Ce, central nucleus of the amydala; L, left hemisphere; R, right hemisphere.

^aWhole‐brain regression analysis (p < .05, whole‐brain Šidák corrected, k ≥ 80 mm³).

^bFor large clusters, subregions were identified using T ≥ 7 and are shown in italics.

^cAreas 10r/m and 11 as described by Ongur, Ferry, and Price (2003).

Table 3.

Regions showing significant functional connectivity with the Cea

x	y	z	t	mm3	Hemisphere	Region(s)/subregions
1	59	19	8.29	504	B	Dorsomedial prefrontal cortex: BA10
1	53	−13	8.7	600	B	Ventromedial prefrontal cortex: OP10r/mc; inferior frontopolar gyrus; rostral gyrus
8	39	−15	7.27	112	R	Ventromedial prefrontal cortex: inferior frontopolar gyrus, straight gyrus
34	37	−13	7.15	96	R	Orbitofrontal cortex: OP11,c anterior orbitalgyrus
−19	37	43	7.92	392	L	Superior frontal sulcus, anterior
39	9	−15	7.84	176	R	Anterior insula: transverse insular gyrus
9	3	3	10.11	424	R	Basal forebrain: caudate, bed nucleus of the stria terminalis, rostrodorsal hypothalamus
−5	1	1	10.56	376	L	Basal forebrain: caudate, bed nucleus of the stria terminalis, rostrodorsal hypothalamus
57	−5	23	12.64	12,736	B	Central cortexb
57	−5	23	12.64	3,024	R	Central sulcus
−3	−22	45	10.87	1,096	B	Cingulate sulcus, posterior; Cingulate sulcus, midcingulate
−1	−31	57	8.44	160	B	Precentral gyrus
−52	−7	25	11.53	6,912	L	Central sulcus
23	−9	−13	22.02	2,696	R	Basal forebrain: piriform cortex, sublenticular extended amygdala (ventral amygdalofugal pathway), amygdala (amygdalohippocampal area, basolateral, basomedial, cortical, lateral, and medial), anterior hippocampus, brainstem
−19	−11	−13	20.91	2,720	L	Basal forebrain: putamen, piriform cortex, sublenticular extended amygdala (ventral amygdalofugal pathway), amygdala (amygdalohippocampal area, basolateral, basomedial, cortical, lateral, and medial), anterior hippocampus, brainstem
51	−12	−13	10.6	4,904	R	Temporal lobe: superior temporal gyrus, planum polare; parietal operculum; superior temporal sulcus
−37	−15	17	10.73	6,400	L	Posterior insula: central operculum, parietal operculum, posterior insula (dorsal portion of the long gyri), planum temporale, Heschl's gyrus, superior temporal sulcus
39	−15	17	10.89	1,096	R	Posterior insula: central operculum, parietal operculum, posterior insula (dorsal portion of the long gyri)
53	−23	45	6.38	80	R	Postcentral sulcus
53	−27	57	6.73	104	R	Postcentral gyrus
25	−37	59	8.3	592	R	Postcentral sulcus
−44	−50	−17	8.1	192	L	Temporal lobe: inferior temporal gyrus, temporooccipital; fusiform, temporooccipital
37	−52	−21	7.99	208	R	Temporal lobe: inferior temporal gyrus, temporooccipital; fusiform, temporooccipital
−1	−53	17	13.43	7,632	B	Posterior cingulate/precuneus
57	−63	11	9.33	2,272	R	Lateral occipital cortex
29	−83	−19	6.39	88	R	Fusiform, occipital

Note. Abbreviations: B, bilateral; BA, Brodmann area; Ce, central nucleus of the amydala; L, left hemisphere; R, right hemisphere.

^aWhole‐brain regression analysis (p < .05, whole‐brain Šidák corrected, k ≥ 80 mm³).

^bFor large clusters, subregions were identified using T ≥ 7 and are shown in italics.

^cAreas 10r/m and 11 as described by Ongur et al. (2003).

Table 4.

Regions showing significant functional connectivity with both the BST and the Cea

x	y	z	mm3	Hemisphere	Region(s)/subregions
1	61	21	48	R	Dorsomedial prefrontal cortex: BA10
3	59	17	16	R	Dorsomedial prefrontal cortex: BA10
1	57	13	24	R	Dorsomedial prefrontal cortex: BA10
1	53	19	80	R	Dorsomedial prefrontal cortex: BA10
−1	49	27	24	L	Dorsomedial prefrontal cortex: BA10
1	39	−15	296	R	Ventromedial prefrontal cortex: OP10r/mb; inferior frontopolar gyrus; rostral gyrus
−21	27	37	304	L	Superior frontal sulcus, anterior
55	7	−3	8	R	Temporal pole
63	7	−1	664	R	Planum temporale
9	5	−1	384	R	Basal forebrain: caudate, bed nucleus of the stria terminalis, rostrodorsal hypothalamus
−9	5	35	3,448	L	Cingulate: cingulate sulcus, posterior midcingulate; cingulate sulcus, posterior
−5	3	−1	312	L	Basal forebrain: caudate, bed nucleus of the stria terminalis
5	1	−3	8	R	Bed nucleus of the stria terminalis
−53	1	−1	40	L	Planum polare
53	1	−1	24	R	Planum polare
−1	1	47	8	L	Juxtapositional lobule
−17	−3	−15	376	L	Dorsal amygdala: amygdalohippocampal area, central, cortical, medial
63	−3	17	2,640	R	Central sulcus
61	−5	−13	392	R	Superior temporal sulcus
29	−11	−23	976	R	Hippocampus
−41	−15	31	2,648	L	Central sulcus
5	−15	73	8	R	Precentral gyrus
63	−17	−7	8	R	Superior temporal sulcus
−53	−17	9	8	L	Heschl's gyrus
−21	−19	−17	616	L	Hippocampus/dorsal amygdala: basolateral, basomedial, central, medial
−57	−19	9	152	L	Planum temporale
13	−19	39	40	R	Cingulate sulcus, posterior
3	−19	67	16	R	Precentral gyrus
−47	−25	3	880	L	Planum temporale
47	−25	7	728	R	Planum temporale
−25	−31	67	96	L	Postcentral gyrus
3	−33	49	16	R	Posterior cingulate
27	−37	55	232	R	Postcentral sulcus
−21	−39	63	128	L	Postcentral gyrus
3	−39	63	8	R	Postcentral gyrus
11	−53	1	6,792	R	Posterior cingulate/precuneus
55	−57	19	136	R	Angular gyrus
45	−59	29	8	R	Lateral occipital cortex
31	−85	−19	88	R	Occipital fusiform

Note. Abbreviations: B, bilateral; BA, Brodmann area; BST, bed nucleus of the stria terminalis; Ce, central nucleus of the amydala; L, left hemisphere; R, right hemisphere.

^aMinimum conjunction (Boolean “AND”) analysis (p < .05, whole‐brain Šidák corrected, k ≥ 80 mm³).

^bArea 10r/m as described by Ongur et al. (2003).

Figure 3.

The BST and the Ce are functionally linked via the SLEA. Clusters in the region of the SLEA (cyan arrowheads). Conventions are similar to Figure 2. Abbreviations: BST = bed nucleus of the stria terminalis; Ce = central nucleus of the amygdala; L = left hemisphere; R = right hemisphere; SLEA = sublenticular extended amygdala. See Figures 2 and 5 for additional views of these contrasts [Color figure can be viewed at http://wileyonlinelibrary.com]

Compared to the Ce, the BST showed significantly stronger coupling with several subcortical regions, including the basal ganglia (i.e., nucleus accumbens, caudate, and putamen), thalamus, and the brainstem in the region of the dorsal periaqueductal gray (PAG) (Figure 4, Supporting Information, Figure S9, and Table 5). The only subcortical regions showing stronger functional connectivity with the Ce were located in the dorsal amygdala and anterior hippocampus, and included the amygdalohippocampal area and basolateral, basomedial, cortical, and medial nuclei.

Figure 4.

Differential functional connectivity of the BST versus Ce. Results of a paired t test comparing the intrinsic functional connectivity of the BST and Ce. The left and right columns depict regions showing significantly stronger coupling with the BST and Ce, respectively. For ease of interpretation, differences were only examined in the subset of 12,004 voxels, where functional connectivity was significant for the BST, the Ce, or both seeds (Figures 2 and 3). Consistent with other analyses, results were thresholded at p < .05 Šidák corrected for the extent of the 12,004‐voxel mask. Results revealed significantly stronger coupling between the BST and the basal ganglia, including the caudate, putamen, and nucleus accumbens (cyan arrowheads). The BST also showed significantly stronger connectivity with the thalamus (magenta arrowheads) and a region of the brainstem consistent with the dorsal periaqueductal gray (green arrowheads; see also Supporting Information, Figure S9). The only regions showing stronger connectivity with the Ce were neighboring regions of the amygdala (white arrowheads), including voxels in the region of the amygdalohippocampal area, anterior hippocampus (not depicted) and the basolateral, basomedial, cortical, and medial nuclei. Note: Results are depicted here and reported in the accompanying tables for clusters of at least 80 mm³. See Figure 5 for additional views of the BST > Ce contrast. Abbreviations: BST = bed nucleus of the stria terminalis; Ce = central nucleus of the amygdala; L = left hemisphere; R = right hemisphere [Color figure can be viewed at http://wileyonlinelibrary.com]

Table 5.

Regions showing significant differences in intrinsic functional connectivity between the BST and the Cea

Effect	x	y	z	t	mm3	Hemisphere	Region(s)/subregions
BST > Ce	−25	55	31	6.8	80	L	Frontal pole: BA9/BA10
	2	45	−1	8.43	344	B	Ventromedial prefrontal cortex: OP10r/mb; inferior frontopolar gyrus; rostral gyrus; anterior cingulate, pregenual
	21	41	31	7.12	96	R	Superior frontal sulcus, anterior
	−25	41	35	5.69	112	L	Superior frontal sulcus, anterior
	11	37	−3	7.11	96	R	Cingulate: cingulate sulcus, pregenual
	7	36	25	9.94	3,504	B	Cingulate: cingulate sulcus, pregenual; cingulate sulcus, anterior midcingulate
	49	23	−9	7.53	80	R	Orbitofrontal cortex: OP47, Basal operculum
	6	5	−2	17.15	10,472	B	Basal forebrain: caudate, putamen, globus pallidus, nucleus accumbens, rostrodorsal hypothalamus, sublenticular extended amygdala (ventral amygdalofugal pathway), thalamus
	3	−11	35	6.73	128	R	Posterior cingulate
	−1	−17	−21	7.06	80	L	Brainstem ventral to the interpeduncular cistern
	−3	−23	−1	7.34	112	L	Periaqueductal gray, dorsolateral
	5	−24	−3	8.38	136	R	Periaqueductal gray, dorsolateral
	3	−27	25	10.17	968	B	Posterior cingulate
	4	−35	47	8.45	800	B	Posterior cingulate
	13	−47	31	5.94	104	R	Posterior cingulate
	−7	−69	33	8.82	288	L	Precuneus
	1	−75	43	6.89	232	B	Precuneus
	−8	−81	3	6.86	216	L	Calcarine sulcus
	9	−87	1	7.59	488	R	Calcarine sulcus
Ce > BST	25	−9	−15	−14.31	536	R	Anterior hippocampus and amygdala: amygdalohippocampal area, anterior hippocampus, basolateral, basomedial, cortical, medial
	−21	−10	−15	−11.19	504	L	Amygdala: amygdalohippocampal area, anterior hippocampus, basolateral, basomedial, cortical, medial

Note. Abbreviations: B, bilateral; BST, bed nucleus of the stria terminalis; Ce, central nucleus of the amygdala; L, left hemisphere; R, right hemisphere.

^aPaired t test for the subset of 12,004 voxels showing significant functional connectivity with the BST, Ce, or both seeds (p < .05, Šidák corrected for the extent of the 12,004‐voxel mask).

^bArea 10r/m as described by Ongur et al. (2003).

3.2. Cortical regions

As shown in Figures 2 and 5, the BST and the Ce showed significant functional connectivity with several cortical regions, including the ventromedial prefrontal cortex (vmPFC), posterior MCC, posterior insula, posterior cingulate/precuneus, and parts of the ventral visual processing stream (e.g., superior temporal sulcus, fusiform cortex) (Tables 2, 3, 4). As shown in Figure 5, relative to the Ce, the BST displayed significantly stronger coupling with a cluster centered on the anterior MCC that extends into the pregenual anterior cingulate cortex (pgACC) and vmPFC (Figure 5, far‐right panels, and Table 5). As detailed in the Supporting Information, Figure S10, control analyses indicated that these effects could not be attributed to regional differences in signal quality, as indexed by several widely used metrics (e.g., the temporal signal‐to‐noise ratio [tSNR]).

Figure 5.

Intrinsic functional connectivity of the EAc and midline cortical regions. The first two columns depict the results of whole‐brain regression analyses for the BST and Ce seed regions, respectively (p < .05, whole‐brain Šidák corrected). The third column depicts the intersection (Boolean “AND”) of the two thresholded maps. The fourth column depicts the results of a paired t test comparing the intrinsic functional connectivity of the BST and Ce (p < .05, small‐volume Šidák corrected). Both seeds show significant functional connectivity with the posterior cingulate/precuneus (green arrowhead), posterior MCC (cyan arrowheads), and vmPFC (magenta arrowhead). Relative to the Ce, the BST shows significantly stronger coupling with the anterior MCC and pgACC (brown arrowheads) and the vmPFC (orange arrowheads). Orange inset depicts a coronal slice through the vmPFC cluster, which extends along the rostral–caudal axis from area 10r/m and the inferior frontopolar gyrus to the rostral gyrus and pgACC. Conventions are similar to Figure 2 (first three columns) and Figure 4 (fourth column). See Figures 2, 3, 4 for additional views of these contrasts. Abbreviations: BST = bed nucleus of the stria terminalis; Ce = central nucleus of the amygdala; EAc = central divisions of the extended amygdala; L = left hemisphere; MCC = midcingulate cortex; pgACC = pregenual anterior cingulate cortex; R = right hemisphere; vmPFC = ventromedial prefrontal cortex [Color figure can be viewed at http://wileyonlinelibrary.com]

4. DISCUSSION

The EAc plays a central role in assembling states of fear and anxiety and is implicated in the development, maintenance, and recurrence of a range of debilitating psychiatric disorders. The present findings provide new insights into the normative architecture of the EAc functional network. Our results indicate that the BST and the Ce are robustly interconnected via the SLEA (Figure 3 and Supporting Information, Figure S8), consistent with anatomical and functional tracing studies in monkeys (Birn et al., 2014; Oler et al., 2012, 2017). By and large, the BST and the Ce showed patterns of functional connectivity that were similar to one another and concordant with prior human imaging research (Table 6). Both regions showed significant coupling with subcortical and cortical regions implicated in fear and anxiety—including the anterior hippocampus, insula, MCC, and vmPFC (Figures 2 and 5)—reinforcing the hypothesis that these regions represent a functionally coherent macro‐circuit (Alheid & Heimer, 1988; Fox et al., 2015a; Fudge et al., 2017; Oler et al., 2012; Shackman & Fox, 2016; Fox & Shackman, in press).

Table 6.

Intrinsic functional connectivity of the EAc in human imaging studiesb, f

Seed	Citation	NAcc	Cd	Putamen	GP	BST	SLEA	Amygdala	Hippocampus	Thalamus	PAG	vmPFC/OFC	pgACC	MCC	Insula	Precuneus
BST	Present study	+	+	+	+	N/A	+	+	+	+	+	+	+	+	+g	+
	Avery et al., 2014, c	+	+	+	+	N/A		+	+	+		+	+	+	+a, g	+
	Torrisi et al., 2015	+	+	+		N/A	+	+	+	+	+	+	+		+g	+
		3/3	3/3	3/3	2/3	N/A	2/3	3/3	3/3	3/3	2/3	3/3	3/3	2/3	3/3	3/3
Ce	Present study		+			+	+	+	+			+		+	+a, g	+
	Gorka et al., 2017, d	+	+	+		+	+	+	+	+	+	+	+		+a, g	+
	Oler et al., 2012, e	+		+		+		+	+	+					+
		2/3	2/3	2/3	0/3	3/3	2/3	3/3	3/3	2/3	1/3	2/3	1/3	2/3	3/3	2/3

Note. Abbreviations: BST = bed nucleus of the stria terminalis; Cd = caudate; Ce = central nucleus of the amygdala; GP = globus pallidus; MCC = midcingulate cortex; NAcc = nucleus accumbens; OFC = orbitofrontal cortex; PAG = periaqueductal gray; pgACC = pregenual anterior cingulate cortex; SLEA = sublenticular extended amygdala; vmPFC = ventromedial prefrontal cortex.

^aAnterior.

^bThis table is not meant to be comprehensive and some regional labels (vmPFC/OFC) encompass multiple subdivisions. Plus signs (+) indicate significant clusters. Empty cells indicate an absence of positive evidence in the published report. In some cases this reflects the absence of significant functional connectivity at the chosen threshold. In other cases, it simply indicates the omission of a specific label (e.g., SLEA). Regardless, empty cells should not be interpreted as indicating an absence of coupling (Fox et al., 2018). Motzkin et al. (2015) do not provide a detailed table of significant clusters and so are not included here, although it merits comment that they do report significant BST connectivity clusters in the pgACC and the vmPFC/OFC. McMenamin et al. (2014) do not provide a detailed table and are also not included, although they too provide visual evidence of a significant BST cluster at the intersection of anterior MCC and pgACC and extending into the edge of vMPFC (rostral gyri). Finally, although Birn et al. (2014) do provide detailed results, their study focused on a large (n = 89) sample of monkeys and so are not included. Nonetheless, it merits comment that they observed significant coupling between the Ce and several relevant regions, including the pgACC, insula, BST, thalamus, and neighboring regions of the amygdala. They also report a significant negative association between Ce–vmPFC functional connectivity and somatomotor responses to human intruder threat, with the cluster encompassing parts of areas 10m, 11, and 14.

^cAlthough Avery et al. (2014) also do not provide a detailed table of significant clusters, they do provide a dense montage of sagittal slices and a brief verbal summary and so are included.

^dGorka et al. (2017) only provide a detailed table for clusters showing significant functional connectivity with both the BST and the Ce. Relative to the Ce, they report significantly greater coupling between the BST and several regions, including the MCC, posterior cingulate, caudate, and NAcc. Conversely, they report significantly greater coupling between the Ce, insula, and neighboring regions of the amygdala.

^eOler et al. (2012) and Birn et al. (2014) did observe significant functional connectivity between the Ce and SLEA in a large sample of anesthetized monkeys.

^fFrom the perspective of generating cumulative knowledge, this table underscores the need to provide detailed cluster tables for every key contrast and/or share data using http://NeuroVault.org.

^gposterior.

Despite their many similarities, it is unlikely that the BST and the Ce are interchangeable (Fox & Shackman, [Link]; Shackman & Fox, 2016). Indeed, the BST showed significantly stronger connectivity with anterior cortical regions (anterior MCC, pgACC, and vmPFC), with the posterior cingulate/precuneus, with the medial temporal lobe (striatum and SLEA), and with the brainstem in the region of the dorsal PAG (Supporting Information, Figure S9), whereas the Ce showed stronger connectivity with neighboring regions of the amygdala, amygdalohippocampal area, and anterior hippocampus (Figures 4 and 5)—observations that largely align with recent high‐resolution fMRI research (Gorka et al., 2017) (cf. Table 1). Supplementary analyses indicated that these effects were not a consequence of regional differences in signal quality (e.g., tSNR).

We also observed significant coupling between the BST, the Ce, and the vmPFC (i.e., inferior frontopolar gyrus, rostral gyrus, and area OP10), although this effect was stronger for the BST seed region (Figure 5). This pattern is consistent with other work leveraging the enhanced resolution afforded by 7T fMRI (Gorka et al., 2017; their figure 2e) and is particularly interesting in light of several recent observations in nonhuman primate models of fear and anxiety. First, intrinsic coupling between the Ce and vmPFC covaries with the intensity of defensive behaviors and neuroendocrine activity elicited by exposure to human intruder threat in monkeys (Birn et al., 2014). Second, metabolic activity in the Ce, BST, and vmPFC, as well as the anterior hippocampus and PAG, covaries with these same anxiety‐related responses (Fox et al., 2015b). Third, vmPFC lesions have been shown to reduce these defensive responses and imaging research suggests that this anxiolytic effect is likely to be mediated by “downstream” alterations in BST metabolism (Fox et al., 2010; Kalin, Shelton, & Davidson, 2007; Motzkin et al., 2015; Rudebeck, Saunders, Prescott, Chau, & Murray, 2013). These and other observations (e.g., Grayson et al., 2016; Kalin et al., 2016, 2004; Meyer et al., 2017; Mobbs et al., 2009, 2007, 2010) motivate the hypothesis that fear and anxiety partially reflect a core neural system encompassing the BST, Ce, vmPFC, anterior hippocampus, and PAG (Fox et al., 2015b; Oler et al., 2016; Shackman et al., 2016b).

Our results revealed evidence of robust coupling between the BST, Ce, and rostral cingulate and they hint at a rostro‐caudal gradient: both seeds showed coupling with the posterior MCC, while the BST showed significantly stronger coupling with a cluster centered on the anterior MCC (Figure 5). Notably, the MCC and a region consistent with the BST are frequently co‐activated in imaging studies of Pavlovian fear conditioning (Fullana et al., 2016; Mechias, Etkin, & Kalisch, 2010) and uncertain threat anticipation (Alvarez et al., 2011, 2015; Choi et al., 2012; Grupe et al., 2013; Herrmann et al., 2016; Klumpers et al., 2015; McMenamin et al., 2014; Meyer et al., 2017; Somerville et al., 2010). We have previously hypothesized that the MCC uses information about pain, negative feedback, punishment, and threat to bias responding in situations where the optimal course of action is uncertain or risky (Cavanagh & Shackman, 2015; Shackman et al., 2011) (see also de la Vega et al., 2016) and the present results highlight the potential importance of communication between the MCC and the EAc, particularly the BST, for this kind of adaptive control. A key challenge for future research will be to more formally characterize the nature of task‐related interactions among these three key regions using graph‐theoretic or related analytic techniques (McMenamin et al., 2014; Najafi et al., 2017).

Clearly, a number of other important challenges remain. As with most brain imaging studies, our analyses do not permit mechanistic inferences and like other studies focused on functional connectivity, our conclusions are tempered by questions about the origins and significance of correlated fluctuations in the blood‐oxygen‐level‐dependent (BOLD) fMRI signal (Akam & Kullmann, 2014; Cabral, Kringelbach, & Deco, 2014; Logothetis, 2008). A key challenge for future research will be to use a combination of mechanistic (e.g., optogenetic) and whole‐brain imaging techniques to clarify the specific causal contributions of the regions highlighted here and more precisely delineate the nature of their functional interactions (Fox & Shackman, in press; Shackman & Fox, 2016; Wiegert, Mahn, Prigge, Printz, & Yizhar, 2017).

Existing treatments for anxiety disorders are inconsistently effective or associated with significant adverse effects (Bystritsky, 2006; Cloos & Ferreira, 2009; Craske et al., 2017; Cuijpers, Cristea, Karyotaki, Reijnders, & Huibers, 2016; James, James, Cowdrey, Soler, & Choke, 2015), highlighting the need to identify and understand the neural mechanisms controlling the experience and expression of fear and anxiety. Building on prior mechanistic and imaging research, the present study indicates that the BST and the Ce are marked by broadly similar patterns of intrinsic functional connectivity, with both regions showing significant coupling with the EAc, anterior hippocampus, insula, MCC, and vmPFC. Despite these similarities, the BST displayed significantly stronger connectivity with the rostral cingulate and vmPFC. These observations provide a baseline against which to compare a range of special populations—including individuals at risk for developing mental illness and patients suffering from psychiatric disorders—and inform our understanding of the role of the EAc in normal and pathological fear and anxiety. The use of a relatively large sample increases our confidence in the robustness of these results (Cremers, Wager, & Yarkoni, 2017; Fox, Lapate, Davidson, & Shackman, 2018; Poldrack et al., 2017). Finally, from a methodological perspective, these results highlight the value of several new techniques for EAc seed prescription and image registration/normalization. The former is likely to be useful for other investigators focused on the BST and Ce, while the latter will be advantageous for any investigator confronted with the problem of spatially normalizing structural images that have been modified—anatomically “anonymized” or “de‐identified”—prior to public release (Holmes et al., 2015; Nooner et al., 2012).

5. CONTRIBUTIONS

R.M.T, A.J.S., and J.F.S. designed the study. M.D.S. coordinated data extraction. B.M.N. developed and implemented the protocol for segmenting the Ce seed and the HyperEdge method. J.F.S. developed and implemented the novel image registration/normalization pipeline. R.M.T. and J.F.S. processed data. J.F.S. and A.J.S. analyzed data. R.M.T., A.J.S., and A.S.F. interpreted data. R.M.T., A.J.S., A.S.F., and B.M.N. wrote the article. A.J.S., R.M.T., B.M.N., and A.S.F. created figures. R.M.T. and A.J.S. created tables. S.T. provided theoretical guidance. A.J.S. funded and supervised all aspects of the study. All authors contributed to reviewing and revising the article and approved the final version.

DATA AVAILABILITY/SHARING

Key statistical maps are available in NeuroVault.org. Raw data are publicly available (http://fcon_1000.projects.nitrc.org/indi/enhanced/).

Supporting information

Additional Supporting Information may be found online in the supporting information tab for this article.

Supporting Table 1

Tillman RM, Stockbridge MD, Nacewicz BM, et al. Intrinsic functional connectivity of the central extended amygdala. Hum Brain Mapp. 2018;39:1291–1312. 10.1002/hbm.23917