Altered effective connectivity of central autonomic network in response to negative facial expression in adults with cannabis use disorder

Abstract

BACKGROUND:

Cannabis use is associated with an increased risk of stress-related adverse cardiovascular events. Because brain regions of the central autonomic network (CAN) largely overlap with brain regions related to the neural response to emotion and stress, the CAN may mediate the autonomic response to negative emotional stimuli. We aimed to obtain evidence whether neural connectivity of the CAN is altered in individuals with cannabis use disorder (CUD) when they are exposed to negative emotional stimuli.

METHODS:

Effective (directional) connectivity (EC) analysis using Dynamic Causal Modeling was applied to functional magnetic resonance imaging data acquired from 23 CUD subjects and 23 controls of the Human Connectome Project while they performed an emotional face-matching task with interleaving periods of negative-face (fearful/angry) and neutral-shape stimuli. The EC difference (modulatory change) was measured during the negative-face trials relative to the neutral-shape trials.

RESULTS:

The CUD group was similar to the controls in non-imaging measures and brain activations, but showed greater modulatory changes in the left amygdala to hypothalamus EC (positively associated with Perceived-Stress score), the right amygdala to bilateral fusiform gyri ECs (positively associated with Perceived-Stress score), and the left ventrolateral prefrontal cortex to bilateral fusiform gyri ECs (negatively associated with Perceived-Stress score).

CONCLUSIONS:

The left amygdala to hypothalamus EC and the right amygdala to bilateral fusiform gyri ECs are possibly part of circuits underlying the risk of CUD individuals to stress-related disorders. Correspondingly, the left ventrolateral prefrontal cortex to bilateral fusiform gyri ECs are possibly part of circuits reflecting a protective mechanism.

INTRODUCTION

Accumulated evidence shows an association between cannabis use and adverse cardiovascular events (1-3). The brain central autonomic network (CAN), composed of brain regions including but not limited to ventromedial prefrontal cortex (VMPFC), amygdala, hypothalamus, insula, hippocampus, and cingulate cortex (4-7), helps regulate cardiovascular function (8-12). The CAN maintains homeostasis by modulating the release of hormonal factors, chemical messengers, and neurotransmitters (11). According to the model proposed by (13), the autonomic response is dependent on inputs to the hypothalamus from the amygdala, VMPFC, and insula. By integrating these inputs, the hypothalamus helps maintain homeostasis by controlling involuntary functions such as breathing, blood pressure, and heart rate (14).

Emotional or psychological stress is one of probable triggers of adverse cardiovascular events (15), possibly related to the fact that the CAN regions largely overlap with neuronal regions which are involved in the neural response to emotion (16, 17) and stress (4). Through these common regions, the CAN may be able to mediate the autonomic response from emotional stress (such as anxiety, and fear). According to previous models (7, 18), when exposed to anger/fear/sadness or environmental/psychological stressors, the relevant negative emotional information was first processed/evaluated by insula, amygdala, and anterior cingulate cortex (7, 18). For the individuals with risk of stress-related disorders, the evaluation could facilitate subsequent response in the subcortical autonomic network consisting of hypothalamus, nucleus tractus solitarius, rostral ventrolateral medulla, and parabrachial nucleus, which is signaled by the amygdala (7). The response in the subcortical autonomic network could then result in subsequent physiological responses (7). The roles of other key CAN regions, i.e., VMPFC and hippocampus, were not specified in these models (7, 18). Other models (4, 19) suggest that the VMPFC, together with other prefrontal regions, may exert top-down regulation of subcortical regions such as hypothalamus in response to negative emotion or stress.

Chronic cannabis use is associated with altered brain-stress pathways in animal models (20, 21). One method of exploring this in humans without causing undue stress to participants is to present negative (fearful or angry) facial expressions or scenes, where viewing negative facial expressions has triggered a physiological response (22-27). For example, when matching fearful/angry faces, healthy individuals showed increased activation in bilateral amygdala which was accompanied by increased physiological response as measured by skin conductance response (27).

Several functional neuroimaging studies have been published demonstrating altered brain networks in response to passive viewing of emotional stimuli in cannabis users. Some of these studies (28-30) found that relative to controls, the cannabis users had lower activation in prefrontal regions by negative emotional stimuli. In addition, greater (31) or lower (28, 29) activation in amygdala, lower functional (non-directional) connectivity between amygdala and dorsolateral prefrontal cortex (32), and lower functional connectivity between amygdala and medial orbitofrontal cortex (33) were reported. As commented by (32), lower functional connectivity between amygdala and dorsolateral prefrontal cortex could reflect unsuccessful emotion regulation in the cannabis users. Similarly, poorer regulation of amygdala by frontal cortex when viewing emotional faces has been found in functional magnetic resonance imaging (fMRI) studies of individuals with increased (social) anxiety (34, 35).

None of these published studies investigated the direction of coupling between the hypothalamus and other CAN regions in response to negative emotional stimuli. Using data from the Human Connectome Project (HCP) (36), we studied the effective (directional) connectivities (ECs) among CAN regions in CUD and non-drug-using controls who performed an emotional face-matching task during fMRI with relatively high spatial and temporal resolution.

Based on existing models about neural circuitry involved in the reaction to unpleasant events (13), stress reactivity (4), drug addiction (37), and previous CUD studies reviewed above, we hypothesized the following: (1) In response to negative emotional stimuli in both groups, the alterations in the amygdala to hypothalamus ECs would be positively associated with stress scores; and the alteration in the VMPFC to hypothalamus EC would be negatively associated with stress scores. (2) Compared to the controls, the CUD individuals in response to negative emotional stimuli would show increased amygdala to hypothalamus EC, which could potentially facilitate the hypothalamic autonomic response; and decreased VMPFC to hypothalamus EC, reflecting reduced top-down emotion regulation.

METHODS AND MATERIALS

Participants

The data used in this study were from the HCP 900 Subjects Data Release. Written informed consent was obtained from all participants. The experiments were performed in accordance with relevant guidelines and regulations, and all experimental protocols were approved by the Virginia Commonwealth University Institutional Review Board.

The subject inclusion criteria for this analysis were: (I) right-handed; (II) lifetime cannabis dependence (for CUD subjects, see below); and (III) positive urine test for delta-9- tetrahydrocannabinol (for CUD subjects). The exclusion criteria for this study were: (I) higher than moderate level of nicotine dependence as determined by the score on the Fagerstrom Test for Nicotine Dependence (38); (II) breath alcohol concentration greater than 0.05 g/210 L; (III) positive urine test for cocaine, opiates, and drugs other than cannabis; (IV) positive urine test for cannabis (for controls only), and (V) meeting DSM-IV (39) criteria for alcohol dependence. Participants with diagnosis of lifetime alcohol abuse (rather than dependence) were not excluded because of comorbidity of cannabis use and alcohol consumption (40). A prior history of other drug abuse or dependence other than alcohol was not used as an exclusion criterion because HCP did not provide these diagnoses. Based on these criteria, 23 CUD subjects (CUD group) and 23 matched non-drug-using controls (CTL group) were included. This identical sample has been used in another study (41) published by our group. However, these two studies investigated different topics and used different data: emotion vs. working memory.

In-scanner emotion processing task

During the task (27, 42), the participants were instructed to decide which of two pictures (fearful/angry faces, or shapes) presented on the bottom of the screen matched the picture (face or shape) at the top of the screen. Face blocks alternated with shape blocks, with a total of 3 face blocks and 3 shape blocks. Each block was 21 s, including a 3 s task cue (“shape” or “face”) following by 6 trials of the same task (face or shape presenting for 2 s), which were separated by 1 s intervals.

Stress rating scale

To determine whether individual differences in negative-face-elicited EC alterations were related to individual differences in stress, we tested the relationship between EC modulation in response to negative faces and the raw score of Perceived-Stress from the Category of Emotion rating scale (43). According to the HCP Data Dictionary, the Perceived-Stress score was defined by individual perceptions about the nature of events and their relationship to the values and coping resources of an individual, and assessed how unpredictable, uncontrollable, and overloading respondents find their lives (43).

Measures of cannabis use and dependence

HCP used the Self-Reported Substance Use and Abuse measures from Semi-Structured Assessment for the Genetics of Alcoholism (SSAGA) (44) to quantify lifetime cannabis use and dependence. Specifically, the four measures were: “Ever used marijuana?”, “DSM Marijuana Dependence”, “Age at first marijuana use”, and “Times used marijuana”. See Table 2 for these measures of the included participants.

Table 2:

Measures of cannabis use and dependence, measures related to marijuana withdrawal, and measures of use of other drugs and corresponding statistical results, for the included CUD and CTL participants. Degrees of freedom were 44 for the Student-t tests. Fisher’s exact test was used to test the group difference in the portion of participants using other drugs more than 6 times. All the p values were two-tail.

Parameter	CUD (n=23)	Control (n=23)	Statistical results
Measures of cannabis use and dependence
Ever used marijuana?	23 Y, 0 N	11 Y, 12 N	N/A
DSM Marijuana Dependence	23 Y, 0 N	0 Y, 23 N	N/A
Age at first marijuana use	≤14 years old (n=6) 15-17 years old (n=10) 18-20 years old (n=5) ≥21years old (n=2)	≤14 years old (n=1) 15-17 years old (n=4) 18-20 years old (n=4) ≥21years old (n=2) N/A (n=12)	N/A
Times used marijuana	101-999 (n=5) ≥1000(n=18)	0(n=12) 1-10 (n=5) 11-100 (n=6)	N/A
Measures related to marijuana withdrawal
Anger-Aggression (range: 43-75)	58.9 ± 12.4 (range: 43.0-73.1)	53.4 ± 8.4 (range: 43.0-70.2)	t=1.760 p=0.090
Sleep (Pittsburgh Sleep Questionnaire) Total Score (range: 0-16)	5.2 ± 2.8 (range: 0-13)	5.2 ± 3.0 (range: 0-13)	t=0.000 p=0.999
ASR DSM Anxiety Problems Raw Score (range: 0-14)	3.6 ± 3.3 (range: 0-13)	3.9 ± 1.9 (range: 1-8)	t=0.378 p=0.707
ASR DSM Depressive Problems Raw Score (range: 0-36)	5.1 ± 4.9 (range: 0-22)	4.6 ± 3.0 (range: 0-12)	t=0.417 p=0.678
ASR Somatic Complaints Raw Score (range: 0-24)	3.5 ± 5.3 (range: 0-23)	3.5 ± 3.7 (range: 0-24)	t=0.000 p=0.999
Measures of use of other drugs
Times used illicit drugs	<6 (n=20) ≥6 (n=3)	<6 (n=23) ≥6 (n=0)	p=0.233
Times used cocaine	<6 (n=18) ≥6 (n=5)	<6 (n=21) ≥6 (n=2)	p=0.414
Times used hallucinogens	<6 (n=18) ≥6 (n=5)	<6 (n=22) ≥6 (n=1)	p=0.187
Times used opiates	<6 (n=18) ≥6 (n=5)	<6 (n=22) >6 (n=1)	p=0.187
Times used sedatives	<6 (n=18) ≥6 (n=5)	<6 (n=22) ≥6 (n=1)	p=0.187
Times used stimulants	<6 (n=18) ≥6 (n=5)	<6 (n=22) ≥6 (n=1)	p=0.187

Measures related to cannabis withdrawal symptoms

HCP provided five measures which are related to cannabis withdrawal symptoms (DSM-5): Anger-Aggression as measured using NIH Toolbox Anger-Physical Aggression Survey, Sleep Total Score as measured using Pittsburgh Sleep Questionnaire, Adult Self-Report (ASR) DSM Anxiety Problems Raw Score (26), ASR DSM Depressive Problems Raw Score (26), ASR Somatic Complaints Raw Score (26). See Table 2 for these measures (mean, standard deviation, and range) of the included participants.

Measures of alcohol and tobacco usage

HCP used the SSAGA to quantify tobacco and alcohol usage. Following (45), a previous study of CUD using HCP data, we quantified alcohol usage using composite measure calculated from the average of the Z-scores from five SSAGA measures. The tobacco use could be similarly quantified (45). However, among the four SSAGA measures used in (45), only one measure “Total times used/smoked any tobacco in past 7 days” was completely available for all the participants included in this study. Thus this measure was used to evaluate tobacco use in the current study.

Measures of use of other drugs

HCP used SSAGA to measure the use of other drugs. Specifically, the numbers of times of using illicit drugs, cocaine, hallucinogens, opiates, sedatives, and stimulants were recorded in the ranges of 0, 1-2, 3-5, 6-10, and >10 times. See Table 2 for these measures of the included participants. For simplicity, only two ranges (<6 and >6) were summarized in Table 2.

Physiological recordings

HCP had released physiological data (cardiac and respiratory signals acquired during MRI scans) (42). However, the physiological recording files were not released for 7 of the 46 included subjects. In addition, as described in the HCP S1200 Release Reference Manual, HCP had identified unknown timing errors in the physiological recordings in the HCP 900 Subjects Data Release, which is the source of the data used in the present study. Therefore, cardiac and respiratory data acquired during MRI scans were not used in the current study.

FMRI acquisition

As described in (42), whole-brain gradient-echo, echo-planar fMRI imaging data was acquired with a 32-channel head coil on a modified 3-T Siemens Skyra scanner (Erlangen, Germany), with repetition time=720 ms, echo time=33.1 ms, flip angle=52 degrees, bandwidth=2290 Hz/pixel, in-plane field-of-view=208×180 mm, 72 slices, 2.0 mm isotropic voxels, multi-band acceleration factor of 8. Two fMRI runs, with left-to-right and right-to-left phase encodings respectively, were acquired, preprocessed, and concatenated for DCM analysis.

fMRI Preprocessing

We performed DCM on fMRI data already “minimally” preprocessed by HCP per their optimized, empirically-tested pipeline (46), including gradient unwarping, motion correction, fieldmap-based EPI distortion correction, brain-boundary-based registration of EPI to structural T1-weighted scan, nonlinear registration into standard MNI152 space, and grand-mean intensity normalization. Following (47), the fMRI data was spatially smoothed with a 4-mm gaussian kernel, using Statistical Parametric Mapping 12 (SPM12) software (http://www.fil.ion.ucl.ac.uk/spm/).

SPM univariate analysis

Based on the contrast of face minus shape, standard SPM univariate analyses (48, 49) were conducted in order to select the DCM nodes based on group differences in task-elicited brain activation. See Supplementary Information for the detailed description. Anatomical labels for regions of activation were determined using the Anatomical Automatic Labeling2 (AAL2) toolbox (50), and hypothalamus atlases (51, 52).

Dynamic causal modeling

FMRI-based Dynamic Causal Modeling (DCM) is a biophysical model of how the neuronal connectivity generates the observed fMRI signal (53). DCM has been described in detail elsewhere (48, 49, 53). In brief, DCM is a dynamical system of bilinear differential state equations with coefficients (in units of Hz) (53). A node in the model that receives driving input is the brain region which first experiences a change in neuronal activity. This node then influences other nodes. The endogenous connectivity measures the EC strengths between nodes, regardless of the moment-to-moment switching on and off of inputs. Experimental conditions can modulate the endogenous connectivities. The parameters of the modulation effects quantify increased or decreased connectivity strength compared to the endogenous ECs.

Following (54), an experimental condition called “All-visual-stimuli” (reflecting common features of the face and shape stimuli) was used as a single driving input to the DCM, and another experimental condition called “emotional-face minus neutral-shape” (face-shape), reflecting the special effect of face over shape, was used as a single modulator of EC. Here, the changes of ECs corresponding to the face–shape modulator (relative to the endogenous ECs) are termed as modulatory changes. By definition, a modulatory change reflects the change in EC corresponding to the emotional-face trials minus the change in EC corresponding to the neutral-shape trials.

DCM nodes

The DCM nodes were selected from the clusters (results of SPM univariate analysis) activated by the task, a priori guided by existing models for neural circuitry involved in the reaction to unpleasant events (13), stress reactivity (4), face perception (55), and drug addiction (37). Within the activated clusters, three CAN regions (i.e., VMPFC, amygdala, and hypothalamus) are also regions in models of neural response to emotion and stress (16, 17). Thus, these regions were considered as candidate DCM nodes. Hippocampus was not considered as a DCM node because it was not considered as a key region in the stress reactivity model (4) that we used in this study. Insula was not considered as a DCM node because it showed only weak activation during the task. Left (L) and right (R) fusiform gyrus (FG) showed very strong activation during the task and are important regions in the neural response to face perception (55); thus they were considered as candidate DCM nodes. L ventrolateral prefrontal cortex (VLPFC) showed strong activation during the task, and imaging studies have revealed baseline decreases in lateral VLPFC function during dependence (37); thus L VLPFC was also considered as a candidate DCM node. In summary, the following seven regions were chosen as DCM nodes: VMPFC, L-VLPFC, L-amygdala, R-amygdala, L-FG, R-FG, and, hypothalamus. Each DCM node was defined as a sphere with its center at the voxel with the local maximum voxel T value within each regional activation (see Figure 1 and Supplementary Information for the detailed information about the DCM nodes). We did not directly use the MNI-based anatomical atlas released by HCP because brain regions in these atlases are generally larger than the area activated by the task. The DCM nodes in this study were constrained by the univariate brain activations such that the noisiest voxels could be excluded. The fMRI time series of each node, i.e., the blood-oxygen-level dependent (BOLD) signal over time in that node, was extracted by using the principal-eigenvariate of that node (49). Each time series was also adjusted by the F-contrast of the effects-of-interest (56).

Figure 1:

Location of all spheres used as DCM nodes, visualized with the BrainNet Viewer (http://www.nitrc.org/projects/bnv/) (85). L = left. R = right. The MNI coordinates (mm) and color codes for these VOIs were: VMPFC (0, 52, −10), L-VLPFC (−36, 32, −14), L-amygdala (−20, −6, −14), R-amygdala (22, −6, −14), L-FG (−38, −72, −14), R-FG (40, −54, −18), and R-hypothalamus (2, −4, −12). Corresponding to the color bar, VMPFC and L-VLPFC were shown in light blue color, L-amygdala and R-amygdala were shown in dark blue color, L-FG and R-FG were shown in dark purple color, and R-hypothalamus was shown in purple color. The radius of the R-hypothalamus node was 3 mm, and the radius for the remaining DCM nodes were 5 mm.

DCM Parametric Empirical Bayes (PEB) analysis

The PEB approach (57) as implemented in SPM12 (Revision 7219), was used to conduct group level analyses for the EC modulatory changes. Three PEB analyses were conducted: (1) testing the mean of each modulatory change across the CTL participants; (2) testing the group difference in each modulatory change between the CUD and CTL participants; and (3) testing the relationship between each modulatory change and the Perceived-Stress score across all participants using linear regression analysis. Because anxiety highly correlated with stress, a supplementary PEB analysis was conducted testing the relationship between each modulatory change and the ASR DSM Anxiety Problems Raw Score (Table 2) across all participants. The advantage of the linear regression analysis within the PEB framework is that the covariance among all DCM parameters in the model is automatically taken into consideration. In linear regression analysis, the beta coefficient is the slope of the linear relationship, i.e., the degree of change in the outcome variable per-unit of the predictor variable. For example, a beta of 0.01 indicates a change of 0.01 Hz in modulatory change per each unit change of Perceived-Stress score.

PEB uses Bayesian posterior inference (58), which has the advantage of lack of false positives; thus removing the need to contend with the multiple-comparison problem (58). In PEB posterior inference, the posterior probability (PP) is used as an indicator of the confidence in whether a modulatory change in a group is different from zero (or different compared to another group) or the confidence in the degree of linear relationship between variables. The PP (0⩽PP⩽1) is the conditional probability that is computed by PEB using Bayes rule after the available information (the likelihood function and the prior probability density of the model’s parameters) is taken into account (61). The higher the PP, the greater the confidence. In this paper, a group difference or linear relationship was considered reliable if PP>0.95.

RESULTS

Non-imaging results

As shown in Table 1, the two groups were not different in age, sex, handedness, education, alcohol usage, tobacco usage, Perceived-Stress score, and task performance. With group (CUD and CTL) as the between-subjects factor and visual stimulus type (face and shape) as the within-subjects factor, an analysis of variance (ANOVA) on reaction time showed no main effects of group (F=0.06; degree-of-freedom [df]=1,91; p=0.81), stimulus type (F=0.61; df=1,91; p=0.44), or their interaction (F=0.20; df=1,91; p= 0.65). Another ANOVA on the percentage of correct responses showed no main effects of group (F=1.64; df=1,91; p= 0.20), or any interaction between group and stimulus type (F=0.01; df=1,91; p=0.91). However, the main effects of stimulus type were statistically significant across both groups (F=11.62; df=1,91; p=0.001), with a greater percentage of correct responses during face trials than shape trials.

Table 1:

Demographic information, alcohol usage, tobacco usage, out-scanner measures of stress, inscanner behavioral performance of the CUD and CTL groups and corresponding statistical results. Degrees of freedom were 44 for the Student-t tests. All the p values were two-tail. F=female, M=male, L=left, R=right, RT=reaction time, PCR=percentage of correct responses.

Parameter	CUD (n=23)	Control (n=23)	Statistical results
Demographics
Age [years] mean and standard deviation (range)	28.2 ± 3.5 (22 - 33)	28.7 ± 3.7 (22 - 35)	t=0.471 p=0.640
Sex	6 F, 17 M	7 F, 16 M	p=1.000
Handedness	23 R, 0 L	23 R, 0 L	p=1.000
Education [years] mean and standard deviation (range)	14.6 ± 1.8 (11 - 17)	14.6 ± 2.1 (11 - 17)	t=0.000 p=1.000
Averaged Z-scores for the SSAGA measures of alcohol use (range)	−0.14 ± 2.26 (−5.62 - 4.05)	0.15 ± 2.40 (−3.89 - 4.93)	t=0.0.412 p=0.682
Total times used/smoked any tobacco in past 7 days (range)	22.9 ± 27.4 (0 - 75)	18.9 ± 25.3 (0 - 70)	t=0.514 p=0.610
Out-scanner measures of stress
Perceived Stress (range)	49.2 ± 10.4 (31.4 – 73.0)	47.3 ± 7.6 (32.8 – 60.5)	t=0.707 p=0.483
In-scanner behavioral performance
RT (emotion)	797.42 ± 155.12	778.47 ± 111.07	See main text
RT (neutral)	763.92 ± 138.76	769.49 ± 109.47
PCR (emotion)	98.41% ± 3.16%	99.42% ± 1.64%	See main text
PCR (neutral)	96.01% ± 5.15%	96.86% ± 3.05%

As shown in Table 2, the two groups were not significantly different in any of the five symptoms related to cannabis withdrawal that were measured (including the Anxiety-Problems Raw Score), nor in the number of times using other drugs.

SPM univariate analysis

There was no statistically significant group difference (FWE corrected cluster p>0.100, two-tail) in regional brain activation for the contrast of face–shape. Across both groups combined, the SPM one-sample t-test analysis found several clusters for the contrast of face–shape with uncorrected two-tail cluster level p < 0.050. These clusters were in portions of CAN regions such as VMPFC, amygdala, hypothalamus, insula, and hippocampus. See Figure 2 and Table 3 for the detailed information regarding these clusters. The clusters were used to constrain the DCM nodes. It is commonly accepted that less conservative statistical criteria can be used when brain activations are used for determining DCM nodes (53). The only purpose of this is to exclude the most noisy voxels from the DCM nodes and not to perform statistical inference.

Figure 2:

Brain clusters detected by the SPM12 second-level random effects one-sample t-test analysis for the contrast of face – shape with the cluster level p < 0.05 (uncorrected, two-tail). The clusters are overlaid in color on axial slices of the MNI brain template image in gray. Scale on color bar represents voxel t values. The reader’s left (L) side of each slice is the subjects’ left brain hemisphere.

Table 3.

The SPM12 second-level random effects one-sample t-test analysis result, across both groups combined, for the contrast of face-shape, with the cluster level p<0.05 (uncorrected, two-tail). x, y, and z = MNI standard space coordinates (mm). Negative x = Left hemisphere. L=left. R=right. The Hypothalamus was located in these clusters as determined by the hypothalamus atlases (55, 56). However, the number of activated voxels within the hypothalamus was unknown because the binary mask for hypothalamus was not available.

AAL2 Label	Number of voxels	Maximal t value within the labeled region	MNI coordinates [x y z] (mm) of voxel with maximal t
L middle frontal gyrus	10	2.86	−34, 20, 30
R middle frontal gyrus	94	3.89	50, 32, 22
L inferior frontal gyrus pars opercularis	38	3.14	−42, 18, 32
R inferior frontal gyrus pars opercularis	198	4.77	52, 18, 32
L inferior frontal gyrus pars triangularis	227	3.75	−44, 18, 26
R inferior frontal gyrus pars triangularis	476	4.67	52, 30, 20
L inferior frontal gyrus par orbitalis	187	6.60	−36, 32, −14
L olfactory gyrus	18	6.67	−26, 6, −18
R olfactory gyrus	23	5.18	30, 10, −20
L medial orbital frontal cortex	104	5.83	−4, 52, −12
R medial orbital frontal cortex	70	4.84	0, 52, −10
L insula	23	6.77	−26, 6, −18
R insula	43	4.25	32, 12, −20
L hippocampus	595	10.02	−18, −8, −14
R hippocampus	495	9.51	20, −6, −14
L paraHippocampal gyrus	101	5.52	−22, −6, −28
R paraHippocampal gyrus	144	6.51	18, −6, −18
L-amygdala	197	9.06	−20, −6, −14
R-amygdala	239	9.04	22, −6, −14
L calcarine fissure	185	7.93	−18, −98, −6
R calcarine fissure	410	12.30	22, −96, −6
L cuneus gyrus	17	2.63	−2, −74, 24
R cuneus gyrus	25	4.29	20, −96, 8
L lingual gyrus	152	11.36	−30, −90, −14
R lingual gyrus	325	12.40	22, −96, −8
R superior occipital gyrus	26	7.36	22, −96, 4
L middle occipital gyrus	410	10.77	−30, −94, −6
R middle occipital gyrus	190	11.03	28, −94, 0
L inferior occipital gyrus	528	11.44	−28, −94, −8
R inferior occipital gyrus	400	13.53	28, −94, −6
L fusiform gyrus	735	10.85	−38, −72, −14
R fusiform gyrus	604	12.48	40, −54, −18
L angular gyrus	63	3.74	−50, −70, 24
R angular gyrus	45	4.11	46, −64, 24
R precuneus	22	2.88	30, −52, 8
L putamen	43	4.30	−24, −4, −8
L pallidum	23	3.65	−22, −4, −6
R thalamus	20	5.09	12, −30, −2
R superior temporal gyrus	160	4.84	54, −4, −14
L superior temporal pole	157	9.88	−32, 6, −22
R superior temporal pole	381	8.03	34, 6, −22
L middle temporal gyrus	112	5.02	−50, −72, 18
R middle temporal gyrus	413	4.92	52, −6, 16
L middle temporal pole	39	5.11	−34, 10, −32
R middle temporal pole	166	4.39	40, 12, −32
L inferior temporal gyrus	153	10.28	−40, −48, −18
R inferior temporal gyrus	191	9.87	44, −58, −16

DCM PEB analysis

The first DCM PEB analysis tested the modulatory changes against zero across the CTL participants. For each EC, the mean modulatory change, and the corresponding PP are shown in Table 4.

Table 4.

The results of the group level PEB DCM analyses. Strength = modulatory change in EC. PP = posterior probability, VMPFC = ventromedial prefrontal cortex, VLPFC = ventrolateral prefrontal cortex, AMY = amygdala, FG = fusiform gyrus, HTH = hypothalamus, L = left, and R = right.

	CTL group		Group difference (CUD – CTL)		Linear regression of Strength on Perceived Stress scores		Linear regression of Strength on Anxiety Problems scores
CONNECTIVITY	Strength (Hz)	PP	Strength (Hz)	PP	beta	PP	beta	PP
VMPFC → L-VLPFC	0.0403	0.5091	0	0	0	0	0	0
VMPFC → L-AMY	0	0	−0.0977	1	0	0	0	0
VMPFC → R-AMY	0.2108	1	−0.0837	1	0	0	−0.0252	0.6924
VMPFC → L-FG	0	0	0	0	0	0	0	0
VMPFC → R-FG	−0.1345	1	0	0	0	0	0	0
VMPFC → R-HTH	−0.1136	0.9670	0.0477	0.6525	−0.0168	1	−0.0229	0.6658
L-VLPFC → VMPFC	0.2772	1	0.1380	1	0.0170	1	0.0700	1
L-VLPFC → L-AMY	−0.2761	1	0.2726	1	0.0243	1	−0.0725	1
L-VLPFC → R-AMY	0	0	0	0	0	0	0.0633	1
L-VLPFC → L-FG	−0.2261	1	0.3525	1	−0.0472	1	0	0
L-VLPFC → R-FG	0	0	0.3314	1	−0.0268	1	0	0
L-VLPFC → R-HTH	0.2968	1	−0.0755	0.7027	0	0	0.0903	1
L-AMY → VMPFC	−0.4841	1	−0.1451	1	0	0	−0.0282	0.6653
L-AMY → L-VLPFC	−0.2561	1	0	0	−0.0297	1	−0.0591	1
L-AMY → R-AMY	0	0	0	0	0	0	−0.1275	1
L-AMY → L-FG	0	0	0	0	0	0	−0.1048	1
L-AMY → R-FG	0	0	0	0	0	0	−0.0965	1
L-AMY → R-HTH	−0.2531	1	0.1061	1	0.0116	0.9751	0	0
R-AMY → VMPFC	0.1894	1	−0.1973	1	−0.0210	1	−0.1079	1
R-AMY → L-VLPFC	−0.2240	1	0	0	0.0166	1	0	0
R-AMY → L-AMY	0	0	0	0	0	0	0.0569	1
R-AMY → L-FG	−1.3848	1	0.5304	1	0.0235	1	0.1595	1
R-AMY → R-FG	−1.3526	1	0.4949	1	0.0265	1	0.1398	1
R-AMY → R-HTH	−0.6737	1	0.2343	1	0	0	0	0
L-FG → VMPFC	0	0	0	0	−0.0361	1	0	0
L-FG → L-VLPFC	−0.3241	1	0.2489	1	0	0	0	0
L-FG → L-AMY	0.1159	1	0.0809	0.8502	0	0	0.0562	1
L-FG → R-AMY	−0.5235	1	0.1217	1	0.0233	1	0.1331	1
L-FG → R-FG	0.2995	1	0	0	0	0	0	0
L-FG → R-HTH	0.1661	0.9635	−0.2835	1	0.0193	1	0	0
R-FG → VMPFC	−0.1600	1	0.1247	1	0.0275	1	−0.0480	1
R-FG → L-VLPFC	0.3827	1	−0.1498	1	0	0	−0.0246	0.6497
R-FG → L-AMY	0	0	−0.1266	1	−0.0067	0.6570	−0.0959	1
R-FG → R-AMY	0.6371	1	−0.0844	0.8564	−0.0276	1	−0.1259	1
R-FG → L-FG	0.6105	1	−0.2598	1	0.0273	1	0.0516	1
R-FG → R-HTH	0.0786	0.6495	0.1790	1	−0.0226	1	0	0
R-HTH → VMPFC	0.6686	1	0	0	0.0157	1	0.1396	1
R-HTH → L-VLPFC	0.5566	1	−0.3288	1	0.0089	0.7980	0.0795	1
R-HTH → L-AMY	0.1993	1	−0.2335	1	0	0	0	0
R-HTH → R-AMY	0.1057	0.8613	−0.1858	1	0	0	0.0882	1
R-HTH → L-FG	0.6067	1	−0.8443	1	−0.0184	1	0	0
R-HTH → R-FG	0.6090	1	−0.6839	1	−0.0175	1	0	0

The second DCM PEB analysis tested if the modulatory changes were different between the two groups. For each EC, the mean group difference in modulatory change (CUD–CTL), and the corresponding PP are shown in Table 4. Compared to the CTL group, the CUD group showed reliable larger modulatory changes in the L-amygdala to R-hypothalamus EC (group difference=0.1061 Hz; PP=1) and in the R-amygdala to R-hypothalamus EC (group difference=0.2343 Hz; PP=1). The group difference in the modulatory change in the VMPFC to R-hypothalamus EC was not reliable (group difference=0.0477 Hz; PP=0.6525). Six non-hypothesized ECs with the largest reliable modulatory changes were all ECs that terminated in bilateral FG (“to FG” ECs), specifically the ECs from R-amygdala, from L-VLPFC, and from R-hypothalamus to bilateral-FG (absolute values of group differences>0.30 Hz, each PP=1). A schematic diagram is shown in Figure 3, representing the results for the three hypothesized ECs and the six non-hypothesized “to FG” ECs.

Figure 3:

Schematic diagram representing the hypothesized ECs and the six non-hypothesized “to FG” ECs with the largest group differences in the modulatory changes. An endogenous EC is shown using lines ending with an arrow. A modulatory change in EC is shown by a line ending with a solid dot. Each number represents the group difference in the modulatory change (CUD minus CTL). An EC with modulatory change positively associated with the Perceived-Stress score is shown by a line in orange color; an EC with the modulatory change negatively associated with the Perceived-Stress score is shown by a line in green color; and an EC with the modulatory change not associated with the Perceived-Stress score is shown by a line in black color. All of the modulatory changes shown were reliable except for VMPFC to R-HTH EC. VMPFC = ventromedial prefrontal cortex, VLPFC = ventrolateral prefrontal cortex, AMY = amygdala, FG = fusiform gyrus, HTH = hypothalamus, L = left, and R = right.

The third DCM PEB analysis tested the linear regression of the EC modulatory changes on the Perceived-Stress scores for all the participants. For each EC, the regression beta and the corresponding PP are shown in Table 4. The L-amygdala to R-hypothalamus EC and the R-amygdala to bilateral-FG ECs (orange color ECs in Figure 3) showed reliable positive betas for the regression of modulatory changes on Perceived-Stress scores (respectively: beta=0.0116, PP=0.9751; beta=0.0235, PP=1; and beta=0.0265, PP=1). The VMPFC to R-hypothalamus EC, the L-VLPFC to bilateral-FG ECs, and the R-hypothalamus to bilateral-FG ECs (green color ECs in Figure 3) all showed reliable negative betas for the regression of modulatory changes on Perceived-Stress scores (respectively, betas=−0.0168, −0.0472, −0.0268, −0.0184, and −0.0175; each PP=1). The modulatory change in the R-amygdala to hypothalamus EC did not show a reliable regression with the Perceived-Stress score (beta=0, PP=0).

The supplementary DCM PEB analysis tested the linear regression of the EC modulatory changes on the Anxiety-Problems scores for all the participants. For each EC, the regression beta and the corresponding PP are shown in Table 4. Among the nine ECs identified above, only the findings on the R-amygdala to bilateral-FG ECs are consistent with above linear regression analysis on the Perceived Stress: these two ECs both showed reliable positive betas for the regression of modulatory changes on Anxiety-Problems scores (respectively, betas=0.1595, and 0.398; each PP=1).

DISCUSSION

We investigated whether CUD individuals show altered ECs between CAN regions when exposed to fearful/angry facial expressions. During the task, the CUD group did not show different brain activation and behavioral performance compared to the CTL group. The main effects of stimulus type were statistically significant across both groups, with a greater percentage of correct responses during face trials than shape trials, possibly because of relatively more visual attention to facial expressions than shapes (59, 60).

Within the modeled neuronal network, the DCM analysis revealed reliable group differences for the modulatory change in the hypothesized L-amygdala to R-hypothalamus EC. Six ECs that terminated on bilateral FG showed the largest reliable group differences in the modulatory changes among the non-hypothesized ECs. The modulatory changes on these ECs were either positively or negatively associated with the Perceived Stress score.

ECs related to the Hypotheses

Amygdala to R-hypothalamus ECs:

Consistent with our hypothesis, in response to the fearful/angry facial expressions, the bilateral-amygdala to R-hypothalamus EC showed greater modulatory change in the CUD group compared to the CTL group. Across all participants, the modulatory change in the L-amygdala to R-hypothalamus EC was positively associated with the Perceived-Stress score, but the modulatory change in the R-amygdala to R-hypothalamus EC was not reliably associated with the Perceived-Stress score. These linear regression results are consistent with previous studies (61, 62) suggesting that L-amygdala and R-amygdala may have different functions in response to emotion stimuli.

Amygdala, hypothalamus, and periaqueductal gray are three brain regions believed to related to fear response, with a circuit from amygdala to hypothalamus to periaqueductal gray (63). During response to fear stimuli, this system could evoke autonomic response, such as increased heart rate (63). Thus the greater modulatory change in the L-amygdala to R-hypothalamus EC found in the CUD group in response to negative stimuli suggests that this EC may be part of a circuit that facilitates enhanced autonomic response in CUD individuals.

Because of incomplete or missing phyiological data released by HCP (50), physiological data was not analyzed in the current study. Thus, in future studies, it would be interesting to test the relationship between the physiological data such as heart and respiratory rates and the EC measures.

VMPFC to Hypothalamus EC:

Consistent with our hypothesis, the modulatory change in VMPFC to R-hypothalamus EC was negatively associated with the Perceived-Stress score across all participants. However, the group difference of the modulatory change in this EC was not reliable (PP=0.6525).

The six largest “to FG” ECs

R-amygdala to L-FG EC.

The modulatory change in R-amygdala to L-FG EC was positively associated with the Perceived-Stress score across all participants. In addition, the modulatory change in this EC was greater in the CUD group than the CTL group. The enhancement of the R-amygdala to L-FG EC in CUD is consistent with a previous DCM study (64) showing that face perception increased the ECs between amygdala and FG, and a recent meta-analysis (65) showing that the reaction to fearful faces increased functional connectivity between the amygdala and FG. Given the function of the amygdala in the neural circuitry involved in the reaction to unpleasant events (in particular fear) (66), and the role of the FG in facial perception/sensitivity (67, 68), especially in detecting fearful faces (69, 70), the greater modulatory change in R-amygdala to L-FG EC suggests that this enhanced EC may mediate the enhanced sensitivity to negative facial expression in the CUD group.

R-amygdala to R-FG EC.

The findings for this EC were similar to the R-amygdala to L-FG EC discussed above.

L-VLPFC to L-FG EC:

The modulatory change in L-VLPFC to L-FG EC was negatively associated with the Perceived-Stress score across all participants. In CUD compared to CTL, the L-VLPFC to L-FG EC showed greater modulatory change in response to fearful/angry faces. Previous studies (71-73) revealed that VLPFC function was related to voluntary emotion regulation. Thus the heightened L-VLPFC to L-FG EC found in the CUD group may suggest that this EC could be involved with reduction of sensitivity to negative facial stimuli in the CUD individuals, possibly through voluntary emotion regulation by the L-VLPFC.

Altered brain activation and/or altered brain connectivity found in CUD during cognitive tasks amid normal behavioral performance has been attributed to compensatory activity (41, 74), including the reaction to emotional stimuli (32). In the current study, the CUD group did not differ from the CTL group in behavioral performance. As discussed above, the enhanced R-amygdala to L-FG EC and the enhanced L-VLPFC to L-FG EC seemed to facilitate enhanced and reduced sensitivity to negative facial stimuli respectively. Thus the enhanced L-VLPFC to L-FG EC could reflect a compensatory mechanism to achieve normal performance. Data from the same participants also suggested a compensatory mechanism in the CUD individuals during a working memory task (41). This compensatory mechanism would be analogous in some respects to the greater dorsolateral prefrontal cortex activation in schizophrenic patients during working memory tasks, which is thought to enable normal performance under certain conditions (89).

L-VLPFC to R-FG EC.

The findings in this EC were similar to the L-VLPFC to L-FG EC.

Hypothalamus to L-FG EC.

The modulatory change in this EC was negatively associated with the Perceived-Stress score across all participants, and the modulatory change in this EC was smaller in the CUD group than the CTL group. These results suggest that the hypothalamus to L-FG EC could be related to the vulnerability to heightened stress observed in CUD individuals (20, 21). Rodent studies (75) suggest that the projections from the hypothalamus to cortical areas are involved in mediating specific behaviors, such as arousal responses. A human study (76) reported that patients with migraine showed lower functional connectivity between hypothalamus and FG.

Hypothalamus to R-FG EC.

The findings on this EC were similar to the hypothalamus to L-FG EC.

Clinical implications

CUD is associated with anxiety disorders (77, 78). Interestingly, our finding for the R-amygdala to bilateral-FG ECs is similar to a previous study (79) showing that in response to fearful faces, patients with social anxiety disorder showed greater functional connectivity between the R-amygdala and the bilateral-FG than the healthy controls. These similarities are supported by the supplementary DCM PEB analysis, which showed that across all participants the greater modulatory changes on the R-amygdala to bilateral-FG ECs were associated with greater anxiety score. These similarities suggest that the enhanced ECs identified in the current study may be neuronal circuits underlying CUD’s vulnerability to anxiety disorders or to stress cardiomyopathy, a kind of neurocardiological disorder associated with heightened anxiety (80, 81). As far as we are aware, there is no published study investigating the amygdala and hypothalamus brain connectivity and the VMPFC and hypothalamus brain connectivity in anxiety disorders or CUD yet, possibly because of the difficulty in localizing the hypothalamus using fMRI techniques (82), which was mitigated by the relatively higher resolution HCP fMRI scans. These three brain regions are components of a neural circuit believed to be related to anxiety (82), and a preclinical study (83) has shown that it is possible to reduce anxiety by targeting the hypothalamus.

Limitations

(I) We relied on a task that may not actually induce acute stress per se, such as by presentation of shock, negative social feedback, or a more difficult task. However other data suggest that facematching nevertheless induces an autonomic response (22-27). (II) As discussed by (82), it is relatively difficult to accurately localize hypothalamus using current fMRI techniques. However the HCP fMRI images are of high resolution; in addition, relatively small spatial smoothing (4-mm) was used; furthermore, atlases were used to guide the location of hypothalamus (described in the Supplementary Materials). All of these factors tend to mitigate the effects of this limitation. (III) The modeled neuronal network consisted of a relatively small number of nodes, which was necessitated by computational constraints in DCM that have been discussed in (84). (IV) Block-design was used, and the imaging analyses were based on the contrast between face-trial and the shape-trial blocks. Thus the imaging results may have been slightly confounded by sporadic incorrect responses (less than 4% for all trials). (V) The CUD participants included in this study all had a positive urine test on THC, which may reflect either recent use, or more distant use up to a few weeks if heavy chronic use. HCP did not release cannabis use information other than the four measures shown in Table 2. Thus we were unable to examine whether there were dose-response relationships with the EC findings. In addition, we were unable to distinguish the acute or long-term effects of cannabis use on the EC findings. For the same reason, the effects of cannabis withdrawal, which can occur in regular heavy users following one or two days of abstinence, cannot be excluded. However, the CUD participants all had a diagnosis of cannabis dependence and their urines were all positive for THC. Treatment of CUD was not an objective of the HCP project, and participants were not instructed to stop or reduce using cannabis. Therefore the CUD participants were unlikely to have had motivation for stopping use of cannabis for this study. In addition, the CUD participants included in this study did not show a significant difference from the controls in the five cannabis withdrawal symptoms that were measured in this study (Table 2), including three psychiatric symptoms (anxiety problems, depressive problems, and somatic complaints). These considerations collectively suggest that cannabis withdrawal was unlikely in the CUD participants included in this study, and that psychiatric symptoms were unlikely to contribute to the group differences in EC. (VI) Other substance use disorders were not used as exclusion criteria because of unavailability of these diagnoses. Thus previous use of other drugs might explain differences in EC. However, participants with positive urine screens on other drugs were excluded. In addition, the majority (40 of 46) of the included participants did not use or only slightly used other drugs (<6 times). Furthermore, the portion of number of participants used other drugs more than 6 times was similar between groups. Thus, although we cannot completely exclude the effects of other drugs, we believe that previous use of other drugs unlikely affected the DCM findings significantly. (VII) Physiological measures of anxiety, which would be important in interpreting the DCM findings, were not available. Although HCP recorded cardiac and respiratory signals during MRI scans, these data were not usable as mentioned above. Future studies inspired by these findings can include electrodermal response, pupillary measures, or cardiac and respiratory data.

Conclusion

The left amygdala to hypothalamus EC and the right amygdala to bilateral fusiform gyri ECs are possibly part of circuits underlying the risk of CUD individuals to stress-related disorders. Correspondingly, the left ventrolateral prefrontal cortex to bilateral fusiform gyri ECs are possibly part of circuits reflecting a protective mechanism.