Network Analysis of Behavioral Approach/Inhibition Systems and Brain Volume in Individuals with and without Major Depressive Disorder or Social Anxiety Disorder

Abstract

Background:

Social anxiety disorder (SAD) and major depressive disorder (MDD) are characterized by behavioral abnormalities in motivational systems, namely, the behavioral inhibition system (BIS) and behavioral activation system (BAS). Limited studies indicate brain volume in regions that support emotion, learning/memory, reward, and cognitive functions relate to BIS/BAS. To increase understanding of BIS/BAS, the current study used a network approach.

Methods:

Patients with SAD (n=59), MDD (n=64) and healthy controls (n=36) completed a BIS/BAS questionnaire and structural MRI scans; volumetric regions of interest comprised cortical and limbic structures based on previous BIS/BAS studies. Bayesian Gaussian graphical model was used for each diagnostic group and groups were compared. Among network metrics, bridge centrality was of primary interest. Analysis of variance (ANOVA) evaluated BIS/BAS behaviors between groups.

Results:

Bridge centrality showed hippocampus positively related to BAS, but not BIS, in MDD; no findings were observed in SAD or control groups. Yet, network density (i.e., overall strength of relationships between variables) and degree centrality (i.e., overall relationship between one variable to all other variables) showed cortical (e.g., precuneus, medial orbitofrontal) and subcortical regions (e.g., amygdala, hippocampus) differed between diagnostic groups. ANOVA results showed BAS was lower in the MDD/SAD groups compared to controls, while BIS was higher in the SAD relative to MDD group, which in turn was higher than controls.

Conclusions:

Preliminary findings indicate network-level aberrations may underlie motivational abnormalities in MDD and SAD. Evidence of BIS/BAS differences builds on previous work that points to shared and distinct motivational differences in internalizing psychopathologies.

Introduction

Major depressive disorder (MDD) and social anxiety disorder (SAD) are pervasive (1), debilitating disorders (2–4), yet, our understanding of their neurobiological underpinnings remains to be established, particularly as they relate to motivational systems. Accruing data from neuroimaging studies point to links between regions implicated in motivational systems and behavior, specifically, the behavioral inhibition system (BIS) and behavioral activation system (BAS) in the context of reinforcement sensitivity theory (RST) (5). RST originally posited that the BIS and BAS modulate aversive and appetitive motivation, respectively (5). Subsequently, considerable research, particularly in animals, led to the inclusion of defensive behaviors, namely, flight-fight-freeze systems (FFFS) (6). These defensive systems rest on the functional distinction between behaviors, which includes defense approach and avoidance behavior and resolving conflict between the BAS and FFFS (e.g., approach-avoidance conflict) to determine course of action (6,7).

As for BIS/BAS behavior, BIS involves sensitivity to anxiety provoking stimuli and BAS involves reward behaviors, such as pursuit of goals (drive), sensitivity to reward (reward responsiveness; ‘reward’), and approach-related impulsivity (fun seeking) (8). These behaviors correspond with brain volume. For example, positive associations have been observed between nucleus accumbens (NAcc) and BAS (reward), medial orbitofrontal cortex (mOFC) and BAS total score (9) and between hippocampus and BAS (reward) (10) in healthy and community samples, respectively. Consistent with RST (6,7), it has been proposed that the NAcc and mOFC may contribute to BAS, or aspects of BAS, via their role in detecting reward-related stimuli and regulating behavior (9).

With regard to BIS and BIS-related behavior (e.g., sensitivity to punishment), studies involving healthy or community samples show higher BIS corresponds with more volume in limbic regions--amygdala, parahippocampus, and hippocampus (10,11). Findings provide support for RST as these regions are consistent with aspects of the BIS and FFFS system (7). Also, evidence of positive associations between hippocampus and BIS and BAS is consistent with its function as a comparator assessing conflict between simultaneous goals (10). More BIS has also been shown to relate to less precuneus and mOFC volume and the negative relationship is proposed to relate to anxiety traits (e.g., neuroticism) (12).

Altogether, despite mixed findings that may be due in part to methodological differences between studies, individual differences in BAS and BIS relate to variance in key structures involved in reward (striatum), emotion (amygdala), learning and memory (parahippocampus, hippocampus), self-related processes (precuneus), and top-down control (mOFC) (13–16) in addition to the role regions play in the context of BIS/BAS (e.g., hippocampus as a comparator assessing conflict between goals) (6,7).

We are not aware of a study that examined brain volume and BIS/BAS in MDD or SAD. However, meta-analytic studies point to volumetric abnormalities in these disorders in regions implicated in BIS/BAS. For example, compared to healthy controls (HC), individuals with MDD have less brain volume in regions involved in reward processes (e.g., dorsal striatum), learning and memory (e.g., hippocampus), and top-down control (e.g., OFC) (17). Less brain volume in MDD is proposed to relate to increased sensitivity in the stress system in depression (18) and pathways include decreased dendritic branching (19), neurogenesis (19,20), loss of neurons, and decreased expression of brain-derived neurotrophic factor (19,20).

Regarding SAD, findings generally point to larger brain volume. For example, compared to HC, a meta-analysis showed individuals with SAD had more brain volume in regions that underlie self-related processes (e.g., precuneus) and top-down control (e.g., dorsolateral prefrontal cortex) (21). In contrast, the putamen, which is involved in reward functions (15,22), was smaller in SAD compared to controls. Since SAD is frequently comorbid with depressive disorders (23,24), comparison was also made between controls and SAD without concurrent depressive disorders, and meta-analytic results showed that only the larger precuneus result continued to differ from controls (21). Altogether, SAD is generally characterized by larger brain volume, which may reflect lack of synaptic pruning (21).

In summary, studies involving healthy individuals and convenience samples suggest volumetric variance in regions involved in emotion and reward processes, learning and memory, self-related processes, and top-down control correspond with BIS/BAS behavior. Regarding MDD and SAD, meta-analytic findings indicate MDD is generally characterized by less volume in frontal and limbic regions whereas in SAD, brain volume tends to be in the direction of greater volume and aberrance is relatively limited. Though direct comparisons cannot be made between studies due to methodological differences, findings suggest MDD, SAD, and HC should exhibit differential relationships between brain volume and BIS/BAS behavior.

At the behavioral level, higher BIS and lower BAS are often observed in MDD (25–27). For example, BAS (reward) was found to negatively correlate with depressive symptoms in undergraduate students (28). In individuals with MDD, BAS negatively related to depression severity cross-sectionally and longitudinally (29). In SAD, more BIS and less BAS (fun seeking) was found compared to HC (30), while lower BAS and higher BIS positively correlated with cognitive biases related to social anxiety (31–34). Grossly, limited data suggest MDD and SAD may be characterized by more BIS and less BAS relative to controls, though this has yet to be tested directly.

Collectively, accumulating data highlight the relevance of BIS/BAS in advancing our understanding of volumetric brain and BIS/BAS relationships in MDD and SAD, which are likely complex as brain regions relate to each other to varying degrees (35). Therefore, the current study used a network approach. Specifically, a Bayesian Gaussian graphical model was used, which is appropriate for imaging studies with limited sample size (36). This approach captures complex relationships between variables that infer an underlying structure (i.e., partial correlational ‘network’) (37,38) and uses the partial correlation coefficient to remove confounding effects. It also accounts for uncertainty in the model and permits evaluation of relationships between variables that may reflect causal relationships (39). The network consists of ‘nodes’ (i.e., variables of interest) and ‘edges’ (i.e., relationship between nodes adjusting for the influence of all other variables) (40,41). Among standard metrics (42), we were most interested in bridge centrality, which represents variables that connect two or more networks (43,44). Specifically, we were most interested in identifying ‘bridges’ between brain regions and BIS/BAS behavior in MDD, SAD, and HC.

Extending previous work that suggests brain volume differs between MDD, SAD, and HC (17,21), we took a disorder-specific approach. Therefore, network analysis was performed for each diagnostic group and groups were compared to each other. ‘Nodes’ comprised brain regions of interest based on previous BIS/BAS volumetric studies (9–12) and BIS/BAS behavior (8).

Beyond network analysis, we expected both MDD and SAD groups would exhibit less BAS and more BIS than the HC group with no BIS/BAS difference between the MDD and SAD groups based on previous studies (25–27,30). Given the dearth of studies, we had no hypotheses regarding specific BAS behaviors (i.e., drive, reward, fun seeking) (8).

Methods and Materials

Participants

This is a secondary analysis of a study investigating neural predictors and mechanisms of psychotherapy outcome (ClinicalTrials.gov: NCT03175068). MRI data was collected from 2017 to 2021 excepting COVID-related shutdowns. The current study is limited to pre-treatment data. All participants were required to meet diagnostic criteria for SAD (n=59) or MDD (n=64), except for healthy controls (HC) (n=36), who could not have current or history of an Axis-I disorder. Also, the SAD group could not have comorbid MDD and vice versa though other comorbidities were permitted (see Table 1). Participants were also required to be 18 to 65 years old and average age was 28.5 (SD=9.7) years. Other exclusion criteria were self-reported history of psychosis (e.g., bipolar disorder), major medical or neurological problems, active suicidal ideation or self-injurious behavior in the past 6 months, psychotropic medications 6 weeks before or during the study, cognitive dysfunction (e.g., traumatic brain injury), developmental disorders (e.g., learning disability), moderate-severe substance abuse or dependence in the last 6 months, and contraindications for MRI. All participants tested negative for pregnancy and substance use before the MRI scan.

Table 1.

Demographic and clinical descriptives and analysis of variance (ANOVA); all values are means unless otherwise indicated and standard deviations are in parentheses.

	Total N=159		SAD n=59		MDD n=64		HC n=36	F	dfb,dfw	p
Age	28.53 (9.70)		24.85 (6.51)		30.12 (10.03)		31.75 (11.64)	7.69	2,156	.001
Liebowitz Social Anxiety Score	47.90 (33.06)		82.51 (18.27)		37.06 (17.98)		10.44 (11.02)	228.19a,b,c	2,156	<.001
Hamilton Depression Rating Score	8.98 (6.47)		8.75 (4.28)		13.89 (4.78)		0.64 (1.31)	123.25 a,b,c	2,156	<.001
BAS Drive	10.25 (2.45)		9.68 (2.38)		9.95 (2.37)		11.77 (2.13)	9.82a,b	2,155	<.001
BAS Fun seeking	11.46 (2.19)		11.12 (2.21)		11.42 (2.35)		12.22 (1.73)	2.32	2,155	.101
BAS Reward	16.78 (2.28)		16.27 (2.28)		16.45 (2.48)		17.89 (1.43)	5.63a,b	2,155	.004
BIS	22.39 (4.02)		24.73 (2.48)		22.88 (3.35)		17.54 (3.07)	64.99a,b,c	2,155	<.001
	n	%	n	%	n	%	n	%	χ2	p
Gender									1.16	.55
Female	105	66%	37	62.71%	46	71.88%	22	61.11%
Male	53	33.33%	21	35.59%	18	28.13%	14	38.89%
Not reported	1	0.63%	1	1.69%	0	0%	0	0%
Race									8.61	.19
American Indian or Alaskan Native	1	0.63%	1	1.69%	0	0%	0	0%
Asian	37	23.27%	15	25.42%	9	14.06%	13	36.11%
Black	18	11.32%	10	16.95%	5	7.81%	3	8.33%
More than one race	8	5.03%	2	3.39%	5	7.81%	1	2.78%
Native Hawaiian or other Pacific Islander	1	0.63%	0	0%	0	0%	1	2.78%
Not reported	3	1.89%	2	3.39%	1	1.56%	0	0%
Other	17	10.69%	4	6.78%	12	18.75%	1	2.78%
White	74	46.54%	25	42.37%	32	50.00%	17	47.22%
Ethnicity									0.07	.78
Hispanic or Latino	42	26.42%	15	25.42%	20	31.25%	7	19.44%
Not Hispanic or Latino	117	73.58%	44	74.58%	44	68.75%	29	80.56%
Not Reported	0	0%	0	0%	0	0%	0	0%

Abbreviations: BAS = behavioral activation system, BIS = behavioral inhibition system, MDD = major depressive disorder, SAD = social anxiety disorder, HC = healthy control

Notes: Post-hoc comparisons:

^aSAD group differed from HC group;

^bMDD group differed from HC group;

^cMDD and SAD groups differed

All participants provided written informed consent prior to study participation and were monetarily compensated. Study procedures including consent were conducted at the University of Illinois at Chicago, approved by the university’s Institutional Review Board, and complied with the Helsinki Declaration.

Clinical Measures

After obtaining consent, participants completed the Structured Clinical Interview for DSM-5 (SCID-5)(45) and interviewer-based Liebowitz Social Anxiety Scale (46) and Hamilton Depression Rating Scale (47) to assess symptom severity in SAD and MDD groups, respectively. All interviewer-based measures were conducted by a trained staff member; details are in Supplemental Materials.

Behaviors were evaluated with the Behavioral Inhibition/Behavioral Approach Scale (BIS/BAS Scale), a 24-item self-report which assesses how individuals react to situations and is shown to have good validity and test-retest reliability (8). The BIS/BAS Scale comprises one BIS score and three BAS scores (drive, reward, fun seeking). Higher scores reflect more BIS/BAS.

Structural MRI acquisition

The structural MRI scans were obtained on a 3T GE Discovery System (General Electric Healthcare, Waukesha, WI) with an 8-channel head coil. Additional acquisition information is provided in the Supplemental Materials.

Structural MRI preprocessing

FreeSurfer Image analysis suite version 7.1.1(http://freesurfer.net/fswiki/FreeSurferWiki) was used to obtain brain measurements. A priori bilateral volumetric regions of interest were as follows: amygdala, striatum (i.e., caudate, putamen, NAcc), hippocampus, precuneus, mOFC, and parahippocampal area for a total of 16 ROIs. See Supplemental Materials for MRI preprocessing details.

Participant Characteristics

One-way analysis of variance (ANOVA) and chi-square analysis were used to test BIS/BAS hypotheses and evaluate potential clinical and demographic differences between groups. Bonferroni correction was used to adjust for multiple comparisons.

Analytic Strategy

All analysis was two-tailed with familywise alpha level set to .05 and performed in R (4.1.2). See details in the Supplemental Materials.

Network construction

Connectivity matrices were constructed with the following variables as nodes: all a priori brain regions and all BIS/BAS scores (8). The ‘bggm’ package in R was used to estimate the network (42). Network nodes represent variables whereas edges represent relationships between two variables adjusting for all other variables in the network (i.e., pairwise partial correlations). Bayesian approach, deriving posterior samples from prior parametric assumptions and observed data via Monte Carlo simulation, was used for inference and estimation. Sparsity of the graphs, namely, fewer relationships than the maximum possible number of relationships in the network, was determined based on significance of the edges, after adjusting for multiple comparisons (36).

Network density

Density metrics reflect the overall strength of relationships between variables (i.e., summed absolute pairwise partial correlations) in each network. We additionally computed density metrics for two ‘subnetworks’, one for brain regions and the other for BIS/BAS behaviors, which represent sum of edges (i.e., sum of pairwise absolute partial correlations) within the respective subnetworks. We also computed cross-loading density, which refers to the overall relationship strength between the two subnetworks (i.e., brain and behavior [BIS/BAS]). Higher cross-loading density means more and/or stronger links in overall brain-behavior relationships.

Network edge comparison

To compare MDD, SAD, and HC groups against each other, between-group differences in edges were compared. Bayesian credible intervals (i.e., an interval within which an unobserved parameter value falls with a particular probability) were used to determine the significance of these differences.

Degree centrality and comparison

Degree centrality represents sum of edge weights of one variable (i.e., node) to all other variables. Therefore, greater degree centrality (i.e., a more central node) represents greater strength of the overall relationship between one variable to all other variables in the network. We ranked the degree centrality qualitatively to identify the most important variables. Using Bayesian credible intervals, we compared between-group differences in centrality for each variable.

Network bridge centrality

Bridge centrality refers to the importance of variables (i.e., nodes) in connecting two or more networks, or put another way, how one variable in a given subnetwork (e.g., brain regions) connects to all variables to the other subnetwork (e.g., BIS/BAS behaviors). Bridge centrality is the sum of partial correlations for one variable in one subnetwork to all variables in the other subnetwork (i.e., cross-loadings). For example, a behavioral variable within the behavioral subnetwork that has high bridge centrality indicates it plays a greater role in connecting to brain variables than other behavioral variables. We computed bridge centrality between brain and behavioral (BIS/BAS) subnetworks.

Results

Clinical and demographic data

As expected, the SAD group was significantly more socially anxious than MDD and HC groups and the MDD group was significantly more depressed than SAD and HC groups. See Table 1 and Supplemental Materials for clinical and demographic details.

With regard to BIS/BAS, one HC participant did not complete the questionnaire (8) due to human error. For BIS, the one-way ANOVA was significant [F(2,155)=64.99, p<.001, η²_g=0.46] and post-hoc Bonferroni corrected results revealed the SAD group endorsed more BIS (M=24.73, SD=2.48) than MDD (M=22.88, SD=3.35, p=.002) and HC groups (M=17.54, SD=3.07, p<.001). Also, the MDD group endorsed more BIS than the HC group (p<.001). BAS drive also differed among groups [F(2,155)=9.82, p<.001, η²_g=0.11] and post-hoc Bonferroni corrected results showed the HC group had higher BAS drive (M=11.77, SD=2.13) than the SAD group (M=9.68, SD=2.38, p<.001) and MDD group (M=9.95, SD=2.37, p=.001); there was no difference between SAD and MDD groups (p=1.00). Findings were similar for BAS reward [F(2,155)=5.63, p=.004, η²_g=0.07] as the HC group had higher BAS reward (M=17.89, SD=1.43) than the SAD (M=16.27, SD=2.28, p=.010) and MDD groups (M=16.45, SD=2.48, p=.007) yet there was no difference between SAD and MDD groups (p=1.00). Conversely, no group effects were observed for BAS fun seeking [F(2,155)=2.32, p=.101, η²_g=0.03].

Bayesian Gaussian graphical models

Figure 1 depicts estimated Gaussian graphical model for MDD, SAD, and HC groups. Edges represents pairwise partial correlations. Nonsignificant edges are omitted from the figure.

Figure 1.

Network density graph including brain (regional volume) and behavioral (Behavioral Inhibition System [BIS] total score and Behavioral Activation System [BAS] drive, reward, fun seeking subscales) variables for HC, MDD, and SAD groups. Green nodes denote behavioral variables whereas orange nodes denote brain variables. Green edges denote positive associations whereas red edges denote negative associations. Top, middle, bottom, respectively, depicts network graph for HC, MDD, and SAD groups. Nonsignificant edges were omitted.

Grayscale figure note: For HC, the relation between Precuenus (R) and Parahippocampal (L) is negative; for SAD, the relation between Caudate (L) and Hippocampus (L), as well as the relation between Hippocampus (R) and Caudate (R), are negative. All other edges are positive associations.

HC = healthy control, MDD = major depressive disorder, SAD = social anxiety disorder.

Network density

To summarize findings for the brain subnetwork, both the HC and SAD groups showed higher density (i.e., more relationships between a priori brain regions) than the MDD group. However, the HC and SAD group did not differ on overall connectivity among brain regions. In contrast, when examining the behavioral (BIS/BAS) subnetwork, only the MDD group showed higher density. Specifically, a positive relationship between BAS drive and BAS reward. No other relationships were observed. In the SAD and HC groups, no BIS/BAS relationships or relationships between BAS subscales were detected.

For overall density of the brain-behavioral network, the HC group showed higher density (i.e., more relationships between brain regions and BIS/BAS behaviors; not shown) than the MDD group, with no difference observed in HC and SAD groups.

For cross-loading density, which assesses relationships between brain subnetwork (i.e., a priori regions) and behavioral subnetwork (i.e., BIS/BAS behaviors), only the MDD group showed higher cross-loading than either the HC or SAD groups. Specifically, in the MDD group, BAS drive was positively associated with right hippocampus. In contrast, no brain-behavior relationships emerged in HC and SAD groups.

Network edge comparison

Network edges based on brain regions showed differences between groups. For brain regions, we report comparisons between groups as follows:

Compared to the HC group, the MDD group had a weaker partial correlation (i.e., negative connection) between the left and right precuneus (r_MDD-control=−0.319[−0.667, −0.014]), while the SAD group had weaker partial correlations (i.e., negative connection) between the left caudate and left hippocampus (r_SAD-control=−0.680[−1.174 −0.010]) and between the right mOFC and right precuneus (r_SAD-control=−0.759[−1.231 −0.184]).

Compared to the MDD group, the SAD group had a stronger partial correlation (i.e., positive connection) between the left and right parahippocampus (r_SAD-MDD=0.536[0.078, 0.986]), the left and right precuneus (r_SAD-MDD=0.295[0.029, 0.628]), as well as a weaker partial correlation (i.e., negative connection) between the right amygdala and right mOFC (r_SAD-MDD=−0.575[−1.017, −0.056]).

Network degree centrality and comparison

Table 2 displays degree centrality for all groups, as well as comparisons between groups. Among findings were differences between MDD and HC groups for certain regions in left hemisphere (amygdala, NAcc, mOFC, parahippocampus, precuneus) and right hemisphere (putamen, amygdala). Differences were also found between SAD and HC for certain regions in left hemisphere (caudate, mOFC) and right hemisphere (hippocampus, mOFC, precuneus). Differences were also observed between SAD and MDD groups. For BIS/BAS behavior, BAS drive and reward differed between MDD and HC group and between MDD and SAD groups. No effects were found for BIS or BAS fun seeking. Figure 2 shows histogram of posterior centrality estimates for each group ranked from the most to least central. The top three central nodes differed for each group and details are reported in Supplemental Materials.

Table 2.

Psychiatric comorbidities

	Total N=123		SAD n=59		MDD n=64
	n	%	n	%	n	%
Psychiatric Comorbidity
Generalized Anxiety Disorder	51	32.70%	25	42.37%	26	40.63%
Persistent Depressive Disorder	24	15.09%	0	0%	24	37.50%
Panic Disorder	4	2.52%	4	6.78%	0	0%
Post-Traumatic Stress Disorder	6	3.78%	2	3.39%	4	6.25%
Phobia	10	6.29%	5	8.47%	5	7.81%
Attention-Deficit/Hyperactivity Disorder	4	2.52%	2	3.39%	2	3.13%
Insomnia	37	23.27%	12	20.34%	25	39.06%
Hypersomnolence	15	9.43%	7	1.19%	8	1.25%
Eating Disorder	2	1.26%	1	1.69%	1	1.56%
Adjustment Disorder	2	1.26%	2	3.39%	0	0%
Obsessive Compulsive Disorder	1	0.06%	0	0%	1	1.56%
Mild Alcohol Use Disorder	1	0.06%	0	0%	1	1.56%

Abbreviations: MDD = major depressive disorder, SAD = social anxiety disorder

Figure 2.

Degree centrality of brain (regional volume) and behavioral (Behavioral Inhibition System [BIS] total score and Behavioral Activation System [BAS] drive, reward, fun seeking subscales) variables in descending order for HC, MDD, and SAD groups. Degree centrality represents the overall connections of one variable (node) to all other variables in the network. X-axis denotes raw degree centrality. Y-axis were ranked ordered from most central to least central nodes for each group.

HC = healthy control, MDD = major depressive disorder, SAD = social anxiety disorder.

Network bridge centrality

Bridge centrality within each group showed no cross-loading between brain and behavioral (BIS/BAS) subnetworks within the HC or SAD group. Therefore, no ‘nodes’ connected these networks within the HC group or the SAD group. However, within the MDD group, there was cross-loading (i.e., positive relationship) between right hippocampus and BAS drive (r _{drive and hippocampus (R)}=0.340[0.012, 0.622]). No other cross-loading was detected within the MDD group.

Discussion

The objective of the current study was to use a Bayesian Gaussian graphical model to advance our understanding of complex relationships between brain volume and BIS/BAS behavior in individuals with and without MDD or SAD. While we had no specific network-related hypotheses, we expected network metrics would be sensitive to diagnostic status (i.e. MDD, SAD, HC) based on meta-analytic studies showing brain volume in MDD and SAD differs from HC in different directions (17,21). Results were partially supported as network edges (i.e., connections between network nodes) and degree centrality (i.e., overall connections of one variable to all other variables) differed between diagnostic groups. Yet only the MDD group showed brain-behavioral effects. Outside of network analysis, we hypothesized MDD and SAD groups would exhibit less BAS and more BIS than the HC group with no BIS/BAS difference between MDD and SAD groups based on previous studies (25–27,30). Hypotheses were partially supported.

Among network metrics, we were most interested in bridge centrality. Results showed a positive relationship between BAS drive, which reflects persistence in pursuing desired goals (8), and right hippocampus in the MDD group, but not SAD or HC groups. Network density cross-loading findings were similar suggesting the BAS drive-right hippocampus link is important in the larger network for MDD. Since Bayesian networks have the potential to identify causal relationships, it will be important for future studies to determine whether BAS drive contributes to variance in hippocampal volume or vice-versa. When considering BAS drive was lower in MDD than controls in the current study along with evidence MDD is characterized by less hippocampal volume relative to controls (17) possibly due to decreased dendritic branching (19), neurogenesis (19,20), loss of neurons, or decreased expression of brain-derived neurotrophic factor (19,20), it may be that hippocampal abnormalities contribute to less BAS drive. Alternatively, less BAS drive may contribute to hippocampal abnormalities. For example, evidence that experience/learning can lead to sustained changes in hippocampal volume (48) would suggest less drive for approach-related goals could lead to less opportunity for learning and memory formation thus impacting hippocampal volume. Also, while hippocampal volume has been associated with BIS (10,11), a study comprising a community sample identified a positive relationship between BAS drive and bilateral hippocampal volume (10). Findings are consistent with the proposal that among its functions, the hippocampus serves as a comparator--assessing conflict between different goal-directed behaviors to facilitate exploratory rather than defensive patterns of behavior (49), which has implications for BAS drive. We hesitate to interpret the laterality finding due to the exploratory nature of network analysis, however, it is possible that the BAS drive-right hippocampal finding reflects hippocampal asymmetry as greater volume in right relative to left hippocampus has been observed in MDD (50,51).

Network edge results showed differences in connections between different brain regions. Specifically, the MDD group exhibited weaker partial correlations for bilateral precuneus compared to controls. The precuneus relates to self-focus and it is a key region in the default mode network implicated in a wide array of self-processing operations (14). It is possible that the negative relationship between right and left precuneus reflects the lateralization of self-processing functions that may underlie increased self-processing (e.g., rumination) observed in MDD (52). Therefore, it will be important for future network studies to include self-processing measures.

Relative to controls, the SAD group had weaker (i.e., negative) relationships between the left caudate and left hippocampus, as well as the right mOFC and right precuneus. The role of the caudate has been well-documented in reward processing and is particularly important for planning goal-oriented actions (53), while the hippocampus is a limbic region involved in learning and retrieval of action-outcome relationships, among other functions (54,55). The negative relationship may reflect aberrant reward processing related to goal-directed behaviors in those with SAD. The mOFC is a regulatory region involved in reward value reinforcers (16) and precuneus is central to self-processing (52). Conceivably, less covariance between the mOFC and precuneus, two crucial regulatory regions, could reflect inefficiencies between regulating behaviors and self-processing. Overall, findings suggest there may be unique differences in structural variance between healthy individuals and those with anxiety and depression.

As for the degree centrality, various regions differed between groups. Namely, several left lateralized regions were less central in the MDD relative to HC group, including regions implicated in reward-related behaviors (i.e., NAcc, mOFC), self-processing (i.e., precuneus), and memory and emotion (i.e., parahippocampus, amygdala). Interestingly, the left amygdala was less central in the MDD group compared to controls, while the right amygdala was more central, in addition to the right putamen in the MDD group compared to controls. Regions less central in SAD compared to controls included those involved in regulation and self-related processes (bilateral mOFC, right precuneus) without the same laterality effect. The SAD group also demonstrated more centrality in the left caudate and right hippocampus, compared to controls. Notably, several regions also differed in terms of degree centrality between MDD and SAD groups. Altogether, findings suggest groups differed in terms of which cortical and limbic regions were central (i.e., important) in the network.

Regarding laterality, there was no discernable pattern across metrics and we hesitate to interpret this finding. It is conceivable that the lack of an asymmetrical effect is due to nuanced relationships between internalizing psychopathology and regions involved in various functions. Given the dearth of research related to the neurobiological underpinnings (i.e., structural variations) of BIS/BAS in clinical populations and the exploratory nature of network analysis, further study is needed.

Concerning degree centrality and behavior, BAS drive and reward were more central in the MDD group compared to SAD and HC groups. No effects were found for BIS or BAS fun seeking. Results suggest that in MDD, BAS drive (i.e., pursuit of goals) and sensitivity to reward have more connections than BIS and BAS fun seeking. Or put another way, BAS drive and sensitivity to reward are more ‘important’ in MDD than BIS and BAS fun seeking. Findings may relate to deficiencies in these behaviors though further study is needed to determine the clinical inference of results. In general, findings are consistent with previous reports that aberrance in BAS plays a role in depression and may serve as a unique vulnerability (27,29,56).

Behaviorally, we hypothesized that both clinical groups would exhibit less BAS and more BIS than the HC group with no BIS/BAS difference between the MDD and SAD groups. ANOVA results showed BAS components (drive, reward) were lower and BIS was higher in the clinical groups relative to the HC group with no BAS differences between MDD and SAD groups. Yet, BIS was also higher in the SAD relative to MDD group. These findings are consistent with a body of literature demonstrating lower BAS and higher BIS in those with SAD and MDD (25–27,30). Though we did not anticipate BIS to be higher in the SAD relative to MDD group, BIS has been proposed to relate to the experience of anxiety in response to anxiety-relevant cues (56) thus the finding is not entirely unexpected. As for BAS fun seeking, it is not clear why there was no group effect though when considering fun seeking reflects approach-related impulsivity, it is not wholly unexpected that impulsivity in MDD and SAD groups was in the normal range.

The present study is not without limitations. First and foremost, MDD and SAD are frequently comorbid (57,58) and by excluding such comorbidity, findings may not generalize to individuals with MDD and comorbid SAD or vice versa. Also, participants were unmedicated and mostly female; these and other clinical and demographic characteristics reduce generalizability. Findings are limited to a priori brain regions; therefore, we cannot rule out the potential that other regions may have yielded findings. While a ‘whole-brain’ approach would address this limitation, it would comprise 85 brain regions, which would considerably exceed the number of participants in each diagnostic group, leading to estimation and inference challenges. Relatedly, there is no consensus on sample size for network models and our sample size was modest; it will be important to replicate findings in a larger sample. Concerning age, the sample mostly consisted of young adults; even so, older adults were not excluded to be as inclusive as possible. Thus, variation in age may have impacted findings. Network brain volume findings may not generalize to functional networks (e.g., default mode network) (59–61). While the BIS/BAS scale (8) is widely used, it is based on the original RST and thus does not separate the BIS and flight-fight-freeze systems (6). An automated approach was used for brain volume, which was visually inspected for quality control. However, we cannot rule out the possibility that minor brain volume errors confounded results. The study was cross-sectional in nature, which precludes causal inferences regarding the observed associations. Lastly, BIS/BAS and exclusion criteria (e.g., cognitive dysfunction, substance abuse/dependence) were based on self-report.

Despite limitations, the current study is the first we are aware of to identify network-level differences that may have behavioral and neurobiological implications for the pathophysiology of MDD and SAD. Given standard treatments are only moderately effective for MDD, SAD, and other internalizing psychopathologies (62–64), increased understanding of motivational systems (i.e., BIS/BAS) in these disorders have the potential to identify crucial targets for treatment. Moreover, evidence of group differences in self-reported BIS and BAS fills an important gap in the literature.

Table 3.

Degree centrality and comparisons

	HC	MDD	SAD	MDD-HC	SAD-HC	SAD-MDD
Caudate (L)	0.92	0.88	1.63	−0.03 [−0.136,0.038]	0.72 [0.085,1.275]	0.76 [0.123,1.303]
Putamen (L)	0.76	0.87	0.90	0.11 [−0.039,0.336]	0.13 [−0.005,0.345]	0.02 [−0.035,0.094]
Hippocampus (L)	1.08	0.57	1.08	−0.52 [−0.943,0.096]	−0.01 [−0.67,0.784]	0.51 [−0.001,1.008]
Amygdala (L)	0.74	0.52	0.55	−0.22 [−0.485,−0.006]	−0.19 [−0.457,0.008]	0.03 [−0.104,0.159]
Accumbens (L)	0.76	0.49	0.66	−0.27 [−0.528,−0.097]	−0.1 [−0.318,0.126]	0.17 [−0.038,0.422]
Medialorbitofrontal (L)	0.95	0.00	0.00	−0.97 [−1.483,−0.222]	−0.97 [−1.483,−0.222]	0 [0,0]
Parahippocampal (L)	1.01	0.00	0.61	−1.02 [−1.492,−0.331]	−0.41 [−0.896,0.253]	0.62 [0.307,0.809]
Precuneus (L)	0.85	0.54	0.83	−0.31 [−0.607,−0.051]	−0.02 [−0.203,0.18]	0.29 [0.093,0.533]
Caudate (R)	0.92	0.88	1.24	−0.03 [−0.136,0.038]	0.32 [−0.086,0.626]	0.36 [−0.031,0.655]
Putamen (R)	0.76	1.26	0.90	0.5 [0.153,0.825]	0.13 [−0.005,0.345]	−0.37 [−0.654,0.011]
Hippocampus (R)	0.63	0.91	1.55	0.28 [−0.152,0.651]	0.93 [0.128,1.645]	0.65 [−0.172,1.461]
Amygdala (R)	0.74	1.32	0.55	0.58 [0.037,1.019]	−0.19 [−0.457,0.008]	−0.77 [−1.216,−0.222]
Accumbens (R)	0.76	0.49	0.66	−0.27 [−0.528,−0.097]	−0.1 [−0.318,0.126]	0.17 [−0.038,0.422]
Medialorbitofrontal (R)	1.08	0.41	0.00	−0.67 [−1.182,−0.088]	−1.08 [−1.487,−0.475]	−0.41 [−0.679,−0.047]
Parahippocampal (R)	0.53	0.00	0.61	−0.54 [−0.804,−0.141]	0.08 [−0.197,0.428]	0.62 [0.307,0.809]
Precuneus (R)	1.90	0.54	0.83	−1.37 [−1.934,−0.633]	−1.09 [−1.587,−0.389]	0.29 [0.093,0.533]
BAS Fun seeking	0.00	0.00	0.00	0 [0,0]	0 [0,0]	0 [0,0]
BAS Drive	0.00	0.75	0.00	0.75 [0.229,1.214]	0 [0,0]	−0.75 [−1.214,−0.229]
BAS Reward	0.00	0.41	0.00	0.42 [0.021,0.723]	0 [0,0]	−0.42 [−0.723,−0.021]
BIS	0.00	0.00	0.00	0 [0,0]	0 [0,0]	0 [0,0]

Abbreviation: (L) = left; (R) = right, HC = healthy control, MDD = major depressive disorder, SAD = social anxiety disorder, BAS = behavioral activation system, BIS = behavioral inhibition system

Note: Credible intervals here are 99.92%. Significant differences are bolded.