Attenuated facial movement in depressed women is associated with symptom severity, and nucleus accumbens functional connectivity

Highlights

• •The facial movements under emotional stimuli were recorded using webcam.
• •Women with MDD showed the attenuated mouth movements compared to HC.
• •In MDD, the mouth movement score was associated with depressive symptom.
• •Mouth movement score was also associated with fMRI connectivity between NAc and pIns.
• •It can be a promising marker for assessing depressive symptoms and brain circuitry.

Abstract

It is assumed that mood can be inferred from one’s facial expression. While this association may prove to be an objective marker for mood disorders, few studies have explicitly evaluated this linkage.

The facial movement responses of women with major depressive disorder (n = 66) and healthy controls (n = 46) under emotional stimuli were recorded using webcam. To boost facial movements, the naturalistic audio-visual stimuli were presented. To assess consistent global patterns across facial movements, scores for facial action units were extracted and projected onto principal component using principal component analysis. The associations of component for facial movements with functional brain circuitry was also investigated.

Clusters of mouth movements, such as lip press and stretch, identified by principal component analysis, were attenuated in depressive patients compared to those in healthy controls. This component of facial movements was associated with depressive symptoms, and the strengths of resting brain functional connectivity between nucleus accumbens and both posterior insular cortex and thalamus.

The evaluation of facial movements may prove to be a promising quantitative marker for assessing depressive symptoms and their underlying brain circuitry.

1. Introduction

The core features of major depressive disorder (MDD) are depressed mood and loss of interest or pleasure (American Psychiatric Association, 2015). Deficits in neural systems involving attention, affective, and motivational processes may be linked to abnormalities in emotional regulation in MDD (Park et al., 2019), which, in turn, can cause impaired interpersonal functioning and social anhedonia (Kupferberg et al., 2016, Surguladze et al., 2005).

Such alterations in behavior may be reflected in facial expressions or movements, which can confer affective state and motivation. In fact, expressions can be evoked unconsciously in response to emotion-eliciting stimuli, such as seeing another’s emotional face (Dimberg et al., 2000, Ellingsen et al., 2022, Ellingsen et al., 2020). In rodents, facial movements (i.e., orofacial reactions) are elicited in response to stimuli with hedonic or aversive valence (Berridge, 2000, Dolensek et al., 2020). Furthermore, animal studies have demonstrated that activity in the insula, thalamus, and basal ganglia, particularly nucleus accumbens (NAc), might mediate facial movements via modulation of monoaminergic neurotransmitters, including acetylcholine, dopamine, and serotonin (Carvalho et al., 2003, Castro et al., 2016, Dolensek et al., 2020, Grill and Norgren, 1978, Salamone et al., 1986).

Human studies have shown that dysfunctions in emotional regulation might affect facial movement. For instance, patients with bipolar disorder exhibit stronger facial responses to neutral stimuli and less facial responses to negative emotional valence (Broch-Due et al., 2018). Patients with unipolar depression showed reduced facial movement for a facial expression imitation task (Trémeau et al., 2005). On the other hand, irregular facial movements can be induced without explicit emotional experiences. In some neurological conditions, such as Tourette syndrome or tardive dyskinesia, involuntary facial movements occur, independent of emotional stimuli, and are associated with aberrant neurotransmission and disrupted brain network topology (Klawans and Rubovits, 1974, Shulman et al., 1995, Worbe et al., 2015). Taken together, these findings suggest that facial movement could be used as behavioral biomarkers reflecting alterations in the monoaminergic system and dysfunction of brain circuitry.

Although facial movement patterns can be an intuitive and efficient tool for assessing symptoms in MDD, its underlying neurophysiological mechanisms have rarely been studied. Here, we investigated the abnormalities of facial movement in MDD based on prominent facial movement features extracted by principal component analysis (PCA) from facial action units, which were identified from facial video recordings of patients with MDD and of healthy controls (HC). Furthermore, we collected fMRI data during rest to assess functional brain connectivity which may underlie any alterations in stimulus-evoked facial movements in MDD.

2. Materials and methods

2.1. Participants

At baseline, 81 women satisfying the DSM-IV criteria for unipolar MDD, and 52 healthy women were enrolled via local bulletin boards and online advertisements (See Table 1 for demographic and clinical characteristics of participants). All participants were between 19 and 65 years old and had not used any psychotropic medications or herbal preparations with psychoactive properties within 4 weeks. Inclusion criteria for the MDD group were a current diagnosis of MDD, a minimum total score of 14 on the 17-item Hamilton Depression Rating Scale (HDRS). In the HC group, inclusion required a maximum total score of 7 on the 17-item HDRS. Exclusion criteria were attempted suicide in the past year or current suicide risk; lifetime diagnoses of bipolar disorder, schizophrenia, schizoaffective or other psychotic disorder, history of alcohol or other substance abuse, pregnancy, breastfeeding, clinically significant liver disease, evidence of untreated or unstable thyroid disorder, brains with structural abnormalities, clinically significant abnormal findings on physical examinations, routine laboratory tests, and electrocardiography. All participants completed the 21-item Beck’s Depression Inventory (BDI).

Table 1.

Demographic and clinical characteristics of participants.

	Women with MDD (Mean ± SD)	Healthy women (Mean ± SD)	p-value
Age (years)	29.8 ± 10.9	24.1 ± 5.3	0.04*
Education (years)	13.4 ± 2.3	13.8 ± 2.1	0.30
Hamilton depression rating scale	18.5 ± 2.8	4.0 ± 1.6	<0.001***
Beck depression inventory score	28.7 ± 7.8	3.5 ± 3.6	<0.001***
Age at onset (years)	25.6 ± 12.0	–	–

* p < 0.05, *** p < 0.01 at Mann–Whitney U-tests.

All participants provided written informed consent in the study protocol. This study was approved by the Institutional Review Board of the Korea Institute of Oriental Medicine (I–1711/001–001–01), Daejeon Korean Medicine Hospital of Daejeon (DJDSKH–16–BM–17), and the Korea National Institute for Bioethics Policy (P01–201609–11–002), and conducted from October 2017 to May 2019. This study was registered in the Clinical Research Information Service, established by the Korea Centers for Disease Control and Prevention and embodied as a part of the Primary Registries in the World Health Organization International Clinical Trials Registry Platform (KCT0002532 and KCT0002747, https://cris.nih.go.kr).

2.2. Assessment of facial expression movements

For all participants, facial expression movements were recorded at 30 Hz using a webcam while watching the emotional stimuli. They were seated 50 cm in front of the laptop computer screen. The display size was 34.0 × 20.3 cm², representing visual angles of 37.6° horizontally and 22.6° vertically. To boost facial movement changes, the naturalistic audio-visual stimuli were presented via a laptop computer screen and headphone. Stimuli were presented in a block design, which consisted of 30-s blocks of emotional stimuli followed by a 10-s fixation cross. Each 30-s block contained 2–3 clips obtained from comedy shows, newscasts, and YouTube. Two blocks of stimuli were included for each category of emotional or hedonic valence (i.e., joyful, sad, appetite-inducing, anxiety-inducing, and neutral; see Supplementary Figure S1). The five emotional contexts (30-s stimuli) were repeated twice in pseudo-randomized sequences. To avoid the carry-over effect, the first joyful block was presented before the blocks with a negative valence. Participants were left alone in facial expression recording, and asked to feel emotions freely on watching the clips. After watching clips, participants rated how much they feel joyful, sad, disgusted, fearful/anxious on a ten-point scale from zero to nine. Participants reported their rating watching the screenshot of each clip using E-Prime 3.0 software (Psychology Software Tools, PA, USA).

The iMotions 7.0 software (iMotions Inc., MA, USA) automatically extracted facial features from facial videos recorded during the task above. This software estimated probabilistic scores of 17 distinct facial action units based on the Facial Action Coding System (EKMAN and ROSENBERG, 1997). Scores for these 17 facial action units (Supplementary Table S1) were responded to emotional stimuli at sparse time-points (Supplementary Table S5), which enhanced variations of the action units within participant. Averaged coefficient of variations (ACV) of 17 action units to the stimuli within participants was 5.34, which was much higher than ACV of mean 17 action units between participants (MDD: 1.67; HC: 0.522). Based on these stimulus-associated variations of the action units, global patterns of facial expression in response to emotional stimuli were assessed using PCA (Joliffe, 1986). Because latency and frequency of the facial expression to emotional stimuli were varied between participants, PCA was performed using scores of 17 action units for whole time-points of the stimulus paradigms without temporal information. Two-dimensional matrix (observations $\times$ scores of 17 action units) was decomposed by PCA using alternating least squares algorithm (MATLAB; The MathWorks Inc., Natick, USA). The scores for the 17 action units for every time-points were projected onto principal component (PC), which generated a time-series of PC scores from both MDD and HC participants. Among the 17 PCs, we used the first 3 PCs in further analysis, which explained 48.64% of all variability (PC1: 19.63%, PC2: 17.30%, PC3: 11.61%). The action unit for “inner brow raise” was not reliable and excluded from the PCA because of high baseline scores observed in 30.6% of the participants, without actual facial movements in the inner brow region, as determined from the facial recordings. Finally, a square root transform was performed on the averaged PC scores of facial movements, in order to reduce the skewness of the data distribution.

We carefully checked the facial movements and excluded 21 participants because of the low accuracy of detection (facial landmark detection error; covering participant’s face with hand; frequent bowing their head), or the low quality of data (being drowsy during the experiment). Thus, 66 patients with MDD and 46 HCs were included in the analysis (See Supplementary Figure S2 for participants flow diagram). As for the medication status, 65 patients out of 66 patients are medication-naïve. Only one patient had been treated with serotonin-norepinephrine reuptake inhibitor and benzodiazepine, 5 months before the facial movements recording.

2.3. Resting-state fMRI acquisition and preprocessing

Of the 112 women enrolled in this study, 80 participants (HC, n = 42; MDD, n = 38) agreed to participate in the MRI portion of this study, and were screened using MRI compatibility criteria. The resting state-fMRI acquisition was conducted on a 3.0-T Philips Achieva scanner (Philips Medical Systems, Best, The Netherlands) with a 32-channel head coil at the Korea Basic Science Institute. Whole brain blood oxygen level-dependent (BOLD) images were acquired with a T2*-weighted echo-planar sequence (TR/TE = 2000/35 $\mathit{ms}$, voxel size = 3 × 3 × 3.5 ${\mathit{mm}}^3$ with a 0.5-$\mathit{mm}$ slice gap, number of volumes = 300). Total scan time for fMRI was 10 min. T1-weighted anatomical images were acquired for spatial normalization of BOLD images using a three-dimensional TFE pulse sequence (TR/TE = 6.7/3.1 $\mathit{ms}$, flip angle = 9°, voxel size = 1 × 1 × 1 ${\mathit{mm}}^3$).

For fMRI data, head motion was reduced by linear realignment of images to the first volume (FSL-MCFLIRT). Next, skull stripping was performed (FSL-BET). For T1-weighted anatomical images, cortical surface reconstruction was completed for co-registration (recon-all, FreeSurfer). BOLD fMRI images in native space were registered to the Montreal Neurological Institute (MNI) space following co-registration between subjects’ anatomical and functional volumes (bbregister, FreeSurfer). Preprocessing also included regressing out the six translation/rotation motion parameters (from FSL-MCFLIRT), motion-outlier confound matrices (FSL_motion-outliers), and the aCompCor regressors, containing four principle components and average of white matter and CSF, from the CONN toolbox (Behzadi et al., 2007, Whitfield-Gabrieli and Nieto-Castanon, 2012) using a generalized linear model (GLM). Spatial smoothing was then performed using a 5-$\mathit{mm}$ full-width-at-half maximum Gaussian kernel, and temporal high-pass filtering was performed with a 0.008-Hz cut-off frequency.

2.4. Seed-based functional connectivity

For connectivity analyses, we used a NAc seed, which was defined by subdividing the NAc region of the Automated Anatomical Labelling atlas (Rolls et al., 2020) based on the parcellation of putative core and shell subdivisions in humans, using diffusion tractography (Baliki et al., 2013) (Fig. 3A). The time series of the NAc core was averaged across all voxels of this ROI, for each participant, which was then fed into the GLM analyses as a regressor of interest (FSL-FEAT). Estimated parameters and their variances from this single subject analysis were then used in a group-level GLM (FSL-FEAT, mixed-effect model).

Fig. 3.

Altered seed-based functional connectivity of the nucleus accumbens (NAc) core in major depressive disorder (MDD) patients as compared to healthy controls (HCs). Functional connectivity of the NAc core with the medical prefrontal cortex (mPFC) was decreased in MDD participants as compared to HCs, while the connectivity with the left posterior insula cortex (pIns) and dorsolateral prefrontal cortex were increased in MDD participants as compared to HCs (z > 2.58, FWE-corrected with p < 0.05). Estimates of the head motions were corrected in group-level analysis for functional connectivity.

A separate GLM was also performed to calculate the difference in NAc core functional connectivity between MDD and HC individuals. To investigate the association between resting NAc connectivity, facial movements, and depressive symptoms, a multivariable regression model was used. The root mean square of the head-motion parameters from FSL-MCFLIRT were calculated within the individuals and added to the group-level GLM as covariate to correct motion effect between subjects. The maps obtained from the group-level GLM analysis were thresholded at z > 2.58, and cluster corrected for multiple comparisons by converting sizes of clusters to p-values using gaussian random field (GRF) theory with significant level at p < 0.05 (thresholded at z > 2.58, family-wise error [FWE]-corrected with p < 0.05; FSL_Cluster). For subsequent analysis with functional connectivity metrics, z-statistics over a 4-mm diameter sphere, centered on the local maxima of the significant cluster, were used.

2.5. Statistical analysis

All tests were two-tailed except fMRI analyses (one-tailed). For parametric testing, all data were first tested for normal distribution, as assessed by the Kolmogorov–Smirnov test. Repeated-measures analyses of variance (ANOVAs) and post-hoc paired t-tests were used to investigate the differences in facial component scores across the category of emotional stimuli. Group differences were evaluated using independent t-tests or Mann–Whitney U-tests. To investigate the associations between variables, Pearson’s or Spearman’s correlations were used. The significance level was set to α = 0.05. All statistical tests, except the fMRI data analyses, were performed using the Statistical Package for Social Sciences 17.0.

3. Results

3.1. Affective ratings of the emotional stimuli

As for affective rating, participants rated they felt joyful more than other feelings for joyful and appetite-inducing stimuli (paired t-tests, p < 0.001 after Bonferroni correction, Supplementary Figure S3). For sad stimuli, participants felt sad more than others, and for anxiety-inducing stimuli, they felt fearful/anxious more than others (paired t-tests, p < 0.001 after Bonferroni correction, Supplementary Figure S3). Each category of video clips induced the proper affect (or feeling) more than other affect. Subsequently, we compared the affective ratings between MDD patients and HC under each emotional stimulus, which were shown in Supplementary Table S2. For joyful and appetite-inducing stimuli, HC rated more joyful than MDD patients (all p < 0.01 after Bonferroni correction at Mann–Whitney U-tests). For joyful and neutral stimuli, MDD patients rated more sad, disgusted, fearful or anxious than HC (all p < 0.01 after Bonferroni correction at Mann–Whitney U-tests). These findings indicated that MDD patients had some biases in emotional experience under the stimuli with positive valence. For sad and anxiety-inducing stimuli, no significant difference was shown in positive (joyful) or negative affect ratings (sad, disgusted, fearful or anxious) between MDD patients and HC.

3.2. Facial movements induced by emotional stimuli

From facial video recordings of participants under naturalistic emotional stimuli, scores for 17 facial action units were extracted and projected onto PCs using PCA, which yielded three PCs, clusters of prominent facial movement features. The first PC encompassed the facial action units of “mouth open”, ”lid tighten“, ”jaw drop“, and ”brow furrow“, overall representing mouth-opening movements, termed the ”AH component“ (Fig. 1A–B). The second PC encompassed ”lip press“, ”dimpler“, ”lip suck“, and ”lip stretch“, representing lip-pursing/stretching movements, termed the ”HM component“ (Fig. 1A–B). The third component was predominantly driven by ”eye closure.“ To evaluate the stability of feature selection using PCA, we repeated PCA for the MDD and HC groups separately, and found results similar to those for the merged (MDD + HC) group (Supplementary Figure S4).

Fig. 1.

Consistent global patterns of facial expression in response to emotional stimuli. (A) Principal components (PCs) of the 17 facial action units represented common facial expressions. Facial action units of “mouth open”, “lid tighten”, “jaw drop”, and “brow furrow” contributed most strongly to PC1 (“AH component”). Facial action units of “lip press”, “dimpler”, “lip suck”, and “lip stretch” contributed most to PC2 (“HM component”). (B) Visualization of PCs using Openface toolkit (Baltrušaitis et al., 2016) and MakeHuman software (http://www.makehumancommunity.org/, MakeHuman Team). Virtual facial expressions for PCs was rendered based on 17 weights of facial action units from PC analysis. The AH component represented mostly mouth-opening movements, whereas the HM component represented mostly lip-pursing/stretching movements. Averaged scores of PCs and major facial action units are shown in terms of their response to the diverse set of emotional stimuli presented to study participants. The AH component was mostly elicited during presentation of joyful emotional stimuli. The HM component was also elicited during the presentation of joyful stimuli as well as during fixation cross presentation.

Both the “AH component” and “HM component” were associated with the joyful stimuli (Fig. 1C). The AH component score for joyful stimuli was higher than for other stimuli (repeated-measures analysis of variance [ANOVA], F_5,555 = 27.51, p < 0.0001, post-hoc paired t-tests, p < 0.0001 after Bonferroni correction, Supplementary Table S3), indicating the AH component was mostly elicited during joyful stimuli. The HM component was elicited during the presentation of fixation cross as well as the joyful stimuli. The HM component scores for responses to the fixation cross and joyful stimuli were higher than that for other stimuli (repeated-measures ANOVA, F_5,555 = 18.30, p < 0.0001, post-hoc paired t-tests, p < 0.05 for others; no significant score difference between fixation cross and joyful stimuli, p > 0.9 after Bonferroni correction, Supplementary Table S3). We presented the individual time courses of the AH and HM components for the representative participants in Supplementary Figure S5, which showed the AH and HM components occurred in turn during presentation of joyful stimuli. In summary, the joyful stimuli were most evocative of facial movements in our participants.

3.3. The HM component is associated with depression

To assess possible depression-specific facial movements, we first compared the scores of AH and HM components between HC and MDD participants, and found a significant difference only for the HM component score (HM score [mean ± SD], 2.23 ± 1.20 in HC, 1.49 ± 1.52 in MDD, t = 2.75, p = 0.007; AH score, 2.91 ± 1.97 in HC, 2.55 ± 2.36 in MDD, t = 0.85, p = 0.40, respectively), indicating that participants with MDD elicited HM components less strongly than did HCs (Fig. 2A).

Fig. 2.

The stimulus-evoked HM component in MDD and HC participants, and its association with depressive symptoms. (A) The stimulus-evoked HM facial expression component score was significantly different between MDD and HC groups (p = 0.007, gray circle denotes mean; error bars, SD). (B) The HM component in MDD was inversely correlated with the Beck Depression Inventory (BDI) scores (r = −0.39, p = 0.001) at baseline. ** p < 0.01.

Moreover, the HM component in MDD participants was inversely correlated with depressive symptoms on the BDI (Fig. 2B, r = −0.39, p = 0.001), indicating that the HM component was less prominent for women with higher scores for depressive symptoms. On the other hand, the AH component did not show a significant correlation with baseline depressive symptoms score (r = −0.04, p = 0.75).

3.4. Alteration of nucleus accumbens functional connectivity in MDD patients

To elucidate the neural correlates associated with the decreased HM component in MDD, seed-based functional connectivity analysis was conducted using resting-state fMRI data of 38 MDD patients and 42 HCs. The HM component was composed of the mouth movements. Previous animal studies have shown that mouth movements, such as the hedonic orofacial reaction, could be modulated by neurotransmitter microinjection into the NAc (Castro et al., 2015, Castro et al., 2016). The NAc is a key region for the motivational process and reward circuitry. The NAc is implicated in the core symptoms of MDD, including anhedonia or loss of interest (Felger et al., 2016, Francis and Lobo, 2017, Russo and Nestler, 2013). Thus, we focused on the NAc circuitry to investigate the neurological basis related to the deficit in the HM component in depressive women.

Difference matrices of whole brain functional connectivity between MDD and HC individuals demonstrated that NAc core connectivity with the medial prefrontal cortex (mPFC) was significantly decreased in MDD patients as compared to HCs (thresholded at z > 2.58, FWE-corrected with p < 0.05, Fig. 3 and Supplementary Table S4). Moreover, we found a significant increase in NAc core connectivity with the left posterior insula cortex (pIns), left dorsolateral prefrontal cortex (dlPFC), left inferior prefrontal gyrus (IFG), and left middle insula cortex (mIns) in MDD patients as compared to HCs (Fig. 3 and Supplementary Table S4). Interestingly, this NAc–pIns connectivity showed a negative correlation with the HM component score (Fig. 4), indicating that increased connectivity between the NAc and the pIns was associated with a decrease in HM components in MDD. However, the NAc core–mPFC, NAc core–dlPFC, NAc core–IFG, or NAc core–mIns connectivity did not show any significant correlation with the HM component score (all p > 0.1).

Fig. 4.

The association between the HM component and nucleus accumbens (NAc) functional connectivity in major depressive disorder (MDD) patients (z > 2.58, FWE-corrected with p < 0.05). HM component score was inversely correlated with the connectivity of the NAc core with the left pIns. Moreover, connectivity of the NAc core with the medial thalamus was inversely correlated with the HM component in MDD participants. Estimates of the head motions were corrected in group-level analysis for functional connectivity.

3.5. NAc functional connectivity associated with the HM component

We then investigated the association between the whole-brain NAc core connectivity matrices and the HM component in MDD participants. The whole-brain correlation matrices showed that NAc core connectivity with the pIns and medial thalamus was significantly inversely correlated with the HM component in MDD patients (thresholded at z > 2.58, FWE-corrected with p < 0.05; Fig. 4 and Supplementary Table S5).

As there was a significant difference in age between the MDD and HC groups (MDD, n = 66; 29.8 ± 10.9 years [mean ± SD]; HC n = 46, 24.1 ± 5.3 years, Mann–Whitney U test, p = 0.04), we re-compared the PC scores and brain connectivity strengths using an age-matched group (MDD, n = 53, 25.8 ± 7.7 years [mean ± SD]; HC, n = 36, 25.2 ± 5.6; Mann–Whitney U test, p = 0.41) for confirmation purposes, which showed similar results (Supplementary Figures S6 and S7). Briefly, the HM component score was significantly different between MDD and HC (independent t-test, p = 0.004). The HM component score was significantly correlated with the depressive symptoms as assessed by BDI (r = −0.37, p = 0.006). The connectivity strengths of NAc with pIns and dlPFC, and with mPFC showed significant difference between the MDD and HC groups (thresholded at z > 2.58, FWE-corrected p < 0.05).

4. Discussion

We identified prominent clusters of facial movements in female participants using PCA: the “AH component (PC1)” and ”HM component (PC2)” (Fig. 1). The AH component was related to mouth-open, jaw-drop, and lid-tighten movements, which are action units known to occur in positive (joyful) or negative (sad, anger, fear) expressions, as well as in surprise (Kohler et al., 2004). In the current study, joyful stimuli elicited more of the AH component in participants than did other stimuli, although the anxiety-inducing and sad stimuli also evoked the AH component to a moderate extent (Fig. 1C). These results indicate that the AH component is primarily induced by joyful stimuli.

On the other hand, the HM component mainly comprised lip suck, lip press, lip stretch, and dimpler. These movements were evoked under joyful stimuli or during presentation of the fixation cross. Thus, joyful stimuli evoked both the AH and HM components. The facial movements of the AH and HM components sometimes occurred in turn during presentation of joyful stimuli (Supplementary Figure S5). Thus, the HM components may be related to a smile without mouth opening, or they may simply be a cluster of mouth movements preceding or followed by a smile (Messinger et al., 1999). In addition, the HM components were frequently observed during presentation of the fixation cross. It is unclear what aspect of the fixation cross elicited the HM components. Mind-wandering, irrespective of the stimuli, rumination on previous stimuli, or waiting for upcoming stimuli may have elicited the HM components and associated activities in brain regions at rest (Carlson et al., 2020, Gilbert et al., 2007, Mason et al., 2007).

We found a significant difference in the HM component score between HC and MDD participants, with mouth movements related to HM components markedly attenuated in patients with MDD. In addition, the higher scores of HM components were associated with a lower score of depressive symptoms. These findings verified the relationship between the HM component and depression, suggesting that the HM component score can be used as a feasible nonverbal tool for assessing depressive symptoms and detecting changes after treatment.

Whole-brain functional connectivity matrices for resting states shed light on the neural mechanism of attenuated HM component in depressed women. We found a significant difference in NAc connectivity between patients with MDD and HCs. The NAc core showed increased connectivity with the left pIns in MDD patients, while mPFC–NAc connectivity was decreased in patients with MDD. We also found that the strength of the connectivity between the NAc and the pIns was associated with the HM component score in MDD patients.

A previous meta-analysis of human fMRI connectivity studies demonstrated that the NAc, insula, and thalamus were consistently co-activated and had rich reciprocal connections (Cauda et al., 2011, Cho et al., 2013). These three brain regions are key components in the reward system (Cauda et al., 2011, Cho et al., 2013). NAc cells respond to appetitive or aversive stimuli (Roitman et al., 2005), and the NAc integrates the information of emotional valence and sends signals to the motor system (Mogenson et al., 1980). The NAc is implicated in orofacial movements, which are accompanied by feeding, drinking, and vocalization (Mogenson et al., 1980). Thus, the NAc and its connectivity with the insula and thalamus are important in elucidating the neural basis of human mouth movements.

Previous rodent studies have shown that rodent mouth movements can be modulated by pharmacological manipulation within the NAc (Castro et al., 2016, Richard and Berridge, 2011). Optogenetic or electrical stimulation of the pIns also contributes to inducing facial movement in mice (Dolensek et al., 2020)^, (Maeda et al., 2014, Tsutsumi et al., 2018, Zhang and Sasamoto, 1990). The insula sends afferents to the NAc, and the insula–NAc circuit is implicated in socioemotional behavior (Gehrlach et al., 2019, Rogers-Carter and Christianson, 2019). A recent study showed that the pIns–NAc core pathway transmitted information of the aversive state of malaise, and finally inhibited ongoing feeding behavior (Gehrlach et al., 2019). Feeding behavior includes mouth movements. These studies indicated that the pIns–NAc core pathway contributed to the inhibition of ongoing behavior, suggesting that its facilitation in MDD could reinforce behavioral suppression.

In addition, animal studies have shown that the paraventricular nuclei, midline nuclei located in the medial thalamus, to NAc projection mediated reward-seeking behavior (Cheng et al., 2018, Choi and McNally, 2017, Do-Monte et al., 2017, Otis et al., 2019). The inhibition of input from the paraventricular nuclei to the NAc increased unproductive reward-seeking (Lafferty et al., 2020), suggesting that this pathway suppresses compulsive behavior by regulating motivational inhibition. Interestingly, the pIns and medial thalamus to NAc connectivity showed an inverse correlation with the HM component in MDD participants. Thus, greater connectivity from the pIns and medial thalamus to the NAc were associated with HM component attenuation. These findings suggest that the suppressed HM component in MDD could be the result of behavioral inhibition modulated by the pathway from the pIns and medial thalamus to the NAc.

We also found that patients with MDD showed decreased functional connectivity between the NAc core and mPFC, indicating suppression of the mPFC–NAc pathway. Neuroimaging studies have shown that participants with depressive symptoms show blunted responses in the mPFC under both reward or aversive stimuli, and also demonstrate abnormalities in the frontal–striatal network (Rzepa et al., 2017, Zhang et al., 2016). These results were consistent with our study and supported the hypothesis that decreased mPFC–NAc connectivity might have some relevance to depressive symptoms, including anhedonia.

An important limitation of the study is the homogeneous sample. Participants were all Korean women, thereby limiting our ability to generalize results to depressed men.

In conclusion, we found attenuated facial movement (HM component) related to lip-stretching in MDD patients, which was associated with depressive symptoms, and with the strength of connectivity of the NAc core to the pIns and thalamus. Thus, we elucidated the mechanism underlying abnormal facial movement in patients with MDD. Evaluation of facial movement can be used as a quantitative marker for assessing depressive symptoms.

Author contributions.

SC, CC, VN, ICJ, and HK conceptualized the study design. SC, YKS, YC, SMC, YS, JoK, GC, ICJ, and HK collected the data. CJ, JiK, SC, and VN, HK analyzed and interpreted the data with contributions from YKS, KP, and OK; CJ and JiK wrote the first draft of the paper. All authors contributed to critically revising the manuscript and approved the final manuscript.

Funding

This work was supported by the Korea Institute of Oriental Medicine grant KSN2212010 (to HK) and C18210 (to HK and ICJ), and the National Research Foundation of Korea grant NRF-2017R1A2B4012546 (to CC).

Declaration of Competing Interests

The authors declare that they have no known competing financial interests or personal relationships that could have appeared to influence the work reported in this paper.

Appendix A. Supplementary data

The following are the Supplementary data to this article:

Data availability

The authors do not have permission to share data.