Association of neuroimaging measures with facial emotional processing in healthy adults: a task fMRI study

Abstract

Investigating the neural processing of emotion-related neural circuits underlying emotional facial processing may help in understanding mental disorders. We used two subscales of the Toronto Alexithymia Scale (TAS) to assess the emotional cognitive of 25 healthy participants. A higher score indicates greater difficulty in emotional perception. In addition, participants completed a n-back task during functional magnetic resonance imaging. Psychophysiological interaction analysis was used to explore the functional connectivity (FC) of neural circuits. Next, we used elastic-net regression analysis for feature selection and conducted correlation analysis between the neuroimaging measures and questionnaire scores. Following a 3-fold cross-validation, five neuroimaging measures emerged as significant features. Results of correlation analysis demonstrated that participants with higher TAS scores exhibited increased FC between the amygdala and occipital face area during facial stimulus processing, but decreased connectivity during emotional processing. These findings suggested that individuals with poor emotional recognition exhibited increased connectivity among face-related brain regions during facial processing. However, during emotional processing, decreasing neural synchronization among neural circuits involved in emotional processing affects facial expression processing. These findings suggest potential neural marker related to subjective emotional perception, which may contribute to the diagnosis and treatment of emotional dysregulation in individuals with psychiatric conditions.

Introduction

Emotion is a complex psychological phenomenon and process that reflects people’s attitudes toward various things. Moreover, it is closely intertwined with our experience of the world and serves as a driving force for individual behavior. With the rapid development of the economy and society, individuals need to quickly recognize emotionally arousing stimuli, such as rewards and threats, and make adaptive emotional and behavioral responses (such as fighting or fleeing), which are crucial for survival and development. Multiple studies have shown that emotions play a critical role in cognition, such as perceptual, attentional, memory, and decision-making processes, etc. (Aldinger et al. 2013, Packard et al. 2021, Goodhew and Edwards 2022). The accurate recognition and expression of emotions are vital for individual survival, environmental adaptation, and the establishment and maintenance of interpersonal relationships. Meanwhile, emotion can be harmful if the intensity, duration, frequency, or type of emotion is inappropriate and leads to improper deviations in cognition and behavior. In addition, when negative emotions persist for a long time, they may lead to the occurrence of mental disorders, such as anxiety, bipolar disorder, and depression (Kendler et al. 1999). Consequently, delving into the neural mechanisms underlying emotional cognition not only enhances our understanding of the biological bases of human emotional processing but may also provide insights for the diagnosis and treatment of emotional disorders, such as depression and anxiety disorders.

Faces are a common visual stimulus in daily life, containing rich emotional information. Facial expressions serve as an important nonverbal means of social communication. In 2000, Haxby et al. proposed a model of facial processing that consisted of face face-selective regions: the occipital face area (OFA), fusiform face area (FFA), and superior temporal sulcus face area (STS-FA) (Haxby et al. 2000). This model has illuminated how the brain processes and interprets the face at different levels. With the increasing application of advanced neuroimaging technologies such as functional magnetic resonance imaging (fMRI), and functional near-infrared spectroscopy (fNIRS), researchers have discovered higher-level facial perception abilities within face-selective regions (Zhang and Kay 2020, Chai et al. 2024). Notably, the OFA has demonstrated a heightened ability to discern different emotional expressions in faces (Rotshtein et al. 2005, Pitcher et al. 2008, Atkinson and Adolphs 2011). In addition to cortical brain regions involved in facial emotional processing, there is also a subcortical face-detection pathway that mainly processes low spatial frequency facial information and participates in emotional information processing (Johnson 2005). This subcortical pathway refers to the transmission of visual information through the superior colliculus, hypothalamus, and amygdala (AMY). This system enables rapid processing of emotional information, especially quick responses to threatening stimuli, bypassing cortical processing in regions such as the primary visual cortex (Tamietto and de Gelder 2010). Many studies have provided evidence that emotional information may be processed rapidly through this subcortical pathway (Morris et al. 1999, Streit et al. 2003). Beyond the involvement of the AMY and visual cortex in processing emotional faces, the interactions between the AMY and prefrontal cortex are also crucial during affective processing. Interestingly, some research on patients with amygdala lesions challenge the essential role of amygdala for fear and threats processing during early stages, indicating that heightened cortical activity is not solely dependent on the amygdala (Tsuchiya et al. 2009, Pessoa and Adolphs 2010). For instance, one functional imaging study revealed that amygdala resection patients exhibit similar or enhanced visual cortex responses to emotionally salient stimuli compared with healthy controls, suggesting non-amygdalar processes contribute to emotional modulation of visual processing (Edmiston et al. 2013). The precise role of amygdala on emotional processing merits further studies. On the other hand, the prefrontal cortex is also involved in emotional information processing. Previous works found that unconscious fearful faces activate the ventral anterior cingulate cortex (ACC), while conscious fearful faces activate the dorsal ACC and medial prefrontal cortex (Phillips et al. 2008b, Ochsner et al. 2009). Willinger’s proposed prefrontal-AMY network for emotional processing encompasses the medial prefrontal cortex, lateral prefrontal cortex, AMY, and fusiform gyrus (Willinger et al. 2019), highlighting the complexity and interconnectedness of these brain regions in emotional processing.

Working memory (WM), which encompasses the temporary storage and processing of information during cognitive tasks, is instrumental in the intricate interplay between emotion and cognition (Baddeley 2003). Prior research underscores that emotional salience modifies cognitive control load, subsequently influencing attention and memory abilities (Brosch et al. 2013). Conversely, the constrained capacity of WM can pose challenges in integrating internal and external sensory information during emotional recognition. Notably, Phillips et al. have corroborated the hypothesis that WM load hinders the accurate recognition of emotional facial expressions (Phillips et al. 2008a). Their study design entailed participants identifying facial emotions concurrently with an n-back task, where varying task complexities were introduced. Both the increased dual-task demands and task complexity significantly hindered participants’ ability to accurately perceive others’ emotions. Therefore, it is clear that WM capacity is intricately linked with emotional cognition abilities. Further exploration of this connection through neuroimaging studies has revealed that WM load is associated with reduced activation in brain regions critical for emotional processing, such as the amygdala (Van Dillen et al. 2009, Kellermann et al. 2012). This reduction in activation may stem from the depletion of attentional resources under high WM load, potentially leading to the suppression of emotional processing and a subsequent decrease in emotional experience. Additionally, research focused on individual differences has indicated that individuals with higher WM capacity, as demonstrated by better performance on tasks like the Operation Span Task (OSPAN), are more capable of effectively suppressing both negative and positive emotional experiences, showcasing superior emotion regulation skills (Schmeichel and Demaree 2010; Schmeichel et al. 2008). In contrast, participants with lower WM capacity are more prone to the interference of negative thoughts. Nonetheless, the precise neural mechanisms of facial emotional processing under high load WM tasks remain unclear and need to be fully elucidated.

Based on the previous discussion, we hypothesize that as the load of WM tasks increases, the neural substrates for processing facial emotions may vary depending on the complexity of the WM challenges. Additionally, under conditions of high memory demand, alterations in FC between different brain regions are expected. These alterations may be evident through a reconfiguration of the FC networks. To verify this hypothesis, in this study, we have crafted an n-back task that incorporates varying degrees of mnemonic demand—specifically, low-memory demand and high-memory demand conditions—and integrated this with fMRI technology to investigate the neurobiological underpinnings of emotional cognition. Our study is poised to uncover the neural dynamics of facial emotion processing by examining the functional connectivity (FC) within the neural circuitry implicated in emotional processing among a cohort of healthy participants. Utilizing the Toronto Alexithymia Scale (TAS), we evaluated the participants’ capacity for emotional recognition and investigated how this capacity correlates with the FC of brain regions engaged in emotional face tasks. The outcomes of this research are anticipated to yield fresh perspectives on the neural processes that govern the interplay between emotion and cognition, thus may lay a theoretical groundwork for the comprehension of affective disorders.

Materials and methods

Participants

Thirty adults were enrolled through posters and advertisements at West China Hospital of Sichuan University. A total of 27 healthy participants completed the MRI examination, and underwent comprehensive psychological assessments, including the Chinese version of the TAS. One subject was removed because of pituitary adenoma, and one participant was excluded from further analysis due to a large amount of head motion (>0.5 mm of mean framewise displacement throughout the scan).

Psychological measurement

The TAS is a psychological tool used to assess an individual’s level of difficulty in expressing emotions. The TAS aimed to assess three factors by factor analysis: (i) difficulties in identifying feelings (DIF), (ii) difficulties in describing feelings (DDF), and (iii) externally oriented thinking (EOT). According to our study purpose, we used the sum of the DIF and DDF subscales to assess the subjects’ cognitive level with higher scores indicating more difficulty on emotional discrimination. The Ethics Committee of the West China Hospital of Sichuan University approved this study, and each subject signed an informed consent before participation.

Neuroimaging data acquisition

All scans were acquired by a 3.0 T system (Tim Trio; Siemens Healthineers, Erlangen, Germany) equipped with a 32-channel phased-array head coil. A three-dimensional T1-weighted image was acquired using a spoiled gradient-recalled echo sequence with the following parameters: repetition time (TR) = 2400 ms, echo time (TE) = 2.01 ms, flip angle = 8°, matrix size = 320 × 320, field of view = 256 × 256 mm², slice thickness = 0.8 mm, and voxel size = 0.8 × 0.8 × 0.8 mm³. A reverse interleaved gradient echo planar imaging (EPI) sequence was acquired covering the whole-brain images with the following parameters: TR = 1400 ms; TE = 30 ms; flip angle = 65°; field of view = 100 × 100 mm²; slice thickness = 2.0 mm; matrix size = 112 × 112; and voxel size = 2. 0 × 2.0 × 2.0 mm³. To reduce scanning noise and head motion, each participant was equipped with earplugs and sponge pads. An experienced radiologist checked the T1-weighted images to exclude participants with any structural abnormalities.

Experimental design

Participants completed an implicit emotional face n-back (EF-N-BACK) task (Acuff et al. 2018) in the presence of emotional distractor stimuli displayed in the peripheral visual field during a WM task (Fig. 1a). In this task, participants were required to respond to prespecified letters in a pseudorandom sequence of letters presented in the center of an LED screen. The task involved two difficulty levels of WM. One was a 0-back condition with a low-memory load (EF-0-back; e.g. pressing the “number 1” key with the index finger of the right hand when the letter “X” was presented or pressing the “number 2” key with the middle finder of the right hand when any other letter appeared). The other was a 2-back condition with a high-memory load (EF-2-back; e.g. pressing the “number 1” button when the letter was identical to the letter presented two trials ago, i.e. L–X–L; otherwise, pressing the “number 2” button). The letters were surrounded by fearful, happy, or neutral Asian face distracters, with a no-face condition controlling for the interference related to presenting a face distractor. Each run contained a total of eight stimuli blocks: two memory load conditions (0-back and 2-back) with four face distractor conditions each (fear, happiness, neutral, or no face). The task consisted of three runs of 7 min and 4 s, with a total of 24 blocks presented in a pseudorandom order. Each block included twelve 500 ms trials, with a single letter presented and flanked by identical facial expressions or no picture. Jittered intertrial intervals (average duration = 3500 ms) included a fixation cross (intermixed with faces). Participants were instructed to respond as quickly as possible, and we recorded the accuracy and reaction time of each run. At the beginning of each block, brief instructions were presented on the screen for 4000 ms. The task description and stimuli presentation were implemented using Psychtoolbox-3 (https://github.com/Psychtoolbox-3/Psychtoolbox-3/tree/3.0.19.0). Task practice and detailed instructions were provided prior to the scanning session.

Figure 1.

Basic experimental scheme. (a) The overview of the experimental block design. (b) Functional and structural MRI data preprocessing using the data for one participant as an example. (c) The extraction of 308 FC predictors across 14 task contrasts. FC values were computed between amygdalae and 11 ROIs to capture neural correlates of task conditions. Each contrast yielded 22 FC values representing neural interactions during the task. (d) gPPI analysis and the extraction of FC values. (e) The application of elastic-net regression with 3-fold cross-validation to select key features for correlation analysis of TAS-12 scores, incorporating 310 predictors including 2 demographic information and 308 FC predictors.

fMRI preprocessing

Image preprocessing (Fig. 1b) was performed with a standard processing stream in Statistical Parametric Mapping (SPM12) (http://www.fil.ion.ucl.ac.uk/spm/doc/) in MATLAB, version 2019a (MathWorks, Natick, MA, USA). Briefly, all image preprocessing can be summarized in the following steps:

1. The DICOM data format was converted to the NIFTI format, and visual inspection was performed to ensure the convenience of data processing.

2. The slice timing was corrected to account for differences in the time of slice acquisition.

3. Head motion was corrected using six parameters, with the first volume serving as a reference slice. In addition, the ArtRepair toolbox (http://spnl.stanford.edu/tools/ArtRepair/ArtRepair.htm) was used to correct for excessive movement.

4. All functional images were normalized to the Montreal Neurological Institute (MNI) space via the rigid body deformation fields derived from tissue segmentation of structural images (resampling voxel size = 2 mm × 2 mm × 2 mm) for accurate spatial localization within the brain.

5. Spatial smoothing was performed after normalization to improve the quality of the group-level statistics using an isotropic Gaussian kernel with a full-width at half-maximum (FWHM) value of 4 mm.

fMRI data analyses

We used SPM’s CONN generalized psychophysiological interaction (gPPI) models (Whitfield-Gabrieli and Nieto-Castanon 2012) to assess task-related connectivity between the right/left amygdala seed and regions of interest (Fig. 1d). We compared conditions with emotional (fearful, happy) and neutral faces to conditions presenting no faces, as well as contrasting conditions with emotional faces to neutral faces in the EF-0-back and EF-2-back tasks. Additionally, a comparison between EF-2-back and EF-0-back was conducted for each stimuli condition: fearful, happy, neutral, and no faces, to investigate the task load effects. These comparisons yielded a total of 14 distinct contrasts. In this study, we selected key brain regions related to facial and emotional processing as regions of interest (ROIs) from Neurosynth (https://neurosynth.org/). Specifically, we defined a spherical ROI with a 5 mm radius, centered on coordinates obtained from the Neurosynth database in the MNI space, for the FFA (Goh et al. 2010), OFA (Baseler et al. 2014), STS (Baseler et al. 2014), dorsolateral prefrontal cortex (dlPFC) (Grimm et al. 2008, Hanslmayr et al. 2012), and vlPFC (Alacreu-Crespo et al. 2020) (Table 1). Additionally, for the regions of the amygdala and ACC, we relied on the Harvard–Oxford atlas to determine their coordinates (Frazier et al. 2005, Desikan et al. 2006). To comprehensively capture the neural correlates of our task conditions, we computed FC values between the selected ROIs. For each of the 14 contrasts described earlier, we extracted FCs between the ROIs of the left/right amygdala and the others (bilateral OFA, FFA, STS, dlPFC, vlPFC, and ACC) in each subject. This resulted in a total of 22 FC values per contrast, representing the strength of the neural interactions between these key brain regions during the task. Across all 14 contrasts, this amounted to a total of 308 predictors (14 contrasts × 22 connectivity values; Fig. 1c). We also incorporated two critical demographic variables: gender and age. The integration of these factors expanded our predictor set to a total of 310 variables, encompassing the initial 308 functional connectivity predictors and the two demographic predictors. Further analysis was conducted using GLMNET in R (https://cran.r-project.org/web/packages/glmnet/glmnet.pdf.) and IBM SPSS software (version 22.0; SPSS, Inc.).

Table 1.

Center coordinates in MNI space for the selected ROIs as a sphere with 5 mm radius.

Region	Hemisphere	x	y	Z
FFA	L	−42	−50	−24
	R	43	−49	−22
OFA	L	−41	−85	−16
	R	42	−80	−16
STS	L	−49	−55	7
	R	53	−51	9
DLPFC	L	−45	6	39
	R	46	16	48
VLPFC	L	−50	34	10
	R	56	12	4

Statistical analysis

A single elastic-net regression analysis was used for feature selection and reduction via GLMNET in the R package glmnet (https://cran.r-project.org/web/packages/glmnet/index.html). The outcome variable of this model was the TAS-12 score. To address the issues of multicollinearity and overfitting, we utilized elastic-net regularization with a cross-validation scheme to select the key features (Kauttonen et al. 2015). This method, which combines ridge regression and LASSO regression, aims to minimize the loss function. We utilized the R package glmnet (https://cran.r-project.org/web/packages/glmnet/index.html) to implement this approach. The main parameter α, which varies between 0 and 1, determines the ratio between ridge regression (α = 0) and least absolute shrinkage and selection operator (LASSO) regression (α = 1); we selected α = 0.80, and the outcome variable was the TAS-12 score (Kauttonen et al. 2015). For cross-validation, we used 3-folds. The data were divided into two subsets, training and test sets, to more accurately assess the generalizability of the model and to identify the optimal regularization parameter λ. SPSS software was used for statistical analyses. In addition, the model included 310 predictor variables in total, demographic information (sex and age), and FC between the left/right AMY and each ROI (the ACC, left/right dlPFC, vlPFC, FFA, STS, and OFA) for each task condition contrast. Correlation analyses were also performed to evaluate the associations between the extracted FC and questionnaire scores. The significance level was set at P < .05.

Results

Demography and behavioral performance

A total of 25 subjects’ [mean (±s.d.) age = 35.28 (±2.99) years, 15 females] fMRI data and TAS scores [mean (±s.d.) score = 28.36 (±5.60)] were included in the analysis. We conducted a 2 (n-back conditions: 2-back and 0-back) by 4 (types of facial emotions: no faces, happy faces, neutral faces, and fearful faces) ANOVA test. The study findings demonstrated that reaction time and accuracy were predominantly influenced by the main effect of n-back conditions. Specifically, the mean reaction time significantly rose with increasing n-back levels (F = 2.122, P < .001; Fig. 2a), while accuracy declined (F = 58.061, P < .001; Fig. 2b). Facial emotion types showed no notable effect on reaction time (F = 0.146, P = .932; Fig. 2c) or accuracy (F = 0.632, P = .595; Fig. 2d). No significant interaction was observed between n-back conditions and facial emotion types for either measure. The adjusted R² for reaction time and accuracy models stood at 0.292 and 0.216, respectively.

Figure 2.

Comparison of behavioral performance metrics under different cognitive load and facial emotion conditions. Significant differences showed in (a) reaction time (P < .001) and (b) accuracy (P < .001) between 0-back and 2-back working memory tasks; no significant differences among (c) reaction time and (d) accuracy across facial emotion types.

TAS-12

The fMRI data and TAS scores [mean (s.d.) = 28.4 (5.60)] were included in the analysis. Using the TAS-12 score as the outcome variable of the model, an optimized model fit (λ = 3.188; Fig. 1e) from the initial 310 predictors, resulting in the selection of 5 features (Table 2). Further correlation analyses for the selected features showed that four of five had significant associations with TAS-12 scores (Fig. 3). Specifically, the TAS-12 score was positively correlated with the FC of the rAMY-rOFA in the EF-2-back task in the condition of a neutral face versus no face (r = 0.612, P = .001, Fig. 4a, Table 2), and with the FC of the lAMY-rOFA in the EF-0-back task in the condition of a fearful face versus no face (r = 0.510, P = .009, Fig. 4b, Table 2). Meanwhile, the TAS-12 score was negatively correlated with the FC of the rAMY-rOFA in the EF-0-back task with conditions of happy faces versus neutral faces (r = −0.537, P = .006; Fig. 4c, Table 2) and with the FC of the lAMY-rdlPFC in the EF-2-back task with conditions of happy faces versus no faces (r = −0.489, P = .013, Fig. 4d, Table 2).

Table 2.

Correlation analysis showed significant associations between TAS-12 scores and five specific features.

Questionnaire	Neuroimaging measures	Spearman correlation	Sig. (2‐tailed)
TAS-12	rAMY-rOFA in EF-2-back neutral face vs. No face	0.612	0.001
	rAMY-rOFA in EF-0-back happy vs. neutral face	−0.537	0.006
	rAMY-rdlPFC in EF-0-back happy face vs. No face	0.396	0.050
	lAMY-rdlPFC in EF-2-back happy face vs. No face	−0.489	0.013
	lAMY-rOFA in EF-0-back fearful face vs. No face	0.510	0.009

Figure 3.

Visualization of significant functional connectivity of ROIs associated with TAS-12 scores. Blue spheres indicate ROIs lacking significant correlation with TAS-12 scores in functional connectivity. Red spheres show ROIs with a significant correlation between connectivity values and TAS-12 scores. Red lines denote connections have a positive correlation with TAS-12 scores, while blue lines denote connections have a negative correlation with TAS-12 scores.

Figure 4.

Correlation analysis between TAS-12 scores and neuroimaging measures. (a) For the emotional face 2-back task, TAS-12 scores had a significantly positive correlation with the FC of right amygdala-right OFA in response to neutral faces versus no face. (b) For the emotional face 0-back task, TAS-12 scores had a significantly positive correlation with the FC of left amygdala–right OFA in response to fearful faces versus no face. (c) For the emotional face 0-back task, TAS-12 scores had a significantly negative correlation with the FC of right amygdala-right OFA in response to happy faces versus neutral faces. (d) For the emotional face 2-back task, TAS-12 scores had a significantly negative correlation with the FC of left amygdala–right dlPFC in response to happy faces versus no face.

Discussion

This study elucidated the associations of patterns of facial emotional processing during WM tasks of varying intensities with emotional processing abilities. We employed the TAS-12 as an assessment tool to quantitatively evaluate participants’ emotional processing abilities. We observed that under low or high memory load conditions, the OFA is primarily involved in facial perception rather than facial expression processing. Our research results indicate that during implicit emotional processing, the strength of FC between the amygdala and the OFA serves as a fundamental neurophysiological indicator for emotional processing.

In this study, we utilized the EF-n-back task to investigate the neural regions involved in emotional processing in healthy adults while performing a WM task. In previous studies, the EF-n-back task was designed to examine the impact of emotionally salient distracters on the attentional control processes involved in WM (Ladouceur et al. 2009, Bertocci et al. 2012). Using facial expressions as distracters has two advantages. First, facial expressions tend to effectively activate brain areas related to facial and emotional processing (Haxby et al. 2002). Second, facial expressions are nonverbal stimuli that carry emotional significance (Kacperek 1997). The EF-n-back task includes two memory-load conditions: the 0-back condition and the 2-back condition. To successfully complete the 2-back condition with a high memory load in this task, participants must focus their attention on the target letter, keeping at least three letters in WM while inhibiting engagement of attention by the emotional facial expressions flanking the target letters. This design investigates the role of task difficulty (i.e. the amount of attention resources recruited) in the impact of emotional interference on attentional control processes involved in WM. In addition, the EF-n-back task was also used to investigate the neural regions involved in emotional regulation during WM tasks. Kowalczyk et al. used the EF-n-back task to investigate the persistent impairment of WM and emotional processing-related brain function in patients at risk of developing postpartum psychiatric disorders. They found that at-risk women, as a group, exhibited excessive activation in the prefrontal cortex, cingulate gyrus, and subcortical regions during the fearful face task, while the connectivity between the left AMY and ipsilateral parietal-occipital regions decreased (Kowalczyk et al. 2021). In research on affective disorders (ADs), the application of the EF-n-back task identified neural markers of the development of AD in at-risk youth. These neural markers can provide new perspectives for future treatment and intervention strategies and help prevent or delay the development of affective disorders (Acuff et al. 2018, Fournier et al. 2021). Therefore, we utilized the emotional n-back WM task to investigate the neural process underlying the relationship between objective emotional process and subjective emotional cognition ability in our study.

Regarding behavioral outcomes, our data indicate that increasing memory load significantly influences both reaction times and accuracy. Specifically, a higher task load is associated with increased reaction times and decreased accuracy. However, the type of facial emotion presented did not significantly alter reaction times or accuracy. It is important to highlight that no significant interaction effect between emotional stimuli and memory load was observed, implying that the emotional content of the stimuli did not significantly alter the behavioral performance on WM. The lack of significant correlations between the accuracy and reaction times from the n-back and emotion tasks with TAS scores may be due to the study’s focus on a healthy participant population. In addition, regarding the subjective emotional cognition ability, the results of TAS scores on DIF [mean (min, max) = 15.76 (9, 24)] and DDF [mean (min, max) = 12.92 (10, 17)] indicated that the participants are all within a normal range referring to other studies for healthy subjects (Peng et al. 2023).

Regarding the measurement of subjective emotional perception ability, we found significant positive associations between TAS-12 scores and right/left AMY to right OFA FC strength during EF-2-back and EF-0-back tasks. The OFA participates in lower-level facial classification skills, such as distinguishing faces from non-face objects (Rossion 2008). Several studies have stimulated the right OFA located near the surface of the brain using transcranial magnetic stimulation (TMS), finding that participants’ facial perception was affected (Pitcher et al. 2009, 2012). Considering our comparison involved conditions where emotional faces contrasted with no faces, notably, participants with poorer emotional recognition exhibited greater sensitivity to facial stimulus. This improvement in facial perception may be due to the increased activity of the OFA to distinguish faces. In addition, we also found the positive correlation persists both in EF-0-back and EF-2-back conditions, indicating that the improvement in facial perception among participants with higher TAS-12 scores is not affected by WM tasks. This may be related to the unconscious facial processing performed by the OFA when adults view faces, which does not consume cognitive resources (Haist et al. 2010). Mattavelli et al.’s fMRI study found that, under unconscious conditions, even if the facial stimulus is unrecognized, the OFA region still elicits an excitatory response within a specific time window (∼180–240 ms after stimulation; Mattavelli et al. 2019). In our study, we employed an implicit emotional task, where participants processed facial information unconsciously. We postulate that participants with poorer emotional recognition exhibit greater sensitivity to facial stimuli, possibly due to an elevated excitatory response in the OFA region when viewing facial stimuli. On the other hand, we also found a negative correlation between TAS-12 scores and the FC value from the right amygdala to the right OFA during EF-0-back condition. This result showed that participants with poorer emotional recognition experience a decrease in their ability to detect positive emotions under low load memory conditions. The OFA not only participates in low-level facial discrimination but is also involved in higher-level facial perception, such as identity, gender, and expression judgment (Atkinson and Adolphs 2011). Facial expressions of emotion convey crucial social information and have the ability to evoke emotions in others. TMS studies have indicated that the right OFA is instrumental in processing common configural cues associated with changes in facial expressions. Cohen Kadosh et al. utilized a task wherein participants were required to ascertain whether a face matched one previously presented (Kadosh et al. 2011). The paired faces might vary solely in terms of identity, emotion, or gaze direction, or in any combination of these three dimensions. After repeated 10 Hz TMS on the right OFA, there was a decline in participants’ accuracy in matching faces based on a combination of identity and expression, while the facial processing remained unaffected per se. Based on the above evidence, we suppose that during low memory load, the decrease in emotional perception among participants with higher TAS scores is associated with reduced emotional expression processing in the OFA. There are two possible explanations for this finding. Firstly, as we have previously discussed, for participants with poorer emotional identification and expression skills, the OFA primarily participates in lower-level facial classification skills during unconscious facial perception, thereby reducing the higher-level facial perception functions of the OFA, affecting expression processing, and consequently influencing emotional processing. Secondly, the OFA involves both configural and featural processing when processing expressions (Cohen Kadosh et al. 2010). Nevertheless, in our study, we utilized an implicit emotional task in which facial emotional stimuli were presented in the peripheral visual field, resulting in reduced visibility of the faces. This may have disrupted the processing of facial configural and featural information, thus impacting the OFA’s proficiency in facial expression processing.

Interestingly, we also found that TAS-12 scores were negatively correlated with the FC between the left amygdala and right dlPFC in the contrast of happy face to no-face conditions under high-memory loads. This finding indicated that participants with poorer emotional recognition exhibited a decreased ability to perceive facial stimuli. Neuroimaging studies have highlighted the involvement and significance of the dlPFC in cognitive control (Vuilleumier et al. 2001, Bunford et al. 2017). Conflicts arise when different streams of information compete for processing resources, necessitating the involvement of the dlPFC to resolve them. In Lavie et al.’s load theory, there exist two mechanisms of selective attention: (i) under conditions of high perceptual load, attentional resources are depleted, making it difficult to process other unattended information; and (ii) under conditions of low cognitive load, attentional resources are more likely to spill over to other unplanned information, allowing it to be processed (Lavie et al. 2004). It is consistent with our findings that participants with poorer emotional abilities experience a decrease in their ability to detect facial stimuli during the EF-2-back task. This could be due to weaker FC between the right amygdala and right dlPFC, leading to reduced recruitment of attentional resources in high-load cognitive tasks. Notably, TAS-12 scores were positively associated with the FC between the left AMY and right dlPFC in the contrast of happy face to no face condition under low memory loads (there’s a marginal significance with a P-value of .05). This observation could also be explained by the load theory that, at lower perceptual loads, participants with poorer emotional recognition are inclined to devote a greater proportion of attentional resources to facial processing.

Limitations of our study

This study has several limitations. First, the sample size was limited, and only healthy participants were recruited. This limitation restricts the generalizability of our findings, as it remains unclear whether the observed neural mechanisms would apply to individuals with psychological disorders or other populations. Future studies should aim to recruit larger and more diverse samples, including patients with depression or anxiety disorders, to further validate the role of the neural mechanisms we have identified in psychological disorders. Second, we did not divide the AMY into subregions. The basolateral amygdala (BLA) is believed to be involved in processing high-level sensory input and stimulus-value associations, while the centromedial amygdala (CMA) is involved in generating attentional, autonomic, and motor responses (Bzdok et al. 2013). Huang et al. found that BLA output neurons receive a diverse range of inputs from throughout the brain, each with different strengths and a strong emphasis on contextual information (Huang et al. 2021). Analyzing the subregions of the amygdala may help to identify the specific neural circuits and pathways involved in emotion processing, providing a more nuanced understanding of how the brain processes emotional stimuli by examining these subregions separately.

Conclusion

This study demonstrates a correlation between individuals’ subjective ability to describe and recognize emotions and the FC between brain regions involved in facial emotional processing. We found that participants with poor emotional recognition abilities exhibit certain brain activation patterns. Specifically, people with harder emotional recognition may experience an increased sensitivity in facial perception demonstrated by stronger FC between the amygdala and OFA. On the other hand, people with poorer emotional recognition tend to demonstrate a lack of adequate allocation of cognitive resources towards emotional processing, resulting in a diminished capacity for emotional cognition. Our findings offer new insights into the neural mechanisms underlying facial emotional processing in healthy participants, potentially aiding in the diagnosis and treatment of emotional abnormalities in patients with mental disorders.

Conflict of interest

None declared.

Funding

This study was supported by the National Natural Science Foundation of China (NSFC 82070996), the 1·3·5 project for disciplines of excellence—Clinical Research Fund, West China Hospital, Sichuan University (2024HXFH044) and The Ministry of Science and Technology of the People’s Republic of China (STI2030-Major Projects 2021ZD0201900).

Data availability

The dataset supporting the findings of this article can be obtained upon request to the corresponding authors.

Author contribution

Gantian Huang (Conceptualization, Methodology, Data curation, Writing—original draft: Writing—review & editing), Chen Qiu (Resources, Data curation, Writing—review & editing), Ping Jiang (Resources, Methodology, Data curation, Writing—review & editing, Supervision, Funding acquisition), Qiyong Gong (Resources, Data curation, Writing - review & editing, Project administration), Meng Liao (Writing—review & editing), and Longqian Liu (Writing—review & editing, Supervision, Funding acquisition).