Participants included MDD patients (MDD group) and healthy controls (HC group). Inclusion criteria for the MDD group were as follows: (a) met the criteria of MDD of the Diagnostic and Statistical Manual of Mental Disorders, Fifth Edition (DSM-5); (b) age ranged from 18 to 65 years old; (c) had a minimum primary-school-level education; (d) had normal or corrected-to-normal vision. Exclusion criteria included: (a) met DSM-5 criteria for any other mental disorders; (b) had electroconvulsive therapy or modified electroconvulsive therapy within 6 months before recruitment; (c) had taken medication that damaged cognitive function, such as atropine, or benzodiazepine, during the last two weeks prior to recruitment; (d) had any significant physical or neurological illnesses; and (e) had difficulty fulfilling the experimental requirements. Inclusion criteria for the HC group included: (a) age ranged from 18 to 65 years; (b) a minimum primary-school-level education; (c) well understood the experimental instructions and had capacity to complete the entire experiment.

MDD patients were selected from the Department of Psychology of the Wuxi Mental Health Center, affiliated with Nanjing Medical University, China. HCs were recruited from Wuxi City, China, via advertisements in local residential communities. Sample-size estimation was performed using G*Power software (latest ver. 3.1.9.7, Heinrich-Heine-Universität Düsseldorf, Düsseldorf, Germany), targeting a statistical power of 0.95 (1-$\beta$ = 0.95) with an $\alpha$ level of 0.05 for F tests, resulting in a minimum requirement of 23 subjects per group. The study ultimately included 31 MDD patients and 32 HCs. However, five MDD patients and four HCs were excluded from subsequent analyses due to poor electrode contact and frequent blink or muscle artifacts.

The present study was conducted at the Wuxi Mental Health Center, affiliated with Nanjing Medical University, China, from December 4, 2020 to July 31, 2023. The research was approved by the Ethics Committee of the Wuxi Mental Health Center, affiliated with Nanjing Medical University (WXMHCIRB2020LLkY001), and the study was carried out strictly in accordance with guidelines outlined in the Declaration of Helsinki. Participants were told about the equipment used in the experiment and their tasks and they all signed informed consent forms prior to their participation.

In the present study, we used 240 emotional pictures (120 negative and 120 neutral) selected from the Chinese Affective Picture System (CAPS) [24]. These materials were divided equally into two sets, each containing 120 images (60 negative and 60 neutral). One of the sets served as the study items, and the other served as the distractors in the recognition test. Both the study items and distractors were matched in terms of valence and arousal. All affective pictures used in this experiment were resized to the identical size and resolution (15.5 cm $\times$ 11.6 cm, 433 $\times$ 325 pixels) by Picture Manager Software. Moreover, the background, contrast, and brightness of all emotional images were consistent. All stimuli were presented on a 19-in computer screen with 1280 $\times$ 1024 resolution and a 60-Hz refresh rate.

The experiment consisted of two parts: study phase (see Fig. 1A) and recognition test (see Fig. 1B). A calculation task was performed between the two phases. Subjects were seated in a moderately lit and sound-attenuated room at approximately 60 cm from a computer screen. They were instructed to keep their eyes fixed on the monitor and avoid blinking their eyes or moving their body as much as possible. Before the formal trial, there was a practice session to ensure that all participants were familiar with the procedure.

Fig. 1.

Schematic illustration of the Emotional Directed Forgetting task. (A) Study phase. ‘记’ cue (Chinese character denoting Remember. (B) Test phase.

During the study phase, 60 negative images and 60 neutral images were presented in pseudo-randomised order for each participant. Initiated by a 500-ms presentation of a black cross in the centre of the screen, a blank screen was displayed with a duration ranging randomly from 200 to 400 ms. After the blank screen, an emotional image was presented for 1000 ms. Subsequently, a second fixation point, a star, was shown for 1200 ms, followed by another blank screen with a duration that varied from 200 to 400 ms, randomly. This was followed for 1500 ms by either a ‘忘’ cue (Chinese character denoting Forget) which meant forget the previous image, or a ‘记’ cue (Chinese character denoting Remember) which meant remember the previous image. The next trial began immediately afterward. It is worth noting that the same instruction was set not to occur three or more times in a row.

After the study phase, participants were instructed to do a calculation test (multiplication of 2- or 3-digit numbers) as a distractor for 5 min.

During the test phase, both the images presented in the first part (60 negative and 60 neutral) and newly presented images (60 negative and 60 neutral) were displayed in pseudo-random order. Each trial began with a blank screen presented for 200–400 ms randomly, followed by presentation of a fixation point, a black cross, lasting for 500 ms. Then an emotional image was displayed for 2000 ms. During the presentation of the image, participants had to decide whether they had seen the picture in the study phase, regardless of the preceding F/R instruction. Subjects were required to click the left or right mouse button with their right index finger to indicate their decisions. The left mouse button was clicked for an image that they had seen before; the right mouse button was clicked for an image that they had not seen before. The assignment of right and left buttons was balanced for all participants.

During the Emotional Directed Forgetting (EDF) task, continuous EEG data were synchronously recorded using a 64-channel EasyCap (Brain Products GmbH, Wörthsee, Bavaria, Germany) and a BrainAmp Standard amplifier (Brain Products GmbH, Wörthsee, Bavaria, Germany). The EEG data were continuously sampled at a rate of 500 Hz and amplified using a 0.1–100-Hz band-pass for offline analysis. The reference electrode was placed in the centre of the forehead and the ground electrode was placed 1–2 cm below the left clavicle. For recording the artifacts produced by blinks and other types of eye movements, two horizontal electrogram electrodes were placed about 1 cm out from the lateral canthus of both the right and left eyes, and one vertical electrogram electrode was placed approximately 1 cm below the centre of the lower margin of the left eye. Electrode impedance was carefully kept below 5 k$\mathrm{\Omega }$ throughout the EEG recording period.

After recording, the EEG data were pre-processed using EEGLAB 2021 toolkit under the MATLAB 2020b (The MathWorks, Inc., Natick, MA, USA) environment platform. The recorded EEG data were filtered by a band-pass of 0.1–30 Hz and re-referenced to the algebraic average of the two mastoids. The atypical artifacts from eye movement, muscle, or cardiac components were rejected and corrected using independent component analysis. Trials contaminated with excessive artifacts exceeding an amplitude of $\pm$100 µV were excluded from further analyses.

For the study phase, two types of trials were left: trials related to emotional images (Neutral or Negative); and trials related to instructive cues (Remember Negative, Remember Neutral, Forget Negative, Forget Neutral).

In accordance with the grand average waveform and previous research [25, 26, 27], image-evoked ERP was selected from four locations along the midline: frontal (FZ), frontocentral (FCZ), central (CZ), and parietal (PZ), for statistical analysis. The average amplitudes of P2 and N2 were calculated after onset of the emotional images within the window of 150–200 ms and 250–300 ms, and the average amplitudes of LPPs were calculated after onset of the emotional images within the window of 500–1000 ms after the onset of image presentation, respectively.

In line with previous study, which analyzed the neural mechanisms underlying directed forgetting in individuals with depressive tendencies [22], we sorted the ERP data into four experimental conditions corresponding to the cues: Remember Negative, Remember Neutral, Forget Negative, and Forget Neutral, to explore the brain activity elicited by the instructions. These conditions were determined by a combination of instructions and valences. The average amplitudes of occipital P1, frontal N2 and parietal P3 components of cue-evoked ERP were analyzed. In particular, the P1 amplitude was calculated as the average amplitude over a window of 80–120 ms after the onset of cues at the electrode sites of O1 and O2. The N2 amplitude was calculated as the average amplitude over a window of 200–250 ms after the onset of cues at the electrode sites of FC1, FC2 and FCZ. The P3 amplitude was calculated as the average amplitude over a window of 300–400 ms after the onset of cues at the electrode sites of CP1, CP2 and CPZ.

The Hamilton Depression Rating Scale (HAM-D) (24-item edition), developed by Hamilton in 1960, is the most widely used scale in the clinical assessment of depression. The total scores are less than 20 points is normal, total scores are more than 20 points must be depression, and total scores are more than 35 points is suffering from severe depression [28].

All data were analyzed using IBM SPSS Statistics version 22 (IBM Corp., Armonk, NY, USA). Quantitative data and qualitative data comparing the MDD group and the HC group used the independent t test (two-tailed) and Pearson chi-square test respectively. Analysis of variance (ANOVA) was performed on both behavioral data and ERP data. The Greenhouse-Geiser corrected method was used to adjust statistical results, if necessary. The Bonferroni technique was used to correct for pairwise comparisons of interactions. A two-tailed Pearson r correlation was performed between HAM-D24 scores and behavioral/ERP data. Effect sizes were reported as partial eta squared ($\eta$_p²) and Cohen’s d. The significance level was set at 0.05.

For behavioral data, hit rate, false alarm rate, and reaction times (RTs) were analyzed. Hit rate was defined as the proportion of old items (images presented in the study phase) correctly recognised as ‘old’, regardless of type of instructions. False alarm rate was defined as the proportion of new items (images never presented in the study phase) incorrectly recognised as ‘old’, regardless of type of instructions. RT was defined as the time spent from the onset of the image to the pressing of the mouse button for a correct response. A three-way mixed, repeated-measures ANOVA was performed for hit rate and RT, with group (HC vs. MDD) as a between-subjects factor and valence (Negative vs. Neutral) and instruction (Remember vs. Forget) as within-subjects factors. Considering that false alarm rate could not be distinguished by instruction, it was analyzed using a 2 $\times$ 2 ANOVA (Group: MDD vs. HC) $\times$ (Valence: Negative vs. Neutral).

For ERP data, image-evoked P2, N2, and LPP were analyzed using a 2 $\times$ 2 $\times$ 4 mixed-design ANOVA (Group: MDD vs. HC) $\times$ (Valence: Negative vs. Neutral) $\times$ (Region: frontal vs. frontocentral vs. central vs. parietal). Furthermore, cue-evoked P1, N2, and P3 were entered into a 2 $\times$ 2 mixed-design ANOVA (Group: MDD vs. HC) $\times$ (Valence: Negative vs. Neutral) $\times$ (instruction: Remember vs. Forget).

As shown in Table 1, there was no significant difference between the MDD group and HC group in terms of gender, age, years of education, or handedness.

Hit rate. The interaction between valence and instruction was significant (F (1, 52) = 10.839, p = 0.002, $\eta$_p² = 0.175). Both the negative and neutral images in the remember-cue condition were recognised at higher rates than were those in the forget-cue condition (Negative: F (1, 52) = 7.788, p = 0.007, $\eta$_p² = 0.132; Neutral: F (1, 52) = 43.835, p = 0.0001, $\eta$_p² = 0.462). However, the negative images were recognized at higher rates than were the neutral images only in the remember-cue condition (F (1, 52) = 5.176, p = 0.027, $\eta$_p² = 0.092). The main effect of group was significant (F (1, 52) = 4.157, p = 0.047, $\eta$_p² = 0.075); the MDD group had lower hit rates than did the HC group (see Table 2).

False Alarm Rate. The main effect of group was significant (F (1, 52) = 4.965, p = 0.03, $\eta$_p² = 0.089). More false alarms occurred in MDD group than in the HC group. Also, a significant effect of valence was found (F (1, 52) = 29.041, p = 0.0001, $\eta$_p² = 0.363), with a larger false alarm rate observed with negative images than with neutral images (see Table 2).

RTs. The main effect of group was significant (F (1, 52) = 4.556, p = 0.038, $\eta$_p² = 0.084); the RTs were longer for the MDD group than for the HC group. The main effect of valence was significant (F (1, 52) = 30.741, p = 0.0001, $\eta$_p² = 0.381), and showed that RTs for the negative images were longer than those for the neutral images. The main effect of instruction was significant (F (1, 52) = 21.130, p = 0.001, $\eta$_p² = 0.297), in that the RTs were significantly longer after the Forget instruction than after the Remember instruction (see Table 2).

To ensure the quality of data, we rejected bad epochs caused by inevitable artifacts (e.g., violent muscle shocks, unavoidable signal interference and poorly conditioned electrodes) that the tool for pre-processing could not remove. Consequently, the number of trials in each condition was disparate. The number of trials [mean (SD)] entered into the statistics in the ‘Neutral’ and ‘Negative’ conditions were 52.31 (8.78) and 52.00 (8.88) for the MDD group, and 51.93 (9.70) and 52.86 (8.03) for the HC group, respectively. Moreover, the number of trials under the ‘Remember Negative’, ‘Remember Neutral’, ‘Forget Negative’, ‘Forget Neutral’ conditions were 24.50 (5.98), 24.81 (6.51), 24.69 (6.57), and 25.23 (6.30) for the MDD group, and 26.04 (4.94), 25.82 (5.70), 26.64 (3.86), and 25.43 (5.12) for the HC group, respectively.

Image-evoked P2. As shown in Fig. 2, in the P2 time window, the main effect of valence was significant (F (1, 52) = 4.64, p = 0.039, $\eta$_p² = 0.079), which indicated that, compared with neutral images, the negative images evoked a higher positive P2 amplitude. There was also a significant main effect of region (F (3, 50) = 12.463, p 0.001, $\eta$_p² = 0.428). Post hoc analysis showed that the P2 amplitudes at the frontal location were more positive than were those at the frontocentral, and parietal locations (all p 0.05). The main effect of group was significant (F (1, 52) = 4.325, p = 0.043, $\eta$_p² = 0.077); the MDD group had more positive P2 amplitudes than did the HC group. As for latencies, there was no significant difference between the two groups.

Fig. 2.

Grand averaged ERPs in frontal (FZ), frontocentral (FCZ), central (CZ), and parietal (PZ) locations and topographical distribution of grand averaged image-evoked P2, N2 and LPP of both MDD and HC groups. The image-evoked P2 time window is 150–200 ms; the image-evoked N2 time window is 250–300 ms, the image-evoked LPP time window is 500–1000 ms. LPP, late positive potential; MDD group, major depressive disorder group; HC, healthy control group; ERP, event-related potentials.

Image-evoked N2. The interaction between group and region was significant (F (3, 50) = 4.098, p = 0.011, $\eta$_p² = 0.197); the simple effect analysis showed that, compared with the HCs, the MDD participants had smaller N2 amplitudes in the frontal, frontocentral, and central regions (all p 0.05). There was also a significant interaction between valence and region (F (3, 50) = 5.937, p = 0.002, $\eta$_p² = 0.263). The simple effect analysis showed that the N2 amplitudes that were evoked under negative conditions were larger than those evoked under neutral conditions in the frontal, frontocentral, central, and parietal regions (all p 0.05). Moreover, mean amplitudes of the N2 evoked by neutral and negative images became increasingly larger from the parietal to the frontal region (all p 0.05) (see Fig. 2).

For N2 latencies, the interaction between valence and region was significant (F (3, 50) = 3.075, p = 0.036, $\eta$_p² = 0.156). Further post hoc analysis showed that, compared with neutral images, the N2 evoked for negative images had significant longer latencies at the frontal, central, and parietal positions (all p 0.001). However, there was no significant difference for N2 latencies between the groups (F (1, 52) = 0.016, p = 0.899, $\eta$_p² = 0.000).

Image-evoked LPP. As presented in Fig. 2, in the LPP time window, there was a significant interaction between emotion and group (F (1, 54) = 8.120, p = 0.006, $\eta$_p² = 0.131). On the one hand, the simple effect analysis reflected that greater LPP amplitudes were evoked for negative images than for neutral images (MDD: F (1, 52) = 6.291, p = 0.015, $\eta$_p² = 0.108; HC: F (1, 52) = 36.017, p = 0.001, $\eta$_p² = 0.409). On the other hand, the MDD group had larger LPP amplitudes than did the HC group both in negative and neutral emotional conditions (Negative: F (1, 52) = 4.779, p = 0.033, $\eta$_p² = 0.084; Neutral: F (1, 52) = 12.979, p = 0.001, $\eta$_p² = 0.200).

Cue-evoked P1. There was no other interaction or main effect found with P1 amplitudes or latencies.

Cue-evoked N2. As shown in Fig. 3, the main effect of group was significant (F (1, 52) = 6.605, p = 0.013, $\eta$_p² = 0.113). The MDD group had the larger N2 amplitudes than did the HC group. The main effect of instruction was significant (F (1, 52) = 6.451, p = 0.014, $\eta$_p² = 0.110), as the Remember cues elicited larger N2 amplitudes than did the Forget cues. For N2 latencies, there was no other difference found to exist.

Fig. 3.

Grand averaged amplitudes and topographical distribution of cue-evoked N2 of both MDD and HC groups. The cue-evoked N2 time window is 200–250 ms.

Cue-evoked P3. As shown in Fig. 4, the main effect of group was significant (F (1, 52) = 8.972, p = 0.004, $\eta$_p² = 0.147). The HC group showed larger P3 amplitudes than did the MDD group. The main effect of instruction was significant (F (1, 52) = 5.815, p = 0.019, $\eta$_p² = 0.101), as the Remember cues evoked larger P3 amplitudes than did Forget cues. For P3 latencies, there was no other interaction or main effect observed.

Fig. 4.

Grand averaged amplitudes and topographical distribution of cue-evoked P3 of both MDD and HC groups. The cue-evoked P3 time window is 300–400 ms.

As shown in Fig. 5, the HAM-D24 scores showed a moderately robust positive correlation with the LPP amplitudes that were evoked by negative images at the central location (r = 0.447, p = 0.022). However, the HAM-D24 scores had no correlation with the other ERP components (all p $>$ 0.05).

Fig. 5.

Scatterplot of image-evoked LPP amplitudes under negative condition in central location and the HAM-D 24 scores in MDD group. MDD, major depressive disorder; HAM-D, Hamilton Depression Rating Scale; LPP, late positive potential.