Depression is a significant global health concern, and understanding its etiology and maintenance is crucial. Cognitive models posit that biases in emotion processing, particularly a preference for negative stimuli, contribute to both the onset and persistence of depression. Behavioral studies have shown negative biases in attention, interpretation, and memory in depressed individuals, even at early, automatic processing stages using subliminal stimuli. Cross-sectional neuroimaging studies have further implicated the amygdala, a key region in the emotion processing network, showing amygdala hyperactivity to subliminal negative stimuli and hypoactivity to positive stimuli in depressed patients. However, the question remains whether these neurofunctional abnormalities are transient, state markers associated with active depression, or enduring, trait markers representing a vulnerability to or consequence of depression. Most longitudinal studies have focused on controlled, later stages of emotion processing, leaving the persistence of automatic emotion processing biases unclear. This study aimed to address this gap by investigating the neural correlates of automatic emotion processing in a naturalistic, two-year longitudinal design, focusing on amygdala activity using a well-established subliminal affective priming paradigm.

Existing literature suggests a strong link between biased emotion processing and depression. Cognitive models highlight the role of cognitive vulnerability and reactivity in maintaining depressive states. Behavioral studies have demonstrated attentional, interpretational, and memory biases towards negative stimuli in depressed individuals, even with subliminal presentation. Neuroimaging research has consistently shown amygdala hyperactivity to masked negative stimuli and hypoactivity to positive stimuli in individuals with depression. However, most longitudinal neuroimaging studies have focused on later stages of emotion processing, with limited investigation into the persistence of these biases in automatic processing over extended periods. The few existing studies suggest some normalization of amygdala dysfunction after treatment, but the long-term stability of this bias remains unclear.

This prospective case-control study recruited 57 patients with acute MDD and 37 healthy controls (HC). Participants underwent fMRI at baseline and after approximately two years. The fMRI scans were conducted while participants completed a subliminal affective priming paradigm using masked emotional faces (sad, happy, neutral). The paradigm involved subliminal presentation (33ms) of an emotional prime followed by a supraliminal neutral target face (467ms). Participants rated the valence of the target faces. Data preprocessing involved realignment, unwarping, spatial normalization to MNI space, and smoothing. Statistical analysis utilized a 3 × 2 × 2 ANCOVA with group (HC, MDD relapse, MDD no-relapse) as a between-subjects factor and time (baseline, follow-up) and condition (sad > neutral, happy > neutral) as within-subjects factors. Age and sex were included as covariates. A region-of-interest (ROI) approach focused on the amygdala, using threshold-free cluster enhancement (TFCE) for multiple comparison correction. Additional analyses explored the effects of comorbid disorders, medication, psychotherapy, current mood state, prior disease progression, and environmental risk factors.

At baseline, MDD patients showed significantly increased amygdala activity to sad primes compared to HC, replicating previous findings. Importantly, this amygdala hyperactivity to sad primes persisted at the two-year follow-up in both the MDD relapse and MDD no-relapse groups, regardless of remission status. Healthy controls showed a positive bias, with higher amygdala responses to happy primes than sad primes at baseline, but this difference was not as pronounced at follow-up. There was no significant difference in amygdala activity between the relapse and no-relapse groups at baseline, indicating that baseline amygdala activity did not predict subsequent relapse. Additional analyses revealed that higher medication loads were associated with increased amygdala activity to happy primes at follow-up. Whole-brain analyses at baseline showed group × condition interactions in several regions involved in visual and attentional processing, but these differences were not significant at follow-up using a strict threshold. Behavioral data did not show a clear affective priming effect, although participants showed longer reaction times to sad primes.

The persistent amygdala hyperactivity to subliminal negative stimuli, irrespective of relapse or remission, suggests that this neurofunctional abnormality is a trait marker of depression, rather than a state marker. This contrasts with findings from studies focusing on controlled processing stages, where normalization of activity is often observed in remission. The persistence of this bias suggests it may represent a vulnerability factor or a lasting consequence of the illness. The lack of predictive value of baseline amygdala activity for relapse suggests that this marker reflects general vulnerability rather than individual prognosis. Medication effects may contribute to observed patterns, particularly the increased amygdala response to happy faces at follow-up in the relapse group. The absence of consistent behavioral priming effects highlights the potential sensitivity of fMRI in detecting subtle biases in emotion processing.

This study provides strong evidence that amygdala hyperactivity to subliminal negative stimuli is a persistent trait marker of depression. This finding has implications for understanding the neurobiological underpinnings of depression and developing more effective treatments. Future research should focus on larger samples to investigate the potential predictive value of this trait marker and explore the impact of specific treatments on these persistent neural patterns. Further investigation into the role of genetic and environmental factors in shaping this vulnerability is also warranted.

The study's naturalistic design, while enhancing generalizability, limits the control over confounding factors. The sample's relatively high clinical severity and the small sample size of the relapse and no-relapse subgroups may limit the generalizability of the findings to less severe cases and reduce the statistical power of some analyses. The lack of a clear behavioral affective priming effect necessitates caution in interpreting the results solely based on neural activity.