Persistence of amygdala hyperactivity to subliminal negative emotion processing in the long-term course of depression

The study addresses whether abnormalities in automatic emotion processing—specifically amygdala hyperreactivity to negative stimuli—constitute a persistent, trait-like marker of depression or are state-dependent changes that normalize with remission. Cognitive models posit that depression involves negative biases throughout attention, memory, and evaluation, with automatic processing stages implicated. Prior cross-sectional neuroimaging indicates amygdala hyperactivity to subliminal negative stimuli and hypoactivity to positive stimuli in acute MDD, but long-term prospective evidence beyond 8 weeks is lacking. This work aims to (a) replicate mood-congruent bias in automatic emotion processing at baseline in acute MDD, (b) test whether these neural alterations normalize or persist over 2 years depending on illness course (relapse vs no-relapse), and (c) evaluate whether baseline amygdala function predicts subsequent relapse; additionally, the influence of medication and psychotherapy is explored. The findings are important for clarifying trait vs state neural markers in depression and potential vulnerability mechanisms.

Behavioral studies show negative biases in attention, interpretation, and memory in depressed individuals and those at risk, including effects under subliminal conditions. Neuroimaging evidence demonstrates mood-congruent amygdala hyperreactivity to masked negative faces and reduced responses to positive stimuli in acute MDD, with amygdala reactivity linked to negative evaluative biases. Short-term interventional studies (SSRIs, ECT) suggest normalization of automatic amygdala responses within weeks, hinting at state-dependence. Conversely, risk factors such as 5-HTTLPR genotype and childhood maltreatment associate with automatic amygdala biases in healthy or at-risk samples, suggesting trait-like vulnerability. Longitudinal neuroimaging over controlled (supraliminal) processing stages indicates normalization with remission, but there has been no long-term prospective study of automatic (subliminal) processing over periods >8 weeks prior to this study.

Design: Prospective, naturalistic longitudinal case-control study within the Münster Neuroimaging Cohort. Baseline assessments between May 2010–June 2015; follow-up ~2 years later (September 2012–August 2017). Participants: n=57 inpatients with acute moderate/severe MDD at baseline; n=37 healthy controls (HC) without lifetime psychiatric disorders at both time points. Exclusions: chronic medical illnesses, neurologic abnormalities, benzodiazepine intake at any time point, MRI contraindications; for MDD: bipolar/psychotic disorders, acute substance dependence, history of ECT. Diagnoses via SCID-I (DSM-IV) at both time points; HDRS administered. Patients categorized by 2-year course: relapse group (ongoing depression at follow-up or ≥1 relapse; n=37) vs no-relapse group (full remission at follow-up, no further episodes; n=20). Medication type/dose recorded; composite medication load index computed per time point; psychotherapy frequency/type documented. Ethics: Approved (University of Münster 2007-307-f-S); written informed consent. Task: Subliminal affective priming paradigm (80 trials). Prime: sad, happy, neutral, or no-face presented for 33 ms (subliminal), masked by a neutral target face (same identity) presented for 467 ms (supraliminal). Participants rated valence of neutral target (4-point scale). Stimuli from Ekman and Friesen set; participants were unaware of primes. MRI acquisition: 3T Philips Gyroscan Intera; T2* EPI; 34 slices; matrix 64×64; voxel 3.6×3.6×3.6 mm; TR 2.1 s; TE 30 ms; flip 90°; interleaved acquisition; slices tilted 25° from AC-PC to reduce OFC/MTL dropout; visual projection rear of scanner. Preprocessing: Realign, unwarp, normalize to MNI, smooth 6 mm FWHM (SPM8). Event-related GLM modeling onsets for prime conditions (happy, sad, neutral, no-face) with canonical HRF; serial correlation correction; high-pass 128 s. First-level contrasts: sad>neutral; happy>neutral. Second-level analyses: SPM12 with bilateral amygdala ROI (WFU PickAtlas AAL3.1). Non-parametric TFCE with 10,000 permutations; combined peak-cluster FWE-corrected p<0.05. ANCOVA 3×2×2 with factors group (HC, MDD relapse, MDD no-relapse), time (baseline, follow-up), condition (happy>neutral, sad>neutral); covariates: age, sex. Tested main and interaction effects. Specific tests: baseline group×condition interaction (objective a) with post-hoc within- and between-group t-tests; follow-up group×condition (objective b); baseline differences between MDD subgroups per condition (objective c). Exploratory whole-brain analyses with voxel-wise FWE p<0.05 (and uncorrected p<0.001 exploratory). Additional analyses: control for comorbid disorders; effects of medication load and psychotherapy; associations with current mood (HDRS), prior disease progression (first vs recurrent), and environmental risk (childhood maltreatment). Behavioral analyses per Supplementary methods.

Amygdala ROI ANCOVA: Significant main effect of group bilaterally (left: k=126, F2,362=26.23, PTFCE-FWE<0.001, ηρ²=0.122; right: k=198, F2,362=24.72, PTFCE-FWE<0.001, ηρ²=0.089). No main effects of condition or time (PTFCE-FWE>0.999). Significant group×condition interaction when patient subgroups combined (right: k=8, F1,362=11.40, PTFCE-FWE=0.040, ηρ²=0.106). No group×time, condition×time, or three-way interactions (all PTFCE-FWE≥0.456). Baseline (objective a): Significant group×condition interaction in bilateral amygdala (left: k=17, t(362)=3.40, PTFCE-FWE=0.027; right: k=14, t(362)=3.14, PTFCE-FWE=0.035). Post-hoc: MDD (both subgroups) > HC for sad primes (e.g., all MDD>HC right: x=32,y=−2,z=−12, t=4.08, k=157, PTFCE-FWE=0.001; left: x=−24,y=0,z=−16, t=3.41, k=110, PTFCE-FWE=0.005). No group differences for happy primes (PTFCE-FWE≥0.644). Within-group: HC showed higher amygdala activity to happy vs sad primes (right: k=45, t(362)=3.46, PTFCE-FWE=0.025); MDD groups showed no significant within-condition difference but numerically higher responses to sad vs happy. Follow-up ~2 years (objective b): Group×condition interaction not significant (PTFCE-FWE=0.107), yet both MDD subgroups remained higher than HC for sad primes (e.g., MDD>HC right: x=28,y=−6,z=−14, t=2.94, k=186, PTFCE-FWE=0.002; left: x=−28,y=−4,z=−22, t=2.58, k=8, PTFCE-FWE=0.036). Happy primes: relapse group > HC at follow-up (left: k=4, t(362)=2.66, PTFCE-FWE=0.046). Relapse vs no-relapse did not differ significantly at follow-up; trend for higher happy-prime activity in relapse (PTFCE-FWE=0.058). Within-group at follow-up: no differences between happy and sad primes (all PTFCE-FWE≥0.222). Prediction of relapse (objective c): No baseline differences between relapse and no-relapse groups in amygdala for sad (PTFCE-FWE=0.522) or happy primes (PTFCE-FWE=0.460). Additional analyses: Controlling for comorbidity (including anxiety) did not alter the main effects/interaction. Medication load×time interaction for happy primes: higher medication loads associated with elevated amygdala activity to happy primes at follow-up, not at baseline. No significant psychotherapy effects on amygdala function. No association between HDRS symptom severity and amygdala activity (p≥0.296); higher amygdala responses to sad primes vs HC persisted irrespective of remission status at follow-up. No baseline amygdala differences between first-episode vs recurrent patients (PTFCE-FWE=0.190). Trend for higher baseline amygdala response to sad primes in those with childhood maltreatment (PTFCE-FWE=0.077). Behavioral: No significant affective priming effect on valence ratings; reaction times were longer for sad vs neutral primes at baseline across groups. Exploratory whole-brain: At baseline, group×condition interaction driven by increased activity to sad primes in MDD vs HC in regions including middle temporal pole/gyrus, precentral/postcentral gyrus, angular gyrus, and superior occipital gyrus. At follow-up, no whole-brain FWE-corrected differences; at p<0.001 uncorrected, elevated activity to sad primes persisted in salience and visual networks, particularly in the relapse group.

Findings replicate a mood-congruent bias at the neural level: acute MDD shows elevated amygdala responses to subliminal sad faces and absence of the positive bias seen in HC. Critically, amygdala hyperreactivity to masked sad faces persists over two years regardless of relapse status or current mood, supporting a trait-like marker of depression for early, automatic emotion processing. This contrasts with studies of controlled (supraliminal) processing stages where limbic activity normalizes with remission, suggesting differential state-dependence across processing stages. The persistence could reflect either a consequence of depressive illness (scar) or a pre-existing vulnerability; trends with childhood maltreatment and genetic risk reported in prior literature support a vulnerability link. Whole-brain differences were less robust longitudinally and may be more state-sensitive, while amygdala responses remain persistently elevated. Medication effects likely contributed to increased happy-prime amygdala activity at follow-up, especially in the relapse group with higher medication loads, consistent with antidepressants shifting automatic processing toward positive valence. Behavioral measures were less sensitive than fMRI to subliminal biases, underscoring the value of neuroimaging for detecting automatic processing alterations.

This first long-term prospective fMRI study of automatic emotion processing in MDD demonstrates that amygdala hyperactivity to subliminal negative stimuli persists over two years and is independent of current mood or relapse, indicating a trait-like neural marker of depression. In contrast to controlled processing stages, early automatic biases appear enduring. Medication may enhance amygdala responses to positive stimuli over time, suggesting a potential mechanism for therapeutic modulation of bias. Future research should: (1) replicate in larger samples with broader severity spectra (including outpatient and first-episode cohorts), (2) incorporate at-risk and never-depressed populations to disentangle vulnerability vs consequence, (3) probe mechanisms linking environmental risk (e.g., childhood maltreatment) to automatic amygdala biases, (4) examine connectivity-level changes, and (5) evaluate targeted interventions that modify automatic processing biases and assess their prognostic significance.

Naturalistic design introduces uncontrolled factors (life events, prior illness course, heterogeneous treatments). Sample skewed to higher clinical severity (all inpatients at baseline; ~two-thirds relapsed or remained depressed at two years), limiting generalizability. Subgroup sizes (relapse vs no-relapse) are relatively small, reducing power for cross-sectional and predictive contrasts. Potential adaptation to the priming paradigm at follow-up may affect within-group condition effects. Whole-brain analyses may have been underpowered/conservative at FWE thresholds. Long-term fMRI studies are challenging; replication in larger, more diverse cohorts is needed.