Real-world stress resilience is associated with the responsivity of the locus coeruleus

The study investigates why some individuals are resilient to prolonged real-life stress while others develop increased anxiety or depression. Prior animal work implicates hyper-responsivity of the locus coeruleus–noradrenergic (LC-NE) system and its coupling with the amygdala in stress vulnerability. Human studies often measure brain activity after stress exposure, limiting prediction of vulnerability. The authors prospectively test whether individual LC-NE responsivity and LC–amygdala coupling measured during an emotional conflict task predict subsequent changes in anxiety and depression during a stressful medical internship, aiming to identify neurobiological markers of stress resilience beyond self-report.

Animal research links LC-NE hyperactivity to anxiety, depression, PTSD, and cardiovascular risks; LC projections to amygdala promote anxiety-like behavior in rodents. Human imaging shows LC involvement in conflict resolution and unexpected uncertainty but prior prospective links to stress outcomes are scarce. Oddball-induced LC responses did not predict anxiety/depression in earlier work, whereas conflict tasks robustly engage arousal mechanisms and pupil-linked arousal. Pupil dilation correlates with LC activity but also reflects cholinergic influences. Thus, LC-NE function during emotional conflict, LC–amygdala coupling, and pupil responses are plausible biomarkers to forecast stress-related psychopathology.

Design: Prospective cohort of medical students before their first emergency-room internship, followed over 6 months. Participants: N=48 (28 women; mean age 24±1.99 years) recruited from 200; inclusion/exclusion followed standard MRI safety and psychopathology criteria. Ethics approved; consent obtained. Timeline: Baseline (t0, pre-internship): fMRI with emotional-Stroop task, simultaneous pupillometry, and questionnaires. Follow-ups at 3 months (t1) and 6 months (t2) during internship: questionnaires repeated. Self-report measures: State-Trait Anxiety Inventory (STAI) for anxiety and PHQ-9 for depression at t0, t1, t2; pre-trauma history via adapted Trauma Checklist (PDS). Symptom change scores computed as t1–t0 and t2–t0; also mean change across t1 and t2. Task: Emotional Stroop. Participants categorized facial expression (happy/fearful) while ignoring overlaid congruent/incongruent words ("HAPPY"/"FEAR"). Conflict-sequence manipulation: current incongruent trials preceded by congruent (CI) vs. preceded by incongruent (II). The CI>II contrast isolates conflict-induced upregulation of control with identical stimuli/responses. fMRI acquisition/analysis: 3T scanner; whole-brain EPI with 2.5 mm isotropic voxels (standard sequence). Preprocessing and GLM in SPM8; regressors for trial types (CI, II, etc.), eye movements/blinks, motion; AR(1) autocorrelation model. Physiological noise control via PCA components from individual CSF masks (first 5 PCs) added as nuisance regressors; analyses performed on smoothed and unsmoothed data, with emphasis on physio-corrected unsmoothed data for brainstem specificity. LC ROI defined using standardized probabilistic masks (Keren et al.) with 1SD and 2SD variants; weighted-average extraction across LC-mask voxels. Additional ROIs for brainstem nuclei (MR, DR, VTA, SN) and amygdala for specificity comparisons. Whole-brain and brainstem temporal SNR quantified. Functional coupling: Psychophysiological interaction (PPI) with LC seed (5 mm sphere around subject-specific peak from leave-one-subject-out, LOSO). Contrast CI>II for LC–amygdala coupling; amygdala ROI tested with small-volume correction; exploratory whole-brain also performed. Pupillometry: EyeLink 1000 (250 Hz) during fMRI; preprocessing included interpolation of signal loss, high-pass 0.05 Hz and low-pass 4 Hz filtering, z-scoring per run. Trial-locked epochs ±5 s. Condition averages for I, C, CI, II, CC, IC. Cluster-based permutation tests marked significant time windows. Defined pre-trial window (−3044 to −1222 ms) and current-trial window (1530 to 4862 ms) for CI–II effects. Pupil dilation distance (PDD) computed as |(CI–II)current − (CI–II)pre| per subject. Behavioral analyses: Trial-wise regressions on congruency and sequence for RT/accuracy to confirm classic Stroop and sequence effects. Prediction and validation: Correlational analyses relating LC CI>II responsivity, LC–amygdala coupling, and PDD to symptom change at t1, t2, and the mean. Robust regressions and Spearman's rho reported. Out-of-sample predictions: (1) Leave-one-subject-out linear models predicting left-out participant’s mean symptom change; (2) Leave-two-subjects-out (LTSO) classification determining which of two left-out participants has higher symptom change; significance via permutation (1000 permutations per pair). Model comparisons via GLMs: base model (t0 symptom + pre-trauma), addition of behavioral CSE, LC CI>II, PDD, and LC–amygdala coupling; stepwise regression to identify parsimonious model. Performance metrics: R2, adjusted R2, AIC/BIC, ROC/AUC, LTSO accuracy. Code/data availability: GitHub repository with data and MATLAB code.

- Emotional conflict and sequence effects replicated behaviorally; CI>II contrast engaged LC and other regions. - LC activation: CI>II revealed robust midbrain/brainstem activity (cluster extent=170; df=47; non-parametric P(FWE)=0.039; peak X/Y/Z: 6/−27/−10). ROI in LC showed significant CI>II responses (LC-right: P(SVC)=0.003; X/Y/Z: 6/−37/−28). Participants with larger subsequent symptom changes showed stronger LC responsivity than those with smaller changes (median split): anxiety p=0.019 (T=2.431); depression p=0.037 (T=2.154). - Prospective correlations (LC CI>II vs. symptom change): • Anxiety: t1 rho=0.30, p=0.018; t2 rho=0.31, p=0.002; mean change rho=0.30, p=0.002. • Depression: t1 rho=0.38, p=0.004; t2 rho=0.26, p=0.034; mean change rho=0.36, p=0.006. Smaller LC upregulation predicted less symptom increase (greater resilience). - Specificity/robustness: Associations strongest for LC compared to nearby brainstem nuclei (MR, DR, VTA, SN) and amygdala activity; effects robust across LC masks (1SD, 2SD) and pipelines, strongest in physio-corrected unsmoothed data. tSNR in LC >30 and highest among brainstem nuclei. - Out-of-sample predictions using LC CI>II: • LOSO regression: predicted vs observed mean change: anxiety rho=0.25, p=0.01; depression rho=0.28, p=0.05. • LTSO classification accuracy: anxiety 60.3% (p<0.001); depression 59.4% (p<0.001). - LC–amygdala coupling (PPI CI>II) related to mean symptom changes: • Anxiety: P(SVC)<0.001; peak −25/1/−23; T=6.72, Z=5.58; LTSO 56.3%, p=0.002. • Depression: P(SVC)<0.001; peak 31/−2/−18; T=3.45, Z=3.24; LTSO 56.0%, p<0.001. - Pupillometry: • Incongruent > congruent pupil dilation significant 945–3668 ms post-stimulus (cluster-corrected p<0.05). • Sequence effects: pre-trial pupil larger for II than CI (−3044 to −1222 ms); current-trial pupil smaller for II vs CI (1530–4862 ms), indicating carry-over/upregulation. • Pre-trial CI–II pupil difference correlated negatively with current-trial CI–II pupil (R=−0.52, p=0.00014) and with LC CI>II BOLD (R=−0.30, p=0.038). • Pupil dilation distance (PDD) correlated with mean symptom changes: anxiety R=0.36, p=0.013 (LTSO 52.6%, p=0.09), depression R=0.30, p=0.04 (LTSO 55.3%, p<0.001). - Model comparisons (anxiety): Base model failed (accuracy 51.86%, p=0.234; AUC=0.558). Full model with LC, pupil, CSE, LC–amygdala improved adjusted R2 to ~0.50 and LTSO accuracy 58.7% (p<0.001; AUC=0.685). Optimal sparse model (LC, CSE, pupil, LC–amygdala) adjusted R2=0.518; accuracy 59.2% (p<0.001; AUC=0.675). LC-only AUC=0.673; LC–amygdala-only AUC=0.624; pupil-only AUC=0.521. - Model comparisons (depression): Base model already predictive (accuracy 67.38%, p<0.001; adjusted R2=0.23; AUC=0.754). Adding LC CI>II improved variance explained (adjusted R2≈0.271; model 3 p=0.046). Full model accuracy 64.36% (p<0.001; AUC=0.742). Optimal model with LC CI>II and baseline PHQ achieved adjusted R2=0.30; accuracy 67.7% (p<0.001; AUC=0.779). LC-only AUC=0.705; LC–amygdala-only AUC=0.535; pupil-only AUC=0.614.

The findings demonstrate that lower responsivity of the LC-NE system during conflict-induced upregulation predicts greater resilience to prolonged real-life stress, whereas higher responsivity predicts increased anxiety and depression symptoms. This provides a mechanistic, prospective biomarker beyond self-report, aligning with animal literature on LC hyperactivity and LC–amygdala pathways in stress-induced anxiety. LC–amygdala functional coupling further relates to symptom changes, suggesting a pathway by which noradrenergic hyper-reactivity may elevate amygdala-driven threat perception, contributing to anxiety and depression. Pupillometry corroborates a putative noradrenergic mechanism in conflict generation and adaptation and offers a portable physiological index, particularly informative for depression predictions. Collectively, the results address key gaps by prospectively linking LC-NE function to real-world stress outcomes with cross-validated predictive utility, supporting the LC-NE system as a target for monitoring and preventive interventions (e.g., regulation training, neurofeedback).

Prospectively measured LC-NE activation during conflict upregulation and LC–amygdala coupling predict subsequent anxiety and depression symptom changes during a stressful medical internship, establishing LC responsivity as a robust biomarker of stress resilience. Pupil-based measures provide complementary predictive value, especially for depression. These biomarkers surpass or complement standard self-reports in predictive accuracy. Future work should leverage higher-resolution brainstem imaging and neuromelanin localization for LC specificity, disentangle neuromodulatory contributions to pupil signals, test larger and more diverse populations, incorporate fuller resilience constructs (e.g., coping strategies, trajectories), and assess LC responsivity longitudinally during stress. Connectivity-based neurofeedback may offer avenues to modulate LC–amygdala dynamics for prevention and intervention.

- Brainstem imaging challenges: LC is small, near ventricles, with low SNR. Although analyses used CSF-based physiological noise correction, unsmoothed data, probabilistic LC masks, and tSNR checks, specialized high-field, high-resolution, neuromelanin-sensitive, and partial-coverage protocols with cardio monitoring would enhance LC specificity. - Pupil specificity: Pupil dilation reflects noradrenergic and cholinergic activity; it is not a specific NE marker. Future studies should parse contributions of multiple neuromodulatory systems. - Sample/generalizability: Modest N=48 Swiss medical students with relatively low symptom levels compared to some cohorts; working conditions differ internationally. Resilience is multifaceted; future work should include richer constructs (e.g., coping, variability, latent trajectories) and repeated LC assessments before and during stress exposure.