Intracranial directed connectivity links subregions of the prefrontal cortex to major depression

Major depressive disorder (MDD) is highly prevalent and its neural basis remains elusive. Symptom severity fluctuates over hours to days, yet neural correlates of these temporal dynamics are underexplored. Prior intracranial EEG studies decoded mood from orbitofrontal and cingulo-limbic activity but did not directly examine directed connectivity or focus on MDD specifically. Neuroimaging work implicates widespread network dysfunction, including hyperconnectivity in the default mode network and hypoconnectivity in frontoparietal attention networks, with altered coupling among prefrontal subregions (dlPFC, OFC, ACC). Electrophysiological and molecular evidence points to compromised GABAergic inhibition in MDD, potentially reflected in delta-band oscillations prominent in frontal cortex. The authors hypothesized that delta-band directed connectivity among PFC subregions, measured via Granger causality, would track transient increases in depression severity and reveal hemispheric imbalances related to mood.

Longitudinal imaging studies have linked MDD severity to structural and functional changes, including increased brainstem white matter volume and decreased prefrontal–limbic connectivity. MDD commonly shows hyperconnectivity in default mode and hypoconnectivity in dorsal attention networks, with reduced coupling between right dlPFC and right ACC. Lesion and stimulation studies implicate OFC in emotion regulation; OFC lesions blunt responses to emotional stimuli and increase MDD risk, while OFC stimulation can acutely improve mood. dlPFC supports executive control and working memory, coactivating with limbic structures in affective tasks; ACC is central to emotional response inhibition. fMRI-based connectivity is challenging to interpret causally due to BOLD–neural oscillation relationships, which are more consistent with gamma power and less so with low-frequency oscillations. Electrophysiology and transcriptomics identify GABAergic deficits—particularly somatostatin- and parvalbumin-positive interneurons—linked to delta oscillations (~1–3 Hz) and cortical inhibition. Delta rhythms are prominent in frontal cortex, relate to wakeful rest and sleep, and increase during pain and panic. These findings motivate targeting delta-band directed connectivity across OFC, dlPFC, and ACC to understand dynamic symptom fluctuations in MDD.

Participants: Six patients with treatment-resistant depression (TRD; 3 male, 3 female; mean age 44.33 years) enrolled in an NIH BRAIN Initiative clinical trial (UH3 NS103549) received therapeutic DBS leads in ventral capsule/ventral striatum and subcallosal cingulate and temporary stereo-EEG (sEEG) electrodes in bilateral dlPFC (BA 10/46/47), OFC (BA 11), and ACC (BA 24/32). All provided informed consent; none were in remission at recording. Symptom assessment: Computerized Adaptive Test for Depression Inventory (CAT-DI) administered 1–2 times/day over 9–11 days (n=78 assessments; ~12 items; ~3 minutes; test–retest reliability r ≥ 0.90), with mean severity 61.8 (SD 9.18), range 27–86. Surveys preceded recordings by ~24 minutes. Methods evaluated opioid use effects; ANOVA showed no effect on CAT-DI (F(1,65)=0.37, p=0.547). Intracranial recordings: Resting-state iEEG (eyes on fixation cross) for 5 minutes, most often mid-afternoon. Signals recorded at 2 kHz (Blackrock Cerebus), online bandpass 0.3–500 Hz (Butterworth). Artifact rejection by visual inspection and automatic thresholding (3× SD of spectral power 1–50 Hz) prior to any symptom analyses; noisy channels excluded. Notch filtering (60 Hz and harmonics) and bipolar re-referencing to mitigate common signals and protect Granger causality. Electrode localization confirmed via fused preoperative MRI/postoperative CT; one gray matter contact per region per hemisphere selected (Area 10/46 dlPFC, Area 11 OFC, Area 24/32 ACC) by expert consensus. Electrode coordinates were statistically similar across hemispheres. Due to trial procedures, 70.5% of recordings occurred within ~2.63 h (±18 min) of DBS stimulation experiments to subgenual cingulate and ventral striatum/ventral capsule; sEEG contacts in ACC, dlPFC, OFC were not stimulated. Directed connectivity: Multivariate vector autoregressive (MVAR) modeling with the MVGC MATLAB toolbox quantified pairwise-conditional spectral Granger causality among PFC subregions. Model order selected via minimum Akaike Information Criterion using up to 250 ms (order ≤ 500). Coefficients estimated with locally weighted regression; stationarity verified by spectral radius < 1. Cross-power spectral density computed via Fourier transform of autocovariance and factored through transfer function to obtain stable frequency-domain GC. Primary analyses focused on delta band (1–3 Hz), with additional analyses in theta (4–7 Hz), alpha (8–14 Hz), beta (15–30 Hz), gamma (31–50 Hz). Power spectral density (PSD): Morlet wavelet convolution (6 cycles) from 1–50 Hz in 0.1 Hz steps; delta-band PSD (1–3 Hz) averaged within each ROI. Statistics: Generalized estimating equations (GEEs) modeled relationships between symptom severity (CAT-DI) and directed connectivity terms (R1→R2, R2→R1, and bidirectional interaction R1↔R2), accounting for within-subject autocorrelation using AR(1)-type estimation and robust fitting (GEEQBOX). Hemispheres analyzed separately to conserve rank and test within-hemisphere interactions; hemispheric differentials computed as right-minus-left directed connectivity or PSD and analyzed via GEE. Pairwise t-tests compared connectivity magnitudes and PSD across hemispheres. p-values FDR-corrected across multiple comparisons (connections, frequencies, hemispheres). Granger causality significance also assessed via Geweke’s chi-square; all sessions showed GC > 0 (p < 0.05). Reliability: Intra-class correlation (ICC; Cronbach’s alpha) assessed test–retest consistency across GC (mean ICC 0.63) and PSD (mean ICC 0.94); GC ICC increased with frequency in left hemisphere (r=0.92, p=0.026).

Delta-band directed connectivity across PFC subregions increased with depression severity. Left hemisphere: OFC_left → dlPFC_left positively correlated with symptom severity (z(74)=3.82, p<0.001); dlPFC_left → OFC_left also positive (z(74)=3.14, p=0.002). Bidirectional interaction between these pathways was significant, indicating highest severity when both directions increased (z(74)=-4.97, p<0.001). dlPFC_left → ACC_left and ACC_left → dlPFC_left were positively correlated (z(74)=2.66, p=0.008; z(74)=2.31, p=0.020), with a significant interaction (z(74)=-2.32, p=0.020). OFC_left ↔ ACC_left showed no significant effects (z(74)=-1.41, p=0.158). Right hemisphere: OFC_right → ACC_right positively correlated (z(74)=4.18, p<0.001), with a marginal bidirectional interaction (z(74)=-1.92, p=0.055); homologous left-direction showed no effect (z(74)=1.08, p=0.279). Hemispheric magnitude comparisons: All ACC-involving pathways were stronger in the left hemisphere (ACC→dlPFC: t(77)=6.20, p<0.001; dlPFC→ACC: t(77)=4.21, p<0.001; OFC→ACC: t(77)=2.52, p=0.014; ACC→OFC: t(77)=3.06, p=0.005). OFC↔dlPFC communication was stronger in the right hemisphere (OFC→dlPFC: t(77)=-2.66, p=0.011; dlPFC→OFC: t(77)=-3.23, p=0.003). Hemispheric differentials: Symptom severity increased when right OFC→dlPFC exceeded left (z(74)=3.69, p<0.001); severity decreased when left OFC→ACC exceeded right (z(74)=-4.96, p<0.001); severity increased with left-bias in dlPFC→ACC (z(74)=-2.06, p=0.039). PSD: Delta power in right OFC (z(71)=2.90, p=0.004) and right dlPFC (z(71)=2.01, p=0.045) positively correlated with severity; right OFC–right dlPFC spectral interaction positive (z(71)=2.23, p=0.025). Right ACC had greater delta power than left (mean 17.99 vs 17.71 dB; t(77)=-5.23, p<0.001). A right-minus-left OFC power differential predicted higher severity (z(74)=2.05, p=0.040). Higher-frequency connectivity: Left hemisphere—ACC_left → OFC_left theta negatively correlated (z(74)=-2.86, p<0.001); OFC_left → ACC_left beta (z(74)=-3.02, p=0.036) and gamma (z(74)=-3.60, p<0.001) negatively correlated, consistent with improved mood. Right hemisphere—OFC_right → ACC_right theta positively correlated (z(74)=6.40, p<0.001) while gamma negatively correlated (z(74)=-3.53, p<0.001); dIPFC_right → OFC_right theta positively correlated (p=0.048); ACC_right → dIPFC_right gamma positively correlated (z(74)=2.69, p=0.048). Alpha-band connectivity showed no relationships (p>0.50). Across bands, delta connectivity was more positively correlated with severity than gamma (delta mean z=1.59±0.49 vs gamma mean z=-0.59±0.56; t(22)=2.07, p=0.008).

Transient increases in MDD symptom severity were associated with heightened directed connectivity among PFC subregions, challenging the common narrative of prefrontal hypoactivation and instead supporting models of diminished cortical inhibition and dysregulated network communication. Delta-band effects align with the role of GABAergic systems in frontal cortex and depression. OFC-to-dlPFC and OFC-to-ACC pathways appear critical during negative mood, potentially reflecting increased demand for inhibitory and cognitive control during rumination. Hemispheric specialization emerged: ACC-related communication was stronger and more symptom-linked in the left PFC, while OFC–dlPFC interactions were stronger and more symptom-linked in the right PFC. Imbalances in interhemispheric connectivity, especially right-biased OFC→dlPFC, tracked higher severity, suggesting that functional lateralization contributes to pathophysiology. Higher-frequency effects revealed an oscillatory hierarchy, with theta and gamma showing opposing correlations depending on direction and hemisphere; notably, increased left ACC→OFC theta and left OFC→ACC beta/gamma connectivity tracked lower severity, consistent with improved emotion regulation. These findings have implications for neuromodulatory therapies: rTMS benefits targeting dlPFC may leverage its bidirectional coupling with ACC, particularly in the left hemisphere; DBS of cingulate circuits may be optimized by promoting beneficial high-frequency prefrontal projections while reducing low-frequency connectivity associated with worse mood. Directed connectivity biomarkers could inform closed-loop strategies or clinician-in-the-loop monitoring to tailor stimulation to favorable network states.

Intracranial iEEG in TRD patients demonstrates that delta-band directed connectivity among OFC, dlPFC, and ACC increases with transient worsening of depression, and that hemispheric imbalances within prefrontal networks track symptom severity. Left ACC-centered communication and right OFC–dlPFC coupling play distinct roles, with right-biased OFC→dlPFC connectivity associated with higher severity. High-frequency (beta/gamma) directed connectivity between left OFC and ACC is inversely related to severity, suggesting potential therapeutic targets. These results advance mechanistic understanding of MDD as a disorder of network disinhibition and lateralized prefrontal dysfunction and provide testable biomarkers for neuromodulation. Future work should examine interhemispheric PFC interactions via corpus callosum, larger limbic–cortical networks, and cross-frequency coupling metrics; and evaluate closed-loop neuromodulation protocols that enhance beneficial high-frequency prefrontal projections while suppressing maladaptive low-frequency connectivity.

Small sample size (n=6) of patients with severe, treatment-resistant depression limits generalizability. Recordings were restricted to selected contacts within dlPFC, OFC, and ACC, potentially omitting broader network influences. Resting-state sessions were brief (5 minutes) and many occurred within hours of DBS stimulation elsewhere in the brain (though PFC contacts were not stimulated), introducing possible state-related confounds. Granger causality estimates showed lower test–retest reliability than PSD, reflecting complexity of directed measures. Analyses separated hemispheres to conserve statistical rank, reducing the ability to model full interhemispheric interactions directly. Data and code are not publicly available due to ongoing clinical trial constraints, limiting external replication.