Brain structural network alterations related to serum cortisol levels in drug-naïve, first-episode major depressive disorder patients: a source-based morphometric study

Major depressive disorder (MDD) is a prevalent and disabling condition linked to functional impairment, self-harm, and suicide risk. Endocrine dysregulation, particularly abnormalities involving the hypothalamic-pituitary-adrenal (HPA) axis and other hormonal systems (thyroid, gonadal hormones, oxytocin/vasopressin), has been implicated in MDD pathophysiology. Elevated cortisol, a hallmark of HPA axis hyperactivity, has been associated with brain structural alterations, including reduced hippocampal volumes and orbitofrontal cortical thinning. Traditional univariate morphometric methods such as voxel-based morphometry (VBM) evaluate voxel-wise differences but do not capture inter-voxel covariance patterns. Diffusion tensor imaging reveals white matter changes related to cortisol but offers limited insight into interregional gray matter network organization. Source-based morphometry (SBM), a multivariate, data-driven approach using independent component analysis, can identify structural covariance networks without a priori assumptions. This study asked whether SBM can detect gray matter network alterations in first-episode, drug-naïve MDD and whether these network alterations are associated with morning serum cortisol levels, thereby clarifying network-level structural correlates of HPA axis dysregulation early in the illness course.

Prior research shows HPA axis alterations in MDD, including cortisol hypersecretion, reduced glucocorticoid receptor (GR) expression, and impaired negative feedback. Animal studies indicate glucocorticoids increase excitotoxic injury and impair neuroplasticity in GR-rich regions. Meta-analytic and neuroimaging evidence links higher cortisol with smaller hippocampal volumes and orbitofrontal cortical thinning, particularly in late-life depression. DTI studies in MDD have related higher cortisol to white matter disruptions (inferior fronto-occipital fasciculus, uncinate fasciculus, anterior thalamic radiation). SBM has identified structural covariance abnormalities in other neuropsychiatric conditions (schizophrenia, autism, criminal offenders, delusional infestation) but had not been applied to MDD prior to this study. Theoretical frameworks emphasize distributed circuit dysfunctions in depression (default mode, positive affect, frontoparietal attention, cognitive control), with the prefrontal cortex as a key hub. These literatures motivated examining network-level gray matter changes and their association with cortisol in early, untreated MDD.

Design and participants: Cross-sectional study of 42 first-episode, drug-naïve MDD patients (21 males, 21 females; mean age 48.1 ± 14.3 years) diagnosed via Structured Clinical Interview for DSM-IV-TR. Exclusion criteria: any past DSM-IV-TR Axis I disorder, medical illnesses, neurological disorders, or use of drugs that may cause depression. Thirty-nine healthy subjects (HS; 26 males, 13 females; mean age 43.3 ± 11.6 years) were recruited via advertisement and screened with SCID-I/NP; no personal or first-degree family history of major psychiatric/neurologic illness. Serum cortisol assay: Morning blood samples were drawn 1 hour after awakening (9–10 AM) to capture peak cortisol. Serum was separated and stored at −20°C. Cortisol was quantified by direct radioimmunoassay after displacement from binding proteins using a highly specific antibody. MRI acquisition: All participants underwent MRI prior to any treatment in patients. Scans were acquired on a 3.0 T GE Signa EXCITE with 8-channel head coil using 3D fast spoiled gradient-recalled acquisition (3D-FSPGR). Parameters: TR/TE/TI = 10/4.1/700 ms; flip angle 10°; FOV 24 cm; slice thickness 1.2 mm; resolution 0.9 × 0.9 × 1.2 mm. Images were corrected for gradient nonlinearity (Grad Warp) and intensity nonuniformity (N3). A neuroradiologist confirmed absence of gross abnormalities. Preprocessing and VBM: Using SPM12, T1 images were normalized, segmented (GM/WM/CSF), and modulated with DARTEL; GM images smoothed with 8-mm FWHM Gaussian kernel. Whole-brain two-sample t-tests compared GM volume between groups with covariates (age, sex, total intracranial volume); voxel-level p < 0.001, cluster-level FWE-corrected p < 0.05. SBM processing: Gray matter images were analyzed with GIFT toolbox. The minimum description length criterion estimated 17 independent components (ICs). ICA (Infomax) was run and repeated 20 times with ICASSO to ensure stability; all extracted components had quality index Iq > 0.97. Artifact ICs (edge/extra-GM) were removed; from remaining 14 ICs, 10 ICs relevant to depression/anxiety were retained (cerebellar-dominant ICs excluded). Source maps were Z-scaled and thresholded at Z > 2.5. The mixing matrix provided subject loading coefficients for each IC. Statistics: Group comparisons for demographics used two-tailed t-test (age), χ2 (sex), and Mann–Whitney U (cortisol). For VBM, FWE-corrected whole-brain inference as above. For SBM, two-tailed t-tests compared loading coefficients (Fisher Z-transformed) between groups per component; Spearman correlations assessed associations between cortisol and loading coefficients within groups. Bonferroni correction applied; significance at corrected p < 0.05. Analyses used EZR (graphical interface for R). Ethics approval obtained; all participants gave written informed consent.

- Sample: 42 MDD, 39 HS; no significant group differences in age (p = 0.10) or sex (p = 0.20). - Serum cortisol: Higher in MDD vs HS (mean ± SD: 12.3 ± 5.1 vs 9.6 ± 3.6 nmol/L; p = 0.03). - VBM: No significant group differences in regional GM volume at FWE-corrected thresholds. - SBM components: 17 ICs estimated; after artifact removal and selection, 10 GM networks analyzed. Two components showed significant group differences after Bonferroni correction: • Prefrontal network: Loading coefficients lower in MDD (−0.25 ± 0.71) vs HS (0.24 ± 0.76); p < 0.01; Cohen’s d = 0.68. • Insula-temporal network: Loading coefficients lower in MDD (−0.27 ± 0.80) vs HS (0.39 ± 0.74); p < 0.01; Cohen’s d = 0.84. - Cortisol-network associations (MDD group): • Prefrontal network: significant negative correlation with serum cortisol (r = −0.354, p = 0.0216; significant after Bonferroni correction). • Insula-temporal network: negative trend, not significant (r = −0.294, p = 0.0588). - No significant cortisol correlations in HS.

The study addressed whether multivariate SBM detects gray matter network abnormalities in first-episode, drug-naïve MDD and whether these alterations relate to HPA axis activity. Two structural covariance networks—the prefrontal and insula-temporal networks—showed reduced loading in MDD versus controls, indicating altered GM covariance patterns early in the illness before treatment exposure. Crucially, only the prefrontal network exhibited a significant inverse association with morning serum cortisol levels in MDD, linking HPA axis hyperactivity to reduced prefrontal network integrity. This aligns with models positing prefrontal circuit dysfunction in depression (involving default mode, reward/positive affect, attention, and cognitive control networks) and with evidence that glucocorticoid excess can impair neuroplasticity in GR-rich cortices. As anatomical covariance may reflect functional connectivity, the SBM findings suggest that structural network vulnerabilities, particularly in prefrontal regions, may be a proximal correlate of elevated cortisol in early MDD, supporting the prefrontal cortex as a target of cortisol’s negative-feedback effects. The absence of significant VBM findings underscores SBM’s sensitivity to distributed network-level alterations that voxel-wise univariate approaches may miss.

SBM revealed altered gray matter network patterns in first-episode, drug-naïve MDD, specifically reduced loading in prefrontal and insula-temporal networks, with only the prefrontal network showing a significant negative association with morning serum cortisol. These results support the prefrontal cortex as a target site for cortisol’s negative-feedback effects on the HPA axis in early MDD and demonstrate SBM’s utility as a multivariate alternative to univariate morphometric analyses for detecting network-level structural abnormalities.

- Modest sample size may limit statistical power and generalizability and introduces potential sampling bias. - Although SBM leverages multivariate covariance and showed high component stability, larger datasets are preferable to maximize robustness. - Cross-sectional design precludes causal inference. - Sample restricted to first-episode, drug-naïve patients enhances internal validity by avoiding treatment effects but may limit generalizability to chronic or treated populations. - Cortisol measured once in the morning; diurnal variability and multiple sampling were not assessed.