Differential increase of hippocampal subfield volume after socio-affective mental training relates to reductions in diurnal cortisol

The study investigates whether distinct contemplative mental training practices differentially affect hippocampal subfield structure and function, and how these changes relate to stress regulation indexed by diurnal and chronic cortisol. Stress-related disorders are highly prevalent, and contemplative training programs have shown promise for stress reduction and brain plasticity, but the most effective practice types and their neurobiological correlates remain unclear. The hippocampus, rich in glucocorticoid receptors and central to HPA-axis negative feedback, is a prime candidate region for stress-related plasticity. Prior findings on relationships between cortisol indices (e.g., cortisol awakening response, diurnal slope, total output) and hippocampal measures are mixed, often cross-sectional, and with small samples. The ReSource Project offers a longitudinal, large-sample framework to test how attention-based mindfulness (Presence), socio-affective (Affect: compassion, care), and socio-cognitive (Perspective: mentalizing, perspective-taking) training impact hippocampal subfield volumes and intrinsic functional connectivity, and how these changes associate with diurnal and hair cortisol as stress markers.

Prior work links hippocampal integrity with stress and HPA-axis regulation, noting high hippocampal glucocorticoid receptor density and its inhibitory role on cortisol release. Studies have associated altered diurnal cortisol indices (e.g., CAR, total output) with hippocampal volume differences across healthy and clinical populations, though results are inconsistent and often cross-sectional. Some evidence suggests smaller hippocampal volume relates to blunted CAR and elevated cortisol to hippocampal atrophy in aging or psychiatric samples. Resting-state functional connectivity changes in hippocampal networks have also been associated with cortisol variations. Mindfulness-based interventions (e.g., MBSR) report reductions in CAR and evening cortisol, but meta-analyses reveal mixed effects likely due to heterogeneous practices and modest samples. In the ReSource cohort, socio-affective training reduced CAR and both socio-affective and socio-cognitive training reduced acute stress reactivity; hair cortisol decreased across modules, suggesting different cortisol indices may respond differently to training content. Hippocampal subfields (SUB, CA1-3, CA4/DG) have distinct microstructure and connectivity patterns, motivating subfield-specific analysis.

Design: Longitudinal, randomized cohort study (ReSource Project) with 332 healthy adults assessed over 9 months. Two training cohorts (TC1, N=80; TC2, N=81) completed three sequential 3-month modules: Presence (attention/interoception), Affect (compassion/care, socio-emotional skills), and Perspective (socio-cognitive, mentalizing), with module order counterbalanced (TC1: Presence→Affect→Perspective; TC2: Presence→Perspective→Affect). A third cohort (TC3, N=81) completed 3 months Affect only. A matched retest control cohort (RCC, N=90) underwent no training. Assessments at baseline (T0) and after each module (T1, T2, T3). Participants: Healthy adults (20–55 years; mean ~40.7 years; 197 women), screened for psychiatric/neurological conditions; exclusions included recent Axis I disorders, severe psychiatric conditions, HPA-axis-influencing medications. Ethics approvals obtained; trial registered (NCT01833104). Training: Daily exercises (~30 min, five days/week) including core meditations (Breathing Meditation, Body Scan; Loving-kindness Meditation, Affect Dyad; Observing-thoughts Meditation, Perspective Dyad) and weekly group sessions. MRI acquisition: 3T Siemens Verio, 32-channel coil. Structural T1-weighted MPRAGE (1 mm isotropic; TR 2300 ms, TE 2.98 ms, TI 900 ms). Resting-state fMRI T2* EPI (3 mm isotropic, TR 2000 ms, TE 27 ms, 210 volumes), eyes-open fixation. Hippocampal subfield volumetry: Automated patch-based surface segmentation (Caldairou et al., 2016) of SUB, CA1-3, CA4/DG on T1w images registered to MNI152 (implicit ICV control). Quality control by blinded raters; reprocessing/exclusion for poor segmentations. Resting-state processing: DPARSF/REST pipeline with discarding first 5 volumes, slice timing, motion correction, coregistration to T1, MNI normalization, nuisance regression (WM/CSF signals, 6 motion parameters), scrubbing for framewise displacement ≥0.5 mm. Seed-based functional connectivity from left/right subfields to Schaefer-400 cortical parcels; Fisher r-to-z transformation and rescaling (0–1). Functional networks defined as top 10% baseline ipsilateral connections per subfield; individuals with mean FD >0.3 mm excluded. Diurnal cortisol: Salivary cortisol collected 7 times/day over 2 consecutive weekdays (awakening, +30, +60, +240, +360, +480, +600 min). Mobile device reminders to enhance adherence; samples assayed via time-resolved fluorescence immunoassay; log-transformed and winsorized (±3 SD). Indices computed: CAR (peak at 30 or 60 min vs awakening), diurnal slope (awakening to 600 min), total daily output AUCg (awakening, 240, 360, 480, 600 min; excluding CAR samples). Hair cortisol/cortisone: 3 cm posterior vertex hair segments per timepoint (reflecting prior 3 months), LC-MS/MS assay; subset N≈44 across modules. Statistical analysis: Mixed-effects models for longitudinal subfield volume and functional connectivity changes (SurfStat), correcting for age, sex, and random subject effects. Main contrasts: Presence vs active control (Affect-only TC3; T0–T1) and Affect vs Perspective (within TC1/TC2; T1–T3); supplementary comparisons vs retest controls and within groups. Multiple comparisons controlled using FDR/Bonferroni. Associations between brain changes and cortisol indices: univariate correlations and multivariate partial least squares (PLS) with 1000 permutations and 100 bootstraps; covariates regressed (age, sex, subject random effects).

• Bilateral CA1-3 volume increases after socio-affective Affect training compared to socio-cognitive Perspective training and retest controls; subiculum (SUB) and CA4/DG did not show similar group-level changes. • Affect vs Perspective: Left CA1-3 t=2.360, p=0.019 (FDR q>0.1), Cohen’s d=0.282; Right CA1-3 t=2.930, p=0.004 (FDR q=0.022), d=0.350. • Within-module changes: Affect showed subtle bilateral CA1-3 volume increases (Left t=2.495, p=0.013, q=0.08; Right t=2.374, p=0.018, q>0.1). Perspective showed subtle right CA1-3 decreases (Right t=−2.118, p=0.035, q>0.1).
• Functional connectivity: Right CA1-3 mean network showed differential change for Affect vs Perspective (t=2.420, p=0.016, q=0.032, d=0.289), attributable to decreased FC in Perspective (t=−2.012, p=0.045). Regionally, Affect vs Perspective showed increased connectivity between right CA1-3 and right mPFC (t=3.262, p=0.002, d=0.389) and decreased connectivity between left CA1-3 and left posterior insula (t=−3.097, p=0.003, d=−0.370).
• Associations with diurnal cortisol (Affect module): CA1-3 volume increases correlated with reductions in total diurnal cortisol output (AUCg): Left t=−2.237, p=0.028, q=0.056; Right t=−2.283, p=0.025, q=0.05. Left CA1-3 mean network change positively associated with diurnal slope (t=2.653, p=0.009, q=0.018) and AUCg (t=2.261, p=0.026, q=0.052).
• Hair cortisol (across modules, subset): Decreases in hair cortisol associated with increases in left CA1-3 volume (t=−2.574, p=0.011, q=0.022) and left CA1-3 functional network change (t=−2.700, p=0.008, q=0.016). Additional associations: Right CA4/DG volume (t=−3.138, p=0.002, q=0.01) and left SUB function (t=−2.890, p=0.005, q=0.03) with hair cortisol.
• Multivariate PLS: Significant latent components linking cortisol indices (especially slope and AUCg) with hippocampal plasticity. CA1-3-focused model: LC1 p<0.01 (67% stability), overall r=−0.20; strongest association in Affect (r=−0.27). All-subfield model: LC1 p<0.01 (71% stability), overall r=0.24; Affect r=0.30, Perspective r=0.26, Presence r=0.16. Results suggest inverse contributions of structure and function and highlight broader subfield involvement beyond CA1-3.

The findings demonstrate that socio-affective (compassion- and care-focused) mental training selectively increases CA1-3 hippocampal subfield volume and modulates its functional connectivity compared to socio-cognitive training or no training. These structural changes were linked to reductions in total diurnal cortisol output, consistent with the hippocampus’s inhibitory role in HPA-axis regulation. Functional connectivity changes in CA1-3—particularly increased coupling with mPFC and decreased coupling with posterior insula—align with cortical regions implicated in stress regulation, suggesting a network-level adaptation accompanying socio-affective training. While causal direction between reduced cortisol and hippocampal plasticity cannot be established, two interpretations are plausible: (1) Affect training reduces daily stress load and cortisol, leading to downstream hippocampal changes; (2) Affect training directly enhances CA1-3 integrity/function, improving HPA-axis inhibition and thereby lowering cortisol. Multimodal and multivariate analyses indicate that, beyond CA1-3, other subfields may contribute to stress-related changes, underscoring system-level hippocampal adaptations. The results integrate with prior ReSource findings showing module-specific effects on dynamic cortisol responses (CAR, acute reactivity) and general reductions in chronic stress (hair cortisol), and emphasize CA1-3 as a central locus in socio-emotional training-related stress reduction.

This longitudinal multimodal study links socio-affective mental training to increases in hippocampal CA1-3 volume, corresponding functional connectivity alterations, and reductions in diurnal cortisol. The results pinpoint CA1-3 as a key subfield in stress-related neuroplasticity following compassion-based practice while indicating contributions from other subfields in multivariate analyses. These insights may inform targeted interventions leveraging socio-emotional training to mitigate stress and advance models of hippocampal subfield roles in socio-emotional and stress processes. Future work should examine causal mechanisms, employ higher-resolution hippocampal mapping (e.g., unfolded surfaces, long-axis gradients), and incorporate computational multivariate models to detail subfield-specific plasticity and its relationship to diverse stress markers.

• Effect sizes for CA1-3 volumetric changes are small; functional connectivity differences are modest and partly driven by decreases during Perspective training.
• Hippocampal subfield segmentation on 1 mm isotropic T1w images and MNI normalization may over/underestimate subfield volumes or blur boundaries; more granular surface-based/unfolding approaches could improve specificity.
• Diurnal cortisol sampling lacked objective verification of sampling times; potential non-adherence may confound CAR and diurnal indices despite mobile reminders.
• The study design cannot determine causal direction between cortisol reductions and hippocampal changes.
• Not all cortisol indices (e.g., CAR, slope) showed consistent univariate associations with CA1-3 volume; sensitivity may vary across stress markers.
• Sex differences in baseline subfield volumes existed, though adjusting for intracranial volume and covariates did not alter main findings.