Fitness is positively associated with hippocampal formation subfield volumes in schizophrenia: a multiparametric magnetic resonance imaging study

The hippocampal formation (HF), comprising subfields such as CA1–CA4, dentate gyrus (DG), presubiculum, and subiculum, is critical for memory and cognition and shows subfield-specific vulnerability in schizophrenia, with early and pronounced atrophy in CA1, CA4/DG, and subiculum. These structural and functional abnormalities relate to cognitive deficits, especially verbal memory. Aerobic exercise is proposed to counteract neurodegeneration and promote neuroregeneration, with prior work in healthy populations showing positive associations between aerobic fitness and HF volume and functional connectivity, and mixed findings in schizophrenia. This study aimed to determine whether aerobic fitness is positively associated with HF subfield volumes and HF-related functional connectivity, and whether these associations mediate verbal memory performance in individuals with schizophrenia.

Prior neuroimaging work indicates HF atrophy and functional abnormalities in schizophrenia, with subfield-specific reductions (notably CA1, CA4/DG, and subiculum) even early in illness and progression with duration. Cognitive studies link HF abnormalities to deficits in verbal memory. In healthy populations, meta-analyses and sMRI studies show aerobic exercise and higher fitness levels are associated with preserved or increased HF volume and enhanced HF functional connectivity with regions such as parahippocampal, middle frontal, and cingulate gyri, often alongside cognitive benefits. In schizophrenia, evidence is limited and inconsistent: an RCT reported improved verbal memory after aerobic exercise without group-level increases in HF or subfield volumes, though post hoc analyses suggested effects in CA3/CA4 for a subset. Few studies have integrated rs-fMRI or examined subfield-specific effects or cognitive mediation, highlighting a gap addressed by the present multiparametric design.

Design and participants: Cross-sectional baseline analysis from the ESPRIT C3 multicenter RCT; only the Munich site (with submillimeter MRI) was included. Fifty-three participants with DSM-IV schizophrenia completed aerobic fitness assessments and MRI; sMRI analyses included n=48 (after quality-based exclusions), rs-fMRI analyses included n=44. Ethics approval and informed consent were obtained. Aerobic fitness: Conducted a lactate threshold test during bicycle ergometry. Identified each participant’s aerobic threshold (approx. 1.8–2.5 mmol/L lactate) and computed aerobic fitness as wattage at threshold divided by body weight (W/kg). MRI acquisition and preprocessing: 3T Siemens Skyra; T1-weighted 3D MPRAGE and two resting-state EPI sequences. Structural preprocessing with FreeSurfer v7.2; functional preprocessing with fMRIPrep including ICA-AROMA, smoothing (6 mm FWHM), and confound regression (global signal, CSF, white matter, ICA-AROMA components). Quality control used MRIQC and VisualQC. MRI post-processing: sMRI—segmented 38 HF subfields (19 per hemisphere) with FreeSurfer’s hippocampal module; volumes were ICV-corrected (proportions method). rs-fMRI—extracted BOLD time series for seeds in HF, parahippocampal gyrus, middle frontal gyrus, and cingulate gyrus (Brainnetome Atlas); computed partial correlations (converted via Fisher r-to-z) to derive 166 functional connectivity measures (within HF and between HF and target regions). Cognition: Administered Verbal Learning and Memory Test (VLMT). Constructed short-term memory (VLMT-STM: z-standardized sum of Trial 1 and interference list) and long-term memory (VLMT-LTM: average of immediate post-interference recall and 20-min delayed recall) scores. Statistics: Outlier detection performed. Predictors included aerobic fitness (z-standardized) and covariates: age, sex, body mass index, disorder duration, education years, and chlorpromazine equivalents (defined daily dose method). Conducted two multivariate multiple linear regressions (MMLR): (1) predicting 38 HF subfield volumes (n=48), and (2) predicting 166 HF-related functional connectivity measures (n=44). For significant aerobic fitness effects, performed causal mediation analyses with aerobic fitness as independent variable, HF measures as mediators, and VLMT-STM/-LTM as outcomes, controlling false discovery rate (FDR). Visualizations via ggplot2.

- HF subfield volumes: The overall MMLR model trended but was not significant (R²pseudo=0.196, T(7)=1.393, p=0.059). Aerobic fitness significantly predicted HF subfield volumes (T(1)=3.639, R²pseudo=0.073, p=0.003), with generally positive associations across subfields. Strongest positive relations were observed in CA1–CA4, dentate gyrus (DG; including GCMLDG), and subiculum. Aerobic fitness showed positive effects on all subfields except for bilateral fimbria and bilateral bodies of presubiculum and molecular layer, where effects were not positive. - Functional connectivity: No significant associations were found between aerobic fitness and static functional connectivity within HF or between HF and parahippocampal, middle frontal, or cingulate seeds. The overall MMLR model was not significant (R²pseudo=0.16, T(7)=1.01, p=0.442); aerobic fitness predictor p=0.399. - Mediation by HF subfields on verbal memory: No mediation effects remained significant after FDR correction. At uncorrected p-levels, positive mediation effects were noted for left GCMLDG body and left CA4 body on VLMT-STM, and for left GCMLDG body on VLMT-LTM; a negative uncorrected mediation was observed for left subiculum head on VLMT-LTM. These did not survive FDR control. - Samples: sMRI analyses n=48; rs-fMRI n=44; overall recruited with valid fitness data n=53.

Findings support the hypothesis that higher aerobic fitness is associated with greater volumes in most HF subfields in schizophrenia, particularly in the subfields known to be most vulnerable (CA1–CA4, DG, subiculum). This suggests aerobic fitness may mitigate or counteract disease-related hippocampal volume loss. The absence of associations with HF functional connectivity contrasts with limited prior interventional findings and may reflect methodological differences, population heterogeneity, or insufficient power for subtle connectivity effects. The lack of robust mediation of verbal memory by subfield volumes indicates that, cross-sectionally, volumetric differences linked to fitness may not directly translate to verbal memory performance, or effects may be small and task-dependent. Comparisons with prior short-duration interventions suggest that longer-term fitness increases may be critical for structural effects; genetic factors (e.g., schizophrenia polygenic risk, cell-specific risk scores) may moderate neuroplasticity, offering a possible explanation for inter-individual variability.

The study demonstrates a positive cross-sectional association between aerobic fitness and volumes of hippocampal formation subfields in schizophrenia, with strongest effects in CA fields, dentate gyrus, and subiculum. No significant relations were observed for HF functional connectivity or for mediation of verbal memory after multiple-comparison correction. Results motivate longitudinal and interventional studies with larger samples to test causality, assess dose–response and duration effects of exercise, incorporate genetic moderators, and examine broader HF-dependent cognitive domains (e.g., spatial memory). Exercise remains a promising adjunctive approach to support hippocampal integrity in schizophrenia.

- Cross-sectional design precludes causal inference. - MRI data at required submillimeter resolution were available only from one center, limiting generalizability; even with high resolution, some anatomical variations may remain unresolved. - Potential unmeasured confounders (e.g., menstrual cycle phase) were not assessed. - Limited statistical power may have obscured small effects; replication in independent, larger samples is needed. - Genetic risk factors and subgroup characteristics were not incorporated and could modulate neuroplastic responses to fitness. - Cognitive assessment focused on verbal memory; other hippocampal-dependent domains were not tested.