An Exploratory Study of Exercise-related Effects on Memory and Hippocampal Connectivity in Schizophrenia

Abstract

Memory impairment in schizophrenia has been linked to abnormal functioning of fronto-temporal networks. In this pilot study, we investigated whether 12-weeks of exercise improved hippocampal-dependent memory functions and resting-state functional connectivity in middle-aged adults with schizophrenia. The exercise regimen was feasible, well-attended, and safe. There was a pre- to post-intervention increase in spatial memory accuracy that was correlated to an increase in hippocampal-prefrontal cortex connectivity. No increase was found in pattern separation performance or hippocampal volume. A controlled trial is needed to replicate these findings and elucidate the functional brain networks underlying exercise-induced cognitive improvement in schizophrenia.

Introduction

Emerging research on the beneficial effects of exercise on brain and cognitive health (1, 2, 3) has provided new opportunities to address the cognitive problems associated with schizophrenia. Studies in this population show that aerobic exercise improves cognitive functioning indexed on neuropsychological measures of attention, visual and verbal learning, and measures of global cognition (4, 5, 6, 7). While these results are promising the majority of these studies have been conducted in young to middle-age samples. However, studies in healthy aging show that exercise improves cognitive functioning across the lifespan into late adulthood (3). Cognitive interventions appear to be less effective later in the course of schizophrenia (8); however, little attention has been directed to this question with exercise. This pilot study addresses the feasibility of aerobic exercise training in adults with schizophrenia in midlife and tests for preliminary effectiveness of exercise to modulate specific cognitive and brain functions.

In non-clinical samples, aerobic training improves hippocampal-dependent memory functions such as spatial memory and pattern separation (9, 10), as well as working memory and executive functions largely dependent on the prefrontal cortex (11, 12). Neuroimaging studies suggest these improvements are related to the influence of cardiorespiratory fitness and aerobic training on hippocampal and prefrontal cortex volume (13, 14), hippocampal blood volume (15), and resting-state functional connectivity in hippocampal (16) and other large-scale networks (17).

Here, we use several of these experimental cognitive tasks coupled with neuroimaging to probe the functioning of fronto-temporal networks to examine the beneficial effects of aerobic exercise in schizophrenia patients across a broad age range. Specifically, we asked whether a 12-week exercise intervention improves performance on tasks of spatial memory and pattern separation, and further, whether enhanced performance in these tasks is correlated to changes in functional connectivity (FC) of hippocampal networks. We focus on the FC between hippocampus and prefrontal cortex given the role of these regions and their functional interaction in declarative and working memory deficits in schizophrenia (18, 19, 20, 21). We also tested whether the exercise intervention increased circulating levels of brain-derived neurotrophic factor (BDNF), which animal studies show plays a central role in mediating the effects of exercise on hippocampal plasticity and learning (22, 23).

Methods

Ten veterans (1 female; median age = 56.0 yrs, range = 39 to 67 yrs) who met Structured Clinical Interview for DSM-IV criteria for schizophrenia or schizoaffective disorder participated. Subjects were receiving stable doses of antipsychotic medication, with five also receiving stable antidepressant medication, for at least six weeks prior to enrollment. Antipsychotic dosage was calculated for each patient in olanzapine equivalents according to Leucht et al. (24) (mean dosage = 13.67 mg/day, SD = 10.82). Patients with significant medical or neurological conditions, or diagnosis of substance use disorder within three months of enrollment, were excluded. The Washington DC VA and Georgetown University Institutional Review Boards approved the study; all subjects provided informed consent.

The intervention was a 12-week (3 days/week) combination aerobic and strength training program supervised by an exercise physiologist. Subjects exercised at the Washington DC VA Wellness Center according to individualized plans. The intervention consisted of 30 minutes on a treadmill, stationary bicycle, or elliptical trainer augmented by weight training for upper and lower extremities. The intensity of aerobic exercise was determined for each participant and was set at 60% to 80% of peak heart rate achieved during a standard treadmill test, taking into consideration the individual’s resting heart rate. Subjects were encouraged to keep their heart rate within the target range using heart rate read-outs on the exercise equipment.

Assessments were performed within 7 days of the start and end of the intervention. Aerobic capacity was assessed using the exercise treadmill test. Peak aerobic capacity was measured by metabolic equivalent (MET), estimated with standardized equations on the basis of peak speed and peak time for the Bruce protocol (25). Serum BDNF levels were quantified using an enzyme immunoassay (Quantikine® ELISA Human BDNF Immunoassay). The Positive and Negative Syndrome Scale (PANSS; 26) was used to assess symptoms. In the spatial memory task (also administered at week 6), subjects viewed one, two, or three dots on a computer screen and after a 3-second delay judged whether a single dot was in the same or different location as one of the previously presented dots. In the pattern separation task, subjects performed an incidental task on pictures of objects followed by a test in which they identified targets (previously presented items), lures (visually similar to targets), and foils (new items) (27). The measure of pattern separation is correct identification of lures.

Magnetic resonance imaging (MRI) was conducted on a 3T Siemens Trio. Data for the high resolution T1-weighted structural scan were acquired with echo time (TE) = 2.52 ms, repetition time (TR) = 1900 ms, field of view (FOV) = 256 × 256 mm2, voxel size (isotropic) = 1mm, and 9° flip angle. For resting-state functional MRI, whole-brain images were acquired using T2-sensitive gradient echo planar imaging with TE = 30 ms, TR = 3000 ms, FOV = 205 × 205 mm, voxel size = 3.2 × 3.2 × 3mm, and 90^o flip angle, during a 5 mins/51secs scan. CONN functional connectivity toolbox was used to perform standard preprocessing and denoising steps, including slice-timing and motion correction, spatial normalization (resampled at 2×2×2mm) into MNI standard space, smoothing (8mm kernel), nuisance regression (6 motion parameters, CSF and white matter signal, and high motion volumes, i.e. > 3 standard deviations of average signal or >.5mm framewise displacement (16.6 +/−17 for pre, and 14.5+/−15.9 for post) “scrubbed”) and bandpass filtering (0.008-0.09 Hz).

Based on findings of functional segregation along the horizontal axis of the hippocampus (28, 29), we used a mask in SPM Marsbar (30) to divide the hippocampus into anterior, body, and posterior regions of interest (ROIs), resulting in three ROIs in each hemisphere (total = 6 ROIs). Time series for each ROI were correlated to that of every other voxel in the brain (Pearson’s correlation r values were transformed to Z scores), which produced a whole-brain connectivity map for each subject at pre and post-intervention that was entered into paired t-tests in CONN to test for second level FC differences. To avoid the potential for inflated false positive inference errors (31) for pre-post intervention differences, we used parametric statistics in CONN that is based on random field theory assumptions (32, 33), using a voxel- based height threshold of P < .001 combined with FDR-corrected cluster-level P < .0083, which is corrected for the number of seeds based on Bonferroni correction (P < .05/6).

Results

Nine of the 10 subjects completed the intervention with a mean attendance rate of 73%. There was no change in psychotrophic medication or dosage throughout the trial. MET level increased 18% from baseline (9.25) to endpoint (10.94), t(7)=3.06, P < 0.05, Cohen’s d_z = −1.08. Serum BDNF levels did not increase between baseline (22.5ng/mL) and endpoint (20.5 ng/mL), P >.05; however, there was a correlation between change in BDNF and change in aerobic fitness (Pearson’s r = .85, P < .01). There was no significant decrease in PANSS scores from baseline to endpoint [Mean (SD) total score: 58.5 (14.27) to 53.7 (6.99)], t(9) = 1.40 , P > 0.05. Accuracy (but not RT) in the spatial memory task improved pre- to post-intervention (Figure 1A), as shown by a significant interaction of Session (baseline, endpoint) by Set Size (1, 2, or 3), F(2, 16) = 3.65, P < .05, partial eta squared = .31. Post-hoc comparisons confirmed a significant increase for the set size of three, t(8) = −2.87, P <.05, Cohen’s d_z = −.99. By contrast, pattern separation performance between baseline (18%) and endpoint (18%) was unchanged, t(8) = .08, P > .05.

Figure 1.

A. Mean accuracy (± SEM) in the spatial memory task at baseline and the 12-week endpoint across set size (1,2, and 3).

B. Upper panel: Example of the right anterior (red), body (blue), and posterior divisions (green) of the hippocampus from one participant’s T1 scan. Following the procedures described by Demaster et al. (26), the anterior and posterior regions of the hippocampus were segmented from the body of the hippocampus in coronal view (Y = −20 and 36 for left anterior-body, body-posterior division; Y = −18 and 34 for right). Lower panel: Summary of FDR-corrected functional connectivity results. The color bar shows peak T values.

C. The correlation between change in mean accuracy (percent correct) in the spatial memory task and change in functional connectivity between the left anterior hippocampus and right MFG (BA 9/10) following the 12-week exercise intervention.

The seed-based FC analysis (available for 8 of the 9 completer-subjects) showed an increase in connectivity between the hippocampus bilaterally and regions in prefrontal and temporoparietal cortex (Figure 1B). Specifically, there was an increase between right anterior hippocampus and right middle frontal gyrus (MFG) [Brodmann’s area (BA) 10]; k = 104; peak T = 10.73; x, y, z coordinates 34, 54, 10, left anterior hippocampus and right MFG (BA 9/10); k=120; peak T = 14.48; x, y, z coordinates 40, 38, 28, and right body of the hippocampus and right supramarginal gyrus (BA 40); k = 112; peak T = 9.07; x, y, z coordinates 56, −22, 16. Coordinates are in MNI space.

In light of increased FC between the hippocampus and MFG, we examined volumetric changes in these regions using Freesurfer software (http://surfer.nmr.mgh.harvard.edu/) (34, 35). Measures were calculated as a percent of total intracranial volume, obtained from FreeSurfer. There was a significant increase for relative volume in right rostral MFG [Mean (SD): .913 (0.89) to .937 (0.89)], t(8) = −2.36, P < 0.05, Cohen’s d_z = −.79, but no change in the hippocampus.

To assess whether improvement in spatial memory performance was correlated to increases in fitness, FC or MFG volume, we performed partial correlation controlling for age. We did not observe a correlation with fitness measured by MET level; however, we found a positive trend between percentage of sessions within prescribed heart rate range and spatial memory in the three-item span condition (r =.51, P =.16). That is, memory improvement was higher for those subjects with the greater proportion of sessions within the exercise target heart rate range. This finding suggests that a workload threshold may be necessary to improve memory performance. There was a significant correlation between improved spatial memory and increased FC between left anterior hippocampus and right MFG (BA 9/10), Pearson’s partial r = .95, P < .01 (Figure 1C), but not with MFG volume change. No measure correlated to antipsychotic dose equivalents or antidepressant treatment, all Ps > .05.

Discussion

We replicated findings that exercise increases aerobic fitness in individuals with schizophrenia (36) and showed a significant association between enhanced fitness and serum BDNF (5, 37). Clinical symptoms, which were low to start at baseline, did not significantly improve but there was no adverse effect from engaging in the exercise program. Importantly, the intensity, frequency, and duration of the aerobic and strength training regimen was feasible and safe in middle-aged adults with schizophrenia. Improvement in spatial memory was associated with increased hippocampal-prefrontal connectivity, suggesting that greater coordination between these regions increases efficiency of encoding and maintenance of spatial representations. No changes were observed in measures of pattern separation or hippocampal structure (38, 39). Exercise seemed to improve connectivity in circuitry implicated in prefrontal cortical-based functions such as short-term spatial memory and working memory, a finding further supported by an increase in right rostral MFG volume.

Limitations of the pilot study include small sample size and lack of a control group. We therefore cannot rule out the influence of practice effects on cognitive performance and other non-specific effects of study participation. Nonetheless, our findings are consistent with research that cardiorespiratory fitness and physical activity have a positive influence on spatial memory and FC in hippocampal and other brain networks. Given the disrupted functional communication between hippocampus and prefrontal cortex in schizophrenia, exercise may provide an effective tool to strengthen connections between these regions. The preliminary data reported here provide support for a controlled study of the effects of exercise on cognition and functional brain connectivity in individuals with schizophrenia in mid to later life.