Long-Term Improvement in Hippocampal-Dependent Learning Ability in Healthy, Aged Individuals Following High Intensity Interval Training

Abstract

Physical exercise may reduce dementia risk in aging, but varying reports on its effectiveness make it challenging to ascribe what level of exercise will have significant longer-term effects on important functions such as hippocampal-based learning and memory. This study compared the effect of three different 6-month exercise regimens on hippocampal-dependent cognition in healthy, elderly individuals. Participants, aged 65-85 with no cognitive deficits, were randomly assigned to one of three exercise interventions (low (LIT), medium (MIT), and High intensity interval training (HIIT), respectively). Each participant attended 72 supervised exercise sessions over a 6-month period. A total of 151 participants completed all sessions. Cognitive testing for hippocampal performance occurred monthly, as did blood collection, and continued for up to 5 years following initiation of the study. Multimodal 7 Tesla MRI scans were taken at commencement, 6 and 12 months. After 6 months, only the HIIT group displayed significant improvement in hippocampal function, as measured by paired associative learning (PAL). MRI from the HIIT group showed abrogation of the age-dependent volumetric decrease within several cortical regions including the hippocampus and improved functional connectivity between multiple neural networks not seen in the other groups. HIIT-mediated changes in the circulating levels of brain-derived neurotrophic factor (BDNF) and cortisol correlated to improved hippocampal-dependent cognitive ability. These findings demonstrate that HIIT significantly improves and prolongs the hippocampal-dependent cognitive health of aged individuals. Importantly, improvement was retained for at least 5 years following initiation of HIIT, suggesting that the changes seen in hippocampal volume and connectivity underpin this long-term maintenance. Sustained improvement in hippocampal function to this extent confirms that such exercise-based interventions can provide significant protection against hippocampal cognitive decline in the aged population. The changes in specific blood factor levels also may provide useful biomarkers for choosing the optimal exercise regimen to promote cognitive improvement.

Figure 1.

Cohort schematic illustrating recruitment and participation for each group during the 6-month exercise trial and beyond. The number of participants who were screened and assigned to each group is provided. The reasons for withdrawal are summarized.

Figure 2.

Learning performance was significantly improved during the exercise intervention period and beyond. (A) Only the HIIT group exhibited significantly improved PALTEA performance (fewer errors) during the exercise intervention. Neither the LIT nor the MIT group exhibited altered PALTEA performance during the exercise intervention. At the 6-month timepoint the HIIT group significantly outperformed the LIT group (mean±SE; two-way RM-ANOVA [time x exercise intensity effect F (10,738) = 2.272, p= 0.012], with Bonferroni post-hoc tests, p= 0.004 and p= 0.008 respectively, effect sizes LIT: HIIT = 0.72, MIT: HIIT = 0.74). (B) The improvement in PALTEA performance for the HIIT group observed following 6 months of exercise was maintained for up to 5 years following exercise initiation. Conversely, both the LIT and MIT groups remained stable at each time point measured (mean±SE; two-way RM-ANOVA [time x exercise intensity effect F (6,317) = 4.076, p= 0.0006], repeated measures with Bonferroni’s post-hoc tests, p= 0.036 and p= 0.017 respectively, 12m effect sizes; LIT: HIIT = 0.5, MIT: HIIT effect size = 0.62. >48m effect sizes; LIT: HIIT = 0.73, MIT: HIIT = 0.77). (C) Comparing cognitive performance for participants who were +1SD above mean PALTEA performance at the start of testing showed that the HIIT group significantly improved and outperformed both the LIT and MIT groups (mean±SD; one-way ANOVA F (2,32) = 10.71, p= 0.0003, with Bonferroni’s post hoc tests, effect sizes: LT: HIIT= 1.4, MIT: HIIT= 1.7). Within group statistical comparisons are represented by significance indicators of the same colour whereas black significance indicators represent differences between groups. ns = non-significant, *p< 0.05, **p< 0.01, ***p< 0.001, ****p< 0.0001.

Figure 3.

HIIT exercise prevents volumetric loss in specific brain regions. (A) Comparison of the % change in volume from baseline to the end of the exercise intervention showed that the hippocampal volume from the right-hand side (RHS) remained stable for the HIIT group. However, both the LIT and MIT groups showed a significant reduction (mean±SD; mixed effects model with post hoc tests, p< 0.001 and p< 0.0001 respectively). (B) Following the intervention, the cortical spinal tract remained stable for both the HIIT and MIT groups; however, there was a significant decrease in the LIT group. Differences in % volume change were also observed between the MIT and LIT groups and the HIIT and LIT groups (mean±SD; mixed effects model with post hoc tests [time x exercise intensity effect F (2,214) = 4.71, p= 0.01], with Bonferroni post-hoc tests). (C) Following exercise there was a significant % difference in the volume of white matter associated with the arcuate fasciculus in the HIIT group compared to both the MIT and LIT groups (mean±SD; mixed effects model with post hoc tests [time x exercise intensity effect F (2,214) = 4.718, p= 0.001], with post-hoc tests. (D) The increased white matter of both the cortical spinal tract and the arcuate fasciculus, as compared between the HIIT and LIT groups, was shown to connect to the premotor cortex (purple) and the primary motor cortex (cyan) by voxel-wise analysis. (E) The amount of volumetric loss in the RHS Hippocampus after 12 months was significantly reduced in the HIIT group compared to both the LIT and MIT groups. (F) At the 12-month time point the HIIT group showed an increase in cortical spinal tract volume whereas the LIT showed a decrease, resulting in a significant difference between these groups (p= 0.004). (G) The arcuate fasciculus remained stable in the HIIT and MIT groups whereas a significant decrease was observed in the LIT group at the 12-month time point (mean±SD; mixed effects model with post hoc tests). Coloured significance indicators are relative to baseline cognitive performance over time for each respective group whereas black significance indicators represent significant differences between groups. *p< 0.05, **p< 0.01, ***p< 0.001, ****p< 0.0001.

Figure 4.

HIIT exercise improves connectivity between multiple resting-state networks. (A) Comparison of baseline to post-exercise scans revealed FC in the HIIT group significantly increased between network pairs including the DMN-FRNT, DMN-ATTN and VIS-MOT networks when FDR was corrected at Q< 0.2. (B) Following exercise, the HIIT group showed increased network FC between the DMN-MOT and VIS-MOT network pairs compared to the LIT group. (C) Comparison of the MIT and HIIT groups following exercise revealed increases in connectivity in the HIIT group, including in the DMN-FRONT, DMN-MOT, VIS-FRNT, ATTN-FRNT and VIS-MOT circuits (FDR corrected at Q <0.1). (D) Comparison of Group x Time (the change in connectivity between baseline and following exercise) interactions between the LIT and HIIT groups showed that the HIIT group had significant FC increases between the DMN-MOT, DMN-VIS, DMN-FRNT, DMN-ATTN, VIS-MOT and VIS-FRNT networks (FDR corrected at Q< 0.1). (E) Representative MRI scans showing the regions of the brain that belong to the independent components (ICs) used to determine the MOT, VIS, DMN, ATTN and FRNT networks. Hot colours represent significant increases in functional connectivity (FC) while cool colours represent a decrease in FC. The X, Y and Z planes are shown for each ICs. Motor (MOT), visual (VIS), default mode (DMN), attention (ATTN) and frontal (FRNT) networks. The numbers listed for each network group represent different anatomical regions of the brain. Significance was calculated as a log p value and only those that survived FDR correction were included.

Figure 5.

Cumulative analyte deltas show biochemical and exercise intensity-specific changes. Tracking individual participants cumulative pre-post exercise deltas (∆) revealed large differences both between and within groups. (A) The majority of individuals in each exercise group had a positive cumulative ΔGH value. The cumulative ∆GH slope was significantly higher for the HIIT group than the LIT group (mean±SE; two-way RM-ANOVA [time x exercise intensity effect F (12,888) = 2.802, p= 0.0009], with post hoc tests, effect size LIT: HIIT = 0.43). (B) Several of the LIT participants had a negative cumulative ∆BHB value whereas the majority of MIT and HIIT individuals had a positive value. The cumulative ∆BHB slope was significantly higher for both the HIIT and MIT groups compared to the LIT group, and the HIIT slope was significantly steeper than that of the MIT group. Following 6 months of exercise there was a significant difference in ∆BHB between groups (mean±SE; two-way RM-ANOVA [time x exercise intensity effect F (12,870) = 20.39, p< 0.0001], with post hoc tests, effect sizes LIT: HIIT = 1.62, LIT: MIT = 1.13). (C) There was a large spread in cumulative ΔBDNF between participants for each exercise intensity. The cumulative ∆BDNF slope for the LIT group was close to zero after 6 months of exercise whereas both the MIT and HIIT groups had positive values (mean±SE; two-way RM-ANOVA [time x exercise intensity effect F (12,883) = 1.814, p= 0.042], with post hoc tests, effect size LIT: HIIT = 0.33). (D) The majority of participants in the MIT and LIT groups had a negative cumulative ∆cortisol value whereas the HIIT group had a large spread in cumulative values. Both the LIT and MIT groups had significant negative cumulative ∆cortisol slopes whereas that of the HIIT group remained steady (mean±SE; two-way RM-ANOVA [time x exercise intensity effect F (12,888) = 16.63, p< 0.0001], with post hoc tests, effect sizes LIT: HIIT = 0.68, MIT: HIIT = 1.18). (E) The LIT and MIT participants were evenly distributed with negative and positive cumulative ∆prolactin values whereas the majority of HIIT participants had positive ∆prolactin levels. Both the LIT and MIT cumulative ∆prolactin levels remained stable over 6 months whereas the HIIT group exhibited a significant positive slope (mean±SE, two-way, RM-ANOVA [time x exercise intensity effect F (12,888) = 17.16, p< 0.0001], with post-hoc tests, effect sizes LIT: HIIT = 0.99, MIT: HIIT = 1.07). * p< 0.05, ** p< 0.01, and *** p< 0.001.

Figure 6.

Pre/post exercise delta values for cortisol and BDNF predict improved learning ability for the HIIT group. (A) The HIIT group showed a significant correlation between the initial exercise-mediated cortisol pre/post delta values (∆) and learning ability (∆PALTEA), with higher cortisol values correlating to improved learning. (B) The total cumulative ∆cortisol levels at 6 months showed that higher ∆cortisol levels correlated to better learning ability for the HIIT group. (C) The 6-month pre/post delta Δcortisol levels correlated to superior endpoint cognitive ability for the HIIT group. (D) The HIIT group showed a significant correlation between the initial, pre/post-exercise-mediated ΔBDNF values and PALTEA learning ability. Spearman correlations were used for analysis.

Figure 7.

Lower prolactin levels and higher cumulative exercise workload correlate to improved cognition in MIT participants. (A) Exercise-mediated changes in total cumulative Δprolactin correlated to improved PALTEA performance for the MIT group. (B) Individual MIT workload levels subdivided based on gender, with solid lines representing males and dotted lines representing females. (C) The total cumulative workload after 6 months of exercise for the MIT group significantly correlated to ∆PALTEA change with higher total workload correlating to better PALTEA improvement. Spearman correlations were used for analysis.