Participants were recruited from the “Mihama-Kiho Project” in the towns of Mihama and Kiho, Mie, Japan. The project was designed to investigate the effect of physical exercise with and without music on cognitive function and was also a general healthcare project of the towns. The research ethics committee of Kinan hospital approved the experimental protocol, and all participants gave written informed consent prior to the experiment. The study was performed according to the Declaration of Helsinki. This study was registered with the University Hospital Medical Information Network Clinical Trials Registry (UMIN000012148).

To recruit participants, public servants distributed paper flyers among inhabitants 65 years and older who lived in areas of Mihama and Kiho town. The inclusion criteria were as follows: (a) over 65 years old; (b) in good physical and psychological health; (c) normal or corrected vision; (d) ability to clearly hear instructions; (e) living independently; and (f) able to attend an exercise session once a week. Participants were excluded if they met any of the following exclusion criteria: (a) apparent history of cerebrovascular attack; (b) presence of chronic exhausting disease such as malignancy or infection; (c) severe cardiac, respiratory, or orthopedic disability that would prevent participants from exercising; (d) medication that might adversely affect cognition (antidepressants or antipsychotics); or (e) a diagnosis of dementia. The inclusion and exclusion criteria for the control (Cont) group were identical to those for the physical exercise with musical accompaniment (ExM) and the same exercise without musical accompaniment (Ex) groups except for (f) able to attend an exercise session once a week. Instead, participants in the control group were simply required to undergo an MRI and neuropsychological and physiological assessments once a year. Due to budget limitations of the towns, there was no music alone group, and the Cont group recruited was half the size of the intervention groups.

Between July 1 and 15, 2011, 166 participants expressed interest in the physical exercise with and without music groups (ExM and Ex, respectively), and 41 participants were interested to be included in the control (Cont) group. Before the intervention, the towns’ public health nurses saw participants and interviewed their family about the subjects’ daily physical and psychological activities. Six participants were excluded because they did not meet the inclusion criteria and declined to participate in the intervention groups. Five participants were excluded from the Cont group because they declined to participate and did not undergo MRI scans. According to age, sex, and activities of daily living (ADL) grade established by the Ministry of Health, Labor and Welfare, the participants were semi-randomly classified into ExM and Ex groups (Figure 1). However, married couples and siblings who wished to exercise together were placed in the same group.

No participants switched group. In the ExM group, 29 participants dropped out [busy with work and due to damage by a major flood (n = 24), dementia onset (n = 1), death (n = 1), and did not complete MRI scans (n = 3)]. In the Ex group, 19 participants dropped out [busy with work and due to damage by a major flood (n = 13), other (n = 2), and did not complete MRI scans (n = 4)]. In the second assessment, four participants were excluded in the Cont group [rejected second assessment (n = 4)]. We analyzed participants whose attendance rate was above 75%. As a result, data from a total of 144 participants were included (Figure 1).

The physical exercise program is described in detail in our previous paper (Satoh et al., 2014). The intervention period was 1 year, and the total number of exercise sessions was 40. Exercise intensity gradually increased with each session. The exercise program and musical accompaniment were developed approximately 10 years ago in collaboration between the Japan Fitness Association, sport medicine experts, and the YAMAHA Music Foundation. The musical accompaniment was synthesizer-heavy dance pop music. The exercise program consists of nine stages, and participants easily and happily performed the exercise. The physical exercise regimen was identical for the ExM and Ex groups and was implemented by professional trainers. To control auditory cue and feedback between the ExM and Ex groups, the ExM group heard music over speakers, while the Ex group only heard a percussive sound that counted the beat over speakers.

Neuropsychological assessment procedures were the same as previously described (Satoh et al., 2014). The Mini Mental State Examination (MMSE) (Folstein et al., 1975) and Raven’s Colored Progressive Matrices (RCPM) (Raven, 1947) were used to screen cognitive ability and quantify intellectual function, respectively. Memory was evaluated using the Logical Memory (LM)-I/-II subtests of the Rivermead Behavioural Memory Test (RBMT) (Wilson et al., 1985), which require immediate and delayed recall of a short story. The RBMT contains four stories with difference levels of difficulty and numbers of words. We used different stories for the pre-and post-testing periods to avoid familiarity with the story content. Visuospatial constructional ability was based on the method described by Strub and Black (2001). Five kinds of figures (vertical diamond, two-dimensional cross, three-dimensional block, three-dimensional pipe, and triangle within a triangle) were shown to the subjects, and they were asked to draw them one by one. Each drawing was scored by assigning one of four possible grades (0: poor, 1: fair, 2: good, and 3: excellent), with a maximum score of 15. Frontal lobe function was assessed using two tasks: word fluency (WF) and Trail-Making Test (TMT)-A/-B (Partington and Leiter, 1949). The WF test consisted of category and letter domains. For the categorical WF, participants were asked to name as many animals as possible in 1 min. For the letter WF task, participants were asked to say the name of objects that begin with each of four phonemes, ka, sa, ta, and te (Dohi et al., 1992). We used the average scores of these four phonemes for statistical analyses. The vital capacity as per cent of predicted (%VC) was used as a physiological assessment.

These neuropsychological and physiological assessments were administered before and after the 1-year intervention period to both the ExM and Ex groups. Cont group participants performed these assessments twice with an interval of 1 year.

Group differences for demographic variables were examined. The data were assessed using ANOVAs for continuous variables, chi-square tests for dichotomous variables, and Kruskal–Wallis tests for non-parametric data. Post-intervention changes in neuropsychological assessment results between the ExM or Ex group and Cont group were examined. These data were assessed using Dunnett’s tests for continuous variables and the Steele test for non-parametric data. Statistical analyses were done using IBM SPSS Statistics software version 20 (IBM Corp., Armonk, NY, United States) and EZR software version 1.29 (Kanda, 2013).

All MRI scans were performed with a 1.5-T MRI scanner (Intera, Royal Philips, Netherlands; ECHELON, Hitachi Medical Corporation, Japan). A T1-weighted gradient echo sequence was used (repetition time [TR] = Shortest [Automatic]; echo time [TE] = 15 [Intera]/11 [ECHELON] ms; flip angle = 90°; field of view = 230 mm × 230 mm; slice thickness = 5 mm; in-plane resolution = 0.45 mm × 0.45 mm). The first scans were taken as a baseline before the intervention, and the second scans were taken approximately 1 year later (average interval between pre- and post-intervention: 422.3 ± 34.6 days).

Magnetic resonance imaging data were analyzed using SPM12 (Wellcome Institute of Neurology, University College London, United Kingdom), running on MATLAB R2012a (MathWorks, Natick, MA, United States). In the pre-processing phase, images were set to match the anterior to posterior commissure (AC-PC) line using an automated MATLAB script. Then, images were visually inspected to check for possible scan issues such as field distortion and movement artifacts. Reoriented images were corrected for the intensity inhomogeneity and segmented into gray matter (GM), white matter (WM), cerebrospinal fluid, and other tissues outside the brain by the SPM12 tissue probability maps. The images were registered to the East Asian brains ICBM (International Consortium for Brain Mapping) space template through affine regularization. We created a population-specific template using the SPM12 DARTEL template procedure to directly compare ExM, Ex, and Cont groups and thus investigate (1) GM and WM differences between groups and (2) the relationship between neuropsychological assessment results and GM in these groups at a whole-brain level. The GM and WM segments were inputted into high-dimensional DARTEL to create non-linear modulated-normalized GM and WM images that were smoothed using a Gaussian kernel of 8 mm FWHM (full width at half maximum). No participants were excluded from the analysis after these steps.

For whole-brain and multiple regression analyses, we assessed the statistical significance at a cluster threshold of p < 0.05 (family wise error corrected) with a voxel threshold of p < 0.001 (uncorrected), and contiguous clusters of at least 10 voxels were reported. We obtained both MNI and Talairach coordinates to detect the anatomical regions of the clusters. We used a transform from Matthew Brett¹ to convert MNI coordinates to Talairach coordinates, and Talairach Client 2.4.3 (Lancaster et al., 2000) was used to identify the anatomical regions corresponding to Talairach coordinates.

Participant demographics are shown in Tables 1A,B. Although there were no significant differences with regard to the male:female ratio, age, years of education, ADL, MMSE score, or GMV, significant between-group differences were found in white matter volume (WMV) before the intervention (p = 0.001). Multiple comparisons revealed that the WMV of the Cont group was larger than that of the ExM and Ex groups (both p = 0.001).

The neuropsychological and physiological assessment results before and after the interventions are shown in Table 2. Intra-group comparisons showed significant improvements in the MMSE score, LM-I and -II subtests of the RBMT, visuospatial assessments, and %VC in both the ExM and Ex groups post-intervention. In addition, significant improvement in the WF category was observed in the Ex group post-intervention. In the Cont group, LM-II and TMT-B scores were significantly improved after 1 year.

Significant between-group differences were found for visuospatial assessment and %VC (p = 0.038 and 0.031). Multiple comparisons revealed significant differences between the ExM and Cont groups for visuospatial assessment (p = 0.037) and between the Ex and Cont groups for %VC (p = 0.036).

Although there were no significant between-group differences in post-intervention GMVs, the volumes were reduced with a declining trend compared to pre-intervention data in each group (Table 1B). The following area volumes were significantly larger (preserved) in the ExM group compared to the Cont group: the frontal gyrus (inferior, superior, and medial), cingulate (anterior and posterior), temporal gyrus (inferior, superior, and transverse), insula, parahippocampal gyrus, hippocampus, uncus, fusiform gyrus, thalamus, amygdala, middle occipital gyrus, and cerebellum (Figures 2A,B and Table 3A). The following area volumes were significantly larger (preserved) in the Ex group compared to the Cont group: the frontal gyrus (inferior, superior, middle, and medial), cingulate (anterior and posterior), temporal gyrus (inferior, superior, middle, and transverse), insula, parahippocampal gyrus, hippocampus, uncus, thalamus, cuneus, precuneus, and cerebellum (Figures 2C,D and Table 3B).

To compare the amount of GMV change between the ExM or Ex group and Cont group, we contrasted the post- and pre-intervention data for each group: (post_ExM > pre_ExM) > (post_Cont > pre_Cont) and (post_Ex > pre_Ex) > (post_Cont > pre_Cont). These contrasts showed larger volumes of the bilateral superior frontal and parahippocampal gyri in the ExM and Ex groups compared to the Cont group (p < 0.05 uncorrected). To distinguish pre- and post-intervention volume differences in different brain areas for each group, we calculated 90% confidence intervals by using these contrasts as a region of interest (ROI) (p < 0.001, uncorrected; Figure 3 and Table 4). The volumes of the ExM and Ex groups were preserved and actually increased. The right superior frontal gyrus volume only increased in the ExM group. In the Cont group, these areas were reduced after 1 year.

Significant between-group differences were found for post-intervention WMV (p = 0.001). Multiple comparisons revealed that the WMV of the Cont group was larger than those of the ExM and Ex groups (p = 0.009 and 0.001, respectively).

To compare the amount of WMV change between the ExM or Ex group and the Cont group, we contrasted the post-intervention and pre-intervention data for each group: (post_ExM > pre ExM) > (post_Cont > pre_Cont) and (post_Ex > pre_Ex) > (post_ Cont > pre_Cont). These contrasts showed a larger right anterior corona radiata volume in the ExM and Ex groups compared to the Cont group (p < 0.001, uncorrected). To determine pre- and post-intervention volume differences, we calculated a 90% confidence interval by using these contrasts as an ROI (Figure 4 and Table 4). WMV was increased in the ExM and Ex groups post-intervention, whereas WMV of the Cont group was reduced compared to baseline.

Based on the visuospatial assessment results only showing a significant difference between the ExM and Cont groups, we found that changes in visuospatial scores were positively correlated with the volumes of the left superior temporal gyrus, right anterior cingulate gyrus, and left insula (Figure 5 and Table 4). We calculated a 90% confidence interval by using these contrasts as an ROI. Although the volumes of the ExM group were preserved, those of the Ex and Cont groups were reduced (Figure 5).