The SUPERBRAIN trial protocol (ClinicalTrials.gov: NCT03980392) (Park et al., 2020) and primary findings (Moon et al., 2021) have been described previously. Briefly, this study was a 24-week randomized controlled trial conducted at three hospitals and five public health centers across South Korea, with a three-parallel-arm design including the FMI, home-based multidomain intervention (HMI), and control groups. Participants were selected from people who visited outpatient clinics or public health centers for memory problems, and those recruited through advertising. The study included 152 participants aged 60–79 years who had no dementia but had one or more modifiable dementia risk factors such as hypertension, diabetes mellitus, dyslipidemia, smoking, obesity, abdominal obesity, metabolic syndrome, educational level of ≤9 years, social isolation, and physical inactivity. In addition, they had a Mini-Mental State Examination z score of ≥ −1.5 based on the means and standard deviations in the age- and education-matched normal elderly Korean population (Han et al., 2008), were able to perform independent activities of daily living and pass a literacy test, and had a reliable informant who can provide investigators with the requested information. The individuals with major psychiatric illness such as major depressive disorder, dementia, Parkinson’s disease, malignancy within the previous 5 years, cardiac stent or revascularization within the previous 1 year, serious or unstable cardiovascular disease, other serious or unstable medical disease such as acute or severe asthma, active gastric ulcer, severe liver disease, or severe renal disease, as well as severe loss of vision, hearing, or communication disability were excluded. The study was conducted in accordance with the International Conference on Harmonization Good Clinical Practice Guidelines, and the institutional review boards of all institutions approved the protocol and consent forms before the start of the study. Written informed consent was obtained from all potential participants prior to enrollment. The SUPERBRAIN MRI exploratory sub-study included 63 participants from three trial sites (Ajou University Hospital, Ewha Womans University Medical Center, and Inha University Hospital) (Figure 1). Brain scans were conducted at baseline and 24-week visits. The present study included 55 participants with both baseline and repeat scans of good quality.

The participants were randomly assigned to the FMI, HMI, and control groups at baseline at a 1:1:1 ratio. The FMI and HMI groups received interventions consisting of five components, including monitoring and management of metabolic and vascular risk factors, cognitive training and social activity, physical exercise (Lee et al., 2020), nutritional guidance, and motivational enhancement, as previously described in detail (Park et al., 2020; Moon et al., 2021).

Before the intervention, metabolic and vascular risk factors were assessed using blood tests and anthropometric measurements. These risk factors included hypertension, diabetes mellitus, dyslipidemia, obesity, abdominal obesity, smoking, and high alcohol consumption, which were monitored and managed. Study doctors met with each participant at baseline and week 12 to inform them of their risk factors and prescribe medications if necessary. Participants were educated about their risk factors and given lifestyle guidelines to prevent dementia by a study nurse at baseline. The nurse also met with participants every 4 weeks for measurements and monitoring of smoking and alcohol consumption. Measurements were recorded in the participants’ SUPERBRAIN notebooks. If a participant’s risk factors did not improve, they were re-educated by the nurse at week 12 using the booklets. The FMI group received facility-based interventions three times a week, while the HMI group received the same cognitive training and physical exercise programs as the FMI group but with a different schedule. In the first 2 months of the trial, the HMI group participated in one group-based cognitive training session (50 min) and one home-based cognitive training session (30–40 min) per week, and one group exercise session (60 min) and two home-based exercise sessions (60 min) per week. For the rest of the 6-month study, the HMI group attended one group cognitive training session and one group exercise session every 2 weeks. On weeks that included group sessions, the participants attended one cognitive training session and two exercise sessions individually at home each week. On weeks that did not include group sessions, participants attended two cognitive training sessions and three exercise sessions alone at home each week. The cognitive training focused on episodic memory, executive function, attention, working memory, calculation, and visuospatial function and was conducted using a tablet-based application or workbooks supervised by trained health professionals. The physical exercise program was designed to include aerobic exercise, exercises to enhance balance and flexibility, muscle-strengthening activities involving major muscle groups, and finger-and-toe movements. Trained exercise professionals guided the exercise programs during group sessions at a gym, using portable tools such as elastic bands, floor plates with numbers, and chairs. The exercise intensity increased every 2 months. During home-based exercise sessions, participants exercised by watching videos on a tablet PC or by following instructions in a poster or booklet. The nutritional intervention consisted of three individual counseling sessions (each lasting 30 min) and seven group sessions (each lasting 50 min) led by study nutritionists. The individual sessions involved tailoring the participant’s daily diet and providing education on customized diets to manage individual vascular risk factors. The group sessions provided discussions, practical exercises, and cooking lessons on making meals with recommended ingredients. Motivational enhancement included four group counseling sessions (each lasting 50 min) led by a psychologist at 1, 2, 13, and 24 weeks. These sessions provided information and support to facilitate dementia prevention program activities and included discussions on the importance of change, ambivalence, and self-efficacy, as well as family education using video clips. The family-coach program involved a family member in reinforcing the motivation of a participant. Encouraging pop-up video messages made by participants’ families and self-rated achievement pop-up messages were provided every week before the tablet-based cognitive intervention. Participants in workbook-type cognitive interventions were sent the encouraging pop-up video messages on their cell phones and performed self-rated achievement assessments on paper. The control group received regular health advice according to established guidelines.

The study used the Repeatable Battery for the Assessment of Neuropsychological Status (RBANS) to evaluate cognitive function. The RBANS is a well-validated and reliable cognitive screening battery comprising 12 subtests that can be administered in 20–30 min with 4 alternate versions (A, B, C, and D) for reevaluation (Randolph et al., 1998). The subtests for each cognitive domain are as follows: digit span and coding for attention, picture naming and semantic fluency for language, figure copy and line orientation for visuospatial-construction, list learning and story memory for immediate memory, and list recall, list recognition, story recall and figure recall for delayed memory. Participants completed versions A and D at baseline and post-intervention, respectively, and a standard score was obtained for each cognitive domain based on same-aged peers. The total scale index score of cognitive functioning was calculated by combining these scores. Assessors received education on outcome measurements prior to the study (Park et al., 2020). Higher scores on the RBANS indicate better cognitive functioning. The assessment was conducted within 4 weeks post-intervention.

Changes in serum BDNF levels were investigated after the multidomain intervention. Fasting blood samples were collected at approximately 9 am using serum separator tubes (SSTs) within 4 weeks before and after the intervention. The SSTs were kept at room temperature for 30 min, centrifuged for 10 min at 3000 rpm, and sent to a central laboratory. Serum samples were stored at −70°C or lower until analysis. Serum BDNF levels were measured using a quantitative sandwich enzyme-linked immunosorbent assay (ELISA) (DBD00; R&D Systems, Inc., Minneapolis, MN, USA) in accordance with the manufacturer’s instructions.

Diffusion-weighted MRI data at each clinical site were acquired with the following MR systems and DTI parameters: 3 T Achieva, Philips with a 8-channel SENSE head coil [32 diffusion sampling directions; b-value, 1,000 s / m m 2 ; 256 × 256 acquisition matrix with 80 slices; repetition time (TR), 11.659 s; echo time (TE), 75 ms; flip angle, 90°; voxel size, 0.89 × 0.89 × 2 m m 3 ; field of view, 230 m m ] at the Ajou University Hospital, 3 T Achieva, Philips with a 8-channel SENSE head coil (32 diffusion sampling directions; b-value, 1,000 s / m m 2 ; 128 × 128 acquisition matrix with 80 slices; TR, 11.659 s; TE, 92.222 ms; flip angle, 90°; voxel size, 1.79 × 1.79 × 2 m m 3 ; field of view, 230 m m ) at the Ewha Woman’s University Medical Center, and 3 T Signa Architect, GE with a 34-channel array head coil, at the Inha University Hospital (30 diffusion sampling directions; b-value, 1,000 s / m m 2 ; 256 × 256 acquisition matrix with 70 slices; TR, 15 s; TE, 79.371 ms; flip angle, 90°; voxel size, 0.89 × 0.89 × 2 m m 3 ; field of view, 230 m m ). The same imaging parameters and MRI scanners were used for the baseline and 24-week scans. Regular phantom scans were performed within each site, although the phantom was not shared among three centers.

The diffusion-weighted images were processed using FMRIB’s Software Library (FSL).¹ The Brain Extraction Tool (BET) was used to remove skull and non-brain tissue to extract non-diffusion-weighted volume (b0 volume) brain regions. The EDDY tool² (Andersson and Sotiropoulos, 2016) was used to estimate and correct for susceptibility-induced off-resonance field, volume movement, and eddy current distortions. The low-frequency MR intensity inhomogeneity and its effects on the diffusion images were estimated for the b0 volume and the estimated bias field map was applied to all diffusion-weighted volumes using the N4BiasFieldCorrection tool from Advanced Normalization Tools (ANTs) (Tustison et al., 2010). To correct for geometric distortions due to MR susceptibility in the diffusion data, previous studies have used additional diffusion-weighted images with reversed phase-encoding direction using the TOPUP tool³ (Andersson et al., 2003). However, most diffusion-weighted studies struggle to apply the TOPUP tool due to a lack of reversed data (Wang et al., 2017). Alternatively, nonlinear registration for a structural T1 MRI was performed to correct for distortions using symmetric regularization (SyN) implemented in ANTs (Avants et al., 2008). The bias-corrected and skull-stripped T1 MRI was inverted prior to non-linear registration for the similar contrast characteristics of the source and the target images. The dtifit tool was used to fit the diffusion tensor model and computed diffusion scalar maps for FA, mean diffusivity (MD), axial diffusivity (AD) and radial diffusivity (RD) using the tensor eigenvalues.⁴ Finally, we performed feature extraction by aligning JHU-ICBM-labels atlas to the native FA image using a SyN and calculating the average FA, MD, AD, and RD values for each region of interest (ROI) (Mori et al., 2005). Based on the results of previous imaging analyses from the SUPERBRAIN studies (Moon et al., 2022a,b), specific ROIs were selected where changes in white matter integrity were anticipated. These ROIs consisted of the cingulum cingulate gyrus (CgC), cingulum hippocampus (CgH), superior longitudinal fasciculus (SLF), as well as the uncinate fasciculus (UF) on both hemispheres (Figure 2).

Group differences at baseline were analyzed by t-test (MRI vs. non-MRI groups) or one-way analyses of variance (ANOVA, FMI vs. HMI vs. controls) for continuous variables and the chi-square test for categorical variables. In addition, two-way repeated measures ANOVA, with time and groups as independent factors, was used to compare changes in the RBANS and serum BDNF levels from baseline to the study endpoint. The groups were considered similar when p > 0.05. Post hoc analyses were done by Dunnett test.

Individual FA, MD, AD, and RD at baseline were compared among groups using one-way analyses of covariance (ANCOVA) with age, sex, years of education, and imaging site effects added as covariates of no interest. In addition, to compare changes in the FA, MD, AD, and RD from baseline to the study endpoint, we used two statistical analyses. First, two-way repeated measures ANOVA was used, with time and groups as independent factors and age, sex, years of education, and imaging site effects added as covariates of no interest. Secondly, ROI-wise paired t-tests were used to analyze longitudinal changes in white matter integrity for each group (FMI, HMI or controls) and combined intervention groups (FMI + HMI), respectively, using age, sex, years of education, and site effects as covariates. The statistical significance level was set at Bonferroni corrected value of p < 0.05. Significant brain image measures in previous paired t-test were compared between each group pair (FMI vs. control; HMI vs. control; FMI + HMI vs. control) using ANCOVA with changes as dependent variables and age, sex, education, and site as covariates. Finally, correlations between clinical parameters (total and standard scores cognitive domains of the RBANS) or the serum BDNF level and changes in significant brain image measures were evaluated after controlling for age, sex, and years of education at a statistical significance level of p < 0.05 (uncorrected for multiple corrections). Specifically, the statistical analyses were conducted using statsmodels.api.OLS package.⁵

The FA, MD, AD, and RD at each ROI at the baseline were not different among groups after Bonferroni correction. In the first statistical analysis using two-way repeated measures ANOVA, any significant difference in longitudinal changes in the FA, MD, AD, and RD was not revealed. The second statistical analysis, which analyzed longitudinal changes in white matter integrity for each group, revealed the results as follows: FA increase (p-value = 0.0003, t value = −4.625) and RD decrease (p-value = 0.0026, t value = 3.599) of the left UF in the control group; MD decrease (p-value = 0.0061, t value = 3.106) of the left CgH in the FMI; AD decrease (p-value = 0.0009, t value = 3.939) of the right CgH in the HMI. However, the difference in these longitudinal changes between groups was not significant. Additionally, in the combined intervention groups (FMI + HMI), ROI-wise paired t-tests showed significant longitudinal changes in white matter integrity of the cingula and right UF (Table 3). Among the significant regions in paired t-test of the combined intervention groups, compared with the control group, the FMI, HMI, and intervention group yielded significantly more beneficial effects on the AD of the CgC (Table 4; Figure 3).

Correlations between AD of the CgC and clinical variables were evaluated in participants. Although longitudinal AD changes of the CgC did not correlate significantly with changes in the RBANS, longitudinal AD changes of the left CgC correlated with the BDNF changes (r = 0.280, p = 0.048).