Patients (aged 50–80 years) with MCI were recruited in Berlin (Memory Clinic of the Department of Neurology of the Charité University Hospital and Neurology specialist practice) and Frankfurt am Main (Institute of General Practice), Germany. MCI patients (amnestic; single and multiple domain) were diagnosed according to Mayo criteria within 12 months before baseline visit. These criteria comprised a subjective cognitive complaint and an objective memory impairment in standardized tests [performing at least one standard deviation below age- and education-specific norm in relevant subtests of the CERAD-Plus or Rey Auditory Verbal Learning Test (AVLT) battery (Total Word List, Delayed Recall Word List/Figures) (Morris et al., 1989)]. Moreover, patients showed relatively preserved general cognition, no impairment in activities of daily living, and no dementia (Petersen et al., 1999). Exclusion criteria comprised MMSE scores <24 at baseline visit, severe untreated medical, neurological or psychiatric diseases and brain pathologies identified in the magnetic resonance imaging (MRI) scan, no right-handedness (Oldfield, 1971), non-fluent German language and BMI <18 kg/m² or >35 kg/m².

One-hundred-ten patients, previously diagnosed with MCI, were screened for study eligibility by telephone (e.g., for additional severe diseases, known brain pathologies, MR ineligibility). Twenty-two had to be excluded on the basis of these criteria. The remaining 88 patients were invited for baseline assessment. From this group, 11 patients had to be excluded either due to a pathological MRI finding (n = 3) or due to comorbidities (n = 8; Parkinson's disease, depression). Eligible patients were allocated in different intervention groups (details below) by a simple randomization approach, based on a computer-generated list of random numbers, carried out by an investigator that was not involved in the study. Thirty-five patients were randomized into a separate study testing the effect of omega-3 fatty acids on brain structure and function. Two patients (resveratrol n = 1; placebo n = 1) did not complete the intervention due to time constraints. In total, 40 MCI patients completed the current study (resveratrol n = 18; placebo n = 22; see Figure 1). MRI scans from follow-up were not available for 10 patients (scheduling problems at follow-up), leaving 30 patients for longitudinal MRI analysis (resveratrol n = 14; placebo n = 16).

During baseline assessment, patients underwent a standardized medical examination, neuropsychological testing, structural and resting-state functional MRI of the brain, as well as fasting blood sampling, and assessments of anthropometric data and mood (baseline assessment; see Figure 1 and Table 1).

MCI patients were randomized in either a resveratrol or placebo group, receiving a daily supplementation of 200 mg resveratrol, plus 350 mg quercetin to increase bioavailability of resveratrol (De Santi et al., 2000), or 1,015 mg olive oil, respectively. Patients were instructed to follow a regular intake once a day (4 capsules) before or at a main meal for 26 weeks. Resveratrol intake was well tolerated with no reported treatment-related serious adverse events that would have led to study termination. Each patient received half of their study medication at baseline and the second half after 13 weeks.

Capsules were provided by VIA Vitamine, Oberhausen, Germany. Following the intervention, baseline measurements were repeated (follow-up assessment; see Figure 1).

The number of remaining resveratrol or placebo capsules was counted after 13 and 26 weeks, showing a >83% compliance rate. Moreover, patients completed a questionnaire on capsule intake at the end of the study. Misses of capsule intake exceeding 5 times/ month would have led to pre-specified exclusion from analysis. However, none of the patients had to be excluded in this study due to capsule misses. Patients were instructed not to change their dietary habits and their physical activity throughout the intervention.

Learning and episodic declarative memory performance of MCI patients was tested, using the German version of the Rey Auditory Verbal Learning Test (AVLT) (Lezak, 2004). Patients were asked to learn a list of 15 words within five immediate recall trials, followed by a 30 min delayed recall and delayed recognition test. Learning ability was defined as the sum of words learned in all five trials (maximum 75 words). Delayed recall represented the total number of remembered words after 30 min (maximum 15 words). For delayed recognition, MCI patients were asked to recognize the 15 original words presented within 35 distractor words subsequent to the delayed recall test (number of correctly recognized words minus false positive words; maximum 15 words). Trained staff members conducted all tests according to standardized procedures.

MRI scanning was conducted at baseline and follow-up, using a 3-Tesla Siemens Trio system with a 12-channel head coil at the Berlin Center for Advanced Neuroimaging.

High resolution T1-weighted scans (3D Magnetization Prepared Rapid Acquisition with Gradient Echoes (MPRAGE); TR = 1,900 ms, TE = 2.52 ms, 192 sagittal slices, voxel-size of 1.0 × 1.0 × 1.0 mm³, flip angle = 9°), and diffusion-weighted spin-echo echo-planar imaging (EPI) scans (TR = 7,500 ms, TE = 86 ms, 61 axial slices, voxel size of 2.3 × 2.3 × 2.3 mm³; 64 directions with a b-value of 1,000 s/mm² and 10 b0) were acquired.

Functional scans were obtained at rest using a T2^*-weighted EPI sequence (TR = 2,300 ms, TE = 30 ms, 34 slices, voxel size of 3.0 × 3.0 × 4.0 mm³, flip angle = 90°). Patients were instructed to keep their eyes closed, relax, think of nothing in particular and move as little as possible during this 6 min scan.

Image preprocessing and analyses were done using the software package FSL 4.1 (http://www.fmrib.ox.ac.uk/fsl), AFNI 2011 (http://afni.nimh.nih.gov/afni), and FreeSurfer 5.3 (http://surfer.nmr.mgh.harvard.edu/), as indicated below.

MRI analyses were performed according to Witte et al. (2014), where the procedure is described in detail.

Briefly, volumetric delineation of the total gray matter, and left and right HC was conducted, using FSL and Freesurfer brain extraction tools (BET and mri_watershed), FMRIB's Automated Segmentation Tool (FAST) and Integrated Registration and Segmentation Tool (FIRST). To determine the respective anterior and posterior parts, the center of gravity of the HC was assessed after rigid body transformation to MNI space and the corresponding y-coordinate then served as a measure for individual anterior/posterior-division.

Individual HC volumes were adjusted for intracranial volume (ICV), according to previous studies (Raz et al., 2005; den Heijer et al., 2012; Kerti et al., 2013), using the following formula: adjusted volume = raw volume − b × (ICV − mean ICV). The coefficient b represents the slope of regression of a region of interest volume on ICV. The results of HC segmentation were superimposed on anatomic images and visually inspected to exclude misregistration or erroneous HC identification.

Hippocampal microstructure was assessed by mean diffusivity (MD), estimated by using diffusion tensor imaging (DTI), in line with previous studies (den Heijer et al., 2012; Kerti et al., 2013). Therefore, a tensor model was fitted to the motion-corrected DTI data at each voxel to create individual 3-dimensional maps of MD. Then, individual T1-weighted images were co-registered to the b0 images, using rigid-body transformation. These registrations were used to transform masks of the left and right anterior and posterior HC (derived from the T1 images) to the MD maps, for extraction of the mean individual hippocampal MD values. For this analysis FSL software was used.

To assess potential changes in RSFC of the HC, we used a customized processing stream based on the 1,000 Functional Connectomes Project (http://www.nitrc.org/projects/fcon_1000) (Biswal et al., 2010). Co-registered masks of the left and right anterior and posterior HC served as seeds for RSFC analysis, in line with previous studies (Rombouts et al., 2005; Andrews-Hanna et al., 2010; Witte et al., 2014). Pre-processing of individual functional scans comprised slice time correction, motion correction, spatial smoothing with a 6 mm full-width-half-maximum (FWHM) Gaussian kernel, temporal filtering (0.01–0.1 Hz), and de-trending, using AFNI and FSL software. The functional scans were normalized to the anatomical image, using affine co-registrations. Noise due to motion, white matter, cerebrospinal fluid, and global change was removed from the functional signal by multiple regressions, creating standardized residual BOLD-signal time series (in FSL). Then, mean time series of the individual HC seeds were correlated with times series of all other gray matter voxels in the brain, using a general linear model approach within FSL (FMRIB's local analysis of mixed effects with Ordinary Least Square option, FLAMEO). The resulting Pearson's r correlation coefficient 3D maps were then Fisher's z-transformed and smoothed, using a kernel of sigma = 1 to improve normality. This produced spatial maps in which the values of voxels represented the strength of the correlation with the individual HC seeds. Registration of individual z-maps for group analysis included a rigid body within-subject registration of both time points to a “halfway space” before affine and non-linear registrations to a study-specific template, as described in detail in previous studies (Witte et al., 2014; Köbe et al., 2016).

DNA was extracted from whole blood, using a blood mini-kit (Qiagen, Hilden, Germany) and stored at −80°C until analysis. Genotyping of the single nucleotide polymorphisms (SNP) apolipoprotein E (APOE) rs429358 and rs7412 that have been previously implicated in cognitive performance (Corder et al., 1993; Egan et al., 2003; Witte and Floel, 2012) was performed, using a pre-designed Taqman assay at the laboratory of Prof. Dr. Dan Rujescu (University of Halle, Germany), following procedures described previously (O'Dwyer et al., 2012).

After fasting overnight of at least 10 h, all subjects underwent venous blood sampling for assessment of serum levels of glucose, glycated hemoglobin A1c as long-term measure of glucose, insulin, high-to-low density lipoprotein (HDL-to-LDL) ratio, and high-sensitive C-reactive protein (hsCRP). All parameters were analyzed by IMD Laboratory, Berlin, Germany. Anthropometric measures included weight and body mass index (BMI). Patients also reported their physical activity and other lifestyle habits, using the Freiburger physical activity questionnaire (Frey et al., 1999). For mood ratings during neuropsychological testing, the Positive and Negative Affective Schedule (PANAS; Krohne et al., 1996) was used. Moreover, to characterize each patient's depression and anxiety level at baseline, the Beck's Depression Inventory (BDI; Kuhner et al., 2007) and the State-Trait Anxiety Inventory (STAI X1; Laux et al., 1981) were used.

Before data analysis, all variables were tested for normal or near-normal distribution (unimodal, |skewness| <1). Accordingly, parametric and non-parametric tests were calculated, as appropriate. Two-sided level of significance was set at α < 0.05. SPSS 23.0 (PASW, SPSS; IBM, Armonk, NY) was used for the analysis.

At baseline, parameters of memory performance, glucose metabolism, HC volume and microstructure were compared between groups, using independent t-tests or Mann-Whitney U-tests, as appropriated. To detect differences between groups with regard to changes over time in these parameters, we performed repeated-measures analysis of variance (ANOVA_RM) with “time” as repeated factor (baseline, follow-up) and “group” as between-subject factor (resveratrol, placebo intervention). Changes over intervention time within groups were evaluated, using paired t-tests or Wilcoxon signed-rank tests, as appropriate. Correction for multiple comparison was applied for our primary hypothesis that resveratrol intake has beneficial effects on memory performance in comparison to placebo intake, using Bonferroni threshold (α = 0.05 divided by the number of tests per category). Secondary hypotheses comprised detection of changes in glucose metabolism and HC RSFC and structure. For all secondary analyses no adjustment for multiple testing was applied.

HC seed-based RSFC group analysis included gray matter-voxelwise GLM statistics, implemented in FSL, between changes in HC RSFC in the resveratrol group compared with those in the placebo group, using a cerebral gray matter mask (cerebellum excluded). Gaussian random field theory was used to correct for multiple comparisons at the cluster level (FSL easythresh, clusterwise correction, z > 2.3, p < 0.05).

Initially all analyses were conducted unadjusted. Additionally, analyses with 40 available datasets (memory performance and glucose control) were corrected for 4 covariates (age, sex, APOE e4 carrier status, and education) and analyses with 30 available data sets (MRI analyses) were corrected for 3 covariates (age, sex, and APOE e4 carrier status), to allow adjustment for important covariates but to avoid over-fitting Stoltzfus (2011).

To detect associations between changes in glucose parameters, HC structure and RSFC, and memory performance after the 26-week intervention, we ran bivariate correlations; Pearson or Spearman's rank correlation analysis, according to distribution of the data.

At baseline, both intervention groups were similar with regard to sex, age, years of education, cardiovascular and psychological risk factors, standard markers for lipid and inflammatory status, physical activity and MMSE scores (Table 1). The placebo group comprised a higher number of APOE e4 carriers (59%) and a higher concentration of high-sensitive CRP (2,1 pg/ml) compared to the resveratrol group (28% and 1,3 pg/ml).

Mean scores of all AVLT subtests, i.e., learning ability, delayed recall, retention and recognition, were comparable between intervention groups at baseline (see Table 2). Against our primary hypothesis no interaction effect of group × time was found for memory performance (ANOVA_RM; all ds ≤ 0.471; p ≥ 0.157, Bonferroni corrected). Adjustment for age, sex, APOE e4-carrier status and education as well as a separate analysis for male and female patients did not change the results. For details see Table 2.

At baseline, parameters of glucose control were similar in both intervention groups (see Table 3).

An interaction effect of group × time was found for HbA1c with a moderate effect size of d = 0.66 [ANOVA_RM, F_{(1, 35)} = 3.80, p = 0.059], similar after full adjustment for age, sex, APOE e4 carrier status and education [ANCOVA_RM, F_{(1, 31)} = 3.6, p = 0.066, d = 0.65]. Here, the long-term glucose marker HbA1c was significantly reduced in MCI patients after resveratrol intervention [−0.15%; paired t-test; t₍₁₇₎ = 3.3, p = 0.005, d = 1.60]; this was not the case in the placebo group (−0.02%) (Figure 2). For glucose and insulin no group × time interaction effects and no changes over time in both groups were observed. For details see Table 3.

The seed-based RSFC analysis revealed that resveratrol intervention significantly increased functional connectivity between the right HC seed and a cluster in the right angular cortex compared to placebo (Figure 3, yellow-red, cluster of 523 voxels, hot voxel: x = 60, y = −56, z = 10, p < 0.001). This effect was primarily driven by the anterior part of the right HC (cluster of 557 voxels, hot voxel: x = 60, y = −58, z = 10, p < 0.001). Results did not change after adjustment for age, sex and APOE e4 carrier status (p < 0.001) and after correction for multiple comparisons (FSL easythresh). The decrease in HbA1c did not significantly correlate with the increase in RSFC in the resveratrol group (r = −0.131, p = 0.665). We did not find significant group differences for the opposite contrast (placebo intervention > resveratrol intervention). No significant group differences were observed in correlations between the left HC seed and other gray matter brain regions.

At baseline, total gray matter volume, and volume and microstructure of the HC were similar between both groups (see Table 4).

At between-group level, total gray matter volume was similar after intervention period in both groups [ANCOVA_RM, F₍₂₈₎ = 0.272, p = 0.606, d = 0.20]. A group × time interaction effect was found for the atrophy rate of the left HC with a moderate effect size of d = 0.68 [ANCOVA_RM; F₍₂₈₎ = 3.2, p = 0.084], particularly for the left anterior HC [ANCOVA_RM; F₍₂₈₎ = 3.8, p = 0.061, d = 0.74] after no adjustment, and was still present after full adjustment for age, sex, and APOE e4 carrier status [ANCOVA_RM; left HC, F₍₂₄₎ = 2.3, p = 0.144, d = 0.57; left anterior HC, F₍₂₄₎ = 3.2, p = 0.088, d = 0.68]. Patients of the placebo group showed a decrease of −4.2% in volume of the left, specifically left anterior (−4.5%) HC, whereas the volume was preserved in the resveratrol group (+0.3%). No group × time interaction effects were found for changes in volume of the left posterior HC and the right HC of both groups. For details see Table 4 and Figure 4.

Turning to HC microstructure, no selective differences were noted between groups for changes in mean diffusivity of the HC over time (see Table 4).

The authors declare that the research was conducted in the absence of any commercial or financial relationships that could be construed as a potential conflict of interest.