Mediterranean Diet, Alzheimer Disease Biomarkers, and Brain Atrophy in Old Age

Abstract

Objective

To determine whether following a Mediterranean-like diet (MeDi) relates to cognitive functions and in vivo biomarkers for Alzheimer disease (AD), we analyzed cross-sectional data from the German DZNE-Longitudinal Cognitive Impairment and Dementia Study.

Method

The sample (n = 512, mean age 69.5 ± 5.9 years) included 169 cognitively normal participants and individuals at higher AD risk (53 with relatives with AD, 209 with subjective cognitive decline, and 81 with mild cognitive impairment). We defined MeDi adherence according to the food frequency questionnaire. Brain volume outcomes were generated via voxel-based morphometry on T1-MRI, and cognitive performance was assessed with an extensive neuropsychological battery. AD-related biomarkers (β-amyloid_42/40 [Aβ_42/40] ratio, phosphorylated tau 181 [pTau181]) in CSF were assessed in n = 226 individuals. We analyzed the associations between MeDi and outcomes with linear regression models controlling for several covariates. In addition, we applied hypothesis-driven mediation and moderation analysis.

Results

Higher MeDi adherence related to larger mediotemporal gray matter volume (p < 0.05 family-wise error corrected), better memory (β ± SE = 0.03 ± 0.02; p = 0.038), and less amyloid (Aβ_42/40 ratio, β ± SE = 0.003 ± 0.001; p = 0.008) and pTau181 (β ± SE = −1.96 ± 0.68; p = 0.004) pathology. Mediotemporal volume mediated the association between MeDi and memory (40% indirect mediation). Finally, MeDi favorably moderated the associations among Aβ_42/40 ratio, pTau181, and mediotemporal atrophy. Results were consistent correcting for APOE-ε4 status.

Conclusion

Our findings corroborate the view of MeDi as a protective factor against memory decline and mediotemporal atrophy. They suggest that these associations might be explained by a decrease of amyloidosis and tau pathology. Longitudinal and dietary intervention studies should further examine this conjecture and its treatment implications.

Healthy dietary patterns such as the Mediterranean diet (MeDi) might reduce the risk of dementia and cognitive decline.^1-4 Although contrasting findings have been reported as well,^5,6 encouraging results were provided by the Prevención con Dieta Mediterránea (PREDIMED) study, a randomized clinical trial in which a MeDi intervention was associated with both improved cognitive functioning⁷ and reduced incident mild cognitive impairment (MCI).⁸ Likewise, adherence to MeDi could diminish the conversion rate from MCI to dementia.^9,10

At the biomarker level, MeDi has been associated with preserved cortical thickness and brain volume in middle-aged^11,12 and older individuals,^13-15 especially in brain regions associated with aging and Alzheimer disease (AD). Moreover, adherence to MeDi has been related to lower amyloid load studied with ¹¹C-Pittsburgh compound B-PET in cognitively unimpaired individuals,^11,16,17 while another study could not find such an association using ¹⁸F-florbetaben-PET.¹⁸ Furthermore, 1 study found an association in volunteers with subjective cognitive decline (SCD) or MCI between MeDi and lower FDDNP-PET, a compound measure of amyloid and tau pathology.¹⁹ Two longitudinal studies reported better MeDi adherence to be associated with less amyloid accumulation over time.^17,20

This initial evidence suggests that MeDi might reduce amyloid deposition since midlife with a probable downstream effect on neurodegeneration and cognition. We in addition hypothesized that MeDi is associated with tau levels and moderates the associations among amyloid, tau, and brain atrophy. Here, we examined these questions by leveraging a large cohort of older individuals at increased risk for AD.

Methods

Standard Protocol Approvals, Registrations, and Patient Consents

At each DZNE-Longitudinal Cognitive Impairment and Dementia Study (DELCODE) site, the local institutional review boards approved the study protocol, and the ethics committees issued local ethics approval. DELCODE is registered at the German Clinical Trials Register (DRKS00007966; April 5, 2015). The study protocol followed the ethics principles for human experimentation in accordance with the Declaration of Helsinki. All participants in the study provided written informed consent.

Participants

As of July 2020, the baseline of the German multicenter DELCODE includes 1,079 individuals. A complete overview of the study design, group definitions, and aims is provided in a previous paper.²¹ Here, we selected 512 individuals (average ± SD age 69.49 ± 5.86 years, 270 female, self-reported sex) according to availability of both the detailed food frequency questionnaire (FFQ) and T1-weighted MRI. The sample was enriched for risk of AD in that it included individuals with SCD (n = 209, 41%) or amnestic MCI (n = 81, 16%) who were referrals to the participating memory clinics. Participants with SCD reported self-perceived cognitive decline with concerns while showing a preserved performance in all tests of the Consortium to Establish a Registry for Alzheimer's Disease (CERAD) neuropsychological battery (above −1.5 SDs compared to age-, sex-, and education-adjusted norms). Conversely, individuals with amnestic MCI performed below −1.5 SDs on the delayed-recall trial of the CERAD word-list episodic memory tests. The clinical diagnoses were part of the clinical workup at each site (not of DELCODE itself) and conformed to published research criteria.^22-24 In addition, first-degree relatives of patients with AD (n = 53, 10%) and cognitively normal volunteers without increased risk for AD (n = 169, 33%) were recruited with an advertisement campaign in local newspapers. Both groups met the requirement for an unimpaired cognitive performance on the CERAD battery (as the SCD group).

Complete demographic information is reported in table 1 and stratified by clinical group in table e-2, doi.org/10.5061/dryad.6t1g1jwxg. A subsample of 226 participants in addition underwent lumbar puncture for assessment of AD-related neuropathologic biomarkers in CSF. Comparing the groups with and without CSF information, we did not find differences in age, sex distribution, prevalence of APOE ε4, body mass index (BMI; kilocalories per day), level of physical activity (as measured with the Physical Activity Scale for the Elderly),²⁵ or MeDi score. However, individuals with CSF data available had a lower educational attainment, a higher prevalence of MCI, and accordingly a lower performance on the Mini-Mental State Examination (table e-1).

Table 1.

Demographic and Basic Clinical Characteristics (n = 512)

MRI Acquisition

The acquisition of structural brain images was performed with 3T MRI scanners mounting 32-channel head array coils. A 3D T1-weighted magnetization prepared rapid gradient echo sequence was used, with an echo time of 4.37 milliseconds, repetition time of 2,500 milliseconds, inversion time of 1,100 milliseconds, and flip angle of 7°. All images had a 1-mm³ isotropic nominal image resolution with a final image matrix of 256 × 256 × 192. Four different MRI scanners from Siemens (Siemens Healthcare, Erlangen, Germany) were used across centers: MAGNETOM TrioTim (n = 209), Verio (n = 163), Skyra (n = 110), and Prisma (n = 30). Image quality assessment is described in the supplements (doi.org/10.5061/dryad.6t1g1jwxg).

Cognitive Assessment

All study participants underwent an in-depth neuropsychological assessment to cover a broad spectrum of cognitive functioning.²¹ Our analysis focused on 5 factor scores derived from a confirmatory factor analysis and capturing the cognitive performance in different domains: memory, language, executive functions, working memory, and visuospatial abilities. Rationale and methods for the definition of factor scores are described in a previous paper.²⁶ A list of the cognitive tests contributing to each cognitive domain is given in table e-3, doi.org/10.5061/dryad.6t1g1jwxg/.

Dietary Assessment and MeDi Score Definition

We administered the German adaptation of the semiquantitative European Prospective Investigation of Cancer (EPIC) FFQ²⁷ (more details in supplements, doi.org/10.5061/dryad.6t1g1jwxg/). Our sample of 512 participants did not include those who reported abnormal daily energy intake, defined as <500 or >5,000 kcal/d (n = 4) and individuals who did not answer more than 20% of the FFQ questions (n = 2).

We computed the a priori MeDi score on the basis of sex-specific medians from this study population. Briefly, food items from the EPIC-FFQ were clustered into 9 food categories. A score of 1 was assigned when the food intake for 1 participant was equal to or above the sex-specific median for 6 food categories typical of MeDi (fish, vegetables, fruits/nuts, legumes, cereals, and higher ratio of monounsaturated/saturated fats) or below the cutoff for foods nontypical of MeDi (meat, dairy products). For alcohol, a moderate consumption (10–50 g/d in men and 5–25 g/d in women) was considered beneficial and scored 1 point. The final MeDi score can span from 0 to 9, with higher values indicating higher adherence.²⁸ Table e-4 and figure e-3 (doi.org/10.5061/dryad.6t1g1jwxg/) display each food category stratified by MeDi score (low, medium, high) and sex.

CSF Sampling and Assessment

A subsample of 226 participants consented to undergo lumbar puncture. All procedures were guided by German Center for Neurodegenerative Diseases standard operating procedures (see supplementary Methods, doi.org/10.5061/dryad.6t1g1jwxg/). We focused our analyses on phosphorylated tau 181 (pTau181), β-amyloid_1–42 (Aβ₄₂), their ratio Aβ₄₂/pTau181, and the Aβ_42/40 ratio to take into account individual differences in overall Aβ peptide concentrations.²⁹

Voxel-Based Morphometry Analysis

We applied voxel-based morphometry³⁰ to study the relationship between gray matter volume and MeDi. All analyses were performed with the Computational Anatomy Toolbox (CAT12) and Statistical Parametric Mapping (SPM12, Wellcome Trust Centre for Neuroimaging, University College London, UK) running on Matlab 2014b (The MathWorks, In., Natick, MA). All T1-MRI images were normalized to the Montreal Neurological Institute standard space and segmented into gray matter, white matter, and CSF compartments. Modulation of preprocessed MRI images included both linear and nonlinear deformations (i.e., jacobian determinants) to account for contractions and expansions during image normalization. Image smoothing was applied with a 8-mm full width at half-maximum gaussian kernel. Total intracranial volume and total gray matter volume were extracted from CAT12 output.

The association between MeDi score and gray matter volume was investigated via application of the general linear model (1-sample t test in SPM12) with age, sex, total intracranial volume, and MRI scanner type entered as nuisance covariates. Heterogeneity in MRI devices was expressed with one-hot encoding for categorical data to avoid order effects. In addition, we reran the analysis correcting also for caloric intake, BMI, physical activity levels, and APOE ε4 status. The model was first applied at the whole-brain level, without any a priori hypothesis, and then restricted to hypothesis-driven regions of interest (ROIs) in the mediotemporal lobe, which shows early changes in AD.³¹ Anatomic ROIs were selected from the Automated Anatomical Labeling atlas using the Wake Forest University Pickatlas tool for SPM (bilateral hippocampi and parahippocampal gyri). Of note, the entorhinal cortex is included in the parahippocampal gyrus ROI as defined in the Automated Anatomical Labeling atlas (figure e-4, doi.org/10.5061/dryad.6t1g1jwxg/). Correction for multiple comparisons was performed with the nonparametric threshold free cluster enhancement approach implemented in SPM (neuro.uni-jena.de/tfce/). We used the threshold free cluster enhancement technique with 5,000 permutations, weighting parameters for cluster extent E = 0.6 and height H = 2 and a significance level of p < 0.05 (family-wise error [FWE] corrected).

Statistical Analysis on CSF Variables and Cognitive Factors

We assessed the associations between MeDi and cognition or CSF variables with linear regression models adjusted for age, sex, and education. The analysis was repeated including supplementary covariates to control for potential confounding effects from BMI, caloric intake, and physical activity, as well as for APOE ε4. Outliers identified on CSF variables were removed from the analysis, leading to the exclusion of 12 individuals who had values at 1.5 multiplied by the interquartile range below or above the 25th or the 75th percentile, respectively. Figure e-2, doi.org/10.5061/dryad.6t1g1jwxg/, displays the distributions of CSF variables. We repeated the analysis without outlier exclusion (applying log transformation to pTau181) and with robust linear regression, which is less sensitive to outliers. Finally, all linear models were corrected for the time distance between baseline visit (when biomarkers and cognitive assessment took place) and FFQ questionnaire (mean ± SD 41.5 ± 43.17 weeks, median 51.7 weeks).

Mediation Analysis

We created hypothesis-driven models and tested them with mediation and moderated mediation analysis. All models were created with processR and estimated with the lavaan package (version 0.6-5, lavaan.ugent.be/) in R version 3.6.3 (R Foundation for Statistical Computing, Vienna, Austria).

The aim of model 1 was to investigate the interplay among MeDi, brain volume, and memory function. Specifically, we hypothesized that the brain changes observed in the bilateral hippocampi and parahippocampal regions mediate the association between MeDi and memory identified in the regression analyses (figure 2). The model included all the 512 individuals in the study. Gray matter values were extracted from the significant cluster from the ROI-based analysis using MarsBaR toolbox for SPM. To assess the specificity of the mediation effect for mediotemporal regions, we replicated a similar mediation model using total gray matter volume as mediator. A parameter to model the indirect effects of MeDi on memory via brain measures was included.

Figure 2. Graphical Display of Mediation and Moderated Mediation Models.

Names of the paths and associated regression estimates are reported. Solid lines represent significant paths according to confidence intervals generated with bias corrected bootstrap with 10,000 replicates. Dashed lines mark nonsignificant regression paths. For models 2.1 and 2.2, in addition to the statistical models, the conceptual models are shown in the upper right corners, and simple slopes representing the interactions effects are shown below. A complete overview of direct and indirect effects is reported in table 4. Aβ_42/40 = ratio between β-amyloid₄₂ and β-amyloid₄₀; MeDi = Mediterranean diet; MEM = memory summary factor score; pTau = phosphorylated tau; rGMV = regional gray matter volume in bilateral hippocampi and parahippocampi.

We then designed additional models to disentangle the moderation effect of MeDi on the associations between Aβ_42/40 ratio and pTau181 and brain volume in mediotemporal regions. In particular, we adopted the theoretical framework of the amyloid cascade hypothesis according to which amyloidosis is the earliest upstream pathologic event that leads to tau phosphorylation and finally to brain atrophy.³² The following models were therefore performed on the subsample with CSF information. The rationale for these models is that MeDi adherence might sustain brain maintenance, thus reducing the development of disease-related brain changes and pathology.³³ In particular, we expected that MeDi moderates the paths connecting neuropathology and brain atrophy as defined by the amyloid cascade model. First, we tested a mediation model reflecting the amyloid cascade hypothesis itself, ie, Aβ_42/40 → pTau181 → brain volume (model 2.0). Then, we tested 2 additional models in which MeDi score was added as a moderator either of the path connecting Aβ_42/40 to pTau181 (model 2.1, first-stage mediation) or of the path connecting pTau181 to brain volume (model 2.2, second-stage mediation). This analysis allows us to test whether the associations between Aβ_42/40 and pTau181 and between pTau181 and brain volume vary at different levels of MeDi. A schematic visualization of the models is presented in figure 2.

In all models, we included age, sex, and education level as background confounds, and brain measures were also corrected for total intracranial volume. In addition, we tested the influence of APOE ε4 as a covariate. The significance of the associations was based on confidence intervals (CIs) generated with bias corrected bootstrap with 10000 replicates. In the moderated mediation models, all predictors were mean centered. For models 2.1 and 2.2, direct and indirect effects were evaluated at different levels of the moderator (i.e., MeDi) using the mean ± 1 SD approach. In addition, we report the index of moderated mediation, which reflects whether the indirect effects vary at different levels of the moderator.

Exploratory Analysis of MeDi Diet Components

To explore the individual contribution of each of the 9 MeDi score components, we ran additional linear regression models. Dependent variables were the memory factor score, brain volume in hippocampal and parahippocampal regions, pTau181, or Aβ_42/40 ratio. In each model, we entered all dichotomous MeDi components at once, correcting for age, sex, education, caloric intake, BMI, and physical activity.

Data Availability

Anonymized data generated and analyzed in the current study will be made available on reasonable request from qualified investigators.

Results

Brain Volume

Whole-Brain Results

The MeDi score showed a significant positive association with brain gray matter volume in the right parahippocampal gyrus and right hippocampus (p < 0.05 FWE corrected). The opposite contrast did not show any negative association. Results are shown in figure 1A and in table 2. Figure e-1 (doi.org/10.5061/dryad.6t1g1jwxg/) shows the results corrected with the less conservative p < 0.05 false discovery rate approach.

Figure 1. Positive Association Between MeDi and Brain Volume.

(A) Positive association between Mediterranean diet (MeDi) score and brain gray matter volume at the whole-brain level. (B) Positive association between MeDi score and gray matter volume in a priori–defined regions of interest (ROIs) covering the bilateral hippocampi and parahippocampal gyri. All results are corrected for age, sex, total intracranial volume, and MRI scanner heterogeneity. Results are shown at p < 0.05 family-wise error corrected. Images are displayed in neurologic convention: left of the brain on the left of the image. Unthresholded T map is available at Neurovault (neurovault.org/collections/KMIELIOW/).

Table 2.

Montreal Neurological Institute Coordinates and Statistics From Neuroimaging Analysis

ROI-Based Results

Restricting the analysis to a priori ROIs revealed a bilateral association between higher MeDi and increased gray matter volume in hippocampi and parahippocampal gyri (p < 0.05 FWE corrected). Of note, we observed also in this analysis a right > left asymmetry (figure 1B and table 2). The opposite contrast did not reveal any negative association. Of note, a 1-point increase in MeDi corresponds to an increase in brain volume in the significant cluster associated with −0.84 years of age. The results of whole-brain and ROI-based analyses were stable correcting for caloric intake, BMI, physical activity, and APOE ε4 status. The unthresholded T maps of whole-brain models are available at Neurovault (neurovault.org/collections/KMIELIOW/).

Cognition

The models adjusted for age, sex, and education showed an association between MeDi and both memory (F_4,507 = 57.87, p < 0.001, R² = 0.31) and language (F_4,507 = 59.22, p < 0.001, R² = 0.32) but not for the other domains (table 3). In the models also corrected for BMI, caloric intake, and physical activity, only the association between an increased adherence to MeDi and an improved memory performance remained (F_7,482 = 30.57, p< 0.001, R² = 0.31). Here, a 1-point increase of MeDi corresponded to an increase of memory performance associated with almost −1 year of age. Correcting for APOE ε4 and time distance between baseline visit and FFQ did not change the results (table 3 and table e-7, doi.org/10.5061/dryad.6t1g1jwxg/).

Table 3.

Associations Among MeDi Score, Cognitive Outcomes, and CSF Biomarkers

CSF Biomarkers

The linear regression models showed significant associations of MeDi with pTau181 (F_4,209 = 6.02, p < 0.001, R² = 0.103), Aβ_42/40 (F_4,209 = 6.15, p < 0.001, R² = 0.105) and Aβ₄₂/pTau181 (F_4,209 = 6.29, p < 0.001, R² = 0.107).

The associations of MeDi with pTau181 (F_7,197 = 4.118, p < 0.001, R² = 0.128), Aβ_42/40 (F_7,197 = 3.509, p = 0.0014, R² = 0.111), and Aβ₄₂/pTau181 (F_7,197 = 3.933, p < 0.001, R² = 0.123) were stable in addition controlling for BMI, caloric intake, and physical activity (table 3). Higher adherence to MeDi showed associations with pTau181 and both Aβ_42/40 and Aβ₄₂/pTau181 ratios. Specifically, in the adjusted models, a unity increase in MeDi score was associated with a decrease of 1.96 pg/mL in pTau181 and with an increase of 0.0027 and 0.71 in Aβ_42/40 and Aβ₄₂/pTau181 ratios, respectively. For comparison, a 1-point increase in MeDi corresponded to a decrease of the neuropathologic burden on Aβ_42/40 and pTau181 associated with >−3 years of age (−3.5 and −3.33 years, respectively). Correcting for APOE ε4 reduced the associations between MeDi and CSF biomarkers for amyloid (but showing a consistent pattern of results, table 3), while the time distance between baseline visit and FFQ did not influence the results (table e-7, doi.org/10.5061/dryad.6t1g1jwxg/). We observed very similar results in the analysis without outlier exclusion and using both linear and robust linear regressions (table e-5, doi.org/10.5061/dryad.6t1g1jwxg/).

Mediation Models

Model 1 revealed a significant indirect effect of MeDi on memory via brain volume in hippocampal and parahippocampal regions (effect estimate (est) 0.017, CI 0.007–0.03). Notably, the direct effect of MeDi on memory was no longer significant (est 0.025, CI −0.005 to 0.056), thus suggesting complete mediation. The indirect pathway representing the effect of MeDi on memory via hippocampal and parahippocampal volume accounted for 40% of the total effect. The replication of model 1 using total gray matter volume showed a significant direct effect, while the indirect effect was weak and accounted for only 4.6% of the total effect (table e-6, doi.org/10.5061/dryad.6t1g1jwxg/).

Model 2.0 showed a complete mediation of Aβ_42/40 on brain volume through pTau181 in that only the indirect effect (est 0.109, CI 0.009–0.0239) was significant and explained 34% of the total effect. In model 2.1, we observed a significant index of moderated mediation (est −0.02, CI −0.065 to −0.001) and significant indirect effects at all levels of the moderator. The indirect effect was larger for lower values of MeDi and decreased for higher MeDi score. The proportion of the total effect mediated by the Aβ_42/40 → pTau181 → brain volume path at different levels of MeDi was 39% at −1 SD, 32% at the mean level, and 23% at +1 SD. Model 2.2 showed a significant index of moderated mediation (est −0.047, CI −0.101 to −0.004) and a significant indirect effect only at the lowest level of the moderator, ie, at −1 SD. Complete details are displayed in table 4. All mediation and moderated-mediation models showed consistent results when corrected for APOE ε4 (table 4).

Table 4.

Result of Mediation and Moderated-Mediation Models

Individual Contributions of MeDi Components

Table e-9 (doi.org/10.5061/dryad.6t1g1jwxg/) displays the results of the exploratory analysis on individual MeDi components. With memory function as a dependent variable, we observed a significant positive association only for cereals (p = 0.013). Congruently, only cereals showed a marginally significant positive association with mediotemporal volume (p = 0.056). For both pTau181 and Aβ_42/40 ratio, a significant association was found with the ratio of monounsaturated/saturated fat (p = 0.021 and p = 0.038, respectively). Specifically, an increased ratio of monounsaturated/saturated fat was associated with increased levels of Aβ_42/40 and decreased burden of pTau181.

Discussion

Overall, our results suggest that the favorable association between MeDi adherence and memory performance, found here as in many previous studies, could be mediated by preservation of brain volume in mediotemporal regions. Moreover, we show that MeDi adherence is inversely associated with both pathologic biomarkers for amyloidosis and tauopathy, which underlie AD. Finally, our data show that a healthier diet moderates the associations among Aβ_42/40, pTau181, and brain atrophy, suggesting that MeDi contributes to brain maintenance.³³

First, we observed a significant association between MeDi and hippocampal and parahippocampal regions in both whole-brain and ROI-based analyses. This is in line with studies that reported positive associations between MeDi and brain morphology in cognitively normal middle- and old-aged individuals and in elderly individuals without dementia.^11-15 However, 1 study reported no significant association between MeDi and brain volume,³⁴ and another reported an association only with meat consumption but not with MeDi as a whole.³⁵Compared to these studies, we analyzed a larger sample enriched for AD risk, thus possibly making our analysis more sensitive to capture brain structural variations related to MeDi. Moreover, in both negative studies, there was a larger temporal distance between dietary and MRI data assessments (5 and 9 years), which might have influenced the results. Several hypotheses could be advanced concerning the link between diet and brain structural integrity. Considering our moderated mediation results, we hypothesize that the adherence to MeDi protects brain structures from the adverse effects of upstream pathologic events, i.e., accumulation of amyloid plaques and tau phosphorylation. This hypothesis would clarify why the association between MeDi and brain structure is specific for the mediotemporal regions: AD-related atrophy starts in these regions and colocalizes with tau accumulation.

The second main finding is the favorable association between MeDi and memory performance. In particular, we show a significant positive association between diet and a composite memory factor score, which, capitalizing on an in-depth memory assessment, was used to quantify the level of memory performance in our sample.²⁶ This finding replicates previous work performed on a smaller interim release of DELCODE³⁶ and is in agreement with the view of MeDi as a protective lifestyle factor against cognitive decline and dementia.^1-3 Despite a protective effect of MeDi being reported for general cognition and for different cognitive domains, memory seems to be the one that benefits more from a healthy diet,^15,37,38 in line with the region-specific association with brain volume. The analysis of the individual MeDi score components showed a significant association between memory and cereals. This supports previous studies showing a protective effect of cereals, in particular whole grains, on cognition.^37,39 We propose that the specificity of our findings for the memory domain should be interpreted in light of the mediation analysis showing that the mediotemporal volume mediates the association between MeDi and memory. Of note, the mediation effect was specific for the mediotemporal regions in that the mediating effect of total gray matter volume was very weak.

Finally, the analysis of the subsample with CSF information allowed us to investigate the associations between MeDi and AD-related biomarkers and to model their interplay with brain volume. First, we reported that MeDi is associated with lower levels of amyloid pathology as expressed by the Aβ_42/40 ratio and with reduced pTau181. In agreement with our observations, previous studies in middle- and old-aged, cognitively normal individuals reported that diet is associated with reduced amyloid pathology levels and amyloid accumulation as studied with ¹¹C-Pittsburgh compound B-PET assessments.^17,20 Of note, we observed a significant association between MeDi and Aβ_42/40 ratio but not with Aβ₄₂. Previous studies suggested that the Aβ_42/40 ratio is a more sensitive biomarker for AD compared to Aβ₄₂.²⁹ Moreover, a recent study on a cell culture model of AD showed the relevance Aβ_42/40 ratio, but not total amyloid, as driver of tau pathology.⁴⁰ The mediation model 2.0 is in line with the amyloid cascade hypothesis, showing a link among Aβ_42/40, pTau181, and brain atrophy.³² Then, in models 2.1 and 2.2, we showed that MeDi exerts a significant moderation effect both on the association between Aβ_42/40 ratio and pTau181 and, to a lesser extent, on that between pTau181 levels and brain atrophy, specifically mitigating their associations. However, these models should be interpreted with caution because they rely on cross-sectional data and cannot therefore prove causal pathways. A possible (and speculative) mechanistic interpretation of these observations is that MeDi acts on the triggers that connect these pathologic events, e.g., inflammation⁴¹ and oxidative stress.⁴² MeDi is indeed based on higher consumption of fruits and vegetables, whole grains, fish, and olive oil, which are known for their anti-inflammatory and antioxidant actions.⁴³ Future studies could include markers for inflammation or oxidative stress to test more fine-grained hypotheses concerning the underlying biological processes.

Notably, the exploratory analysis of the individual MeDi components showed a beneficial association between the ratio of monounsaturated/saturated fat and both pTau181 and Aβ_42/40 ratio. Monounsaturated fats are found in many food sources such as plant oils, nuts, seeds, and animal products, and a combination of them likely accounted for the total level in our study. In Mediterranean regions, higher scores of monounsaturated/saturated fat ratio most likely reflect higher consumption of extravirgin olive, which has been associated with reduced AD pathology in mice⁴⁴ and with better cognitive performance in human participants in the PREDIMED trial.⁸

A strength of the present study is the availability of multiple data types, which enabled the integration of dietary information, cognitive data, brain morphometry, and CSF biomarkers. This allowed us to model not only the associations between MeDi and the single variables of interest but also their interplay. Another strength is that the sample is enriched for AD risk. While this constrains generalization to the older population at large, it allows studying the interaction of diet with substantial variation of amyloid, tau, and brain neurodegeneration in a group who could be a target for nutritional intervention trials. We also repeated the regression models excluding individuals with MCI, the highest-risk clinical group. This showed a stable association of MeDi with mediotemporal brain volume but not with other outcomes, pTau181, Aβ_42/40 ratio, and memory (table e-10, doi.org/10.5061/dryad.6t1g1jwxg/). This might indicate that the beneficial associations between MeDi and AD-related biomarkers and cognition are more pronounced in the prodromal AD stages. However, these negative findings might also be attributable to reduced power in the subsample analysis and to lower variability in the outcomes.

A limitation of the present cross-sectional study is that it does not allow causal inference. However, MeDi scores are stable over years in older adults, even in the years before a diagnosis of incident dementia,^1,45 and Wagner et al⁴⁶ showed that the longitudinal trajectories of MeDi over 15 years are comparable between women who showed cognitive decline and those who did not in the Nurses' Health Study. Therefore, we posit that MeDi adherence reflects the past aggregate exposure to the MeDi ingredients, so the statistical associations with MeDi described above could result from accumulated long-term causal effects of diet. The extension to longitudinal data, including data from DELCODE follow-ups, should be the next step to address this limitation and to validate the proposed models. Moreover, it has to be noted that the analysis of the single components presented here is exploratory and should be validated by more focused studies. Future studies in humans and animal models could focus on specific hypothesis-driven dietary components and leverage on modern techniques to directly measure their effects on the metabolome and microbiome.⁴⁷ Along the same line, recent efforts to map the chemical complexity of diets provide a promising avenue for a deeper understanding of the effects of diet on health and disease.⁴⁸ It has to be mentioned that previous studies reported an association between different dietary patterns (i.e., Western diet and the Alternative Healthy Eating Index 2010) and risk of dementia and cognitive decline⁴⁹ or AD-related markers such as hippocampal volume.⁵ This might bring into question whether the results reported in our study are specific for MeDi or rather reflect a more general advantage of a healthy diet. This is linked to another limitation of our and similar studies in which MeDi adherence is defined on sample medians, thus representing the relative adherence to dietary guidelines and not the high consumption of beneficial foods in absolute terms as in Mediterranean regions. Moreover, it is possible that MeDi has systemic effects on health (e.g., modulating inflammation or cardiovascular health⁵⁰) that might in turn influence AD-specific mechanisms. Our results were stable when we controlled for factors associated with cardiovascular risk (BMI, physical activity, and smoking, see table e-8, doi.org/10.5061/dryad.6t1g1jwxg/), but a deeper investigation of this topic is needed. The study of many other biomarkers such as diffusion tensor imaging, resting-state functional connectivity, and markers for neuroinflammation, especially in longitudinal study design, could help generate a more comprehensive and mechanistic understanding of the effects of MeDi on cognition in old age and early AD.

Our study supports the view of MeDi as a protective lifestyle factor against AD-related neurodegeneration and memory impairment. Longitudinal studies with AD biomarker outcomes could further examine this conjecture and pave the way for dietary interventions to delay AD.

Glossary

Aβ

β-amyloid

AD

Alzheimer disease

BMI

body mass index

CAT12

Computational Anatomy Toolbox

CERAD

Consortium to Establish a Registry for Alzheimer's Disease

CI

confidence interval

DELCODE

DZNE-Longitudinal Cognitive Impairment and Dementia Study

EPIC

European Prospective Investigation of Cancer

est

effect estimate

FFQ

food frequency questionnaire

FWE

family-wise error

MCI

mild cognitive impairment

MeDi

Mediterranean diet

PREDIMED

Prevención con Dieta Mediterránea

pTau181

phosphorylated tau 181

ROI

region of interest

SCD

subjective cognitive decline

SPM12

Statistical Parametric Mapping

Appendix. Authors

Appendix 2. Coinvestigators

Study Funding

Study funded by the German Center for Neurodegenerative Diseases (Deutsches zentrum für Neurodegenerative Erkrankungen [DZNE]) and by the Diet Body Brain research cluster funded by the German Federal Ministry of Education and Research (grant 01EA1809C).

Disclosure

T. Ballarini, D. Melo van Lent, J. Brunner, A. Schröder, S. Wolfsgruber, S. Altenstein, F. Brosseron, K. Buerger, P. Dechent, L. Dobisch, E. Düzel, B. Ertl-Wagner, K. Fliessbach, S. Dawn Freiesleben, I. Frommann, W. Glanz, D. Hauser, J. D. Haynes, M. T. Heneka, D. Janowitz, I. Kilimann, C. Laske, F. Maier, C. D. Metzger, M. H. Munk, R. Perneczky, and O. Peters report no disclosures relevant to the manuscript; J. Priller received fees for consultation, lectures, and patents from Neurimmune, Axon, Desitin, and Epomedics. A. Ramirez, B. Rauchmann, N. Roy, K. Scheffler, A. Schneider, A. Spottke, E. J. Spruth, S. Teipel, R. Vukovich, and J. Wiltfang report no disclosures relevant to the manuscript; F. Jessen received fees for consultation from Eli Lilly, Novartis, Roche, BioGene, MSD, Piramal, Janssen, and Lundbeck. M. Wagner reports no disclosures relevant to the manuscript. Go to Neurology.org/N for full disclosures.