Aiginition Longitudinal Biomarker Investigation of Neurodegeneration (ALBION) is a longitudinal study taking place in the Cognitive Disorders Clinic of Aiginition Hospital of the National and Kapodistrian University of Athens, and is designed to address research questions regarding the preclinical and prodromal stages of AD. A detailed description of the study protocol has been published previously (Kalligerou et al., 2019; Scarmeas et al., 2022). Briefly, study participants include people aged 40 years or older visiting the cognitive disorders' outpatient clinic of a tertiary university hospital. These participants may have concerns about their cognitive status or may be asymptomatic but committed to contributing to medical science. Exclusion criteria are diagnosis of dementia, neurological, psychiatric or medical conditions associated with a high risk of cognitive impairment or dementia, MRI contraindications, as well as the use of anticoagulant medication.

Global cognitive status was assessed using the Mini Mental State Examination (MMSE) (Folstein et al., 1975) and the Addrenbrooke's Cognitive Examination-revised (ACE-R) (Mioshi et al., 2006), while the pre-morbid level was estimated based on vocabulary. Attention, executive function, memory, language and visuospatial cognitive domains were examined from different neuropsychological tests. A detailed description is presented elsewhere (Brikou et al., 2023). Participants' raw scores on the individual neuropsychological tests for each cognitive domain were transformed to z-scores using mean and standard deviation values derived from the healthy controls of the study sample. An average domain score for each cognitive domain (i.e. attention, executive functions, memory, language and visual-spatial functioning) was produced (with a higher score indicating better cognitive performance). All tests were administered by trained neuropsychologists.

In the analyses of the current study, 120 volunteers without dementia who had full dietary and MRI data at baseline (first visit evaluation) were included.

Dietary intake was evaluated through four 24-h recalls using the multiple-pass method (Conway et al., 2004). Participants were interviewed by appropriately trained registered dietitians, and they were asked to report in detail all foods and beverages consumed the day before. The first recall was conducted in person and the subsequent ones over the phone. Three of the recalls were conducted once per week for the next 3 weeks, on weekdays and one on a weekend day to more accurately estimate usual intake throughout the week. Participants were not aware of the day of the recall in advance, so they could not change their diet in anticipation of the interview. Recall data were grouped into specific food groups, namely full- and low- fat dairy products, non-refined cereals (whole bread, pasta, rice, other grain), fruits, vegetables, potatoes, red meat and products, poultry, fish, legumes, added fats, alcoholic beverages, sweets etc.

Based on this food grouping, adherence to the MeDi was estimated using the MedDietScore (Panagiotakos et al., 2006). The MedDietScore was based on the weekly consumption of 11 food groups. For the food groups that are presumed to closely characterize the Mediterranean pattern (i.e., non-refined cereals, fruits, vegetables, legumes, potatoes, fish and olive oil), individuals who reported no consumption were assigned a score of 0, and scores of 1–5 are assigned for rare to daily consumption. For those foods that are presumed to diverge from this diet pattern (i.e., meat and meat products, poultry and full-fat dairy products), participants were assigned scores on a reverse scale. For alcohol intake, a score of 5 was assigned for consumption of <300 ml of alcohol/day, a score of 0 was assigned for no consumption or for consumption of 700 ml/day and scores of 4–1 were assigned for consumption of 600–700, 500–600, 400–500, and 300–400 mL/day, respectively. Total MedDietScore ranges from 0 to 55, with higher MedDietScore values indicating greater adherence to the MeDi pattern.

All participants underwent a brain MRI on a 3T Achieva TX Philips manufactured MRI scanner (Philips, Best, the Netherlands) equipped with an eight-channel head coil using the same imaging protocol. The imaging protocol included an anatomical high resolution 3D high T1 (3D-HR-T1) weighted sequence (repetition time (TR): 6.7 ms, time echo (TE): 3.1 ms, flip angle: 9, voxel size 1.1 × 1.1 × 1.2 mm³, sagittal orientation), a whole-brain functional T2^* weighted gradient echo combined with echo planar imaging (TR: 1,900 ms, TE: 30 ms, flip angle: 90°, acquisition voxel size 3.3 × 3.3 × 3.3 mm³ and sensitivity encoding reduction factor of two, 204 dynamic scans ~ 6 min 34 s), as well as clinically used imaging techniques such as 2D T2 turbo spin echo and 2D T2 weighted with fluid attenuated inversion recovery (FLAIR) sequence. Foam pads were positioned to participants' head to minimize voluntary motion and the participants were instructed to lie still with their eyes closed. T2, 3D T1 and T2 Flair images were reviewed by an experienced neuroradiologist (G.V.) for unexpected findings whereas fMRI data were reviewed by an MR physicist (E.K.) for motion or scanner related artifacts. Imaging data of 105 participants (out of 120) were finally analyzed due to the presence of motion artifacts and unexpected findings in 10 and five participants, respectively. Motion artifacts were identified by an MR physicist with >15 years of experience in neuroimaging data acquisition and analysis (EK) whereas unexpected findings were identified by an experience neuroradiologist (GV).

Rs-fMRI imaging data were pre- and post-processed using the Functional Connectivity (CONN) toolbox v22 of Matlab (http://www.nitrc.org/projects/conn).

Demographic, clinical, cognitive and dietary data are presented with descriptive values. For the purpose of the main aim of the study, a linear regression model was applied to investigate any association between MeDi score and FC values between different brain anatomical areas using age, sex, education, and total daily calorie intake as nuisance parameters. For these neuroimaging analyses conducted CONN toolbox v22 of Matlab, we used a voxel threshold p < 0.001 uncorrected and a cluster threshold p < 0.05 family-wise error (FWE) – corrected to account for Type I errors (Woo et al., 2014). As a supplementary exploratory analysis, we further examined whether any resulted FC were associated with participants' cognitive performance, using linear regression models (with cognitive measures as dependent variables and MedDietScore, functional connections related to MedDietScore, and their interactions as predictors). In case the final model was significant, we considered any significant independent variables that contributed to the model whereas we further ran post-hoc analysis for interactions. For this exploratory analysis conducted using IBM SPSS v. 29.0, the level of statistical significance was set at p < 0.05.

After exclusion of participants with motion artifacts (n = 10) and unexpected MRI findings (n = 5), the final sample consisted of 105 participants without dementia. The participants who were excluded from the study did not differ from the 105 participants in terms of age, sex, education, composite cognitive z-score, and total daily calorie intake. The group demographic, clinical and dietary data are summarized in Table 1. The scores of 11 food groups are presented in Supplementary Table 1.

FC analysis revealed significant associations between the MedDietScore and the FC of several anatomical areas (Table 2).

The FC between the medial prefrontal cortex and a cluster located in left postcentral gyrus and in the left anterior division of supramarginal gyrus was positively associated with the MedDietScore. Further inspection of the connectivity values revealed that the negative FC between these anatomical regions was weakened as the adherence to MeDi increased (Figure 1).

On the other hand, the FC between the medial visual and the right posterior division of both middle and superior temporal gyrus was negatively associated with the MedDietScore. Specifically, the negative FC between these anatomical regions was strengthened as the adherence to MeDi increased (Figure 2).

Furthermore, the FC between the right posterior segment of superior temporal gyrus and right occipital lobe was inversely related to the MedDietSore; the negative FC between these anatomical regions was strengthened as the adherence to MedDietSore increased (Figure 3).

The functional connections related to MeDi were significantly associated with attention, executive functions, and memory. Specifically, the functional connection between right posterior superior temporal gyrus and right occipital pole was associated with attention z-score (B = −1.474, p = 0.037), executive functions z-score (B = −1.156, p = 0.015), and memory z-score (B = −1.533, p = 0.018); the stronger the anticorrelation between these two anatomical regions the higher the cognitive performance. No other associations between functional connections related to MeDi and cognitive scores were significant at p < 0.05. Of note, the MedDietScore was not related individual cognitive domains (p > 0.05).

We found a significant interaction between MeDi and the functional connection of right posterior superior temporal gyrus and right occipital pole on attention (p-interaction = 0.042) and executive functions (p-interaction = 0.044) (Table 3). Further analysis revealed that rsFC had a significant effect on attention (B = −3.525, p = 0.005) and executive functions (B = −2.461, p < 0.001) in the high MeDi group but not in the low MeDi group (attention: B = −0.586, p = 0.523; executive functions: B = −0.198, p = 0.783) (Figure 4). No other interactions were significant at p < 0.05.

To the best of our knowledge, this is the first study analyzing adherence to MeDi and both positive and negative correlations between anatomical regions on rs-fMRI. Previous studies focused only on positive inter-network correlations (Gaynor et al., 2022a,b) but yet did not find any significant association between MeDi and overall inter-network FC. In our study, the FC values between the different anatomical areas in all the associations with MedDietScore were negative, indicating the temporal inverse correlation in spontaneous BOLD fluctuations (anti-correlation).

While positive correlations between different brain networks may play a unifying role by combining neuronal activity that supports similar goals or representations, anti-correlations might play a distinguishing role by separating neuronal processes that support opposite goals or competing representations and by facilitate task-switching from rs conditions to task performance (Dosenbach et al., 2006; Chiong et al., 2013). Although researchers have spent significant efforts to comprehend the positive connectivity in different anatomical areas, the available studies investigating the negative connectivity values are limited (Kelly et al., 2008; Hampson et al., 2010). Anti-correlation was first identified between the task-activated brain regions and the nodes of the DMN which were deactivated during the task in normal subjects (Fox et al., 2005). The existence of inverse correlation has previously been a subject of debate in terms of artificially anti-correlations introduced by a pre-processing approach to remove noise from BOLD imaging data (global signal regression) (Aguirre et al., 1998; Desjardins et al., 2001; Macey et al., 2004; Chai et al., 2012). However, the observed anticorrelations in our study were not artifacts of the global signal regression since we applied a valid component base noise reduction method (CompCor), implemented in CONN toolbox that eliminates the above artificial effect (Behzadi et al., 2007). Of note, the temporal anti-correlation between task-positive networks engaged during task executions and task-negative networks, such as the DMN, is normally found in healthy population and it is reduced in aging and along the progression of AD (Fox et al., 2005; Dennis and Thompson, 2014; Meskaldji et al., 2016; Weiler et al., 2017; Lin et al., 2018). In line with this theoretical point of view and the existence of considerable ambiguity in the interpretation of negative correlations (Murphy et al., 2009; Chai et al., 2012), the only available studies in the literature which investigated the association between MeDi and overall inter-network FC either cross-sectionally (Gaynor et al., 2022b) or longitudinally (Gaynor et al., 2022a), excluded all negative correlation values from their analysis.

Having established the validity of negative correlations in rs-fMRI analysis (Behzadi et al., 2007), we assume that our results have a neurobiological basis and potentially contribute to the understanding of neurophysiological mechanisms in aging, providing evidence for the role of MeDi in brain FC. Of note, we did not find any significant association between MedDietSore and participants' cognitive measures probably due to power issues or participants' characteristics. The fact that (a) MedDietSore was associated with FC but not directly to cognitive measures and (b) some of these functional connections were associated with participants' attention, executive and memory performance, and these associations were significantly stronger in participants with high adherence to MeDi compared to those with low adherence to MeDi, may indicate that MeDi potentially impact cognition through changes in FC and that rs-fMRI is a sensitive method to capture brain alterations. Changes in FC may reflect an intermediate phenotype before the emergence of cognitive changes. Previous studies have noted patterns of functional connectivity measured through rs-fMRI predict novel individuals' cognitive changes in a highly heterogeneous aging populating, highlighting that rs-fMRI may provide clinically relevant information about cognitive functions and add to the identification of markers that capture cognitive changes in normal and pathological aging (Lin et al., 2018). A similar pattern favoring our current findings was recently observed in another study focusing on MeDi, memory function, and structural connectivity (Christidi et al., 2024).

In our study, the negative correlation between occipital regions (medial visual, right occipital pole) and right temporal regions (right posterior segment of superior and middle temporal gyrus) was strengthened with higher adherence to MeDi. Of note the stronger the anticorrelation between superior temporal gyrus and occipital lobe, the higher the cognitive performance on tests of attention, executive functions, and memory. These functions are among the core cognitive functions that show alterations in normal and pathological aging (Petersen et al., 1992; Fernandez-Duque and Black, 2006; Bisiacchi et al., 2008). The FC between temporal and occipital lobes mediates in audiovisual integration processes, and the ability to combine information from the spoken message with the visual information from a speaker's face (Beauchamp et al., 2004; Noesselt et al., 2007, 2012; Uno et al., 2015). Right temporal regions are implicated in the integration of phonetic detail with talker information (Formisano et al., 2008; von Kriegstein et al., 2010; Evans and Davis, 2015; Luthra et al., 2023). Therefore, we could argue that distinct neural segregation of occipito-temporal networks may also be important to voice identity. There is also evidence that the role of the occipito-temporal network is crucial during high-order language processing. For example, superior and middle temporal and occipital anatomical regions are activated during reading while deactivation in the occipito-temporal network is associated with increased reading speed (Sun et al., 2024). In the same context, higher adherence to MeDi has been associated with higher language scores (Anastasiou et al., 2017) and better verbal abilities (Corley et al., 2013). We also noted a positive association between the MedDietScore and the FC of medial prefrontal cortex and left postcentral gyrus/anterior segment of the supramarginal gyrus; higher the adherence to MeDi was associated with reduced anti-correlation between these two anatomical regions. To the best of our knowledge, adherence to the MeDi has never been linked to negative effects on brain function. In addition, this pattern of FC was not related to any cognitive domain. Therefore, although these regions are implicated in several higher-order cognitive processes, including integration of sensory information, cognitive and motor response inhibition, word processing and phonological decision, ability to use tools, episodic memory encoding, and social cognition (Hartwigsen et al., 2010; Oberhuber et al., 2016; Lesourd et al., 2017; Christidi et al., 2018; Rubinstein et al., 2021; Friedman and Robbins, 2022), any direct interpretation of these findings would be speculative and thus future longitudinal studies are necessary to replicate our findings and provide a more plausible explanation.

In contrast to previous cross-sectional and longitudinal studies on MeDi and FC (Gaynor et al., 2022a,b) which focused only on positive correlation values and did not find any direct relation between MeDi and FC, we identified significant associations between adherence to MeDi and FC of specific anatomical regions. In their first study, Gaynor and colleagues did not find MeDi to be directly associated with rs-FC, yet they identified a moderating role on the relationship between inter-network FC and cognition, mostly fluid reasoning and to a lesser degree, Vocabulary (Gaynor et al., 2022b). Furthermore, in their longitudinal study, the same group found that the MeDi significantly moderated the effect of change in overall between-network and within-network FC on change in memory performance, suggestion that a healthy dietary pattern may protect memory function by attenuating the negative effects of alterations in FC over time (Gaynor et al., 2022a). Therefore, our findings are partially in line with the previous related studies and further highlight the role of MeDi on FC, since not only we found a direct association between MeDi and anticorrelating patterns of FC but we also support a moderating role of MeDi on the relationship between temporo-occipital FC and attention and executive functions. The mechanism by which MeDi exerts a role on cognition and brain structural and functional status has not been clearly defined, even though reduced inflammation and oxidative stress (Siervo et al., 2021), as well as protection against cardiovascular and metabolic diseases (Wu and Sun, 2017), have been proposed as potential mechanisms.

Our study has several strengths. Regarding the evaluation of dietary intake, four 24-h recalls were made using the multiple-pass method (Conway et al., 2004). Moreover, each of our participants underwent thorough examinations by neurologists, and their statuses were determined through diagnostic consensus meetings and after detailed neuropsychological evaluation. Additionally, all imaging data acquisition and protocols were supervised by experienced neuroscientists. On the other hand, the present study suffers several limitations. Our study includes a relatively small sample posing power limitations and not allowing the generalizability of the results to a larger population. Due to power issues, complex models including additional covariates (e.g., lifestyle factors) or testing interaction effects (e.g., with age and/or sex) were not applied. Future studies may also examine the moderating role of other lifestyly factors, including physical activity. The participants were selected from a referral-based recruitment of a specific clinic, which may not be indicative of the population at large and may cause possible bias. Residual confounding cannot be excluded although we took into consideration important demographic confounders in the relationships explored. Another limitation is the cross-sectional design; we included baseline data from a longitudinal study and although significant, the observed findings provide associations but not causality between MedDietScore and patterns of FC.