Introduction

More than 35.6 million people were living with dementia worldwide in 2010 (Prince et al. 2013). Due to the increased life-expectancy, worldwide prevalence of dementia is projected to be 115.4 million in 2050 (Prince et al. 2013). Alzheimer′s disease (AD), vascular dementia (VD) and a mixed neurodegenerative–vascular dementia are the most common types of dementia (Jellinger and Attems 2010). There is not cure for dementia. Vascular dementia can be delayed or even prevented by treatment or management of the predisposing vascular factors (Chang-Quan et al. 2011; Exalto et al. 2012), but currently, there is no available treatment for delaying the onset or the progression of Alzheimer disease. Moreover, the only dementia with specific therapies is the dementia caused by Alzheimer′s Disease, and these treatments offer only symptomatic relief (Birks 2006; McShane et al. 2006). Thus, dementia is a public health priority, and prevention is an essential strategy to deal with it.

Apolipoprotein E (ApoE) is the most important susceptibility locus for late-onset AD (Burke and Roses 1991), and it has been also related to a higher risk of VD (Liu et al. 2012; Yin et al. 2012). Most recently, genome-wide association studies (GWAS) have identified single nucleotide polymorphisms (SNPs) at the CLU, CR1 and PICALM loci that increase the risk of late-onset AD (Harold et al. 2009; Lambert et al. 2009). A higher educational level and higher engagement in physical, cognitive and social activities are lifestyle factors that may reduce the risk of dementia or cognitive decline (Hamer and Chida 2009; Kareholt et al. 2011; Morgan et al. 2012; Niti et al. 2008; Sattler et al. 2012). Nutritional epidemiology has examined the potential benefit on cognition of fatty acids (van Gelder et al. 2007), vitamins (Mangialasche et al. 2010; Tangney et al. 2009), fish (van Gelder et al. 2007), fruit and vegetables (Hughes et al. 2010) with favorable but not fully consistent results. Thus, both genetic and lifestyle-related factors are involved in cognitive decline and dementia, but these factors do not contribute independently to neurodegeneration. In fact, the development and also the phenotype of neurodegenerative processes are a result of complex gene–environment interactions. The interaction between ApoE genotype and physical and cognitive activities (Niti et al. 2008) and some nutrients (Barberger-Gateau et al. 2007) has been examined with conflicting results.

Over the last few years, interest in overall dietary patterns has risen to a level comparable to the interest in nutrients and foods. The Mediterranean diet (MedDiet) pattern is characterized by olive oil as the main culinary fat, high intake of plant-based, moderate-to-high consumption of fish and seafood, moderate-to-low intake of dairy products, low intake of meat or meat products and regular but moderate intake of red wine during meals (Willett et al. 1995). Compelling evidence worldwide supports a benefit effect of this emerging nutritional pattern on health (Roman et al. 2008). The MedDiet has provided favorable results on cognitive function, cognitive decline and dementia prevention along several observational studies (Scarmeas et al. 2006; Psaltopoulou et al. 2008; Gu et al. 2010; Feart et al. 2009; Scarmeas et al. 2009a, b; Roberts et al. 2010; Tangney et al. 2011; Kesse-Guyot et al. 2013; Samieri et al. 2013; Tsivgoulis et al. 2013). Recently, the PREDIMED-NAVARRA primary prevention trial has confirmed the benefit on cognitive function and mild cognitive impairment development (MCI) of a nutritional intervention with MedDiet supplemented with extra-virgin olive oil (EVOO) or nuts compared with a low-fat control diet (Martinez-Lapiscina et al. 2013a, b). However, MedDiet–gene interactions have not been broadly evaluated in observational studies, and when they were assessed, this assessment only included populations outside of the Mediterranean basin, only evaluated ApoE4 genotypes, and the results were negative (Cherbuin and Anstey 2012; Roberts et al. 2010). To our knowledge, no previous study has ever assessed gene–diet interactions on cognition in the context of a long-term trial, where subjects were randomly allocated to dietary patterns.

In our study, we aimed to explore the effect of an intervention with MedDiet on cognitive function of participants included in the PREDIMED-NAVARRA randomized trial across different genetic variants not only of ApoE, but also of the CLU, CR1 and PICALM genotypes, previously identified as related to cognitive decline through GWAS studies.

Methods

Trial design

The PREDIMED (PREvención con DIeta MEDiterránea) study was a randomized, parallel-group, vascular primary prevention trial conducted in Spain from October 2003 to December 2010 testing two interventions with MedDiet (supplemented with EVOO or with mixed nuts) versus a control group advised to follow low-fat diet for outcomes on cardiovascular disease events in a high-risk population. The design and methods of the PREDIMED trial have been described in detail elsewhere (Martinez-Gonzalez et al. 2012a). On July 2011, the Data Safety Monitoring Board recommended stopping the trial since the pre-specified stopping boundary for the benefit of the MedDiets on the primary end point had been reached. The primary results have been recently published (Estruch et al. 2013). No relevant diet-related adverse effects were reported (Estruch et al. 2013). The study population was drawn from 1 of 11 recruitment centers (PREDIMED-NAVARRA). At this site, recruitment was completed by 2005, offering thus a longer intervention and follow-up period than other centers.

Study population

Participants were community-dwelling men (55–80 years) and women (60–80 years), without cardiovascular disease (CVD) at baseline but at high vascular risk. The inclusion criteria were the presence of either type-2 diabetes or at least three major risk factors: current smoking, hypertension, dyslipidemia, overweight or family history of premature coronary heart disease. The presence of prior history of CVD, any severe chronic condition or illiteracy was considered as an exclusion criterion (Martinez-Gonzalez et al. 2012a). All procedures followed the Declaration of Helsinki. The Institutional Review Board of the Navarra recruitment center approved the study protocol (protocol 50/2005). All participants signed an informed consent. The PREDIMED-NAVARRA center recruited 1,055 participants between 2003 and 2005.

The cognitive study was conducted over 8 months in the primary care centers in the same day that the nurse performed the blood analyses or the dietitians administered the group or individual sessions for the PREDIMED trial. This routine provides the participants the opportunity to complete different task of the PREDIMED study on the same day and promotes compliance. Participants who did not attend the visits on their scheduled days were considered non-eligible for the cognitive study. Those who attended the visits but did not accept undergoing the neuropsychological study were excluded. Finally, 522 subjects accepted to participate among 969 participants who were alive at the moment of the cognitive evaluation (Fig. 1).

Fig. 1
figure 1

Flowchart of participants

Full size image

Nutrition interventions and dietary assessment

A nutritional intervention promoting the MedDiet or advising to follow a lower fat diet was implemented among participants allocated to the two MedDiet groups or control group, respectively. At inclusion and quarterly thereafter, dietitians administrated individual and group sessions, separately for each group to provide an intensive education to follow the intended intervention. Additionally, participants allocated the MedDiet groups received free allotments of either EVOO (1 l/week) or 30 g/day of raw, unprocessed mixed nuts (15 g walnuts, 7.5 g almonds and 7.5 g hazelnuts). Non-food gifts were provided to the control group to improve compliance. Energy restriction was not advised, nor physical activity promoted (Martinez-Gonzalez et al. 2012a; Zazpe et al. 2008).

At baseline and yearly thereafter, a trained dietitian administered a validated 137-item food-frequency questionnaire and a 14-item short questionnaire of adherence to the MedDiet in a face-to-face interview. Both questionnaires have been validated (Fernandez-Ballart et al. 2010; Martinez-Gonzalez et al. 2012b).

Cognitive assessment

Two brief cognitive tests were administered after a mean follow-up of 6.5 years: the Mini-Mental State Examination (MMSE) and the Clock Drawing Test (CDT). MMSE evaluates orientation, registration, calculation, immediate and differed verbal recall, language and visual construction. MMSE ranges from 0 to 30 (Folstein et al. 1975). There is a validated Spanish version (Blesa et al. 2001). The CDT is especially useful for assessing executive function and visuospatial skills that are under-evaluated in the MMSE (Freedman et al. 1994). We used a validated Spanish version ranging from 0 to 7 (del Ser Quijano et al. 2004). Both cognitive tests are valid indicators of cognitive impairment (Crook et al. 1986; Paganini-Hill and Clark 2011). Researchers who assessed the outcome were blinded to group assignment.

Genotyping

CR1, CLU and PICALM genotyping was performed for subjects included in the NAVARRA recruitment center of the PREDIMED trial. DNA was extracted from overnight fasting venous blood samples, collected in EDTA added tubes. A commercial kit (Master PureTM; Epicentre, Madison, WI, USA) was used to extract DNA from the buffy coat fraction. All the subjects were genotyped for the rs3818361 of the CR1 gene, rs11136000 of CLU gene and rs3851179 of PICALM gene using Taqman with allele-specific probes on the ABI Prism 7300HT Sequence Detection System (Applied Biosystems, Foster City, CA, USA) according to standardized laboratory protocols.

ApoE genotyping was carried out blind at the Genetic and Molecular Epidemiology Unit, Valencia, Spain, for all participants of the PREDIMED trial. Genomic DNA was extracted from buffy coat with the MagNaPure LC DNA Isolation Kit (Roche Diagnostics, Mannheim, Germany). ApoE genotyping was carried out by a validated single-tube protocol using fluorescent probes in the LightTyper instrument (Roche), as previously reported (Frances et al. 2008). Quality control procedures including positive and negative controls as well as replication of a random 15 % of samples were applied. The duplicate concordance rate was higher than 98 %.

Covariate assessment

The baseline general questionnaire provided information about socio-demographic, lifestyle features and health-related conditions and family history of cognitive impairment or dementia (Martinez-Gonzalez et al. 2012a). Anthropometric measurements were performed by standard methods, as previously described (Martinez-Gonzalez et al. 2012a). Physical activity was assessed with the validated Spanish version of the Minnesota Leisure-Time Physical Activity questionnaire and expressed in minutes at a given metabolic equivalent per day (MET-min/day) (Elosua et al. 1994, 2000).

Statistical analyses

Hardy–Weinberg equilibrium was tested using a chi-square test. First, we explored differences in baseline characteristics of participants according to the allocated intervention groups. The association between the genotypes and cognitive tests (MMSE and CDT) according to the intervention groups (MedDiet groups vs. control group) were analyzed using linear regression models after adjusting for confounding variables: age (years), sex, education (years of formal education), ApoE genotype, family history of cognitive impairment or dementia, hypertension, dyslipidemia, type-2 diabetes mellitus, smoking status, alcohol intake (g/day), body mass index (BMI) (kg/m2), total energy intake (kcal/day) and physical activity (MET-min/day). Interaction terms were not included in the models. Individual models, one for each SNP (CLU, CR1 and PICALM) and for the ApoE genotype, were fitted. Interactions between gene variants (dominant models) and the intervention groups (MedDiet groups vs. control group) were estimated with the likelihood ratio test in separated models for each SNPs and adjusted for sex, age, education, ApoE genotype (when pertinent), family history of cognitive impairment or dementia and BMI. The false discovery rate method from Benjamini and Hochberg was used to control for multiple testing in the subgroups analysis (Benjamini and Hochberg 1995). Values in the text are means and SDs unless otherwise indicated. All p values were two-tailed at the <0.05 level. Statistical analyses were performed with STATA version 12.0 software.

Results

In our population, the frequencies of these three SNPs did fulfill the Hardy–Weinberg equilibrium (Table 1). Anthropometric and lifestyle baseline features of participants of the PREDIMED-NAVARRA study according to the intervention group are displayed in Table 2. As expected from the randomized design, the groups were well balanced with respect to all these baseline characteristics. Since we evaluated the effect of MedDiet as a whole dietary pattern, we have merged participants of MedDiet + EVOO and MedDiet + Nuts groups in a single MedDiet intervention group to compare this intervention fostering adherence to the MedDiet pattern to a low-fat intervention of control group.

Table 1 Genomic characteristics of the SNP markers used in the analysis
Full size table
Table 2 Baseline characteristics of the participants according to intervention group
Full size table

As shown in Tables 3 and 4, we did not find statistically significant differences in cognitive function scores (MMSE and CDT) across the considered genetic variants. Nevertheless, the MedDiet intervention seemed to modulate the effect of these genetic variants. In Table 3, the likelihood ratio test showed a statistically significant interaction between the CLU gene rs11136000 variant and the intervention status for the MMSE test (p = 0.041). For this SNP, we observed a beneficial effect of the MedDiet on MMSE score among subjects carrying the T minor allele (with a previously described protective effect). For these participants, in those carrying at least one copy of the T minor allele, the intervention with the MedDiet was associated with an increase in 0.97 points of the MMSE score (B = 0.97, 95 % CI 0.45–1.49, p < 0.001). This observation remained statistically significant after the Benjamini–Hochberg multiple comparison analysis. This effect was not observed for those without the T minor allele (B = 0.18, 95 % CI −0.52 to 0.88, p = 0.612). A similar effect was observed for the CR1 gene rs3818361 variant, even if in this case, the likelihood ratio test did not show a statistically significant interaction (p = 0.335). Within this frame, the MedDiet intervention significantly improved the cognitive function measured by MMSE, 0.76 points, in those subjects without the A minor risk allele (B = 0.76, 95 % CI 0.32–1.20, p = 0.001). No significant interaction was observed between the PICALM gene rs3851179 polymorphisms (p = 0.872); however, when stratifying our population according to the dominant model, we observed a beneficial effect of the MedDiet intervention in both genotype groups, in subjects without the T minor allele (protective allele) and in subjects bearing at least one T allele (B = 0.76, 95 % CI 0.10–1.43, p = 0.024 and B = 0.51, 95 % CI 0.01–1.00, p = 0.046, respectively). However, this observation did not remain statistically significant after the Benjamini–Hochberg multiple comparison correction.

Table 3 Multivariable-adjusted means and differences (95 % CI) for the association between each of the SNPs and cognitive assessment (Mini-Mental State Examination) according to intervention status
Full size table
Table 4 Multivariable-adjusted means and differences (95 % CI) for the association between each of the SNPs and cognitive assessment (Clock Drawing Test) according to intervention status
Full size table

Table 4 shows similar analyses for the CDT. In this case, no interaction was found between the MedDiet intervention and any of these gene variants. However, differences between genotype groups were found. For the CLU gene rs11136000 variant, among subjects with the T minor allele, MedDiet intervention significantly improved CDT scores in 0.60 points (B = 0.60, 95 % CI 0.24–0.96, p = 0.001). Similar effects were observed for the CR1 gene rs3818361 variant within subjects without the A risk allele (B = 0.46, 95 % CI 0.13–0.79, p = 0.006) and for the PICALM gene rs3851179 polymorphism. For this SNP, subjects with at least one T minor allele with a probable protective effect on cognition benefited from the MedDiet intervention and their score in the CDT was significantly increased in 0.59 points (B = 0.59, 95 % CI 0.18–1.00, p = 0.005). All these observations remained statistically significant after the Benjamini–Hochberg multiple comparison correction.

Finally, Table 5 displays the effects of the MedDiet intervention on cognition (MMSE and CDT) in our population according to the ApoE genotype. In the first general analysis, the linear regression model revealed that E4 isoform was associated with a significantly decreased score for both cognitive tests, MMSE (B = −0.96, 95 % CI −1.47 to −0.45, p < 0.001) and CDT (B = −0.50, 95 % CI −0.87 to −0.14, p = 0.007). Interestingly, when stratifying our subjects accordingly to their genotype for ApoE4, we observed that the MedDiet intervention had a protective effect on cognitive function for both non-ApoE4 and ApoE4 carriers. For the MMSE score, we found statistically differences in participants allocated to the MedDiet intervention among non-ApoE4 carriers. The MedDiet intervention was associated with a 0.56-point higher average MMSE score (B = 0.56 95 % CI 0.15–0.97, p = 0.007). Specifically, in those carrying the E4 risk isoform, the MedDiet was associated with an increased score in more than 1.6 points (B = 1.61, 95 % CI 0.10–3.13, p = 0.037). However, this observation did not remain statistically significant after de Benjamini–Hochberg multiple comparison correction. A higher CDT score was also found among subjects randomly allocated to the MedDiet intervention, but this difference was only significant for non-ApoE4 carriers (B = 0.55, 95 % CI 0.25–0.85, p < 0.001).

Table 5 Multivariable-adjusted means and differences (95 % CI) for the association between the SNPs and cognitive assessment according to intervention status
Full size table

Discussion

In this study, an intervention with a MedDiet modulated the association of genetic risk factors on cognition. The interaction between MedDiet and the CLU gene rs11136000 variant is a novel finding. Overall, the protective effect on cognition of an intervention with MedDiet was greater in participants with a favorable genetic profile of the CR1 gene (without A minor allele of the rs3818361); the CLU gene (with T minor allele of the rs11136000); and the PICALM gene (with T minor allele of the rs3851179). The only exception was the higher CDT score found in participants without the T minor protective allele of the rs3851179 polymorphism of the PICALM gene compared to those with the T minor allele of this polymorphism. Considering the ApoE genotype, a MedDiet intervention benefited both non-ApoE4 and ApoE4 carriers. Our results strongly suggest a MedDiet effect on both MMSE and CDT assessments among non-ApoE4 carriers, but the relatively low number of ApoE4 carriers in our study hinders definite conclusions regarding this group.

For more than 20 years, it has been known that ApoE4 genotype is a genetic risk factor for Alzheimer’s disease (Burke and Roses 1991). Years later, GWAS studies identified variants at other genes such as CLU, PICALM and CR1 as risk factors for Alzheimer’s disease (Lambert et al. 2009; Harold et al. 2009). These results have been confirmed in a meta-analysis (Jun et al. 2010). More recently, some of these risk gene variants have been associated with cognitive decline, more concretely with poor memory performance in population-based studies (Barral et al. 2012; Pedraza et al. 2013). Parallel with these studies, there is accruing evidence about the benefits in cognition of lifestyle-related habits such as physical and cognitive and social activities (Hamer and Chida 2009; Kareholt et al. 2011; Morgan et al. 2012; Niti et al. 2008; Sattler et al. 2012) and nutrition in terms of nutrients and foods with conflicting results (Coley et al. 2008). Regarding dietary patterns as a whole, a protective effect of MedDiet on cognitive function (Kesse-Guyot et al. 2013), cognitive decline (Tsivgoulis et al. 2013; Tangney et al. 2011; Feart et al. 2009; Scarmeas et al. 2006) and development of mild cognitive impairment (Scarmeas et al. 2009b; Roberts et al. 2010) and dementia (Roberts et al. 2010; Gu et al. 2010; Scarmeas et al. 2006, 2009a, b) has been found in several observational studies but not in other cohort studies (Psaltopoulou et al. 2008; Samieri et al. 2013; Cherbuin and Anstey 2012). Thus, there is not definitive evidence from observational studies to ensure the benefits of an intervention with MedDiet against cognitive impairment. However, a protective effect has been recently suggested by the PREDIMED-NAVARRA randomized trial conducted in Spain (Martinez-Lapiscina et al. 2013a, b). This is important since some observational studies may not find a protective association between the MedDiet and cognitive function because their MedDiet-like dietary pattern did not truly represent the traditional MedDiet due to sociocultural factors (Cherbuin and Anstey 2012). Nevertheless, the benefit of MedDiet on cognitive impairment is far from being resolved. Going a step beyond, the interaction between MedDiet and risk factors has been poorly studied. In a study conducted in Australia, the MedDiet did not show any protective association with cognitive changes and no significant interaction was observed for the ApoE4 genotype and the MedDiet (Cherbuin and Anstey 2012). We did not identify any studies assessing the interaction between MedDiet and CR1, CLU and PICALM gene variants.

We acknowledge that our study may have some limitations. First, cognitive function of participants was not assessed at baseline. Nevertheless, the presence of any chronic condition that may limit the compliance with the protocol was an exclusion criterion, and this is a good reason to confide that a relevant cognitive decline was not present at baseline. In addition, the randomized design of the trial resulted in well-balanced groups with respect to the overall baseline features, and this argues for similar cognitive status at baseline. Moreover, we performed our analyses after adjusting for a wide range of potential confounders. Second, the sample size was small. Third, the study has a single blinded nature, but there is no possibility of conducting true double-blind long-term trials in nutrition. The selection of a low-fat dietary pattern as comparator might have introduced a limitation in our results, because of the potential protection afforded by educating the control group on a low-fat diet. The role of this dietary pattern on cognition is not completely understood. Should the low-fat diet have a detrimental effect, the allocation to a low-fat diet for comparison in the control group may represent an alternative explanation to the protective effect observed for the MedDiet. However, the long-term sustainability of low-fat diet is suboptimal, and in any case, controls may have experienced a benefit from their education on a low-fat diet that would have not existed should they have followed a Western-type diet. Another limitation is that we evaluated the interaction between MedDiet and gene risk variants in a high-risk vascular population so generalization of our findings to the average general population is uncertain. Finally, we acknowledge that our results are relevant in terms of statistical significance, but they do not represent a large difference from the clinical or practical point of view. However, we think that our manuscript is scientifically relevant because it points to a biological interaction between MedDiet and gene profile beyond ApoE. The introduction of this interaction is a novelty, and our exploratory results may favor further and deeper studies of gene–nutrition interactions in larger cohorts that may consider not only cognitive function but also, more importantly, the incidence of clinical cases of mild cognitive impairment or dementia.

Our study has also strengths. First, this is a long-term randomized controlled trial with supplementation with hallmark components of the traditional MedDiet such as EVOO and nuts. Second, we included several potential confounders in our analyses that help us to rule out residual confounding.

In conclusion, we provide here, for the first time, the evidence of an interaction between the MedDiet and CLU on cognitive function in the context of a randomized trial. A MedDiet intervention modulates the effect of the genetic risk factors on cognition, and overall, the protective effect of MedDiet may be greater for subjects with favorable genetic profile. Future research is needed to obtain firmed evidence of the complex MedDiet–gene interaction.