Establish working mechanisms of the combination and contributions of each individual compound

In the past decade, the effect of nutrition and dietary components on brain structure and function has been investigated intensively. This research has shown that dietary patterns, and their impact on possible mechanisms such as neuroinflammation and cerebro/cardiovascular impairment (e.g. chronic cerebral hypoperfusion), may influence the development of particular types of dementia such as Alzheimer’s disease (AD)^{(Reference Gardener, Scarmeas and Gu12–Reference Kennedy, Wightman and Reay21)}. The combination of dietary components that simultaneously, and possibly synergistically, target an underlying risk factor and/or physiological process could have an even larger effect on inhibiting the development of neurodegenerative disorders than single nutrients targeting that process^{(Reference Joris, Mensink and Adam17,Reference Gorelick, Scuteri and Black22)} . Therefore, one approach is to identify working mechanisms of food component interactions, and how the contribution of each individual compound may affect those mechanisms.

Dietary nutritional components may influence numerous biological pathways which could be beneficial at the cellular and systems level. Although a single ingredient may be able to influence such a pathway, it is likely that food component combination may provide a better result than any compound in isolation. Previous studies have identified effects of dietary nutrients on several specific cellular pathways that have systemic effects. For example, docosahexaenoic acid (DHA) and eicosapentaenoic acid (EPA) found in high concentration in oily fish and fish oil have shown in both in vitro brain cell culture experiments and in vivo animal studies to mitigate inflammatory pathways, through activation of peroxisome proliferator-activated receptor γ (PPARγ)^{(Reference Mita, Beaulieu and Field23)} and the NF-κb transcription factor^{(Reference Paterniti, Impellizzeri and Di Paola24)}, both of which are involved in many biological pathways including inflammation. Notably, other nutrients such as glutamine, selenium, spices or flavonoids are also able to influence PPARγ and NF-κB^{(Reference Marion-Letellier, Dechelotte and Iacucci25,Reference Youn, Lim and Choi26)} , indicating the potential for additive or synergistic effects.

Indeed, the clinical evidence for the beneficial effects of DHA supplementation on cognitive decline is mixed^{(Reference Wu, Ding and Wu27)}, suggesting that long-chain omega-3 PUFA may need to be given in combination with additional nutrients to synergistically produce an effect at the cellular and systems level. In agreement with this scenario, one example is the VITACOG study in which the effect of B vitamins on slowing down cognitive decline is enhanced when omega-3 fatty acid concentrations are in the upper range^{(Reference Oulhaj, Jerneren and Refsum28)}. Moreover, antioxidant nutrients can also protect the highly susceptible long-chain omega-3 PUFA from peroxidation when co-supplemented. Accordingly, it has been shown that response to omega-3 PUFA supplementation was enhanced in individuals when consumed with dark-green vegetables which are good dietary sources of folate, lutein, carotenoid, vitamin K and vitamin B6, suggesting a potential interaction between these nutrients and the absorption or bioavailability of omega-3 PUFA^{(Reference O’Sullivan, Armstrong and Schuster29–Reference Harris and Von Schacky33)}.

Another type of interaction could involve activating a detoxification pathway. For example, pairing red meat with fresh garlic containing allicin was able to block dietary carnitine derived trimethylamine production in the gut^{(Reference Barabási, Menichetti and Loscalzo34)}. This ultimately leads to lowering of the trimethylamine N-oxide (TMAO) concentration in plasma, therefore protecting against the development of cardiovascular diseases. Notably, other components found in the Mediterranean diet such as wine, olive oil and grapeseed oil are also able to lower TMAO^{(Reference Wang, Klipfell and Bennett35–Reference Wu, Panyod and Ho38)}. Indeed, one advantage of a nutritional approach to improve brain health is that, unlike specific pharmacological approaches, dietary interventions have the possibility to include numerous components simultaneously that may act in synergy^{(Reference Kardum and Glibetic39–Reference Zhang, Zheng and Liu41)}. Nonetheless, it is equally important to recognise that nutrients may also have antagonistic effects on one or multiple pathways of interest, or impact the absorption of each other, making the understanding of a combinatorial approach more challenging than expected. For example, dietary polyphenols such as flavonols, anthocyanins and isoflavones exert many positive actions, including anti-proliferative, anti-inflammatory, immuno-regulatory and neuroprotective effects^{(Reference Forbes-Hernandez, Gasparrini and Afrin42)}. However, when they are ingested with macronutrients, the bioavailability and bioactivity can be significantly changed. Polyphenols such as those found in coffee or tea are often consumed with milk. The binding of polyphenols to milk proteins negatively affects the bioefficacy and may decrease, for example, vascular protective effects and antioxidative properties of polyphenols^{(Reference Kardum and Glibetic39–Reference Zhang, Zheng and Liu41)}. Polyphenols themselves affect the uptake and bioavailability of vitamins and minerals, especially iron, zinc and calcium^{(Reference Cianciosi, Forbes-Hernandez and Regolo43,Reference Jakobek and Matić44)} . For example, in Wistar rats it has been demonstrated that polyphenolic extracts of green tea, chokeberry or honeysuckle fruits decrease zinc and calcium absorption^{(Reference Frejnagel and Wroblewska45)} .

In conclusion, rather than focusing on specific nutrients, it may be more advantageous to concentrate instead on identifying specific mechanisms. It may then be possible to find a group of ingredients that influence such pathways leading to desirable behavioural outcomes^{(Reference Kardum and Glibetic39–Reference Zhang, Zheng and Liu41,Reference Frejnagel and Wroblewska45–Reference Wurtman48)} . Nonetheless, to design a multi-nutrient approach it is important to understand the contribution of each individual component on the hypothesised working mechanism to favour synergistic effects^{(Reference McNamara, Kalt and Shidler49)}. A caveat is that whole diets include dozens of foods providing thousands of molecules, which results in complex biological activity. Such complex systems may not simply dissect into constituent parts. Focusing on one or two of these molecules or potential pathways means that the vast majority are ignored when they may have active influence. The consequence is unpredictable ‘emergent’ properties. Therefore, whilst it may be possible to define isolated functions and linear contributions of some nutrients in some small part of the system, the approach is likely to be limited, and complementary approaches (discussed later) may be required.

Validate the relevance of the mechanisms for the targeted human condition

Often, neurological diseases are influenced by biological pathways, including those outside of the brain, which could be targeted with nutrition interventions to enable greater treatment efficacy. However, verifying the association between specific biological mechanisms and cognitive and brain health is crucial. For example, it was reported that type II diabetes is a risk factor for developing AD^{(Reference Sridhar, Lakshmi and Nagamani50,Reference Steen, Terry and Rivera51)} . Both conditions are associated with impaired glucose metabolism and insulin resistance^{(Reference Zhao and Townsend52)}. Therefore, assuming that insulin resistance is a contributing factor to the pathogenesis of AD, therapeutic interventions aiming to improve insulin resistance in combination with brain-specific pathways may be beneficial. A second example comes from recent work indicating a possible relationship between blood pressure and AD pathology^{(Reference Arvanitakis, Capuano and Lamar53)}. Remarkably, a report from the SPRINT MIND trial indicated that intensive blood pressure reduction in older adults resulted in a significant reduction in the number of individuals developing mild cognitive impairment over the 8-year study^{(Reference Williamson and Pajewski54)}. This suggests that dietary intervention to reduce blood pressure throughout the lifespan, coupled with nutrients targeting the brain, may act in cooperation to reduce the risk of developing dementia. However, often biological pathways are poorly understood and causal links to the behavioural outcome are missing. When using biological mechanisms to inform the design of multi-nutrient trials, it is necessary to differentiate risk factors from mere correlates. While causal risk factors can be manipulated to change the probability that an outcome will occur, correlates cannot. One reason for non-significant results in nutrition trials based on prior significant epidemiological observations might be a failure to differentiate between risk factors for the development of disease and correlates that change because of the disease process. Therefore, validating the relevance of biological pathways for the human condition is a key aspect that needs to be addressed early on, to have real synergies or even additive outcomes.

Biology is complex; however, some modifiable risk factors for cognitive decline and dementia have been identified. For example, changes in plasma levels of the amino acid homocysteine is considered a risk factor for dementia, and related changes in brain structure^{(Reference Seshadri, Beiser and Selhub55)}. There are two pathways by which homocysteine is broken down: re-methylation that requires folate and vitamin B12, and the trans-sulphuration pathway, which requires vitamin B6. A supplement of these three co-enzymes was given for 2 years and reduced the levels of homocysteine^{(Reference Seshadri, Beiser and Selhub55)}. Structural MRI was used to measure any increase in the size of the ventricles, a reflection of the shrinking of the brain. The rate at which the brain was shrinking was less in those taking the vitamin supplement. These findings illustrate that, if multi-nutrient interventions target well-validated causal pathways, beneficial behavioural outcome can be observed.

An example of a successful multi-component approach is the specific formulation of nutrients, registered as Fortasyn Connect. Owing to the additive effects of the combination of DHA, uridine, choline, folate, vitamin B12, vitamin B6, vitamin E, vitamin C and selenium^{(Reference Beeri, Haroutunian and Schmeidler46,Reference Chen, Mecca and Naganawa47)} , this supplement promotes synaptogenesis^{(Reference Wurtman48)} to counteract the synaptic loss occurring in cognitive impairment and AD. These ingredients may act via several synergistic mechanisms including, but not limited to, providing precursors for the Kennedy cycle, lowering homocysteine levels, and increasing cerebral blood flow and perfusion^{(Reference Rasmussen56)}. Many of these nutrients are also common in the Mediterranean diet, for example, fish oil (DHA), tomatoes (uridine), choline in eggs, fish and nuts, and vitamins and antioxidants in onions, lentils and beans. The Mediterranean diet has been found to have numerous beneficial physiological effects, including reducing blood pressure^{(Reference Filippou, Thomopoulos and Kouremeti57)} preventing type 2 diabetes and improving glycaemic control^{(Reference Esposito, Maiorino and Ceriello58)}, and lowering inflammation^{(Reference Bonaccio, Pounis and Cerletti59)} and cholesterol levels^{(Reference Esposito, Maiorino and Bellastella60)}. Notably, these are all well-established prospective risk factors that may increase the odds of developing dementia in a dose-dependent way. It was reported that, compared with no risk factors, the relative risk for dementia was 1·20 for one risk factor, 1·65 for two and 2·21 for three or more risk factors^{(Reference Peters, Booth and Rockwood61)}. Therefore, the combination of these dietary components found in the Mediterranean diet and Fortasyn Connect may protect against cognitive decline, and ultimately AD, by targeting inflammation, stimulating neuronal cell membrane production and improving vascular and metabolic health, which cannot be achieved by single ingredients/dietary components by themselves^{(Reference Scarmeas, Stern and Tang62,Reference Bianchi, Herrera and Laura63)} . It should, however, be noted that in one of the main Fortasyn Connect trials the effect on the primary endpoint (change in a neuropsychological test battery (NTB; composite z-score based on the Consortium to Establish a Registry for Alzheimer’s Disease (CERAD) ten-word list learning immediate recall, CERAD ten-word delayed recall, CERAD ten-word recognition, category fluency, and letter digit substitution test) was not significant, dividing scientific opinion regarding the effectiveness of this multi-nutrient formula in those with mild cognitive impairment^{(Reference Soininen, Solomon and Visser64)}.

It is also important to consider that other multi-nutrient studies have failed to demonstrate synergistic or additive effects on cognition. For example, as mentioned previously, the combination of fish oil and anthocyanin-rich blueberries in older adults did not interact to produce cognitive enhancement^{(Reference McNamara, Kalt and Shidler49)}. Instead, those who consumed blueberries alone had an improvement in discrimination memory^{(Reference McNamara, Kalt and Shidler49)}. The authors speculated that the combined treatment may have subverted the beneficial effect that was observed when blueberries were consumed only by excessively upregulating the transcription factor NF-E-2 related factor 2. This finding emphasises how, in complex systems, nutrients may not always interact in a predictable way, highlighting the need for a clear understanding of the proposed mechanistic synergies to produce positive outcomes.

Include current nutrient status, intake or dietary pattern as inclusion/exclusion criteria in the study design

One key consideration should be to include current nutrient status, intake or dietary pattern as an inclusion/exclusion criterion in the study design. Indeed nutrient–nutrient interactions may involve an interaction between the experimental nutrient, and the diet habitually consumed by the individual/population. Therefore, in intervention studies, it is important to quantify the intake of relevant nutrients in the individuals’ diet. Not doing so risks masking any possible beneficial effects of the nutrient supplementation, and limits understanding of the ability to metabolise the nutrients in the specific population. For example, interventional studies in older adults involving dietary supplementation with omega-3 PUFA have produced less convincing and more mixed results^{(Reference Baleztena, Ruiz-Canela and Sayon-Orea68,Reference Vakhapova, Cohen and Richter69)} ; for review, see Weiser et al. ^{(Reference Weiser, Butt and Mohajeri70)}. For example, in the OmegAD cohort, supplementation with DHA and EPA for 6 months did not delay the rate of cognitive decline^{(Reference Freund-Levi, Eriksdotter-Jonhagen and Cederholm71)}. However, in a follow-up analysis the authors report an inverse correlation between plasma levels of omega-3 PUFA and the rate of cognitive decline^{(Reference Eriksdotter, Vedin and Falahati72)}. Furthermore, in the Multidomain Alzheimer Preventive Trial (MAPT), DHA and EPA supplementation over 3 years had no significant effects on cognition in elderly individuals^{(Reference Andrieu, Guyonnet and Coley8)}. Like the OmegaAD study, the authors also noted that individuals with lower levels of DHA and EPA in blood exhibited a larger decline in cognitive measures than those individuals with normal concentrations. Together these data suggest that systemic levels of DHA may be important for slowing cognitive decline. However, as noted above, supplementation with the single compound, or in combination with EPA, may not be enough, as the beneficial effects of omega-3 fatty acids may be due to interactions with existing factors within the diet. Selection of the population should be guided towards individuals with low intake of foods containing all the components of interest.

Regarding habitual diet, individual variation in microbiota and microbiome is also an important aspect to consider. Much attention is now being focused on the bidirectional communication between gut microbiota and the brain, particularly with regard to host brain activity modulation via a microbiome–gut–brain axis^{(Reference Bastiaanssen, Cowan and Claesson73)}. For example, an increase in the microbial load and the metabolites released can increase the permeability of the intestinal barrier, allowing the stimulation of innate immunity cells by lipopolysaccharide. The subsequent circulation of pro-inflammatory cytokines or short-chain fatty acids derived from bacterial metabolism^{(Reference Schönfeld and Wojtczak74)} may affect the blood–brain barrier and lead to their entry into the brain^{(Reference Borre, O’Keeffe and Clarke75)}, where they stimulate microglial cells and cause neuroinflammation. The products of bacterial metabolism and pro-inflammatory mediators produced after immune stimulation induced by the microbiota can reach the brain through a leaky blood–brain barrier.

It is now recognised that foods and nutrients influence the microbiome in different ways. For instance, a Western diet with high fat and sugar intake reduces the microbial diversity and affects cognition by triggering chronic low-grade inflammation^{(Reference Komanduri, Gondalia and Scholey76–Reference Proctor, Thiennimitr and Chattipakorn79)}. In addition, essential fats, proteins, non-digestible carbohydrates, probiotics and polyphenols all induce shifts in the microbiome with consequences for host metabolic and immunologic markers^{(Reference Singh, Chang and Yan80)} Therefore, attempts to investigate effects of a specific diet in a population already adhering to that diet may be of low yield and/or limited by small changes in their microbial diversity. Notably, data from the Personalised Responses to Dietary Composition Trial (PREDICT) indicated that the gut microbiome composition predicted several post-prandial glycaemic, lipaemic and inflammatory indices^{(Reference Berry, Valdes and Drew81)}. In addition, the nature of intestinal microbiota contributes to the inter-individual variations in the metabolism of various phytochemicals^{(Reference Setchell, Brown and Lydeking-Olsen82)}. Essentially, an individual’s response to a food, or a group of foods, is likely to depend on the pre-existing composition of their microbiome. Therefore, for dietary interventions based on nutrients, foods and dietary patterns, the participants’ background dietary intake along with the consideration of the baseline microbiota is critical.

Select a participant population that is clinically and biologically appropriate for all nutritional components of the combination

Consider a range of cognitive outcomes

Many symptoms of the ageing of the brain are cognitive. The assessment of cognition raises a range of general methodological issues that have been reviewed previously^{(Reference Devore, Stampfer and Breteler91)}, although when considering several nutrients at the same time there is additional complexity. Cognitive functioning clusters into six major domains: executive functioning, memory, attention, perception, psychomotor and language. Each domain can be further subdivided: for example, memory, amongst other distinctions, includes short-term, long-term and working memory, and involves processes such as encoding, consolidation and retrieval. You may be interested in a particular aspect of cognition, but there may be unpredictable outcomes such that other aspects of cognition or mood, that were not being experimentally examined, are influenced. Given the complexity of the diet and the complexity of the brain, the ‘law of unintended consequences’ comes into play.

Varying several aspects of nutrition may influence the functioning of several areas of the brain and hence different aspects of its functioning; thus, changes in a measure of one aspect of cognition may in practice reflect changes in another. Without measuring a range of aspects of cognition, you cannot know if any change reflects an improvement in the one aspect of cognition that is of interest, rather than another aspect of functioning that facilitates the first. An example is an association between stress, depression and memory. A systematic review of those with mild memory problems found that those with anxiety, but not depressive symptoms, were later more likely to become cognitively impaired^{(Reference Desai, Whitfield and Said95)}. When under stress, the body releases cortisol, and over time an area of the brain associated with memory, the hippocampus, shrinks^{(Reference Ouanes and Popp96)}. Therefore, when studying memory as a measure of cognitive decline, we need to also monitor stress. There is no point in addressing the neural mechanisms associated with memory if the problem is stress related. We must address the cause and not the consequences.

A need to consider a range of aspects of brain functioning is illustrated by a review of forty-six studies that found a higher intake of flavonoids improved depression^{(Reference Ali, Corbi and Maes97)}, although it has also been said that ‘flavonoids may exert particularly powerful actions on mammalian cognition and. may reverse age-related declines in memory’^{(Reference Spencer98)}. Unless you monitor both, how can you be sure that a dietary-induced memory improvement is any more than better mood? The message is that we need to monitor a range of aspects of cognition functioning and the more that aspects of diet are varied the greater the likelihood that a range of measures will be influenced. The performance of a psychological test demands an ability to maintain attention, remember, and develop strategies (executive functioning), motivation, mood and language skills. If you measure only one of these mechanisms, the way diet was influential will be unclear, a consideration that becomes more important when several aspects of diet are being examined.

The problem of examining a single primary outcome is illustrated by studies that were hoping to improve memory. A post-hoc analysis of the Multi-domain Alzheimer Preventive Trial (MAPT) found that supplementation with DHA, in older individuals with low blood omega-3 fatty acids, benefited executive functioning but no other cognitive domain^{(Reference Zwilling, Talukdar and Zamroziewicz99)}, that is, the expected change in memory was not observed. Similarly, in older individuals, the frequent consumption of seafood for five years was associated with a lesser decline in semantic memory and perceptual speed, but not episodic or working memory, visual spatial functioning or global cognition^{(Reference Breitling, Perna and Muller90)}. This was an unexpected finding as the initial memory problems associated with dementia reflect episodic rather than semantic memory^{(Reference Prentice and Huang100)}. If there are unpredictable consequences with one nutrient, then the potential is greater when many nutrients are combined, as they may interact in nonlinear ways. Thus, a consideration of psychological phenomena demands a test battery that assesses all cognitive domains that are potentially influential.

Only if we show that an improvement in functioning reflects memory as such, and not a change in mood, arousal, attention, executive functioning or anything else, will we have established that an intervention has influenced disease progression. Clearly, any positive response to an intervention is to be welcomed. However, unless you benefit memory as such, you are not slowing the progress of a disease; rather, you are optimising any residual capacity, but only while some capacity remains. The take-home message is that cognition, in general, should be measured rather than only memory or some composite measure that adds together things with little in common. Of course, when measuring multiple outcomes, appropriate corrections to prevent inflating the changes of type 1 error, should also be applied.

Consider the limits of reductionism and the ‘gold standard’ randomised controlled trial

There is a well-established and rigorous way of designing experiments: you vary an independent variable and predict the consequences for a dependent variable, stating a priori the primary outcome measure. Although this approach will hopefully be used in this area, in some cases it might not be the most efficient way of proceeding. The problem is complexity which results from the need for detailed study of the linear and non-linear interactions between various parameters. The present state of knowledge makes it difficult to develop precise predictions. Therefore, although randomised controlled trials (RCT) are considered the ‘gold standard’, when it comes to dietary interventions, we may need to acknowledge that well-conducted epidemiological studies have several advantages^{(Reference Lichtenstein, Petersen and Barger101)}. In fact, there is substantial evidence from epidemiology that many aspects of diet are associated with the risk of dementia. One explanation is that observational prospective studies benefit from the ability to assess the aggregate effects of multiple nutrients over an extended period – something especially pertinent when considering the effects of diet on the development of dementia^{(Reference Benton85)}. For example, by combining data from three longitudinal cohorts, Moore et al.^{(Reference Moore, Ames and Mander102)} observed that those with low plasma concentrations of vitamin B12 and high concentrations of folate in the blood have a substantially higher risk of cognitive impairment compared with those with low blood folate and low plasma vitamin B12. Therefore, observational data may also be able to address questions concerning the influence of multiple nutrient deficiencies that may occur with age, questions unlikely to be assessed in RCT owing to practical and ethical considerations. Additionally, Zwilling et al. (2019) combined nutrient biomarker pattern analysis with functional brain network analysis and cognitive testing. It was reported that high plasma ω-3 and ω-6 PUFAs, lycopene, carotenoids and vitamins B (riboflavin, folate, B12) and D were associated with better cognitive functioning^{(Reference Zwilling, Talukdar and Zamroziewicz99)}. Furthermore, plasma levels of ω-3 and ω-6 PUFAs and lycopene moderated the association between brain network connectivity and cognition. This study illustrates how innovative methods from epidemiology, cognitive and network neuroscience can be combined to enhance understanding of how nutrient interactions might influence brain ageing.

Importantly, epidemiology tends not to look at single nutrients, or specific foods, but rather at dietary styles combining a range of nutrients. The advantage of this approach is that it might help capture the complexity of nutrient–nutrient interactions, although establishing the active molecules and the mechanism of action is difficult. Without such an understanding, precise dietary recommendations are not possible, and the development of functional foods is precluded when it is unlikely that any benefit will reflect a single nutrient or food. We need to know if the benefit of a style of eating reflects the small influence of many foods and, if so, which ones. If there are synergistic, additive or antagonist effects, which foods are involved? Additionally, a thorough understanding of nutritional epidemiology is required to ensure that research delivers reliable and robust results on which to base the development of RCTs. A complete discussion of epidemiological methods is beyond the scope of this article and can be found elsewhere^{(Reference Prentice and Huang100,Reference Almirall, McCaffrey and Ramchand103–Reference Fairchild and McDaniel106)} . Dietary pattern analysis is considered in the section: decomposing the complexity of diet.

In performing a multi-nutrient intervention study there are many considerations to consider. Broadly speaking, interventions can incorporate the multi-nutrient dimension in one of two ways: through a food-based approach where foods enriched in the components are consumed as part of a daily diet or in a capsule form where the components are incorporated into the capsule. In the latter scenario, the control group receives an intervention identical in appearance and sensory characteristics and blinding is easy to implement. The biggest disadvantage is the loss of the food matrix and potential contributions^{(Reference Aguilera107,Reference Wahlqvist108)} . With respect to whole diets and food-based approaches, it is more difficult to develop a control intervention identical to the test intervention minus the multi-nutrient bioactive components. Furthermore, blinding is more difficult but not impossible. There are examples of food-based interventions that have successfully used a food substitution approach to alter multiple components of the diet in a controlled fashion^{(Reference Weech, Vafeiadou and Hasaj109,Reference Tierney, McMonagle and Shaw110)} . Notwithstanding the limitations of either approach, positive response to a nutritional intervention can be interpreted only in comparison with the group with which it is compared. Thus, the negative consequences of the control diet need to be as clearly defined as the active intervention. It is as useful to conclude that there is a negative response to a poor diet as well as a beneficial response to a good diet.

To conclude, such is the importance of RCTs that, when claiming a causal relationship, epidemiological data can be downplayed. However, for practical and ethical reasons there are many situations where it is not possible to run a RCT. It is not questioned that smoking is a risk factor for lung cancer, although nobody has randomly required half their subjects to smoke for a lifetime. Similarly, nobody is going to randomly allocate people to a diet that is suggested to increase the risk of dementia.