Skip to main content

The Neurobiology of Decision-Making in Eating - Innovative Tools

Final Report Summary - NUDGE-IT (The Neurobiology of Decision-Making in Eating - Innovative Tools)

Executive Summary:
Nudge-it was a multidisciplinary, multinational project funded by the European Union FP7 programme to address the need for a better understanding of the determinants of food choice. The project engaged experts in the neurobiology of motivational behaviour, the neuroscience of reward pathways, the neuroendocrinology of homeostatic regulation of appetite, experimental psychology, functional brain imaging, behavioural economics (the scientific study of decision making) and computational neurobiology.
In the developed world, food is abundant and we must make choices about what we eat. But these choices can bring problems - it can be hard to make healthy decisions. Often, foods that are inexpensive and convenient contain a large number of calories, and as a result the levels of overweight and obesity are high and rising. The various determinants of food choice include dietary components, but also cultural and social pressures, cognitive factors (such as perceived stress), and familial, genetic and epigenetic influences. Choices are also influenced by how foods are marketed and labelled, by economic factors, and choices reflect both habits and impulses, moderated by an understanding of what constitutes ‘healthy eating’.

Nudge-it focussed on developing new tools to better understand food choice behaviour, particularly as it affects health. In formulating policy, it is essential to make optimal use of a large and complex evidence base. In particular, the key to robust evidence-based policy is to harness evidence from what can be three very different domains: mechanistic evidence of the fundamental neurobiological processes involved in food choice (typically studies in rodents); associational evidence linking regional brain activity with food choice behaviour; and evidence from intervention studies.

A particular focus of Nudge-it was to develop tools to help ‘bridge the gaps’ between these domains. We approached this by working collaboratively across disciplines to design studies that could be conducted analogously in both humans and animal models, and to develop model frameworks to allow findings in one domain to be interpreted in an adjacent domain. We also developed tools for use by the broader food research community to facilitate data sharing and rigorous evaluation of data, and have made these openly available.

Amongst the main outputs of the project, we
1. Developed novel approaches to understand how early-life experiences and environmental determinants affect food choice.
2. Developed novel neuroimaging technologies to understand the brain mechanisms underlying food choice and understand how the lifelong learning process contributes to this.
3. Modelled how physiological, psychological, and emotional factors (like impulsivity and stress) predispose people to unhealthy eating.
4. Advanced knowledge and developed new tools to model and test the decision processes involved in dietary choices, with a specific focus on low socioeconomic status groups.
5. Developed a formal predictive systems model of the interactions between physiological, psychological and emotional factors related to food choice.
6. Constructed a ‘policy toolbox’ for making optimal use of evidence in the formulation of policy.

The outputs of the project include about a hundred articles in the academic literature. The consortium also undertook extensive dissemination activities to spread awareness of the project to diverse stakeholder groups, including importantly the general public. These activities included a Massive Open Online Course on ‘Understanding Obesity’.


Project Context and Objectives:
Summary description or project context and objectives

The prevalence of overweight and obesity across Europe has increased dramatically in recent decades. The full consequences of this epidemic have yet to unfold, with an expected increase in cardiovascular disease, hypertension, type 2 diabetes, stroke, certain cancers, musculo-skeletal disorders and a range of mental health conditions including anxiety and depression. These are accompanied by huge social, health service and economic costs: they affect individuals in the midst of their working lives, impoverishing families through time lost at work and impaired employment prospects. Stress, compounded by social stigma and prejudice in workplace, educational and health care contexts, adds to the pressures on families and employment, and can enhance the vicious cycle of weight-gain through “comfort eating”.
For effective policies on healthy eating, we must be able to make meaningful predictions about the consequences of particular behaviours and interventions. For this, we need to better understand the determinants of food choice. These determinants include dietary components, but also cultural and social pressures, cognitive-affective factors (such as perceived stress), and familial, genetic and epigenetic influences. Choices are influenced by how foods are marketed and labelled, and by economic factors, and they reflect both habits and impulses, moderated, albeit imperfectly, by an understanding of what constitutes healthy eating.
To develop the evidence base that is needed for good policies, bridges must be built across different levels of knowledge and understanding. This requires new experimental models that can fill in gaps in our understanding, new translational models that link mechanistic understanding from the laboratory to the real life human condition, and new formal models that embed understanding in a way that facilitates policy-relevant predictions. This focus on building bridges in understanding offers a unique potential for identifying objective measures of stimuli for food intake, satiety and even restraint in eating, in a way that will contribute to studying obesity and weight management, eating and malnutrition disorders from a completely different perspective.
Nudge-it was a multidisciplinary project; it engaged experts in the neurobiology of motivational behaviour, in the neuroscience of reward pathways, in the neuroendocrinology of homeostatic regulation of appetite, in experimental psychology, in functional brain imaging, in behavioural economics, and in computational modelling. Together, our aim was to understand:
• The importance of early life experience: how the choices we make as adults are influenced by stress and poor nutrition in early life.
• Habitual eating behaviour: the life-long learning process and how it is moderated by homeostatic mechanisms.
• Impulsive choice behaviour: the momentary choices we make to eat high energy or low energy foods, depending on hunger state and reward value, and upon price and availability.
• The importance of environmental context on decision-making processes: how day-to-day choices and preferences are affected by socio-economic status, stress, and exercise levels.
Our focus was on tools that led to knowledge that could be translated into policy. We set out to develop new tools and experimental approaches to support the integration of behavioural and observational studies with neurobiological studies in a way that can lead to advances in consumer and nutrition research, providing the evidence base needed to educate stakeholders and inform policy. We need to understand what unconscious factors bias our choices and subvert our best intentions – or to look at the problem another way, how complex psychological and physiological mechanisms support healthy eating, and how these mechanisms come to be undermined. Policies on healthy eating must be framed in this setting if they are to be effective.

The general objectives of the Nudge-it project were:
1. To develop novel approaches to understand how early life experiences affect food choice. Early life experience has a major impact upon health throughout life, but little is known about how stress or poor nutrition in early life influences later food selection and valuation, and this is key to defining the timing and nature of policy interventions.
2. To develop experimental models of the environmental determinants of eating habits and food choice. We need new experimental approaches that help us to understand the life-long learning process - how habit-like and impulsive behaviours are controlled by physiological and endocrine regulators of food preference. These are essential for designing interventions that work with homeostatic control systems, rather than against them.
3. To optimise the neuroimaging of food choice. Human neuroimaging is an important emerging technology that can be used to define the neural circuits involved in food valuation and selection. We need to optimise the methodologies used in studies of food choice, address the needs for standardisation of protocols to facilitate data-sharing and meta-analysis, and build links with mechanistic studies in rodents.
4. To develop novel technologies to understand the brain mechanisms underlying food choice. We need to develop technology that addresses how molecular and cellular events, initiated by the exposure to food, translate into changes at the neuronal circuit level and how these changes translate to decisions on food choice.
5. To develop, optimise, and validate new and existing tools and technologies to understand how the life-long learning process contributes to food selection and valuation. To understand how healthy and unhealthy dietary decisions are learned and expressed we need to deconstruct food choice, to answer fundamental questions about the nature of post-ingestive effects, and how these are combined with sensory signalling from the mouth to the brain. We need new tools to help us understand the role of attention and interoception in dietary learning.
6. To model how physiological, psychological, and emotional factors predispose people to unhealthy eating. This includes overeating in response to external food cues or negative emotional states, and choosing ‘‘junk foods’’ over healthier food options.
7. To advance knowledge and develop new tools to model and test the decision processes involved in dietary choices, with a specific focus on low SES groups. Advice designed to help consumers make healthier choices has not achieved its goal. We must investigate methods of altering diet- and food choice-related behaviour. We need new models that explore decision-making and new tools to trigger behavioural change. We must focus on adults and children in lower socio-economic status (SES) groups as these groups are particularly resistant to campaigns aimed at promoting healthy behaviour.
8. To develop a predictive systems model of the interactions between physiological, psychological and emotional factors related to feeding behaviour. We need to integrate knowledge from diverse disciplines and approaches, exploiting the power of modern computational approaches to link mechanistic and systems level understanding in a way that can properly inform policy.
9. To construct a policy toolbox. It is essential that advice is supported by rigorous science, and that what is a complex message is fully understood by those likely to be delivering it to vulnerable target groups. We need policy tools based on frontier knowledge of the neurological, physiological and psychological mechanisms involved in dietary choices, and better ways of testing the effectiveness of policy interventions.


Project Results:
Description of the main results and findings

In this section we outline the main results from the project. In addition to showing how we attained our original objectives, we highlight additional or unexpected outcomes, and value added by the consortium’s collaborative approach.
Objective: To develop novel approaches to understand how early life experiences affect food choice.

Key outcome: Children’s dietary habits can be improved with simple interventions, this occurs independently of parent’s habits

To understand the role of early exposure, we conducted a randomized controlled experiment for 12 weeks with low income families who had at least one child between the age of 2 and 6. Low SES families were recruited in two different areas in the UK and were allocated to a control group of a treatment group. Families in the treatment group received weekly groceries at home, as well as recipes, to cook 5 family meals per week. These meals were designed by a professional nutritionist, and the intervention lasted for 12 weeks. We collected information about dietary preferences, dietary intake, and collected anthropometric measurements (height and weight) from the main adult carer and the youngest child in the family who was between the age of 2 and 6. Families were asked to come to the university for 6 visits in total (before the intervention, half way through the intervention, after the intervention, and then every year after the start of the intervention for three consecutive years). We were also able to collect information on a range of blood biomarkers for a subsample of the study before and immediately after the intervention.
The main conclusion from the study is that early exposure does seem to matter, but we did not find strong evidence of changes in food preferences of children. We did however find that their intake of sugar was significantly lower after the intervention, and so was their average BMI. In contrast, we found little evidence of any change in dietary habits among their parents. These results suggest that it is possible to affect children’s dietary habits even if their parents are unaffected.

Key outcome: Children’s preferences for fat but not sugar are associated with maternal consumption during pregnancy

Regarding the effects of in-utero exposure to obesogenic foods, we found that a reduction in the exposure to foods high in fat during pregnancy was associated with weaker preferences of the child for processed foods. We did not find such association for foods high in sugar. We also documented an association between the maternal consumption of foods high in fat and the probability of the child being overweight and obese where halving of the consumption of high fat foods eaten during pregnancy corresponds to a 0.2% reduction in the probability of being overweight and obese. These associations control for current maternal dietary intake and BMI, as well as socio-economic factors. These results suggest that there may be a mechanism of transmission of food preferences that operates in-utero, particularly when these preferences relate to foods that are high in fat. Further research is needed to understand whether it is possible to exploit this mechanism for designing interventions aimed at reducing the exposure of pregnant mothers to foods that are high in fat.

Key outcome: Children’s unhealthy food preferences are associated with maternal stress during pregnancy

We found that maternal stress during pregnancy significantly affected children’s food and taste preferences as well as their diet, even after controlling for maternal diet, current maternal stress and demographics of child and mother. Higher average stress during pregnancy is linked with food preferences and a diet that are significantly less healthy, and with weaker preferences for sour and bitter foods. Further research should aim at understanding the mechanisms driving the relationship between stress during pregnancy and food preferences of the offspring. Future research should also explore whether stress reduction interventions, based on psychological techniques for example, may be useful avenues to design policy interventions.

Key outcome: Development of translational models of early-life experience

To understand whether early-life experience can affect food choice in adulthood we developed a series of translational models of early-life (prenatal) stress and of early-life (postnatal) over-nutrition. Using two distinct models of prenatal stress (a physical stressor or a social stressor experienced by the pregnant rat), we found no evidence for changes in food choice in the adult male or female offspring of these stressed rats. This calls into question the reliability or relevance of these translational models in mimicking the influence of early-life experience on appetite control.
To model early-life over-nutrition, we adjusted downwards the litter size of rats, giving each pup greater access to the mother’s milk. Small litter (SL) pups’ body weight diverged upwards from pups in control litters before weaning and this increased body weight was maintained into adulthood. In adulthood we tested food choice in SL rats by offering normal chow, sucrose pellets and lard. Adult SL rats altered their food choice, unexpectedly showing a preference for normal chow over sucrose pellets and lard. This choice preference was maintained during refeeding after a fast. Interestingly, this behaviour is reminiscent of control rats administered ghrelin (see below). We only observed increased consumption of obesogenic foods by SL rats if access was restricted in a binge-like fashion.
These data indicate that early-life over-nutrition does not necessarily drive unhealthy food choices in rats, unless access to food is restricted.

Objective: To develop experimental models of the environmental determinants of eating habits and food choice.

Key outcome: Oxytocin neurons respond to a high-sugar food, but not to an isocaloric high-fat food

The neurohormone oxytocin is a potent inhibitor of appetite and has an important role in satiety signalling. We developed a novel translational model of gut-brain signalling in rats to study the activity of oxytocin neurons in response to voluntary consumption or oral gavage of foods. Using c-Fos expression as a marker of neural activation, we showed that oxytocin neurons are strongly activated during voluntary consumption of a high-sugar food (sweetened condensed milk; SCM). To model the effect of food in the stomach, we gavaged anesthetized rats with SCM, and showed that this activated the oxytocin neurons, and that this was not purely a result of gastric distention. We then recorded the electrical activity of oxytocin neurons in anesthetized rats during gavage with SCM (high-sugar) or isocaloric single cream (high-fat). Whereas oxytocin neurons responded to SCM gavage with a linear, proportional, and sustained increase in firing rate, cream gavage resulted in a reduction in firing rate. Thus the presence of specific foods with different macronutrient profiles in the stomach differentially regulates the activity of oxytocin neurons with potentially different effects on brain network signalling and food choice behaviours.
Next we investigated possible mechanisms linking gut-brain signalling. Given that we observed an increase in insulin secretion after SCM gavage, we focused on this as a potential mediator. Using rats, we showed that systemic administration of insulin enhances the electrical activity of oxytocin neurons, and the magnitude of this response is modified by feeding condition. We also found that activation of brain insulin receptors mediates these responses as prior receptor blockade abolished the excitatory responses of the oxytocin neurons. Thus, one mechanism by which the intake of sweet food is selectively signalled to the brain is by the induction of insulin secretion which specifically activates oxytocin neurones.

Key outcome: A translational model of snacking and caloric compensation

To complement the consortium’s human studies on snacking, we developed a translational model of snacking, with a view to understanding mechanisms of caloric compensation. Consumption of palatable foods has been implicated in weight gain, but this assumes that homeostatic control systems cannot accurately detect this hedonically-driven energy intake. This study tested this assumption, hypothesizing that satiated rats would reduce their voluntary food intake and maintain a stable body weight after consuming a palatable food.
Lean rats (or rats previously exposed to an obesogenic diet) were schedule-fed with fixed or varying amounts of sweetened condensed milk (SCM) daily, and their voluntary energy intake and body weight were monitored. During scheduled feeding of SCM, rats reduced bland food consumption and maintained a stable body weight. This behaviour was also seen in rats with access to an obesogenic diet and was independent of the predictability of SCM access, i.e. irregular access did not disrupt these homeostatic responses. However, rats offered large amounts of SCM showed an increase in energy intake. Thus, powerful compensatory reductions in voluntary energy intake were seen in this model, but under-compensation was observed if large amounts of SCM were consumed. We found no evidence that under-compensation (i.e. over-eating of bland food) results from eating being driven by requirements of food components absent from the snack, but this needs further investigation.

Key outcome: Ghrelin is a regulator of food choice in rodents

We sought to determine whether the orexigenic hormone, ghrelin, is involved in the regulation of food choice in rats. We hypothesised that ghrelin is suited to serve such a role given that it signals hunger information from the stomach to brain areas important for feeding control, including the hypothalamus and reward system. Rats were offered a choice of palatable foods (sucrose pellets and lard) superimposed on regular chow for 2 weeks, and we explored whether central delivery of ghrelin (intracerebroventricular (ICV) or intra-VTA) is able to redirect their dietary choice.
The major unexpected finding was that, in rats with high baseline lard intake, acute ICV ghrelin injection increased their chow intake over 3-fold, relative to vehicle-injected controls, measured at both 3 hr and 6 hr after injection. Similar effects were observed when ghrelin was delivered to the VTA, thereby identifying the VTA as a likely contributing neurobiological substrate for these effects. We also explored food choice after an overnight fast, when endogenous ghrelin levels are elevated, and found similar effects of dietary choice to those described for ghrelin. These effects of fasting on food choice were suppressed in models of suppressed ghrelin signalling (i.e. peripheral injection of a ghrelin receptor antagonist to rats and ghrelin receptor (GHSR) knock-out mice), implicating a role for endogenous ghrelin in the changes in food choice that occur after an overnight fast. Thus, in line with its role as a gut-brain hunger hormone, ghrelin appears to be able to acutely alter food choice, with notable effects to promote "healthy" chow intake, and identify the VTA as a likely contributing neurobiological substrate for these effects.
We also used an additional model to explore the effects of ghrelin on food choice, one in which it ought to be extremely difficult to change food choice away from a highly palatable food: we used rats trained to binge on a high-fat diet. Thus we explored whether ghrelin is able to influence binge-like eating behaviour in rodents. We used a palatable scheduled feeding paradigm in which ad lib chow-fed rodents are trained to 'binge' on a high-fat diet (HFD) offered each day for a limited period. In this model, rats can eat up to 80% of their entire daily caloric intake from the HFD that is eaten during the access period. After 2 weeks of habituation to this paradigm, rats were given ghrelin or vehicle solution directly into the brain. Remarkably and unexpectedly, during the palatable scheduled feed, when rats normally only binge on the HFD, those injected with ghrelin started to eat more chow and chow intake remained above baseline for the rest of the 24-hour day. We identified the ventral tegmental area (VTA) (a key brain area involved in food reward) as a substrate involved because these effects could be reproduced, in part, by intra-VTA delivery of ghrelin. Fasting (to increase endogenous ghrelin) immediately before a palatable schedule feed also increased chow intake during/after the schedule feed but, in contrast to ghrelin injection, did not reduce HFD intake. Chronic continuous central ghrelin infusion over several weeks enhanced binge-like behaviour in palatable schedule fed rats. Over 4 weeks, mice lacking the ghrelin receptor were able to adapt and maintain large meals of HFD in a manner similar to wild-type mice, suggesting that ghrelin signalling may not have a critical role in the acquisition or maintenance in this kind of feeding behaviour. In conclusion, ghrelin modulates binge-like eating behaviour by shifting food preference towards a more nutritious choice (from HFD to chow), with these effects being divergent from fasting.

Key outcome: Ghrelin conditions an avoidance in rodents (i.e. induces an unpleasant sensation)

We provided evidence that elevated ghrelin levels may feel unpleasant. Feelings of hunger carry a negative-valence (emotion) signal that is conveyed through agouti-related peptide (AgRP) neurons in the hypothalamus. Ghrelin, a hunger hormone, activates these neurons although it remains unclear whether it also carries a negative-valence signal. Given that ghrelin also activates pathways in the midbrain that are important for reward, it remains possible that ghrelin could act as a positive reinforcer and hence, carry a positive-valence signal. In the project we used condition preference/avoidance tests to explore the reinforcing/aversive properties of ghrelin, delivered by intracerebroventricular (ICV) injection (2µg/injection once a day for 4 days). We found that ICV ghrelin produced conditioned avoidance, both in a conditioned place preference/avoidance test (CPP/CPA, in which the animals avoid a chamber previously paired to ghrelin injection) and in a conditioned flavour preference/avoidance test (CFP/CFA, in which the animals consume/avoid a taste previously paired to ghrelin injection). These effects were observed whether conditioning to ghrelin occurred in the absence or presence of food, and were found in both mice and rats. We did not, however, find evidence that brain ghrelin delivery to rats induces malaise. Our data indicate that ICV ghrelin carries a negative-valence signal consistent with its role as a circulating hunger hormone.

Key outcome: Identification of the supramammilary nucleus (SUM) as a novel metabolic target for ghrelin and a glucagon-like 1 (GLP-1)-oestrogen conjugate molecule

We identified a novel neural target for circulating metabolic signals - the supramammilary nucleus, a hypothalamic area involved in motivation and reinforcement. First, we used a conjugate molecule that combines the gut/brain peptide, GLP-1, with oestrogen, this molecule has been suggested to represent a novel strategy to treat diabetes and obesity. Central administration of this molecule reduced food reward, food intake, and body weight in rats. To determine the brain location of the interaction of GLP-1 with oestrogen, we used single-photon emission computed tomography imaging of regional cerebral blood flow and pinpointed the supramammillary nucleus (SUM) as a potential target of the conjugated GLP-1-estrogen. We confirmed that conjugated GLP-1 and oestrogen directly target the SUM with site-specific microinjections. Additional microinjections of GLP-1-estrogen into classic energy balance controlling nuclei, the lateral hypothalamus (LH) and the nucleus of the solitary tract (NTS) revealed that the metabolic benefits resulting from GLP-1-estrogen injections are mediated through the LH and to some extent by the NTS. In contrast, no additional benefit of the conjugate was noted on food reward when the compound was microinjected into the LH or the NTS, identifying the SUM as the only neural substrate identified here to underlie the reward reducing benefits of GLP-1 and oestrogen conjugate. Collectively we discovered a surprising neural substrate underlying food intake and reward effects of GLP-1 and oestrogen and uncover a new brain area capable of regulating energy balance and reward.
We also demonstrated that the SUM is also a target for ghrelin. Given that ghrelin binds to the SuM, we explored whether SuM cells respond to ghrelin and/or are activated when endogenous ghrelin levels are elevated. We found that peripheral ghrelin injection activates SuM cells in rats, reflected by an increase in the number of cells expressing c-Fos in this area and also by the predominantly excitatory responses of SuM cells recorded in in vivo electrophysiological studies. Further c-Fos mapping studies reveal that this area is also activated in rats in situations when circulating ghrelin levels are known to be elevated: in food-restricted rats anticipating the consumption of food and in fed rats anticipating the consumption of an energy-dense food. We also showed that injection of ghrelin into the SuM induces feeding in rats. Collectively, our data demonstrate that the SuM is activated when peripheral ghrelin levels are high, further supporting the emerging role for this brain area in metabolic and feeding control.
Using cutting-edge neuronal tracing and viral-mapping approaches we investigated the connectivity of the SuM with other key appetite- and reward-related brain pathways. We showed that after consumption of SCM there were positive correlations between the level of c-Fos expression in the medial SuM and the LH, DMH and VTA. Using tracers, we have shown that the medial SUM receives inputs from the LH, DMH, VTA and VMH. Conversely, the medial SuM projects to the LH and DMH, and we identified a dopaminergic projection from SUMM to the DMH. Viral mapping in TH-Cre rats confirmed the existence of a reciprocal SuM-DMH connection and also showed that dopamine cells project from the SuMM to the lateral septum and cingulate cortex. An understanding of these connections establishes the place of the SuM in the known appetite- and reward-related circuits in the brain, and points to this region being a novel node in behavioural control of food choice.

Key outcome: Discovery of the “gravitostat”, a body weight homeostat that regulates fat mass independently of leptin

We contributed to a study identifying a new body weight sensing/regulating mechanism. Human subjects who spend a large amount of time sitting have increased risk of obesity, but the mechanism for the anti-obesity effect of standing is unknown. We hypothesized that there is a homeostatic regulation of body weight. We demonstrated that increased loading of rodents, achieved using capsules with different weights implanted in the abdomen or under the skin on the animal’s back, reversibly decreases the biological body weight via a reduction in food intake. Importantly, loading relieves diet-induced obesity and improves glucose tolerance. The identified homeostat for body weight regulates body fat mass independently of the fat cell-derived hormone leptin, revealing two independent negative feedback systems for fat mass regulation. It is known that bone cells called osteocytes can sense changes in bone strain. In this study, the body weight-reducing effect of increased loading was lost in mice depleted of osteocytes. We propose that increased body weight activates a sensor dependent on osteocytes of the weight-bearing bones. This induces an afferent signal, which reduces body weight.
These findings demonstrate that a leptin-independent body weight homeostat (a "gravitostat") that regulates fat mass. This has potentially very important translational implications, including the prospect of novel behavioral interventions to support maintenance of a healthy body weight. Future work will focus on defining the signaling mechanisms between bone and brain and defining the target actions in the brain.

Key outcome: “Healthy snacking” in children has only a short term effect on BMI, no effect on food preferences, and is difficult to implement

We tested the impact of regular versus irregular rewards (sweet energy-dense snacks, <200 calories per item) in a randomized controlled field experiment involving low income families with children between the age of 2 and 6. Families were randomly assigned “regular snacking protocol”, requesting them to eat three meals at regular times during the day and avoid snacks between meals, except for one day per week). Children were allowed to consume a morning and afternoon healthy snack in addition (provided), at regular times. The goal was to evaluate the impact of the intervention on reported diet and snacking behaviour (quantity of snacks consumed, number of meals, fruit and vegetable consumption, total calories consumed), food preferences, anthropometric measurements and health status measures (based on blood samples in the adult population to measure overnight fasting glucose). We found no evidence of changes in dietary habits and health outcomes in the adult population, but the children of the treated families had a lower BMI than the control group up to two years after the intervention. The difference was no longer significant in the third year. We found no evidence that such protocol led to a long-term change in the preferences of children, specifically for obesogenic foods. Implementing this protocol was experienced as difficult or very difficult by most families. Therefore, it is unclear if the lack of significant changes observed is due to the absence of a relationship between regularity of eating patterns and dietary preferences, or to the weak adherence to the recommended protocol.
This work has important implications. It provides strong evidence that gentle behavioural interventions in children’s diet can produce a relatively prolonged beneficial impact, but whether these beneficial changes are persistent enough to make an enduring difference over the lifespan seems unlikely.

Objective: To optimise the neuroimaging of food choice.

Key outcome: Imaging tools to study attentional networks

To further investigate inhibitory processes we adapted a food-related version of the attention network task and used functional magnetic resonance imaging to study the effects of disinhibition on attentional networks in normal-weight participants. High disinhibition scores were associated with a rapid reorienting response to food pictures after invalid cueing and with an enhanced alerting effect of a warning cue signalizing the upcoming appearance of a food picture. Imaging data revealed activation of a right-lateralized ventral attention network during reorienting. The faster the reorienting and the higher the disinhibition score, the less activation of this network was observed. The alerting contrast showed activation in visual, temporo-parietal and anterior sites. These modulations of attentional networks by food-related disinhibition might be related to an attentional bias to energy dense and palatable food and increased intake of food in disinhibited individuals.
We implemented in collaboration with the SMG Artinis Near-Infrared Spectroscopy (NIRS) system for the assessment of food choices. We showed that this technique is a suitable method to assess food choice related neural signals in healthy participants (lean and obese) and in patients. We finalized the investigation of neural and behavioural effects due to attentional bias modification by mind-set induction during food choices in lean and obese subjects. We used our newly developed fMRI measurement protocols and analysis techniques to assess neural alterations in functional dyspepsia patients during eating related behaviour.

Key outcome: Shared tools for generating standardised fMRI tasks in the assessment of food choice-related brain responses

To foster better standardization of measures and more efficient data pooling we created an online Nutritional Neuroscience Toolbox for generating standardised fMRI tasks for assessing food-related brain responses. This toolbox includes a large set of standardised food images together with the fMRI paradigms and relevant presentation software scripts, covering a diverse range of food-related tasks, including food viewing, food choice, tasting, ingestion and olfaction. These resources can be found at http://nutritionalneuroscience.eu/index.php/resources/nutneurotoolbox. It also features an open measurement toolbox for standardized collection and processing of questionnaire data. It is freely available at http://sourceforge.net/projects/forcplatform/ and has been downloaded more than 250 times.
We implemented a neuroimaging data-sharing and analysis platform tailored to nutritional neuroscience, building on the successful NeuroVault platform. This platform allows easy sharing of nutritional neuroimaging outcomes in the form of subject or group level statistical maps. Moreover, a key novel feature is the possibility to add image meta-data such as age, gender and BMI, but also fields tailored for the nutritional neuroscience community such as degree of hunger, body fat, personality characteristics such as impulsivity and reward sensitivity, and blood parameters like glucose. Together this allows unprecedented pooled analysis of individual food response profiles across multiple studies while taking relevant factors into account. We are currently using this to perform a unique mega-analysis on food-viewing fMRI data to establish the effects of age, gender and BMI on food cue reactivity in a very large sample of aggregated data. To foster such initiatives we created a dedicated Nutritional Neuroscience community page to solicit maps from the international nutritional neuroscience community. In addition, the ability to perform meta-analyses on group level maps was added at https://neurovault.org/my_metaanalyses/. This feature makes it easy for users to select group level maps while browsing NeuroVault and perform a Stouffer’s meta-analysis on them. The functionality we were able to develop and deploy exceeds our initial plans and we believe that the much enhanced functionality of NeuroVault will not only greatly benefit the nutritional neuroscience community as intended, but also other neuroscience communities.

Objective: To develop novel technologies to understand the brain mechanisms underlying food choice.

Key outcome: Translational tools to model decision-making in food choice

We developed translational models that make it possible to determine the impact of palatability and caloric value on food choice. These models also make it possible to determine endogenous states as impulsivity and willingness to endure uncertainty and punishment in food related decision making.
We unraveled a neuronal mechanism underlying food-related decision-making deficits during hyperdopaminergic states. To do so, we implemented fiber photometry to record from dopamine neurons in awake rats while they performed food-decision making tasks where delivery of food was uncertain or which required waiting longer to obtain larger food reward. In addition, we used models where a premature response to obtain a food reward was punished by a mild electric food shocks. We showed that alterations in the dopamine neural circuit impact on food-related decision making. We also showed that food related decision making in which cognitive capacity was taken into account was affected by the female rat estrous cycle, suggesting an impact of hormones on food-related decision making. Importantly, females differ in their food-related decision making processing depending on the phase of the estrous cycle. These data resonate with the well-known differences in food intake in woman across the menstrual cycle and during menopause. We also found in anorexia nervosa patients that when uncertainty is introduced in decision making tasks, learning from losses is reduced, likely due to increased dopamine neuronal activity in brain reward pathways. This latter study was inspired by the discoveries we made in rodents in which we chemogenetically activated dopamine neurons.
We determined the impact of leptin and ghrelin on food cue-induced firing of VTA dopamine neurons – a key node in the brain reward pathway. We found that reduced leptin signaling is a key factor in recruiting VTA dopamine neuronal activity in food-related decision making during states of negative energy balance. This inspired us to focus more on how neural circuits implicated in energy balance connect and influence the dopamine system. We demonstrated that endogenous need states (hunger or salt deprivation) recruit the dopamine system in changing food choice. In the control state, salt is aversive to rats. However, when rats are salt-depleted, salt becomes preferred and this is associated with a shift in VTA dopamine neuronal activity from reduced activity when tasting salt towards increased activity in the salt-depleted state. Thus endogenous states have a major impact on food choice and which neural circuits are engaged in a change in food choice.
We developed novel tools to understand the brain mechanisms underlying food choice from the cellular towards neural circuit level. We used fibre photometry to measure the activity of dopamine and other neurons during food choice and found that leptin sensitive neurons affect neuronal activity. We used chemogenetics and optogenetics to manipulate specific neuronal activity in order to unravel the precise role of specific neurons in food related decision making indirectly via increasing GABAergic input onto VTA dopamine neurons. Using chemogenetics, we found that these GABA-ergic leptin sensitive neurons impact on the motivation to obtain a food reward. We also found differential roles for subsets of lateral hypothalamic on food-related decision making. Using chemogenetics, we found that inputs from neurons that originate in the zona incerta and that project to the VTA, enhance the motivation to obtain food reward.
We demonstrated that enhancing excitability of dopamine neurons promotes motivational behaviour through increased action initiation and that this explained the increased motivation to obtain a food reward. Increasing the activity of midbrain dopamine neurons with chemogenetics disrupted feeding patterns, likely because of an effect on attention. We also developed a novel viral vector system that is dependent on two different recombinases to achieve neuronal subtype and projection specificity.
To bridge the gap between cellular processes and neural network activity, we developed technology to activate specific neurons with chemogenetics and record neural network activity as assessed by neuroimaging in an fMRI scanner. We also determined the effect of hormones, hunger and satiation processes on this neuroimaging response.
We developed novel models to assess the role of different mood states in food-related decision making. We introduced a novel model to assess the impact of a negative mood state induced by presenting a tone that was previously associated with a food shock on food-related decision making. We also found that chemogenetic activation of midbrain dopamine neurons affects attention, but not impulsivity, in a task where rats need to be attentive and suppress premature responding (impulsivity) in order to obtain a food reward.

Objective: To develop, optimise, and validate new and existing tools and technologies to understand how the life-long learning process contributes to food selection and valuation.

Key outcome: Tools to investigate hypothalamic networks in humans

Innovative brain imaging tools for obesity research are exceptionally important for understanding how obesity develops, and for identifying which brain-oriented intervention may help to prevent obesity and improve weight management in particular weight loss and its maintenance. We developed and implemented advanced imaging technology and innovative experimental designs with functional magnetic resonance imaging (fMRI) and magnetoencephalography. In contrast to animal models, the neural circuitry of the human hypothalamus has not been delineated. Hence, we aimed to map the hypothalamus network using resting state functional connectivity (FC) analyses from the medial hypothalamus (MH) and lateral hypothalamus (LH) in healthy normal-weight adults. In a separate sample, we examined differences within the LH and MH networks between healthy normal-weight versus overweight/obese adults. FC patterns from the LH and MH revealed significant connections to the striatum, thalamus, brainstem, orbitofrontal cortex, middle and posterior cingulum and temporal brain regions. We found subtler distinctions within hypothalamic subregions. The LH was functionally stronger connected to the dorsal striatum, anterior cingulum, and frontal operculum, while the MH showed stronger functional connections to the nucleus accumbens and medial orbitofrontal cortex. Furthermore, overweight/obese participants revealed heightened FC in the orbitofrontal cortex and nucleus accumbens within the MH network. Our results indicate that the MH and LH network tap into different parts of the dopaminergic reward circuitry of the brain, potentially modulating food reward based on the functional connections to the ventral and dorsal striatum, respectively.
In obese adults, FC changes were observed in the MH network. In addition twenty-five lean and twenty-two overweight/obese participants underwent functional magnetic resonance imaging, on two separate days, before and after intranasal insulin or placebo application. Compared to placebo administration, insulin resulted in increased FC between the prefrontal regions of the default-mode network and the hypothalamus as well as the hippocampus. The insulin-induced change in hypothalamic FC showed a significant interaction with peripheral insulin sensitivity. Only participants with high peripheral insulin sensitivity showed a boost in hypothalamic FC. The change in hippocampal functional connectivity significantly correlated with visceral adipose tissue and the change in subjective feeling of hunger after intranasal insulin. Mediation analysis revealed that the intranasal insulin induced hippocampal FC increase served as a mediator, suppressing the relationship between visceral adipose tissue and hunger. Hence, brain insulin action may facilitate weight loss by modifying brain FC within and between cognitive and homeostatic brain regions.

Key outcome: Quantitative tools to study white matter in the human brain

Structural brain imaging studies have shown that obesity is associated with reduced grey matter (GM) volume, however the knowledge on changes in white matter are rather limited. The investigation of white matter changes is of special importance for the investigation of possible changes in brain network structure. To evaluate the applicability of advanced quantitative imaging techniques for white matter changes in obesity, we explored brain tissue properties by combining diffusion tensor imaging (DTI) with quantitative multi-parameter mapping (MPM) in lean, overweight and obese young adults. Most studies until now have shown a loss of white matter integrity with obesity-related factors in several brain regions quantified by DTI. However, the variety of obesity-related factors resulted in competing influences on the DTI indices. By MPM, white matter structures showed differences in MRI parameters consistent with reduced myelin, increased water and both increased and decreased iron content with increasing BMI in the superior longitudinal fasciculus (SLF), anterior thalamic radiation (ATR), internal capsule and corpus callosum. BMI-related changes in DTI parameters revealed mainly alterations in mean and axial diffusivity with increasing BMI in the corticospinal tract, ATR and SLF. These alterations, including mainly fiber tracts linking limbic structures with prefrontal regions, could potentially promote accelerated aging in obese individuals leading to an increased risk for cognitive decline.

Key outcome: Brain reward processing may be different when key reward signals are depleted

Reward sensitivity and possible alterations in the dopaminergic-reward system are associated with obesity. We investigated the influence of dopamine depletion on food-reward processing. We investigated female subjects in a randomized placebo-controlled, within subject design using an acute phenylalanine/ tyrosine depletion drink representing dopamine depletion and a balanced amino acid drink as the control condition. Brain activity was measured with fMRI during a ‘wanting’ and ‘liking’ rating of food items. Eating behavior-related traits and states were assessed on the basis of questionnaires. Dopamine depletion resulted in reduced activation in the striatum and higher activation in the superior frontal gyrus, independent of BMI. Brain activity during the wanting task activated a more distributed network than during the liking task. This network included gustatory, memory, visual, reward, and frontal regions. An interaction effect of dopamine depletion and the wanting/liking task was observed in the hippocampus. The interaction with the covariate BMI was significant in motor and control regions but not in the striatum. Our results support the notion of altered brain activity in the reward and prefrontal network with blunted dopaminergic action during food-reward processing. This effect is, however, independent of BMI, which contradicts the reward-deficiency hypothesis. This hints at a different or more complex mechanism underlying the dopamine reward function in obesity.

Key outcome: Nutritional information can influence pathophysiological responses to food

High-fat meals are associated with dyspeptic symptoms in functional dyspepsia (FD) patients. It is still unclear how fat is processed, or how FD symptoms and neuronal activities are modulated by psychological factors. We investigated brain activity by fMRI after ingestion of high- and low-fat foods with correct/incorrect fat information. 12 FD patients and 14 healthy controls (HCs) were compared. We recorded resting-state fMRI on four days before and after ingestion of four yogurts (200 mL, 10% or 0.1% fat, “low fat” or “high fat” label). FD patients showed more dyspeptic symptoms than HCs, and symptoms were relieved less after consuming high fat–labeled yogurt than low fat–labeled yogurt, irrespective of the actual fat content. This is indicative of either a placebo effect of low-fat information or a nocebo effect of high-fat information on symptom expression. FD patients showed greater activity than did HCs in occipital areas before and after ingestion regardless of fat content and label, as well as greater activity in the middle frontal gyrus before ingestion. In addition, functional connectivity (FC) from the insula to the occipital cortex (I-O) increased after high fat ingestion and decreased after low fat in FD patients.

Key outcome: Tools to study the effect of portion size on food choice in different mind-sets

Obesity develops due to an imbalance between energy intake and expenditure. Besides the decision about what to eat, daily energy intake might be even more dependent on the decision about the portion size to be consumed. For decisions between foods, attentional focus is considered to play a key role. We investigated the attentional modulation of portion size selection during pre-meal planning. Thus, we designed an fMRI task in which healthy, lean participants selected their ideal portion size for lunch in dependence of different mind-set instructions directing their attention to different aspects of the meals. Compared with a free choice condition, participants reduced their portion sizes when they had to consider health aspects and when they planned to eat with pleasure and showed increased activity in left prefrontal cortex and left orbitofrontal cortex, respectively. When they were planning to be full until dinner, they selected larger portion sizes and we observed increased activity in left insula. These results provide first evidence that the cognitive process of pre-meal planning is influenced by the attentional focus at the time of choice, which might provide a key opportunity to influence the control of meal size selection by mind-set manipulation. To extend our knowledge in this respect we investigated the neural underpinnings of pre-meal planning in lean, overweight and obese adults. Participants of all weight groups reduced their portion size when adopting a health mind-set, accompanied by enhanced left prefrontal cortex activation. Fullness and pleasure mind-sets resulted in altered reward responses in the overweight/obese compared to lean. Under the pleasure mind-set, overweight/obese persons responded with heightened right frontal operculum activity, while the fullness mind-set induced a decreased response in the ventral striatum. Overweight/obese participants did not modify their behaviour under the pleasure mind-set and selected larger portions than the lean group. We identified specific brain response patterns as participants made a final choice of a portion size. Thus, different brain responses and behaviour during pre-meal planning can inform the development of strategies for healthy weight management.

Key outcome: Magnetoencephalography-based tools to study decision-making

To assess the temporal dynamics of cortical decision-making processes we used magnetoencephalography (MEG) to investigate eating related cortical processes. The prefrontal cortex has a pivotal role in top-down control of cognitive and sensory functions. In complex go-nogo tasks, the right dorsolateral prefrontal cortex is considered to be important for guiding the response inhibition. However, little is known about the temporal dynamics and neurophysiological nature of this activity. To address this issue, we recorded magnetoencephalographic brain activity in 20 women during a visual go-nogo task. The right dorsolateral prefrontal cortex showed an increase for the amplitude of the event-related fields and an increase in induced alpha frequency band activity for nogo in comparison to go trials. The peak of prefrontal activity preceded the mean reaction time of around 360 ms for go-trials, supporting the proposed role of right dorsolateral prefrontal cortex in gating the response inhibition and suggesting that right prefrontal alpha band activity is involved in this gating. However, the results in right dorsolateral prefrontal cortex were similar for both successful and unsuccessful response inhibition. In these conditions, we observed pre- and poststimulus differences in alpha band activity in occipital and central areas. Thus, successful response inhibition also depend on prestimulus anticipatory alpha desynchronization in sensory areas. In conclusion, we suggest a role for functional inhibition by alpha synchronization not only in sensory areas, but also in prefrontal areas.

Key outcome: MRI-based tools to study perceived satiety

Little is known about how properties of foods such as thickness (viscosity), taste and energy-density affect our perceptions and food choice behaviours. We developed tools to investigate this. We found using MRI that increasing viscosity is less effective in slowing gastric emptying than increasing energy density. However, viscosity is more important to increase perceived fullness. This underscores the lack of satiating efficiency of “empty calories” in quickly ingested drinks such as sodas. The increase in perceived fullness due solely to increased viscosity, a phenomenon we refer to as “phantom fullness”, may be used to lower energy intake.
We found that our MRI measure of gastric emptying is more sensitive than the more commonly used breath test (13C tracer method). Although the breath test can assess relative gastric emptying differences on a group level, individual differences between different liquid foods were not accurately captured. This finding underscores that food matrix effects should be considered when interpreting the results of breath tests and that MRI techniques are better for research on gastric emptying dynamics. Another reason that MRI is to be preferred is that MRI provides additional information on the behaviour of the food in the stomach, such as layer formation and gastric sieving. In indirect gastric emptying measurements such as breath tests, the behaviour of labelling agents may be affected by the layering and confound emptying measurements. We showed that gastric sieving can occur for liquid foods; water can drain from the stomach while a layer of nutrient rich liquid is retained. This shows that when water is blended into a food (as in a soup or meal shake e.g.) gastric emptying is slower compared to separate consumption of water after the meal, since this will sieve through and thus yield less stomach distention and lower satiety.
We pioneered the combination of MRI measurements of gastric emptying in conjunction with brain activity measures and showed how this can be used to study gut-brain interactions associated with satiety. We examined with MRI how several individual characteristics such as age, gender, BMI and interoceptive awareness impact on gastric interoceptive ability. This provides an excellent stepping stone for further research on how differences in interoceptive ability may be exploited to affect appetite with personalized interventions.
Using a novel combination of measures we extended novel insights in the area of sweetness and carbohydrate metabolism in collaboration with Prof. Dana Small (Yale, member of the consortium’s External Advisory Group) on flavour-nutrient learning. We showed that sweetness interacts with caloric content to produce greater metabolic responses for sweeter drinks, which are driven by underlying differences in gastric emptying. This may have important implications for glucose homeostasis after the consumption of ‘light’ soft drinks which are too sweet relative to the amount of calories they contain. This approach provides an excellent foundation for further work, which should help resolve the controversies about artificial sweeteners and their physiological effects.

Objective: To model how physiological, psychological, and emotional factors predispose people to unhealthy eating.

Key outcome: A translational model of positive mood and food choice in rats

It is often asserted that positive mood is associated with healthier food choices, and that unhealthy food choices may improve mood. We mirrored the consortium’s studies on emotional eating in humans by developing a model of positive mood in rats and studying the effects on food choice. We established used ultrasonic vocalisations (USVs) of a specific frequency range as an indirect measure of mood in rats. We confirmed previous findings that heterospecific social interactions with rats (i.e. a human experimenter playing with a rat) increase the number of USVs emitted by the rat, implying that positive mood can be quantified by measuring these USVs. However, we found that access to palatable food had no effect on USVs regardless of whether access was predictable. We also induced positive mood in rats (using play) and studied food choice. We saw no effect of mood on sucrose choice, indicating that positive mood does not reduce unhealthy food choices. Nor did we see an effect of fasting or access to palatable food on rats’ preference for heterospecific social interactions, indicating that the motivation for contact is not affected by metabolic state or access to food. These data indicate at most a weak link between positive mood and food choice in rats.

Key outcome: “Real-world”-level stressors impair dietary self-control

We focused on the impact of acute stress on food choices and the brain using stress induction procedures that generate a moderate level of stress. These stressors are consistent with events that occur often in daily life. We found that moderate stress makes it less likely people will forego making healthier food choices. In other words, acute stress impairs dietary self-control. These results are consistent with common notions of the use of “comfort foods” to alleviate stress or unhappiness. However, we also sought to understand how the brain is influenced by stress and what types of changes neural activity are associated with the changes in food choice behaviour.
Our analyses of fMRI revealed distinctions between how the brain responds to high levels of perceived stress versus high levels of the stress hormone cortisol. In other words, mental and physical reactions to a moderate stressor impact brain function in different ways. This is both good and bad news in regard to buffering negative impacts of acute stressors on eating behaviour. The bad news is that mental and physical reactions to stress can synergise to have a stronger influence on brain function and food choices. The good news is that we have multiple pathways we can try to target to reduce the effects of stress. For example, increased cortisol release may be an unavoidable response to stress, but if we can find ways to prevent or quickly reduce the perception of psychological stress, then we may reduce the negative impact of acute stressors on food choices.

Key outcome: Validation of a novel tool to measure individual self-control behaviour

We demonstrated that individual self-control behavior and its underlying mechanisms can be better quantified using an individual trait measure of resting heart rate variability (HRV), a marker of physiological and psychosocial health. While a positive association between HRV and cognitive functions related to self-control had been reported in numerous behavioral studies, evidence on the underlying neural mechanisms was still very sparse. Our work showed that dietary self-control success and trait HRV are positively correlated. Extending these behavioral associations, we demonstrated a link between a behavioral mechanism of dietary self-control, down-weighing of taste information at the time of choice, and neural activity measured with BOLD fMRI in the ventromedial prefrontal cortex (vmPFC). We found a negative relationship between HRV and the strength of the taste attribute representation in vmPFC: the higher the HRV, the lower the individual represents the taste of tempting items when facing challenging food choices. This is noteworthy because many neuroimaging studies have shown that activity levels in the vmPFC are correlated with the value a person assigns to a good or bad outcome. Thus, people with higher HRV may be able to better shift the way they evaluate foods away from tastiness alone in order to encompass other factors such as calories, nutrients, and other factors related to healthiness.
These results indicate that efforts to link cognition with central and peripheral neurophysiology may promote a better understanding of individual differences in health and cognitive behaviors, and provide opportunities for prediction and early intervention against dysfunctions. HRV may be a useful tool for scientists seeking to capture individual variance in cognitive control capacities. By including this trait characteristic into models of self-control, we can robustly improve the amount of explained variance in choice behaviour.

Key outcome: A novel computational tool to understand decision-making in food choice

We used computational modeling of food choice outcomes and response times to better understand the mechanisms of decision-making. Traditionally, models of decision making and self-control have assumed that all attributes of the choice options are considered simultaneously or these ignored temporal dynamics altogether. We demonstrated that this assumption is often incorrect, by applying a novel time-varying sequential sampling model to data from several different food-choice experiments. Most datasets we analyzed show a substantial difference when participants begin to process the tastiness relative to the healthiness attributes of food rewards. We found that how quickly and how strongly an attribute impacts the decision are the results of two, dissociable processes. There is considerable individual variability in whether tastiness or healthiness is considered first: about half of the participants considered healthiness first, and half tastiness first. By estimating when and how strongly these influence choices, we can better explain and predict individual differences in dietary self-control under baseline conditions as well as how attention cues and brain stimulation alter self-control.
Our results show that weighting and timing aspects explain unique aspects of the observed changes in behavior following attention cues. Cues drawing attention to the healthiness of food items caused 85% of participants to make healthier choices more frequently. At the group level, this change in behavior was accompanied by significant shifts in both when and how strongly those attributes were considered. However, whether the cues changed when or how strongly attributes were weighted varied greatly across individuals. Only 33% of participants changed how strongly they weighted both attributes as well as the relative timing of when they were considered in favor of making the healthier choice (i.e. healthiness stronger and faster than tastiness) relative to the baseline condition. Most participants changed only one or the other of the two processes and our analysis allowed us to detect this heterogeneity in what would otherwise be mistaken for a uniform behavioral response to the healthiness cues. This has implications for interventions designed to foster self-control in food choices and many other domains. Examining relative differences in when the attributes begin to be processed as well as their weighting strength may be useful in understanding why interventions and policies work in some cases (e.g. for specific types of individuals) but not in others, and may help to increase the effectiveness of these interventions.
In addition to revealing individual differences in the mechanistic response to attention cues, we used the time-varying sequential sampling model to elucidate the effects of transcranial direct current stimulation over the left dorsolateral prefrontal cortex (dlPFC). Although several neuroimaging studies have reported correlations between left dlPFC activity and dietary self-control, this is the first test of its causal role in dietary self-control. We found that cathodal, but not anodal stimulation over the left dlPFC caused a reduction in self-control relative to sham stimulation. Our computational modeling showed that this reduction in self-control was the result of a selective increase in how strongly tastiness attributes were weighted in the absence of a change in how quickly tastiness or healthiness was processed. In other words, our tDCS experiment reveals a dissociation between the neural systems that underlie the processes that determine when and how strongly different attributes are weighted during a decision. This work provides both a concrete advancement in our knowledge of decision mechanisms and a computational modeling approach that can be applied to extract deeper mechanistic insights from data on choice outcomes and response times. We made code for fitting this type of model to new datasets publicly and freely available at https://github.com/galombardi/method_HtSSM_aDDM

Key outcome: Tools to investigate the influence of food dimensions on choice in humans – a “food decision profile”

In humans, weight gain typically occurs over many years, hence the impact of any single dietary decision is likely to be trivial relative to the accumulative effect of subtle and chronic differences in food selection strategies. To expose these differences researchers have tended to rely on large data sets, e.g. diet-diaries, dietary recall, or grocery shopping history. Although such approaches can reveal important associations between dietary patterns and %fat mass, they lack information about the underlying motivations that drive choices. To address this problem, we developed a paradigm designed to isolate and quantify a) the relative importance of different food dimensions in human food choice, and b) relate individual differences in food-choice strategies to %fat mass. A community sample of participants with a wide BMI range completed six different judgment tasks and made hundreds of decisions between pairwise presented foods.
Because food options varied along important food dimensions, choices involved trade-offs. A person who regards portion size as especially important will, for example, preferentially choose foods that are presented in larger serving sizes – even though they may be less palatable and less healthy than other options. Under these conditions, the relative importance of different food dimensions can be quantified for individual participants, thus measuring each participant’s ‘food decision profile’. Lean, overweight, and obese participants individually evaluated 30 lunchtime foods for palatability, perceived healthiness, expected satiation, and perceived energy content, and made 250 decisions between pairwise presented foods. Individual food-choice models correctly predicted 84% of participants’ choices. Critically, some participants placed more emphasis on some dimensions than others. In a second stage, participant decision profiles were entered in a regression model to predict %fat mass.
Importantly, although palatability strongly predicted choice, palatability was a non-significant predictor of fat mass. Instead, those with higher %fat mass placed greater emphasis on amount of food (through calorie content, expected fullness and food weight). Participants’ ability to discriminate the energy content of foods was measured with an independent psychophysical task. Remarkably, participants with greater fat mass showed better performance. Thus, the tendency for overweight and obese participants to “select more when it is available” may be mediated by improved discrimination of energy content.
We developed novel methods for identifying subtle differences in dietary decision-making that chronically impacts %fat mass, and we have demonstrated how subtle differences can be measured and used to predict chronic health outcomes. In a follow-on project we have applied these tools to understand how preference for protein-containing foods can impact lean muscle mass in an aging population (https://research.ncl.ac.uk/proteinforlife/)
Key outcome: Tools for understanding whether humans are adapted to the modern food environment and are accepting of food reformulation
Using methods developed for our food-choice architecture, we studied the relationship between food choice and food energy density (kcal/g). Although previous studies had explored brain and behavioural responses to low and high energy-dense foods, no studies had considered how adult humans differentiate foods that vary across a broad range of energy densities. In two studies, participants completed several tasks to assess the ‘value’ of a range of foods with different energy densities, ranging from vegetables (~0.1 kcal/g) to chocolate (~5 kcal/g). Across studies and in several different tasks the same compressive power function was observed. This is important, because it indicates that human physiology is poorly adapted to evaluate foods that have an unusual (high) energy density. This has implications both for our understanding of how “modern” energy-dense foods affect choice and energy intake, and for strategies aimed at removing calories from highly energy-rich foods. The results show that energy-rich foods are poorly discriminated based on energy content. In other words, modifying or creating foods to be even more energy-dense is unlikely to yield a significant increase in consumer acceptance. Equally, these data have been used to make the case to the food industry that, for already energy-rich products, reformulations aimed at reducing calorie content are unlikely to dramatically reduce consumer acceptance.

Key outcome: Tools to investigate the relationship between portion size and food choice

Although portion size is an important driver of larger meals it remains unclear how serving larger portions might impact food choice. Two studies were conducted to study the trade-off between portion size, palatability and expected satiety in food choice. We approached this problem using the novel methods outlined above. Across a range of portion sizes, palatability remained a consistent and positive predictor of food choice. By contrast, expected satiety was favoured, but only when small portions were compared. Together, these findings highlight an added complexity to food choice. Specifically, they show how the roles of palatability and expected satiety can be isolated and quantified, and how their importance varies with portion size and context. For the first time, this work showed that larger portions not only promote the consumption of larger meals, but also encourage the adoption of food choice strategies motivated solely by palatability.

Key outcome: Tools to investigate the relationship between chaotic eating and BMI

In several countries, including the UK, Australia, and Canada, regular meal timings are recommended for weight loss. Similarly, cognitive behavioural therapies for binge eating and obesity prescribe a regular, structured, meal pattern. However, the evidence supporting a relationship between BMI and irregularity of meal timings had not been explored previously. We characterised “chaotic eating” as the tendency to eat at variable times of day. In two studies, we used a novel measure to explore the relationship between BMI and chaotic eating. In the first study, we measured BMI and used a self-report measure to assess the usual range of times that meals and snacks are consumed over a seven-day period, as well as meal and snack frequency. After adjusting for age, gender, and dietary habits we found no relationship between BMI and chaotic eating of meals or snacks. In the second study, we calculated the same chaotic eating index (meals and snacks) using extensive data from the UK National Diet and Nutrition Survey of adults 2000–2001 (seven-day diet diaries). Again, we found little evidence that BMI is associated with chaotic eating of meals or snacks. Together, these results suggest that eating at irregular times of day does not promote weight gain. These findings challenge public-health guidelines that recommend regularity in meal timings for weight loss.

Objective: To advance knowledge and develop new tools to model and test the decision processes involved in dietary choices, with a specific focus on low SES groups.

Key outcome: Behavioural economics tools to study dietary choices

We aimed to develop a behavioural economics model of dietary choices based on the idea that poor dietary choices may be driven by rational inattention in a low SES population. In a world with competing demands, we hypothesized that dietary choices may not be prioritized because their consequences are less immediate. We developed a model that allowed us to derive precise predictions and identify possible anchors for policy interventions that we then tested experimentally. We also aimed to evaluate the effectiveness of behavioural economics tools targeting important triggers to behavioural change in a low SES adult and children population. We conducted two studies, one with a low SES adult population and one with low SES children.
The first study involved an adult population. We hypothesized that if poor dietary choices are (partly) driven by the fact that they are not prioritized, providing easy-to-grasp information to make healthy choices and/or nudging people into spending more time contemplating their choices may affect their choices. To study dietary choices, we developed an easy-to-use tool, called “Food Choice Tool”, which measures how people allocate a specific budget across different food items. They have 100 items to choose from, which correspond to the most popular items in a UK supermarket. The measures of dietary choices obtained from the food choice tool appears more correlated with anthropometric measurements than other measures, based on self-reports (food frequency questionnaire or 24h dietary recall).
We invited around 300 low SES individuals to the laboratory and measured their dietary choices. We varied the nature of information they received and the time they had to make dietary choices. They were randomly allocated to one of three groups. Either they received tailored dietary advice, generic dietary advice or no advice. We found that a large share of participants who received tailored health information received good news. This group tended to become more optimistic about their health after having received the information than a group who did not receive any information and a group who received generic health information. However, it is that latter group that responded most to the information intervention. This group was less likely to choose foods high in fat in the food choice tool. We also varied the time participants had to make their food choices. The hypothesis was that those who had little time would rely more on heuristics and habits they have and would not respond to the information we gave them. In practice, we found no significant differences according to the decision time. These results are important in the current policy context of design of tools providing tailored health advice. Our results suggest that people are not underestimating the health risks they face; on the contrary, they recognise these risks. Important research avenues include understanding better when and how tailored advice may be suited and for which populations.
The second study relates to children. We evaluated the effects of a behavioural tool aiming at encouraging children to taste and eat “healthy” foods. We conducted a randomized experiment in primary schools in Germany. We combined this study with an experimental study on rodents conducted by the Gothenburg group, which complements our results (see above). The specific tool or mechanism we look into is the use of food as a reward to encourage the development of a taste for that food. We find evidence supporting this mechanism both in the human and in the rodent studies. For children, we find that when children are rewarded with a specific food item (in our case, a dried apple), they develop a taste for it. They are more likely to report liking it and more likely to choose it among alternatives, relative to a control group. Importantly, this is only true when the task for which they are rewarded does not require much effort.
In rodents we sought to determine whether animals increase consumption of a food offered previously as a reward. Rats were trained in a progressive ratio operant responding apparatus to press a lever for a sucrose reward. In this paradigm, they have to work increasingly hard for each sucrose pellet earned. Control rats received the same number of sugar pellets but did not have to work hard to get them. On the next day, we explored spontaneous sucrose pellet consumption, expecting that those who had worked hard to get the sugar pellets would eat more. An important insight from this study is that the increase in consumption of food used as a reward leads to an increase in overall calorie intake over. Such an effect is almost impossible to document in human studies. Further research is need to understand whether this mechanism could be exploited to design interventions aimed at strengthening preferences for non-obesogenic foods among children.

Objective: To develop a predictive systems model of the interactions between physiological, psychological and emotional factors related to feeding behaviour.

Key outcome: A spiking neuronal network model of decision-making hypothalamic systems involved in feeding behaviour

Neurons of the ventromedial nucleus (VMN) are essential for energy homeostasis. We have made extensive recordings from these neurons to study how they respond to appetite related signals, including leptin, ghrelin and cholecystokinin (CCK). The neurons display a variety of intriguing activity patterns. Analysis of spike rate and interspike interval data reveals varied features amongst different subpopulations, including, of particular interest, bistable activity, where neurons switch between prolonged states of slow or fast spiking in response to external stimuli such as injection of ghrelin or CCK. Bi-stability is the kind of feature we expect to find in a decision making network. We used modelling to determine whether these features are based on intrinsic electrophysiological properties, or are a result of network interactions.
We began with minimalist assumptions about the intrinsic properties of VMN neurons inferred from electrophysiological recordings, using an integrate-and-fire based model modified to simulate activity-dependent post-spike changes in neuronal excitability. We developed a genetic algorithm based method to fit model parameters to the statistical features of spike patterning (this methodological innovation is one that has potentially wide application in neuronal modelling). The spike patterns in both recorded cells and model cells were assessed by analysis of interspike interval distributions and of the index of dispersion of firing rate over different bin-widths (again this innovative approach to validating models is potentially of wide application). Simpler patterned cells could be closely matched by single neuron models, but many others could not. We then constructed network models to explain the more complex patterns. We assumed that neurons of a given type (with heterogeneity introduced by independently random patterns of external input) were mutually interconnected at random by excitatory synaptic connections (with a variable delay and a random chance of failure). Simple network models of one or two cell types could explain the more complex patterns. We then explored the information processing features of such networks that might be relevant for decision-making. We concluded that rhythm generation (in the slow theta range) and bi-stability arise as emergent properties of networks of heterogeneous VMN neurons.

Key outcome: A network model of reward signalling systems involved in feeding behaviour

Oxytocin neurons have an important role in appetite regulation, including, for example, signalling satiety in response to the gut hormone CCK1. They respond to multiple signals, secreting oxytocin both peripherally into plasma, from axonal terminals in the posterior pituitary, and centrally, via dendritic secretion. Blood plasma oxytocin levels give the most accessible measure of oxytocin neuronal activity, but the relationship between input stimulus and plasma output response is complex and highly non-linear. In response to appetite-related signals, the neurons act as a population of independent cells collectively producing an output signal. The input signals consist of many small excitatory and inhibitory pulses from thousands of input neurons. The oxytocin neurons must act individually and collectively to process the noisy input signal into a robust output signal.
We built an integrated model coupling stimulus, spiking, secretion, and plasma. For spiking and secretion we used an integrate-and-fire based spiking model coupled to an experimentally-validated model of stimulus-secretion coupling, developed a way of modelling the CCK input, and added a model of plasma diffusion and clearance, mapping secretion rate to plasma level. The model was fitted to multiple sources of experimental data, including spiking and secretory responses to CCK, plasma diffusion and clearance following oxytocin infusion, and secretion response to spike stimulation. It can accurately simulate spiking activity in response to CCK injection, and the resulting plasma oxytocin levels, allowing the model to be used to infer neuronal activity from experimental hormone measurements. We used the integrated model to study the functional effect of particular intrinsic mechanisms such as activity-dependent after-hyperpolarisation on signal processing. For example, the model shows that this intrinsic property reduces the variability of response to a single stimulus, improving the information processing of the neurons.

Key outcome: A predictive systems model of the interactions between physiological, psychological and emotional factors related to feeding behaviour’.

Our system model is distinguished by its focus on testing what has become one of the main hypotheses of Nudge IT, that the control of food choice and appetite is centred on the reward system. It maintains the principle of only adding complexity as required, so that every element of the model is functionally justified, taking the initial approach of an engineering problem, rather than attempting to immediately replicate all the complexity of the biological mechanisms.
The model is based on the Artificial Intelligence techniques of Behaviour Based Robotics, where simple modular behaviours are coupled or layered together to produce a system that generate complex behaviour. The basic version implements a framework proposed by Nudge-it members Rogers PJ, Brunstrom JM (2016) Appetite and Energy Balancing Physiol Behav.164(Pt B):465-471. The main idea is that ‘hunger’ is driven primarily by the gut rather than energy stores, and that appetite is driven by the competing positive ‘reward’ signals for eating and negative gut signals for ‘fullness’. In the simplest version of the model, a single food type is made available at regular or random intervals. Its appearance triggers feeding which continues until the fullness signal exceeds the reward signal. The fullness signal is generated by the gut, proportional to how much food it contains, with a small contribution from energy stores. During feeding, food units are transferred to the gut, where it is ‘digested’ at a fixed rate, transferring gut content into energy in the energy store. The energy store is depleted at a fixed rate, calibrated to match typical resting energy use in a human male, 1770 kcal per day. The reward signal when eating has a fixed value. When the fullness value exceeds this and eating stops, the remaining available food is removed. The gut then continues to digest food until empty. As food is digested the fullness signal falls. The model is tested by running for 100 days and observing the evolution of energy stores. In this model, a smaller reward value results in less food consumed in each meal, leading to negative energy balance and weight loss. A high reward value results in more food consumed in each meal, leading to positive energy balance and weight gain. This depends however on sufficient food being available. If the food presentations are limited to 500 kcal then the higher reward value has much less effect.
This basic model has no adaptive ability: if the balanced model is changed by for example increasing meal frequency then it will go into positive energy balance, and reduced meal frequency will generate negative energy balance. It thus requires the ability to adjust the reward value of food based on energy balance. From physiological and imaging studies in Nudge-it we now know that peripheral hormones that report on energy stores (leptin and ghrelin) do indeed alter the reward value of food. We designed a model mechanism that nudges the target reward value up or down in proportion to current energy balance but only shifts the actual reward value towards the target value using a slow time constant. Overall, in this model, the reward value is able to adjust well to maintain stable long term energy stores.
We thus have a basic model of energy intake and appetite regulation that can maintain energy homeostasis, adapting reward values and resulting meal sizes in response to a proposed mechanism that measures short term energy balance. We then added to the model the ability to deal with multiple food types. In the software, each food type is represented by its own set of values and parameters, including the amount available to eat, the amount in the gut, taste value, reward value, and energy density, presentation frequency, and presentation amount. A new ‘food choice’ module was added to the model so that, at each time step after meal initiation, it chooses what to eat from the range of available foods. The choice is random based on probabilities generated from the reward values of each available food type. For digestion, the fixed absorption rate is maintained but it is divided proportionally between the different food types. Energy intake also becomes dependent on the food types’ energy density as well as absorption rate.
The experimental target for the model is data from WP2, measuring changes in calorie intake and body mass in rats in response to varied presentations of bland and palatable food. The experiments showed that rats adapt their food intake to maintain overall calorie intake when the palatable food ‘SCM’ is available. Stable energy intake is maintained by reducing bland food intake. The rats readily eat SCM because it has a higher reward value. Rats avoid overconsumption, but only up to a certain amount of SCM consumption. This suggests that only part of the reward value is subject to modulation by homeostatic control. We added this to the model by making ‘reward factor’ the sum of a basal component and a modifiable component.
An important aspect of reward signalling is the detection of nutritional value in the gut. It is necessary to add this to the model to match data on the effect of artificial sweeteners, which generate a strong taste reward without matching nutritional reward. The current version of the model includes a gut based reward signal that encodes the rate of energy intake: This gives a feedback measure of a food’s energy density, but is not directly attached to a particular food type. If two or more food types are consumed within a single meal then the gut reward signal gives an imperfect measure of their nutritional value. With this, the model can closely match the compensatory change in bland food intake, and maintain stable energy stores despite the appearance and consumption of high reward food. The model data shows some delay in the compensatory changes, consistent with the experimental data. The model can also match the under-compensation observed with larger consumption of palatable food:

Objective: To construct a policy toolbox.

Specific aims were:
(i) to develop and test communication tools for healthier food choices generally and with low SES consumers specifically;
(ii) to design and test examples of such behaviourally informed tools together with partners of practice (e.g. canteens, supermarkets, restaurants, schools) in empirical studies in both, the lab and the field;
(iii) to disseminate and discuss the results of these findings (and other findings from the Nudge-it consortium) with relevant academic communities outside the consortium, with public health actors, policy makers, administrators and policy “shapers”, regarding the opportunities and limits of such behaviourally informed policies in practice.

Key outcome: A systematic review on behavioural mechanisms and unhealthy dietary choices.

This “review of reviews” of behavioural tools for healthier food choices was one of the first of its kind. The main outcome was that most studies in this area provide supporting evidence that behavioural tools are able to change food choice. However, the policy insights provided by the current state of evidence is limited as these academic studies suffer from several shortcomings. Specifically, we do not yet understand the boundary conditions for behavioural tools and their effect on long-term health effects. The existing evidence does no paint a comprehensive picture that is able to inform policy makers where to implement which specific tools on which target population. This lack of knowledge calls for a more coordinated approach between policy-makers and academic research to translate research on behavioural insights into actionable behavioural tools that make real-life impact by fostering healthier and more sustainable diets. Supported by the current “replication crisis” in related academic areas, we emphasize the importance of incrementalism and a theory-guided empirical search to find out what approaches are effective. Equally important is to share the results within a policy-interested research community also to better understand the limits and culture-boundedness of the tools.

Key outcome: Novel tools from a qualitative in-depth study with low SES households

We conducted a qualitative study with 18 low SES households that we visited in their homes. During these visits and by personal interviews, we co-created – using a participative design approach - an information booklet (also available as online tool) that reflected the information needs of low SES (and migrant) households. It contained simple recommendations for healthier eating and lifestyles in families, used infographics and limited text. We tested and improved this tool and made it attractive by adding quizzes and games, following the EAST (easy, attractive, social, timely) approach that many behavioural policy units use. Since the sample was small and the time span of the intervention short, results in the sense of participating families choosing healthier diets were modest. What we did learn, however, was which barriers hinder those families to try out or adapt healthier lifestyles, namely very practical issues such as scarce resources, lack of language competence, and awareness of unhealthy habits. It also became clear that participative design approaches have potential, particularly, if the participants would be willing to continue and teach other low SES (particularly migrant) households with the developed material. Co-creation helps since neither policy makers nor health administrators might be aware of the specific needs, norms, dislikes and preferences of the respective target groups. Also, there is hope to gain more “buy-in” and agency if the target group is invited to optimize the tool. As we know from earlier research, nudges work best when they are transparent and well designed. Information tools must not only adhere to the “EAST rule” but must also be easily available and only “one click away” from the point of decision making. The booklet and the digital material are still prototypes that should be further improved and adapted.

Key outcome: Empirical studies on behavioural intervention tools designed to change food choice

Based on the qualitative study and the literature review, we designed studies to test several concrete examples of behavioural insights based tools together with partners of practice. Not all of the designed studies could be run in the end (but designs exist for future research), mostly due to practice partners declining shortly before the field time would have started (lesson learned: it is difficult to motivate corporations to cooperate on more than a short-term trial, if at all). We successfully conducted eight empirical studies in cooperation with restaurants, corporate cafeterias, supermarkets, online food shops and more in real world and online settings. In brief:
• A supermarket intervention study in Denmark with the goal to increase fruit and vegetable purchase by the use of social norm messages and choice architecture, in cooperation with the Danish “INudgeYou” network and a corporate partner (a Danish supermarket chain). We collected consumer sales data from more than 200,000 transactions to evaluate our intervention. An incremental improvement of previously tested social norm messaging using various in-store signs yielded a significant increase of fruit and vegetable purchase over several months.
• An intervention in a corporate cafeteria in Germany focusing on young employees with the goal to increase their choice of healthy lunches by using choice architecture, priming, and text reminders: Based on data from more than 100,000 meals, we compared three different nudges (i.e. reminders, convenience, and salience) between different target groups. The main outcome was that we could not identify a “one size fits all” nudge. While regular employees benefited from increased convenience of healthy choices and lower educated young employers from increased salience, higher-educated interns even showed a reduction in healthy choices during some of our interventions, presumably a sign of reactance.
• A field study in a Danish Sushi restaurant tested the effectiveness of a nudge to affect the consumer willingness-to-pay a 5% premium for organic and sustainable ingredients. The study revealed a huge discrepancy between consumers’ reported willingness-to-pay vs. their actual behavior. Highlighting the value of more sustainable diets by using a priming table tent had mitigating effects on this discrepancy and boosted the share of customers willing to pay more.
• A study comparing different healthy eating nudges (i.e. priming, ordering, labelling) applied to different social strata in an U.S. online shop with the goal to find out which nudges work best for which social strata. Eliciting the effects of different food nudges compared to a control in a hypothetical food choice task mimicking an online supermarket, we obtained population parameters from a representative sample of 1,200 individuals. Initial results show that ordering food offers by health yields the largest effect compared to the control. Subgroup analyses showed that positive effects of labels are observed for women alone and priming only works when we account for the fact that people misidentify healthy foods (i.e. limited knowledge about what is actually healthy impairs the effectiveness of priming). Concerning the social strata, we find that effects are largely driven by responses of the lowest income quantile.
• An online experiment focused on the effects of competitive priming in the Netherlands where we tested how previously effective health primes work in a representative sample. A task and prime that were already pretested earlier was supplemented and contrasted with supplemental neutral and hedonic priming banners. This experiment provided insights in the interplay of different simultaneous goal activations. The results have also some practical relevance to assess the possibility of “counter-nudging” public priming interventions using visual clutter or hedonic primes. Initial results yield overall small effects of health priming on a population level and show that only hedonic primes lead to significant changes regarding food choice (in the direction of less healthier options).
• A framing study on calorie information updating and food choice in an online setting in the U.K. with the goal to investigate whether people “update asymmetrically” when given new information about calorie content. Being interested in potential side-effects of behavioural interventions, we developed a protocol to test whether the effect of calorie labelling depends on the prior beliefs about the caloric content of a portion size. Following the theory of asymmetric response to positive and negative information, our initial analysis suggests that despite an overall underestimation of calories, people who actually overestimate them are more like to change their portion size and select larger portions in response to calorie labels.
• An experience sampling study on the barriers and drivers for healthy food choices with the goal to gain insights into internal and external barriers that hinder people in making healthier food choices. We employed a sample of 400 adult individuals who had the expressed goal to eat healthier. Several times per day, using a smart phone app, participants answered questions about the physical and social surroundings, their visceral and emotional state as well as their food choices. Based on more than 6,000 individual choice assessments, we can deconstruct barriers and conditions for unhealthy food choices. This data also contains rich information about attitudes towards classical consumer food policy and the use of behavioral methods.
• An online survey study in Germany, the Netherlands and Spain inquiring into the number of daily food choices. Inspired by assumptions of the scale and scope of daily food choices (and the hypothesis that the sheer number prohibits active healthy choices), we aimed at getting a better idea on how many food choices people make on a regular week day. The data is yet to be fully explored.
In these studies, we pilot-tested behavioural policies for healthier food choices, including choice architecture, priming, framing, reminders, social norms, and defaults. Overall, while some of these behavioural instruments showed effectiveness, it depends on the context, the target group, and the social and cultural setting whether the benefits outweigh the economic and social costs. Each intervention should therefore not only be carefully designed and but also pilot tested in a real-world setting with all key actors involved, before a wider rollout of the policy takes place. A cost-benefit analysis or similar ex ante procedure is highly recommended to estimate the cost-effectiveness of the policy. In addition, unintended side effects (i.e. rebound, moral licencing, boomerang, reactance, wear-out, and distributional effects) should be carefully monitored, evaluated, and possibly countered. A stepwise feedback and feedforward procedural approach – as suggested by behavioural public health policy in general – that emphasizes empirical testing of health policies and stepwise improvements (”test, learn, adapt, share the results, improve the tool”) is key.

Key outcome: Compilation of key policy-relevant results from across the consortium

In the two final years of the project, the consortium distilled, collected and compiled policy-relevant results from all research partners. We compiled this in an accessible report that was discussed within the consortium and with policy makers and “policy shapers” (i.e. consultants and NGOS who are potentially important influencers and gatekeepers). We also sought dialogue with practitioners such as representatives from the educational sector, from the Danish Regions (responsible for the Public Health System) and Ministries of Consumer Affairs in Denmark and Germany to test the ideas created. Besides implications for policy and practice, we also discussed implications for future research and worked on a research agenda.
The internal debate took place at several meetings with the whole consortium. The debate with external stakeholders and actors from public health policy took place at several Symposia organized by the consortium as well as at an international “Research meets policy workshop” in Copenhagen in May 2018. The purpose was to discuss evidence-based insights from the Nudge-it project in a diverse forum. We gathered a diverse mix of stakeholders involved in policy-making and food-decision research: 29 participants from different countries, from neuroscientists to public administrators. The workshop sparked discussions and reflections about linking scientific evidence about the determinants of food choice with the practical application of real-life food policy. The main focus were five potential anchors for policy (calories vs nutrients; liquid vs. solid food; formation of tastes and habits; reduce stress; cognitive mechanisms, attention and salience) extracted from the evidence from data collected in the Nudge-it project; these were the base for six suggested policies:
• Encourage the consumption of solid foods over liquids (other than water or breast milk) –within the same food group (i.e. eating an apple is preferable to drinking its juice).
• Avoid drinks with added sugar. Vegetable- and fruit-based juices or smoothies with no added sugar could be a good way of consuming necessary nutrients, perhaps particularly for children.
• Decrease exposure to non-obesogenic foods early on in life, to foster the development of non-obesogenic tastes, which may play a key role in the establishment of unhealthy dietary habits.
• Reduce stress or, perhaps more feasibly, increase the ability to cope with stress with stress management methods.
• Facilitate healthy decision-making overall, and in stressful situations in particular. For example, it reiterates the importance of establishing good habits early on life to which one can fall back on during stressful situations, as well as encouraging pre-planning of future consumption while in low-stress situations.
• Facilitate access to information on individual needs and foods. Information should be both highly salient and easily processed. The goal should be to reduce the effort required to incorporate health concerns into the choice process.


Potential Impact:
Strategy for impact

The broad aim of this project was to develop tools and knowledge than can be used to fill gaps in our understanding of food choice. As these gaps are filled, policies can be developed that work with, not against, physiological, psychological, behavioural and environmental drives.
The project’s impact can be thought of in several dimensions: as conceptual impact, capacity-building, network effects and societal improvement. It is important to note that the development of policies was not an aim of this project so we do not expect this project to lead directly to societal improvement. Instead we aimed primarily for capacity-building, predominantly by developing new tools in fields from in vivo translational studies to behavioural economics.
We took a multidisciplinary and collaborative approach in our efforts to build bridges between disciplines. This is evidenced by, for example, our parallel studies on snacking in humans and animal models, and computational models of behaviour developed with insights from Partners expert in the psychology and physiology of food choice.
At each consortium meeting, time was set aside for plenary discussion. Topics ranged from defining “hunger” to conceptualising processes needed to provide evidence for policies that will work in the real world. This collaborative approach was important as it let us combine numerous points of view into key outcomes for dissemination.
The consortium sought to maximise the impact of its activities through a range of dissemination and exploitation activities targeted at a range of end-users and stakeholders. Nudge-it has provided added value by building up the necessary critical mass across diverse fields of the neurobiology of motivational behaviour, the neuroscience of reward, the neuroendocrinology of homeostatic regulation of appetite, experimental psychology, functional brain and gut imaging, behavioural economics and computational neurobiology. Collectively we have strengthened European research capacity, evidenced by the large number of publications and dissemination activities. The co-operative aspect of the consortium’s work has been a key aspect of the success of this project. This co-operation is evidenced by joint publications (particularly a review article published in Proceedings of the Nutrition Society outlining the consortium’s consensus on key conceptual issues; Leng et al. The determinants of food choice. Proc Nutr Soc. 2017 76:316), the shared development and deployment of new technologies and tools, and the provision of a large number of early-career scientists with training that allows them to contribute to the scientific community and wider society.
All partners in the Nudge-it consortium took part in activities that had substantial impact in the scientific community and/or in wider society. Some highlights are given below.
Targeting the scientific community
Much of the consortium’s work has been carried out by ECRs, and career development for this group was a specific focus of the project. This has been successful in the majority of cases. To give three examples, Dr Jan Bauer was recruited as a postdoc to the project, and after just four years was appointed as an Associate Professor at Copenhagen Business School. Dr John Menzies also worked as a postdoc on the project and is now a research-active lecturer at the University of Edinburgh and Undergraduate Programme Director at the University of Edinburgh-Zhejiang University Joint Institute in Haining, China. Dr Amy Warnock was recruited as a PhD student. She was a driving force in the project’s public engagement activities and now, after graduating with her PhD, is the Communications Officer for The Physiological Society.
The consortium has published almost 100 manuscripts in peer-reviewed journals, with more in press and under review. In a number of cases, these publications have been picked up by national or international media. These data have also been disseminated via numerous posters and invited talks at many international meetings.
Some examples of high-impact engagement with the scientific community are given below.
1. Partners from the University of Gothenburg presented their work at a plenary talk at REGPEP2018, a major international congress with a focus on the metabolic/endocrine regulators of energy balance. The new knowledge disseminated were the discoveries that ghrelin alters food choice (in ways that promote the intake of healthier foods) and is aversive (that speaks towards a role in hunger). This information is important because ghrelin has been gaining status as a hormone that targets to reward system; this new knowledge suggests that, while the reward system is a target for ghrelin, its effects to increase food motivation more likely reflect an effect to avoid hunger.
2. Partners from the University of Gothenburg contributed to an article (Jansson J-O, et al. 2018. Body weight homeostat that regulates fat mass independently of leptin in rats and mice. Proc Natl Acad Sci USA 115:427) describing an entirely new body weight homeostatic mechanism, including a new regulatory system for the control of food intake. The first regulatory system controlling fat mass was discovered in 1994 by Friedman and colleagues, who discovered a new hormone, leptin, as a major regulator of fat mass. This “gravitostat” mechanism is leptin independent. This work was published in articles all over the world including The New York Times. It is important because obese animals can lose weight by having a weight implanted. The newly-discovered mechanism could not only provide a new non-invasive treatment for obesity (wearing a weighted vest, for example) but may also lead to the discovery of new mechanisms for body weight control.
3. The project coordinator gave a keynote plenary lecture on the neurobiology of food choice to an audience of ~5000 psychiatrists in Barcelona at the Congress of the European College of Neuropsychiatry. This was followed by a book-signing event to promote his recent book “The Heart of the Brain,”
4. Partners from UMCU implemented a neuroimaging data-sharing and analysis platform tailored to nutritional neuroscience by significantly enhancing the functionality of NeuroVault in collaboration with the Poldrack lab at Stanford University (EAG member). The new functionality for performing pooled and group image-based analysis and meta-analysis with the option to include relevant meta-data on state and trait factors benefits not only the nutritional neuroscience community but also the neuroscience community at large.
5. Consortium partners have written an important good practice paper for the field of nutritional neuroimaging. This is currently in press at The American Journal of Clinical Nutrition, which is among the top 10% journals in Nutrition (Smeets PA, Dagher A, Hare TA, Kullmann S, van der Laan LN, Poldrack RA, Preissl H, Small D, Stice E, Veldhuizen MG. Good practice in food-related neuroimaging. American Journal of Clinical Nutrition, in press).
6. Partners from UZH produced new tools for studying food choices. Specifically, they developed a computational modeling approach that can be applied to extract deeper mechanistic insights from data on choice outcomes and response times. This tool is available at https://github.com/galombardi/method_HtSSM_aDDM
7. Partners in Bristol and UNIBRIS developed tools that enable researchers to measure people’s ideal portion size. The custom-software is now being used by several research groups (based in the USA, Spain, Singapore, UK, and Germany). At Bristol we have used this approach to study child-parent dyads, to determine whether children’s BMI is related to their own preferred portion sizes or those of their parents. Partners in UNIBRIS are currently collaborating with clinicians in Bristol to explore ways to apply these tools in an intervention aimed at reducing the portion sizes of children with obesity. The same tools have been adapted by collaborators at Columbia University (USA), who are working on an NIH-funded project exploring pre- and post-operative changes in preferred portion sizes, and whether changes in portion-selection (using our tools) can predict weight loss outcomes following bariatric surgery.

Research tools

The Nudge-it project developed a number of specific tools which have been made openly available for use by the community of food researchers. These resources include:
1. A collection of nutritional neuroscience resources such as software, standardized images and fMRI tasks that are freely shared for research purposes. http://nutritionalneuroscience.eu/index.php/resources/nutneurotoolbox.

2. An open measurement toolbox for standardized collection and processing of questionnaire data. It is freely available at http://sourceforge.net/projects/forcplatform/ and has been downloaded more than 250 times.

3. A neuroimaging data-sharing and analysis platform tailored to nutritional neuroscience was implemented building on the successful NeuroVault platform (D3.5 MS3.3 https://neurovault.org/communities/nutritional

4. All source code for the computational work has also been made openly available (Project deliverable 8.3)

Targeting key stakeholders

Based upon the expertise gathered in Nudge-it and other projects, a partner from UMCU was recruited as an expert for the European Food Safety Authority’s working group “Added Sugars”. Knowledge gained in these projects will likely impact on society as it will influence policy makers as they prepare advice on the use of added sugars.
Partners from CBS produced a “review of reviews”. This was developed into two distinct publications with different target audiences: first, a book chapter in an edited book, focussing on research gaps and developing a research agenda (publication in Spring 2019); and second, a scientific journal article in the Journal of Consumer Policy, focussing on the existing evidence of efficacy of the tools in focus and the implications of these findings for consumer and health policy. The Journal has a broad readership including policy makers (and their academic staff). The paper was downloaded over 200 times before being assigned publication.
The UK study on asymmetric response to calorie information will be co-authored by partners from UNIBRIS and CBS, and our external partner, Prof Cass Sunstein (Harvard University; one of the “fathers” of nudging) who has written extensively on asymmetric updating.
A key dissemination event with research, policy, public health administration, consultants, NGOs and others was the Nudge-it Copenhagen Workshop in May 2018. The prep document on the five policy anchors and the seven suggested policies has been discussed and used widely since then. The feedback we received from the workshop was very positive and reflected the need of such “bridge events”. For instance, the delegation from Danish Veterinary and Food Administration expressed: “We expected hear presentations concerning effect-full tools for us as policy makers and presentations concerning nutrition and nudging. The five anchor points were well presented at the workshop and the knowledge about nudging - what works and what does not work - was interesting”. One of the studies presented at the workshop contained additional support for the use of the labels, such as the Keyhole label, which is used in the Nordic countries to make healthier choices easier for consumers.
Partners from CBS have been invited to give talks and have meetings with public health policy makers in the UK (Cambridge), Denmark, Sweden, Germany, and France to present the consortium’s results. We have also been contacted by researchers in Arab countries (Oman, Qatar), Argentina and Mexico with potential follow-up projects and consultancies resulting from these contacts.

Targeting the public and wider stakeholders

The “flagship” engagement element in Nudge-it was a Massive Open Online Course (MOOC) called Understanding Obesity (www.coursera.org/learn/understanding-obesity). The MOOC’s key objective was the wide and effective dissemination of ideas to people and groups usually considered to be outside the normal sphere of influence of academic researchers. This target audience included the general public and health-related professional groups that have direct contact with the public – groups that have an ability to inform and alter behaviour. The course comprised films and podcasts featuring consortium members discussing scientific evidence behind the major themes of the Nudge-it project. This was supported by a discussion forum and a number of “citizen science” projects, where course participants undertook scientific work in collaboration with professional scientists in the consortium. In these projects, the course instructors formulated questions, developed hypotheses and designed studies to test these hypotheses. Course participants were asked to predict the outcome of the study then provide data to test the hypotheses. The MOOC team analysed these data and shared and discussed it with participants. In this way we encouraged participants to step beyond received wisdom about food choice and obesity, and begin to ask and answer their own questions.
The MOOC first ran in 2015. In this run, almost 19,000 participants from 175 countries signed up the course. Over 7,000 participants visited the course pages and over 2000 participants visited the discussion forums. 2,292 posts were made on the discussion forum, ~300 by course instructors. 74% of participants rated the course as very good or excellent and 82% said it met or exceeded their expectations. Some participant feedback is given below.

“I was impressed by everything: all the assessments ingenious as learning exercises in themselves”
“Not only did you participate more actively in the forums than any other instructor I've encountered. You also gave us the chance not only to learn, but to participate in the scientific process”
“I'm really much impressed with your active participation in the forums. Honestly I do not remember [an]other Professor so active in the forums”

The host platform (Coursera) shifted to an “on-demand” model in 2017 where participants take the course at their own pace. Since 2017, a further 17,391 participants have enrolled on the course. The course will remain open for the foreseeable future.
In addition to the MOOC, consortium members took part in a diverse range of public-facing activities. These included:
• The publication of a popular science book (The Heart of the Brain) by the Project Co-ordinator.
• Partners from WUR engaged in two public science festivals (Oct 5-6, Betweter Science Festival Utrecht, and Science Weekend) where data on estimated intake of sweet and savoury solid and liquid foods were collected.
• UMCU organized a very successful public symposium entitled “Healthy food choice - State of the Art, Challenges & Solutions”. The audience was a mix of civil society, academics, industry, and policy makers. A report was published on the Nudge-it website http://www.nudge-it.eu/news/report-of-the-nudge-it-symposium-healthy-food-choice-state-of-the-art-challenges-solutions.html.
• The UZH paper “Acute Stress Impairs Self-Control in Goal-Directed Choice by Altering Multiple Functional Connections within the Brain’s Decision Circuits in Neuron gained widespread attention in newspapers, news websites, and magazines all over Europe (Switzerland, Austria, Germany, France, UK, Italy, Spain) as well as in the USA, Australia and India. It was featured on CBS news and ABC news in the USA as well on radio and TV in Europe. In total 83 features appeared in various forms of news reporting. ( Please see uploaded document)

• A Nature Communications article (Verharen et al Nat Commun. 2018 Feb 21; 9(1):731) received attention in national newspapers as it explained what underlies the tendency to display addictive behaviours towards food, gambling and drugs of abuse.
• Partners from WUR published a paper on measuring gastric emptying with MRI, and the occurrence of ‘phantom fullness’ (Camps et al. 2016 AJCN doi: 10.3945/ajcn.115.129064). It was highlighted on the journal website and subsequently attracted wide national (>20 news websites) and international media attention, including from Time Magazine. (Please see uploaded document)

• Partners from Bristol featured in several large-scale public engagement activities, including two British Science Festival presentations and a month-long science experiment, hosted at the London Science Museum. (Please see uploaded document)

• In collaboration with the BBC, they produced a public interactive-science web page that focuses on Nudge-it related activities and findings Please see uploaded document

The Nudge-it website was a critical interface between the consortium and our stakeholders. It featured sections oriented to the scientific community, to stakeholders and to the general public. Over the lifetime of the project, the website had over 100,000 page views and has been visited by over 30,000 unique users. There was a notable spike in activity (up to 700 users per day) that coincided with the first run of the MOOC, indicating that this is a powerful way to drive users to learn more about a parent project. (Please see uploaded document)

Dissemination activities will not end with the end of the project. Partners will continue to build on and share work done in the project with our target audiences. For example, the MOOC and BBC-hosted material will remain online for the foreseeable future, and Partners in behavioural economics will use the anchors developed in the project to engage with policy makers and “shapers”.
Exploitation
It is still too early to fully realise the extent to which the results from our project will be exploited. The main outcomes is the generation of new knowledge and new tools, and exploitation of this will likely inform scientific research, public understanding of science, and, in the longer term, the development of effective policy. Examples of some of these routes to exploitation are detailed below.

Drug targets:

1. While discovery of new drug targets was not a main objective of the Nudge-it project, the research undertaken nevertheless led to some findings of potential translational importance in this respect. Amongst these, work on oxytocin neurons in the consortium has helped raise the status of this system to that of a major potential target for development of new therapies (see the recent review by Ding et al. Oxytocin in metabolic homeostasis: implications for obesity and diabetes management. Obes Rev. 2019 20:22-40.). Several clinical trials are now underway to assess the potential of intranasal administration of oxytocin to support weight management in humans.

2. The identification of the supramammilary nucleus as a major node in food choice decision making and particularly the role there of ghrelin highlights this site as one that is engageable by systemic delivery of gut hormone analogues. In these studies, we used a conjugate molecule that combines the gut/brain peptide, GLP-1, with oestrogen, this molecule has been suggested to represent a novel strategy to treat diabetes and obesity.

Behavioural interventions:

1. Work in Nudge-it contributed to the discovery of a novel homeostatic mechanism (the gravitostat) by which bone loading influences energy balance. This opens up a wholly novel range of possibilities for behavioural interventions to support the maintenance of a healthy body weight (see http://www.bariatricnews.net/?q=tags/gravitostat; Please see uploaded document

2. The psychological studies in Nudge-it deconstructed food choice behaviour in ways that can be used to inform future behavioural strategies. Specifically, it was clearly established that meal planning in humans involves accounting for reward value of the food, but also importantly involves accounting to avoid hunger between meals. Thus foods that have been learnt from experience to lead to prolonged sensations of fullness are prioritised when there is expected to be a long gap until food is next available. This highlights the importance of “food learning” and the importance of clearly identifying healthy foods with an extended satiety profile.

3. Work identifying heart rate variability as a valid objective measure correlating with self-control in food choice indicates the possibility that this relatively simple and non-invasive measure could be used to identify sub-populations of individuals most amenable to certain types of behavioural interventions. Targeting sub-populations of individuals may be more effective in driving dietary change than population-wide interventions.

Policy:

In the two final years of the project, the consortium distilled, collected and compiled policy-relevant results from all research partners. We compiled this in an accessible report that was discussed within the consortium and with policy makers and “policy shapers” (i.e. consultants and NGOS who are potentially important influencers and gatekeepers). We also sought dialogue with practitioners such as representatives from the educational sector, from the Danish Regions (responsible for the Public Health System) and Ministries of Consumer Affairs in Denmark and Germany to test the ideas created. Besides implications for policy and practice, we also discussed implications for future research and worked on a research agenda.
There is an extended account of this above under the outcome Compilation of key policy-relevant results from across the consortium.

More information on the project can be found at www.nudge-it.eu.

List of Websites:
www.nudge-it.eu
email Nudge-it@ed.ac.uk
Contact Carol Wollaston, Centre for Discovery brain Sciences, Edinburgh Medical School:Biomedical Sciences, Hugh Robson Building, George Square, Edinburgh EH8 9XD
Tel +44 131 650 3551 Moble +44 7917 688050