Food for thought

According to a recent article in The Atlantic Monthly “Tasting a Flavor That Doesn't Exist”, food companies are now implementing the findings of recent research on “phantom aromas” that somehow trick the brain into “manufacturing a taste”. While we can’t really expect non-technical magazines to necessarily convey the complexity of most research, this type of article nevertheless annoyingly perpetuates the idea that somehow the brain is an organ that “we” can easily confuse by feeding into it contradictory or incomplete information. This idea has a long history in the realm of visual illusions. Lines can seem longer than they are, rooms smaller, and the moon larger if we provide the brain with visual information in the right way or under certain constraints. What a silly old brain! 

But of course, our brains (that is, us) do not process sensory information out of context. Since the context may tell us something quite important about what we are sensing, why would they? The context may be environmental (Why is the moon so big? Well, things near the horizon tend to be further away) or it may come from the different sources of sensory information that arrive at the same time. This latter type of context provides the basis for the recent fascination with the multisensory nature of food, especially the findings that one sense can influence another, e.g. [1]. 

But this shouldn’t be surprising – the brain takes in and integrates whatever sources of information are needed to help us survive. In the food realm, this means identifying those things that are edible. Animal studies have even identified single nerve cells in the brain that receive information from touch, hearing and vision [2], because together such information might reliably tell us where something edible might be found.

But why do brains sometimes get it wrong? The answer is straightforward: if the information is incomplete or perhaps different senses contradict one another, the brain takes its best guess. A lot of the time, this guess relies on searching for, and identification of, patterns that are statistically more commonly experienced. Sometimes the best guess allows one sense to dominate others - visual dominance responsible for the illusion of the speaking ventriloquist’s dummy – because it is usually a more reliable source of information. At other times, such as when it’s easier to understand what someone is saying in a noisy environment if you can see their lips move, information from one sense enhances that from another. 

This is exactly the same phenomenon that we see in foods when we find that an aroma can enhance the intensity of a taste. Beginning in the late 1980s, Frank and colleagues [3] explored the phenomenon of sweet-smelling odours. From these studies, two things stood out. The first of these was that such odours seemed to be mostly those that were repeatedly experience together with sweet tastes in foods. Secondly, a sweet smell would only enhance a sweet taste and not, for example, a salty taste. They noted that the odour and taste needed to share a common property – the sweetness – whether tasted or smelled.

A smell with a taste? A phantom aroma indeed! Except …. we now need to consider the one other source of information that the brain uses. This comes from …. wait for it .... Inside our heads! The fact that sweet aromas are those experienced previously with sweet tastes (and salty aromas with salty tastes, and so on) is the key. In integrating information about food, the brain only really cares about the fact that together the tastes, aromas and tactile qualities uniquely identify something useful to us. Ah, this flavour is sweet, say you (and your brain), and hence the food is good to eat. Because we respond to the overall flavour, and not which sense provides what information, the odour is not distinguished from the sweetness of the food and this is encoded in memory – subsequently sniffing the odour can then activate this memory and … hey presto! … a phantom aroma (except what The Atlantic Monthly presumably meant was a phantom taste).*

What this all means is that flavours are assembled from information that comes from the mouth and nose, together with other information resulting from our experiences with foods and extracted, consciously or not, from memory. In other words, flavours are at least partly cognitive. Some knowledge of this cognitive landscape, traditionally largely ignored by food scientists and flavour chemists, is crucial to understand perception, whether of foods or of anything else. Cognitive psychology – or as it is known now by those who want to ask the same questions using large, expensive machines, cognitive neuroscience – has provided models and techniques for studying memory, attention, decision making and, increasingly, emotions.

The crucial role that cognitive processes play helping us understand food perceptions and preferences extends beyond explanations of flavour, and odour/taste interactions. Consider the activity of wine tasting. In describing the wine, the taster selectively attends to some qualities (berry-like, tobacco notes) by somehow mentally extracting them from the complex mixture that is wine flavour. These notes can be weak or intense, but the only way of knowing this is by comparing in memory what is perceived to what is typical. Then there is the decision that might have to be made: is this a good wine or a wine typical of the region.

In fact, we don’t have to rely on expert wine tasters to illustrate our reliance on cognitive processes in tasting. Everyday evaluations of food carried out within the food industry or in food research environments illustrate the point equally well. Sensory scientists routinely administer tests to trained panelists or consumers to evaluate foods perceptions or preferences – what sensory qualities does a food possess, how strong are these qualities, are two products different, or perhaps which product is liked more. Alternatively, we can see the administration of sensory tests to panelists or preference tests to consumers as requests to pay attention to product attributes, make decisions about differences or preferences, to analyse complex flavours into their elements, remember previous product experiences for comparison, or to describe emotions arising from product experiences. 

Even the way in which sensory tests are carried out is a function of what we know about cognitive biases. If you want people to attend to the overall flavour, don't ask them about different aspects of the flavour. In a group of products varying in intensity on an attribute, the first sample will be held in memory to act as the point of comparison for the other samples. Asking about the intensity of a single product will activate long-term memory for the intensity of what they usually consume. The answer to whether two products are the same or different will depend on the cognitive criterion for difference that the person adopts. And so on.

Cogito ergo … er, gustavero? **

___________________________________________________________________

*  Those wishing a more detailed account of this process should see one of the following reviews: [4-6]

** Even if you don’t know Latin, you should know Descartes – look it up!

___________________________________________________________

1. Piqueras-Fiszman, B. and C. Spence, Sensory expectations based on product-extrinsic food cues: An interdisciplinary review of the empirical evidence and theoretical accounts. Food Qual Prefer, 2015. 40: 165-179.

2. Meredith, A. and B.E. Stein, Interactions among converging sensory inputs in the superior colliculus. Science, 1983. 221: 389-391.

3. Frank, R.A. and J. Byram, Taste-smell interactions are tastant and odorant dependent. Chemical Senses, 1988. 13(3): 445-455.

4. Prescott, J., Chemosensory learning and flavour: Perception, preference and intake. Physiol Behav, 2012. 107: 553-559.

5. Prescott, J. and R.J. Stevenson, Chemosensory Integration and the Perception of Flavor, in Handbook of Olfaction & Gustation: Modern Perspectives, R.L. Doty, Editor. 2015, John Wiley & Sons. 1008-1028.

6.  Stevenson, R.J. and R.A. Boakes, Sweet and sour smells: the acquisition of taste-like qualities by odors, in Handbook of Multisensory Processes, G. Calvert, C.B. Spence, and B. Stein, Editors. 2004, MIT Press: Cambridge. 69-83.

(De)constructing flavours

… and in breaking news, chocolate flavour and vanilla flavour are different from one another. But what actually makes flavours different? Clearly, both of these flavours are sweet, so we recognize that the non-sweet bit – the inherent chocolaty-ness or vanilla-ness – is the essential difference. And these qualities are, of course, due to volatiles – in other words, they are smells, perceived via the retronasal (back of the mouth) route to our smell receptors in the nose. So, clearly flavours are combinations of smells and tastes.

But what about other sensory information? If we wanted to accurately describe the flavour of a curry we’d also no doubt also say the flavour is hot or spicy; similarly, the flavour of Coca-Cola is fizzy and that of vodka has ‘bite’. The inclusion of this type of sensory information in our definition of flavours is not especially controversial – after all, a curry without the spiciness seems not a curry at all. More to the point, our mouths contain endings of the trigeminal nerve that are responsible for picking up those chemical signals that lead to perceptions of chemical heat, temperature, and sensations of dryness, prickliness and bite. These same nerve endings may also contribute to the experience of taste.

Stepping outside the mouth though, we are confronted with vast quantities of sensory information that contributes to our experience of a food’s sensory properties. Several overviews in recent years have discussed the importance of vision, hearing, and touch (outside the mouth) in influencing our perceptions of what goes on in the mouth during eating [1, 2](see also previous posting). In a new book by Spence and Piqueras-Fiszman [3], the recent research on the impact of properties like colour, shape and weight in the cutlery, crockery and physical environment of food perceptions and preferences are detailed. Does that mean each of these influences must be included within our definition of flavour?

Some researchers do, in fact, suggest this. They propose, for example, that food features such as crunchiness, the result of both sounds and feedback from pressure sensors in the teeth and gums, are intrinsic to the flavour of some foods. Previously, I discussed the role of expectations – essentially the product of our memory for what belongs with what – as another important influence on flavour experiences. So, potentially, if we wished to be broad enough, flavours could be perceptions involving multiple near and far senses, as well as input from memories. Suddenly, the mouth and what goes on inside it becomes only part of our concept of flavours.

Does it really matter how broadly we define flavours? We can distinguish between the sound of an orchestra and the impact that the acoustics and seat comfort of the hall in which it is heard have on our experience of the music. Similarly, there are good reasons to divide our food experiences into mouth-based sensations, on the one hand, and other sensory information that impinge on our perception or enjoyment of these sensations, on the other. Thus, we are able to experience flavours without any input from any senses except smell and taste. To be chocolate flavour, a little square of fat, sugar and miscellaneous odour compounds does not require the colour brown, or even the sensation of melting, but it does require chocolate aroma and sweet taste in all cases.

Recent studies of the brain’s processing of smell and taste have identified a network of neural structures that appears to encode for flavour, as distinct from odours and tastes separately [4]. Indeed, the brain appears to process the same odour differently, depending on whether it is experienced in the mouth as part of a flavour or external to the mouth, when sniffed. The brain is fundamentally a processor of multisensory signals, largely because integration of different sources of sensory information is biologically useful. An important question, though, relates to the adaptive significance of the ‘construction’ of flavours – why do discrete neural circuits, for example, represent flavours rather than simply odours and tastes separately? It is generally proposed that, in the case of foods, it is the combination of tastes and odours together that reliably tell us whether an object is a food that is fit to eat. However, it is clearly not only about identifying foods. While it can be argued that it is taste and odour together that allow us recognize pear as a pear, in practice, once it is familiar the pear odour is sufficient. In a world without taste, trial and error would allow one to distinguish pears from apples and could even tell you whether or not pears were safe to eat.

The most important consequence of integrating odours and tastes may be primarily about pleasure. From the perspective of food preferences, flavours seem to be fundamental units. This is because, at birth (or in the case of salt, shortly thereafter), we are hedonically inflexible when it comes to basic tastes – sweet, sour, salty, bitter, umami. Our likes and dislikes appear to be pre-set as an adaptive mechanism to ensure intake of nutrients (sweetness, saltiness, umami) and avoid toxins or otherwise harmful substances (bitterness, sourness). On the other hand, there is little evidence that odour preferences are other than the result of experience, a process that may begin in the womb.

Repeatedly experiencing odours with tastes attaches additional meaning to the odour that is primarily hedonic, that is, pleasure-related. The pear flavour that is not bitter, not too sour, and quite sweet provides pleasure in the eating. In other words, we are motivated to consume it because of its sweet taste and the prior associations with the calories that the sweetness, and subsequently (through repeated experience), that the pear flavour itself signals. And, of course, this occurs even prior to eating: pear odour repeatedly paired with a sweet taste itself becomes pleasant.

The perceptual consequences of integration of odours and tastes can be interpreted in the same way. The well-known phenomena of food odours being described in terms of tastes – the sweet smell of vanilla or the sour smell of vinegar – also arise from the repeated pairing of the odour with a taste, sweetness or sourness, respectively. But, these perceptual qualities also have hedonic consequences: sweet-smelling odours are pleasant and this quality may in itself motivate consumption even if we cannot identify the actual odour or its source. Conversely, something with a bitter or sour smelling odour is unlikely to be eaten, especially if we cannot recognize the odour. As such, these perceptual changes to odours may help compensate for the fact that odour identification is particularly difficult even for common foods.

Hence, the key purpose of odour/taste integration is not that it aids identification per se (although it might), but rather that it gives an hedonic value to the flavour, which crucially it the defining characteristic of the food. Thus, flavours can be most accurately seen as “objects” constructed for their hedonic qualities. Initial, “gut” responses to foods are almost always hedonic, and this naturally precedes accepting or rejecting the food. Thus, what we perceive when we sit down to dinner are, thankfully, integrated pleasure-inducing perceptions – spaghetti al pomodoro and a nice Chianti – rather than a collection of independent, hedonically-diverse tastes, odours and textures.

                                                                                                                       

1.         Delwiche, J., The impact of perceptual interactions on perceived flavor. Food Qual Pref, 2004. 15(2): p. 137-146.

2.         Prescott, J. and R.J. Stevenson, Chemosensory Integration and the Perception of Flavor., in Handbook of Olfaction & Gustation: Modern Perspectives., R.L. Doty, Editor 2014. p. In press.

3.         Spence, C. and B. Piqueras-Fiszman, The Perfect Meal. The multisensory science of food and dining.2014, Chichester: John Wiley & Sons.

4.         Small, D.M., et al., The role of the human orbitofrontal cortex in taste and flavor processing. Ann. N.Y. Acad. Sci., 2007. 1121: p. 136-151.

Feeling all emotional

Leaving aside issues of nutrition, we usually judge a food by how much we like its flavour. But what is ‘like’? Ask a social psychologist and, after prolonged qualifications, they will probably tell you that ‘like’ is an attitude or a positive disposition towards something. So, if you wish to compare two foods in terms of liking or acceptability or preference, then degrees of positive disposition are what we are after. This doesn't sound very much like pure pleasure of the type that food can produce and yet ratings of liking are what are used by food companies to discover how consumers respond to foods and even to make decisions about whether or not a particular product should be launched into the marketplace.

Ratings of liking are thus a proxy for an emotion – pleasure – and for a behavior, food choice. This would be fine if we could be sure that ‘liking’ was a good proxy for either, but there is little evidence that this is the case. Here’s the problem. There is a substantial body of research and practical experience showing that liking ratings fail to predict later food choices. Very many reasons can be found to explain this, including the fact that rated samples are not representative of meals, that liking varies as a function of many variables – time of day, mood, and so on, and that choice is influenced by other variables apart from liking.

A relatively recent approach to this problem has been to ask if we are measuring the right thing. Are positive dispositions the sort of thing that are likely to reveal the next big product? I have posted previously about comfort and wanting. We recognize that a search for comfort is a particular feeling that will direct us towards specific types of foods, at least some of the time. Wanting, too, is a strong motivator that might underlie actual choices, irrespective of degree of liking. Of course, there are a multitude of emotions that we could measure, some of them perhaps quite important to us. Clinical psychologists and psychiatrists have been using standardized measures of emotion for some time, but the emphasis is unsurprisingly on negative emotions that are unlikely to be elicited by eating in most of us. 

Compiling lists of independent emotional terms that might be more related to foods can of course be done and there are examples in the recent literature (see for example [1]). There is certainly evidence that rated emotions do discriminate between products, and perhaps in some cases better than liking ratings [2]. But the issues to be addressed are many. Thus, how do we know that a food-elicited emotion is not mediated by the pleasure given by the food (as reflected by ‘liking’) rather than being a direct result of the food experience? This would be consistent with a great deal of research showing that pleasant stimuli of all kinds produce positive moods, which in turn might be described using words such as ‘cheerful’, ‘energetic’ or ‘friendly’ – all terms that have been used in emotion questionnaires.

Which brings us to another important issue: just how many emotions are there and how do we decide which ones to measure? On the basis of detailed cross-cultural comparisons, Ekman [3] for example has argued for the universality of a small number of emotions – happiness, anger, disgust, fear, sadness and surprise – and their associated facial expressions. If this is true, then it may be that your ‘cheerful’ is the same as my ‘friendly’, and that we actually both mean ‘happy’. In other words, perhaps these terms are all highly correlated with something called ‘positivity’. Searching for such key descriptors is not unlike the search for non-redundant terms in describing the sensory properties of products. But it is a more complex task to determine if a consumer’s rating of an emotion as irrelevant to their food experience represents actual lack of feeling or a poor fit between the word and what is actually felt. And that’s even without considering such seemingly bizarre concepts as “guilty pleasure”.

An emotion, though, is not just an internal feeling that needs a label. William James raised the question over a century ago of whether the other aspects of emotions – physiological responses and facial expressions – might not actually be primary, in that they precede the subjective feeling. To paraphrase his argument, do we run from a bear because we are afraid or are we afraid because we run from the bear? If the latter (and see [4] for evidence of this), then either facial expressions or the physiological correlates of emotions might be more direct measures of responses to foods.

While measuring facial expressions could be considered a more objective means of getting directly at emotions because they can be objectively verified, it is not true that expressions are spontaneous. We all manage our expressions, masking or modulating them depending on the context (or culture) we are in – a skill that is evident even in 3 year olds. Moreover, just how finely are smiles/frowns gradated? More than a 9 point liking scale? In a recent study [5], automated readings of positive (happy), neutral and negative (angry, disgusted) emotions in response to juice samples tracked ratings on a 9-point hedonic scale quite closely. In fact, the facial readings were able to differentiate between the samples, but not to a degree better than the scale.

Measures of physiology in response to foods are at a similar state of progress. Food scientists are now in many ways tackling some of the same issues that William James did at the end of the 19^th century. Psychologists have recognized for the past century that quite different emotions share common physiological traits – e.g., both fear and anger are associated with increased heart rate. Following James, later theories of emotion emphasized the role that cognitions – interpretations – had in determining which emotional quality was associated with particular physiological changes for a given context. This was compellingly shown in Schacter and Singer’s classic 1962 experiment on the role of expectations in emotions [6]. They showed that physiological arousal produced by an injection of epinephrine (adrenalin) could be interpreted as consistent with either anger or excitement, depending entirely on the context to which the subject was subsequently exposed. While there are good data showing that foods do elicit changes in physiology [7], once again the ability of these changes to predict food choice is unknown. In any case, the context in which the food is experienced is likely to be crucial.

My feeling is that there are a few years left in old liking scale yet. If we don’t receive sensory pleasure from a food, it is more than likely doomed to be a single purchase item; but the converse is not true. Liking therefore might best be seen as a measure that is necessary, but not sufficient, to predict food choices. The search is really for complementary measures that add some predictive power. The fact that emotion research has yet to add this tells us only that some key questions are in need of answers and that it is premature to expect successful applications without more basic research.

_____________________________________________________________________

1. Cardello, A.V., et al., Measuring emotional responses to foods and food names using questionnaires. Food Qual Pref, 2012. 24: p. 243-250.

2. Ng, M., C. Chaya, and J. Hort, Beyond liking: Comparing the measurement of emotional response using EsSense Profile and consumer defined check-all-that-apply methodologies. Food Qual Pref, 2013. 28: p. 193-205.

3. Ekman, P. and W.V. Friesen, Constants across cultures in the face and emotions. J. Pers.Soc. Psychol., 1971. 17(2): p. 124-129.

4. Strack, F., L.L. Martin, and S. Stepper, Inhibiting and Facilitating Conditions of the Human Smile: A Nonobtrusive Test of the Facial Feedback Hypothesis. J Pers Soc Psychol, 1988. 54(5): p. 768-777.

5. Danner, L., et al., Make a face! Implicit and explicit measurement of facial expressions elicited by orange juices using face reading technology. Food Qual Pref, 2013 (in press).

6. Schachter, S. and J. Singer, Cognitive, Social, and Physiological Determinants of Emotional State. Psych Rev, 1962. 69: p. 379-399.

7. de Wijk, R.A., et al., Autonomic nervous system responses on and facial expressions to the sight, smell, and taste of liked and disliked foods. Food Qual Pref, 2012. 26: p. 196-203.

Learning to want

The sensory properties of foods – their tastes, odours, textures – are crucial to determining what we eat.  This is because these qualities, together and apart, evoke pleasure. So, when we talk about motivations to consume foods, it is often taken for granted that food acceptability and preferences underlie our behaviours. Of course, this is not to ignore a variety of other motivations – nutrition, convenience, and so on – but foods that are not liked are generally not eaten. And if we find a food is especially palatable, we will eat more of it.

Consider though if you were very hungry and your food choices were limited. A plate of something that we would otherwise regard as unpalatable might still be gratefully eaten if that was all that was available. Our motivation here is driven not by liking, but by wanting.

Although this distinction between liking and wanting was first seen in the drug addiction literature, it is increasingly seen as important in helping to explain motivations to consume foods. On the majority of eating occasions, we want what we like, and vice versa. One major reason for this is that foods that are high in energy, either from fats or carbohydrates, are those foods that are both highly liked and stimulate wanting.

It is possible to distinguish between liking and wanting, and in some studies this has been done by contrasting ratings of liking for a food with ratings of desire to eat it. One other way is to observe facial expressions. A study by Julie Mennella [1] in which infants were fed a novel food, green beans, demonstrated how infants’ facial expressions clearly indicated dislike for this food. Following repeated exposure to eating the beans, the infants were willing to eat increasing amounts of the beans – a clear sign of wanting. However, their facial expressions did not change with repeated exposure to the beans – unless the infant had been fed peaches after the beans, in which case the facial expressions were much more positive. Pairing the sweet peaches had conditioned a liking for the beans, resulting in a changed facial expression.

This finding, and the dissociation between indicators of liking (facial expressions) and wanting (consumption), can be understood in terms of the everyday processes of flavor-flavour learning and flavor-calorie learning. As shown in humans by Yeomans [2] ingestion of energy or other wanted nutrients, especially while hungry, conditions a liking for a food flavor. In addition, however, experiencing the conditioned food odour/flavor can also elicit increased appetite and consumption. This contrasts with the pairing of an odour/flavour with just a sweet taste (which may or may not be associated with calories), which only reliably conditions liking for that odour. While, as mentioned above, we seldom want to eat what we do not find palatable, it is highly likely that it is not this preference that pushes us to eat, but rather the engagement of wanting. In studies by Yeomans and others, odours have been conditioned through pairing with nutrients. In everyday situations, not only odours but also sights, sounds and contexts can become associated with foods.

The relevance of conditioned wanting is evident when we try to understand why we eat particular foods at particular times. Most of our eating is not done because we are severely depleted of energy or other nutrients. It is done in response to a particular amount of time having passed, or the presence of cues that remind us of food. If your stomach rumbles when you enter a kitchen where something delicious is being cooked, or when passing a bakery from which the aroma of fresh bread wafts, it is a signal that your gastric juices have been conditioned to the food odours by prior pairing of those odours with the calories that followed them.

One very plausible reason why people in affluent societies are nowadays eating so much is that our worlds are filled with a multitude of such cues to wanting that occur without our being consciously aware of them: odours, flavours, sights, sounds associated with eating.  There is a good reason, for example, why television advertising of snack foods and confectionary is effective, and this is because highly realistic cues for foods can elicit wanting. In a study of these effects, Ferriday & Brunstrom [3] showed that exposure to the sight and smell of pizza in a laboratory setting increased consumption of freely available pizza after participants had already consumed a fixed amount.

Another construct – hedonic hunger – has also recently been discussed as a major motivation for eating. Hedonic hunger is seen as a drive towards food pleasure-seeking that coexists with hunger driven by energy needs. It is a very similar idea to both food craving and conditioned wanting in that it is elicited by food sensory cues. It is, by definition, satisfied only by highly preferred foods. Of course, as noted above these are the foods that are most of the time both liked and wanted.

The idea of hedonic hunger appears to be useful in helping to explain the drive to consume highly palatable foods when we are trying to eat a ‘healthy’ diet or one that leads to weight-loss. Dietary restriction reduces both energy intake and food pleasure, and so if we are genuinely motivated by pleasure-seeking in our eating, then this helps to explain why diets so often fail.

On the face of it, this problem ought to be addressed by good tasting, low calorie foods, and of course the food industry is working hard to provide these. However, widespread use of low calorie foods may be a problem in itself. Because wanting is driven by conditioned associations between energy and flavours, we may find that we start to selectively want only high calorie versions of foods. Just such a finding was suggested recently by O’Sullivan [4] who showed that a low calorie, but familiar, version of a pasta dish became less and less liked relative to the regular version over repeated eating occasions. Whether this would have led to reduced amount consumed or desire to consume was not measured. Another inadvertent consequence of proliferation of low-calorie foods may be a reduced ability to estimate energy intake. At present, sensory properties provide important information about the calories in what we consume. So, thick, sweet, rich foods tend to be higher in calories; these same, palatable qualities uncoupled from their calorie consequences may limit our ability to implicitly monitor our energy intake. Since most of us – but especially those trying to restrict intake – rely heavily on such cues, we may be losing an important part of our ability to monitor what we eat and, for example, compensate for high energy intake at one meal a with lower intake at another.

________________________________________________________________

   1.     Forestell, C.A. and J.A. Mennella, Early Determinants of Fruit and Vegetable Acceptance. Pediatrics, 2007. 120(6): p. 1247-1254.

   2.     Yeomans, M.R., et al., Differential hedonic, sensory and behavioral changes associated with flavor-nutrient and flavor-flavor learning. Physiol. Behav., 2008. 93: p. 798-806.

   3.     Ferriday, D. and J.M. Brunstrom, How does food-cue exposure lead to larger meal sizes? Brit. J. Nutr., 2008. 100: p. 1325-1332.

   4.     O’Sullivan, H.L., et al., Effects of repeated exposure on liking for a reduced-energy-dense food. Am J Clin Nutr, 2010. 91: p. 1584-1589.