Food reward derives from nutrient content and sensory qualities

A large body of work has identified value signals in a number of brain areas and explored their role in decision-making and learning (1–3). However, what is currently not well understood is how these value signals originate and how they are related to the specific sensory experiences that generate them. Taste is famously subjective and is typically treated as a given variable beyond the scope of the experiment. However, in any decision maker, be it a human or an animal, a particular pattern of sensory inputs must directly generate specific value signals in the reward circuit of that subject. This is an objective process, and, in principle at least, we should be able to understand it. Is it possible in practice? In PNAS, Huang et al. (4) describes work that starts to answer some of these questions in the domain of food preferences.

Value is a fundamental concept in the behavioral sciences. Humans and animals can select between options with very different attributes. For example, we can easily choose between buying a new computer or going on an expensive vacation, even though these options are very different in type (an object and an activity) and will evoke very different sensory experiences. The behavioral sciences explain this ability by postulating an internal unidimensional signal, called “value,” that can be attributed to any distinguishable option and that collapses all attributes of that option into a single variable. This internal value signal can then be easily compared across all available options, so that the one associated with the largest value signal is chosen. While economics treats value as just a hypothetical variable that explains pattern of preference, neuroscience has demonstrated that value is, in fact, encoded in specific brain circuits. Single-unit recordings have shown the existence of neuronal activity that encodes the value of available options as well as the value of actions that will obtain these options (e.g., ref. 5). A different line of research has explored how value signals and their prediction can be used to guide learning (6, 7).

Despite these successes, we do not know how value signals originate, and we cannot predict the value of an option based on its attributes alone. Value has to be inferred, either as “revealed preference” in economic choice experiments or as an outcome for which an animal is willing to work and that drives learning (8). However, in the brain, the final value signal is, of course, directly computed based on the initial sensory input. If we could understand this process, we should be able to predict the value of sensory experiences based on their attributes.

There are no specific reward receptors. Instead, specific sets of sensory perceptions give rise to reward signals of a specific size. Very different combinations of sensory input across all sensory modalities can be perceived as valuable. Thus, the specific rules will likely be different for different types of sensory experiences. In PNAS, Huang et al. (4) begin in one specific domain, namely, food. Food preferences are of particular importance because of their great clinical relevance given the worldwide obesity epidemic and the numerous health challenges it creates (9). In addition, they are also of great scientific interest, because they expand the range of reward options that are used and investigated in economic experiments in primates. Most studies use fluid rewards (typically water), where the only changes in reward are the amount or probability of the fluid delivery. Notable exceptions are the work by Padoa-Schioppa and Conen (10), using fluid rewards with different tastes, and work by Platt and coworkers (11) using visual stimuli that are sexually attractive to male macaques. However, even in these studies, while the value of the sensory experiences was measured behaviorally and neuronally, how the sensory signals generate the reward was not investigated.

To better understand food preferences, Huang et al. (4) start by quantifying the relationship between food choice, macronutrient content, and energy content of a small set of foods offered to macaques. In order to allow better control of texture and for more precise control of delivery and amount of the food, the monkeys did not receive solid food but a dairy-based liquid reward (similar to a milkshake). Food contains several macronutrients, such as carbohydrates, protein, and fat. The animals in the study received a normal diet in their home cages but could choose between different nutritious fluids that contained different concentrations of two nutrients, namely, sugar and fat. These were chosen because of their relevance for human overeating and obesity, and their role in determining sensory food qualities.

The design of the choice tasks is clever and builds on years of experience in using behavioral tasks to measure the preference of animals for options with differing attributes. Basically, the monkeys were offered choices between low-fat, low-sugar (LFLS) fluids and either high-fat, low-sugar (HFLS) fluids or low-fat, high-sugar (LFHS) fluids. This allowed the authors first to demonstrate that the monkeys were aware of the differences in nutrient content and preferred more nutrient-rich fluids. By varying the quantities of the less preferred LFLS fluid offers, they asked the monkeys to trade off fluid amount against nutrient content. In this way, they could determine the amount of less preferred fluid that was equivalent in value to the preferred fluid.

To better understand food preferences, Huang et al. start by quantifying the relationship between food choice, macronutrient content, and energy content of a small set of foods offered to macaques.

Importantly, the two high-energy fluids (HFLS and LFHS) contained the same amount of energy, which allowed determination of whether the monkeys’ choices were guided by energy only (calories) or by nutrient content (sugar or fat amount). By examining the monkeys’ choices across an entire session, the authors could show that the monkeys did not try to maximize energy but rather tried to maximize the intake of their preferred nutrient. Consistently, all three tested monkeys preferred the sugar-rich over the fat-rich fluid. Thus, monkeys (like humans) have nutrient preferences that result in food intake that is different from what would be predicted by optimal foraging theory (12) or by homeostatic setpoint models (13).

How do the monkeys sense the sugar and fat content? Sugar is directly sensed by taste receptors, but the mechanism for fat sensing remains unclear (14). Existing evidence points to a somatosensory, oral texture mechanism (15). By extensively testing the mechanical attributes of the fat-containing fluids, the authors confirm that viscosity and sliding friction explain the monkeys' fat preferences.

What can be done with these results? By quantitatively relating the nutritional attributes of fluids to value, the authors can now predict and test the value of new, unknown fluids with novel combinations of nutrient concentrations. This will allow the identification of neuronal signals that represent the constituent attributes (nutrient levels) and of the resulting combined food value. The obvious next step is to record from neurons in the brain of macaques and to start unraveling the circuit that links primary gustatory and somatosensory representations with the final representation of the food value. Huang et al. developed a range of choice models that identify candidate neuronal mechanisms for food preferences. This will allow a detailed description of the mechanism by which objective attributes of objects in the world are translated into subjective hedonic value.

The PNAS article by Huang et al. (4) not only answers some important questions but also opens additional ones. One relates to the role of the background diet in this experiment. The monkeys were fed a standard diet for macaques, that fulfilled all their nutritional needs. On a given testing day, the animals had free access to their standard diet before and after the experiments and received their main liquid intake in the laboratory. Thus, the main motivation for participating in the experiments came from fulfilling their fluid requirements, not from fulfilling nutritional needs. This might change, if the monkeys were deprived either in overall energy or with respect to a particular nutrient. For example, the authors show that the monkeys prefer sugar over fat. It would therefore be interesting to deprive the monkeys, for 1 d to 2 d, of fat in their diet, without affecting their overall caloric intake. The preference experiments could then be repeated under this deprivation condition. Likely, the relative value of fat as a nutrient would be affected under this condition. Similar deprivation experiments could be performed with sugar and, more importantly, energy. The authors convincingly describe evidence that the monkeys’ choices are not based on an energy-maximizing strategy. Again, this might change, if the monkey has experienced hunger, that is, a deprivation in overall caloric intake, without a change in the overall balance of nutrient intake. The upshot of all these experiments would be to investigate the possibility that nutrient value functions are dependent on the internal homeostatic state. This would link the current work with ongoing research into internal mechanisms regulating states of hunger and satiety (16, 17).

Another interesting question regards the relationship between the intrinsic attributes of an option investigated in the current work and the more abstract cognitive factors (e.g., probability of reward, delay until reward delivery, menu of currently available options) investigated in previous work. Clearly, both “bottom-up” intrinsic and “top-down” cognitive attributes influence the value of an option. By experimentally manipulating both factors, it would be possible to determine whether they mutually influence each other, or whether their influence on value is independent of each other. In addition to these functional considerations, there is also the question of whether these factors are handled by independent brain circuits or whether both factors are intermixed in the reward circuitry. At the moment, there is evidence for both possibilities. For example, orbitofrontal cortex lesions impair macaques’ ability to adjust their food choices after food-specific satiation, which requires assessing the food’s intrinsic attributes, but do not impair the tracking of stimulus−reward probabilities (18). This would argue for separate representation of internal and cognitive influences on value. On the other hand, neuroimaging experiments in humans have shown that the ventromedial prefrontal cortex is correlated with the subjective value of a wide range of goods (food, nonfood consumables, and monetary gambles) (19). This supports the idea that at least some brain areas encode a “common currency” that allows for a shared valuation for different options taking into account both “bottom-up” and “top-down” influences on value (20).