The prospect that memory plays an important role in the regulation of food intake is consistent with an emerging literature on ‘memory for recent eating’ in humans [13], [14]. In a series of studies, Higgs and colleagues have shown that reminding people of a recent meal can decrease the amount that is consumed at a subsequent meal [13], [15]. This effect persists for several hours into the inter-meal interval and, importantly, it is evident only when memory of a very recent meal is recalled.

With this paradigm, a potential concern is that the effect on food intake is revealed only after an instruction to recall a recent meal. Whether humans routinely retrieve explicit memories of recent meals remains unclear. An alternative strategy is to disrupt memory encoding during a meal and then measure appetite and intake at a subsequent meal. This can be achieved by asking volunteers to engage in a distracting task (e.g., watching television or playing a computer game) while eating. This not only impairs the quality of memory for the meal but it is associated with elevated hunger in the inter-meal interval and with greater consumption at a subsequent meal [16], [17]. Again, a difficulty with this approach is that the effect of distraction on memory formation is very difficult to measure or confirm with certainty. Moreover, distraction has the potential to influence eating rate, mood, and level of stress, all of which are known to moderate appetite and food intake [18]–[20].

Perhaps the most striking evidence for impaired encoding is found in patients with retrograde amnesia. Again, consistent with Higgs’s interpretation, bilateral hippocampal damage is associated with hyperphagia - after eating one meal to fullness, an amnesic may go on to consume further meals and to report no memory for recent eating and little change in hunger [21]–[23]. Albeit dramatic, to date, these behaviours have only been documented in a small number of individuals, many of whom had structural deficits that extended beyond the hippocampus (amygdala and insula). More generally, neuroscientists widely acknowledge the limitations of inferring normal cognitive function from examples of neurological impairment [24].

Together, these findings highlight a potentially important role for episodic memory in the control of meal size and appetite in humans. This merits attention, not least because it challenges our understanding of short-term energy regulation and, in particular, the long-standing assumption that cognition plays a role primarily around a meal and during the early stages of the ‘satiety cascade’ [25]. However, to test this hypothesis directly, a novel paradigm is required. Specifically, one that can both isolate and quantify the effects of episodic memory on satiety, relative to other psychobiological influences. For the first time, we present an empirical approach that creates this dissociation, under relatively normal eating conditions, and in neurologically intact humans.

Previously, Wansink et al. have used a passive self-refilling soup bowl to remove visual information about the amount of soup that is consumed in a meal [26]. In the present study we used an adapted version of this paradigm. Specifically, we developed a process that enables the experimenter to either increase or decrease a predetermined amount of soup into a soup bowl. This was achieved covertly, thereby enabling us to systematically increase or decrease the volume that a participant consumes relative to the amount that he or she observes they have consumed. In so doing, a mismatch can be achieved between i) the proximal effects of the soup in the GI tract and, ii) an episodic memory that forms around the amount consumed during a meal. By exploring the interaction between actual and remembered amounts we sought to quantify their relative contribution to hunger and fullness over a three-hour period.

A second objective was to establish the extent to which our memory manipulation impacts beliefs about the soup at a subsequent test session. Recently, we and others have explored a range of phenomena relating to ‘expected satiation’ – the extent to which a food is expected to deliver fullness when compared with other foods on a calorie-for-calorie basis [27]–[32]. Expected satiation is an excellent predictor of the energy content of self-selected meals and more important than how much a food is liked [27].

Further into the inter-meal interval, a different pattern of results was observed. Two and three hours after the meal, hunger was no longer predicted by the actual amount consumed. This was the case despite the fact that participants ate either 300 ml or 500 ml of soup. Instead, where differences in hunger were observed, these related to the perceived amount at the beginning of the meal. Specifically, participants who were shown 500 ml of soup experienced greater satiety than those who were shown 300 ml. This result accords with Higgs’ original proposition that satiety is influenced by memory for a recently consumed meal [13]. In particular, it fits well with a previous observation that effects of memory recall are evident three hours after eating but not after only one hour [15]. More generally, our findings are noteworthy because they reveal that the role for memory processes is substantial and that it can be exposed even without the need to explicitly cue a memory of recent eating.

The prospect that satiety is influenced by memory for recent eating is consistent with studies exploring the role of expectations around mealtime. Several studies show that beliefs about the content or energy density of a meal can have a marked effect on subsequent hunger and fullness [37], [39], [42], [43]. One suggestion is that information about a meal may trigger ‘meal schema’ that influences intake at a subsequent meal [44]. For example, Capaldi and colleagues have reported differences in subsequent intake after eating a food described as either a ‘meal’ or as a ‘snack’ [45]. Our findings also add to emerging evidence that the retrieval of food-related imagery can impact appetite and energy intake directly [46].

One possibility is that memory for recent eating serves a purpose. Specifically, it may help to interpret post-ingestive signals by attributing them to a recently consumed meal [15], [23]. We suggest that this process enables a ‘tuning’ or contextualisation of the interoceptive signals that are generated by ‘satiety hormones’ such as cholecystokinin (CCK) [47]. One of the important advantages of our approach is that it enables us to isolate and estimate the impact of post-ingestive feedback after controlling for effects of memory for recent eating. Two and three hours into the inter-meal interval, participants who consumed very different amounts of soup (300 ml or 500 ml) reported broadly similar hunger, suggesting blunted sensitivity based on post-ingestive feedback alone. Consistent with this interpretation, amnesiacs appear to experience a disconnect between feelings of hunger and appetitive behaviour [22]. This failure to recognise and integrate visceral sensations also accords with animal hippocampal-lesion studies showing an impaired ability to use interoceptive states to predict a shock or a reward in ambiguous environments [48], [49]. In particular, one suggestion is that the hippocampus is responsible for inhibiting food intake and that this process can be conceptualised as an example of negative occasion setting [9].

These ideas are important and well grounded. Nevertheless, they remain largely untested in humans. In particular, an opportunity exists to explore obese/lean differences in memory function and appetite control. One hypothesis is that diets that are high in saturated fat impair hippocampal function and that this leads to a deficit in memory performance [10]. This results in weakened inhibitory control, leading to greater consumption of high-fat foods, further impairment, and further weight gain [9]. In future, this proposition might be explored by comparing the effects of our manipulation in consumers of high- and low-fat diets and/or in patients with Alzheimer’s disease.

A potential concern is that the manipulation of perceived intake amounts to a form of deception that tells us little about normal appetite regulation. In response, we note that our volunteers were unaware that the volume of soup had been manipulated. This makes it very difficult to attribute our findings to a simple demand characteristic. An important issue relates to whether the memory for recent eating can be modified. Many everyday behaviours are supported by implicit memory [50] - in other words, memories that facilitate performance without the need for conscious or intentional recollection of those experiences (e.g., walking to work). In this context, learning is often regarded as ‘incidental’ because it occurs without conscious effort. Memory for recent eating might also be influenced by explicit processes that are under conscious control. If this is the case then an opportunity exists to enhance satiety by avoiding distraction and encouraging encoding during a meal. In a recent study, Higgs and Donohoe showed that focusing on the sensory characteristics of a food (while eating) leads to a reduction in intake at a subsequent meal [51]. They do not distinguish between implicit and explicit processes. Nevertheless, this is an exciting finding and one that holds promise as the basis for a novel therapeutic intervention.

Our manipulation check indicates that participants are able to discriminate between a 300 ml and a 500 ml portion of soup and that this ability is also expressed in memory for these portions, both immediately after exposure and after a two-hour interval. This is critical, because it shows that memory is differentially influenced by our manipulation, even though participants are never instructed to encode the amount that they have consumed. In of itself, this does not demonstrate a causal relationship between hunger and memory for recent eating. However, this would seem a parsimonious explanation for our findings. Nevertheless, two alternatives merit consideration. First, the effect of perceived volume reflects subtle differences in the capacity of the (perceived) large and small portion to elicit a conditioned cephalic phase response at the time of ingestion [52], [53]. In relation to this idea, we note that the effect of perceived volume was evident only after a delay of two hours and not immediately after eating, as might be expected were this the case. Nevertheless, it is clear that ‘high level’ beliefs and cognition can influence stomach emptying and endocrine responses to foods and beverages [54] and we recommend that these measures should be included in future research. A second possibility is that perceived volume influenced mood. Specifically, participants who saw a 300-ml portion may have been disappointed by its small size and responded to this negative response by rating their hunger higher than those who saw a 500-ml portion. In relation to this proposition, again, we suggest that a ‘protest vote’ would be more likely immediately after receiving the small portion. Instead, both immediately and at one-hour post-consumption, participants experienced a similar reduction in hunger irrespective of whether they saw 300 ml or 500 ml. Notwithstanding this point, to eliminate this hypothesis with certainty we would recommend that measures of mood and ‘portion satisfaction’ should be included in future protocols.

In addition to the immediate effects of memory on post-meal hunger and fullness, we also assessed effects on the expected satiation of a fixed portion of soup (400 ml) at a subsequent test session. Regardless of amount eaten, those participants who initially saw a smaller portion of soup (day 1) then went on to expect the 400 ml portion to be relatively less satiating (day 2). A likely explanation is that participants were biased by their recent post-ingestive experience. Those who initially saw a large portion then went on to experience a greater reduction in hunger. This memory for hunger then biased estimates of expected satiation 24-hours later. Although this interpretation remains to be tested formally, it is consistent with models that characterise the retrieval of abstract knowledge (expected satiation included) in terms of multiple activation of episodic memory traces [55]. Indeed, recent models emphasis the importance of recency in this form of learning [56].

Previously, we have shown that expected satiation is dynamic and it ‘drifts’ over time [29], [31]. Shifts are more likely to involve an increase than a decrease in expectations. However, once learned, these expectations may be preserved over long periods [57], perhaps even permanently. In future, it would be interesting to measure the extent to which our manipulation leads to sustained changes in expected satiation and the extent to which this generalises to other types of soup. We also note that memories are more likely to be retrieved if retrieval takes place in the same environment in which a memory was encoded [58]. In our experiment, participants were tested in the same environment on both days. It remains to be determined whether a shift in expected satiation is dependent on this kind of context-dependent memory.