Efficiency of post-meal memory inhibition predicts subsequent food intake

Memory processes may contribute to appetite regulation. When people look at palatable foods, their desire to consume them depends upon memory retrieval (i.e., recalling if it will taste good). If memory inhibition occurs during satiety, then pleasant eating-related memories will not be retrieved, making eating less likely. In contrast, if memory inhibition is less efficient, pleasant food-related memories will be retrieved, the food will appear desirable, and the chance of consumption increases. Here we tested whether a putative measure of memory inhibition could predict post-meal snack food intake. Study participants looked at palatable snacks and judged their desire to eat them (i.e., a memory-dependent process), and then ate a small sample of each food, and rated them for liking (i.e., an orosensory-dependent process) – all using category rating scales. Following a filling meal, this test was repeated, alongside others. Finally, participants were given the opportunity for ad libitum snack food consumption, in addition to collecting measures such as impulsivity. Poorer memory inhibition (i.e., smaller changes in wanting relative to liking from pre-to post-meal) was associated with greater consumption of snacks on the ad libitum test (Sr 2 % = 4.4, p = 0.006) after controlling for other variables likely to influence eating (e.g., impulsivity). This effect was maintained even when the memory inhibition measure was based on foods different to those being consumed on the ad libitum snacking test. In conclusion, memory inhibition may contribute to food intake regulation, and when this is less efficient, more palatable food is likely to be eaten in the post-meal period.


Introduction
A series of studies using both animals and humans have suggested that two complementary forms of inhibition, both of which involve memory, may be an important contributor to the regulation of food intake (e.g., Davidson et al., 2005;Davidson et al., 2019;Hannapel et al., 2019;Henderson et al., 2013;Kanoski & Grill, 2017;Parent, 2016Parent, , 2022)).In one, episodic memories of recent food intake serve to inhibit behaviour -namely further food intake -with this process being mediated via the hippocampus (e.g., Parent, 2016Parent, , 2022)).In the other, the presence of physiological satiety signals indicate that food will not be rewarding in this state (e.g., Davidson et al., 2005;Hargrave et al., 2016).This state-based contextual modulation is also supported by the hippocampus, which then serves to inhibit retrieval of excitatory food memories (e.g., Davidson et al., 2019;Kanoski & Grill, 2017).We refer here to this latter process as memory inhibition.In humans, one paradigm -the wanting and liking test -has been used in several studies to index putative inter-individual variation in memory inhibition (e.g., Attuquayefio et al., 2016;Pender et al., 2019;Stevenson et al., 2017;2020).An obvious prediction, and one which has not been tested, is that this index of memory inhibition should predict intake of palatable (i.e., excitatory) foods following a meal.This study provides an initial test of this prediction.
The study of memory inhibition emerged from the animal literature (Davidson et al., 2005(Davidson et al., , 2019;;Kanoski & Grill, 2017).Physiological signals associated with food intake and digestion form an interoceptive context, and this is used by the hippocampus to resolve decision conflicts as to whether food will be rewarding to eat now or not (i.e., should or should not the food cue retrieve an excitatory association -a memory).Retrieval of the excitatory association would then facilitate ingestion, as it would bring to mind rewarding memories of eating that food.In contrast, if memory inhibition occurs, this would prevent such memories coming to mind, and so reduce the likelihood of eating.In humans, a series of studies on intact participants have been used to explore this model.These have been motivated by an additional finding, namely that a Western-style diet, one rich in saturated fat, added salt and sugar, rapidly impairs hippocampal function in both animals and humans (Abbott et al., 2019;Taylor et al., 2021).Consequently, Western-style diets would be expected to impair memory inhibition.
To examine this idea, we developed a paradigm called the wanting and liking test (e.g., Attuquayefio et al., 2016;Pender et al., 2019;Stevenson et al., 2017;2020).In this test, hungry participants are asked to look at palatable snack foods (i.e., excitatory cues) and rate how much they would like to eat them.Responses on this test are presumably based on prior memories of eating these foods.Then, in the next part of the test, participants are asked to eat a small amount of each palatable food and judge how much they like it.This judgment is primarily dependent on orosensory perception of the food's flavour (i.e., its sweetness, its fattiness, etc).The wanting and liking test is then repeated after a meal.
A consistent finding has been that memory-based wanting declines significantly more from pre-to post-meal, than orosensory based liking (e.g., Attuquayefio et al., 2016;Pender et al., 2019;Stevenson et al., 2017;2020).Our argument has been that this greater decline in wanting relative to liking reflects the influence of memory inhibition.Not only does consumption of a Western-style diet impair performance on neuropsychological tests sensitive to hippocampal function, it also impairs memory inhibition as indexed by the wanting and liking test (Attuquayefio et al., 2016;Stevenson et al., 2020).People who habitually consume a Western-style diet have a smaller decline in wanting relative to liking from pre-to post-meal, when compared to those who consume a healthier diet (Attuquayefio et al., 2016).This diet-dependent difference on the wanting and liking test correlates with performance on neuropsychological measures of hippocampal-dependent learning and memory (Attuquayefio et al., 2016).Moreover, experimental manipulations of diet, namely getting people to shift from a healthier diet to a Western-style diet, result in the same type of changes on the wanting and liking test as observed in habitual consumers of a Western-style diet (Stevenson et al., 2020).In addition, these changes correlate with a decline in hippocampal dependent learning and memory (Stevenson et al., 2020).Thus, changes in wanting relative to liking from pre-to post-meal (i.e., change in state) seem to index of hippocampal-dependent memory inhibition.
We have suggested before that effective memory inhibition may reduce the likelihood of eating palatable food in the post-meal period, by inhibiting retrieval of rewarding (i.e., excitatory) memories of food (Davidson et al., 2019;Davidson & Stevenson, 2022).If this is the case, then our putative index of memory inhibition should predict food intake on a post-meal food intake test, where excitatory cues are present (i.e., palatable snack foods).This prediction has not been tested.To undertake such a test, would require assessing several other variables that may also contribute to eating under conditions of satiety.Some of these variables may be more prevalent in people with poorer memory inhibition, such as in those who consume a Western-style diet.One important variable in this regard concerns impulsive traits, whose presence may facilitate choosing highly palatable Western-style foods (e.g., Limbers & Young, 2015;Lumley et al., 2016;Yeomans et al., 2023).Impulsive traits have been consistently identified as an a priori risk-factor for weight gain and obesity (e.g., Davis et al., 2010;Schag et al., 2013).Consequently, a proportion of people with poorer memory inhibition, may also be more impulsive, and hence more likely to eat when palatable food is available.
In sum, the primary aim was to determine if a putative measure of memory inhibition derived from the wanting and liking test (the change in wanting relative to liking for palatable snack foods, from pre-to postmeal) would predict post-meal intake of other palatable snack foodswhen controlling for variables that are also likely to predict snack food intake (e.g., impulsivity).

Overview and design
A summary of the procedure is presented in Fig. 1.Hungry participants completed the key test, the pre-meal wanting and liking test, before eating a filling test meal.This test was then repeated after the test meal, with the change in wanting relative to liking for palatable snack foods, from pre-to post-meal, serving as our putative measure of memory inhibition.Towards the end of the study, participants were provided the opportunity to snack on sweet palatable foods in the two food evaluation tests.Amount consumed on these tests formed the key dependent variable, which we attempted to predict using our putative measure of memory inhibition.
This design included another type of memory procedure -the food description task -to see if changes from pre-to post-meal on this task were also predictive of food intake on the food evaluation tests.Previously (Stevenson et al., 2022(Stevenson et al., , 2023)), we have asked participants to Fig. 1.Summary of the procedure.R.J. Stevenson et al. Appetite 203 (2024) 107686 provide written descriptions of what it would be like to eat various foods, with this task completed pre-and post-meal.Participant food descriptions are then coded for hedonic tone (mentions of positive and negative phrases/words) and mentions of how filling the food would be to eat.We found that food descriptions become less positive and more negative after a meal (Stevenson et al., 2023), and that mentions of how filling a food would be to eat also increase post-meal (Stevenson et al., 2022(Stevenson et al., , 2023)).As both judgments draw on recollections of eating in each state, they should also be susceptible to memory inhibition.Several control measures that might affect snack food intake were also measured.These included two questionnaires to assess for impulsive choice, both generally (Barratt impulsivity scale; Patton et al., 1995) and more specifically in relation to food (Three factor eating questionnaire; Karlsson et al., 2000).These were completed during the second food evaluation test.In addition, several other potentially relevant variables were collected, including gender, body mass index, changes in hunger and fullness ratings, the amount of food consumed during the test meal, and participants liking for the snacks used in the food evaluation tests.

Participants
The study was powered to detect an effect in the small-to-moderate range, namely a Pearson correlation of around 0.2.Consequently, with d set to 0.45, and at a power of 0.80, this would require a total of 154 participants.
One hundred and fifty-nine participants were recruited from the Macquarie University undergraduate psychology pool and took part for course credit.On the study advertisement, potential participants were instructed to self-exclude if they had a current or past eating disorder, significant food restrictions (i.e., vegan, gluten-free diet, currently dieting), or a body weight outside of the normal-to-overweight range (i.e., BMI 17.0-29.9).They were also asked to avoid eating in the 3 h prior to their scheduled test session (be it either lunch or dinner time).Data from twelve participants were excluded (see Fig. 2); five refused to eat one or more of the test foods; three were lost through procedural errors; three were removed as their body mass index was in the obese range; and one overseas student was excluded as they did not understand the tasks and were unfamiliar with the test foods.This left 147 participants for analysis.
All participants consented to take part in the study, which was described generically as examining the effects of 'diet on behaviour'.Participants were informed as to the broad nature of the tasks but were not told that the foods they were going to be exposed to would be weighed before and after the experiment.Participants were debriefed at the end of the study.The protocol was approved by the Macquarie University Human Research Ethics Committee (approval reference number 52023506444635).

Test meal
The test meal consisted of either beef ('On the menu' Woolworths; 260g, 1570 Kj; 15.3g protein, 18.7g fat, 35.1g carbohydrate) or vegetarian lasagne ('On the menu' Woolworths; 260g, 1140 Kj; 7.3g protein, 9.4g fat, 38.5g carbohydrate).The vegetarian lasagne was offered as an alternative only if participants reported being vegetarian, they did not eat beef, or disliked beef lasagne (note that the two meals were matched for weight and volume).Twelve participants (8% of the sample) ate the vegetarian lasagne for their test meal.Both foods have been used in several prior studies (e.g., Attuquayefio et al., 2016;Stevenson et al., 2023) and were selected due to their high acceptability.The first sixteen participants were also offered ice-cream (60g vanilla; Bulla; 430 Kj), but this was discontinued as several participants reported that they did not want more food at the meal.

Food evaluation tests
This consisted of two tests.In the first, participants were given three white bowls of confectionary, with each filled to the maximum (approximate diameter 15 cm, depth 10 cm at peak; see top part of Fig. 3).One white bowl contained Maltesers (approximately 290g; 20 Kj/g), one contained Fruit Basket jellies (Natural Confectionary Company; approximately 450g; 15 Kj/g) and one contained Smarties (Nestle; approximately 430g; 20 Kj/g).These foods were presented in this way (i.e., heaped in each bowl), such that if one or two items were taken it would not be readily noticed.This feature was important as the instructions for this task were ambiguous about whether food could or could not be eaten here.
The second Food evaluation test used the same confectionary items, but this time presented in orange bowls of similar size, but that were just comfortably filled (see lower part of Fig. 3).The Maltesers orange bowl contained approximately 210g of these sweets, the Fruit Basket jellies orange bowl approximately 340g, and the Smarties orange bowl approximately 360g.

Procedure
Prior to participant arrival, each of the three-white bowls (respectively, Maltesers, Smarties, Jellies) and each of the three-orange bowls (respectively, Maltesers, Smarties, Jellies) were weighed to establish their pre-weights.

Pre-meal testing
On arrival participants were instructed to leave their phones/bags by the door.Only one participant was tested at a time.They were then asked both when and what they last ate and drank, and to complete the first set of general rating scales.In total, seven ratings were made; how hungry, thirsty, full, bloated, empty, nauseous, and happy they felt, with each using a seven-point category scale (anchors 'Not at all' and 'Very').Only hunger and fullness are reported, as these were the study focus.
The pre-meal food description test followed.A practice was undertaken first.Participants were instructed to write a description of what it would be like to eat 'a meat pie and sauce' right now.They were told they would be given 1-min to write their description and just to put down what most readily came to mind.If they had not finished by the end of 1min they were told to stop writing after a further 20 s had elapsed.The test proper then began, using two further food descriptions.Which two food descriptions participants were asked to write about was counterbalanced, so approximately half received one set of two food descriptions before the test meal and the other set after or vice versa.One set was composed of 'A prime steak, fries and a salad' and 'A bowl of boiled rice and soy sauce'.The other was composed of 'A rack of ribs, fries and onion rings' and 'A bowl of noodles and soy sauce'.These foods were selected as they had been found to be familiar to most participants in our prior studies and represented meals of differing palatability (Stevenson et al., 2022(Stevenson et al., , 2023)).
The pre-meal wanting and liking test was then completed.This was composed of two phases.In the first, participants were passed a single food item in a plastic transparent sample cup and were asked to look at it.They then wrote down the name of the food and judged, using a seven-point category scale (anchors Not at all, A lot), how much they would like to eat it now.This process was then repeated for each of the five foods used on this test.Presentation order of the test foods was counterbalanced across participants.
In the second phase, participants were again passed a sample cup containing one of the five test foods.This time they were asked to eat it, and then judge how much they liked it (7 point category scale, anchors Not at all, A lot) and how much more of this food would they like to eat now (7 point category scale, anchors None, A lot).They were then asked to rinse their mouth with water.This process was then repeated for each of the remaining foodsnamely the same ones used in the first phase and presented in the same counterbalanced order (to maintain judgmental context).

Meal
The pre-meal general scales were completed next, followed by the test meal.After the test meal was served participants were left alone for 5-mins to eat, followed by an additional 3-mins if they had not finished at the 5-min mark.These time intervals were based on prior use of this form of test meal (e.g., Attuquayefio et al., 2016;Stevenson et al., 2023).Ad libitum water was available throughout.At the end of the test meal participants were asked if they were comfortably full, and whether they wanted any more lasagne.Only one participant requested a second portion.The post-meal general scales were then completed.

Post-meal testing
The post-meal wanting and liking test was undertaken next, with this being identical to the pre-meal test.This was followed by the post-meal food description task.This was procedurally identical to the pre-meal test, but this time without the practice phase, and using the other set of food descriptions.
The first food evaluation test followed.Participants were brought a tray containing the three-white bowls (see Fig. 1).They were told "I would like you to imagine eating each food.As you imagine eating it, I want you to write down on the questionnaire what eating it would be like for that food.Please start a new questionnaire for each food".It was left ambiguous as to whether participants were or were not allowed to eat any of the foods in front of them and no specific instruction regarding this were provided.If participants asked, they were told to follow the instructions they had been given.
Participants were told that the experimenter would leave the roomas they had for the meal -returning in 5-mins.Participants then completed a new questionnaire for each food.Each questionnaire asked them to imagine eating each confectionary item in the white bowl in front of them.They then had to write a description of its flavour, texture, and how filling it would be.Rating scales were then completed for how frequently they normally ate it (5 selections; Daily, Weekly, Monthly, More than yearly, Never) and how much they normally enjoyed it (7 point category scale, anchors Not at all, A lot).Data from these questionnaires is not reported, as the primary purpose of this task was to increase temptation to take food from the white bowls.
When the experimenter returned, the tray containing the white bowls of confectionary was immediately removed.The second food evaluation test then followed.The experimenter brought a tray to the participant containing the three-orange bowls.They were told "You are welcome to snack on these if you wish, while completing the final questionnaires".Participants then completed the 18-item revised form of the Three Factor Eating Questionnaire (Karlsson et al., 2000) and the 30-item Barratt impulsivity questionnaire (BIS; Patton et al., 1995).The final questionnaire asked how tempted they had been by these foods on the earlier 'white bowls' test (7 point category scale, anchors; Not at all, Very tempted), whether it was OK to eat the foods on the earlier 'white bowls' test (7 point category scale, anchors; Not OK, Definitely OK), and whether they remembered eating any of the foods from the 'white bowls' (7 point category scale, anchors; No, Definitely did).They were then asked to rate how much they liked each of the three test foods using seven-point category scales (anchors, Not at all, A lot).Completion of this final part of the study was self-paced, with the experimenter in the room, but sitting out of sight of the participant.Participants were asked to call the experimenter when they had finished.
Finally, participants' height and weight were measured, followed by a debriefing/question-answering session about the study (noting that participants were not asked what they thought the aim was).After the participant had left, any remaining lasagne was weighed to calculate amount eaten.The white and orange bowls were then reweighed to establish if anything had been taken from them.The whole experiment took around 1-h, with little variability, as each component was either presented by the research assistant (e.g., wanting and liking test) or time delimited by them (e.g., food description task).

Analysis
Bootstrapped hierarchical multiple regression with predictor variables entered in blocks was used to test the primary aim, namely whether our putative measure of memory inhibition derived from the wanting and liking test (i.e., the change in wanting relative to liking for palatable snack foods, from pre-to post-meal) could predict the amount eaten on the food evaluation tests (alongside other potential predictor variables).
The following predictor variables were selected and organised in the following way.Gender and BMI formed the first block as both these variables can influence food intake.They were followed by kJ eaten on the test meal and change in hunger and fullness from pre-to post-meal, all of which may affect appetite.Liking for the three foods used on the food evaluation test was also included as a predictor, as this may impact whether any snacks are consumed at all.Two individual differences measures linked to ingestive behaviour were also used, namely the three sub-scales from the impulsivity measure -BIS attention, BIS motor and BIS non-planning, and the three subscales from the TFEQ-R18 -uncontrolled eating, cognitive restraint and emotional eating.The final predictor was the change in wanting relative to liking from pre-to post-meal (i.e., the interaction of State by Rating from the wanting and liking test) -the putative measure of memory inhibition.
These data were suitable for regression, with satisfactory case-topredictor variable ratio (i.e., >10:1), no multicollinearity, and no cases identified as exerting undue influence.The same approach was used to examine if the food description task variables could also predict food intake on the food evaluation tests, with these too being suitable for this approach.
The analysis of the wanting and liking test data focussed on just two of the three rating types obtained on this test, namely the wanting ratings made when looking at the food and the liking ratings made when tasting it.This reflects both the focus of past work using this test (Attuquayefio et al., 2016;Pender et al., 2019;Stevenson et al., 2017;2020) and the fact that it allows us to contrast a wanting measure based upon memory of the food (i.e., the judgment made when looking at the food) with the liking measure based upon the food's orosensory properties (i.e., the judgment made when/just after tasting it).Consequently, we do not report the want more ratings, which were made immediately after tasting.The wanting and liking ratings were analysed using a repeated measures ANOVA, with State (pre-lunch vs. post-lunch) and Rating (Wanting vs. Liking) as factors.
The food description task data were coded by FW, a researcher familiar with the coding categories used here -namely identifying positively and negatively valenced words/phrases, and words/phrases indicative of how filling the food would be to eat.We did not establish the validity of these coding categories as this had been done previously with this coder on this procedure (see Stevenson et al., 2022; finding a high degree of reliability).These data were then analysed using repeated measures ANOVA, with State (pre-lunch vs. post-lunch) and Affect (positive vs negative phrases/words) as factors.
Data employed in these analyses were suitable for parametric testing (as established by examination of skew and kurtosis, in conjunction with the Shapiro-Wilks test) except as identified in the Results.Alpha was set at 0.05 for all analyses.

Participant characteristics
The sample consisted of 57 males and 90 female, with participant details and their BIS and TFEQ-R18 scores presented in Table 1.

Food evaluation tests
In the first Food evaluation test (white bowls) only 14 participants took food (see Fig. 4, upper panel).Of the 14 who took food, the mean amount consumed was 12.9g (SD = 17.9).
In the second Food evaluation test (orange bowls), 81 participants took food, leaving 66 who did not (see Fig. 4, lower panel).Of the 81 taking food, the mean amount consumed was 28.8g (SD = 28.9).
Due to the low consumption rate on the first test, and that the same test foods were used in the second test, we combined intake across both for the analysis.Food intake from the food evaluation test was not normally distributed, and so we undertook a square root transformation of grams of food consumed (after adding 1 to each score), which substantially improved the skewness and kurtosis of these data.We use this transformed variable for the later bootstrapped regression analyses.
Participants were asked at the end of the study to report their liking for each of the three foods used in the food evaluation tests.Scores for all three foods were at the midpoint of the scales (Smarties 3.7/7, SD = 1.9;Jellies 4.0/7, SD = 2.0; Maltesers 4.8/7, SD = 1.9), suggesting they were generally liked, and these were averaged to produce one liking score for the later bootstrapped regression analyses.

Wanting and liking test
This analysis used a two-way mixed design ANOVA, with State (pre-  mealpost-meal; M = 0.9, SD = 0.9), which yields the State by Rating interaction effect (M = 0.4, SD = 1.1) -our putative measure of memory inhibition.

Regression analyses
Using bootstrapped hierarchical multiple regression, with the square-root transformed consumption data from the food evaluation tests as the dependent variable, we entered the predictor variables in a series of successive blocks (gender and BMI; then kJ eaten on the test meal and change in hunger and fullness from pre-to post-meal; then mean liking for the three foods used on the food evaluation test; then BIS attention, BIS motor and BIS non-planning; then TFEQ-R18 uncontrolled eating, cognitive restraint and emotional eating; and finally, change in wanting relative to liking from pre-to post-meal (i.e., the interaction of State by Rating from the wanting and liking test) -the putative measure of memory inhibition.
The outcome is presented in Table 2.The final model was significant, F(13,133) = 3.21, p = 0.001, accounting for 16.4% of the variance in the amount of food consumed on the food evaluation tests.Looking at whether the entry of any blocks of variables made a significant improvement to the model, two blocks were identified.One was liking for the foods used on the food evaluation tests (F Change(1,140) = 17.53, p < 0.001) and the other was change from pre-to post-meal in wanting relative to liking, from the wanting and liking test, which also significantly improved the model fit (F Change(1,133) = 7.67, p = 0.006) -the putative measure of memory inhibitionwith this effect being illustrated in Fig. 6.Both variables, not surprisingly, accounted for unique variance in food consumed on the food evaluation tests.
The wanting and liking test utilised five snack foods.One of these foods was identical to a food used in the food evaluation task (Maltesers), one was very similar (Jelly), one was similar (Oreo cookie' namely being sweet), but two were different (the savoury snacksa Pringle and a Cheetoh).To determine if food similarity was a contributory factor to the predictive ability of the change in wanting relative to liking from pre-to post-meal, we recomputed the wanting minus liking change score, but now basing this solely on the two savoury food items.We then re-ran

Table 2
Final bootstrapped regression model predicting the amount eaten on the food evaluation tests.Finally, we conducted four sensitivity analyses.The first checked the impact of participants who consumed significantly less (>2 SDs) than the mean amount eaten on the test meal, with the regression analysis repeated with them excluded.The second, third and fourth analyses, due to their larger numbers, involved adding new bivariate variables to the regression analysis.Here we tested if the type of meal impacted the outcome (i.e., beef lasagne vs. vegetarian lasagne), if having consumed ice-cream in the meal impacted the outcome (i.e., consumed vs. did not consume ice-cream) and if the time the experiment was conducted impacted the outcome (at lunch vs. at dinner).None of these additional analyses changed the outcome.For each analysis, the overall model remained significant (all p's < 0.001), and the only significant predictors in each case were liking for the evaluation test foods (all p's < 0.001) and change in wanting relative to liking from pre-to post-meal (i.e., memory inhibition effect; all p's < 0.01).

Food description test
Participants descriptions of what it would be like to eat the foods named in this test were coded for mention of positive and negative words/phrases.Number of mentions were then analysed using a twoway mixed design ANOVA, with State (pre-lunch vs. post-lunch) and Affect (positive vs negative phrases/words) as the within-subject variables.
The ANOVA revealed a main effect of Affect (F(1,146) = 8.02, p = 0.005, partial-eta squared = 0.05), with a greater dominance of positive over negative words/phrases in all descriptions (M positive = 1.2 vs. M negative = 0.9), which was qualified by an interaction of State and Affect (F(1,144) = 38.81,p = 0.001, partial-eta squared = 0.21).The State by Affect interaction reflects more positive (M = 1.4) and less negative statements (M = 0.7) before the meal and less positive (M = 1.0) and more negative (M = 1.1) statements following the meal.
As with the Wanting and liking test data, we then explored using boot-strapped regression, whether the State by Affect interaction could predict consumption on the food evaluation tests, in the presence of the same set of variables used in the preceding regression analyses.However, inclusion of this variable (i.e., State by Affect interaction) made no significant contribution to the model (F Change <1).
Participants descriptions were also coded for how filling they thought the food would be.Reports of fillingness significantly increased from before (M = 0.3 reports) to after the meal (M = 0.4 reports; t(146) = 3.78, p < 0.001).Inclusion of the change in fillingness in the same regression model described above revealed that it made no significant contribution to the model.

Final questionnaire
Participants were asked how tempted they were to consume the foods offered in the First food evaluation test (white bowls).A moderate amount of temptation was reported (M = 3.5/7).This temptation score was positively correlated with taking food (vs.not taking food) from the white bowls (Spearman's rho (147) = 0.24, p = 0.003).Similarly, this temptation score was also positively correlated with taking food from the orange bowls in the second Food evaluation test (r(147) = 0.44, p < 0.001).
A second question asked if it was OK to take food from the white bowls on the first food evaluation test, with the mean response reflecting some uncertainty over this (M = 3.3; 1 = Not OK, 7 = Definitely OK).Scores on this rating were not correlated with whether participants took anything from the white bowls on the first food evaluation test.
A third question asked if participants could remember taking any food from the white bowls (M response = 2.2; 1 = No, 7 = Definitely did).Scores on this rating were positively correlated with whether food was taken from the white bowls (Spearman's rho (146) = 0.19, p = 0.025).

Discussion
The primary aim of this experiment was to determine if a putative measure of memory inhibition, derived from the wanting and liking test, would predict intake of palatable snack foods following a test meal.The measure of memory inhibition was the change in wanting (based on memory) relative to liking (based on orosensory perception) for palatable snacks, with these ratings obtained before and after the test meal.Intake of palatable food following the test meal was measured using two food evaluation tests that both used the same foods.On the first (white bowls), very few participants ate food, so we combined these data with consumption from the second (orange bowls) where a majority did take food.Here, using multiple regression to control for other factors likely to influence food intake (e.g., impulsivity), we found that the putative measure of memory inhibition was predictive of food intake on the combined food evaluation tests.Moreover, the predictive ability of the putative memory inhibition measure was retained under two further sets of conditions.First, when based on palatable foods different to those used on the food evaluation test (i.e., savoury vs. sweet) -although noting of course that all these foods are still highly palatable snacks.Second, from sensitivity testing, which revealed that the result was not influenced by variations in the study design (e.g., time of testing).These findings suggest that individual variation on our putative measure of memory inhibition is predictive of palatable snack food intake following a meal.Moreover, this suggests that memory inhibition may have a role in the routine regulation of human food intake, and especially so under conditions conducive to excess energy intake (i.e., eating palatable snack foods after a meal).
The study also had a secondary aim.This involved the food description task, which was completed before and after the test meal, with participants providing written descriptions of what eating particular foods would be like.Two variables were derived from these descriptions.One was the number of positive and negative words and phrases, and the other was reports of words and phrases indicating how filling the food would be.We then examined if these variables could also predict food intake on the combined food evaluation test data.Consistent with previous studies, both these variables were affected by a change in state (i.e., eating the test meal), with more negative words/ phrases and more reports of food fillingness when sated (Stevenson et al., 2022(Stevenson et al., , 2023)).However, neither could predict palatable food intake on the food evaluation test.
The current design included two tests that allowed us to assess intake of palatable snack foods (i.e., the food evaluation tests).Two tests were used as we were uncertain what specific conditions would be needed to favour palatable snack food intake following a meal.The first food evaluation test (white bowls) attempted to maximise temptation, with no experimenter present, heaped bowls of food where some missing would not be obvious, instructions to imagine eating the food, and ambiguity as to whether food could be taken or not.However, under these temptation conditions, only 10% of participants took food.A final questionnaire surveyed participants about their decisions to eat on this first 'white bowls' test.It found that while many participants were tempted, they were uncertain as to whether eating was allowed.While such uncertainty was not linked to food intake, it seems likely that the absence of an explicit instruction to eat probably explained why most participants did not take food.The main reason for this assertion comes from participants' behaviour on the second food evaluation test (orange bowls).Here, the majority (55%) took food, and notably on this test participants were explicitly instructed to help themselves to food if they wished to do so.As both tests used the same snack foods, the combined intake was consequently used in the principal analyses.
The primary purpose of this study was to see if our putative measure of memory inhibition, derived from the wanting and liking test, could predict palatable food intake on the food evaluation test.The inhibition measure was derived from the finding that wanting ratings, which must be based primarily on mnemonic information as the food is just being viewed, decrease far more from pre-to post-meal (i.e., across state) than ratings of food liking obtained contemporaneously with tasting the food (i.e., more dependent on orosensory properties).Not only was this greater decline in wanting relative to liking present in this data set, as we have consistently observed before (Attuquayefio et al., 2016;Pender et al., 2019;Stevenson et al., 2017;2020), it was also able to predict food intake on a later and unrelated task.Specifically, and after controlling for the impact of other variables known to affect food intake, greater declines in wanting relative to liking were associated with reduced intake on the food evaluation tests.
In the Introduction we outlined some of the research linking memory inhibition to food intake regulation.The practical importance of this account derives from findings concerning consumption of a Westernstyle diet.This type of diet in known to cause a rapid and selective impairment to the hippocampus in both animals and people (Abbott et al., 2019;Taylor et al., 2021).One consequence of this impairment are changes to performance on the wanting and liking task.This has been demonstrated both cross-sectionally (i.e., contrasting groups who differ in their adherence to a Western-style diet; Attuquayefio et al., 2016) and experimentally (i.e., exposing participants who eat a healthy diet to a week of a Western-style diet; Stevenson et al., 2020).Both study types have convergent findings.On the wanting and liking test, people who consume a Western-style diet or have been experimentally exposed to it, demonstrate smaller changes in wanting relative to liking, from pre-to post-meala change in our putative measure of memory inhibition.The practical consequences of this change are now potentially apparent.Based on the findings obtained in this study, participants who demonstrate smaller changes in wanting relative to liking are more likely to eat palatable snack foods on a subsequent experimental test.This lends credence to the idea that a Western-style diet dysregulates appetite by impairing hippocampal function (Davidson et al., 2019;Davidson & Stevenson, 2022), and this then manifests as poorer inhibitory control of appetite, resulting in over-consumption especially of palatable foods (i.e., those that are most tempting/excitatory).This in turn may create a vicious circle of over-consumption of palatable Western-style foods, further hippocampal damage, further impairment of appetite regulation and weight gainprecisely the pattern predicted by Davidson's vicious circle model of obesity (e.g., Hargrave et al., 2016;Kanoski & Davidson, 2011).
The present study also included measures of impulsivity, notably the Barratt impulsivity scale (BIS; Patton et al., 1995).This self-report measure was used as previous data have revealed that greater impulsivity is weakly linked to poorer (less healthy) dietary choices (e.g., Limbers & Young, 2015;Lumley et al., 2016;Yeomans et al., 2023).Behavioural measures of impulsivity have also obtained similar findings, with poorer self-control linked to dietary reports of less healthy food intake (e.g., Ames et al., 2014;Lumley et al., 2016).We did not find evidence of such relationship here, as BIS scores did not predict intake on the food evaluation tests.This may have resulted from the typically small effect size of these relationships, and the presence of other predictor variables.
We also included an additional measure related to impulsivity, the TFEQ.This too was not predictive of intake on the food evaluation tests.As we did not inquire in detail about participants disordered eating (other than asking those with a current or past eating disorder to selfexclude) the failure to find a significant TFEQ component is reassuring.This is because higher scores on this measure, especially uncontrolled and emotional eating, have been linked to binge eating (e.g., Abdo et al., 2020) and a propensity to binge eat might have influenced behaviour on the food evaluation tests.
As noted above, other variables may have contributed to participants consumption of the snacks on the food evaluation tests.One variable that significantly explained variance in intake was liking for the snacks used on these tests, with greater liking predicting greater intake.Other factors too are likely to be important, some of which were not directly measured here.One such factor is food cue reactivity (Kanoski & Boutelle, 2022).This has both a genetic and an environmental component and impacts participants response to reinforcing food cues.Participants with heightened food cue reactivity, are likely to respond more strongly to the presence of food cues (i.e., salivating, likelihood of eating etc), and as such it may be that the liking ratings made when they looked at the snacks used in the food evaluation tests, reflected, to some extent, variation in reactivity.This may further explain this variable's (snack food liking) ability to predict food intake on the food evaluation tests.
There are two other potential factors that could have contributed to food intake on the food evaluation tests.First, participants may have believed that eating the snacks was the goal of the study, and so social desirability could have motivated intake.As noted earlier, we did not systematically examine whether participants knew that the white and orange bowls would be weighed at the end of the study, and so it is not possible to know how relevant this may be.However, if social desirability did enhance food intake, this effect is likely to be independent of any influence of memory inhibition and so would be expected to just add noise.Second, another influence may have been the impact of the questionnaires being completed with the second food evaluation test (orange bowls).These may have prompted recollections of prior food intake from earlier in the experiment (i.e., white bowls), which may have suppressed intake in some participants who had eaten earlier (e.g., Collins & Stafford, 2014;Higgs, 2002;Higgs, Williamson & Attwood, 2008).This too would likely have added noise, thus making it harder to detect a relationship with performance on the wanting and liking test.
The current study afforded an opportunity to examine the predictive ability of another task, the food description test (Stevenson et al., 2022(Stevenson et al., , 2023)).This task relies solely on recollection of food-related memories, and so in theory one might expect that memory inhibition would operate here as well.In this case it might be expected to reduce the number of positive memories and increase the number of negative memories from pre-to post-meal.Similarly, it might also increase recollection of unrewarding aspects of consumption when sated, such as adding to feelings of fullness by consuming more filling foods.While these properties of memory were indeed present, replicating prior findings (Stevenson et al., 2022(Stevenson et al., , 2023)), there was however little evidence for any relationship with food intake on the food evaluation tests.There are three possible reasons for this.First, this measure relies on response coding, which could be less reliable than self-report ratings of wanting and liking.However, this would seem unlikely as previous assessments have suggested a high degree of coding reliability (Stevenson et al., 2022).Second, the food description task used foods that were markedly different to those used on the food evaluation tests, which may have weakened its predictive power.However, this should not be important, as the memory inhibition effect should be independent of food type, as was evident for the wanting and liking test data (i.e., where use of unrelated savoury snack foods still predicted intake of sweet snack foods).Third, the absence of a sensory liking measure for the actual foods used on the food description task may be particularly relevant.From prior work it appears that it is only the change in wanting (memory) relative to liking (orosensory perception) that is sensitive to impairment by a Western-style diet and correlated with hippocampal dependent learning and memory (Attuquayefio et al., 2016;Stevenson et al., 2020).This is because the effects of memory inhibition may only be revealed relative to the more general reduction in food-liking that occurs following a mealnamely alliesthesia (Cabanac, 1971).Consequently, following a meal, wanting judgments of food may be affected by both alliesthesia and memory inhibition, while liking judgments based primarily on orosensory perception would be affected mainly by alliesthesia.Therefore, it may require the combination of wanting and likingnamely the change in wanting minus change in liking from pre-to post-meal -to detect a memory inhibition effect.
As with all studies this one had its limitations.The study used a homogenous sample drawn from undergraduate students, and hence less representative of the broader population.The controlled nature of the study, and its laboratory setting, limit its ecological validity.In addition, the study foods were limited, both at the test meal, and for assessing additional food intake.Nonetheless, these findings provide initial support to the idea that memory inhibition can influence palatable energydense snack food intake.Such foods, which are widely eaten between meals, may significantly contribute to excess weight gain (e.g., Njike et al., 2016).
In conclusion, this study examined whether a putative measure of memory inhibition derived from the wanting and liking test, could predict later palatable snack food intake.We found that after controlling for other variables known to affect food intake, greater declines in wanting relative to liking, pre-to post-meal, were associated with reduced intake on the food evaluation tests.The predictive ability of this putative measure of memory inhibition was maintained even when it was based on foods different to those being consumed on the later food intake test, and when accounting for variations in the procedure.In sum, this study suggests that memory inhibition contribute to food intake regulation, and that when memory inhibition is less effective, then more is eaten in a test situation where sweet palatable snack foods are available following a meal.

Ethics statement
All participants consented to take part in the study, which was described generically as examining the effects of 'diet on behaviour'.Participants were informed as to the broad nature of the tasks but were not told that the foods they were going to be exposed to would be weighed before and after the experiment.Participants were debriefed at the end of the study.The protocol was approved by the Macquarie University Human Research Ethics Committee (approval reference number 52023506444635).

Declaration of competing interest
The authors report no conflicts of interest.

Fig. 3 .
Fig. 3.The white and orange bowls used in the first (upper) and second (lower) food evaluation tests.
lunch vs. post-lunch) and Rating (Wanting vs. Liking) as within-subject variables.There was a significant main effect of State (F(1,146) = 175.21,p < 0.001, partial eta-squared = 0.55), and Rating (F(1,146) = 168.66,p < 0.001, partial eta-squared = 0.54) and a significant interaction between these variables, with wanting declining more pre-to post-meal than liking (F(1,146) = 21.90, p < 0.001, partial etasquared = 0.13) -as observed before.The interaction effect is illustrated in Fig.5, where the variable of key interest for subsequent analyses is the change in wanting from pre-to post-meal ([pre-mealpost-meal; M = 1.3, SD = 1.3) minus the change in liking from pre-to post-meal ([pre-

Fig. 4 .
Fig. 4. Upper: Histogram of the amount of food eaten on the first Food evaluation test (white bowls).Lower: Histogram of the amount of food eaten on the second Food evaluation test (orange bowls).

Fig. 5 .
Fig. 5. Change in wanting and liking ratings from pre-to post-meal, with wanting changing significantly more than liking (State by Rating interactionthe putative measure of memory inhibition).
in wanting relative to liking from pre-to post-meal (i.e., the memory inhibition effect) − 0.51 (0.19) − 0.86 to − 0.13 0.006 the regression model.The outcome was the same, with a significant overall model (F(13,133) = 2.98, p = 0.001, accounting for 15% of the variance [adjusted R 2 ]), and with the savoury based change in wanting relative to liking from pre-to post-meal score and liking for the foods used on the food evaluation tests, as the only significant unique predictors.

Fig. 6 .
Fig. 6.Partial regression plot showing the relationship between amount of food consumed on the food evaluation tests and the change in wanting relative to liking from pre-to post-mealthe putative measure of memory inhibition.

Table 1
Participant characteristics and scores on the Barratt Impulsivity Scale and the Three-Factor Eating Questionnaire (R18).