Fat and Carbohydrate Interact to Potentiate Food Reward in Healthy Weight but Not in Overweight or Obesity

Prior work suggests that actual, but not estimated, energy density drives the reinforcing value of food and that energy from fat and carbohydrate can interact to potentiate reward. Here we sought to replicate these findings in an American sample and to determine if the effects are influenced by body mass index (BMI). Thirty participants with healthy weight (HW; BMI 21.92 ± 1.77; M ± SD) and 30 participants with overweight/obesity (OW/OB; BMI 29.42 ± 4.44) rated pictures of common American snacks in 120-kcal portions for liking, familiarity, frequency of consumption, expected satiety, healthiness, energy content, energy density, and price. Participants then completed an auction task where they bid for the opportunity to consume each food. Snacks contained either primarily carbohydrate, primarily fat, or roughly equal portions of fat and carbohydrate (combo). Replicating prior work, we found that participants with HW bid the most for combo foods in linear mixed model analyses. This effect was not observed among individuals with OW/OB. Additionally, in contrast with previous reports, our linear regression analyses revealed a negative relationship between the actual energy density of the snacks and bid amount that was mediated by food price. Our findings support altered macronutrient reinforcement in obesity and highlight potential influences of the food environment on the regulation of food reward.

then completed an auction task where they bid for the opportunity to consume each food. Snacks 23 contained either primarily carbohydrate, primarily fat, or roughly equal portions of fat and 24 carbohydrate (combo). Replicating prior work, we found that participants with HW bid the most for 25 combo foods in linear mixed model analyses. This effect was not observed among individuals with 26 OW/OB. Additionally, in contrast with previous reports [1,2], our linear regression analyses revealed 27 a negative relationship between the actual energy density of the snacks and bid amount that was 28 mediated by food price. Our findings support altered macronutrient reinforcement in obesity and 29 highlight potential influences of the food environment on the regulation of food reward. 34 Food choices depend on a sophisticated interaction of biology with the social, economic, 35 perceptual, and nutritional characteristics of food. Biological signals conveying nutritional information 36 may be conscious, such as the sweetness of watermelon, or they may be unconscious, as in the case 37 of peripherally-derived signals generated during nutrient metabolism, such as glucose oxidation [3]. 38 Understanding how these conscious and unconscious signals are integrated to regulate food choice 39 may reveal new insights into the mechanisms by which the modern food environment promotes 40 overeating. For example, processed foods are specifically associated with a number of deleterious 41 health outcomes including not only obesity and diabetes [4,5], but also depression [6], cardiovascular anxiety [35,36]. All participants provided written informed consent and study procedures were 105 approved by the Yale Human Investigations Committee. 106 To determine sample size for our primary goal of replicating macronutrient effects on food 107 reinforcement, a power analysis was conducted with G*Power 3 [37] using previously collected data 108 from the auction task [2]. We first calculated the effect sizes from the means and standard deviations 109 of the differences in WTP as d = 1.103 for combo versus carbohydrate foods and as d = 0.536 for 110 combo versus fat foods from the German study. For a conservative estimate using the smaller effect 111 size, we determined that a total sample of 30 participants would be needed for a two-tailed t-test of 112 two dependent means at an alpha of .05 to achieve a power of 0.80. We therefore aimed to include 113 n = 30 participants with HW (defined as BMI < 25 kg/m 2 ) as well as n = 30 with overweight/obesity 114 (OW/OB; BMI ≥ 25 kg/m 2 ) in order to test the role of BMI. Recruitment into the low and high BMI 115 groups was counterbalanced to minimize demographic differences between them. More specifically, 116 age and household income were stratified such that randomization was broken in order to include 117 participants of older age and with lower household income into the HW group. A total of 5 participants 118 were excluded from final analyses (see Section 2.5), leaving n = 30 in each BMI group.  120 MacroPics is a 36-item picture set of American snack foods [38]. Each food image portrays a 121 120-kcal portion that is classified into one of three categories by macronutrient content: (1) 122 predominantly carbohydrate, (2) predominantly fat, (3) or a combination of carbohydrate and fat 123 (combo) [38]. Food images in these carbohydrate, fat, and combo categories differ minimally in a 124 number of visual properties such as color and intensity, as well as in objective qualities (e.g., food 125 energy density, price, and sodium content) and subjective (e.g., perceived liking, familiarity, and 126 estimated energy content) characteristics using ratings provided by 128 participants with a range of 127 BMIs in a prior study [38]. Examples of the MacroPics stimuli are presented in Figure 1a. fixation cross, observed a food image, and bid for the snack from 0-5 USD. 132 The study consisted of two session days. The majority of sessions took place between 12pm 193 and 6pm and lasted ~1 hour. On day 1, participants were instructed to arrive at least 1-hour fasted.

194
To start, participants were trained to use the rating scales (Table 2). They were given definitions of a 195 "calorie" as "a unit of energy," of "energy content" as "the amount of energy people get from the foods 196 or drinks they consume," and of "energy density" as "the number of calories stored in the food per 197 unit volume." They were also given an example of the difference between energy content and energy 198 density and were allowed to ask questions for clarification. Participants then made subjective ratings 199 (Table 2)   On day 2, participants were instructed to arrive in a hungry state and at least 4-hours fasted.

204
After providing internal state ratings (Table 1), participants completed the auction task to bid up to 5 205 USD for each food item. Afterward, a computer bid from 0-5 USD was randomly generated for the 206 selected trial. Participants were given time to consume their snack if won; 15% of participants 207 obtained a snack by outbidding the computer on the selected trial. Lastly, participants rated what they 208 believed the grocery store price of each item would be ("estimated price" in Table 2). All participants 209 were compensated with 20 USD in addition to their earnings from the auction task (up to 5 USD) upon 210 completion of the second session.  Demographic and anthropometric characteristics of the final sample are described in Table 3.
235  238 We first tested whether macronutrient category and BMI group (< 25 or ≥ 25) influence WTP for 239 foods in our American sample. Guided by the analyses used in our earlier study of the effect of 240 macronutrient on bidding behavior (i.e., WTP) in German participants with HW [2], we ran a LMM 241 across all participants with bid/WTP as the outcome variable; BMI group, macronutrient category, the 242 interaction of BMI group × macronutrient category, actual energy density, estimated energy density, 243 estimated energy content, portion size, and liking as fixed effects; and participant as a random effect. 244 We found a significant main effect of macronutrient category (F(2,2149) = 5.733, p = 0.003) and a    (Figure 3a). However, we unexpectedly found that WTP was negatively associated with actual energy 302 density across all food items (r 2 = 0.209, p = 0.005; Figure 3b). This result was driven by a strong   317 We aimed to better understand the unexpected negative relationship between actual energy 318 density and WTP. As this effect was driven by a strong association among combo foods, we wanted 319 to determine if there was a third variable related to energy density in this macronutrient category (but 320 not in the fat or carbohydrate categories) that helped to account for our results. Using only combo 321 food items, we first correlated actual energy density with all remaining food characteristics and 322 subjective ratings (Table S5). Actual energy density was significantly related to volume, actual price, 323 and participant ratings of estimated price, expected satiety, and healthiness after correction for 324 multiple comparisons at this step (Table S5). Using all food items, we then performed an ANOVA to items. Indeed, actual energy density was related to actual price across all food items (r 2 = 0.417, p < 332 0.001) and in the combo category (r 2 = 0.859, p < 0.001), but not in the carbohydrate or fat categories 333 (Figure 4a). The associations between energy density and volume or expected satiety did not follow 334 this pattern (Table S6), indicating that they were unlikely to account for the strong negative association 335 between energy density and WTP specific to the combo category. We therefore reasoned that price could be a likely candidate to explain the strong negative 344 association observed between WTP and actual energy density among combo, but not carbohydrate 345 or fat, snacks. To test this, we first confirmed that actual price was positively associated with bidding 346 behavior across all food stimuli (r 2 = 0.352, p < 0.001) and in the combo category (r 2 = 0.748, p < 347 0.001; Figure 4b). We then performed a formal mediation analysis [48] and found that food price fully 348 mediates the relationship between WTP and actual energy density across all food items (Figure 5a) 349 and within the combo category (Figure 5b). After identifying the role of food price in bidding behavior, 350 we sought to ensure that the supra-additive effect of fat and carbohydrate on WTP that we observed 351 in participants with HW would remain significant after including actual price as a covariate. We tested 352 the same LMMs as before (see Section 3.2), but with food price added as a fixed effect. The   Table S7. We again observed a negative association between WTP (for the portion shown) and actual 378 energy density (r 2 = 0.143, p = 0.023; Figure 7a). We also found that food price was negatively related 379 to actual energy density (r 2 = 0.109, p = 0.049; Figure 7b) and positively correlated with WTP (r 2 = 380 0.389, p < 0.001; Figure 7c). Ultimately, our formal mediation analysis revealed that food price fully 381 mediates the relationship between WTP and actual energy density across these food items in varying 382 portions (Figure 7d). For full data visibility, we report fitted scatter plots comparing each food 383 characteristic and subjective rating with actual energy density ( Figure S1) and WTP ( Figure S2) for energy density, price, and WTP were each significant, we tested whether price mediates the negative 392 association between energy density and WTP. As predicted, the relationship between energy density 393 and WTP ( 3') was no longer significant when the indirect effect of price was accounted for in the 394 regression model, which itself remained significant ( 2'). Each data point depicts a single food item 395

Food price mediates the negative relationship between actual energy density and WTP
(averaged across all participants) and shading indicates 95% CI. * p < 0.05. 396

397
Our study had three main objectives. The first two were to test if we could replicate prior reports 398 demonstrating that participants with HW will pay more (1) for equally liked, familiar, and energetic 399 snacks containing fat and carbohydrate compared to those containing primarily fat or primarily 400 carbohydrate alone [2]; and (2) for foods with greater energy density [1,2]. The third objective was to 401 determine if WTP depends on BMI. In line with previous work [2], we found that bid amounts were 402 greatest for the combo foods and that calorie-for-calorie, combining fat and carbohydrate has a supra-403 additive effect on WTP in participants with HW. Food price did not impact these results; a key feature 404 of MacroPics is that snack items in the three macronutrient categories do not significantly differ in 405 cost [38]. We also replicated the supra-additive effect after directly accounting for price. These 406 findings collectively build upon existing evidence that fat and carbohydrate interact to potentiate food 407 reward [2]. In addition, we found that these effects were specific to the HW group, with macronutrient 408 content exerting no significant influence on WTP in individuals with OW/OB. Finally, unlike in the prior 409 Canadian and German samples, our American participants bid more for foods with lower energy 410 density, and this negative relationship was mediated by food price.  . However, the Canadian study did, and found that the positive association between 493 actual energy density and bidding behavior was still significant after including a covariate for price [1]. 494 We believe that this is evidence for population-level differences between Germany, Canada, and the 495 US that may encompass factors such as cost of living or disparities in the typical price of unprocessed 496 versus processed foods. Future work directly comparing WTP or another metric of food reward with 497 price across different cultures will be required to formally test these hypotheses.

517
In summary, we demonstrate that the combination of fat and carbohydrate has a supra-additive 518 effect on WTP among participants with HW, but not with OW/OB, as defined by BMI. We also highlight the importance of considering the role of the food environment in food choice. Finally, we speculate 520 that common US diets made up of cheap, processed foods may disrupt the capacity of unconscious 521 energy signals to drive food reinforcement.