Log in
Dietary patterns worldwide have shifted toward increased consumption of ultraprocessed foods (UPFs), which has been linked to higher disease burden. One proposed mechanism underlying both UPF consumption and metabolic disease is altered post-ingestive responses relative to nutritionally similar foods. Here we report the effects of food processing on post-ingestive metabolism and brain response in a randomized, crossover study involving 57 healthy-weight adults who consumed a nutritionally matched UPF or non-UPF meal. We show that despite being nutritionally similar, UPF meals evoke greater insulinaemic and energetic responses, with attenuated carbohydrate oxidation relative to non-UPF meals. Between-condition differences in peak carbohydrate oxidation are associated with striatal activation in response to food cues. We also find that although subjective food value does not differ between conditions, brain responses correlated with food valuation are positive for non-UPF but negative for UPF in the visual cortex and striatum. Overall, these findings suggest that food processing influences post-ingestive metabolic and neural responses through mechanisms beyond calories and macronutrients alone.
The modern food environment has undergone a shift towards widespread availability of UPFs1,2. Under the Nova classification system, UPFs are industrially manufactured products that undergo multiple physical and chemical transformations and typically contain added industrial ingredients, specifically additives not commonly used in home cooking3. Concerningly, higher UPF intake is associated with poorer health outcomes, including cancer, overweight and obesity, and metabolic dysfunction4,5. Despite these adverse health associations, UPFs comprise the majority of calories consumed in the United States and most of the food market1,2. Although the mechanistic basis for UPF-associated health effects is not fully established, the incorporation of manufacturing processes that alter nutritional availability has been proposed as a potential factor5.
Even with the same ingredients, micronutrient bioavailability differs after food processing6. For macronutrients, industrial extrusion divides starches to produce a spectrum of fractions, granules and crystalline structures, augmenting digestibility and resulting in distinct glycaemic and insulinaemic properties7. For example, compared with steel-cut, instant oats that have been cooked at higher temperatures to partially gelatinize the starch elicit a higher glycaemic index despite still only containing oats8. Varying the cellular microstructure of chickpeas, including disrupting cell walls as seen in many UPFs, produced larger postprandial glucose and insulin excursions and greater concentrations of glucose and maltose in the duodenum9. Although evidence that food processing could lead to altered metabolic function is accumulating, much of it comes from studies that did not explicitly test foods that differ on the Nova score while controlling for nutritional composition.
If UPFs elicit augmented physiological responses owing to altered nutritional availability relative to nutritionally matched foods that are not ultraprocessed, not only could this provide a potential mechanism for their effects on metabolic health but also may represent a key mechanism contributing to their overconsumption. Foods high in both fat and sugar, as seen in many UPFs, are valued more than foods high in fat or sugar alone and elicit responses in regions critical for reward valuation, such as the striatum10,11. Moreover, this difference in food value is thought to be learned in part through gut-to-brain signalling driven by post-ingestive metabolic responses, known as flavour-nutrient learning12,13,14. The magnitude of the post-ingestive metabolic response to sugar-sweetened beverages has been associated with changes in rated liking and food-cue reactivity in the striatum, evidence for flavour-nutrient learning in humans15,16. This nutrient information is thought to be relayed from the duodenum, where nutrients cross the brush-border membrane, to the brain17,18,19. Collectively, these findings support a framework in which UPFs may amplify gut-to-brain signalling by increasing nutrient exposure at the proximal small intestine, the site of greatest nutrient absorption19, and leading to changed food valuation and consumption12,13,20.
To test this hypothesis, we characterized the post-ingestive response to nutrient-matched meals that differed in level of processing according to the Nova classification system. Next, we tested whether these processing-related differences in post-ingestive metabolism are associated with neural responses to food pictures and subjective food value. These findings identify potential post-ingestive mechanisms for overconsumption of foods, including UPFs, that extend beyond caloric and macronutrient composition alone.
The full study comprised 57 participants; of these, 32 completed both metabolic sessions in a randomized crossover design and 52 completed the functional magnetic resonance imaging (fMRI) session (Extended Data Fig. 1). Every participant who completed the metabolic sessions also completed the fMRI session. The overall participant sample was 31.6% male, with a mean age of 26.21 ± 6.85 years and body mass index (BMI) of 22.75 ± 1.87 kg m−2 (Extended Data Table 1). Participants’ habitual energy intake on average comprised 54.4 ± 18.4% UPFs, similar to the national average of 55.0% (ref. 2). Additional participant characteristics are described in Extended Data Table 1. Exclusions and final analytic samples for each modality are summarized in the CONSORT diagram (Extended Data Fig. 1).
Acute metabolic effects of UPFs and non-UPFs
To investigate the effect of processing level on post-ingestive metabolic response to foods with matched nutrient content, 32 participants consumed, and were required to finish within 10 min, ~300 kcal nutritionally matched meals composed entirely of either UPF or non-UPF while undergoing 4 h (50 min baseline followed by a 3 h postprandial period) of whole-room indirect calorimetry (WRIC) and concomitant blood collection by intravenous catheter (Fig. 1a). The non-UPF and UPF test meals were closely matched on meal weight, energy, energy density, macronutrients, available carbohydrate, glycaemic index and load, total dietary fibre, sodium and water (all ≤1.6% deviation between meals; Fig. 1b). The meals were consumed on separate days in a randomized crossover design, allowing within-participant comparison of indirect calorimetry and blood-derived metabolic outcomes (see Methods for session timing).
a, Session flow and timing of blood draws with a picture of each meal as it was given to participants. b, Meals were composed of different food sources and were nutritionally matched. c, Blood glucose AUC did not differ between conditions (t(27.15) = −0.11; P = 0.92; 95% CI, −738.2 to 813.21); however, there was a significant time-by-diet interaction, such that non-UPF blood glucose level was higher than UPF at minute 20 (χ2(7) = 23.60; P = 0.001). d, Blood insulin AUC was greater in the UPF condition (t(28.67) = 5.48; P < 0.001; 95% CI, −2254.02 to −1032.03), and a significant time-by-diet interaction was observed such that insulin levels were higher between 40 and 120 min after consumption of the UPF compared with non-UPF meals (χ2(7) = 45.43; P < 0.001). e, Metabolic rate AUC was greater in the UPF condition (t(30.73) = 2.39; P = 0.02; 95% CI, −0.036 to −0.003). f, RER AUC was greater in the non-UPF condition (t(30.75) = −3.04; P = 0.01; 95% CI, 0.004–0.022). g,h, This difference is reflected in lower carbohydrate oxidation AUC after consumption of non-UPF compared with UPF meals (t(30.76) = −2.47; P = 0.02; 95% CI, 0.002–0.019) (g) and greater fat oxidation AUC in the UPF condition (t(30.84) = 3.26; P = 0.003; 95% CI, −0.01 to −0.002) (h). Each dot represents a participant; error bars, s.e.m. n = 32 participants for all analyses except blood insulin (n = 31 participants). *P < 0.05; all tests are two-tailed. Created in BioRender; Hutelin, Z. https://biorender.com/8mxvats (2026).
Data collection was performed in the morning, and participants were instructed to fast overnight. Time since last meal was recorded before data collection. Self-reported time since last meal did not differ between the non-UPF and UPF conditions (t(30.46) = −1.31; P = 0.20; 95% CI, −0.21 to 0.96). Total energy intake on the day preceding the metabolic session did not differ between conditions (t(30.90) = 1.30; P = 0.20, 95% CI, −440.2 to 97.55); macronutrient intake also did not differ between conditions (carbohydrate, P = 0.07; fat, P = 0.58; protein, P = 0.37).
During the 50 min baseline resting metabolic measures, wrist and ankle accelerometry indicated low movement and no statistically significant differences in movement between conditions (t(28.46) = 1.73; P = 0.10; 95% CI, −1771.61 to 150.57; Extended Data Fig. 2a). Baseline blood glucose (t(26.46) = 1.43; P = 0.17; 95% CI, −5.11 to 0.93) and insulin (t(28.91) = 0.04; P = 0.97; 95% CI, −2.27 to 2.09) did not differ between conditions (Extended Data Fig. 2b,c). WRIC baseline measures were calculated by averaging the 15 min before test-meal consumption (minutes −30 to −15). Baseline metabolic rate (t(30.10) = 0.06; P = 0.95; 95% CI, −0.02 to 0.02), respiratory exchange ratio (RER) (t(29.99) = 1.41; P = 0.17; 95% CI, −0.024 to 0.004), carbohydrate oxidation (t(30.49) = 1.40; P = 0.17; 95% CI, −0.023 to 0.004) and fat oxidation (t(29.95) = −1.22; P = 0.23; 95% CI, −0.003 to 0.01) did not differ between conditions (Extended Data Fig. 2d–g).
After the 50 min baseline measures, participants were instructed to consume all of the test meal in under 10 min. Ratings of the participants’ internal state (hunger, fullness and thirst) were collected at the start of the baseline measurement, immediately after meal consumption, 60 min post meal and at the end of the metabolic measurement (180 min). The UPF meal was consumed more slowly than the non-UPF meal by an average of 1.5 min (t(30.43) = 5.13; P < 0.001; 95% CI, −1.82 to −0.78; Extended Data Fig. 3a). Immediately after meal consumption, participants rated meal liking (t(30.13) = 0.61; P = 0.55; 95% CI, −15.07 to 8.12) and wanting (t(30.87) = 0.73; P = 0.47; 95% CI, −12.41 to 5.88), which were not statistically different between conditions (Extended Data Fig. 3b,c). There were no time-by-condition interactions for rated hunger (χ2(3) = 2.05; P = 0.56), fullness (χ2(3) = 6.19; P = 0.10) or thirst (χ2(3) = 0.76; P = 0.86; Extended Data Fig. 3d–f).
Blood samples were obtained intravenously at minutes −60 (baseline), 5, 20, 40, 60, 90, 120 and 180 after consuming the test meal (Fig. 1a). Consistent with the meals being matched on glycaemic index, blood glucose area under the curve (AUC) did not differ significantly between conditions (t(27.15) = −0.11; P = 0.92; 95% CI, −738.2 to 813.21; Fig. 1c). However, when we examined blood glucose trajectories over time, we observed a significant time-by-condition interaction (χ2(7) = 23.60; P = 0.001). Specifically, blood glucose rose more rapidly in the non-UPF condition, with higher concentrations at 20 min post meal (t(426.30) = −3.10; P = 0.02; 95% CI, −17.58 to −1.05, corrected; Fig. 1c), and both conditions peaked at 40 min. However, glucose remained elevated later in the UPF condition, trending higher at 120 and 180 min and failing to return to baseline in the same manner as the non-UPF condition. Despite comparable blood glucose AUC, blood insulin AUC differed significantly between conditions, with the UPF condition evoking a markedly greater insulin response (t(28.67) = 5.48; P < 0.001; 95% CI, −2254.02 to −1032.03; Fig. 1d). Insulin also showed a significant time-by-condition interaction (χ2(7) = 45.43; P < 0.001). Concentrations were similar at 5min and 20 min; however, insulin was greater in the UPF condition at minutes 40, 60, 90 and 120 (all P < 0.05), with both conditions returning to near-baseline values at 180 min (Fig. 1d). Together, these findings indicate that although peak blood glucose concentrations were similar between meals, the UPF condition evoked a greater insulinaemic response.
We next evaluated 3 h postprandial metabolic responses collected using WRIC. Despite both conditions having the same caloric load, post-ingestive metabolic rate AUC was greater after the UPF meal (t(30.73) = 2.39; P = 0.02; 95% CI, −0.036 to −0.003; Fig. 1e). Metabolic rate increased similarly in both conditions and peaked at ~40 min; however, after this peak, metabolic rate declined more in the non-UPF condition than in the UPF condition. By contrast, RER AUC was higher after the non-UPF meal (t(30.75) = −3.04; P = 0.01; 95% CI, 0.004–0.022; Fig. 1f). Notably, the meals were matched with respect to macronutrient composition, yet the RER response indicates differential substrate partitioning. Consistent with this observation, carbohydrate oxidation followed the same pattern as RER, with a greater response at minute 40 following the non-UPF meal compared to the UPF meal (t(30.76) = −2.47; P = 0.02; 95% CI, 0.002–0.019; Fig. 1g), whereas fat oxidation demonstrated the reciprocal pattern (t(30.84) = 3.26; P = 0.003; 95% CI, −0.01 to −0.002; Fig. 1h). To explore the effects of habitual dietary intake on metabolic outcomes, total grams of each macronutrient and %kcal from UPFs were entered as covariates and found to have no effects. Jointly, these findings suggest that degree of food processing alters post-ingestive metabolic rate dynamics, and compared to a nutritionally matched non-UPF meal, the UPF meal blunts the post-ingestive shift from fat to carbohydrate oxidation. The attenuated post-ingestive increase in carbohydrate oxidation observed in the UPF condition may also contribute to the increased insulin response.
Associations between metabolic and brain responses
Prior evidence suggests that post-ingestive metabolic signals are associated with, and are hypothesized to shape, the learned value of food through gut-to-brain signalling and repeated exposure12,13,20. We selected foods for the picture set that are generally familiar and frequently consumed across the population, as confirmed in our validation study21 and in our current population (Fig. 3). We reasoned that the metabolic effects of these foods would be well-learned through exposures before participation in this research study, as prior studies using food picture sets and fMRI paradigms have demonstrated10,11,22. Therefore, we sought to examine whether post-ingestive metabolic responses were associated with food-cue reactivity, a measure of brain response to food pictures that relies on the learned subjective value of food. To test this hypothesis, participants included in the analysis of metabolic variables also completed a Becker–DeGroot–Marschak (BDM) auction task during fMRI on a separate day; notably, the non-UPF and UPF meals used in the metabolic sessions were derived from a ~300 kcal subset of the foods used in the BDM task. To examine the relationship between metabolic responses and food-cue reactivity, we focused on the picture-viewing period by bringing the first-level non-UPF > UPF contrast for the 5 s picture-viewing epoch forward to the group-level analysis. We also calculated the difference in peak metabolic change between conditions (non-UPF − UPF). Prior work in both humans and rodents has shown that peak metabolic responses to sugar-sweetened beverages are associated with striatal activation15,16,23; therefore, we examined both whole-brain-corrected and small-volume-corrected results within an a priori striatal mask (see Methods for details; Supplementary Fig. 2).
Immediately before and after the fMRI session, participants rated hunger, fullness and thirst (Extended Data Fig. 4). Pre-scan hunger ratings were relatively neutral (M = 56.46) and increased modestly to a post-scan mean of 70.00 (t(51) = 4.69; P < 0.001; 95% CI, −19.34 to −7.74). Thirst ratings showed a similar pattern, rising from a pre-scan mean of 53.23 to 63.96 post-scan (t(51) = 4.98; P < 0.001; 95% CI, −15.06 to −6.4). As expected, fullness decreased over the session (t(51) = − 4.70; P < 0.001; 95% CI, 5.64–14.05).
Regressing the between-condition change in peak carbohydrate oxidation on participant-level non-UPF > UPF contrast estimates revealed a whole-brain family-wise error (FWE)-corrected association in the left superior temporal gyrus ((−44, −24, −8); t = 7.33; PFWE = 0.013; Fig. 2b and Extended Data Table 2). Within our a priori striatal mask, small volume correction (SVC) identified significant associations in the right caudate ((6, 10, 4); t = 6.11; SVC PFWE = 0.007; Fig. 2c and Extended Data Table 2), and the left ventral striatum ((−8, 6, −10); t = 5.20; SVC PFWE = 0.045; Fig. 2d and Extended Data Table 2). Specifically, a larger increase in peak carbohydrate oxidation for non-UPF relative to UPF was negatively associated with the neural response (non-UPF > UPF) (Fig. 2b–d and Extended Data Table 2). As expected, given the close correspondence between these measures, peak RER exhibited a similar relationship in the left superior temporal gyrus ((−44, −24, −8); t = 7.63; PFWE = 0.007) and SVC right caudate ((4, 8, 2); t = 5.77; SVC PFWE = 0.014; Fig. 2e and Extended Data Table 2). In addition, peak RER was also associated with a second locus of activation in the right caudate ((22, −12, 24); t = 5.48; SVC PFWE = 0.026; Fig. 2f and Extended Data Table 2). Between-condition changes in peak metabolic rate, fat oxidation, blood glucose and blood insulin were not significantly associated with neural responses. Together, these findings are consistent with the possibility that a food processing-related shift in post-ingestive carbohydrate oxidation may be associated with food-cue reactivity, learned through exposure to these foods, extending previous findings from beverages to whole foods and providing additional evidence consistent with pre-clinical findings that carbohydrate metabolism is important for neural response15,16,23.
a, Participants saw 14 non-UPF and 14 UPF items in a random order with a jittered intertrial interval (ITI) before rating their willingness to pay for that item. b, A higher response in the superior temporal gyrus to non-UPF pictures relative to UPF was associated with less carbohydrate oxidation to UPF ((−44, −24, −8); t = 7.33; PFWE = 0.013). c,d, After SVC of an a priori striatal region of interest (ROI), associations in the same direction were also observed in caudate ((6, 10, 4); t = 6.11; SVC PFWE = 0.007) (c) and ventral striatum ((−8, 6, −10); t = 5.20; SVC PFWE = 0.045) (d) for carbohydrate oxidation, and in e,f, caudate for respiratory exchange ratio ((4, 8, 2); t = 5.77; SVC PFWE = 0.014; (e) (22, −12, 24); t = 5.48; SVC PFWE = 0.026 (f)). Arrows point to the significant FWE-corrected peak displayed within a cluster threshold of Puncorrected < 0.001. n = 29 participants. All r2, P < 0.0001; error bands, 95% CI. PE, parameter estimate. Created in BioRender; Hutelin, Z. https://biorender.com/msvyhzj (2026).
Hunger, fullness and thirst changed significantly over the session; therefore, we explored additional models of all fMRI findings with these factors entered as covariates and found the pattern of results unchanged (Supplementary Tables 1–4). The results also remained similar after adjustment for BMI (Supplementary Table 5). Age and sex were included as covariates in all models.
Willingness to pay and brain correlates of subjective value
We next evaluated processing-related differences in subjective food value in the full fMRI cohort (n = 52). First, to identify attributes of the food picture set that could influence valuation, participants rated the food image set on liking, frequency of consumption, familiarity, expected satiety, perceived healthiness, estimated energy density and estimated calories before the fMRI scan, and provided estimated price ratings after the scan so as to not influence bidding behaviour during the scan. Notably, the food image set was designed and validated on an independent out-of-sample cohort to be matched on these rated attributes, nutritional properties and visual properties of the picture stimuli21 (Supplementary Fig. 1). All attributes except perceived healthiness were rated similarly between non-UPF and UPF foods21.
Consistent with the prior validation21, ratings did not significantly differ for estimated calories (t(26) = 0.99; P = 0.33; 95% CI, −51.55 to 18.08; Fig. 3a), estimated energy density (t(26) = −1.21; P = 0.24; 95% CI, −23.28 to 90.03; Fig. 3b), estimated price (t(26) = −0.35; P = 0.73; 95% CI, −0.43 to 0.61; Fig. 3c), liking (t(26) = −0.46; P = 0.65; 95% CI, −6.29 to 9.89; Fig. 3d), expected satiety (t(26) = −1.99; P = 0.06; 95% CI, −0.37 to 21.28; Fig. 3e) or familiarity (t(26) = −0.09; P = 0.93; 95% CI, −5.31 to 5.79; Fig. 3f). As in the previous study, participants rated non-UPF as healthier than UPF (t(26) = −5.28; P < 0.001; 95% CI, 17.1–38.93; Fig. 3g). Unexpectedly, frequency of consumption also differed by processing category in this sample (t(26) = −4.03; P < 0.001; 95% CI, 5.55–17.11; Fig. 3h), with non-UPF rated as being consumed more often than UPF. Therefore, we adjusted for healthiness and frequency of consumption by adding it as a trial-by-trial parametric modulator or covariate in subsequent models comparing UPF to non-UPF.
a–f, Participants rated estimated calories (t(26) = 0.99; P = 0.33; 95% CI, −51.55 to 18.08) (a), estimated energy density (t(26) = −1.21; P = 0.24; 95% CI, −23.28 to 90.03) (b), estimated price (t(26) = −0.35; P = 0.73; 95% CI, −0.43 to 0.61) (c), liking (t(26) = −0.46; P = 0.65; 95% CI, −6.29 to 9.89) (d), expected satiety (t(26) = −1.99; P = 0.06; 95% CI, −0.37 to 21.28) (e) and familiarity (t(26) = −0.09; P = 0.93; 95% CI, −5.31 to 5.79) (f) similarly between conditions. g,h, In this sample, participants’ ratings of healthiness (t(26) = −5.28; P < 0.001; 95% CI, 17.1–38.93) (g) and frequency of consumption (t(26) = −4.03; P < 0.001; 95% CI, 5.55–17.11) (h) differed between conditions. Each dot represents the mean for that food (28) from n = 52 participants; error bars, s.e.m. *P < 0.05; all tests are two-tailed.
To quantify subjective food value, participants completed a BDM auction task concomitant with fMRI. During the task, participants viewed all 28 food images and bid their willingness to pay (WTP; $0–$5) for each item (see Methods and Fig. 2a). The measure of WTP, while adjusting for perceived healthiness and frequency of consumption in the model, did not differ between the non-UPF and UPF conditions (t(32.41) = 0.96; P = 0.34; Fig. 4a and Supplementary Table 6). When sex was entered as a covariate in the model, the results were unchanged.
a, Participants bid similar amounts for non-UPF and UPF pictures after controlling for frequency of consumption and healthiness (t(32.41) = 0.96; P = 0.34). b, Overall value, as measured by willingness to pay for each food item, was associated with activity in brain areas consistent with prior work using the BDM auction task. c,d, In a whole-brain analysis, value for non-UPF items was positively associated with activity in the fusiform gyrus ((−32, −52, −14); t = 7.89; PFWE < 0.00010) (c) and lingual gyrus ((−22, −84, −14); t = 7.32; PFWE < 0.0001) (d), whereas value for UPF items was negatively associated with activity in these same regions. e,f, In an a priori ROI analysis, value was again oppositely associated with non-UPF versus UPF items in the putamen ((32, −4, 12); t = 4.62; SVC PFWE = 0.017) (e) and caudate ((14, 16, 4); t = 4.34; SVC PFWE = 0.043) (f). Arrows point to the significant FWE-corrected peak displayed within a cluster threshold of Puncorrected < 0.001. Each dot represents a participant; error bars, s.e.m. n = 52 participants. *P < 0.001; tests are two-sided.
Although WTP did not differ by processing category, we next tested whether processing level could modulate the neural representation of subjective value by entering WTP as a parametric modulator. Consistent with prior work, measures that differed by condition were included as additional parametric modulators10,22; in our case, these were frequency of consumption and perceived healthiness, as the photo set did not differ by condition in nutritional or visual property measures.
First, as a check of our procedures, we examined the WTP model regardless of condition. Consistent with meta-analysis findings from BDM auction tasks24, WTP was associated with activation in the left orbitofrontal cortex ((−28, 24, −34); t = 6.27; PFWE = 0.001; (−24, 30, −20); t = 5.78; PFWE = 0.005; Fig. 4b) and the left pregenual anterior cingulate cortex ((−14, 52, 4); t = 5.40; PFWE = 0.021; (−6, 52, 2); t = 5.20; PFWE = 0.043). WTP-related activity was also observed in regions implicated in higher-order visual processing, including the right fusiform gyrus ((46, −62, −14); t = 5.64; PFWE = 0.008) and the right lateral occipital cortex ((44, −78, −6); t = 5.50; PFWE = 0.014). A complete list of activations is provided in Extended Data Table 2.
Next, we tested whether activity was associated differently across UPF versus non-UPF conditions, while adjusting for frequency of consumption and perceived healthiness. Peak differences were observed in the left occipitotemporal gyri, specifically the left fusiform gyrus ((−32, −52, −14); t = 7.89; PFWE < 0.0001; Fig. 4c and Extended Data Table 2) and two peaks in the left lingual gyrus ((−22, −84, −14); t = 7.32; PFWE < 0.0001; (−24, −74, −8); t = 6.69; PFWE = 0.0001; Fig. 4d and Extended Data Table 2). Parameter estimates of the left fusiform gyrus and the left lingual gyrus indicated opposing responses, with positive associations with value in the non-UPF condition and negative in the UPF condition (Fig. 4c,d and Extended Data Table 2). We next examined effects within our single a priori striatal mask using an SVC. This analysis revealed significant activation in the right putamen ((32, −4, 12); t = 4.62; SVC PFWE = 0.017; Fig. 4e) and in the right caudate ((14, 16, 4); t = 4.34; SVC PFWE = 0.043; Fig. 4f and Extended Data Table 2). Both regions showed a similar pattern to the activation in the occipitotemporal gyri, with positive parameter estimates—and therefore a positive correlation between brain activity and bid value—for non-UPF cues and negative parameter estimates—indicating a negative correlation—for UPF cues (Fig. 4e,f). Interestingly, the contrast UPF > non-UPF did not yield any significant voxels at either the whole-brain or SVC level. Collectively, WTP did not differ between non-UPF and UPF conditions, even after controlling for frequency of consumption and perceived healthiness (Supplementary Table 6); however, neural differences were still observed. This could reflect differences at the individual level in how subjective value is encoded in these foods despite no overall group difference in bidding behaviour.
Although our stimuli were matched on nine visual properties, we further explored a model accounting for trial-by-trial variation in visual properties (complexity and object size) of our stimuli and found they did not alter the pattern of results (Supplementary Table 5).
Here, we provide evidence that UPFs may influence metabolic and learned neural responses differently from nutrient-matched non-UPFs. First, we demonstrated that even when meals were matched across multiple nutritional factors, the degree of food processing classified by Nova significantly altered post-ingestive metabolic responses. Next, we showed that these metabolic differences were associated with neural responses to non-UPF versus UPF food cues. Finally, although behavioural indices of subjective value did not differ between non-UPF and UPF, the neural representation of subjective value did, with condition-dependent value signals observed in higher-order visual and striatal regions. Taken together, these findings provide experimental evidence that Nova processing level may influence post-ingestive metabolic responses and, in turn, the neural response to food cues in ways not explained by macronutrient composition alone, supporting the hypothesis that altered nutritional availability is a candidate mechanism linking UPF consumption to metabolic health and overeating5,13,20.
Food processing can modify post-ingestive physiology, often by increasing carbohydrate availability and elevating glycaemic index, thereby producing larger postprandial blood glucose and insulin responses7,8. Prior work has reported differences in metabolic rate and respiratory quotient following meals differing in degree of processing; however, these effects are difficult to attribute directly to processing, as the test meals also differed in total energy and macronutrient composition25. The present findings address this limitation directly: the test meals were matched on energy, macronutrients and glycaemic index and yielded comparable blood glucose AUC, while the UPF condition elicited larger insulinaemic and energetic responses with attenuated carbohydrate oxidation. Notably, homogenized meals can produce similar blood glucose responses to their intact counterparts while altering other postprandial metabolic responses26, paralleling the pattern observed across our conditions. Importantly, many UPF products, despite being eaten as solids, are formulated to transition quickly into a lubricated, semi-fluid bolus. For brittle snacks and cereals, processing promotes extensive first-bite fracturing and rapid saliva incorporation, while added lipids increase lubrication, together facilitating faster, more homogeneous bolus formation27. These manufacturing techniques are primarily intended to optimize the oral sensory experience; however, they may unintendedly also influence digestion kinetics and downstream metabolic responses, such as the observed higher insulinaemic response but similar peak blood glucose levels, a pattern consistent with a relative reduction in insulin sensitivity. We also observe a potential delayed shift in substrate oxidation, which is also observed in insulin resistance and type 2 diabetes28,29. Although this study was not designed to assess chronic outcomes, the present data show that nutrient-matched non-UPF and UPF meals elicit acutely different post-ingestive glycaemic and insulinaemic responses and alter substrate oxidation, suggesting a plausible physiological mechanism through which repeated exposures could contribute to the metabolic dysfunction reported in epidemiological studies4,5.
These metabolic differences may also shape neural responses to food cues. Post-ingestive metabolic signals are a key reinforcing stimulus linking orosensory food cues to their nutritional value12,14, and these signals are thought to be integrated in the striatum (both dorsal and ventral) and guide eating behaviour13,20. This gut-to-brain signalling is thought to be macronutrient specific, with fat signalling through PPARα (peroxisome proliferator-activated receptor alpha) receptors on the vagus nerve19. Although carbohydrate sensing has multiple proposed mechanisms, including activation of the sodium-glucose transporter14, vagal and spinal afferent-dependent pathways17,18 and an unknown carbohydrate oxidation sensor23, insulin secretion is not macronutrient specific and can be evoked by carbohydrates and proteins. In animal models, there is some evidence that it influences food preference30, and in humans, intranasal inulin increases response to sweet tastes31. Although we observed differences in multiple metabolic measures, our data provide additional evidence for the importance of carbohydrate oxidation in conveying nutrient information to the brain.
Although we replicate previous findings that carbohydrate oxidation is associated with ventral striatal activation, the direction of the relationship differs. Notably, the other studies reported a positive association between post-ingestive metabolic responses and striatal activation15,16. This discrepancy may reflect a key methodological difference: prior studies measured neural responses during oral flavour receipt, whereas the present study assessed responses to visual food cues, which have been shown to evoke distinct and sometimes opposite neural responses32. However, we do replicate the finding that peripheral post-ingestive metabolic signals are associated with striatal activation, although previous studies15,16, including the present one, differ in which specific metabolic signal drives this relationship. This heterogeneity may reflect the fact that these multiple metabolic measures could index a shared underlying metabolic process. Overall, we provide evidence for the framework in which differences in nutrient availability can guide eating through learning12,13,20.
In addition to striatal regions, processing-related individual differences in carbohydrate oxidation were also associated with responses in the superior temporal gyrus (STG). Although primarily known for its role in auditory and language processing, the STG is also responsive to food cues33. More work is needed to determine whether the STG is involved in food-specific multisensory integration, including interoceptive metabolic signals.
The subjective value of food as measured by our auction task did not differ between conditions, which is consistent with the meals being matched with respect to energy density and macronutrient composition, attributes that have previously been shown to drive differences in subjective food valuation10,11,22, but was contrary to our hypothesis. Although we did not find large differences in subjective food value across processing, differences in the neural encoding of subjective food value were observed. Notably, prior work has shown that the subjective value of foods high in both fat and sugar, relative to foods high in either nutrient alone, is associated with different activity in the putamen and caudate10, and although our conditions were matched in macronutrient composition, we observed a similar pattern of striatal activation, suggesting that processing level, and potentially nutrient availability, may engage these regions through mechanisms independent of macronutrient content.
Interestingly, we observed condition differences in subjective value associations in the occipitotemporal gyrus, specifically the fusiform and lingual gyri. Although the fusiform gyrus is best known for its role in face processing within the ventral visual stream, a distinct region identified for food-specific visual processing has also been found and termed the fusiform food area34. Response in the fusiform food area is modulated by circulating glucose, insulin and ghrelin35,36. The lingual gyrus has been reported in studies of food cues37, and the visual association cortex in mice can be sensitive to reward associations38.
Limitations and future directions
There are a few limitations that should be acknowledged in this work. First, we assembled the test meals from a diverse set of foods to ensure that metabolic responses were not driven by any single item. Although total protein was matched, the sources of protein in each meal differed and could have influenced insulin response39. Second, inherent in the Nova definition are additives and alterations in the food matrix; the UPF meal contained more additives, and we did not explicitly measure physical and chemical structures of foods comprising the food matrix, beyond total fibre. Third, responses may also differ to larger caloric loads. Additional tests of these factors will be important to establish the generalizability of these findings. Fourth, prior work indicates that UPFs are consumed more rapidly under ad libitum conditions25,40. By contrast, in the present study, the meals were isocaloric (∼300 kcal) and fully consumed, and the non-UPF meal was eaten slightly faster than the UPF meal. However, the modest 1.5 min difference in eating time is unlikely to produce a physiologically meaningful effect41. Finally, WTP did not differ between conditions. Local grocery prices changed rapidly during the time of data collection; future studies should explicitly collect fluctuating food prices.
In summary, we provide evidence that non-UPF and UPF meals matched across a variety of nutritional factors elicit distinct post-ingestive metabolic responses, specifically amplified insulinaemic and energetic responses with attenuated carbohydrate oxidation after UPF consumption. These processing-related metabolic differences were also associated with neural metrics of food-cue reactivity within the striatum across foods that differ in degree of processing. Collectively, these findings support the concept that food processing could influence physiology and brain function through mechanisms extending beyond calories and macronutrient composition alone, while identifying altered nutritional availability from altered physical structure as a plausible contributor to both their overconsumption and effects on metabolic health.
All study protocols were approved by the Virginia Tech Institutional Review Board (21-1052) and registered at ClinicalTrials.gov (identifier: NCT06017986). The study consisted of a pair of metabolic sessions, a behavioural session and an fMRI session. The two metabolic sessions were held on separate days in a randomized crossover design to test whether degree of processing alters postprandial metabolic response independent of nutrient content. During each session, participants resided in a whole-room indirect calorimeter for 4 h and consumed a nutritionally matched non-UPF or UPF meal. A 50 min baseline measurement and blood draw were collected before meal consumption, followed by 3 h of postprandial metabolic measurements and serial blood draws to characterize postprandial metabolic responses to UPF versus non-UPF meals. As these sessions had a high participant burden, we completed them in a subset (n = 32) of the total (n = 57) sample.
In the behavioural session, anthropometric measures were collected, and participants were trained on the rating scales before rating 28 food images (14 non-UPF and 14 UPF). In the fMRI session, participants completed four runs of a BDM auction task during fMRI acquisition to quantify food value and test whether valuation-related brain responses differed as a function of processing level for the same 28 images presented in the behavioural session.
Picture stimuli
For this study, a picture set of 28 foods that are commonly consumed in the United States but systematically differ in degree of processing was used. As previously reported, using the Nova classification system5, the picture set was divided into two groups differing in the degree of processing: 14 non-UPFs (Nova 1–3) and 14 UPFs (Nova 4). To verify that the food picture set was Nova-scored correctly, 67 raters, including 17 registered dietitians, Nova-scored the picture set and showed strong inter-rater reliability, as reported in a prior publication21. These foods were additionally matched on 26 characteristics consisting of nine visual properties (for example, pixel colour, object size and brightness), 11 nutritional characteristics (for example, calories, energy density and macronutrients) and six perceptual properties (for example, perceived liking, estimated energy density, estimated cost and expected satiation)21.
Meal stimuli
The non-ultraprocessed and ultraprocessed meals used in the metabolic sessions were derived from a 300 kcal subset of the 28 foods used in the picture set21. The nutritional information for all 28 foods was derived from the Nutrition Data System for Research (version 2022) and then entered into an algorithmic process of testing combinations of different food pairings with different portion sizes to identify the most nutritionally matched set of non-UPF and UPF meals. Through this process, the final non-UPF and UPF meals were matched on weight, energy, energy density, total carbohydrates, total fats, total proteins, available carbohydrates, glycaemic index, glycaemic load, total dietary fibre, sodium and water, all with <1.6% error between meals. To match weight, energy, energy density and water, a 12 g serving of water was added to the UPF meal. Nutritional information and a list of the foods are available in Fig. 1b.
Recruitment and screening
Study participants were recruited from the university and surrounding community in Roanoke, Virginia, through social media advertisements, campus advertisements and flyers. Interested individuals filled out a general online screening form on Ripple Science software from June 2023 to December 2025 to provide self-report information for the determination of study eligibility. To be eligible, individuals had to be between 18 and 45 years old with a BMI between 18.5 and 25 (calculated from reported height and weight). Exclusion criteria included having dietary restrictions (food allergies, vegetarian, keto and so on), medical conditions (including metabolic, neurologic or psychiatric disorders) or medications used to treat these disorders; being pregnant; currently using inhaled nicotine or recreational drugs other than periodic cannabis use; having impaired taste or smell; having contraindications for MRI; or having vision not sufficiently correctable to see the MRI screen. Eligible individuals were invited for in-person screening and consent. Data collection started in June 2023 and ended in December 2025, and the metabolic sessions spanned June 2023 to March 2025. Participants were compensated for their time.
Statistics and reproducibility
Power analysis (behaviour)
The overall sample size was estimated based on behavioural data from a previous publication10. Sample size was calculated in R42 based on a repeated-measures ANOVA model, and detecting differences in WTP (bid amount in USD) among the degree of food processing condition groups43. With a total sample size of 52 participants, pairwise comparisons among the two groups yielded 80% power to detect an effect size of Cohen’s f = 0.40 with a type I error rate of 5%. Based on this total sample size of 52, power was estimated for a generalized linear mixed-effects model, regressing the primary outcome (WTP) on food processing, sex, their two-way interaction and controlling for potential confounders (hunger, energy expenditure, blood glucose, blood insulin). This was accomplished using 5,000 iterations of a Monte Carlo simulation; expected assumptions were based on published bidding data10. In these simulations, the model parameters were estimated by restricted maximum likelihood, such that the specified model was compared to the null model using the likelihood ratio test, yielding 99% power to detect a statistically significant effect at the 5% level of significance. In summary, a sample size of 52 participants provides sufficient power to detect a main effect difference in WTP by food processing group.
Power analysis (associations between metabolism and fMRI)
An ad hoc power analysis was conducted to ensure that power was achieved for the subset of participants in the correlations between metabolic and brain responses. The most relevant published data16 reported a correlation of r = 0.83, substantially exceeding the effect sizes used to power the study. We performed a power analysis based on a test of Pearson’s correlation with that value. To achieve 95% power, 16 participants were needed at a significance level of 0.0001.
Power analysis (metabolism)
A separate power analysis was conducted to evaluate the sample size needed to detect a difference in metabolic response to the meals consumed in the whole-room indirect calorimeter. There was over 80% power to detect a difference, based on a paired t-test, in the AUC for energy expenditure, assuming a Cohen’s d of 0.57 (medium size) and a sample size of 27. Data collection and analysis were not performed blind to the conditions of the experiment.
Participant inclusion and eligibility
Study-wide participant inclusion and exclusion are presented in a CONSORT diagram (Extended Data Fig. 1). A total of 61 participants completed in-lab eligibility screening. Four participants were excluded before enrollment (two were ineligible for the study payment system and two were not interested). The remaining 57 participants enrolled, and all completed the fMRI session. Five participants were subsequently excluded from the fMRI analyses: two because of excessive head motion (>2 mm in more than two of the four runs), two because of task performance (>25% of bids <$0.25 or missing) and one because of equipment failure. Of the 52 participants included in the fMRI analyses, four had one run excluded, two were excluded because of excessive head motion and two were excluded because of equipment failure. Two additional participants were missing two runs; one because of excessive head motion and one because of equipment malfunction.
Of the 57 enrolled participants, 32 completed the metabolic sessions. All participants completed both the non-UPF and UPF conditions; however, one participant experienced a vasovagal response after the 60 min blood draw, resulting in early termination of that session. Owing to IV patency issues, glucose samples could not be obtained for one condition in five participants. Insulin samples could not be obtained for one condition in three participants and for both conditions in one participant. All 32 participants who completed the chamber sessions also underwent MRI scanning: two were excluded because of excessive head motion (>2 mm in more than two of the four runs) and one because of equipment failure.
Missing data from metabolic sessions
There was minimal missing data across the study. Data were analysed using linear mixed-effects models, which allow for unequal observations per participant (Supplementary Materials).
Anthropometric
Body weight was measured with an electronic scale (Health O Meter ProPlus digital scale), and height was measured with a wall-mounted stadiometer. Weight was recorded to the nearest 0.1 kg and height to the nearest 1 cm. BMI was calculated as kg m−2. Waist and hip circumference were measured using a Gulick tape measure following World Health Organization protocols for the calculation of waist-to-hip ratio.
Habitual diet
Information on habitual diet measures can be found in the Supplementary Materials.
Metabolic session
Participants completed two separate metabolic sessions on different days in a randomized crossover design. To reduce participant burden, we did not require a specific session spacing. Seven participants had sessions within 3 days, 14 participants had sessions between 4 days and 2 weeks, seven participants had sessions between 15 days and 4 weeks and four participants had sessions with a greater than 4-week gap. For each session, participants were instructed to arrive after an overnight fast and to refrain from strenuous exercise on the preceding day. Sessions were scheduled to begin in the morning, after participants’ habitual wake-up time. A small-volume whole-room indirect calorimeter (MEI Research) was used to measure participants’ metabolic rate, RER, carbohydrate oxidation and fat oxidation for 4 h (ref. 44). More details on chamber configuration can be found in the Supplementary Materials.
After arrival and entry into the whole-room indirect calorimeter, participants completed a 50 min baseline measurement. The average of the 15 min before test meal consumption (minutes −30 to −15) was used as the baseline to ensure a stable reference for calculating change from baseline. At minute 50, without disrupting the whole-room indirect calorimeter, participants uncovered and then consumed the test meal in under 10 min. After meal consumption, the postprandial metabolic response was measured for 3 h. Concomitant with the indirect calorimeter measurement, blood draws were performed at baseline (60 min before meal consumption) and after meal consumption at minutes 5, 20, 40, 60, 90, 120 and 180. Blood draws were performed through custom ports so as not to disturb the WRIC measurements. For blood sample analyses, glucose concentrations were measured in duplicate using a point-of-care system (HemoCue Glucose 201 System), insulin concentrations were quantified using an enzyme-linked immunosorbent assay (ALPCO) and haemoglobin A1c was determined with a point-of-care analyser (Afinion HbA1c, Abbott Laboratories). Throughout the metabolic session, participants were also fitted with accelerometers (ActiGraph wGT3X-BT) on their right wrist and left ankle to measure movement. Internal states (hunger, fullness and thirst) were assessed before and after the WRIC measurements, as well as immediately after meal consumption and 60 min after the meal. During the post-meal measurement, participants also rated the meal for perceived liking using the labelled hedonic scale, and wanting on a visual analogue scale.
Behavioural session
All ratings were collected using PsychoPy3 (v.2023.1.0)45. Before providing liking ratings, participants completed training on proper use of the labelled hedonic scale46 and practised rating a variety of remembered sensations. Following this training, participants rated an example food image (not included in the 28-food picture set) to practise the full set of ratings (liking, frequency of consumption, familiarity, expected satiety, perceived healthiness, estimated energy density and estimated calories) before assessing all 28 foods (Supplementary Materials).
At the end of the behavioural session, all participants practised a run of the fMRI task (described below) in a mock MRI simulator (Psychology Science Tools) equipped with a mock 64-channel head coil.
fMRI session
Participants were instructed to arrive neither hungry nor full, having fasted for 3 h, and to refrain from strenuous exercise the day before. In relation to the metabolic sessions, six participants completed the fMRI session as their last session, seven participants completed the fMRI session between their metabolic sessions and 16 participants completed the fMRI session before their metabolic sessions. During this session, participants completed four runs of a BDM auction task to assess food value47 while undergoing fMRI scanning24. Additional details can be found in Supplementary Materials.
Immediately before and after scanning, participants rated their hunger, fullness and thirst. Ratings were made on visual analogue scales anchored with ‘not hungry at all’ to ‘very hungry’, ‘not full at all’ to ‘very full’ and ‘not thirsty at all’ to ‘very thirsty’. After completing the fMRI scan, participants rated the estimated price of each of the 28 foods using the same rating scale as in the scanner. This assessment was administered post-scan to avoid biasing bidding behaviour during the task.
MRI acquisition parameters
All MRI data were acquired with a 3 Tesla Siemens MAGNETOM Prisma scanner using a 64-channel head coil. Acquisition parameters for the functional echo-planar images were TE = 34 ms, echo spacing = 0.66 ms, TR = 1.5 s, flip angle = 70, voxel size = 2 × 2 × 2 mm, number of slices = 72 and multiband acceleration factor = 4. To allow distortion correction, sets of identical images were acquired with reverse-phase (posterior-to-anterior and anterior-to-posterior) encoding polarity. A high-quality T1-weighted anatomical image (TE = 2.32 ms, TR = 2.3 s, flip angle = 8, slices = 192, field of view = 240 mm, voxel size = 0.9 × 0.9 × 0.9 mm) was acquired for registration of functional images.
Data analysis
Metabolic analysis
All statistical analyses on metabolic data were conducted with R version 4.4.1 (2024-06-14). Internal state ratings before and after the metabolic measurements, as well as immediately after meal consumption and 60 min after the meal, were compared using linear mixed-effects models with internal state (hunger, fullness or thirst) as the outcome, time as a fixed effect and participant as a random intercept. In addition, post-meal liking and wanting ratings were compared across the non-UPF and UPF conditions using linear mixed-effects models. In these models, condition was included as a fixed effect and random intercept for participant to account for within-subject correlations. Where multiple comparisons were performed, appropriate corrections were applied.
Baseline measures were compared across conditions to check for any systematic differences. Baseline measures included resting metabolic rate, resting RER, resting carbohydrate oxidation, resting fat oxidation, baseline blood glucose, baseline blood insulin and pre-meal movement AUC, calculated as the summed ActiGraph arm and leg movement signals. These outcomes were compared across the non-UPF and UPF conditions using linear mixed-effects models. In these models, condition was included as a fixed effect and a random intercept for participant to account for within-subject correlations.
Blood outcomes of interest included AUC of change from baseline in insulin and glucose levels. Each of these outcomes was compared across the non-UPF and UPF conditions using linear mixed-effects models. In these models, condition was included as a fixed effect and a random intercept for participant to account for within-subject correlations. Additionally, each insulin and glucose measure at each of the eight time points was examined using a linear mixed-effects model with fixed effects of time, condition and a time-by-condition interaction. A random intercept for participant was included to account for within-subject correlation. Two-sided post hoc Wald-type tests were conducted to examine differences between time points. Holm–Bonferroni correction was used to control the FWE rate.
WRIC outcomes of interest included AUC of change from baseline in metabolic rate, RER, carbohydrate oxidation and fat oxidation. Each of these four outcomes was compared across the UPF and non-UPF conditions using linear mixed-effects models. In these models, condition was included as a fixed effect and a random intercept for participant to account for within-subject correlations.
For each model, residuals were visualized, and assumptions of normality and homoscedasticity were determined to be met.
Behavioural analysis
All statistical analyses on behavioural data were conducted with R version 4.4.1 (2024-06-14). Where multiple comparisons were performed, appropriate corrections were applied. All tests were two-sided. To test whether the average estimated price, liking, perceived healthiness, familiarity, perceived satiety and frequency of consumption were different across non-UPF and UPF foods, linear regression models were used. In these models, the dependent variable was the mean rating for each food (averaged across participants) for each outcome, and food type (non-UPF vs UPF) was the independent variable10,11,22. To examine differences in WTP, linear mixed-effects models were performed with WTP as the outcome and food type as the fixed effect. Perceived healthiness and frequency ratings were entered as covariates, as they differed between conditions. Random intercepts for food item and participant, as well as a random slope for food type within participant, were included to account for any within-subject or within-food correlations. This covariance structure was selected using Akaike’s information criteria. For each model, residuals were visualized, and assumptions of normality and homoscedasticity were determined to be met.
fMRI analyses
Internal state ratings pre-MRI and post-MRI session were compared using linear mixed-effects models with internal state (hunger, fullness or thirst) as the outcome, time (pre or post) as a fixed effect and participant as a random intercept.
fMRI data were preprocessed with FSL version 6.0.7.17 (FMRIB Software Library) and SPM version 25.01.02 (Statistical Parametric Mapping) implemented in MATLAB R2021b. Susceptibility-induced distortions were estimated using opposite phase-encoded images and corrected with a Jacobian transformation using FSL48 topup49. Preprocessing was then completed with SPM and included realignment, coregistration of the functional images to each participant’s T1-weighted anatomical image, spatial normalization to standard space and spatial smoothing with a 6 mm full-width half-maximum Gaussian kernel.
First and group-level models
Two first-level general linear models (GLMs) were specified. For both GLMs, the interval from the onset of the post-picture fixation period through bid submission (that is, the participant-specific reaction time during each bidding period) was modelled as a regressor of no interest. Additional fMRI analysis information can be found in the Supplementary Information.
The first GLM was designed to capture food-cue reactivity by modelling the 5 s picture-viewing period with a boxcar regressor, with separate regressors for non-UPF and UPF trials10,22. For the first GLM, the non-UPF > UPF contrast was brought to the group level and used to test the association with the between-condition difference in peak metabolic response to the test meals (non-UPF − UPF) measured during the metabolic sessions. To determine the peak metabolic response, we identified local maxima as time points exceeding their immediately preceding and following values. Candidate peaks were retained if their magnitude was at least 75% of the subsequent local maximum and if they occurred at least four units earlier. The peak metabolic response was defined as the first local maximum that met these criteria. Separate models were estimated for each metabolic measure (carbohydrate oxidation, RER, metabolic rate, fat oxidation, blood glucose and blood insulin). No additional correction for multiple comparisons across metabolic measures was applied, as each represented an a priori physiological outcome of interest and was evaluated independently, consistent with previous studies examining metabolism–brain associations15,16.
For the second GLM that examined value, group-level analyses assessed value-related activity by examining the WTP parametric modulator during picture viewing and testing differences in value-related responses between the non-UPF and UPF cues. The second GLM was designed to capture value-related responses while accounting for measures that differed across conditions (frequency of consumption and perceived healthiness), such that effects attributed to food type were not confounded by these factors. Accordingly, the second model was identical to the first but additionally included trial-wise WTP, frequency of consumption and perceived healthiness as parametric modulators of the 5 s picture-viewing regressor. The contrast of non-UPF > UPF and UPF>non-UPF was then tested on the second level to examine the interaction, or the difference, between the association of WTP across UPF and non-UPF items.
Prior work has shown that peak metabolic responses are associated with striatal activation15,16, and recent meta-analyses have also associated the striatum with differences in WTP24. Accordingly, we examined both GLMs using whole-brain corrected analyses and SVC within a single a priori striatal mask (caudate, putamen and nucleus accumbens) derived from the Harvard–Oxford atlas50. These regions were extracted bilaterally using the most lenient probability threshold (largest mask), added and then binarized to create a single large mask. This single large mask was used for all region of interest analyses to reduce alpha inflation (Supplementary Fig. 2). Sex and age were entered as covariates for both GLM group-level analyses.
Corrections for multiple comparisons were applied using a peak-level FWE-corrected threshold of P < 0.05. For visualization of clusters containing the peak voxel, statistical maps are displayed at uncorrected P < 0.001, k > 10. Two-tailed t-tests were performed on extracted parameter estimates where appropriate.
Reporting summary
Further information on research design is available in the Nature Portfolio Reporting Summary linked to this article.
Data underlying this manuscript are available from the Virginia Tech Data Repository (https://doi.org/10.7294/32077803)51. Neuroimaging data have been deposited on OpenNeuro, accession number ds007693 (ref. 52). Source data are provided with this paper.
Code underlying this manuscript is available from the Virginia Tech Data Repository (https://doi.org/10.7294/32077803)51
Ravandi, B. et al. Prevalence of processed foods in major U.S. grocery stores. Nat. Food 6, 296–308 (2025).
Williams, A. M., Couch, C. A., Emmerich, S. D. & Ogburn, D. F. Ultra-processed food consumption in youth and adults: United States, August 2021–August 2023. NCHS Data Brief https://doi.org/10.15620/cdc/174612 (2025).
Monteiro, C. A. et al. Ultra-processed foods: what they are and how to identify them. Public Health Nutr. 22, 936–941 (2019).
Lane, M. M. et al. Ultra-processed food exposure and adverse health outcomes: umbrella review of epidemiological meta-analyses. Brit. Med. J. 384, e077310 (2024).
Monteiro, C. A. et al. Ultra-processed foods and human health: the main thesis and the evidence. Lancet 406, 2667–2684 (2025).
Clarke, J. D. et al. Bioavailability and inter-conversion of sulforaphane and erucin in human subjects consuming broccoli sprouts or broccoli supplement in a cross-over study design. Pharmacol. Res. 64, 456–463 (2011).
Huang, X., Liu, H., Ma, Y., Mai, S. & Li, C. Effects of extrusion on starch molecular degradation, order–disorder structural transition and digestibility—a review. Foods 11, 2538 (2022).
Tosh, S. M. & Chu, Y. Systematic review of the effect of processing of whole-grain oat cereals on glycaemic response. Br. J. Nutr. 114, 1256–1262 (2015).
Cai, M. et al. Upper-gastrointestinal tract metabolite profile regulates glycaemic and satiety responses to meals with contrasting structure: a pilot study. Nat. Metab. 7, 1459–1475 (2025).
DiFeliceantonio, A. G. et al. Supra-additive effects of combining fat and carbohydrate on food reward. Cell Metab. 28, 33–44.e3 (2018).
Perszyk, E. E. et al. Fat and carbohydrate interact to potentiate food reward in healthy weight but not in overweight or obesity. Nutrients 13, 1203 (2021).
Myers, K. P. The convergence of psychology and neurobiology in flavor-nutrient learning. Appetite 122, 36–43 (2018).
de Araujo, I. E., Schatzker, M. & Small, D. M. Rethinking food reward. Annu. Rev. Psychol. 71, 139–164 (2020).
Sclafani, A. & Ackroff, K. Role of gut nutrient sensing in stimulating appetite and conditioning food preferences. Am. J. Physiol. Regul. Integr. Comp. Physiol. 302, R1119–R1133 (2012).
de Araujo, I. E., Lin, T., Veldhuizen, M. G. & Small, D. M. Metabolic regulation of brain response to food cues. Curr. Biol. 23, 878–883 (2013).
Veldhuizen, M. G. et al. Integration of sweet taste and metabolism determines carbohydrate reward. Curr. Biol. 27, 2476–2485.e6 (2017).
Fernandes, A. B. et al. Postingestive modulation of food seeking depends on vagus-mediated dopamine neuron activity. Neuron 106, 778–788.e6 (2020).
Goldstein, N. et al. Hypothalamic detection of macronutrients via multiple gut–brain pathways. Cell Metab. 33, 676–687.e5 (2021).
Han, W. et al. Striatal dopamine links gastrointestinal rerouting to altered sweet appetite. Cell Metab. 23, 103–112 (2016).
Kelly, A. L., Baugh, M. E., Oster, M. E. & DiFeliceantonio, A. G. The impact of caloric availability on eating behavior and ultra-processed food reward. Appetite 178, 106274 (2022).
Hutelin, Z. et al. Creation and validation of a NOVA scored picture set to evaluate ultra-processed foods. Appetite 198, 107358 (2024).
Tang, D. W., Fellows, L. K. & Dagher, A. Behavioral and neural valuation of foods is driven by implicit knowledge of caloric content. Psychol. Sci. 25, 2168–2176 (2014).
Tellez, L. A. et al. Glucose utilization rates regulate intake levels of artificial sweeteners. J. Physiol. 591, 5727–5744 (2013).
Newton-Fenner, A. et al. Economic value in the brain: a meta-analysis of willingness-to-pay using the Becker–DeGroot–Marschak auction. PLoS ONE 18, e0286969 (2023).
Hall, K. D. et al. Ultra-processed diets cause excess calorie intake and weight gain: an inpatient randomized controlled trial of ad libitum food intake. Cell Metab. 30, 67–77.e3 (2019).
Peracchi, M. et al. The physical state of a meal affects hormone release and postprandial thermogenesis. Br. J. Nutr. 83, 623–628 (2000).
Witt, T. & Stokes, J. R. Physics of food structure breakdown and bolus formation during oral processing of hard and soft solids. Curr. Opin. Food Sci. 3, 110–117 (2015).
Corpeleijn, E., Saris, W. H. M. & Blaak, E. E. Metabolic flexibility in the development of insulin resistance and type 2 diabetes: effects of lifestyle. Obes. Rev. 10, 178–193 (2009).
Koves, T. R. et al. Mitochondrial overload and incomplete fatty acid oxidation contribute to skeletal muscle insulin resistance. Cell Metab. 7, 45–56 (2008).
Stouffer, M. A. et al. Insulin enhances striatal dopamine release by activating cholinergic interneurons and thereby signals reward. Nat. Commun. 6, 8543 (2015).
Wingrove, J. et al. The influence of insulin on anticipation and consummatory reward to food intake: a functional imaging study on healthy normal weight and overweight subjects employing intranasal insulin delivery. Hum Brain Mapp. 43, 5432–5451 https://doi.org/10.1002/hbm.26019 (2022).
O’Doherty, J. P., Deichmann, R., Critchley, H. D. & Dolan, R. J. Neural responses during anticipation of a primary taste reward. Neuron 33, 815–826 (2002).
Wang, G.-J. et al. Exposure to appetitive food stimuli markedly activates the human brain. NeuroImage 21, 1790–1797 (2004).
Adamson, K. & Troiani, V. Distinct and overlapping fusiform activation to faces and food. NeuroImage 174, 393–406 (2018).
Kroemer, N. B. et al. Fasting levels of ghrelin covary with the brain response to food pictures. Addict. Biol. 18, 855–862 (2013).
Kroemer, N. B. et al. (Still) longing for food: insulin reactivity modulates response to food pictures. Hum. Brain Mapp. 34, 2367–2380 (2013).
Stice, E., Burger, K. & Yokum, S. Caloric deprivation increases responsivity of attention and reward brain regions to intake, anticipated intake, and images of palatable foods. NeuroImage 67, 322–330 (2013).
Nguyen, N. D. et al. Cortical reactivations predict future sensory responses. Nature 625, 110–118 (2024).
Wolever, T. M., Zurbau, A., Koecher, K. & Au-Yeung, F. The effect of adding protein to a carbohydrate meal on postprandial glucose and insulin responses: a systematic review and meta-analysis of acute controlled feeding trials. J. Nutr. 154, 2640–2654 (2024).
Forde, C. G., Mars, M. & De Graaf, K. Ultra-processing or oral processing? A role for energy density and eating rate in moderating energy intake from processed foods. Curr. Dev. Nutr. 4, nzaa019 (2020).
Angelopoulos, T. et al. The effect of slow spaced eating on hunger and satiety in overweight and obese patients with type 2 diabetes mellitus. BMJ Open Diabetes Res. Care 2, e000013 (2014).
Bunn, A. et al. dplR: Dendrochronology program library in R. Version 1.7.8. CRAN https://doi.org/10.32614/CRAN.package.dplR (2007).
Zhang, Z., Mai, Y., Yang, M., Xu, Z. & McNamara, C. WebPower: basic and advanced statistical power analysis. R package version 0.9.4. CRAN https://cran.r-project.org/web/packages/WebPower/index.html (2023).
Baugh, M. E. et al. Validity and reliability of a new whole room indirect calorimeter to assess metabolic response to small calorie loads. PLoS ONE 19, e0304030 (2024).
Peirce, J. et al. PsychoPy2: experiments in behavior made easy. Behav. Res. Methods 51, 195–203 (2019).
Lim, J., Wood, A. & Green, B. G. Derivation and evaluation of a labeled hedonic scale. Chem. Senses 34, 739–751 (2009).
Becker, G. M., Degroot, M. H. & Marschak, J. Measuring utility by a single-response sequential method. Behav. Sci. 9, 226–232 (1964).
Smith, S. M. et al. Advances in functional and structural MR image analysis and implementation as FSL. NeuroImage 23, S208–S219 (2004).
Andersson, J. L. R., Skare, S. & Ashburner, J. How to correct susceptibility distortions in spin-echo echo-planar images: application to diffusion tensor imaging. NeuroImage 20, 870–888 (2003).
Desikan, R. S. et al. An automated labeling system for subdividing the human cerebral cortex on MRI scans into gyral based regions of interest. NeuroImage 31, 968–980 (2006).
Hutelin, Z. Metabolic and neural responses to ultraprocessed foods: a randomized, controlled, crossover study. University Libraries, Virginia Tech https://doi.org/10.7294/32077803 (2026).
Hutelin, Z. Ultraprocessed foods elicit distinct metabolic and neural responses when compared to non-ultraprocessed foods. [Dataset]. OpenNeuro https://doi.org/10.18112/openneuro.ds007693.v1.0.0 (2026).
We thank M. Fowler and B. Carter for performing the blood draws, Brenda M. Davy for guidance throughout the study, as well as R. McMillian, H. Zhang, and C. Najt and J. Drake of the Metabolism Core at Virginia Tech for assaying the blood samples.
This study was supported by R01 DK132389 to A.G.D. and a National Science Foundation Graduate Research Fellowship (2235205) to Z.H. The funders had no role in study design, data collection and analysis, decision to publish or preparation of the manuscript.
Authors
Z.H. was responsible for the conceptualization, validation, data curation, formal analysis, visualization, investigation, methodology, project administration and writing of the original draft of the manuscript. M.A. performed data curation, formal analysis, visualization and manuscript review and editing. M.E.B. was involved with conceptualization, validation, methodology and manuscript review and editing. E.N. performed formal analysis along with manuscript review and editing. D.L.H. assisted with validation, methodology and manuscript review and editing. A.L.H. was responsible for funding acquisition, supervision, formal analysis and manuscript review and editing. A.G.D. was responsible for funding acquisition, conceptualization, creating methodology, project administration, formal analysis and manuscript review and editing.
The authors declare no competing interests.
Nature Metabolism thanks Carlos Monteiro, Marc Tittgemeyer and the other, anonymous, reviewer(s) for their contribution to the peer review of this work. Primary Handling Editor: Jean Nakhle, in collaboration with the Nature Metabolism team. Peer reviewer reports are available.
Extended Data Table 1 Participant Characteristics
Extended Data Table 2 fMRI results from all analyses
A general participant screening survey is used for the lab, resulting in a high number of pre-screens. After assessing eligibility, 133 participants were contacted and 61 scheduled for in-person screening. Participants were included in multiple analyses, and reasons for exclusion are noted for each analysis type. Final samples for each analysis type are in the bottom row. BMI: body mass index.
There were no differences across conditions among all baseline measures assessed: a) pre meal movement as measured by accelerometers. t(28.46) = 1.73, p = 0.10, 95% CI [−1771.61, 150.57]) b.) blood glucose (mg/dL) (t(26.46) = 1.43, p = 0.17, 95% CI [−5.11, 0.93]) c.) blood insulin (µIU/mL) (t(28.91) = 0.04, p = 0.97, 95% CI [−2.27, 2.09]) d.) metabolic rate (kcal/min) (t(30.10) = 0.06, p = 0.95, 95% CI [−0.02, 0.02]) e.) respiratory exchange ratio (t(29.99) = 1.41, p = 0.17, 95% CI [−0.024, 0.004]) f.) carbohydrate oxidation (g/min) (t(30.49) = 1.40, p = 0.17, 95% CI [−0.023, 0.004])g.) fat oxidation g/min (t(29.95) = −1.22, p = 0.23, 95% CI [−0.003, 0.01]). Each dot represents a participant and error bars represent standard error of the mean. (n = 32 participants for all analysis except blood insulin which is n = 31 participants). All tests are two-tailed.
a.) The ultraprocessed meal was consumed more slowly than the non-ultraprocessed meal (t(30.43) = 5.13, p < 0.001, 95% CI [−1.82, −0.78]). All other measures including b.) liking (t(30.13) = 0.61, p = 0.55, 95% CI [−15.07, 8.12]) c.) wanting (t(30.87) = 0.73, p = 0.47, 95% CI [−12.41, 5.88]) d.) hunger (χ2(3) = 2.05, p = 0.56), e.) fullness (χ2(3) = 6.19, p = 0.10), f.) thirst (χ2(3) = 0.76, p = 0.86) did not have a significantly time by condition interaction. Each dot represents a participant, and error bars represent standard error of the mean. (n = 32 participants) *p < 0.05 and all tests are two-tailed.
a.) Hunger was moderate and increased across the fMRI session (t(51) = 4.69, p < 0.001, 95% CI [−19.34, −7.74]), b.) fullness decreased over the course of the session (t(51) = 4.98, p < 0.001, 95% CI [−15.06, −6.4]), and c.) thirst increased (t(51) = − 4.70, p < 0.001, 95% CI [5.64, 14.05]). Each dot represents a participant, and error bars represent standard error of the mean. (n = 52 participants) *p < 0.05 and all tests are two-tailed.
Supplementary Figs. 1 and 2, Supplementary Tables 1–6 and Supplementary Protocols.
Reporting Summary (download PDF )
Peer Review file (download PDF )
Means for line and bar plots for Fig. 1.
Data for scatter plots in Fig. 2.
Data for bar plots and means for Fig. 4.
