Study overview

This study was conducted on 272 participants tested in one of 3 main experiments. Conditions for optimal use of maltodextrin in FNC, with identification and masking of the orosensory cues resulting from consumption of maltodextrin solutions, were tested in 159 healthy participants (“Maltodextrin optimization” group), 57 of whom to improve definition of maltodextrin concentration and dextrose equivalents, and 102 to define optimal conditions to mask cues allowing for oral identification of maltodextrin. These initial experiments allowed for the definition of a protocol for FNC, with control for the orosensory identification of maltodextrin. This was applied to 52 other healthy volunteers (the “FNC development” group) to assess behavior in the FNC protocol and perform further optimization. In a final group of 61 participants, to compare patients with obesity either before or after bariatric surgery and a new group of healthy controls, data was collected with the final FNC protocol and nuclear medicine imaging of DD2lR availability. Please see Table 1 for a detailed demographic and clinical description of participants.

Download:

• PPT
  PowerPoint slide
• PNG
  larger image
• TIFF
  original image

Table 1. Demographic, gustatory, and psychometric measures of feeding behavior in healthy subjects.

https://doi.org/10.1371/journal.pbio.3002936.t001

Maltodextrin is identified through orosensory cues

While maltodextrin is typically used in FNC protocols as a source of calories due to insipid taste, we had evidence from preliminary qualitative experiments that sweetness and texture could be cues for detection of maltodextrin solutions. We thus developed a series of experiments to determine conditions in which maltodextrin, when dissolved in low-fat yogurt sweetened with sucralose at 0.01% (w/v), would not be discriminated from the base low-fat yogurt solution without maltodextrin added. Across 39 participants, we started by testing different maltodextrin concentrations and dextrose equivalents (DEs) since there are reports of increasing sweetness for higher DE [26]. Indeed, we found that yogurt intensity ratings, normalized relative to the base yogurt solution without maltodextrin, varied according to maltodextrin DE (4–7, 13–17, and 16.5–20; F_(2,72) = 3.4, P = 0.04), but not concentration (17%, 25%, and 33% w/v; F_(2,36) = 0.5, P = 0.6) nor their interaction (F_(4,72) = 0.7, P = 0.6; n = 39; Fig 1A). Furthermore, in a separate group of 18 healthy volunteers, even at the lowest maltodextrin DE (4–7) and concentration (17%), in 3-alternative forced-choice (3-AFC) tests with 1 or 2 stimuli consisting of maltodextrin yogurt and the remaining (respectively 2 or 1) of base yogurt, participants very easily identified maltodextrin above chance level (P < 0.0001; n = 18; Fig 1B). Since all yogurts were similarly sweetened with sucralose, and we had evidence from preliminary experiments that texture could be the major cue for maltodextrin detection, we addressed the problem of maltodextrin identification by adding carboxymethylcellulose (CMC; 0.4% w/v), a flavorless and low energy food thickener, to the base yogurt solution. This was tested in additional 3-AFC tests conducted in a new group of 102 healthy volunteers. Discrimination of the CMC-enriched base yogurt was first tested against 17% (n = 24), 25% (n = 22), and 33% (n = 25) of the lowest DE maltodextrin. While 25% and 33%, maltodextrin was still identified significantly above chance level (25.0%, P = 0.02, n = 22; 33.0%, P = 0.001, n = 25), the lowest concentration (17%) maltodextrin yogurt was not discriminated from CMC yogurt solutions (P = 0.41, n = 24). This was conserved when tested in a final group of 33 participants using flavored yogurts, as planned for FNC experiments (P = 0.42, n = 33; Fig 1C). Thus, in subsequent experiments, we used the contrast 17% maltodextrin/0.4% CMC.

Download:

• PPT
  PowerPoint slide
• PNG
  larger image
• TIFF
  original image

Fig 1. Development of a novel FNC protocol.

(A) Intensity and pleasantness ratings of unflavored maltodextrin yogurt were tested across 3 different groups of healthy individuals, each at a different maltodextrin concentration (17%, 25%, or 33% w/v). In each group, participants consumed a base yogurt solution without maltodextrin and 3 solutions with maltodextrin at distinct dextrose equivalents (DE 4–7, DE 13–17, and DE 16.5–20). Intensity ratings of maltodextrin yogurts, normalized to ratings of base yogurt, varied according to DE (F_(2,72) = 3.4, P = 0.04), independently of the concentration tested (F_(2,36) = 0.5, P = 0.6; interaction: F_(4,72) = 0.7, P = 0.6; mixed-model two-way ANOVA; n = 39). (B) In 3-AFC tests, 1 or 2 of 3 yogurt samples contained maltodextrin (DE4-7), and the other(s) were control yogurts. The percentage of participants that detected maltodextrin was significantly above the chance level (94.4%, p < 0.0001; binominal test vs. 33%; n = 18). (C) In other 3-AFC tests, testing discrimination relative to CMC (0.4% w/v) rather than base yogurt, the higher concentrations of maltodextrin (25% and 33%) were identified significantly above chance (54.5%, P = 0.02, n = 22; 60%, P = 0.001, n = 25, respectively), while 17% maltodextrin was not discriminated from CMC (37.5%, P = 0.41, n = 24), as confirmed when yogurts were flavored (36.4%, P = 0.42, n = 33). (D) After conditioning, while intake of both CS+ and CS- flavors increased significantly (time: F_(1,51) = 17.1, P = 0.0001; stimulus: F_(1,51) = 1.2, P = 0.3; post hoc tests: CS⁺, P < 0.0001, CS^-_, P = 0.02), the interaction between factors suggested differential effects between stimuli (F_(1,51) = 4.3, P = 0.04; repeated measures two-way ANOVA; n = 52). (E) Post-conditioning, preference for CS⁺, as measured according to intake, increased significantly (t₍₅₁₎ = 2.6, P = 0.02, paired t test, n = 52). (F) In participants with low baseline intake preference for CS⁺ (≤50%), there was a significant post-conditioning increase in preference (t₍₂₇₎ = 3.7, P = 0.002, n = 28), but not in those with high baseline preference (>50%; t₍₂₃₎ = 0.08, P = 0.9, n = 24, one-sample t test vs. 0). (G) We found similar pleasantness ratings both for CS^- and CS⁺ before and after conditioning, and unchanged preference for CS⁺, as measured by pleasantness (H), irrespective of the baseline preference (I). Notes: In panels A, D, F, G, and I, data is presented as mean ± standard error of the mean (SEM). *P ≤ 0.05; **P ≤ 0.01; ***P ≤ 0.001; ****P ≤ 0.0001; ns P > 0.05. 3-AFC, 3-alternative forced-choice; CMC, carboxymethylcellulose; DE, dextrose equivalent; FNC, flavor-nutrient conditioning; gLMS/gLHS, general labeled magnitude/hedonic scales. The data supporting this figure is available in S1 Data.

https://doi.org/10.1371/journal.pbio.3002936.g001

Flavor-nutrient conditioning occurs through increased intake but not pleasantness ratings

In another group of 63 eligible healthy volunteers, we then tested a FNC protocol using the optimized sweetened low-fat yogurt solutions, i.e., with either 17% DE 4.0–7.0 maltodextrin (+0.68 kcal/ml; unconditioned stimulus) or 0.4% CMC (+0.012 kcal/ml; control stimulus), paired to 2 distinct flavors (0.3% w/v, respectively CS⁺ and CS^-). Conditioning was conducted at home, during 2 days with 150 g CS⁺/Maltodextrin, alternated with 2 days of 150 g CS^-/CMC. Among these volunteers, 9 were excluded due to errors in the application of or compliance with protocol, and 2 due to low at-home consumption of yogurt, resulting in data from 52 participants for analysis (n = 52, Table 1). In this group, during at-home conditioning, no effects were found for day or stimulus in this group on hunger, thirst, novelty, intensity, or pleasantness ratings (S1A–S1E Fig). However, there was overall greater intake of CS⁺ than CS^- (F_(1,102) = 5.5, P = 0.02), with less consumption in the second conditioning day (F_(1,102) = 8.1, P = 0.005; interaction: F_(1,102) = 0.5, P = 0.5; S1F Fig). Hunger, thirst, and intensity ratings also did not differ significantly between pre-and post-conditioning days (S1G, S1H, and S1J Fig), while novelty ratings decreased significantly after conditioning (F_(1,51) = 10.21, P = 0.002), similarly for both stimuli (F_{(1, 51)} = 0.17, P = 0.7; interaction: F_{(1, 51)} = 1.14, P = 0.29; S1I Fig), presumably as a result of multiple exposures during conditioning. Importantly, intake increased significantly in post-conditioning tests, relative to pre-conditioning (F_(1,51) = 17.1, P = 0.0001), without significant differences between CS^- and CS⁺ (F_{(1,51) =} 1.2, P = 0.3). However, there was a significant interaction between time and stimulus (F_(1,51) = 4.3, P = 0.04; Fig 1D), showing that this change was differential between stimuli. Indeed, the %preference for CS⁺, as calculated according to intake (intake %preference), significantly increased from pre to post-conditioning (t₍₅₁₎ = 2.61, P = 0.02; Fig 1E), suggesting that change in this measure (ΔCS⁺ preference = post-test minus pre-test CS⁺ %preference) may be used to assess the efficacy of conditioning per individual. Moreover, the increase in ΔCS⁺ intake preference was particularly evident for participants with low (≤50%) baseline intake %preference for CS⁺ (t₍₂₇₎ = 3.4, P = 0.002, n = 28), while in those with >50% CS⁺ intake %preference at baseline, ΔCS⁺ preference did not increase nor reduce significantly after conditioning (t₍₂₃₎ = 0.08, P = 0.9, n = 24; Fig 1F). Since the definition of the CS+ yogurt was random, participants were randomly distributed between the high and low preference groups and had similar demographic, gustatory, and psychometric variables (S1 and S2 Tables) that are thus not expected to have contributed to differences between the groups. Our results rather suggest that preference could not be further increased by conditioning when it was already high at baseline, with the fact that it also did not decrease arguing against effects resulting simply from regression to the mean [27], and thus supporting that conditioning was restricted to the low baseline preference subgroup. Pleasantness ratings, on the other hand, did not change significantly after conditioning (F_{(1, 51)} = 0.05, P = 0.8) and were similar for CS^- and CS⁺ (F_(1,51) = 0.4, P = 0.5; interaction: F_(1,51) = 0.2, P = 0.8; Fig 1G). Consistently, CS⁺ %preference, when calculated according to pleasantness ratings (pleasantness %preference), did not differ between pre- and post-conditioning (t₅₁ = 0.5, P = 0.6; Fig 1H). Furthermore, in calculations according to pleasantness measurements, ΔCS⁺ preference did not differ from zero in those with low baseline pleasantness %preference for CS⁺ (t₍₂₅₎ = 1.3, P = 0.2, n = 26), and had a close to significant reduction, rather than increase, in those with high baseline pleasantness %preference (t₍₂₅₎ = 1.9, P = 0.07, n = 26; Fig 1I). Importantly, a sensitivity analysis excluding participants with BMI 25 kg/m² or greater revealed overlapping results to those obtained in the full sample (S2A and S2B Fig). Additionally, the ΔCS⁺ preference measure calculated according to intake was not associated with baseline consumption of milk and yogurt, as self-reported according to average daily number of cups, namely in participants most sensitive to the effects of conditioning (low baseline preference for CS+: r = −0.1, P = 0.6, n = 28). Overall, these findings support that human flavor nutrient conditioning, when performed controlling for orosensory discrimination of maltodextrin, influences primarily implicit feeding decisions (i.e., increase in intake) rather than explicit assessments of stimuli (i.e., increase of pleasantness).

Flavor-nutrient conditioning is conserved in obesity and after bariatric surgery

We then used the optimized FNC protocol to test a clinical cohort of patients from a bariatric surgery program, where we recruited 34 eligible patients, 11 of whom with obesity approved for bariatric surgery and 23 after bariatric surgery (gastric bypass, n = 13; sleeve gastrectomy, n = 10), that were compared with a group of 27 healthy controls (S3A and S3B Fig). Groups differed significantly according to age (F_(2,60) = 12.7, P = 0.00003), BMI (F_(2,60) = 106.3, P < 0.00001) and years of formal education (F_(2,60) = 10.9, P = 0.0001; Table 1), but significant differences were not found across most gustatory and psychometric variables, except PFS scores (aggregate and food available; both P = 0.04), addiction-like feeding behavior (YFAS - number of symptoms, P = 0.004; YFAS - diagnosis rate P = 0.01), and external eating (P = 0.001; Table 1).

Here, for FNC, flavors with lower baseline intake preference were selected to pair with maltodextrin, given that conditioning was not observed during protocol development when maltodextrin was paired with flavors with high baseline preference (please see Fig 1F). Consequently, CMC was paired with the flavor with a higher pre-conditioning intake preference, and maltodextrin was paired with the flavor with a lower preference (S4 Fig). During conditioning, significant differences were not found between the groups or CS types regarding hunger, thirst, novelty, and pleasantness ratings (S5A–S5C and S5E Fig). However, intensity ratings varied according to CS type (F_{(1, 50)} = 5.2, P = 0.03; S5D Fig), while intake varied according to group (F_{(1, 50)} = 7.5, P = 0.001; interaction: F_{(2, 50)} = 1.0, P = 0.4; S5F Fig). Consistently with data from the original healthy volunteer group, while there were no effects for hunger, thirst, and intensity, novelty ratings decreased from pre- to post-conditioning (F_(1,50) = 12.9, P = 0.001; S5G–S5J Fig). Regarding the effects of conditioning on flavor preference, we found that ΔCS⁺ preference, as assessed by intake, increased significantly after conditioning (t₍₅₂₎ = 3.6, P < 0.001; n = 53), with no significant differences significantly across healthy (n = 24), obese (n = 9), and surgical groups (n = 20; F_{(2, 50)} = 1.9, P = 0.2; Fig 2A). Regarding pleasantness ratings, ΔCS⁺ preference also did not vary significantly according to group (F_{(2, 50)} = 0.2, P = 0.8; Fig 2B) and, as observed in the initial experiments, did not reflect any significant effects of conditioning (t₍₅₂₎ = −1.4, P = 0.2; n = 53). In exploratory analyses, ΔCS⁺ preference did not differ between the sleeve and bypass groups for either intake (t₍₁₈₎ = 1.1, P = 0.3) or pleasantness (t₍₁₈₎ = −0.2, P = 0.9; n = 20). Sensitivity analyses excluding participants with BMI 25 kg/m² or greater from the healthy volunteer group revealed overlapping results to those obtained in the entire sample (S2C and S2D Fig). Finally, although there was a significant effect across groups for at-home intake during conditioning, with lower volumes consumed by participants in the surgical group (S5F Fig), across all participants as well as healthy, obesity, and surgical groups, we did not find any significant correlation between ΔCS⁺ preference intake and total mean consumption (all: r = 0.2, p = 0 .3; healthy: r = 0.2, p = 0.3; obese: r = 0.3, p = 0.4; surgical: r = 0.1, p = 0.6), mean CS+ consumption (all: r = 0.2, p = 0.1; healthy: r = 0.2, p = 0.3; obese: r = 0.3, p = 0.4; surgical: r = 0.3, p = 0.2) or mean CS-consumption (all: r = 0.1, p = 0.5; healthy: r = 0.2, p = 0.5; obese: r = 0.3, p = 0.5; surgical: r = −0.1, p = 0.8).

Download:

• PPT
  PowerPoint slide
• PNG
  larger image
• TIFF
  original image

Fig 2. Measures of FNC across the clinical study groups.

(A) ΔCS⁺ preference, measured by intake, across the clinical study groups. A one-way ANOVA revealed a nonsignificant group effect (F_{(2, 50)} = 1.9, P = 0.2) across healthy (n = 24), obese (n = 9), and surgical (n = 20) groups. (B) ΔCS⁺ preference, measured by pleasantness ratings, across the clinical study groups. A one-way ANOVA revealed a nonsignificant group effect (F_{(2, 50)} = 0.2, P = 0.8) across groups. Notes: Data is presented as mean ± SEM. FNC, flavor-nutrient conditioning; gLMS, general labeled magnitude scales; SEM, standard error of the mean. The data supporting this figure is available in S1 Data.

https://doi.org/10.1371/journal.pbio.3002936.g002

Reduced striatal DD2lR availability in obesity is recovered after bariatric surgery and may be related to postingestive conditioning after gastric bypass

While reduced availability of DD2lR is associated with extreme obesity [14,28–30], there is controversy relative to the possibility that bariatric surgery may normalize this effect [31–34], with a need for further studies and some evidence of advantages in the use of [¹²³I]IBZM SPECT [15]. We used this method to assess DD2lR availability in the groups of the clinical study and performed exploratory analyses to test associations with FNC in the surgical group. We found significant differences in DD2lR availability across study groups (F_(2,52) = 9.8, P = 0.0002), which was lower in patients with obesity (n = 11) when compared both with healthy volunteers (P = 0.02, n = 21) and surgical patients (P = 0.0001, n = 23), but did not differ between the surgical and healthy groups (P = 0.2; S3A and S3B Fig), with globally overlapping results in sensitivity analyses excluding participants with BMI 25 kg/m² or greater from the healthy volunteer group (S2E Fig). While we did not find differences in DD2lR availability between the gastric bypass (n = 13) and sleeve gastrectomy (n = 10) subgroups (P = 1.0), within the former group DD2lR availability was negatively correlated with ΔCS⁺ preference (intake) (r = −0.7, P = 0.02, n = 11; Fig 3C), and positively correlated with DEBQ—restrained eating (r = 0.8, P = 0.01, n = 11; Fig 3D). Neither of these correlations was found in sleeve gastrectomy (Fig 3C and 3D) or other groups (see S3 Table for details). Our results corroborate previous findings of decreased DD2lR availability in obesity and support the possibility that these obesity-related effects are reverted by bariatric surgery, similarly following gastric bypass and sleeve gastrectomy. Nevertheless, and in support of specificities of gastric bypass effects on postingestive conditioning, only after gastric bypass was there evidence of associations of DD2lR availability with FNC conditioning strength, as well as with restrained eating.

Download:

• PPT
  PowerPoint slide
• PNG
  larger image
• TIFF
  original image

Fig 3. Striatal DD2lR availability in obesity and bariatric surgery and associations with postingestive conditioning and restrained eating.

(A) Average [¹²³I]IBZM group images in the striatal central transverse plane of healthy subjects (n = 21; upper panel), patients with obesity (n = 11; lower left panel), and surgical patients (n = 23; lower right panel). (B) There was a group effect for striatal DD2lR availability (F_{(2, 52)} = 9.81, P = 0.0002), with post hoc tests supporting lower striatal BP for the obesity group relative to both surgical (P = 0.0001) and healthy (P = 0.02) groups, but not between surgical and healthy groups (P = 0.19). (C) Association between striatal DD2lR availability and ΔCS⁺ preference (intake) across surgery types (Surgical group, r = −0.14, P = 0.6, n = 20; Bypass, r = −0.68, P = 0.02, n = 11; Sleeve, r = 0.49, P = 0.18, n = 9). (D) Association between striatal DD2lR availability and the Dutch Eating Behaviour Questionnaire—restrained eating scores across surgery types (Surgical group, r = 0.19, P = 0.4, n = 19; Bypass, r = 0.77, P = 0.01, n = 11; Sleeve, r = −0.26, P = 0.5, n = 8). Notes: In panel B, data is presented as the mean ± SEM. *P ≤ 0.05; ***P ≤ 0.001. The data supporting this figure is available in S1 Data. DD2lR, dopamine D2-like receptor; SEM, standard error of the mean.

https://doi.org/10.1371/journal.pbio.3002936.g003

Limitations

The results of this study should be interpreted according to its limitations. Our FNC protocol is limited by the fact that the 4 conditioning days were conducted at home to avoid loss of follow-up, but it also limits experimental control over this phase of the experiment. Despite our efforts to minimize these limitations, namely the use of saliva samples to increase compliance for fasting and exclusion of participants with low adherence to conditioning procedures, there is a possibility for self-report bias that we cannot account for. On the other hand, we did not perform a priori group matching in the clinical study for age, gender, or education. A strictly matched case-control design in the clinical study would have been an asset but is hindered due to the challenges in recruitment within the clinical groups, particularly for a complex protocol as described here. Furthermore, the obesity and surgical groups were not assessed prospectively to avoid the effects of learning on repeated exposures to the FNC paradigm and due to the known challenges of longitudinal follow-up in this clinical population [52]. Challenges in recruitment of the clinical population were also reflected in small sample sizes that may have limited statistical power. Larger studies addressing more restricted hypotheses (e.g., differences in obesity versus controls or the impact of gastric bypass) are needed to replicate and expand these results.

It is also noteworthy that we applied exclusion criteria directly related to metabolic health, namely diabetes, and treatment with antidiabetic medication, contributing to a sample not fully representative of bariatric populations. Indeed, we have previously described that approximately 20% of pre-bariatric patients recruited from the Portuguese healthcare system have T2DM [24]. This decision resulted from concerns that experimental instructions for fasting would not be safe in patients with T2DM. However, glycemic metabolism may likely impact FNC processes, and future studies will benefit from including a more extensive metabolic characterization, including measures of glucose homeostasis, among others, in addition to direct measures of ingestive behavior and other variables that allow phenotyping of the patients. Finally, we assessed DD2lR availability in a static protocol, after the FNC protocol. Ideally, future research should quantify brain responses to postingestive conditioning in real time.

Study design and participants

Healthy volunteers were recruited from the community to optimize the conditions for using maltodextrin in FNC (“Maltodextrin optimization” group) and then to test and optimize a controlled protocol for FNC (“FNC development” group). Inclusion criteria were age between 18 and 65 years and general good health as determined by the investigator. Exclusion criteria assessed at entry into the study were active acute respiratory infection, active neurological or psychiatric disease; active gastrointestinal, hepatic, or pancreatic disease; diabetes, illicit substance use or alcohol abuse; use of any neuropsychiatric medication (including anxiolytics, antipsychotics, antidepressants, anticonvulsants, stimulants, anti-dementia medication, dopamine agonists, and opioid analgesics) or antidiabetic medication (including glp-1 agonists); illiteracy, or otherwise not understanding instructions for the study; prior major gastrointestinal surgery and intra-gastric balloon in the previous 12 months; history of food allergies, including any allergy to milk components or lactose intolerance; pregnancy or breastfeeding. The clinical cohort consisted of consecutive patients at a tertiary care outpatient center specialized in the surgical treatment of obesity, belonging to Centro Hospitalar de Lisboa Ocidental, E.P.E., in Lisbon, Portugal. The cohort included patients approved for bariatric surgery and on the waiting list (obese group) and patients who had received bariatric surgery (surgical group). The latter were recruited no less than 1.5 and no more than 4 years after either gastric bypass or sleeve gastrectomy, when patients are expected to be weight stable and capable of consuming small volumes of liquid. Approval for bariatric surgery followed standard criteria defined by the Portuguese National Health Service [24]. Exclusion criteria for patients were equivalent to those mentioned above, except for BMI and prior major gastrointestinal surgery for the surgical group only. Patients were identified by the clinical team, and those consenting to be contacted were screened by phone. Those not excluded were further assessed for eligibility at admission into the study. For patients, we retrieved the surgery date and type from clinical files to avoid self-report bias. An additional group of healthy volunteers was recruited for comparison with patients. Exclusion criteria were equivalent to those mentioned above, as well as obesity (BMI ≥ 30 kg/m²) and underweight (BMI < 18.5 kg/m²). Participants of the study were recruited between 11/01/2013 and 06/12/2021, with the recruitment for the clinical study starting on 30/12/2016. The study was conducted according to the principles expressed in the Declaration of Helsinki. Approval for the study protocol was granted by Ethics Committees of the Champalimaud Foundation (ref: none available), the Lisbon Academic Medical Centre (ref: 124/16), and Centro Hospitalar Lisboa Ocidental (ref: none available). Written informed consent was obtained from each participant. The possibility of discontinuing participation at any time during the study was given to all participants. All data were de-identified.

Solutions

Yogurt (34 kcal, 0.1 g of fat, 4.3 g of carbohydrates, and 4.0g of protein per 100 g) was purchased from a national commercial provider (Continente, Portugal) and was stored at 4°C. Sucralose, maltodextrin, carboxymethylcellulose (Sigma Aldrich), and flavors (Nature’s Flavors, Orange, California, United States of America) were stored at room temperature. Milli-Q water was obtained from our institutional distilled water system. All yogurt-based solutions were prepared daily under sterile conditions and stored at 4°C until 1 h before each experiment when they were transferred to room temperature. Maltodextrin was first diluted in 1/3 of the intended final yogurt solution volume in Milli-Q water. The solution was dissolved using a plate heater and a magnetic stirrer at 90°C (Ohaus, USA). Once this solution was again at room temperature, the final intended volume was completed with 2/3 yogurt to obtain final maltodextrin yogurt at concentrations of 17%, 25%, or 33% (w/v; respectively 0.68 kcal/ml, 1 kcal/ml, 1.32 kcal/ml). CMC (carboxymethylcellulose) yogurt solutions were similarly prepared to obtain a final concentration of 0.4% (w/v; (0.012 kcal/ml). A base yogurt solution was prepared with 1/3 Milli-Q water and 2/3 yogurt (v/v). All yogurt solutions had sucralose added at a concentration of 0.01% (w/v). For flavored solutions (cashew, lychee, tamarind, cider, black currant, and pomegranate), the selected flavor was added at a concentration of 0.3% (w/v).

Optimizing maltodextrin concentration and dextrose equivalents

In the first cohort of healthy volunteers, psychophysical assessments of distinct maltodextrin concentrations and dextrose equivalents were performed. Participants were divided into 3 groups, each testing one concentration of maltodextrin (17%, 25%, or 33%). For each concentration, in addition to a base yogurt solution, participants sampled 3 solutions of maltodextrin-enriched yogurt, all at the same concentration but prepared with maltodextrin of distinct DEs, namely DE 4–7, DE 13–17, and DE 16.5–20. The 4 yogurt solutions were presented in randomized order and immediately rated according to intensity (0 to 100 mm general Labelled Magnitude Scale—gLMS [53]) and pleasantness (−100 to 100 mm, general Labelled Hedonic Scale—gLHS [54]).

Discrimination tests to assess maltodextrin detection

Additional groups of healthy volunteers performed a 3-AFC test to determine the discriminability of sweetened maltodextrin yogurt solutions. In an initial test, discrimination of 17% DE 4–7 maltodextrin was tested against a base yogurt solution. The 3-AFC test presented 1 or 2 cups with 17% maltodextrin yogurt solution, with the remaining (respectively 2 or 1) cup(s) containing the base yogurt solution. Participants were asked to sample all 3 of the yogurts and then select the one that was different from the other two according to any sensory attribute that was salient for that participant (e.g., taste, smell, texture, among others). In 3 additional cohorts of healthy volunteers, each of the 3 different maltodextrin yogurt concentrations (17%, 25%, or 33%), all prepared with DE 4–7 maltodextrin, were contrasted in 3-AFC tests with CMC yogurt solution (0.4%). In a final group of healthy participants, the contrast between 17% maltodextrin and 0.4% CMC yogurt solutions was repeated, but with one of 6 flavors (cashew, lychee, tamarind, cider, black currant, and pomegranate; 0.3%) added to the solutions used in each 3-AFC test.

Flavor-nutrient conditioning

Experimental sessions occurred on 6 consecutive days following an overnight fast of 8 to 10 h. Participants attended the laboratory on the first (pre-conditioning) and last (post-conditioning) days, with 4 conditioning days performed at home between the first and last test days. During the experiments, Milli-Q water at room temperature was available for consumption if desired by the participant. On the first day, participants were assessed for height and weight with a digital scale and a mechanical stadiometer (Kern & Sohn GmbH, Balingen, Germany) with light clothes and without shoes. At the end of that day, each participant completed gustatory psychophysics (taste strips method for citric acid, sodium chloride, sucrose, and quinine hydrochloride; taste thresholds assessed with electrogustometry [55]) and psychometric evaluation of reward-related feeding behavior (Power of Food Scale [13,49,56,57], Yale Food Addiction Scale [58,59], Dutch Eating Behavior Questionnaire [60,61], and Food Action Rating Scale [62]), as described previously (please see Ribeiro and colleagues [24] for details). On the pre-conditioning day, after collecting ratings of hunger and thirst on 0 to 100 mm Visual Analogue Scales (VAS), participants were presented with samples of 6 differently flavored (cashew, lychee, tamarind, cider, black currant, and pomegranate) CMC yogurt solutions, presented in random order for ratings of stimulus novelty (0 to 100 mm VAS, with higher numbers indicating greater novelty), intensity (0 to 100 mm gLMS, with higher numbers indicating greater intensity), and pleasantness (−100 to 100 mm gLHS, with more positive numbers indicating greater pleasantness, and more negative numbers indicating greater unpleasantness). We selected 2 flavored beverages for each participant based on high novelty and similar moderate pleasantness. Each subject then performed 6 trials of 3-AFC discrimination tests contrasting the 2 selected flavors, with a revision of the flavors selected if correct discrimination was not obtained in at least 4 trials. Participants were excluded if 2 flavors with pleasantness rated above 0 in the gLHS, and that were correctly discriminated in 3-AFC tests, could not be identified. We then presented the 2 flavored CMC yogurt solution in 2 large white cups for ad libitum consumption/intake and measured weight (g) before and after consumption (intake measurement). In the initial cohort of healthy volunteers, one of these flavors was randomly chosen to pair with maltodextrin during home conditioning (CS⁺, +102 kcal), while the alternate flavor was paired with CMC (CS^-, +1.8 kcal). For the clinical experiment, maltodextrin was always paired with the least preferred flavor, as assessed according to intake during ad libitum consumption. In the following 4 conditioning days, at home, subjects were instructed to maintain overnight fasting, after which they should consume yogurt solutions, and refrain from eating anything else for the following hour. Yogurts for at-home consumption were distributed to participants in the pre-conditioning day in sterilized glass bottles, containing 150 g of yogurt, that they were instructed to conserve at 4°C. Bottles were labeled with the consumption date and a letter code assigned to CS⁺ or CS^- flavors, so that they consumed either a 17% maltodextrin yogurt solution, paired with the CS⁺ flavor, in 2 non-consecutive days, or a 0.4% CMC yogurt solution, paired with the alternate flavor (CS^-), in the 2 alternate days, with the order of flavors randomized. Participants were instructed to consume the maximum yogurt possible and return any yogurt that they did not consume for quantification of intake (g). Pre-conditioning, conditioning, and post-conditioning data were excluded from analysis in participants for whom mean home consumption was, on average, less than 25 g in total or for any of the 2 flavors. They were also asked to perform ratings of hunger and thirst prior to consumption, and of stimulus novelty, intensity, and pleasantness after consumption. In the post-conditioning day, we presented the same six-flavor sequence of CMC yogurt solutions, as on the first day, for the same process of flavor rating. Then, the 2 flavors selected for conditioning were given for ad libitum consumption. The main outcomes of this protocol were changes in %preference for CS⁺, assessed according to intake (CS⁺ intake %preference) or pleasantness ratings (CS⁺ pleasantness %preference) from pre- to post-conditioning. In participants of the clinical study, at the end of the last day, single-photon emission computed tomography (SPECT) scans were performed.

Striatal DD2lR availability imaging

We assessed striatal DD2lR availability using SPECT with [¹²³I]-Iodobenzamide ([¹²³I]IBZM, GE Healthcare, Eindhoven, NL). Participants were scanned early in the afternoon for approximately 30 min, 2 h after a bolus injection of 185 MBq of [¹²³I]IBZM. Each participant was pretreated with potassium iodide to block thyroid uptake of free radioactive iodine (¹²³I). SPECT imaging was performed using a Philips BrightView gamma camera (Philips Healthcare, Eindhoven, NL) with low-energy and high-resolution collimators. Image reconstruction was performed using the Astonish algorithm (Philips Healthcare, Eindhoven, NL), an optimized 3D ordered subsets expectation maximization (3D-OSEM) algorithm. After reconstruction, images were corrected for attenuation using the Chang method (linear attenuation coefficient of 0.11 cm^-1) and the Hanning filter (cut-off 1.0). SPECT images were reconstructed with cubic voxels of 4.664 mm width, and a region of interest (ROI) analysis was performed for quantification based on an automated software validated for brain [¹²³I]FP-CIT SPECT scans [63]. This software was adapted for brain [¹²³I]IBZM and validated against semiautomated quantification performed by experienced nuclear medicine physicians. The primary outcome of ROI analysis was the striatal binding potential (BP) in the ROIs, which is a proxy for striatal DD2lR (i.e., D2 and D3 receptors) binding. When the radiopharmaceutical reaches equilibrium, the BP can be obtained as the ratio of the specific uptake in the target region and the nonspecific uptake, as shown in Eq 1.

Eq 1. Binding potential.

The reference region was a portion of the occipital lobe where D2 and D3 receptors are absent. The software quantifies the BP in 6 striatal ROIs (right caudate, right putamen, right striatum, left caudate, left putamen, and left striatum). For statistical analyses, the left and right striatum mean values were considered.

Data analysis

Analyses were performed using SPSS version 29 (SPSS Inc., Chicago, Illinois, USA). Graphs were produced in GraphPad Prism version 8.0 (GraphPad Software, La Jolla, California, USA) and edited in Adobe Illustrator version 2022 (Adobe Inc., San Jose, Califonia, USA). Data for continuous measurements is presented as mean ± standard error of the mean (SEM). Assessment of the normal distribution of continuous measurements was performed according to visual inspection of distribution as well as analysis of kurtosis, skewness, and comparison between mean and median. To address items with missing values, for scales where this approach is admissible (PFS, DEBQ, and FARS), simple imputation was performed with the mean score of the respective subscale when missing values represented 10% or less of the total items in that subscale. Demographic, clinical, psychometric, and psychophysical data was compared between groups using independent samples t tests or one-way analysis of variance (ANOVA) for continuous variables and χ2 tests for categorical variables. In the maltodextrin optimization group, normalized intensity ratings were analyzed using mixed model two-way ANOVA according to concentration (between subjects) and maltodextrin DE (within subjects). The proportion of participants correctly discriminating maltodextrin yoghurts in 3-AFC tests was contrasted to 1/3 (chance level) using binomial tests. To determine changes in raw intake and pleasantness ratings from pre- to post-conditioning in the healthy group (“FNC development”), we used repeated-measures two-way ANOVA with intake (g) or pleasantness ratings (mm) as independent variables, according to a time factor pairing pre- to post-conditioning days (pre-post) and a stimulus factor comprising CS^- or CS⁺ flavors (CS^- versus CS⁺). Intake %preference for CS⁺ was calculated as [CS⁺ intake/(CS^- + CS⁺ intake) *100]. The same formula, but using gLHS pleasantness ratings, was used to calculate pleasantness %preference for CS⁺. In this case, we transformed pleasantness ratings by adding the amount needed for the minimum value to be 1 (“+101”). To determine changes in preference for CS⁺, we used paired t tests to compare pre- to post-preferences according to intake or pleasantness. The difference between the pre-and post-conditioning preference for CS⁺ (Δpreference CS⁺), calculated according to either intake or pleasantness, was analyzed separately according to respective baseline %preference, namely low (<50%) and high (≥50%), using one-sample t tests contrasting against zero, to test whether there were significant changes in preference in each group. To compare groups in the clinical cohort for Δpreference CS⁺ (intake) or DD2lR availability, we used a one-way ANOVA. Other data from FNC experiments either comparing pre- versus post-conditioning days (hunger, thirst, novelty, intensity) or data from home-conditioning (hunger, thirst, novelty, intensity, pleasantness, intake) was analyzed using repeated-measures two-way ANOVA, paired t tests or mixed-model two-way ANOVA in the case of between-group analyses. Across ANOVA analyses, Bonferroni post hoc tests were performed as planned. Exploratory associations between DD2lR BP, Δ intake preference for CS⁺, and gustatory and psychometric feeding behavior variables were determined using Pearson’s correlation (r). A two-tailed p-value of 0.05 was selected as the significance level for all analyses. Further details on the primary statistical models performed are described in S4 Table. As shown in S5 Table, we determined Eta-squared (η [2]) to measure effect size in the analysis of variance (ANOVA) and mean differences with 95% CIs and Cohen’s d to estimate the effect size of the mean’s differences (Cohen’s d = (M₂—M₁) / SD_pooled) [64]. Regarding correlations, the r² was calculated as the measure of the effect size of correlations. The magnitude of effect size according to Cohen’s (1988) classification, namely, small (d ≤ 0.2), medium (0.2 < d < 0.8), and large (d ≥ 0.8) [64], is shown in S5 Table.