The Institutional Review Boards of the University of Alabama at Birmingham (141121006) and Texas Tech University (505571) approved this study. This trial was pre-registered on clinicaltrials.gov as NCT02613065. The reporting of this article complies with the Consolidated Standards of Reporting Trials (CONSORT) guidelines (see Additional file 2).
Participants
Participants were recruited from the Birmingham, AL metro area. The study population was males and females 30–50 years old with a body mass index (BMI) of 20–30 kg/m2. Interested persons were screened over the phone and, if eligible, invited for an in-person screen. The in-person screen included staff-measured height and weight, a urine pregnancy test, and completion of the Brief Symptom Inventory [22] and EAT-26 [23] questionnaires to measure psychological distress and eating disorder symptoms, respectively. Participants then provided written informed consent.
Participants were excluded for self-reported high levels of physical activity, major illness, smoking, statin use, use of an unstable dose of anxiety, depression, or steroid medications, food allergies or restrictions, claustrophobia, drug or alcohol abuse, participation in a weight loss program or special diet in previous 3 months, weight change > 5% in previous 6 months, use of medication that affects appetite, prior surgical procedure for weight control, EAT-26 score ≥ 20, BSI score ≥ 90th percentile, pregnancy, anticipating pregnancy, an unwillingness to take contraceptive measures, and nursing.
Study design
This double-blind randomized crossover trial sought to compare energy balance between two 5-day (Monday- Friday) treatment periods with a 2-week wash out between treatments (see Fig. 1). During each treatment period, participants ate breakfast, lunch and dinner (henceforth called major meals) at UAB’s Bionutrition Unit. Participants lived in their normal environments and came and went from the Bionutrition Unit for their major meals. Participants were asked to arrive at the Bionutrition Unit for each major meal between 6:45–8:30 am, 11 am – 12:30 pm and 4–5:30 pm, respectively. Participants often chose to sit together during mealtimes, but independent vs. group eating was not prescribed or recorded. Participants were required to finish either an egg white protein shake (PS) or a maltodextrin carbohydrate shake (CS) prior to consuming a buffet-style major meal. No time restrictions were placed on shake consumption, the transition from shake to buffet, or buffet consumption. Although buffets are known to cause greater food intake than meals with less variety [24], we wanted to test the practicability of the treatment in a quasi-real-world setting. Participants consumed the same shake for the entirety of each treatment period and were randomly assigned the order in which they received PS and CS (n = 48 randomized, see Fig. 2). We hypothesized that weight status may confer differing degrees of appetite regulation and therefore randomization was stratified by BMI (normal weight versus overweight). Allocation within each stratum was determined by block randomization with a block size of 4. The randomization scheme was created by the study statistician and provided to the study coordinator, who enrolled participants and assigned them to treatment allocation.
To strengthen the blinding process and minimize bias, participants and data collectors were unaware of the true study hypothesis and were told the purpose of the study was to determine the effects of a high-fiber and low-fiber shake on mood, which was self-reported by questionnaire on the first, third, and fifth day of each treatment period. The primary purpose of the questionnaire was not to measure mood per se but to increase the believability of the study rationale provided to participants. Participants were debriefed about the true study hypothesis at the end of their participation in the study. At that time, participants were also asked about any GI discomfort experienced during their participation, since the first participant withdrew for this reason.
Basal metabolic rate, resting metabolic rate, and thermic effect of food
Basal metabolic rate (BMR), resting metabolic rate (RMR) and thermic effect of food (TEF) were assessed by indirect calorimetry [25] within 14 days prior to the first treatment period. Briefly, BMR was measured between 7 and 9 am, after a 12 h fast. Afterwards, participants consumed 300kcals of PS or CS, whichever they were assigned to for the first treatment period, and underwent RMR measurements for 10 min every 30 min for 6 h [25]. The first participant consumed 500kcals of PS; thereafter, the recipe was reduced to 300kcals to ease consumption during the allotted 10-min period. The 300 kcal dose, which was larger than what was given before each major meal, was chosen to illicit a measurable response according to standard TEF protocols. Participants were instructed to rest, but remain awake, in a reclined position during these 6-h. TEF was calculated as the area under the curve for energy expenditure, adjusting for baseline BMR.
Preload shakes
Baseline BMR determined the caloric dose of each participant’s shakes. Shake dose was given as a percent of energy needs because energy balance was the primary outcome of the trial. Giving a large dose of supplement to someone with low energy needs would therefore bias energy intake positively relative to a person’s needs, or vice versa, and we wanted to eliminate this bias. In addition, meals where at least 20% of energy is derived from protein induces greater acute satiety after the meal compared to meals with less than 20% energy from protein [26]. Therefore, providing a protein preload with approximately 20% of energy needs over the course of the day should theoretically ensure that total daily energy consumed is at least 20% protein. This dose would likely also reflect what someone might take naturally if they were to consume the supplement on their own, outside a clinical setting or strict recommendation.
Participants were categorized into three BMR groups to minimize error when making the shakes: ≤1199 kcals/day, 1200–1599 kcals/day and ≥ 1600 kcals/day. Participants were assigned a 93 kcal, 130 kcal or 168 kcal dose, respectively, so that preload dose was relative to overall energy needs. For each BMR range, we multiplied the median value of the range by a physical activity level of 1.4. Twenty percent of this calculated value was provided through the preload shake at each meal. For the 93 kcal dose (henceforth called “low dose”), this translated to 18.6 g of protein for PS and 23.4 g of carbohydrate for CS. For the 130 kcal dose (henceforth called “medium dose”), this translated to 26 g of protein for PS and 32.8 g of carbohydrate for CS. For the 168 kcal dose (henceforth called “high dose”), this translated to 33.6 g of protein for PS and 42.4 g of carbohydrate for CS. Caloric doses and total shake volume were held constant across treatments. Shakes were made from water, Crystal Light no-calorie sweetener, and either NOW® Eggwhite Protein (PS) or NOW® Carbo Gain (CS). Energy density (kcal/g) was 4 kcal/g for PS and 3.73 kcal/g for CS. Bionutrition Unit staff developed shake recipes and experimented with dose of crystal light flavoring to sensorially match the shakes as much as possible. Crystal Light flavor (chosen by participant) and dose were also constant across treatments and were served in opaque containers. If participants detected differences between shakes, the differences were likely attributed to differences in fiber content, as they were told that shakes would have high and low fiber content. Shake recipes are provided in Table 1.
Energy intake
Major meals were served ad libitum, buffet-style in UAB’s Bionutrition Unit. The buffet provided mixed meals of “typical” American food in excess, to ensure that quantity did not limit intake. The macronutrient distribution of the buffet was 45–65% carbohydrate, 20–30% fat, and 10–35% protein, in keeping with USDA recommendations [27]. Several snack options were available to be taken between each major meal. Participants were asked to return snack packages, empty or otherwise, at the next major meal. Menus were designed by registered dietitians. Meals were prepared by trained staff who also weighed the food items before and after each meal or snack to obtain the total weight of foods consumed. After the first cohort (n = 5), the meal and snack menus were slightly adjusted to reduce food waste; nevertheless, the menus were consistent for each cohort. All cohorts after the first received the same meal and snack menu (see Additional file 1).
Physical activity energy expenditure
During both treatment periods, participants were instructed to wear a tri-axial accelerometer (AntiGraph GT3X+, Pensacola, FL) over their right hip during waking hours. Accelerometers were distributed at breakfast on Monday and collected at dinner on Friday. Accelerations were summed over 1-min epochs and 60 min were used to determine non-wear time. Using the Freedson VM3 Combination algorithm [28] in Actilife v6.13.3, accelerometer data were processed to yield estimates of daily and weekly physical activity energy expenditure (PAEE) for each treatment period.
Satiety and meal-liking
A standard satiety questionnaire was administered immediately after shake consumption (0 min) and then 15, 90, and 180 min after buffet lunch completion on the first and fifth day of each treatment period. A 100 mm Visual analogue scale (VAS) anchored by “not at all” and “extremely” assessed fullness, hunger, ability to eat more and desire to eat more (see Fig. 3 legend for complete satiety questions). The 0 min satiety questionnaire was completed before partaking the buffet; the 15, 90, and 180 min post-meal questionnaire was completed elsewhere and returned to the Bionutrition Unit at the next meal. A second questionnaire was administered after every major meal to evaluate how much participants liked the shake and buffet meal. A nine-point Likert scale from “dislike extremely” to “like extremely” was used to answer the question, “How much did you like or dislike the meal you just had?”. Shake and buffet meal likings were evaluated together. All VAS ratings were recorded by pen and paper and subsequently scored and entered twice.
Energy balance
Energy intake (EI) for each food item was calculated as follows: EI = total grams of the food item consumed * calories present in 1 g of that food item. For items that had a label, nutrition information was taken from the Nutrition Facts. For items without a label (e.g., fresh apples) the Nutrition Data System for Research was used. EI for each meal and snack were obtained by summing the calories of all food items consumed during each meal, including calories from the preload shake, and snack period, respectively. Calories from each day’s meals (n = 3) and snacks (n = 3) were summed for daily EI. Weekly EI was the sum of calories from all meals (n = 15) and snacks (n = 15) during the treatment period. Total daily energy expenditure was the sum of daily PAEE and BMR, while total weekly energy expenditure was the sum of daily PAEE and BMR*5. Energy Balance (EB) was calculated as the net difference between measured total daily energy intake and total daily energy expenditure per day over the treatment period.
Statistical analysis
Based on a previous study [25], our sample size (n = 48) had 80% power to detect a significant difference in energy balance between the two conditions at the 0.05 2-tailed alpha level. The crossover design provided power to detect treatment effects explaining as little as 8% of the variance in energy balance. Due to the crossover design, outcomes were assessed using linear mixed effects models, adjusting for caloric dose and including subject as a random effect. The primary analysis followed the intent-to-treat principle (n = 48). The secondary analysis included all subjects who completed the intervention (n = 43) and one subject who completed the first treatment period only (n = 44 total). Multiple imputation with 1000 imputations per analysis was used to account for missing data in accordance with the intent-to-treat principle [29]. While this is a high number of imputations, modern computation obviated any need to be frugal in this regard. Missing data in the secondary analyses were also treated with multiple imputation. The smoothed TEF curves of Fig. 4 are based on loess local polynomial regression [30]. The primary and secondary analyses were performed using R version 3.1.2 25, with specific use of the loess function and the lme4 package.
A post hoc analysis sought to determine if the macronutrient content of the preload (protein vs. carbohydrate) impacted the macronutrient content of the major meal and snack items consumed ad libitum. During data processing for this analysis, we made several assumptions: If the final weight of a food item or a beverage was missing, we assumed that the participant consumed all the food or beverage provided; If the serving weight was missing the average served weight was used; When the served weight for a beverage was missing, average weight was calculated based on 10 random selections of each beverage consumed by random participants and if 10 random selections were not found, the average of as many as were recorded was taken. If the type of beverage was missing, the average caloric value of all the beverages provided during the study was used. Along with the four participants who were excluded from the secondary analysis, the participant who did not attend the second treatment period was excluded from this analysis. Furthermore, we considered data for the last day of one participant as missing since the individual took unreasonably large amounts of food without recorded consumption amounts causing that data point to be an outlier. The difference in ad libitum carbohydrate and protein intake between the two treatment periods was calculated. Linear mixed models with repeated measures including treatment period, time, and caloric dose as fixed factors were used to study the differences in ad libitum protein and carbohydrate intake. Percent of daily calories from protein and carbohydrate was calculated as: (daily calories from macronutrient / total daily calories)*100. Student’s paired t-test compared treatment periods. This post hoc analysis was conducted using IBM SPSS software version 25 and Microsoft Excel.