Study design and sample size

Twenty-five men and women (Table 1) participated in an open-label (participants blinded to primary outcome but not intervention), randomized, crossover design. Participants completed 1 week of habitual diet and physical activity monitoring, before completing 3 laboratory visits in a random order, during which they standardised diet for 24-h prior, then received breakfast in the laboratory with the 4-h postprandial responses measured, before they received an ad libitum lunch. Participants left the laboratory for the remainder of the day immediately following lunch, were provided with ad libitum dinner to consume outside of the laboratory, and returned the following morning. No restrictions were placed on free-living physical activity outside the laboratory, and participants did not know this was the primary outcome as confirmed using an exit questionnaire during the final visit. A schematic of the study design is presented in Fig. 1. The study was approved by the Research Ethics Approval Committee for Health at the University of Bath (EP 17/18 78) and all measures were conducted in accordance with the Declaration of Helsinki with participants providing written informed consent. The study was registered at clinicaltrials.gov (NCT03509610). Trial order randomization was completed using randomizer.org. Inclusion criteria were body mass index 18.5–29.9 kg∙m−2, age 18–65 years, and no anticipated changes in diet and physical activity during the study (e.g., holidays or diet plans). Exclusion criteria were any reported condition or behaviour that might pose undue personal risk or introduce bias, diagnosed metabolic disease (e.g. type 2 diabetes), lifestyle not conforming to standard sleep–wake cycle (e.g. shift worker), and any reported change in body mass greater than 3% in the previous 6 months [21].

Table 1 Participant characteristics at time of preliminary measures Full size table

Fig. 1 Schematic of study design. RMR, resting metabolic rate; MODSUG, moderate-sugar diet; LOWSUG, low-sugar diet; LOWCHO, low-carbohydrate diet Full size image

The required sample size was estimated based on the Bath Breakfast Project [8] using G*Power 3.1 software [22]. Mean ± SD physical activity energy expenditure (PAEE) for the fasting vs. breakfast groups during the morning (when differences in carbohydrate availability between groups were present) were 311 ± 124 kcal vs. 492 ± 227 kcal. Based on this effect size (d = 0.998), a two-tailed matched-pairs design with 15 participants would provide > 90% chance (power) of detecting the stated effect with an α-level of 0.05. Due to a technical failure, data loss occurred for the first 10 participants for the primary outcome measure. As stated in the original protocol, a rolling recruitment to achieve the sample size of 15 for primary outcome measure (24-h physical activity energy expenditure) was continued which resulted in n = 15 for the primary outcome measure but a total sample of n = 25. The study team were unable to cannulate one participant due to small vasculature so blood samples could not be obtained. The higher sample size than initially planned enabled exploratory analyses of sex differences in response to the different diets.

Diets

We investigated 3 diets (Fig. 2). The moderate-sugar (MODSUG) diet was designed to be reflective of macronutrient and sugar intake in European populations [1, 23], the low sugar (LOWSUG) diet was designed to meet UK public health guidelines that advocate reducing free sugar intake to < 5% of total energy intake [10, 24], and the low-carbohydrate (LOWCHO) diet was designed to restrict carbohydrate availability and promote ketogenesis, consistent with the definition of a ‘very low-carbohydrate ketogenic diet’ [25]. Estimated caloric values for each nutrient were used to calculate energy intake: carbohydrates 3.75 kcal g−1, sucrose 3.94 kcal g−1, fat 8.94 kcal g−1, protein 4.02 kcal g−1, and alcohol 6.93 kcal g−1 [26]. The UK labelling system currently requires the reporting of ‘total sugars’ rather than ‘free sugars’ [27], therefore whilst we have aimed to manipulate free sugars between the diets, it can only assumed that most of the sugars in the present study are free sugars, so we refer to ‘sugars’ throughout. A description and nutritional information of the breakfast meals given to participants to consume in full within the laboratory is provided in Table 2. A description and nutritional information of the lunch and dinner meals given to participants to consume ad libitum within the laboratory (lunch) and outside of the laboratory (dinner) is provided in Table 3. Photographs of meals provided to participants are shown in Supplemental Fig. 1. The energy content of the breakfast meal was calculated on the first laboratory visit and replicated in subsequent trials. We aimed to provide 20% of total energy requirements as a typically-representative breakfast intake [28] and factor in confinement to the laboratory during the testing phase by using measured resting metabolic rate (RMR) and measured habitual physical activity energy expenditure (PAEE). To achieve this, we estimated energy requirements by combining 8 h of resting metabolic rate (RMR) for sleep, 6 h of RMR for the laboratory component of the trial (14 h total), and combined resting and physical activity energy expenditure (PAEE) for the remainder of waking hours (10 h) using the following equation:

$${\text{Total}}\,{\text{energy}}\,{\text{requirements}}\,\left( {{\text{kcal}}} \right) = \left[ {\left( {\frac{{{\text{RMR}}\left( {{\text{kcal}}} \right)}}{24}} \right) \times 14} \right] + \left\{ {\left[ {\left( {\frac{{{\text{RMR}}\left( {{\text{kcal}}} \right)}}{24}} \right) + \left( {\frac{{{\text{PAEE}}\left( {{\text{kcal}}} \right)}}{16}} \right)} \right] \times 100} \right\}.$$ (1)

Fig. 2 Dietary macronutrient intake as a percentage of total energy intake for habitual diet and each of the three experimental conditions. MODSUG moderate-sugar diet, LOWSUG low-sugar diet, LOWCHO low-carbohydrate diet, CHO carbohydrates, EtOH, alcohol. For habitual mean and SD are shown Full size image

Table 2 Description and nutritional information of each breakfast meal Full size table

Table 3 Description and nutritional information of each lunch/dinner meal Full size table

Lunch and dinner were ad libitum with 2000 kcal prepared for each meal. All lunch and dinner meals were prepared the evening before the trial day and refrigerated overnight and all bread was refrigerated but not frozen, as these practices influence resistant starch production and glycaemic responses to the ingested carbohydrates [29, 30]. Palatability of the breakfast and lunch meals was assessed by asking participants to strike a line through a 0–100 mm scale (0 = bad, 100 = good) shortly following ingestion of the meal. A palatability score was calculated by combining mean scores for ‘Visual Appeal’, ‘Smell’, ‘Taste’, and ‘Palatability’.

Preliminary measures

Participants attended the laboratory for eligibility screening and a treadmill walk to calibrate the physical activity monitors. Then, participants completed 7 days of habitual lifestyle monitoring, which comprised a weighed food diary and wearing a combined accelerometer and heart rate monitor to measure free-living physical activity energy expenditure (Actiheart 4™, CamNtech Ltd., UK). Participants were provided with weighing scales to weigh food items (SmartWeigh, China) and food diaries were analysed using diet analysis software (Nutritics, Ireland).

Laboratory visit standardization

Participants chose a menu aligned to the macronutrient intake of the MODSUG diet and were provided with food, weighing scales, and a physical activity monitor for 24 h leading into trial days. Participants were asked to record actual intake for 24 h leading into the first trial and replicate before this each subsequent trial, which was confirmed by writing on a printed menu. Participants were also asked to refrain from strenuous physical activity in the 24 h before each laboratory visit. Median (interquartile range) time between main visits was 13 (7–21) days for males and females using oral contraception, and 28 (28–34) days for menstruating females.

Laboratory visits

Participants arrived at the laboratory following an overnight fast (duration 11:38 ± 00:57 hh:mm). They were asked to consume a glass of water and take an inactive transport mode rather than walk or cycle to the laboratory. Anthropometric measures of height, body mass, waist circumference, and hip circumference were obtained, and body fat percentage was estimated using digital scales (Tanita, Japan). Resting metabolic rate was measured using the Douglas bag technique by averaging three 5-min gas samples, with guidelines for best practice followed [31]. Expired fractions of O 2 and CO 2 were determined via paramagnetic and infrared analysers (Mini HF 5200, Servomex Group Ltd., UK), respectively, and the volume expired was measured using a dry gas meter (Harvard Apparatus, UK). Inspired O 2 and CO 2 were measured concurrently to account for ambient fluctuations [32]. Energy expenditure and substrate oxidation in the postprandial period was calculated using stoichiometric equations [33, 34], assuming urinary nitrogen excretion was negligible.

A cannula (BD Venflon™ Pro, Becton Dickenson & Co., Sweden) was inserted into a hand vein or antecubital forearm vein if hand cannulation was unsuitable and the arm was placed in a heated box (University of Vermont, USA) set to 55 °C to arterialize venous blood [35]. Participants completed a computer-based food preference task, which consisted of 18 plates of food individually photographed on a white plate or transparent bowl. Two foods were placed side-by-side and the participant selected which food they would ‘choose to eat right now’. Foods were distinguished into three categories: sweet high-carbohydrate foods, non-sweet high-carbohydrate foods, non-sweet low-carbohydrate foods which were matched at six levels for energy density and content. The task determines relative preference for these 3 food categories at each measurement. Participants completed 0–100 mm visual analogue scales to measure appetite; marking a line through the scale relating to how hungry, full, or thirsty they were and how strong their desire for sweet, savoury, rich, or creamy food with 0 corresponding to ‘not at all’ and 100 corresponding to ‘extremely’. A baseline blood sample was obtained, and the cannula was flushed with 10 mL sterile NaCl 0.9% (B. Braun, Pennsylvania, USA) to maintain patency throughout the trial (repeated at each blood sample). Participants were provided with breakfast at 09:12 ± 00:19 hh:mm and asked to ingest the whole meal within 15 min. A timer was started upon ingestion of the first bite of breakfast and metabolic responses were measured for 4 h. Blood samples were taken at 15-min intervals for the first hour and then every 30 min thereafter. A five-minute expired gas samples was collected within the final 10 min of each hour. Visual analogue scales were repeated hourly. Immediately following the 4-h postprandial period, participants completed a second computer task for food preferences with the hand still being heated. Following this the cannula was removed and participants were served with the lunch meal. Participants remained in the bed whilst bowls of the lunch meal were served to them in ~ 500 kcal portions. They were asked to eat until they were comfortably full. Food was served at 52.8 ± 3.7 °C (mean ± SD). Bowls were replaced at random time intervals with the aim that participants did not finish a portion and so could not estimate the quantity consumed.

Participants left the laboratory following lunch and returned the following morning using a similar mode of transport. Food was provided ad libitum between laboratory visits with no constraints on free-living energy expenditure. Participants were instructed to eat only the food provided and to drink only water for the rest of the day. Upon arrival the following morning, participants completed a visual analogue scale, a food preference task, and a 5-mL arterialized venous blood sample was obtained from an antecubital vein.

Physical activity energy expenditure

Physical activity energy expenditure was measured using branched-equation modelling of heart rate and accelerometry across 24 h. Physical activity monitors were individually calibrated for each participant by using a treadmill protocol modified from Brage et al. [36]; participants attended the laboratory following a minimum 5 h fast and walked at a speed of 5.2 km h−1 for 20 min on a treadmill with incremental 5-min stages at gradients 0, 3, 6 and 9%. Expired gas samples were obtained in the final minute of each stage and analysed via indirect calorimetry to measure energy expenditure. Heart rate was obtained during the final minute of each stage using a chest-worn monitor (Polar Electro, Finland). Sleeping heart rate was measured during the 7 days of preliminary wear time. Mean resting metabolic rate from laboratory visits was entered as energy expenditure at sleeping heart rate. A linear model was fitted for energy expenditure at a range of heart rates from sleeping heart rate to the final stage of the treadmill walk and was extrapolated beyond this point for higher intensity activity. Thresholds for physical activity intensities were defined and calculated for each participant as sedentary < 1.5 METs, light ≥ 1.5–< 3.0 METs, moderate ≥ 3.0–< 6.0 METs, vigorous ≥ 6.0–< 10.2 METs, and very vigorous ≥ 10.2 METs [37, 38].

Blood sampling and analyses

Blood samples were collected into tubes containing clotting activator (Sarstedt, Germany) and left at room temperature for 15 min before being centrifuged at 4000 × g for 10 min at 4 °C. Serum was aliquoted in duplicate into sterile tubes, placed on dry ice, and stored at − 80 °C. Serum glucose, triglycerides (TAG), glycerol, non-esterified fatty acids (NEFA), lactate, beta-hydroxybutyrate (βOHB), total cholesterol, high-density lipoprotein cholesterol (HDL-c), and low-density lipoprotein cholesterol (LDL-c) concentrations were measured using an automated analyser (RX Daytona, Randox Laboratories, UK). Reported TAG values in the present paper have been blanked for glycerol based on recommendations for clinical research [39]. Serum insulin and leptin were measured using enzyme-linked immunosorbent assay (ELISA) kits (Mercodia AB, Sweden). Fibroblast growth factor 21 (FGF21) was measured using a U-plex electro-chemiluminescent kit (U-Plex, Mesoscale Discovery, USA). Incremental area under the curve (iAUC) or total area under the curve (tAUC) for postprandial responses were calculated with the trapezoid method using the Time Series Response Analyser [40]. Inter-assay coefficients of variation were < 3% for glucose, < 2% for TAG, < 6% for glycerol, < 7% for NEFA, < 3% for lactate, < 6% for βOHB, < 4% for total cholesterol, < 5% for HDL-c, < 6% for LDL-c, < 7% for insulin, < 6% for leptin, and < 3% for FGF21. All samples for a participant were measured on the same run or plate. Samples producing values below the lower limit of detection were assigned the value of the lower detectable concentration, which was necessary for some samples with βOHB, insulin, and leptin.

Statistical analyses

Descriptive statistics were calculated using Microsoft Excel (Microsoft, Washington, USA). GraphPad Prism was used for other statistical analyses and producing figures (GraphPad Software Inc., California, USA). The distribution of residuals was checked using Shapiro–Wilk tests and visual inspection of residual plots. Single-variable outcomes were analysed using one-way repeated measures ANOVA with post-hoc Bonferroni corrections applied. Outcomes with multiple time-points for each condition were analysed using two-way repeated measures ANOVA or mixed-effects models (depending on missing data points) to detect significant time, condition, or time x condition interactions, with post hoc Bonferroni corrections applied. Pearson correlation coefficients were used to assess linear associations between βOHB and NEFA, and LDL-cholesterol and NEFA across 4 and 24 h in the LOWCHO condition. The larger sample size than originally planned offered the opportunity to perform post-hoc tests exploring sex differences in physiological outcomes. Firstly, we ran two-way ANOVA to detect significant sex x condition effects for summative outcomes (iAUC and tAUC across 4 or across 24 h for outcomes with only 3 time points) and secondly, we disaggregated data by sex and ran the same analyses as the whole cohort to identify changes in interpretation compared to the whole sample. Significance was accepted at P ≤ 0.05. Data are presented as mean and 95% confidence intervals (CI) unless otherwise stated.