Trial design and participants
We completed a parallel-groups randomised trial with participants allocated (1:1:1) to one of three groups: two groups who performed exercise training either in the morning (morning exercise [EXam] group) or in the evening (evening exercise [EXpm] group); and a third group who did not undertake any exercise training (control group [CON]) but consumed the same diet throughout the intervention as participants who exercised. Participants were recruited via social media and community advertisements. Eligibility criteria were as follows: male sex; aged 30–45 years; BMI 27.0–35.0 kg/m2; and sedentary lifestyle (<150 min/week moderate-intensity exercise for >3 months and sitting for >5 h each day). Potential participants were excluded if they met any of the following criteria: known CVD or type 2 diabetes; major chronic illness that impairs mobility or eating/digestion; taking prescription medications (i.e. β-blockers, anti-arrhythmic drugs, statins or insulin sensitising drugs); previous bariatric surgery; shift-work; smoking; strict dietary intake regimes (i.e. vegan, avoidances of principal study foods, not regularly consuming a breakfast meal, not regularly consuming three meals per day, actively trying to lose weight); or not being weight stable (±5 kg) for the last 3 months. Data were collected at the St Patrick’s (Fitzroy, VIC) campus of Australian Catholic University. The trial was approved by the Human Research Ethics Committee of the Australian Catholic University (2016-254H) and was registered with the Australian New Zealand Clinical Trials Registry (registration no. ACTRN12617000304336). Eligible participants received written information and provided informed consent before participation. The allocation sequence for participant randomisation was generated using a computer random number generator developed and administrated at the Unit for Applied Clinical Research at the Norwegian University of Science and Technology. The randomisation had varying block sizes, with the first, the smallest and the largest block defined by the computer technician. The researcher enrolling the participants (TM) received the allocation results onscreen and by email after registration of each new participant into the study and was unaware of the size of the blocks.
Experimental protocol and interventions
Before commencement of dietary or exercise interventions, participants recorded their habitual intake of food and liquid in a food diary for three weekdays and one weekend day to allow for comparison with the macronutrient composition and energy intake of the study diet. Habitual dietary intake in the pre-intervention period was analysed using FoodWorks (Version 8; Xyris, QLD, Australia). The first 5 days of the investigation were the same for all participants and consisted of the introduction of an HFD while they remained sedentary (Fig. 1). For each of the 11 days of the HFD, participants received three pre-packed meals (breakfast, lunch and dinner), each containing 33.3% of the total daily energy intake. The macronutrient content of meals was 65% of energy from fat, 15% from carbohydrate and 20% from protein, with individualised total energy intake based on measurements of resting metabolic rate (detailed below). Participants were instructed to consume meals at prescribed times (07:30, 13:00 and 19:30 hours) and to abstain from alcohol. Water and coffee/tea (without sugar and milk) could be consumed ad libitum.
After the 5-day dietary harmonisation period, all participants consumed the HFD for a further 6 days at the prescribed times and with the same restrictions on liquids. On days 6–10, participants allocated to an exercise group performed once-daily exercise training at either 06:30 hours (EXam group) or 18:30 hours (EXpm group). Participants allocated to the control group (CON) did not perform any exercise but maintained their normal activities of daily living while remaining on the HFD (Fig. 1). Exercise training was performed on an electronically braked cycle ergometer (Lode Excalibur Sport, Lode, Groningen, the Netherlands). On days 6, 8 and 10, participants performed HIT (10 × 1 min work-bouts at 95–120% of individual peak power output (PPO), separated by 1 min low-intensity cycling). On days 7 and 9, participants in the exercise groups undertook a continuous bout of moderate-intensity cycling (60–65% of PPO) for 40 and 60 min, respectively. Participants allocated to an exercise training group received an extra 419 kJ snack (65% fat, 15% carbohydrate, 20% protein) following each exercise session.
Assessments and outcomes
Prior to intervention, participants underwent the following procedures: (1) a dual-energy x-ray absorptiometry (DXA) (GE Lunar [USA] iDXA Pro, encore software Version 16) scan to estimate body composition; (2) measurement of resting energy expenditure (TrueOneRMR; Parvo Medic, Sandy, UT, USA) to estimate daily energy requirements; (3) girth measurements of waist and hip; (4) measurement of seated systolic and diastolic BP and resting heart rate using an automated oscillometric blood pressure monitor (Welch Allyn, NY, USA); and (5) a maximal cycling test with measurement of expired gases (Parvo Medic, Sandy, UT, USA) to determine peak oxygen uptake (\( \dot{V}{\mathrm{O}}_{2\mathrm{peak}} \)) and PPO. Apart from resting energy expenditure, all assessments were repeated after the intervention period, with the \( \dot{V}{\mathrm{O}}_{2\mathrm{peak}} \) test and the DXA scan undertaken 48–72 h after participants had completed the 11 days of the HFD.
Blood samples (18 ml) were drawn using a single forearm venepuncture before breakfast and after dinner, at baseline (day 0), after 5 days on the HFD (day 5) and after a further 5 days of the HFD, either with daily exercise or no exercise (day 11). None of the participants exercised on day 11 (Fig. 1). Participants were fitted with a continuous glucose monitor (CGM) system (iPro2 with Enlite sensor; Medtronic, Dublin, Ireland) placed on the lower back adipose site, and worn throughout the study. We calculated hourly values and defined daytime glucose as the mean glucose concentration from 06:00 hours to 22:00 hours and nocturnal glucose as the mean glucose concentration from 22:00 hours to 06:00 hours.
Participants wore an activPAL inclinometer (activPAL3TMtri-axial physical activity monitor; PAL-technologies, Glasgow, UK) on the thigh and an ActiGraph accelerometer (ActiGraph GTX3+; Pensacola, FL, USA; during waking hours only) around the waist to estimate physical activity and movement patterns. Additionally, a SenseWear armband (Bodymedia, Pittsburgh, PA, USA) was worn on the upper arm for estimates of energy expenditure.
Biochemical analyses
Total cholesterol, HDL-cholesterol, LDL-cholesterol and triacylglycerol concentrations were analysed in whole blood using Cobas b 101 (Roche Diagnostics, Basel, Switzerland) and plasma glucose concentration using the hexokinase methods with a YSI 2900 STAT Plus (YSI Life Sciences, Yellow Springs, OH, USA). Serum insulin concentrations were determined in duplicate using ELISA (ALPCO, Salem, NH, USA). HOMA-IR was calculated according to the following formula: fasting serum insulin (μU/ml) × fasting plasma glucose (mmol/l) / 22.5 [11].
Serum metabolomics
Metabolomics analysis was undertaken by Metabolon (Durham, NC, USA) [12]. Thawed serum was methanol extracted and analysed using ultra-HPLC–tandem MS (UPLC-MS/MS) positive mode, UPLC-MS/MS negative mode and GC-MS. The UPLC-MS/MS platform used a Waters Acquity UPLC with Waters UPLC BEH C18–2.1× 100 mm, 1.7 mm columns and a ThermoFisher LTQ MS, including an electrospray ionisation source and a linear ion-trap mass analyser. Samples for GC-MS analysis were dried under vacuum desiccation for a minimum of 18 h prior to being derivatised using bis(trimethylsilyl) trifluoroacetamide. Using helium as carrier gas and at a temperature ramp from 60 °C to 340 °C within a 17 min period, derivatised samples were separated on a 5% phenyldimethyl silicone column. Samples were analysed on a Thermo-Finnigan Trace DSQ fast-scanning single-quadrupole MS operated at unit mass resolving power with electron impact ionisation and a 50e750 atomic mass unit scan range. Metabolites were identified by automated comparison of the ion features in the experimental samples to a reference library of chemical standard entries. The reference library included retention time, molecular mass (m/z), preferred adducts and in-source fragments, as well as associated MS spectra, and were curated by visual inspection for quality control using software developed at Metabolon [13].
Statistical methods
A formal sample size calculation was not undertaken due to the exploratory nature of the research question. Data are expressed as means (±SD) and estimates with 95% CIs. We used paired t tests to assess differences in traditional biomarkers of cardiometabolic health between baseline and after 5 days of HFD, for the entire cohort. To determine between-group differences in traditional biomarkers we used mixed models with participant as random effect, and with two dummy variables to uniquely identify the two exercise groups (EXam and EXpm) and their interactions with time (categorical) as fixed effects. We assumed no systematic differences between groups at baseline or after 5 days of HFD. The analyses were repeated to test for differences between the EXam and EXpm groups. We adjusted for baseline values for each outcome variable as recommended by Twisk et al. [14]. For the traditional clinical biomarkers, p values <0.05 were considered statistically significant (without any adjustment for multiple comparisons).
Metabolomics analysis
The raw metabolomic data consisted of 897 quantified metabolites. For the data analysis, the original scale was used, with values normalised in terms of raw area counts by Metabolon. Metabolites with >30% missing values (n = 105, of which 93 were xenobiotics) were removed prior to analysis, as was lidocaine, resulting in 791 metabolites. Values below the limit of quantification were imputed with a value equal to half of the lowest detected value of the corresponding metabolite. Principal component analysis (PCA) was undertaken on serum metabolites to assess the presence of a systematic trend in the change of the metabolic profiles of the participants during the harmonisation period (habitual diet vs HFD), and to assess differences in metabolic profiles induced by exercise training. PCA was performed using PLS_Toolbox 8.7.2 (Eigenvector Research, Wenatchee, WA, USA) in Matlab R2017b. Percentage changes of the metabolite concentrations as induced by the HFD were calculated. The significance of the change between habitual diet and HFD was determined by paired t tests. Mixed models were used to test between-group differences while adjusting for baseline values (metabolite values at day 5, prior to the exercise intervention), with participant as random effect, time and time × group interactions as fixed effects, and metabolite concentration as the dependent variable. Separate models were built for comparison of EXam vs CON, EXpm vs CON, and EXam vs EXpm. The p values were adjusted using the Benjamini–Hochberg procedure [15] and significance was considered for q values <0.05. Analyses were performed on the morning and evening samples separately.
Heatmaps within each class of metabolites (lipids, amino acids, carbohydrates, peptides, co-factors and vitamins, energy, nucleotides, and xenobiotics) were ordered by hierarchical clustering using Spearman correlation as a similarity measure and average linkage as a distance measure. Clustering was performed on morning samples, and the same metabolite order was used for evening samples.