Study design and protocol overview

The study was a randomised, controlled study, comprised of two parallel dietary intervention groups: a TRF intervention group (three-hour contraction of habitual eating window) and a control (no intervention) group. The groups were balanced for age, BMI and gender. The dietary intervention period was 10 weeks, with study measurements (anthropometry and biochemistry) taken before and after. Food diaries were completed at baseline and final (tenth) week of the intervention. The study was conducted between September and November 2015.

5.3.2.1 Study design considerations

Due to inter-individual variation in habitual eating duration, participants were asked to decrease the duration of their eating window by a set amount relative to their habitual intake. A modest three-hour total temporal restriction was chosen to ensure acceptability of the TRF intervention, and because it was likely to have some effect based on findings from a previous TRF intervention

study which assessed anthropometric outcomes. This study found that a ~3-4 hour contraction in habitual eating window (comparable to the present study) led to a 3.3 kg reduction in body weight after 16 weeks in overweight/obese individuals (138).

5.3.2.2 Dietary interventions

5.3.2.2.1 Time restricted feeding intervention

Participants assigned to the TRF intervention were asked to alter their daily feed-fast cycle by delaying their first energy intake of the day and advancing their last energy intake of the day, each by 1.5 hours. This symmetrical compression of feeding duration avoided any bias of the intervention to one end of the day compared to the other. TRF participants were asked to maintain their usual physical activity patterns throughout the intervention period.

5.3.2.2.2 Control intervention

Participants assigned to the control intervention were asked to maintain their usual dietary and physical activity patterns throughout the intervention period.

5.3.2.3 Clinical visits

All participants initially undertook a one-week baseline period, during which they were required to maintain habitual feed-fast cycles. The timing and composition of energy intake was recorded in diet diaries over the final four days. At the end of the baseline period, participants attended the Clinical Investigation Unit (University of Surrey) for initial measurements. Participants were instructed to abstain from alcohol and strenuous exercise for 24 hours before the visit, and consumed their preceding evening meal before 20:00. Participants arrived at the unit at 08:00 following a 12-hour overnight water only fast. Body weight and composition were measured via bioimpedance (Section 2.1). Following this, a fasted blood sample was taken via venepuncture which was used to measure fasting glucose, insulin and lipid profiles. After the initial clinical visit, both groups commenced their respective dietary interventions whilst maintaining habitual activity patterns. At the end of the 10-week intervention period, participants returned to the

Clinical Investigations Unit for repeated measurements (pre- and post-intervention clinical visits were identical). Before the second clinical visit, participants received the same instructions given before the initial visit regarding alcohol and exercise restrictions. In addition, participants consumed the same evening meal as they did prior to the first clinical visit, as it has previously been shown that the macronutrient composition of an evening meal can affect metabolic responses on the following day (173).

5.3.3 Experimental techniques and analyses

5.3.3.1 Blood biochemistry

Blood samples were taken at each clinical visit via venepuncture. Please refer to Sections 2.5 for full description of blood collection, processing and laboratory techniques. Plasma samples were batch analysed upon study completion with all samples from an individual participant included in the same assay. Metabolites were analysed using the following methods: insulin using ELISA (Millipore, Billerica, MA, USA; intra-assay CV 5%); glucose, TAG, total and HDL cholesterol using commercially available kits (Instrumentation Laboratory, Warrington, UK) for the ILAB650 (Instrumentation Laboratory, Warrington, UK; intra/inter-assay CV all <10%). LDL- cholesterol was calculated using the Friedewald equation (Section 2.5.2.8). HOMA-IR was additionally assessed as a proxy for hepatic insulin sensitivity (Section 2.5.2.4).

5.3.3.2 Dietary analyses

Participants recorded intake in diet diaries for four days at two time points: at baseline (prior to starting their dietary intervention) and in the final (tenth) week of the intervention. Four-day intakes were then averaged. Recorded time of first and last energy containing food/drink was used to calculate each participant’s eating window. Target eating window was calculated by subtracting 180 minutes from an individual participant’s baseline eating window. Compliance in the final week of the intervention was assessed by comparing the reported eating window (recorded in the second diet diary) to the individual’s prescribed eating window. Please refer to Section 2.3 for a full explanation of methods relating to diet diary analyses.

5.3.4 Statistical analyses

Data were checked for normality using the Shapiro-Wilks test, with non-normally distributed data normalised where possible via log transformation to permit parametric testing (Section 2.7). The primary analysis was a one-factor ANCOVA between the dietary intervention groups with post- treatment values as the dependent variable, and baseline values of each parameter as the covariate. This is recommended statistical method (in terms of bias, precision and power) for the analysis of continuous outcomes in randomised studies with a single post-treatment measurement previously measured at baseline. An additional benefit of using the ANCOVA method is that it adjusts for the presence of baseline imbalances by including baseline values in the model (181). Prior to analyses, it was ensured that assumptions made by the ANCOVA model with regards to independence of the covariate and treatment effect and homogeneity of regression slopes (treatment x covariate interaction) were met (157). A significant treatment x covariate interaction existed for plasma glucose responses, and so a general linear model (GLM) was created which incorporated this interaction term. Using this GLM, further exploratory analyses were conducted to explore this interaction, and to predict treatment responses (estimated marginal means [EMM]) to the TRF or control interventions at the varying baseline levels of fasting glucose observed within the cohort. This analysis was conducted as the study was a pilot designed to inform future work, potentially within pre-T2DM and T2DM populations. To construct this model and obtain the EMMs, the following SPSS syntax was inputted, with rx representing treatment groups:

Unianova change by rx with baseline/print=parameter/emmeans=tables(rx)/ emmeans=tables(rx) with (baseline=3.5) / emmeans=tables(rx) with (baseline=4) / emmeans=tables(rx) with (baseline=4.5)/emmeans=tables(rx) with baseline=5) / emmeans=tables(rx) with (baseline=5.5) /

The Mann Whitney U test was used as the non-parametric alternative to ANCOVA to compare change scores between intervention groups. Differences between intervention groups at baseline were assessed using independent t-tests for continuous variables or the Chi squared test for categorical variables. No significant baseline differences were found. A paired t-test or non- parametric Wilcoxon signed-rank test were used to assess the change between baseline and post-

intervention values within each dietary intervention group. Results are presented as mean ± SEM.

5.3.5 Data omissions

One control participant did not complete their second diet diary, so dietary intake data are presented for the remaining 12 participants who completed both diaries.

5.4 Results

Of the 16 participants initially recruited into the study, three participants (TRF=2, Control=1) did not complete the study. Reasons for dropouts included faintness during blood collection at the first clinical visit (Control=1), travel across multiple time zones as well as change to habitual diet through participation in another research study (protocol violation, TRF=1), or loss of contact (TRF=1). The consort diagram is presented in Figure 5.1.

Baseline characteristics of the 13 study completers (TRF=7, Control=6) are reported in Table 5.1. The groups were matched for age, BMI, adiposity, gender and were all Caucasian.

Figure 5.1 Consort diagram for the time-restricted feeding study

TRF (n=7) Control (n=6) TRF vs. Control Age (years) 47 ± 3 45 ± 4 NS1 Gender (M/F) 1 / 6 0 / 6 NS2 Weight (kg) 86.2 ± 5.2 77.8 ± 7.6 NS1 BMI (kg/m2_{) 29.0 ± 1.7 28.6 ± 2.8 NS}1 Lean 1 / 7 3 / 6 NS2 Overweight 3 / 7 1 / 6 Obese 3 / 7 2 / 6 Body fat (%)* All 36.0 ± 2.9 34.6 ± 3.5 NS1

Males 21.9 n/a n/a

Females 38.4 ± 1.9 34.6 ± 3.5 NS1

Table 5.1 Baseline characteristics for study completers

*Bioimpedance.

Statistics and data presentation: 1 _{Unpaired t-test,} 2 _{Chi squared. NS –}

Non-significant. Presented as mean ± SEM.

Abbreviations: BMI – Body mass index; F – Female; M – Male; TRF –

Time-restricted feeding.