Out of the 219 individuals screened, 54 patients with type 2 diabetes (with disease duration of more than 1 year) treated by oral hypoglycaemic agents (both men and women), age 30–70 years, BMI 27–50 kg/m2 and HbA1c 6–11.8% (42–105 mmol/mol), met all the inclusion criteria, gave their written informed consent and underwent randomisation. Exclusion criteria comprised alcohol or drug abuse, pregnancy or lactation, unstable medication or weight in the last 3 months, a diagnosis of type 1 diabetes and the presence of a cardiostimulant.
We used a randomised crossover study design. The study protocol was approved by the Institutional Ethical Committee. In a single-centre study, after a 1 month run-in period (when the patients learned how to write their food diaries and use the pedometers and glucometers), the participants began a 12 week regimen of either six (A6) or two (B2) meals a day. The A6 regimen consisted of three main meals (breakfast, lunch and dinner), and three smaller snacks in between. The B2 regimen consisted of breakfast (eaten between 06:00 and 10:00 hours) and lunch (eaten between 12:00 and 16:00 hours). The regimens were switched for the subsequent 12 weeks. All measurements were performed at weeks 0 (baseline), 12 and 24 (Fig. 1 and Table 1).
The composition of the diet in both regimens followed the Study Group on Diabetes and Nutrition of the European Association for the Study of Diabetes guidelines  with the same caloric restriction: a restriction of 2,092 kJ/day (500 kcal/day) based on the measurement of each individual’s resting energy expenditure (REE) by indirect calorimetry (metabolic monitor VMAX; SensorMedics, Anaheim, CA, USA) . Individual calculations of energy requirements for both regimens were based on the formula: (REE × 1.5) − 2,092 kJ. The diet derived 50–55% of its total energy from carbohydrates, 20–25% from protein and less than 30% from fat (≤7% saturated fat, less than 200 mg/day of cholesterol), with 30–40 g/day of fibre. Alcoholic beverages were limited to one per day for women and two per day for men. Participants were asked not to alter their exercise habits during the study. Each regimen started with a 4 day tutorial where they learned in detail how to compose and prepare their diet, with follow-up 1 h weekly meetings with lectures and cooking classes throughout the whole study. All the meals during the entire 24 weeks of the study were provided for one half of the participants (randomised within each study arm with an equal number of participants) while the other half of the participants prepared their meals by themselves.
At weeks 0, 12, and 24, a 3 day dietary record (2 weekdays and 1 weekend day) was completed by each participant. A registered dietitian analysed all these dietary records using a country-specific food-nutrient database NutriDan 1.2 (www.institut-danone.cz/cz/odborna-sekce/nutridan).
This was assessed with an Omron HJ-720IT pedometer (Omron, Kyoto, Japan; using a 1 month average step count for evaluation) and two questionnaires: the International Physical Activity Questionnaire  and the Baecke questionnaire  at weeks 0, 12, and 24.
Participants were asked to continue their pre-existing medication regimens, except when hypoglycaemia occurred repeatedly (fasting plasma glucose determined at the laboratory <4.4 mmol/l or a capillary glucose reading <3.4 mmol/l accompanied by hypoglycaemic symptoms). In such cases, medications were reduced by a study physician following the medication protocol. All participants were given an Accu-Chek Performa glucometer (Roche, Basel, Switzerland) and instructed how to use it.
All measurements were performed on an outpatient basis at weeks 0, 12 and 24, after a 10–12 h overnight fasting with tap water ad libitum. Height and weight were measured using a periodically calibrated scale accurate to 0.1 kg. Waist circumference was measured with a tape measure placed at the midpoint between the lowest rib and the upper part of the iliac bone. Blood pressure and heart rate were measured after 5 min in a seated position at rest, using a digital M6 Comfort monitor (Omron, Kyoto, Japan). Three measurements were taken 2 min apart. The first measurement was discarded, and the mean of the remaining two measurements was recorded.
Gas exchange measurements were taken during a 45 min basal period before the clamp. Air flow and O2 and CO2 concentrations in expired and inspired air were measured by a continuous open-circuit system (metabolic monitor VMAX; SensorMedics, Anaheim, CA, USA).
Plasma concentrations of glucose, immunoreactive insulin and C-peptide were measured at 0, 30, 60, 120 and 180 min after a standard breakfast (1,895 kJ, 45% carbohydrates, 17% proteins, 38% lipids). Insulin secretion and whole-body insulin sensitivity were calculated by mathematical modelling (described below).
Hyperinsulinaemic isoglycaemic clamp
The hyperinsulinaemic (1 mU kg−1 min−1) isoglycaemic clamp, lasting 3 h, was conducted as previously described . Insulin sensitivity was estimated as the metabolic clearance rate of glucose (MCR) .
Proton magnetic resonance spectroscopy
HFC was measured by proton magnetic resonance spectroscopy on a 3 T MR scanner (Magnetom Trio, Siemens, Erlangen, Germany) with an eight-channel body array coil. This method has been validated at our institution . The measurement protocol included conventional MRI using a localiser and HASTE sequence with breath-holding in the coronal and transversal planes. Spectra were obtained from three different segments of the right lobe of the liver—volume of interest, 30 ml each and evaluated using the LCModel (www.s-provencher.com/pages/lcmodel.shtml) and MestReC (Mestrelab Research, Santiago de Compostela, Spain) programs. The signal intensities of water and hepatic lipids were used to determine the fat to total signal peak area ratio and then converted to absolute concentrations expressed as a percentage of fat using equations validated by Longo et al . Fourteen individuals did not undergo an HFC measurement due to the patient’s refusal, claustrophobia or the patient’s weight exceeding the limit of the equipment.
Modelling analysis of beta cell function was performed during standard meal tests. Insulin secretory rates were calculated from plasma C-peptide levels by deconvolution  and expressed per square meter of estimated body surface area. The dependence of insulin secretory rates on glucose levels was modelled separately for each patient and each study day. The beta cell model used in the present study, describing the relationship between insulin secretion and glucose concentration, has previously been described in detail [23–25].
Briefly, insulin secretion consists of two components. The first component represents the dependence of insulin secretion on absolute glucose concentration at any time point and is characterised by a dose–response function. Characteristic variables of the dose–response are insulin secretion at a fixed glucose concentration and the mean slope in the observed glucose range. The dose–response was modulated by a potentiation factor that accounts for several agents (prolonged exposure to hyperglycaemia, non-glucose substrates, gastrointestinal hormones and neurotransmitters). The potentiation factor was set to be a positive function of time and to be an average of 1 during the experiment. It thus expresses a relative potentiation of the secretory response to glucose.
The second insulin secretion component represents a dynamic dependence of insulin secretion on the rate of change of glucose concentration. Termed the derivative component, it is described by a single variable, rate sensitivity. This secretion component is related to early insulin release [23, 24].
The model variables (the variables of the dose–response, the rate sensitivity and the potentiation factor) were estimated from the glucose and C-peptide concentrations by regularised least squares, as previously described [23, 24]. Estimation of the individual model variables was performed blinded for the randomisation of the patients for treatment.
Whole-body insulin sensitivity was estimated in two ways: (1) as the MCR calculated during the last 20 min of the isoglycaemic hyperinsulinaemic clamp after correction for changes in glucose pool size , and (2) by a glucose–insulin model to derive an oral glucose insulin sensitivity (OGIS) index, validated against the clamp data .
Serum glucose was analysed using the Beckman Analyser glucose-oxidase method (Beckman Instruments, Fullerton, CA, USA). Plasma immunoreactive insulin and C-peptide concentrations were determined using insulin and C-peptide IRMA kits (Immunotech, Prague, Czech Republic). HbA1c was measured by HPLC (Tosoh, Tokyo, Japan). Plasma concentrations of glucagon were measured using ELISA kits (BioVendor, Brno, Czech Republic). Plasma lipids concentrations were measured by enzymatic methods (Roche, Basel, Switzerland). HDL-cholesterol was measured after double precipitation with dextran and MgCl2. LDL-cholesterol was estimated using the Friedewald equation if the triacylglycerol concentration was <4.53 mmol/l.
The intention-to-treat analysis included all participants. We tested the distributions of the data. If the distribution was skewed, we used the Box-Cox transformation to attain data symmetry and homoscedasticity . Non-homogeneities in the data were detected using residual analysis as described elsewhere . 2 × 2 crossover ANOVA was used for data evaluation. The model consisted of the between-subject factor ‘sequence’, the factor ‘subject’ and within-subject factors of ‘period’ and ‘treatment’. In a subsequent subanalysis, the factor for prepared meals that were collected by patients was added. The relationships between continuous variables were evaluated using Pearson’s correlation and BMI-adjusted partial correlations.