Individuals with type 2 diabetes (age 35–65 years, HbA1c 6.5–9.0% [48–75 mmol/mol], diabetes duration <4 years, stable BMI 25–45 kg/m2) were recruited. Participants were excluded if being treated with thiazolidinediones, insulin, steroids or beta-blockers, with a serum creatinine >150 mmol/l, with a serum alanine transaminase level >2.5-fold above the upper limit of the reference range, or if there were contraindications for MRI. Statin therapy was continued. The study protocol was approved by the Newcastle upon Tyne Ethics Committee No. 2, and all participants gave their informed consent. Sulfonylurea (two individuals) was discontinued 2 months, and metformin (seven individuals) 1 week, before the baseline study. Dietary adherence was assessed using capillary ketone levels (Xceed Optium; Abbott Diabetes Care, Maidenhead, UK). Three individuals failed to comply with the diet (two during the first week and one during weeks 4–8), and one left the study for an unrelated medical reason. Hence 11 individuals (nine male and two female, age 49.5 ± 2.5 years) completed the study.
Nine control participants matched for weight, age and sex were also studied (seven male, two female, age 49.7 ± 2.5 years; Table 1). These participants had no family history of diabetes, were taking no medication and had normal glucose metabolism as confirmed by a standard 75 g OGTT.
Participants were asked to continue their habitual pattern of eating until the start of the study. Assessments of beta cell function, insulin sensitivity, liver and pancreatic fat content and total body fat were carried out at baseline immediately prior to dietary intervention (day −1), and after 1, 4 and 8 weeks of the very-low-energy diet. A group of matched non-diabetic controls were studied on one occasion only, without dietary intervention.
After the baseline measurements, individuals with type 2 diabetes started the diet, which consisted of a liquid diet formula (46.4% carbohydrate, 32.5% protein and 20.1% fat; vitamins, minerals and trace elements; 2.1 MJ/day [510 kcal/day]; Optifast; Nestlé Nutrition, Croydon, UK). This was supplemented with three portions of non-starchy vegetables such that total energy intake was about 2.5 MJ (600 kcal)/day. Participants were provided with suggestions of vegetable recipes to enhance compliance by varying daily eating. They were also encouraged to drink at least 2 l of water or other energy-free beverages each day, and asked to maintain their habitual level of physical activity. Ongoing support and encouragement was provided by means of regular telephone contact. At the end of the 8 week intervention participants returned to normal eating but were provided with information about portion size and healthy eating.
The striking results seen at 8 weeks demanded experimental follow-up, and additional ethics permission was obtained to repeat the MRI studies and carry out OGTTs 12 weeks after completing the dietary intervention.
Hepatic glucose production and insulin sensitivity
After an overnight fast, cannulae were inserted into an antecubital vein for infusion and the contralateral wrist vein for arterialised blood sampling. [6′6′-2H]glucose (98% enriched; Cambridge Isotope Laboratories, Andover, MA, USA) was used to determine hepatic glucose production [12, 13], and basal rates were calculated during the last 30 min of the 150 min basal period. Preinfusion enrichment of isotope was insignificant throughout. An isoglycaemic–hyperinsulinaemic clamp (insulin infusion rate 40 mU m−2 min−1) was initiated at 0 min . Isoglycaemia was used to ensure that the true fasting condition of each participant could be observed at each study time point. Each participant was clamped at the glucose level observed at the end of the basal period. Whole-body insulin sensitivity was determined during the last 30 min of the hyperinsulinaemic glucose clamp as whole-body glucose disposal corrected for glucose space and urinary loss [14, 15]. Glucose metabolic clearance rates during steady-state conditions were calculated by dividing whole-body insulin sensitivity by steady-state plasma glucose.
Assessment of beta cell function
Sixty minutes after the clamp test, two consecutive 30 min square-wave steps of hyperglycaemia (2.8 and 5.6 mmol/l above baseline) were achieved by a priming glucose dose followed by variable 20% glucose infusion . Blood samples for determination of plasma glucose, insulin and C-peptide concentrations were obtained every 2 min for the first 10 min and every 5 min for the other 20 min of each step. An arginine bolus was administered during the second step of hyperglycaemia, followed by sampling every 2 min for 10 min. Insulin secretion rate was calculated using a computerised program implementing a regularisation method of deconvolution  and using a population model of C-peptide kinetics .
Magnetic resonance measurements
A Philips 3.0 T Achieva scanner and six-channel cardiac coil (Philips Healthcare, Best, the Netherlands) were used to acquire three gradient-echo scans with adjacent out-of-phase and in-phase echoes (time to repetition/time to echo/averages/flip angle = 50 ms/3.45, 4.60, 5.75 ms/1/30°, matrix 160 × 109, field of view 400–480 mm to suit participant size with 70% phase field of view). Six slices were acquired within a 17 s breath-hold to cover the liver with slice thickness 10 mm and the pancreas with slice thickness 5 mm. The data were analysed in MATLAB (MathWorks, Cambridge, UK) to produce separate fat and water images . The fat content of the image was expressed as a percentage of the total signal per voxel. The intraorgan fat percentage was evaluated from five liver regions of interest and two pancreatic regions of interest, defined and averaged in a blinded fashion by one observer (K. G. Hollingsworth). The pancreatic fat data represent solely intrapancreatic fat, and the liver data avoid contamination from blood vessels and gallbladder. This was achievable because of the data processing after image acquisition, allowing manual selection of wholly intraorgan volume from the anatomical slice. The interscan Bland–Altman repeatability coefficients were 0.5% for the liver and 0.9% for the pancreas.
Body composition and anthropometry
Percentage body fat was measured following an overnight fast using air-displacement plethysmography (BOD POD Express; Life Measurement, Concord, CA, USA). Waist and hip circumferences were measured with the participants in a relaxed standing posture. Waist circumference was taken at the mid-point between the anterior superior iliac spine and the lower edge of the rib cage, and hip circumference at the level of the greater trochanter. All measurements were made throughout the study period by a single observer (E. L. Lim).
Plasma glucose concentration was measured by the glucose oxidase method (YSI glucose analyser; Yellow Springs, OH, USA), plasma insulin and C-peptide concentrations by ELISA (Dako, Ely, UK), plasma triacylglycerol by lipase with released glycerol measured by a Roche Cobas centrifugal analyser using a colorimetric assay (ABX Diagnostics, Montpellier, France) and HbA1c by a Biorad HPLC (TOSOH Corporation, Tokyo, Japan). 2H atom percent excess in plasma glucose was determined using a Thermo ‘Voyager’ single quadrupole mass spectrometer with Thermo ‘Trace’ gas chromatograph (Thermo Scientific, Waltham, MA, USA).
Statistical analyses were performed using SPSS 17.0 software (SPSS Inc., Chicago, IL, USA). Data are presented as mean ± SE. Statistical comparisons between diabetes and control groups were performed using the Student’s t test, while within-group differences were determined using a paired t test. Changes of sequential data within experiments were evaluated by repeated measures one-way ANOVA with post hoc Bonferroni testing where appropriate. Correlations were examined using the Spearman rank test. Statistical significance was accepted at p < 0.05.