Patients with type 2 diabetes mellitus treated with diet alone or metformin, with or without a low dose sulfonylurea, were recruited from January to November 1995. Participants were included if their fasting plasma glucose was less than 11 mmol/l on diet treatment alone, or having stopped a low dose of sulfonylurea. They were excluded if they suffered from active cardiac disease, severe diabetic complications or other major illnesses. Participants were randomly allocated to one of two groups who received either ascending or descending glucose clamp levels, using minimisation on standardised Mahalanobis distances to match the groups for sex, BMI and 7-day mean fasting blood glucose. Participants stopped oral glucose-lowering medications for 3 weeks prior to the study to ensure complete washout of their effects. All participants were on a weight-maintaining diet and were asked not to change their dietary habits during the study. The success of this was not verified by food diaries, however. All participants gave their written informed consent. The study was approved by the Central Oxford Ethics Committee.
The study was a randomised, double-blind, placebo-controlled, crossover study. Participants received 14 days treatment with glibenclamide (2.5 mg twice daily) and placebo in random order, with a washout period of at least 14 days with no additional glucose-lowering medication treatment periods. A three-step hyperinsulinaemic glucose clamp was performed at the end of each treatment period, with plasma glucose clamped at 4.0, 8.0 and 12.0 mmol/l using a fixed rate insulin infusion and a variable rate glucose infusion. Plasma C-peptide concentration was measured at intervals and deconvoluted to estimate C-peptide (and hence insulin) secretion rates.
Hyperinsulinaemic three-step ascending and descending clamp procedure
Participants were admitted to the Diabetes Trials Unit clinical research facility at 7:30 am on two separate mornings after an overnight fast, apart from water to drink. A concentric double-lumen Teflon cannula was inserted into a wrist vein. The hand was warmed with an electric heating pad at 47°C to arterialise the venous blood. A second cannula was placed in an antecubital vein for glucose and insulin infusion. Heparinised blood was withdrawn continuously throughout the clamp using a peristaltic pump and fraction collector. Three integrated 6 min samples were taken, following which participants took either the glibenclamide 2.5 mg or placebo tablets that they had been taking the previous 14 mornings.
At time 0 and 30 min after taking the tablets, an insulin infusion was started at an initial priming rate of 480 mU m−2 min−1 for 10 min, then reduced to 80 mU m−2 min−1 for the rest of the procedure . The prime and the sustained insulin infusion rates were double those described in the original report of the hyperinsulinaemic clamp to ensure that a blood glucose concentration of 4 mmol/l would be achieved in these overweight or obese patients with a mean HOMA2 insulin sensitivity [%S] (www.dtu.ox.ac.uk/homacalculator/, last accessed 17 September 2014) of approximately 50%. The glucose infusion rate was adjusted to achieve and maintain each of the targeted blood glucose levels. Successive 3 min blood glucose results were entered into an iterative computer program that used current and previous glucose concentrations and glucose infusion rates to recommend a glucose infusion rate appropriate to maintain desired levels of glycaemia . Ascending stepped changes in blood glucose were achieved with boluses of 25% glucose diluted in normal saline, as recommended by the computer program.
Both ascending and descending clamp orders were used and were studied in the parallel groups to account for the possibility that the insulin secretory response to glucose could differ in conditions of rising or falling glucose concentrations and could be influenced by antecedent glucose concentrations during the investigative procedure.
For ascending clamps, blood glucose was stabilised at 4 mmol/l over the first 120 min. The blood glucose was then raised acutely to 8 mmol/l and clamped at this level until 210 min. The blood glucose was then again raised acutely to 12 mmol/l and clamped at this level until 300 min.
For descending clamps, blood glucose was initially raised at time 0 to 12 mmol/l by glucose bolus injection. The blood glucose was clamped at this level until 120 min. The blood glucose was then allowed to fall and clamped at 8 mmol/l until 210 min. The blood glucose was then again allowed to fall and clamped at 4 mmol/l until 300 min.
At the end of the study, participants were provided with lunch and the glucose infusion was maintained as required until blood glucose levels were rising or stable for at least 15 min without glucose infusion.
A rapid glucose oxidase method (Hemocue B Glucose Photometer; Hemocue AB, Ängelholm, Sweden) was used to measure glucose concentration every 3 min during the clamps. Integrated 6 min plasma samples (pooled samples from two 3 min aliquots) were also assayed by a laboratory hexokinase method (Boehringer Mannheim, Lewes, UK).
Fasting insulin was measured using a double antibody RIA (Pharmacia, Milton Keynes, UK) with Sepharose attached to the second antibody for separation by decanting. Human insulin standard was calibrated against ‘Research Standard A for insulin, human, for immunoassay 66/304’ from WHO International Laboratory for Biological Standards.
Plasma C-peptide was assayed on each 6 min pooled aliquot by RIA (Linco, Biogenesis, Poole, UK), using a competitive, equilibrium RIA using 125I-labelled human C-peptide with guinea pig anti-human C-peptide antibody and polyethylene glycol precipitation.
The C-peptide secretion rate (CPSR) was calculated from the plasma C-peptide concentration in 6 min aliquots of blood that was continuously withdrawn throughout the clamp procedure. The mean CPSR at the end of each clamp stage was calculated by taking the arithmetic mean of the five values covering the final 30 min of each clamp stage (90–120, 180–210 and 270–300 min from the start of the clamp). Similarly, the plasma glucose concentration was calculated for each participant as the mean of the five calculated secretion rates during the last 30 min of each clamp stage. The first phase response to a 4 mmol/l step increase in plasma glucose was calculated for the 4 mmol/l to 8 mmol/l and 8 mmol/l to 12 mmol/l transitions in the ascending clamp group. This was calculated as the difference between mean CPSR of the first three values after the step-up points (120–138 and 210–228 min) and the CPSR during the last 30 min at the end of the preceding clamp stage (as detailed above).
The HOMA2 computer model of the homeostatic basal insulin:glucose interaction was used to assess the contribution of two variables, beta cell function (HOMA2_%B) and insulin sensitivity (HOMA2_%S), to the fasting glucose and insulin or C-peptide concentrations . HOMA2  is calibrated to a reference population of normal weight healthy individuals aged <35 years who were defined as having ‘normal’ values of %B and %S equal to 100%. Three consecutive 6 min blood samples were withdrawn continuously from −50, −44 and −38 min for measurement of plasma glucose, insulin and C-peptide. HOMA2_%S was determined from glucose-insulin pairs and HOMA2_%B from glucose-C-peptide pairs.
Beta cell stimulus–response curve assessed by CPSR
The two-compartment model and deconvolution equations of Eaton et al were used to calculate estimates of insulin secretion rates from plasma C-peptide concentrations . The variables of the model were derived from Van Cauter's adjusted population values taking into account sex, age, obesity and diabetic status of the individual participants .
Data were collected on clinical report forms and double entry was performed and stored in a read-only file on the departmental server blind to randomisation codes. Non-parametric variables were transformed logarithmically. These included fasting plasma insulin, HOMA2_%B and HOMA2_%S indices, CPSRs and their proportional increments on glibenclamide. Second phase CPSR and the incremental impact of glibenclamide between treatment periods were compared using ANOVA with treatment order, treatment, ascending vs descending clamp order and clamp glucose steps as fixed factors. Pairwise comparisons were performed using Dunnett’s test. Statistical analyses were performed using SPSS for Windows, Rel. 18.0.0 (IBM, Portsmouth, UK). Descriptive statistics are expressed either as mean (1 SD) for symmetrical distributions, or geometric mean (SD range) for positively skewed distributions. Statistical estimates are expressed either as mean or geometric mean (95% CI). Geometric mean and error bars for SEM range are presented in Figs 1 and 2.