Glucose-sensitive insulin with attenuation of hypoglycaemia

Abstract

The risk of inducing hypoglycaemia (low blood glucose) constitutes the main challenge associated with insulin therapy for diabetes^1,2. Insulin doses must be adjusted to ensure that blood glucose values are within the normal range, but matching insulin doses to fluctuating glucose levels is difficult because even a slightly higher insulin dose than needed can lead to a hypoglycaemic incidence, which can be anything from uncomfortable to life-threatening. It has therefore been a long-standing goal to engineer a glucose-sensitive insulin that can auto-adjust its bioactivity in a reversible manner according to ambient glucose levels to ultimately achieve better glycaemic control while lowering the risk of hypoglycaemia³. Here we report the design and properties of NNC2215, an insulin conjugate with bioactivity that is reversibly responsive to a glucose range relevant for diabetes, as demonstrated in vitro and in vivo. NNC2215 was engineered by conjugating a glucose-binding macrocycle⁴ and a glucoside to insulin, thereby introducing a switch that can open and close in response to glucose and thereby equilibrate insulin between active and less-active conformations. The insulin receptor affinity for NNC2215 increased 3.2-fold when the glucose concentration was increased from 3 to 20 mM. In animal studies, the glucose-sensitive bioactivity of NNC2215 was demonstrated to lead to protection against hypoglycaemia while partially covering glucose excursions.

Fig. 1. Functional principle and 3D model of NNC2215.

a, NNC2215 is an insulin molecule with a glucose-sensitive switch. At increasing glucose concentrations, the switch equilibrates towards an open state and the insulin receptor affinity of NNC2215 is high, thereby contributing to preventing hyperglycaemia. When glucose levels decrease, the switch equilibrates towards a closed state, interfering with the ability of NNC2215 to bind to the insulin receptor, thereby contributing to preventing hypoglycaemia. Insulin backbone, macrocycle, glucoside and glucose models were prepared using BIOVIA Discovery Studio (Dassault Systèmes). b, 3D models of NNC2215 in the open and closed forms. The insulin backbone is shown as ribbons and the switch elements (glucoside and macrocycle) are shown as stick representations. Insulin receptor chains A and C from PDB 6PXV are shown as white and grey surface representations. The open form of NNC2215 (yellow) has free glucose (orange; top left corner) bound to the macrocycle at B29. The B1–glucoside of NNC2215 in the open form is shown in orange (on the right). The closed form of NNC2215 (cyan) has the glucoside bound in the macrocycle and shows a clash between the C-terminal part of the insulin B-chain, including the switch and the C-terminal domain of the insulin receptor (α-CT, purple).

Fig. 2. Chemical structure and glucose-binding properties of NNC2215.

a, NNC2215 with dual conjugation of the macrocycle at B29 and the glucoside at B1. The control compound NNC2215a has only the B29–macrocycle. Prepared using BIOVIA Draw (Dassault Systèmes). b, ITC measurement of the affinity of glucose to the free macrocycle. A K_d of 98 μM was obtained by fitting the data to a 1:1 binding model. c, Analysis of the binding of glucose towards NNC2215 (K_d = 2.1 mM) and NNC2215a (K_d = 0.5 mM) using native MS (the raw native mass spectra are shown in the Supplementary Data). R² values from nonlinear regression are 0.9989 for NNC2215 and 0.9994 for NNC2215a. Data are mean ± s.d. n = 3 technical replicates. The s.d. error bars are shorter than the size of the symbols. Individual values are shown as black squares/circles. Source Data

Fig. 3. Glucose-dependent hIR-A affinity of NNC2215.

a–c, Representative displacement curves of ¹²⁵I-insulin from hIR-A for NNC2215 (a), human insulin (b) and insulin degludec (c) in the presence of 0 to 20 mM d-glucose. Data are mean ± s.d. n = 3 technical replicates. For some datapoints, the s.d. error bars are shorter than the size of the symbols. d, hIR-A affinity of NNC2215 and insulin degludec relative to human insulin over increasing glucose concentrations. Data are mean ± s.d. n = 3 independent replicates. For some datapoints, the s.d. error bars are shorter than the size of the symbols. e, Representative curves of NNC2215 and insulin degludec dose-dependent conversion of ³H-d-glucose into lipid in isolated rat adipocytes at low (3 mM) and high (20 mM) l-glucose concentrations. Data are mean. n = 2 technical replicates. CPM, counts per minute. Source Data

Fig. 4. Glucose-concentration-sensitive activation and deactivation of NNC2215 in vivo.

a, Triggering of NNC2215 by dosing rats i.v. with 0 to 2 g kg^–1 l-glucose, resulting in lowering of d-glucose. Data are mean ± s.e.m. n = 7 animals per group. b, Glucose dose-dependent clearance of NNC2215 measured by the NNC2215 concentration at 60 min. Data are mean ± s.e.m. n = 7 animals per group. Statistical analysis was performed using a two-sided Student’s t-test; *P = 0.013. c, Representative glucose profiles in LYD pigs during constant i.v. infusion of NNC2215 (1.86 pmol kg⁻¹ min⁻¹) or insulin degludec (0.9 pmol kg⁻¹ min⁻¹) after stopping and restarting the constant i.v. glucose infusion (6 mg kg⁻¹ min⁻¹). Data are mean ± s.d. n = 7 (NNC2215) and n = 8 (insulin degludec) pigs. d, Pharmacokinetic/pharmacodynamic modelling results for a representative LYD pig dosed with 1.72 pmol kg⁻¹ min⁻¹ NNC2215. SI, insulin sensitivity index. Source Data

Fig. 5. Glucose-induced activation of NNC2215 during a GTT in streptozotocin-diabetic rats.

a,b, The glucose infusion rate (GIR) (a) and plasma glucose (b) profiles before (0–150 min), during (150–210 min) and after (210–300 min) a GTT in streptozotocin-diabetic rats receiving i.v. infusion of NNC2215 (84 pmol kg⁻¹ min⁻¹), human insulin (20 pmol kg⁻¹ min⁻¹) or human insulin (20 pmol kg⁻¹ min⁻¹) plus additional 10 pmol kg⁻¹ min⁻¹ (+50%) during the GTT. Data are mean ± s.e.m. n = 5 (NNC2215 and human insulin) and n = 7 (human insulin +50%) rats. Source Data

Extended Data Fig. 1. In vitro biology.

a, Glucose-dependent hIR-A affinity of NNC2215 with no HSA in the assay. Data are mean ± SD (n = 3 independent replicates). In each set of experiments the measurements were performed with three technical replicates. ^aDetermined from the IC₅₀ in the presence of 0 or 20 mM D-glucose (without HSA). b, Glucose-dependent hIR-A affinity of NNC2215a with 1.5% HSA in the assay. Data are mean ± SD (n = 3 independent replicates). In each set of experiments the measurements were performed with three technical replicates. ^bDetermined from the IC₅₀ in the presence of 0 or 20 mM D-glucose (with 1.5% HSA). c, Representative displacement curves of ¹²⁵I-IGF-1 from hIGF-1R for NNC2215 and human insulin in the presence of 0 or 20 mM D-glucose. Data are mean ± SD (n = 3 technical replicates). For some data points, the SD error bars are shorter than the size of the symbols. d, Glucose-dependent IGF-1R affinity of NNC2215 with no HSA in the assay. Data are mean ± SD (n = 3 independent replicates). In each set of experiments the measurements were performed with three technical replicates. ^cDetermined from the IC₅₀ in the presence of 0 or 20 mM D-glucose (without HSA). e-g, Representative curves of IR (e), Akt (f) and ERK (g) phosphorylation in CHO-hIR cells induced by increasing concentrations of NNC2215 or human insulin. Data are mean (n = 2 technical replicates). Source Data

Extended Data Fig. 2. L-glucose triggering experiment in vivo with the glucose non-sensitive insulin analogue, insulin degludec, as negative control.

a, D-glucose levels did not decrease after dosing rats i.v. with increasing L-glucose doses (0 to 2 g/kg). Data are mean ± SEM (n = 7 animals per group). b, Glucose dose-independent clearance of insulin degludec measured by the insulin degludec concentration at 60 min. Data are mean ± SEM (n = 7 animals per group). Source Data

Extended Data Fig. 3. Hypoglycaemia study in pigs - Analysis of protection against hypoglycaemia and C-peptide concentrations.

a-c, Exploratory correlation analyses of the lowest glucose concentration after turning off the glucose infusion vs. the glucose concentration immediately prior to turning off the glucose infusion. Nonlinear least squares regression analysis was performed using GraphPad Prism. All doses are contained in the plots. In (a) the insulin degludec data set has been truncated so the range of glucose concentrations before turning off glucose infusion is the same in the two groups. Slopes for NNC2215 and insulin degludec are not statistically significantly different (p = 0.293) as tested using an extra sum-of-squares F test. Intercepts for NNC2215 and insulin degludec are statistically significantly different (p < 0.0001) as tested using an extra sum-of-squares F test. The estimated difference in the drop in glucose between NNC2215 and insulin degludec was 1.8 mM. In (b) data from all pigs are shown, including those insulin degludec treated pigs with lower starting glucose values, illustrating that counterregulation sets in in those pigs and the estimated slope becomes significantly different (p = 0.024) as tested using an extra sum-of-squares F test. In (c) six data sets corresponding to all six experimental days in a single animal with high C-peptide levels were omitted, which does not alter the conclusion that can be drawn from (b). d, C-peptide concentrations. Data are geometric mean±geometric SD (n = 31 animals for NNC2215 and n = 15 animals for insulin degludec). Data are plotted across all dose levels documenting suppression of C-peptide despite relative hyperglycaemia in the pigs during the experiment. Suppression was not different in NNC2215 and insulin degludec treated pigs. During the experiment, most values were below the lower limit of quantification of 45 pM and those were represented as 22.5 pM. Occasional poor replicate C-peptide measurements were treated conservatively in relation to demonstrating C-peptide suppression by selection of the higher of the two replicates. Geometric mean was chosen due to one pig having markedly higher C-peptide levels than the rest, albeit still with lower values during infusion than at baseline. Source Data

Extended Data Fig. 4. Pharmacokinetic profile and glucose concentration following s.c. and i.v. administration of NNC2215 in LYD pigs.

a, NNC2215 concentrations in plasma. b, Plasma glucose levels. Data are mean ± SEM (n = 4 animals for s.c. and n = 3 animals for i.v.). Source Data

Extended Data Fig. 5. Minimal model based on the hypoglycaemic study in LYD pigs quantifying the change in NNC2215 activity depending on glucose concentration.

a, Parameters for the minimal model. Data are typical values from a non-linear mixed effects model. Endogenous insulin was confirmed by bioanalysis to have been suppressed by the somatostatin infusion, and hence this contribution was set to 0 in the model. ^aInterindividual variation: 86%. ^bInterindividual variation: 38%. Correlation with P2: 44%. b, Model-based SI index for NNC2215 as a function of plasma glucose. The simulation was based on typical values from the pharmacokinetic/pharmacodynamic model. SI was expressed as 0.00673 • Glucose concentration/(50 + Glucose concentration). The SI index for a given glucose concentration was calculated as the SI for that glucose concentration in percent of the SI for a glucose concentration of 5.5 mM.

Extended Data Fig. 6. Linear regression of maximal plasma glucose concentration during the GTT vs. steady-state GIR prior to the GTT in rats.

For each animal, the maximal plasma glucose concentration during the GTT was plotted against the steady-state GIR prior to the GTT, which represents the quantitative effect of the constant insulin infusion. Nonlinear least squares regression analysis was performed using GraphPad Prism. Slopes for NNC2215, human insulin and human insulin +50% are not statistically significantly different (p = 0.908) as tested using an extra sum-of-squares F test. The slope for the three groups pooled is −0.311 mM/(mg/kg/min). Importantly, the Y-axis intercepts (i.e. the maximal plasma glucose concentration during the GTT) for NNC2215, human insulin and human insulin +50% are statistically significantly different (p < 0.0001), as tested using an extra sum-of-squares F test, with the intercept of the NNC2215 group being in between the human insulin group, which was higher, and the human insulin +50% group, which was lower. Quantitatively, the maximal plasma glucose concentration during the GTT evaluated at GIR = 0 mg/kg/min prior to the GTT was reduced by 18% for NNC2215 vs. human insulin and by 30% for NNC2215 vs. human insulin +50%. Thus, the glucose sensitive effect of NNC2215 was less than the effect of 50% additional human insulin but corresponded more closely to 30% additional human insulin (50% * 18/30). Evaluated at GIR = 15 mg/kg/min prior to the GTT, the effect of NNC2215 during the GTT corresponded to 32% additional human insulin.

Extended Data Fig. 7. O-succinimidyl pentyn-1-oxycarbonyl and beta-carboxymethyl-oxyethyl-O-peracetyl-D-glucoside active ester.

a, O-succinimidyl pentyn-1-oxycarbonyl. b, beta-carboxymethyl-oxyethyl-O-peracetyl-D-glucoside active ester. Carbon atoms are numbered to allow identification of the NMR assignments provided in the Supplementary Data.

Extended Data Fig. 8. Synthesis of the macrocycle roof propyl azide.

a, Trisbromomethyl acid. b, Carboxy tris-azide. c, Alcohol tris-azide. d, Alcohol tris-Boc. e, O-mesylate tris-Boc. f, Azide tris-Boc. g, Propyl azide tris-isocyanate. h, Compound 25c. i, Macrocycle roof propyl azide. Carbon atoms in (i) are numbered to allow identification of the NMR assignments provided in the Supplementary Data. Yields are given in percent.

Extended Data Fig. 9. Control compound NNC2215a.

The structure of the control compound NNC2215a.