Functionally selective signaling and broad metabolic benefits by novel insulin receptor partial agonists

Insulin analogs have been developed to treat diabetes with focus primarily on improving the time action profile without affecting ligand-receptor interaction or functional selectivity. As a result, inherent liabilities (e.g. hypoglycemia) of injectable insulin continue to limit the true therapeutic potential of related agents. Insulin dimers were synthesized to investigate whether partial agonism of the insulin receptor (IR) tyrosine kinase is achievable, and to explore the potential for tissue-selective systemic insulin pharmacology. The insulin dimers induced distinct IR conformational changes compared to native monomeric insulin and substrate phosphorylation assays demonstrated partial agonism. Structurally distinct dimers with differences in conjugation sites and linkers were prepared to deliver desirable IR partial agonist (IRPA). Systemic infusions of a B29-B29 dimer in vivo revealed sharp differences compared to native insulin. Suppression of hepatic glucose production and lipolysis were like that attained with regular insulin, albeit with a distinctly shallower dose-response. In contrast, there was highly attenuated stimulation of glucose uptake into muscle. Mechanistic studies indicated that IRPAs exploit tissue differences in receptor density and have additional distinctions pertaining to drug clearance and distribution. The hepato-adipose selective action of IRPAs is a potentially safer approach for treatment of diabetes.


Synthesis of IRPA
RHI was used for the synthesis of dimers, except for IRPA-4, for which desB30 B3K B29R RHI was used. Briefly, IRPA-1 was synthesized in a two-step process, where the first step included biscarbamylation of A1 and B1 sites of insulin with potassium isocyanate in water. The second step included dimerization using disuccinimidyl suberate in organic solvent with triethylamine as the base. Similar chemistry was used for the synthesis of IRPA-2, IRPA-3, IRPA-7, IRPA-8. Compounds with triazole (Trz) linker, IRPA-4, IRPA-5, IRPA-6, and IRPA-9 were assembled by copper (I) catalyzed "click" reaction from individually prepared alkyne-and azide-decorated insulins. The synthesis of IRPA-1 and IRPA-4 are described below in detail. The experimental details for the preparation of the rest of the IRPA dimers are similar to those described for IRPA-1 and IRPA-4 and can be found in literature 1,2 as well as medicinal chemistry publications being prepared.
To a suspension of RHI (1g, 0.172 mmol) in water (50 mL) was added a solution of potassium phosphate, dibasic (0.249 g, 1.429 mmol) in water (5.0 mL). After stirring at room temperature for 30 minutes, to the resulting mixture was added potassium cyanate (0.279 g, 3.44 mmol). The reaction mixture was allowed to stir for 16 hours. To stop the reaction, unreacted potassium cyanate was removed by TFF using MWCO 3K diafiltration device, and the product was isolated as a solid by lyophilization. The product contained about 10-35% of A1/B1/B29-tris-urea-RHI, which optionally could be removed by preparative HPLC with gradient 26-34% (Solvent A: Water-0.05%TFA; Solvent B: AcN-0.05%TFA). UPLC-MS Method C: Rt = 4.29 min, m/z = 1474.6 (z = 4). Average yields of desired material ranged from 30 to 70%.
Dissolved the product of previous step (1.488 g, 0.252 mmol) in DMSO (21 mL) and added 1.06 mL (7.57 mmol) of triethylamine followed by dropwise solution of disuccinimidyl suberate (0.037 g, 0.101 mmol) in 1.0 mL of DMSO. The mixture was stirred for 30 min and dimerization was verified by UPLC. The reaction mixture was added dropwise to 20 volumes of water and pH adjusted to 7.4 by addition of 1M HCl. Concentrated the reaction mixture by diafiltration to volume of ~50 mL. The product was isolated by reverse-phase chromatography using 26-40% gradient (Solvent A: Water-0.05%TFA; Solvent B: AcN-0.05%TFA) in 30 min, 10 sequential portionwise injections were done for the purification of the total batch. 550 mg (36%) was obtained after lyophilization of fractions. UPLC-MS Method D: Rt = 3.59 min, m/z = 1988.4.6 (z = 6).
Step 2. Synthesis of the B3 azide component.
A mixture containing 50 mg (8.6 μmol) of alkyne component, 50 mg (8.6 μmol) of azide component, 12 mL of DMSO, 18 mL of water, and 7.0 mL of 2M of triethylammonium acetate buffer, PH=7.0, was degassed by nitrogen purge for 10 min. Added 4.0 mL of freshly prepared 5.0 mM aq. ascorbic acid solution and degassed by nitrogen purge for 1 min. Added 2.0 mL of 10 mM CuSO4-TBTA (55% DMSO-water) solution and degassed by nitrogen purge for 1 min. Gently shook the mixture overnight. The precipitated product was re-dissolved by addition of 80 mL of 30% AcN-water solution and adjustment pH to 3.5. The mixture was concentrated to total volume of 8.0 mL in Amicon tubes and the product was isolated by preparative HPLC using 25-45% gradient

Cellular assays to profile site-specific phosphorylation
CHO-hIR cells were serum starved overnight, then treated with compounds at 37 o C for 10 minutes. The cells were lysed with lysis buffer (MSD) on ice in the presence of protease inhibitors and phosphatase inhibitors for 30 minutes. The lysates were applied to the modified MSD Elisa assays to measure phosphorylation of different tyrosine residues on insulin receptor. Briefly, lysates were added to custom MSD IR plates (MSD cat# N45CA-1). Phosphorylation of tyrosine on IR were measured using antibody against pY960 (rabbit mAb, Cell Applications #CB4378) or pY1345 (rabbit mAb 14A4, Cell Signaling #3026), respectively in combination with SULFO-tag anti rabbit antibody (MSD cat# R32AB). For pY1150, cell lysate was added to MSD high binding plate (MSD cat # L15XB) coated with pY1150/1151 antibody (mouse monoclonal antibody 19H7, Cell Signaling #3024). The bound insulin receptor was detected using combination of rabbit antiinsulin receptor b (C-19, Santa Cruz sc-711) and SULFO-tag anti rabbit antibody.

IGF-1R binding, cellular proliferation in SaoS2 cells and 2-deoxyglucose (2DG) update in mouse adipocytes
IGF-1R affinity of IRPAs was assessed in a whole cell binding assay using a cell line stably expressing human IGF-1R with radiolabeled native IGF-1 as the competitor ligand. Cellular proliferation activities of IRPAs as measured by EdU incorporation to SAOS2/B10 cells using the Click-it EdU (5-Ethynyl-2′-deoxyuridine; ThermoFisher Scientific) assay. To assess IRPA function downstream of the signaling cascade, [ 14 C]-2-deoxyglucose cellular uptake in the differentiated mouse 3T3 L1 adipocytes was measured using standard protocol ( 3 ).

Electron microscopy
The preparation, collection and processing of the sample of IR in complex with RHI was performed as described in 4 . The complexes with IRPA-3 and IRPA-9 were made by mixing the purified IR ectodomain at a concentration of 0.3 mg/mL with a 10 molar excess of the different synthetic insulin molecules and incubating at 4°C for 1 h. For every sample, 3 L were applied to grids previously plasma cleaned with a Solarus (Gatan). To prevent the preferred orientation problem typically observed with this protein, a combination of different grids were frozen by using different available techniques: both for IRPA-3 and IRPA-9, nanowire-Lacey grids were ultra-fast frozen with SpotItOn (at the Simons Electron Microscopy Center in the New York Structural Biology Center); additionally for IRPA-3, grids were plunge-frozen using both 1.2/1.3 C/Cu C-flat or UltraAu Au/Au and a manual cryoplunger (at NanoImaging Services). The C/Cu and SpotItOn grids were used to collect untilted data and the Au/Au grids to collect -30 deg tilted data at NIS in a 300 kV Titan Krios (ThermoFisher) equipped with a GIF Quantum 967 LS imaging filter (Gatan) and a K2 Summit direct detector (Gatan). Movies were recorded in counting mode at a calibrated pixel size of 1.04 Å/pix. The movies were fractionated into 30 frames of 200 ms (yielding movies of 6 s duration) and a total accumulated exposure of 45 e -/Å 2 . For IRPA-3 a total of 1673 movies were collected without tilt for C/Cu grids, 5778 movies from SpotItOn grids and finally 5633 movies for Au/Au applying -30 deg tilt. In each case, they were necessary several sessions to collect these amounts. For IRPA-9, a total of 2974 images were collected in a single session.

Image processing and structure determination
Images for all datasets were collected using Leginon. The frame alignment was carried out using MotionCor2 and their CTF was calculated per-particle using gCTF. After a strict data evaluation based on their CTF information and visual inspection 46%, 38% and 66% of the movies were discarded for IRPA-3 grids of C/Cu, SpotItOn and Au/Au respectively, while 5% of the movies were discarded for IRPA-9 grids. For IRPA-9, particle picking done using Gautomatch (https://www2.mrc-lmb.cam.ac.uk/research/locally-developed-software/zhang-software/) resulting in 1,236k particles that were extracted by using Relion. The extracted particles were imported and processed (from 2D classification to final reconstruction) in cryoSPARC v0.6.5. After 2 rounds of 2D classification, 212k particles were selected for an ab-initio reconstruction with 5 classes and without imposing symmetry, in which one of the resulting volumes accumulated 47% of the particles and showed the best quality and features of the expected C2 symmetry. The corresponding volume was refined thorough a single homogenous refinement job and using the associated particles to a 5.2 Å resolution map.
For IRPA-3, all the remaining movies after CTF evaluation were reimported into cryoSPARC v2, realigning their frames and re-estimating their CTF with GCTF. A resulting dataset of 692k particles was selected for local motion correction and after 3 rounds of 2D classification, reduced to 258k particles that were fed to an ab-initio job for 3 classes without imposing symmetry. The 2 best classes were selected based on their 3D features and angular coverage and their particles used to start an iterative series of several ab-initio jobs with the same parameters than the initial one. For every step of that series, the particles were used both to run a subsequent ab-initio job and a homogenous refinement, and the refined volume inspected in terms of FSC resolution and 3D features. At the end of the 5 th round, 92k particles until a final dataset of 92k particles resulted in the best reconstruction at 5.0 Å of resolution. The resulting volume and particles were refined by C2 non-uniform refinement generating the final map at 4.5 Å of resolution, calculated by goldstandard FSC.
The previously solved cryoEM complex of IR bound to insulin was positioned into both maps using MolRep and rigid-body refinement using Coot. The final structures were subjected to several cycles of global real-space refinement in Phenix with the parameters NCS, rotamer, Ramachandran plot and C-beta deviations restraints enabled. There was no density in the map of IRPA-9 for the C-terminal residues 590 to 806 of one of the chains, being that part of the polypeptide chain modeled based on similarity with the insulin bound complex structure. Map visualization was done by using Chimera and figures both by using Chimera and Pymol (The PyMOL Molecular Graphics System v.1.8). Data collection, refinement and validation statistics are summarized in Table S3.

HDX-MS Studies
The HDX-MS experiments were carried out using a soluble, fully glycosylated IR-ectodomain b isoform. IR-ectodomain an isoform was also tested (R&D Systems Cat# 1544-IR/CF). Both isoforms yielded similar H/D signatures upon insulin binding (data not shown), demonstrating that both isoforms have similar conformation in solution. Thus, for structural studies, both isoforms can be used to understand the interaction behaviors of human insulin and its analogues. Typical sequence coverage for IR(b) or IR(a) is approximately 77% with greater than 250 peptic peptides detected (Fig. S2). Reasonable sequence coverage was obtained for almost all their-ECD domains except cysteine-rich domain.
Difference in deuterium uptake rate was detected when comparing IR-ECD alone to IR-ECD bound with RHI, IRPA-3 or IRPA-9. The sequence regions that showed deuterium uptake difference for RHI or IRPA-9 bound IR-ECD are listed in Table S1, and for IRPA-3 bound in Table S2. A H/D difference plot is generated by the sum of difference in deuterium uptake rate for each peptic peptide at each deuterium labeling time point between bound and unbound IR-ECD (y-axis) and peptic peptides are arranged from N to C terminus along X-axis (Fig. 2i). The grey bar pointing downward indicates that the deuterium uptake rate is reduced upon ligand binding. The magnitude of the grey bar (y-axis) greater than (± 0.5 Da) is considered statistically significant. Beagle dogs were rested for minimally 4 weeks before reuse. Unknown insulins were initially tested at infusion rates of 6 & 9 pmols/kg/min (pkm); expanded to 3, 6, 9 12, 15-16 and 20 pkm as necessary. C-peptide and glucagon measurements were made using Millipore's C-peptide & Glucagon RIA kits (CCP-24HK & GL-32K, respectively). Glucagon assays were altered by adding an additional 24hrs incubation prior to the delayed addition of labeled tracer. Table   Table S1.

Comparison between HDX Signatures and Cryo-EM Map for Insulin Receptor Ectodomain
Complexed with RHI or IRPA-9.