Efficient 18F-Labeling of Synthetic Exendin-4 Analogues for Imaging Beta Cells

A number of exendin derivatives have been developed to target glucagon-like peptide 1 (GLP-1) receptors on beta cells in vivo. Modifications of exendin analogues have been shown to have significant effects on pharmacokinetics and, as such, have been used to develop a variety of therapeutic compounds. Here, we show that an exendin-4, modified at position 12 with a cysteine conjugated to a tetrazine, can be labeled with 18F-trans-cyclooctene and converted into a PET imaging agent at high yields and with good selectivity. The agent accumulates in beta cells in vivo and has sufficiently high accumulation in mouse models of insulinomas to enable in vivo imaging.


Introduction
The ability to visualize beta cells noninvasively could have far reaching implications for both biomedical research and clinical practice. Progressive loss of functional beta cell mass (BCM) is the underlying cause of autoimmune type 1 diabetes mellitus, and is also responsible for the secondary failure of clinical drugs in type 2 diabetes. It is widely believed that noninvasive imaging of beta cells could ultimately facilitate not only our understanding of the natural history of islet formation but also the pathophysiology of diabetes. In turn, we would have the capability to diagnose diabetes earlier, monitor the efficacy of widely used drugs, as well as advance the discovery of new therapies. Furthermore, beta cell-specific imaging approaches could be used to diagnose and localize insulinomas and aid the assessment of transplanted islets or pancreata.
In a previous report, we described the development and validation of near infrared fluorescent exendin-4 analogues for imaging beta cells at single cell resolutions, [1] and for fiberoptic, endoscopic or intraoperative imaging. [2] We showed that one lead agent, derived from exendin-4 (E4 K12 -FL), had subnanomolar EC 50 binding concentrations and high specificity. In addition, its binding could be inhibited by glucagon-like peptide 1 (GLP-1) receptor agonists. Following intravenous administration to mice, pancreatic islets could be readily distinguished from exocrine pancreas, achieving target-to-background ratios of 6:1. Serial imaging subsequently revealed rapid accumulation kinetics (with initial signal in the islets detectable within 3 min and peak fluorescence occurring within 20 min of injection). Such properties make this an ideal agent for in vivo imaging. Together with other reports of various exendins labeled with chelates, [3][4][5][6][7][8][9] we hypothesized that 18 F-labeled exendin-4 analogues could be used for noninvasive imaging with positron emission tomography-computed tomography (PET-CT). While two approaches of 18 F-labeling have been recently reported, [10,11] the 18 F-conjugation methods used in these studies do not appear to have been used in concert with removal of unreacted material via bioorthogonal scavenging resins. [12] In this study, we started with a cysteine (C12) version of our previously validated exendin-4 (E4 K12 ), by exchanging the lysine at position 12 with a cysteine. Using bioorthogonal labeling strategies employing 18 F-trans-cyclooctene ( 18 F-TCO) and tetrazine (Tz) modified molecules, [13][14][15] we report the facile synthesis and purification of 18 F-labeled exendin-4. The described reaction demonstrated fast reaction times (20 min), high purity as well as specific activity. Given that the ultimate goal is to translate this technology to the clinic, a lead 18 F-labeled compound was subsequently applied to PET-CT imaging of insulinoma in a mouse model. Pharmacokinetic modeling, the plasma clearance and tracer-uptake data obtained from these experiments were subsequently used for extrapolation to humans.

Results and Discussion
We previously demonstrated that modification of the exendin-4 amino acid sequence at position 12 does not result in perturbation of the molecule's intrapancreatic binding, selectivity or specificity for the GLP-1 receptor. In order to translate this finding into a noninvasive 18 F-PET probe, we designed the cysteine-tetrazine (Tz) cross-linker, maleimide-Tz 3 (Scheme 1). The A number of exendin derivatives have been developed to target glucagon-like peptide 1 (GLP-1) receptors on beta cells in vivo. Modifications of exendin analogues have been shown to have significant effects on pharmacokinetics and, as such, have been used to develop a variety of therapeutic compounds. Here, we show that an exendin-4, modified at position 12 with a cysteine conjugated to a tetrazine, can be labeled with 18 F-trans-cyclooctene and converted into a PET imaging agent at high yields and with good selectivity. The agent accumulates in beta cells in vivo and has sufficiently high accumulation in mouse models of insulinomas to enable in vivo imaging.
The blood half-life of 18 F-E4 Tz12 7 was determined through serial retro-orbital bleeds, and the individual data points were then fitted using a biexponential decay curve. This resulted in a weighted half-life (t 1/2 ) for 18 F-E4 Tz12 7 of 6.  (Figure 2 C) which support the ex vivo excretion profiles. Tissue levels of the compound were highest in the lungs (4.1 AE1.5 % ID g À1 ) and pancreas (1.2 AE 0.1 % ID g À1 ); although, uptake of 18 F-E4 Tz12 7 was found to be significantly lower in the pancreata of mice that had been preinjected with cold exenatide (Byetta , 0.36 AE 0.05 % ID g À1 ). Accumulation in the bone was low (0.6 AE 0.1 % ID g À1 ), indicating minimal defluorination of 18 F-E4 Tz12 7.
To determine the intra-pancreatic distribution of the compound (islets of Langerhans comprise only 1-2 % of the pancreatic mass), we performed autoradiography. We injected 18   green fluorescent protein (GFP)]. [18] After 3 h, the mice were euthanized, and their pancreata excised. The pancreata were then imaged using surface reflectance imaging (to show the islet distribution) before being exposed for autoradiography (to show the distribution of 18 F-E4 Tz12 7). Figure 3 shows good colocalization between the fluorescence of the GFP islet and the autoradiographic signal from 18 F-E4 Tz12 7 with a Pearson's coefficient of 0.83 AE 0.04 (R coloc. ). Based on micro-dissected specimens and target-to-background ratios, we calculated a concentration of approximately 40 % ID g À1 ) in the islets.
Ultimately, these agents are being developed for their clinical application. While their clearance is very rapid in mice (80 % with a 1.9 min half-life and 20 % with a 26.8 min half-life for 18 F-E4 Tz12 7), we were interested in determining the optimal clearance kinetics in humans. A compartmental pharmacokinetic model was thus developed to extrapolate our results from mice. The advantage of this model is that some of the parameters (e.g., plasma clearance) that vary between species can be scaled up, while others (e.g., the binding rate constants and radioactive decay half-life) are kept constant.
Using clinical data available for exenatide, the plasma concentration after continuous infusion [19] was fit to a two-compartmental model, in order to predict the percent clearance of a bolus imaging dose. The results indicated that 73 % of the imaging agent dose redistributes to peripheral tissues with a rapid 1 min half-life, while the remaining 27 % clears with a 63 min half-life. This is close to the percent clearance observed with inulin in humans following an intravenous bolus injection (76 % with a 10 min half-life and 24 % with a 86 min half-life [20] ); the model therefore provides a reasonable estimate of clearance.
The exchange rate of the compound between the plasma and extracellular space was subsequently estimated from literature values [21] and adjusted to fit our experimental results in mice (Figure 5 A). The results in Figure 5 B show estimates of human uptake and clearance, based on clinical data, and using mechanistic rate constants from mice. In both cases, the specific uptake of the compound in islets is significantly higher than in the exocrine pancreas due to its specific target binding.
The GLP-1 receptor is highly expressed in beta cells within the islets of Langerhans as well as in functioning beta cell islet tumors (insulinomas). GLP-1 analogues are a new class of peptide-based drugs used for the treatment of diabetes. Exenatide, the first FDA approved GLP-1 analogue, is a synthetic version of exendin-4. It is a 39-amino acid peptide isolated from the saliva of the Gila monster (Heloderma suspectum) and contains 53 % sequence homology with GLP-1. A recent crystal structure of the extracellular domain of the GLP-1 receptor showed the binding mode of exendin-4 (amino acids 9-39). [22] From this crystal structure, it was clear that lysine 12 (K12) is not involved in binding to the GLP-1 receptor domain. Moreover, it explains why K12-modified exendins retain high affinity for the receptor. [1,2] Our results further demonstrate that K12 modification with tetrazines are not only stable but allow rapid site-specific and high-yield fluorinations. Tetrazine functionalization of the peptide also allows removal of unreacted starting material with the complimentary trans-cyclooctene beads, an option not available to other current 18 F or metal chelation labeling strategies. The resulting compounds exhibit appropriate pharmacokinetics for PET imaging of beta cells in a mouse model.
In an effort to predict the compound's kinetics in humans, we applied pharmacokinetic modeling and allometric scaling. [23,24] In the mouse, the synthesized compound had a weighted half-life (t 1/2 ) of 6.8 min. Using our modeling and scaling approach, we predicted a t 1/2 value of 18 min in human. Importantly, we found that this molecule size has a beta phase clearance half-life of 63 min. Agents with clearance rates that are much slower than the radioactive half-life could have a background that is too high during the imaging window. Conversely, agents that clear much faster than the radioactive half-life could have inefficient accumulation within the target tissue. Given that the decay of 18 F is 109.8 min, our modeling indicates that this compound would have close to ideal clearance for human imaging. The pharmacokinetic modeling also indicates that further improvements in linker modification could reduce exocrine uptake and improve detection sensitivity. For example, by using bioorthogonal chemistry, which allows facile modulation of the linkers, further improvements in the reaction kinetics, stability and biocompatibility of the compound could be achieved. [25,26] Experimental Section Chemistry General: Unless otherwise noted, all reagents were purchased from Sigma-Aldrich (St. Louis, MO, USA) and used without further purification. Exendin-4 (exenatide, Byetta ) was obtained from Amylin/Eli Lilly (San Diego, CA, USA). E4 C12 (4163 g mol À1 ; HGEGTFTSDLSCQMEEEAVRLFIEWLKNGGPSSGAPPPS) was obtained from Genscript (Piscataway, NJ, USA). [ 18 F]-Fluoride (n.c.a.) was purchased from PETNET Solutions (Woburn, MA, USA). 3-maleimidopropanoic acid succinimidyl ester 1, tetrazine (Tz) amine 2 and 18 Ftrans-cyclooctene ( 18 F-TCO) 4 were synthesized as described else- Figure 5. Extrapolation of uptake in human pancreata using pharmacokinetic modeling. A) The efficiency of uptake in the islet (c) versus the exocrine pancreas (c) is shown for the mouse using a compartmental model. Initial rate constants were adjusted to fit experimental data. B) The mechanistic rate constants (from mice) were combined with a fitted clearance in humans (based on clinical data), to estimate uptake and clearance of the compound in islets (c) and exocrine pancreas (c) following injection of a bolus dose [injected dose per gram of tissue (% ID g À1 )]. Due to differences in body weight, the human data is expressed as injected dose per kilogram of tissue (% ID kg À1 ).
Mice: Experiments were performed in Nu/Nu mice (from Massachusetts General Hospital, Boston, MA; for tumor implantations and imaging; n = 6), C57BL/6 (B6) mice (from The Jackson Laboratory, Bar Harbor, ME; for biodistribution and pharmacokinetics; n = 8), or B6.Cg-Tg(Ins1-EGFP)1Hara/J mice (from The Jackson Laboratory, Bar Harbor, ME; for autoradiography/surface reflectance imaging; n = 3). [18] B6.Cg-Tg(Ins1-EGFP)1Hara/J mice express the en- ing longer than 1 h, the isoflurane flow rate was reduced to~1.0 % isoflurane in O 2 at 2.0 L min À1 . Surgeries were conducted under sterile conditions with a zoom stereomicroscope (Olympus SZ61). All procedures and animal protocols were approved by the Massachusetts General Hospital subcommittee on research animal care.
Whole pancreas islet imaging: B6.Cg-Tg(Ins1-EGFP)1Hara/J (MIP-GFP) mice [18] were administered 18 F-E4 Tz12 7 (92 AE 12 mCi [3.40 AE 0.44 MBq]) via intravenous tail-vein injection, and the GPL-1 receptor-specific probe was allowed to accumulate and clear for 3 h. Mice were then euthanized, their organs perfused using 1 PBS (30 mL) and the pancreata harvested. They were subsequently weighed and placed between two glass cover slides using a 1 mm rubber gasket, maintaining a constant thickness. Initially, fluorescence reflectance was recorded by imaging the entire pancreas on an OV110 epifluorescence imager (Olympus America, Center Valley, PA, USA). The pancreata were then transferred to an autoradiography phosphor imaging plate (SI, Molecular Dynamics) and exposed at À20 8C for 12 h before the plate was analyzed using a Typhoon scanner (GE Healthcare). Image analysis was conducted using Im-ageJA 1.45 software. 18 F-E4 Tz12 7 biodistribution studies: C57BL/6 (B6) mice were used for blood half-life determinations. Mice were administered 18 F-E4 Tz12 7 (68 AE 12 mCi [2.52 AE 0.44 MBq]) by intravenous tail-vein injection. Blood sampling was performed by retro-orbital puncture using tared, heparinized capillary tubes. Samples were subsequently weighed and activity measured using a Wallac Wizard 3" 1480 Automatic Gamma Counter (PerkinElmer). Blood half-life data were fitted to a biexponential model using Graphpad Prism 4.0c software (GraphPad Software Inc., San Diego, CA), and results were reported as the weighted average of the distribution and clearance phases. For biodistributions, (B6) mice were intravenously injected via tail vein with 18 F-E4 Tz12 7 (131 AE18 mCi [4.85 AE 0.67 MBq]). Animals were euthanized at 3 h and their organs perfused using 1 PBS (30 mL). Tissues were subsequently harvested, weighed and their radioactivity counted using a Wallac Wizard 3" 1480 Automatic Gamma Counter. Statistical analysis was performed using Graphpad Prism 4.0c.
MicroPET-CT imaging: Mice were imaged by PET-CT using an Inveon small animal microPET scanner (Siemens Medical Solutions). Mice were injected with 18 F-E4 Tz12 7 (557 AE 38 mCi [20.61 AE 1.41 MBq]) via tail-vein injection under isoflurane anesthesia (see above). Acquisition for static microPET images started 2 h post injection and acquisition took approximately 30 min. For dynamic microPET imaging, mice were injected approximately 30 s after the start of microPET acquisition, and data was collected for 2 h. The radioactivity concentration for a tissue was determined by measuring within regions of interest (ROIs) for a given tissue with the units of Bq mL À1 min À1 . A tissue density of 1 g mL À1 was assumed and ROIs were converted to Bq g À1 min À1 and divided by the injected activity to obtain an imaging ROI-derived % ID g À1 . For GLP-1 receptor blocking experiments, unlabeled exenatide (250 mL, 60 mm) was preinjected 45 min prior to injection of 18 F-E4 Tz12 7. A high-resolution Fourier rebinning algorithm was used, followed by a filtered back-projection algorithm using a ramp filter, to reconstruct 3D images without attenuation correction. The image voxel size was 0.796 0.861 0.861 mm, for a total of 128 128 159 voxels. Peak sensitivity of the Inveon accounts for 11.1 % of positron emission, with a mean resolution of 1.65 mm. The total counts acquired was 600 million per PET scan. Calibration of the PET signal with a cylindrical phantom containing 18 F was performed before all scans. CT images were reconstructed using a modified Feldkamp reconstruction algorithm (COBRA) from 360 cone-beam X-ray projections (80 kVp and 500 mA X-ray tube). The isotropic voxel size of the CT images was 60 mm. The reconstruction of data sets, PET-CT fusion, and image analysis were performed using Inveon Research Workplace (IRW) software (Siemens). 3D visualizations were produced using a digital imaging and communications in medicine (DICOM) viewer (OsiriX Foundation, Geneva, Switzerland).

Modeling
A compartmental model was used to extrapolate results from mouse-imaging studies to humans. The model includes biexponential loss from the plasma compartment (due to redistribution and clearance), and separate compartments for the endocrine and exocrine pancreas. Exchange with the endocrine tissue (islets) was estimated as a function of the vascular surface area-to-volume ratio (measured at 505 AE 146 cm À1 using CD31 stained histology slides), [28] and permeability was estimated at 30 mm s À1 (for this sized molecule in the fenestrated capillary bed). [21] Exocrine pancreas was modeled in a similar manner, while the exchange parameters were adjusted to fit experimental data. Within the compartments, the imaging agent is able to bind the target, dissociate, internalize, and be degraded and washed out. [24] These rate constants were assumed constant between species. For plasma clearance in humans, the rate constants for exchange and clearance from a two-compartmental model were fit to experimental data taken from patients undergoing an intravenous infusion of exenatide [19] using a least-squares fitting algorithm in Matlab (Mathworks, Natick, MA, USA). Estimates for humans were obtained by entering the plasma clearance values from human clinical data into the model together with the microscopic transport rates obtained from mouse experiments.