Development of an orally delivered GLP-1 receptor agonist through peptide engineering and drug delivery to treat chronic disease

Peptide therapeutics are increasingly used in the treatment of disease, but their administration by injection reduces patient compliance and convenience, especially for chronic diseases. Thus, oral administration of a peptide therapeutic represents a significant advance in medicine, but is challenged by gastrointestinal instability and ineffective uptake into the circulation. Here, we have used glucagon-like peptide-1 (GLP-1) as a model peptide therapeutic for treating obesity-linked type 2 diabetes, a common chronic disease. We describe a comprehensive multidisciplinary approach leading to the development of MEDI7219, a GLP-1 receptor agonist (GLP-1RA) specifically engineered for oral delivery. Sites of protease/peptidase vulnerabilities in GLP-1 were removed by amino acid substitution and the peptide backbone was bis-lipidated to promote MEDI7219 reversible plasma protein binding without affecting potency. A combination of sodium chenodeoxycholate and propyl gallate was used to enhance bioavailability of MEDI7219 at the site of maximal gastrointestinal absorption, targeted by enteric-coated tablets. This synergistic approach resulted in MEDI7219 bioavailability of ~ 6% in dogs receiving oral tablets. In a dog model of obesity and insulin resistance, MEDI7219 oral tablets significantly decreased food intake, body weight and glucose excursions, validating the approach. This novel approach to the development of MEDI7219 provides a template for the development of other oral peptide therapeutics.

The in vitro potency of MEDI7219 was similar to semaglutide in the CHO cell cAMP accumulation assay in the presence of 0.1% bovine serum albumin (BSA), with low picomolar EC 50 values for both peptides (MEDI7219, 3.4 pM; semaglutide, 12 pM; Supplementary Fig. S1). As anticipated, the potency of these peptides was reduced when a physiological level of 4.4% human serum albumin (HSA) was used in the CHO cell assay (MEDI7219, 398 pM; semaglutide, 2630 pM; Supplementary Fig. S1), but the EC 50  www.nature.com/scientificreports/ semaglutide was similar across assays. Potencies were comparable for both peptides when the assay was performed in the EndoC-βH1 human pancreatic β-cell line, which endogenously expresses GLP-1R ( Supplementary  Fig. S1). To confirm that bis-lipidation conferred a level of plasma protein binding expected to improve circulating half-life, we compared the peptides in an in vitro EScalate equilibrium shift assay. MEDI7219 and semaglutide both bound to human, dog, monkey and rat plasma proteins, with over 97% peptide bound in all cases. Higher unbound peptide proportions were observed for MEDI7219 than for semaglutide (Fig. 2b). Taken together, amino acid modification and bis-lipidation of MEDI7219 resulted in a proteolytically stable and potent peptide with bioactivity toward the GLP-1R in the presence of physiological HSA levels, supporting further development.
In the diet-induced obese (DIO) mouse model, MEDI7219 10 nmol/kg reduced body weight by 17.0% after 21 days, compared with an 18.7% reduction for semaglutide 10 nmol/kg and a 7.2% increase for placebo (Fig. 3b). Fasting glucose and insulin levels at day 14 were also reduced in DIO mice receiving once daily subcutaneous administration of MEDI7219 10 nmol/kg compared with placebo, and were similar to levels in mice receiving semaglutide 10 nmol/kg (Fig. 3c,d). In the diabetic db/db mouse model, MEDI7219 dose-dependently reduced glucose levels following once-daily subcutaneous administration for 28 days, with statistically significantly lower glucose levels at doses of 3-30 nmol/kg (P < 0.001) than placebo (Fig. 3e). Significant reductions in glucose levels were observed as early as day 7 in mice receiving MEDI7219 10 nmol/kg or 30 nmol/kg. MEDI7219 also dose-dependently reduced HbA 1C levels compared with placebo (P < 0.001 for doses 1-30 nmol/kg), with similar effects to those of semaglutide injected at doses of 3 nmol/kg and 30 nmol/kg (Fig. 3f).

Selection of permeation enhancers for oral administration.
To optimize gastrointestinal absorption of our GLP-1 peptide analogues, we screened comprehensive panels of permeation enhancers (Supplementary  Table S1) in vitro using Caco-2 cell monolayers (Fig. 4a)      www.nature.com/scientificreports/ (Fig. 4b). These experiments used mono-lipidated J229 as a model peptide with physicochemical characteristics similar to those of the bis-lipidated lead peptide MEDI7219. The in vitro screen identified a novel combination of sodium chenodeoxycholate (NaCDC) and propyl gallate (PG) as the optimal permeation enhancers in the panel tested (Fig. 4a). In rats, the largest improvement in bioavailability following intraduodenal administration was with 50 mg/kg NaCDC and 25 mg/kg PG, among the panel tested. This combination increased bioavailability to a systemic fraction (%F) of 0.39 compared with 0.02 for J229 alone, at a dose of 1 mg/kg (Fig. 4b). Switching to MEDI7219, mean bioavailability was 13-fold higher than for semaglutide when administered intraduodenally with permeation enhancers in rats (%F, 1.01 vs 0.08) (Fig. 4c).
Site of gastrointestinal absorption of stabilized GLP-1RA peptides in dogs. We used Intelli-Cap controlled release capsules 46 to determine the site of maximal gastrointestinal absorption of our stabilized GLP-1RAs in dogs. Capsules filled with J229 formulated with permeation enhancers were actuated at various pH-dependent points along the gastrointestinal tract following oral administration ( Table 2). Bioavailability of J229 was highest when released in the proximal colon (%F, 3.8) and the small intestine (%F, 2.2), but was low following oral gavage of the same liquid formulation used in the capsules (%F, 0.2). IntelliCap capsule transit times through each compartment of the gastrointestinal tract were recorded, and the effect of site of peptide release was investigated. When J229 was released in the small intestine, transit times through site of the release and through downstream compartments were slowed (Supplementary Table S2). Therefore, we aimed to formulate MEDI7219 tablets for peptide delivery to sites of maximal absorption in the small intestine and proximal colon.
Oral bioavailability of MEDI7219 tablets. We formulated MEDI7219 as enteric-coated oral tablets containing 20 mg of peptide with 300 mg of permeation enhancers (100 mg of NaCDC and 200 mg of PG) for pharmacokinetic studies in dogs. The enteric coating was chosen to protect the tablet from the low pH of the stomach, and to release the drug by dissolution in the neutral pH of the intestine. For comparison, semaglutide was formulated as uncoated tablets containing 20 mg of peptide and 300 mg of SNAC permeation enhancer. After oral administration, the mean bioavailability of MEDI7219 was considerably higher than that of semaglutide (%F, 5.92 vs 0.08) and the mean plasma half-life of MEDI7219 was shorter than that of semaglutide (9.8 h vs 60.5 h) ( Table 3). These half-life values were consistent with the lower levels of in vitro plasma protein binding already observed for MEDI7219 compared with semaglutide (Fig. 2b). These findings confirmed pharmacokinetic parameters suitable for once-daily oral dosing of MEDI7219 in this tablet formulation.

Effects of MEDI7219 oral tablets on weight and glucose control in a dog model of obesity and insulin resistance.
The potential anti-diabetic efficacy of MEDI7219 oral tablets was evaluated in dogs with obesity, insulin resistance and mild hyperglycemia induced by a high-fructose/high-fat (HFHF) diet 47,48 . Plasma glucose excursions were statistically significantly improved 10-40 min after an oral glucose challenge following single doses of MEDI7219 1 mg or 10 mg oral tablets, compared with placebo (P < 0.05; Fig. 5a,b). Over 14 days, dogs receiving once-daily MEDI7219 10 mg oral tablets showed significantly higher body weight loss than those receiving placebo (-4.86% vs + 0.71%, P < 0.05; Fig. 5c). This correlated with significantly decreased cumulative food intake (P < 0.05) over the 14-day period (Fig. 5d). An oral glucose tolerance test performed in fasted dogs postdose on day 14 also revealed significant improvements in glucose excursions in MEDI7219-treated dogs compared with placebo control (P < 0.01; Fig. 5e).
To investigate the potential effect of oral MEDI7219 on gastric emptying, a typical GLP-1R mediated effect 49 , an oral dose of acetaminophen, which is absorbed in the small intestine 50 , was administered during the glucose tolerance test in dogs on day 14. Peak plasma acetaminophen concentration was significantly reduced at 20-40 min (P < 0.05) and delayed in the MEDI7219 group compared with the placebo group ( Fig. 5f), indicating that MEDI7219 treatment led to a delay in gastric emptying.

Discussion
Oral delivery of peptide therapeutics represents a significant advance in the ability to administer medicines in a convenient and patient-friendly way, with the potential to improve adherence and consequently treatment outcomes. Overcoming the challenges of proteolytic degradation, gastrointestinal permeability, and rapid renal clearance to ensure high bioavailability, whilst maintaining biological function, requires a rational design process. We have presented a template for this process using a systematic multidisciplinary approach to develop a novel GLP-1RA peptide therapeutic that exhibited potent agonism, a circulating half-life suitable for once-daily dosing and significant oral bioavailability. The improved absolute oral bioavailability of MEDI7219, ~ 6% in dogs, was achieved without compromising potency by stepwise engineering based on structure-activity relationships.
The rational approach in the design and development of MEDI7219, starting with engineering the peptide to remove sites of vulnerability to both serum and gastrointestinal proteases through amino acid substitution, contrasts with previous approaches to oral peptide delivery, which have relied on retrofitting injectables to the oral route without stabilizing peptide sequences against gastrointestinal proteases. Furthermore, entericcoated MEDI7219 tablets were designed to withstand transit through the stomach releasing the peptide at the site of maximal absorption in the small intestine, which was identified with the use of the IntelliCap system. Taken together, the extensive screening identified an effective combination of permeation enhancers that greatly improved the inherently low uptake of a peptide from the gastrointestinal tract, resulting in high oral bioavailability (~ 6%) of MEDI7219 in dogs.
MEDI7219 is the first bis-lipidated GLP-1RA peptide, and this bis-lipidation approach was crucial for promoting plasma protein binding to reduce renal clearance without sacrificing agonist activity, while minimizing the use of α-methyl amino acids to stabilize the molecule. The pair of C 12   www.nature.com/scientificreports/ shorter circulating half-life than the single C 18 lipid dicarboxylic acid in semaglutide when tested in dogs. This correlated to plasma protein binding data, in which a higher fraction of unbound MEDI7219 than semaglutide was observed for plasma proteins of all species tested. Although the half-life of MEDI7219 is shorter than that of semaglutide in dogs, allometric scaling ( Supplementary Fig. S2) supports once-daily dosing in humans (t ½ 15 h). In a dog model of obesity and insulin resistance, once-daily administration of MEDI7219 oral tablets led to both decreased glucose excursion, when challenged with an oral glucose bolus, and body weight loss, consistent with the action of other GLP-1RA peptide therapeutics 23,27 . Administration of acetaminophen during the oral glucose challenge indicated that MEDI7219 delayed gastric emptying, as expected with GLP-1R agonism [51][52][53] , and this also probably contributed to the decreased glucose levels. These findings confirmed that we achieved a preclinical efficacy profile for MEDI7219 supportive of progression to clinical trials.
The average bioavailability of MEDI7219 was approximately fivefold higher than the published value for semaglutide in dogs (%F, 5.92 vs 1.22) 45 . We observed lower semaglutide bioavailability with our in-house tablets than published values, resulting in an even greater magnitude of improvement in bioavailability for MEDI7219. This suggests that some methodological differences in the manufacture of oral semaglutide tablets may have led to lower bioavailability than that reported in previously published data. Nevertheless, our results confirm a dramatic and unprecedented high oral bioavailability for MEDI7219 compared with other linear peptide therapeutics.
The dog model used to test the activity of MEDI7219 had advantages for the in vivo characterization because it allowed for oral administration of tablets, which is not possible in rodents. In addition, because dogs gained excess body weight on a high-fat diet and became insulin resistant, it was possible to ascertain GLP-1R-mediated pharmacological effects in the model following repeated oral dosing. Pharmacokinetic parameters are influenced by characteristics such as gastrointestinal barriers and plasma protein binding, and may vary among species. Furthermore, interspecies physiological differences, such as gastrointestinal pH values and transit time, may also affect formulation performance. Bioavailability variability was ~ 50% coefficient of variation in the wellcontrolled dog study but may be different in a clinical population, potentially affecting the therapeutic window.   www.nature.com/scientificreports/ Therefore, it is crucial to consider potential translational difference when moving from dog to human studies, and to monitor these concerns closely in the clinical setting.
In conclusion, here we demonstrate the very first peptide synthesized by biotechnology routes to address the intrinsic limitation of oral peptide delivery and first pass effects. Our data demonstrated reasonable bioavailability in preclinical models and therefore a very robust approach when coupled with the site of absorption and controlled drug delivery. Our systematic multidisciplinary approach for engineering a novel oral GLP-1RA offers a valuable model for the development of future oral peptide therapeutics for diabetes and other chronic conditions. This has the potential to make these peptides more accessible to patients worldwide, with the opportunity for improved patient adherence and hence reduced hospitalization or other side effects resulting from non-compliance. We leveraged advances in peptide chemistry and solid dosage formulation to develop a oncedaily oral tablet GLP-1RA with high bioavailability, and demonstrated efficacy in a preclinical model of insulin resistance and obesity. Our findings warrant clinical studies of MEDI7219 to confirm translatability of observed bioavailability and pharmacology from dogs to humans.

Materials and methods
In vivo studies. The development program involved peptide optimization using in vitro models, and in vivo potency testing in mouse models and formulation testing using rat models.

Systemic and gastrointestinal stabilization of GLP-1 receptor agonists. Amino acid substitution
and lipidation. N-α-Fmoc-L-amino acids were from Bachem (Bubendorf, Switzerland), and α-methyl and other unnatural amino acids were from Iris Biotech (Marktredwitz, Germany), Pharmaron (Beijing, China) or PepTech (Burlington, MA, USA); solvents were from Merck (Darmstadt, Germany). Peptides J211, J229 and MEDI7219 (Fig. 1) were prepared as C-terminal carboxamides on Novabiochem NovaSyn TentaGel Rink (Merck) synthesis resin using standard chemistry and coupling procedures 54 , and reagents from Sigma-Aldrich (Gillingham, UK). Amino acids following α-methyl residues were coupled twice to ensure full incorporation. The N-terminal histidine residue of GLP-1 was incorporated as Boc-His(Trt)-OH to simplify peptide cleavage. At designated lipidation positions, Fmoc-L-Lys(Mmt)-OH was incorporated into the peptide backbone during automated assembly and the Mmt protecting groups were removed upon completion. The acidified resin was quenched, and the exposed epsilon amino functions were selectively lipidated as required before peptide cleavage. Crude peptides were purified chromatographically using 5 µm Agilent Polaris C8-A (mono-lipidated; Agilent, Santa Clara, CA, USA) or Waters XBridge C18 stationary phases (bis-lipidated; Waters, Antwerp, Belgium), and lyophilized. Purified peptides were characterized by single quadrupole liquid chromatography/mass spectrometry (LC/MS) with a Waters XBridge C18 stationary phase, using a generic linear binary gradient of 10-90% methyl cyanide (MeCN; 0.1% trifluoroacetic acid [TFA] v/v) in water and the Waters MassLynx 3100 platform (ESI + mode, monitoring 3 M + H and 4 M + H ions) to verify molecular mass. Analytical reverse-phase high-performance liquid chromatography (RP-HPLC) was conducted using an Agilent Polaris C8-A stationary phase (3 µm) at 1.5 mL min -1 with a linear binary gradient of 10-90% MeCN (0.1% TFA v/v) in water, and monitored by UV absorption at 210 nm. Overall yields of J211, J229 and MEDI7219 were greater than 50% based on initial resin functionalisation (0.24 mmol/g). Mass spectrometry data for all three peptides was consistent with calculated values (given below). J211 • Requires: 3228.  55 . and used immediately. Peptides were dissolved in pre-warmed FaSSIF before addition of pre-warmed FaSSIF/pancreatin with agitation. Samples were analyzed by analytical RP-HPLC to determine remaining intact peptide by area under the curve.
EScalate equilibrium shift assay. Binding of peptides to human plasma proteins was determined by the EScalate equilibrium shift assay (Sovicell, Leipzig, Germany), using five dilutions of plasma in phosphate buffered saline and an assumed binding protein concentration of 600 µM. Samples were incubated for 1 h at room temperature then centrifuged to remove the HSA-coated beads. Supernatant samples were analyzed using HPLC (3 µm Phenomenex Aeris Widepore XB-C18 stationary phase) coupled to an electrospray ionization quadrupole time-offlight mass spectrometer in high resolution accurate mass mode (Agilent, Santa Clara, CA, USA). Unbound fractions were calculated from the concentration-dependent shift in binding equilibrium according to where P is the concentration of the binding protein in plasma and K Plasma D is the dissociation constant of the compound from plasma proteins in solution.
Peptide serial dilutions were prepared in assay buffer (Hank's Balanced Salt Solution [HBSS] containing 25 mM HEPES and 0.5 mM IBMX; pH 7.4) supplemented with 0.1% BSA or 4.4% HSA, using an Echo 550 acoustic liquid handler (Labcyte Inc., Sunnyvale, CA, USA) to obtain an 11-point concentration-response curve. Cells were suspended in assay buffer and combined with serially diluted peptides at room temperature for 30 min. cAMP levels were measured using a cAMP dynamic 2 HTRF kit (Cisbio, Codolet, France), following the manufacturer's two-step protocol, on an EnVision plate reader (PerkinElmer, Waltham, MA, USA). Data were transformed to % Delta F, as described in the manufacturer's guidelines, and expressed as % activation, in which 80 nM native GLP-1 peptide (Bachem) defines maximum effect. The transformed data were analyzed by four-parameter logistic fit to determine EC 50 values using GraphPad Prism 6 (GraphPad Software, San Diego, CA, USA).

In vivo potency of subcutaneously administered bis-lipidated peptides. Acute food intake in
C57Bl/6J mice. Male C57Bl/6 J mice 8-10 weeks of age (Jackson Laboratories, Bar Harbor, ME, USA) were single-housed in the BioDAQ (Research Diets, New Brunswick, NJ, USA) food monitoring system with ad libitum access to standard chow and water. Mice were randomized into groups of 6-9 per group on baseline 24-h food intake. On the first study day, mice were fasted for 6 h and then received a single subcutaneous dose of test peptide or placebo (50 mM Tris-HCl, 150 mM mannitol, 0.02% polysorbate 80; pH 8.0). Food intake was monitored over the next 48 h.
Body weight, blood glucose and plasma insulin in diet-induced obese mouse model. Male 20-week-old C57Bl/6 J mice were single-housed for approximately 14 weeks with ad libitum access to water and 60% high-fat diet (D12492, Research Diets). Mice were randomized into groups of 12 mice per group on baseline body weight, 6 h fasting glucose and 6 h fasting insulin. Over 21 days, mice received daily subcutaneous injections of 10 nmol/kg of test peptides or placebo (50 mM Tris-HCl, 150 mM mannitol, 0.02% polysorbate 80; pH 8.0). Body weight was measured daily and fasting glucose and insulin levels were measured on day 14 following a 6-h fast. Blood was collected via tail-snip and glucose was measured with an Ascensia Breeze 2 glucometer (Bayer, Mishawaka, IN, USA). Insulin was measured in plasma using a Meso Scale rat/insulin kit (Meso Scale Discovery, Rockville, MD, USA).

Selection of permeation enhancers for oral administration. Caco-2 screen. Caco-2 cells
were trypsinized, suspended in medium and seeded to wells of a Millipore 96-well plate following standard procedures 56 . Cells were fed at 2-day intervals for 3 weeks until a transepithelial electric resistance of ~ 1000 ohms/ cm 2 was achieved. Test samples containing peptide and permeation enhancers (Supplementary Table S1 For the single-dose study, dogs were randomized on baseline glucose levels to receive MEDI7219 1 mg or 10 mg or placebo tablets in the morning following an overnight fast (7-11 dogs per group). Ninety minutes after tablet administration, dogs received an oral glucose bolus (200 mg/mL at a 5 mL/kg volume calculated on body weight prior to HFHF feeding), and blood glucose (blood collected via peripheral vein) was measured with an AlphaTRAK glucometer system (Zoetis, Parsippany, NJ, USA) at -90, 0, 10, 20, 30, 40, 50, 60, 75, 90, 120, 180 and 240 min after glucose administration. Approximately 12 weeks later, dogs entered the multiple-dose study and received either MEDI7219 10 mg (n = 12) or placebo tablets (n = 6) once daily for 14 days. Dogs received two cups of HFHF canine diet per day, with feeding 3 h postdose during the 14-day study. Dogs had access to water ad libitum throughout the study. Food intake and body weight were monitored daily. On day 14, after overnight fasting, dogs received their final dose followed by an oral glucose and acetaminophen bolus 90 min after administration. Blood glucose was measured as described previously, and plasma acetaminophen levels were measured by LC/MS method at the same time points.
Quantification of plasma concentrations of peptides. Fifty or seventy μL aliquots of K 2 EDTA plasma samples were precipitated with 75% acetonitrile (Sigma-Aldrich) in water (J.T. Baker, Phillipsburg, NJ, USA) (v/v) and centrifuged at 2000 × g. The supernatant was dried under nitrogen at 60 °C, then cooled. Samples were reconstituted in 20% acetonitrile in water (v/v). The extracted samples were separated using a Waters Acquity UPLC BEH C18 column (2.1 × 100 mm) on a Shimadzu Nexera UHPLC at 60 °C with 0.7 mL/min flow rate, and detected using a SCIEX TripleTOF 6600 operating in full scan + MS2 positive ion mode or a SCIEX 5500 operating in MRM positive ion mode, or a Thermo TSQ Vantage operating in MRM mode. Gradient separation was www.nature.com/scientificreports/ performed with water and 0.2% formic acid (Thermo Fisher Scientific, Waltham, MA, USA) as mobile phase A, and acetonitrile with 0.2% formic acid as mobile phase B. The method was qualified for the quantification range of 1-1000 ng/mL with accuracy and precision of ± 20%, except for at the lower limit of quantification when the accuracy and precision were ± 25%. Plasma concentrations were subject to noncompartmental analysis consistent with the route of administration using Phoenix WinNonlin (version 7.0, Certara, Princeton, NJ, USA).

Statistical analyses.
Sample sizes were calculated based on power analyses for pharmacodynamic endpoints. Statistical analyses were performed using GraphPad software (San Diego, CA, USA). Results are expressed as mean ± standard error of the mean unless otherwise stated. In vivo data were analyzed with one-way analysis of variance (ANOVA) followed by Tukey post hoc analysis (acute food intake, db/db HbA 1C data). Longitudinal data were analyzed by two-way ANOVA followed by Tukey post hoc analysis (db/db glucose and DIO mouse data) or two-way ANOVA followed by Sidak or Bonferroni post hoc analysis (HFHF canine acute oral glucose tolerance test data and 14-day data). In all statistical tests P < 0.05 was considered significant. One animal in the single MEDI7219 10 mg dose group for the glucose tolerance test was excluded from analysis due to a likely error in dosing. In the analysis of the effect of MEDI7219 on body weight in HFHF fed dogs, one animal in the placebo group and one datapoint on day 10 in the MEDI7219 group were excluded due to errors in measurement. No other animals or data points were excluded from any of the in vivo study analyses.