FGF21 Analogs of Sustained Action Enabled by Orthogonal Biosynthesis Demonstrate Enhanced Antidiabetic Pharmacology in Rodents

A human FGF21 theoretical model was constructed by homology modeling using the crystal structures of FGF19 (Protein Data Bank code 2P23 and 2PWA) (7,8) and FGF23 (Protein Data Bank code 2P39) (7); all three belong to the FGF19 subfamily. The FGF19 subfamily members use Klotho proteins to facilitate their interactions with FGF receptors (FGFRs), rather than heparin sulfate as other FGF family members (7). Building on previous biochemical and structural work characterizing the FGF family receptor-binding interface (7–10), amino acids with surface-exposed side chains distal from the putative receptor binding regions were selected for substitution with pAcF and subsequent PEG attachment.

Using previously described methods (6), an Escherichia coli K-12 W3110 cell line was generated to create a temperature-sensitive cell line with an additional W375R point mutation to confer lethal host phenotype at temperatures ≥37°C, which was subsequently transformed with an FGF21Q108amber plasmid into DH5α-T1 cells (Invitrogen Corp., Carlsbad, CA). FGF21Q108pAcF protein production was confirmed through SDS-PAGE and liquid chromatography-mass spectrometry.

Production and purification of PEGylated His₆-FGF21 variants was accomplished as described earlier (6). The PEGylated His₆-tagged FGF21 protein was >95% pure, with endotoxin levels in each variant <2 EU/mg protein by LAL analysis with EndoSafe-PTS LAL cartridges (Charles River Laboratories, Wilmington, MA).

For production of PEGylated FGF21 variants lacking the His₆ tag, frozen cell paste was resuspended in 9 mL buffer per gram of paste in 50 mmol/L Tris, 100 mmol/L NaCl, 1 mmol/L EDTA, 1% Triton X-100, pH 8.0. The lysate was centrifuged at 10,000 g for 15 min at 4°C, and the pellet was washed using the same resuspension and centrifugation steps. Two additional washes were performed in 50 mmol/L Tris, 100 mmol/L NaCl, 1 mmol/L EDTA, pH 8.0. The pellet containing purified inclusion bodies was solubilized in 9 mL buffer per gram of inclusion bodies in 20 mmol/L Tris, 8 mol/L urea, 10 mmol/L β-mercaptoethanol for 1 h at room temperature, and refolded in 20 mmol/L Tris, pH 8.0, at a 9:1 (v/v) ratio of refold buffer:solubilized inclusion bodies at 4°C for 18–24 h. The refolded solution was processed through 0.2-μm filtration, anion-exchange capture (Q-Sepharose high-performance column; 20 mmol/L Tris, pH 8.0, 0–250 mmol/L NaCl gradient) and a hydrophobic interaction column (phenyl Sepharose high-performance column; binding buffer: 20 mmol/L Tris, 0.55 mol/L ammonium sulfate, pH 8.0; elution buffer: 20 mmol/L Tris, 2 mol/L urea, pH 8.0 gradient elution). The PEGylation and post-PEGylation purification methods were as previously described (6). All variant masses for in vitro and in vivo testing are reported as milligrams of protein without inclusion of the mass of any conjugated PEG.

HEK293 β-Klotho clone #4 cells were seeded at 100,000 cells/well in a poly-d-Lys–coated 96-well plate and incubated overnight at 37°C and 5% CO₂ to allow cells to adhere. The next day, fivefold serially diluted wild-type (WT) His₆-tagged FGF21 and His₆-tagged PEGylated FGF21 proteins were added to the cells in duplicate for 9 min at 37°C and 5% CO₂. After stimulation, cold lysis buffer containing protease and phosphatase inhibitors was added to the wells and incubated on ice for 25 min to allow complete cell lysis. After centrifugation, the lysate was transferred to a 96-well Phospho/Total-ERK1/2 MultiSpot plate (Meso Scale Discovery, Gaithersburg, MD). Lysates were analyzed on a Meso Scale Discovery SI2400 reader according to protocol, and half-maximal effective concentration (EC₅₀) values were calculated for all proteins. His₆-tagged PEGylated FGF21 proteins were rank ordered by their EC₅₀ values. The top five PEGylated FGF21 proteins, lacking the His₆ tag, were also prepared and demonstrated comparable EC₅₀s.

Jugular vein–catheterized, male Sprague Dawley rats (Charles River Laboratories) were housed in individual cages in a temperature-controlled environment with a 12-h light-dark cycle with ad libitum access to water and standard rodent chow (Diet 5001; Purina). Rats were administered a single, subcutaneous injection of WT FGF21 or PEG30-FGF21 variants lacking the His₆ tag (0.25 mg/kg). Blood samples were collected at various time points and analyzed by a bridging ELISA (see below) to determine compound concentrations. All experiments were conducted under protocols approved by the Institutional Animal Care and Use Committee of Ambrx, Inc. Pharmacokinetic analyses were performed using WinNonLin software (version 5.0.1; Pharsight) and noncompartmental modeling.

Standards, quality controls, and samples were spiked into 100% Sprague Dawley rat serum. All standards, quality controls, and study samples were then loaded onto the wells after pretreating 1:50 with assay buffer. The ARX618 in the standards, quality controls, and samples were captured by immobilized anti-PEG monoclonal antibody that had been coated onto the plate. After a wash step to remove unbound materials, biotinylated goat anti–FGF21 IgG polyclonal antibody was added to the wells. After another wash step, streptavidin sulfo tag was added for detection of the captured ARX618. After a final wash step, Meso Scale Discovery read buffer was added to all wells, and light was generated in proportion to the amount of ARX618 bound by the capture and detection reagents. The conversion of relative light units for the study samples and the quality controls to concentration was achieved through comparison with a standard curve on the same plate, which was regressed according to a four-parameter logistic regression model with a weighting factor of 1/Y². Results are reported in ng/mL concentration units.

Male db/db mice were purchased at 5 weeks of age from The Jackson Laboratory and housed eight per cage in temperature-, humidity-, and light-controlled rooms with ad libitum access to autoclaved water and food (Diet 5008; Purina). At the beginning of the experiment, db/db mice (8 weeks of age) were grouped based on BW and ambient glucose levels. All db/db studies were performed with WT FGF21 or PEG30-FGF21 variants lacking the His₆ tag.

For chronic efficacy evaluation, animals were dosed subcutaneously twice weekly (BIW; WT or PEG30-FGF21) or twice daily (BID; WT FGF21) and studied for 12 days. Glucose, BW, and food consumption were monitored as indicated after commencement of treatment. Plasma triglyceride, free fatty acid, insulin, glucagon, and liver triglyceride levels, pancreas staining/quantification, and islet functional studies were performed as previously described (11). Glucagon and insulin were measured using commercial ELISA kits (Linco Research Immunoassay, St. Charles, MO, and ALPCO Diagnostics, Windham, NH, respectively).

For the insulin tolerance test, db/db mice (12 weeks of age) were sorted based on blood glucose. Diet was removed at 9:00 a.m. and mice were injected subcutaneously with vehicle or PEG30-FGF21. Humulin R (3 units/kg; Eli Lilly) was administered 6 h afterward. Blood glucose from tail vein was measured by glucometer (Ultra LifeScan; OneTouch, Milpitas, CA) at the indicated time points pre- and postchallenge.

To compare insulin signaling using phospho-Akt as a marker, 12-week-old male db/db mice were grouped based on blood glucose. Diet was removed at 9:00 p.m. and mice were injected subcutaneously with vehicle or PEG30-FGF21 at 0.25 mg/kg. Saline or Humulin R (2 units/kg) was injected by tail vein the next morning at 9:00 a.m.. Liver, gastrocnemius muscle, and white and brown adipose tissue were collected 5 min postinjection and snap-frozen in liquid nitrogen for phospho-Akt analysis with the whole-cell lysate phospho-Akt (Ser473) assay kit according to the manufacturer's protocol (Meso Scale Discovery).

Nine-week-old db/db mice received a subcutaneous injection of either vehicle or PEG30-FGF21 at 0.75 mg/kg. One and three hours postdose, mice received an intraperitoneal injection of ³H-H₂O at 2.5 mCi/kg. Mice were killed 1 h post–³H-H₂O injection, and hepatic de novo lipogenesis (DNL) was determined as described previously (12).

Six-week-old, male db/db mice were acclimated to a pathogen-free animal facility for 1 week, where they were maintained at 22°C on a 14-h:10-h light-dark cycle. They were placed in individual boxes within the temperature-controlled chambers of an OxyMax metabolic system (Columbus Instruments, Columbus, OH). The mice were randomized according to their glucose levels and divided into two groups (n = 8); one cohort received 0.25 mg/kg PEG30-R131 subcutaneously once daily for 12 days while the other was injected with vehicle. The OxyMax metabolic chamber system ran 24 hours a day, 7 days a week to collect oxygen consumption (Vo₂) and CO₂ production (Vco₂) data nonstop except when the mice were dosed right after their dark cycle ended. The respiratory exchange ratio (RER) was calculated as Vco₂/Vo₂.

All data are expressed as means ± SEM. Statistical analysis was conducted using ANOVA or unpaired Student t test where appropriate. Statistical significance was defined as P < 0.05.

A human FGF21 theoretical model was constructed by homology modeling, using the crystal structures of FGF19 and FGF23 (7,10). Building on previous biochemical and structural work characterizing the FGF family receptor-binding interface (9), amino acids with surface-exposed side chains distal from the putative receptor binding regions were selected for substitution with pAcF and subsequent PEG attachment (see 2research design and methods).

On the basis of the aforementioned modeling, a series of human FGF21 variants was constructed by inserting the novel amino acid pAcF at single specific sites within the protein using the ReCode methodology. These variants were then biosynthesized, PEGylated with a 30-kDa PEG, and purified as described in 2research design and methods. The variants that were generated are listed in Table 1.

His₆-tagged versions of the PEGylated FGF21 variants, designated by the location of the novel amino acid insertion, were tested for in vitro activity by examining their ability to augment extracellular signal–related kinase phosphorylation (pERK) within a HEK293 cell line engineered to stably express β-Klotho. As previously described (7), FGF21 intracellular signaling activity required the endogenous expression of FGFRs and the forced expression of β-Klotho. Table 1 shows the potencies with which His₆-tagged WT FGF21 and the corresponding variants induced pERK levels. There was a good correlation among the variants between pERK activity and location within the homology model. Variants Y22, K56, L99, and Y104, which are all within the modeled FGFR binding sites, proved to be entirely inactive. In contrast, those variants most distal to the binding sites, R77, L86, Q108, and R131, proved to be most potent, whereas other variants were found to be of intermediate activity. The potency of the most active PEGylated proteins was diminished approximately fivefold compared with WT FGF21. Glucose uptake assays performed in 3T3-L1 adipocytes demonstrated the superior potency of a largely overlapping set of variants, R77, L86, E91, Q108, and R131, whereas K56, L99, and Y104 were once again found to be inactive (not shown). Therefore, we found good concordance of PEGylated variant potency using different means of assessing their functional activity in vitro.

The most potent PEGylated variants, R77, L86, E91, Q108, and R131, were selected for single-dose subcutaneous testing in male rats and prepared without the His₆ tag. As shown in Supplementary Fig. 1, all of these PEGylated proteins had substantially prolonged pharmacokinetic profiles versus the WT protein. Beyond this general PEG effect on circulating FGF21 levels, important differences in pharmacokinetics were seen among the variants conjugated to the same size PEG. For example, compared with the WT FGF21–terminal half-life of 1.19 h, 30-kDa PEG differentially prolonged the half-lives of PEGylated variants across a range extending from 14.72 (R72) to 33.9 h (L86), a greater than twofold difference based solely on the site of attachment. The areas under the curve (AUCs) showed a similar range, with the WT protein displaying an AUC of 259.8 ng ⋅ h/mL, whereas those of the PEGylated variants ranged from 3,216 ng ⋅ h/mL (L86) to 13,238 ng ⋅ h/mL (R126). These differences in circulation time were not predicted from the protein homology modeling, nor did they correlate with measured in vitro activities of the variants. Thus, the specific site of PEGylation can alter pharmacokinetics beyond the choice of PEG size itself. The combination of well-preserved in vitro activity and enhanced plasma levels warranted the further study of these five FGF21 variants in a setting where in vivo efficacy could be determined.

Several previous studies characterizing the actions of WT FGF21 (1,4) have demonstrated its glucose-lowering activity in rodent models that displayed concurrent BW reduction, making it difficult to determine the direct antihyperglycemic activity of the protein. Additionally, these earlier studies had to use aggressive dosing regimens in order to maintain pharmacologically active levels of WT FGF21. To avoid alterations in BW and concerns with the short half-life of WT FGF21, we conducted studies with the PEG30-FGF21 variants Q108 and R131 in db/db mice, an obese, hyperphagic model of T2D that lacks a functional leptin receptor. During a 12-day experiment, once-daily subcutaneous dosing of either PEG30-FGF21 analogs lowered ambient glucose robustly at doses as small as 0.1 mg/kg (data not shown). To examine efficacy with an extended dosing interval, mice were dosed subcutaneously BIW with Q108 and R131 and monitored for decreases in glucose over 12 days. When administered at 0.75 and 2.5 mg/kg, both variants dose dependently lowered circulating glucose levels and essentially normalized hyperglycemia at the higher dose (Fig. 3A). Unlike results reported in other animal models (1,2,4,5), db/db mouse BW was maintained during the course of the study (Fig. 3B), supporting the proposition that the decrease in ambient glucose seen in this T2D model was a direct effect and not secondary to BW reduction. Reduction of fasting glucose (Fig. 3C) and improved glucose tolerance (data not shown) were also detected during the study, notably without hypoglycemia even at the highest doses. BID dosing of WT FGF21 at 1.25 mg/kg resulted in glucose-lowering efficacy similar to PEG30-Q108 and PEG30-R131 (Fig. 3D), whereas BIW dosing of WT FGF21 at 2.5 mg/kg had no effect on ambient glucose; neither dosing paradigm caused BW change (data not shown). Taken together, these results indicate that potent PEG30-FGF21 variants displayed robust antihyperglycemic activity in diabetic mice without affecting their BW. Furthermore, these PEGylated proteins provided comparable efficacy to WT FGF21 when dosed far less frequently and at lower cumulative doses, a likely result of their significant intrinsic activity and greatly increased in vivo half-life.

In addition to mitigating elevated glucose levels, PEGylated Q108 and R131 both reduced plasma and liver triglyceride levels in db/db mice at the termination of the 12-day study (Supplementary Fig. 2A and B). Additionally, a single injection of PEGylated R131 significantly reduced hepatic DNL as soon as 1 h postdose (Supplementary Fig. 2C), thereby providing a potential mechanism for the observed reductions in dyslipidemia and hepatic steatosis.

At the termination of the 12-day study, glucagon levels were significantly reduced in db/db mice treated with PEGylated Q108 and R131 (Fig. 4A). Pancreatic insulin content (Fig. 4B) generally trended higher, with the effect of Q108 reaching statistical significance at 2.5 mg/kg. In comparison with vehicle treatment, PEG30-Q108 increased total islet number from 51.2 ± 6.8 to 98.2 ± 4.2 10E6/mm²; this augmentation was distributed across the range of observed islet sizes (Fig. 4C). In a separate study, db/db mice were dosed BIW with PEG30-Q108 at 2.5 mg/kg for 6 days, and islets were subsequently isolated to measure insulin content and secretion; the same number of size-matched islets were isolated and examined from each animal. Significant increases in static islet insulin content (data not shown) and ex vivo glucose-dependent insulin secretion (Fig. 4D) were found in islets from PEGylated Q108– versus vehicle-treated mice.

To determine whether PEG30-FGF21 affects insulin sensitivity in diabetic db/db mice before glucose lowering is observed, the animals were given a single injection of PEG30-R131 at 0.25, 0.75, or 2.5 mg/kg. This treatment resulted in reduced fasting plasma insulin and improved insulin tolerance in a dose-responsive manner 12 h postdose (Fig. 5A–C). We also profiled phospho-Akt levels in four major metabolic tissues 15 min after an intravenous injection of vehicle or insulin. The results (Fig. 5D–G) indicated that a single dose of PEGylated R131, provided 12 h in advance of insulin administration, significantly improved insulin signal transduction in liver, skeletal muscle, and white and brown adipose tissue. Together, these data demonstrated that a single dose of a potent PEG30-FGF21 variant led to a rapid improvement of insulin sensitivity in db/db mice without causing hypoglycemia, even at higher doses.

After observing the ability of PEG30-FGF21 to dramatically reduce hyperglycemia in db/db mice without increasing BW, we hypothesized that the glucose was being taken up by various metabolic tissues and oxidatively catabolized. In order to test this proposition, we placed db/db mice in an OxyMax metabolic chamber system and injected them daily for 12 days with 0.25 mg/kg PEG30-R131 while constantly monitoring their Vo₂ and Vco₂. These measurements were used to calculate the RER of the mice. We found that although both cohorts began the study with an RER of ∼0.86, the RER of those treated with the PEG30-FGF21 variant rose to ∼0.91 over the course of 4 days and maintained this elevated level throughout the treatment regimen (Fig. 6). In contrast, the control group's RER trended down to ∼0.83 during the course of the study. These results demonstrated that treatment of db/db mice with PEGylated R131 rapidly and durably increased oxidation of glucose, thereby allowing it to be removed from circulation without accumulating in insulin-responsive tissues as glycogen or lipids that would increase BW.

Recent studies in preclinical species have shown that FGF21, an atypical endocrine hormone–like member of the FGF superfamily, can address multiple pathophysiologic elements of T2D. These include mitigation of hyperglycemia and insulin resistance, the preservation of insulin production and pancreatic β-cell function, amelioration of dyslipidemia and altered lipid storage, and improvement of intermediary metabolism substrate fluxes (1–5). To our knowledge, no adverse effects of FGF21, including hypoglycemia at high doses, have yet been described in the literature. Thus, FGF21 has emerged as a promising candidate for the treatment of T2D.

Unfortunately, the short duration of circulation and action of native FGF21 undermines its therapeutic utility. Thus, the search for a means of prolonging the pharmacokinetics of FGF21 while maintaining its intrinsic activity led to the utilization of a novel, selective PEGylation strategy. The importance of such selectivity has been amply demonstrated, because conventional PEGylation of therapeutic proteins typically leads to the generation of heterogeneous mixtures, or less than optimal site selection (6). As a result, the specific activity is often reduced relative to the native protein, thereby necessitating higher clinical doses that lead to increased concerns about increased immunogenicity, renal vacuolation, and treatment cost.

Homology modeling allowed directed selection of pAcF site incorporation predicted to be distant from the receptor binding regions of human FGF21. Functional characterization of the PEGylated variants using cell-based assays measuring induction of ERK phosphorylation and glucose uptake validated the structural model we had produced, demonstrating that mono-PEGylation of FGF21 with a linear PEG30 polymer at putative receptor binding sites caused large losses in protein potency, whereas PEGylation at sites distal to those involved in FGFR interactions maintained high potency. These findings indicated that site matters in prolonging time of action while maintaining specific activity. Recode methodology enables selection based on biological properties as opposed to conventional approaches rooted in the native placement of lysine. In addition, we showed that the same PEG30 polymer differentially improved the pharmacokinetic character of FGF21, when conjugated to different sites, perhaps due to differential absorption and proteolytic lability. These data collectively illustrated the pharmacological power in being unconstrained in site selection through biosynthesis with unique chemical functionality.

With no detectable weight change upon exposure to WT FGF21 or its PEGylated variants, db/db mice served as a good model to assess the direct metabolic actions of this therapy in a stringent rodent model of T2D. We found that BIW administration of the five most potent PEGylated FGF21 variants led to impressive glucose lowering with no sign of hypoglycemia during the 12-day study period. The rank ordering of their glucose-lowering potencies was similar to that observed in the pharmacokinetic study, with the best overall performance demonstrated by variants Q108 and R131. Such data provide strong evidence that FGF21 has direct antihyperglycemic activity and that site-specific mono-PEGylation of FGF21, as described in this work, provided molecular entities with superior pharmacological properties.

Accompanying the antihyperglycemic efficacy, treatment with PEGylated FGF21 analogs preserved pancreatic islet morphology, size, and number and improved glucose-dependent insulin secretion. These results are important because loss of islet function and mass are hallmarks of T2D (13), and the ability to mitigate endocrine pancreas deterioration promises durable antidiabetic therapy.

PEGylated FGF21 variants enhanced insulin sensitivity, as was detectable by diminished fasting insulin and improved insulin tolerance as early as 12 h after a single administration and long before significant glucose-lowering efficacy was observed. Additionally, increased insulin sensitivity was manifested by potentiation of insulin-induced Akt phosphorylation in major metabolic tissues, including liver, skeletal muscle, and adipose, which suggests that these tissues mediate the observed FGF21-induced improvements in whole-body insulin sensitivity and glucose disposal.

Acute administration of PEG30-FGF21 resulted in a rapid diminution in DNL. Though modest in scale, this change probably contributed to the reduction in hepatic steatosis seen after chronic treatment with PEGylated FGF21, which in turn would fortify its aforementioned insulin-sensitizing activity. The inhibition of hepatic DNL by PEG30-FGF21 might also mediate the observed improvement in dyslipidemia.

Our work demonstrated that FGF21 can modify nutrient utilization based on the availability of fuel substrate. In particular, when provided to hyperglycemic db/db mice, PEG30-FGF21 durably increased whole-body carbohydrate oxidative catabolism as indicated by an elevation in RER. Therefore, we proffer that PEGylated FGF21 enhances metabolic flexibility, i.e., the switch to insulin-stimulated glucose oxidation during periods of elevated circulating glucose, which has previously been shown to be impaired in T2D (14). Alleviating such a metabolic derangement could aid in the mitigation of insulin resistance.

In summary, our findings demonstrate that site-specific PEGylation of FGF21 results in the generation of potent, long-acting molecules with highly desirable antidiabetic therapeutic profiles. Such uniquely modified proteins hold significant promise in addressing the complex syndrome of T2D.

This study was funded by Ambrx, Inc. and Merck Research Laboratories. During the course of the research presented in this article, J.P., L.S., H.M., A.-M.P., S.B., T.C.N., B.K., A.M., D.L., and D.W.A. were employed by Ambrx, Inc. and J.M., Z.L., N.L., Q.D.-Y., J.Y., M.W., C.G.R., J.M.M., D.M.Z., B.B.Z., and J.P.B. were employed by Merck Research Laboratories. R.D.D. is professionally related to Ambrx, Inc. No other potential conflicts of interest relevant to this article were reported.

J.M. and J.P. designed and performed research, analyzed data, and wrote the manuscript. Z.L. performed research and analyzed data. L.S. and N.L. designed and performed research and analyzed data. H.M. designed and performed research, contributed new reagents/analytic tools, and analyzed data. Q.D.-Y. performed research and analyzed data. A.-M.P. designed and performed research and analyzed data. J.Y. performed research and analyzed data. S.B. designed and performed research. M.W. performed research and analyzed data. T.C.N. designed research, analyzed data, and wrote the manuscript. C.G.R. and A.M. performed research and analyzed data. B.K. designed research and analyzed data. J.M.M. designed and performed research and analyzed data. D.L. performed research. D.M.Z. and B.B.Z. designed research. R.D.D. wrote the manuscript. J.P.B. designed research, analyzed data, and wrote the manuscript. D.W.A. analyzed data and wrote the manuscript. J.P. is the guarantor of this work and, as such, had full access to all the data in the study and takes responsibility for the integrity of the data and the accuracy of the data analysis.

The authors gratefully acknowledge Drs. Guoqiang Jiang, Cai Li, and Laurent Audoly (Merck Research Laboratories) for their invaluable assistance during the course of these studies. The authors also thank Nancy Thornberry and Dr. Catherine Straeder (Merck Research Laboratories) for their support in this endeavor.