Design and Pharmacokinetic Characterization of Novel Antibody Formats for Ocular Therapeutics

Abstract

Purpose: To design and select the next generation of ocular therapeutics, we performed a comprehensive ocular and systemic pharmacokinetic (PK) analysis of a variety of antibodies and antibody fragments, including a novel-designed bispecific antibody.

Methods: Molecules were administrated via intravitreal (IVT) or intravenous (IV) injections in rabbits, and antibody concentrations in each tissue were determined by ELISA. A novel mathematical model was developed to quantitate the structure–PK relationship

Results: After IVT injection, differences in vitreal half-life observed across all molecules ranged between 3.2 and 5.2 days. Modification or elimination of the fragment crystallizable (Fc) region reduced serum half-life from 9 days for the IgG to 5 days for the neonatal Fc receptor (FcRn) null mAb, to 3.1 to 3.4 days for the other formats. The F(ab')₂ was the optimal format for ocular therapeutics with comparable vitreal half-life to full-length antibodies, but with minimized systemic exposure. Concomitantly, the consistency among mathematical model predictions and observed data validated the model for future PK predictions. In addition, we showed a novel design to develop bispecific antibodies, here with activity targeting multiple angiogenesis pathways.

Conclusions: We demonstrated that protein molecular weight and Fc region do not play a critical role in ocular PK, as they do systemically. Moreover, the mathematical model supports the selection of the “ideal therapeutic” by predicting ocular and systemic PK of any antibody format for any dose regimen. These findings have important implications for the design and selection of ocular therapeutics according to treatment needs, such as maximizing ocular half-life and minimizing systemic exposure.

Eye diseases contributing to loss of visual acuity have major societal impacts, accounting for loss of productivity in the workplace and of personal independence.¹ Some of the most debilitating diseases of the eye occur in the elderly (>55 years) and in diabetic patients. In response, research efforts in biotechnology have grown rapidly in recent years to address unmet medical needs such as lack of efficacy and dose frequency. Age-related macular degeneration (AMD) and diabetic macular edema (DME) are complex diseases involving the central retina, specifically the macular region that is responsible for sharp central vision. Hallmarks of these include atrophy or pathologies of the vascular beds that service the retinal tissue.^1,2 

Biologics breakthrough therapy in treating AMD and subsequently DME was ushered in with the off-label use of bevacizumab and subsequent approval of ranibizumab (Genentech, Inc., South San Francisco, CA, USA), an antigen-binding fragment (Fab) against the vascular endothelial growth factor (VEGF) administered by intravitreal (IVT) injection.³ The use of a Fab in the eye has been highly successful from a therapeutic protein due to short systemic exposure.⁴ The effectiveness of anti-VEGF therapy has been followed with the use of different anti-VEGF molecules including aflibercept, a fragment crystallizable (Fc) region fusion (Regeneron, Tarrytown, NY, USA). Using biologics in the eye is a relatively new advance, and therefore little is understood about antibody and antibody fragment pharmacokinetics after IVT administration. With an ongoing interest across the research community to develop more effective, next-generation ocular protein-based therapeutics, we characterized a comprehensive set of molecules in rabbits to better understand the pharmacokinetic impact of molecular size, structure and relationship with neonatal Fc-receptor (FcRn). 

A major goal of this study was to provide reliable data to guide the design and selection of new molecules for intraocular treatment, including bispecific antibodies. Improving upon the efficacy of current therapeutics is a key strategy in the ophthalmology field, and is a highly sought-after goal by many pharmaceutical and biotechnology firms. Examples of single-agent preclinical efficacy of anti-angiopoietin2 (ANG2) activities exist in the literature and could be considered for potential combination with anti-VEGF using a bispecific.^5–7 

Beyond improving efficacy and/or durability, understanding the implications of systemic exposure after IVT injection was a secondary goal of the study. Molecular design may also help mitigate safety concerns and the side effects of chronic monthly administration of anti-VEGF therapy. Serious systemic adverse events primarily related to cardiovascular health have been suggested in database meta-analyses and in pivotal and comparative trials (e.g., CATT).⁸ Ultimately, an understanding of pharmacokinetics in the eye of various antibody fragments derived from the native IgG structure will guide development of potent and durable VEGF-based bispecific molecules for optimal residence in the intraocular compartment, while limiting systemic exposure to minimize the potential of adverse events. 

The success of ranibizumab has prompted interest in exploring the potential of the F(ab')₂ as an attractive structure (Fig. 1A) due to the availability of two Fabs and the lack of the Fc region, enhancing the targeting properties of these molecules and retaining the favorable systemic PK profile seen with Fabs.⁹ To characterize the key structure–PK relationships of these protein drug forms, we conducted a side-by-side comparison of the pharmacokinetics of a mAb, Fab, conventional F(ab')₂ (double-disulfide bond [dDS]), F(ab')₂ with a single-disulfide link (sDS), a mAb with Fc mutations that prevent binding to FcRn (FcRn null mAb),¹⁰ and an Fc-only construct following IVT and IV administration in rabbits. These studies allowed us to assess two standing hypotheses in ocular PK: that molecular weight has a major impact on PK, and that FcRn impacts antibody clearance from the eye.^11–16 A mathematical model was developed and validated for characterization of the pharmacokinetics of the different drug formats in both the ocular tissues and systemic circulation. The model is utilized for quantifying the PK differences to support selection of the “ideal” drug format. Our analyses suggest that balancing systemic liabilities by reducing systemic exposure, molecular stability, and longer half-life in the vitreous can be achieved with a F(ab')₂ fragment. 

Figure 1

 

Different antibody forms and fragments used for ocular pharmacokinetics assessment. (A) Multiple antibody and antibody formats were developed as test molecules (nontarget binding) to evaluate the impact of size and FcRn on ocular and systemic pharmacokinetics. C_H, constant domain heavy chain; C_L, constant domain light chain; DS, disulfide bond; Fab, antigen-binding fragment; Fc, constant region; kDa, kilodalton; mAb, monoclonal antibody; V_H, variable-domain heavy chain; V_L, variable-domain light chain. (B) Model schematic describing ocular and systemic PK following intravitreal or intravenous administration.

These systematic analyses further support the design of different formats to be characterized by future preclinical animal studies and subsequently translated to humans. The impact of this work is not only on the design of the bispecific but also in the clear and thorough description of the PK behavior in different relevant components of the eye. 

Methods

Generation of F(ab)s

All biological materials were made within Genentech facilities. Human IgG1 half-antibodies (anti-VEGF or anti-angiopoeitin-2 [ANG2] and human anti-glycoprotein D [anti-gD]) IgG1 were digested 1/1000 (wt/wt) with the endoproteinase Lys-C (Wako Chemicals, Richmond, VA, USA) at 37°C and pH 7.5 for 3 hours or until the Fc was completely cleaved. The protease was then deactivated by the addition of a 100 M excess inhibitor-to-enzyme ratio with N-tosyl-L-lysine chloromethyl ketone hydrochloride. The digest products were subsequently purified by cation exchange using SP HP (GE Healthcare, Pittsburgh, PA, USA) with a 30 column volume (CV) gradient from 0 to 300 mM NaCl in 50 mM sodium acetate in pH 5.0. 

Generation of F(ab')₂ With Two Inter-Heavy-Chain Disulfides

Bispecific human anti-ANG2/anti-VEGF human anti-gD antibodies were digested with pepsin at 1/500 at pH 3.0 for 2 hours or until the Fc had been removed. The bispecific antibody was produced by a half-antibody expression and assembly method. The pH of the material was then adjusted to 5.0, and the digests were purified by cation exchange using SP HP (GE Healthcare) and a 30 CV gradient from 0 to 300 mM NaCl in 50 mM sodium acetate at pH 5. The material was subsequently purified by S75 Superdex Gel Filtration (GE Healthcare) to formulate and remove any undigested antibody or fragments. 

Generation of F(ab')₂ With One Inter-Heavy-Chain Disulfide

The heavy chain in anti-gD was changed at position C229 (EU numbering) to S229 by QuikChange Site-directed mutagenesis. The molecule was then expressed by transient transfection in CHO cells and purified by standard Protein A capture. The purified antibody was digested with pepsin as described for F(ab')₂ with two inter-heavy-chain disulfides and purified in a similar manner. The anti-VEGF, anti-alpha5beta1, and anti-ANG2 Fab and Fab' were cloned into Escherichia coli expression vectors and expressed as previously described.¹⁷ The C-terminal residues for the Fab and Fab' construct are T225 and C226 (both EU nomenclature), respectively. Purification of anti-VEGF was done with Protein G FF (GE Healthcare), and purification of anti-ANG2 was done with CH1 Select resin (Life Technologies, Grand Island, NY, USA). The affinity-purified anti-VEGF and anti-ANG2 Fabs containing a single C-terminal cysteine were dialyzed into 25 mM Tris, 150 mM NaCl, 5 mM dithiothreitol, pH 8.0 to remove adducts on C231. The proteins were then dialyzed into 10 mM Tris, 150 mM NaCl, pH 8.0 to allow the interchain disulfide between the heavy and light chains to reform. Oxidation was monitored by mass spectrometry (Agilent, Santa Clara, CA, USA) until the heavy and light chains had reoxidized, after which 2 mM EDTA was added and the pH of the solution was lowered to 5.0. 

The anti-VEGF Fab' was reacted with a 30 M excess of 2,2′-dipyridyl disulfide (DPDS) and monitored by mass spectrometry for complete adduction by the disulfide exchange reagent. The disulfide-activated Fab', anti-VEGF-pyridyl disulfide, was purified on an S75 Superdex in 25 mM sodium succinate, 150 mM NaCl, 5 mM EDTA, pH 6.0 to remove unreacted DPDS and unreacted Fab'. The anti-ANG2 was purified similarly but without DPDS. The purified Fab' molecules were mixed and concentrated to 10 mg/mL and incubated at room temperature overnight. After overnight incubation, the conjugate was complete and the product was isolated using SP HP cation exchange chromatography as described for the F(ab')₂ purification. A final purification step was done using S75 Superdex Gel Filtration in 25 mM sodium succinate, 150 mM NaCl, pH 6.0. All antibodies or fragments were formulated for IVT injection by concentrating to 10 mg/mL in 20 mM histadine, pH 5.0, and 150 mM NaCl. 

Generation of Anti-gD Fc

The anti-gD antibody was digested using lysine-C as described above. Anti-gD (250 mg) at 5 mg/mL was digested with lysine-C at 1/1000 ratio overnight at room temperature. After digestion, the reaction mixture was purified by cation exchange chromatography using SP HP (GE Healthcare) with a 30 CV gradient from 0 to 300 mM NaCl in 50 mM sodium acetate in pH 5.0. 

Animals

Male New Zealand White rabbits weighing 2.5 to 3.5 kg (Covance, Princeton, NJ, USA) were randomly assigned to one of four separate PK studies: anti-gD antibody and Fab (n = 39), anti-gD F(ab')₂ in single- and double-disulfide formats (n = 42), anti-ANG2/anti-VEGF bispecific F(ab')₂ in single- and double-disulfide formats (n = 45), and alpha5beta1/anti-VEGF (n = 22). Animals were treated and handled in accordance with the Animal Welfare Act, the Guide for the Care and Use of Laboratory Animals with protocols approved by Genentech, Inc., the Institutional Animal Care and Use Committee, and the ARVO Statement for the Use of Animals in Ophthalmic and Vision Research. All animals were single housed with water and food available ad libitum. Animals were acclimatized for 4 or 5 days prior to dose administration. 

At 3 days prior to dosing, animals were randomly assigned to experimental groups, ear marked, weighed, and prepared for dosing (IV groups only). Pharmacokinetic studies were designed to incorporate matched IV and IVT experimental groups irrespective of the test molecule. Serial blood samples were collected from IV experimental groups (n = 3 per group) throughout the study. In contrast, IVT experimental groups (n = 12–18, depending on the test molecule and study duration) were terminally sampled for blood and ocular tissues at predetermined time points. The number of animals used in the study was determined on the basis of the minimum needed for calibration of the PK model. 

Animal Dosing

For IV dosing, all test articles were formulated under sterile conditions in 25 mM histidine/150 mM sodium chloride buffer (pH 5.8), and were administered as a single IV dose to the marginal ear vein at 500 mg/1 mL per animal. 

For IVT dose administration, animals were anesthetized using isoflurane (induction: 5%; maintenance for entire dose administration protocol: 3%) and study eyes were dilated with 1% tropicamide ophthalmic solution. Topical ophthalmic anesthetic (0.5% proparacaine) was administered to each eye, followed by 0.2 mL 0.02 to 0.1 mg/kg subcutaneous (SC) buprenorphine in the scapular region. 

All IVT test articles were formulated under sterile conditions in 25 mM histidine/150 mM sodium chloride buffer (pH 5.8), and were administered via single bilateral IVT injections at 500 μg/50 μL/eye. All IVT dosing was conducted in the left eye first. A sterile insulin 28-gauge needle was introduced into the inferior temporal quadrant of the ocular globe and dose administered into the inferior vitreous body. This procedure was then repeated for the right eye. From day 1 following dosing, animals received twice-daily 0.02 to 0.1 mg/kg SC buprenorphine for a minimum of 3 days, and Puralube veterinary ophthalmic ointment (Fera Pharmaceuticals, Locust Valley, NY, USA) was administered as needed if excessive squinting was observed on day 1. 

Blood and Ocular Tissue Sample Collection

For serum analysis following IV administration, 0.2 mL blood was serially collected via the central ear artery in triplicate at 5 and 30 minutes, and at 1, 2, 4, 8, 24, 32, 48, 72, and 96 hours post dose. For IVT-dosed groups, 0.2 mL or 1.0 mL blood was collected via the central ear artery in quadruplicate at 2 and 6 hours*, and at 1, 2*, 4, 7*, 14*, 21* and/or 28 days*. Sample volume was dependent on whether the collection procedure was terminal (1.0 mL/sample; terminal time points indicated by asterisks) or not (0.2 mL/sample). 

Ocular tissue was collected from IVT groups only, at terminal time points (indicated by asterisks). Following blood collection, animals scheduled for ocular tissue harvesting were administered 2 mL ketamine/xylazine (5:1 volume ratio of 100 mg/mL ketamine to 100 mg/kg xylazine) via intramuscular injection. At 5 to 10 minutes following anesthesia, animals were euthanized with 1 mL of a 1:1 solution of EUTHASOL (Vibrac AH, Inc., Fort Worth, TX, USA) and saline via the marginal ear vein. Aqueous humor, vitreous humor, and retinal tissue were harvested at 6 hours, and at 2, 7, 14, 21, and/or 28 days post dosing as described below. 

Briefly, the ocular globe was enucleated. All the aqueous humor (approximately 0.3 mL) was then withdrawn and placed into a 2-mL polypropylene vial on dry ice. Using a hemostat, the optic nerve was clamped and the whole ocular globe was slowly submerged into liquid nitrogen for 20 to 30 seconds. The globe was then placed on a bed of crushed dry ice and, using a scalpel blade, an incision was made through the sclera at approximately 2 to 3 mm from the limbus and the whole anterior chamber thus removed. Frozen vitreous humor was collected using a scalpel blade to make an incision through the sclera; the sclera and additional tissues were removed from the vitreous humor (approximately 1.4 mL) and placed in a 2-mL polypropylene vial on dry ice. All harvested ocular tissues were stored at −80°C until analysis. This procedure was repeated for each eye and for two or three animals per harvesting time point. This procedure has been used successfully in the past with no apparent impact on antibody and fragment stability.^18–21 

Determination of Antibodies and Antibody Fragment Concentrations

Enzyme-linked immunoabsorbent assays were developed at Genentech, Inc., to measure concentrations of anti-ANG2/anti-VEGF and anti-alpha5beta1/anti-VEGF bispecifics, as well as anti-gD antibodies and fragments in serum, aqueous, retina lysates, and vitreous samples (Table). Intact anti-ANG2/anti-VEGF and anti-alpha5beta1/anti-VEGF bispecific molecules were detected using the basic procedure described here. Enzyme-linked immunoabsorbent assay plates (384-well MaxiSorp; Thermo Scientific Nunc, Roskilde, Denmark) were coated with either 0.5 μg/mL soluble recombinant human angiopoietin-2 (Genentech, Inc.) or alpha5beta1 in 50 mM carbonate, pH 9.6 at 4°C overnight. After the plates were washed with wash buffer (0.05% polysorbate 20 in PBS, pH 7.4), they were blocked with block buffer (0.5% BSA, 15 parts per million proclin in PBS, pH 7.4) at room temperature for 1 hour with gentle shaking and washed. Standards (3.125–0.024 ng/mL in 2-fold serial dilution) and samples (in 3-fold serial dilution) in sample buffer (PBS containing 0.5% bovine serum albumin, 0.35 M NaCl, 0.25% CHAPS, 5 mM EDTA, 0.05% Tween-20, and 15 parts per million proclin) were incubated on the plates for 2 hours at room temperature. Bound antibody was detected with biotinylated VEGF at 0.1 μg/mL, followed by horseradish peroxidase-conjugated streptavidin (GE Healthcare, Piscataway, NJ, USA). Plates were developed using 3,3′,5,5′-tetramethyl benzidine (Moss, Pasadena, MD, USA) as the substrate. The reaction was stopped with 1 M phosphoric acid and absorbance was read at 450 nm on a Multiskan Ascent reader (Thermo Scientific, Hudson, NH, USA). The standard curve was fit using a four-parameter regression curve-fitting program (Genentech, Inc.). Procedures similar to those previously described were used to detect the other antibodies and fragments included in these studies (Supplementary Table S1). 

Table

 

Rabbit PK Parameters of the Anti-gD and Bispecific Variants After IVT Administration

 

Pharmacokinetic Analysis

A PK model was utilized to represent concentration–time profiles. The model includes first-order clearance from the vitreous or the central compartment for all fragments and first-order dissociation of F(ab')₂ into Fab fragments. The dosing is either in the vitreous (IVT) or in the central compartment (IV). Intravitreal dosing was performed in both eyes (bilateral injection), and the data from each eye (vitreous, aqueous, and retina) were analyzed independently and pooled for PK analysis The contribution from both eyes in each animal was considered for evaluating the serum exposures following IVT injections. Clearance from each eye is assumed to be identical and additive when modeling serum exposure. 

The dose administered in each study is nominally 500 μg per eye (1000 μg total per animal treated IVT) or 500 μg IV; given the variability in the dose solution preparation, the dose used for the model was corrected by the assay recovery obtained for the dosing. The model was developed in SimBiology (MathWorks, Inc., Natick, MA, USA). See Supplementary Table S2A for model equations. 

Results

Production of Anti-gD Forms

Five different forms of anti-gD, our nontargeting molecule raised against herpes simplex virus 1 gD, were created for the rabbit ocular and systemic PK study. The full-length antibody, used routinely as a control reagent, was produced in a mammalian expression system. The resultant antibody is glycosylated in the Fc region and fully capable of binding to Fc-gamma receptors and the FcRn. This form of the antibody was used directly in the study, and was also used as a source to create the other proteins. The Fab was made by digesting the antibody with lysine-C to yield an intact Fab with a single proteolysis site in the linker region between the first constant domain of the Fab and the hinge of the Fc. The Fc fragment was derived from the same method, but in a separate reaction. Both the Fab and Fc were purified to homogeneity after the reaction and analyzed by protein analytical techniques to ensure the quality of the protein. The F(ab')₂ containing a dDS bond, which is the native human IgG1 structure, was created by proteolytic digestion with pepsin. The F(ab')₂ from digestion was purified to homogeneity and characterized. The sDS-containing F(ab')₂ molecule was derived from a similar method. In this case, however, the expression construct for the intact (full-length) antibody was modified to substitute the second cysteine of the hinge with a serine residue. The protein product was full-length antibody with a single inter-heavy-chain disulfide that, after digestion with pepsin, yielded a F(ab')₂ with Fab domains covalently attached by a single, native disulfide bond (Supplementary Table S1). 

In vitro, all protein fragments showed optimal properties for an intraocular test article. After storage for several weeks at 4°C in formulation buffer with a pH of 5.5 to 6.0, no dissociation of the Fab domains in either the sDS- or the dDS-containing F(ab')₂ was observed. None of the molecules aggregated or showed insolubility, but clear differences were observed among the molecules in size and hydrodynamic volume as determined by size exclusion chromatography coupled to static light scatter (Supplementary Fig. S1). Because all of the molecules are stable under our storage conditions, and because they are derived from the exact same IgG and do not bind target protein, interpretation of the data from subsequent studies allowed for conclusions to be based solely on the structure and overall architecture of the molecule. 

PK Following IVT Administration

The concentration–time profiles for all nontargeted antibody molecules are presented in Figure 2. Following IVT administration, the clearance from the vitreous followed first-order kinetics for all molecules. The half-life of the molecules in the vitreous is shown in the Table. There are modest differences in the half-life across the molecules, with the smaller molecules of size 50 kDa (Fab and Fc) having a half-life of ∼3.2 days and the full-length antibodies (mAb and FcRn null mAb) with half-life of ∼5.2 days. The less stable F(ab')₂ sDS showed a half-life similar to the Fab, whereas the more stable F(ab')₂ dDS had a slightly higher half-life (4.4 days). The systemic exposures following IVT dosing were significantly influenced by the clearance kinetics in circulation (PK with IV in the next section), resulting in notable differences in the exposures for the different molecules. The presence of an intact FcRn-binding Fc increased the systemic exposures of the molecules, with 3-fold greater exposure for the mAb than the FcRn null mAb; similarly, the Fc fragment had higher systemic exposures compared to the Fab. The F(ab')₂ dDS had higher exposures than the Fab, but still lower than the Fc fragment. Finally, the less stable F(ab')₂ sDS had systemic exposures similar to the Fab, suggesting a rapid cleavage of the DS bond. The area under the curve (AUC) of the systemic exposure is shown in Figure 5A. The terminal half-life in circulation followed the vitreal half-life for all molecules except the Ab, where the half-life was higher given the slow systemic clearance for the full-length Ab. The exposures in the aqueous and the retina tissue (when available) followed kinetic profiles similar to those observed in the vitreous. 

Figure 2

 

Pharmacokinetics of nonbinding control antibody and antibody fragments in rabbits. (A) Ocular PK following intravitreal administration. (B) Serum PK following either intravitreal (left) or intravenous (right) administration. Solid lines are model simulations and open circles are observed values. Data measured as lower limit of quantitation not included in figure and analysis.

PK Following IV Administration

The concentration–time profiles following intravenous administration are shown in Figure 2. As expected, the Fab formats were rapidly cleared from the circulation. In contrast, the serum concentration of the Fc-containing molecules, mAb, FcRn null mAb, and Fc fragment decreased at a significantly slower rate (Fig. 2B; Supplementary Table S3). The model equations and the PK parameters for all the molecules are shown in Supplementary Table S2. 

Synthesis of Bispecific F(ab')₂ Molecules Containing Single or Double-Disulfide Bonds

Two methods were developed for the production of bispecific F(ab')₂ molecules that maintain natural antibody architecture. The first method relies on bispecific antibody assembly as the first step in synthesis, followed by proteolytic digestion of the antibody to produce a F(ab')₂ (Fig. 3A). We used two enzymes, pepsin and immunoglobulin G-degrading enzyme of Streptococcus pyogenes (IdeS), to obtain intact F(ab')₂ with two disulfide bonds. Both enzymes efficiently produced F(ab')₂ without impacting the integrity of the Fab domains. We found that pepsin digestion resulted in a product that was not homogeneous and, after longer periods of digestion, began to produce additional proteolytic products. The heterogeneity shown by mass spectrometry analysis in Figure 3B reflects alternative sites for cleavage by pepsin but results in only minor biochemical changes to the protein, so these variants are difficult to separate. In contrast, reactions with IdeS-produced molecules were completely homogeneous. Even after 48 hours of digestion with IdeS at 37°C, only a single heavy-chain clip site was observed. In addition, IdeS derived from recombinant sources offers advantages over porcine-derived pepsin. The availability of recombinant protease, the ability to control the reaction, and the resulting product homogeneity may be preferable for large-scale production and manufacturing. Synthesis of F(ab')₂ containing a sDS was accomplished using a method distinct from that used for anti-gD sDS (Fig. 3C). In part, we were interested in developing a nonenzymatic process. Furthermore, production levels of Fabs are typically higher than for half-antibodies, and this production scheme increases the specific product yield because 30% to 40% of the product will not be inherently lost after digestion. Three important steps in the reaction are as follows: (1) reduction and oxidation of Fab' to produce a reactive Fab-SH with a free thiol; (2) reaction of one Fab-SH with a disulfide exchange reagent, DPDS, and isolation of the Fab-S-S-pyridyl product; and (3) reaction of Fab-S-S-pyridyl with the other Fab-SH and isolation of the Fab-S-S-Fab, or F(ab')₂ sDS product. 

Figure 3

 

Synthesis of bispecific F(ab')₂ with either one or two disulfide bonds linking the two Fab'. (A) Production of F(ab')₂ with two intra-Fab disulfide bonds was done by half-antibody assembly followed by site-specific proteolytic digestion using IdeS or pepsin. (B) Mass spectometry analysis of heavy-chain species following digestion by IdeS and pepsin. (C) Production of F(ab')₂ with one intra-Fab disulfide bond was done using free thiol chemistry and a two-step process. (D) The product of the first and second steps of the synthesis was analyzed by gel filtration (top) and mass spectometry indicating the desired bispecific product (lower).

The production of Fab' in E. coli generally results in the addition of adducts, such as glutathione or other thiol-containing small molecules, to the unpaired cysteine or dimerization between two Fabs. These unreactive molecules are therefore treated with reducing agent followed by oxidation through dialysis in buffer at pH 8. We found that certain extraction conditions may limit disulfide formation, but did not employ these procedures for this study. Reaction with DPDS resulted in >98% of the desired product formation with no detectable homodimer product by mass spectrometry and size exclusion chromatography. The second reaction yielded approximately 85% product as the dimer. Primarily unreacted Fab' accounted for the remaining 15% of product due to a slight stoichiometry difference in the reaction. Overall, the yield and purity were acceptable, with only the heterodimer species observed by mass spectrometry analysis (Fig. 3D). 

PK Following IVT and IV Administration of Bispecific Molecules

Two different bispecific molecules were evaluated for IVT and IV PK. The data from anti-ANG2/anti-VEGF molecules were used for model calibration (in addition to the gD data used previously), whereas the data for anti-alpha5beta1/anti-VEGF were used entirely for testing the predictions of the model. Given that the full-length mAb displayed increased systemic exposures that potentially could lead to safety concerns for the two therapeutic candidates, only the Fab and the F(ab')₂ formats were considered. 

The exposure in the ocular tissues and in circulation following IVT and IV dosing with the anti-ANG2/anti-VEGF bispecific is shown in Figure 4A. The vitreal half-life followed the same trend as observed with the control molecules with half-lives of 3.3 to 3.9 days for the molecules evaluated. The disassociation of the sDS occurs faster than that of the dDS, as demonstrated by the data and the corresponding model calibration from the F(ab')₂ intact bispecific (Supplementary Fig. S2). The systemic exposures also followed similar trends, with the more stable F(ab')₂ dDS having slightly increased exposure compared to the F(ab')₂ sDS and the Fab. 

Figure 4

 

Pharmacokinetics of binding antibody and antibody fragments in rabbits. (A) Ocular PK following intravitreal administration of anti-ANG2/anti-VEGF single- and double- disulfide bond, and anti-ANG2 Fab. (B) Serum PK following either intravitreal (left) or intravenous (right) administration of the same compounds as in (A). (C) Ocular and serum PK following intravitreous administration of anti-alpha5beta1/anti-VEGF and individual Fab' (each Fab dosed at 250 μg). Solid lines are model simulations and open circles are observed values.

Figure 5

 

Comparison of vitreous and serum exposures for antibody and antibody fragments. (A) Data for AUC obtained from single intravitreal administration also described in the Table. (B, C) Simulated systemic PK following repeated intravitreal dosing, once-a-month or every two months, respectively. Color reference corresponding to cartoons in (A).

For the test dataset of anti-alpha5beta1/anti-VEGF bispecific, the model parameterization of the anti-ANG2/anti-VEGF was utilized for model prediction and data comparison. The predictions for the exposures in both the ocular tissues and the systemic circulation were consistent with the data. This suggests that the trends observed between the different molecule formats are general and reproducible. 

Simulated Anti-gD Serum PK Following Repeated IVT Dosing

In clinical trials, antibody molecules intended for IVT use have been administered on a fixed monthly, bimonthly, quarterly, or as needed basis. A simulation of anti-gD serum PK following monthly or bimonthly IVT dosing is shown in Figure 5. For both dosing regimens, the antibody platform is associated with much higher systemic exposure compared to the Fab, F(ab')₂ sDS, and F(ab')₂ dDS platforms. 

Discussion

Anatomic and dynamic physiologic barriers restrict access to intraocular tissues, including the retinal tissue targeted in indications such as diabetic retinopathy. The most effective delivery of biotherapeutics in the intraocular compartment is IVT microdosing (range, 50–100 μL). Following distribution in the vitreous humor and the retina, biotherapeutics egress into the systemic circulation via the choroidal vasculature and aqueous humor outflow (Fig. 1B). The exposures in the aqueous and the retina tissues follow trends similar to those for the vitreous. Also, the ratio of retina-to-vitreous exposure is uniform across all the molecules evaluated, suggesting that the vitreous PK is a sufficient indicator of exposure at the site of therapeutic action (Fig. 2). 

The efficacy and safety profile of therapeutic antibodies and Fabs for intraocular use may be improved by engineering more potent or multispecific inhibitors to limit the number of molecules/injections required to achieve the desired therapeutic effect. Molecule design may extend half-life in the vitreous and/or combine different mechanisms of action to reduce the frequency of IVT injections. Minimizing unwanted systemic exposure may be another goal of molecular design and selection. 

In this study we examined the PK of several different antibody drug platforms. The data indicate that the elimination of the Fc region is associated with a significant reduction of systemic exposures following IVT dosing. The F(ab')₂ formats allow this reduction in systemic exposure while maintaining exposures in the ocular tissues comparable to that observed with the full-length antibody. Molecules containing an Fc have greater systemic exposures because they are protected from proteolytic catabolism, by binding to FcRn expressed by endothelial cells. 

Despite the 3-fold differences in molecular weight, Fab and IgG have only slight differences in ocular pharmacokinetics after IVT administration. To explain this phenomenon, previous reports have suggested that endothelial expression of FcRn plays a significant role to increase the rate of elimination of IgG from the ocular compartment.¹⁵ There are other routes for elimination by the aqueous humor outflow pathways and bulk flow through the trabecular meshwork and Schlemm's canal that could play a more significant role.²² Our data suggest that these routes are in fact dominant over FcRn recycling in determining the rate of clearance from the eye. In our study, IgG and IgG-FcRn null show similar kinetics in the rabbit eye after IVT injection. These molecules are identical in molecular weight and differ only in their ability to bind FcRn. We also demonstrate that the Fc alone, which is nearly identical in size to a Fab yet binds to FcRn, shows Fab-like ocular kinetics. Thus, the presence of the Fc region and subsequent binding to FcRn has no measurable impact on the overall ocular pharmacokinetics. Previous reports that have examined the role of FcRn in ocular pharmacokinetics using FcRn-knockout mice did not measure vitreous PK parameters of IgG in wild-type versus knockout mice, which could have revealed this phenomemon.¹⁵ We speculate that the reason FcRn may not play a major role in the eye when compared to the central compartment could be due to the expression levels of FcRn in the eye relative to the expression in the systemic epithelium. 

The difference in molecular weight of 50 to 150 kDa had a small impact on the vitreous half-life, which ranged from 3.2 to 5.2 days, comparable to previous reports.^12,18,20,21 Systemic elimination that utilizes a glomerular filtration cutoff of around 70 kDa MWt shows greater than 10-fold difference in clearance rate from 20 to 150 kDa.²³ In our study we were able to isolate the impact of molecular weight by removing FcRn-binding as a variable. Elimination of proteins from the eye does not appear to have a similar dependency on molecular weight. These results complement the basic understanding from previous reports comparing pharmacokinetics of different-size proteins, but not independent of FcRn.^12,18,20,21 

Overall the PK profile of F(ab')₂ molecules was characterized by longer half-life in the vitreous and rapid elimination from the systemic circulation. The vitreous half-life of F(ab')₂ dDS format was higher than the vitreous half-life of F(ab')₂ sDS and Fab molecules, indicating that the presence of a second disulfide bond has a stabilizing effect. The format also allows for the development of bispecific molecules and may be the most versatile antibody platform for the development of future agents for intraocular use.^7,24 However, this format represents a new challenge for antibody development due to the potential impact of preexisting antibodies on the PK and safety of F(ab')₂ therapeutics.^25,26 This topic remains to be further investigated. 

The consistency of the model predictions with the data for both bispecific molecules evaluated showed that the PK characteristics were conserved for the different molecules, and the learning from these could potentially guide PK characterization for future molecules. The model developed in this work supports predictions for multiple dosing regimens. Simulations suggested that increased exposures for the full-length antibody compared to the Fab and F(ab')₂ formats are driven by the Fc and FcRn binding in the systemic compartment. 

In summary, we characterized the ocular and systemic PK in the rabbit for various antibody-derived protein fragments and developed a comprehensive mathematical model, which together can be used to guide ocular protein therapeutic design and development. The mathematical model can be generalized for all protein fragments derived from an antibody regardless of the presence of an Fc region, and herein has been applied to the development of a bispecific F(ab')₂ fragment with activity targeting multiple angiogenesis pathways. It is exciting to anticipate the further development of these molecules and the potential for impact on the unmet medical need. 

Acknowledgments

The authors thank Enzo Palma, Saileta Prabhu, Lu Xu, Dan Yansura, Richard Vandlen, Sheila Ulufatu, Margaret Kenrick, Nicole Vega, Chris Schuetz, Ivo Stoilov, Saroja Ramanujan, Betsy Kitchens, and Paul Fielder for their contributions to this work. 

All authors are employees and shareholders of Genentech, Inc. 

Disclosure: K. Gadkar, Genentech, Inc. (E); C.V. Pastuskovas, Genentech, Inc. (E); J.E. Le Couter, Genentech, Inc. (E); J.M. Elliott, Genentech, Inc. (E); J. Zhang, Genentech, Inc. (E); C.V. Lee, Genentech, Inc. (E); S. Sanowar, Genentech, Inc. (E); G. Fuh, Genentech, Inc. (E); H.S. Kim, Genentech, Inc. (E); T.N. Lombana, Genentech, Inc. (E); C. Spiess, Genentech, Inc. (E); M. Nakamura, Genentech, Inc. (E); P. Hass, Genentech, Inc. (E); W. Shatz, Genentech, Inc. (E); Y.G. Meng, Genentech, Inc. (E); J.M. Scheer, Genentech, Inc. (E)