The PK-Eye: A Novel In Vitro Ocular Flow Model for Use in Preclinical Drug Development

Materials

Kenalog® (40 mg/mL) was purchased from Squibb. Bevacizumab (Avastin®, 25 mg/mL; Genentech, South San Francisco, California) and ranibizumab (Lucentis®, 10 mg/mL; Genentech, South San Francisco, California) were obtained from the pooled remaining contents of vials that had been used clinically. Sodium hyaluronate (solubility: 5.0 mg/mL; 1.8 MDa) was purchased from Aston Chemicals (Aylesbury, UK) and agar (solubility: 15.0 mg/mL) was obtained from Fluka Analytical (Gillingham, Dorset, UK). Dye (Coomassie Brilliant Blue, MW = 825.97 g/mol) was used to validate flow in the model. Visking dialysis membrane tubing (MWCO of 12–14 kDa) was purchased from Medicell International Ltd. (London, UK). Bovine serum albumin (BSA) was obtained from Sigma–Aldrich. A 16-channel Ismatec peristaltic pump (Michael Smith Engineers Ltd., Woking, Surrey, UK) was used to generate fluid flow into the PK-Eye. UV measurements were analyzed using Hitachi U-2800A spectrometer and a Wallac Victor2 1420 plate reader. HPLC was conducted using a Agilent 1200 series (Agilent, Wokingham, Berkshire, UK) equipped with Chemstation software (Agilent, Wokingham, Berkshire, UK) using a reverse-phase Synergi Polar-RP Phenomenex 4 μm, 15 cm column.

Methods

Design of the PK-Eye Flow Model

Efforts to develop the model were focused at the outset (1) to have the anterior and posterior cavities to be a similar size as in the human eye and (2) to use an aqueous inflow that mimics the aqueous outflow in humans. The model comprises two plastic pieces that are secured together with four screws (Fig. 1). The overall dimensions and volumes of the posterior (4.2 mL) and anterior (0.2 mL) cavities of the model are comparable to that of the human eye (Table 1). Several prototypes of the model were evaluated (Table 2) to best mimic the mass transfer caused by aqueous outflow. The prototyping efforts listed in Table 2 were focused to design a model that was both simple to fabricate and to use. It was found that a single inlet port (1.5 mm ID) with a flow of 2.0 μL/min15 was the most practical way to mimic aqueous flow in this model. The aqueous inlet port was placed in the vitreous cavity near the membrane barrier. An injection port was placed at the top of the vitreous cavity. A widely available dialysis membrane (MWCO 12–14 kDa) was fixed between the posterior and anterior cavities (Fig. 1). This membrane was used to provide a barrier to the anterior cavity for solubilized molecules residing in the posterior cavity. For outflow, a single outlet port in the anterior cavity extended upwards from the model to provide a small amount of backpressure. This was used to avoid the formation of air bubbles and to maintain the total liquid volume within the model.

Table 1. Average Human Eye Dimensions Versus the Dimensions of the PK-Eye Model Used in This Study

		Average	PK-Eye
Dimensions	Anatomy	Human Eye56	Model
Length (mm)	Cornea vertical length	11.0	9.0
	Cornea horizontal length	12.0	9.0
Volume (mL)	Posterior cavity	3.9–5.0	4.2
	Anterior cavity	0.2	0.2

Table 2. Summary of PK-Eye Prototype Development

	Ports
Prototype	In	Out	Vitreous (mL)	Design Information	Observations During Validation
A	1	1	5.6	A spherical globe made of polyacrylate plastic	The volume was too large compared with the human eye
B	2	1	4.2	Volume was adjusted to match the average value reported in human eyes	Cavity lacked spherical shape
C	2	4	4.2	Incorporation of anterior (0.2 mL) and posterior cavities (4.2 mL)	Could not maintain a gel to water interface between posterior and anterior cavities
D	2	10	4.2	Fit cavity separating membrane and increase outlet ports	Unequal and insufficient flow in each port observed
E	2	4	4.2	Reduce the number of outlet ports	Input port diameters too large and too far away from the cavity separating membrane
F	4	4	4.2	Move input ports to replicate ciliary body location and flow	Inadequate and inconsistent flow from each outlet port
G	4	1	4.2	Incorporation of new outlet port in the anterior cavity	Too difficult to maintain an accurate flow rate using four inlet ports
H	1	1	4.2	One inlet to replicate ciliary body flow	Much more practical to use than with two inlet ports

• The prototype that was used for these studies reported herein is H.

Dye Experiments to Illustrate Flow Properties of the Model

Dye studies were conducted to illustrate the need for flow and compartmentalization of the model. With the aqueous inlet flow set at 2.0 μL/min, Coomassie blue dye was injected (100 μL, 1.2 mM) into the anterior cavity (Fig. 2). The amount of dye used was in large excess to what can be administered in an in vivo experiment by either injection or the use of eye drops. The dye cleared through the outlet port with only a trace amount being observed to migrate into the posterior cavity at about 24 h after injection (Fig. 2). In a second experiment, the dye was injected into the posterior cavity. With aqueous inflow (2.0 μL/min), the dye cleared from the posterior cavity into the anterior cavity and then through the outlet port of the model (Fig. 3, upper panel). When there was no flow (Fig. 3, lower panel), the dye remained concentrated only in the posterior cavity.

These dye experiments broadly mimic mass transfer in the eye that is caused by aqueous flow. When a drug is in the anterior chamber, very little of it will pass into the posterior cavity because of the aqueous outflow. And if a drug is in the posterior cavity, flow is necessary for the drug to clear anteriorly via aqueous outflow. It was conspicuous that the dye did not migrate between the two compartments unless flow was induced (Figs. 2 and 3), so the visking membrane did offer barrier properties to the dye. Coomassie blue has a relatively large molecular weight (∼825 g/mol) and is partially soluble in water because of its hydrophobic nature. The visking membrane is made from reconstituted cellulose. Water is thought to swell the membrane creating a hydrophilic barrier layer that can act to inhibit the diffusion rate of hydrophobic molecules through the membrane.58, 59

In addition to in vitro permeation models,26, 29 other in vitro models have been described for the preclinical development of IVT ophthalmic preparations.45, 60-63 For example, the dissolution of poorly soluble drugs and extended-release tablets62 were evaluated using a USP4 dissolution apparatus with a flow through of 1.5 mL/min in a chamber of 8–19 mL.64 Aspects of drug distribution in the vitreous have been evaluated using a model for saccades (i.e., eye movements), which can influence drug distribution in the vitreous.65 This single compartment, nonflow model was used to evaluate dye distribution in glycerol and polyacrylamide gels as vitreous substitutes.66 These in vitro models do not mimic the human eye in terms of its compartmentalization, scale, or mass transfer properties because of aqueous flow.

Computational modeling of the eye has also been used to evaluate drug distribution within the vitreous of animals and humans18, 67-74 after injection or release by implants,75, 76 and to develop vitreous diffusion IVIVCs.77, 78 The need to understand how a drug is distributed within the vitreous to access the retina has been a key driver for many of the studies focused on drug distribution within the posterior segment. However, knowledge of vitreous distribution can only be complete by first determining the residence time of a drug within the vitreous. Any preclinical candidate designed for prolonged action in the posterior cavity must be able to achieve extended PK properties in the presence of intraocular aqueous flow.

Preincubation of the Model with Albumin

There was a concern that a protein undergoing evaluation would adversely adsorb onto the internal plastic surfaces of the model. The model was preincubated with a 0.1% albumin solution, and then with the inflow set at 2.0 μL/min, bevacizumab was injected (1.25 mg in 50 μL) into the posterior cavity of the model.

This experiment was conducted using PBS in the model to allow rapid and complete distribution of bevacizumab throughout the posterior cavity. The use of PBS in the posterior cavity also acted to mimic the reduced viscosity of the vitreous after a vitrectomy. The viscoelastic properties of the vitreous are incompletely understood15, 34, 79 and there are many differences in the composition and microstructure of the vitreous.24, 30, 36-38 As a large molecule, the diffusion of a protein will be faster80 and will clear more rapidly from a vitrectomized eye.28, 81 It was also thought that faster and probably more complete antibody diffusion would occur in a low viscous solution that would make it easier to observe its propensity to adsorb onto the plastic surface.

The observed residence time of bevacizumab from the model after preincubation with albumin followed first-order kinetics and was 1.5 ± 0.3 days (Fig. 4a). When the model was not coated with albumin, the bevacizumab clearance half-life was 1.2 ± 0.1 days (Fig. 4b). Although this was not significantly different with the clearance half-life when the model was preincubated with albumin, there appeared a less discernable peak concentration, which suggests the bevacizumab reversibly adsorbed onto the plastic surface of the uncoated model. The presence of a peak concentration is characteristic for an intraocular injection73 and was more evident when bevacizumab clearance was measured in the albumin-coated model (Fig. 4a). If the model is fabricated from plastic and PBS is used in the posterior cavity, consideration should be made to preincubate with albumin or another blocking protein prior to evaluating a candidate therapeutic protein.

Bevacizumab is a monoclonal antibody that is widely used as an unlicensed, but more cost-effective anti-VEGF medicine3, 4 that is considered to be clinically equivalent to ranibizumab for the treatment of some indications.82, 83 It was encouraging that the half-life of 1.5 ± 0.3 days from the albumin-coated model (Fig. 4a) is comparable to the half-life of bevacizumab at the same dose in vitrectomized monkey eyes (1.5 ± 0.6 days; n = 6).84 There is a report that the half-life of bevacizumab at a 1.25-mg dose from a victrectomized human eye of a single patient was 0.7 days.85 In nonvitrectomized human eyes at this dose, the half-life of bevacizumab has mean values ranging from 4.9 to 10 days.81, 85-88 In contrast, the half-life of bevacizumab at this dose is 2.8 ± 0.6 days (n = 3)89 in normal monkey eyes that have an intact vitreous. Although less human data exists for the clearance of bevacizumab in vitrectomized eyes, clearance is expected to be faster than in nonvitrectomized eyes. There is also a significant variation in the half-lives for nonvictrectomized human patients, so the monkey and human data are also significantly different. On the basis of the more complete monkey data, the model could be used to estimate the vitrectomized clearance times by using PBS in the vitreous cavity. It is thought this would provide the fastest clearance time possible for a given preclinical formulation.

Clearance Profile of Ranibizumab Using Simulated Vitreous Fluid

Protein therapeutics will diffuse more slowly in vitreous fluid, which is generally more viscous than PBS. A simulated vitreous liquid was then used in the posterior cavity of the model. Reduced diffusion will prolong the time it takes for the protein to clear from the vitreous cavity. Simulated biological fluids are widely used to mimic aspects of an endogenous fluid during preclinical research.32, 90 Vitreous substitutes are also widely used, both clinically and for preclinical research.91, 92 We used a simulated vitreous fluid prepared from HA and agar56 that had a dynamic viscosity of approximately 0.6 Pa·s, which is comparable to human vitreous (∼0.5 Pa·s).93 Initial experiments established that none of the components of the simulated vitreous diffused into the anterior cavity of the model and that the simulated vitreous maintained its viscosity over time (∼2–3 weeks). No substrates were observed in the outflow unless ranibizumab was first injected.

Ranibizumab was then injected into the vitreous cavity of the model using the usual clinical dose (0.5 mg, 50 μL). The ranibizumab clearance half-life in the model followed first-order kinetics and was 8.1 ± 3.1 days (Fig. 5a), which is similar to the half-life in humans at this dose (∼7.294 and 9.095 days). Using simulated vitreous provides the means to estimate the time needed for a therapeutic protein to clear via the aqueous outflow pathway from a human eye. When PBS was used in the posterior cavity, the clearance half-life of ranibizumab was 1.5 ± 0.6 days (Fig. 5b). This was comparable to bevacizumab (Fig. 4a). When the model was pre-incubated with albumin and then rinsed, no protein was observed from the outflow. Protein was only observed when the sample, i.e. ranibizumab, was injected into the posterior cavity of the model.

The visking membrane had a molecular weight cut off value of 12–14 kDa. This refers to the smallest number average molecular weight of a globular macromolecule that will be about 90% retained when dialyzed overnight,96 although there are no strict specification standards.55 It was anticipated that with experiments lasting several days to weeks, that passage through the membrane of small amounts of molecules with larger molecular weight would be possible. This is what was observed for both the ranibizumab and bevacizumab experiments (Figs. 4 and 5). It was also necessary for the membrane to ensure the simulated vitreous solution remained in the posterior cavity without allowing the gelled material to pass into the anterior cavity. The simulated vitreous should be made using very high-molecular-weight gelling components that form a stable viscous solution.

As a Fab, ranibizumab (∼48 kDa) is about 1/3 the molecular weight of bevacizumab (∼150 kDa), which is a full monoclonal antibody. In humans, these molecules display comparable mean half-lives with a similar therapeutic effect.21, 86, 94 The half-lives for ranbizumab span from about 7.2 to 9.0 days and for bevacizumab, the range is about 6.7–10.0 days.81, 85-88 As mentioned, these half-lives can be reduced because of medical conditions such as vitrectomy and retinal detachment.97 The usual clinical dose for ranibizumab (0.5 mg) is about 20% more than bevacizumab (1.25 mg) on a molar basis. Of the other biologics registered for ophthalmic use by IVT injection,1 there is no reported human half-life data for the Fc fragment fusion (alfibercept, ∼115 kDa),1, 81 but the PEGylated-aptamer (Pegaptanib, ∼50 kDa) displays a mean half-life of about 7–8 days for a IVT dose of 0.3 mg as determined by plasma measurements.98 Although the half-life of Pegaptanib is essentially the same as ranibizumab, its dose is about 60% of that is used for ranibizumab on a molar basis. Increasing the dose of Pegaptanib by 10 times of the US FDA-approved dose to 3.0 mg only increases the half-life to about 10 (±4) days.99, 100

These observations suggest that there is probably a “molecular weight threshold” after which most proteins and macromolecular drugs will display comparable half-lives in the posterior cavity due to clearance by the aqueous outflow pathway. New strategies are being developed to increase the vitreous resident times of macromolecular drugs.7, 10, 19, 34, 101-103 Some of these strategies may employ the use of implants (for example) where a release or dissolution step is required prior to clearance of the protein therapeutic. It is for these new strategies that we believe our model will find real utility in preclinical research. The strategy to increase the dose of a molecule as a means to increase its “therapeutic tail,” which is used with aflibercept, can also be evaluated in a realistic 2-compartment outflow model such as the PK-Eye. As more stable protein therapeutics are developed, much higher concentrated doses may be possible where exploitation of the therapeutic tail is difficult to evaluate in an animal model simply because of the generation of ADAs.

The need during preclinical research to obtain reliable pharmacokinetic information using animal models alone for protein therapeutics cannot be adequately met. The use of our model will avoid the problems associated with ADAs.39-43 The model also avoids the need to consider the anatomical differences28, 33-35 that exist between animal and human eyes. For example, globe sizes of the rabbit (New Zealand white) and monkey (cynomolgus) are similar, but the relative sizes of the anterior chamber and lens of the rabbit are much larger than the monkey.33 The rabbit vitreous is about 50% the size of the monkey vitreous (and about 30%–40% that of the human vitreous). But the aqueous flow of the rabbit (∼2.0 L/min104) and monkey (∼2.8 L/min105) are similar to that of the human. Although clearance of ranibizumab will be primarily driven by its diffusion through the vitreous,28 its clearance should be faster in the rabbit (less vitreous, larger anterior cavity, shorter diffusion path). However, the half-life of ranibizumab (0.5 mg) from the rabbit and monkey are about the same, being about 2.8–2.9 days22, 106, 107 and about 3 days,35 respectively. Even though the monkey eye is more similar to the human eye than is the rabbit eye,33 the half-life for IVT-injected ranibizumab (0.5 mg) from the monkey is much less than that observed in humans (∼7.294 and 9.095 days).

Although three of the four clinically used macromolecular therapeutics broadly display similar mean half-lives in humans, there is still range of clearance times that is indicative of how intraocular pharmacokinetics can vary in human eyes that are being treated. This variation may be because of how the IVT injection procedure is conducted,28, 108, 109 but more often there are complicating medical conditions86, 110, 111 in patients that require IVT injections that could facilitate greater RCS uptake of nonpermeable molecules. There are also variations in vitreous composition37 that can influence flow in the posterior chamber of human eyes68, 112 where aqueous outflow through the retina may increase.30 Variations in the vitreous occur because of (1) age (i.e., liquefaction),24, 28, 38 (2) whether a vitrectomy has been performed, and (3) due to the nature of the medical condition afflicting the eye. For example, small breaks in the inner-limiting membrane (ILM)113, 114 can cause retinal uptake of non-RCS permeable molecules such as therapeutic proteins.21, 87, 110, 111, 115 The intact ILM that is between the vitreous and retina can act as a barrier for macromolecules to the retina.7, 34 The pharmacokinetics of antibodies can further vary because of the presence of Fc receptors in the retinal tissue.115-117 Although some posterior aqueous outflow occurs through the retina, the relative amount may be higher in animal models than in humans17 and radiolabelled ranibizumab can be detected within the retina of the rabbit.118 However, quantifying the total dose that may clear through the retina119 or extrapolating data for retinal uptake from an animal model after IVT administration is not yet possible. The use of the model with PBS in the vitreous cavity can provide data for the fastest vitreous clearance time to be expected for a preclinical candidate. If considered along with the clearance times obtained with simulated vitreous, it will be possible to obtain minimum and maximum estimates for posterior cavity clearance times very early in preclinical research.

It is not possible with in vivo animal studies to mimic the wide range of conditions encountered in human eyes that may cause pharmacokinetic variation. For example, ranibizumab (and bevacizumab) clearance from vitrectomized rabbit eyes is about the same107 or at best slightly reduced compared with nonvitrectomized rabbit eyes.106 This may be because of the large rabbit lens making it difficult to remove a high enough proportion of the vitreous.107, 109 As longer-lasting preparations are developed, differences in human ocular pharmacokinetics will persist to obscure clinical outcomes. It is important in preclinical research to obtain robust PK data based on the aqueous outflow pathway because it appears to change less than RCS pathways for many of the intraocular diseases that require treatment by IVT injections. It is possible with our model to obtain good estimates of the upper limit for clearance early in preclinical studies. The many medical conditions that contribute to accelerate clearance times can then be accounted for to estimate the different clearance times that might occur clinically.120 Furthermore, as longer-lasting therapeutic protein preparations are developed, the use of our outflow model will allow for easy sampling to evaluate protein stability that will be important for preparations designed to last 2–3 months inside the vitreous cavity.

Clearance Profile of TA Suspension (Kenalog®)

The clinical use of steroid implants (e.g., Ozurdex® and Iluvien®) that release drug over several months into the vitreous are examples of IVT implants19 that may become increasingly more common in the clinic.19, 24, 34, 102, 103, 121-123 Many current implants utilize poorly soluble steroids124 so the nonsink conditions of the posterior cavity can be exploited to prolong the vitreous resident time. Steroids tend to be lipophilic and permeable, and so a significant amount of material solubilized in the vitreous would be expected to clear by both RCS and anterior outflow pathways. Once a permeable, low-molecular-weight molecule is solubilized in the vitreous, clearance would be expected to be relatively quick.18, 24, 27, 28 For example, the half-life of intravitreally injected dexamethasone phosphate, which is water soluble, is 3.48 h.27

Prior to the development of implants, the IVT injection of a suspension of a poorly soluble drug such as TA (Kenalog®)125, 126 was the best way to clinically prolong the presence of a drug in the vitreous cavity. TA solubility is in the range of approximately 21–30 μg/mL depending on solution media and temperature (25°C–37°C).127, 128 Kenalog® is the most commonly used TA preparation and has long been used for ophthalmic conditions125, 126, 129; however, there are other, preservative-free, TA formulations available that include Triesence® (Alcon) and Trivaris® (Allergan).129

Triamcinolone acetonide (Kenalog®, 4.0 mg in 100 μL) was injected into the posterior cavity of the model, which was heated to 37°C. The dissolution rate of the TA suspension is solubility controlled in the total liquid volume of the model and a near constant-release concentration (∼25–30 μg/mL) in both PBS and simulated vitreous was observed (Fig. 6). The clearance half-life of TA from the model was in the range of 26–28 days.

The mean half-life in humans for the IVT administration of TA at the same dose spans 15.4130–18.6 days.131 This clearance time is faster than what was observed in the model because of contributing clearance through RCS pathways. RCS permeable molecules once in solution will clear through both RCS and anterior outflow pathways. Although active transport by the RCS can be significant for some molecules,72 our model will provide an estimate of the upper time period for clearance. The model can be used to estimate drug release from implants or the dissolution of suspensions as a first step in their clearance. This will often be the rate-limiting step that defines the overall clearance time of a candidate preparation. Once the active is in solution, the model has the capacity to provide a good estimate of clearance by the anterior outflow route. As for other preparations designed for different modes of administration (e.g., oral, pulmonary), this information can be used with an in vitro26 or computational18, 29-31 permeability model to develop IVIVCs.49, 52, 53 There is intense effort to develop implants and other long-acting preparations from proteins, peptides and nucleotides,7, 10, 19, 34, 101, 103 and possibly other low-molecular-weight substances (e.g., the antiviral agent, Foscarnet18) that do not clear to a great extent by the RCS. In these cases, the preclinical use of our model would be expected to provide good estimates of useful clearance times.