Structural, Kinetic, and Thermodynamic Analysis of ... - ACS Publications

Structural, Kinetic, and Thermodynamic Analysis of the Binding of the 40 kDa PEG−Interferon-α2a and Its Individual Positional Isomers to the Extrac...
0 downloads 0 Views 290KB Size
Bioconjugate Chem. 2005, 16, 518−527

518

Structural, Kinetic, and Thermodynamic Analysis of the Binding of the 40 kDa PEG-Interferon-r2a and Its Individual Positional Isomers to the Extracellular Domain of the Receptor IFNAR2 Christophe Dhalluin,‡ Alfred Ross,‡ Walter Huber, Paul Gerber, Doris Brugger, Bernard Gsell, and Hans Senn* F. Hoffmann-La Roche AG Department of Pharma Research CH-4070 Basel, Switzerland. Received September 15, 2004; Revised Manuscript Received March 8, 2005

Type-I Interferons exert antiviral and antiproliferative activities through the binding to a common cell surface receptor comprising two subunits, IFNAR1 and IFNAR2. Human recombinant InterferonR2a (IFNR2a) is a potent drug (Roferon-A) used to treat various cancers and viral diseases including Hepatitis B/C infections. To significantly improve the pharmacological properties of the drug, a pegylated form of IFNR2a was developed (PEGASYS). This 40 kDa PEG-conjugated IFNR2a (40PEGIFNR2a) is obtained by the covalent binding of one 40 kDa branched PEG-polymer to a lysine sidechain of IFNR2a. Here, we report the detailed structural, kinetic, and thermodynamic analysis of the binding to the extracellular domain of the receptor IFNAR2 of 40PEG-IFNR2a and its isolated positional isomers modified at K31, K134, K131, K121, K164, and K70, respectively, in comparison with unmodified IFNR2a. Our binding studies, using the surface plasmon resonance technique, show that the pegylation does not abolish the binding to the receptor, but significantly reduces the affinity mainly due to a change of the association rate. The results are supported by modeling and simulation of the binding, using Self-Avoiding-Walk calculations for the polymer conformations. A correlation between the structural parameters and the kinetic and thermodynamic parameters of the binding of the positional isomers could be established. For the Isomer-K31 and -K164, the PEG-polymer attachment point is located in proximity to the binding interface, and the isomers display affinity in the range 150-520 nM in an enthalpy-driven binding process. In contrast for the Isomer-K134, -K131, -K121, and -K70, the PEG-polymer is attached remotely from the binding interface, and the isomers exhibit a higher affinity (32-76 nM) in an entropy-driven binding process. This study constitutes an essential collection of knowledge on which the interaction of 40PEG-IFNR2a and its positional isomers with its cellular receptors can be better understood.

INTRODUCTION

Type-I Interferons (IFNs)1 are a family of homologous cytokines that potently elicit an antiviral and antiproliferative state in cells (1). Human type-I IFN R/β bind to a common cell surface receptor consisting of two transmembrane subunits, IFNAR1 and IFNAR2 (2, 3), to form a ternary complex (Figure 1). IFNAR2 is the major ligand binding component of the receptor complex, exhibiting nanomolar affinity to both IFNR and IFNβ (4). While IFNAR2 binds IFNs, such as Interferon-R2a (IFNR2a) * To whom correspondence should be addressed. Tel: +41 (0)61 68 82028; fax: +41 (0)61 68 87408; e-mail: hans.senn@ roche.com. ‡ These authors contributed equally to this work. 1 Abbreviations: IFN, Interferon; HCV, Hepatitis C Virus; IFNR2a, Interferon-R2a; IFNAR2EC, extracellular domain of the receptor IFNAR2; PEG, poly(ethylene glycol); 40PEG-IFNR2a (PEGASYS) 40 kDa branched PEG-conjugated Interferon-R2a, composed of monopegylated positional isomers; monopegylated positional isomer, one Interferon-R2a molecule with one 40 kDa branched PEG molecule covalently attached at the -amino group of a lysine residue of Interferon-R2a; 5PEG-IFNR2a, Interferon-R2a modified at the -amino group of a lysine residue with a 5 kDa linear PEG-polymer; SAW, Self-Avoiding-Walk; SEC, size-exclusion chromatography; LS, light-scattering; RPHPLC, reverse-phase high-performance liquid chromatography; SPR, surface plasmon resonance; NMR, nuclear magnetic resonance.

Figure 1. Schematic diagram showing the ternary complex formed by the binding of INFR2a to the human IFN R/β receptor comprising two transmenbrane subunits, IFNAR1 and IFNAR2. The IFNR2a structure (35) is reported as a ribbon with R-helices in orange and connecting loops and turns in light gray.

with high affinity (KD ∼ 2-10 nM), the intrinsic binding affinity of IFNAR1 to IFNR2a is low (KD > 100 nM). The ternary complex mediates activation of numerous genes

10.1021/bc049780h CCC: $30.25 © 2005 American Chemical Society Published on Web 05/03/2005

PEG−Interferon-R2a/IFNAR2EC Receptor Complex

through different transduction pathways (5). Various expression profiles are induced by IFNs, apparently through the same receptor with particular differences between IFNR and IFNβ. However, the molecular mechanisms underlying the functional differences between IFNs remain elusive. IFNs are currently one of the classes of human proteins most widely used as therapeutic agents, approved to treat various types of cancer and viral diseases. Since 1986, the medical potential of recombinant human IFNR2a has been recognized and the drug (Roferon-A) approved for the treatment of malignant melanoma, renal cell carcinoma, chronic myelogenous leukaemia, non-Hodgkin’s lymphoma, and Hepatitis B/C infections (8, 9). Today, a major worldwide health concern is infection by Hepatitis C Virus (HCV). The factors limiting the efficacy of IFNR in the treatment of HCV are the pharmacokinetic properties of IFNR, that lead to a short serum half-life of the drug: the protein is rapidly degraded, diffuses widely throughout the entire body, and has a high rate of renal clearance (10, 11). One successful way to overcome these limitations has been the use of pegylation, the process by which an inert, nontoxic, and water soluble poly(ethylene glycol) (PEG)polymer is covalently attached to a therapeutic protein (12-15). An earlier developed pegylated IFNR2a was obtained with a 5 kDa linear PEG-polymer (16). However, this first generation of pegylated IFNR2a, 5PEG-IFNR2a, did not exhibit significant improvements, in terms of pharmacological properties and efficacy, with respect to unmodified IFNR2a (13). Based on the latest advancements in pegylation technology and a better empirical basis of the correlation existing between the PEG-polymer characteristics and the resulting pharmacological properties of the modified protein (12, 17, 18), a second generation of pegylated IFNR2a drug was developed, 40PEG-IFNR2a (PEGASYS) (13). Here, by the covalent binding of one 40 kDa branched PEG-polymer to a lysine side-chain of the protein, a mixture of mainly six monopegylated positional isomers with distinct in vitro antiviral activity is obtained (19). The drug has a superior efficacy over IFNR2a, with increased serum half-life and reduced renal clearance (9, 20). PEGASYS received FDA approval for the HCV treatment in late 2002. Understanding the molecular basis of the in vitro biological activity of 40PEG-IFNR2a requires quantitative and qualitative interpretation of the structural, kinetic, and thermodynamic binding parameters for the interaction of the individual positional isomers with the receptors IFNAR1 and IFNAR2. Very recently, the NMR solution structure of the extracellular domain of the receptor IFNAR2 (IFNAR2EC) has been solved and a model of the IFNR2a/IFNAR2EC complex reported, presenting a molecular basis for the IFN binding mechanism (21). The IFNAR2EC binding site is formed by an extensive and predominantly hydrophobic patch, which interacts with a matching hydrophobic surface of IFNR2a, while complementary adjacent motifs of alternating charged side-chains provide specific recognition. In an accompanying publication we present the structural and biophysical characterization of the individual positional isomers of 40PEG-IFNR2a (22). In this study, we establish the correlation between the structural parameters and the kinetic and thermodynamic parameters of the binding of the positional isomers to IFNAR2EC. The results are compared to those obtained for IFNR2a and 5PEG-IFNR2a. In addition, we present results on the

Bioconjugate Chem., Vol. 16, No. 3, 2005 519

time-course of the digestion of IFNR2a and 40PEG-IFNR2a by trypsin that provide evidence for the PEG-polymer protection against proteolytic degradation. The interpretation of the experimental results is supported by modeling and simulation of the binding of the positional isomers with the polymer conformations mimicked as Self-Avoiding-Walk (SAW) constructs (23). The data presented here constitute an important step toward understand the rules that govern the in vitro antiviral activity of the different positional isomers of 40PEGIFNR2a and may support in the future the development of a third generation of pegylated IFNR2a or other pegylated proteins. MATERIALS AND METHODS

IFNR2a, Pegylated IFNR2a and Polymer Materials. The standard recombinant human Interferon-R2a, Roferon-A (IFNR2a) (19.2 kDa, 165 amino acids) (8, 24), the 40 kDa branched PEG-conjugated IFNR2a, PEGASYS (40PEG-IFNR2a), and the 5 kDa linear PEG-conjugated IFNR2a (5PEG-IFNR2a) were obtained from the Biopharmaceuticals Production Department of Hoffmann-La Roche Inc., as well the PEG-polymer reagents used for the pegylation reactions, the N-hydroxysuccinimide ester derivative of a 40 kDa branched PEG-polymer, and the pyridinyl carbamate derivative of a 5 kDa linear PEGpolymer (13, 16, 25). In 5PEG-IFNR2a and 40PEGIFNR2a, the PEG-polymer is attached nonspecifically on surface exposed -amino group of lysine residues via an urea bond and an amide bond, respectively (13, 16). The final product of the respective pegylation reactions is a mixture of monopegylated IFNR2a conjugates with potentially 11 lysine positions of modification (plus the N-terminus). Very recently, an ion-exchange HPLC separation procedure of the positional isomers of 40PEG-IFNR2a has been developed, and it has been shown that 40PEGIFNR2a is composed of nine bioactive isomers (19). Using this isolation method, the four major isomers modified at K31, K134, K131, K121 and the two minor isomers modified at K164 and K70 were separated with a purity of over 97% as determined by analytical ion-exchange HPLC. These six positional isomers have been extensively analyzed structurally (22) and constitute approximately ∼98% of the protein mixture. The molecular weight of IFNR2a protein, based on the amino acid content, is 19241 Da. The calculated extinction coefficients of IFNR2a and 40PEG-IFNR2a are 1.00 mL‚mg-1‚cm-1 and 1.05 mL‚mg-1‚cm-1, respectively. The sample concentration for the pegylated IFNR2as is given in mass of protein per unit of volume. In this work, the positional isomers refer to the four major and two minor monopegylated positional isomers of 40PEG-IFNR2a modified at K31, K134, K131, K121, K164, and K70, respectively. Production of the Extracellular Domain of IFNAR2 Receptor and Complex Formation with IFNR2a. IFNAR2EC was produced in an E. coli expression system and purified by ion-exchange and size-exclusion chromatography as previously described (26). The expression yield was ∼5 mg of protein per gram of biomass. The molecular weight of IFNAR2EC protein (216 amino acids ending with the sequence SESAES) determined by electro-spray-ionization mass specrometry was 24621 Da. This is in agreement with the expected mass of 24628 Da. Analytical Size-Exclusion Chromatography (SEC) on a Superdex-75 HR10/30 (Pharmacia Biotech) in 20 mM TRIS/HCl, 100 mM NaCl at pH 8.4 coupled to a Minidawn

520 Bioconjugate Chem., Vol. 16, No. 3, 2005

Light-Scattering (LS) detector system (Wyatt Technology, Germany) showed that IFNAR2EC was monomeric and aggregate-free in solution. As previously described (27), a (1:1) stoichiometric complex IFNR2a/IFNAR2EC was prepared and purified by SEC-LS with a molecular weight of ∼44.5 kDa. By performing an analytical RP-HPLC run of the complex IFNR2a/IFNAR2EC, the activity of IFNAR2EC was determined to be over 95%. Experimental Details of Surface Plasmon Resonance. The binding affinity of IFNR2a, 5PEG-IFNR2a, 40 PEG-IFNR2a, and its positional isomers for IFNAR2EC was determined by SPR measurements on a Biacore-3000 (Biacore, Sweden) instrument at 10, 17, 25, 30, and 37 °C. For all the samples, the measurement was repeated in duplicate or triplicate at nine different concentrations ranging from 0.6 nM to 1500 nM. The BIAevaluation program (Biacore, Sweden) was used to analyze and fit the experimental data to a (1:1) biomolecular interaction model with mass transport limitation. The binding affinity of the free 5 kDa linear and 40 kDa branched PEGpolymers was evaluated at 25 °C. The receptor IFNAR2EC was labeled with biotin before being immobilized on a streptavidin containing sensor surface (SA sensor, Biacore-3000). Biotinylation was performed by mixing a 0.8 mM solution of biotinamidocaproate-N-hydrosuccinimide ester (Fluka) in DMSO to the IFNAR2EC solution (0.1 mg‚mL-1 in 10 mM HEPES buffer at pH 8.4) to get a final concentration of 20 µM of biotinylated reagent. After 1 h of reaction, the solution was dialyzed overnight in 10 mM HEPES buffer at pH 8.4 to remove excess biotin. Immobilization was performed by contacting the biotinylated protein solution, diluted to 0.01 mg‚mL-1, with the SA sensor surface. For the binding experiments, the amount of immobilized protein was controlled via the contact time: 200 RU were immobilized, corresponding to approximately 0.2 ng‚mm-2 of IFNAR2EC. The sample and running buffer for the binding experiments was 10 mM HEPES, 150 mM NaCl, 3 mM CaCl2, 0.005% polysorbate-20 at pH 8.4 with a flow velocity of 50 µL‚min-1. The acquisition sampling rate was 5 points per second. Between the measurements no regeneration of the sensor surface was needed because of the fast dissociation process. Typical sensorgram responses are shown in Figure 2. Data Analysis of SPR: Equilibrium Binding and Kinetic Rate Constants. The equilibrium dissociation constant KDeq. was determined after analysis of the equilibrium SPR response with the Scatchard analysis: the ratio of the sensor response to the ligand concentration was plotted versus the sensor response, and the resulting slope of the straight line provided the equilibrium association constant KAeq. ) 1/KDeq.. The association and dissociation rates, kon and koff, were determined by fitting the time dependent association and dissociation phases with a mathematical model for a (1:1) complex formation. KDkin. was calculated as koff /kon. In case of fast association, kon could not be evaluated with a reliable fitting of the association phase. In such a case, only koff was obtained by the fitting of the dissociation phase by a monoexponential decay curve. The KDeq. value determined by the Scatchard analysis of the equilibrium sensor response was then used to calculate kon via the expression kon ) koff /KDeq.. Results obtained are presented in Table 1. Equilibrium Thermodynamics of the Binding to IFNAR2EC. Upon binding of IFNR2a and the pegylated IFNR2as to IFNAR2EC, the change in free energy ∆G° is related to the enthalpy change ∆H°, the entropy change

Dhalluin et al.

Figure 2. Sensorgram response of binding affinity measured by SPR with immobilized IFNAR2EC at 25 °C in contact with: (A) IFNR2a at concentrations 100, 80, 40, 20, 10, 5, and 2.5 nM, respectively (black line); (B) Isomer-K31 at concentrations 1000, 800, 600, 400, 300, 200 and 150 nM, respectively (blue line); (C) Isomer-K70 at concentrations 300, 290, 150, 120, 70, 40, 15, and 7.5 nM, respectively (red line). For panels A, B, and C, the green dashed lines represent the result of global fits using a (1:1) biomolecular interaction model. Panel D presents the overlay of the 100 nM sensorgram response of binding kinetics with immobilized IFNAR2EC in contact with IFNR2a, the IsomerK31 and -K70, colored in black, blue, and red lines, respectively. For comparison, the maximum sensorgram response at equilibrium was adjusted to the same level. Table 1. Kinetic Parameters kon, koff, KDkin. () koff/kon), and KDeq. of the Binding to IFNAR2EC of IFNr2a, 40PEG-IFNr 2a and Its Individual Positional Isomers, and 5PEG-IFNr a 2a kon 105 M-1‚s-1 koff 10-2 s-1 KDkin., nM KDeq., nM IFNR2a PEG-IFNR2ab

40

27.0 1.6

1.3 2.0

5.0 121

5.5 124

Isomer-K31 Isomer-K164

Low Affinity (LA) Isomer 1.0 5.2 2.3 3.3

520 150

500 150

Isomer-K134 Isomer-K131 Isomer-K121 Isomer-K70 5 PEG-IFNR2ac

High Affinity (HA) Isomer 2.7 2.1 4.8 1.9 5.3 2.1 5.5 1.8 7.4 1.3

76 40 39 32 17

77 43 40 32 17

a SPR measurements were done at 25 °C. For all the values, the standard error is (10%. b 40PEG-IFNR2a is a mixture of mainly six monopegylated positional isomers modified at K31, K134, K131, K121, K164, and K70, respectively, representing ∼98% of the mixture (19). c 5PEG-IFNR2a is a mixture of 11 monopegylated positional isomers modified at all the lysine positions (16).

∆S°, and the equilibrium association constant KAeq. according to the eq 1 with R the gas constant (1.99 × 10-3 kcal‚K-1‚mol-1).

∆G° ) ∆H° - T∆S° ) -RT ln KAeq.

(1)

If the enthalpy change does not vary with the temperature, the van’t Hoff plot of ln KAeq. versus 1/T results in a straight line of slope ∆H°/R.

ln KAeq ) -∆H°/RT + ∆S°/R

(2)

Although the intercept, ∆S°/R, provides the entropy, a better alternative is to use eq 1. However ∆H° and ∆S°

PEG−Interferon-R2a/IFNAR2EC Receptor Complex

Bioconjugate Chem., Vol. 16, No. 3, 2005 521

Table 2. Thermodynamic Parameters ∆G°, ∆H°, and T∆S°, for the Binding to IFNAR2EC of IFNr2a, 40PEG-IFNr 2a and Its Individual Positional Isomers, and 5PEG-IFNr a 2a ∆G°, ∆H°, T∆S°, ∆Cp°, kcal‚mol-1 kcal‚mol-1 kcal‚mol-1 kcal‚mol-1‚K-1 IFNR2a 40 PEG-IFNR2ab Isomer-K31 Isomer-K164 Isomer-K134 Isomer-K131 Isomer-K121 Isomer-K70 5 PEG-IFNR2ac

-11.2 -9.4

-5.7 -3.7

5.5 5.7

-0.2 n.d.d

Low Affinity (LA) Isomer -8.6 -6.5 2.1 -9.3 -6.0 3.3

-0.5 -0.6

High Affinity (HA) Isomer -9.7 -3.0 6.7 -10.0 -4.6 5.4 -10.1 -3.6 6.5 -10.2 -4.6 5.6 -10.6 -5.4 5.2

-0.2 -0.5 -0.1 -0.4 -0.4

a

The values were derived from the van’t Hoff analysis of the temperature dependence of the binding affinity data at equilibrium. T∆S° is calculated at 25 °C. The temperature dependence of the enthalpy and entropy was evaluated by the determination of ∆Cp°. The standard errors for ∆G°, ∆H°, T∆S° were estimated to be in the 10% range, and that of ∆Cp° was estimated to be in the 100% range. This estimated was done by a replicate of the Kd(T) series of IFNR2a which showed a experimental error on Kd in the 10% range. This experimental error was used as noise in a Monte Carlo simulation of the fit parameters of the van’t Hoff equation. b 40PEG-IFNR 2a is a mixture of mainly six monopegylated positional isomers modified at K31, K134, K131, K121, K164, and K70, respectively, representing ∼98% of the mixture (19). c 5PEGIFNR2a is a mixture of 11 monopegylated positional isomers modified at all the lysine positions (16). d n.d. ) not determined.

very often show temperature variations, resulting in a non-zero value of the heat capacity change ∆Cp°. Within the temperature range relevant for the study of such associations, ∆Cp° itself does not vary with temperature. The temperature dependence of the equilibrium association constant, KAeq., is then taken into account by introducing ∆Cp° in the integrated van’t Hoff eq 3.

ln(KAeq./K°) ) [(∆H° - To.∆Cp°) × (1/To - 1/T) + ∆Cp° ln(To/T)]/R (3) Here, a nonlinear fitting of the experimental data provides ∆H° and ∆Cp°. Thus, ∆G° and ∆S° were calculated using eq 1. The standard temperature To was taken to 25 °C. Thermodynamic parameters are summarized in Table 2. Monitoring of the in Vitro Digestion of IFNR2a and 40PEG-IFNR2a by Trypsin. The in vitro digestion of IFNR2a and 40PEG-IFNR2a by trypsin at room temperature was monitored by successive RP-HPLC runs spaced in time. RP-HPLC experiments were conducted with a Hewlett-Packard 1050 HPLC system equipped with a Supelco-C5 column (4.6 mm inner diameter × 150 mm length) from Sigma-Aldrich. Proteins were eluted using a gradient of acetonitrile in 0.08% trifluoroacetic acid and water in 0.1% trifluoroacetic acid at a flow rate of 0.25 mL‚min-1 and monitored by absorbance at 214 nm. Prior to measurement, IFNR2a and 40PEG-IFNR2a solutions were dialyzed in 20 mM TRIS 0.02% NaN3 at pH 7.4 and subjected to a preparative centrifugation to ensure the removal of aggregates. The protein solutions were subsequently diluted to a concentration of 0.25 mg‚mL-1 measured by UV absorbance (280 nm). Bovine pancreas trypsin of sequencing grade quality was purchased from Roche Diagnostics (Penzberg, Germany). This serine protease specifically cleaves peptide bonds at C-terminally arginine and lysine at an optimum

Figure 3. Time-course of the in vitro digestion of INFR2a and 40PEG-INFR , respectively by trypsin monitored with successive 2a RP-HPLC runs. For each time point, an aliquot (60 µL) of the incubation solution was injected. For both proteins tested, the concentration was 0.25 mg‚mL-1 in 20 mM TRIS 0.02% NaN3 at pH 7.4. The molar ratio (trypsin:protein) was (1:40). In the RP-HPLC chromatograms, the area of the elution peak, derived from the intact protein, was normalized with respect to the area of elution peak for the first time-point, i.e., 1 h after incubation.

working pH of 8.0. A stock solution of 1 mg‚mL-1 of trypsin was prepared in the buffer specified above. In individual sealed vials, the digestion was triggered by mixing a specific amount of trypsin solution to 1 mL of IFNR2a and 40PEG-IFNR2a solutions at 0.25 mg‚mL-1, respectively, to make a molar ratio (1:40; trypsin:protein). After 1 h of incubation, the first RP-HPLC chromatogram was recorded by injecting an aliquot of 60 µL from the vials. The monitoring of the in vitro digestion of IFNR2a and 40PEG-IFNR2a by trypsin was subsequently performed by time-spaced injections of 60 µL from the content of the vials over the time-course of the digestion. Each chromatogram provided a snapshot of the digestion at different times with decreasing signals derived from intact IFNR2a and 40PEG-IFNR2a, respectively, and new increasing signals derived from peptide fragments. IFNR2a and 40PEG-IFNR2a proteins displayed the same retention times in all their respective chromatograms. Digestive time-courses are shown in Figure 3. Self-Avoiding-Walk and Simulation of the Binding to IFNAR2EC. To structurally understand the differences observed for IFNAR2EC binding data of 40PEGIFNR2a and its positional isomers with respect to IFNR2a, a simple kinetic model (28) and numerical simulations of the protein-polymer construct were used. In the simulation, proteins are modeled as hard spheres of defined radii and the polymer-chains represented as an ensemble of on-lattice SAWs. All molecular modeling calculations were performed using the molecular modeling package MOLOC on a PC-workstation (29, 30). In a brute force on-lattice representation of a single PEG-polymer chain the grid in use has to encompass as many shells as there are bonds in the chain to prevent undesired compaction of the SAW. This requires, for a 40 kDa single PEG chain of 40PEG-IFNR2a a grid dimension of 465 (units) × 3 (bonds) ) 1395 shells, not tractable by use of the computational power available. Basic polymer theory (31) shows that a long polymerchain can be represented by a much smaller number of nodes if the unit length of the grid is defined by a properly selected statistical length that is related to the typical direction correlation distance of the polymer. To define this statistical length, 1000 SAWs of 72 steps were enumerated on a diamond lattice (dmd) of 72 shells;

522 Bioconjugate Chem., Vol. 16, No. 3, 2005

thus, the SAWs are never restricted by hitting the boundary as they start at the origin. Artifactual compaction of SAWs is excluded. Direction correlation distance was extracted from distance, i.e., number of steps, dependent walk direction correlation function, evaluated from the SAW ensemble. The number of steps over which these correlation functions decayed to e-1 was 1.163. For the definition of the statistical length needed, connection with real PEG chain was made by the experimental finding that the ratios for populations of gauche and trans steps is 90% for the O-C-C-O conformation and 10% for the C-O-C-C links (32). Thus, the real PEG chain loses its direction of propagation approximately every third step, i.e., every monomer. From this estimate and the direction correlation length of 1.163, diamond lattice walks of ∼465 steps (equal to the mean number of units in a 20 kDa PEG-chain) were considered. Thus, a reduction in the number of lattice shells by a factor of 3, corresponding to a reduction in computational time by a factor of 33 was achieved. The size of the SAWs is well below the dimension of the corresponding lattices. Errors introduced from walks bouncing off the lattice surface are prevented. Due to the scaling of lengths described above, IFNR2a has to be represented by a sphere of size 4.36 lattice bond distances. The spheres were placed such that the lattice origin was at the surface of the sphere. All walks started from this origin and could not penetrate to lattice vertexes inside the sphere. 4000 SAWs were generated. To evaluate the activation entropic contribution to the free energy of protein-protein association, a purely entropic model was used. In this model, a test sphere representing an approaching protein, e.g. a receptor or a protease, of varying radius was placed in the lattice box in surface-contact with IFNR2a sphere. This was done at various azimuth angles θ with respect to the starting point of the SAWs at the surface of the IFNR2a sphere. Sphere size and angle dependent contributions were obtained by counting the fraction of SAWs that did not penetrate the test sphere. Thus, the likelihood that a protein in surface-contact with the protein part of the pegylated construct intersects with the polymer was evaluated. This measure was calculated as the fraction of an ensemble of SAWs intersecting with a test sphere in contact with the pegylated protein but positioned at different locations with respect to the origin of the SAWs. The result can be displayed as an intersection probability p(θ) expressed as an activation entropic contribution ∆S‡(θ) to the free energy of association by a simple Boltzmann like relationship given by p(θ) ∝ exp(-∆S‡(θ)/ kB), with kB the Boltzmann constant. To test the influence of the lattice type used, we repeated all simulations with a simple (sc) a body (bcc) and a face-centered (fcc) cubic lattice. It turned out that the type of lattice has only a minor influence on the fraction of intersecting walks. Averages over all lattice types were used for further discussion. Results are shown in Figures 4 and 5. RESULTS AND DISCUSSION

Experimental Results: Kinetics and Thermodynamics of the Binding to IFNAR2EC. SPR experiments provided essential data for dissecting out the kinetics and mechanisms involved in the binding of IFNR2a, 5PEGIFNR2a, 40PEG-IFNR2a, and its positional isomers to IFNAR2EC. All the proteins, including the receptor IFNAR2EC, were obtained and used in a highly pure monomeric and aggregate-free form (see Materials and

Dhalluin et al.

Figure 4. Normalized probability of nonintersection [1 - p(θ)] of an ensemble of SAWs of the PEG-chains with a test sphere positioned in surface contact with the INFR2a sphere at an azimuth angle θ with respect to the origin of the SAW (see Materials and Methods). The origin is defined at the center of the INFR2a binding epitopes of IFNAR2EC (see text). The normalization is done with respect to a SAW starting on the opposite side of the sphere contact point given by [1 - p(180°)]. Two simulations are included: diamond-shaped points connected by a solid line, the diameter of the test sphere was the same as the INFR2a sphere; triangles connected by a dashed line, the diameter of the test sphere was doubled. Lines are shown to guide the eye. Filled circles represent angle corrected ratios kon/kon(180°) of experimental kon values of the positional isomers. kon(180°) is defined by kon(180°) ) 121kon × p(113°)/ p(180°), with 121kon the association rate of the Isomer-K121. The θ-angles are extracted from the literature (21). The definition of the x- and y-error bars is given in the text.

Figure 5. Simulation of the intersection probability p(30°) of an ensemble of SAWs of various PEG-polymers with a test sphere (same radius as the INFR2a sphere) positioned in surfacecontact with the INFR2a sphere at an azimuth angle of 30° normalized to p(180°) in dependence on the molecular weight of a PEG-polymer (see Materials and Methods). Clearly seen is the plateau of this ratio for a molecular weight beyond 20 kDa. Line is shown to guide the eye. The filled square and triangle correspond to a 5 kDa and 40 kDa PEG-polymer, respectively.

Methods section) (22). For all the samples, the SPR sensograms displayed a very good signal-to-noise ratio with an increase in the rate and extent of the binding with increasing protein concentrations. Typical sensorgrams for the binding to IFNAR2EC of varying concentrations of IFNR2a and the positional isomers modified at -K31 and -K70 are illustrated in Figure 2. For all the samples, KDkin., kon, and koff could be calculated. The binding affinity data extracted from the SPR measurements at 25 °C for IFNR2a, 5PEG-IFNR2a, 40PEGIFNR2a, and its individual positional isomers are reported in Table 1. Neither the free 5 kDa linear PEG-polymer nor the free 40 kDa branched PEG-polymer displayed any SPR response indicative of an affinity to the receptor (data not shown). IFNR2a binds to IFNAR2EC with a very high affinity (KDkin. ) 5 nM). This is in agreement with previously reported values and the model recently proposed for the complex IFNR2a/IFNAR2EC (21, 27, 33, 34). Here, the binding involves hydrophobic patches of IFNR2a interacting with matching hydrophobic surfaces of IFNAR2EC,

PEG−Interferon-R2a/IFNAR2EC Receptor Complex

Bioconjugate Chem., Vol. 16, No. 3, 2005 523

Figure 6. Representation of the complex structure INFR2a/IFNAR2EC obtained by the docking of the two individual NMR structures (21, 35). The model has been described and documented in ref 21. The IFNAR2EC structure is reported as a space-fill representation. The IFNR2a structure is reported as a ribbon with R-helices in orange and connecting loops and turns in light gray. At the lysine positions known to be pegylated in the positional isomers of 40PEG-INFR2a, two short PEG-chains are represented as on-lattice SAWs of 20 steps using the molecular modeling package MOLOC (29, 30). The PEG-chains of the class-LA isomers at K31 and K164 are colored in yellow. The PEG-chains of the class-HA isomers at K134, K131, K121, and K70 are colored in light green, light blue, dark blue, and dark green, respectively. Separation between the class-LA and -HA isomers is indicated by a dashed line. These figures were generated using the RasMol program (48).

while complementary adjacent motifs of alternating charged side-chains guide the two proteins in a fast association to form a tight complex. Compared to IFNR2a, the binding affinity of 5PEGIFNR2a to the receptor is ∼3.4-fold reduced, mainly due to a ∼3.6-fold decrease of the association rate. This must be attributed to the 5 kDa linear PEG-polymer. Once bound, 5PEG-IFNR2a dissociates with the same koff as IFNR2a. For 40PEG-IFNR2a, the binding affinity for the receptor is ∼24-fold reduced with respect to IFNR2a. Again, this is mainly due to a change of the association rate: kon decreases by ∼16.8-fold while koff remains essentially unaltered. All these results clearly show that the pegylation significantly reduces the affinity of the pegylated IFNR2a for IFNAR2EC without abolishment of the binding, and that the affinity is PEG-size dependent. However, as 40PEG-IFNR2a as well as 5PEG-IFNR2a is a mixture of monopegylated positional isomers (16, 19), the kinetic parameters extracted results from the weighted contributions of all the components in the mixture. Therefore, only a careful analysis of the binding of the individual isomers to the receptor can provide a clear interpretation of the kinetic binding parameters for 40 PEG-IFNR2a (see below). For 5PEG-IFNR2a, the isolated positional isomers were not available. Each positional isomer of 40PEG-IFNR2a exhibits different binding affinity for IFNAR2EC, ranging from 32 nM to 520 nM (Table 1). The losses of binding affinity are clearly attributable to the 40 kDa branched PEGpolymer and are dependent on the pegylation site. One can identify among the positional isomers two classes with respect to the kinetic binding process. The first class, called the class-LA (LA for low affinity), comprises the Isomer-K31 and -K164, that bind to the receptor with lower affinities (150-520 nM). In contrast, the second class, called the class-HA (HA for high affinity), comprises the Isomer-K134, -K131, -K121, and -K70, that interact with the receptor with higher affinities (32-76 nM). The two classes are directly related to two different areas of location of the pegylation sites (Figure 6). For the class-LA isomers, the PEG-polymer attachment point is located in proximity to the IFNAR2EC binding interface (21). A high steric hindrance of the protein-protein contacts due to the PEG-polymer seems therefore to be responsible for the reduced binding affinity of these isomers. This especially holds true for the Isomer-K31

that carries the polymer close to a hot-spot region for the interaction with IFNAR2EC (33, 34). For the class-HA isomers, the PEG-polymer is attached remotely with respect to the binding interface. This leads to lower reduction of binding affinities, certainly due to lower steric hindrance by the polymer. In all the isomers, the affinity losses are controlled by a substantial change of the association rate combined with a moderate change of the dissociation rate. The models, presented in the molecular modeling and simulation section, explain nicely the reduction in kon experimentally observed for the class-LA and -HA isomers by ∼16-fold and ∼6-fold, respectively. As far as the dissociation rate koff is concerned, only the class-LA isomers present notable differences with respect to IFNR2a: the Isomer-K31 and -K164 are released from the complex ∼4fold and ∼2.5-fold faster than IFNR2a, respectively. Again these differences are attributed to the location sites of the PEG-polymer. The attachment of the PEG-polymer at the binding interface may favor a faster release of the isomer than in the case where the polymer is located on the opposite side of the interface (Figure 6). Moreover, it is known that the charges contribute to the binding affinity to IFNAR2EC (21, 33, 34). Thus for the IsomerK31, the removal of one charge by pegylation of K31 might increase the dissociation rate. In the Isomer-K164, the high mobility of the C-terminal segment of IFNR2a (35) may bring the pegylated residue K164 near to the binding interface. Thus, the removal of one charge might also contribute in this isomer to increase koff. Experimental Results: Temperature Dependence of the Equilibrium Thermodynamics of the Binding to IFNAR2EC. The thermodynamic analysis of the temperature dependence of the binding affinity data at equilibrium, derived from SPR measurements, was performed using the van’t Hoff analysis. The free energy ∆G°, the enthalpy change ∆H°, the entropy change ∆S°, and the specific heat capacity ∆Cp° of the binding of IFNR2a and pegylated IFNR2as to IFNAR2EC at equilibrium could be determined. The thermodynamic parameters at equilibrium ∆G°, ∆H°, ∆S°, and ∆Cp° are reported in Table 2. The high affinity binding of IFNR2a to IFNAR2EC is a slightly enthalpy-driven process as the free energy of binding, ∆G° ) -11.2 kcal‚mol-1, results from an enthalpy contribution of ∆H° ) -5.7 kcal‚mol-1. The

524 Bioconjugate Chem., Vol. 16, No. 3, 2005

formation of electrostatic and hydrophobic contacts and hydrogen bonds between the two molecules account for this exothermic interaction (∆H° < 0), accompanied by an increase of entropy ∆S° also favorable for binding. This large entropy contribution suggests that the binding may involve a burial of hydrophobic surface, an increase in the conformation freedom of the proteins and/or a release of bound water molecules (36, 37). Water release is consistent with a charge compensation occurring at the binding interface upon complex formation. In the IFNR2a/ IFNAR2EC model, important electrostatic contacts are responsible for driving the two proteins in a tight complex (21). The binding of 5PEG-IFNR2a to IFNAR2EC is also a slightly enthalpy-driven process, with enthalpy and entropy losses of ∼5% and ∼10%, respectively, when compared to IFNR2a. These low ∆H° and ∆S° variations show the similarities of the thermodynamic binding process. In contrast for 40PEG-IFNR2a, the binding to IFNAR2EC is entropy-driven. But it must be noted that this change in character is mostly caused by a loss of enthalpy by ∼35% while the entropy remains unaltered. To provide a detailed description of the thermodynamics of binding of 40PEG-IFNR2a, each individual positional isomer was analyzed (see below). Upon binding to IFNAR2EC, all the positional isomers of 40PEG-IFNR2a gained enthalpy and entropy. With respect to the relative importance of electrostatic and hydrophobic contacts involved, this suggests that the isomers interact with IFNAR2EC globally in a similar manner as IFNR2a. Interestingly, the class-LA binds to the receptor, like IFNR2a, in an enthalpy-driven binding process. In contrast the binding for the class-HA members, like 40PEG-IFNR2a, is entropy-driven. As compared to IFNR2a, the class-LA isomers present a mean enthalpy gain of ∼10%. In contrast, the class-HA isomers display a mean enthalpy loss of ∼30%. Whether these trends of ∆H° change are due to alterations of remotely located surface charges and/or charges involved in the binding cannot be decided with the data available. The differences in ∆H° can also be attributed to differences in the details of the hydration of the proteins and the PEG-polymer involved (38, 39). An obvious possibility to explain the significant loss of entropy by ∼50% for the class-LA isomers, is the reduced number of conformations accessible to the PEG-polymer due to the steric hindrance of mobility arising from IFNR2a and IFNAR2EC, and viceversa (36, 40). Experimental Results: PEG-Polymer Protection against Proteolytic Degradation. To evaluate the PEG-polymer protection against proteolytic degradation, the in vitro digestion of IFNR2a and 40PEG-IFNR2a by trypsin was monitored by RP-HPLC. As illustrated in Figure 3, the area of the normalized signal of intact 40 PEG-IFNR2a in the chromatograms decreases ∼7-fold slower than IFNR2a. This shows a reduction of the in vitro proteolytic degradation due to the pegylation of IFNR2a. By imposing a steric hindrance on the protein, the PEGpolymer limits with a large excluded volume the accessibility for protease. The increased stability against proteolytic degradation was also reported for other pegylated proteins (41, 42). Molecular Modeling and Simulation of Binding to IFNAR2EC. The reduction in binding affinity by a factor of ∼24 observed for the 40PEG-IFNR2a/IFNAR2EC complex, as compared to its IFNR2a counterpart, is mainly determined by a corresponding difference in the association rate kon (Table 1). In general the kinetics of the formation of a protein-protein complex can be explained

Dhalluin et al.

by the following determinants. The collision rate, at which two partners meet, is controlled by the concentration of the two molecules. Under diffusion-limited conditions, the collision rate is independent of the molecular weight of the partners and their relative orientation. This rate constitutes an upper limit and is defined in water at 300 K by kcollision ≈ 6.6 × 109 M-1‚s-1 (28). If each hit of two molecules would lead to a complex, this value is the expected kon of the binding process. However, kon values for real proteins are often substantially lower, as productive complex formation depends on a correct position and relative orientation of the two partners with respect to each other. A correct preorientation of the binding partners can, for example, be achieved by longrange electrostatic interactions assuming a freely accessible binding area. In this case, the two binding partners can interact and lower the free-energy of the complex compared to the unbound state. In the case of the IFNR2a/ IFNAR2EC complex, it is estimated that only one out of approximately 2400 collisions leads to an active complex. This reduction is related to the ratio of the interaction surface to the overall surface of the proteins involved. It has been shown that kon scaling is expected to be proportional to RH-2, with RH being the hydrodynamic radius (28). With the RH scaling argument given above and using experimental RH values, kon for 40PEG-IFNR2a was estimated to be reduced ∼10-fold compared to IFNR2a (22). Thereby, it is assumed that the size of the contact area and the long-range interactions between the proteins are not altered by the presence and location of the PEG-polymer. Already, this simple kinetic model provides a rough estimate for the expected drop of kon and KD for pegylated proteins. It has to be emphasized that this reduction in kon proportional in RH-2 cannot be circumvented by optimized positioning of the polymer with respect to the binding interface. Further reduction of kon is expected if the polymer is attached near to or at the edge of the binding interface, due to steric hindrance of the formation of specific protein-protein contacts. This holds true in the case of the class-LA isomers (Table 1). But also the variations in kon between all the isomers in both cases can be rationalized as a consequence of different steric interactions in the process of complex formation due to spatial differences among the positional isomers. In our model elaborated in this work, hindrance is represented as the likelihood that a protein, e.g. IFNAR2EC, in surfacecontact with the protein domain of the pegylated construct intersects with the polymer. This likelihood was calculated as the fraction of an ensemble of SAWs intersecting with a test sphere in contact with the pegylated protein. The test sphere is positioned at different locations defined by an azimuth angle θ with respect to the origin of the SAWs. The result is displayed as an intersection probability p(θ) expressed as an activation entropic contribution ∆S‡(θ) to the free energy of association (see Materials and Methods). Thus [1 - p(θ)] defines the fraction of an approaching protein that can form an active complex. Pictorially, it is clear that the protection of a surfaceelement located on the surface of the IFNR2a sphere against a contact with an approaching protein is more pronounced if the PEG-polymer attachment point is located close in space. The normalized function [1 - p(θ)]/ [1 - p(180°)] is reported for two test sphere radii in Figure 4. It is a measure to rate the additional repulsive effect of the different isomers of 40PEG-IFNR2a as compared to the hypothetical most remotely attached polymer position, i.e., for θ ) 180°. In the complex of

PEG−Interferon-R2a/IFNAR2EC Receptor Complex

IFNAR2EC with the positional isomers, the starting points for the SAWs are defined by the azimuth angle θ between the PEG-polymer attachment point and the center of the binding interface (21). The reference kon value for the hypothetical isomer at θ ) 180°, kon(180°), was calculated by scaling kon of the Isomer-K121 (θ ) 113°) by the factor p(113°)/p(180°). A relative contribution for the destabilization of the complex is reported for each individual positional isomer with kon in Figure 4 by kon/kon(180°). It must be noted that the definition of the center of the binding interface is somewhat arbitrary. This reflects itself in angular θ-error bars given for the experimental data points in Figure 4. This SAW based-model can semiquantitatively explain the experimental differences in kon observed for the various isomers (Table 1). Especially, the experimental factor ∼5.5 for the isomer dependent kon variations is nicely reproduced with this simulation. In addition, the model predicts that only a factor below ∼10 in kon can be gained for hypothetical pegylation at θ ∼180° compared to pegylation at the rim (θ ∼30°) of the binding interface. Of high importance is the finding that if the pegylation is performed at an angle of at least ∼40° away from the ridge of the binding interface, i.e., θ ∼70°, only a ∼2-fold reduction in kon is expected compared to the sterically most favorable case with θ ) 180° (Figure 4). The azimuth angles of all the isomers, with the exception of the Isomer-K31 and -K164, lie in this range. These two isomers display a small θ angle, as their pegylation sites are located at or close to the rim of the binding interface. However, for the IsomerK164, it must be noted that the pegylation site resides in the highly flexible C-terminal region of the protein (35); therefore, it is not clear whether the extracted angle really represents an accurate value valid for the complex. One is tempted to increase the θ-angle for the IsomerK164 by approximately ∼15° still representing a conformation accessible for the flexible C-terminus without steric clash. Only the Isomer-K31 and -K134 experience a kon reduced by a factor ∼2.0 compared to the prediction of the SAW model (Figure 4). For the Isomer-K31, the deviation is in line with the argument that the removal of a charge on the side-chain influences the long-range interactions needed for the prepositioning of the proteins to support complex formation. Whether the same holds true for the Isomer-K134 is not clear. To summarize, the variations in kon observed for IFNR2a and the individual positional isomers of 40PEG-IFNR2a (Table 1) can be explained with reasonable accuracy by taking into account (a) the size (RH) of the unpegylated and pegylated IFNR2a (22) and (b) the differences in steric hindrance encountered between IFNAR2EC and the positional isomers in the process of complex formation. The first effect can be semiquantitatively explained by the rigid-body association model (28), and the second effect by the SAW based-model elaborated above. Only this latter model can convincingly explain the observed variations in binding affinities and association rates for the positional isomers. The SAW based-model can also make a prediction of the steric effect for different sizes of PEG-polymer attached at various θ-angles on intermolecular interaction. Figure 5 shows the ratio p(30°)/p(180°), expressing the relative hindrance at two extreme angles of PEG attachment site (see above) in dependence on the polymer molecular weight. This dependence becomes independent of the molecular weight of the polymer above 20 kDa. Therefore, very similar if not identical angle dependencies are expected for kon of proteins modified with

Bioconjugate Chem., Vol. 16, No. 3, 2005 525

polymer-chains larger than 20 kDa. For smaller PEGchains a more pronounced θ-dependence is observed (Figure 5). A substantially higher reduction of kon relative to the relaxed position at θ ) 180° is expected, if PEGchains smaller than 20 kDa were attached close to the binding interface, e.g. in case of a ∼5 kDa PEG-chain, the variations p(30°)/p(180°) are ∼6-times larger than that expected for a 20 kDa PEG-chain. In all the positional isomers investigated here, the PEG-polymer attachment site lies well outside or close to the rim of the binding epitopes of IFNR2a. If the attachment site were to be within the binding region, very pronounced effects on the binding affinity, association and dissociation rates would be expected. In this particular case, the flexible polymer-chain would severely prevent the correct alignment and interaction of the two binding partners, by its steric bulk. In IFNR2a, no lysine (with the exception of K31 known to lie close to a hotspot region of binding) is directly within this critical region (27, 33). For the stability and endurance of the complex, the strength of the electrostatic, hydrogen bond, van der Waals and hydrophobic (dehydration) interactions are of high importance. These effects, however, have only a minor effect on kon (see above), having a higher importance once the complex is formed. Pharmacological Effects and Molecular Properties. The reduction in receptor binding affinity of pegylated IFNR2a has direct implications on the in vitro biological activity. In the in vitro antiviral assay an approximately ∼3-fold and ∼30-fold lower specific activity was observed for 5PEG-IFNR2a and 40PEG-IFNR2a, respectively (16, 19). In the in vivo situation, this PEGsize dependent effect is more than counterbalanced by the two following positive findings: (a) the prolongation of PEG-chains goes in hand with an increase in stability and protection against proteolytic degradation, and (b) bigger particles experience a reduction in renal clearance by glomerular filtration (41-43). Both effects also depend critically on the size of the attached-PEG-polymer. The size of the pegylated protein particle is given by the hydrodynamic radius RH, and as it is dominated by a long polymer chain, it increases with MW3/5, with MW the polymer molecular weight (22, 31, 44). The simple kinetic model described above predicts a increased protection against proteolytic degradation roughly proportional in ∼RH2 (∼MW1.2). An estimate of the protection against the proteolytic degradation thus predicts a decrease of the digestion rate of 40PEG-IFNR2a by a factor ∼10 as compared to IFNR2a, in good agreement with the experimentally observed reduction of the in vitro digestion rate by a factor of ∼7 (Figure 3). Previously, it has been shown that the pegylation of IFNR2a reduces the tendency of the protein for aggregation (22). Here, the same rigid-body association model predicts that the self-aggregation of pegylated IFNR2a by protein-protein interactions is reduced by a scaling factor ∼RH-4. This remains to be confirmed experimentally. Due to the low concentration of the drug in the body, aggregation is expected to be of minor importance after the substance is administered. However, reduced aggregational property is essential while the protein is maintained in its formulation at higher concentration. It is well conceivable that this RH-4 dependence also relates to the substantially reduced rate of self-digestion if a polymer is attached on proteases (45, 46). Steric hindrance of protein approach brought about by the large hydrodynamic volume of the PEG-polymer (22) can also account for the significantly improved immunological properties of 40PEG-IFNR2a (9), as the immune system requires the recognition of the antigen.

526 Bioconjugate Chem., Vol. 16, No. 3, 2005

Clinical data show that the half-life of 40PEG-IFNR2a in the blood is at least by a factor of ∼8 longer than IFNR2a (20). The new drug PEGASYS has superior efficacy over IFNR2a. The reduced receptor affinity of 40 PEG-IFNR2a and its correspondingly reduced in vitro antiviral activity is thus more than compensated by the dramatically increased efficacy due to the significantly improved half-life in the blood. Impact for Future Molecular Design of Pegylated Proteins. The data, presented in this work and the accompanying paper (22), provide a solid collection of knowledge on variously pegylated IFNR2a. They contribute to a better understanding of the molecular framework responsible for improved structural stability, binding interaction, and pharmacological properties of pegylated proteins in general. Thus, future research and technology to optimize and tailor the pharmacological properties of proteins by pegylation may be rationally guided. Some of the methods presented already constitute new indispensable tools in the quality control of pegylated IFNR2as (22). For design purposes, the dependence of the intermolecular association rate kon on polymer size found (∼1/ RH2 ∝ 1/MW1.2), and its modulation by the spatial location of the PEG attachment site are of great interest. If possible the PEG-polymer attachment sites should be located as remotely as possible from the protein binding interface, but angles above ∼40° away from the rim of the binding interface already lead to only a minor decrease of the association rate kon as compared to the maximum value achievable by a given size of the PEGpolymer. The angle dependent effect is expected to be more pronounced for polymer chains of lower molecular weight (