Cytokine Activation by Antibody Fragments Targeted to Cytokine-Receptor Signaling Complexes*

Background: Cytokines are administered to patients to eliminate viral infections and cancer yet often have side effects. Results: Antibody fragments have been designed that recognize cytokine-receptor complexes and change cytokine biological activity profiles. Conclusion: Designed proteins enhance interferon antiviral activity without inducing antiproliferative signaling pathways. Significance: Antibody fragments targeted to cytokine-receptor complexes provide new tools for manipulating cytokine signaling. Exogenous cytokine therapy can induce systemic toxicity, which might be prevented by activating endogenously produced cytokines in local cell niches. Here we developed antibody-based activators of cytokine signaling (AcCS), which recognize cytokines only when they are bound to their cell surface receptors. AcCS were developed for type I interferons (IFNs), which induce cellular activities by binding to cell surface receptors IFNAR1 and IFNAR2. As a potential alternative to exogenous IFN therapy, AcCS were shown to potentiate the biological activities of natural IFNs by ∼100-fold. Biochemical and structural characterization demonstrates that the AcCS stabilize the IFN-IFNAR2 binary complex by recognizing an IFN-induced conformational change in IFNAR2. Using IFN mutants that disrupt IFNAR1 binding, AcCS were able to enhance IFN antiviral potency without activating antiproliferative responses. This suggests AcCS can be used to manipulate cytokine signaling for basic science and possibly for therapeutic applications.

Exogenous cytokine therapy can induce systemic toxicity, which might be prevented by activating endogenously produced cytokines in local cell niches. Here we developed antibodybased activators of cytokine signaling (AcCS), which recognize cytokines only when they are bound to their cell surface receptors. AcCS were developed for type I interferons (IFNs), which induce cellular activities by binding to cell surface receptors IFNAR1 and IFNAR2. As a potential alternative to exogenous IFN therapy, AcCS were shown to potentiate the biological activities of natural IFNs by ϳ100-fold. Biochemical and structural characterization demonstrates that the AcCS stabilize the IFN-IFNAR2 binary complex by recognizing an IFN-induced conformational change in IFNAR2. Using IFN mutants that disrupt IFNAR1 binding, AcCS were able to enhance IFN antiviral potency without activating antiproliferative responses. This suggests AcCS can be used to manipulate cytokine signaling for basic science and possibly for therapeutic applications.
Cytokines are ␣-helical proteins that regulate cell growth as well as innate and adaptive immunity (1,2). Thus, cytokinebased therapies have the potential to stimulate immune responses that eliminate pathogens and/or enhance antitumor activities. However, bolus injections of exogenous cytokines often induce toxicity in patients at concentrations below that required for therapeutic efficacy (3,4). Toxicity is thought to be associated with systemic activation of signaling pathways in cell types that are not associated with the infection or malignancy. Cytokine toxicity and/or poor efficacy is also associated with the activation of multiple cellular responses in the same cells (e.g. cytokine pleiotropy) (5). To address these problems, cytokines have been engineered to target specific cell types and exhibit different biological activity profiles that could improve their therapeutic profiles (6 -9). In contrast to these efforts, we sought to identify alternative proteins, called activators of cytokine signaling (AcCS), 3 which change the activity profiles of natural, or mutant cytokines, by changing the stability of the cytokine/receptor signaling complex.
It has been established that cytokines control cellular activation through a two-step engagement of cell surface receptors (10). In the first step cytokine binds to a high affinity receptor chain. In the second step, the binary complex interacts with a low affinity receptor chain resulting in the formation of a ternary cytokine-receptor signaling complex that initiates cellular responses. We hypothesized AcCS, which selectively bind to cytokine-receptor cell surface complexes (e.g. cytokine high affinity receptor chain, cytokine low affinity receptor chain, or the ternary cytokine-receptor signaling complex (TCRSC)), would enhance the biological potency of natural cytokines by increasing the stability of the TCRSC. Due to their ability to bind complex epitopes and escape host immune detection, humanized antibody scaffolds were chosen to develop as AcCS. Notably, AcCS are mechanistically and functional distinct from agonistic antibodies, which function as cytokine mimics with similar toxicity issues of normal cytokines (11,12). In contrast, AcCS are designed to increase cytokine activity only when cytokines are bound to their cell surface receptors. Thus, AcCS may provide a novel strategy to enhance the bioactivity of endogenously produced cytokines. AcCS were developed for the type I interferons (IFN), which are already used in the clinic (13,14). The IFN family consists of 16 IFN subtypes that adopt conserved ␣-helical bundle topologies and bind to the same cell surface receptors, IFNAR1 and IFNAR2 (15,16). In the interferon system, IFNAR2 is the high affinity receptor chain that exhibits ϳnM affinity for IFNs. The IFN low affinity receptor chain, IFNAR1, binds to the IFN-IFNAR2 binary complex with ϳM affinity to form the ternary IFN-IFNAR1-IFNAR2 signaling complex (17). IFN-induced ternary complex formation activates the JAK/STAT pathway and the expression interferon-stimulated genes (ISGs), which ultimately give rise to antiviral (AV), antiproliferative (AP), and immunomodulatory cellular responses (18). Based on these activities, IFNs have been administered as therapies for viral infections and cancer (IFN␣2) as well as for multiple sclerosis (IFN␤) (14,19). However, IFNs often exhibit dose-limiting toxicities and limited therapeutic effectiveness, which is disappointing considering the observation of potent tumor rejection and elimination of viral infections in animal models (20 -22).
The development of AcCS was initiated by the continued need to identify improved ways of activating IFN signaling without the associated toxicity problems. Toward this goal, we report the identification of two AcCS (AcCS1 and AcCS4) that potentiate IFN␣2a biological activity by ϳ100-fold. Biochemical and structural analysis demonstrate activation occurs due to the ability of the AcCS to stabilize the IFN-IFNAR2 binary complex. Based on this mechanism of action, we explored AcCS-mediated activation of IFN␣2a mutants with disrupted IFNAR1 binding sites. In the presence of these mutants, AcCS selectively activated IFN AV activity without inducing AP activity. Thus, AcCS provide a novel strategy for modulating IFN activity that can be extended to other cytokine systems.

Experimental Procedures
Phage Display and Protein Expression-IFN␣2a and IFNARs fused to an engineered immunoglobulin heterodimer (IFNAR1-FChk, IFNAR2-FChk, and IFNAR1-IFNAR2-FChk) were prepared as previously described (23). IFN␣2a mutants were made by quick change mutagenesis and expressed as previously described for IFN␣2a (24). For phage display, proteins were immobilized at 2 g/ml on Nunc Maxisorp microplates. All wells were blocked with a 0.5% bovine serum albumin (BSA) in phosphate-buffered saline (PBS), pH 7.4, for 1 h at room temperature with shaking. Purified phage library F (diversity of ϳ1 ϫ 10 10 at a concentration of 10 13 virus particles in PBS, 0.2% BSA, 0.5% Tween (PBT)) was incubated sequentially with each immobilized protein for 30 min to remove phage with affinity for the individual components of the complex. The depleted library was subsequently incubated for 2 h with the assembled IFN␣2a⅐IFNAR1-IFNAR2-FChk ternary complex. After incubation, the plate was washed eight times with PBT buffer. Remaining phages were eluted from the plate with 100 mM HCl. The pH of the eluted phage solution was neutralized with 1 M Tris, pH 11.0, and used to infect XL1-blue Escherichia coli cells (Stratagene). Phages were either amplified for use in further selections or plated as individual clone colonies for isolation, sequencing, and manipulation of clone DNA as previously described (25).
ELISA-Proteins (2 g/ml) used in the negative and positive selections were immobilized by passive adsorption by incubation overnight at 4°C in Maxisorp microplates (Nunc). The wells were subsequently blocked with 100 l of PBS containing 0.2% BSA (PB) for 1 h at room temperature before washing four times with PBS containing 0.05% Tween (PT). The IFN␣2a⅐IFNAR1-IFNAR2-FChk ternary complex was assembled in wells containing immobilized IFNAR1-IFNAR2-FChk heterodimer by incubating with 20 g/ml IFN␣2a in PBT for 30 min. Wells were then exposed to PBT-diluted phage (10 13 phage/ml) for 15 min, washed 8ϫ with PT buffer, and then incubated with a 1:5000 dilution of anti-M13 antibody (HRP conjugate, GE healthcare) for 30 min at room temperature. Microplates were washed again 6ϫ with PT buffer and 2ϫ with PBS and then developed by the addition of TMB substrate (KPL, Gaithersburg, MD). After 5 min, the reaction was stopped by the addition of an equal volume of 1 M H 3 PO 4 . The plates were read at 450 nm in a Biotek Powerwave XS microtiter plate reader (Biotek, Winooski, VT). Clones were considered specific for the interferon ternary complex if the signal was Ͼ10-fold above the individual proteins of the complex.
Subcloning, Expression, and Purification of AcCS-AcCSphage DNA was converted into expression plasmids by Kunkel mutagenesis of a C-terminal residue Cys (TGC) to a stop codon (TAG) to encode a soluble AcCS molecule containing a C-terminal FLAG tag. Plasmids were sequence-verified, transformed into 55244 cells by standard KCM methodology, and cultured overnight in 2ϫ yeast extract tryptone media containing carbenicillian. The overnight culture was gently pelleted and transferred to phosphate-depleted carbon-rich alkaline phosphatase (CRAP) media for expression overnight at 16°C. Cells were harvested by centrifugation and immediately lysed in buffer containing 50 mM Tris, 150 mM NaCl, 1%Triton X-100, 1 mg/ml lysozyme, 2 mM MgCl 2 , and 10 units of benzonase for 4 h at 4°C. Lysates were cleared by centrifugation at 16,000 ϫ g for 30 min and applied to rProtein A-Sepharose columns (GE Healthcare), washed with 10 column volumes of 50 mM Tris, 150 mM NaCl, pH 7.4, and eluted with 100 mM phosphoric acid buffer, pH 2.5 (50 mM NaH 2 PO 4 , 140 mM NaCl, 100 mM H 3 PO 4 ), into neutralizing buffer consisting of 1 M Tris, pH 8.0. Eluted AcCS were characterized for purity, concentration, and activity by SDS-PAGE gels, absorbance at 280 nm, and ELISA, respectively.
Surface Plasmon Resonance (SPR)-SPR experiments were performed on a Biacore T200 (GE Healthcare) at 25°C using a running buffer consisting of 10 mM Hepes, 150 mM NaCl, 0.0125% P20 (GE Healthcare), and 125 g/ml BSA. IFNAR-FChk fusion proteins were captured onto CM-5 sensor chips using an anti-murine FC antibody (GE Healthcare). All SPR experiments were performed in duplicate and double referenced (e.g. sensorgram data were subtracted from a control surface and from a buffer blank injection) as previously described (26). The control surface for all experiments consisted of the capture antibody. Approximately 100 -250 relative units (RU) of the various IFNAR-FChks were captured onto the chip surfaces. Fresh IFNAR-FChks were coupled to the surfaces for each injection. The surfaces were regenerated between injections with a 3-min injection of 10 mM glycine, pH 1.7. The buffer flow rate for all studies was 50 l/min. Sensorgrams were globally fit to a 1:1 binding model using Biacore T-200 evaluation software version 1.0. Antiviral Assays-Antiviral activity was performed using Huh-7 cells infected with vesicular stomatitis virus (VSV). Huh-7 cells were maintained at 37°C in DMEM/F-12 media (Mediatech Inc.) supplemented with 10% fetal calf serum. Cultures before the 11th passage were trypsinized and plated at 2 ϫ 10 5 cells/ml (100 l/well) in 96-well flat bottom microplates followed by overnight incubation at 37°C. The following day, AcCS/IFNs were added to the cells, and 24 h later the cells were challenged with VSV at a dilution, which induced 100% cell death in non-treated cells. Viable cells were quantified 40 h after viral challenge using Cell Titer 96 Aqueous One solution Cell Proliferation Assay kit (Promega) based on manufacturer's instructions. Dose-response curves were analyzed using PRISM (Graphpad Inc.) to derive EC 50 values.
Antiproliferative Assays-Anti-proliferative assays were performed using Huh-7 cells. Cells were maintained at 37°C in DMEM/F-12 media (Mediatech Inc.) supplemented with 10% fetal calf serum. Cultures of 5 ϫ 10 4 cells/ml (100 l/well) were added to 96-well plates and incubated overnight at 37°C followed by the addition of IFNs/AcCS. After a 72-h incubation, cell proliferation was measured using CellTiter-Glo (Promega) using the manufacturer's instructions. Dose-response curves were analyzed using PRISM (Graphpad Inc.) to derive EC 50 values.
Reporter Cell Assays-HL116 cells (27) were plated in white opaque plates (Corning) at 4 ϫ 10 5 cells/ml (100 l/well) and incubated overnight at 37°C. Dilutions of IFNs or AcCS were prepared in the DMEM-glutamax, HAT (Sigma), 10% FBS media and incubated for 20 min at 37°C followed by incubation with cells for 5 h at 37°C. After incubation, the plates were moved to room temperature for 10 min followed by the addition of 50 l of luciferase assay reagent (Steady-Glo, Promega) to each well. Luminescence was measured on a Biotek Synergy 2 plate reader and analyzed with PRISM software using a four parameter fit with variable slope (Graphpad Inc.).
Western Blot Analysis-Levels of tyrosine-phosphorylated STAT1 were measured by Western blot. After treatment cells were washed 3 times with Dulbecco's PBS containing 100 M sodium orthovanadate (Na 3 VO 4 ). Whole-cell lysates were prepared and quantified using the bicinchoninic acid assay (BCA, Pierce). Lysates were resolved by electrophoresis on 12% SDS-PAGE gels and then transferred to polyvinylidene difluoride membranes. The levels of tyrosine-phosphorylated STAT1 and unphosphorylated STAT1 were visualized by enhanced chemiluminescence (ECL) with anti-phospho-Y701-STAT1 (catalog #9167) and anti-STAT1 (catalog #9175) antibodies (Cell Signaling), respectively. An anti-␤-actin antibody (Sigma, catalog #A5316) was used to confirm equivalent sample loading.
Quantitative Real-time PCR (qPCR)-HuS-E/2 cells were grown as previously described (28). RNA isolation was performed using the RNeasy micro kit (Qiagen). RNA was quantified using the Quant-iT RiboGreen RNA Assay kit (Life Technologies) and a Rotor-Gene Instrument (Qiagen). Reverse transcription was performed from 500 ng of RNA in 10-l volumes using the Primescript RT master mix (Takara/Clontech) according to the manufacturer's instructions. qPCR was performed with a Roto-Gene using Premix EX Taq 2x master mix (Clontech/Takara) and Taqman MGB primer and probe sets (Life Technologies). All qPCR experiments were performed using duplex reactions, two replicates for each sample, containing the primer/probe of the gene of interest and a VIC-labeled GAPDH probe (primer pair-limited) as a housekeeping control. Relative gene expression was reported using the ⌬⌬CT method with the untreated control as reference (29).
Single Particle Electron Microscopy-AcCS-IFN␣2a-IFNAR2 complexes were formed by incubating the three proteins at a 1:1:1 molar ratio for 1 h followed purification by gel filtration chromatography. AcCS-IFN␣2a-IFNAR2 complexes (3 l) were applied to glow discharged holey carbon grids and stained with 2% uranyl formate. The grids were characterized using a Tecnai T12 microscope operating at 120 kV. EM tilt pair images (0°and Ϫ55°) were recorded at 52,000Ϫ magnification on a Tietz 4k ϫ 4k CCD camera with an electron dose of 40 e Ϫ /Å 2 using the LEGINON software system (30).
Electron microscopy (EM) images were processed using the APPION pipeline (31). Image defocus was estimated using ctffind3 (32) and applied by phase flipping the whole micrograph in EMAN (33). Particles were selected automatically using dog/tilt Picker from 330 (AcCS1 complex) and 371 (AcCS4 complex) images (34). Phase-flipped, 2ϫ binned stacks of particles (AcCS1 ϭ 105,598 particles, AcCS4 ϭ 72,637 particles) were created using a box size of 56 pixels and a pixel size of 4.1Å. After stack creation, the particles were classified using two-dimensional maximum likelihood procedures (35) implemented in XMIPP (36). Class averages that did not show all components of the complex, exhibited high noise, aggregation, or extensive distortion were rejected, resulting in stacks of 10,196 AcCS1 complex particles and 16,564 AcCS4 complex particles. Initial three-dimensional constructions were obtained using random conical tilt procedures from a subset of two-dimensional class averages, correlated in the 0°and Ϫ55°t ilt images, using the back projection and refinement procedures in APPION. A 50 Å low band pass filter was applied to the resulting random conical tilt volumes of each complex and used as initial three-dimensional reference models for refinement. Particle stacks were subjected to one additional round of twodimensional classification in RELION-1.3 (37). A total of 6,303 AcCS1 complex particles and 9,787 AcCS4 particles were used in gold-standard refinement and Fourier shell correlation (FSC) calculations in RELION-1.3 (37).
Modeling-IFN␣2a-IFNAR2 (38) and Herceptin (39) crystal structures were fit into the EM reconstructions using Chimera (40). AcCS1 and AcCS4 complimentarity determining regions were modeled using Rosetta antibody 3.0 (41) implemented on the Rosie server. The top 10 low energy FV models from Rosetta were superimposed onto the Herceptin structures that had beenpreviouslypositionedintotheAcCS1andAcCS4EMreconstructions. The lowest energy FV models that exhibited essentially no steric clashes with IFN␣2a-IFNAR2 were chosen from the 10 possible models. For both AcCS1 and AcCS4 models, the FV with the second lowest overall energy was chosen. No additional model refinement was performed on the docked FV models. All figures were generated using PyMOL (42).

Identification of AcCS-Prior studies have demonstrated IFN
biological activity is controlled by the affinity of IFNs for the individual IFNARs as well as the overall stability of the ternary signaling complex (43). Based on these data, we sought to obtain humanized antibody scaffolds that would specifically bind to the IFN-IFNAR1-IFNAR2 signaling complex and potentiate IFN activity. This was accomplished by screening a human synthetic antibody library (44) for antibody scaffolds that would bind to an IFN␣2/IFNAR1-IFNAR2-FChk ternary signaling complex mimic but not to any component individually (Fig. 1A). Scaffolds with affinity for the monomeric components of the complex were removed from the naïve library by negative selections, performed against IFN␣2a, the soluble receptors (IFNAR1-FCkh, IFNAR2-FCkh, IFNAR1-IFNAR2-FChk), and FChk (23). The remaining antibody scaffolds were subjected to four cycles of positive selection against IFN␣2a⅐IFNAR1-IFNAR2-FChk. Use of an IFNAR1-IFNAR2-FChk heterodimer (23) in the selection process was essential as it produced a highly stable ternary complex, in contrast to the complexes formed with monomeric IFNAR proteins lacking the FChk domains. Following the selection procedure, the specificity of 75 antibody scaffolds were characterized by ELISA. Two scaffolds (AcCS1 and AcCS4) exhibiting the greatest affinity for the IFN␣2a⅐IFNAR1-IFNAR2-FChk complex by ELISA (Fig. 1B) were purified for functional characterization.
AcCS Enhance IFN␣2a Antiviral and Antiproliferative Activity-The ability of the AcCS to stimulate IFN␣2a biological activity was determined by performing AV assays using VSV infection of Huh-7 cells (Fig. 1C). The addition of AcCS1 or AcCS4 to Huh-7 cells at concentrations as high as 1 M did not protect the cells from virus-induced death (viral cytopathic effect). However, in the presence of low levels of IFN␣2a, both AcCS enhanced AV activity in a concentration-dependent manner, with AcCS4 increasing IFN␣2a potency to a greater extent than AcCS1. The AcCS were responsible for increasing the potency of IFN␣2a, as increased potency was not observed when AcCS preparations were heat-denatured before addition to the cells (Fig. 1D).
IFN␣2a induces AV and AP activity in Huh-7 cells at effective concentrations (EC 50 ) of 29 pM and 330 pM, respectively ( Table 1, Fig. 1E). To further evaluate AcCS activity, IFN doseresponse curves were performed in the presence of constant concentrations of AcCS. Under these conditions, AcCS1 increased IFN␣2a AV activity 6-fold, relative to IFN␣2a alone, whereas AcCS4 increased IFN␣2a AV activity 29-fold. Thus, the AV potency of IFN␣2aϩAcCS4 mixtures is greater than the activity of IFN␤ in the same assay. AcCS enhanced IFN␣2a AP activity (53-83-fold) to a greater extent than AV activity. In fact, both AcCS increased the AP potency of IFN␣2a beyond that observed for IFN␤. In the presence of AcCS1, IFN␣2a exhibited essentially identical EC 50 values in the AV and AP assays, which is reflected in an AP/AV ratio of ϳ1 (Table 1). Interestingly, AcCS4 increased the AV and AP potency of IFN␣2a more than AcCS1 but maintained an AP/AV ratio of 4, as observed for IFN␤. Overall, these studies confirm AcCS1 and AcCS4 potently activate wild type IFN␣2a biological activity, often to levels greater than observed for IFN␤.
Sensorgrams derived from AcCS4ϩIFN␣2a mixtures were more complex due to the ability of AcCS4 and IFN␣2a to independently bind to IFNAR2 (Fig. 3D). However, formation of AcCS4-IFN␣2a-IFNAR2⅐FChk complexes could be identified with increasing IFN␣2a concentrations by the observation of a very slow off-rate in the sensorgrams that were distinct from sensorgrams for IFN␣2a (Fig. 3B) or AcCS4 (Fig. 3A) alone. Thus, AcCS4 is able to bind to IFNAR2 as well as stabilize the IFN␣2a-IFNAR2 complex with an apparent affinity of 0.19 nM ( Table 2). These data further define the distinct binding mechanisms of AcCS1 and AcCS4. Specifically, AcCS1 efficiently binds and stabilizes IFN␣2a-IFNAR2 complexes once they form, whereas AcCS4 binds to IFNAR2 before IFN␣2a binding yet is also able to stabilize the IFN␣2a-IFNAR2 complex.
AcCS Recognize an IFN␣2a-induced Conformational Change in IFNAR2-To further define the AcCS binding epitopes, structures of AcCS-IFN␣2a-IFNAR2 complexes were obtained using negative stain electron microscopy (Fig. 4). The structure of AcCS1-IFN␣2a-IFNAR2 Changed from FCS. (FSC1 ⁄ 2 ϭ 16.4 Å; Figs. 4C and 3D) revealed the AcCS1 binding site is located within the IFNAR2 D1 domain where it forms putative interactions with ␤-strands CЈ (residues Asp-51-Val-54) and F (resi-  Table 1. dues Glu-84 -Thr-91) (Fig. 5A). Based on the EM-derived model, prominent van der Waals contacts in the interface are made by tryptophan residues within AcCS1 complimentarity determining regions and residues on the IFNAR2 ␤-strand CЈ. In addition, IFNAR2 residues Asp-51 and Glu-84 may form an extensive hydrogen bond network with AcCS1 (Fig. 5A). To test the accuracy of this model, Asp-51 and Glu-84 were mutated to alanine (D51A, E84A, and the D51A/E84A double mutant), and AcCS1 binding affinity was evaluated (Fig. 5B). An additional mutation in IFNAR2 residue E28A, which is located on the opposite side of the AcCS1/IFNAR2 interface, was also made as a control mutation. Consistent with the EM-derived model, D51A and E84A mutants exhibited 139-fold and 35-fold reductions in AcCS1 binding affinity, respectively (Fig. 5B). No AcCS1 binding was detected to the D51A/E84A double mutant, whereas AcCS1 exhibited essentially wild type affinity for the E28A control mutation. Although the AcCS1 epitope is located within IFNAR2, AcCS1 does not bind to IFNAR2 unless it is bound to IFN␣2a (Figs. 2 and 3). This suggests AcCS1 increases IFN␣2a affinity by recognizing and stabilizing an IFN␣2a-induced conformational change in IFNAR2. Consistent with this hypothesis, NMR studies demonstrate IFN␣2a induces conformational changes in IFNAR2 residues 44 -53, which includes Asp-51, that is essential for AcCS1 binding (46).
The structure of AcCS4-IFN␣2a-IFNAR2 (FSC1 ⁄ 2 ϭ 16.4 Å, Fig. 4, E and F) revealed AcCS1 and AcCS4 share overlapping binding epitopes (Glu-51 and Glu-84) located within the IFNAR2 D1 domain (Fig. 6A). However, AcCS4 appears to form additional contacts with the IFNAR2 FG loop that are not present in the AcCS1 complex structure. These additional contacts are consistent with the ability of AcCS4, but not AcCS1, to bind to IFNAR2 in the absence of IFN␣2a (Figs. 2 and 3). Because AcCS4 and AcCS1 share overlapping binding epitopes, the IFNAR2 mutants used to characterize the AcCS1-IFN␣2a-   Table 2. RU, relative units. IFNAR2 complex were also used to characterize the AcCS4 binding epitope (Fig. 6B). Consistent with the AcCS4-IFN␣2a-IFNAR2 structural model, the IFNAR2 E84A mutant disrupted AcCS4 binding, whereas AcCS4 bound to the E28A control mutation with wild type affinity (Fig. 6). In contrast to the expectation that AcCS4 would not bind to the IFNAR2 D51A mutation, AcCS4 bound tighter to the D51A mutant than wild type IFNAR2 (Fig. 6). At this time we cannot explain the AcCS4 binding phenotype for the D51A mutation, which is located on the edge of the AcCS4/IFNAR2 interface. However, our data emphasize the energetic importance of IFNAR2 Glu-84 for AcCS4 binding, which is ϳ14 Å from the IFN␣2a/IFNAR2 interface and shown to undergo IFN␣2-induced conformational changes upon complex formation (46). Thus, IFN␣2ainduced conformational changes in IFNAR2 appear to be the mechanism by which AcCS1 and AcCS4 distinguish IFN␣2abound IFNAR2 from IFN␣2a-unbound IFNAR2. AcCS Binding to the IFN␣2a-IFNAR2 Complex Does Not Change IFNAR1 Affinity-AcCS binding to the IFN␣2a-IFNAR2 complex may increase the affinity of IFN␣2a for the IFNAR1 chain, which would explain how the AcCS enhance IFN␣2a biological activity. To evaluate this possibility, AcCS-IFN␣2a-IFNAR2 complexes were captured on SPR chips followed by the injection of soluble IFNAR1 chain over the complexes (Fig. 7). Using this assay, the affinity of IFNAR1 for the AcCS1-IFN␣2a-IFNAR2 and AcCS4-IFN␣2-IFNAR2 complexes was 3 M and 3.1 M, respectively, which is similar to previously reported affinities (K D ϭ 3.5 M) of the IFN␣2a-IFNAR1 interaction (16,23) These results suggest that AcCS1 and AcCS4 both enhance IFN␣2a biological potency exclusively by stabilizing the IFN␣2a-IFNAR2 binary complex.
AcCS Allow Decoupling of IFN Cell Potency and Gene Expression on HL116 Cells-IFNs induce complex biological responses in cells, which have been shown to depend on the stability of the IFN-IFNAR1-IFNAR2 ternary complex (16,43). Given the ability of the AcCS to form very stable IFN-IFNAR2 complexes, we sought to determine the influence of the AcCS on IFN␣2a-induced gene expression using two IFN␣2a mutants (NLYY and R120E) that exhibit wild type binding affinity for IFNAR2 but reduced affinity (R120E more than NLYY) for IFNAR1 (47,48). To measure gene expression, the activity of NLYY and R120E were characterized on HL116 cells that are stably transfected with an IFN-inducible reporter gene (27). Experiments performed with HL116 cells revealed IFN␣2a mutants with weak IFNAR1 binding properties exhibit decreased potency (e.g. increased EC 50 values) and reduced levels of gene expression relative to IFN␣2a (Fig. 8, Table 3). The addition of AcCS to the assay enhanced IFN␣2a or IFN␣2a mutant EC 50 values ϳ2-120-fold. However, neither AcCS1 nor AcCS4 was able to increase IFN-induced gene expression levels beyond that observed in the absence of the AcCS (Fig. 8). Thus, by stabilizing the IFN/IFNAR2 interaction, AcCS can enhance IFN potency without changing gene expression levels in HL116 reporter cells.
AcCS Potentiate the Antiviral Activity of IFN Mutants without Activating Antiproliferative Activity-Genes required for AV activity have been shown to have robust promoters that respond to weak IFN signaling, whereas promoters in genes associated with AP activity require stronger IFN induction (7). Studies on HL116 cells suggest the IFN-IFNAR2 interaction mediates IFN potency, whereas IFNAR1 regulates IFN potency and gene expression levels. If these findings are correct, AcCS should be able to enhance the AV potency of the IFN␣2a mutants with disrupted IFNAR1 binding sites without activating AP activity. To test this hypothesis, the AV (Fig. 9A) and AP (Fig. 9B) activity of NLYY and R120E were determined on Huh-7 cells with and without AcCS. On Huh-7 cells, R120E was unable to induce significant antiviral activity at concentrations as high as 2 M in the presence or absence of AcCS (Fig. 9A). In contrast, NLYY was able to almost fully protect Huh-7 cells from VSV infection at a concentration of ϳ1 M with an EC 50 value of 28 nM. AcCS1 had little impact on NLYY antiviral activity. However, AcCS4 increased NLYY AV potency by 93-fold (EC 50 ϭ 300 pM), which allowed NLYY to fully protect Huh-7 cells from VSV infection at concentrations of 1.6 nM. NLYY alone or in the presence of AcCS1 or AcCS4 was unable to induce AP activity on Huh-7 cells at the highest concentrations tested (2 M, Fig. 9B). Thus, in contrast to IFN␣2a, combinations of NLYY and AcCS4 confer complete AV protection to VSV without activating Huh-7 AP activity.
ISG Expression Profiles with and without AcCS4 -To determine how AcCS4 influences ISG expression, Huh-7 cells were treated with NLYY, IFN␣2a, NLYYϩAcCS4, IFN␣2aϩAcCS4,  Table 2. SDS-PAGE gel inset shows starting complex (S) and fractions from the column that were pooled for analysis. Fraction 25ϩ corresponds to fraction 25 that was incubated with ␤-mercaptoethanol before the addition to the gel. Final map/model (C) and gold-standard FSC plots (D) for AcCS1-IFNAR2-IFN␣2a and AcCS4-IFNAR2-IFN␣2a (E and F). AcCS1 and AcCS4 are shown in yellow and green, respectively. IFN␣2a is colored cyan, and IFNAR2 is in magenta.
The gene expression profiles of the six ISGs (Fig. 10) mimicked the potency of the IFN treatments in the AV and AP assays (Fig. 9, Table 1). Thus, IFN␣2aϩAcCS4 induced the highest levels of gene expression followed by IFN␣2a, NLYYϩAcCS4, and NLYY. Importantly, the gene expression profiles (e.g. the level of ISG induction at each time point) of NLYYϩAcCS4 are closer to IFN␣2a and IFN␣2aϩAcCS4 treatments than NLYY alone. This suggests NLYYϩAcCS4 may reach a gene expression threshold necessary for AV activity that is not reached by NLYY alone. Consistent with this hypothesis, NLYY-mediated induction of the antiviral gene isg20 is 10-fold lower than its induction by IFN␣2a, whereas NLYYϩAcCS4-stimulated isg20 levels are only 1.7-fold lower than in IFN␣2a-treated cells (Fig. 10D).  Distinct features of the gene expression profiles may also explain why NLYYϩAcCS4 treatment does not induce AP activity, whereas IFN␣2a or IFN␣2aϩAcCS4 treatment induces AP activity (Fig. 9B). Of the 6 ISG profiles measured, the greatest difference (10-fold) in gene expression between NLYYϩAcCS4 and IFN␣2a treatments occurs for cxcl11 (Fig.  10E). This is important because cxcl11 gene expression levels have been shown to correlate with IFN-mediated AP activity (52). Furthermore, IFN␣2aϩAcCS4 treatment, which induces the greatest AP activity (Fig. 9B), increases cxcl11 levels 30-fold over NLYYϩAcCS4 (Fig. 10E).
Gene expression levels of tnfrsf10a are also expected to correlate with Huh-7 AP activity (49,51). However, IFN␣2aϩAcCS4, IFN␣2a, and NLYYϩAcCS4 induce similar levels of tnfrsf10a (Fig. 10F). Although tnfrsf10a gene induction levels are very similar in all three conditions, in NLYYϩAcCS4treated samples, tnfrsf10a mRNA levels return to baseline after 24 h (Fig. 10F). In contrast, tnfrsf10a mRNA levels remain high in IFN␣2a-treated cells and are even higher with IFN␣2aϩAcCS4 treatment. Thus, fast decay of tnfrsf10a mRNA levels at 24/48 h post NLYYϩAcCS4 treatment are correlated with the loss of AP activity, whereas slow decay of   Table 3.

TABLE 3 IFN biological activity (EC 50 ) on reporter cells Ϯ AcCS
Shown is activity of IFN␣2a and IFN␣2a mutants on HL116 cells, which correspond to dose-response curves shown in Fig. 8 tnfrsf10a observed in IFN␣2a-and IFN␣2aϩAcCS4-treated cells (Fig. 10F) correlates with the strength of AP activity (Fig. 9B). AcCS4 Alters the Kinetics of IFN-induced STAT1 Phosphorylation-STAT1 phosphorylation in Huh-7 cells was evaluated by Western blotting (Fig. 11). Overall, pSTAT1 levels correlated with the potency of the IFN or IFNϩAcCS4 treatment observed in the AV and AP assays (Fig. 9, Table 1). Specifically, NLYY exhibited the weakest induction of pSTAT1 (Fig. 11A), which was increased by the addition of AcCS4 (Fig.  11B). The highest levels of pSTAT1 were induced by IFN␣2a and IFN␣2aϩAcCS4, although the differences between these two conditions were minor (Fig. 11, C and D).
IFNϩAcCS4 induced pSTAT1 with different kinetics from IFNs alone. For example, IFN␣2a and NLYY induced maximal pSTAT1 levels after 30 min that were slightly reduced at 5 h and then extremely weak (IFN␣2a) or completely absent (NLYY) at the 10-and 24-h time points. In contrast to the IFNs alone, IFNϩAcCS4 mixtures exhibited maximal pSTAT1 levels at 5 h rather than 30 min. However, despite this change, the decay of pSTAT1 appeared to occur with the same kinetics as the IFNs alone. Thus, even though AcCS4 substantially increased the level of pSTAT1 induced by NLYY at 5 h, pSTAT1 was not detected at the 10 or 24 h time points (Fig. 11, A and B). Thus, AcCS4 potentiates NLYY-induced pSTAT1 to levels similar to IFN␣2a or IFN␣2aϩAcCS4 treatment. However, NLYYϩAcCS4-induced pSTAT1 is absent 10 h post treat-ment, whereas pSTAT1 levels persist at 24 h with IFN␣2a and IFN␣2aϩAcCS4 treatment.
AcCS4 Does Not Significantly Activate Cells That Produce Constitutive Levels of IFN-Many cells produce low levels of IFNs (constitutive IFN) to establish a cellular threshold for subsequent robust IFN signaling (53,54). To determine if AcCS could activate endogenously produced IFN, AcCS4 was added to an immortalized human hepatocyte cell line (HuS-E/2 cells) that constitutively produces IFN␣ (28). The AcCS4-treated HuS-E/2 cells were evaluated for mxa gene expression at 5, 10, and 24 h after the addition of AcCS4 (Fig. 10G). AcCS4 induced mxa expression in HuS-E/2 cells by 28% relative to untreated cells, which did not reach statistical significance. Although this level of ISG activation might impact tonic IFN signaling (e.g. enhance activation upon the subsequent addition of IFN), AcCS4 cannot induce a robust IFN response from constitutive levels of IFN.

Discussion
Using IFNs as a model system, a method to identify AcCS that stabilize protein-protein interfaces of cytokine receptor complexes is described. Using a hybrid structural approach, we determined that AcCS1 and AcCS4 recognize an IFN␣2a-induced conformational change in the IFNAR2 chain and stabilize the IFN␣2a/IFNAR2 interaction. Thus, the AcCS are allosteric effectors of IFNAR2 that enhance IFN␣2a biological potency to levels greater than or equal to IFN␤. The properties of the AcCS were used to optimize IFN AV potency without activating off-target AP activity. These results might be important for the further clinical development of IFNs, as reducing the pleiotropic actions of IFNs may reduce the associated toxicity. Because the method of generating AcCS is quite straight forward, it should be easily adapted to other cytokines of interest.
The key to the selection of AcCS was the ability to screen antibody scaffolds against a stable IFN␣2a⅐IFNAR1-IFNAR2-FChk ternary complex, which reasonably mimics the cell surface signaling complex (23). Furthermore, the half-life of the IFN␣2a-IFNAR2 and IFN␣2a-IFNAR1 binary complexes are too fast (ϳ17 s or shorter) to use in the screening procedure. In contrast, IFN␣2a binds to the IFNAR1-IFNAR2-FChk complex with ϳ30 pM affinity and exhibits a half-life of Ͼ1 h, providing a highly stable antigen for the selections (23). Although the selections were performed against the entire IFN␣2a-IFNAR1-IFNAR2 complex, the two antibody scaffolds with the highest ELISA signals both recognize essentially the same epitope within the IFN␣2a-IFNAR2 complex. This is interesting because additional conformational changes have been reported in IFNAR1 upon IFN␣2a binding (38,55). Future changes in the library and/or the selection process may identify other epitopes unique to IFN-IFNAR1-IFNAR2 complex formation, which have been observed by structural and biophysical approaches.
Despite overlapping binding epitopes, AcCS1 and AcCS4 have distinct binding and IFN-activating properties. The current model of receptor-mediated IFN activation separates IFN ternary complex formation into an IFN-IFNAR2 binding step (K B ) followed by a translational interaction (K T ) of IFNAR1, which results in ternary complex formation and signaling (17).  Table 1.
Our studies show AcCS1 and AcCS4 impact IFN-IFNAR2 binding differently. AcCS1 binding is entirely dependent on IFN-IFNAR2 complex formation, whereas AcCS4 binds to IFNAR2 in the absence of IFN and subsequently stabilizes the IFN-IFNAR2 complex. The data suggest AcCS1 and AcCS4 both decrease the off-rate of K B to similar levels. However, FIGURE 10. Influence of AcCS4 on ISG expression in Huh-7 cells. A-F, Huh-7 cells were untreated (cyan) or treated with NLYY (gray), NLYYϩAcCS4 (blue), IFN␣2a (orange), IFN␣2aϩNLYY (yellow), or AcCS4 alone (green). At the time intervals of 5, 10, 24, and 48 h, ISG expression was measured by qPCR. The data are the mean Ϯ S.E. of mRNA levels from two biological replicates that were each measured twice by qPCR. The amounts of mRNA were normalized to the untreated control cells and plotted as relative mRNA versus time. The concentration of NLYY and IFN␣2a was 1 nM, and the concentration of AcCS4 was 200 nM in all experiments. G, HuS-E/2 cells, which constitutively produce IFN, were left untreated or treated with 200 nM AcCS4. Mxa gene expression was measured and reported as described for Huh-7 cells at the time intervals of 5, 10, and 24 h. AcCS1 is at a great disadvantage relative to AcCS4, as its binding is dependent on finding IFN-IFNAR2 complexes. The unique ability of AcCS4 to bind IFNAR2 in the absence of IFN␣2a allows AcCS to efficiently "capture" IFN on the cell surface for subsequent IFNAR1 binding and IFN-IFNAR1-IFNAR2 signaling complex formation.
Elucidation of the AcCS4 mechanism of action provided an opportunity to revisit the impact of IFN-IFNAR1 and IFN-IFNAR2 affinity on IFN biological activity. These studies demonstrated stabilizing the IFN␣2a-IFNAR2 interaction with AcCS4 could shift the AV and AP dose responses by 29-and 83-fold, respectively. In addition, AV activity was selectively induced (e.g. no AP activity was observed) in Huh-7 cells by using an IFN␣2a mutant (NLYY), with a weakened IFNAR1 binding site, but stabilizing the NLYY/IFNAR2 interaction with AcCS4. The gene expression studies suggest IFN potency (e.g. reduction in EC 50 values) is dependent on increased ISG levels, which is dependent on increased numbers of ternary signaling complexes for a given concentration of IFN. However, the selective activation of AV activity (e.g. no induction of AP activity), observed with NLYYϩAcCS4 treatment, appears to depend not only on reduced ISG expression but also on the rapid down-regulation of genes that induces AP activity. Our data suggest NLYYϩAcCS4 treatment induces a pulse of STAT1 phosphorylation and ISG expression that returns to pretreatment levels faster than with IFN␣2a and IFN␣2aϩAcCS4 treatments, which both induce potent AP activity. At this time it remains unclear if NLYYϩAcCS4 enhances negative signaling mechanisms that restore the cell to homeostasis or lower ISG levels induced by NLYYϩAcCS4 allow the cell to return to pre-stimulation levels faster. Mechanistically, the data suggest IFN-IFNAR2 stabilization increases ternary complex formation, gene expression, and biological activity, which are also promoted, to a lesser extent, by IFN-IFNAR1 interactions. However, the IFN-IFNAR1 interaction appears to play a special role in regulating time-dependent IFN signal termination. These findings may help to explain the unique biological profiles of some IFN␣ subtypes (16). Furthermore, these studies suggest tuning IFN-IFNAR2 stability with AcCS, and modulating IFNAR1 affinity could lead to the design of AV treatments with limited off-target effects.
The binding mechanism of the AcCS highlights the fact that they are not cytokine mimics, as they do not activate cytokine receptors in the absence of cytokine. Rather, they potentiate the biological activity of IFNs by stabilizing IFN/IFNAR interactions. The question arises: can't these properties be engineered into the IFNs themselves, eliminating the need for AcCS? Yes, some properties of the AcCS should be amenable to IFN design. For example, it should be possible to engineer an IFN that exhibits ultra-high binding affinity for the IFNAR2 chain. However, although some binding properties might be engineered into an IFN scaffold, further tuning of IFN signaling might require two distinct proteins (.e.g. IFN and AcCS) to achieve appropriate results for a particular application. For example, AcCS might be used to increase the activity of natural IFNs produced locally at the site of a viral infection. In addition, several recent studies have demonstrated targeting IFNs to specific cells can be accomplished using IFN fusion proteins, which target IFNs to various cell surface markers by a variety of binding proteins (9,20,45). Improved targeting efficiency of IFN fusion proteins was recently demonstrated by reducing IFN-IFNAR2 binding affinity (9). Combining IFN fusion proteins with AcCS may allow the targeted IFNs to be further inactivated to further improve targeting specificity but allow them to be re-activated using AcCS. Thus, AcCS provide an opportunity to independently control IFN targeting and activation steps. Although IFNs are the focus of this report, AcCS can be developed for a wide variety of cytokines where novel therapeutics or reagents are needed (1).