An Interferon α2 Mutant Optimized by Phage Display for IFNAR1 Binding Confers Specifically Enhanced Antitumor Activities*

All α-interferons (IFNα) bind the IFNAR1 receptor subunit with low affinity. Increasing the binding affinity was shown to specifically increase the antiproliferative potency of IFNα2. Here, we constructed a phage display library by randomizing three positions on IFNα2 previously shown to confer weak binding to IFNAR1. The tightest binding variant selected, comprised of mutations H57Y, E58N, and Q61S (YNS), was shown to bind IFNAR1 60-fold tighter compared with wild-type IFNα2, and 3-fold tighter compared with IFNβ. Binding of YNS to IFNAR2 was comparable with wild-type IFNα2. The YNS mutant conferred a 150-fold higher antiproliferative potency in WISH cells compared with wild-type IFNα2, whereas its antiviral activity was increased by only 3.5-fold. The high antiproliferative activity was related to an induction of apoptosis, as demonstrated by annexin V binding assays, and to specific gene induction, particularly TRAIL. To determine the potency of the YNS mutant in a xenograft cancer model, we injected it twice a week to nude mice carrying transplanted MDA231 human breast cancer cells. After 5 weeks, no tumors remained in mice treated with YNS, whereas most mice treated with wild-type IFNα2 showed visible tumors. Histological analysis of these tumors showed a significant anti-angiogenic effect of YNS, compared with wild-type IFNα2. This work demonstrates the application of detailed biophysical understanding in the process of protein engineering, yielding an interferon variant with highly increased biological potency.

Type I interferons (IFNs) 3 are multifunctional cytokines that orchestrate the antiviral innate immunity in vertebrates, acting in virtually every nucleated cell. They belong to the helical cytokine superfamily (1), and consist of at least 17 different members in humans: 13 different IFN␣ subtypes, IFN␤, IFN, and IFN, all of which share significant amino acid homology (2).
All type I interferons induce their biological effects through a common receptor, composed of two distinct transmembrane proteins, namely IFNAR1 and IFNAR2 (3)(4)(5). Upon formation of the ternary complex, the interferon signal is transduced through receptor-associated Janus family kinases (JAK), which activate the signal transducers and activators of transcription (STAT) proteins. These, in turn, form homo-and heterodimers that translocate to the nucleus, where they directly regulate the transcription of specific interferon responsive genes (6).
In addition to their important roles at the first line of defense against viral infections (7) and in modulation of the adaptive immune system (8), type I interferons are capable of inducing a significant antiproliferative activity (9). Consequently, interferons are widely used in the clinic for the treatment of multiple sclerosis and chronic hepatitis (10,11), as well as for a number of malignancies (12). However, the efficiency of interferon treatment in the majority of cancers is modest, limited by side effects and varying resistance of malignant cells to this treatment (13,14).
Despite their sequence homology and shared receptor, type I interferons are not redundant, but rather induce their activities differentially (15,16). One important example is the substantially higher antiproliferative response induced by IFN␤ relative to IFN␣. The exact molecular mechanisms underlying this differential signaling, which is initiated through the same receptor complex, remains unclear. However, we have recently shown that whereas the binding affinity of interferon toward IFNAR2 correlates with both antiviral and antiproliferative activities (17), a change in the affinity toward IFNAR1 translates differently between these two activities (18). Specifically, we have found that increasing the affinity of IFN␣2 toward IFNAR1 is accompanied by a significant increase in its antiproliferative activity, whereas its antiviral activity is minimally affected. Moreover, sequence alignment of the different human IFN␣ proteins showed that their weak binding toward IFNAR1 is a conserved characteristic, as manifested by three conserved positions on IFN␣2 (His 57 , Glu 58 , and Gln 61 ), of which mutations to Ala caused a significant increase in the binding affinity and antiproliferative activity of this mutant (termed HEQ).
Here, we exploited this important aspect of type I interferon behavior by using phage display to further optimize the IFNAR1-binding affinity of IFN␣2. Indeed, an optimized IFNAR1 binder, screened out of a mutant library, exhibited a 150-fold enhanced antiproliferative activity on cell cultures, accompanied by only a minor (3.5-fold) change in antiviral activity. In light of the importance of IFN in cancer therapy, we further discuss the implications of the YNS mutant in improving this mostly unfulfilled promise, and show how it confers a significantly stronger anti-tumor activity in a mouse model.

EXPERIMENTAL PROCEDURES
Cell Lines and Antibodies-Human epithelial WISH cells and human MDA-MB-231 breast cancer cells were kindly provided by Daniela Novick (Weizmann Institute of Science) and Lea Eisenbach (Weizmann Institute of Science), respectively. Monoclonal anti-IFNAR1-EC DB2 and AA3 antibodies were a gift from Biogene. Monoclonal anti-IFNAR2-EC 46.10 and 117.7 were a gift from Daniela Novick.
Phage Display-Screening for strong IFNAR1 binders was carried out using a hybrid phage-display system comprised of a phagemid encoding for a P3 viral-coat protein fused to an IFN␣2 gene followed by a FLAG tag, which encodes for an epitope peptide used to detect expression of this construct on the phage envelope by ELISA. A helper phage is added to encode for the necessary proteins for phage assembly (19). The library was produced by site-directed mutagenesis at positions 57, 58, and 61 of the fused IFN, using T7 DNA polymerase for primer extension on a single-stranded DNA template. The template was produced in Escherichia coli CJ236 cells (dutϪ/ ungϪ) incorporating uracils. The resulting phagemid library was transformed into ss320 bacteria by electrophoresis, and a phage library was subsequently produced. Proper IFN protein expression on the phages was confirmed by binding to Maxisorp plates (Nunc) coated with anti-FLAG antibodies, followed by ELISA using horseradish peroxidase-conjugated anti-phage antibodies. The proper folding of the expressed IFN was assessed according to the ability to bind IFNAR2-EC, using a similar ELISA sandwich assay where Maxisorp plates were coated with non-neutralizing anti-IFNAR2 46.10 antibody, followed by addition of IFNAR2 and IFN-producing phages. These were detected using horseradish peroxidase-conjugated anti-phage antibodies. Specific IFNAR1 library screening was done similarly, with Maxisorp plates coated with anti-IFNAR1-EC DB2 antibody (20), followed by addition of IFNAR1 and IFN producing phages. Control wells consisted of antibody coat only. Phage elution was carried out in three sequential steps, using 0.05, 0.1, and 0.2 M HCl. Phage numbers were calculated from the number of bacterial colonies formed after phage transfection at serial dilutions. After the fifth round of panning, individual phages were tested individually for IFNAR1-EC binding by ELISA. Phagemids from positive clones were extracted for sequencing.
Protein Expression and Purification-IFN␣2 and IFNAR2-EC were expressed using the expression vector pT7T318U in E. coli BL21 strain and purified by ion-exchange and size-exclusion chromatography (21). YNS was expressed in the E. coli Rosetta strain, where it was found mainly in inclusion bodies. Following four washing cycles with Triton wash solution (0.5% Triton X-100, 50 mM Tris, pH 8.0, and 100 mM NaCl) and one additional wash without Triton, the protein was unfolded in 6 M guanidine and then refolded by 1:20 dilution in 0.8 M arginine, pH 9.3, followed by dialysis in 25 mM Tris, pH 7.4. The protein was then purified on an ion-exchange column. Proteins were analyzed by analytical gel filtration chromatography to assess purity and binding activity. Protein concentrations were determined both by analytical gel filtration chromatography and from the absorbance at 280 nm with ⑀ 280 ϭ 18,500 cm Ϫ1 M Ϫ1 for IFN␣2 and ⑀ 280 ϭ 26,500 cm Ϫ1 M Ϫ1 for IFNAR2-EC. IFNAR1-EC was expressed in Sf9 insect cells and purified as described before (22), and was a generous gift from Jacob Piehler.
Binding Measurements-Binding affinities of IFN toward IFNAR1-EC or IFNAR2-EC were measured using the ProteOn XPR36 Protein Interaction Array system (Bio-Rad), based on surface plasmon resonance technology. A solution of 0.005% Tween 20 in PBS, pH 7.4, was used as running buffer at a flow rate of 30 l/min. For immobilization, an activated EDC/NHS surface was covered with the non-neutralizing antibodies DB2 and 46.10 against IFNAR1-EC and IFNAR2-EC, respectively, and blocked with ethanolamine. Thereafter, five of the six channels were reacted with IFNAR1-EC or IFNAR2-EC (180 l at a concentration of 0.5 M), leaving one channel free as reference. This was followed by cross-linking a second antibody, AA3 for IFNAR1-EC and 117.7 for IFNAR2-EC (23). Interferons were then injected perpendicular to ligands, at six different concentrations within a range of 37 to 8,000 nM for IFNAR1 binding and 3.12 to 100 nM for IFNAR2 binding. Data were analyzed with BIAeval 4.1 software, using the standard Langmuir models for fitting kinetic data. Dissociation constants K D were determined from the rate constants according to, or from the equilibrium response at six different analyte concentrations, fitted to the mass-action equation.
Determination of IC 50 Binding Values on WISH Cells-Wildtype (WT) IFN␣2 was labeled with 125 I using the chloramine T iodination method (24). After iodination, the radiolabeled interferon was cleaned using a homemade Sephadex column. For the competition assay, WISH cells were grown on 24-well plates, washed once with PBS ϩ 0.1% sodium azide, and then incubated for 10 min with the same solution. Cells were then incubated for 1 h at room temperature (20°C) with labeled IFN␣2 (200,000 cpm/well) in the presence of unlabeled IFN␣2 WT or YNS at different concentrations (50 nM to 25 pM), in culture medium ϩ 0.1% sodium azide. Thereafter, cells were washed twice in PBS, trypsinated, and transferred into test tubes for measuring of bound, 125 I-labeled IFN␣2 WT, using a ␥-counter (Packard). The experiment was repeated three times, each time in duplicates. IC 50 values were calculated using Kaleidagraph Synergy Software.
Antiviral and Antiproliferative Assays-The antiproliferative assay (25) on WISH and MDA-MB-231 cells was conducted by adding IFN (IFN␣2 WT, YNS, or IFN␤) at serial dilutions to the growth medium in flat-bottomed microtiter plates, and monitoring cell density after 72 h by staining with crystal violet. The 50% activity concentrations (EC 50 ), as well as the sensitivity of cells to increasing amounts of interferon, were deduced from an IFN dose-response curve (Kaleidagraph, Synergy Software) using Equation 2, where y is the absorbance and reflects the relative number of cells, A 0 is the offset, A is the amplitude, c is the IFN concentration, and s is the slope (25). Antiviral activity of IFN␣2 WT, YNS, and IFN␤ was assayed as the inhibition of the cytopathic effect of vesicular stomatitis virus on human WISH and MDA-MB-231 cells, as described previously (25,26). In general, IFN was added at serial dilutions to cells grown on flat-bottomed microtiter plates. Four hours later vesicular stomatitis virus was added to all wells, and after 17 h of incubation cell density was measured by crystal violet staining. EC 50 was calculated as described for the antiproliferative experiment. Both the antiviral and antiproliferative assays were repeated at least 3 times for each protein. The experimental error () for both assays was 35%. Therefore, a 2 confidence level would suggest that differences smaller than 2-fold between interferons are within the experimental error.
Annexin V/PI Assay-Apoptosis was monitored by the phosphatidylserine content on the outer leaflet of the cell membrane with the annexin V-FITC/PI assay kit (Bender MedSystems). Cells were detached with 5 mM EDTA in PBS and labeled with annexin V following the kit manufacturer's instructions. Samples were analyzed with FACSCalibur flow cytometer (BD Biosciences).
Quantitative PCR-Selected human interferon-stimulated gene (ISG) expression levels were measured with the ABI Prism 7300 Real-Time PCR System, using the SYBR Green PCR Master Mix (Applied Biosystems) and cDNA samples were produced with Superscript II (Invitrogen) from 1 g of total RNA, extracted with the RNeasy kit (Qiagen). Primer sequences were designed using the Primer Express software (Applied Biosystems). Standard curves were generated per primer pair using serial dilutions of an IFN treatment sample, to ensure reaction efficiencies of 100 Ϯ 5%: only primers that yielded a Ϫ3.3 slope (Ϯ 0.3), corresponding to an amplification efficiency of 2 (efficiency ϭ 10 Ϫ1/slope ) were used. Both primer and template concentrations were optimized. Quantitative PCR was performed using 1.25 ng of cDNA for HLA-F, ISG15, and MDA5, and 2.5 ng for TRAIL and CASP1 transcript levels, in a total volume of 20 l and at least in triplicate per sample. Relative expression levels were calculated by the ␦-␦ cycle threshold (Ct) relative quantification (RQ) method (ddCt, RQ ϭ 2 ϪddCt ), using the untreated cells control as the calibrator sample, and glyceraldehyde-3-phosphate dehydrogenase as the endogenous control detector (reference gene levels measured on the same sample), at least in triplicate for each sample. Standard error of ddCt, calculated per sample, was calculated as the square root of the sum of squares of four S.E. Ϯ mean Ct; S.E. of tested gene in treatment sample; S.E. of tested gene in control sample; S.E. of reference gene in treatment sample; and S.E. of reference gene in control sample.
In Vivo Experiments-All animal procedures were carried out in accordance with the Guidelines for the Care and Use of Research Animals at the Weizmann Institute of Science. Nude mice (n ϭ 23, 8-week-old) were injected subcutaneously with MDA-MB-231 human breast cancer cells (10 7 in 0.2 ml of PBS). Mice were then divided randomly into four treatment groups (IFN␣2, YNS, and control n ϭ 6; IFN␤, n ϭ 5). Interferon (20 g in 0.2 ml of PBS) was injected intraperitoneally; control mice were injected with 0.2 ml of PBS. Treatments were given twice a week, starting at day 3 up to day 35, after which mice were sacrificed; tumor remainders and suspected scar tissues were surgically removed, fixed in 4% formaldehyde, and embedded in paraffin blocks. Sections were stained with hematoxylin-eosin (H&E) for histological analysis. The protocol closely followed that used successfully by Wagner et al. (27).

Screening a Phage-displayed IFN␣2 Library for Tight IFNAR1
Binding-We have previously identified three conserved residues at the IFNAR1-binding region of IFN␣2, namely His 57 , Glu 58 , and Gln 61 , whose combined alanine substitution (termed HEQ) resulted both in a large increase of affinity toward IFNAR1 and a greatly enhanced antiproliferative but not antiviral activity in WISH cells (18). As it is difficult to envision alanine as an optimal binding residue, we established a phage-display system to isolate even tighter IFNAR1-binding interferon variants out of a library of IFN␣2 comprised of random mutations at these three positions. The gene encoding for IFN␣2 was inserted into a phagemid encoding for a fusion protein consisting of a phage P3-coat protein at the C-terminal end and a FLAG tag at the N terminus. The phagemid was transfected into E. coli bacteria, and expressed as part of the capsule of M13 phages following infection with K07M13 helper phages. To verify the proper translation of this fusion protein, we conducted an ELISA using anti-FLAG antibody-coated wells, and horseradish peroxidase-conjugated anti-phage antibodies for detection of specific binding. Positive signals were indicative of both a correct translation of the entire fusion protein in the correct reading frame, and a sufficient level of integration within the phage capsule. Next, the proper folding of the fused IFN␣2 was verified by its ability to bind IFNAR2-EC (K D ϳ 5 nM). This was done by a similar ELISA, using IFNAR2-ECcoated wells, which resulted in a significant signal of specifically bound phages.
The phagemid library containing all possible codon combinations at positions 57, 58, and 61 (8000 different clones) was constructed using degenerate primers, starting from a template containing a stop codon at each of these positions. The use of stop codons was necessary to avoid a high percentage of wildtype clones in the library, because the digestion of the singlestranded DNA template by the host SS320 bacteria is incomplete. The resulting library contained 10 11 phages/ml (according to serial dilutions), ensuring a high copy representation (ϳ10 7 phages/ml) of each clone. The first enrichment round of the library included a relatively mild washing step prior to elution with 0.1 mM HCl. As of the second round, the washing steps were increasingly aggressive, and starting from the third round the elution step was made up of three sequential elutions using 0.05, 0.1, and 0.2 mM HCl. A graphic representation of specific population enrichment along the panning rounds and the following decay thereof, which is a typical feature of this type of screen, is given in Fig. 1. Specific binding levels peaked at the fifth selection round, from which 20 single phage clones were tested for binding to IFNAR1-EC by ELISA. Four of them produced a significant signal (Ͼ3-fold higher than background). The residues replacing His 57 , Glu 58 , and Gln 61 in these clones were YNS, MDL, YLD, and YAS. A clear preference for Tyr is observed for position 57, but no clear preference was observed at the other two positions (except for a preference for polar over hydrophobic residues at a 5:2 ratio). Repeated phage ELISA has shown that the YLD and YAS clones were not stable, as they rapidly lost their binding activity toward both IFNAR1-EC and IFNAR2-EC. We therefore concentrated our efforts on expression and purification of the MDL and YNS clones.
Expression, Purification, and Characterization of the IFN␣2-YNS Mutant-The standard protocol for expression and purification of IFN␣2 mutants (21) did not work properly for the expression of YNS. We therefore elaborated a new expression protocol, based on unfolding in guanidine HCl and refolding in arginine (see "Experimental Procedures"). Using this protocol, we were able to produce and purify 7.5 mg/liter of the IFN␣2-YNS mutant. To examine its activity, we injected this mutant alone or together with IFNAR1-EC, IFNAR2-EC, or both receptor subunits to an analytical gel-filtration column (Fig. 2). The peak corresponding to YNS was occluded by either receptor separately (orange and red lines), as well as by both receptors together (dotted black line), while forming higher M r heterodimers and ternary complexes, respectively. It should be noted that the retention times of IFNAR1-EC and its complexes are shorter than expected from their calculated molecular mass (R1, 47 kDa; R2, 24.5 kDa; IFN, 18.5 kDa; R2 ϩ IFN, 43 kDa), due to glycosylation of the protein during production (22). The indicated receptor binding and ternary complex formation for YNS is possible due to its higher affinity to IFNAR1-EC, whereas occlusion of IFN␣2 WT by IFNAR1 was not detectable in this type of assay.
YNS Binds IFNAR1 with 60-Fold Higher Affinity than Wildtype IFN␣2-The binding affinities of various IFNs toward IFNAR1-EC and IFNAR2-EC were measured using the multichannel ProteOn XPR36 instrument (Bio-Rad) ( Fig. 3 and Table 1). The binding affinity of the YNS mutant toward IFNAR1-EC was 60-fold higher than that of the IFN␣2 WT and about 3-fold higher than measured for IFN␤, HEQ, or MDL. In contrast, the affinity of YNS, HEQ, and MDL toward the IFNAR2 subunit was about equal to that of the IFN␣2 WT protein. The binding parameters given here for IFN␤ and the HEQ mutant were similar to those reported by us previously   (18), but the affinity of IFN␣2 WT toward IFNAR1-EC is somewhat higher (1.9 versus 5 M). Due to the weak binding of IFN␣2 WT to IFNAR1-EC, its calculated affinity relies on mass action data rather than kinetic measurements as for the other interferons. Moreover, the current measurements were done on a ProteOn, whereas the previous measurements were done using RIfS, with a different receptor immobilization method applied in each study.
To verify the higher affinity of YNS also toward the full IFNAR receptor complex within the cellular environment, a binding competition assay on WISH cells was initiated. 125 I-Labeled IFN␣2 WT was mixed with cold IFN (either WT or YNS) at concentrations from 0.01 pM to 100 nM and binding of 125 I to the cells was measured (see "Experimental Procedures"). YNS exhibited a 20-fold lower IC 50 value compared with the wild-type protein ( Table 1). The somewhat lower increase in affinity compared with the in vitro assays can be attributed to specific factors of the cellular environment, such as the effect of receptor surface concentrations on ligand-receptor affinity (28), the relative abundance of the receptors on the cell membrane, and the measuring of ternary complex stability rather than single receptor subunit binding. Quantitative differences between IC 50 values and binding affinities toward purified receptor subunits are frequently observed.
Antiproliferative and Antiviral Potencies of YNS-Next, the biological activities of YNS, MDL, HEQ, IFN␣2 WT, and IFN␤ were assessed on human WISH cells. Table 2 shows the antiproliferative and antiviral potencies of these interferons, which were determined by their EC 50 from a dose-response curve (Equation 2). Consistent with its high affinity toward IFNAR1, the YNS mutant exhibited the highest antiproliferative potency, about 150-fold higher than the wild-type protein. Importantly, the antiviral potency of YNS was only 3.5-fold higher. This finding is in line with the small increase in antiviral activity measured for the HEQ and MDL mutants and for IFN␤, relative to their high antiproliferative potency, and further stresses the notion that the antiproliferative activity is differentially affected by the binding affinity toward IFNAR1.
The activity of YNS was also assessed on MDA MB231 human breast cancer cells. The antiproliferative potency of IFN␣2 on this cell line is similar to that measured on WISH cells (Table 2). However, the increase in activity determined for YNS and IFN␤ is somewhat smaller (80-and 20-fold, relative to 150-and 60-fold, respectively, Table 2). Interestingly, the antiviral activity of YNS was 16-fold higher compared with IFN␣2 WT on MDA cells. This is still much below the increase in the antiproliferative activity of this mutant, but above the small increase observed in WISH cells. However, comparing the absolute molar EC 50 values for YNS in both cell lines shows that they are virtually identical (0.2 versus 0.14 pM), suggesting that the antiviral potency is close to its theoretical limit for IFN␣2 WT on WISH, but not on MDA231 cells. Differences in activities of interferon on different cell lines were commonly found before. Nonetheless, both cell lines exhibit the same trend, with the highest antiproliferative potency being measured for IFN␣2-YNS, followed by IFN␤ and a much lower potency for IFN␣2 WT. Because YNS was a much more potent IFN mutant than MDL (whose performance was similar to HEQ), further characterization concentrated only on YNS.

The Higher Antiproliferative Potency of YNS Is Reflected by Higher Levels of Apoptosis at Low Protein Concentrations-
The antiproliferative activity pattern of YNS presented an interesting phenomenon in which relatively low concentrations of the YNS mutant were more efficient compared with IFN␤, reflected by its lower EC 50 value (Table 2), whereas at high concentrations YNS reached a maximal activity (as measured by cell density) similar to that observed for wild-type IFN␣2 and the HEQ mutant, which is somewhat lower than the maximal activity observed for IFN␤ (18). The antiproliferative activity of IFNs was previously reported to result from both growth arrest (29 -31) and apoptosis (32). We therefore turned to focus on the apoptotic activity of YNS. For this purpose, WISH cells were incubated for 72 h with IFN␣2 WT, YNS, or IFN␤ at different concentrations, labeled with annexin V-FITC and PI,

TABLE 1 Kinetic and affinity binding values of interferons with their receptor subunits IFNAR1 and IFNAR2
Binding was measured using the SPR-based ProteOn XPR36 instrument. K D ratios are relative to wild-type IFN␣2 and determined from the ratio of k d /k a .  and analyzed by flow cytometry. Annexin V is a specific apoptotic marker, whereas PI binds nucleic acids, thus detecting necrotic cells in which the cell membrane is damaged. At this point, we should mention that in this kind of cell culture it is expected that the majority of apoptotic cells are already PI positive, i.e. the state of annexin V-positive/PI-negative is transient (33). At low IFN concentrations, the apoptotic fraction was highest in cells incubated with 30 pM YNS (Fig. 4), in agreement with its EC 50 (Table 2). Remarkably, at 1.5 nM IFN concentration, IFN␤ slightly surpassed YNS, which has already reached maximal activity at about 150 pM (similar apoptotic levels as with 1.5 nM YNS), consistent with the maximal antiproliferative activity of these IFNs after 3 days of incubation. However, after longer incubation times the fraction of apoptotic cells induced by YNS closely matches that of IFN␤. 4 YNS Is a Stronger ISG Inducer Than IFN␤-We have previously shown that the HEQ mutant induces a similar gene expression profile as IFN␤ (18). Here, we wanted to determine whether the optimized YNS mutant could further increase the induction of ISG expression. Accordingly, we monitored ISG expression levels by quantitative PCR using the SYBR Green method in human WISH cells treated with different concentra-tions of IFN␣2 WT, IFN␤, or YNS, for 8, 16, and 36 h. Fig. 5 shows the expression patterns of five genes selected from the set of up-regulated genes in our microarray experiments (18). Both YNS and IFN␤ induced significantly higher expression levels than IFN␣2 WT, but with the same expression pattern. Notably, the differences between IFNs and their different concentrations increased with treatment duration. At 8 h, all IFN treatments induced the expression of the genes tested at much the same levels, whereas differences became apparent at the longer treatments. Indeed, as the treatment length increased, YNS was slightly more potent than IFN␤ at the lower concentrations (30 pM), whereas IFN␤ was more potent at the higher concentrations, in striking correlation with the antiproliferative activities of these interferons. Of note, the expression levels of TRAIL, a gene involved in the induction of apoptosis mainly by IFN␤ and less so by IFN␣ (34), paralleled the apoptosis levels seen in Fig.  4, much like the relation observed between apoptosis and their antiproliferation parameters: at the 30 pM concentrations, levels of TRAIL in cells incubated with YNS were the highest (particularly after 36 h), in agreement with the highest levels of apoptosis at the same concentration; at the higher concentrations, IFN␤ (150 pM) slightly surpasses the performance of YNS (300 pM), and IFN␣2 WT at even 3 nM.

IFNAR1-EC
Phosphorylation of STAT1 Is Transient and Limited to the Initial Few Hours of IFN Signaling-STAT1 is one of the main IFN signal transducers, activated upon phosphorylation of tyrosine at position 701. As the antiproliferative activity of IFN on cell culture relies on continuous IFN treatment (data not shown), we asked whether this activity, as well as ISG induction at longer IFN treatments, requires STAT1 activity. Fig. 6A shows that phosphorylated STAT1 rapidly accumulates at the onset of the IFN treatment, reaching a maximum around 1 h of 4 D. A. Jaitin, R. Abramovich, and G. Schreiber, unpublished results.  30 pM IFN␣2 induction, and around 30 min of 30 pM YNS, but then gradually declines. Despite the strong induction of STAT1 protein levels at longer time points, in agreement with its gene expression levels (18), no significant levels of pSTAT1 were observed at prolonged IFN treatments, irrespective of the type of IFN and its concentration used (Fig. 6B). These results seem to indicate that other signaling elements mediate the IFN antiproliferative activity, in addition to or independently of the canonical JAK/STAT signal transduction pathway.
The YNS Mutant Exhibits Increased Anti-tumor Activity in a Mouse Model-Following the high antiproliferative activity of YNS in tissue-cultured cells, we turned to assess its biological potency as a tumor inhibitor in vivo. As human interferons are active only in human cells, we extended the experiment to breast cancer xenografts transplanted into nude mice. MDA MB231 cells were subcutaneously injected to nude mice, and the animals were divided into four treatment groups: YNS (n ϭ 6), wild type IFN␣2 (n ϭ 6), IFN␤ (n ϭ 5), and buffer (PBS, n ϭ 6) as a negative control. These four treatments were injected intraperitoneally twice a week, for a total of 5 weeks. Numbers of tumor lumps and their relative size were evaluated and recorded prior to each injection (Table 3). At the end of the experiment (after 5 weeks), the animals were sacrificed, the size of the lumps was measured, and samples were taken for histological analysis. Following 1 week of MDA231 cell injection, all mice developed xenograft tumors. Within 26 days following the first IFN injection, all xenograft tumors in the YNS-treated mice were cleared, whereas in the wild-type treated group five of six mice still displayed visible tumors. IFN␤-treated mice were free of tumors 22 days after the first injection, with the exception of a single lump. The PBS-treated group displayed visible tumors in all mice, which were larger compared with tumors in the wild-type mice (Table 3 and Fig. 7A). Notably, the tumors of four of six mice in the YNS-treated group went through a necrotic process, as was the case for five of five tumors in the IFN␤ group. This phenomenon was absent in the wild-type IFN␣2 and control groups. The remaining tumors, as well as scars, were subjected to histological analysis (Fig. 7B). All sections from the YNS-treated group were clean of tumor cells, displaying healthy scar tissues. The same picture was apparent for sections from IFN␤-treated mice. The single visible lump in the IFN␤ group turned out to be an inflammatory response, probably due to bacterial infection (implied by the large presence of neutrophils), and was clean of cancerous cells. Out of the five visible tumors in the wild type-treated group, three exhibited clear tumor morphology, and two consisted mainly of fat cells.

DISCUSSION
IFN␣2 is the first approved biotherapy agent against human malignancy, and is currently used in the clinic for the treatment of different types of cancer, including hairy cell leukemia, malignant melanoma, AIDS-related Kaposi sarcoma, follicular lymphoma, and more. However, the efficiency of interferons in the treatment of these malignancies varies, and for most solid tumors remains poor. Treatments are limited by the varying levels of cellular resistance exhibited by the tumors, and significant side effects that are most likely to stem from the pleiotropic nature of interferons. Our current understanding of the molecular mechanisms that govern the diverse effects of interferons is incomplete. However, in a system consisting of 17 different ligands and a single shared receptor complex, receptor-ligand interactions are bound to play a significant functional role. Due to its low binding affinity toward IFN, IFNAR1 is widely referred to as the "signaling" subunit, confined to the IFN signal transduction (35). In a recent work, we identified three conserved positions on IFN␣2 that confer low binding affinities toward IFNAR1 (25). Coupled to an enhanced IFNAR1-binding affinity, combined alanine substitutions of these three positions exhibited a higher antiproliferative activity in tissue culture, but only a minor change (ϳ2-fold) in antiviral activity (18). Given this substantial effect of the neutralized positions, we assumed that an optimized IFNAR1 binder could be engineered that will be used for further investigation of the specific contributions of IFNAR1 to interferon activity. In the work presented here, we successfully isolated such an optimized mutant using a phage display system. This elegant method allows the rapid and simultaneous screen of extremely diverse mutant libraries, whereas maintaining a physical link

TABLE 3 Lump size of MDA231 breast cancer xenografts transplanted into nude mice
Mice were injected with MDA231 cells on day 0, followed by injections of 20 g/mice of the different treatments at the indicated times. Six mice were treated for each experimental condition, except five for IFN␤. Growth of tumors was estimated by eye for all but day 34. On the last day (34) mice were sacrificed, and tumors were extracted and measured. L, M, S and N stand for tumor size (in parentheses is the measured diameter at day 34), L, large (Ͼ5 mm); M, medium (2.5-5 mm); S, small (Ͻ2.5 mm); N, none. between the displayed mutant and the DNA sequence encoding it. Notably, within the most specific phage population along our screen (following the fifth panning round) only four of 20 indi-vidually tested phages gave a significantly specific signal. This was sufficient, however, for the identification of a clone with a 60-fold tighter binding affinity toward IFNAR1, bearing mutation Tyr 57 -Asn 58 -Ser 61 (designated YNS). This binding affinity, which is 3-fold stronger than that of IFN␤, is accompanied by an affinity toward IFNAR2 similar to IFN␣2 wild-type, but 25-fold bellow that of IFN␤. The tighter binding of IFN␤ to IFNAR2 that is not accompanied by enhanced activity, suggests that IFNAR1, and not IFNAR2 binding, is limiting the IFN biological activity. However, we have also shown that a mutation on IFN␣2 where we replaced its tail with that of IFN␣8 resulted in 20-fold increased binding affinity to IFNAR2 and 10-fold increased antiproliferative potency (36). Thus, it seems that activity cannot be simply described as the sum of binding affinities to the two receptors.
Notably, a unique role for IFNAR1 in IFN signaling has been proposed by Pfeffer et al. (37) and later by Rani et al. (38), who showed that this subunit can associate with PI3K in an indirect or direct manner, respectively. A more defined role for PI3K as a parallel pathway in IFN signaling was suggested by Thyrell et al. (39), who showed that the treatment of multiple myeloma cells with LY294002, an inhibitor of PI3K activity, inhibits IFNinduced apoptosis but not antiviral activity and STAT activation. However, in our system (using WISH cells and IFN␣2 WT, YNS, or IFN␤) the inhibition of PI3K with LY294002 did not have any effect on the antiproliferative activity of IFN. Indeed, LY294002 inhibited cell growth in a dose-dependant manner, but it did not alter the effects of interferon on growth as previously suggested (data not shown). In addition, we found no apparent correlation between the IFN antiproliferative effect and the phosphorylation of STAT1. This is particularly true after longer times of IFN induction, which are absolutely required for the antiproliferative response. 4 Even though STAT1 levels significantly rise during the long term induction, they are not accompanied by phosphorylation (Fig. 6B). Yet, a correlation was observed between apoptosis, the interferon subtype used, its concentration and duration of treatment, and specific gene expression patterns. This is particularly conspicuous in the TRAIL expression behavior (Fig. 5, TRAIL panel), a gene highly implicated in interferon-mediated apoptosis (33). A similar correlation was found by us also between IFNAR1 surface expression and antiproliferative activity of interferons. 5 These observations suggest that additional factors are involved in the transduction of the interferon antiproliferative signaling.
The YNS mutation changes the electrostatic profile of the IFNAR1 binding surface (Fig. 8), as well as reduces the volume of the side chains by 24 Å 2 . The reduction in size seems to be an important feature, as already the AAA mutation (termed HEQ in Table 1) confers a 20-fold higher binding affinity compared with the wild-type His, Glu, and Gln residues. However, the YNS mutation adds additional binding energy over the AAA mutation (whereas the MDL mutant does not), suggesting that these residues also make new specific interactions with IFNAR1, in addition to removing negatively contributing side chains on the wild-type protein. One should note that the wild-  type HEQ residues are evolutionary conserved, suggesting that weak binding is important for differential IFN␣ action (18,25).
As we expected, the antiproliferative potency of the YNS mutant as measured in WISH cells substantially increased to an EC 50 value of 0.02 nM, which is 150-fold greater than for the wild-type protein. The antiproliferative activity of YNS was much higher compared with that of IFN␣2 wild-type or IFN␤, also in MDA231 cells, suggesting that the enhanced activity is not limited to WISH cells. Moreover, as MDA231 cells originate from human breast cancer, the anti-malignant activity of YNS becomes evident. In marked contrast, the antiviral activity of YNS was only slightly enhanced (3.5-fold). The significant differences in biological responses within this unprecedented range of affinity values support former data regarding the role of IFNAR1 in facilitating the differential activities of IFNs in cells, particularly that of IFN␤ (see also Ref. 18).
In line with the observed activities in tissue culture, the mouse xenograft experiment indicated that the efficiency of YNS in eradicating MDA231 tumor cells is significantly high also in an animal model. No tumors were visible following 26 and 22 days of treatment with YNS and IFN␤, respectively, as was also confirmed by a following histological analysis. Tumor morphology throughout the experiment provided us with important information regarding an additional aspect of YNS activity: a necrotic process accompanied tumor regression in mice treated with YNS, as opposed to the wild-type IFN␣2 that failed to induce this effect within the tested concentrations. A similar necrotic process was evident in the IFN␤-treated group. The observed process was identified as resulting from ischemia, following a failure of the tumor cells to induce neovascularization within their microenvironment (40). The formation of new blood vessels around tumors reflects a balance between pro-and anti-angiogenic factors, which are produced by both the tumor cells and the host (41); because murine cells are not responsive to human IFNs (42), the observed effects can be attributed only to the inhibition of pro-angiogenesis processes within the tumor cells. Indeed, the inhibition of angiogenesis in tumors following exposure to type I IFNs is well documented (42)(43)(44), and is ascribed to an IFN-induced down-regulation of pro-angiogenic genes such as bFGF, VEGF, and MMP-9 (13,45). The ability of YNS to exert a significant anti-angiogenic effect, which cannot be achieved using similar doses of the wild-type protein IFN␣2, adds to its enhanced antiproliferative potency in the overall anti-tumor activity observed in vivo. These data are of particular value when considering the relatively low effectiveness of IFN␣2 on most solid tumors in vivo, and might provide the grounds for the design of novel therapeutic agents of the IFN␣ family. At the molecular level, enhancement of the antiproliferative activity of IFN␣2 in a discrete manner will make corresponding signaling pathways easier to monitor, based on observed signal augmentation specifically relayed by YNS.
In this work we went from the IFN␣2 screening toward an optimized IFNAR1-binding mutant, through its biophysical characterization, evaluation of its biological activity showing increased antiproliferative potency, and eventually demonstration of its enhanced antitumor activity in a mouse xenograft model. Apart from the gained insights on the molecular mechanisms of IFN action, we demonstrated the importance of connecting proper biophysical understanding of a biological process along with a profound knowledge of the proteins involved, to obtain a specifically optimized protein that is highly active, and can potentially be used as a drug.