Identification of Critical Residues in Bovine IFNAR-1 Responsible for Interferon Binding*

Interferons have antiviral, antigrowth and immuno-modulatory effects. The human type I interferons, IFN- a , IFN- b , and IFN- v , induce somewhat different cellular effects but act through a common receptor complex, IFNAR, composed of subunits IFNAR-1 and IF-NAR-2. Human IFNAR-2 binds all type I IFNs but with lower affinity and different specificity than the IFNAR complex. Human IFNAR-1 has low intrinsic binding of human IFNs but strongly affects the affinity and differential ligand specificity of the IFNAR complex. Under-standing IFNAR-1 interactions with the interferons is critical to elucidating the differential ligand specificity and activation by type I IFNs. However, studies of ligand interactions with human IFNAR-1 are compromised by its low affinity. The homologous bovine IFNAR-1 seren-dipitously binds human IFN- a s with nanomolar affinity. Exploiting its strong binding of human IFN- a 2, we have identified residues important for ligand binding. Mutagenesis of any of five aromatic residues of bovine IF-NAR-1 caused strong decreases in ligand binding, whereas mutagenesis of proximal neutral or charged residues had smaller effects. These residues were mapped onto a homology model of IFNAR-1 to identify the ligand-binding face of IFNAR-1, which is consistent with previous structure/function studies of human IF-NAR-1. The topology of IFNAR-1/IFN interactions appears novel when compared with previously studied cytokine receptors. The type I interferon of BoIFNAR-1, cysteines IFNAR-1 reasonable formation of four disulfide bonds homologous individual b -strands of IFNAR-1 IFNGR-1. the predicted -sheet quite some as to the precise or orientation of in the

The low intrinsic affinity of HuIFNAR-1 for IFNs has hampered studies seeking to identify residues involved in ligand binding and specificity. Previously, the identity of elements of the ligand binding site of HuIFNAR-1 could only be deduced indirectly from studies involving antibody epitope mapping (40 -42) or homology modeling based on other cytokine receptors (40,43). Without large-scale mutagenesis and ligand binding analysis, key residues could not be identified in HuIFNAR-1.
The bovine IFNAR-1 homologue is an attractive target for mutagenesis and analysis of the IFN binding site. Although the type I interferons are predominantly species-specific, human interferons display uniformly high binding and biological activity on bovine cells (3, 44 -46). This appears to reflect the ability of bovine IFNAR-1 (BoIFNAR-1) to bind human type I IFNs with moderately high affinity. Thus, human or murine cells expressing BoIFNAR-1 greatly increase their responsiveness to a variety of human type I IFNs (18,47,48). The nanomolar binding affinity of BoIFNAR-1 for HuIFN-␣s provides an elegant way to circumvent difficulties in the studies of the human IFN type I receptor complex (47)(48)(49)(50).
BoIFNAR-1 cDNAs were cloned by independent laboratories (47,48) and found to encode a transmembrane protein of 560 amino acids. The protein consists of a 24-amino acid signal sequence, a 414-amino acid extracellular domain, a 24-amino acid transmembrane sequence, and a 98-amino acid cytoplasmic domain. The human and murine IFNAR-1 proteins share 68 and 46% amino acid sequence identity, respectively, with BoIFNAR-1 (6,48,51). This high sequence identity, combined with an overall structural conservation, has allowed us to use BoIFNAR-1 in place of HuIFNAR-1 in studies directed at identifying structural features contributing to HuIFN ligand binding.
The ability of BoIFNAR-1 to bind HuIFNs in the nanomolar range has been well characterized using a variety of systems (18,23). COS-1 cells transfected with BoIFNAR-1 cDNA express very high levels (0.5-1.0 ϫ 10 6 /cell) of nanomolar affinity binding sites for HuIFN-␣2a, -␣8b, -␣1b, -␤, and - (11,48,49). Consistent with these results, a soluble BoIFNAR-1/Fc fusion protein bound HuIFN-␣2a with an affinity of at most 10 nM (the true affinity is expected to be closer to a K i of 0.14 nM (23)). All human type I IFNs tested were able to compete with HuIFN-␣2a for binding to BoIFNAR-1/Fc, although they varied in affinity. Thus, the extracellular domain of BoIFNAR-1 by itself displays moderate affinity and differential binding of a broad range of human type I IFNs.
Previously, we used BoIFNAR-1 to regionally localize the determinants that confer strong binding of IFN (49). A series of 14 HuIFNAR-1/BoIFNAR-1 chimeric receptors representing various human/bovine subdomain substitutions were assayed for their ligand binding properties. Only when the two central domains of BoIFNAR-1 (SD2 and SD3) were simultaneously substituted into the HuIFNAR-1 was a significant increase in the binding of HuIFN-␣2a measured over the low affinity binding characteristic of HuIFNAR-1, indicating that ligand binding was focused in this region. The BoIFNAR-1 N-terminal (SD1) and membrane-proximal (SD4) domains each further enhanced binding when transferred with SD2 and SD3 into HuIFNAR-1 (49). Thus, the IFN binding site is not localized on one FNIII subdomain or even on one tandem pair of FNIII subdomains, as previously predicted from other cytokine family members (43); instead the ligand-binding determinants of IF-NAR-1 appear to be distributed in a more complex array centered on SD2 and SD3. Kumaran et al. (52) used murine/human IFNAR-1 hybrids to confirm that SD1 plays at most a minor role in species-specific ligand binding.
The current study of BoIFNAR-1 identifies amino acids that are critical for IFN binding, using a series of alanine substitutions throughout its extracellular domain. By presenting these results in the context of a new three-dimensional homology model of IFNAR-1, we shed light on the interaction of type I IFNs with IFNAR-1.
Phosphorylation of IFN-The HuIFN-␣2a analogue IFN-␣2a-P1 (53) was phosphorylated to a radiospecific activity of ϳ1 ϫ 10 3 Ci/mmol with [␥-32 P]ATP (6000 Ci/mmol, PerkinElmer Life Sciences) and bovine heart cAMP-dependent protein kinase (Sigma). This corresponds to a fractional labeling of ϳ0.15 mol of 32  Parental Plasmids and Expression-Bovine and Human IFNAR-1 were transiently expressed from the EF-1␣ promoter in the vector pcDEF3 (54). Unique restriction sites were engineered into the bovine IFNAR-1 cDNA by using oligonucleotides containing silent mutations RsrII, BspEI, AgeI, and NheI. These restriction sites were placed be-tween each respective subdomain and at the beginning of the transmembrane domain, resulting in a modified version of BoIFNAR-1. This version of BoIFNAR-1 was also tagged with the FLAG epitope (55) between the signal peptide and the beginning of the first subdomain of the protein, allowing for its recognition on the cell surface.
Mutagenesis-Mutations were engineered into the receptor using either a two-step splice overlap extension (SOE) polymerase chain reaction method (56,57) or the QuikChange site-directed mutagenesis kit (Stratagene). All clones were screened by restriction digests and confirmed by sequencing. In most cases altered receptors were analyzed using two independent clones, to eliminate false decreases in binding due to reasons other than the engineered mutation. Mutants were made from the BoIFNAR-1 cDNA template in three different versions of pcDEF-3 at various stages in this project. The BoIFNAR-1 gene was the same for all three; however, the first had an HA tag at the N terminus (which was not effective for flow cytometric detection); in the second, the HA tag was replaced with a FLAG epitope to allow for cell surface confirmation by flow cytometry; the third version contained an internal ribosome entry site coupled to enhanced green fluorescent protein (IRES-EGFP) (CLONTECH) after the FLAG-bovine IFNAR-1, and a zeocin marker in place of the original neomycin marker. These plasmids, when transfected into COS cells, were indistinguishable in their transient BoIFNAR-1 expression and IFN binding.
The two-step Splice Overlap Extension (SOE) polymerase chain reaction method (56,57) was carried out using ELongase (Life Technologies, Inc.). SOE fragments were restriction-digested and then ligated into the proper vector. All reactions were carried out in a 50-l volume in 0.5-ml thin-wall tubes in a PerkinElmer Life Sciences DNA thermal cycler. The QuikChange site-directed mutagenesis kit was used according to the manufacturer's directions (Stratagene), except that template and primer concentrations were doubled; extension time was increased to 3.5 min per kilobase; and DH10B Electromax competent cells (Life Technologies, Inc.) were used for transformations.
DNA Transfections-COS-1 cells (58), derived from a simian kidney CV-1 line, were transfected with 5-10 g of plasmid using the DEAEdextran/Me 2 SO shock protocol (48,59,60). Briefly, tissue culture dishes (10-cm dish, Falcon) containing 10 ml of 10% cosmic calf (Hyclone) serum-supplemented DMEM were seeded with 1.75-2.0 ϫ 10 6 cells. Cells were incubated overnight at 37°C before transfecting. Cells were harvested for assays after 48 -72 h. Under our transfection conditions, we generally see about 0.25-1 ϫ 10 6 receptors per cell, averaged over the entire cell population, as measured by ligand binding to BoIF-NAR-1. This arises from high level expression on 20 -50% of the cell population, as demonstrated by flow cytometry detection of the common FLAG epitope at the N terminus. Much of this expression variability is between experiments; within any experiment, the variation of expression between constructs has a much smaller range. Each clone was transfected at least two times, and for most mutations binding was measured with transfections of two independent DNA clones.
Saturation Binding-Saturation binding assays were done as previously described (48). Briefly, COS cells were trypsinized and resuspended to 1 ϫ 10 6 cells/ml. Aliquots of cells, with and without excess cold IFN (1-3 g/ml), were combined with [ 32 P]IFN-␣2 (maximum concentration of 1.5-4 ϫ 10 Ϫ9 M), serially diluted, and then incubated while rocking for 1 h at room temperature. Cell-bound [ 32 P]IFN-␣2 was separated from unbound by brief centrifugation of 100 l of sample through a cushion of 10% (w/v) sucrose in PBS. Tubes were then frozen and cut, and the tips and tops were counted separately. Suppliers' determinations of IFN concentration (in mg/ml) were used for all calculations. Data were analyzed by non-linear regression to one-site binding using the program Prism v.2.01 (GraphPad Software, Inc., San Diego, CA). A mutant receptor that lost 75% of the binding seen for unaltered BoIFNAR-1 was defined as having a significant loss of binding activity. Among the mutants, there is a clear demarcation between those mutants that lost 75-100% and a secondary group of mutants, which lost 0 -56% of binding. This criterion is consistent with thresholds used in studies of other receptors (61)(62)(63) and with the accuracy and precision of our data, thereby guarding against over-interpretation of marginal effects.
Flow Cytometric Analysis of Receptor Surface Expression-Transfected cells were resuspended in DMEM supplemented with 5% cosmic calf serum (Life Technologies, Inc.). Cells (7.5 ϫ 10 5 to 1.25 ϫ 10 6 in 25 l) were incubated for 1 h at 4°C with 25 l (0.12-0.25 g) of primary antibody (M2 anti-FLAG antibody, Sigma) or medium. The cells were washed with PBS and resuspended in 50 l (0.125 g) of secondary antibody (R-phycoerythrin-conjugated F(abЈ) 2 goat anti-mouse IgG, Jackson ImmunoResearch) and incubated for 1 h. Cells were again washed with PBS and then incubated in 100 l of 3% paraformaldehyde at 4°C for 1 h. The cells were washed once with 1 ml of PBS containing 50 mM Tris and then resuspended in 500 l of PBS with 50 mM Tris and stored at 4°C until analysis with a Coulter Epics Profile II cell sorter.
Model Building-Homology models were generated for the extracellular domains of bovine and human IFNAR-1, based on the coordinates of the extracellular domain of IFNGR-1 (30,70) graciously provided by Drs. Mark Walter (University of Alabama, Birmingham) and Steve Ealick (Cornell University). The homology model was created using GeneMine (LOOK v.3, Molecular Applications Group) and manipulated with GeneMine and Sybyl (Tripos, Inc.).
The IFNAR-1 (4-domain receptor) was modeled against IFNGR-1 (2-domain receptor) in two steps. Because of the size disparity, it was necessary to separately model the N-terminal half (SD1 and SD2) and the C-terminal half (SD3 and SD4) of the receptor. This procedure is consistent with the finding that the four domains of IFNAR-1 likely evolved from a tandem duplication of a two-domain ancestor, which diverged to generate the cytokine receptor superfamily (27,38,39).
The sequence alignment of the N-and C-terminal halves of bovine and human IFNAR-1 with HuIFNGR-1 is shown in Fig. 1, with ␤-strands from the crystal structure of IFNGR-1 indicated (underlined in the sequence) and sequentially labeled as B1-B7 within a FNIII domain. The alignments are facilitated by the similar sequence lengths and conserved characteristic residues of IFNGR-1 and the N-and C-terminal halves of IFNAR-1. Sequence alignments were similar with GCG version 9.0 and with ClustalW. (Expected loops are designated by the flanking ␤-strands: e.g. the loop connecting ␤-strands 4 and 5 is denoted "L4 -5.") As commonly observed within families of homologous proteins, the "core" ␤-sheet structure is well conserved, and much of the variability arises in the interstrand loops, particularly the loops L2-3 and L4 -5 in SD2 and SD4 of IFNAR-1. In our homology model of BoIFNAR-1, the positions of the cysteines in the IFNAR-1 are within a reasonable distance for the formation of four disulfide bonds homologous to those found in IFNGR-1. Similarly, individual ␤-strands of IFNAR-1 aligned closely with those of IFNGR-1. Thus, the predicted core ␤-sheet should be quite reliable, with some ambiguity as to the precise location or orientation of residues in the loops.

RESULTS
General Considerations-With 414 amino acids predicted for its extracellular domain, BoIFNAR-1 presented a large target for mutagenesis. We therefore created clusters of alanine mutations first, followed by individual alanine substitutions. Initially we investigated loops whose amino acid sequences differ between BoIFNAR-1 and HuIFNAR-1, because their affinities for IFNs differ by Ͼ100-fold and because residues implicated in ligand binding are most often located in the loops of the cyto-kine receptors, rather than in the ␤-strands (31,33,62,64,65). The search for functionally important residues was aided by information about ligand/receptor interactions in other members of the cytokine receptor families, and by our three-dimensional homology model (27, 30, 37-39, 61, 66).
Transfected COS cells transiently expressing the receptors on the cell surface at very high levels serve as a reliable, efficient, and convenient platform for ligand binding experiments. With such high expression, the binding of IFN to BoIF-NAR-1-transfected cells increases 15-to 40-fold over mocktransfected cells (48), providing a very strong signal. In the same system, surface expression of HuIFNAR-1, at levels comparable to that of BoIFNAR-1, increases binding of [ 32 P]IFN-␣2 to cells by 1-to 2-fold over COS cells transfected with empty vector (Table I; see also Ref. 49,and Ref. 48). Thus, any excess endogenous (simian) IFNAR-2 does not make a substantial contribution to ligand binding. Cell surface expression levels of different constructs were compared through a common FLAG epitope. We thereby demonstrated that high binding and low binding IFNAR-1 variants express at comparable levels (Table I). Previously we had shown for both BoIF-NAR-1 and HuIFNAR-2 that the affinity and ligand specificity of the receptors expressed on COS cells is essentially the same as their affinity and specificity when produced as purified fusion proteins of the receptor extracellular domain with an IgG Fc domain (11,23). 2 Thus, the low background IFN binding contributed by endogenous COS cell receptors (Ͻ5000 receptors/cell) does not significantly contribute to the observed high levels of IFN binding, which is attributable to the high level expression of ectopic BoIFNAR-1. However, because the Simian-derived COS cells are biochemically responsive to human IFNs through their endogenous receptors, this system is useful only for ligand binding assays. Finally, the expressed receptors on the cell surface are likely to be correctly folded, because misfolded proteins in eukaryotic cells are generally not efficiently presented on the cell surface; thus, questions about protein misfolding, common with bacterial expression systems, are largely mitigated when the proteins are demonstrated to 2 J. A. Langer, unpublished data. express efficiently on a eukaryotic cell surface.
Mutagenesis-Previous work had emphasized the importance of subdomains 2 and 3 in generating the nanomolar affinity of BoIFNAR-1 (49), so our mutagenesis focused on these subdomains. In addition, a region corresponding to a well-characterized epitope of SD1 was examined in detail, as were several areas of SD4 whose sequences differ significantly between the bovine and human homologues. Finally, differences between human and bovine IFNAR-1 were investigated with several substitutions incorporating the amino acids found in HuIFNAR-1 into the BoIFNAR-1 structure ("homologue mutants"). The list of mutants (Tables I and II) is organized by subdomain, and denoted by the secondary structure location (e.g. 2L4 -5 is a mutation in SD2 in the loop Mutants are arranged by subdomain. Column 1 has the designation of mutants based on their originally predicted secondary structure. Residue numbers in Column 2 are those of the initial amino acid when a cluster was mutated. Numbering is based on the mature protein. Results are tabulated as the percent decrease from the level of IFN binding by BoIFNAR-1 (normalized within each experiment), with the standard deviation (S.D.) for the total number of experiments (N; in parentheses). The last column presents cell surface expression based on flow cytometry using the M2 anti-FLAG antibody for N-terminally FLAG-tagged constructs. This is presented as the percent of cells with fluorescence that is increased by Ͼ0.5 logarithms above background (mock-transfected) levels, with the expression of wild-type BoIFNAR-1 within each experiment defined as 100%. Receptors without the FLAG epitope were not analyzed by flow (NT). Asterisks indicate mutants with increases in binding of 60 -100% over control BoIFNAR-1.

Interferon Ligand-binding Determinants of Bovine IFNAR-1
connecting ␤-strands 4 and 5; the suffix "H" refers to human homologue mutants). Mutagenesis resulted in altered receptors whose abilities to bind HuIFN-␣2 ranged from the high levels equivalent to (or occasionally higher than) parental BoIFNAR-1, to receptors whose binding was low and indistinguishable from COS background (Table I, Fig. 2). However, all receptors were present on the cell surface at levels similar to parental (70 -140% of BoIF-NAR-1), as monitored by flow cytometric detection of a common FLAG epitope (55) at the N terminus (the FLAG epitope was previously shown to have no effect on ligand binding (49)). The expression of BoIFNAR-1 gave very high levels of ligand binding as shown previously; unfortunately, like parental BoIF-NAR-1, many mutants with strong binding did not approach saturation so that K d values could not be accurately estimated. Therefore, results are compared as the percent binding relative to the unaltered BoIFNAR-1 control at the maximal IFN input (1.3-4 nM). For comparison, mock-transfected and HuIFNAR-1transfected COS cells were shown to bind an average of 3 and 6% of the level of COS cells transfected with BoIFNAR-1 (Table I).
For clarity of interpretation, and consistent with mutagenesis studies of other cytokine receptors (33,34,65), we focused on those mutants that showed decreases in ligand binding of Ն75% while retaining high cell surface expression. Mutants with low binding of IFN were consistently expressed at levels comparable to parental BoIFNAR-1 and other high binding mutants (Table I).
SD3 was examined with a series of 12 cluster mutants, 3 single alanine mutants, and 3 human substitutions. The majority of mutations showed minimal changes in ligand binding (0 -25% decrease; see Table I). Three alanine clusters, 3B6, 3L3-4:3, and 3B7, resulted in drastic decreases of 76 -91%, as did the single point mutation 3L3-4W (Trp 253 ) ( Because our previous studies implicated SD1 as having a lesser effect on ligand binding than SD2 and SD3 (49), a full study of this domain was not undertaken. A more focused series of mutants came from the finding that the epitope for the anti-human IFNAR-1-neutralizing monoclonal antibody 64G12 (67) maps to the human IFNAR-1 sequence 62 FSSLKLNVY 70 in SD1 (sequence numbering of the mature protein; Fig. 1) (41,42). We therefore explored the effect of mutations in the equivalent BoIFNAR-1 sequence 62 FSSVELENVF 71 , focusing on the large aromatic and charged residues (Table I). Alanine substitutions for Phe 62 or Val 65 actually produced an increase in ligand binding (Table I). A single alanine substitution at Phe 71 (1L5-6F2) caused a moderate decrease of ligand binding. A slightly larger decrease (57%) was seen for the 64G12 epitope cluster mutant 1L5-6, where the ELEN sequence is substituted by four alanines. Because of our interest in this region, we also explored the charged and large aromatic residues in loop 3-4, which in our three-dimensional homology model is proximal to the 64G12 epitope. Substitution of alanine for Asp 44 or for Trp 46 (mutants 1L3-4D and 1L3-4W) did not dramatically decrease IFN binding. Thus, mutants in and near the bovine homologue of the epitope for mAb 64G12 in SD1 showed moderate but not strong effects on IFN-␣2 binding.
In SD4, seven mutants were generated, based on differences between the bovine and human sequences. This series carried the N-terminal HA epitope tag, which was not accessible by flow cytometry, compromising the ability to quantitate surface expression. However, mutants 4L2-3, 4L2-3V, and 4L6 -7:2 produced no decrease in IFN binding, and 4L4 -5 showed some increase in ligand binding. Therefore, these mutants do not require independent estimates of surface expression. Mutants 4L1-2, 4B5:1, and 4B5:2 showed moderate decreases of binding (about 50%). Because of the lack of reliable expression data, the precise interpretation of these results is uncertain. Nevertheless, it can be said that none of these mutants in SD4 strongly (4-fold) decreases ligand binding.
Thus, the alanine-scanning mutagenesis identified several large hydrophobic residues in SD2 and SD3 of BoIFNAR-1 whose substitution to alanine dramatically decreased the binding of IFN-␣2. All of these residues are conserved in the human and murine IFNAR-1. This was surprising, because the initial experiments assumed that differences in ligand binding between human and BoIFNAR-1 would reflect a small number of residues that differed between them. We therefore examined some loops in SD2 and SD3 with significant sequence differ- ences between HuIFNAR-1 and BoIFNAR-1, by substituting clusters of residues found in HuIFNAR-1 for the homologous residues in BoIFNAR-1 (Table II; Fig. 4). The human homologue mutants in SD2 (identified by the suffix "H"), 2L4 -5H and 2L6 -7H, caused small or insignificant decreases in ligand binding (31 and 13%, respectively). However, when both of these homologous substitutions were engineered into the same mutant (2L4 -5/6 -7H), the binding of IFN-␣2 was significantly decreased by 70% of that seen with BoIFNAR-1 while retaining expression equivalent to the BoIFNAR-1 control. Similarly in SD3, the homologue cluster mutations 3L3-4:2H and 3L5-6: 2H, produced decreases of 21 and 0%, respectively, whereas the combination of both homologous substitutions produced a larger, but still modest, decrease in binding of about 48%, slightly more that seen with 3L3-4:2H alone. However, when combinations of three human cluster substitutions ("Bo3H", composed of 2L6 -7H, 3L3-4:2H and 3L5-6:2H) or all four human clusters ("Bo4H") were placed in SD2 and SD3, these multiple mutants decreased ligand binding by a minimum of 75% of the original binding. Thus, the substitution of homologous amino acids found in HuIFNAR-1 in any one of these loops produced small or insignificant decreases in ligand binding, whereas several combinations of changes led to very strong decreases in ligand binding, approaching a level similar to HuIFNAR-1.

DISCUSSION
BoIFNAR-1, whether as a purified BoIFNAR-1/Fc fusion protein or when expressed at high levels on COS cells, binds HuIFN-␣2 with 1-10 nM affinity, and binds a broad selection of other type I IFNs (23). Based on subdomain substitutions between the human and bovine homologues, this strong binding, relative to the weak ligand binding by human IFNAR-1, was shown to reside predominantly in the two central subdomains, SD2 and SD3 (49). Here, through a series of alanine and homologue substitution mutations, we have identified residues in SD2 and SD3 of BoIFNAR-1 that strongly affect the binding of HuIFN-␣2. The strongest decreases in binding came from the substitution of any of five hydrophobic residues, conserved in bovine, human, and murine IFNAR-1. Smaller effects came from mutating charged residues proximal to those hydrophobics.
The mutagenesis results are best interpreted in the context of our three-dimensional homology model of IFNAR-1 derived from the atomic coordinates of the closely related, two-subdo-main IFNGR-1 3 (70) (Fig. 5). As there is no related crystal structure on which to base the interaction of the two separate domains of IFNAR-1 (specifically, the SD2-SD3 junction), we have chosen to illustrate the four subdomains in an extended array (rather than, for instance, a semi-circular or strongly bent arrangement). To orient the two halves of IFNAR-1 (SD1ϩSD2 and SD3ϩSD4) to each other, we logically assume that residues of SD2 and SD3, which strongly affect ligand binding, are on the same face of the molecule.
Most mutations resulted in little significant change in binding of IFN-␣2 (Table I; Fig. 5, blue). Where cluster mutations showed strong decreases in binding, the localization was generally refined to a single residue, which decreased IFN binding by 80% of native BoIFNAR-1 (red). Several clusters, which produced large decreases in IFN binding, were not refined to single residues (orange).
Alanine mutagenesis identified five hydrophobic residues that individually decreased the binding of HuIFN-␣2 by 82-96% while retaining cell surface expression levels of 69 -125% of parental receptor. The homology model presents the functionally important aromatic residues Trp 132 , Phe 139 , Tyr 160 , and Trp 253 as well exposed on the receptor surface, easily accessible for protein interactions. The three residues in SD2 (Trp 132 , Phe 139 , Tyr 160 ) are localized to a common region. Trp 253 , in SD3, is on the protein surface and can be aligned in 3   this extended view with the residues of SD2 to define a common binding surface for IFNs. In contrast, Tyr 141 appears buried beneath Phe 139 and Tyr 160 , forming extensive ␤-strand contacts within SD2. Therefore Tyr 141 , although not directly involved in ligand binding, is likely to be required for the proper presentation of Phe 139 and Tyr 160 . Each of these critical hydrophobic residues is conserved in the human and murine IF-NAR-1 homologues. Two other alanine clusters involving residues 113 EDK 115 (2L1-2) and 157 ETV 159 (2L4 -5:2) produced decreases of IFN-␣2 binding of 90% (orange) (Fig. 5). 157 ETV 159 (ENI in HuIF-NAR-1 and INS in murine) is proximal (4 -14 Å) to both Phe 139 and Tyr 160 and could be involved in either positioning them or in direct ligand interactions. The highly charged 113 EDK 115 cluster is near the interface with SD3 and is conserved in human and murine IFNAR-1. This could be involved in important contacts with SD3 or with IFNs. Mapping within these clusters was not further refined.
The conservation of the critical large hydrophobic residues in bovine, human, and murine IFNAR-1 was surprising, because the strategy originated in our studies of bovine/human IF-NAR-1 chimeras and had presumed that the key residues would be those that differ between bovine and human IF-NAR-1. However, like IFNAR-1, IFNGR-1 and growth hormone receptor demonstrate the importance of aromatic residues (predominantly W and Y) combined with charged residues (K, R, E, and D) packing against the critical aromatics in forming a binding interface (65,67,68). In BoIFNAR-1, the aromatic/ charged pairs found appear to be Asp 128 /Trp 132 , Arg 136 /Phe 139 / Arg 140 , Glu 157 /Tyr 160 /Glu 162 /Asp 163 , Phe 241 /Lys 243 , and Lys 252 / Trp 253 /Lys 254 . Therefore, we hypothesize that the conserved aromatic residues are the foci of the receptor/ligand interface and that the clusters of residues surrounding the key hydrophobic residues modulate ligand binding and distinguish the low affinity human from the high affinity bovine IFNAR-1. These surrounding residues may also help differentiate the binding affinities of the diverse type I ligands.
This hypothesis led us to substitute homologous human residues, which were predicted from the three-dimensional model to be proximal to the conserved critical hydrophobic residues. Substituting individual clusters of human residues (2L4 -5H or 2L6 -7H, 3L3-4:2H or 3L5-6:2H) for their homologous bovine residues produced minimal or no changes in IFN-␣2 binding (Table II). However, various combinations of two to four homology clusters produced a cumulative decrease of 50 -80% in ligand binding. These effects are consistent with our hypothesis that residues surrounding the critical aromatic residues may, when taken together, be responsible for the differences between human and bovine affinities.
Because of its low affinity for IFN, HuIFNAR-1 is a poor target for structure/function studies. What little is known about the ligand binding site of HuIFNAR-1 has been derived Red, single alanine substitutions of BoIFNAR-1 producing decreases of Ն75%. Yellow, orange, and green, residues within the epitope. Yellow, not mutated in BoIFNAR-1; orange, mutated singly or as a cluster in BoIFNAR-1 with some decrease of ligand binding (see Table I); green, mutated singly or as a cluster in BoIFNAR-1 with no significant decrease in IFN binding. from indirect studies. Two neutralizing mAbs against HuIF-NAR-1, EA12 (50) and 4A7 (40), seem to recognize epitopes that are difficult to map to a single domain. The epitopes for two other neutralizing anti-HuIFNAR-1 mAbs have been identified. The high affinity mAb 2E1 appears to recognize epitopes within the non-contiguous sequences 244 HLYKWK 249 and 291 EEIKFDTE 298 , with strong contributions to antibody binding attributed to Lys 249 , Glu 291 , and Asp 296 (40). This epitope includes Trp 248 , the human homologue of our critical bovine Trp 253 (Fig. 6). The epitope for the anti-human IFNAR-1 64G12 mAb (67) that neutralizes the activity of all tested human type I IFNs on human cells has been localized to a sequence in SD1, 62 FSSLKLNVY 70 (41,42). However, our mutations in the homologous sequence ( 62 FSSVELENVF 71 ) in BoIFNAR-1 produced only insignificant to moderate effects on binding of IFN-␣2 (Fig. 6). This is consistent with our observations and those of Kumaran et al. that SD1 plays at most a minor role in the species specificity of human versus murine IFN binding (49,52). Because the three-dimensional model places the 64G12 epitope within 9 -17 Å of the critical Trp 132 residue in SD2, we speculate that mAb 64G12 sterically occludes Trp 132 and any neighboring residues involved in IFN binding.
On the three-dimensional IFNAR-1 model, the epitopes for these two independently produced neutralizing mAbs (40 -42, 50, 67) are proximal to and located on the same face of IFNAR-1 as the residues that we identify in BoIFNAR-1 as being critical to the binding of IFN-␣2; i.e. in both BoIFNAR-1 and HuIF-NAR-1 there is one face of IFNAR-1 that is involved in ligand binding. Thus, our mutagenesis study of BoIFNAR-1 significantly enhances the interpretation of the currently available studies of HuIFNAR-1 and provides focus for future studies of HuIFNAR-1.
In conclusion, in BoIFNAR-1 we have identified five large hydrophobic residues, conserved in the human and murine homologues, whose substitution with alanine decreases ligand binding by 82-96%. Several clusters of amino acids, proximal to the important aromatic residues, have additionally been identified whose coordinated substitution by alanine produces large decreases in binding of human IFN-␣2. The critical residues of BoIFNAR-1 map to the same face and are proximal to amino acids independently identified as part of epitopes for anti-HuIFNAR-1 mAbs that efficiently neutralize IFN activity and block IFN binding, supporting the relevance of the current results to the human IFNAR-1.
The most straightforward interpretation of our binding, domain swapping, and alanine mutagenesis data (23,48,49) is that the single nanomolar affinity site for HuIFN-␣2 on BoIF-NAR-1 primarily, but not exclusively, involves ligand interactions with SD2 and SD3. The finding that ligand binding appears focused in SD2 and SD3 is novel in terms of previous models (39,43,69) and presents a unique pattern of ligandreceptor interactions for the cytokine superfamily. Although confirmation of any model must await a crystal structure, this extensive analysis of BoIFNAR-1 enables the focused analysis of the human IFNAR-1.