Fig. 1. The IFN-g D' dimer with the 3.5 A MIR electron density.STEVEN E. EALICK, WILLIAM J. COOK, SENADHI VIJAY-KUMAR,
MIKE CARSON, TATTANAHALLI L. NAGABHUSHAN,
PAUL P. TROTTA, CHARLES E. BUGG
The x-ray crystal structure of recombinant human interferon-gamma has been determined with the use of multiple-isomorphous-replacement techniques. Interferon-gamma, which is dimeric in solution, crystallizes with two dimers related by a noncrystallographic twofold axis in the asymmetric unit. The protein is primarily a helical, with six helices in each subunit that comprise ~62 percent of the structure; there is no sheet. The dimeric structure of human interferon-gamma is stabilized by the intertwining of helices across the subunit interface with multiple intersubunit interactions.
Interferon-gamma (IFN-g) is a product of activated T lymphocytes and natural killer (NK) cells that was originally described as an antiviral agent (l). IFN-g exhibits pleiotropic biological activities (2) and specifically has been shown to regulate expression of class II major histocompatibility antigens (3) and Fc receptors (4), activate human monocyte cytotoxicity (5), enhance NK cell activity (6), and regulate immunoglobulin production and class switching (7). Expression of biological activity appears to be mediated through binding to specific cell-surface receptors (8), which have been cloned (9). The expression of human IFN-g in Escherichia coli (10, 11) has resulted in the preparation of large quantities of highly purified recombinant human IFN-g on which detailed studies of structure-function relations have been initiated (12-16).
We reported a preliminary crystallographic study of an E. coli derived recombinant form of human IFN-g, designated IFN-g D', in which the five COOH-terminal residues are deleted (17, 18). We report here the determination of the three-dimensional (3-D) structure of recombinant human IFN-g D' with the use of multiple isomorphous replacement (MIR) techniques (19, 20).
Screening for heavy-atom derivatives was done initially by film methods with synchrotron radiation and subsequently with a Nicolet X-lOOA area detector (21). Eight heavy-atom derivatives were identified, all of which contained multiple sites that made the difference Patterson maps difficult to interpret. Fortunately, one of the derivatives [l mM KAu(CN)2] contained one site of much higher relative occupancy that could be identified with certainty from the Patterson map. Phases were calculated based on the Au derivative, and the heavy-atom sites of the other derivatives were then located from cross-difference Fourier maps. Heavy-atom parameters were refined with the use of the centric data (Table l) (22); none of the eight derivatives was of high quality. The overall figure-of-merit for data to 3.5 A resolution for all eight derivatives was 0.74.
Prior to calculation of an electron density map, the phases were improved by solvent-flattening techniques with the ISIR/ISAS package of programs by Wang (23). The initial molecular envelope was determined by assuming a solvent fraction of 50%. After convergence, the overall figure-of-merit was 0.85, the average accumulated phase shift was 21°, the R factor between calculated structure factors (Fc) after map inversion and observed structure factors (F)) was 0.282, and the correlation coefficient between F, and Fc was 0.95. The electron density map calculated with these phases was quite noisy, and the noncrystallographic symmetry was poor. Therefore, the noncrystallographic twofold axes were derived by averaging the positions of the heavy atoms in the various derivatives, and the electron density was then averaged based on the noncrystallographic symmetry.
This map clearly showed the protein-solvent boundary and two dimers that were related by a noncrystallographic twofold axis. All 12 a helices in the dimer could be seen in this map. Four loop regions linking five of the helices were also visible in the map. Two regions of the polypeptide chain, the long loop between helices A and B and the last 15 residues at the COOH-terminus, appear to be disordered or highly flexible because the electron density is weak.
Preliminary refinement of a starting model (24) with the method of simulated annealing with the program XPLOR (25) gave an R factor of 0.25 with 6.0 to 2.8 data (13,192 reflections > 2 sigma; 4,072 atoms). Portions of helices B, C, and D with the associated 3.5 A MIR electron density that has been solvent flattened and symmetry averaged are shown in Fig. 1. The final model contains 123 residues and is complete except for the COOH-terminal residues 124 and 138; residues 122 to 138 probably extend away from the dimer. The root-mean-square differences between C-alpha carbons of individual subunits after refinement without the use of noncrystallographic symmetry restraints ranged from 0.8 to 1.2 A. There were no significant differences between subunits in the same dimer compared to subunits in separate dimers.
Fig. 1. The IFN-g D' dimer with the 3.5 A MIR electron density.
Recombinant human IFN-g D' is a dimer with identical subunits that are related by a noncrystallographic twofold axis (Fig. 2). The dimer is globular with overall dimensions of approximately 60 A by 40 A by 30 A. The asymmetric unit contains two dimers related by a noncrystallographic twofold axis. Thus, the overall noncrystallographic symmetry is described by the point group 222 (26).
Stereo views of the IFN-g D' dimer.
Each subunit of IFN-g contains six helices that comprise ~62% of the structure (Fig. 3). The helices range in length from 9 to 21 residues. All 12 helices in the dimer are generally parallel to the dimer twofold axis. There are no clear antiparallel four-helix domains in the molecule. Within the dimer, both parallel and antiparallel interactions are observed between adjacent helices. In addition, extensive interhelical contacts occur between helices in different subunits. The first four helices from one subunit form a cleft that accommodates the COOH-terminal helix from the other subunit (Fig. 4A). The COOH-terminal helix shows a bend in the center at residue Glu 112 with an angle between the two segments of ~125 degrees. This pronounced bend may result from the large number of contacts between the helix and the cleft. There are hydrophobic contacts among all of the helices, but most are between the C and D helices. The C helix is the most hydrophobic helix in the subunit and is essentially buried in the core of the dimer.
Fig. 3. Amino acid sequences of IFN-g
The structure of the subunit is extended and has a flattened prolate elliptical shape. However, the two subunits are intimately related, and the overall structure is compact and globular. It is difficult to see how the dimer could be separated without significant disruptions in the tertiary structures of the individual subunits. The dimer interface is centered on helix C. The two symmetry-related C helices pack against each other with an angle of ~55 degrees. Helices E and F from the other subunit flank either side of helix C. Intersubunit contacts also occur between the NH2-terminal helix of one subunit and the COOH-terminal helix of the other. There is no beta-sheet structure within the subunit or across the dimer interface.
Human IFN-g contains two Asn-X-Ser/Thr glycosylation sites in the mature protein (27). One occurs at residues 25 to 27 at the end of the long flexible loop between helices A and B. The other occurs at residues 97 to 99 at the end of helix E. Both of these regions are on the surface of the molecule and are exposed to the solvent.
The dimeric structure of IFN-g D' is consistent with sedimentation equilibrium experiments which show that a similar form of human IFN-g is dimeric in solution (28,29), rather than trimeric or tetrameric (30). Unlike the E. coli-derived material described here, native human IFN-g contains N-linked carbohydrate and is also heterogeneous at the COOH-terminus, exhibiting variable degrees of COOH-terminal processing (27). However, since the absence of carbohydrate does not significantly alter the conformation of human IFN-g, and because the removal of up to 13 amino acid residues from the (COOH-terminus does not affect either self-association or conformation (14), the dimeric structure and high-helical content noted for recombinant human IFN-g D' are probably characteristic of native human IFN-g. The dimeric structure of IFN-g D' is unusual among globular proteins in the way in which the subunits are so intimately linked. This type of intertwining between subunits has been reported for Trp repressor, which is also a dimer in which the subunits are primarily alpha-helical (31).
Based on the current assignments of secondary structure, human IFN-g D' contains 62% alpha-helix and no beta-sheet These data agree with analyses of far-ultraviolet circular dichroic spectra that estimate an (alpha-helix content of 40 to 66% and a low percentage of beta-sheet (12,29,32). However, IFN-g D' has no apparent homology with any previously reported alpha-helical proteins. On the basis of the location of the exon boundaries, loops are predicted to occur at or near residues 15, 38, and 99 (33), which agrees well with the observed secondary structure assignments (Fig. 3).
Residues at both the NH2- and COOH-termini of human IFN-g may be critical in the interaction with the receptor or triggering of biological response or both. For example, forms of recombinant human IFN-g truncated at the COOH-terminus demonstrate substantially reduced antiviral activity (13-16), and antibodies directed to both the NH2-terminus (34, 35) and COOH-terminus (15,35) neutralize in vitro bioactivity. We conclude that the receptor-binding region requires an intact dimer and may include the COOH-terminus of one subunit and the NH2-terminus of the other.
Truncation at the NH2-terminus appears to result in a dramatic loss in secondary structure (12). The NH2-terminus of one subunit has multiple contacts with helix D of the same subunit and helix F of the other subunit, and hence this region is necessary for maintenance of the overall structure. In addition, the NH2-terminal helix of one subunit is required to maintain the cleft that accommodates the COOH-terminal helix of the other subunit.
Comparison of the amino acid sequences for IFN-g from different mammalian species shows considerable homology (Fig. 3). Based on this homology, we expect that these IFN-g's would have similar tertiary structures. Three of the most highly conserved regions in the sequences occur in helices C and F, which are the two most buried helices in the dimer, and a short basic stretch at the beginning of the COOH-terminal tail. Deletion mutants of human IFN-g truncated at the COOH-terminus begin to show loss of activity when residues in this region are removed (13-16). One of the most variable regions is the loop between helices A and B. This loop may account for the high species specificity of IFN-g; that is, IFN-g from one species generally displays poor affinity for the receptor from another species.
To date, x-ray structures of three monomeric alpha-helical cytokines, growth hormone (36), interleukin-2 (37), and interferon-beta (IFN-b) (38), have been reported. As a dimeric cytokine IFN-g represents a new structural class. Even so, comparison of the 3-D structures of IFN-g and IFN-b reveals a striking similarity in folding topology (Fig. 4). Five of the 12 helices in the IFN-g dimer (A, C, D, E', and F') form a structural domain that corresponds to the five helices of the IFN-b molecule. In addition, the short helix B of IFN-g corresponds to a kinked region of IFN-b (B*) that displays some helical structure. Although the helices of IFN-b are somewhat more parallel to each other than are the helices of IFN-g, the overall geometrical arrangement and connectivity are conserved. A long loop between helices D and E facilitates the monomeric structure of IFN-b, whereas in IFN-g this loop is replaced by a tight turn that causes helices E and F to extend into the second subunit. This striking similarity in the 3-D structures of IFN-b and IFN-g provides evidence for a common ancestral gene, even though no significant homology exists between their amino acid sequences and each molecule has a separate cellular receptor and biological function.
Fig. 4. Schematic drawings of IFN-gamma and IFN-beta