View metadata, citation and similar papers at core.ac.uk brought to you by CORE

provided by Elsevier - Publisher Connector

Research Article 863

Crystal structure and functional dissection of the cytostatic oncostatin M Marc C Deller1,2†#, Keith R Hudson2‡#, Shinji Ikemizu1, Jerónimo Bravo1§, E Yvonne Jones1* and John K Heath2

Background: The cytokine oncostatin M (OSM) inhibits growth of Addresses: 1Cancer Research Campaign Receptor certain tumour-derived cells, induces proliferation in other cell types Structure Group, Division of Structural Biology, The Wellcome Trust Centre for Human Genetics, (e.g. haemangioblasts) and is a mediator of inflammatory responses. Its University of Oxford, Roosevelt Drive, Oxford OX3 mechanism of action is via specific binding to gp130 and either the leukaemia 7BN, UK and 2Cancer Research Campaign Growth inhibitory factor receptor (LIFR) or (OSMR) systems Factors Group, School of Biochemistry, University of at the cell surface to form an active signalling complex. Birmingham, Edgbaston, Birmingham B15 2TT, UK. Present addresses: †Yale University School of Results: We report here the crystal structure of human oncostatin M (hOSM) Medicine, Department of Pharmacology, Sterling along with mutagenesis data which map the receptor-binding epitopes of the Hall of Medicine, PO Box 208066, New Haven, molecule. The structure was determined to a resolution of 2.2 Å and conforms CT 06520-8066, USA, ‡Genesis Research and to the haematopoietin cytokine up-up-down-down four- topology. Development, 1 Fox Street, Parnell, Auckland, New Zealand and §Laboratory of Molecular Biology, The site 2 epitope, responsible for gp130 binding, is centred around Gly120 Medical Research Council Centre, Hills Road, which forms a ‘dimple’ on the surface of the molecule located on helices A and Cambridge CB2 2QH, UK. C. The site 3 motif, responsible for LIFR and OSMR binding, consists of a protruding Phe160/Lys163 pair located at the start of helix D. *Corresponding author. E-mail: [email protected]

Conclusions: The data presented allow functional dissection of the #These authors contributed equally to this work. receptor-binding interfaces to atomic resolution. Modelling suggests that the gp130 residue Phe169 packs into the site 2 dimple in an analogous fashion Key words: four-helix bundle cytokine, gp130 binding, oncostatin M, structure–function to structurally equivalent residues at the growth hormone–growth hormone receptor interface, implying that certain key features may underlie recognition Received: 26 April 2000 across the whole cytokine/receptor superfamily. Conversely, detailed Revisions requested: 7 June 2000 comparison of the available structures suggests that variations on a common Revisions received: 22 June 2000 Accepted: 28 June 2000 theme dictate the specificity of receptor–ligand interactions within the gp130 family of . Published: 20 July 2000

Structure 2000, 8:863–874

0969-2126/00/$ – see front matter © 2000 Elsevier Science Ltd. All rights reserved.

Introduction hOSM is a member of the family of gp130 cytokines that Human oncostatin M (hOSM) is a secreted 252 are characterised by the use of the shared transmembrane polypeptide cytokine that was originally isolated by virtue transducing receptor gp130 in their mechanism of action of its ability to suppress the proliferation of human [11]. hOSM is therefore functionally related to leukemia cells in vitro [1]. Subsequent biological investiga- inhibitory factor (LIF), (IL)-6 and IL-11, car- tions have revealed that hOSM and mouse OSM (mOSM) diotrophin-1 (CT-1) and ciliary neurotrophic factor exhibit a diversity of effects on different target cell types. (CNTF). A key feature of the gp130 cytokines is that These include the ability to inhibit the multiplication of a intracellular signalling results from the ligand-mediated variety of different tumour-derived cell lines (reviewed in formation of oligomeric receptor complexes which, [2]) as well as inducing the proliferation of certain cells depending upon the identity of the ligand, can differ in including Kaposi sarcoma-derived spindle cells [3] and hae- both composition and stoichiometry [12]. In all cases, mangioblasts derived from the aorta–gonad–mesonephros however, receptor complexes include one or more mol- region of the developing embryo [4]. hOSM, released from ecules of the shared transducer gp130 [13]. gp130 macrophages and neutrophils [5], has also been implicated cytokines can therefore exhibit both unique and shared as an important mediator of pro-inflammatory responses in biological activities in vivo and in vitro depending upon the joint destruction [6] and central nervous system inflamma- composition of the receptor signalling complex that is tion and neurodegeneration [7,8] via, inter alia, the induc- formed [11]. hOSM signals via formation of an active tion of acute- phase expression [9,10]. receptor signalling complex that comprises gp130 and one 864 Structure 2000, Vol 8 No 8

of two partner receptors: either the LIF receptor (LIFR) cases, receptor specificity is controlled by subtle structural [14] or the structurally related OSM-specific receptor, changes to a common recognition framework. OSMR [15]. mOSM [16] differs from hOSM in that it only signals via association with gp130 and the murine OSMR Results [17,18]; this means that there is a significant divergence in Expression and structure determination action between mOSM and hOSM and that the LIFR A soluble 21.4 kDa fragment of hOSM was expressed in recognition elements of OSM are not conserved. This Escherichia coli as a glutathione-S-transferase (GST) fusion finding also indicates that the unique biological activities protein. The resulting purified protein consists of residues of hOSM, such as cytostatic effects on tumour cells, are 1–187 of hOSM and a further four residues at the N termi- mediated via association with OSMR. Those activities of nus corresponding to the 3C protease site linker (GPGS). hOSM that are shared with other cytokines, such as LIF or Ba/F3 cell-survival assays and receptor-binding assays con- CNTF, may reflect a common association with the LIFR. firmed that this recombinant form binds gp130 with an affinity equivalent to wild-type (WT) hOSM. Knowledge of the structural basis of receptor recognition by gp130 cytokines is a central issue in understanding both Crystallisation trials produced Bragg diffraction quality the specificity of their biological functions and developing crystals belonging to the orthogonal space group P212121, methods for therapeutic intervention in gp130-mediated with unit-cell dimensions a = 35.8 Å, b = 53.1 Å, c = 106.8 Å. signalling. The crystal structures of three gp130 cytokines, The structure was determined from a single ethylmer- LIF [19], CNTF [20] and IL-6 [21], have been previously curyphosphate (EMP) derivative using the multiple described. These reveal that all three cytokines share a anomalous dispersion (MAD) phasing technique with common three-dimensional fold, the four-helix bundle X-ray diffraction data collected on beam line BM14 of the [22], which comprises four α helices ranging from 15 to 22 European Synchrotron Radiation Facility (ESRF). The amino acids in length (termed A, B, C and D) and linked structure has been refined to a crystallographic R factor of by polypeptide loops. Detailed structure–function studies 20.5% (Rfree 26.1%) for all data between 20.0 and 2.2 Å, of human LIF, which is most closely related to hOSM in with stereochemistry typified by root mean squared devia- sequence and receptor repertoire, have defined two recep- tions (rmsd) in bond lengths of 0.015 Å and in bond angles tor recognition epitopes: sites 2 and 3. Site 2 (so-termed by of 1.8°. Crystallographic statistics are reported in Tables 1 analogy to the receptor-binding sites in growth hormone and 2. Residues 1–3 and 135–155 appear to be disordered [23]), is required for recognition of gp130 and has been and are not included in the final model. With the excep- narrowed down to a cluster of candidate amino acid tion of these regions, and a flexible loop between residues residues located in helices A and C [24]. Site 3 is required Glu90 and Lys110, the electron density is good through- for recognition of LIFR and is dominated by a pair of con- out the course of the mainchain (a representative portion served amino acid residues located at the N-terminal tip of of electron density is shown in Figure 1a). Residue num- helix D [24]. The combined activity of these two recogni- bering is that of the human native mature protein tion epitopes leads to the formation of a trimeric, LIF- (GenBank accession number M27286). mediated, high-affinity signalling complex [24]. Structure description Structural comparisons of ligands with differing recognition hOSM is a compact barrel-shaped molecule with dimen- properties present a valuable opportunity to analyse the sions of approximately 20 Å × 27 Å × 56 Å that conforms molecular determinants that dictate receptor specificity. to the up-up-down-down four-helix bundle topology Here, we report the determination of the three-dimensional (Figure 1b,c). The basic fold consists of the four main structure of hOSM by crystallographic techniques and con- helical regions (helix A, residues 10–37; helix B, residues comitant functional dissection by mutagenesis. Because 67–90; helix C, residues 105–131; helix D, residues hOSM binds to gp130 with a higher affinity than LIF [24], 159–185) connected by two long overhand loops (AB loop, this allows structure–function analysis of the site 2 gp130 residues 38–66; CD loop, residues 130–158) and one short recognition epitope at the resolution of individual amino loop (BC loop, residues 91–104). The overall arrangement acid residues. Combining these findings with the high-reso- may be considered as two pairs of antiparallel helices that lution crystal structure of the complementary ligand-recog- pack together, with A–D forming one pair and B–C nition epitope located in the homology forming the other. domain (CHD) of gp130 [25] results, for the first time, in a definition of the structure of the interface between hOSM Both helices A and C exhibit breaks in the classical hydro- and gp130. In addition a detailed comparison of the three- gen-bonding pattern, with substitute hydrogen bonds dimensional structure of the site 3 recognition epitope of formed to tightly bound water molecules. In helix C this hOSM and hLIF also offers insights into the molecular results in a slight kink (between residues Gln112 and basis by which related ligands exhibit different receptor Pro116). The kink in helix A is more pronounced and is recognition properties. These findings reveal that, in both caused by a break in the helical conformation, as residues Research Article Crystal structure of oncostatin M Deller et al. 865

Table 1

Data collection statistics.

λ λ λ Native EMP 1 EMP 2 EMP 3

Wavelength (Å) 0.979273 1.00914 0.826516 1.00669 Unit cell a,b,c (Å) 35.8, 53.1, 106.8 35.9, 53.3, 106.7 α = β = γ (°) 90 90 Resolution range (Å) 25–2.2 25–2.6 Total reflections 43,901 24,040 26,223 24,159 Unique reflections 10,513 11,650 12,069 11,691 Redundancy 4.2 2.1 2.2 2.1 Completeness (%) 96.5 (98.2) 96.4 (63.1) 99.9 (100) 96.8 (66.7) 13.2 (4.3) 12.1 (2.9) 12.5 (4.7) 12.3 (3.3)

Rmerge (%) 8.3 (37.6) 4.6 (18.7) 4.6 (16.4) 4.3 (17.3) λ Σ Σ EMP denotes MAD data collected from EMP-derivatised hOSM. Rmerge = |I–|/ . Values in parentheses correspond to the highest resolution shell (2.66–2.60 Å for EMP, 2.25–2.20 Å for native).

Gln25 and Leu30 adopt a hydrogen-bonded turn confor- LIF [19,27]. Structural superimpositions with these and mation, with four tightly bound water molecules substitut- other four-helix bundle cytokines show that the gp130 ing for the classical hydrogen-bonding pattern (Figure 1a). cytokines form a distinct family both structurally and

Residues Thr27–Ile37 of helix A adopt a 310 helix confor- functionally. Within the gp130 cytokine family hOSM has mation. The curved nature of helices A and C facilitates the greatest structural similarity to LIF (rmsd of 2.1 Å close packing of the A–D and B–C helix pairs, so promot- over 145 pairs of structurally equivalent Cα atoms). The ing an extensive solvent-inaccessible core. more distant structural similarity to the four-helix bundles of the growth hormone family is at its best for superimpo- The core of the protein is composed of two groups of aro- sition of hOSM and (EPO; [28]) (rmsd of matic stacking interactions in which Phe56, Tyr173, 2.8 Å for 120 pairs of structurally equivalent Cα atoms). Phe169 and Phe176 form one group and Phe70, Phe185 and Trp187 form another. This distribution emphasizes The addition of hOSM to the gp130 cytokine structural the hydrophobic nature of helix D that donates all of the database highlights several features of potential functional aromatics to these groupings, with the exception of Phe56 (AB loop) and Phe70 (B helix). Table 2

The N-terminal loop preceding helix A (Gly4–Glu9) is Refinement statistics. tethered to the C terminus of helix C by one of the two disulphide bridges (Cys6 and Cys127). The second disul- Resolution range (Å) 20–2.2 (2.28–2.20) R (%) 20.5 (27.1) phide bridge, between Cys49 and Cys167, anchors the cryst Rfree (%) 26.1 (33.6) start of the AB loop to the N-terminal region of helix D. Completeness The AB loop itself consists of two separate α-helical working set (%) 86.1 (81.0) regions running from Pro43 to Arg46 and Glu59 to Gly64. test set (%) 9.9 (8.6) Residues between these two helical regions adopt an No. of reflections (F > 0) working set 9410 (870) extended conformation and pack closely against helix D. test set 1082 (92) By contrast, the second two loops (BC and CD) are less No. of non-hydrogen atoms closely associated with the core structure. The BC loop protein 1319 protrudes above the four-helix bundle and exhibits higher water 116 than average B factors. Despite this relative mobility it Rmsd from ideality bond lengths (Å) 0.0147 does, however, contain a high percentage of regular sec- bond angles (Å) 1.7997 ondary structure with a 310 helix running between residues dihedrals (°) 19.8882 Ala95 and Asp97 followed by an α helix up to residue improper (°) 1.0381 Ser101. The long CD loop appears to be highly mobile Average B factors mainchain (Å2) 27.1 and could not be modelled as a single conformation. sidechain (Å2) 32.4 water (Å2) 38.4 Structural comparison with other helical cytokines all atoms (range) (Å2) 30.5 (6.3–84.9) hOSM is the fourth member of the gp130 cytokine family Σ Σ Rcryst = ||Fobs|–|Fcalc||/ |Fobs|. Rfree is as for Rcryst but calculated using a to be analysed structurally; crystal and/or solution struc- 10% test set of reflections excluded from the refinement. Values in tures have been reported for IL-6 [21,26], CNTF [20] and parentheses correspond to the highest resolution shell (2.28–2.20 Å). 866 Structure 2000, Vol 8 No 8

Figure 1

The structure of hOSM. (a) Refined α coordinates and 2|Fo|–|Fc| calc electron-density map of hOSM contoured at 1σ showing density for the kinked region of helix A centred on Thr27. CNTF and LIF also have kinks owing to disruption of the mainchain hydrogen bonding around residues Thr32 and Ser36, respectively [19,20]. Both these residues bind water molecules through the Oγ1 atom to provide a substitute hydrogen-bond network similar to the pattern observed here for hOSM involving residue Thr27. (b) Ribbon diagram of hOSM coloured from blue at the N terminus through to red at the C terminus. Disulphide bonds are represented as ball-and-stick models with the sulphur atoms highlighted as yellow spheres. The transparent dotted section represents the CD loop as observed in LIF. (c) Stereodiagram of the Cα trace for hOSM.

importance (Figure 2). First, the kinked helix A is as having the dominant role in ligand binding. Comparison revealed as a distinctive feature not just of the gp130 of loop structures between family members appears to cytokines but also specifically of the members of the bear this out as there are few clearly conserved characteris- family of cytokines that bind to LIFR; IL-6 is the sole tics. Each crystal structure, except LIF, indicates highly gp130 cytokine lacking this feature (Figures 1a,2a). The flexible regions in the long interhelix loops; however, kink results in a repositioning of the C-terminal end of these vary in location between the AB loop in CNTF and helix A relative to the ‘top’ of the four-helix bundle, a IL-6, and the CD loop in CNTF and hOSM. Similarly the region carrying the major epitope for LIFR, and OSMR, positioning of disulphides tethering loop regions to the binding (at the start of helix D). This feature is therefore core four-helix bundle is not conserved. It has been noted, suggestive of the positioning of the C-terminal end of however, that the positioning, relative to helix D, of the helix A impacting on LIFR and OSMR binding. Con- N-terminal portion of the AB loop in LIF differs from that versely, the positioning of the N-terminal portion of in the distantly related four-helix bundle cytokines [19]. helix A, which is implicated in gp130 binding, is relatively The current structural comparisons may point to some conserved across all members of the family. functional correlation of this feature with LIFR binding as both LIF and hOSM have a disulphide bridge anchoring In general, functional studies on the gp130 cytokines have an α-helical N-terminal portion of the AB loop to the top implicated the helices, rather than the connecting loops, of helix D (Figure 2a). Research Article Crystal structure of oncostatin M Deller et al. 867

Figure 2

Structural comparisons. (a) From left to right the structures of human OSM, LIF (PDB entry code 1EMR), CNTF (1CNT) and IL-6 (1ALU) are shown in equivalent orientations. Each is represented as a ribbon coloured blue at the N terminus through to red at the C terminus. Helix A (blue) is kinked in OSM, LIF and CNTF whereas helix C ( green/yellow) is only significantly kinked in OSM. Of the interhelical loops, only the relatively short BC loop is well determined in all the structures. This region in hOSM has a five-residue insertion that protrudes from the top of the four-helix bundle. Both LIF and hOSM have a disulphide bridge anchoring an α-helical N-terminal portion of the AB loop to the top of helix D, whereas in IL-6 the loop is anchored to helix B. The AB loop in CNTF lacks any such tether and was thus too mobile to be accurately modelled in the crystal structure. The CD loop has not been fully modelled for hOSM and CNTF, although the orientation of residues preceding this loop suggests a similar route is taken to that of LIF, IL-6 and CNTF. (b) Comparison of surface charge between hOSM, LIF, CNTF and IL-6. Blue denotes positive and red negative; electrostatic potential is contoured at ±8.5 kT using the program GRASP [53]. The orientation is as in (a). (c) Surface charge comparisons are displayed as in (b) but viewed directly onto the top of the four-helix bundle.

Sequence conservation across the gp130 cytokines is rela- (KRH and JKH, unpublished observations) derived using tively low (i.e., typically some 20% sequence identity); the crystal coordinates of human LIF (hLIF; Protein Data therefore very limited conservation of surface character is Bank [PDB] ID 1EMR) as a template and previously pub- to be expected. Figures 2b and 2c illustrate a comparison lished work on the location of receptor-recognition epi- of the surface characteristics in two regions of functional topes in human LIF [24] and CNTF [29]. Only mutation significance; for both regions there is notable conservation of those residues that proved to exhibit significant solvent between three of the four gp130 cytokines. At the puta- exposure (Table 3) in the crystal structure of hOSM are tive gp130-binding region involving helix A (Figure 2b), considered here. CNTF exhibits a significantly different electrostatic surface potential, whereas for the putative LIFR binding Mutant were tested for their ability to inhibit the site 3 on the top of the four-helix bundle (Figure 2c), IL-6 binding of biotin-labelled hOSM to immobilised human is the outlier. gp130 or human LIFR ectodomains in solid-phase binding assays (Figure 3). It was not possible to analyse Site-directed mutagenesis of hOSM binding to hOSMR in this assay format owing to the sig- A panel of alanine scanning mutants of hOSM was gener- nificantly reduced affinity of hOSM for OSMR compared ated in order to identify critical residues involved in recep- to LIFR [18]. Mutant proteins were also tested for their tor recognition. The selection of residues for mutagenesis ability to promote the multiplication of BAF cells trans- was initially on the basis of a structural model of hOSM fected with either human LIFR/human gp130 [24] or 868 Structure 2000, Vol 8 No 8

Table 3

Mutagenesis data for hOSM.

Location/mutant Site RSA (%)* Receptor-binding assays Ba/F3 cell-survival assays

gp130 LIFR LIFR/gp130 hOSMR/gp130

Wild-type – – 1.0 1.0 1.0 1.0 N terminus of helix D F160A 3 65 1.1 >100 >10,000 >10,000 K163A 3 42 0.9 >100 >100 >10,000 F160A/K163A 3 – 1.0 >100 >10,000 >10,000 N-terminal loop S7A 2 100 1.1 0.9 0.8 0.5 K8A 2 8 0.8 1.1 0.8 0.5 E9A 2 96 1.2 0.8 1.4 1.2 N terminus of helix A Y10A 2 35 3.4 7.2 2.1 0.9 R11A 2 74 1.0 1.9 0.8 0.5 V12A 2 55 1.0 1.2 0.7 0.7 Q16A 2 56 75.0 Q18 2 36 4.0 5.0 1.2 1.1 K19 2 79 1.0 2.0 1.4 2.0 Q20A 2 27 90.0 1.0 20.0 30.0 D22A 2 40 1.0 1.0 0.8 1.0 L23A 2 61 3.0 6.7 2.5 5.0 D26A 2 63 4.0 5.0 3.0 4.1 Helix C M113A 2 66 0.7 1.0 1.0 1.0 P116A 2 52 0.7 2.3 1.0 0.9 N117A 2 29 2.0 1.0 2.6 8.0 L119A 2 46 1.7 2.2 1.2 1.7 G120A 2 76 >100 1.8 60.0 60.0 G120Y 2 76 >100 1.8 >10,000 >10,000 N123A 2 61 8.0 1.0 3.3 10.0 N124A 2 31 >100 1.0 >10,000 >10,000

*RSA, the relative solvent accessibility of the sidechain. Mutants were tested for biological activity on Ba/F3 cells transfected with tested for inhibition of binding to immobilized gp130 and hLIFR. gp130/LIFR or hOSMR/gp130. Activity is expressed as the ratio EC50 Activity is expressed as the ratio IC50 mutant/IC50 WT hOSM. IC50 mutant/EC50 WT hOSM. EC50 refers to the concentration of mutant or refers to the concentration of mutant or WT protein required to elicit WT hOSM required to elicit a response equivalent to 50% of the 50% inhibition of binding of WT biotinylated hOSM. Mutants were also maximal response to hOSM. human OSMR/human gp130. The results for each mutant WT hOSM in the ability to interact with LIFR (data not are summarised in Table 3 and data from selected mutants shown), but greatly reduced activity in the ability to bind are shown in Figure 3. gp130 or stimulate the multiplication of gp130/LIFR (Figure 3b) or gp130/OSMR (Figure 3c) transfected BAF Identification of site 2 cells. A comparable order of activity was observed in The interaction of hLIF with gp130 (via site 2) has been both gp130-based solid-phase assays and bioassays; the previously shown to depend on a cluster of adjacent mutant Asn124Ala exhibited the greatest reduction in amino acids in the A and C helices, although the relative activity followed by Gly120Ala and Gln20Ala. We also significance of each amino acid could not be determined constructed a mutant in which Gly120 was replaced by [24]. This issue can be addressed by analysis of hOSM tyrosine (Gly120Tyr) thereby emulating the identity of which binds to gp130 with higher affinity than hLIF. the equivalent residue in CNTF that binds gp130 with Alanine mutations in the individual amino acids compris- very low affinity. This mutant also exhibited reduced ing the topologically equivalent region of hOSM ability to bind gp130 and stimulate the multiplication of revealed that the interaction between hOSM and gp130 transfected BAF cells and was significantly less active was critically dependent upon the identity of four struc- than the mutant Gly120Ala in both assay formats turally adjacent solvent-exposed residues, Gln20, Gly120 (Figures 3a–c). Collectively, these findings reveal that and Asn124 (Figure 3a), and, to a lesser degree, Gln16 the interaction of hOSM with gp130, the site 2 (Table 3). Each mutant exhibited equivalent activity to epitope, is governed by four residues located in a cluster Research Article Crystal structure of oncostatin M Deller et al. 869

Figure 3

(a) (d) (b) (e)

(c) (f)

Structure

Functional binding data for hOSM muteins. (a) Competitive inhibition Ba/F3–OSMR/gp130 assay. (d) Competitive inhibition of biotinylated of biotinylated OSM binding to gp130–Fc by site 2 OSM muteins. OSM binding to LIFR–Fc by site 3 OSM muteins. (e) Biological activity (b) Biological activity of OSM muteins in the Ba/F3–LIFR/gp130 of OSM muteins in the Ba/F3–LIFR/gp130 assay. (f) Biological activity assay. (c) Biological activity of OSM muteins in the of OSM muteins in the Ba/F3–OSMR/gp130 assay.

on the A and C helices (Figure 4). Of these four residues plays a critical role in receptor recognition, in that a the most significant, in terms of the consequences of single substitution to the bulky tyrosine residue (found alanine substitution, is Asn124 (in helix C) followed by in CNTF) almost completely ablates the ability of Gly120 and Gln20. This study also reveals that Gly120 hOSM to interact with gp130. 870 Structure 2000, Vol 8 No 8

Figure 4

Structural mapping of binding sites 2 and 3 for hOSM. (a) Ribbon coloured patches. For site 2 the orientation is as in (a) for site 3 the representation of hOSM with solvent-exposed residues studied by view is on to the top of the four-helix bundle. Residues not implicated mutagenesis depicted in ball-and-stick representation. The solvent- in gp130 binding are in light-blue, those with some effect are yellow, accessible surface of hOSM in the regions of (b) site 2 and (c) site 3 residues critical for gp130 binding are green and those critical for LIFR with areas contributed by the mutagenised residues highlighted as binding are red.

Identification of site 3 conserved. This implies that the difference in receptor The location of residues that govern the interaction with specificity between hLIF and hOSM and mOSM and LIFR (site 3) has been defined for hLIF [24] and CNTF hOSM is the result of structural differences between these [29]. These comprise a conserved, solvent-exposed, ligands in the vicinity of the core Phe–Lys motif. phenylalanine and lysine pair located at the N-terminal end of helix D. These two residues are conserved in all Discussion cytokines that bind LIFR [19], including hOSM (Phe160 The gp130 family of cytokines represents a situation in and Lys163). The availability of a high-resolution structure which ligands exhibit both communality and divergence of hOSM now allows examination of the mechanism by in ligand-recognition properties. The crystal structure of which hOSM, unlike hLIF, is able to bind both LIFR and hOSM reported here represents an important advance in OSMR, and therefore the extent to which the site 3 recog- understanding the molecular basis of this receptor speci- nition mechanism for different receptors is conserved. Two ficity. When combined with the previously reported struc- single point substitutions, Phe160Ala and Lys163Ala, and tures of murine, human and inter-species chimeric LIF the compound mutant, were tested for biological activity in (PDB entries 1LKI, 1EMR, 1A7M), as well as CNTF the assay systems described above (Figures 3d–f; Table 3). (1CNT), it is possible to define structural features that Whereas all three mutants exhibited comparable activity to represent shared recognition properties, such as the inter- WT hOSM in their ability to interact with gp130 (data not action with gp130 and LIFR and divergent properties shown), both point mutants and the compound mutant such as the interaction between hOSM and OSMR or the were unable to interact with LIFR (Figure 3d; Table 3). requirement for the accessory CNTF receptor for the Analysis of the effect of these mutations on the ability to interaction between CNTF and gp130. stimulate the multiplication of transfected BAF cells (Figures 3e,f) revealed that for both LIFR- and OSMR- All gp130 cytokine structures reported to date, including dependent BAF cells, mutations of either Phe160 or hOSM, conform to a similar overall structural fold. This is Lys163 resulted in a dramatic loss of activity. the archetypal four-helix bundle form of the long-chain cytokine class [30]. On the basis of this paper, two struc- These findings show that the ability of hOSM to interact turally distinct subfamilies can be defined within the with LIFR and OSMR is critically dependent upon the gp130 cytokines. All cytokines that interact with LIFR identity of this pair of conserved residues and that the prove to have a marked kink in the A helix. This is most mechanism of interaction with both LIFR and OSMR is pronounced in the present hOSM structure (Figure 2). Research Article Crystal structure of oncostatin M Deller et al. 871

Figure 5

Complimentarity between the interaction surfaces of hOSM and gp130. The solvent- accessible surfaces of site 2 on hOSM (left) and the cognate binding site on gp130 (right) are displayed with areas contributed by residues implicated in binding highlighted as coloured patches. The orientation for hOSM site 2 is rotated 90° from that in Figure 4 whereas gp130 is rotated by 180° from its orientation relative to hOSM in the putative interaction complex, as in the opening of a book.

By contrast, in IL-6 [21], which does not interact with immediately suggests an explanation for the molecular LIFR, the A helix conforms to the standard hydrogen- mechanism of ligand recognition. This indicates that the bonding pattern throughout its length. This indicates exposed phenylalanine residue on gp130 may be docked that the A helix kink may represent a structural feature of into the hydrophobic ‘dimple’ formed on the hOSM LIFR recognition. In this respect it is notable that the surface from the Cα backbone of Gly120. In this configu- LIFR contains two copies of the CHD in the extracellu- ration the hydroxyl group of gp130 Tyr196 would be able lar domain and that the A helix kink may represent a to hydrogen bond with the exposed amide group of structural feature which permits a low-affinity interaction Asn124 from hOSM. This explanation is consistent with with the N-terminal CHD of LIFR. The existence of mutagenesis of hOSM Gly120 reported here in which sub- such an interaction has been inferred from homology stitution to alanine, which has a relatively small non-polar scanning studies of the interaction between LIF and sidechain, has less effect on binding than substitution to LIFR [31,32]. the bulkier tyrosine residue (Table 3). The substitution of a bulky sidechain would be expected to occlude the non- The interaction with gp130 polar dimple and thereby inhibit docking of the Phe169 of The detailed mutagenesis study of hOSM reported here, gp130 and formation of the hydrogen bond between the combined with earlier studies of hLIF [24], has identified Tyr196 of gp130 and the Asn124 of hOSM. those amino acid residues that contribute the majority of binding energy to the interaction with gp130, as defined This proposed mode of docking between hOSM site 2 and by the consequences of alanine scanning analysis. This gp130 has strong similarities to the interaction between reveals a quartet of residues in the A and C helices that growth hormone and the growth hormone receptor [23] form a distinct patch in the crystal structure (Figure 4). In and EPO/EPO receptor [28]. In both these cases the site 2 terms of the interaction between hOSM and gp130, the interaction also involves burying a bulky, solvent exposed, most important residues are Asn124 and Gly120 in helix C, aromatic residue in a hydrophobic patch formed from the with lesser contributions from Gln16 and Gln20 in helix A. Cα backbone of helix C. This suggests that the basic The crystal structure of the CHD of gp130 [25] combined site 2 recognition mechanism may exhibit strong structural with mutagenesis data of gp130 ([33]; KRH, E Raulo, O similarities across the whole cytokine/receptor family. The McKenzie and JKH unpublished observations) provides a point of divergence between gp130 and these other structural explanation for this interaction. The most promi- systems appears to arise from the use of polar interactions, nent feature of gp130 is a solvent-exposed hydrophobic for example the putative Tyr100–Asn124 hydrogen bond, residue (Phe169) in an interstrand loop of the N-terminal that contribute additional specificity to complex forma- half of the bipartite CHD (Figure 5). The second signifi- tion. Thus a major contribution to binding affinity from a cant determinant of gp130 recognition is a prominent distinctive hydrophobic interaction is conserved whilst solvent-exposed tyrosine residue (Tyr196), located close to specificity is controlled through polar interactions. A the junction between the two elements of the CHD, the similar pattern of conservation has recently been noted for hydroxyl group of which faces away from the main body of a second set of cytokine–receptor interactions, those of the the molecule into solution (Figure 5). Inspection of the tumour necrosis factor (TNF) like cytokines with their two functional epitopes in open book format (Figure 5) cognate receptors [34]. 872 Structure 2000, Vol 8 No 8

This hypothesis also suggests why CNTF interacts with cytokine–receptor interaction is of fundamental biologi- gp130 at a very much lower affinity than either LIF or cal significance as it dictates the resulting intracellular hOSM. In CNTF the equivalent residue to Gly120 is a signal-transduction processes and, accordingly, the asso- tyrosine which, as argued above, would be predicted to ciated biological responses. The family of cytokines disrupt the core interaction. The employment of the acces- which mediate their effects through binding to the gp130 sory receptor CNTF-R in this scheme is required to com- receptor — namely oncostatin M (OSM), leukaemia pensate for this loss of affinity by introducing additional inhibitory factor, 6 and 11, cardiotrophin-1 sites of receptor–receptor interaction into the complex. and ciliary neurotrophic factor (CNTF) — are of partic- ularly broad biomedical importance. Of these gp130- The interaction with LIFR and OSMR binding cytokines, human OSM (hOSM) is of potential A distinctive feature of the gp130 cytokine family is the interest for novel cancer therapeutics as well as provid- use of a novel receptor recognition site 3 located at the N ing a target for the treatment of inflammatory responses terminus of helix D. This epitope engages the second sig- in the joints and nervous system. nalling receptor in the complex: either LIFR/OSMR in the case of OSM, LIF and CNTF, or gp130 in the case of IL-6 The combined crystal structure and mutagenesis analy- and IL-11. Previous studies [24,29] have revealed that, for sis of hOSM reported here has provided important the interaction with LIFR, this site 3 epitope involves a insights into the molecular basis of biological specificity solvent-exposed pair of residues which are conserved in all in the gp130 cytokine system. Particular value accrues LIFR-binding cytokines described to date. The availabil- from the structural comparisons these data now allow ity of the crystal structure of hOSM has enabled us to with a panel of gp130 cytokines that exhibit both determine both the extent to which this recognition common and divergent receptor-recognition properties. scheme holds for additional members of the family and to The study confirms the concept [36] that receptor recog- examine the involvement of this epitope in the interaction nition involves a conserved non-polar ‘core’ inter- between hOSM and OSMR. Mutagenesis of the two key action with specificity contributed by the identity of site 3 residues in hOSM, Phe160 and Lys163, reveals that residues surrounding the core. This study also suggests each residue is required for the interaction with both LIFR a second aspect to recognition specificity: the require- and OSMR. This epitope (Figure 4) therefore represents a ment for soluble receptors in some gp130 signalling com- core structural element in site 3 recognition, although the plexes (e.g. CNTF receptor in the CNTF-mediated precise significance of each residue in the epitope differs recognition complex) acting to compensate for low-affin- between LIFR and OSMR. It follows that the ability of ity core interactions by the introduction of additional OSM to interact with OSMR must result from either the receptor–receptor interactions. Finally, this study pro- involvement of additional residues in the vicinity of this vides the structural basis for the development of small- epitope or the presence of residues in LIF which block molecule compounds to intervene in hOSM-mediated some feature of the interaction with OSMR. signalling events.

It is intriguing that the structure of the site 3 recognition Materials and methods epitope is similar to the site 2 recognition site of gp130: Production and expression of hOSM proteins both employ a prominent aromatic sidechain as a ‘core’ A 21.4 kDa fragment of mature hOSM (residues 1–187) was expressed as a GST fusion protein using the pGex-2T expression with additional affinity contributed by the sidechain of a vector (Pharmacia). The hOSM and GST proteins were joined by a residue in close physical proximity. This suggests that recognition site for rhinovirus 3C protease. All OSM mutants were gen- site 3 receptor recognition may involve a similar overall erated by overlap polymerase chain reaction from the pGEX–hOSM scheme to that of site 2 except that the complementary template GST fusion using specific oligonucleotides for each mutant; the sequences of primers used are available on request. hOSM epitopes on receptor and ligand are reversed. This suggests (1–187) and mutant proteins were expressed as GST fusions in E. coli that the site 3 recognition site in LIFR, which is located in strain JM109. Soluble GST fusion proteins were bound to glu- the immunoglobulin-like domain [31,32,35], may involve tathione–sepharose resin and the OSM protein released by overnight an analogous hydrophobic dimple or patch formed from cleavage with 3C protease at 37°C and purified by high-resolution gel- filtration on a Sephadex-75 column (Pharmacia). All proteins had a residues with small sidechains. In this respect it is notable purity of greater than 80% as determined by SDS PAGE and silver that alanine scanning studies of the LIFR site 3 hOSM staining. Protein concentrations were determined by the Coomassie recognition site imply a short sequence motif with a con- Protein Plus assay (Bio-rad). served glycine residue at its core [32]. The identity of the purified hOSM (1–187) protein employed for crys- tallisation was confirmed using electrospray ionisation on a Micromass Biological implications BioQ II triple, quadrapole, atmospheric pressure mass spectropho- The interaction of cytokine ligands with transmembrane tometer equipped with an electrospray interface operating in the posi- receptors to form signalling complexes is a major mecha- tive ion mode (Robin Aplin, Oxford Centre for Molecular Science, Oxford). N-terminal sequencing was also carried out for 10 cycles nism for intercellular signal transduction and cell regula- using an Applied Biosystems 494A/473A Sequencer (Tony Willis, tion. The molecular specificity of the extracellular MRC Immunochemistry Unit, Oxford). Research Article Crystal structure of oncostatin M Deller et al. 873

Bioassays of hOSM mutants for the EMP derivative were determined by manual inspection of differ- Mutant forms of hOSM were tested for biological activity on the Ba/F3- ence and anomalous Patterson syntheses and confirmed using the pro- hLIFR/hgp130 cell line as described previously [24]. The BA/F3- grams RSPS and SOLVE [42]. The sites were refined using the hOSMR/hgp130 cell line was generated by co-transfecting Ba/F3 cells program SHARP [43], and MAD phases were calculated to 2.6 Å reso- with a full-length hgp130 cDNA expressed under the control of the lution (figure of merit = 0.46, phasing power = 1.84). Initial maps based λ cytomegalovirus (CMV) promoter in plasmid pCDNA3 and the full-length on these phases and Fobs from the EMP 2 data set proved uninter- hOSMR cDNA [15] expressed under the control of the CMV promoter in pretable. Hendrickson-Lattman coefficients for SIRAS phases derived plasmid pCDNA3. Transfected cells were cultured overnight in 50 ng/ml using the EMP derivative and native data were combined with the MAD λ IL-3 and then selected for co-expression of hgp130 and hOSMR in the phases. Maps based on these combined phases and the EMP 2 Fobs presence of 50 ng/ml hOSM. Co-expression of both receptors in co- (figure of merit 0.51, phasing power = 1.90) showed improved main transfectants was confirmed by FACS analysis after staining with anti- chain density. Phase improvement was carried out using solvent flat- bodies directed against hgp130 [37] and hOSMR (KRH and JKH, tening, as implemented by the program SOLOMON [44], and the unpublished observations). Bioassays were performed as described pre- structure was traced using the interactive graphics program O [45]. viously [19]. All assays were performed in triplicate in at least three inde- pendent experiments. The data presented are the results from a single All refinement steps (positional, torsional dynamics, grouped B factor experiment, as it was noted that absolute response to defined concentra- and individual B factor) were carried out using the program CNS- tions of ligand varied between different batches of cells, although the rel- SOLVE [46]. A bulk-solvent correction was employed (solvent ative potencies of the WT and mutant proteins were preserved. The SEM density = 0.29 e/Å3, B-factor = 31.1 Å2), so allowing all data in the res- of triplicate determinations was always less than 15% of the mean. olution range 20.0–2.2 Å to be included. In the final stages of refine- ment 116 waters were located. The current model consists of residues Ligand-binding assays 4–134 and 156–187 of the mature protein with no residues falling in Human gp130 and hLIFR were expressed as fusion proteins with disallowed regions of the Ramachandran plot. Refinement and model human Fc as described previously [24]. The fusions were purified by statistics are reported in Table 2. protein A sepharose affinity chromatography and employed in solid- phase ligand-binding assays using biotin-conjugated hOSM as the The quality of the final model was assessed using the program tracer [24]. The ability of mutant forms of hOSM to compete with PROCHECK [47], secondary structure assignments were calculated biotin–hOSM tracer was analysed using the PRISM software package using the program DSSP [48] and solvent accessibilities calculated (Graphpad Software) using a single-site binding model. using NACCESS [49]. Structural superimpositions were performed using the program SHP [50] and structural figures were produced Crystallization and derivatisation using BOBSCRIPT [51], RASTER3D [52], VOLUMES (R Esnouf, per- For crystallisation trials, hOSM(1–187) was concentrated to 8 mg/ml in sonal communication) and GRASP [53]. 50 mM Tris-HCl containing 150 mM sodium chloride, pH 7.5. Crystals grew by vapour diffusion within 1–3 days at 22°C from hanging drops Accession numbers typically comprising 2 µl of protein solution plus 2 µl of reservoir solu- Atomic coordinates for hOSM have been deposited with the PDB tion (30% w/v PEG 35,000, 0.25 M sodium acetate, pH 7.5). Crystals (accession number 1EVS). measured approximately 0.3 × 0.2 × 0.2 mm3 and showed ordered Bragg diffraction to 2.2 Å when exposed to synchrotron radiation. Pre- Acknowledgements liminary characterisation indicated that the crystals belong to the This work was supported by the Cancer Research Campaign. We thank K. orthogonal spacegroup P212121 with unit cell dimensions a = 35.8 Å, Harlos for help with ‘in house’ data collection; the EMBL and ESRF staff, in b = 53.1 Å, c = 106.8 Å and α = β = γ = 90°. There is one molecule particular V. Stroganoff, for assistance on BM14 at the ESRF, R. Esnouf for per asymmetric unit and the crystal solvent content is ~48.0% provision of computing facilities, V. Barton and L. McGovern for discussion. (Matthews coefficient, Vm = 2.3 Å/Da). Tests using a number of heavy We are grateful to T. Willis and R. Aplin for performing the protein sequenc- atom compounds highlighted EMP as a putative heavy atom derivative. ing and mass spectrometry, respectively. EYJ is a Royal Society University Derivatisation was carried out by soaking crystals in pre-equilibrated Research Fellow. SI is supported by the Wellcome Trust and MCD was drops of mother liquor containing 14 mM EMP for 16–24 h. supported by the MRC and CRC.

Data collection References MAD data were collected to a resolution of 2.6 Å from a single EMP deriv- 1. Malik, N., et al., & Wei, C.M. (1989). Molecular cloning, sequence analysis, and functional expression of a novel growth regulator, ative crystal on beam line BM14 at the ESRF, Grenoble, France. Crystals 9 µ oncostatin M. Mol. Cell Biol. , 2847-2853. were briefly passed through a cryoprotectant solution consisting of a 2 l 2. Grant, S.L. & Begley, C.G. (1999). The oncostatin M signalling pathway: drop of 30% (w/v) PEG 35,000, 0.25 M sodium acetate, pH 7.5 over- reversing the neoplastic phenotype? Mol. Medicine Today 5, 406-412. layed with 2 µl of 30% (w/v) PEG 8000, 0.25 M sodium acetate, pH 7.5 3. Miles, S.A., et al., & Linsley, P.S. (1992). Oncostatin M as a potent and then flash-cooled directly in liquid propane. The crystal was main- mitogen for AIDS-Kaposi’s sarcoma-derived cells. Science tained at 100K in a dry air stream and data recorded using a MAR 345 255, 1432-1434. image plate detector (MAR Research, Hamburg, Germany). The mercury 4. Mukouyama, Y., et al., & Miyajima, A. (1998). In vitro expansion of LΙΙΙ edge of the derivative crystal was characterised by a fluorescence murine multipotential hematopoietic progenitors from the embryonic scan and X-ray data sets collected at three wavelengths (λ = 1.00669 Å, aorta–gonad–mesonephros region. Immunity 8, 105-114. 5. Grenier, A., et al., & Chollet Martin, S. (1999). Oncostatin M 1.00914 Å and 0.826516 Å). A 2.2 Å resolution native data set was also production and regulation by human polymorphonuclear neutrophils. collected at the same beamline from a single cryo-cooled crystal. Blood 93, 1413-1421. 6. Kerr, C., Langdon, C., Graham, F., Gauldie, J., Hara, T. & Richards, Individual data sets were auto-indexed, integrated and scaled using the C.D. (1999). Adenovirus vector expressing mouse oncostatin M programs DENZO and SCALEPACK [38]. Structure factor amplitudes induces acute-phase proteins and TIMP-1 expression in vivo in mice. were calculated using the program TRUNCATE [39], the data sets J. Cytokine Res. 19, 1195-1205. scaled using the program SCALEIT [40], and normalised anomalous 7. Ensoli, F., et al., & Aiuti, F. (1999). Inflammatory cytokines and HIV-1- scattering magnitudes estimated using the program REVISE [41]. Data associated neurodegeneration: Oncostatin-M produced by collection statistics are reported in Table 1. mononuclear cells from HIV-1-infected individuals induces apoptosis of primary neurons. J. Immunol. 162, 6268-6277. 8. Kordula, T., Rydel, R.E., Brigham, E.F., Horn, F., Heinrich, P.C. & Travis, Structure determination and analysis J. (1998). Oncostatin M and the interleukin-6 and soluble interleukin-6 Scaling of MAD data sets yielded maximal dispersive differences of receptor complex regulate α1-antichymotrypsin expression in human 10.6% and maximal anomalous differences of 9.7%. The mercury sites cortical astrocytes. J. Biol. Chem. 273, 4112-4118. 874 Structure 2000, Vol 8 No 8

9. Heinrich, P.C., et al., & Wollmer, A. (1998). Interleukin-6 and related 35. Taupin, J.L., et al., & Moreau J.F. (1999). Binding of leukemia inhibitory cytokines: effect on the acute phase reaction. Zeitschrift factor (LIF) to mutants of its low affinity receptor, gp 190, reveals a LIF Ernährungswissenschaft 37, 43-49. binding site outside and interactions between the two cytokine 10. Kuropatwinski, K.K., De Imus, C., Gearing, D., Baumann, H. & Mosley, binding domains. J. Biol. Chem. 274, 14482-14489. B. (1997). Influence of subunit combinations on signaling by receptors 36. Clackson, T. & Wells, J.A. (1995). A hot spot of binding energy in a for oncostatin M, leukemia inhibitory factor, and interleukin-6. J. Biol. hormone-receptor interface. Science 267, 383-386. Chem. 272, 15135-15144. 37. Vernallis, A.B., Hudson, K.R. and Heath, J.K. (1997). An antagonist for 11. Taga, T. & Kishimoto, T. (1997). Gp130 and the interleukin-6 family of the leukemia inhibitory factor receptor inhibits leukemia inhibitory cytokines. Annu. Rev. Immunol. 15, 797-819. factor, cardiotrophin-1, ciliary neurotrophic factor, and oncostatin M. 12. Heinrich, P.C., Behrmann, I., Muller-Newen, G., Schaper, F. & Graeve, J. Biol. Chem. 272, 26947-26952. L. (1998). Interleukin-6-type cytokine signalling through the 38. Otwinowski, Z. & Minor, W. (1997). Processing of X-ray diffraction gp130/Jak/STAT pathway. Biochem. J. 334, 297-314. data collected in oscillation mode. Meth. Enzymol. 276, 307-326. 13. Hibi, M., Murakami, M., Saito, M., Hirano, T., Taga, T. & Kishimoto, T. 39. French, S. & Wilson, K. (1978). On the treatment of negative intensity (1990). Molecular cloning and expression of an IL-6 signal transducer, observations. Acta Crystallogr. A 34, 517-525. gp130. Cell 63, 1149-1157. 40. Howell, P.L. & Smith, G.D. (1992). Identification of heavy-atom 14. Gearing, D.P. & Bruce, A.G. (1992). Oncostatin M binds the high- derivatives by normal probability methods. J. Appl. Crystallogr. 25, 81-86. affinity leukemia inhibitory factor receptor. New Biol. 4, 61-65. 41. Woolfson, M.M., Yao, J.X. & Fan, H.F. (1997). New techniques for 15. Mosley, B., et al., & Cosman, D. (1996). Dual oncostatin M (OSM) applying anomalous-scattering and isomorphous-replacement data receptors. Cloning and characterization of an alternative signaling incorporated in ANOMIR - a general application package. Acta subunit conferring OSM-specific receptor activation. J. Biol. Chem. Crystallgr. 53, 673-681. 271, 32635-32643. 42. Terwilliger, T.C. & Berendzen, J. (1999). Automated MAD and MIR 16. Yoshimura, A., et al., & Miyajima, A. (1996). Mouse oncostatin M: an structure solution. Acta Crystallgr. 55, 849-861. immediate early gene induced by multiple cytokines through the JAK- 43. de La Fortelle, E. & Bricogne, G. (1997). Maximum-likelihood heavy- STAT5 pathway. EMBO J. 15, 1055-1063. atom parameter refinement for multiple isomorphous replacement and 17. Ichihara, M., Hara, T., Kim, H., Murate, T. and Miyajima, A. (1997). multiwavelength anomalous diffraction methods. Meth. Enzymol. Oncostatin M and leukemia inhibitory factor do not use the same 276, 472-494. functional receptor in mice. Blood 90, 165-173. 44. Abrahams, J.P. & Leslie, A.G.W. (1996). Methods used in the 18. Lindberg, R.A et al., & Fletcher, F.A. (1998). Cloning and structure determination of bovine mitochondrial F1 ATPase. Acta characterization of a specific receptor for mouse oncostatin M. Mol. Crystallogr. D 52, 30-42. Cell Biol. 18, 3357-3367. 45. Jones, T.A., Zou, J.-Y., Cowan, S.W. & Kjeldgaard, M. (1991). Improved 19. Robinson, R.C., et al., & Jones, E.Y. (1994). The crystal structure and methods for building protein models in electron density maps and the biological function of leukemia inhibitory factor: implications for location of errors in these models. Acta Crystallgr. A 47, 110-119. receptor binding. Cell 77, 1101-1116. 46. Brünger, A.T., et al., & Warren, G.L. (1998). Crystallography & NMR 20. McDonald, N.Q., Panayotatos, N. & Hendrickson, W.A. (1995). system: a new software suite for macromolecular structure Crystal structure of dimeric human ciliary neurotrophic factor determination. Acta Crystallogr., 54, 905-921. determined by MAD phasing. EMBO J. 14, 2689-2699. 47. Laskowski, R.A., MacArthur, M.W., Moss, D.S. & Thornton, J.M. 21. Somers, W., Stahl, M. & Seehra, J.S. (1997). 1.9 Å crystal structure of (1993). PROCHECK: A program to check the stereochemical quality : implications for a novel mode of receptor dimerization of protein structures. J. Appl. Crystallogr. 26, 283-291. and signaling. EMBO J. 16, 989-997. 48. Kabsch, W. & Sander, C. (1983). Dictionary of protein secondary 22. Bazan, J.F. (1991). Neuropoietic cytokines in the hematopoietic fold. structure: pattern recognition of hydrogen-bonded and geometrical Neuron 7, 197-208. features. Biopolymers 22, 2577-2637. 23. de Vos, A.M., Ultsch, M. & Kossiakoff, A.A. (1992). Human growth 49. Hubbard, S.J. & Thornton, J.M. (1993). ‘NACCESS’. Computer hormone and extracellular domain of its receptor: crystal structure of program. Department of Biochemistry and Molecular Biology, the complex. Science 255, 306-312. University College, London. 24. Hudson, K.R., Vernallis, A.B. & Heath, J.K. (1996). Characterization of 50. Stuart, D.I., Levine, M., Muirhead, H. & Stammers, D.K. (1979). Crystal the receptor binding sites of human leukemia inhibitory factor and structure of cat muscle pyruvate kinase at a resolution of 2.6 A. J. Mol. 271 creation of antagonists. J. Biol. Chem. , 11971-11978. Biol. 134, 109-142. 25. Bravo, J., Staunton, D., Heath, J.K. & Jones, E.Y. (1998). Crystal structure 51. Esnouf, R.M. (1997). An extensively modified version of Molscript that of a cytokine-binding region of gp130. EMBO J. 17, 1665-1674. includes greatly enhanced colouring capabilities. J. Mol. Graphics 26. Xu, G.Y., et al., & Cumming, D.A. (1997). Solution structure of 15, 133-138. recombinant human interleukin-6. J. Mol. Biol. 268, 468-481. 52. Merritt, E.A. & Murphy, M.E.P. (1994). Raster3D version 2.0. A program 27. Hinds, M.G., Maurer, T., Zhang, J.G., Nicola, N.A. & Norton, R.S. for photorealistic molecular graphics. Acta Crystallogr. D 50, 869-873. (1998). Solution structure of leukemia inhibitory factor. J. Biol. Chem. 53. Nicholls, A., Sharp, K.A. and Honig, B. (1991). Protein folding and 273, 13738-13745. association: insights from the interfacial and thermodynamic 28. Syed, R.S., et al., & Stroud, R.M. (1998). Efficiency of signalling properties of hydrocarbons. Proteins 11, 281-296. through cytokine receptors depends critically on receptor orientation. Nature 395, 511-516. 29. Di Marco, A., et al., & Laufer, R. (1996). Identification of ciliary neurotrophic factor (CNTF) residues essential for leukemia inhibitory factor receptor binding and generation of CNTF receptor antagonists. Proc. Natl Acad. Sci. USA 93, 9247-9252. 30. Boulay, J. & Paul, W.E. (1993). Hemapoietin sub-family classification based on size, gene organization and . Curr. Biol. 3, 573-575 31. Owczarek, C.M., Zhang, Y., Layton, M.J., Metcalf, D., Roberts, B. & Nicola, N.A. (1997). The unusual species cross-reactivity of the leukemia inhibitory factor receptor α-chain is determined primarily by the immunoglobulin-like domain. J. Biol. Chem. 272, 23976-23985 32. Chobotova, K., Hudson, K.R. and Heath, J.K. (2000). The LIGHT motif: a ligand recognition site in the immunoglobulin-like domain of leukemia inhibitory factor receptor, submitted. 33. Horsten, U., et al., & Grotzinger, J. (1997). Molecular modeling-guided Because Structure with Folding & Design operates a mutagenesis of the extracellular part of gp130 leads to the identification of contact sites in the interleukin-6 (IL-6).IL-6 ‘Continuous Publication System’ for Research Papers, this receptor.gp130 complex. J. Biol. Chem. 272, 23748-23757. paper has been published on the internet before being printed 34. Mongkolsapaya, J., et al., & Screaton, G.R. (1999). Structure of the TRAIL-DR5 complex reveals mechanisms conferring specificity in (accessed from http://biomednet.com/cbiology/str). For apoptotic initiation. Nat. Struct. Biol. 6, 1048-1053. further information, see the explanation on the contents page.