Structural insights into the assembly and activation of IL-27 signalling complex

Interleukin 27 (IL-27) is a heterodimeric cytokine that elicits potent immuno-suppressive responses. Comprised of EBI3 and p28 subunits, IL-27 binds GP130 and IL-27Rα receptor chains to activate the JAK/STAT signalling cascade. However, how these receptors recognize IL-27 and form a complex capable of phosphorylating JAK proteins remains unclear. Here, we used cryo electron microscopy (cryoEM) to solve the structure of the IL-27 receptor recognition complex. Our data show how IL-27 serves as a bridge connecting IL-27Rα with GP130 to initiate signalling. While both receptors weakly bind the p28 component of the heterodimeric cytokine, EBI3 stabilizes the complex by binding a positively charged surface of IL-27Rα. We observe a large degree of flexibility in the rotation of D1 and the two CHR domains of GP130 contacting p28, which could contribute to GP130 binding degeneracy. We find that assembly of the IL-27 receptor recognition complex is distinct from both IL-12 and IL-6 cytokine families and provides a mechanistic blueprint for tuning IL-27 pleiotropic actions.


Introduction
IL-27 is an immunosuppressive cytokine involved in resolving T cell-mediated inflammation.

IL-27 inhibits Th-17 responses 1-3 and induces differentiation of T regulatory cells 4 . T cell
stimulation by IL-27 promotes secretion of the anti-inflammatory cytokine, IL-10, further contributing to a reduction in the inflammatory response 5 . Together, these properties make IL-27 an attractive drug target to treat T cell-mediated inflammatory disease. IL-27 also contributes to immune exhaustion by regulating expression of co-inhibitory receptors 6,7 .
Elevated levels of IL-27 gene signatures are found in cancer and are associated with poor prognoses 8 . Understanding how IL-27 engages its cellular receptors to form a signalling complex will provide fundamental insight into immune regulation, which could facilitate the development of new therapeutics that target IL-27 responses.
IL-27 is a heterodimeric cytokine composed of two gene products, Epstein-Barr virusinduced gene 3 (EBI3) and IL-27p28 (p28) 9 . Grouped within the IL-12 family of heterocytokines, p28 and EBI3 share sequence homology with two components of IL-12: p35 and p40, respectively 10 . Unlike other heterodimeric cytokines (IL-12 and IL-23) which activate Signal Transducer and Activator of Transcription (STAT) STAT3 and STAT4, IL-27 primarily induces activation of STAT1 and STAT3 pathways 11 . These differences can be attributed to the engagement of GP130 by IL-27, which is shared across the IL-6 cytokine family 12 .
IL-27 bridges two major cytokine families, raising the question of whether competition or synergism could occur in dimer formation between them. Under some conditions, p28 and EBI3 can be secreted independently and associate with other proteins to induce differential responses [13][14][15][16][17] . Promiscuity in chain combinations is characteristic of the IL-6/IL-12 family and allows the system to produce different biologically active factors starting from relatively few precursor molecules.
GP130 binds a range of cytokines to elicit diverse cellular responses 18,19 . The receptor has two cytokine binding sites. The first is located within the elbow between its cytokine homology regions (CHR) and has been shown to engage site 2 of cytokines, including IL-6 20 , Leukemia inhibitory factor (LIF) 21 and Ciliary neurotrophic factor (CNTF) 22 . The Nterminal immunoglobulin (Ig) domain of GP130 comprises an additional cytokine-binding interface that recognises an epitope at site 3, resulting in higher order assemblies with different stoichiometries 20,23 . In the hexameric GP130:IL-6 signalling complex, two molecules of GP130 dimerize to trigger JAK/STAT activation 23 . Here, the two low affinity site 3 interfaces are necessary to stabilize the complex and trigger receptor activation by IL-6. For GP130 complexes with LIF and CNTF cytokines, site 3 is occupied by the co-receptor LIF-R 22,24 . Structural details for how GP130 forms a heterodimeric signalling complex by binding site 3 of a cytokine remains unresolved.
Here, we report a 3.8 Å resolution cryo-EM structure of IL-27 in complex with cytokine binding domains of GP130 and its co-receptor IL-27Rα. This structure reveals a modular assembly mechanism that differs from those described for other members of the IL-12 and IL-6 families. IL-27 binds with high affinity to IL-27Rα at site 2, which is further stabilized by electrostatic interactions between IL-27Rα and EBI3. Subsequently, GP130 is recruited to site 3 of the cytokine in a second step. Our structural and biochemical data explain how IL-27 coordinates two signalling receptors to modulate T cell mediated inflammation. Our results provide a blueprint for developing new therapeutics that target and tune IL-27 responses.

Molecular architecture of the IL-27 cytokine recognition complex
IL-27 signals through dimerization of the cognate receptor IL-27Rα and the shared receptor GP130 25 . Current models for the assembly of complexes mediated by IL-27 have been largely based on structural principles derived from the IL-12/IL-23 or IL-6 systems. However, given differences in signalling outcomes between IL-27 and these two families, together with the growing importance of IL-27 as a therapeutic target, we used cryoEM to solve the structure of the IL-27 cytokine recognition complex.
To overcome challenges in cytokine stability, we engineered an IL-27 cytokine variant in which the two monomeric components (EBI3 and p28) were fused by a short linker 25 . The fusion cytokine was expressed in insect cells and purified together with Domains 1-3 (D1, D2, D3) of GP130 and the first two domains of IL-27Rα ( Supplementary Fig. 1). We then used cryoEM to solve the structure of the complete IL-27 cytokine-recognition complex ( Fig.   1, Supplementary Fig. 2 and Supplementary Fig.3). We used the ab initio reconstruction protocols within cryoSPARC 26 to generate an initial model of the complex. Maps were further refined using a combination of homogenous and heterogenous refinement procedures. The complex was refined to a resolution of 3.8 Å and models were built starting from Alpha fold predictions 27 (Supplementary Table 1). The chirality of the helical bundle central to the p28 cytokine was used to assign the handedness of the reconstruction.
The overall structure of the IL-27 cytokine recognition complex exhibits a classical architecture with a helical cytokine bundle central to the assembly (Fig. 1). For the IL-27 complex, the helical bundle is sandwiched by two 'L'-shaped densities, analogous to other cytokine complexes. However, unlike previous structures, the IL-27 cytokine recognition complex has an additional flexible prong that contacts the back face of the helical bundle.

Principle binding interfaces for IL27 cytokine recognition
Cytokines from the IL-12/IL-6 family have three highly-conserved principle binding interfaces: site 1, site 2 and site 3 (Figs. 1b and 2a) 23 . For the IL-27 recognition complex, site 1 is occupied by the second half of the heterodimeric cytokine, EBI3 (Figs. 1 and 2b).
EBI3 is comprised of two Fibronectin type III (FNIII) domains bent into an 'L'-shaped arrangement. While we observe some flexibility in the way EBI3 rotates with respect to p28, the interaction interface between the two components of the cytokine is well-resolved. In our map, the bend of the EBI3 elbow is formed by a cluster of aromatic residues (EBI3: Y39, P40, F96, F159, and Y209), which create a hydrophobic groove recognized by p28:W93 ( Fig. 2b). This feature of knobs and holes shape recognition is consistent with other cytokine binding interfaces 23 . Our p28:EBI3 interface is supported by mutagenesis studies showing that equivalent residues in the human cytokine (EBI3:F97; p28:W97) are essential for cytokine complex formation 28 . The same study identified an aspartic acid on EBI3 that also influences stability of the heterodimer 28 . In our structure, the equivalent murine residue (EBI3:D205) is nearby an arginine on p28 (p28:R217) and may form an additional salt bridge to stabilize the orientation of the heterodimer (Fig. 2b).
For the IL-6:GP130 cytokine recognition complex, site 2 of IL-6 is occupied by the two CHR domains of GP130 (D2 and D3) 20 . To understand how GP130 recognizes chemically unique cytokines, we next investigated the IL-27 site 2 interface. By contrast to the IL-6 recognition complex, we find that site 2 of p28 is formed by the apex of the elbow region between the two CHR domains of IL-27Rα (Figs. 1 and 2c). Here we find that the knob and hole pattern is encoded by aromatic residues of IL-27Rα (IL-27Rα: P153, P152, and W151), which form a pocket to bind a tyrosine sidechain extending from p28 (p28:Y48) (Fig. 2c). In the IL-6:GP130 cytokine recognition complex, the equivalent tyrosine on IL-6 slots into a hydrophobic groove formed by the interface between D2 and D3 of GP130 20 .
The position of EBI3 at site 1 and IL-27Rα at site 2 are coordinated by a second interaction interface. EBI3 directly contacts the second CHR domain of IL-27Rα. By contrast to the hydrophobic knob and hole recognition motifs observed for p28, this secondary interface is

Kinetic drivers of IL-27 signalling complex assembly
To understand the kinetic drivers underpinning assembly of the IL-27 signalling complex we defined the binding affinities for each of the subcomponents using surface plasmon resonance (SPR). First, we set out to identify which of the two signalling receptors bound the IL-27 heterodimer with higher affinity. We immobilized biotinylated IL-27Rα or GP130 on a streptavidin SPR surface and passed a range of IL-27 concentrations to measure rates of association and dissociation (Figs. 3 a,b). We found that IL-27Rα has a higher affinity (K D : 0.29 nM) for the heterodimeric cytokine than GP130 (K D : 3.1 nM), in agreement with our structural data showing that GP130 occupies the low-affinity binding site 3. These results suggest a two-step binding mode for IL-27 complex formation, where IL-27 binds IL-27Rα with high affinity in a first step, and subsequently recruits GP130 with lower affinity to form the signalling complex. We next wanted to understand the roles of EBI3 and p28 in the kinetics of complex formation. As we were unable to produce recombinant EBI3, we focused our analysis on p28. We quantified the binding affinity of p28 for each signalling receptor immobilized to the streptavidin SPR chip (Fig. 3c, d). We observed only weak binding of p28 to IL-27Rα (K D : 2.8 μM) and no binding to GP130 at the doses tested, consistent with weak signalling responses elicited by p28 when compared to IL-27 in CD8 T cells (Fig. 3e, f).
Taken together, these data support a stabilizing role of EBI3 in formation of the IL-27 signalling complex.

Discussion
IL-27 is a model system for signalling by heterodimeric cytokines, including IL-12 and IL-23.
A comparison of the heterodimeric IL-23 receptor complex (PDB: 6WDQ) 30 with our structure shows that EBI3 and the IL-12β subunit (p40) overlay directly with site 1 (Fig. 4a).
In addition, we observe that the position of GP130 at IL-27 site 3 is similar to the orientation of the IL-23 signalling receptor. The arrangement of these extracellular domains likely directs the orientation of intracellular regions that activate the JAK/STAT pathway. Therefore, it may be possible that the bend between D2 and D3 of GP130 in the signalling receptor is a topological requirement for activation by IL-12, IL-23 and IL-27.
IL-27 is also a central member of the IL-6 family of cytokines that signal through GP130 31 .
The diversity in cellular responses within the family stems in part from the strict transcriptional control of cytokines secreted by different cell types 32 . Although IL-27 is a heterodimeric cytokine, there are some instances whereby the two components are independently expressed 33,34 . In the absence of EBI3, p28 can act as an antagonist for IL-6mediated GP130 signalling 13 . Our biophysical data show that binding of p28 alone to GP130 is negligible (Fig. 3). A comparison of the IL-6 cytokine recognition complex (PDB:1P9M) with our structure shows that binding of EBI3 to site 1 overlaps with site 1 of the IL-6 ( Fig.   4b). Previous studies have reported that p28 can bind IL-6Rα 17 . Though signalling by the putative p28:IL-6Rα is weaker compared to IL-6 or IL-27, it may be possible that IL-6Rα engages p28 at site 1 with low affinity to produce weak agonistic or antagonistic activities in certain contexts.
Chain sharing is a common theme among cytokine receptors 9 and contributes to the layers of complexity derived from relatively few building blocks. Although both IL-27 and IL-6 engage GP130, they do so through different interaction interfaces. One important difference between these two structures occurs at site 2 (Fig. 4b). While the two CHR domains of GP130 (D2 and D3) bind IL-6 at site 2, the equivalent position in the IL-27 system is occupied by IL-27Rα. Unlike the IL-6 co-receptor (IL-6Rα), the intracellular domains of GP130 and IL-27Rα both associate with JAKs proteins and activate signalling upon dimer formation. In the case of IL-27, the cognate co-receptor IL-27Rα could fill an analogous role to GP130 at site 2 in the IL-6-signaling complex. In both systems site 1 is occupied by nonsignalling proteins: EBI3 and IL-6Rα. Our structural and biophysical data support role for auxiliary proteins at site 1 to enhance the stability of the signalling complex, while receptors occupying site 2 are responsible for signal transduction.
Signalling through the JAK/STAT pathway requires dimerization of signalling receptors. How that dimerization is achieved, and how the geometry of the signalling complex contributes to the functional diversity of the cascade, remains an open question. There is a wide range of oligomeric assemblies observed for GP130-cytokine complexes 20,23 . For IL-6, dimerization of GP130 occurs through a hexameric assembly (2 copies of the IL-6:IL-6Rα:GP130 complex). Within the hexamer, GP130 bound to IL-6 site 2 dimerizes with a second copy bound to site 3 of the additional cytokine (Fig. 4c). In the IL-27 signalling complex, dimerization of IL-27Rα and GP130 may occur from receptors occupying sites 2 and 3 from the same cytokine, though additional cytokine-receptor stoichiometries cannot be ruled out.

Protein expression and purification
Murine IL-27 was cloned as a linker-connected single-chain variant (p28+EBI3) as described in 36 (Supplementary Fig. 1a). In the text we refer to this variant only as IL-27. Murine IL-27Rα (amino acids 28-224) and GP130 (amino acid 23-319) ectodomains were cloned and expressed as described in 37 . Briefly, protein sequences were cloned into the pAcGP67-A vector (BD Biosciences) in frame with an N-terminal gp67 signal sequence, driving protein secretion, and a C-terminal hexahistidine tag. Baculovirus stocks were produced by transfection and amplification in Spodoptera frugiperda (Sf9) cells grown in SF900III media (Invitrogen) and protein expression was carried out in suspension Trichoplusiani ni (High

Five) cells grown in InsectXpress media (Lonza).
Protein purification was carried out using the method described in 38  To generate biotinylated proteins for surface plasmon resonance studies the GP130 sequence was subcloned into the pAcGP67-A vector with a C-terminal biotin acceptor peptide (BAP)-LNDIFEAQKIEWHW followed by a hexa-histidine tag. The purified protein was biotinylated with BirA ligase following a previously described protocol 38 . IL-27Rα was Nterminal biotinylated in vitro using EZ-Link Sulfo-NHS-SS-Biotin (Pierce) at pH 6.5.

Surface plasmon resonance
Surface plasmon resonance was used to determine the binding affinity of the recombinantly produced IL-27 and p28 to IL-27Rα and GP130. Biotinylated IL-27Rα and GP130 were immobilised onto the chip surface with streptavidin. Series S Sensor SA (GE Healthcare) chips were primed in 10 mM HEPES, 150 mM NaCl, 0.02% TWEEN-20, prior to immobilisation of the biotinylated receptor. Analysis runs were performed in 10 mM HEPES, 150 mM NaCl, 0.05% TWEEN-20 and 0.5% BSA. A Biacore T100 (T200 Sensitivity Enhanced) was used for measurement and Biacore T200 Evaluation Software 3.0 was used for data analysis.

CryoEM sample preparation and data collection
Pre-formed IL-27:IL-27Rα:GP130 complex sample was diluted to 0.1 mg/ml and vitrified in liquid ethane using Vitrobot Mark IV (Thermo Fisher Scientific). Lacey Carbon Au 300 grids (EMS) were glow discharged in air using a Cressington 208 for 60 seconds prior to application of the sample. 4µL of sample were applied to the grids at a temperature of 20°C and a humidity level of 100%. Grids were then immediately blotted (force -2, time 2 seconds) and plunge-frozen into liquid ethane cooled to liquid nitrogen temperature.
Grids were imaged using a 300kV a Titan Krios transmission electron microscope (Thermo Fisher Scientific) equipped with K3 camera (Gatan) operated in super resolution mode.
Movies were collected at 81,000x magnification and binned by two on the camera (calibrated pixel size of 1.06 Å/pixel). Images were taken over a defocus range of -0.5 µm to -2.25 µm with a total accumulated dose of 50 electrons per Å 2 using EPU (Thermo Fisher Scientific, version 2.11.1.11) automated data software. A summary of imaging conditions is presented in Supplementary Table 1.

CryoEM data processing
28,437 micrographs were pre-processing using cryoSPARC (v.3.3.1) 26 patch motion correction and patch CTF. The datasets were manually curated to remove movies with substantial drift and crystalline ice, resulting in a total of 17,228 micrographs going on for further processing. Deep learning models used in Topaz (v.0.2.5) 40 were used to automatically pick particles from these micrographs within the cryoSPARC workflow. We used the Topaz pre-trained model (ResnNet16) for particle picking in the first instance.
During the initial 2D classification, rare views were combined and used as input for a second round of Topaz training. Particles from both rounds were combined and duplicates were removed. A total of 3,575,367 particles were extracted at 2.12 Å/pixel (bin by 2) and subject to iterative 2D classification to remove ice contamination, carbon edges and broken particles. Of these, 628,386 particles were used to create an initial model using cryoSPARC's ab initio reconstruction procedure with three classes. By comparing the reconstructions, it was obvious that one class represented a subcomponent of the complex.
Particles from the other two classes (486,648 particles) showing similar structural features were merged and refined using nonuniform refinement. Particles were then re-extracted at 1.06 Å/pixel and movement within the ice was corrected using cryoSPARCs own implementation of local motion correction. The dataset was then subjected to heterogenous refinement using six classes. All six classes were individually refined using homogeneous refinement. The three classes that refined to the highest resolution were pooled (289,489 particles) and were further refined using local refinement procedures manually centered on the core. The final map resolution 3.8 Å was assessed using the gold standard FSC at a threshold of 0.143 and locally filtered using cryoSPARCs own implementation.

Model building and refinement
An initial model of IL-27 was generated by rigid body fitting each protein into the locally sharpened map generated by DeepEMhancer 41 . Rigid body fitting was performed in Chimera 42 . The models used were murine p28 (AlphaFold entry ID: Q8K3I6), murine IL-27Rα (AlphaFold entry ID: O703940), murine EBI3 (AlphaFold entry ID: O35228), and murine GP130 (AlphaFold entry ID: Q00560). Alpha fold models were trimmed to reflect the domain boundaries in the constructs used. A rigid body fit of the GP130 model resulted in clashing of GP130-D1 with p28. We therefore fitted GP130 as two rigid bodies; D1 (residues 28-123) and D2D3 (residues 124-319). The local fit of D1 and D2D3 was then optimised and the linker between the two rigid bodies was restored.
The initial model of IL-27 was refined into the cryoEM map using ISOLDE (v.1.3) 43 implemented in ChimeraX (v.1.3) 44 . During refinement we applied adaptive distance restraints to each subunit. A loop region of p28 (residues 178-194) was removed to prevent steric clashes with GP130. The atomics models were refined using phenix.real_space_refine in Phenix (v.1.20.1-4487) 45 with secondary structure, reference model, and geometry restraints. B-factors were refined in Phenix. Model FSC validation tools and the overall quality of the model were assessed in the map using the cryoEM validation tools in Phenix and MolProbity 46 (Supplementary Table 1).  While this interface is not well resolved, p28:W195 which is essential for GP130-mediated signalling 28 , is facing D1. The key residues that mediate the interactions at interfaces are shown as sticks.

Data Availability
Data supporting the findings of this manuscript are available from the corresponding authors upon reasonable request. The accession numbers for the EM map and models of the IL27receptor recognition complex reported in this paper are XXX and XXX.