Structural basis of Cullin 2 RING E3 ligase regulation by the COP9 signalosome

Cullin-Ring E3 Ligases (CRLs) regulate a multitude of cellular pathways through specific substrate receptors. The COP9 signalosome (CSN) deactivates CRLs by removing NEDD8 from activated Cullins. Here we present structures of the neddylated and deneddylated CSN-CRL2 complexes by combining single-particle cryo-electron microscopy (cryo-EM) with chemical cross-linking mass spectrometry (XL-MS). These structures suggest a conserved mechanism of CSN activation, consisting of conformational clamping of the CRL2 substrate by CSN2/CSN4, release of the catalytic CSN5/CSN6 heterodimer and finally activation of the CSN5 deneddylation machinery. Using hydrogen-deuterium exchange (HDX)-MS we show that CRL2 activates CSN5/CSN6 in a neddylation-independent manner. The presence of NEDD8 is required to activate the CSN5 active site. Overall, by synergising cryo-EM with MS, we identify sensory regions of the CSN that mediate its stepwise activation and provide a framework for understanding the regulatory mechanism of other Cullin family members.


SUMMARY
Cullin-Ring E3 Ligases (CRLs) regulate a multitude of cellular pathways through specific substrate receptors. The COP9 signalosome (CSN) deactivates CRLs by removing NEDD8 (N8) from activated Cullins. The structure of stable CSN-CRL can be used to understand this mechanism of regulation. Here we present the first structures of the neddylated and deneddylated CSN-CRL2 complexes by combining single particle cryo-electron microscopy (cryo-EM) with chemical cross-linking mass spectrometry (MS). These structures reveal a conserved mechanism of CSN activation, consisting of conformational clamping of the CRL2 substrate by CSN2/CSN4, release of the catalytic CSN5/CSN6 heterodimer and finally activation of the CSN5 deneddylation machinery. Using hydrogen deuterium exchange-MS we show that CRL2 binding and conformational activation of CSN5/CSN6 occur in a neddylation-independent manner. The presence of NEDD8 is required to activate the CSN5 active site. Overall, by synergising cryo-EM with MS, we identified novel sensory regions of the CSN that mediate its stepwise activation mechanism and provide a framework for better understanding the regulatory mechanism of other Cullin family members.
(designated as CSN1-8 by decreasing molecular weights of 57-22 kDa), and is organized in a "splayed hand" architecture, which has high sequence and structural homology to the proteasome lid 13,14,[16][17][18] . CSN1, 2, 3, 4, 7 and 8 are structurally homologous to each other and together contribute to the fingers of the "splayed hand" which arise from extended N-terminal α-helical repeats 16,18 . CSN5 and 6 are also closely related structurally and form a globular heterodimer located on the palm of the hand. CSN5 is responsible for the deneddylase activity of the CSN. A ninth subunit, CSNAP, has recently been identified, and is thought to play a role in stabilising the CSN complex 18 .
Electron microscopy (EM) based structural analysis has provided important insights into the mechanism of CRL1 regulation by CSN 16,19 . CRL4A and CRL3 have also been observed to form such complexes 20 . However, despite intense interest, structural information of the CSN bound to CRLs remains limited to CRL1 12,16,19 , CRL4A 20 , and a low-resolution map of a dimeric CSN-CRL3-N8 20 complex. In particular the analysis of the CSN-CRL4A complex 20 identified at least three major steps by which CRL~N8 is deneddylated by the CSN. In the first step, the extended N-terminal helical modules of CSN2 and CSN4 conformationally "clamp" the C-terminal domain of the CRL4A~N8 and RBX1 16,19,20 . The second step involves the release and consequent relocation of CSN5/CSN6 closer to NEDD8, brought about by disruption of the CSN4/CSN6 interface 20 . Disrupting the binding interface between CSN4/CSN6 through removal of the CSN6 insertion-2 loop (Ins-2), resulted in enhanced deneddylase activity 13 , presumably due to more complete release of CSN5/CSN6. In the final step, the mobile CSN5 binds to NEDD8, leading to deneddylation via its JAB1/MPN/MOV34 (JAMM) metalloprotease domain 21 . The JAMM motif consists of H138, H140 and D151 zinc-coordinating residues, and residue E104 of the CSN5 insertion-1 loop (Ins-1) 13 . In apo-CSN, the Ins-1 loop occludes the CSN5 active site, auto-inhibiting the deneddylase 13,22 . Deneddylation is also severely diminished by a H138A point mutation in CSN5 13 .
Surprisingly, the CSN can also form complexes with each of the Cullin 1-5 family members even without NEDD8 23 . Free CRLs such as CRL1 have been reported to readily bind and inhibit the CSN, albeit at relatively lower affinity than the neddylated CRL1 24 . While the exact role of CSN-CRL complexes remains unclear, it has been hypothesised that these complexes may function to regulate the cellular level of ubiquitin ligase activity of CRLs once they have been deneddylated, effectively sequestering E3 ligases from the intracellular environment 24 .
Building on the existing knowledge of the CSN-CRL systems, here we pose the question: are similar structural changes to be found in other CSN-CRL complexes, and how does binding of neddylated CRLs lead to activation of the CSN5 catalytic site? To address this, here, we present novel structures of the CSN-CRL2~N8 complex, together with the first structure of the CSN-CRL2 deneddylation product. We complemented our cryo-EM analysis with chemical cross-linking mass spectrometry (XL-MS) allowing us to clarify the positions of particularly dynamic regions in the complexes. We used hydrogen-deuterium exchange mass spectrometry (HDX-MS) to interrogate the role of the CSN4/CSN6 interface in communicating CRL binding to the CSN5 active site. Overall, our structures of the CSN-CRL2~N8 and its deneddylation product, the CSN-CRL2, provide a molecular level understanding of how deneddylation is performed by the CSN.

Cryo-EM structures of the CSN-CRL2~N8 complex
To study the molecular interactions between neddylated CRL2 (CRL2~N8) and the CSN, we performed single-particle cryo-EM to resolve a structure of the assembled CSN-CRL2~N8 (referred to as the holocomplex) ( Fig. S1-S2). The mutation H138A in the catalytic site of CSN5 subunit makes it possible to assemble CSN-CRL2~N8 complexes in which NEDD8 remains covalently attached over the time scale for experimentation 16,19 . This mutant form of CSN was used throughout the work described below, unless otherwise specified.
Using three-dimensional classification, we were able to generate maps of three different structures: a) a holocomplex map at 8.2 Å, b) a map of the complex with little or no density for VHL at 8.0 Å, and c) a map of the complex with little or no density for CSN5/CSN6/VHL at 6.5 Å (Fig. S2, Fig. S3). The two partial complexes likely arise from compositional heterogeneity in the original samples from which the structural analysis has succeeded in isolating subpopulations.
Next, we fitted into each map, the crystallographic CSN (PDB 4D10) and a homology model of the CRL2 (including the VHL-ELOB-ELOC) using molecular dynamics flexible fitting 25 (Table 1; Methods). In the model of the holocomplex (8.2 Å), the main interactions occur between the C-terminal end of Cullin-2 and the extended N-terminal helical repeats of CSN2 and CSN4 ( Fig. 1a-b). Compared to their apo-CSN crystal structure conformations (PDB 4D10), CSN2 and CSN4 are moved by 30 and 51 Å, respectively, towards Cullin-2 (Fig. S4a). In each case, the motion is a swinging rotation about hinges located close to the CSN2 and CSN4 winged helix domains: in the case of CSN2 this is coupled with an additional rotation about the axis of the superhelix formed from its N-terminal helical repeats. In the case of CSN4 the movement is coupled to the detachment of CSN4 from the Ins-2 loop of CSN6 by ~30 Å and leads to a ~12 Å shift in CSN5 (Fig. 1c, Fig. S4b). Only minor conformational changes were found in the CSN1, CSN3, CSN7B or CSN8 subunits. In the CRL2~N8 moiety, a number of relatively small rearrangements of Cullin-2, RBX1, ELOB, ELOC and VHL subunits were observed compared to its crystal structure 26 (Fig. S4c-d).
In our EM maps and the other published CSN-CRL structures 16,19,20 , the exact position of NEDD8 and the Cullin-2 Winged-Helix B (WHB) domain were difficult to determine.
To address this limitation, we carried out XL-MS experiments on the CSN-CRL2~N8 complex using the bis(sulfosuccinimidyl)suberate (BS3) cross-linker which targets lysine sidechains (Methods). We identified a total of 24 inter-and 60 intra-protein cross-links (Table S1, Fig. S5a-b). To generate a model of the CSN-CRL2~N8, we performed cross-link guided modelling which allows the placement of the WHB, NEDD8 and VHL subunits using identified cross-links from XL-MS (Methods). We imposed a cross-link distance threshold of 35 Å which takes into account the length of two lysine side chains (15 Å), the BS3 cross-linker length (10 Å) and an extra 10 Å to allow for domain-level flexibility (Methods). Our model of the CSN-CRL2~N8 satisfies all cross-link distances (Fig. S5c). Three cross-links between Cullin-2-WHB (Cullin-2 K382 -WHB K720 , Cullin-2 K382 -WHB K677 and Cullin-2 K433 -WHB K677 ) were used for the positioning of the WHB domain (Fig. S5d, red text). A further two cross-links between Cullin-2 and NEDD8 (Cullin-2 K382 -N8 K33 and Cullin-2 K433 -N8 K6 ) allowed us to place NEDD8 near CSN5 (Fig. 1d, Fig. S5d, green text). In this conformation, the isopeptide bond of NEDD8 is juxtaposed towards to the CSN5 active site. For the isopeptide bond of NEDD8 to reach the CSN5 active site, the Cullin-2 WHB domain must be extended from its crystallographic conformation towards the CSN5 by 19 Å (Fig. S5e).

Structure of the deneddylated CSN-CRL2 complex
Having determined the structure of the CSN-CRL2~N8 complex, we next sought to detail any conformational differences in the deneddylated CSN-CRL2. The affinity of CSN for the non-neddylated CRLs is significantly lower than for the neddylated forms, limiting the yield of the desired product 19 . We used native MS experiments to confirm the formation of the complex (Fig. S6). We next resolved a cryo-EM map of the CSN-CRL2 to 8.8 Å resolution (Fig. S7). Although the resolution of the CSN-CRL2 map is similar to that for CSN-CRL2~N8 holocomplex, we only observed partial density for VHL and CSN4. This is most likely due to the reduced stability of the CSN-CRL2 assembly in the absence of NEDD8. We were however able to use the density obtained to model the CSN-CRL2 structure, which has principally the same topology as the neddylated holocomplex (Fig. 2a). Using the same procedure as for the neddylated CSN-CRL2~N8 complex, we fitted CSN and CRL2 subunits into the density map of the CSN-CRL2 complex (Methods). We then utilised cross-link guided modelling to establish the position of the WHB which lacked clear density, similar to the neddylated holocomplex ( Fig. S8; Methods). To determine any local changes across the CSN-CRL2~N8 in the absence of NEDD8, we aligned the cryo-EM models of neddylated and deneddylated holocomplexes using the C-terminal helical bundle as a reference point (Fig. S9). We systematically compared the conformations of each subunit ( Fig. S9b-i). Compared to its structure in the CSN-CRL2~N8, the N-terminal helices of CSN2 are shifted by ~21 Å towards the Cullin-2 C-terminal domain (Fig. 2b, Fig. S9c). This change in CSN2 in the absence of NEDD8, leads to a structural difference in Cullin-2 which rotates upwards towards the rest of the CSN by 20 Å (Fig. 2c, Fig. S9a). The position adopted by Cullin-2 in the deneddylated holocomplex, places ELOB closer to CSN1, forming a CSN1-ELOB interface (Fig. 2d). Interactions between substrate adaptor complex and CSN1 have similarly been reported for the CSN-CRL1~N8 16 and CSN-CRL4A~N8 20 . RBX1 remains clamped between CSN2 and CSN4 (Fig. 2e).
The most striking conformational differences were observed in CSN6 (Fig. 2f, Fig.   S9g). In the absence of NEDD8, CSN6 is dramatically shifted away from its position in the neddylated holocomplex by ~40 Å (Fig. 2f). This previously unknown conformation of CSN6 appears to be unique, differing from the conformation captured in our neddylated holocomplex, and the CSN-CRL4A~N8 and CSN-CRL3~N8 19 structures. Similar to the neddylated holocomplex, no significant changes were identified in CSN3, CSN7B and CSN8 subunits. Overall, comparison between the CSN-CRL2~N8 and CSN-CRL2 structure reveal significant conformational rearrangements in CSN5/CSN6 and the N-terminal domain of CSN2.

HDX-MS reveals a stepwise mechanism of CSN activation
Having . These observations likely identify the hinge points which permit the bending of CSN2 to clamp onto the Cullin-2 C-terminus ( Fig. 3b; Fig. S12a-b). Colour scheme follows that of (b).

Conformational remodelling of the CSN5 active site is achieved in the presence of NEDD8
We next considered the release mechanism of the CSN5/CSN6 subunits of both the neddylated and deneddylated holocomplexes. In both Δ(CSN-CRL2~N8 -CSN) and Δ(CSN WT -CRL2 -CSN WT ) experiments, the Ins-2 loop of CSN6 was destabilised, correlating with the release of CSN6 from its interface with CSN4 and in line with the allosteric activation mechanism of CSN by CRL4A 19 (Fig. 4a-b i). An interesting difference between the neddylated and deneddylated complexes is that the CSN6 α4 helix is destabilised only in the absence of NEDD8 (Fig. 4b i). Similarly, the CSN5 α7 helix is also destabilised in both neddylated and deneddylated conditions ( Fig. 4a-b ii). The CSN6 α4 and CSN α7 helices are topologically knotted in the CSN5/CSN6 heterodimer and tether the globular domains of CSN5/CSN6 to the C-terminal helical bundle 13 . These observations suggest that structural changes are required in the helical knot to bring about release of the CSN5/CSN6 globular domains from their apo conformation.
Another finding is that we identified destabilisation in the Ins-2 loop of CSN5 ( Fig. 4ab ii). The Ins-2 loop of CSN5 has a lesser understood role in CSN activation. In isolated CSN5, the Ins-2 loop is highly disordered 22 (Fig. S13a), while it folds into a helical-loop structure when incorporated into the CSN 13 (Fig. S13b). Accompanying the changes in the CSN5 Ins-2 loop, in both comparative HDX-MS experiments, we detected destabilisation of the α5 helix area which surrounds the CSN5 active site ( Fig. 4a-b ii). The changes in both the CSN5 Ins-2 and α5 helix indicate a major conformational remodelling in the area adjacent to the CSN5 active site, which can be triggered through the binding of both CRL2 or CRL2~N8 to the CSN in a NEDD8independent manner. It is only in the presence of NEDD8, that the CSN5 active site is further destabilised suggesting that in a final activation step, NEDD8 induces conformational changes in the active site itself (Fig. 4a-b iii, Fig. S14).

DISCUSSION
Here we have combined EM and MS analyses to provide new insights into the mediation of CRL2 by the CSN. We have described the first structures of CSN-CRL2~N8 and a novel deneddylated CSN-CRL2 structure. As far as we are aware, this is also the first structural description of a non-neddylated CSN-CRL complex.
Furthermore, we combined cryo-EM maps with comparative HDX-MS to expand on the stepwise activation mechanism of the CSN, involving a conformational network of both NEDD8-independent and dependent stages. We suggest that the steps which lead to deneddylation are mostly NEDD8-independent, except for the remodelling of the CSN5 active site which requires NEDD8 to encounter the CSN5 active site.
Our map of the deneddylated CSN-CRL2 holocomplex represents a complex in which the CSN is still associated with its CRL2 reaction product. Resolving this structure has provided several important details into how activation of the CSN is achieved. Our comparison of the neddylated and deneddylated holocomplex structures indicated that the CSN2 contacts the CRL2 C-terminal domain in a slightly different conformation to when the CRL2 is modified with NEDD8. Between both neddylated and deneddylated conformations, the clamping by CSN2 involves destabilisation of the CSN2 helical modules 6-9 which function possibly as a hinge that allows the CSN2 to bend upwards towards the CRL2. The plasticity of these N-terminal helices presumably permits the binding of deneddylated and alternative Cullin isoforms. HDX-MS further indicates that RBX1 and CSN4 form an interface, which is more prominent in the absence of NEDD8, as shown through stabilisation of the two interfaces (Fig. S12b). Overall, the conformational variations seen in the CSN2 N-terminal helical modules (Fig. 2b), the bend of CRL2 (Fig. 2c) and the HDX differences in CSN4/RBX1 (Fig. S12b) (Fig. 5). In the first activation step, the CSN and CRL2~N8 associate through major conformational changes in CSN2 and CSN4, which clamp onto the CRL2 (Fig. 5a-b). The dramatic change in CSN4 breaks its interface with CSN6 through the CSN6 Ins-2 loop and releases the CSN5/CSN6 heterodimer (Fig. 5c) Up to here, the changes experienced by the CSN can be brought about in a NEDD8independent manner. In the next stage, the presence of NEDD8 acts as a selectivity filter which results in remodelling of the CSN5 active site itself (Fig. 5d). These changes presumably expose the CSN5 JAMM ligands of the metalloprotease site and allow subsequent deneddylation to occur (Fig. 5e). Finally, deneddylation ensues with the cleavage of NEDD8 from CRL2 and the dissociation of the complex (Fig. 5f). The fact that CSN can then re-associate with its CRL2 reaction product following dissociation, as shown by our study and structure of the CSN-CRL2, suggests that this complex may serve a lesser understood role in the CSN-CRL network. Furthermore, there may be additional roles for the CSN as suggested by the compositional plasticity seen in our CSN-CRL2~N8 classes. In our map of the CSN-CRL2~N8, the Cullin arm was observed to shift downwards away from CSN3 in maps deficient of the VHL substrate receptor (Fig. S15) Overall, our study has provided unprecedented level of detail at the role of CRL2 and NEDD8 on regulating the activation mechanism of CSN. We have revealed that the series of mechanistic responses of the CSN that lead up to deneddylation, can be triggered even by the CRL2 reaction product in a completely NEDD8-independent manner. The presence of NEDD8 on the activated CRL2 substrate, triggers remodelling within the catalytic site of the CSN5 subunit during the final stage of CSN activation. We envision that our results will have implications for the entire family of CRL proteins and their regulatory relationship with the CSN. Our study therefore provides a template not only for assisting investigations of other CRL-based systems but also for bringing together data from different structural biology techniques that otherwise will be reported independently.

Cryo-EM of CSN-CRL2~N8
The CSN-CRL2~N8 complex was formed by incubation between CRL2~N8 (1.1x molar excess) and CSN at room temperature for 90 min. The preparation (~0.5 MDa) was subjected to size exclusion chromatography using a Superose 6 Increase 3.  37 and subsequent CTF estimation of micrographs was performed using CTFFIND4 38 (Fig. S1). Auto-picking selected ~317,000 particles from ~3100 micrographs. Particles were subjected to reference free 2D classification to assess data quality and to remove contaminants selected by auto-picking. This process reduced the particle number to ~250,000. Following particle selection through 2D classification, particles were divided into 15 3D classes. Three of these classes (~69,000 particles) were selected for further classification and processing, as described in Fig. S2.

Cryo-EM of CSN-CRL2
The CSN-CRL2 complex was formed by incubation between CRL2 (1. CTF estimation of micrographs was performed using CTFFIND4 38 . Auto-picking selected ~309,000 particles from ~6800 micrographs. Particles were subjected to reference free 2D classification to assess data quality and to remove contaminants selected by auto-picking. This process reduced the particle number to ~208,000. Following particle selection through 2D classification, particles were divided into 6 3D classes first with alignment, then subsequently without alignment with a mask around CSN5/CSN6 in order to perform focused classification on this area. The map that showed the greatest recovery of detail for CSN5/CSN6 (Fig. S7) was then subjected to 3D auto refinement and post-processing. Local resolution was estimates using ResMap 40 as part of the RELION wrapper.

Homology modelling of the CRL2
Homology modelling of the CRL2 was necessary, due to a combination of missing

Model fitting of EM maps
All models were fitted using CSN subunits sourced from the 4D10 crystal structure  Table 1 of the manuscript.

Native mass spectrometry
All spectral data were collected using a SYNAPT G2-Si (Waters Corp., Manchester, UK) high-definition mass spectrometer and samples were ionized using a  47 .

Data analysis for XL-MS
Raw files were converted into Mascot generic format (mgf) files using pXtract

XL-MS guided placement of the WHB, NEDD8 and VHL subunits
Cross-links determined from XL-MS for the CSN-CRL2~N8 and CSN-CRL2 complexes were used to clarify the position of the WHB, NEDD8 and VHL subunits which lacked clear density in our cryo-EM maps. We performed XL-guided placement using the Integrative Modelling Platform (IMP) 50  Our modelling procedure utilized two types of cross-links. The first type are pseudocross-links which maintain the correct topology of the complex: a single pseudo-crosslink between Cullin-2 T655 to WHB T656 of 5 Å to mimic a covalent bond, and connections between VHL-ELOB, VHL-ELOC and VHL-Cullin-2 to maintain integrity of the VHL-ELOB-ELOC adaptor complex and its interface with Cullin-2. A single pseudo-crosslink of 10 Å was used to mimic the isopeptide bond of WHB K689~N 8 G76 (7.5 Å lysine side chain + ~3 Å glycine C-terminus). The second type are cross-links determined experimentally between WHB, NEDD8 and VHL with its surrounding subunits (Table S1-S2) which utilized a distance threshold of 35 Å (two lysine side chains at 15 Å, BS3 linker length at 10 Å, plus 10 Å for flexibility). IMP was parametrized to perform 1000 iterations, with each iteration randomly moving WHB, NEDD8 and VHL relative to the stationary CSN and CRL subunits. IMP parameters used were num_mc_steps = 10, rb_max_trans = 2, rb_max_rot = 0.1, bead_max_trans = 0.5 and excluded volume restraint resolution = 20. The single best model was evaluated by projecting all crosslinks for the complex onto the structure and confirming that all distances were below the 35 Å distance threshold. A table of cross-links can be found in Supplementary   Table S1 for the CSN-CRL2~N8 and Table S2 CSN-CRL2. Additionally see Fig. S5 and S8.