Structure of CRL7FBXW8 reveals coupling with CUL1–RBX1/ROC1 for multi-cullin-RING E3-catalyzed ubiquitin ligation

Most cullin-RING ubiquitin ligases (CRLs) form homologous assemblies between a neddylated cullin-RING catalytic module and a variable substrate-binding receptor (for example, an F-box protein). However, the vertebrate-specific CRL7FBXW8 is of interest because it eludes existing models, yet its constituent cullin CUL7 and F-box protein FBXW8 are essential for development, and CUL7 mutations cause 3M syndrome. In this study, cryo-EM and biochemical analyses reveal the CRL7FBXW8 assembly. CUL7’s exclusivity for FBXW8 among all F-box proteins is explained by its unique F-box-independent binding mode. In CRL7FBXW8, the RBX1 (also known as ROC1) RING domain is constrained in an orientation incompatible with binding E2~NEDD8 or E2~ubiquitin intermediates. Accordingly, purified recombinant CRL7FBXW8 lacks auto-neddylation and ubiquitination activities. Instead, our data indicate that CRL7 serves as a substrate receptor linked via SKP1–FBXW8 to a neddylated CUL1–RBX1 catalytic module mediating ubiquitination. The structure reveals a distinctive CRL–CRL partnership, and provides a framework for understanding CUL7 assemblies safeguarding human health.

in various assemblies remains unclear. What is clear is that CUL7, RBX1 and CUL9 interact. Also, CUL7-RBX1 binds OBSL1 and CCDC8 in a complex named '3M' in reference to mutations in the CUL7, OBSL1 or CCDC8 genes causing 3M syndrome. The 3M syndrome is characterized by short stature, and distinctive facial and skeletal abnormalities. The CCDC8 protein was named for its predicted coiled-coil domain, is thought to contain several intrinsically unstructured regions that mediate protein interactions, and localizes at the plasma membrane 15,[21][22][23][24] . OBSL1 is a putative cytoskeletal adapter protein with multiple immunoglobulin-like domains and a fibronectin type-3 domain 25 . CCDC8 binds to OBSL1, which leads to assembly of CUL7-OBSL1-CCDC8 into a membrane-associated 3M E3 ligase complex and ubiquitination of the plasma membrane cell migration regulatory protein PHLDB2 (also known as LL5β) 26 . In addition, CUL7 and CUL9 also interact with TP53 (refs. 21,27,28 ). Notably, CUL7 and/or FBXW8 have been implicated in the ubiquitination of a variety of proteins with diverse roles in cellular signaling and proliferation, including TP53 (refs. 17,21,29 ). It is unknown whether CUL7's interaction and functions in partnership with SKP1-FBXW8, with OBSL1-CCDC8, with CUL9, and with TP53 are independent of each other, but the distribution of these proteins copurifying with each other in various immunoprecipitations suggests that there are multiple distinct CUL7-containing E3 complexes in vivo 26 , none of which are understood structurally. CRL7 FBXW8 might seem to parallel canonical CRLs, in that it comprises both a CUL-RBX subcomplex and a SKP1-F-box protein subcomplex. However, CUL7 is unusually large (≅190 kDa), and the CRL7 FBXW8 complex displays many atypical features. CUL7 has a distinctive ≅100 kDa N-terminal region, homologous only to a similar region in CUL9. This region is predicted to contain armadillo repeats, a ≅15 kDa DOC domain of unknown function and a ≅11 kDa CPH domain (a conserved domain within CUL7, CUL9 and HERC2 proteins) that binds TP53 (ref. 28 ). Beyond the structure of the CPH domain and characterization of its interactions with TP53 by NMR 30 , little is known of the detailed interactions constituting or made by CRL7 FBXW8 . Moreover, features of CRL7 FBXW8 differ substantially from the well-studied CRL1 family wherein SKP1 connects CUL1-RBX1 and an F-box protein. First, whereas CUL1-RBX1 binds nearly 70 different human SKP1-F-box protein modules, CUL7-RBX1 exclusively associates with SKP1 bound to FBXW8 (ref. 10 ). Second, the sequence of CUL7 is devoid of the CUL1 domain that binds the SKP1-F-box motif raising the question of how it binds SKP1-FBXW8 (ref. 6 ). Third, assays of CUL7 immunoprecipitates from cells cotransfected with wild-type and mutant expression plasmids for CUL7 and TP53 showed CUL7-mediates ubiquitination of TP53 in a manner that, perplexingly, does not require its cullin domain-bound RING-domain partner RBX1 (ref. 21 ). Fourth, along the same lines, CUL7 copurifies with NEDD8 but does not appear to be neddylated itself 14,31 . Finally, in vivo, CRL7 FBXW8 curiously associates with CUL1-RBX1. This interaction requires FBXW8: in wild-type mouse embryonic fibroblasts, exogenously expressed Flag-tagged CUL1 immunoprecipitates endogenous CUL7 and Flag-tagged CUL7 immunoprecipitates endogenous CUL1. However, the CUL7-CUL1 interactions were not observed in FBXW8 null cells 18 . Although a number of models have been proposed 18,32,33 , it remains unknown how FBXW8 brings together multiple different cullin molecules, and why specifically CUL7 and CUL1. This raises questions of whether and how CUL7 could serve as an E3 ligase. Here, we answer these questions with the cryo-EM structure of CRL7 FBXW8 and structure-based biochemical studies showing its unique architecture and assembly with neddylated CUL1-RBX1 into an active E3 ubiquitin ligase.

Results
CRL7 FBXW8 forms a unique cullin-RING ligase assembly. A cryo-EM reconstruction of a recombinant human CRL7 FBXW8 complex, prepared by coexpressing CUL7, RBX1, SKP1 and FBXW8 in human human embryonic kidney 293S (HEK293S) cells, refined to 2.8 Å resolution overall. The density allowed fitting previous structures of RBX1 and SKP1, and building atomic models of CUL7 and FBXW8 de novo. The final coordinates were generated after multiple cycles of manual rebuilding and refinement (Fig. 1, Table 1, Extended Data Fig. 1 and Extended Data Table 1). Although AlphaFold was released subsequent to structure determination, it is noteworthy that most of the individual domains in CUL7 and FBXW8 superimpose with those predicted with Cα-r.m.s.d. values of <1.3 and <0.8 Å, respectively (Extended Data Fig. 2) 34 .
CRL7 FBXW8 adopts a diagonal T-shaped structure. CUL7-RBX1 forms a ≅160 Å long bar that crosses the ≅115 Å long CUL7-FBXW8-SKP1 base of the 'T' at a ≅50° angle (Fig. 1a). The T-junction is formed by five domains from CUL7 wrapping around FBXW8's WD40 β-propeller and stalk helix (Fig. 1b). SKP1's N-terminal region projects away from CUL7-RBX1 and was poorly resolved in the maps, but was visualized at the edge of the complex by docking previous SKP1-F-box protein complex structures into the low-resolution density (Extended Data Fig. 3a) RBX's RING domain and building of residues for most of the protein complex (Extended Data Fig. 3b-e). The CRL7 FBXW8 T-shape differs from the more ovoid assemblies formed by canonical CRLs (Fig. 1c). CUL7 does not directly bind SKP1. This is explained by CUL7 lacking a cullin-repeat 1 (CR1) domain that is used by cullins 1-5 to bind their cognate substrate receptor modules. Notably, CR1 domain interactions are central to the characteristic architecture of prototypic CUL1-based E3 ligases 6 . CUL1-RBX1 forms half an oval. At one end, CUL1's CR1 domain and SKP1 curve around to place the F-box protein's substrate-binding domain facing RBX1 bound at the other end of CUL1. The oval is closed when RBX1's RING domain binds a ubiquitin-carrying enzyme to deliver ubiquitin to an F-box protein-bound substrate 35,36 (Fig. 1c). Below, we describe the distinct structural details of CRL7 FBXW8 , and how E3 ligase activity can be achieved despite its unique assembly.

SKP1-FBXW8 resembles canonical SKP1-FBXW assemblies.
Within CRL7 FBXW8 , although SKP1 does not contact with CUL7, the SKP1-FBXW8 subcomplex shows all the hallmark features of canonical SKP1 complexes with WD40 domain-containing F-box proteins: (1) the compact three-helix F-box is encased by SKP1's C-terminal helices (Extended Data Fig. 4a); (2) a ≅30 Å long stalk helix rigidly projects from the F-box and (3) the stalk helix continues into a β-propeller (Extended Data Fig. 4b,c). FBXW8's eight-bladed β-propeller overall superimposes well with those of other F-box proteins, most notably human FBXW7 (Protein Data Bank (PDB) 2OVP) and yeast Cdc4p (PDB 1NEX) (Extended Data Fig. 4b) 37,38 . As is typical of WD40 proteins, the β-propeller is closed by the N-terminal strand completing the otherwise C-terminal-most blade (Fig. 1d). However, FBXW8 is unique in that its C-terminal ≅35 residues extend beyond the propeller to form a triangular platform between the propeller and stalk helix. This C-terminal structure makes three-way interactions between FBXW8's first propeller blade and two disparate CUL7 domains (Fig. 1b).
Unique domains and cullin-RING arrangement in CUL7-RBX1. The CUL7-RBX1 portion of the structure contains four distinct regions (Fig. 2a,b). CUL7's atypical N-terminal region and C-terminal cullin region bound to RBX1 together form the elongated bar shape that crosses the 'T' . CUL7's N-terminal region comprises a unique collection of domains that cap the N-terminal edge of the cullin region. A small beta-domain (SBD) connects via a poorly resolved 50-residue linker to the first of three short armadillo repeat units (ARM1, ARM2 and ARM3). The ARM3 unit covers the seam between the ARM1 and ARM2 domains, and proceeds into the first helix of the cullin region. The junction between the N-and C-terminal regions is mediated by a 64 Å long bent helix that is both the last helix of the ARM3 unit and the first helix of the cullin region. Altogether, the SBD, ARM1, ARM2 and ARM3 domains pack against each other to wrap around and completely encircle this junction helix. The N-terminal-most domain in the cullin region (the CR2 domain) is embraced on one side by the SBD and on the other by the ARM2 domain (Fig. 2b).
RBX1's N-terminal region forms a β-strand, which is inserted in CUL7's α/β region. Together, these CUL7 and RBX1 elements form an intermolecular cullin/RBX (C/R) domain that requires both proteins for proper folding. However, the relative orientation of RBX1's ensuing RING domain differs from the model arrangement in CUL1-RBX1 (Fig. 2d). RBX1's ten C-terminal residues extend beyond the RING domain, fill a crevice between CUL7's 4HB and WHB domains, and cement a CRL7-specific domain arrangement (Fig. 2e). Although the WHB domain is the C terminus of canonical cullins, CUL7 has an additional 50 residues not overtly visible in the cryo-EM density. It is tempting to speculate that CUL7's C terminus corresponds to patchy cryo-EM density covering RBX1's RING domain, although the map does not allow unambiguous attribution of this region (Extended Data Fig. 3b). The two other structured domains from CUL7-the ARM2-Ins (an extended loop insertion in ARM2) and the DOC domainboth bind FBXW8 (Fig. 3a). The CPH domain is not visible in the cryo-EM maps of CRL7 FBXW8 and is presumably flexibly tethered to the ARM2 domain.
Unique cullin-F-box protein assembly. CUL7 binds FBXW8 through an elaborate multipart interface that differs entirely from CUL1 interactions with F-box proteins. First, the concave surface from CUL7's CR2, CR3 and 4HB domains encases roughly half the rim of the top side of FBXW8's β-propeller (Fig. 3a). Grooves between CUL7 CR3 and 4HB domain helices embrace intra-and inter-blade loops from blades 5-8 of FBXW8 (Fig. 3b, c). Extended loops between the central strands of blades 5 and 8 secure the two edges of this interface. Second, CUL7's ARM2-Ins protrudes nearly 40 Å from the ARM2 domain, fills the groove between CUL7's CR2 and CR3 domains, meanders around the crevice between the FBXW8's β-propeller and C-terminal extension, and then returns to complete the ARM2 domain (Fig. 3c). Finally, CUL7's DOC domain nestles in a furrow between the stalk helix and C-terminal extension of FBXW8 (Fig. 3d). The atypical assembly explains why CUL7 uniquely binds FBXW8 but no other F-box proteins: the interactions are with domains, domain insertions and sequences unique to FBXW8, and even those with the WD40 domain involve residues that are not conserved even with its closest homolog FBXW7 (Extended Data Fig. 5b,c).
The structure also provides a rationale for why our purification of FBXW8 required coexpression not only with SKP1 to bind its F-box, but also CUL7-RBX1 (Methods). The interface is highly hydrophobic; it seems likely that exposure of the hydrophobic surfaces would lead to aggregation of SKP1-FBXW8 purified on its own. In total, 85 residues from CUL7 interact with 91 residuesnearly 20%-from FBXW8 to bury roughly 3,500 Å 2 of surface area (Extended Data Fig. 5d).
Comparing interface residues between the cryo-EM structure and those predicted using AlphaFold raises the possibility some key interacting regions in CUL7 and FBXW8 may fold and/or rearrange to form the interaction. In particular, the differences from the models suggest the CUL7 ARM2-Ins and DOC domain, and FBXW8 blade 5 and blade 8 loops and the C-terminal extension are considerably remodeled and/or reoriented in the complex (Extended Data Fig. 2b).
Purified recombinant CRL7 FBXW8 assembly is inactive. Canonical CRLs exhibit a wide range of E3 ligase activities, in partnership with various NEDD8-and ubiquitin-linked carrying enzymes. These activities depend on multivalent interactions, at minimum involving the RBX RING domain, an E2 and its active site-linked NEDD8 or ubiquitin, and also other contacts, for example between a cullin-linked NEDD8 and a ubiquitin-carrying enzyme 35 6 . e, C terminus of rBX1 anchored to WHB and 4HB of CUL7 shown in DeepeMhancer map.
The structure of CRL7 FBXW8 is inconsistent with even the most rudimentary of these canonical CRL NEDD8 and ubiquitin E3 ligase activities. A previous structure showed how CUL1 is neddylated on a specific lysine in the WHB domain 40 . Comparing the structure of CRL7 FBXW8 showed two features that prevent it from attaining the arrangement required for neddylation despite CUL7's WHB domain displaying a lysine corresponding to the CUL1 neddylation site. First, a long, rigid 'Helix29'-which is conserved in structures of unneddylated cullins 1-5-acts like a wand to direct the targeted lysine into the active site of the RBX1-E2~NEDD8 intermediate (Fig. 4a) 40 . However, CUL7 lacks this long, rigid Helix29. Instead, the corresponding region forms the α/β-to-WHB linker, an extended amphipathic structure not found in the other cullins (Figs. 2a,c and 4a,b). The hydrophobic face of the α/β-to-WHB linker is embedded in a hydrophobic surface from CUL7's 4HB domain (Fig. 4b,c). These interactions, along with additional contacts between CUL7's WHB domain with the 4HB domain and with RBX1's RING domain, fixes CUL7's WHB domain in an orientation incompatible with neddylation ( Fig. 4d-e). Second, previous studies also showed that neddylation requires a specific RBX1 RING-domain orientation. This is required to bind and position the E2~NEDD8 intermediate with its active site adjacent to the cullin's target lysine (K720 in CUL1, Fig. 4d) 41 . However, in CRL7 FBXW8 , RBX1's RING domain is secured in an inactive arrangement by intercalation of its C-terminal residues between CUL7's 4HB and WHB domains (Fig. 2e) Fig. 6a,b). E3 ligase activities have not been reconstituted with biochemically pure CRL7 FBXW8 . We therefore tested activity with our purified recombinant components. In canonical CRLs, RBX1 promotes NEDD8 transfer to the cullin's WHB domain lysine from either neddylating E2 UBE2M or UBE2F, while the homologous RBX2 only functions with UBE2F (ref. 42 ). Examining neddylation side-by-side with canonical cullin-RBX complexes showed that in control reactions, by the first time-point, fluorescent NEDD8 was fully transferred from UBE2M or UBE2F to CUL1-RBX1, and from UBE2F to CUL5-RBX2. However, CRL7 FBXW8 was not substantially modified by either NEDD8 E2 (Fig. 4f and Extended Data Fig. 6c).
We next probed automodification as a readout for intrinsic ubiquitin E3 activity. Under conditions when the control canonical CUL-RBX1 complexes (wild-type or neddylated) were autoubiquitinated in the presence of E1 (UBA1) and E2 (UBE2D3), CRL7 FBXW8 was relatively inactive, similar to the nonneddylatable CUL1 K720R mutant (Extended Data Fig. 6d). One possibility could be that CRL7 FBXW8 preferentially uses an alternative E2. However, autoubiquitination was also not observed with 26 tested E2s (Extended Data Fig. 6e).
In complexes with neddylated canonical cullins, RBX1's RING domain can also promote ubiquitin chain formation by UBE2G1 and by UBE2R-family E2s 43 . RBX1 associated with unneddylated canonical cullins can also promote such polyubiquitination with UBE2R1 (ref. 44 ), which is stimulated by interactions between an acidic C-terminal region of UBE2R1 and a basic canyon on canonical cullins 45 that is notably lacking in CUL7 (Extended Data Fig. 6f). These activities can be assayed even in the absence of a substrate, by monitoring transfer of fluorescent ubiquitin from a preformed E2~ubiquitin intermediate to an unlabeled acceptor ubiquitin (~ here refers to the thioester linkage between E2 catalytic cysteine and ubiquitin C terminus). However, CUL7-RBX1 was inactive under conditions when RBX1 complexes with CUL1 promote di-ubiquitin synthesis (Extended Data Fig. 6g). Taken together, the results are consistent with the structural findings that RBX1's RING domain in CRL7 FBXW8 is hindered from binding an active E2~NEDD8 or E2~ubiquitin intermediate.

FBXW8-SKP1 bridges CUL7-RBX1 and CUL1-RBX1.
In vivo, CUL7 was found to associate with CUL1 in an FBXW8-dependent manner 18 . The CRL7 FBXW8 structure suggests the mechanism: the canonical CUL1-binding residues in CUL7-bound SKP1 and the F-box are fully exposed and thus could mediate the interaction of CUL7 with CUL1. Comigration in size-exclusion chromatography (SEC) showed CRL7 FBXW8 directly binds full-length CUL1-RBX1 (Fig. 5a), but not an N-terminally truncated version lacking the CR1 domain (Extended Data Fig. 7a). Furthermore, when bound to CRL7 FBXW8 , CUL1-RBX1 is excluded from binding another SKP1-F-box protein complex (harboring a monomeric version of the F-box protein β-TRCP1) (Extended Data Fig. 7b-d). Indeed, a cryo-EM map of a complex between CRL7 FBXW8 and the N-terminal region of CUL1 (comprising the CR1, CR2 and CR3 domains), refined to a resolution of 4.6 Å visualized the interactions: SKP1-FBXW8 within the CRL7 FBXW8 complex binds CUL1's CR1 domain in a canonical manner (Fig. 5b).

CRL7 FBXW8 can recruit a neddylated CRL1 substrate in vitro.
Given that CRL7 FBXW8 lacks E3 activity in our assays, but it binds CUL1-RBX1, we asked if CRL7 FBXW8 could in principle serve in a substrate receptor manner for a neddylated CRL1 complex. If this were the case, then a CRL7 FBXW8 -interacting protein could potentially serve as a substrate in vitro. The only such interaction that has been reconstituted with pure proteins is CUL7's CPH domain binding to TP53. Indeed, we found that CRL7 FBXW8 , TP53 and CUL1-RBX1 can form a complex based on comigration by SEC, and this  depends on CUL7's TP53-binding CPH domain (Extended Data Fig. 7e-j). The CPH domain is connected to the ARM2 domain by equal to or roughly 45-and 28-residue long linker sequences, which are presumably flexible given that this domain is not observed in the cryo-EM map. In principle, the CPH domain-which itself spans equal to or roughly 25 Å-could extend over a wide radius relative to the rest of the complex to deliver TP53 to the ubiquitination active site (Extended Data Fig. 8a). Previously, we found that immunoprecipitated CUL7 was shown to mediate mono-or oligoubiquitination of TP53 (ref. 21 ). Notably, a CUL7 mutant unable to bind RBX1 showed similar activity 21 . We were able to reconstitute such TP53 ubiquitination using entirely purified components, dependent on CRL7 FBXW8 and neddylated CUL1-RBX1 (Fig. 5c, lanes 1-4). We thus, used TP53 as model substrate to probe mechanistic implications of the CRL7 FBXW8 -CUL1-RBX1 complex. Several observations are consistent with CUL7 recruiting TP53 for neddylated CUL1-RBX1-mediated ubiquitination. First, ubiquitination depends on CUL7's TP53-binding CPH domain (Fig. 5c, lane 5). Second, neddylated CUL5-RBX2, which does not bind SKP1-F-box protein substrate receptor modules, cannot substitute for neddylated CUL1-RBX1 (Fig. 5c, lane 6). Finally, the CRL7 FBXW8 -dependent TP53 ubiquitination is decreased with catalytically impaired neddylated CUL1-RBX1 mutants (nonneddylatable CUL1 K720R, Fig. 5c, lane 7), or CUL1-RBX1 modified with Q40E mutant NEDD8 (the Q40E mutation blocks noncovalent interactions between NEDD8 and CUL1 that allosterically activate CUL1) 35 (Fig. 5c, lane 8). Future studies will be required to investigate potential cellular functions of TP53 ubiquitination by the distinctive CRL7 FBXW8 -neddylated CUL1-RBX1 assembly. Nonetheless, the data demonstrate that CRL7 FBXW8 can biochemically act as a CRL1 substrate receptor (Fig. 5b,c and Extended Data Fig. 8a,b).
3M syndrome mutations. The physiological importance of CUL7 is underscored by its mutation causing developmental defects resulting in the human hereditary short stature disorder 3M syndrome 23 . CUL7 mutations account for 70% of 3M syndrome cases, with the remainder caused by mutations in OBSL1 and CCDC8 occurring in a mutually exclusively manner between these three genes.   40 in a with CrL7 FBXW8 in b, views are aligned on α/β domain. PDB 4P5O required use of a CUL1 version with the neddylation site K720 mutated to arginine to prevent transfer of NeDD8 and visualize the catalytic architecture. K720 was modeled in the structure here for clarity. c, Interface between CUL7's α/β-to-WHB linker region and 4HB shown in cryo-eM map calculated with DeepeMhancer. d, two views of structure visualizing CUL1 neddylation (PDB 4P5O). e, Modeling activated rBX1 rING-UBe2M~NeDD8 (PDB 4P5O) shows this clash with CrL7 FBXW8 , depicted by aligning rBX1 rING domain from both structures. f, In vitro neddylation assays using the indicated cullin-rING complex and NeDD8 e2 (UBe2M or UBe2F), NeDD8 e1 (NAe1-UBA3) and detecting fluorescently labeled NeDD8 in SDS-PAGe gels (samples of both gels derived from the same experiment and gels were processed in parallel, n = 2 technically independent experiments).
CUL7 interacting with OBSL1 and CCDC8 may be compatible with CUL7 incorporation into CRL7 FBXW8 or binding to TP53 (refs. 15,17,21,23,24,46,47 ), although it is unclear if such interactions are functionally related to the 3M complex since neither FBXW8 nor TP53 are mutated in the disease (Extended Data Fig. 8c). Nonetheless, the CRL7 FBXW8 assembly allows obtaining insights into potential structural effects of 3M syndrome mutations. Mapping mutations onto the structure shows that most are premature truncations that would affect CUL7 binding to RBX1 and presumably formation of the 3M complex, as this depends on C-terminal regions of CUL7 binding to OBSL1 (ref. 15 ). Although missense mutations map throughout the structure, several do map to interprotein interfaces in CRL7 FBXW8 , including with FBXW8 (P861S in the DOC domain, and Q1246G in the 4HB) and RBX1 (W1448R, G1452D, H1464P, Q1469R in the α/β-domain), as well as the interdomain interface between the CR2 and CR3 domains (L1014R) (Extended Data Fig. 8d,e). In agreement with the structure, mutation of H1464 and Q1469 located in CUL7's β-sheet that binds RBX1's N terminus, or the frame-shift at E1594 that truncates the RBX1-binding region of the WHB domain, were shown to eliminate binding to RBX1 (ref. 24 ). Thus, although FBXW8 has not been found to be mutated in 3M syndrome patients, the CUL7 mutations at the interface with FBXW8 may suggest a role in the function of the 3M E3 ligase complex.

Discussion
The structure of CRL7 FBXW8 reveals an atypical assembly between a cullin and F-box protein, and provides rationales for its perplexing features and the curious interaction between CUL7 and CUL1.   CRL7 FBXW8 displays many fundamental differences from canonical CRLs. First, extensive interactions anchor CUL7's WHB domain and RBX1's RING domain in an arrangement that is incompatible with both neddylation and ubiquitination (Figs. 2c-e and 4a-e, and Extended Data Fig. 6a,b). In particular, RBX1's RING domain is uniquely lodged against CUL7 in a way that would prevent binding to an activated E2~NEDD8 or E2~ubiquitin intermediate. Accordingly, our purified CRL7 FBXW8 complex lacks intrinsic NEDD8 or ubiquitin ligase activities in our assays (Fig. 4f and Extended Data Fig. 6c-e), agreeing with published proteomics studies that identified NEDD8 modification of other cullins but not CUL7 (refs. 15,48 ). Second, five CUL7 domains directly bind FBXW8 elements not found in other F-box proteins (Fig. 3). This explains CUL7's striking specificity for FBXW8 (refs. 10,15 ). Finally, the unique CRL7 FBXW8 assembly exposes the F-box and SKP1. These elements are thus available to bridge CUL7 and CUL1-RBX1 (Fig. 5b and Extended Data Fig. 8a). CRL7 FBXW8 binding to neddylated CUL1 may also solve the riddle of how affinity-tagged NEDD8 copurifies CUL7 despite lack of neddylation of CUL7 itself 14,31 .
To address why CRL7 FBXW8 binds another cullin-RING complex, our structural modeling suggested that CRL7 FBXW8 resembles a multiprotein substrate receptor for a neddylated CRL1 complex (Extended Data Fig. 8a). There has long been precedent for CRLs relying on multiprotein complexes to recruit substrates, from one of the first human F-box proteins discovered: the F-box protein SKP2 binds a cyclin-CDK2-CKSHS1 complex to recruit the substrate phosphorylated p27 (ref. 49 ). Indeed, we were able to biochemically reconstitute ubiquitination of TP53 recruited by CUL7 (Fig. 5c). Although we were not able to reconstitute ubiquitination of other CRL7 FBXW8 substrates, it seems likely that some substrates may be avidly recruited by interaction with other domains of CUL7 and/ or FBXW8 domains, much like multivalent interactions underlie recruitment of substrates to CRL1 SKP2-CKSHS1-Cyclin-CDK2 . CRL7 FBXW8 joins a growing list of E3 ligases that function in partnership with other E3 enzymes. For some E3 pairs that coassemble in E3-E3 complexes-for example, the two RBR-family E3s HOIL-1L and HOIP in the linear ubiquitin assembly complex, and the HECT-type E3 Ufd4 with RING or RING-like E3s Ubr1 and Ufd4, respectively-each E3 mediates sequential steps in polyubiquitination 50,51 . However, each E3 within these complexes is thought to achieve activity by binding an E2~ubiquitin intermediate, whereas we did not observe E3 ligase activity with purified CRL7 FBXW8 (Extended Data Fig. 6d,e). Rather, CRL7 FBXW8 partnering with neddylated CUL1-RBX1 may share some conceptual parallels with canonical neddylated CRLs partnering with ARIH-family RBR E3s such that one E3 (the CRL) recruits substrate while the other (the ARIH-family member) mediates ubiquitination 36,41,52 .
At this point, many questions linger about the cellular functions of CRL7 FBXW8 and other complexes formed by CUL7-RBX1. First, can CUL7-RBX1 directly ubiquitinate a substrate without CUL1-RBX1? The identities of direct substrates and functional roles of CRL7 FBXW8 ubiquitination-including for TP53-remain poorly understood. Although deletion of subunits of CRL7 FBXW8 affects the cellular levels of some proteins, it is not clear which of those bind to CRL7 FBXW8 directly, or whether their ubiquitination depends on CRL7 FBXW8 forming a complex with neddylated CUL1-RBX1 (ref. 17 ). Second, what is the role of RBX1 in CUL7 complex given that the purified recombinant CRL7 FBXW8 complex appears to be catalytically inactive? From a structural perspective, RBX1 appears to serve as a crucial stabilizing support for CUL7's cullin region (Fig. 2). But does RBX1 play additional functions beyond promoting proper CUL7 folding in the context of other binding partners? For example, in the context of the 3M complex, does RBX1 display NEDD8 or ubiquitin ligase activity? Do currently unknown mechanisms activate neddylation? Are there contexts in which the RING in CUL7-RBX1 is an active E3 ligase? Third, CUL7 associates with many other proteins, including CUL9, which is homologous to CUL7 but with an additional ARIH1-family E3 ligase domain, yet little is known about how and why these proteins interact and mediate regulation 13,14,16 . Is the CUL7-CUL9 complex catalytically active? Fourth, does the CUL7-FBXW8 interaction play a role in preventing 3M disease? Although SKP1-FBXW8 copurifies with the 3M complex from some cell types 15,24,53 , it is perplexing that mutations in these proteins or RBX1 have not been found in patients with 3M syndrome. And if so, are there essential functions of FBXW8 explaining lack of disease mutations, and by extension does CUL1 also play a role in 3M syndrome? Finally, residue-specific details of how CUL7-RBX1 binds with OBSL1 and CCDC8 and how this contributes to disease are important questions for the future. Our structure of the unusual multidomain CRL7 FBXW8 complex, and finding that this can mediate recruitment of another protein for ubiquitination by neddylated CUL1-RBX1, provide a platform for future studies aimed at understanding these other atypical cullin interactions and mechanisms underlying 3M syndrome (Fig. 5d).

Online content
Any methods, additional references, Nature Research reporting summaries, source data, extended data, supplementary information, acknowledgements, peer review information; details of author contributions and competing interests; and statements of data and code availability are available at https://doi.org/10.1038/ s41594-022-00815-6.

CRL7 FBXW8 expression constructs.
With the aim of recombinant coexpression of CUL7, RBX1, FBXW8 and SKP1 in mammalian cells we used an approach combining the pEG plasmids from the BacMam system 55 with an adjusted version of the biGBac cloning system 56 . All proteins sequences are of human origin, complementary DNAs were synthesized from Twist Bioscience and variants were produced and verified by standard molecular biology techniques, a list of used primers is in Supplementary Table 1. pEG plasmids for each of the four proteins were cloned, with CUL7 carrying a N-terminal Twin-Strep-tag and FBXW8 a N-terminal GFP-tag, both having a PreScission-cleavage-site between protein of interest and tag. All constructs were full length except RBX1 that was residue 5 to C terminus. Gene-expression cassettes were amplified by PCR with standard forward primers from biGBac system and reverse primers adjusted to be complementary to a sequence downstream of the SV40 terminator. Linearized pBig1 vector and two gene-expression cassettes were fused by a Gibson assembly reaction, resulting in the following circular pBig1 constructs: pBig1a carrying N-terminal twin-Strep-tag CUL7 and RBX1; pBig1a carrying N-terminal twin-Strep-tag CUL7 ΔCPH (residues 389-404 exchanged by a GGGSGGGGSGGGS linker) and RBX1; pBig1b carrying N-terminal GFP-tagged FBXW8 and SKP1. pBig1 constructs were then used to generate bacmids from DH10 EMBacY, which were used to transfect Sf9 cells and amplify baculovirus up to P3. Virus supernatant was sterilized by filtration, then used for subsequent infection for expression. CRL7 FBXW8 expression and purification. Canonical CRL1s are typically assembled from separately purified CUL1-RBX1 and SKP1-F-box protein complexes. Although we were able to express and purify stable CUL7-RBX1 alone, we were not able to obtain a stable and pure SKP1-FBXW8 subcomplex as this readily associated with chaperones. Such chaperone copurification often reflects improper folding that can be overcome by coexpression with other binding proteins. Indeed, we were able to purify homogeneous and stoichiometric CRL7 FBXW8 by coexpressing all four subunits in HEK293S cells in suspension. For expression of CRL7 FBXW8 , CRL7 ΔCPH-FBXW8 , CUL7-RBX1, 800 ml of HEK cultures were grown to a cellular density of >3 × 10 6 cells per ml and infected with 10% (v/v) P3 of corresponding viruses and incubated at 37 °C. 6 h posttransduction, 10 mM butyrate was added and incubated for 60 h at 30 °C. After that, the collected cells were lysed by sonication in lysis buffer containing 50 mM Tris pH 7.5, 150 mM NaCl, 5 mM DTT, 0.001 mg ml −1 Benzonase, EDTA-free protease inhibitor cocktail (one tablet per 50 ml of buffer). Insoluble material was removed by centrifugation at 50,000g and supernatant was incubated with Strep-Tactin resin in-batch rocking for 15 min at 4 °C. The resin was then transferred to a column and washed with 25 CV of Wash Buffer containing 50 mM Tris pH 7.5, 150 mM NaCl, 5 mM DTT. Protein was eluted with 10 CV of Elution Buffer containing 2.5 mM d-desthiobiotin in the wash buffer and sequentially cleaved with 3C protease overnight at 4 °C, followed by Anion exchange chromatography. Protein containing fractions were pooled and concentrated to <500 μl using a 100 kDa cutoff concentrator (Amicon) and final polishing was performed by SEC on a Superose 6 Increase 10/300 GL column (GE Healthcare) with Wash Buffer. Peak fractions containing CRL7 FBXW8 were pooled, concentrated to around 1.7 mg ml −1 and used immediately for cryo-EM grid preparation.
Cloning, expression and purification of TP53. His-lipoyl domain-tagged TP53 was cloned into a pRSF vector, expressed in Escherichia coli BL21 Rosetta (DE3) at 20 °C overnight. Cells were lysed in 50 mM Tris pH 8.0, 300 mM NaCl, 0.01% NP-40, EDTA-free protease inhibitor cocktail (one tablet per 50 ml of buffer), 5 mM β-mercaptoethanol. After sonication, the lysate was centrifugated for 20 min at 15,000g and the supernatant was loaded onto a Ni-NTA column. His-tagged TP53 was eluted with increasing imidazole gradient and dialyzed overnight in wash buffer (50 mM Tris pH 7.5, 150 mM NaCl, 5 mM DTT) in the presence of tobacco etch virus (TEV) protease. Cleaved TP53 ran as a tetramer on a Superpose 6 Increase 10/300 GL column (GE Healthcare) and was separated from the His-tagged lipoyl domain.
Generation of TAMRA-TP53 fusion via Sortase-mediated transpeptidation reaction. TAMRA-labeled TP53 was generated in a Sortase-catalyzed reaction by fusing a TAMRA-LPETGG peptide (obtained from Max Planck Institute of Biochemistry Core Facility) with the N terminus of TP53, which retains a glycine after TEV cleavage. Reaction was incubated with concentrations of 50 μM TP53, 300 μM TAMRA-LPETGG peptide and 5 μM SortaseA-6xHis harboring activity enhancing mutations 57 for 2 h in 50 mM Tris pH 8.0 150 mM NaCl 10 mM CaCl 2 , removed SortaseA by retention on nickel resin and further purified by SEC in 50 mM Tris pH 7.5, 150 mM NaCl, 1 mM DTT.
Fusion proteins were subjected to GST-affinity chromatography, eluted by overnight protease cleavage on beads and further purified by ion-exchange and SEC in 25 mM HEPES, 150 mM NaCl, 1 mM DTT 36,40 .
For fluorescent labeling GST-TEV fusion proteins of NEDD8 and ubiquitin (wild-type and K48R) carried an additional cysteine in the N-terminal region were purified similarly 40 . Purified NEDD8 and ubiquitin variants were then incubated 30 min on ice with addition of 10 mM DTT to ensure full reduction for single cysteine labeling. Proteins were buffer exchanged into 25 mM HEPES pH 7.5, 150 mM NaCl using NAP-5 columns, 1 mM Fluorecein-5-maleimide was added to 200 μM protein and incubated for 2 h at room temperature. To remove free unreacted dye the reaction mix was buffer exchanged a second time using a NAP-5 column in 25 mM HEPES 7.5, 150 mM NaCl, 1 mM DTT and fluorescently tagged protein were purified by SEC in the same buffer 40 .
His-MBP-TEV β-TRCP1 (monomeric form, from residue 175 to C terminus) was expressed and purified in complex with 'SKP1 ΔΔ ' (SKP1 harboring two internal deletions, of residues 38-43 and 71-82). The complex was purified by nickel affinity chromatography, followed by TEV cleavage, ion-exchange and SEC in 25 mM HEPES 7.5, 150 mM NaCl, 1 mM DTT 35 . This complex is referred to as SKP-β-TRCP1 in the main text and figures.
Cryo-EM sample preparation and data collection for CRL7 FBXW8 . R1.2/1.3, 200 mesh holey carbon grids (Quantifoil) were glow discharged, 4 μl of freshly purified CRL7 FBXW8 was applied at 100% humidity and 4 °C in a Vitrobot Mark IV (Thermo) and plunge-frozen into liquid ethane (5 s incubation time, blot force 3-4, blot time 4 s). High-resolution cryo-EM data were collected on a Titan Krios electron microscope operating at 300 kV, equipped with a post-GIF Gatan K3 Summit direct electron detector (counting mode) using SerialEM 58 . Videos were recorded at a nominal magnification of ×105,000 (0.8512 Å per pixel at the specimen level), with target defocus ranging between −0.7 and −2.8 μm and total exposure of roughly 60 e − /Å 2 fractionated over 40 frames.
Cryo-EM data processing. RELION 3.0 (ref. 59 ) was used for motion-correction and dose weighting, the contrast transfer function was estimated using CTFFIND-4.1 (ref. 60 ) and particles were picked using Gautomatch (K. Zhang, MRC Laboratory of Molecular Biology) with a template based on previous screening datasets. 19,499 micrographs with a maximum resolution estimate better than 5 Å were imported into RELION v.3.1 (ref. 59 ), from which roughly 6.05 million particles were extracted applying sixfold binning. These were subjected to several rounds of 2D classification, followed by initial model generation and 3D classification. Then 355,547 particles from the best 3D class were reextracted at full pixel size, followed by masked 3D autorefinement, producing a reconstruction at 2.82 Å overall resolution. RELION 59 postprocessing and DeepEMhancer 54 were used for sharpening of the final map.
Model building and refinement. The CRL7 FBXW8 structure was traced de novo using Coot (v.0.8.9.2) 61 and Phenix.refine was used for real space refinement 62 . Residues were placed starting with the best resolved cullin repeats and WD40 domain of FBXW8, and sequentially filled in the rest of the density by seeking landmark residues (tryptophan, phenylalanine, tyrosine, proline) in the sequences with corresponding well-resolved densities. These were used in addition to secondary structure prediction to guide sequence assignment, loop building and helical packing. RBX1 was built based on several of the published structures in the Research Collaboratory for Structural Bioinformatics (RCSB) (for example, PDB 4P5O) 40 . The density representing SKP1 allowed placement of helices but was not resolved well enough for unambiguous placement of side chains except in the interaction surface between the F-box domain of FBXW8 (Extended Data Fig. 3a). A SKP1 crystal structure was fitted in the density and kept unchanged except deleting loops that had no density in the cryo-EM map. The resolution of cullin repeats and WD40 domain of FBXW8 was better than 3 Å in most parts allowing for side-chain placement (Extended Data Fig. 3d,e). However, there was an additional poorly resolved domain near the N terminus of CUL7 domain, which was not assigned to any part of CRL7 FBXW8 (Extended Data Fig. 3c). The same is true for a patch of density adjacent to the RBX1 RING domain that could resemble a helix (Extended Data Fig. 3b). After the structure was determined, AlphaFold was released and enabled comparison of the AlphaFold-derived model to the experimentally derived structure 34 .