CUL5-ARIH2 E3-E3 ubiquitin ligase structure reveals cullin-specific NEDD8 activation

An emerging mechanism of ubiquitylation involves partnering of two distinct E3 ligases. In the best-characterized E3-E3 pathways, ARIH-family RING-between-RING (RBR) E3s ligate ubiquitin to substrates of neddylated cullin-RING E3s. The E3 ARIH2 has been implicated in ubiquitylation of substrates of neddylated CUL5-RBX2-based E3s, including APOBEC3-family substrates of the host E3 hijacked by HIV-1 virion infectivity factor (Vif). However, the structural mechanisms remained elusive. Here structural and biochemical analyses reveal distinctive ARIH2 autoinhibition, and activation on assembly with neddylated CUL5-RBX2. Comparison to structures of E3-E3 assemblies comprising ARIH1 and neddylated CUL1-RBX1-based E3s shows cullin-specific regulation by NEDD8. Whereas CUL1-linked NEDD8 directly recruits ARIH1, CUL5-linked NEDD8 does not bind ARIH2. Instead, the data reveal an allosteric mechanism. NEDD8 uniquely contacts covalently linked CUL5, and elicits structural rearrangements that unveil cryptic ARIH2-binding sites. The data reveal how a ubiquitin-like protein induces protein-protein interactions indirectly, through allostery. Allosteric specificity of ubiquitin-like protein modifications may offer opportunities for therapeutic targeting.

We obtained cryo-EM maps for two ARIH2* complexes with neddylated CRL5 Vif-CBFβ , one with the substrate APOBEC3C and the other with APOBEC3G (hereafter A3C and A3G, respectively) (Extended Data Figs. 2-4 and Supplementary Table 2). Because both complexes showed similar properties, only the higher resolution reconstructions with A3C are described. A 7.5 Å resolution low-pass filtered map allowed fitting with published atomic coordinates of A3C, Vif-CBFβ-ELOBC bound to CUL5's N-terminal domain, other domains of CUL5, NEDD8 and nearly all the ARIH2 crystal structure 12,22,[36][37][38][39] (Fig. 1c,d and Extended Data Fig. 5). However, lack of density unambiguously attributable to ARIH2*'s Rcat suggests that this domain is relatively mobile compared to the rest of the E3-E3 complex (Extended Data Fig. 2c).
NEDD8 also binds the edge of CUL5 in the intermolecular C/R domain. NEDD8's Lys6 and His68 form a three-way interface with Leu710 from CUL5's H29-helix and a stripe of Glu617, Leu621 and Glu624 side-chains from CUL5 in the C/R domain (Fig. 2d). Retrospective analysis revealed the same three-way interactions in the previous crystal of neddylated CUL5's C-terminal region, albeit with a twist: in the crystal, rather than occurring within a single complex, these interactions mediate packing between the C/R domain from one molecule of CUL5 and NEDD8 and its linked WHB domain from an adjacent complex in the lattice 12 (Extended Data Fig. 6b,c). Mutation of the key NEDD8 binding surfaces on CUL5's WHB and C/R domain impaired ARIH2-mediated ubiquitylation of neddylated CRL5 substrates (Fig. 2e).
of the helices (Fig. 3c). In particular, the first Ariadne domain helix, which we term a 'switch-helix' , displays a roughly central 15° kink when bound to CUL5. ARIH2 residues 380 and 381 (alanines in ARIH2*) at the center of the kink along with nearby side chains are rotated outward. The switch-helix kink precludes autoinhibitory interactions with the Rcat domain's catalytic Cys (Fig. 3c). Thus, it seems that when ARIH2 is bound to neddylated CUL5-RBX2, kinking of the switch-helix would relieve autoinhibition. This rationalizes the previous finding that neddylated CUL5-RBX2 stimulates reactivity of ARIH2's catalytic Cys with the electrophilic UB probe, UB-VME 21 .
Remodeled CUL5 groove cradles ARIH2 N terminus. The second crucial portion of ARIH2 is its N-terminal region, which is not present in the crystal structure. In complex with a neddylated CRL5, ARIH2*'s N terminus mediates interactions extending more than 50 Å across the structurally remodeled CUL5 (Fig. 3d).
The central portion of ARIH2*'s N-terminal region (residues 35-46) forms a kinked amphipathic helix that docks in a CUL5 groove. One side of the groove is formed by CUL5's 4HB domain. The other side involves CUL5 elements from the C/R domain and the loop preceding its H29-helix, which we term 'gate/groove loop' (Fig.  3d). In an unneddylated CRL5 complex, the gate/groove loop (CUL5 residues 691-695) restricts access to CUL5's 4HB. However, in neddylated CUL5 the H29-helix rotation is accompanied by gate/groove loop remodeling, which generates the ARIH2-binding groove.
Ile35, Tyr38, Tyr39, Val42 and Val46 from ARIH2* form a stripe of hydrophobic knobs that fit into hydrophobic pockets between CUL5 4HB domain helices (Fig. 3e,f and Extended Data Fig. 7b,c). A pair of neddylated CUL5 arginines-Arg460 from the 4HB and Arg691 from the gate/groove loop-seal ARIH2*'s Tyr39 in the groove. CUL5's Arg691 additionally interacts with the backbone carbonyl from ARIH2*'s Gly41 at the kink and Asp45 at the C terminus of the helix (Fig. 3f).
Density was poorly visible for the elements connected to ARIH2*'s N-terminal region. Residues 50-55 comprise a linker to the canonical RBR elements (Fig. 3d). At the opposite, extreme N-terminal end, additional density was visible only at low contour, in a basic groove from CUL5 (Extended Data Fig. 7b).
To interrogate roles of ARIH2's N-terminal region, we performed alanine scanning mutagenesis. Substituting residues individually, or three or four at a time, confirmed key roles of the region containing ARIH2's Tyr38 and Tyr39, and intermediate effects at the junction to the acidic stretch ( Fig. 3g and Extended Data Fig.  7d). We also tested effects of deletions (Extended Data Fig. 7e). The most destructive effects arose from removing portions of ARIH2's N-terminal region-either residues 35-39 or 40-44-that dock in the remodeled CUL5 groove. Deleting the N-terminal 20 residues, which were not observed by cryo-EM, did not overtly impair ubiquitylation. However, deletions within an ARIH2 acidic stretch (residues 25-29 or 30-34) impaired ubiquitylation of neddylated CRL5 substrates. We speculate that the ARIH2 acidic residues could contact basic residues at the entrance to the CUL5 groove. Notably, Ala substitution for either CUL5 Arg460 or Arg691, or four basic residues at the edge of the CUL5 groove (Lys418, Lys423, Lys676 and Lys685) substantially impair ARIH2-mediated ubiquitylation of CRL5 Vif-CBFβ substrates (Fig. 3h). None of the mutations impaired CUL5 neddylation, suggesting they did not affect protein folding (Extended Data Fig. 7f 22 ) was superimposed on the neddylated CUL5-rBX2-ArIH2* structure, but is not shown for simplification. In unneddylated CUL5, the H29-helix encompasses residues 697-725, whereas the first turn is unfolded and the H29-helix encompasses residues 700-725 in neddylated CUL5. c, Close-up of interactions surrounding the covalent isopeptide linkage between NeDD8 (yellow) and CUL5 H29-helix portion of the WHB domain (dark green). d, three-way interface between NeDD8, its covalently linked CUL5 H29 helix and WHB domain, and a CUL5 surface from the C/r domain (relative to c, rotated 80° in x and 40° in y). e, ArIH2-catalyzed fluorescent UB (*UB) transfer to A3G in 10 min, mediated by Wt unneddylated (no NeDD8), Wt neddylated CrL5 Vif-CBFβ (Wt) or versions with indicated mutations of CUL5 residues making noncovalent interactions with NeDD8. N = 2 independent experiments. For samples from same experiment, gels were processed in parallel (Source Data Fig. 2).
structurally activates the E3-E3 assembly in several ways. Gate/ groove loop remodeling not only provides a binding site for ARIH2's N-terminal region, but also eliminates blockage of the groove (Fig. 4a). Also, when unneddylated, the WHB domain packs against CUL5's 4HB and RBX2's RING domains so as to block access of ARIH2's Ariadne domain (Fig. 4b). This unneddylated arrangement is also incompatible with noncovalent interactions between CUL5's WHB domain and NEDD8 (Fig. 4c) 12,22 . Considering that the neddylation reaction requires yet another distinct relative arrangement of RBX RING and cullin WHB domains 41 , we speculate that after NEDD8 linkage to CUL5, formation of the structurally observed noncovalent interactions hinders these domains from adopting the orientations in unneddylated CRL5. The structurally observed conformational changes explain previously reported HDX properties of an unneddylated and neddylated CRL5, ARIH2, and the neddylated CRL5-ARIH2 complex 33 . Deuterium incorporation was measured by mass spectrometry of peptides generated after HDX was quenched 33 . Peptides corresponding to several regions of neddylated CRL5 and ARIH2 remodeled in the cryo-EM structure, for example NEDD8 and the regions of CUL5 it binds, were not detected by mass spectrometry 33 . Nonetheless, the detectable regions that showed greatest HDX differences upon complex formation correlate with the conformational changes indicated by the cryo-EM structure (Supplementary Video 1). In particular, the HDX differences 33 between unneddylated and neddylated CRL5 primarily map to CUL5 4HB and C/R domain regions exposed by the structurally observed relocation of neddylated CUL5's WHB domain. Comparing HDX properties of ARIH2 alone versus bound to a neddylated CRL5 showed greatest differences in the Ariadne domain 33 . The regions showing increased HDX correspond to the switch-and subsequent Ariadne domain helices that become exposed in the structural transition between autoinhibited ARIH2 and neddylated CRL5 Vif-CBFβ -bound ARIH2*. Meanwhile, Ariadne domain regions that were protected from HDX in the complex 33 correspond to ARIH2* elements that bind CUL5-RBX2. Likewise, the CUL5 regions whose HDX properties differ in the complex with ARIH2 (ref. 33 ) correspond to those bound to ARIH2* in the cryo-EM structure.
CUL5's WHB domain and/or the H29-helix, would be sufficient to activate ARIH2 ubiquitylation of a CRL5 substrate. Such deletion mutants would in principle remove the barrier blocking ARIH2's Ariadne domain, although they would preclude interactions that stabilize the remodeled the gate/groove loop. The deletions did increase ARIH2-mediated ubiquitylation of CKB compared to unneddylated CRL5 ASB9 , although not to the level observed with neddylation ( Fig. 4d,e). We thus inspected the structures of unneddylated 12 and neddylated CUL5 to identify residues potentially anchoring the inactive conformation, but whose mutation would not hinder relocation of the H29-helix. In unneddylated CUL5, four CUL5 H29-helix glutamates (Glu701, Glu702, Glu703, Glu705) either directly contact the C/R domain or establish H29-helix-C/R domain electrostatic networks (Fig.  4f). A flanking glutamate (Glu697) also may contribute to the inactive conformation. All five of these H29-helix glutamates are solvent-exposed-their side chains not visible in the mapsin neddylated CRL5 bound to ARIH2*. Thus, we hypothesized that charge-swap mutants could expunge unneddylated CUL5's H29-helix and WHB domain, while allowing the active conformation. Indeed, an 'E-to-K' mutant version of unneddylated CRL5 ASB9 with these CUL5 H29-helix glutamates replaced with lysines enabled ARIH2-dependent CKB ubiquitylation at a level similar to that achieved with neddylated CRL5 ASB9 in our assay (Fig. 4e).
To gain further insights into similarities and differences between E3-E3 ligases, we compared structures of ARIH2 and ARIH1, and their complexes with a neddylated CRL5 or CRL1, respectively ( Fig. 5a,b). The comparison showed similar roles of the ARIH2 and ARIH1 Ariadne domains. The Ariadne-Rcat domain arrangements superimpose in autoinhibited ARIH2 and ARIH1 (0.8 Å r.m.s.d., Extended Data Fig. 8c), and the E3-E3act domains also superimpose for both families (1.1 Å r.m.s.d., Fig. 5c). Notably, the Ariadne domain switch-helix kink observed for ARIH2* is shared by both WT ARIH1 and the corresponding ARIH1* mutant when bound to the cognate neddylated CRL 23 .

Discussion
Structures of ARIH2 alone and bound to neddylated CRL5 Vif-CBFβ reveal the HIV-1 hijacked E3-E3 ligase assembly that overcomes host restriction and defines mechanisms by which NEDD8-linked CUL5-RBX2 activates ARIH2. Due to high sequence and functional homology, we anticipated that neddylated CRL5-ARIH2 and neddylated CRL1-ARIH1 would form superimposable but sequence-specific E3-E3 assemblies. Indeed, for both, CRL5s and CRL1s, the neddylated conformations remove barriers that mask ARIH E3-binding sites in their unneddylated counterparts. Moreover, for both ARIH2 and ARIH1, the Ariadne domains mediate homologous autoinhibitory interactions with the Rcat domains, and homologous interactions with their cognate CUL-RBX partners.
Why might NEDD8 uniquely modulate the structure of CUL5-RBX2 and its interactions with ARIH2? Although answering this will require future studies, we speculate that additional regulation co-evolved with emergence of CUL5-RBX2 in metazoan lineages. For example, CUL5-RBX2 and/or ARIH2-specific metazoan-specific posttranslational modifications or binding partners awaiting discovery may require a distinctive assembly from that formed by ARIH1 and neddylated CRL1s.
The indirect, allosteric mechanism by which NEDD8 stimulates binding to ARIH2 differs from most characterized interactions between UB and UBLs and their downstream recognition machineries. UB-, SUMO-and LC3-interacting motifs in different proteins often form structurally superimposable complexes with their UB or UBL partners 1,2 . Moreover, to our knowledge, UBL (or UB)-driven protein-protein interactions mediated by conformational changes-without direct binding to the UB or UBL itself-have not been structurally defined before. However, SUMO and UB have been shown to induce conformational changes that inhibit interactions of their targets. For example, a SUMO-interacting motif in thymine DNA glycosylase interacts with a linked SUMO to stabilize a conformation incompatible with DNA-binding 48 . Also, UB-binding domains of the yeast transcription factor Met4 engage a K48-linked polyUB chain modification so as to counteract interactions required to activate transcriptional targets 49 . Such unique allosteric switches, as revealed by our structural analyses, may provide opportunities for therapeutic targeting specificity distinguishing otherwise homologous complexes. This may be particularly relevant for CRL5s and ARIH2, which regulate immune pathways, and are conscripted by several viruses to promote infection 29-32,50 .

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/ s41589-021-00858-8.

Methods
Cloning, protein expression and purification. For all expression constructs described in this study, standard molecular biology techniques were used for preparation and verification. Except for HIV-1 Vif (viral infectivity factor), coding sequences of the described proteins are of human origin. Mutant versions of ARIH2, CUL5 and UBE2L3 were generated using the Quikchange system (Agilent) and verified by sequencing.
Open reading frames encoding CUL5 (untagged), RBX2 (a glutathione S-transferase (GST) fusion with an intervening tobacco etch virus (TEV) protease site and encompassing RBX2 residues 5 to the C terminus), UBA1 (a GST fusion with an intervening TEV protease site), APOBEC3C (a GST fusion with an intervening TEV protease site) and APOBEC3G (a GST fusion with an intervening TEV protease site) were subcloned into pLIB vectors for expression in Trichoplusia ni High-Five insect cells. CUL5 and RBX2 were coexpressed via baculoviral co-infection in a manner similar to that described previously for CUL1-RBX1 (refs. 17,23 ). Briefly, CUL5-RBX2 and UBA1 were initially purified by GST-affinity chromatography, subjected to TEV protease cleavage of the fusions overnight at 4 °C, and further purified by anion exchange chromatography using a HiTrap Q HP column (Cytiva Life Sciences) and then by size-exclusion chromatography. CUL5-RBX2 variants were purified using the same procedure, and are indicated by the residue numbers mutated and/or the ranges encompassed with the following exceptions: the CUL5 mutant 'Δ29 helix' lacks residues 694-726; 'K-to-D' is K418D K423D K676D K685D; 'E-to-K' is E697K E701K E702K E703K E705K. APOBEC3C and APOBEC3G (hereafter referred to as A3C and A3G, respectively) were purified as previously described 37 . Neddylated CUL1-RBX1, SKP1-FBXW7 (the ∆D version lacking the dimerization domain), and ARIH1 were expressed and purified as previously described 16,17,23 . The buffer used for the final size-exclusion chromatography purification of all these proteins and complexes was 25 mM HEPES pH 7.5, 150 mM NaCl and 1 mM DTT. Purity of all protein samples was verified by intact mass spectrometry provided by the Max Planck Institute of Biochemistry Core Facility.
Two base versions of ARIH2 are used. A near full-length version lacking only the first 50 residues (encompassing residues 51 to the C terminus) was used for obtaining the crystal structure of autoinhibited ARIH2. All biochemical assays and cryo-EM studies used full-length versions of either WT ARIH2 or the mutant versions of the full-length construct. ARIH2* refers to a mutant version of full-length ARIH2, relieved from autoinhibition through three residue substitutions: L381A E382A E455A. Deletion mutant versions of ARIH2 are indicated by 'Δ' followed by residues excluded from the construct. All versions of ARIH2 were expressed using a common protocol. The constructs, in pRSF vector, contain an N-terminal His 6 -tag followed by maltose binding protein and a TEV protease cleavage site fused to the N terminus of ARIH2 (or residue 51 in the version used in the crystal structure). The various versions of ARIH2 were expressed in in Escherichia coli (Rosetta, DE3). Cultures were grown to an optical density of 0.6-0.8 on which expression was induced with 0.1 mM isopropyl-β-d-thiogalactopyranoside (Sigma) and 0.1 mM ZnCl 2 (Sigma). The various versions of ARIH2 were initially purified by nickel-affinity chromatography, subjected to TEV protease cleavage of the fusions overnight at 4 °C and further purified by anion exchange chromatography using a HiTrap Q HP column (Cytiva Life Sciences) and then by size-exclusion chromatography in 25 mM HEPES pH 7.5, 150 mM NaCl and 1 mM DTT.
An open reading frame encoding HIV-1 Vif (UniProt sequence P12504), codon-optimized for expression in E. coli, was obtained from GeneArt/Thermo Fisher. This and the gene encoding human CBFβ were subcloned into pRSF duet vector (MCS1 and MCS2, respectively). Vif was expressed with an N-terminal His 6 -tag followed by a TEV protease cleavage site. Full-length untagged Elongin B (ELOB) and Elongin C (ELOC) were subcloned into MCS1 and MCS2 of pACYCDuet-1 (Novagen), respectively. The Vif, CBFβ and ELOB and ELOC expression plasmids were cotransformed into E. coli (BL21 Gold, DE3) and the proteins coexpressed and purified as previously described 37 . The ASB9 complex with ELOBC was expressed similarly, except ASB9 was subcloned into the pET3a vector with an N-terminal His 6 -tag followed by a TEV protease cleavage site. These complexes were initially purified by nickel-affinity chromatography, subjected to TEV protease cleavage of the fusions overnight at 4 °C, and further purified by ion exchange chromatography and then by size-exclusion chromatography in 25 mM HEPES pH 7.5, 150 mM NaCl and 1 mM DTT.
CKB was expressed as an N-terminal GST fusion-with a TEV protease between GST and CKB-in E. coli (Rosetta, DE3) cells. Neddylation components NEDD8, UBE2F, UBE2M and APPBP1-UBA3 were expressed in either E. coli (Rosetta, DE3) or BL21 Gold, DE3) cells as GST fusions with either thrombin or TEV as intervening protease cleavage sites. These proteins were expressed and purified as previously described 16,41 , with the exception of APPBP1-UBA3, where all fractions containing the neddylation E1 enzyme were pooled to maximize yield rather than purity. The CUL5-RBX2 complex was neddylated by mixing 12 µM CUL5-RBX2, 1 µM UBE2F, 0.2 µM APPBP1-UBA3, 25 µM NEDD8 in 25 mM HEPES pH 7.5, 150 mM NaCl, 10 mM MgCl 2 and 1 mM ATP. NEDD8 was added at room temperature. Neddylation was quenched after 8 min by adding 10 mM DTT to suppress activity of APPBP1-UBA3 and UBE2F. After microcentrifugation at 13,000 r.p.m. for 10 min, the NEDD8-CUL5-RBX2 was purified using a Superdex SD200 column, in 25 mM HEPES pH 7.5, 150 mM NaCl, 1 mM DTT (-refers to the isopeptide linkage between NEDD8 or UB and a lysine on the target, here a cullin).
*UB refers to human ubiquitin expressed from pGEX-2TK, but with the N-terminal RRASV sequence replaced with RRACV, where the Cys serves as the site for fluorescent labeling with fluorescein maleimide. *UB was expressed in E. coli (BL21 RIL, DE3), purified and fluorescently labeled as previously described 41 .
Peptide. The peptide used as substrate in ubiquitylation assays corresponds to phosphorylated Cyclin E (pCycE) and has sequence KAMLSEQNRASPLPSGLL(pT)PPQ(pS)GRRASY. The peptide was synthesized in the Max Planck Institute of Biochemistry Core Facility, and purified to greater than 95% purify by high-performance liquid chromatography.
In vitro ubiquitylation assays. *UB transfer was monitored using a pulse-chase format. Briefly, the thioester-bonded UBE2L3~*UB intermediate (the '~' refers to thioester linkage between two proteins) was produced in the pulse reaction, and various proteins were added to initiate the chase reaction in which *UB is transferred from UBE2L3 through the ARIH2-dependent ubiquitylation cascade. Pulse reaction conditions were optimized such that when examined by Coommassie-stained SDS-PAGE, all visibly detectable UBE2L3 was thioester-bonded to UB. The pulse reaction producing UBE2L3~*UB was carried out by incubating 15 μM UBE2L3, 0.3 μM UBA1 and 15 μM *UB in 25 mM HEPES pH 7.5, 100 mM NaCl, 2.5 mM MgCl 2 , 1 mM ATP at room temperature for 30 min. The reaction was quenched with 2 U ml −1 apyrase and incubated on ice for at least 5 min. The quenched solution was further diluted to 5 µM with 25 mM HEPES pH 7.5, 100 mM NaCl.
The chase reaction was initiated by adding a premade mixture of various components to the pulse reaction. Chase reaction mixes are described here by the mixture of components, but in final concentrations in the reactions, after addition to the pulse mix. To examine CRL5 activities, the concentration of UBE2L3~*UB generated in the pulse reaction was 0.4 µM. To examine CRL5 substrate ubiquitylation, the chase reaction mix consisted of 0.1 µM ARIH2, 4 µM substrate (A3G, A3C or CKB as indicated), and a neddylated CRL5 assembled in the mix from two parts: 0.4 µM NEDD8-CUL5-RBX2, and either Vif-CBFβ-ELOBC or ASB9-ELOBC as indicated. To examine autoubiquitylation, the chase reaction mix contained 0.4 µM ARIH2 and 0.4 µM NEDD8-CUL5-RBX2 or CRL5. To examine CRL1 activities, the substrate was a phosphopeptide derived from Cyclin E (pCycE, sequence provided above). The chase reaction mix consisted of 0.4 µM UBE2L3~*UB, 2 µM pCycE and 0.5 µM each of ARIH1, NEDD8-CUL1-RBX1 and SKP1FBXW7∆D.
All reactions were carried out at room temperature and quenched at the indicated time points by adding 2× nonreducing SDS-PAGE sample buffer. SDS-PAGE was performed under nonreducing conditions. Gels were scanned using an Amersham Typhoon imager (GE Healthcare). Graphs shown in the main and extended data figures were generated from the 10 min timepoint. Band intensities for ubiquitylated products (*UB linked to A3G or CKB) were measured by ImageQuant TL v.8.2.0.0 and normalized relative to intensities for products generated with WT CUL5 and ARIH2. For samples derived from the same experiment, gels were processed in parallel. Data were processed in Microsoft Excel v.16.16.25 and data points plotted in GraphPad Prism v.8.4.1 (GraphPad Software). All reactions were performed as technical duplicates. All proteins used in assays were roughly 95% pure as judged by Coommassie-stained SDS-PAGE, and molecular weights confirmed by mass spectrometry. For fluorescent ubiquitin, and for ubiquitin-linked proteins, electrophoretic migration was determined by SDS-PAGE and detection by Coommassie staining and fluorescence scanning of the same gel. On this basis, *UB, UBE2L3~*UB, A3G~*UB, CKB~*UB and ARIH2~*UB served as markers of molecular weights 8, 26, 54, 56 and 66 kDa, respectively. The source data files contain the uncropped gel images.
In vitro neddylation assay. Neddylation of CUL5-RBX2 variants was monitored by using a previously described pulse-chase format 41 , except with UBE2F as E2, RBX2 as E3 and CUL5 as substrate. For the chase reaction, the final concentration of UBE2F~*NEDD8 was 0.2 µM, and CUL5-RBX2 (or indicated variant) was 0.5 µM. Reactions were performed at room temperature in 25 mM HEPES, 100 mM NaCl pH 7.5. Samples were taken the indicated time points, quenched with nonreducing 2× SDS-PAGE sample buffer, run on SDS-PAGE and scanned with an Amersham Typhoon imager (GE Healthcare).
Crystallization of autoinhibited ARIH2. The N-terminal 50 residues of ARIH2 are predicted to be disordered 21 and were not included in the version used for crystallization. Crystals of ARIH2 (a version encompassing residues 51 to the C terminus) were grown at 4 °C by the sitting drop vapor diffusion method. Then 10 mg ml −1 protein was mixed in a 1:1 ratio with 0.2 M sodium nitrate, 0.1 M Bis-Tris propane pH 8.5 and 20% PEG 3350 at 4 °C. Crystals typically appeared within 5-10 d. Crystals were cryoprotected in the reservoir solution supplemented with 35% ethylene glycol and flash-frozen in liquid nitrogen.
Crystallographic data collection and structure determination. Diffraction data were collected at the PXIII beamline Swiss Light Source, at −173 °C with a wavelength of 1.2783 Å. The crystal had a rod-like shape. With the goal of obtaining a strong anomalous signal arising from bound zinc atoms in the RING1, IBR and Rcat domains, 360° of data were collected at three different translations across the length of the crystal. The three independent datasets from the single crystal (1,080° of data total) were merged into a single reflection file using XSCALE. Data were indexed, integrated and scaled using XDS 51 . The structure was determined by single anomalous diffraction from zinc atoms, which were located using SHELXC/D/E 52 followed by phase extension using PHENIX Autosol. The resultant map was used in Buccaneer 53 for automatic chain tracing to build the initial model. Further iterative rounds of manual building and refinement were done using COOT 54 and PHENIX 55 . Initially, waters were placed manually in unmodeled density observed in both the 2Fo-Fc map at 1σ and the Fo-Fc map at 3σ contour levels. Next, the 'Update waters' option in PHENIX was used to monitor, add and/or remove waters during the refinement. All waters in the final coordinate file were manually inspected to confirm density in the 2Fo-Fc map. In the final structure, the N-terminal residues of the construct (ARIH2 residues 51-57) and a loop connecting residues 128 and 138 were not visible in the density and hence were not modeled. Pro267 is modeled with a cis peptide bond in both molecules in the asymmetric unit. For the final model, the Clashscore is 6.2 and the Ramachandran statistics are 96% favored, 4% allowed and no outliers. Data collection and refinement statistics are listed in Supplementary Table 1.
Data collection. Cryo-EM datasets were collected using SerialEM v.3.8.0-b5 (ref. 56 ) on a Titan Krios electron microscope at 300 kV with a Quantum-LS energy filter, using a K3 direct detector in counting mode. In total, 9,271 images were collected for the A3C complex and 7,830 images for the A3G complex at a pixel size of 0.8512 and 1.094 Å, respectively. The total exposure ranged from 75 to 90 e − Å −2 and defocus values from −0.7 to −2.5 μm. The data collection statistics are listed in Supplementary Table 2.
For the reconstruction of A3C-bound neddylated CRL5 Vif-CBFβ -ARIH2*, 5,030,529 particles were initially picked using Gautomatch v.056 (K. Zhang, MRC Laboratory of Molecular Biology), followed by two-dimensional (2D) and 3D classification. Initially, particles were binned by a factor of 5, resulting in 4.26 Å per pixel. Subsequent 2D and 3D classifications were done to remove particles belonging to poorly resolved classes. Cryo-EM data for both the A3C-and the A3G-bound neddylated CRL5 Vif-CBFβ -ARIH2* assemblies refined to several similar but nonidentical classes. Even during initial processing, it was apparent that density corresponding to Vif-CBFβ-A3C was relatively poorly resolved, presumably due to heterogeneous orientations of these subunits relative to CUL5. One class with 7,689 particles showed features for the entire complex during 3D classification. This class refined to 6.8-Å resolution, and was low-pass filtered to 7.5 Å to enable docking of subunits (Extended Data Fig. 3a).
Two masks were used for initial consensus refinement, both encompassing the visible density for the entire CRL5 Vif-CBFβ -A3C-ARIH2* complex (Extended Data Fig. 3a). The narrower mask showed more density for the Vif-CBFβ-A3C subunits, resulting in a reconstruction with global resolution of 3.7 Å. The portion of the map corresponding to neddylated CUL5-RBX2-ARIH2* showed clear density with distinct features. Focused 3D classification using a mask covering only ARIH2* bound to neddylated CUL5 (C-terminal region)-RBX2 was used to further improve reconstruction. After iterative rounds of classification, a final set of 191,792 particles were used for final refinement with full pixel size, resulting in a 3.4-Å reconstruction with well-resolved interfaces between ARIH2* and neddylated CUL5-RBX2. The flow chart shown in Extended Data Fig. 3a shows the data processing schemes, including for maps with density for all proteins in the A3C-bound neddylated CRL5 Vif-CBFβ -ARIH2* at 6.8-Å resolution (shown low-pass filtered to 7.5-Å resolution in Fig. 1d and Extended Data Fig. 5a), and for the ARIH2* assembly with a portion of neddylated CUL5 at RBX2 at 3.4-Å resolution. Reported resolution is based on the gold-standard Fourier shell correlation using the 0.143 criterion (Extended Data Fig. 3b,c). Final maps were sharpened using RELION 57 postprocessing or DeepEMhancer 40 . To facilitate model building by improved map quality by local sharpening and noise reduction, two half maps from the final refinement were provided without a mask as input to DeepEMhancer 40 . This resulted in reduced anisotropy and better overall map connectivity. The flow chart for A3G-bound complex processing is shown in Extended Data Fig. 4.
Model building and refinement. A number of maps were used as a guide for model building and refinement. Initially, coordinates from existing crystal structures were fit into the 6.8-Å resolution map (low-pass filtered to 7.5-Å resolution, Electron Microscopy Data Bank (EMDB) EMD-12998), which showed density for all subunits as follows: A3C (PDB ID 3VOW, ref. 39 ); Vif-CBFβ-ELOBC and CUL5 N-terminal domain (chains l, m, n, o and p from PDB ID 4N9F, ref. 36 ), the CR3 domain of CUL5 (PDB ID 6V9I), the crystal structure of ARIH2 from this study (PDB ID 7OD1), and domains from neddylated CUL5 C-terminal domain RBX1 (PDB ID 3DQV) split into two units (the 4HB and C/R domain were fit together as one unit, NEDD8 and its linked CUL5 H29-helix and WHB domain were fit together as another unit) 12 .
The coordinates for a neddylated CUL5-RBX2-ARIH2* subcomplex of the A3C-bound neddylated CRL5 Vif-CBFβ -ARIH2* E3-E3 were subjected to rebuilding, guided by the map processed with DeepEMhancer, and refined using the 3.4-Å resolution postprocessed map arising from focused refinement (A3C E3-E3 catalytic focused in Extended Data Fig. 3a, EMD-12995) 40 . Initial regions of crystal structures were docked in the focus refined map using Chimera v.1.14 (ref. 60 ) and they were allowed to move independently of each other in rigid body refinements using PHENIX 55 . ARIH2*'s UBAL, RING1 and RTI helix region was visible at relatively lower resolution in all cryo-EM maps. Thus, this region of the model was only further subjected to rigid body refinement. The remainder of the structure was subjected to manual model building (including converting the original RBX1 model into the RBX2 protein in this complex) using COOT 54 . Real space refinements were performed in an iterative manner to improve the fit using Phenix.refine 55 . Other than ARIH2* Rcat domain (residues 283-351), most parts of the complex could be resolved except residues 1-34, 51-53, 128-133 and 492-493 of ARIH2*, 6-27 of RBX2 and 1-151, 170-173, 189-193, 386-400 and 675-679 of CUL5, which are not modeled in the final structure. Because the ARIH2* UBAL, RING1 and RTI helix were less well-resolved in the maps, side chains were maintained largely during refinement by restraining them using reference model restraints based on the crystal structures of autoinhibited ARIH2 reported in this study. The side chain of CUL5 Glu717, which is not visible in the density but faces the center of the interface with NEDD8, was modeled based on the crystal structure showing the corresponding portion of neddylated CUL5 (PDB ID 3DQV) 12 , but with zero occupancy. Structures and maps in the figures were rendered with PyMOL or ChimeraX v.1.0.
Analysis of published HDX-mass spectrometry data based on structures of autoinhibited ARIH2 and the ARIH2* complex with neddylated CUL5-RBX2. A previous study compared HDX, quantified by mass spectrometry of tryptic peptides generated after HDX was quenched, for CUL5-RBX2-ELOBC, NEDD8-CUL5-RBX2-ELOBC, ARIH2 and ARIH2-NEDD8-CUL5-RBX2-ELOBC 33 . The authors compared HDX properties between CUL5-RBX2-ELOBC and NEDD8-CUL5-RBX2-ELOBC, between ARIH2 and ARIH2-NEDD8-CUL5-RBX2-ELOBC, and between NEDD8-CUL5-RBX2-ELOBC and ARIH2-NEDD8-CUL5-RBX2-ELOBC and determined peptides that after 0.25 min of HDX showed either statistically significant protection or deprotection on complex formation 33 . Because HDX-mass spectrometry data serve as an independent test of changes in protein conformation, the peptide sequences that show statistically significant protection or deprotection on complex formation were color-coded on the structures of the individual CUL5-RBX2 (ref. 22 ) and autoinhibited ARIH2 E3s, and the neddylated CRL5 Vif-CBFβ -ARIH2* E3-E3 complex (deprotected red, protected-blue, no significant difference white and sequences not detected in the experiments black, Supplementary Video 1). Morphs showing potential trajectories between different conformations were generated using Chimera 60 , and movies were made using PyMOL (The PyMOL Molecular Graphics System, v.2.0 Schrödinger, LLC). Fig. 3 | Cryo-EM image processing flowchart for neddylated CRL5 Vif-CBFβ -ARIH2*-A3C complex. a, Cryo-eM image processing flowchart for the complex containing ArIH2*, neddylated CrL5 Vif-CBFβ , and A3C. Masks are shown in transparent yellow surface superimposed with grey maps. b, From left to right: two views showing local resolution colored on the A3C consensus map for the complex containing ArIH2*, neddylated CrL5 Vif-CBFβ , and A3C; Fourier Shell Correlation (FSC) curve showing the overall resolution of 3.7 Å with the FSC = 0.143 criterion; and angular orientation distribution of the final 3D reconstruction. c, As in b, for the focused refined map (A3C e3-e3catalytic focused) over the catalytic assembly containing ArIH2* and a portion of neddylated CUL5 bound to rBX2. Fig. 4 | Cryo-EM image processing flowchart for neddylated CRL5 Vif-CBFβ -ARIH2*-A3G complex. a, Cryo-eM image processing flowchart for the complex containing ArIH2*, neddylated CrL5 Vif-CBFβ , and A3G. the portion of the reconstructions corresponding to A3G-Vif-CBFβ-eLOBC is visible only at low contour. this subcomplex is presumably mobile relative to the assembly between ArIH2* and neddylated CUL5-rBX2. Masks are shown in transparent yellow surface superimposed with grey maps. b, From top to bottom: two views showing local resolution colored on the A3G consensus map for the complex containing ArIH2*, neddylated CrL5 Vif-CBFβ , and A3G; Fourier Shell Correlation (FSC) curve showing the overall resolution of 3.8 Å with the FSC = 0.143 criterion; and angular orientation distribution of the final 3D reconstruction. Fig. 7 | Interactions between ARIH2 N-terminal region and remodeled CUL5 groove. a, Neddylated CrL5 Vif-CBFβ -dependent fluorescent UB (*UB) transfer to A3G in 10 minutes, catalyzed by Wt ArIH2 or indicated Ariadne domain mutants. (Source Data extended Data Fig. 7). N = 2 independent experiments. For samples from same experiment, gels were processed in parallel. b, Close-up showing cryo-eM density at low contour, over ArIH2*'s N-terminal region (magenta) and neddylated CrL5 groove. Basic residues from neddylated CUL5 (K418, K423, K676 and K685) line the groove and are poised to contact the largely acidic ArIH2 N-terminal stretch (residues 22-34) not modeled in the structure. c, Close-up of cryo-eM structure, showing complementarity between ArIH2* N-terminal region (magenta) and CUL5 groove shown as surface colored by electrostatic potential (red negative, white neutral, blue positive). d, Fluorescent scans of gels showing neddylated CrL5 Vif-CBFβ and ArIH2-dependent *UB transfer from UBe2L3 through e3-e3 cascade for ubiquitylating A3G substrate over time (Source Data extended Data Fig. 7). the assay follows the scheme shown in extended Data Fig. 1d. the assays test effects of indicated multi-Ala mutants in ArIH2 N-terminal domain. residues substituted with Ala are encompassed by the indicated region, for example 22-24 A is 22 A 23 A 24 A. the data are representative from N = 2 independent experiments. e, Assay as b, except with indicated deletion-mutant versions of ArIH2 (Source Data extended Data Fig. 7). f, Chase assays monitoring neddylation of various CUL5-rBX2 mutants as quality control for proper folding. Assays were performed in pulse-chase format, and detect fluorescent *NeDD8. *NeDD8 harbors an N-terminal tag wherein a cysteine is conjugated to a fluorescein-5-maleimide. First, the reactive, thioester-linked UBe2F~*NeDD8 intermediate was generated in a pulse reaction catalyzed by the NeDD8-specific e1 enzyme (the APPBP1-UBA3 complex). After the pulse reaction was quenched, the chase reaction was initiated by adding either Wt or indicated mutant versions of CUL5-rBX2, and *NeDD8 transfer from UBe2F to CUL5 was monitored over time. the deletion mutants of CUL5 all lack the neddylation site (K724) and are: ΔH29 lacking residues 693-725, 1-696 lacking residues 697-780, and 1-712 lacking residues 713-780. Shown are fluorescent scans of SDS-PAGe gels of reaction products (Source Data extended Data Fig. 7). the data are representative from N = 2 independent experiments.