Structural basis for Cul3 protein assembly with the BTB-Kelch family of E3 ubiquitin ligases.

Users may access full items free of charge; copies of full text items generally can be reproduced, displayed or performed and given to third parties in any format or medium for personal research or study, educational or not­for­profit purposes without prior permission or charge, provided: • The authors, title and full bibliographic details is credited in any copy; • A hyperlink and/or URL is included for the original metadata page; and • The content is not changed in any way. Background: BTB-Kelch proteins, including KLHL11, are proposed to bind Cul3 through a " 3-box " motif to form E3 ubiquitin ligases. Results: We solved crystal structures of the KLHL11-Cul3 complex and four Kelch domains. Conclusion: The 3-box forms a hydrophobic groove that binds a specific N-terminal extension of Cul3. Significance: Dimeric BTB-Kelch proteins bind two Cul3 molecules and support a two-site model for substrate recognition.

Ubiquitylation proceeds through a cascade of enzymatic reactions catalyzed by the E1, E2, and E3 enzymes (1,2). The E1 ubiquitin-activating enzyme uses ATP to catalyze the covalent transfer of ubiquitin to the active site cysteine of an E2 ubiquitin-conjugating enzyme. An E3 ubiquitin ligase further catalyzes the transfer of ubiquitin from the E2 to a substrate lysine. Cullin-RING ligases (CRLs) 3 are the largest family of multisubunit E3 ubiquitin ligases and adopt a modular assembly that facilitates the ubiquitylation of divergent substrates. The Cullin subunit (Cul1-5 or Cul7) forms a central stalk-like scaffold that orients and constrains the substrate binding and catalytic centers (3,4). The N-terminal domain (NTD) binds a specific substrate-recognition domain, usually through an adaptor protein, whereas the C-terminal domain (CTD) binds a RING (Really Interesting New Gene) protein (Rbx1 or Rbx2), which in turn recruits an E2-ubiquitin conjugate. Neddylation of the CTD is additionally required to induce conformational changes in the CRL that bring the substrate and E2-ubiquitin into juxtaposition (5,6). The crystal structure of an entire CRL1 complex (also known as a SCF (Skp1-Cul1-F-box) E3 ligase) (7) suggests that different CRLs confer different spacings to allow substrates of varying sizes to be ubiquitylated.
The CRL3 subclass utilizes Cul3, which combines exclusively with BTB-containing proteins as substrate-specific adaptors (8). The BTB domain (Bric-a-brac, Tramtrack, and Broad complex), first characterized by the crystal structure of the promyelocytic leukemia zinc finger protein (9), shares a conserved fold with the Cul1 adaptor Skp1 (10) and the Cul2/5 adaptor ElonginC (11). Moreover, the structure of the SPOP BTB domain in complex with the Cul3 NTD shows an assembly similar to the CRL1 complex of Skp1 and Cul1 (12). However, two features make the BTB adaptors unique among CRLs. First, the BTB adaptor domain dimerizes and is therefore capable of recruiting two Cul3 subunits into the CRL3 complex. Second, the BTB-containing proteins typically host a second proteinprotein interaction domain, so that a single subunit functions as both adaptor and substrate-recognition module. The latter include the C-terminal Kelch, PHR (PAM, Highwire, and RPM-1), or zinc finger domains, whereas SPOP contains an N-terminal MATH (meprin and TRAF homology) domain (13).
The most common substrate-recognition domain in the CRL3 subclass is the Kelch ␤-propeller domain (14), which occurs C-terminal to BTB and BACK (for BTB and C-terminal Kelch") domains (15). The best known example of a BTB-Kelch protein is KEAP1, which associates with Cul3 to regulate cellular levels of the transcription factor Nrf2, a master regulator of the anti-oxidant response (16,17). KEAP1 demonstrates how important the dimeric CRL3 architecture is for substrate ubiquitylation as it requires two Kelch domains to engage two distinct epitopes in Nrf2 simultaneously (18 -21). KEAP1 mutations that disrupt Nrf2 ubiquitylation are associated with lung cancer progression and chemoresistance (22,23). Mutations in the BTB-Kelch proteins KLHL3, KLHL7, and KLHL9 are additionally associated with hypertension (24), retinitis pigmentosa (25), and distal myopathy (26), respectively, although their corresponding substrates have yet to be identified. After KEAP1, the best characterized protein is KLHL12, which modulates COPII assembly for collagen export (27) and also ubiquitylates both the dopamine D4 receptor (28) and dishevelled (29).
Recent structures of SPOP-substrate complexes identify a two-helix extension of the BTB domain that is critical for high affinity Cul3 interaction (30). Defined as the "3-box," this motif appears common to all BTB adaptor proteins (30). To understand the molecular basis for this assembly and to establish structural models for the dimeric BTB-Kelch family, we determined the crystal structures of the BTB-BACK domains of KLHL11 both alone and in complex with Cul3, along with the structures of four representative Kelch domains. The presented data provide a structural model for the understanding of the specific assembly of the BTB-Kelch E3 adaptor proteins with Cul3.
Crystallization and Data Collection-Crystals were grown using the sitting-drop vapor-diffusion technique and cryoprotected before being vitrified in liquid nitrogen (full conditions are listed in Table 1). Diffraction experiments were conducted at 100 K. Data were collected using synchrotron radiation with the exception of KLHL7, for which a full dataset was collected using a Rigaku FR-E SuperBright rotating-anode x-ray generator. Crystallographic data are provided in Table 2. To calculate experimental phases, datasets were collected from crystals of selenomethionine-substituted KLHL11 at a wavelength of 0.9794 Å and from crystals of the KLHL11-Cul3 NTD⌬N22 complex soaked with 2 mM thimerosal at a wavelength of 0.9686 Å.
Structure Determination-Data were integrated using Mosflm (32) or XDS (33) and scaled with SCALA (34) or AIMLESS as part of the CCP4 software suite (35). Experimental phases were calculated, and density modification was carried out with autoSHARP (36). Alternatively, phases were calculated using molecular replacement with PHASER (37) and density modification conducted with PARROT (38). Automated model building tasks were conducted with BUCCANEER (39,40) or PHENIX.AUTOBUILD (41). Manual model building was performed with COOT (42) and the models refined with CNS (43,44), REFMAC (45,46), or BUSTER (47) using TLS and NCS restraints as appropriate. Experimental phase restraints were included in the refinement of the KLHL11-Cul3 NTD⌬N22 complex until the final round of refinement. The model of this Cul3 complex was used as a molecular replacement solution for the higher resolution KLHL11-Cul3 NTD structure. The final model was completed manually and refined to completion. Models were validated using the PHENIX (41) validation tools and/or MOLPROBITY (48).
Isothermal Titration Calorimetry-ITC experiments were performed at 15°C using a Microcal VP-ITC microcalorimeter. Proteins were dialyzed into a buffer containing 50 mM HEPES, pH 7.5, 150 mM NaCl, 0.5 mM tris(2-carboxyethyl)phosphine. Cullin proteins (90 -125 M) were titrated into KLHL11 (10 M). Data were analyzed using a single binding site model implemented in the Origin software package provided with the instrument.

RESULTS
BTB-Kelch Family-The human genome contains some 52 BTB-Kelch family proteins ( Fig. 1 and supplemental Table S1). Their nomenclature is varied, but the protein family can be subdivided into the Kelch-like (KLHL(1-39)) and the Kelch repeat and BTB domain-containing proteins (KBTBD(1-13)). To derive structural models for their function as Cul3-based E3 ligases, members of this family were subcloned and screened for bacterial expression. Soluble expression constructs were identified that contained the BTB-BACK or Kelch domains but not the BACK-Kelch or BTB-BACK-Kelch domains. Crystal structures of the BTB-BACK domains of human KLHL11 were subsequently solved alone and in complex with the N-terminal Cullin-repeat domain of Cul3. Additional structures were solved of the Kelch domains of human KLHL2 (Mayven), KLHL7, KLHL12, and KBTBD5 (see Table 1 for crystallization conditions). A summary of statistics for data collection and refinement is reported in Table 2.
Structure of the BTB-BACK Domains of KLHL11-The structure of the BTB-BACK domains of KLHL11 ( Fig. 2A, residues 67-340) was determined at 2.6-Å resolution using phases calculated from single-wavelength anomalous diffraction collected from selenomethionine-incorporated protein crystals. Two protein chains were present in the asymmetric unit forming a homodimer with an elongated shape of overall dimensions 150 ϫ 35 ϫ 25 Å (Fig. 2B). The BTB domain of KLHL11 closely resembles the common BTB fold of promyelocytic leukemia zinc finger protein (PDB code 1BUO, root mean square deviation 4.65 Å for 118 C␣ atoms) (9). An interchain ␤1-␤5Ј antiparallel ␤-sheet establishes the typical domain-swapped dimer, which is additionally stabilized by the interactions between helices ␣1 and ␣2Ј (where Ј denotes the second chain) (Fig. 2C).
Portions of the all ␣-helical BACK domain, also known as the intervening region, have been previously structurally determined for Gigaxonin (KLHL16) (30) and KBTBD4 (PDB code 2EQX). KLHL11 is the first structure to span the entire domain and contains eight helices in total. The two N-terminal helices (KLHL11 ␣7-␣8) make up the 3-box motif and bind helices ␣5 and ␣6 of the BTB domain in an antiparallel four helix bundle configuration (Fig. 2C). Significantly for Cul3 recognition, all three structures are closely conserved despite limited sequence identity. The remaining C-terminal helices ␣9-␣14 form a distinct subdomain, packing perpendicular to the 3-box. Notably, this arrangement creates a significant cleft some 16 Å deep and  (65) and manually refined with reference to publicly available structures. A phylogenetic tree was created from this alignment using the N-J Tree export functionality of ClustalX and a radial tree figure prepared in PhyloDraw (version 0.8) (66). Further descriptions of each protein are given in supplemental Table S1.    (Fig. 2B). Because the BTB-Kelch proteins are single chain analogs of other CRL substrate adaptors, such as Skp1/␤-TrCP1, the BACK domain was expected to fold similarly to the F-box and helical linker regions of ␤-TrCP1 (15,49). However, superposition of KLHL11 reveals a significant deviation from this arrangement (Fig. 2D). In particular, the long C-terminal helices that support the substrate-binding WD40 (␤-TrCP1) or Kelch (KLHL11) domain show a 50°change in orientation (Fig.  2D). In part, this results from the distinct orientation of the 3-box, which packs perpendicular to the equivalent F-box or SOCS box motifs of other CRL substrate-recognition domains (Fig. 2E). These changes likely reflect the distinct architecture of the CRL3 class imposed by the BTB dimer.

Structural Basis for Cul3 Assembly with BTB-Kelch E3 Ligases the 3-box for Cullin interaction
Structural Diversity of the Kelch Substrate-Recognition Domain-The Kelch domain is the most widespread of the CRL3 substrate-recognition domains and recruits a diverse range of substrates. Structures are previously known only for the Kelch-like protein KEAP1 (KLHL19) (Fig. 3A) (22,50,51) and the Kelch-related protein KBTBD10 (KRP1) from Rattus norvegicus (52). To further characterize the diversity of these substrate-recognition domains, we determined the Kelch domain structures of the Kelch-like proteins KLHL2, KLHL7, and KLHL12, as well as the Kelch-related protein KBTBD5. All structures were refined at high resolution, ranging between 1.6 and 2.0 Å.
Although all four proteins display the typical ␤-propeller structure (Fig. 3B), there are systematic differences between KLHL and KBTBD proteins. The six Kelch repeats form the six "blades" of the propeller, each consisting of a four-stranded antiparallel ␤-sheet (Fig. 3A), with a C-terminal ␤A strand closing the propeller by completing blade I. This site shows diver-gence between the KLHL and KBTBD proteins. The Kelch-like (KLHL) proteins are characterized by consensus repeats, including a Gly-Gly pair that terminates strand ␤B, and hydrophobic Tyr (␤C) and Trp (␤D) residues that pack between blades (supplemental Fig. S1). In contrast, KBTBD5 exhibits an atypical repeat resulting in a more twisted blade I structure, as observed previously for KBTBD10 (52). Both KBTBD proteins also possess an extended C terminus that contributes a short ␤E strand to blade I as well as the usual ␤A (Fig. 3). Sequence comparisons suggest that this altered structure is a common feature of the KBTBD proteins (supplemental Fig. S1).
The surface properties of the six structures are strikingly diverse, reflecting their limited sequence identity. The substrate-binding site, first identified in the KEAP1-Nrf2 complex (51), lies on the narrower upper face of the Kelch domain (Fig.  3A). This surface displays a distinct electrostatic potential in each structure (Fig. 3B). The shape and size of the substrate pocket is determined by the inter-blade DA loops as well as the variable BC loops, which protrude significantly out of the upper propeller face (Fig. 3, A and C). Interestingly, KBTBD5 and KBTBD10 display a rather atypical loop arrangement, with an unusually short BC loop (three residues) in blade I flanked by greatly extended loops (18 and 20 residues, respectively) in blades II and VI (Fig. 3C). By contrast, the BC loops in the KLHL structures are more evenly distributed, with an average loop length of 12 residues. These variant features are expected to contribute to the precise substrate specificity of each BTB-Kelch family member.
Cul3 Binding to the 3-Box Involves a Specific N-terminal Extension Sequence-Substrate ubiquitylation requires the BTB-Kelch proteins to assemble with Cul3. To establish the molecular basis for Cul3 interaction, we adopted a split and express strategy for soluble protein expression (53). The Cul3  MARCH 15, 2013 • VOLUME 288 • NUMBER 11

JOURNAL OF BIOLOGICAL CHEMISTRY 7807
NTD was subcloned with the solubilizing mutations I342R/ L346D and co-purified with the previously characterized BTB-BACK domains of KLHL11. Crystals of the complex were first obtained in space group C2 using a Cul3 construct containing only the three consecutive Cullin-repeat domains (Cul3 NTD⌬22 , Fig. 4A). Diffraction data were collected to 3.1-Å resolution, and the structure was solved by single isomorphous replacement with anomalous scattering (SIRAS) using a mercury derivative. The structural model indicated the potential for an unexpected interaction from the deleted Cul3 N-terminal extension. To address this, we identified a new crystal form in space group I121 containing the full Cul3 NTD (Fig. 4A). The structure of this complex was solved using molecular replacement and refined at 2.8-Å resolution (Fig. 4B). Importantly, the electron density map allowed unambiguous tracing of a further eight N-terminal residues from Cul3 that packed in the hydrophobic groove located with the 3-box between the BTB and BACK domains of KLHL11 (Fig. 4C). An analogous groove was identified in the SCF complexes of Skp1 but was bound to an N-terminal sequence preceding the F-box (10,49).
We were interested to assess the contribution of these Cul3 residues to the interaction and therefore measured the binding affinity of different Cullin proteins by ITC. Cul3 NTD⌬22 containing only the Cullin-repeat domains bound KLHL11 relatively weakly (K D ϭ 0.65 M, Fig. 4D). However, binding affinity was increased 30-fold upon inclusion of the Cul3 N-terminal extension (Cul3 NTD ; K D ϭ 20 nM, Fig. 4D). As expected, there was essentially no binding of KLHL11 to the Cul1 NTD or Cul5 NTD (Fig. 4D), demonstrating the specificity of the interaction.
Structure of the KLHL11-Cul3 Complex-The two crystallized KLHL11-Cul3 complexes exhibited a 2-fold symmetry axis across the BTB dimer, resulting in only one KLHL11 and one Cul3 chain in each asymmetric unit. Generation of the symmetry-related molecules revealed the expected heterotetrameric assemblies, with each subunit in the KLHL11 homodimer binding one molecule of Cul3 (Fig. 4B). KLHL11 was clearly defined by electron density between residues 67 and 336 and the Cul3 NTD between residues 17 and 381. Overall, the contact surface area of the Cul3 NTD interface (1508 Å 2 ) was significantly larger than that of the Cul3 NTD⌬22 complex (1018 Å 2 ), highlighting the importance of the N-terminal extension for Cul3 binding (Figs. 5, A and B, and 6A). By comparison, the SPOP-Cul3 structure (12), which lacks both the N-terminal The sequence coverage of the Cul3 NTD and Cul3 NTD⌬22 constructs used for crystallization and ITC is indicated below. B, ribbon representation of the human KLHL11-Cul3 structure highlighting different functional domains. C, surface representation of KLHL11 showing the interface with Cul3 (cylinder representation) as viewed from below with respect to B. Residues preceding Cul3 H1 (the N-terminal extension) bind the 3-box in a hydrophobic groove located between the BTB and BACK domains and are shown as sticks with the corresponding electron density (2F o Ϫ F c map, contoured at 1) shown in blue. The structures of Cul1 (PDB code 1LDK) (7) and mouse Cul5 (PDB code 2WZK) are superimposed revealing the shorter H2 helix in Cul3. D, ITC measurements of the binding of KLHL11 to different human Cullin N-terminal domains demonstrating the absence of binding to non-Cul3 domains. extension and the 3-box/BACK domain, had a contact area of just 830 Å 2 . Despite its stalk-like appearance, the Cul3 structure is remarkably conserved between the SPOP and KLHL11 complexes (0.8 -1.3 Å root mean square deviation over 300 C␣ atoms), suggesting that its conformation is relatively rigid.
The primary interface in the Cul3 complexes is formed by helices H2 and H5 of the first Cullin repeat, which pack against the BTB and 3-box domains of KLHL11 (Fig. 6B). Sequence conservation in the H2 and H5 helices is strong between Cul3 orthologs but weaker between paralogs (7). Furthermore, the H2 helix of Cul3 is a turn shorter than either Cul1 (7) or Cul5 (PDB code 2WZK) and therefore inserts comfortably into a shallow cleft in the BTB surface (Fig. 4C). In both KLHL11 and SPOP, this cleft forms via an induced fit mechanism facilitated by conformational changes in the ␣3-␤4 loop (Fig. 7). This loop is disordered, or associated with higher crystallographic B-factors, in the unbound crystal structures but becomes helical upon Cul3 binding enabling hydrogen bonds from KLHL11 Ser-131 and Glu-132 to the Cul3 H2 helix (Figs. 6B and 7). In addition, Phe-130 (SPOP Met233) is shifted by some5Åt o insert into a deep hydrophobic pocket formed between Cul3 helices H2 and H4 (Figs. 6B and 7).
The KLHL11 complex reveals for the first time how the 3-box binds to Cul3. The motif, including KLHL11 ␣7 and ␣8, contributes both to the interface with the first Cullin repeat (Fig. 6B) and to the hydrophobic groove that accommodates the Cul3 N-terminal extension sequence (Fig. 6A). The Cullin-repeat interaction is mediated largely by the 3-box ␣7-␣8 loop that packs against the C-terminal regions of Cul3 H2 and H5 (Fig. 6B). Here, the backbone carbonyl of KLHL11 His-213 (␣7) forms a single hydrogen bond with Cul3 Lys-68. The Cul3 N-terminal extension sequence runs antiparallel to the 3-box ␣7 helix and is flanked by the ␣5/␣6 and ␣9/␣10 helices of the BTB and BACK domains, respectively (Fig. 6A). A single hydrogen bond is formed from the side chain oxygen of Cul3 Thr-24 to the backbone nitrogen of KLHL11 Phe-246 (␣10). The remaining N-terminal Cul3 contacts are hydrophobic. In particular, the side chains of Cul3 Ile-18, Ala-20, and Pro-22 insert directly into the hydrophobic groove of KLHL11 to contribute substantially to the overall contact area (Fig. 5A).
Model for the Dimeric CRL3 Complex-The KLHL11-Cul3 structure establishes the core architecture of the dimeric BTB-  Kelch class of E3 ligase. To generate a working model of the complete E3, we built the missing structural domains using other available structures. The Cul3 CTD was built initially from the Cul1 structure (PDB 1LDK) (7) and then modified to fit the active conformation of the neddylated Cul5 CTD -Rbx1 complex (PDB 3DQV) (5). An E2-ubiquitin intermediate was modeled from the UbcH5A-ubiquitin structure (PDB 4AP4) (54) and docked onto Rbx1 by its homology to the Cbl-UbcH7 complex (PDB 1FBV) (55). Finally, the Kelch domain and substrate from the KEAP1-Nrf2 complex (PDB 2FLU) (51) were modeled atop helix ␣14 of the BACK domain of KLHL11. The final model places the two E2-ubiquitin intermediates at the center of the complex where they dissect the axis between the two Kelch domains (Fig. 8A). Flexibility in the linker between the BACK and Kelch domains as well as the limited freedom of the RING domain of Rbx1 may help to bridge the substrate-ubiquitin gap and to break the overall symmetry specified by KLHL11.

DISCUSSION
Here, we show that the interaction of Cul3 with the BTB-Kelch family is unexpectedly two parts. In addition to the expected Cullin-repeat-BTB interaction, we define a novel interaction between the specific N-terminal extension sequence of Cul3 and the 3-box of KLHL11. Previously, the interaction of an N-terminal extension sequence was considered unique to Cul4 (56,57). The surprise interaction of the Cul3 N-terminal region is facilitated by a proximal hydrophobic groove located at the interface of the BTB, 3-box, and BACK domains. Occupying this site enables the N-terminal extension to form contacts across all three KLHL11 domains and therefore to contribute significantly to the overall binding affinity. Notably, the bound Cullin-repeat is also enveloped by the BTB ␣3-␤4 loop that refolds around the relatively short H2 helix like a molecular clamp. A similar BTB adaptor interface is found in the SPOP-Cul3 complex (12) suggesting that this is indeed a conserved mechanism of Cul3 interaction.
Upon assembly with Cul3, the BTB-Kelch proteins may direct dynamic control of ubiquitylation through the bivalent coordination of a single substrate molecule. A "fixed-ends"  model is proposed for Nrf2 recruitment in which high (ETGE) and low (DLG) affinity recognition motifs are tethered to the two Kelch domains of a KEAP1 homodimer to promote ubiquitylation of a central lysine-rich ␣-helix (18 -21). The presented structures offer a molecular model to support this hypothesis (Fig. 8B). The elongated BTB-BACK domains establish a spacer suitable for Nrf2 recruitment while orienting the associated Cullin-RING complexes to position the E2 molecules centrally for ubiquitin transfer. The architecture of the KEAP1 protein has also been determined by single particle electron microscopy (EM) (20). Consistent with this study, the reconstruction at 24 Å resolution revealed an elongated structure with a 2-fold symmetry axis. Two distinct globular domains were attached by short linker arms to a central stem. The crystal structures indicate that the linker likely corresponds to the 3-box that separates the BTB stem from the globular domains, each comprising a Kelch domain atop the BACK domain helices ␣9-␣14. In the EM reconstruction, the positions of the two substrate-binding sites were somewhat flexible with an average separation of 80 Å (20). Some 47 residues separate the ETGE and DLG motifs, giving a theoretic span of 98 Å, assuming a 33-residue helix (18). Our structural model provides a span of 95 Å, broadly consistent with the EM data.
Dimerization is also required for the E3 activity of some SCF complexes, including those of ␤-TrCP1 and Cdc4 (58). In contrast to the BTB adaptors, these CRL1 complexes dimerize through the D domain of the substrate-recognition module (58). Their predicted assemblies also position the substrate centrally to two catalytic centers, although a distinct configuration is enforced due to the alternative mode of dimerization. In addition to robust substrate capture, dimeric E3s may confer greater spatial variability to enable efficient ubiquitylation of diverse substrate lysine acceptor sites. E3 interaction is also thought to bias dynamic E2-ubiquitin ensembles toward a conformation with enhanced reactivity for substrate lysines. In some E3 classes, this reaction is catalyzed by dimerization of the RING domain (54,59,60), although it remains unclear how this mechanism could be utilized by the CRL families. Given the relative positions of the Rbx1 subunits in the structural model, such a mechanism would likely require higher order CRL3 assembly, as suggested for the MATH-BTB protein SPOP (12).
Uniquely, the BTB-Kelch family proteins integrate the functions of both CRL adaptors and substrate receptors. Cul3 must therefore assemble with a large number of distinct BTB adaptor domains, whereas other Cullins bind a common adaptor protein such as Skp1 or ElonginC. As a consequence, BTB E3 ligases may offer a greater diversity of drug-targeting sites for investigation. In this respect, several sites in the KLHL11 structure are of interest, including the BTB-dimer interface, the Cullin-repeat interface, and the hydrophobic groove of the 3-box. Indeed, disruption of Cul3 N-terminal interactions with the hydrophobic groove resulted in a 30-fold loss in affinity. A similar decrease was observed upon deletion of the 3-box in SPOP (30). Promisingly, small molecules have been identified previously that target the corepressor binding groove of the BCL6 BTB domain (61) as well as the substrate binding grooves of E3 ligases, including MDM2 (62), VHL (63), and Skp2 (64).
The presented structures reveal the novel architecture of BTB-Cul3 assembly and impact our mechanistic understanding of CRL3 activity with potential therapeutic implications.