Dimerization of Substrate Adaptors Can Facilitate Cullin-mediated Ubiquitylation of Proteins by a “Tethering” Mechanism

The prevalence and mechanistic significance of self-association among substrate adaptors for the Cul-Rbx family of ubiquitin ligases remain unclear. We now report that it is as a homodimer that the substrate adaptor Keap1 interacts with Cul3. The resulting complex facilitates ubiquitylation of the Nrf2 transcription factor but only when this substrate possesses within its Neh2 domain a second cryptic Keap1-binding site, the DLG motif, in addition to its previously described ETGE site. Both motifs recognize overlapping surfaces on Keap1, and the seven lysine residues of Nrf2 that act as ubiquitin acceptors lie between them. Based on these data, we propose a “fixed-ends” model for Nrf2 ubiquitylation in which each binding site becomes tethered to a separate subunit of the Keap1 homodimer. This two-site interaction between Keap1 and Nrf2 constrains the mobility of the target lysine residues in the Neh2 domain, increasing their average concentration in the vicinity of the Rbx-bound ubiquitin-conjugating enzyme, and thus the rate at which the transcription factor is ubiquitylated. We show that self-association is a general feature of Cul3 substrate adaptors and propose that the fixed-ends mechanism is commonly utilized to recruit, orientate, and ubiquitylate substrates upon this family of ubiquitin ligases.

Ubiquitylation underpins virtually all biological processes as it is the major mechanism regulating the stability of critical effector molecules in eukaryotic cells. It refers to the formation of an isopeptide bond between the C terminus of ubiquitin and the ⑀-NH 2 group of a lysine residue in a target protein. The reaction proceeds in three stages with the final critical step, transfer of activated ubiquitin from an E2 2 ubiquitin-conjugat-ing enzyme to an acceptor lysine, facilitated by E3 ubiquitin ligases (1). Modular complexes based around at least four Cullin-RING box (Cul-Rbx) holoenzymes constitute the largest family of E3 ligases identified to date. These holoenzymes, Cul1-Rbx1, Cul2-Rbx1, Cul3-Rbx1, and Cul5-Rbx2, differ in the nature of the substrate adaptors they bind (2)(3)(4)(5). For example, Cul1-Rbx1 recruits protein dimers comprising the S-phase kinase-associated protein 1 (Skp1) bound to the eponymous domain found in over 40 F-box proteins (2). The resulting E3 ligase is termed SCF F-box (Skp1, Cul1, and F-box, with the specific F-box protein identified in supercript). Cul3-Rbx1 recruits Broad complex, Tramtrack, and Bric-a-brac (BTB) proteins to generate a large family of ligases that, by analogy with SCF ligases, are referred to as BC 3 B BTB ubiquitin ligases (6 -9). 3 Structural similarities exist among substrate adaptors. For example, Skp1 and BTB proteins all utilize BTB folds to interact with Cul proteins (3). Additionally, the domains utilized to recruit substrates can adopt analogous super-secondary structures; whereas F-box proteins frequently use WD40 domains to recruit substrates and BTB proteins commonly exploit Kelchrepeat domains for this purpose, both domains fold to give sixbladed ␤-propeller structures (3,10,11). In conjunction with the seminal crystallographic study on the structure of the SCF F-box of Skp2 complex (12), these common features among adaptors suggest that probably all Cul-Rbx ligases function by recruiting and juxtaposing an E2 enzyme, via a C-terminal Rbx protein, and a substrate, via an N-terminal adaptor.
The above model, in which a single substrate adaptor interacts with an E3 holoenzyme, may be incomplete as two examples exist supporting the idea that self-association of F-box proteins is required for ubiquitylation of some substrates by Cul1-Rbx1. First, association of two F-box proteins, Pop1p and Pop2p, is necessary for polyubiquitylation of Rum1p in fission yeast (13,14). Second, two F-box proteins, ␤-transducin repeatcontaining protein 1 (␤TrCP1) and ␤TrCP2, form hetero-and homo-oligomers with each other, but only the homo-oligomers could target phosphorylated IB␣ for ubiquitylation (15). The various rationales proposed to explain the necessity for adaptor self-association (discussed in Ref. 3) remain unsupported by evidence, and it is not clear whether these findings are idiosyncratic or of general significance. In particular, an analogous role for BTB protein dimers in substrate ubiquitylation by Cul3-Rbx1 has not been reported. Yet the BTB adaptor protein Keap1 (Kelch-like ECH-associated protein 1) homodimerizes in bacteria (16,17). Additionally, evidence from the BTB domains of both B-cell lymphoma 6 (BCL6) and promyelocytic leukemia zinc finger (PLZF) transcription factors (18,19) suggests dimerization of BTB domains may be obligatory (20).
Keap1 recruits the antioxidant NF-E2 p45-related factor 2 (Nrf2) transcription factor to the BC 3 B holoenzyme (21)(22)(23)(24)(25). The domain structure of this substrate is depicted in Fig. 1A with particular emphasis placed on its Nrf2-ECH homology 2 (Neh2) domain. For it is the ETGE motif found therein that directly interacts with the Kelch-repeat domain of Keap1 ( Fig.  1D) (26). Under normal redox conditions, this interaction results in ubiquitylation of the Nrf2 protein and, as a consequence, repression of its steady-state level. The ability of BC 3 B Keap1 to ubiquitylate the factor is inhibited by oxidative modification of specific cysteine residues in Keap1 (17,(27)(28)(29)(30)(31), and this allows rapid repletion of Nrf2 protein levels in stressed cells. As Nrf2 promotes the transcription of genes whose products promote reductive chemistries, Keap1 provides the cell with a negative feedback control loop that plays a pivotal role in maintaining redox homeostasis (32)(33)(34)(35).
We now report that Keap1 in mammalian cells exists as a dimeric protein and that it interacts with Cul3-Rbx1 in this oligomeric form. Additionally, we demonstrate that ubiquitylation of Nrf2 by BC 3 B Keap1/Keap1 4 requires a second previously FIGURE 1. Domain structure of Nrf2 and Keap1. A, sequence conservation in the Neh2 domain of Nrf2 in mouse (m), human (h), rat (r), chicken (g), and Zebrafish (z). The DIDLID element (amino acids (a.a.) [15][16][17][18][19][20][21][22][23][24][25][26][27][28], the DLG (amino acids 29 -31), and ETGE (amino acids 79 -82) motifs are in green, blue, and pink, respectively. CncC and Skn-1 are Drosophila melanogaster and C. elegans proteins that are considered functionally equivalent to Nrf2 (44); note that Skn-1 contains neither an ETGE or DLG motif, whereas CncC contains both. B, various deletions and mutations were made to the Neh2 domain during the course of the study, and the following schematics are provided to aid the reader. The ⌬ is used to signify deletions, such as ⌬17-32 or ⌬ETGE. Multiple deletions are separated by a comma, e.g. ⌬17-32,ETGE. When mutations are targeted to a specific motif, this is signified by the abbreviation mut immediately followed by the name of the motif, e.g. mutDLG or mutDIDLID. These can be combined with deletions so that mutDLG⌬ETGE is used to indicate a protein with a mutated DLG motif and a deleted ETGE motif. C, the mutations targeted to the DLG motif and DIDLID element of mNrf2 are highlighted in red and in the case of mNrf2 mutDLG reflect the Skn-1 sequence. D, domain structure of mKeap1 based on in silico analyses.
unrecognized Keap1-binding site, the DLG motif, to be present in its Neh2 domain (Fig. 1A). Based on these findings, we present a "fixed-ends" model for recruitment and orientation of Nrf2 upon BC 3 B Keap1/Keap1 . The model accounts for the target lysine specificity displayed by the E3 complex and also provides a generally applicable rationale for substrate adaptor dimerization.
Bacterial Expression and Purification of Protein-To purify recombinant hexahistidine-mKeap1, the pET15bmKeap1 plasmid was transformed into Escherichia coli BL21(DE3)pLysS (Novagen) and selected on LB agar containing 100 g/ml ampicillin and 34 g/ml chloramphenicol. One colony was used to inoculate 200 ml of LB broth containing 500 g/ml ampicillin and 34 g/ml chloramphenicol, and the culture was grown at 37°C to an A 600 of 0.4. Expression of hexahistidine-mKeap1 protein was induced by addition of isopropyl ␤-D-thiogalactopyranoside to a final concentration of 1 mM, and after a further 2 h at 30°C the bacteria were harvested, resuspended in 5 ml of binding buffer (20 mM Tris-Cl, pH 7.9, 5 mM imidazole, 0.5 mM NaCl, and 0.01% (v/v) Nonidet P-40), lysed by addition of 1 mg of lysozyme, and sonicated. Hexahistidine-mKeap1 inclusion bodies were harvested by centrifugation (10,000 ϫ g, 15 min, 4°C), and washed with two sequential volumes of 20 mM Tris-Cl, pH 7.5, 10 mM EDTA, and 1% (v/v) Triton X-100. The final pellet was solubilized in Laemmli sample buffer. The purity of the hexahistidine-mKeap1 (Ͼ95%) and concentration were determined by SDS-PAGE followed by Coomassie staining. To purify MBP-mKeap1 ⌬1-307 , pMal-mKeap1 ⌬1-307 was transformed into E. coli BL21(DE3)pLysS, and MBP-mKeap1 ⌬1-307 was induced as described above. Post-induction, the bacteria were resuspended in 5 ml of Column buffer (20 mM Tris, pH 7.6, 200 mM NaCl, 1 mM EDTA), lysed by addition of 1 mg of lysozyme, and sonicated. The lysate was clarified by centrifugation (10,000 ϫ g, 15 min, 4°C), sterile-filtered, and passed through an amylose column (New England Biolabs). After extensively washing the column, the MBP fusion protein was eluted in Column buffer supplemented with 10 mM maltose. Fractions containing protein were dialyzed against 2 volumes of 10 mM HEPES, pH 7.6, 150 mM NaCl, and 0.01% (v/v) Nonidet P-40, and stored at Ϫ80°C.
Whole-cell Extracts, in Vivo Ubiquitylation Assay, Immunoprecipitation, and Immunoblots-For immunoblots, whole-cell lysates were prepared by scraping cell monolayers into ice-cold radioimmune precipitation assay (RIPA) buffer (50 mM Tris-Cl, pH 7.4, 150 mM NaCl, 1% (v/v) Nonidet P-40, 0.5% (w/v) deoxycholate, 0.1% (w/v) SDS). Lysates were clarified by centrifugation (16,000 ϫ g, 15 min, 4°C). The in vivo ubiquitylation assay was carried out as described previously (36). To immunoprecipitate EGFP-, V5-, or Myc-tagged proteins, clarified wholecell lysates were prepared as described above, except that deoxycholate and SDS were omitted from the RIPA buffer. On occasion, clarified lysates that contained different ectopic proteins were mixed together. 1 g of either rabbit anti-GFP, goat anti-V5, or mouse anti-c-Myc (clone 9E10) (all from Abcam) was added to each clarified lysate and mixed by end-over-end tumbling at 4°C for 2 h. At this point, immunocomplexes were gathered with protein G-Sepharose 4B (Sigma) at 4°C for 15 min. Material that remained bound to the resin after washing with modified RIPA buffer was eluted in 50 l of Laemmli reducing sample buffer. Biochemical analyses were carried out by standard methods (36). Antibodies used included mouse anti-V5 and mouse anti-Xpress (Invitrogen), goat anti-hKeap1 (E20; Santa Cruz Biotechnology), mouse anti-HA, (Roche Applied Science), rabbit anti-GFP, goat anti-c-Myc, and rabbit anti-FLAG (the latter three were all from Abcam). During this study, a rabbit anti-mKeap1 antiserum was generated by immunization of a female New Zealand White rabbit with bacterially expressed MBP-mKeap1 ⌬1-307 protein and was used at a dilution of 1:5000.
Gel Filtration-Confluent monolayers of cells in 60-mm dishes were lysed by scraping them into 300 l of ice-cold 50 mM sodium phosphate, pH 7.2, 150 mM NaCl, containing 1% (v/v) Tween 20. The clarified lysate was applied to a Superdex-200 10/300 GL 10 ϫ 300-mm column (Amersham Biosciences) and was eluted at 0.25 ml/min with lysis buffer. Fractions of 250 l were collected.

RESULTS AND DISCUSSION
Keap1 Protein Exists as a Stable Dimer, but Not as a Monomer, in Mammalian Cells-To determine whether Keap1 oligomerizes in mammalian cells, we co-expressed EGFP-and V5-tagged forms of the protein in COS1 cells. After 24 h, wholecell lysates were prepared from each dish of cells. EGFP-tagged mKeap1 was immunoprecipitated from these lysates and probed for the presence of V5-tagged adaptor by immunoblotting. The V5-tagged protein specifically co-immunoprecipitated with its EGFP-tagged counterpart ( Fig. 2A, cf. lanes 1 and 2 with lanes 7 and 8), indicating that Keap1 self-associates.
These data also show that the interaction of the adaptor protein with itself is unaffected by treatment of COS1 cells with doses of the oxidative stressor Sul that are sufficient to inhibit BC 3 B Keap1 activity (36,41).
Inspection of the data in Fig. 2A revealed that the proportion of mKeap1 in the immunoprecipitate that contained a V5 tag was similar to that in the corresponding wholecell lysate ( Fig. 2A, lanes 1 and 2). This is consistent with self-association of the adaptor proceeding to completion. In agreement with this, when we size-fractionated lysates from homeostatic COS1 cells expressing untagged mKeap1, the protein eluted as a single peak ( Fig.  2B (i) and (ii)), with an estimated mass of 147-162 kDa, a value that is approximately twice the theoretical molecular weight of a Keap1 polypeptide (69.5 kDa). These data reveal that overexpressed Keap1 exists solely as a stable dimer under steady-state conditions. Self-association is not an artifact of overexpression as endogenous Keap1 also exhibits this property. To unequivocally identify the endogenous protein, we used two immunological reagents that interact with separate domains of the adaptor. Goat anti-hKeap1 is available commercially and was raised to a peptide within the BTB domain of the adaptor protein. In addition, we generated an antiserum against the Kelch-repeat domain of mKeap1. Both of these reagents recognized hexahistidine-mKeap1 standards (Fig. 3, A and B, lanes [1][2][3]. They also bound to ectopic untagged and EGFP-tagged forms of mKeap1 (Fig. 3, A and B, lanes 5 and 6). Critically, only one endogenous protein in COS1 lysates was recognized by both reagents, and it comigrated with untagged mKeap1 (Fig. 3, A and B, lane 4). This protein, which we identify as Keap1, was present at approximately only 10 ppm of extracted COS1 protein. In an immortalized rat liver epithelial cell line, RL34, and MEFs, the adaptor is present in even lower amounts being below the limit of detection of our assay (approximately 2 ppm of extracted protein) (Fig. 3, A and B, lanes [7][8][9]. Endogenous Keap1 eluted as a single peak from a Superdex-200 column with proteins of 141-155 kDa (Fig. 3C), indicating that it exists solely as a dimer.
Dimeric Keap1 Associates with Cul3-The above experiments suggest it is dimeric and not monomeric Keap1 that interacts with Cul3. We developed a co-immunoprecipitationbased assay to further investigate this issue.
First, we confirmed that Xpress-tagged mCul3 co-immunoprecipitated with V5-tagged mKeap1 (Fig. 4, cf. lanes 1 and 2, and lanes 3 and 4). Next, we generated a mutant form of Keap1 that was unable to bind Cul3. To do this, we created a model of . V e indicates the elution volume for protein of interest; V o indicates the void volume (7.7 ml); V t indicates the total column volume (24 ml).
the Keap1 BTB homodimer structure based on those of BCL6 and PLZF (supplemental Fig. 1, A and B). Based on this model and its structural homology with Skp1, it was predicted that Val-123, Ile-125, Gly-127, Met-161, and Tyr-162 of mKeap1 (see supplemental Fig. 1C) correspond to those residues in Skp1 that make contact with Cul1 (7). In support of this interpretation, when alanines were substituted for these five residues, the resulting mutant protein, mKeap1 mutVIGMY -V5-his, had attenuated affinity for Xpress-tagged mCul3 (Fig. 4, cf. lanes 5 and 6  with lanes 1 and 2). Finally, we reasoned that if Keap1 interacted with Cul3 as a dimer, then expression of untagged wild-type mKeap1 should rescue the failure of the V5-tagged mutant counterpart to associate with Cul3. This was verified to be the case (Fig. 4, cf. lanes 7  and 8 with lanes 5 and 6). These findings strongly support the idea that the Keap1 homodimer interacts with Cul3. Note that these coimmunoprecipitation experiments utilized mCul3 lacking amino acids 597-617, which are essential for the protein to bind Rbx1 (9). This mutation enhances the expression level of mCul3 (data not shown).
Keap1 Function, but Not Its Tendency to Self-associate, Is Sensitive to Amino Acid Substitutions in Its BTB Domain-We extrapolated from the above data that BC 3 B Keap1/Keap1 complexes form in intact cells. This raised the possibility that dimerization of the adaptor plays a mechanistic role during ubiquitylation of Nrf2 by the BC 3 B holoenzyme. One approach to test this hypothesis is to engineer monomeric forms of Keap1 and evaluate their ability to destabilize Nrf2. To generate a monomer from Keap1, we substituted alternative amino acids for residues Asn-68, Arg-71, Ser-102, Ala-140, or Ala-143 of the adaptor. These residues were chosen by virtue of their presence at the interface of the predicted Keap1 BTB dimer (supplemental Fig.  2, A and B); additionally, these residues are moderately well conserved in, and occur at the interface of, the PLZF and BCL6 dimers. However, our studies on the resulting mutant proteins, summarized in Fig. 5A, reveal that this approach is unsuited to testing the hypothesis in question for two reasons.
First, dimerization of Keap1 is a robust property of the protein. Even the presence of multiple amino acid substitutions was insufficient to inhibit adaptor self-association (Fig. 5A). Thus, it is likely that considerable protein engineering would be required to derive a monomeric protein from wild-type Keap1. This conclusion is supported by the unsuccessful efforts to engineer monomeric forms of another BTB protein, PLZF. Melnick et al. (20) found that mutations introduced into PLZF aimed at generating a monomeric protein either failed to inhibit dimerization or, alternatively, caused aggregation and misfolding of the resulting proteins. Based apparently on this work, Zipper and Mulcahy (42) have reported that substituting an alanine for Ser-104 of Keap1 is sufficient to inhibit dimerization and function of Nrf2. In fact, Melnick et al. (20) reported that the corresponding mutation in PLZF was functionally silent, having no effect on dimerization of PLZF or its transcriptional competency. Congruent with this latter result, we found that mutating Ser-104 to an alanine had no effect upon Keap1 dimerization and only mildly affected its ability to mediate ubiquitylation of Nrf2 (data not shown).
Second, some of the mutant Keap1 proteins were nonfunctional, but critically, they fully retained the capacity to selfassociate, bind Cul3, and interact with Nrf2. For example,  . mKeap1 interacts with mCul3 as a homodimer. The indicated proteins were expressed heterologously in COS1 cells. One day later, wholecell lysates were prepared from duplicate dishes of cells; V5-tagged proteins were immunoprecipitated, and both whole-cell lysates (input) and immunoprecipitation (IP) fractions were blotted with mouse anti-V5 and mouse anti-Xpress. Note that for this co-immunoprecipitation experiment (and all others involving Xpress-mCul3), the blot for immunoprecipitated samples was exposed for considerably longer than that for the input samples because of the weakness of the interaction of mKeap1 and mCul3. If exposed for the same length of time as the immunoprecipitation blot, the input signals became saturated and uninformative. WB, Western blot. mKeap1 bearing an Arg-71 to glycine substitution was unable to destabilize Nrf2 (Fig. 5B, cf. lanes 9 and 10 with 5 and 6). This deficiency is unrelated to its reduced expression as it was expressed at similar levels to the Asn-68 to glycine-bearing mutant that remained competent to destabilize mNrf2-V5 (Fig.  5B, lanes 7 and 8), and neither can it be ascribed to a reduction in its tendency to self-associate. Although this mutant protein migrated through a Superdex-200 column with proteins of 94 -104 kDa, as opposed to 141-155 kDa for mKeap1 (Fig. 5C), our co-immunoprecipitation assay for Keap1 self-association indicated that the presence of the Arg-71 to glycine mutation had no effect on adaptor dimerization (Fig. 5D, cf. lanes 7 and 8  with lanes 5 and 6). Finally, the mutant also retained the ability to interact with both Xpress-mCul3 (Fig. 5E) and mNrf2-V5 (Fig. 5F). It remains unclear why this mutant is nonfunctional. However, this result is significant because it highlights the fact that even if we succeeded in creating a monomeric protein from Keap1 that was nonfunctional, there would be no logical basis FIGURE 5. The functional integrity of Keap1 is sensitive to mutations in its BTB domain. A, summary of the consequences of mutating the indicated residues of Keap1. B, plasmids encoding the indicated proteins were co-transfected into COS1 cells. After 24 h, cell lysates were prepared from duplicate dishes of cells and blotted with mouse anti-V5 or rabbit anti-mKeap1. WB, Western blot. C, lysates from COS1 cells expressing mKeap1 or mKeap1 R71G were sizefractionated on a Superdex-200 column. A series of 0.25-ml fractions were collected between elution volumes 7.36 and 16.36 ml, and portions of each were blotted with rabbit anti-mKeap1; to save space, only fractions from 10.36 to 13.86 ml are shown. wt, wild type. D, plasmids expressing the indicated proteins were co-transfected into duplicate dishes of COS1 cells. After 24 h, cell lysates were prepared from each dish of cells, and a portion of each was retained as an input sample. EGFP-tagged proteins were immunoprecipitated (IP) from the remaining portion; both input and immunoprecipitation samples were blotted with rabbit anti-GFP and mouse anti-V5. E and F, plasmids encoding the indicated proteins were transfected individually into COS1 cells, and these were subsequently used to prepare cell lysates containing only one specific ectopic protein. Lysates containing individual ectopic proteins were mixed as indicated, and a portion of each mixture was retained as an input sample. EGFP-tagged proteins were immunoprecipitated from the remainder of each mixture, as described under "Materials and Methods," and both input and IP fractions were blotted with rabbit anti-GFP (E and F), mouse anti-Xpress (E), and mouse anti-V5 (F). The asterisk (E, input samples re-blotted with rabbit anti-EGFP) indicates a residual signal for Xpress-mCul3 from the previous immunoblot of this membrane.
to conclude that the loss-of-function phenotype was directly because of the inability to dimerize.
Consequently, we chose an alternative approach to study the role, if any, played by dimerization of Keap1 during ubiquitylation of Nrf2 by BC 3 B Keap1/Keap1 . The strategy adopted was to investigate the structural features of the substrate required for its ubiquitylation by the E3 complex.
The DLG Motif Is Essential for Ubiquitylation of Nrf2 by BC 3 B Keap1/Keap1 -Deletion of amino acids 17-32 from the Neh2 domain of mNrf2 generates a protein that Keap1 cannot destabilize despite the overall avidity of the substrate for the adaptor remaining unchanged (43). One explanation for this finding is that residues 17-32 might contain a second comparatively weak Keap1-binding site. Consequently, its deletion would not reduce measurably the avidity of Nrf2 for the BTB protein, although its presence might be essential for ubiquitylation of the factor by BC 3 B Keap1/Keap1 . To test this hypothesis, we mixed COS1 lysates containing mutant forms of mNrf2-V5 with lysates that contained EGFP-mKeap1. As shown in Fig.  6A, mNrf2-V5 specifically co-immunoprecipitated with the mKeap1 fusion protein (cf. lanes 3 and 4 with lanes 1 and 2), and deletion of amino acids 17-32 did not obviously impair this interaction (Fig. 6A, lanes 5 and 6). By contrast, removal of the ETGE motif greatly diminished the interaction between substrate and adaptor. Nonetheless, a comparatively weak interaction between the two proteins was consistently detected in the absence of this motif (Fig. 6A, lanes 7  and 8). Amino acids 17-32 account for this residual interaction (Fig. 6A, lanes 9 and 10).
Based on phylogenetic conservation, residues 17-32 can be subdivided into two regions (44). As shown in Fig. 1A, the DLG motif is distinguished from the DIDLID element by the fact that it is absent from Skn-1, the distant Caenorhabditis elegans orthologue of Nrf2 that lacks also the ETGE motif (45). Introduction of mutations into mNrf2 ⌬ETGE -V5 that selectively target one or the other of these two regions (see Fig. 1C), revealed that the remaining affinity of the mutant substrate for mKeap1 was associated specifically with the DLG motif (Fig. 6B).
The DLG motif is indispensable for BC 3 B Keap1/Keap1 -mediated ubiquitylation of Nrf2. For example, V5-tagged mNrf2 and mNrf2 mutDIDLID are destabilized by Keap1 in homeostatic cells and accumulate in oxidatively stressed COS1 cells. However, mNrf2 mutDLG -V5 is resistant to such degradation (Fig. 7A). The enhanced stability of this latter protein, compared with its wildtype counterpart, correlates with a reduced rate of ubiquitylation by BC 3 B Keap1/Keap1 in homeostatic cells (Fig. 7B, cf. lanes 3  and 4 with lanes 5 and 6).
Critically, mutation of the DLG motif has no effect on any property of Nrf2 other than its ability to act as a substrate for BC 3 B keap1/Keap1 . The ability of the transcription factor to transactivate gene expression is unaffected by mutation of the DLG motif (data not shown). The activity of its Neh6 degron, which determines the half-life of the protein in redox-stressed cells (43), is likewise unaffected (data not shown). More convincingly, mutation of DLG does not affect the capacity of the adjacent DIDLID element to mediate protein degradation. This previously unrecognized activity of the DIDLID element was revealed by studying the turnover in keap1 Ϫ/Ϫ MEFs of a fusion protein composed of the mNeh2 domain attached to an HAtagged Gal4 DNA-binding domain (Gal4(HA)mNeh2). CHX chase analysis of this fusion protein revealed it to be degraded by a mechanism requiring the DIDLID element (Fig. 7C). This pathway of degradation was unaffected by mutation of the adjacent DLG motif (Fig. 7C). Taken in total, these observations suggest that the evolution of the DLG motif is specifically related to the mechanism by which BC 3 B Keap1/Keap1 ubiquitylates Nrf2.
The DLG Motif Is a Second Keap1-binding Site-As the DLG motif interacts (albeit comparatively weakly) with Keap1, we hypothesized it is essential for BC 3 B Keap1/Keap1 -mediated ubiquitylation of Nrf2 because it acts as a second Keap1-binding site. Two additional observations also support this hypothesis. First, the sequences surrounding both motifs share certain similarities (Fig. 8A) as follows: a Gly in position 0 is conserved, as is an Asp in position Ϫ4; both sequences contain a negatively FIGURE 6. The DLG motif constitutes a second Keap1-interaction site in Neh2. Whole-cell lysates were prepared from duplicate dishes of COS1 cells heterologously expressing mNrf2-V5, mNrf2 ⌬17-32 -V5, mNrf2 ⌬ETGE -V5, and mNrf2 ⌬17-32,ETGE -V5 (A) or mNrf2 ⌬ETGE -V5, mNrf2 mutDIDLID⌬ETGE -V5, mNrf2 mutDLG⌬ETGE -V5, and mNrf2 ⌬17-32,ETGE -V5 (B) (see Fig. 1B for nomenclature and schematics describing these mutations). These were combined with an equal volume of whole-cell lysates from COS1 cells expressing EGFP-mKeap1 or mock-transfected, as indicated. wt, wild type. A portion of each mixed extract was kept as an input sample. EGFP-mKeap1 was immunoprecipitated from the remaining portions, and both input and immunoprecipitation (IP) fractions were blotted with goat anti-hKeap1 and mouse anti-V5. WB, Western blot. charged amino acid in position Ϫ2. Second, NMR analysis reveals the Neh2 domain to be an extended molten globule-like protein lacking tertiary structure (46). According to these data, the ETGE motif occurs in a loop connecting two antiparallel ␤-strands, and the DLG motif occurs in a flexible, unstructured loop. The lack of a rigid structure around the DLG motif suggests it may be able to adopt a similar conformation to that of the ETGE motif and may utilize the same binding interface on Keap1.
At least two testable predictions stem from the above hypothesis and allowed its evaluation. First, if the DLG motif represents an authentic Keap1-binding site, it should be possible to evolve it into a high affinity Keap1-binding site without compromising the capacity of the resulting protein to be degraded by BC 3 B Keap1/Keap1 . We tested this by introducing mutations sequentially and cumulatively into residues 27-32 of mNrf2 ⌬ETGE -V5, guided by the sequence around the ETGE motif (Fig. 8A). Neither the protein harboring the D29E single mutation (M1) or that containing a D29E/L30T double mutation (M2) exhibited an observable gain in affin-ity for Keap1 as assessed by a co-immunoprecipitation assay (this was apparent from longer exposures of the blot shown in Fig. 8B). By contrast, an additional V32E mutation resulted in a protein (M3, containing three amino acid substitutions) with considerably increased affinity for mKeap1, and this was further augmented by also mutating Ile-28 to Glu (M4; Fig. 8B). Crucially, when these four mutations were introduced into wild-type mNrf2, the resulting protein, mNrf2 M4 -V5, remained subject to BC 3 B Keap1/Keap1 -dependent degradation in homeostatic COS1 cells (Fig. 8C).
Second, the recent solution of the structure of an ETGEcontaining peptide complexed with the Kelch-repeat domain of mKeap1 has revealed that this motif binds to an arginine triad (Arg-380, Arg-415, or Arg-483) at the top face of the Keap1 ␤-propeller (see Fig. 9, A and B) (47). In agreement with this, mNrf2 mutDLG -V5, which interacts with mKeap1 via the ETGE motif, exhibited attenuated affinity for the adaptor protein when any one of these three arginines was substituted with a Met residue (Fig. 9C). Our hypothesis predicts that these mutations should equally affect binding of the DLG motif to Keap1. When tested, the outcome was in excellent agreement with this FIGURE 7. Mutation of the DLG motif abrogates specifically Keap1-dependent degradation of Nrf2. A, plasmids expressing the indicated proteins were co-transfected into COS1 cells with pcDNA3.1/mKeap1 and pCMV␤-gal. After 24 h, duplicate dishes of cells were treated for 2 h with either vehicle (0.1% (v/v) Me 2 SO) or 15 M Sul before whole-cell lysates were prepared and blotted with mouse anti-V5. B, the indicated proteins were co-expressed in COS1 cells. Twenty four h later, a whole-cell lysate (input) and affinity-purified His-tagged protein fraction (IP) were prepared from each dish of cells and blotted with mouse anti-V5. C, plasmids expressing the indicated proteins were co-transfected with pCMV␤-gal into keap1 Ϫ/Ϫ MEFs. The following day, duplicate dishes of cells were treated with vehicle or 40 g/ml CHX for 2 h before whole-cell lysates were prepared and blotted with mouse anti-HA. WB, Western blot. FIGURE 8. Directed evolution of the DLG motif to a high affinity Keap1binding site. A, comparison of the DLG and ETGE motifs, and mutations introduced into the sequence around the DLG motif. B, whole-cell lysates were prepared from duplicate dishes of COS1 cells expressing either mNrf2-V5 or mNrf2 ⌬ETGE -V5 bearing no additional mutations (wt), a single D29E mutation (M1), a double D29E/L30T mutation (M2), a triple D29E/L30T/V32E mutation (M3), or a quadruple I28E/D29E/L30T/V32E mutation (M4). These were mixed with an equal volume of whole-cell lysates from COS1 cells expressing EGFP-mKeap1. A portion of each mixed extract was kept as an input sample. EGFP-mKeap1 was immunoprecipitated from the remaining portions. Both input and immunoprecipitation (IP) fractions were blotted with goat anti-hKeap1 and mouse anti-V5. WB, Western blot. C, plasmids expressing either mNrf2-V5 or mNrf2-V5 bearing a quadruple I28E/D29E/L30T/V32E mutation (mNrf2 M4 -V5) were co-transfected into COS1 cells with pcDNA3.1/mKeap1 and pCMV␤gal. After 24 h, duplicate dishes of cells were treated with either vehicle or 15 M Sul for 2 h before whole-cell lysates were prepared and blotted with mouse anti-V5.
prediction; all three arginine mutations affected the interaction of mNrf2 ⌬ETGE -V5 with mKeap1 to the same extent as they affected that of V5-tagged mNrf2 mutDLG (Fig. 9D).
The inference from the above experiment is that the DLG and ETGE motifs bind to overlapping surfaces on Keap1. To provide independent support for this conclusion, we used synthetic peptides in a competitive binding assay. First, we tested whether an ETGE-containing nonapeptide representing amino acids 76 -84 from mNrf2 (Fig. 10A) competed with the ETGE motif of mNrf2 mutDLG -V5 for mKeap1. This proved to be the case, although the peptide had to be included in the co-immu-noprecipitation assay at a concentration of at least 5 M (Fig.  10B), which greatly exceeds the dissociation constant for the Nrf2-Keap1 interaction of ϳ9 nM (48). Crucially, this peptide also competed with the DLG motif of Nrf2 for the adaptor. For example, when included at 5 M in the co-immunoprecipitation assay, the ETGE-containing peptide efficiently inhibited the interaction of mNrf2 ⌬ETGE -V5 with mKeap1 (Fig. 10C, cf.  lanes 5 and 6 and lanes 7 and 8). These observations are consistent with the ETGE and DLG motifs of Nrf2 competing for the same surface on Keap1.
The ability of the ETGE peptide to compete with mNrf2 ⌬ETGE -V5 for mKeap1 was absolutely dependent upon its embedded DEETGE sequence. When some or all of these residues were replaced, the resulting peptides (DLG-and Skn-1 FIGURE 9. An arginine triad on Keap1 is important for the Nrf2-Keap1 interaction. A, molecular surface of the Kelch-repeat domain from mKeap1 (Protein Data Bank accession code 1X2J) color-coded by electrostatic potential. An electropositive surface is evident, and this attribute is because of the presence of three arginines, Arg-380, Arg-415, and Arg-483, which are shown as space-filled residues in B. The images in A and B were rendered with Deep-View. WB, Western blot. C and D, whole-cell lysates were prepared from duplicate dishes of COS1 cells heterologously expressing mNrf2 mutDLG -V5 (C ) or mNrf2 ⌬ETGE -V5 (D). These were mixed with an equal volume of whole-cell lysate from COS1 cells that were either mock-transfected (Ϫ) or that expressed EGFP-mKeap1 bearing either no mutations (wt), an Arg-380 to Met mutation (R380M), an Arg-415 to Met mutation (R415M), or an Arg-483 to Met mutation (R483M), as indicated. A portion of each mixed extract was kept as an input sample. EGFP-tagged proteins were immunoprecipitated from the remaining portions, and both input and immunoprecipitation (IP) fractions were blotted with goat anti-hKeap1 and mouse anti-V5. The relationship between these peptides is as follows: the ETGE peptide consists of LDEETGEFL; the DLG peptide includes LDIDLGVFL, and the underlined residues represent amino acids 27-32 of mNrf2; the Skn-1 peptide was derived from the DLG peptide by replacing the DLG residues with the sequence AGE. These final three substitutions were chosen based on the sequence alignment presented in Fig. 1A of the DIDLID/DLG region of mNrf2 with Skn-1. B and C, whole-cell lysates were prepared from duplicate dishes of COS1 cells heterologously expressing mNrf2 mutDLG -V5 (B) or mNrf2 ⌬ETGE -V5 (C). These were mixed with an equal volume of whole-cell lysate from COS1 cells that were either mock-transfected (Ϫ) or that expressed EGFP-mKeap1, as indicated. The ETGE nonapeptide or vehicle (Ϫ) was added to the mixed lysates at the indicated final concentrations (B). The ETGE peptide or the related peptides, DLG and Skn-1, was added to the mixed lysates to a final concentration of 5 M (C). A portion of each mixed extract was kept as an input sample. EGFP-tagged proteins were immunoprecipitated from the remaining portions, and both input and immunoprecipitation fractions were blotted with goat anti-hKeap1 and mouse anti-V5. WB, Western blot.
peptides; see Fig. 10A) failed to inhibit the interaction of mNrf2 ⌬ETGE -V5 with adaptor protein even when included in the co-immunoprecipitation assay at concentrations of up to 2.5 ϫ 10 Ϫ4 M (Fig. 10C, lanes 9 and 10 and lanes 11 and 12; and data not shown). Thus, the effects of the ETGE-containing peptide are sequence-specific.
Collectively, these data indicate that the Neh2 domain contains two related Keap1-binding sites. Both are required for maximal destabilization of Nrf2 by BC 3 B Keap1/Keap1 , and both interact with overlapping surfaces on the ␤-propeller structure of Keap1. The comparatively low affinity of the second Keap1binding site correlates with the limited correspondence of its sequence to that of the high affinity ETGE motif.
A Fixed-ends Model for Nrf2 Ubiquitylation by BC 3 B Keap1/Keap1 -The requirement that the lower affinity Keap1-binding site be present in order for Nrf2 to be ubiquitylated by BC 3 B Keap1/Keap1 cannot be explained in terms of simple avidity effects. In its absence, Nrf2 binds Keap1 as avidly as does the wild-type protein, and an increase in its affinity for Keap1 correlates with a weak decrease in turnover of the factor (Fig.  8C). We therefore propose that the significance of the second binding site lies in the fact that it enables Nrf2 to dock simultaneously onto the two separate ␤-propeller structures present in BC 3 B Keap1/Keap1 . This will result in the freedom of movement of the Neh2 domain in Nrf2 upon the multiprotein ligase complex becoming substantially more constrained than if it were bound at only one site. This geometric constraint is necessary for ubiquitylation of the transcription factor and, given that both Keap1-binding sites utilize overlapping surfaces on Keap1, can only be achieved by a Keap1 homodimer.
To illustrate this principle, we have constructed a model for BC 3 B Keap1/Keap1 based on our current understanding of its composition. It is unclear how the BTB and Kelch-repeat domains of mKeap1 are positioned relative to each other, as the structure of the intervening region (IVR; see Fig. 1D) has not been determined. One plausible orientation of these domains was chosen based on the relative positions of the BTB fold (Skp1) and the ␤-propeller structure (WD40 domain of ␤TrCP1) in the crystal structure of Skp1-␤TrCP1 (49). This structure was superimposed on the Skp1-F-box portion of the crystal structure of SCF F-box of Skp2 (12) to generate a model for BC 3 B Keap1/Keap1 (Fig. 11, A and B).
We propose that the ETGE motif binds an arginine triad on one Keap1 polypeptide (Fig. 11C). The resulting localization of the DLG motif in the vicinity of the triad on the second Keap1 subunit compensates for their lower affinity for each other and they too interact (Fig. 11D). The distances between the guanidinium nitrogens of Arg-380, Arg-415, and Arg-483 on both Keap1 subunits are 55, 50, and 55 Å apart, respectively, in our model of BC 3 B Keap1/Keap1 . As the region between the DLG and ETGE motifs is composed primarily of a nine-turn ␣-helix (49 Å), there is no steric impediment to this second interaction. Because the Neh2 domain is inherently flexible (46), this second binding event will substantially constrain the conformational space sampled by any lysine residues lying between them from that which they would sample were only one end or the other fixed in space. Thus, their effective concentration averaged over time in the vicinity of the Rbx1-bound E2 enzyme is increased.
Given the current belief that Cul-Rbx ligases function by increasing the local concentration of reactants, this principle on its own may be sufficient to explain the role played by the DLG motif in ubiquitylation of Nrf2 by BC 3 B Keap1/Keap1 . However, more detailed structural studies will be required to determine whether the second binding event might also result in induced ordering or provoke more subtle geometric alterations in the secondary structure elements of which the Neh2 domain is composed.
The fixed-ends model predicts that the lysine residues in Nrf2 targeted by BC 3 B Keap1/Keap1 will lie between the two Keap1-binding sites. There are 27 lysines in Nrf2, but BC 3 B Keap1/Keap1 only targets seven of them as indicated by the fact that their replacement with arginines is sufficient to completely stabilize Nrf2 (24). All seven lysines implicated in ubiquitylation are located in the ␣-helix that lies between the two Keap1-binding sites. Additionally, six of these residues are positioned on the same side of the helix.
The two arginine triads in Keap1 are not equidistant from Rbx1 as depicted in Fig. 11A, suggesting they might not optimally position the target lysines with respect to the E2 enzyme. Crucially, a model for BC 3 B Keap1/Keap1 in which this condition is satisfied is equally plausible as that shown. This ambiguity in our model reflects the absence of a structure for the IVR of Keap1. Solving the IVR structure remains a critical objective in order to further our understanding of this system.
Self-association Is a Common Feature of BC 3 B Substrate Adaptors-The model outlined above explains why two Keap1binding sites are required for destabilization of Nrf2 by BC 3 B Keap1/Keap1 and accounts for the lysine specificity displayed by the E3 ligase. It also provides a rationale for substrate adaptor self-association that might be generally applicable to other BC 3 B ligases, and perhaps SCF ligases. This assertion is supported by the following two observations. First, three other BTB proteins besides Keap1 have been reported in detail to interact with Cul3, and all appear to associate with themselves. Maternal effect lethal-26, which targets meiosis-1 for ubiquitylation by the BC 3 B holoenzyme in C. elegans, self-associates (50). RhoBTB2 interacts with Cul3 (37), and we have found that it interacts with itself (supplemental Fig. 3A). In our experiments, the HA:Myc ratio, indicative of mixed RhoBTB2 dimers, is much lower in the immunoprecipitate than the corresponding input samples suggesting that oligomerization of this protein may not be complete. As this protein is unique among those under discussion in that it contains two BTB domains per polypeptide, we speculate that intramolecular association of BTB domains competing with intermolecular association of these domains may account for this finding. Finally, nuclear speckled type protein (SPOP) recruits MacroH2A1 for monoubiquitylation by the BCB holoenzyme (38). We find it exists in a complex with itself (supplemental Fig.  3B), and the FLAG:Myc ratios suggest that oligomerization proceeds to completion. As these four BTB proteins (including Keap1) are members of different subclasses of BTB proteins (i.e. they utilize different classes of substrate interaction domains), self-association would appear to be a common attribute of BTB proteins that interact with Cul3.
Second, a common feature among proteins that are ubiquitylated through interactions with substrate adaptors is that the domains involved in binding to the adaptor tend to lack structure (3), as typified by the Neh2 domain of Nrf2. Thus, there is little constraint to the evolution of multiple binding sites composed of short contiguous stretches of amino acids. Consequently, it is improbable that Nrf2 alone has evolved a two-site interaction mechanism to take advantage of the self-association of BC 3 B substrate adaptors. We hypothesize that similar fixed-ends models will be found to apply to many substrates of BC 3 B BTB E3 ligases. The fact that the DLG motif in Nrf2 is weak and nonobvious when compared with the ETGE motif suggests that even in substrates with ostensibly one binding site for a particular BTB protein, additional binding sites may remain to be identified.
Perhaps the most compelling evidence to support a more general application of the fixed-ends model comes not from the BTB family but from the F-box family of substrate adaptors.
With regard to the F-box family of adaptors, the literature only supports the more modest claim that some members of the WD-40 subfamily self-associate, including ␤TrCP, as noted earlier (15). This particular protein recruits substrates bearing destruction motifs (51) to Cul1-Rbx1 holoenzymes. It was recently reported that one of its substrates, cell division cycle 25A, bears two destruction motifs, both of which are required for its ubiquitylation by SCF ␤TrCP (52).
Concluding Comments-This paper presents evidence that the homodimeric structure of the Keap1 substrate adaptor and the presence of two interaction sites in the Neh2 domain are both crucial to tether Nrf2 upon BC 3 B Keap1/Keap1 in order to facilitate substrate ubiquitylation. Furthermore, we have emphasized the fact that similar mechanisms might be utilized more generally to recruit and orientate substrates upon Cul3-Rbx1 holoenzymes. We conclude by highlighting a critical issue relating to Nrf2 upon which our data impinge. Cul3 is in white, and Rbx1 is in green. Arg-380, Arg-415, and Arg-483 are depicted as space-filled residues. C and D, schematics of the fixed-ends model for recruitment of Nrf2 to BC 3 B Keap1/Keap1 . Nrf2 contains an ␣-helix (black filled rectangle) bearing seven lysine residues that are targets for ubiquitylation by BC 3 B Keap1/Keap1 . The ␣-helix is bounded by two regions lacking observable secondary structure, as indicated by zig-zag lines. The DLG and ETGE motifs flank the ␣-helix, and there is some evidence that the ETGE motif lies in a loop of a short mini ␤-sheet (indicated by arrowed lines). The six-bladed ␤-propellers of Keap1 are shown as hexagons, and the arginines involved in binding Nrf2 are indicated within these structures. It is proposed that Nrf2 interacts with one ␤-propeller through its ETGE motif. This single-site interaction does not substantially constrain the target lysines in space because of the inherently flexible nature of the Neh2 domain (C). In order to constrain, and thus present the target lysines for ubiquitylation, the DLG motif must bind to the second ␤-propeller present in dimeric Keap1; this is facilitated by the prior binding of the ETGE motif (D). See text for further details.
The ability of Nrf2 to control the cellular redox state is based on its ability to evade BC 3 B Keap1/Keap1 -dependent ubiquitylation in oxidatively stressed cells. The mechanism of evasion remains controversial. Although it is widely presumed to rely upon a redox-sensitive interaction between Nrf2 and Keap1, a recent paper (48) suggests that the modification of cysteines in Keap1 does not result in release of the transcription factor, and our data support this notion. 5 The fixed-ends model advanced in this paper can accommodate these findings, as it suggests that the avidity of Nrf2 for Keap1 does not have to change substantially in order for the transcription factor to avoid ubiquitylation. Rather, the critical issue is whether or not Nrf2 can dock simultaneously onto both Keap1 subunits. Our data suggest that merely altering the relative positions of the two ␤-propeller domains in the Keap1 homodimer is probably sufficient to inhibit its ability to ubiquitylate Nrf2. With this perspective in mind, we believe it is highly significant that the most reactive cysteines of Keap1 cluster within the IVR of Keap1, which links its BTB-and ␤-propeller structures (17). As the two BTB domains in the Keap1 homodimer are expected to be rigidly configured with respect to each other, any conformational change in the IVR caused by oxidation of the reactive cysteines would alter the relative orientation of the two ␤-propeller structures of Keap1. This could preclude simultaneous binding of Nrf2 to both adaptor subunits, thus allowing Nrf2 to escape ubiquitylation by BC 3 B Keap1/Keap1 in oxidatively stressed cells.