Crystal Structure of KLHL3 in Complex with Cullin3

KLHL3 is a BTB-BACK-Kelch family protein that serves as a substrate adapter in Cullin3 (Cul3) E3 ubiquitin ligase complexes. KLHL3 is highly expressed in distal nephron tubules where it is involved in the regulation of electrolyte homeostasis and blood pressure. Mutations in KLHL3 have been identified in patients with inherited hypertension disorders, and several of the disease-associated mutations are located in the presumed Cul3 binding region. Here, we report the crystal structure of a complex between the KLHL3 BTB-BACK domain dimer and two copies of an N terminal fragment of Cul3. We use isothermal titration calorimetry to directly demonstrate that several of the disease mutations in the KLHL3 BTB-BACK domains disrupt the association with Cul3. Both the BTB and BACK domains contribute to the Cul3 interaction surface, and an extended model of the dimeric CRL3 complex places the two E2 binding sites in a suprafacial arrangement with respect to the presumed substrate-binding sites.


Introduction
Targeted ubiquitination can direct substrate proteins to a variety of functional fates, including proteosomal degradation, the modulation of protein interaction networks, and altered subcellular localizations. The largest class of E3 ligases are the Cullin-RING Ligases (CRLs), which are further identified according to the type of cullin chain that constitutes the central scaffolding unit [1,2]. For example, the CRL3 complex is built around a Cullin3 (Cul3) component. While many CRL complexes interact with bipartite substrate adaptor proteins, such as the Skp1/F-Box adaptor protein complexes in CRL1, CRL3s differ by forming a complex with single-chain substrate adapters that bind directly to both the cullin and substrate [3][4][5][6].
Many of the known CRL3 substrate adapters belong to the BTB-BACK-Kelch family of proteins. These proteins are made almost entirely of three concatenated structural domains: the BTB and BACK domains form the platform that engages the Nterminal region of Cul3, while the Kelch repeat domain forms a bpropeller structure for substrate binding. In humans, most of the 52 BTB-BACK-Kelch proteins can be classified into two named groups, namely the 39 KLHL proteins and the 11 KBTBD proteins [7][8][9]. Notable proteins from this family include Keap1 (KLHL19), an electrophile-sensing regulator of Nrf2 [10][11][12], KLHL9, which is associated with an autosomal distal myopathy [13], KLHL12, a regulator of the dopamine D4 receptor [14] Dishevelled [15,16] and CPOII coat function [17], KLHL20, a regulator of hypoxia-inducible factors [18,19], and KLHL3, a regulator of hypertension with mutations identified in pseudohypoaldosteronism type II (PHAII) [20][21][22].
PHAII, also known as familial hyperkalemic hypertension, is a rare autosomal dominant disease characterized in part by elevated electrolyte and reduced bicarbonate levels in the blood.
The characterization of the molecular defect in this disease has provided key insights into the mechanisms of blood pressure regulation [23][24][25][26][27]. Exome sequencing of affected individuals has established that PHAII can be caused by mutations in either KLHL3 or Cul3, and these mutations are correlated with an increased activity of the NaCl cotransporter (NCC) in the distal convoluted tube (DCT) [21,22]. Because E3 ligases can regulate the endocytosis of integral membrane proteins [28][29][30][31][32][33], KLHL3 may regulate electrolyte homeostasis by regulating NCC trafficking via CRL3 KLHL3 -dependent ubiquitination.
Here, we report the crystal structure of a KLHL3 BTB-BACK / Cul3 NTD complex and characterize the Cul3 binding properties of a series of KLHL3 PHAII mutations. We previously reported the structure of the SPOP BTB /Cul3 NTD complex [9]. In addition, a structure of KLHL11 BTB-BACK in complex with a Cul3 Nterminal domain has been recently reported [34]. An analysis of the three available BTB/Cul3 complexes provides insight into how Cul3 is able to bind to a large number of different adaptor proteins. The structure of the KLHL3 BTB-BACK /Cul3 NTD complex allows an expanded comparison of the BTB-BACK/Cul3 binding interface and allows a more accurate modeling of other intensely studied BTB-BACK-Kelch proteins such as Keap1.

Cloning, Protein Expression and Purification
An expression construct for human KLHL3 comprising residues 27-276 (KLHL3 BTB-BACK ) was designed using the web-based Crystallization Construct Designer [35] and cloned into a pMCSG7 vector via ligation independent cloning [36], producing a protein with an N-terminal 6His tag. The version of KLHL3 BTB-BACK used in crystallization was further modified by surface entropy reduction (SER). The Surface Entropy Reduction prediction (SERp) web server [37] identified K87, K89 and K90 as three nonconserved residues predicted to be exposed at the protein surface. These three lysines were mutated to alanine residues by PCR mediated site directed mutagenesis for the protein used in crystallization. The KLHL3 BTB-BACK protein used in the solution-based experiments did not include the SER mutations, and the PHAII mutations were introduced into the wild-type KLHL3 BTB-BACK expression plasmid by PCR mediated site directed mutagenesis. The N-terminal domain of Cul3 comprising residues 20-381 incorporating the stabilizing mutations I342R/L346D (Cul3 NTD ) was cloned into a pET32a vector as described previously [9].
KLHL3 BTB-BACK and Cul3 NTD were expressed separately in E. coli BL21 DE3 Codon+ cells. Cultures were grown at 37uC to an OD600 of 0.8. The temperature was then reduced to 15uC and the cultures were induced with 1 mM IPTG and grown overnight. Cells were harvested, lysed and the His-tagged proteins were purified by metal ion chelate chromatography on NiNTA resin. The N-terminal thioredoxin-His tag on the Cul3 NTD protein was removed with TEV protease. The final purification step for both proteins was size exclusion chromatography on a Superdex S75 column in 20 mM Tris pH 7.5, 150 mM NaCl, 1 mM TCEP, and 10% v/v glycerol (buffer A).

Crystallization, Data Collection, Structure Solution and Refinement
Crystals of the KLHL3 BTB-BACK /Cul3 NTD complex were grown by hanging drop vapor diffusion at room temperature. KLHL3 BTB-BACK and Cul3 NTD were mixed in a 1:1 molar ratio at a total protein concentration of 10 mg/ml. Hanging drops were set by mixing 1 ml of the protein solution with 1 ml of a reservoir solution containing 0.1 M sodium tartrate and 17% w/v PEG 3350, and incubating against 1 mL of reservoir solution. Crystals were soaked in a reservoir solution containing 15% ethylene glycol for 30 s prior to flash freezing. Diffraction data were collected at 100 K on beamline 19-ID at the Advanced Photon Source at a wavelength of 0.98 Å using ADSC Quantum 315r detector. Data were processed with HKL3000 [38,39]. The KLHL3 BTB-BACK /Cul3 NTD structure was solved by molecular replacement with PHENIX AutoMR [40] using a search model based on a KLHL11/Cul3 complex (PDB ID 4AP2) [34]. The top solution placed one chain of KLHL3 BTB-BACK and one chain of Cul3 NTD in the asymmetric unit, with a solvent content of 66%. The solution generated a KLHL3 BTB-BTB homodimer via a crystallographic 2-fold symmetry operator, even though no information about this expected dimer was included in the molecular replacement procedure. As a further test of the solution, an independent molecular replacement search was carried out using the same search model, except that residues corresponding to helices H1 and H2 of Cul3 were deleted. Following rigid body refinement of this partial model, difference density maps clearly showed the density for the deleted residues ( Figure S1A). Similarly, a search model based on the SPOP BTB /Cul3 NTD (PDB ID 4EOZ) [9] gave an equivalent solution. In this case, positive difference density was observed for KLHL3 helices in the BACK domain, despite the fact that residues in this region were not present in the search model.
Refinement and model building were performed using PHE-NIX [40] and Coot [41]. Ramachandran dihedral restraints were used in the final stages of refinement. The final model consisted of residues 32-223 of KLHL3 BTB-BACK and residues 26-379 of Cul3 NTD . The quality of the final model was verified with a composite omit map calculated with PHENIX [40] ( Figure S1B). Structural superpositions and renderings were performed in PyMol [42]. KLHL3 BTB-BACK /Cul3 NTD interface residues and buried surface area were determined using the EMBL PISA web server [43]. Transformation functions for BTB dimer and BTB-Cul3 interfaces were calculated using EMBL Dali Lite web server [44,45]. Atomic coordinates and structure factors have been deposited in the Protein Data Bank under accession code 4HXI.

Size Exclusion Chromatography
Protein samples were injected onto a Superdex S75 size exclusion column equilibrated with buffer A and elution was monitored at 280 nm.

Isothermal Titration Calorimetry
Isothermal titration calorimetry (ITC) binding experiments were performed using a VP-ITC Micro Calorimeter at 25uC. All proteins were dialyzed in buffer A for three days prior to analysis. Aliquots of KLHL3 BTB-BACK at 150 mM were injected into Cul3 NTD solutions at 15 mM. Data were processed using Origin, and binding isotherms were calculated based on a one-site binding model. A single titration was conducted for wild type KLHL3 BTB-BACK and each mutant. K d error values are based on the sum of square deviations between the non-linear regression curve and the experimental data.

Homology Modeling
A model of the complete SCF3 KLHL3 ubiquitin ligase complex was generated following the approach used in making the SCF3 SPOP model [9]. The complex is based in part on the known structures of Cul1-Rbx1-Skp1-Skp2 [4]. The Cul1 chain was replaced by a full length Cul3 model in which the C-terminal domain was based on Cul1. The E2 Ubch7 was positioned onto Rbx1 by superposing the RING domains from Rbx1 and c-Cbl from the c-Cbl-Ubch7 complex [46]. Ubiquitin was positioned onto Ubch7 by superposing E2-24 from the E2-24-ubiquitin complex on Ubch7 [47]. I-Tasser [48] was used to generate Keap1 BTB-BACK and KLHL9 BTB-BACK homology models, and these were superimposed onto KLHL3 BTB-BACK in the Cul3 complex structure. No insertions or deletions are present near Keap1 C151 and KLHL9 L95 relative to KLHL3, and the backbones of the three BTB-BACK domains were in excellent agreement in the areas near the mutation sites.

Results and Discussion
Crystal Structure of the KLHL3/Cul3 Complex The structure of human KLHL3 BTB-BACK /Cul3 NTD complex reveals a 2:2 complex, with two chains of Cul3 NTD bound independently to two equivalent and non-overlapping surfaces of a KLHL3 BTB-BACK homodimer ( Figure 1, Table 1). The Cul3 NTD chains are positioned to form a cup enclosing the space where the substrate-binding Kelch domains of KLHL3 are predicted to be located. These findings are consistent with suprafacial arrangements [49] observed in two other dimeric Cul3 complexes: SPOP BTB /Cul3 NTD [9] and KLHL11 BTB-BACK/ Cul3 NTD [34]. In solution, KLHL3 BTB-BACK elutes as a dimer by size exclusion chromatography, Cul3 NTD elutes as a monomer, and an equimolar mixture of the two proteins elutes as a single peak with the expected size for a 2:2 complex ( Figure S2).
The structure of the BTB domain of KLHL3 resembles previously solved long-form BTB domains [50], and consists of a three-stranded b-sheet flanked by seven a-helices. The first two helices form the majority of the dimerization interface [50,51]. The strand-exchanged interchain b-sheet involving an N-terminal ''b10 strand has been observed in many long-form BTB domain dimers [50], but is not present in the KLHL3 dimer. This element is not universally present in every BTB dimer [51]. We designate the first beta strand in the KLHL3 BTB domain as b2 in order to remain consistent with the naming convention used in other BTB structures [50][51][52][53]. A short turn of a-helix is present in the loop between a3 and b4, and we designate this helix as a3.1, since this structural element is not observed in unliganded BTB structures (see below). The BACK domain of KLHL3 is similar to the BACK domains from KLHL11 [34] and Gigaxonin [54], with no evidence for the BACK-mediated higher order structures as seen with SPOP BTB-BACK in solution [9]. The KLHL3 BACK domain spans residues 150-276 and is expected to consist four sets of helical hairpins, however only the first 5 helices (two and one half hairpins) could be reliably modeled into the electron density maps. The first BACK hairpin, residues 150-176, corresponds to the 3box region [9,54]. Cul3 NTD is made up of three, 5-helix cullin repeats similar to those previously reported in Cul1 [4], Cul4A [5], Cul4B [55], SPOP/Cul3 [9], and KLHL11/Cul3 [34].

KLHL3/Cul3 Interaction Interface
The Cul3 binding interface on KLHL3 consists mostly of surfaces in the BTB domain, with important contributions from the BACK domain. Four distinct regions of KLHL3 make up the Cul3 binding region: i) helices a3 and a3.1 in the a3/b4 loop, ii) b4 and residues in the b4/a4 loop, iii) a5 and the a5/a6 loop, and iv) a7 and the a7/a8 loop (Figures 2, 3). The first three regions are in the BTB domain and the fourth is in the 3-box element of the BACK domain. With the exception of region ii, all of the KLHL3 binding elements correspond the C-terminal end of an a-helix followed by several of the following loop residues (Figure 2). Overall, the KLHL3 BTB-BACK /Cul3 NTD binding interface consists of 25 residues from KLHL3 and 26 residues from Cul3, and collectively bury 1066 Å 2 of surface area. Approximately 80% of the KLHL3 binding surface can be attributed the BTB domain region, with the remainder coming from the BACK domain. This is consistent with findings from the SPOP BTB /Cul3 NTD crystal structure, however in that case, the contributions from the 3-box/ BACK region were deduced from a predicted model [9]. Solution studies showed that the SPOP BACK domain was required for full binding to Cul3: the SPOP BTB /Cul3 NTD dissociation constant was 1.0 mM, but the SPOP BTB-BACK /Cul3 NTD dissociation constant was 13 nM as measured by ITC [9]. A multiple sequence alignment of BTB-BACK-Kelch proteins reveals that the a3/a3.1 region is relatively well conserved, while other Cul3 binding regions have lower sequence similarity ( Figure 2). As discussed in a later section, PHAII mutations at positions A77, M78, and E85 are found in this location [22], foreshadowing the impact of these mutations on Cul3 binding. The w-x-E motif that was first identified in the SPOP BTB / Cul3 NTD structure [9] is preserved in the KLHL3, where the large hydrophobic residue designated by w is M83, the charged/polar ''x'' residue is S84, and the conserved glutamate is E85. The w residue buries the most surface area of any amino acid in the three known BTB/Cul3 structures (Figure 2), and is nested in a deep pocket in Cul3 that is formed between the H1-H2 loop and helix H5 (Figure 3) [9]. Residue E85 is located in the short a3.1 helix that is disordered in several uncomplexed BTB-BACK structures including KLHL11 [34] and Gigaxonin [54]. Cul3 binding induces an ordering of this region between a3 and b4, including the formation of a3.1, in all three BTB/Cul3 complexes. The added buried surface area in KLHL11 in the regions preceding a4, a5 and a7 (Figure 2) is due to contacts with additional Nterminal residues in the Cul3 construct used in reference 34, which were not present in our construct. While this region of Cul3 affects the affinity of complex in KLHL11, it does not have an effect on the overall geometry of the Cul3 assembly, as shown by the similarity of the PDB structures 4AP2 and 4APF [34]. The overall shape of the Cul3-interacting surface formed by the BTB and BACK regions is fairly consistent, however, the characteristics of these surfaces are remarkably varied (Figure 4). For example, the electrostatic potential in the Cul3 binding region of KLHL3, KLHL11 and SPOP does not reveal a consistent pattern apart from the electronegative region near the conserved w-x-E motif, reflecting the lower sequence identity in regions ii, iii and iv. The Cul3 helix H2 makes several important contributions to the interaction, and the hydrophobic Cul3 residues F54, Y58, Y62 and L66 lie approximately along one side of recognition helix H2 and forms contacts with regions i, iii and iv from KLHL3 (Figures 3 and S1) [9].

Overall Architecture of Dimeric CRL3 Complexes
Overall, the individual subunits in the three available BTB/ Cul3 complexes (KLHL3 BTB-BACK /Cul3 NTD (this work), SPOP BTB /Cul3 NTD [9] and KLHL11 BTB-BACK /Cul3 NTD (PDB ID 4AP2)) are similar. The three Cul3 chains can be superposed with an average Ca RMSD of 0.6 Å , and the three BTB domains can be superposed with an average Ca RMSD of 1.2 Å The BACK domain of KLHL3 and KLHL11 are less similar, and superpose with a Ca RMSD of 2.2 Å .
There are larger differences at the quaternary level, and we observe small but significant differences at the interchain interfaces. Because of the similarities of the structures at the single chain level, we measured these differences as rigid body motions between the subunits. First, at the level of the BTB/BTB interfaces, there is a relative rotation of 4u between the BTB dimer interfaces of KLHL3 and KLHL11, and a much larger 14u rotation between the dimers from SPOP and KLHL3 ( Figure 5A). Similar changes have been observed in the BTB interfaces of domains from BTB-Zinc finger transcription factors [51]. At the BTB/Cul3 level, a superposition of the three BTB domains reveals rotations of the Cul3 subunits by 3.1u and 11.5u in the KLHL11 and SPOP complexes, respectively, relative to the KLHL3 complex ( Figure 5B).
The cumulative result of these interface differences may result in larger changes in the position of the ubiquitin-linked E2 in intact CRL3 complexes ( Figure 5C). The structures of Cul1/Skp1 [4], CBL-UBCH7 [46] and UbcH5b-ubiquitin [56] were used to model the C terminus of Cul3, Skp1 and an E2/Ubiquitin complex. Assuming a rigid association from the Cul3 N-terminal region to the E2-ubiquitin region, the alterations in the BTB/BTB and BTB/Cul3 interfaces may result in a wider opening between the E2 regions in SPOP relative to KLHL3 or KLHL11. The relevance of this modeling study on the effects of substrate ubiquitination are is not clear, however, since activation of CRLs by NEDD8 results in a open, dynamic state for the complexes with conformational variability of the cullin C-terminal domain/Rbx1tethered E2 relative to the rest of the complex [57,58]. This allows the E2/ubiquitin moiety to sample a larger region of space, possibly erasing any differences that we deduce for the rigid complexes.

PHAII Mutations and Cul3 Binding
Mutations in KLHL3 have been found in patients with PHAII [21,22], and while a majority of these are found in the Kelch domain of the protein, mutations in the BTB and BACK domain have also been identified [22]. We used ITC to measure the affinity between KLHL3 BTB-BACK and Cul3 NTD and tested all of the identified missense mutations in the KLHL3 BTB and BACK domains from the study by Boyden et al. [22] (Figure 6). We measured a K d of 10868 nM and a stoichiometry of 1:1 for the association between wild-type KLHL3 BTB-BACK and Cul3 NTD . PHAII mutations A77E and M78V map to KLHL3 helix a3 and mutation E85A localizes to the w-x-E motif in helix a3.1 (Figure 2). In the wild-type protein, all three of these positions are in direct contact with Cul3 (Figure 2 and 6A). All three of these substitutions severely disrupted Cul3 binding and K d values could not be determined for these proteins ( Figure 6C). The C164F mutation was the least disruptive and produced a K d of 4.160.4 mM, a 37 fold decrease in affinity. C164 is near the interaction interface, but is not in direct contact with Cul3. This residue is located in the a7/a8 helical hairpin of the 3-box/BACK domain and is flanked by several residues that interact directly with Cul3 (Figure 2 and 6B). A mutation of this cysteine to a bulky phenylalanine may weaken the association by altering the conformation of the neighboring residues.
All of the PHAII point mutants could be produced as stable, well-folded proteins with biochemical properties that were indistinguishable from those of wild-type KLHL3 BTB-BACK . In vitro, we found that the KLHL3 BTB-BACK point mutations were properly folded and dimeric by size exclusion chromatography, but had reduced affinity for Cul3. In homozygotes, these mutant proteins would preserve the bivalent binding of substrate through the two C-terminal Kelch motifs, but with reduced or no affinity for Cul3/ Rbx1/E2,Ub. In contrast, the expected protein produced by the Q144STOP mutation [22] would generate a truncated protein with an intact BTB domain but with no BACK or Kelch motifs (Figure 2 and 6B). This KLHL3 BTB chain would be competent for BTB-dimerization, but would lack the domains required for Cul3 and substrate binding. In heterozygotes, we propose that BTBdriven heterodimers could form between the missense or truncated copy of the protein and the wild-type copy of the protein.
Overall, we conclude that the KLHL3 BTB-BACK domain mutations found in PHAII patients disrupt Cul3 binding and likely reduces or abrogates the ubiquitination of substrate protein(s) bound to the Kelch domain of KLHL3. In PHAII, the relevant KLHL3 substrate appears to be the NaCl cotransporter (NCC) [21][22][23]. Failure to regulate NCC levels and subcellular localization via a functional CRL3 KLHL3 complex may result in an overabundance and increased activity of NCC, disrupting electrolyte homeostasis and contributing to the hypertensive phenotype.

Key residues in other BTB-BACK-Kelch Proteins
We modeled the Cul3 complex of Keap1, a redox stress sensing BTB-BACK-Kelch protein, based on our KLHL3 BTB-BACK /Cul3 NTD structure. Keap1 and KLHL3 are near-neighbors in sequence space and share 33% sequence identity. This makes KLHL3 a preferred template for modeling relative to KLHL11 or Gigaxonin, which only share 19% and 21% identity to Keap1, respectively. Keap1 residue C151 (equivalent to residue N119 in KLHL3, Figure 2A) has been shown to be covalently modified by electrophiles, resulting in the disruption of Cul3 binding [11,[59][60][61]. Our model places Keap1 C151 in  a loop preceding BTB helix a5, in the vicinity of Cul3, but not in direct contact ( Figure S3A). By similarity to KLHL3, we predict that C151 in a well-ordered region of Keap1 and is solvent accessible. Thus, it is reasonable to assume that electrophile adduction to C151 would affect the position of nearby residues, producing a conformational change that could be transmitted to the a5 and a5/a6 loop region which is in direct contact with Cul3 ( Figure S3B).
In the case of KLHL9, a L95F mutation is associated with an autosomal dominant distal myopathy [13]. This position is equivalent to KLHL3 residue I93, a residue that is partly buried in the BTB domain and also partly exposed at the Cul3 binding surface. Thus, it is thus very likely that substitutions at this position in KLHL9 would perturb the Cul3 binding interaction and affect substrate ubiquitination.

Conclusions
The crystal structure of the BTB-BACK domains of KLHL3 in complex with an N terminal domain of Cul3 reveals the basis for the association between these two proteins. The BTB dimer generates a 2:2 complex in which two Cul3 chains bind independently to the BTB-BACK regions of each KLHL3 subunit. Several hypertension disease mutations in KLHL3 map to the Cul3-binding region and disrupt complex formation. These results provide a molecular basis for understanding the defects in diseases involving CRL3 complexes. Figure S1 Electron density maps. (A) As a test for the molecular replacement solution, an |Fo-Fc| map was calculated from a molecular replacement model which did not include Cul3 helix H2 and contoured at 2s. (B) An |Fo-Fc| omit map was calculated based on the final refined structure, and shows good agreement between the model and the density.