Crystal structure and interaction studies of the human FBxo3 ApaG domain

Abstract

Transcriptional activation of proinflammatory cytokines, mediated by tumor necrosis factor receptor-associated factors (TRAFs), is in part triggered by the degradation of the F-box protein, FBxl2, via an E3 ligase that contains another F-box protein, FBxo3. The ApaG domain of FBxo3 is required for the interaction with and degradation of FBxl2 [Mallampalli RK et al., (2013) J Immunol 191, 5247–5255]. Here, we report the X-ray structure of the human FBxo3 ApaG domain, residues 278–407, at 2.0 Å resolution. Like bacterial ApaG proteins, this domain is characterized by a classic Immunoglobin/Fibronectin III-type fold, comprising a seven-stranded β-sheet core, surrounded by four extended loops. Although cation binding had been proposed for bacterial ApaG proteins, no interactions with Mg²⁺ or Co²⁺ were detected for the human ApaG domain. In addition, dinucleotide polyphosphates, which have been reported to be second messengers in the inflammation response and targets of the bacterial apaG-containing operon, are not bound by the human ApaG domain. In the context of the full-length protein, loop 1, comprising residues 294–303, is critical for the interaction with FBxl2. However, titration of the individual ApaG domain with a 15-mer FBxl2 peptide that was phosphorylated on the crucial T404, as well as the inability of the ApaG domain to interact with full-length FBxl2, assessed by coimmunoprecipitation, indicate that the ApaG domain alone is necessary, but not sufficient for binding and degradation of FBxl2.

Introduction

In response to cellular injury or contact with a pathogen, the human body’s normal immune response is to trigger a burst of cytokine release from proinflammatory macrophages, lymphocytes, and polymorphonuclear leukocytes [1,2]. As part of the associated molecular signal transduction pathway, the tumor necrosis factor receptor (TNFR)-associated factor (TRAF) proteins act as molecular messengers and transduce signals from TNFR receptors through the NF-_ΚB pathway to the nucleus, resulting in the induction of cytokine gene transcription [3]. Under normal conditions, an ubiquitin E3 ligase of the Skp1-Cullin1-F-box (SCF) family, containing F-box leucine rich 2 (FBxl2) as the F-box component, targets TRAFs 1–6 for proteosome-mediated degradation, curbing cytokine levels [4]. However, under conditions where an inflammatory response from the immune system is necessary, FBxl2 becomes phosphorylated on T404, and a second ubiquitin E3 ligase, with FBxo3 as the F-box component, targets FBxl2 for degradation, thereby promoting TRAF signaling and cytokine gene transcription [4]. In many instances, this excessive release of cytokines, termed ‘cytokine storm,’ leads to hypercytokinemia, which can result in capillary leakage, tissue edema, organ failure, and fatality [5,6].

Currently, very little is known about FBxo3. While most human F-box proteins contain either a leucine-rich repeat or a WD40 domain at their C terminus, FBxo3 contains neither, resulting in its ‘F-box only’ designation [7]. Based on sequence similarities with the bacterial proteins, FBxo3 contains two putative protein–protein interaction motifs, located C-terminal to its F-box. One domain possesses sequence similarity to the SUKH (Syd, US22, Knr4 Homology) protein superfamily, a diverse and poorly characterized collection of proteins present in all organisms, which may be involved in immunity to various toxins [8]. A second domain bears similarity to the bacterial ApaG protein [7]. ApaG proteins are poorly biochemically characterized, although both a solution structure (PDB ID 2F1E) [9] and a crystal structure (PDB ID 1XVS) from Xanthomonas axonopodis and Vibrio cholerae, respectively, have been determined. In Escherichia coli, apaG belongs to an operon important for the metabolism of dinucleotide polyphosphates [10,11], but its role in this operon does not appear to involve nucleotide binding [9]. Mutation in the Salmonella typhimurium ApaG homolog, CorD, has been associated with Co²⁺ resistance and a reduction in Mg²⁺ efflux, suggesting a potential role in metal binding [12]. Only one other human protein, PDIP38, also known as POLDIP2 [13], contains a region similar to the bacterial ApaG proteins. This renders the ApaG domain a rare motif in the human genome. However, it appears to be critical for correctly activating and tempering the human immune response. Immunological in vitro assays that used truncated FBxo3 constructs, lacking the ApaG domain, revealed abrogated FBxl2 binding and degradation [14]. Furthermore, a suite of small molecule benzathine derivatives designed against a modeled ApaG domain from FBxo3 have proven effective at preventing FBxo3/FBxl2 binding and degradation, as well as curbing inflammation in animal models [14].

In order to elucidate the basis of how the ApaG domain of human FBxo3 plays such a critical role in the immune response, we determined the X-ray structure to 2.0 Å resolution, as well as tested several hypotheses regarding potential binding partners for the domain. Most prominently, we probed the FBxl2/FBxo3 interaction, implicated by prior studies [4,14], by immunoprecipitation experiments and ¹H–¹⁵N HSQC NMR spectroscopy, using a FBxl2 peptide phosphorylated on T404. Convincing evidence for this interaction was not observed using the FBxo3 ApaG domain alone. In contrast, altering the sequence of loop 1 in the native structure of FBxo3 severely impaired FBxl2 binding. Replacement of each loop in the ApaG domain of FBxo3 by the corresponding E. coli sequence demonstrated that residues 294–303, constituting loop 1, are critical for FBxl2 binding. Therefore, the currently available data suggest that the ApaG domain of FBxo3 is necessary for the interaction with FBxl2, but not sufficient.

Results

Crystal structure of FBxo3 ApaG domain

The amino acid sequence of FBxo3 was aligned with those of the E. coli, V. cholera, and X. axonopodis ApaG proteins (Fig. 1) using the clustalx2 program to identify the domain boundaries of the corresponding human FBxo3 ApaG domain. The protein construct chosen for structure determination, comprising residues 278–407, is 38% identical to the E. coli sequence, and highly expressed in E. coli. The purified protein displayed a propensity for intermolecular interaction. Indeed, recording a series of 1D ¹H NMR spectra at increasing concentrations failed to show an increase in resonance intensities for increased protein concentration, indicating that higher molecular mass species were formed whose signals are broadened beyond detection (Fig. 2). This phenomenon was reversible at pH > 7.5; at lower pH, however, precipitation occurred. Exploiting the aggregation propensity of this domain, crystal screening was undertaken and diffracting crystals (2.8 Å) formed overnight. Optimization of the crystallization conditions resulted in crystals that diffracted to 1.8 Å. The structure of the FBxo3 ApaG domain was solved by molecular replacement to 2.0 Å, using the X-ray structure of the V. cholera ApaG protein (PDB ID 1XVS) as the search model. Like the solution structure of the X. axonopodis ApaG protein [9] and the X-ray structure of the V. cholera ApaG protein, the human FBxo3 domain exhibits an IgG/Fibronectin III-type fold comprising seven β-strands in the core. Structural superposition of the human FBxo3 ApaG domain structure onto the two bacterial structures highlights the overall structural similarity with backbone RMSD values of 1.1 Å and 1.06 Å for the X. axonopodis and V. cholera structures, respectively.

Fig. 1.

Alignment of the amino acid sequences of prokaryotic ApaG proteins with FBxo3. The E. coli, V. cholera, and X. axonopodis ApaG proteins were aligned with the human FBxo3 protein using clustalx2. Identical and conserved residues are shown in black and gray boxes, respectively. The secondary structure elements in the X-ray structure of the FBxo3 ApaG domain are depicted above the sequences. (β, beta strand; L, loop).

Fig. 2.

Oligomerization analysis of the FBxo3 ApaG domain—(A, B) 1D ¹H NMR spectra of FBxo3 278–407 (A) and 263–407 (B) illustrate the effect of intermolecular interaction as a function of protein concentration. (C, D) Size-exclusion chromatography/multi-angle light scattering data (SEC-MALS) are displayed with the elution profile shown as continuous colored traces with the corresponding estimated molecular masses across the peaks. (C) FBxo3 263–407 (100 μl) was injected into a system equilibrated in 20 mm HEPES pH 7.0, 100 mm NaCl, and 0.5 mm TCEP. Panel (D) summarizes the experimental masses from SEC-MALS experiments as a function of injection concentration. The theoretical mass of FBxo3 263–407 is 17 109.74 Da. For all experiments, the protein concentration is represented by 50 μm (black), 100 μm (blue), 200 μm (green), or 400 μm (red).

Despite their similarities, the three ApaG structures are not identical (Fig. 3A–F). In both bacterial structures, the N terminus is in a random coil conformation, while in the human structure, an N-terminal β-strand, β-1, is present. This β-strand appeared incomplete, stopping in the middle of the sheet, suggesting that the N-terminal boundary selected by sequence comparison alone resulted in a truncated start of the domain. This may be related to the fact that the human FBxo3 ApaG domain is a domain residing in a larger, multidomain protein and not an independently transcribed protein like the prokaryotic counterparts (Fig. 3A,D). Given the common Immunoglobin/Fibronectin III-type fold, the structured β-sheet core of all the ApaG domains is essentially identical, and only the most C-terminal β-strand, β7, possesses a disrupting P397 in the FBxo3 ApaG domain, while both prokaryotic structures contain continuous β7 strands (Fig. 3C,F). As Fibronectin III-type proteins engage in intermolecular interactions using flexible loops [15–17], the four long, unstructured loops of human FBxo3 ApaG sets it apart from the other known ApaG structures. Loops 2 and 4 contain insertions of 4 and 2 residues, respectively, and loop 1 exhibits α-helical character in the prokaryotic structures but is unstructured in the human domain (Fig. 3B,E).

Fig. 3.

Comparison of the FBxo3 ApaG crystal structure with prokaryotic ApaG structures. The crystal structure of FBxo3 278–407 was superimposed on the V. cholera ApaG crystal structure (A–C) or the solution structure of ApaG from X. axonopodis (D–F). For all superimpositions, FBxo3 278–407 is in orange and V. cholera and X. axonopodis structures are in green and blue, respectively. The orientation depicted in (B) and (E) of the FBxo3 ApaG X-ray structure is shown in space-filling representations in (G) and (H). To highlight the cleft that was targeted by Mallampalli et al. [14], in the design of compounds exhibiting anti-inflammatory effects, the bracketing secondary structure elements of β4–Loop 3–β5 have been colored purple (H, I) from light to dark with respect to the structural order from N terminus to C terminus. W331 that is located at the base of the pocket is in green. (I) The FBxo3 ApaG domain has been rotated 50° about the y-axis to give a head-on view of the cleft.

The central β-sheet core of the ApaG domain has been the target of drug design, aimed at interfering with cytokine-driven inflammation. Derivatives of benzathine with anti-inflammatory properties were originally identified by in silico screening against a homology model of the FBxo3 ApaG domain, derived from the X. axonopodis solution structure [14]. In that model, a binding pocket had been identified and ligand screening was performed against this pocket. Gratifyingly, the pocket is also present in the experimentally determined X-ray structure of the FBxo3 ApaG domain (Fig. 3G–I). The bottom of the pocket is lined by the aromatic ring of the W331 side chain, and the walls are created by β4, the first half of loop 3, and the latter half of β5, with overall dimensions of ~ 19.2 Å long, 4.6 Å wide, and 6 Å deep. Many of the residues that form this pocket are conserved between prokaryotic ApaG proteins and the human domain (Fig. 1), and as such, this pocket is common to all known ApaG structures.

Interaction of the FBxo3 ApaG domain with divalent cations

As previously alluded to, the crystal structure revealed that the protein construct beginning at V278 had an incomplete first β-strand leading into the domain. Through systematic extension of the N terminus, a construct, comprising amino acids 263–407, was identified that exhibited enhanced stability (> 2 weeks at 25 °C, at pH 6–8) and solubility (≥ 1 mm). Unfortunately, the improved construct did not eliminate the aggregation propensity of the FBxo3 ApaG domain (Fig. 2). Size-exclusion chromatography with in-line multi-angle light scattering revealed a gradual concentration-dependent increase in the observed averaged molecular mass, which was more pronounced at lower pH (Fig. 2C,D). The reversible aggregation for concentrations > 50 μm precluded backbone assignment, but the enhanced stability permitted the use of ¹H–¹⁵N HSQC spectra at low protein concentrations for screening interactions between the FBxo3 ApaG domain and potential ligands.

While little is known about the activities associated with bacterial ApaG proteins, it has been suggested that they may regulate the amounts of divalent cations inside the bacterium, particularly Mg²⁺ and Co²⁺ [12]. To determine whether the human FBxo3 ApaG domain is sensitive to the presence of Mg²⁺ or Co²⁺, titrations with both cations were performed and followed by ¹H–¹⁵N HSQC spectroscopy (Fig. 4). No changes in the ¹H–¹⁵N HSQC spectrum of 100 μm FBxo3 ApaG were observed for Mg²⁺ additions in excess of 1 mm, suggesting that no specific binding site exists on the protein (Fig. 4A). Even if Mg²⁺ binding would involve one of the mobile loop regions, whose resonances are broadened, this would alter the exchange regime and cause sharpening of the resonances. Addition of Co²⁺ to the FBxo3 ApaG domain sample resulted in the precipitation of protein, beginning around 25 μm CoCl₂. The remaining soluble protein did not exhibit any backbone amide resonance changes, nor did any new resonances emerge (Fig. 4B). This demonstrated that no specific binding pocket for either cation exists on the protein. Thus, while prokaryotic ApaG proteins may bind divalent cations, the ¹H–¹⁵N HSQC titrations clearly show that this functionality is not retained by the human FBxo3 ApaG domain.

Fig. 4.

Metal ion and diadenine polyphosphate binding to FBxo3 263–407. Superpositions of the ¹H–¹⁵N HSQC spectra of 75 μm FBxo3 263–407 in the absence (black) and presence of (A) 1.4 mm MgSO4 (red), (B) 100 μm CoCl2 (red), (C) 1 mm diadenine triphosphate (blue), (D) 1 mm diadenine tetraphosphate (green), and (E) 1 mm diadenine pentaphosphate (red) are shown.

Interaction of the FBxo3 ApaG domain with diadenine polyphosphates

Earlier investigation into the role of prokaryotic ApaG proteins identified a GXGXXG motif [9], which has been reported to be involved with nucleotide binding [18,19]. It had been hypothesized that as the apaG-containing operon was responsible for dinucleotide polyphosphate metabolism, this motif would allow ApaG to bind to nucleotides. This hypothesis could not be substantiated experimentally for ATP or GTP [9]. However, diadenine polyphosphates have been reported to act as second messengers in mammalian cells [20] and are synthesized by amino-acyl tRNA synthetases, known modulators of the inflammatory response [21]. To examine whether the ApaG domain of human FBxo3 is capable of binding dinucleotide polyphosphates, diadenosine tri-, tetra-, and pentaphosphate (A3P, A4P, and A5P) were titrated into a sample of ¹⁵N-labeled FBxo3 263–407 and followed by ¹H–¹⁵N HSQC spectroscopy. No changes in the spectra were observed (Fig. 4C–E), suggesting that diadenine polyphosphates are not ligands for the FBxo3 ApaG domain.

Interaction of the FBxo3 ApaG domain with FBxl2

TRAF signaling, resulting in cytokine gene transcription, involves targeting of the panreactive TRAF repressor, FBxl2, for degradation by FBxo3 [4,14]. This interaction has been shown to be dependent on the FBxo3 ApaG domain, as FBxo3 deletions lacking the ApaG domain are defective in FBxl2 binding and ubiquitination. Immunoglobin/Fibronectin III-type folds are robust scaffolds and their ligand binding generally involves the loop regions, with loop-swaps between different molecules altering ligand specificities [17]. We reasoned that one of the FBxo3 ApaG loops may be responsible for FBxl2 binding and that this binding may be affected by sequence changes. Therefore, each of the four extended loops of the ApaG domain was replaced with the corresponding loop from the E. coli ApaG protein in the context of the full-length FBxo3 protein. These chimeric FBxo3 proteins were expected to maintain structural integrity while possibly losing functionality. Binding of each FBxo3 chimera, generated by in vitro transcription/translation, to FBxl2, from murine lung epithelial (MLE-12) cells, was assessed by immunoprecipitation. As previously reported [4], FBxl2 obtained from these cells contains at least one phosphorylated threonine, T404, and should be competent for binding FBxo3 (Figs 5 and 7). Figure 5 clearly demonstrates that the chimera in which loop 1, residues 294–303 of FBxo3 (PELSSVHPPH), was replaced with the equivalent loop, residues 17–26 from the E. coli protein (EAQSSPDNER), has a substantially reduced ability to interact with FBxl2.

Fig. 5.

Identification of the important loop for FBxo3/FBxl2 interaction. Each loop in the FBxo3 278–407 structure was replaced with the corresponding sequence from the E. coli ApaG protein. FBxl2 coupled to agarose beads was incubated with in vitro synthesized and V5-tagged WT FBxo3 or FBxo3 loop chimeras. FBxl2 and any bound proteins were then pelleted, washed to remove any nonspecifically bound protein, and then blotted against α-V5 or α-FBxl2. The FBxl2 immunoprecipitated from the MLE-12 cells was blotted against α-phosphothreonine (P-Thr 47 kDa) indicating that it is in a FBxo3-binding competent state. The electrophoretic difference observed for the WT FBxo3 and the loop chimeras is due to the WT construct containing a 24-amino acid linker separating the V5 tag and main body of the protein that is not present in the loop chimeras.

Fig. 7.

Interaction of the FBxo3 ApaG domain with full-length FBxl2. FBxl2 coupled to agarose beads was incubated with recombinant N-terminally or C-terminally V5-tagged FBxo3 263–407. The beads were pelleted, washed to remove any nonspecifically bound protein, and blotted against α-V5 or α-FBxl2. FBxl2 protein that was immunoprecipitated from the MLE-12 cells was also blotted against an α-phosphothreonine antibody (P-Thr 47 kDa), for confirming that it is in a FBxo3-binding competent state.

Having established that residues 294–303 influence the interaction with FBxl2, the next step was to determine whether loop 1 directly contacted FBxl2 or whether this loop contributed to the regulation of FBxl2 binding through other interactions. FBxl2, phosphorylated on T404, had been shown to be the primary determinant for recognition by FBxo3 [4]. Attempts to produce the C-terminal FBxl2 region in E. coli, which contains the purported FBxo3-binding site, were unsuccessful due to extensive proteolysis (C. Troy & T.C.K. Krzysiak, unpublished data). Therefore, chemically synthesized peptides with and without phosphorylation on T404 were employed for binding studies. The design of these peptides was based on the phosphothreonine-binding determinants of forkhead-associated domains. Most phosphoserine/threonine-binding proteins bind their targets using structured regions; only the forkhead-associated domain has been demonstrated to use unstructured loops to bind phosphothreonine-containing proteins [22], and the phosphorylated residue as well as the +3 residue are the binding determinants. We therefore designed the FBxl2 peptides to contain FBxl2 amino acids 401–415, a region, according to secondary structure prediction algorithms, that contains the phosphorylated residue as well as the final structural element of the protein, an α-helix. Both peptides were readily soluble in aqueous solution and their addition to the FBxo3 ApaG domain was followed by ¹H–¹⁵N HSQC spectroscopy. Titration of ¹⁵N-labeled FBxo3 ApaG domain with the unphosphorylated peptide, up to a concentration of 1 mm, did not result in any changes in the ¹H–¹⁵N HSQC spectrum (Fig. 6A). Surprisingly, the same was observed when the phosphorylated peptide was used (Fig. 6B), indicating that it also did not bind to the FBxo3 ApaG domain.

Fig. 6.

Interaction of the FBxo3 ApaG domain with FBxl2 peptides. Superposition of the ¹H–¹⁵N HSQC spectra of FBxo3 263–407 in (A) the absence (black) and presence of 1 mm unphosphorylated FBxl2 peptide (red) or (B) 1 mm T404 phosphorylated FBxl2 peptide (green).

In the canonical model for substrate recognition by a F-box protein, the site of phosphorylation serves as the anchor point between the F-box protein and its target; however, many more possible modes of interaction exist and cannot be excluded a priori [23]. Therefore, in order to evaluate whether the FBxo3 ApaG domain is able to interact with FBxl2 using another region of the protein, FBxl2 from murine lung epithelial (MLE-12) cells was incubated with recombinant FBxo3 263–407, containing either an N-terminal or C-terminal V5 tag, and the interaction was assessed by coimmunoprecipitation. Phosphorylated FBxl2 was not able to coimmunoprecipitate with the FBxo3 ApaG domain, irrespective of the position of the V5 tag (Fig. 7). Therefore, these data suggest that the ApaG domain of human FBxo3, particularly loop 1, is critical for the interaction with FBxl2, although no direct contacts between this domain and FBxl2 were detected.

Discussion

Only two proteins in the human genome contain a domain with sequence similarity to bacterial ApaG proteins [13]. Interestingly, this domain plays a critical immunological role in the regulation of cytokine-driven inflammation by targeting a suppressor of TRAF signaling, FBxl2, for proteosome-mediated degradation [14]. Initial characterization of the interaction between FBxl2 and FBxo3 revealed that the interaction was dependent on the phosphorylation of FBxl2 T404 and the presence of the ApaG domain of FBxo3 [4,14]. In order to elucidate the structural basis of the interaction, the X-ray structure of the ApaG domain of FBxo3 was solved and direct contacts between the two proteins were explored. Given the 38% sequence similarity between the E. coli ApaG protein and the FBxo3 ApaG domain, not surprisingly, the structure of FBxo3 ApaG exhibits a Immunoglobin/Fibronectin III-type domain. The only differences in secondary structure elements observed are the addition of an N-terminal β-strand and a proline-induced break in β7. More pronounced differences are seen in the loop regions, with loops 2 and 4 containing insertions and loop 1 adopting a random coil conformation in the FBxo3 ApaG domain, while it is α-helical in the bacterial proteins. As can be appreciated from the amino acid sequence comparison (Fig. 1), only loop 3 has more than two residues in common with the E. coli sequence, suggesting that the ApaG domain, like Immunoglobin/Fibronectin III-type structures, encode different functionalities via variable loops that are grafted onto a common core.

Two activities that were proposed for prokaryotic ApaG proteins were evaluated for the ApaG domain of Fbxo3. The first is the link between the bacterial ApaG proteins and divalent cation homeostasis. Specifically Mg²⁺ and Co²⁺ have been implicated, as disruption of the apaG locus in S. typhimurium leads to lowered Mg²⁺ efflux and Co²⁺ resistance [12]. Titrations of the ApaG domain of FBxo3 with these cations did not indicate that specific binding sites are present (Fig. 4), suggesting no direct involvement of these cations with the FBxo3 ApaG domain. The second prokaryotic activity that was examined for the human FBxo3 ApaG domain was a possible interaction with dinucleotide polyphosphates. In prokaryotes, the ApaG protein is expressed as part of an operon that also expresses the ApaH protein. ApaH is a hydrolase responsible for the degradation of dinucleotide polyphosphates that accumulate as a result of metabolic stress [10,11]. Diadenine polyphosphates act as second messengers in mammalian cells and are produced by amino-acyl tRNA synthetases that are known to contribute to inflammatory responses. As described above, however, titration of the FBxo3 ApaG domain with three of these molecules: A3P, A4P, and A5P did not reveal any interaction (Fig. 4C–E).

Having established that Mg²⁺, Co²⁺, and diadenine polyphosphates do not directly interact with the FBxo3 ApaG domain, the role of the ApaG domain in regulating interactions between FBxl2 and FBxo3 was evaluated [14]. To clearly define the contact regions of the interaction between FBxl2 and Fbxo3, each loop sequence of FBxo3 was systematically replaced with the corresponding sequence from the E. coli ApaG protein. It was observed that loop 1, amino acids 294–303, was critically important for the FBxl2/FBxo3 interaction (Fig. 5). On the FBxl2 side, the canonical model of substrate recognition by F-box proteins evoked phosphothreonine 404 as the target for recognition by the FBxo3 ApaG domain. Surprisingly, however, no binding to a C-terminal peptide of FBxl2 was detected in either the unphosphorylated or the phosphorylated form (Fig. 6). Given that it is possible by NMR to detect even very weak binding (micro-millimolar), it seems likely that other regions of the two proteins may be involved in the interaction. This hypothesis was confirmed by the failure of the FBxo3 ApaG domain to coimmunoprecipitate with full-length, phosphorylated FBxl2 (Fig. 7). In lieu of confirming a direct physical interaction between the FBxo3 ApaG domain and Fbxl2, three possible scenarios for explaining the current observations exist: (a) the ApaG domain of FBxo3 governs cellular localization and contact with the C terminus of FBxl2 involves a second region of FBxo3; (b) an unknown factor binds to loop 1 of the ApaG domain and bridges the interaction with FBxl2; (c) loop 1 of the ApaG domain allosterically regulates the availability of a second region in FBxo3 for FBxl2 binding.

All the available data suggest that scenario (c), namely an allosteric regulation by loop 1 of the ApaG domain and the involvement of a second interaction site is likely. A secondary site, important for the regulation of FBxl2 degradation by FBxo3, has been previously identified. The naturally occurring V221I mutation acts as a loss of function mutation and leads to elevated FBxl2 levels and lower levels of the TRAF proteins [4]. Furthermore, this residue is also located in another putative protein–protein interaction motif, belonging to the SUKH superfamily [8], and a region of FBxo3 that secondary structure predictions indicate is predominantly unstructured. Unlike the three other loops of the FBxo3 ApaG domain, loop 1 is located on the opposite end of the domain facing the same direction as the burgeoning N terminus, and hence the SUKH domain. We propose that the ApaG domain regulates FBxo3 binding of FBxl2 by positioning the region around V221 for efficient interaction with FBxl2. In other words, the function of the ApaG domain is to provide a scaffold, such that loop 1 can position critical motifs in the FBxl2/FBxo3 interaction.

Materials and methods

Cloning and protein expression

The cDNA for human FBxo3 was acquired from Thermo-Fisher (Waltham, MA, USA; MHS6278-202758944). To determine the boundaries of the FBxo3 ApaG domain, the FBxo3 and E. coli ApaG protein sequences were aligned using clustalx2 [24]. FBxo3 278–407 and 263–407 were PCR amplified using the forward primers 5′-CTGGAAGAATTCGTAGCAACAACTGGGGATATTACT and 5′-CTGAAGGAATTCATCATCAGAGACCAAATTTTCAGATATG, respectively, with the common reverse primer 5′-ATCATCCTCGAGTTACCTGAATGTTGGACATGCCATATG. Both N-terminally and C-terminally V5-tagged FBxo3 263–407 constructs were ordered as gBlocks from Integrated DNA Technologies (Coralville, IA, USA). The amplified regions and gBlocks were cloned into a modified pET41a plasmid containing a TEV site immediately preceding the cloning site. Plasmids were transformed into the E. coli BL21(DE3)-STAR cell line for expression.

Proteins were produced by expression in modified M9 minimal media using ¹⁵N-ammonium chloride as the nitrogen source. Cells were grown to OD₆₀₀ ~ 0.8, and protein expression was induced with 400 μm IPTG at 18 °C for 16– 18 h. Cells were harvested by centrifugation (4600 g; 10 min; 4 °C), resuspended in 20 mm HEPES pH 7.0, 500 mm NaCl, 5 mm DTT, and lysed using a mircrofluidizer (Microfluidics). DNase (80 μg·mL⁻¹) and RNase A (64 μg·mL⁻¹) were added to the lysate and the reaction was incubated with stirring at 4 °C for 1 h. The lysate was clarified by centrifugation (38 000 g/50 min/4 °C) and applied to a GSTrap column (GE Life Sciences, Marlborough, MA, USA). Bound protein was eluted with 20 mm HEPES pH 8.0, 500 mm NaCl, 40 mm reduced glutathione, and immediately subjected to gel filtration over a Superdex 200 26/60 column (GE Life Sciences) equilibrated in 20 mm HEPES pH 7.5, 150 mm NaCl, and 0.5 mm TCEP. The fusion protein was then cleaved with TEV protease overnight at 4 °C. Further chromatography over a Ni²⁺ affinity column removed the cleaved tag and the TEV protease, and the FBxo3 ApaG domain containing flow through was passed over a Superdex 75 16/60 column (GE Life Sciences) equilibrated in 20 mm HEPES pH 7.5, 100 mm NaCl, and 0.5 mm TCEP for buffer exchange and final purification. Protein purity was estimated > 99% by SDS/PAGE.

Protein crystallization and structure determination

FBxo3 278–407 was initially crystallized using the micro-batch method, mixing 400 μm protein in 20 mm HEPES pH 7.5, 100 mm NaCl, and 0.5 mm TCEP with 100 mm Tris/HCl pH 8.5, 20% w/v PEG 8000, 200 mm MgCl₂•6H₂0 (Molecular Dimensions JCSG-1 screen #42) at a 1 : 1 ratio under 50 μL of a 70 : 30 mix of light paraffin oil (EMD, Darmstadt, Germany; PX0047-1) and polydimethylsiloxane (Sigma-Aldrich, St. Louis, MO, USA; 46319). Crystals used for structure determination were grown using the hanging drop method, mixing 200 μm protein in 20 mm HEPES pH 8.0, 100 mm NaCl, and 0.5 mm TCEP with 100 mm Tris/HCl pH 8.5, 20% w/v PEG 8000, 200 mm MgCl₂•6H₂0 at a 1 : 2 ratio. Crystals were cryoprotected by soaking for a few seconds in well solution supplemented with 25% ethylene glycol prior to flash-cooling (−180 °C).

Diffraction data were collected using the Southeast Regional Collaborative Access Team (SER-CAT) 22-ID beamline at the Advanced Photon Source, Argonne National Laboratory. d*trek software was used to process, integrate, and scale the diffraction data [25], and the CCP4 software package was used to convert the data into mtz format [26]. The structure of FBxo3 278–407 was determined by molecular replacement using the structure of the ApaG protein from V. cholera (PDB ID 1XVS) as the structural probe in phaser [27]. By alternating between manual rebuilding of the structure in Coot [28] and computational refinement using the phenix software package [29,30], the structure was iteratively refined. Only residues 406 and 407 are not accounted for by the electron density, and all other residues exhibiting electron density lie in the favored and allowed regions of the Ramachandran plot according to molprobity [31] evaluation. All structural statistics are provided in Table 1. Structural figures were generated using pymol [32].

Table 1.

Crystallographic statistics for FBxo3 278–407.

Data collection
Space group	P21212
Cell dimensions
a, b, c (Å)	96.813, 30.477, 40.517
α, β, γ (°)	90, 90, 90
Resolution (Å)	37.38–2.0 (2.07–2.00)
R merge	0.076 (0.417)
<//σ/>	10.3 (2.7)
Completeness (%)	99.35 (97.8)
<Redundancy>	5.41 (5.57)
Refinement
Resolution (Å)	37.38–2.0
No. reflections	8559
Rwork/RFree	0.1996/0.2381
No. atoms
Protein	1053
Water	39
B-factors
Protein	55.84
Water	52.47
R.M.S. deviations
Bond lengths (Å)	0.013
Bond angles (°)	1.329

NMR spectroscopy

2D ¹H–¹⁵N HSQC spectra were recorded at 25 °C using a Bruker AVANCE 700 MHz NMR spectrometer (Bruker, Billerica, MA, USA) equipped with a z-axis gradient cryoprobe. For metal titrations, MgSO₄ (final concentration: 50 μm–1.4 mm) or CoCl₂ (final concentration: 10 μm–100 μm) was added to 100 μm FBxo3 ApaG domain in 20 mm HEPES pH 7.0, 100 mm NaCl, and 0.5 mm TCEP. For dinucleotide polyphosphate titrations, diadenine triphosphate (A3P), diadenine tetraphosphate (A4P), and diadosine pentaphosphate (A5P) were purchased from Sigma-Aldrich and dissolved in 20 mm HEPES, 100 mm NaCl, and 0.5 mm TCEP, and the solutions were adjusted to pH 7.0. Each diadenine polyphosphate was titrated on to 75 μm FBxo3 ApaG domain in the same buffer (final concentration: 200 μm–1 mm). 15-mer FBxl2 peptides (APVTPPTAVAGSGQR) containing T404, either unmodified or phosphorylated, were purchased from CHI Scientific (Maynard, MA, USA) at > 95% purity form. Lyophilized peptide was dissolved to 10 mm in 20 mm HEPES pH 7.5, 100 mm NaCl, and 0.5 mm TCEP, and added in aliquots to 50 μm protein in the identical buffer. The final concentrations of unphosphorylated and phosphorylated peptides ranged from 25 μm to 1 mm and 5 μm to 1 mm, respectively.

Light scattering

Size-exclusion multi-angle light scattering (SEC-MALS) measurements were performed using a Superdex 75 10/300 column (GE Life Sciences) with in-line multi-angle light scattering (HELIOS, Wyatt Technology, Goleta, CA, USA), UV (Agilent 1100; Agilent Technology, Santa Clara, CA, USA), and refractive index (OptilabrEX; Wyatt Technology) detectors. astra software V5.3.1.4 (Wyatt Technology) was used for data analysis. For each experiment, the system was equilibrated in 20 mm HEPES pH 7.0, 7.5, or 8.0, 100 mm NaCl, 0.5 mm TCEP, and 100 μl of FBxo3 ApaG domain (residues 263–407). Starting concentrations of 50, 100, 200, or 400 μm were injected.

in vitro FBxo3/FBxl2 immunoprecipitation assays

FBXL2 protein was immunoprecipitated (4 h; 25 °C) from 10⁷ MLE-12 cells lysed in IP buffer (50 mm Tris pH 7.6, 150 mm NaCl, 0.1% v/v Triton X-100, 1 mm PMSF) using α-FBxl2 antibodies (Aviva Biosciences, San Diego, CA, USA; rabbit OAAB02444) coupled to Protein A/G agarose resin (Thermo Fisher, 20422) to a final concentration of 50 μg·mL⁻¹. Sources of the murine lung epithelial (MLE-12) cells were described previously [33,34]. Full-length FBxo3 prey was prepared via in vitro transcription–translation (Promega, Madison, WI, USA; L1170) and the FBxo3 ApaG domain was purified as described above. Full-length FBxo3 wild-type and chimera plasmids were incubated with the kit components following the manufacturer’s instructions to a final concentration of 0.01 mg·mL⁻¹ for 1 h at 30 °C. The immunoprecipitated and resin-coupled FBxl2 bait was incubated with the in vitro synthesized FBxo3 or the purified FBxo3 ApaG domain prey for 16 h at 4 °C. Bait and prey were washed with IP buffer and eluted from resin via heating at 88 °C for 5 min in 60 μl of 1× Laemmli sample buffer. Samples were resolved via 12% acrylamide SDS/PAGE and subjected to immunoblotting for α-FBxl2, α-phosphothreonine (Cell Signaling Technologies, Danvers, MA, USA; 9381), and α-V5 (ThermoFisher; R960CUS) using chemiluminescent detection (Advansta, Menlo Park, CA, USA; K-12043). All binding assays were performed in triplicate.

Abbreviations

ApaG

adenine tetraphosphate adenine G

ApaH

adenine tetraphosphate adenine H

FBxl2

F box leucine rich 2

FBxo3

F box only 3

HEPES

2-[4-(2-hydroxyethyl)piperazine-1-yl]ethanesulfonic acid

NF-_ΚB

nuclear factor kappa B

PDIP38/PolyDip2

polymerase delta-interacting protein 38/2

SCF

Skp1-Cullin1-F-box

SEC-MALS

size-exclusion multi-angle light scattering

SERCAT

Southeast Regional Collaborative Access Team

SUKH

Syd US22 Knr4 homology

TCEP

tris(2-carboxyethyl)phosphine

TNFR

tumor necrosis factor receptor

TRAF

tumor necrosis factor receptor-associated factor

WT

wild-type