Perturbation of the monomer-monomer interfaces of the benzoylformate decarboxylase tetramer.

The X-ray structure of benzoylformate decarboxylase (BFDC) from Pseudomonas putida ATCC 12633 shows it to be a tetramer. This was believed to be typical of all thiamin diphosphate-dependent decarboxylases until recently when the structure of KdcA, a branched-chain 2-keto acid decarboxylase from Lactococcus lactis, showed it to be a homodimer. This lent credence to earlier unfolding experiments on pyruvate decarboxylase from Saccharomyces cerevisiae that indicated that it might be active as a dimer. To investigate this possibility in BFDC, we sought to shift the equilibrium toward dimer formation. Point mutations were made in the noncatalytic monomer-monomer interfaces, but these had a minimal effect on both tetramer formation and catalytic activity. Subsequently, the R141E/Y288A/A306F variant was shown by analytical ultracentrifugation to be partially dimeric. It was also found to be catalytically inactive. Further experiments revealed that just two mutations, R141E and A306F, were sufficient to markedly alter the dimer-tetramer equilibrium and to provide an ~450-fold decrease in kcat. Equilibrium denaturation studies suggested that the residual activity was possibly due to the presence of residual tetramer. The structures of the R141E and A306F variants, determined to <1.5 Å resolution, hinted that disruption of the monomer interfaces will be accompanied by movement of a loop containing Leu109 and Leu110. As these residues contribute to the hydrophobicity of the active site and the correct positioning of the substrate, it seems that tetramer formation may well be critical to the catalytic activity of BFDC.

Oligomerization of proteins in biological systems is a common phenomenon, and protein–protein interactions are often involved in critical cellular processes. The destabilization of interfaces in multimeric protein complexes is thought to contribute to diseases ranging from neurodegenerative disorders to muscular dystrophy.[1−3] Because proteins rarely act alone in vivo, predicting networks of protein–protein interactions, i.e., the “interactome”, has become fundamental to the development of systems biology.[4−6] As a consequence, predicting how any given protein will interact with other biomolecules, understanding the forces stabilizing protein–protein interfaces, and learning how to destabilize an individual interface have become the focus of much research.[7−9]

One of the most common protein–protein interactions is self-association.[10,11] In essence, this allows a single gene of relatively short length to encode a large protein system by allowing multiple peptide sequences encoded from that gene to interact and form a large multimeric protein. In addition to a reduced genome and fewer problems with transcription, this system provides benefits such as an increase in the effective concentration within the cell, enhanced protein stability, and additional regulatory flexibility.[11] While self-association is not necessarily a prerequisite for activity, the active sites of perhaps one-sixth of all oligomeric enzymes are located at highly stabilized interfaces between two peptide chains.[11]

Such is the case with thiamin diphosphate (ThDP)-dependent enzymes.[12,13] While this group of enzymes has evolved to catalyze a wide range of chemical reactions,[14] X-ray structures have shown that overwhelmingly they have developed only one way to bind the ThDP cofactor, namely at the interface between two monomers. This creates two active sites per homodimer with each monomer contributing either a pyrimidine binding (PYR) domain or a diphosphate binding (PP) domain to a given active site (Figure 1). At a minimum, therefore, any ThDP-dependent enzyme must form a dimer.[12,13] Curiously, until recently, all the ThDP-dependent enzymes whose sole in vivo function is to catalyze a decarboxylation reaction were found to be homotetramers, albeit more accurately described as dimers of active dimers.[15−21] The branched-chain 2-keto acid decarboxylase from Lactococcus lactis (KdcA, EC 4.1.1.72) was the first to deviate from this trend when its X-ray structure [Protein Data Bank (PDB) entry 2VBG] showed to it be dimeric.[22] Intriguingly, although the archetypal ThDP-dependent decarboxylase, pyruvate decarboxylase from Saccharomyces cerevisiae (ScPDC, EC 4.1.1.1), is a tetramer in the crystal lattice, analytical ultracentrifugation (AUC) experiments showed it to be a dimer at low solution concentrations.[23] Further, using urea as a denaturant, it was possible to shift the dimer–tetramer equilibrium in favor of the dimer and to demonstrate that the dimer was catalytically active.[24,25]

Figure 1

ThDP (yellow) in the active site of BFDC. The cofactor interacts with both the phosphate-binding domain of monomer A (green) and the pyrimidine-binding domain of monomer B (orange). Prepared using PyMOL using data from PDB entry 1BFD.

Benzoylformate decarboxylase (BFDC, EC 4.1.1.7) is another ThDP-dependent decarboxylase that has been shown to be a homotetramer.[18] First identified as part of the mandelate pathway of Pseudomonas putida,[26] BFDC catalyzes the nonoxidative decarboxylation of benzoylformate to yield benzaldehyde and carbon dioxide.[27] Not surprisingly, given the similarity of their reactions, there is considerable structural homology between BFDC and PDC, including the conservation of several catalytic residues. As part of a project examining the evolution of ThDP-dependent enzymes, we have been attempting to interconvert the two enzymes, but with limited success.[28,29] Unlike ScPDC, it has never been determined whether BFDC is active as a dimer, nor has it been shown what advantage, if any, the tetramer provides to BFDC. Here, we report the use of site-directed mutagenesis coupled with analytical centrifugation and X-ray crystallography to answer two questions. (i) Can BFDC be converted into a dimer? (ii) Is the tetramer required for enzymatic activity?

Materials and Methods

Primers were synthesized by Integrated DNA Technologies. Recombinant alcohol dehydrogenase from Equus caballus (HLADH) was obtained as described previously.[30] Reduced β-nicotinamide adenine dinucleotide (NADH), isopropyl β-d-thiogalactopyranoside (IPTG), and benzoylformate were purchased from Sigma-Aldrich. A gel filtration calibration kit (high molecular mass) and HiPrep 16/60 Sephacryl S-200 High Resolution column were purchased from GE Healthcare. Nickel-nitrilotriacetic acid (NTA) resin was purchased from Qiagen. Pfu DNA polymerase was purchased from Stratagene. Buffers and other assay reagents were purchased from either Sigma-Aldrich or Fisher Scientific and were of the highest grade commercially available. Sequencing was conducted by the University of Michigan DNA Sequencing Core Facility.

Analysis of Dimer and Tetramer Interfaces

The X-ray structure of wt BFDC at 1.60 Å resolution (PDB entry 1BFD) was used as the model for analysis of the interfaces. The SPPIDER server (http://sppider.cchmc.org) was used to calculate the total surface area and the interface surface area and to verify that residues selected for mutagenesis were not involved in the formation of the active dimer. The PDB visualization program PyMOL (Schrödinger, LLC) was used to identify targets for mutagenesis and for the preparation of figures.

Plasmids

Mutagenesis was performed on BFDC expression vector pET24dBFDC-His.[31,32] Primers (Table 1 of the Supporting Information) were designed, and site-directed mutagenesis was performed according to the QuikChange mutagenesis protocol (Agilent) utilizing Pfu DNA polymerase. DpnI was used to digest parental DNA prior to its transformation into chemically competent TOP10 cells. Each plasmid used for protein expression had the complete mdlC gene sequenced to ensure that only the desired mutations were present.

Protein Expression and Purification

Expression of wt BFDC and its variants was conducted in Escherichia coli strain BL21(DE3). Induction of the recombinant protein was initiated with 1 mM IPTG. Following induction, the cultures were grown at room temperature overnight (18–20 h). Enzymes were purified as described previously.[31] The protein was exchanged into BFDC storage buffer [100 mM KHPO4, 1 mM MgSO4, 0.5 mM ThDP (pH 6.0), and 10% (v/v) glycerol] and concentrated using Amicon Ultra centrifugal filters (Millipore). Purity was assessed by sodium dodecyl sulfate–polyacrylamide gel electrophoresis, and the protein concentration was determined from the absorbance at 280 nm using a molar extinction coefficient of 62220, calculated from the sequence of wt BFDC with the ExPASy ProtParam tool.[33] Enzymes were stored at −80 °C until they were required.

Steady-State Kinetic Analysis of Wild-Type and Variant BFDC

The coupled activity assay of the purified enzymes was performed as described previously.[34] The assay mixture contained bovine serum albumin (BSA, 1 mg/mL), NADH, and HLADH in 100 mM potassium phosphate buffer (pH 6.0), and varying concentrations of benzoylformate in a final volume of 1 mL. Reactions were conducted at 30 °C and were initiated by the addition of the enzyme. Assays were performed in triplicate. Steady-state kinetic parameters were determined by fitting the initial rate data to the Michaelis–Menten equation, using the enzyme kinetics module of SigmaPlot version 9.0.1 (Systat Software, Inc.).

For the equilibrium denaturation studies, assays were conducted in denaturation buffer using the direct UV assay of Hegeman.[26] The decrease in benzoylformate concentration was monitored at 334 nm.

Molecular Mass Predicted by Size-Exclusion Chromatography (SEC)

Size-exclusion chromatography (SEC) was performed using an AKTA fast performance liquid chromatography system (GE Healthcare) equipped with a HiPrep 16/60 Sephacryl S-200 High Resolution column (1.6 cm × 94 cm). The column was equilibrated with running buffer [50 mM NaPO4 and 150 mM NaCl (pH 7.5)] prior to calibration. Blue Dextran was used to determine the void volume of the column. A standard curve was obtained using a high-molecular mass gel filtration calibration kit (GE Healthcare). Protein standards were loaded according to the manufacturer’s protocol. BFDC variants were loaded onto the column via injection into a 2 mL loop at concentrations ranging from 0.1 to 1 mg/mL in a total volume of 400 μL. Kav values were calculated and used to predict the molecular mass and oligomerization state of the BFDC variants

SEC-MALS Analysis

Size-exclusion chromatography multiangle light scattering (SEC-MALS) was performed on wt BFDC and the R141E/A306F and R141E/Y288A/A306F variants. Three concentrations, 3.1, 1.5, and 0.75 mg/mL, were used for the mutant BFDC proteins. For the wt enzyme, concentrations of 2.0, 1.0, and 0.75 mg/mL were used. A Superdex 200 column coupled to the three-angle miniDawn TREOS and Optilab-Trex instrument (Wyatt Technologies) was used for collection of the light scattering and refractive index data. The system is coupled to an Agilent 1100 high-performance liquid chromatography system at a flow rate of 0.45 mL/min with an in-line spectrophotometer. Astra (version 6.0.3) was used to analyze the collected data.

Analytical Ultracentrifugation

Sedimentation velocity experiments were conducted on Beckman-Coulter XLA and XLI analytical ultracentrifuges. wt BFDC and the R141E/A306F and R141E/Y288A/A306F variants were dialyzed extensively against a buffer containing 100 mM NaPO4 (pH 7.5), 300 mM NaCl, 1 mM MgCl2, and 35 μM ThDP. The sample concentrations ranged from 0.25 to 1.0 mg/mL. The samples were then centrifuged at 42000 rpm using two-sector 1.2 cm path-length carbon-filled Epon centerpieces. The experiments were conducted on an An-50 Ti rotor at 20 °C. The density and relative viscosity of the buffers were calculated with SEDNTERP version 1.09 (http://www.jphilo.mailway.com/download.htm#SEDNTERP) to be 1.01949 g mL–1 and 0.01069 P, respectively. On the basis of its sequence, the partial specific volume (vbar) of the protein was calculated with SEDNTERP to be 0.7340 mL g–1. The samples were monitored at 280 nm with a 4 min delay and 150 scans. The c(s) distributions were analyzed using SEDFIT version 14.3e.[35] Molecular masses were calculated from the c(s) data using a shape assumption.

Urea Unfolding Monitored by Circular Dichroism (CD) Spectroscopy and Fluorescence Spectrophotometry

wt BFDC and the R141E/A306F variant (0.15 mg/mL) were incubated overnight at 4 °C with varying urea concentrations in 50 mM phosphate buffer (pH 6.0). Far-UV CD experiments were conducted on a Jasco J-810 spectropolarimeter. Unfolding was monitored by changes in ellipticity at 222 nm in a cell with a path length of 0.1 cm at 20 °C. Spectra were averaged over five scans, and the scan speed was set at 20 nm/min with a response time of 1 s and a slit width of 1 nm.

Fluorescence emission spectra were recorded on a Varian Cary Eclipse fluorescence spectrophotometer. The intrinsic fluorescence of wt BFDC and the R141E/A306F variant, both at 0.15 mg/mL in varying concentrations of urea in 50 mM phosphate buffer (pH 6.0) at 20 °C, was recorded from 295 to 450 nm using an excitation wavelength of 280 nm. The bandpass for excitation and emission was set at 5 nm with a scanning rate of 200 nm/min.

Crystallization of the R141E and A306F Variants

Crystals of BFDC R141E and BFDC A306F variants were grown by the hanging drop diffusion method under the same conditions used for the crystallization of wt BFDC.[18] Storage buffer was exchanged for crystallization buffer [0.1 mM MgCl2, 0.2 mM ThDP, and 25 mM NaHEPES (pH 7.0)]. The well solution consisted of 0.1 M Tris (pH 8.5), 0.15 M CaCl2, and 22% PEG400 (v/v). Equal volumes of protein (10 mg/mL) and well solution were pipetted onto a silanized glass slide and mixed. A heavy precipitate immediately formed from this mixture, and single crystals emerged from the precipitate within days. Crystals were transferred to fresh crystallization buffer containing 36% (v/v) glycerol as a cryoprotectant and mounted on Hampton CryoLoops immediately prior to being flash-frozen in liquid nitrogen.

X-ray Data Collection

Diffraction data were obtained at 100 K on the 23-ID-D and 23-ID-B beamlines administered by GM/CA-CAT at the Advanced Photon Source at Argonne National Laboratory (Argonne, IL). Data sets for the R141E and A306F variants were scaled to the I222 space group. Data reduction and processing of data sets were achieved using the HKL2000 software package and the CCP4 suite of programs.[36] Molecular replacement was performed using the search model wt BFDC (PDB entry 1BFD) with metals and waters removed. The asymmetric unit for each variant contained a single monomer.

Structure Solutions and Refinements

PHENIX was used for model refinement.[37] After each round of refinement, the electron density was manually inspected and models were built using Coot.[38] Refinement continued until Rfree and the crystallographic R factor could no longer be improved. The validity of the model was checked using the MolProbity server.[39] All images were generated using PyMOL (Schrödinger, LLC).

Results

Comparison of the Quaternary Structures of KdcA, ScPDC, and BFDC

It is clear from their respective X-ray structures that KdcA[22] is a homodimer and that BFDC[18] is a homotetramer (Figure 2). Although ScPDC was initially reported to be a homotetramer,[40] structural analysis suggests that its tetrameric form may arise simply by crystal packing.[40] To examine this possibility, we turned to the Solvent accessibility-based Protein-Protein Interface iDEntification and Recognition (SPPIDER) server that can be used to predict the total surface area of interfaces from a PDB file.[41] Analyses were conducted on three ThDP-dependent decarboxylases, KdcA (PDB entry 2VBF), ScPDC (PDB entry 1PYD), and BFDC (PDB entry 1BFD). In each case, the interface forming the active site, defined as the A–B interface (Figure 2), was clearly identified (Table 1). There was no evidence of any dimer–dimer interface in the KdcA structure, supporting the crystallographic finding of a homodimer. Interestingly, no evidence was found for any dimer–dimer interface when the ScPDC structure was used as the search model. This reinforces the suggestion that the tetramer observed in the crystal lattice of ScPDC might be an artifact of protein crowding at higher enzyme concentrations. Conversely, analysis of the X-ray structure of BFDC with SPPIDER indicates a robust dimer–dimer interface accounting for almost 10% of the total surface are of the enzyme (Table 1).

Figure 2

Defining the interfaces of BFDC monomers. (A) BFDC tetramer. Monomer A is colored green, monomer B orange, monomer C cyan, and monomer D magenta. (B) Cartoon with lines indicating interactions between the monomers (circles) and ThDP (diamonds). The active sites (one per monomer) are found at the A–B interface. (C–E) A–B, A–C, and A–D interfaces, respectively. Figures based on data from PDB entry 1BFD.

Table 1

SPPIDER Analysis of Monomer–Monomer Interfacesa

interface	KdcA surface area (Å2)	ScPDC surface area (Å2)	BFDC surface area (Å2)	ZmPDC surface area (Å2)
monomer	21856	21819	20611	22364
chain A–chain B	3458 (16%)	2819 (13%)	3475 (17%)	4204 (19%)
chain A–chain C	ndb	ndb	488 (2%)	611 (3%)
chain A–chain D	ndb	ndb	1601 (8%)	1226 (5%)

Surface areas were calculated from PDB entries 2VBF (KdcA), 1PYD (ScPDC), 1BFD (BFDC), and 1ZPD (ZmPDC) using the SPPIDER server.

No interface can be detected by SPPIDER.

Analysis of the Dimer–Dimer Interfaces of BFDC

SPPIDER analysis of the quaternary structure of BFDC shows that each chain has significant interactions with every other chain in the tetramer (Table 1). The interface that forms the active dimer, the A–B interface, was calculated to have the largest total area of contact (3540 Å2), ∼17% of the solvent accessible surface area of the monomer. It seems logical that disruption of the A–B interface would result in a loss of enzyme activity; therefore, our attempts to convert BFDC into a dimer were focused on disrupting the A–C and A–D interfaces (Figure 2).

The A/D interface was the next largest, with a total surface area of ∼1600 Å2 or about 8% of the total monomer A surface area. Here Tyr288 and Ala306 were identified as possible candidates for mutagenesis. Tyr288 of monomer A has 12 points of contact within 5 Å of monomer D (Figure 3), and it was reasoned that elimination of most of these interactions by removal of the phenyl ring should result in a less stable A/D interface. Additionally, the methyl side chain of Ala306 on monomer A is located in a shallow hydrophobic cleft of monomer D (Figure 3), and it was thought that replacement by a bulkier residue such as phenylalanine would not be tolerated, and may help disrupt the interface.

Figure 3

Interactions at the A–D interface. (A) Tyr288 of monomer A is found to have 12 points of contact (≤5 Å) with monomer D. (B) Ala306 of monomer A is located in a shallow hydrophobic cleft of monomer D.

At ∼500 Å2, the A–C interface accounts for ∼2.5% of the total surface area of monomer A and is mostly comprised of residues in charge–charge interactions. The side chains of Glu107 and Asp114 of monomer A, for example, form salt bridges with Arg141 and Arg120, respectively, of monomer C (Figure 4). Because of the 2-fold symmetry, Arg120 and Arg141 of monomer A also interact with Asp114 and Glu107 of monomer C. These four positions were selected as targets for mutagenesis, specifically to introduce like-charge repulsions in an attempt to destabilize the dimer–dimer interface. A summary of the various interactions, and the rationale for the mutagenesis experiments can be found in Table S2 of the Supporting Information.

Figure 4

Interactions at the A–C interface. View of charged residues highlighting those with interatomic distances of ∼3 Å.

Kinetic and SEC Analysis of A–C and A–D Single-Mutation Variants

All variants expressed as soluble protein. Using the typical assay conditions described in Materials and Methods, Km and kcat values of the A–D variants were very similar to those of wt BFDC (Table 2). The A–C variants also exhibited no change in their Km values, but compared to that of wt BFDC, there was a modest 2–4-fold decrease in their kcat values. All variants eluted from the SEC column at a volume indicative of a tetramer (Table 2). On the basis of these results, it appeared that the single mutations did not noticeably alter the dimer–tetramer equilibrium or greatly affect the overall catalytic activity of the enzyme.

Table 2

Steady-State Kinetic Parametersa,b and Oligomerization Statesc of the BFDC Variants

variant	kcat (s–1)b	Km (mM)	kcat/Km (mM–1 s–1)b	oligomerization state at ≤0.5 mg/mL
wt BFDC	350 ± 30 (1)	0.30 ± 0.03	1200 (1)	tetramer
Y288A	310 ± 20 (1)	0.25 ± 0.03	1200 (1)	tetramer
A306F	250 ± 23 (1)	0.31 ± 0.09	890 (1.3)	tetramer
R141E	180 ± 22 (2)	0.19 ± 0.05	950 (1.3)	tetramer
R120E	130 ± 26 (3)	0.29 ± 0.10	450 (2.7)	tetramer
E107R	130 ± 10 (3)	0.33 ± 0.10	390 (3)	tetramer
D114R	83 ± 12 (4)	0.25 ± 0.10	330 (3.6)	tetramer
E107R/D114R	0.8 ± 0.04 (440)	0.35 ± 0.05	2 (585)	tetramer
R141E/Y288A	15 ± 2 (23)	0.21 ± 0.03	70 (17)	tetramer
R141E/A306F	0.8 ± 0.3 (440)	0.25 ± 0.02	3 (400)	evidence of dissociation
Y288A/A306F	0.1 ± 0.01 (3500)	0.33 ± 0.03	0.3 (4000)	tetramer
R141E/Y288A/A306F	nadd	nadd	nadd	evidence of dissociation
L109A	3.9 ± 0.3 (90)	0.80 ± 0.1	4.8 (250)	nde
L110A	1.5 ± 0.3 (230)	2.8 ± 0.3	0.5 (2400)	nde

Values are means of three independent determinations ± the standard error (SE). All variants were prepared with C-terminal hexahistidine tags and assayed as described in Materials and Methods.

In parentheses are fold decreases over the wt value.

Oligomerization states for variants were determined by SEC as described in Materials and Methods.

No activity detected.

Not determined.

Generation and Analysis of the R141E/Y288A/A306F Variant

The inability of a single mutation to produce a dimeric BFDC prompted a reanalysis of the A–D and A–C interfaces. The A–D interface accounts for nearly 10% of the total surface area; therefore, it was not surprising that more than one unfavorable interaction would be required to disrupt it. It was noted that the hydrophobic pocket that accommodates Ala306 comprised Met145, Met148, and the hydrophobic region of the Arg141 side chain. It was reasoned that replacement of Arg141 with a glutamic acid would have a dual effect in that it would set up an unfavorable interaction with Glu107 and reduce the hydrophobicity of the Ala306 binding pocket. To maximize the destabilizing interactions, the R141E/Y288A/A306F variant was constructed. The triple mutant expressed well and was purified as a soluble protein. When subjected to SEC chromatography at 1 mg/mL, it eluted at a volume similar to that of wt BFDC but, at ≤0.5 mg/mL, it eluted at a larger volume, suggestive of a change in oligomerization state (Table 2). Unfortunately, it also was inactive in the assay.

Verification of the Quaternary Structure of wt BFDC and the R141E/Y288A/A306F Variant

To validate the SEC observations and to confirm that the oligomerization state of the triple mutant was indeed concentration-dependent, we turned to a combination of analytical ultracentrifugation and SEC-MALS. Sedimentation velocity experiments were performed, and the sedimentation coefficients (S) for wt BFDC were determined at concentrations of 0.25, 0.5, and 1.0 mg/mL (Figure S1 of the Supporting Information). At each concentration, an S value of 8.4 was calculated from a c(s) distribution (Table S3 of the Supporting Information). The calculated molecular mass from these experiments was 241 ± 22 kDa, which agrees with the theoretical molecular mass of ∼229 kDa for the BFDC tetramer. Additional SEC-MALS analysis provided a molecular mass of 235.1 ± 16.5 kDa, also corresponding to a tetrameric species (Table S4 of the Supporting Information). Similar experiments were performed on the triple mutant, and by contrast, the S value determined at the same concentrations was found to be 7.9 (Figure S1 of the Supporting Information). The SEC-MALS analysis for the R141E/Y288A/A306F variant showed a pronounced concentration-dependent shift in both molecular mass and retention time (Figure S2 and Table S4 of the Supporting Information). All these changes pointed to a shift in the dimer–tetramer equilibrium.

Analysis of the R141E/Y288A/A306F Permutation Variants

Having seemingly changed the dimer–tetramer equilibrium, we thought it was important to determine which combination of mutations was crucial in bringing about the change. Accordingly, the three permutations of the R141E/Y288A/A306F mutant were constructed. Once again, the variants were expressed as soluble proteins and could be assayed under standard conditions. Like the single mutants, R141E/Y288A, R141E/A306F, and Y288A/A306F variants all had Km values similar to that of wt BFDC. Conversely, they all showed considerable decreases in kcat values, with the greatest decrease (∼3500-fold) seen for Y288A/A306F (Table 2). Each of the variants was subjected to SEC chromatography at concentrations ranging from 0.1 to 1 mg/mL. While R141E/Y288A and Y288A/A306F eluted at volumes indicative of a tetramer regardless of concentration, the R141E/A306F variant exhibited concentration-dependent changes similar to those shown previously by the R141E/Y288A/A306F variant. In some ways, this was a surprising result as the kcat value for R141E/A306F was 0.8 s–1, admittedly a ∼450-fold decrease over that of wt BFDC but still much higher than that of the Y288A/A306F variant that remained a tetramer. The Y288A and A306F single variants retained nearly wild-type activity (Table 2), so it is not unreasonable to assume that the double mutant exhibiting the greatest decrease in activity would be the one most likely to show alterations in its oligomerization state. Clearly, that is not the case here, and the reasons for the reduced activity in this variant will become the focus of a future investigation.

To explore this further, the R141E/A306F variant was analyzed, in turn, by analytical centrifugation (Figure S1 of the Supporting Information) and SEC-MALS (Figure S2 of the Supporting Information). The calculated S value for the double mutant was 7.3, significantly lower than that of wt BFDC. This number was obtained at all protein concentrations (Table S3 of the Supporting Information). It was also apparent from the SEC-MALS data that the R141E/A306F variant showed concentration-dependent changes in retention time, which are reflected in its molecular mass (Table S4 of the Supporting Information).

Figure 5 provides a ready comparison of the AUC and SEC-MALS data for wt BFDC and the two variants, R141E/A306F and R141E/Y288A/A306F. The continuous size distribution analysis data, obtained at a protein concentration of 0.5 mg/mL (8.6 μM), highlight the trend toward lower S values, indicative of a smaller species. In addition, at 0.75 mg/mL, both variants have a clear increase in retention time over that of wt BFDC (Figure 5B), confirming a movement toward dimer formation.

Figure 5

(A) Distribution of normalized c(s) values for wt BFDC and the R141E/A306F and R141E/Y288A/A306F variants. In each case, the protein concentration was 0.5 mg/mL (8.6 μM). At that concentration, the S values are 8.4, 7.9, and 7.3 for the wt, triple mutant, and double mutant, respectively. (B) SEC-MALS analysis of wt BFDC and the two variants (0.75 mg/mL) shows the change in the dimer–tetramer equilibrium for the latter.

Equilibrium Denaturation Measurements

The urea-mediated unfolding of wt BFDC and the R141E/A306F variant was followed by both intrinsic fluorescence and CD spectroscopy. To highlight any differences, the unfolding experiments were conducted at an enzyme concentration of 0.15 mg/mL. At this concentration, the wt enzyme is tetrameric whereas the R141E/A306F variant is expected to be primarily dimeric. As shown in Figure 6A, when monitored using the intrinsic fluorescence signal at 327 nm, the unfolding of wt BFDC appears to have three transitions. Initially, between 0 and 0.5 M urea, there is a small increase in fluorescence. Another small transition is observed on going from 0.5 to 2 M urea, while the largest transition takes place between 2 and 6 M urea. Potentially, the first corresponds to the dissociation of tetramers to dimers; the second is that of the dimers to monomers, and the third results from the unfolding of the monomers. Such a pattern was also observed for ScPDC.[25] By contrast, the unfolding of the R141E/A306F variant showed only two transitions (Figure 6A), most likely due to the initial unfolding of the dimer followed by unfolding of the monomer.

Figure 6

Equilibrium unfolding of BFDC by urea. (A) Relative changes in intrinsic fluorescence at 327 nm (λex at 280 nm) for wt BFDC (●) and the R141E/A306F variant (◇). (B) Relative changes in ellipticity at 222 nm (●) and enzymatic activity (○) for wt BFDC. All changes were monitored at 20 °C. The results shown are the average of at least two independent measurements.

Similar results were observed for wt BFDC when unfolding was monitored by far UV-CD (Figure 6B). Further, the urea-induced inactivation of wt BFDC is complete by 3.5 M urea (Figure 6B), and it is evident that some degree of inactivation of the wt enzyme has occurred even at urea concentrations as low as 0.25 M. This decrease in the activity of wt BFDC is presumably caused by the dissociation of the tetramer. This result is in stark contrast to that found in the same experiment conducted with ScPDC. In that case, the enzyme was fully active at 1 M urea, and dimers and tetramers were equally active.[25] The unfolding of the R141E/A306F variant, monitored by CD, followed a pattern almost identical to that observed in the fluorescence experiments (Figure S3 of the Supporting Information). Unfortunately, because of low levels of activity, it was not possible to follow the inactivation of the R141E/A306F variant. Nonetheless, the experiments did demonstrate clearly that there were distinct differences in the urea-mediated denaturation of wt BFDC and the R141E/A306F variant, providing additional evidence that the equilibrium has shifted toward dimer formation in the latter.

X-ray Structures of the R141E and A306F Variants

Attempts to crystallize the double and triple variants were unsuccessful. However, the X-ray structures of the R141E (PDB entry 4MPR) and A306F (PDB entry 4MQ5) variants were determined to resolutions of 1.40 and 1.50 Å, respectively (Table 3). Both exhibited the same fold as the wt enzyme; their space group and cell dimensions were also identical, and as previously observed with other BFDC structures, residues 461–468 had the highest B factors of any region of the protein.[18,42] The active sites were also unchanged, and given that the kinetic parameters for these variants were similar to those of wt BFDC (Table 2), this result was not entirely unexpected. However, the mutations did bring about some unanticipated structural changes that are worthy of comment.

Table 3

Data, Model, and Crystallographic Statistics for the Structures of R141E and A306F BFDC

	R141E BFDCa	A306F BFDCb
Data Collectionc
beamline	APS, GM/CA-CAT, 23-ID-D	APS, GM/CA-CAT, 23-ID-B
wavelength (Å)	1.03	1.03
space group	I222	I222
cell constants	a = 80.98 Å	a = 81.63 Å
	b = 95.97 Å	b = 95.54 Å
	c = 137.3 Å	c = 137.6 Å
	α = β = γ = 90°	α = β = γ = 90°
no. of unique reflections	104731	85254
resolution limit (Å)	1.40 (1.40–1.45)	1.50 (1.50–1.53)
completeness (%)	99.8 (99.8)	99.4 (100)
redundancy	7.2 (7.1)	5.2 (5.0)
I/σI	27 (2.9)	46 (8.7)
Rmerge (%)	8.5 (78)	5.0 (24)
Refinement
resolution range (Å)	1.40–48.0	1.50–26.4
Rfree test set size	1997	2000
Rcryst (%)	12.45	11.96
Rfree (%)	14.96	14.72
no. of atoms
total	4738	4730
protein	4116	4108
water	591	591
overall B factor	12.15	12.78
rmsd
bond lengths (Å)	0.015	0.009
bond angles (deg)	1.563	1.289

PDB entry 4MPR.

PDB entry 4MQ5.

Value in parentheses are for the highest-resolution shell.

It was predicted that mutating Arg141 to a glutamate would result in a variant that not only would be unable to form a salt bridge with Glu107 but also would create like-charge repulsion between Glu107 and Glu141. Surprisingly, the crystal structure of the R141E variant revealed the carboxylate group of Glu141 is complexed to a Na+ ion, as well as the indole moiety of Trp125 (Figure 7). Four water molecules were also coordinated to the sodium ion, as was the carbonyl of Gly105. One of these waters was located within 3 Å of the carboxylate of Glu107. It seems that these unexpected interactions were sufficient to alleviate the predicted charge–charge repulsion that was expected to tilt the equilibrium toward the dimeric state.

Figure 7

Potential Glu141–Glu107 interaction avoided in the R141E variant. Instead, Glu141 is found coordinated to Trp125 and an additional Na+ ion. Blue dashes indicate distances of ≤3 Å.

Perhaps the most surprising observation was that the A306F substitution resulted in the displacement of the side chain of Arg141. This meant that the guanidinium group of arginine forms an intramonomer salt bridge with the side chain of Glu134 rather than the intermonomer interaction with Glu107 (Figure 8). The net result was that the additional bulk of the new phenylalanine residue was readily accommodated.

Figure 8

A306F variant provided an unexpected rearrangement in the A–D interface. (A) wt BFDC showing the original positions of Met145 and Arg141. (B) Phe306 causes Met145 and Arg141 to adopt new rotamers, the latter stabilized by interaction with Glu134. Blue dashes indicate a distance of ∼3 Å. Red dashes indicate distances of ≤4 Å.

Discussion

In his structure-based analysis of the domain relationships inherent to the ThDP-dependent enzymes, Duggleby[12] proposed that the original enzyme was a protein that contained binding sites for both ends of the ThDP molecule. Dimerization provided an advance by which the intrinsic activity of the cofactor was enhanced. Duplication of the gene for the primitive decarboxylase resulted in an α2β2 heterotetramer, likely a still more efficient enzyme. Subsequently, gene fusion led to the combination of the PP and PYR domains, giving rise to, effectively, an (αβ)2 protein that can be seen to this day in phosphonopyruvate decarboxylase.[12,43] The more modern ThDP-dependent decarboxylases at some point acquired a third domain, the TH3 domain, which fused between the PYR and PP domains.[12] A broadly similar evolutionary pattern has also been postulated on the basis of sequence analysis, with the added suggestion that recruitment of the TH3 domain was accompanied by tetramer formation.[13]

What was not considered in any of these discussions is whether tetramer formation is essential or, if not, what advantage accrues due to tetramer formation. Allosteric regulation may have been a possible candidate, particularly as KdcA is a dimer and exhibits Michaelis–Menten kinetics whereas ScPDC is subject to allosteric activation by its substrate.[44] However, this explanation is made less likely by the X-ray structures of ScPDC in its unliganded[45] and substrate-activated forms,[46] which both show the enzyme to be tetrameric. Further, studies by Killenberg-Jabs et al.[24,25] have found that ScPDC exists as an equilibrium mixture of dimers and tetramers under physiological conditions, and that there was no difference in the steady-state activity between the two states. Altogether, it would appear that the ability to form a tetramer is not a prerequisite for allosteric activation.

In contrast to ScPDC, BFDC is tetrameric and obeys Michaelis–Menten kinetics. In many ways, it is more comparable to the PDC from Zymomonas mobiliz (ZmPDC), which has similar properties.[17] The similarity was confirmed by SPPIDER analysis, which showed that an individual monomer in both BFDC and ZmPDC makes significant contact with the other three monomers in the tetramer (Table 1). This is quite different from the cases of KdcA and ScPDC, which show contacts only between the two monomers comprising the catalytic dimer.

Attempts to alter the dimer–tetramer equilibrium of BFDC by disrupting the noncatalytic monomer–monomer interfaces with point mutations were unsuccessful. Initially, it took the introduction of three mutations before any change was observed by SEC. This shift in equilibrium was confirmed by analytical ultracentrifugation. The R141E/Y288A/A306F variant showed no activity in the standard assay or in those containing 500 μg/mL enzyme. Subsequently, it was shown that two mutations, R141E and A306F, were sufficient to alter the dimer–tetramer equilibrium. At an assay concentration of 40 μg/mL, this variant showed a considerable (∼450-fold) reduction in its kcat from that of wt BFDC, but it was impossible to say whether this was a reflection of the presence of tetramer in solution or the dimer indeed retained some activity. To investigate the latter possibility, we turned to equilibrium denaturation experiments.

It has been shown that, in a 0.5 M urea solution, ScPDC forms dimers, and those dimers have the same specific activity as the tetramers.[24] Here we found that BFDC has a urea unfolding curve similar to that of ScPDC, regardless of whether unfolding was measured by intrinsic fluorescence or by CD. However, unlike ScPDC, BFDC showed a rapid loss of activity, even at urea concentrations as low as 0.25 M. Taken together, our data suggest that, to be catalytically active, it is necessary for BFDC to be a tetramer.

Although that may be a reasonable conclusion, it is still not clear why the tetramer is necessary. Although all the single variants were very active, indicating that they were still tetrameric, the X-ray structures of both R141E and A306F provided some clues, with the most insightful piece of information coming from the crystal structure of the A306F variant. It was predicted that replacing Ala306 with phenylalanine would result in a steric clash that would help destabilize the tetramer. Surprisingly, the A306F structure revealed a rearrangement of the shallow pocket, by which the additional steric bulk of the phenylalanine was accommodated (Figure 8). This was due to the displacement of the side chain of Arg141 and Met145, which resulted in a larger binding pocket. The new rotamer of Arg141 saw the loss of the intermonomer interaction with Glu107 and the formation of a new intramonomer bond with Glu134 (Figure 8). Extrapolating to the R141E/A306F variant, we could assume that rather than Glu141 forming a salt bridge with Glu134, the interaction would be unfavorable. Potentially, this could cause Glu141 to spring back into its “original” position, making the accommodation of Phe306 more difficult. In the R141E structure, the enzyme used an additional sodium ion to help adapt to the potential Glu107–Glu141 contact (Figure 7). This may not be feasible in the R141E/A306F variant. The net result would be that the A–C and A–D interfaces are both destabilized, pushing the equilibrium toward the dimer as suggested by the SEC and equilibrium denaturation experiments.

Again, this is speculation, and it does not provide any insight into the apparent necessity for BFDC to be tetrameric to be catalytically active. Still, it is possible to take the speculation a little further by looking more closely at structural differences among BFDC, KdcA, and PDC. The three all belong to the decarboxylase-like (DC) structural clade of ThDP-dependent enzymes.[12] The vast majority of enzymes of the DC clade possess an inner active site loop that contains residues known to be critical to their catalytic mechanism. For example, the loops of ScPDC, 3-indole-pyruvate decarboxylase, phenylpyruvate decarboxylase, and KdcA all include two adjoining histidine residues sometimes termed the HH motif.[47] Mutagenesis of either histidine results in a decrease in catalytic activity of orders of magnitude.[48−50] Conversely, although the active site of BFDC also contains two catalytically important histidine residues, these are not located on the inner loop. Instead, the loop of BFDC houses two contiguous leucine residues, Leu109 and Leu110. Alignment of the sequences of enzymes annotated as BFDCs in the thiamin enzyme engineering database (TEED) shows that Leu110 is completely conserved, and that leucine, or another hydrophobic residue, is invariably found at position 109.[51]

Examination of the X-ray structure of BFDC in complex with benzoyl phosphonic acid methyl ester (MBP), an analogue of benzoylformate, shows that Leu109 lines the phenyl-binding pocket of BFDC and that Leu110 has five points of contact within 5 Å of the glyoxylate analogue portion of MBP (Figure 9). From this structure, it appears that Leu110 acts as a clamp to lock the glyoxylate moiety into the perpendicular arrangement of the carboxylate group to the thiazolium–C2α bond. This geometry is thought to promote decarboxylation by allowing the maximal overlap of the π electrons of the thiazolium ring and the p orbital of the scissile bond.[52,53] Intriguingly, the active site loop of BFDC is bookended by Glu107 and Asp114, which, of course, interact with Arg141 and Arg120, respectively, in the A–C interface (Figure 4). It is not unreasonable to imagine that these two salt bridges may be responsible for maintaining the correct positioning of Leu109 and Leu110 within the active site of BFDC, thereby contributing to both the overall hydrophobicity of the active site and the correct position of the substrate within it.

Figure 9

Leu109 and Leu110 are positioned by the interaction of Arg141 and Arg120 of monomer A with Glu107 and Asp114 of monomer C, respectively. Blue dashes indicate distances of ∼3 Å. Red dashes indicate distances of ≤5 Å.

To assess the importance of the two leucine residues, both Leu109 and Leu110 were replaced, in turn, with alanine. Both variants showed decreases in kcat of 2 orders of magnitude, while L110A also exhibited a 9-fold increase in Km (Table 2). Overall, the changes were broadly in line with those observed with the R141E/A306F variant. It would appear that the decrease in activity observed with the latter may be due to movement in the inner loop resulting in the misalignment of Leu109 and Leu110 within the active site. To test this assumption, the E107R/D114R double variant was generated to completely disrupt the A–C interface, simultaneously preventing the correct positioning of both Leu109 and Leu110. While the individual substitutions produced only a modest effect on catalysis, the catalytic efficiency of the E107R/D114R variant was ∼600-fold lower than that of wt BFDC (Table 2). It is notable that the kcat and Km values for this variant were very similar to those of R141E/A306F, yet SEC data suggested that the E107R/D114R variant remained a tetramer. On that basis, it may be that the integrity of the dimer interface, rather than the tetrameric structure per se, is of paramount importance.

Conclusion

We have demonstrated that, in some cases, only two point mutations are required to alter the dimer–tetramer equilibrium of BFDC. In addition, it is apparent that the shift in equilibrium toward dimer formation is accompanied by a loss of catalytic activity. Finally, we provide evidence that suggests the tetrameric structure of BFDC has evolved as a means of positioning and stabilizing the active site loop residues, Leu109 and Leu110.