The Two Active Sites in Human Branched-chain α-Keto Acid Dehydrogenase Operate Independently without an Obligatory Alternating-site Mechanism*

A long standing controversy is whether an alternating activesite mechanism occurs during catalysis in thiamine diphosphate (ThDP)-dependent enzymes. We address this question by investigating the ThDP-dependent decarboxylase/dehydrogenase (E1b) component of the mitochondrial branched-chain α-keto acid dehydrogenase complex (BCKDC). Our crystal structure reveals that conformations of the two active sites in the human E1b heterotetramer harboring the reaction intermediate are identical. Acidic residues in the core of the E1b heterotetramer, which align with the proton-wire residues proposed to participate in active-site communication in the related pyruvate dehydrogenase from Bacillus stearothermophilus, are mutated. Enzyme kinetic data show that, except in a few cases because of protein misfolding, these alterations are largely without effect on overall activity of BCKDC, ruling out the requirement of a proton-relay mechanism in E1b. BCKDC overall activity is nullified at 50% phosphorylation of E1b, but it is restored to nearly half of the pre-phosphorylation level after dissociation and reconstitution of BCKDC with the same phosphorylated E1b. The results suggest that the abolition of overall activity likely results from the specific geometry of the half-phosphorylated E1b in the BCKDC assembly and not due to a disruption of the alternating active-site mechanism. Finally, we show that a mutant E1b containing only one functional active site exhibits half of the wild-type BCKDC activity, which directly argues against the obligatory communication between active sites. The above results provide evidence that the two active sites in the E1b heterotetramer operate independently during the ThDP-dependent decarboxylation reaction.

MDa in size, are organized around a transacylase (E2) core, to which multiple copies of a decarboxylase (E1), a dehydrogenase (E3), and in the case of mammalian species a specific kinase and specific phosphatase are attached through noncovalent interactions. The lipoyl-bearing domain (LBD) of the E2 core imparts substrate channeling by sequentially visiting different active sites in each of the three E1, E2, and E3 catalytic components (14), leading to the oxidative decarboxylation of ␣-keto acids (Reaction 1). The E1 components of ␣-keto acid dehydrogenase complexes are ThDP-dependent enzymes that catalyze both the ThDP-dependent decarboxylation of ␣-keto acids to produce the ␣-carbanion-ThDP intermediate (Reaction 2) and the reductive acylation of oxidized lipoylated LBD (lip-LBD-S 2 ) on E2, forming S-acyldihydrolipoamide (Reaction 3). REACTION 3 The E1p component of the mammalian and Gram-positive bacterial PDC belongs to the ␣ 2 ␤ 2 -type of ThDP-dependent enzymes with two active sites; each of these active sites catalyzes both Reaction 2 and Reaction 3. The half-of-the-sites reactivity or the alternating active-site mechanism was suggested to function in the E1p component of the pig heart PDC (15). This model was prompted by the observation that the phosphorylation of the complex corresponding to a molar ratio of one phosphate group per two ␣ chains, i.e. one phosphoryl group per ␣ 2 ␤ 2 heterotetramer, results in the complete inactivation of the PDC. Likewise, phosphorylation of half of the available sites in the ␣ 2 ␤ 2 heterotetramer of the related E1b component inactivates reconstituted BCKDC (16). These results are consistent with the thesis that the alternating activesite mechanism occurs in the E1p or E1b component. Thus, the inactivation of one of the two active sites disrupts this mode of action, resulting in the inactivation of the mammalian PDC or BCKDC. A model for the communication between the two active sites of the E1p component of the Bacillus stearothermophilus PDC has been described, in which a proton is reversibly shuttling through an acidic channel in the protein (17). The authors suggested that this proton wire allows the cofactors to function as a general acid/base catalyst to switch the conformation of the active-site loops. It is further hypothesized that this proton-relay mechanism may extend to other dimeric ThDPdependent enzymes, in particular, those requiring an activated cofactor for catalysis. Chemical nonequivalence in the two active sites of the E1p component of human PDC has recently been demonstrated in the forms of ThDP C2 ionization rate, the ability to bind a substrate analog, and the decarboxylation rate constant of the 2-lactyl-ThDP intermediate (18). These results are consistent with the half-of-the-sites reactivity but shed no light on whether the proton-wire mechanism described for the B. stearothermophilus E1p (17) is responsible for the observed chemical nonequivalence between the two active sites in human E1p.

E2-[R-CO-S-lipLBD-SH] ϩ E1-ThDP
We have previously described crystal structures of wild-type and mutant human E1b in the absence and presence of decarboxylation reaction intermediates (16, 19 -22). For these structure determinations, we used a crystal form that has an ␣␤Ј heterodimer in the asymmetric unit and exhibited the symmetry of space group P3 1 21. The full heterotetramer is generated by a crystallographic 2-fold symmetry operator. This symmetry persists to the observed high resolution limit (1.4 Å) of these crystals. As a result of this packing arrangement, the two halftetramers must be structurally identical. The electron densities showed that both active sites are occupied with intermediates. We did not detect any structural asymmetry involving the active sites or surrounding regions, either for human E1b in the absence of substrate or when intermediates were bound. Likewise, the crystal structure of the E1b component from Thermus thermophilus BCKDC showed similar occupancy of the ␣-carbanion-ThDP intermediate in both active sites (23). Therefore, these crystallographic properties of both human and Thermus E1bs are inconsistent with the half-of-the-sites model for these ThDP-dependent enzymes.
To investigate whether the alternating active-site mechanism occurs in the ThDP-dependent E1b component of human BCKDC, we carried out crystallographic and biochemical studies with both wild-type and mutant E1b heterotetramers. We provide evidence that the two active sites in human E1b, unlike B. stearothermophilus and human E1p, can operate independently without the obligatory communication between active sites. Moreover, we find that the completely inactive BCKDC, harboring 50% phosphorylated E1b, regains a significant amount of overall activity when the covalently modified E1b is dissociated and re-associated with E2b. We suggest that the inactive BCKDC observed with the phosphorylated E1b results from the specific geometry associated with the BCKDC assembly.

EXPERIMENTAL PROCEDURES
Protein Expression and Purification-N-terminally and C-terminally His 6 -tagged wild-type and mutant human E1b proteins were generated as described previously (20). The mutations were introduced using the QuikChange site-directed mutagenesis kit provided by Stratagene (La Jolla, CA). Expression of C-terminally His 6 -tagged LBD (residues 1-84 of the human E2b subunit) and in vitro lipoylation was carried out also as described previously (24). To produce an E1b-hybrid protein with only one functional active site, two plasmids were constructed. The pHisT hE1b plasmid containing the N-terminally His 6 -tagged S292Q␣ subunit and the nontagged wild-type ␤ subunit coding regions were as described previously (16). The pStrep hE1b plasmid was derived from the pACYCDuet-1 plasmid (Novagen, San Diego) with the mature E1b␣ and E1b␤ coding sequences, both of the wild type, inserted into its two multiple cloning sites. The Strep tag contained a short sequence of eight amino acid residues (WSHPQFEK) that allowed binding to Strep-Tactin resin (IBA GmbH, Göttingen, Germany). The pHisT hE1b and pStrep hE1b plasmids were co-transformed into BL-21 cells; the cells were selected against both ampicillin and chloramphenicol. The hybrid E1b protein (S292Q␣/WT-␣)␤ 2 , consisting of one His 6 -tagged mutant ␣ subunit and one Strep-tagged wild-type subunit, was purified from the lysate of co-transformed BL-21 cells by consecutive use of the HisTrap HP column (GE Healthcare) connected to an FPLC system and the Strep-Tactin Superflow gravity column. Elution conditions were chosen based on control experiments with His 6 -tagged E1b (with both ␣ subunits carrying the His 6 tag) on a Strep-Tactin column, and with Strep-tagged E1b (with both ␣ subunits bearing the Strep tag) on a HisTrap column to avoid possible contamination by the carryover of single tagged species.
X-ray Crystallography of Mutant Human E1b Proteins-Crystallization of the C-terminally His 6 -tagged human S302P␣ E1b protein was carried out as described previously (22,25). The crystals exhibited the symmetry of space group P3 2 21 with cell parameters of approximately a ϭ b ϭ 145 Å and c ϭ 138 Å, contained an entire heterotetramer per asymmetric unit, and diffracted x-rays to a minimum Bragg spacing, d min , of 1.88 Å. This crystal form is related to a previously published crystal form that exhibited the symmetry of space group P3 1 21 with cell parameters of approximately a ϭ b ϭ 145 Å and c ϭ 69 Å and that contained an ␣/␤Ј heterodimer per asymmetric unit (22). For x-ray diffraction experiments, crystals were soaked with Mn-ThDP and KIV and flash-cooled in liquid propane and kept at about 100 K during data collection at beamlines 19ID and 19BM of the Advanced Photon Source (APS), Argonne National Laboratory, Argonne, IL. Data were processed with the HKL2000 package (26).
The crystal structure was determined by molecular replacement using the program Phaser (27) and the hE1b wild-type structure (Protein Data Bank code 1ols) as the search model. Refinement was carried out with the program Refmac5 (28) of the CCP4 package (29) with a random subset of all data set aside for the calculation of free R factors. No noncrystallographic symmetry restraints or constraints were used during refinement. Manual adjustments to the models were carried out with the program Coot (30).
The electron density clearly showed the presence of 2-(␣hydroxyisobutyl)-ThDP (VThDP), the decarboxylation intermediate from ␣-ketoisovalerate (KIV) in both active sites of the human E1b heterotetramer. A model for this intermediate was included only at the very end of the modeling/refinement process. After the refinement of the protein portions was complete, solvent molecules were added where chemically reasonable. Ramachandran analysis shows that all but 2 out of 1,439 residues in the model fall into the allowed regions. Tyr-113␣ in both copies in the asymmetric unit is in the disallowed region. The electron density for these residues is well defined. Tyr-113␣ adopts a unique conformation (referred to as the P-conformation) after a decarboxylation reaction intermediate has been formed in the active sites as described previously (22). Data collection and refinement statistics are listed in Table 1.
Enzyme Assays-The overall activity of BCKDC (Reaction 1) reconstituted with E1b, E2b, and E3 was assayed with KIV, CoA, and NAD ϩ as substrates and ThDP and MgCl 2 as cofactors; the progress of the reaction was followed spectrophotometrically by an increase in absorbance at 340 nm. The activity of E1b-mediated decarboxylation (Reaction 2) was assayed with KIV as a substrate and 2,6-dichlorophenolindophenol (DCPIP) as an artificial electron acceptor; the reaction was monitored by the decrease in absorbance at 600 nm. The E1b-catalyzed reductive acylation reaction (Reaction 3) was measured using [U-14 C]KIV and oxidized lip-LBD as substrates. The methods for the above three activity measurements have been described previously (20).
Differential Scanning Calorimetry (DSC)-All E1b proteins were dialyzed against 50 mM potassium phosphate buffer, pH 7.5, containing 250 mM KCl, 5% (v/v) glycerol, 1 mM ␤-mercaptoethanol, 0.2 mM EDTA, followed by DSC runs in a VP-capillary DSC system (MicroCal). Each E1b has a final concentration of 5.8 M, which was used to normalize the scans. Final traces were produced after subtraction of the reference scan and subsequent base-line corrections to determine the melting temperature (T m ) using the MicroCal ORIGIN 7 software package.
Phosphorylation of E1b and Dissociation of E1b from E2b after Phosphorylation-Phosphorylation of S302A␣ E1b with the BCKD kinase was carried out in the presence of E2b as described previously (20). Aliquots were taken from the reaction mixture at different time points and mixed with 0.5 mM EDTA and the SDS sample loading buffer, followed by analysis by SDS-PAGE. The amount of 32 P-phosphoryl group incorporated into the E1b␣ subunit was determined by Phosphor-Imaging. The calculation was done by comparing the individual PhosphorImaging bands to a band from standard spotting. To resolve E1b, E2b, and the BCKD kinase proteins after phosphorylation, NaCl was added to the reaction mixture to a final concentration of 1.0 M. After incubation overnight at 4°C, the mixture was concentrated, and E1b was separated from E2b and the BCKD kinase by size exclusion chromatography using a Superdex 200 (HR 10/30) column.
Western Blotting-Column fractions containing His 6 -tagged E1b, Strep-tagged E1b, and His 6 /Strep-tagged hybrid E1b proteins were separated by SDS-PAGE in 12% gels; protein bands were transferred to the Immobilon P membrane (Millipore, Bellerica, MA). The protein bands were probed following the manufacturers' protocol, with either the horseradish peroxidase-conjugated His 6 tag probe supplied by Pierce or the horseradish peroxidase-conjugated Strep probe (Strep-Tactin) provided by IBA GmbH (Göttingen, Germany).

The Presence of Two Identical Active-site Conformations in
Human E1b-We have determined the crystal structure of human E1b in the presence of 2-(␣-hydroxyisobutyl)-ThDP (VThDP), the decarboxylation intermediate from the substrate ␣-ketoisovalerate (KIV), in a novel crystal form that contains an entire human E1b heterotetramer in the asymmetric unit ( Fig.  1). This crystal form was obtained with a human E1b variant (S302P␣) harboring a substitution at phosphorylation site 2 (Ser-302␣). The mutation appears to lead to a small structural change in the phosphorylation loop, which alters the intermolecular interactions in this region and results in a new packing arrangement in the crystals, which is different from that reported previously for human E1b (16, 19 -22). The new crystal form exhibits the symmetry of space group P3 2 21 and shares the same a and b unit cell lengths with the previously obtained counterparts (22), but its c unit cell length is twice as long. Data collection and refinement statistics are shown in Table 1. Noncrystallographic symmetry was not imposed during the refinement. The results show no structural asymmetry between the ␣␤Ј and the ␣Ј␤ heterodimers; the root-mean-square deviation for all compared 711 C-␣ atoms is 0.17 Å. The two active sites and their neighboring regions are structurally virtually identical. The B factor for the cabanion-ThDP intermediate in both active sites is 11.8 Å 2 , compared with the average of 16.8 Å 2 . When this heterotetramer is compared with the one generated from the previously reported heterodimeric structure (Protein Data Bank code 2beu) based on crystallographic 2-fold symmetry, the resulting root-mean-square deviation for all 1422 compared C-␣ carbon atoms is 0.24 Å. Thus, the structures of human E1b in the presence of the intermediate VThDP are virtually identical, irrespective of whether an ␣␤Ј heterodimer (22) or an entire heterotetramer (the present study) is in the asymmetric unit. Taken together, there is no crystallographic evidence for a structural asymmetry between the two active sites in human E1b.
Site-directed Mutagenesis of Candidate Proton-wire Residues in E1b- Fig. 2A shows the alignment of the potential protonwire residues among sequences from human E1b, Pseudomonas putida E1b, human E1p, and B. stearothermophilus E1p. In human E1b, these residues include Asp-200␣, Glu-46␤, Glu-76␤, and Asp-108␤. These residues are located in a channel between the two symmetrical E1b active sites, each showing good density for the occupied VThDP intermediate (Fig. 2B). Glu-198␣ is a candidate proton-wire residue in human E1b, but the equivalent Gln-178␣ in B. stearothermophilus E1p has not been implicated as a proton-wire residue (17). Ala-203␣ in human E1b aligns with the proton-wire residue Glu-183␣ of B. stearothermophilus E1p, but it cannot assume the putative function because of the absence of negative charges. Moreover, there is no magnesium ion present in the center of the channel, unlike in B. stearothermophilus E1p. There is also no residue equivalent to this magnesium ion that could fulfill its role. Instead, there is a pair of Gln-77␤ residues whose side chains are hydrogen-bonded to each other (Fig. 2B). The 2-fold symmetry axis relating the ␣␤Ј and ␣Ј␤ heterodimers in human E1b is located exactly between these two residues. Because of the presence of these nonacidic residues, it is difficult to envision how the shuttling of a proton from one active site to another should occur in human E1b.
The putative role of active-site communication involving candidate proton-wire residues in human E1b was investigated. These residues were altered by site-directed mutagenesis, and enzyme kinetic parameters of these variants were determined. The overall activity of BCKDC was assayed by the reconstitution of the wild-type or mutant E1b protein with E2b and E3 without DTT. As shown in Fig. 2C, D200N␣, E198Q␣/D200N␣ (a double mutant), and E46Q␤ substitutions reduced overall activities of the mutants to 25, 45, and 65%, respectively, compared with the wild type. D200A␣ and D108N␤ variants show 86 and 116% of wild-type activity, respectively. The K m values of these mutants for substrate KIV are similar to that of the wild type (56.5 Ϯ 3.6 M), except for the K m value of the D108N␤ variant (127 Ϯ 15 M), which is twice as high as that of the wild type. The D200A␣/D108N␤ double mutant exhibits 6.8% of residual wild-type overall activity. The rate for E1b-catalyzed decarboxylation of KIV was also measured in the presence of the electron acceptor DCPIP. The rates of E1b-catalyzed decarboxylation measured by decreased absorbance at 600 nm because DCPIP reduction are identical to those determined radiochemically by 14 CO 2 evolution using [U-14 C]KIV as a substrate in the presence of DCPIP (20). As shown in Fig. 3C, rates of E1b-catalyzed decarboxylation are also decreased in the D200N␣, E198Q␣/D200N␣, and E46Q␤ mutants. The D200A␣ and D108N␤ variants show DCPIP-dependent E1b activity similar to wild-type E1b. The D200A␣/D108N␤ double mutant exhibits only trace E1b-catalyzed decarboxylation activity, in concordance with its very low residual overall activity. The proton-wire candidate residue Glu-76␤ is the invariant catalytic acid-base for human E1b (3,25). The substitution of  this residue with an alanine thwarts a proton abstraction at the C-2 atom of ThDP, thereby abolishing E1b activity (25).

The D200A␣/D108N␤ Mutant Is Defective in Folding and
Assembly-In the study by Frank et al. (17), double mutants D180A␣/E183A␣ and D180N␣/E183Q␣ of the B. stearothermophilus E1p exhibit markedly reduced E1p and overall PDC activity. It was suggested that reduced activities were because of a disruption of the proton-wire channel, preventing communication between the two active sites. To decipher the role of Asp-200␣ and Asp-108␤ in the putative proton wiring of E1b, we subject the D200A␣/D108N␤ E1b double mutant to treatment with a naturally occurring osmolyte TMAO. We reported previously that the incubation of E1b proteins carrying certain disease-causing human mutations with TMAO partially correct assembly defects, resulting in the restoration of 30 -50% of BCKDC overall activity (31). Fig. 2D shows that the incubation of the D200A␣/D108N␤ mutant, which has no detectable overall activity, with TMAO leads to the recovery of ϳ50% overall activity compared with the wild type subjected to the same treatment. These results do not support roles of Asp-200␣ and Asp-108␤ in a putative proton-wire channel. Instead, these two residues are likely to be involved in the folding and assembly of the E1b heterotetramer. The moderately reduced overall activity of the remaining D200N␣, E198Q␣/D200N␣, and E46Q␤ E1b variants compared with wild-type E1b is essentially unchanged after TMAO treatment (Fig. 2D). The data indicate that these mutations may not significantly impede the folding and assembly of the E1b heterotetramer. The higher percentage of BCKDC overall activities reconstituted with D200N␣ (65%) and E198Q␣/D200N␣ (63%) mutants in the absence of TMAO in Fig. 2D, compared with those of 25 and 45%, respectively, in Fig. 2C, were because of the addition of DTT in the former set of assays. The results suggest that the reducing equivalence preferentially stabilizes the E1b component carrying these two mutations.
The putative adverse effects of the above mutations on the stability of human E1b were dissected by DSC. Wild-type E1b shows two distinct melting temperatures (T m ) at 52.6 and 65.4°C, which likely represent transitions from heterotetramers to heterodimers and from heterodimers to the random-coil structure, respectively (32). The E46Q␤ E1b mutant shows two T m values at 53.3 and 64.0°C, similar to the wild-type E1b. Notably, both E198Q␣/D200N␣ and D200A␣/D108N␤ double mutants exhibit a single T m value at 55.4 and 57.4°C, respectively. The results confirm that these mutant proteins are defective in folding and assembly and undergo a single transition directly from heterotetramers to the random-coil structure. The remaining E1b mutants, D200A␣, D200N␣, and D108N␤ also show decreased T m values of 53.5, 52.4/59.2, and 54.8°C, respectively (data not shown), indicating that these E1b proteins are also less stable than wild-type E1b.
Overall Activity Is Abolished at 50% E1b Phosphorylation-The human E1b contains two phosphorylation sites (site 1 at Ser-292␣; site 2 at Ser-302␣) on the phosphorylation loop (residues Tyr-286␣ to Pro-315␣) located within each active site channel (20). The rapid phosphorylation of site 1 (the major site) alone in E1b is sufficient to completely inactivate BCKDC. Phosphorylation of site 2 (the minor site) is much slower than FIGURE 3. Overall and reductive acylation activities of 50% phosphorylated human E1b. A, time course for 32 P incorporation and the reduction of overall activity for S302A␣ E1b. The S302A␣ mutant with phosphorylation site 2 inactivated was incubated at 23°C with 0.4 mM Mg-[␥-32 P]ATP, BCKD kinase, and E2b. Aliquots were taken at different times; the ratio of 32 P-phosphate incorporated per E1b tetramer and the overall activity of reconstituted BCKDC were measured. B, dissociation and purification of phosphorylated E1b. The S302A␣ mutant E1b was incubated with 0.4 mM Mg-ATP and BCKD kinase for 20 min. The 50% phosphorylated mutant E1b with no overall activity was dissociated from E2b and BCKD kinase in 1.0 M NaCl, followed by separation on Superdex 200 column. Upper panel, elution profile; lower panel, SDS-PAGE of column fractions. MBP-BCK, maltose-binding protein-fused BCKD kinase. C, activities for the overall reaction (solid bars) and reductive acylation (shaded bars) in the presence of excess exogenous lip-LBD were assayed with nonphosphorylated S302A␣ E1b (lane 1) and 50% phosphorylated S302A␣ E1b in the phosphorylation mixture prior to dissociation and separation from E2b and BCKD kinase (lane 2). Both activities were also assayed with 50% phosphorylated S302A␣ E1b in column fractions 24 -27 from B upon reconstitution with fresh batches of E2b and E3 (lane 3). Activities of 100% phosphorylated S302A␣ E1b after column separation and reconstitution were also measured (lane 4). Relative activity is expressed as percentage of that with the starting nonphosphorylated E1b. site 1 and is without effect on BCKDC activity (16,33,34). To dissect the effect of phosphorylation on the hypothetical communication between the two active sites, the S302A␣ E1b variant with phosphorylation site 2 abolished was phosphorylated by the BCKD kinase in the presence of E2b (33). Fig. 3A shows that phosphorylation of S302A␣ E1b at site 1 reaches 50% within 10 min, with BCKDC activity largely abolished (Ͻ5% initial activity). The second half of phosphorylation at site 1 is sluggish, with close to 100% phosphorylation reached and overall activity completely lost in 16 h.
The overall activity of the pig heart PDC is also lost at 50% phosphorylation of the cognate E1p component (15). Based on this earlier study, it was proposed that phosphorylation at one of the two active sites abrogates communication with the other site in E1p, resulting in the complete inactivation of PDC activity. To entertain this hypothesis, we dissociated E1b of the human BCKDC at the 50% phosphorylation level from the E2b core by incubation in 1 M sodium chloride at pH 7.5, followed by separation by gel filtration (Fig. 3B, upper panel). The separated 50% phosphorylated E1b (Fig. 3B, lower panel) was assayed for overall activity by reconstitution with a fresh batch of E2b and E3. Surprisingly, the BCKDC reconstituted with the 50% phosphorylated E1b shows about 40% of wild-type overall activity (Fig. 3C). The result is consistent with the notion that the 50% phosphorylated E1b still possesses half of wild-type E1b activity imparted by the remaining nonphosphorylated active site. The loss of BCKDC activity at 50% E1b phosphorylation prior to dissociation and re-assembly may result from the geometric constraints imposed on E1b and LBD, both tethered to the E2b scaffold, which prevent the LBD substrate from gaining access to the nonphosphorylated functional E1b active site (see "Discussion").
A Hybrid E1b with One Site Inactivated Exhibits Nearly 40% Activity-To dissect the requirement for the communication between the two active sites, a hybrid E1b was produced. To accomplish this, we generated two plasmids as follows: one plasmid contained the His 6 -tagged ␣ subunit sequence harboring the S292Q␣ mutation that inactivates BCKDC activity (16) and the nontagged wild-type ␤ subunit sequence; the other plasmid consisted of the Strep-tagged wild-type ␣ subunit sequence and the nontagged wild-type ␤ subunit sequence. Both plasmids were co-transformed into the same Escherichia coli BL21 cells and selected simultaneously against appropriate antibiotics. The hybrid E1b heterotetramer carrying one wild type, one S292Q␣ ␣ subunit, and two wild-type ␤ subunits was purified sequentially using nickel-affinity (Fig. 4A) and Strepaffinity (Fig. 4B) chromatography. The presence of the His 6 tag and Strep tag in the purified hybrid E1b was indicated by Western blotting using the respective probes (Fig. 4C). E1b-catalyzed decarboxylation rates for wild-type and hybrid E1b were measured in the presence of DCPIP. The hybrid E1b shows 40% of wild-type E1b activity (Table 2). Assays for reductive acylation show a k cat value of 4.0 s Ϫ1 for the hybrid E1b, accounting for 43% of wild-type activity. The K m values for lip-LBD are similar between wild-type and the hybrid E1b. When assayed for overall activity in the presence of E2b and E3, the hybrid E1b shows a k cat of 73.9 min Ϫ1 , which is nearly 40% of that reconstituted with wild-type E1b ( Table 2). The K m value for KIV of the hybrid E1b (118 M) is twice that of the wild-type E1b (56.5 M). The S292Q␣ E1b mutant, which shows no activity for E1b-catalyzed decarboxylation, reductive acylation, and the overall reaction reconstituted with E2b and E3 (16), served as a negative control. Significantly, the catalytic parameters of this hybrid, despite through a different mechanism of inactivation, are similar to the catalytic parameters of 50% phosphorylated E1b. The above results provide direct evidence that one active site alone can function in the hybrid E1b.

DISCUSSION
The question whether ThDP-dependent enzymes employ an alternating active-site mechanism during catalysis has been of intense interest and is controversial in the literature. This work was prompted by the recent study with B. stearothermophilus E1p, which suggests the presence of proton wires to mediate communication between the two active sites (17). We undertook multiple reinforcing approaches to decipher whether such active-site communication is also required for ThDP-dependent decarboxylation catalyzed by human E1b. At present, it is not known whether the presence of two independently operated active sites in human E1b derived from our studies can be generalized to other ThDP-dependent enzymes. Nonetheless, the recent NMR study by Seifert et al. (18) has shown no evidence for chemical nonequivalence in several related thiamine enzymes, including yeast transketolase, Lactobacillus plantarum pyruvate oxidase, and Zymomonas mobilis pyruvate decarboxylase. The proposed "slinky cycle" mediated by the proton-relay channel in B. stearothermophilus E1p, and therefore an inherent asymmetry, is supported by the presence of both the outer and inner loops in only one of the two active sites upon the binding of ThDP to the E1p heterotetramer in the crystallographic asymmetric unit (17). In contrast, in this study, high occupancies of both ThDP molecules (or the reaction intermediate VThDP; Fig. 2B) and the high degree of order for the phosphorylation loop (equivalent to the outer loop in Bacillus E1p) were observed in the two active sites of the E1b of human BCKDC (Fig. 1). Thus, the human E1b structure does not show the two distinct active-site conformations depicted in the Bacillus E1p structure. The holo-E1b component studied here exists as a single heterotetramer in the asymmetric unit. In our earlier studies, the holo-E1b component with or without the carbanion-ThDP intermediate is present as heterodimers in the asymmetric unit (16, 19 -22). However, the active-site conformation in the heterotetrameric E1b structure built on crystallographic 2-fold symmetry is practically indistinguishable from that in the present structure derived from the heterotetramer present in the asymmetric unit (Fig. 1).
In a related study, the heterotetrameric E1b component of T. thermophilus BCKDC was shown to bind a substrate analog, 4-methylpentanoate (MPA), in only one of its two active sites (23). The binding of MPA promotes the movement of a loop region (Gly-121 to Gln-131) in the ␤ subunit to the active site so as to interact with the bound MPA. The authors attribute the above different active site conformations in Thermus E1b to crystal packing effects. The regions surrounding the bound MPA are in contact with another symmetry-related E1b heterotetramer; this contact is absent in the active site devoid of MPA. The same study shows that the soaking of the Thermus E1b crystal with a keto acid substrate 4-methyl-2-oxopentanoate results in a high occupancy of the reaction intermediate ␣-carbanion-ThDP in both active sites of the heterotetramer, similar to that observed with human E1b. This result is also consistent with the presence of two identical active sites in the crystal lattice prior to the substrate soak.
As shown in Fig. 2A, Asp-200␣, Glu-46␤, Glu-76␤, and Asp-108␤ in human E1b align with Asp-180␣, Glu-28␤, Glu-59␤, and Asp-91␤ in Bacillus E1p, respectively. The negatively charged side chains of the Bacillus E1p residues are suggested to participate in proton relay between the two active sites (17). A double mutant of Bacillus E1p, D180N␣/E183Q␣, exhibits little DCPIP-mediated E1p activity and overall activity of the reconstituted Bacillus PDC. It is proposed that removal of the negatively charged side chains disrupts proton relay in the proton-wire channel, which in turn abrogates communication and the switching of their conformations between the two E1p active sites necessary for the alternating reaction mechanism (17). In this study, we show that mutations in the above putative proton-wire residues in human E1b largely result in moderate reductions (D200A␣, D200N␣, E198Q␣/D200N␣, and E46Q␤) and, in some case, even a slightly increase (D108N␤) in BCKDC overall activity (Fig. 2C). Similar conclusions were reached in a previous study of transketolase (35). Residue Glu-162 in transketolase is part of the glutamic acid cluster at the subunit interface linking the two ThDP molecules. Replacement of this residue results in a mutant with most catalytic properties similar to wild type, which, in retrospect, argues against the presence of a proton wire to mediate communication between two active sites.
Remarkably, the nearly complete inactivation of BCKDC overall activity in the D200A␣/D108N␤ mutant can be rescued to about 50% of the wild type by treatment with the naturally occurring osmolyte TMAO (Fig. 2D). This result suggests that the double mutations in E1b adversely affect the folding and assembly of the mutant protein, similar to that observed with a subset of human mutations causing maple syrup urine disease (31). The putative defective assembly in the D200A␣/D108N␤ mutant is corroborated by the DSC result that shows a significantly reduced single T m (57.4°C) as opposed to the higher end of the double E1b values (52.6/65.4°C) shown by wild-type E1b. The single transition exhibited by this mutant strongly suggests that these mutations produce highly unstable heterotetramers that denature directly into random-coil structure without going through the heterodimeric intermediate.
From a structural perspective, side chains of Asp-200␣ and Asp-108␤ in human E1b are connected by a water-mediated hydrogen-bonding network in an almost perfect tetrahedral arrangement. This water molecule (Fig. 2B) also participates in a hydrogen-bonding network with two additional water molecules (not shown). Alterations of the Asp-200␣ and Asp-108␤ side chains likely disrupt this hydrogen-bonding network required for maintaining the integrity of the ␣/␤ interface. The TMAO treatment possibly imparts a compacting effect at the subunit interface via a "solvophobic" mechanism, resulting in a forced assembly and partial restoration of E1b catalysis (31). Therefore, results of the TMAO incubation and DSC measurements make the participation of Asp-200␣ and

and kinetic constants for hybrid (S292Q␣/WT) E1b
The hybrid E1b containing the S292Q␣ mutation in one of the two ␣ subunits was prepared as described under "Experimental Procedures." E1b-catalyzed decarboxylation was assayed in the presence of DCPIP. E1b-catalyzed reductive acylation was measured with ͓U-14 C͔KIV and free LBD as substrates. Overall activity of BCKDC was assayed spectrophotometrically after reconstitution with E2b and E3. Wild-type E1b and S292Q␣ E1b with both subunits harboring the mutation served as positive and negative controls, respectively.