Protein Biotinylation Visualized by a Complex Structure of Biotin Protein Ligase with a Substrate*

Biotin protein ligase (BPL) catalyzes the biotinylation of the biotin carboxyl carrier protein (BCCP) only at a special lysine residue. Here we report the first structure of BPL·BCCP complex crystals, which are prepared using two BPL mutants: R48A and R48A/K111A. From a detailed structural characterization, it is likely that the mutants retain functionality as enzymes but have a reduced activity to produce the reaction intermediate biotinyl-5′-AMP. The observed biotin and partly disordered ATP in the mutant structures may act as a non-reactive analog of the substrates or biotinyl-5′-AMP, thereby providing the complex crystals. The four crystallographically independent BPL·BCCP complexes obtained can be classified structurally into three groups: the formation stages 1 and 2 with apo-BCCP and the product stage with biotinylated holo-BCCP. Residues responsible for the complex formation as well as for the biotinylation reaction have been identified. The C-terminal domain of BPL shows especially large conformational changes to accommodate BCCP, suggesting its functional importance. The formation stage 1 complex shows the closest distance between the carboxyl carbon of biotin and the special lysine of BCCP, suggesting its relevance to the unobserved reaction stage. Interestingly, bound ATP and biotin are also seen in the product stage, indicating that the substrates may be recruited into the product stage complex before the release of holo-BCCP, probably for the next reaction cycle. The existence of formation and product stages before and after the reaction stage would be favorable to ensure both the reaction efficiency and the extreme substrate specificity of the biotinylation reaction.

Biotin-dependent carboxylases constitute a ubiquitous family of enzymes that catalyze the transfer of carbon dioxide between metabolites using the biotin moiety as a carboxyl carrier (1,2). The attachment of biotin to the biotin-dependent enzymes is catalyzed by the biotin protein ligase (BPL) 2 in two steps (Reactions 1 and 2).
Biotin ϩ ATP 7 biotinyl-5Ј-AMP ϩ PP i REACTION 1 Biotinyl-5Ј-AMP ϩ apo-BCCP 3 holo-BCCP ϩ AMP REACTION 2 Firstly, BPL activates biotin at the expense of ATP to the reaction intermediate biotinyl-5Ј-AMP in which the carboxyl group of inert biotin is activated by the addition of an adenylate group. Subsequently, the biotin moiety of biotinyl-5Ј-AMP is transferred to the ⑀-amino group of a specific lysine residue of the target protein (e.g. the biotin carboxyl carrier protein (BCCP) subunit of acetyl-CoA carboxylase (3)). The biotinylated holo-BCCP subunit carries a covalently bound carboxyl unit between different active sites of the multienzyme biotin-dependent carboxylase complexes, which play essential roles in the fatty acid synthesis, the amino acid degradation, and the CO 2 fixation (4,5). In bacteria and eukaryotes, biotinylation is essential to initiate the first step of fatty acid biosynthesis, which is catalyzed by acetyl-CoA carboxylase. Because human biotin-dependent carboxylases are the direct targets for the development of antiobesity and anti-diabetes agents, structural information on these enzymes is important for drug discovery research (6). In archaea, the function of acetyl-CoA/propionyl-CoA carboxylases is different due to the absence of usual fatty acids in membranes: they act as CO 2 fixation enzymes in the modified 3-hydroxypropionate cycle to assimilate CO 2 into the cell (5). In addition, the structural basis of biotinylation is important to develop useful applications in protein engineering, such as the high affinity biotin tagging for protein purification (7) and the molecular imaging by quantum dots (8).
The understanding of the first step of BPL reaction at the atomic level was reached based on the crystal structures of BPL from Pyrococcus horikoshii OT3 (PhBPL) in complex with various biological ligands including biotinyl-5Ј-AMP (9) as well as on the crystal structure of BPL from Escherichia coli (EcBirA) in complex with the reaction intermediate analog biotinol-5Ј-AMP (10). However, scientists have not yet been able to resolve the structural details in the second step of biotinylation reaction in which the activated biotin is transferred to the special lysine of BCCP. This exceptionally selective post-translational modification makes understanding how the proteins BPL and BCCP carry out the biotin transfer of particular interest. To elucidate the biotin transfer reaction, several biophysical or biochemical studies on the BPL⅐BCCP complex have been performed (e.g. an NMR study (11), mutagenesis studies (12,13), and a chemical cross-linking study (14)). Despite these efforts, the structural mode of the biotin transfer reaction is not fully understood due to the absence of three-dimensional structures for the BPL⅐BCCP complex.

EXPERIMENTAL PROCEDURES
Protein Expression, Purification, Crystallization, and Data Collection-The expression and purification of PhBPL, PhBPL*, PhBPL**, and PhBCCP⌬N76 were performed as described elsewhere (15,16). Crystals of the ATP liganded form of wild-type PhBPL (PhBPL⅐ATP) were prepared by adding 5 mM ATP to the previously reported crystallization condition for the unliganded wild-type PhBPL (15). Crystals of the liganded forms of PhBPL mutants were prepared using the wild-type crystallization condition except for adding ligands: 5 mM ATP and 5 mM biotin for PhBPL*⅐biotinyl-5Ј-AMP and PhBPL**⅐biotin⅐adenosine; 5 mM biotinol-5Ј-AMP for PhBPL**⅐biotinol-5Ј-AMP. Other forms of crystals were prepared as described elsewhere (16). All data were collected at 100 K using synchrotron radiation on a Jupiter 210 charge-coupled device at the beamline BL26B1 of SPring-8, Japan (17). Diffraction data were processed and scaled with the HKL-2000 program suite (18). Data collection statistics are summarized in tables (Table 1 and supplemental Tables  S1 and S2).
Model Building and Refinement-The structures of the mutated and/or liganded forms of PhBPL were determined by the difference Fourier synthesis based on its wild-type structure. The structures of PhBCCP⌬N76 in two different crystal forms were determined by the molecular replacement with the program MOLREP (19), using coordinates of the C-terminal domain of E. coli BCCP (EcBCCP, PDB ID 1BDO) as a search model (20). The initial phases of the protein⅐protein complexes were obtained by the molecular replacement with MOLREP, using the crystal structures of PhBPL*, PhBPL**, and PhBCCP⌬N76 as search models. After rebuilding the initial models using QUANTA (Accelrys, San Diego, CA), several rounds of the manual model revision and the structure refinement using CNS (21) were performed for all the structures. The electron densities for biotinyl-5Ј-AMP and biocytin clearly indicate a covalent linkage between biotin and AMP and between biotin and the Lys 115 of BCCP, respectively. Refinement statistics are summarized in tables (Table 1 and supplemental Tables S1 and S2). All of the models have excellent stereochemistry, as evaluated by the program PROCHECK (22). The structural superposition analysis was performed using the program LSQKAB (23) and a subsequent classification of structural changes according to the multiple superposition method (24). The figures illustrating these structures were prepared using the program PyMOL. 3 Sequence alignments were generated by ClustalX (26) and displayed with ESPript (27).

RESULTS AND DISCUSSION
Structures of PhBPL-To examine the effect of mutations R48A and R48A/K111A on the structure and reactivity of PhBPL, we determined the crystal structures of PhBPL* and PhBPL** and their liganded forms PhBPL*⅐biotinyl-5Ј-AMP, PhBPL**⅐biotin⅐adenosine, and PhBPL**⅐biotinol-5Ј-AMP (supplemental Table S1). Overall, all the structures were found isomorphous to the wild-type PhBPL (9). The crystals contain one dimer in the asymmetric unit. The mutants have essentially the same overall protomer conformation as the wild-type PhBPL: a C ␣ superposition between any pair of the protomers results in the root-mean-square deviation (r.m.s.d.) value of Ͻ1 Å. The liganded forms of crystals were prepared by a cocrystallization with ligands, to avoid crystal packing effects. The apo structures of PhBPL mutants show that the active site loop Gly 45 -Trp 53 is disordered. In the liganded structures, the loop becomes ordered and closes over the bound substrates to form the main hole of the active site. This ligand-induced ordering of the active site loop is also observed in the structures of wild-type PhBPL (9). The cocrystallization of PhBPL* with ATP and biotin provided the PhBPL*⅐biotinyl-5Ј-AMP complex where the U-shaped biotinyl-5Ј-AMP was found in the bifurcated main hole of BPL*, indicating the retained functionality of the single mutant (Fig. 1A, and supplemental Table S3 Fig. S1). However, the same cocrystallization condition using PhBPL** resulted in a substrate complex where the active site hole was occupied by biotin and adenosine. Because ATP dominates the adenosine nucleotide species in the crystallization solution, it is likely that the modeled adenosine is an ordered part of ATP. The nucleotide binding mode in the double mutant was compared with that in the wild-type enzyme (supplemental Table S3 and Fig. S2): PhBPL⅐ATP (PDB ID 1X01) and PhBPL⅐ADP (PDB ID 1WNL). In the wild-type enzyme, the phosphate part of nucleotides is recognized by polar interactions with the basic residues Arg 48 , Arg 51 , Lys 111 , and Arg 233 . Thus the mutations R48A and K111A would induce the disordering of the triphosphate moiety of ATP and tend to prevent it from reacting with biotin in the mutant crystals. Most likely, this reduced reactivity of bound ATP allowed the successful cocrystallization of single and double mutants with BCCP. As described in the later section of BPL⅐BCCP complex, PhBPL** can produce holo-BCCP, indicating the retained functionality of the double mutant. However, the lowest ability in the reaction intermediate production may hinder the formation of biotinyl-5Ј-AMP-liganded crystals in the double mutant. The binding mode of the reaction intermediate in the double mutant could be estimated from the PhBPL**⅐biotinol-5Ј-AMP structure, which was determined from a cocrystal with the reaction intermediate analog biotinol-5Ј-AMP (supplemental Table S3 and Fig. S1). The conformation of bound biotinol-5Ј-AMP in PhBPL** is quite similar to that of bound biotinyl-5Ј-AMP in PhBPL* as well as in wild-type PhBPL, suggesting that the mutations used are not essentially defective in the formation of biotinyl-5Ј-AMP.
Structures of PhBCCP⌬N76-We designed and expressed a truncated version of PhBCCP lacking N-terminal 76 residues and containing 73 C-terminal amino acids from Val 77 (PhBCCP⌬N76, residues 76 -149), to improve its handling and crystallizability. It is well known that the C-terminal half of biotinyl domain is expressed as a stable protein, which can be biotinylated normally both in vivo and in vitro (28 -33). Located upstream of the biotinyl domain sequence are proline/alanine-rich sequences of varying lengths, which have been proposed to act as flexible linkers (34). The crystal structure of PhBCCP⌬N76 (form I; modeled residues 76 -149) has been solved by the molecular replacement using BCCP from EcBCCP (PDB ID 1BDO) as a search model and refined at a resolution of 1.55 Å (supplemental Table S2). The second form of crystal structure was determined by molecular replacement using the form I structure and refined at a resolution of 1.55 Å (form II; modeled residues 80 -149). In the form I crystal, there is one protomer in the crystallographic asymmetric unit, whereas the form II crystal has two protomers in the asymmetric unit. The r.m.s.d. values from a C ␣ superposition between the observed three crystallographically independent protomers are Ͻ1 Å, indicating an essentially identical protomer fold. The overall fold of PhBCCP⌬76 is described as a flattened ␤-barrel structure comprising two four-stranded ␤-sheets with the N-and C-terminal residues close together at one end of the structure (Fig. 1B). The biotinyl domain has an internal 2-fold symmetry. The C ␣ atoms in two halves of the molecule (residues 80 -112 and 117-149) can be aligned with an r.m.s.d. of 0.48 Å after a rotation about the 2-fold axis, clearly indicating that this molecule was created by a duplication of two identical ancestor genes as first suggested by Toh et al. (35) and structurally confirmed by Athappilly and Hendrickson (20). From a structural perspective, it is likely that the ancestor protein is a dimer of two identical hammerhead motifs ␤ 6 -␤ 7 -␤ 8 -␤ 1 and ␤ 2 -␤ 3 -␤ 4 -␤ 5 . The biotinylation target Lys 115 is located at the type I' hairpin ␤-turn involving two residues Met 114 and Lys 115 , which connects the N-terminal and C-terminal halves of the biotinyl domain. The sequences of C-terminal regions corresponding to the biotinyl domain are well conserved among different BCCPs ( Fig. 2A). Accordingly, the crystal structure of PhBCCP⌬76 confirms the same overall folding with the reported crystal structures of the C-terminal fragment of EcBCCP (20, 36) and the 1.3 S subunit of the Propionibacterium shermanii transcarboxylase complex (37). Furthermore, biotinyl domains have been shown to bear sequence and structural similarities to the lipoyl domains of 2-oxo-acid dehydrogenase multienzyme complexes, which undergo an analogous post-translational modification (1,38,39). There are two main differences among these structures: PhBCCP and many other BCCP structures lack the ␤ 2 -␤ 3 "thumb" loop, which is observed in EcBCCP (20); the two symmetric halves of all biotinyl domain structures are connected by a type IЈ ␤-turn, whereas the lipoyl domains adopt a type I conformation for this turn. PhBCCP may be an evolutionally more primitive form as compared with EcBCCP, because PhBCCP is more symmetric in terms of the duplication on hammerhead motifs and lacks the thumb loop providing additional recognition for the biotinyl moiety.
Structures of PhBPL⅐PhBCCP Complex-The complex structures of PhBPL*⅐biotin⅐adenosine⅐PhBCCP⌬N76 and PhBPL**⅐biotin⅐adenosine⅐holo-PhBCCP⌬N76 were determined by molecular replacement using the coordinates of PhBPL*, PhBPL**, and PhBCCP⌬N76 structures and refined at resolutions of 2.7 and 2.0 Å, respectively ( Fig. 1C and Table 1). These two forms of PhBPL⅐PhBCCP complex crystals are nearly isomorphous and both have a 2:2 heterotetramer in the asymmetric unit. The PhBPL⅐PhBCCP association involves the formation of a large intermolecular ␤-sheet, which is solvent-exposed on one side to house the biotinyl-5Ј-AMP. Notably, the bifurcated main holes of both the complexes are occupied by biotin and adenosine. As mentioned in the first section, we suppose that the modeled adenosine is an ordered part of ATP. The B-factors for the where F obs and F calc are the observed and calculated structure factors, respectively. c R free is the R factor for a subset of 5% of the reflections that were omitted from refinement.
ligand atoms are comparable to those of other atoms in the crystal structure; averaged B-factors for the ligand atoms and all atoms are 43.7 Å 2 and 42.1 Å 2 in the single mutant complex and 24.8 Å 2 and 28.4 Å 2 in the double mutant complex, respectively (Table 1 and supplemental Table S3). Judging from this fact, it is likely that the bound ligands have some biological role in the reaction mechanism rather than a mutation/crystallization artifact from the binding of low affinity ligands. In both the complexes, although the binding mode of the biotin and the adenine ring is similar to that of the corresponding part of biotinyl-5Ј-AMP seen in the other liganded forms, the ribose ring shows distinct conformation when compared with that of biotinyl-5Ј-AMP (supplemental Fig. S3). This suggests that the mutations tend to prevent ATP from reacting with biotin in the complex crystals. However, because the single mutant can provide the biotinyl-5Ј-AMP liganded form and the biotinylated holo-PhBCCP is found in the double mutant complex, the functionality of the mutants would be retained. Therefore, the snapshots fortunately captured at points before and after the biotinylation of PhBCCP may shed light on the interaction of proteins and the target protein/lysine residue specificity of BPL.
The association of two 1:1 complexes of PhBPL⅐PhBCCP to make the 2:2 complex is mediated by the PhBPL dimer interface analogous to one observed in the wild-type PhBPL (9). At the PhBPL⅐PhBCCP interface, buried solvent-accessible surface area is ϳ900 Å 2 per protomer, which is comparable to the PhBPL dimer interface area of 1030 Å 2 in the complex. The intermolecular interface is mainly hydrophobic (over 60% of interface atoms are non-polar) and defined by a number of hydrogen bonds ( Table 2). The electrostatic potential on the surfaces of proteins shows a heterogeneous charge distribution that reflects a good charge complementarity between the interacting surfaces (Fig. 3). This indicates that electrostatic interactions are important for the complex formation in addition to a pronounced molecular surface complementarity. The structures of the apo and holo forms of PhBCCP⌬N76 in free and complex states are generally similar, suggesting that binding to PhBPL and biotinylation causes few significant changes in the overall fold of the biotinyl domain, except for the ␤ 4 -␤ 5 hairpin turn. In contrast, structures of PhBPL* and PhBPL** show more extensive local conformational changes. Although the catalytic domain is relatively similar to the one in free state, the C-ter-  S4). In the present two complex structures, we observed four crystallographically independent BPL⅐BCCP complexes. From the overall structural similarity in the orientational relationship between BPL and BCCP, the four independent complexes can be classified into three groups ( Fig. 4A and supplemental Table S4). The B and D subunits of the complexes are in a nearly identical spatial relationship, in which the substrate lysine of BCCP enters close to the BPL active site. We designate this state as the formation stage 1.
The A and C subunits of both the complexes are also in closely related orientations. In the single mutant complex, BCCP seems to approach the BPL active site in another way; this state is referred to as the formation stage 2. The designation of "stage 1" or "stage 2" does not imply an order in which these putative snapshots of the reaction occur along the reaction coordinate. On the other hand, in the double mutant complex, the biotinylated holo-BCCP is observed as the reaction product; this state is referred to as the product stage. Thus, there is a certain amount of plasticity around the active site of PhBPL to accommodate its substrate PhBCCP. Especially, a conformational flexibility of a C-terminal domain loop (Ile 226 -Asp 229 ) in the side hole to the active site provides an entry for the substrate lysine residue and an exit pathway for the product biocytin residue of holo-BCCP (Fig.  1C). Notably, the loop residue Tyr 227 undergoes an open/ close motion, which may act as a lid of the side hole to regulate the traffic of ligands: open in the formation stage 1 and closed in the other stages (Fig. 4B). In addition, the inducedfit ordering of the active site loop (Gly 45 -Trp 53 ) upon binding substrates seems to be required to build the side hole. The association-induced local conformational change in PhBCCP is limited in the ␤ 4 -␤ 5 hairpin turn that displays the target lysine residue Lys 115 . In the formation stage 2 and the product stage, the conformation of this type IЈ turn is similar to that in the free state structure. However, in the formation stage 1, the positions of Met 114 and Lys 115 in PhBCCP clearly shift toward the active site of PhBPL due to both the local deformation of the turn and the rigidbody intermolecular relocation. In the formation stage 1, the strictly conserved Glu 112 of PhBCCP makes a direct or water-mediated hydrogen bond with the mainchain nitrogen of the target lysine Lys 115 , which may be a key interaction to promote the local conformational change of the hairpin turn. The interatomic distance between the carboxyl carbon of biotin and the C ␣ of Lys 115 is ϳ11.9 Å in the formation stage 2 and the product stage, whereas it is ϳ8.4 Å in the formation stage 1. This indicates that the formation stage 1 is more favorable for the biotinylation reaction to take place (supplemental Table S3).

Recognition of Biotinyl Domain by PhBPL-In the PhBPL⅐PhBCCP complex, the biotinyl domain interacts with
PhBPL mainly in two regions ( Table 2 and Figs. 2A and 5). The first cluster comprises its strand ␤ 2 and the loop residues that precede it. The second cluster contains the strands ␤ 4 and ␤ 5 , including the biotinylation target Lys 115 . On the other hand, PhBPL residues interacts with the biotinyl domain mainly in six regions: the ␤ 2 -␤ 3 active site loop, the N-terminal part of the ␣ 2 helix, the ␤ 4 -␤ 5 turn involving the invariant residue Asn 103 , the C-terminal half of the strand ␤ 6 , the ␤ 9 -␤ 10 loop, and the ␤ 11 -␤ 12 loop (Table 2 and Figs. 2B and 5). Although these residue clusters are remote in the primary sequence, the tertiary fold brings them into close proximity. Interestingly, the ␤ 2 strand in PhBCCP and the ␤ 4 -␤ 5 turn in PhBPL show distinct pattern of interactions depending upon the stage of reaction (Table 2 and Fig. 2). Particularly, in the ␤ 4 -␤ 5 turn of PhBPL, only Trp 101 participates in the PhBCCP recognition at the formation stage 2 or the product stage. However, additional ␤ 4 -␤ 5 turn residues Pro 102 -Asp 104 of PhBPL are also buried by the biotinyl domain at the formation stage 1, indicating a significant structural difference.
Because BPL can biotinylate the biotinyl domains from diverse sources (28), key molecular recognition in the BPL⅐BCCP complex formation should be common. Thus, conserved residues among BCCP orthologs ( Fig. 2A), Met 87 , Gly 89 , Gly 105 , Glu 112 , Met 114 , Lys 115 , and Met 116 may have unique roles in the interaction with BPL. The role of the target lysine Lys 115 is obviously crucial. Several mutagenesis studies provided an indication of the importance of the other conserved residues. For example, in EcBCCP, the Glu to Lys substitution at Glu 119 , which corresponds to Glu 112 in Ph-BCCP, makes it virtually inactive as a substrate for BPL (40), and the mutations at either of the methionine residues flanking the specific lysine severely reduce the ability of biotinyl domains to accept biotin moiety (41). In addition, a mutation at the site equivalent to Met 87 in the biotinyl domain of human propionyl-CoA carboxylase (Met 641 3 Lys) dramatically reduces the rate of biotinylation by EcBirA (32). Similarly, important roles should be assigned to the conserved residues of PhBPL (Fig. 2B): Trp 101 -Asp 104 for directing of target Lys 115 in the biotinyl domain. When the substrate Lys 115 enters into the active site of PhBPL, it forms a hydrogen bond with Asn 103 in the formation stage 1 (Fig. 5A). Notably, these two residues are invariant even in the level of protein family, among biotinyl and lipoyl domains as well as among BPL and LplA and LipB enzymes (Fig. 2). The electrostatic interaction between the substrate Lys 115 and the conserved Asp 104 of PhBPL would also be important. Other conserved PhBPL residues, Leu 116 and Glu 118 , on the ␤ 6 strand associate with the ␤ 5 strand of PhBCCP, including a conserved residue Met 116 . Correlated interaction of these two strands may determine the settlement of PhBPL and PhBCCP during the biotinylation process. Only a moderate conservation of interaction residues in the level of protein family, except for the invariant Asn 103 in PhBPL and Lys 115 in PhBCCP, may indicate that mainchain interactions dominate the PhBPL-PhBCCP association.
Implications for the Reaction Mechanism-From the present results, it appears that the observed biotin and partly disordered ATP in the mutant structures act as a nonreactive analog of substrates or biotinyl-5Ј-AMP, thereby allowing us to visualize the snapshots of protein⅐protein complexes. Unfortunately, the non-hydrolyzable intermediate analog biotinol-5Ј-AMP did not provide analyzable complex crystals. To further characterize the observed three stages of the PhBPL⅐PhBCCP complex, structures of bound ligands are compared among various crystal forms (supplemental Table S3). Although the wild type and two mutants of PhBPL bind biotinyl-5Ј-AMP or biotinol-5Ј-AMP in essentially the same conformation, the ligand in the wild-type seems to be more flexible at the ribose-5Ј-phosphate part, which would reflect the higher reactivity of biotinyl-5Ј-AMP in the wild type. The averaged B-factor of ligand atoms is 33.8 Å 2 at the ribose-5Ј-phosphate part and 16.3 Å 2 at the remaining part in the PhBPL⅐biotinyl-5Ј-AMP structure (PDB ID 1WQW), whereas they are 16.2 Å 2 and 11.2 Å 2 at the respective parts in PhBPL*⅐biotinyl-5Ј-AMP, and 21.6 Å 2 and 15.6 Å 2 at the respective parts in PhBPL**⅐biotinol-5Ј-AMP. From a structural comparison, the interatomic distance between the N9 atom of adenine moiety and the carboxyl carbon of biotin moiety (biotin-adenine ring distance) is longer in the wild-type structure: 6.9 Å in the wild-type, 6.6 Å in the single mutant, and 6.5 Å in the double mutant. This fact may indicate that the biotin-adenine ring distance is a good indicator of the reactivity of the intermediate biotinyl-5Ј-AMP. We also calculated the biotin-adenine ring distance as well as the interatomic distance between the carboxyl carbon of biotin and the C ␣ of Lys 115 in BCCP (biotin-BCCP distance), in the BPL⅐BCCP complex structures. Interestingly, the biotin-adenine ring distance and the biotin-BCCP distance show a complementary relationship: 7.2 and 8.4 Å in the formation stage 1, 6.7 and 12.0 Å in the formation stage 2, and 6.9 and 11.7 Å in the product stage, respectively. Assuming that the observed biotin and partly disordered ATP in the mutant complex structures mimic the biotinyl-5Ј-AMP in the formation stages, this analysis of interatomic distance may indicate the higher reactivity of biotinyl-5Ј-AMP in the formation stage 1. Due to the limited resolution of 2.7 Å in the single mutant complex, the structural difference between the formation stage 2 and the product stage is unclear. Because the product stage shows a medium value in the biotin-adenine ring distance, it is not conclusive whether the putatively recruited ATP and biotin probably for the next reaction cycle in the wild-type PhBPL⅐holo-PhBCCP complex can react to produce biotinyl-5Ј-AMP or not. However, the fact that the cocrystallization of wild-type PhBPL with ATP, biotin, and PhBCCP⌬N76 did not provide the expected PhBPL⅐biotinyl-5Ј-AMP⅐holo-PhBCCP⌬N76 complex crystals may indicate that the recruited ATP and biotin to the product stage complex is not reactive until the dissociation of holo-BCCP from BPL or the product releasing and the biotinyl-5Ј-AMP formation is coupled. This hypothesis is reasonable in terms of the efficiency of the biotinylation reaction; the assumption of biotinyl-5Ј-AMP formation in the product stage means the occurrence of competition with the formation stage 2, which would be unfavorable for the complete biotinylation of BCCP molecules in solution due to the product inhibition.
The structures obtained provide a putative scenario for the biotin transfer reaction. The apo BPL exists with the disordered active site loop, which enlarges the entrance to the active site. In the presence of ATP and biotin, BPL forms the reaction intermediate biotinyl-5Ј-AMP and the side hole to accommodate the biotinyl domain of BCCP. In the formation stages 1 and 2, the negative charges of the BPL molecule from the conserved acidic residues (Asp 104 and Asp 229 in PhBPL) and phosphate group of biotinyl-5Ј-AMP may attract the target lysine of apo BCCP (Lys 115 in Ph-BCCP) to make the BPL⅐BCCP complex, which is observed in the present crystal structures of PhBPL**⅐biotin⅐adenosine⅐holo-PhBCCP⌬N76 (chains B and D in the formation stage 1) and PhBPL*⅐biotin⅐adenosine⅐PhBCCP⌬N76 (chains B and D in the formation stage 1; chains A and C in the formation stage 2). Once PhBPL⅐biotinyl-5Ј-AMP and the apo biotinyl domain are associated, the complex should proceed to the reaction stage producing the holo-biotinyl domain. The nucleophilic attack on the carbonyl carbon of activated biotin by the ⑀-amino group of the substrate lysine results in the transfer of biotin onto the apo-biotinyl domain and the concomitant formation of AMP. The side chain of the substrate lysine must be precisely positioned into the active site of BPL to bring the electron donating nitrogen close enough to the reactive carbonyl carbon of biotinyl-5Ј-AMP to permit the chemical reaction, which is in agreement with a mutational study reporting that a translocation of the lysine by one place to either side in the exposed ␤ 4 -␤ 5 turn abolishes biotinylation (41,42). Thus, the biotinyl domain functions as a protein scaffold that displays the biotin-accepting lysine to BPL. On the other hand, it is likely that the conserved residues Asn 103 and Asp 104 of PhBPL form a catalytic dyad, in which Asn 103 is important for the precise positioning of the target lysine by forming a hydrogen bond, and Asp 104 acts as a general base to deprotonate the ⑀-amino group of the lysine to render it sufficiently nucleophilic to attack the carbonyl carbon of the valeric acid side chain of biotin. After the reaction stage, the biotinylated BCCP may not be released immediately but maintained in the product stage, which is observed in the chains A and C of the PhBPL**⅐biotin⅐ adenosine⅐holo-PhBCCP⌬N76 structure. At the product stage, the peptide group of the biocytin residue is recognized by BPL with a hydrogenbonding network involving the main-chain atoms of Phe 210 and Tyr 227 in PhBPL, and the biocytin side chain is packed by van der Waals interactions with Met 114 in PhBCCP, and Pro 76 , Trp 101 , Phe 210 , Gly 211 , Arg 212 , Ile 226 , and Tyr 227 in PhBPL, which allowed us to observe the biocytin residue clearly in the electron density map.
A superposition of the before and after biotinylated subunits (Fig. 4A) provides insights into the unobserved reaction stage. Unfortunately, it is clear that both the formation stages 1 and 2 do not provide suitable geometry for the nucleophilic attack by the special lysine of BCCP, although a slight remodeling of the formation stage 1 seems to allow the reaction. In the chains B and D of the PhBPL**⅐biotin⅐ adenosine⅐holo-PhBCCP⌬N76 structure in the formation stage 1, we observed the closest distance of 6.1 Å between the N of Lys 115 and the biotin carboxyl carbon, which is the putative site for the target carbonyl carbon of biotinyl-5Ј-AMP. In the putative reaction stage, the carbonyl carbon of biotinyl-5Ј-AMP should be located just adjacent to the ⑀-amino group of target lysine, which may be provided by an additional structural change of either the active site loop of BPL or the target ␤-turn of BCCP toward the active site. It is possible that the restoring mutations at Arg 48 and Lys 111 or the binding of physiological intermediate biotinyl-5Ј-AMP instead of the observed biotin and partly disordered ATP provides this putative conformational change. In this case, the current formation stage 1 in the mutants would correspond to the reaction stage in the wild type. Further investigation is required to clarify this point. In addition, it should be noted that multiple conformations are observed in the present crystal structures (Fig. 4A), suggesting that the free energy level of these conformations are comparable. The existence of formation and product stages before and after the reaction stage would be favorable to ensure both the reaction efficiency and the extreme substrate specificity in the biotinylation reaction; the high entropic energy barrier due to the strict substrate specificity is resolved by adding the intermediate steps, and the multistep reaction sequence structurally avoid the error of reaction.
Molecular Discrimination between Biotinyl and Lipoyl Domains-The biotinyl domains of biotin-dependent enzymes and the lipoyl domains of 2-oxo acid dehydrogenase multienzyme complexes have homologous structures, but the post-translational modification of the target lysine residue in each domain is correctly selected by BPL and lipoyl protein ligases LplA/LipB, respectively. Although the reactions catalyzed by these ligases are analogous, the BPL and LplA/LipB bear poor sequence conservation (Fig. 2B), suggesting an evolutionarily distant relationship among the family members. BPLs from various sources have been found to recognize/biotinylate various BCCPs from different organisms (2,12,14,(42)(43)(44)(45). Of all the PhBCCP⌬N76 residues that make interactions with PhBPL, only the MKM motif at the tip of the hairpin loop is well exposed to the solvent. Therefore, the discrimination of this Lys 115 from the other four lysine residues in PhBCCP⌬N76 may be understood from the PhBPL⅐PhBCCP⌬N76 structures showing that only the exposed ␤-turn can fit into the side hole connecting to the biotinyl-5Ј-AMP binding main hole of PhBPL (Fig. 1C). The local sequence surrounding the receptive lysine within the context of a folded BCCP should be important for the BPL recognition, given that the lipoyl domains containing DKA, DKV, and AKA motifs at the tip of the corresponding hairpin loop cannot accept biotinylation. Also it is reported that BPL does not biotinylate a biotinyl domain with DKA replacing the MKM motif (41) and S. tokodaii BCCP, which has the MKS motif does not serve as a substrate for the heterologous BPL or EcBirA (46). However, the MKM motif of the biotinyl domain by itself does not necessarily specify biotinylation, because it has been shown that the two methionines flanking the specific lysine can be replaced by leucines (32,47) and that replacing the DKA sequence at the hairpin loop of lipoyl domain with MKM does not make it a substrate for BPL (48). Therefore, there must be key differences between the biotinyl and lipoyl domains that allow the relevant protein ligases to distinguish them for the different purposes of post-translational modification. To understand how discrimination between lipoylation and biotinylation is achieved, we superposed the selected biotinyl and lipoyl domains to the biotinyl domain in the PhBPL**⅐biotin⅐adenosine⅐holo-PhBCCP⌬N76 complex. The biotinyl domains are well overlaid and show good shape/charge complementarities at the BPL⅐BCCP interfaces. The ␤ 2 -␤ 3 thumb loop of EcBCCP does not collide with the BPL structural elements (Fig. 6A). Although the PhBPL⅐lipoyl domains show overall shape complementarities at their interfaces, the loop ␤ 1 -␤ 2 of lipoyl domains tend to interfere with the N terminus of ␣ 2 of PhBPL (Fig. 6B). Additionally, the charge complementarities at the interfaces are destroyed in the putative complexes with lipoyl domains. The position of conserved positively charged Arg 93 of Ph-BCCP is occupied by negatively charged Glu in the lipoyl domain, and the position of conserved nonpolar Gly 105 of the biotinyl domain is occupied by negatively charged Asp or Glu in the lipoyl domain. The negatively charged aspartate side chain in the motif DKV or DKA of the lipoyl domain instead of nonionic methionines in BCCP also prevents its recognition by the side hole of BPL. Because the attraction of BCCP would be affected by long range electrostatic interactions, we suggest that the difference in surface charge complementarities is one of the factors contributing to the molecular discrimination between the biotinyl and lipoyl domains. The other difference comes from diversity in the type of the target ␤-turn to be modified. In BCCP, the turn loop contains two residues and has the type IЈ, whereas in the lipoyl domain, it consists of four residues and has the type I. Therefore, the conformation of the ␤-turn in the lipoyl domain is different from that in the biotinyl domain, which would affect the positioning of the specific lysine to the biotin moiety (Fig. 6C).
In conclusion, it is apparent that the recognition of the biotinyl domain is a rather complex process. This is in contrast with most other post-translational modifications in which the primary structure surrounding the target residue can be of crucial importance. BPL and BCCP have several mobile parts that are essential for the biotinylation reaction in multiple stages: BCCP binding (formation stages 1 and 2), catalysis (reaction stage), and product release (product stage). Learning how these factors are choreographed to accomplish the protein biotinylation with extreme substrate specificity remains an unmet challenge for investigators.