X-ray Structure of Pyruvate Formate-Lyase in Complex with Pyruvate and CoA

The glycyl radical enzyme pyruvate formate-lyase (PFL) synthesizes acetyl-CoA and formate from pyruvate and CoA. With the crystal structure of the non-radical form of PFL in complex with its two substrates, we have trapped the moment prior to pyruvate cleavage. The structure reveals how the active site aligns the scissile bond of pyruvate for radical attack, prevents non-radical side reactions of the pyruvate, and confines radical migration. The structure shows CoA in a syn conformation awaiting pyruvate cleavage. By changing to an anti conformation, without affecting the adenine binding mode of CoA, the thiol of CoA could pick up the acetyl group resulting from pyruvate cleavage.

form) in complex with oxamate, an isosteric and chemically inert analog of pyruvate, shows the close spatial proximity of the catalytic triad Gly-734, Cys-419, Cys-418 that could accommodate direct radical transfers from Gly-734 to Cys-418 via Cys-419. Oxamate fits into a compact pocket where its carboxamide C is juxtaposed with Cys-418 S␥ (3.3 Å). Based on the structure and biochemical information, along with theoretical data (13), a mechanism was proposed (10) that involves the following steps (Fig. 1a). First half-reaction: pyruvate binding triggers generation of the Cys-418 thiyl radical by hydrogen transfer from Cys-418 S-H via Cys-419 to the glycyl radical; thiyl addition to the carbonyl C of pyruvate produces the tetrahedral oxyradical intermediate, which collapses into the acetyl thioester of Cys-418 and the formyl radical; the formyl radical is quenched to formate by the SH of Cys-419, thus producing the 419 thiyl. Second half-reaction: 419 thiyl generates the thiyl radical of CoA; the radical is replaced with the 418-linked acetyl resulting in acetyl-CoA.
An earlier mechanistic proposal (14) assumed the formation of a thiohemiketal between pyruvate and one of the active site cysteines. This possibility could not be ruled out by the available structure (10), as it contained the inert oxamate, which is principally unable to form a thiohemiketal-like compound. Moreover, the reported structures did not provide information on the binding site of the cosubstrate CoA/acetyl-CoA, since PFL shows no relationship to known folds of CoA binding proteins (15). We describe here the crystal structures of the non-radical form of PFL from E. coli in complex with pyruvate, with pyruvate and CoA, and for comparison, with oxamate and CoA. The new data support our view of the catalytic cycle ( Fig.  1a) and show the binding site of the cosubstrate CoA for the first time.

EXPERIMENTAL PROCEDURES
Protein Preparation-PFL was isolated from overproducing E. coli cells using strain 234M1 transformed with the expression vector p153E1 as described (16).
Data Collection-Each diffraction data set (Table I) was collected from one crystal at 100 K using synchrotron radiation or x-rays generated by a GX-18 rotating anode (Elliot/Enraf-Nonius, Delft). PFLϩpyrϩCoA: EMBL outstation at DESY (Hamburg, Germany), * The costs of publication of this article were defrayed in part by the payment of page charges. This article must therefore be hereby marked "advertisement" in accordance with 18 U.S.C. Section 1734 solely to indicate this fact.
The beamline X13, ϭ 0.8045 Å, MARCCD detector; PFLϩoxaϩCoA: GX-18 rotating anode, CuK␣, MAR345 detector; PFLϩpyr: high resolution data at the European Synchrotron Radiation Facility (Grenoble, France), beamline ID29, ϭ 0.915 Å, ADSC detector, low resolution data at rotating anode. Data were processed by the program package XDS (17). The quality of the data sets and the derived structures is shown in Table I.
Structure Determination-The structure of PFLϩpyrϩCoA was solved by molecular replacement using the published structure PFLϩoxa (Protein Data Bank code 3pfl) (10) as the starting model. By comparison with PFLϩpyrϩCoA, refined at 1.53 Å, a minor error in the starting model was discovered and corrected by a peptide flip of residue Gly-433 (distance (Gly-433, Cys-418 S␥) ϭ 11 Å). Since the monoclinic structures are highly isomorphic, the structure of PFLϩoxaϩCoA could be obtained from PFLϩpyrϩCoA by the F o (1) Ϫ F o (2) difference Fourier technique. The same technique was applied to solve PFLϩpyr by comparison with the re-refined structure PFLϩoxa. The CoA-containing structures contain a density peak near the pyrophosphate that was interpreted as Mg 2ϩ , although no divalent cations were included in the crystallization buffer. Molecular replacement and refinement were done with the program package CNS (18). Model corrections were carried out using the graphics program O (19). Stereochemical correctness of the atomic models was verified by the program PROCHECK (20). Except for residue Cys-418, which is found in all structures in an unusual backbone conformation ( ϭ 59°, ϭ Ϫ109°), all backbone torsion angles are within the allowed regions of the Ramachandran plot. Rendered figures were prepared with BobScript (21,22) and Raster3D (23).
Sequence Comparisons-Using the BLAST(R) 2 program (24), amino acid sequences that are similar ("expect value" Ͻ 100) to our E. coli sequence of PFL were retrieved from the NIH data base (www.ncbi. nlm.nih.gov/blast/). Only sequences that contain two adjacent cysteine residues and the glycyl radical motif were kept and subsequently aligned using the CLUSTAL W program (25). The alignment shows 155 strictly conserved residues.
Coordinates-Structure factor amplitudes and model coordinates for PFLϩpyrϩCoA, PFLϩoxaϩCoA, and PFLϩpyr have been deposited in the Protein Data Bank (accession codes 1h16, 1h17, and 1h18, respectively).

RESULTS
Overview of PFL Structures-We have determined by x-ray crystallography the structure of PFL in complex with both of its substrates CoA and pyruvate at 1.53-Å resolution (R free ϭ 16.3%; coordinate error ϭ 0.13 Å). In addition, we have solved the structure of PFL in complex with pyruvate at 2.3 Å resolution and of PFL with CoA and oxamate at 1.75 Å resolution. All atoms of the structures, except for the four N-terminal residues and very few side chains, were clearly defined by the electron densities. No significant differences in the protein atomic coordinates between these new and the previously reported (10) structures of PFL were detected. Crystallographic data and refinement statistics are compiled in Table I. The structure of PFLϩpyrϩCoA is depicted in Fig. 1b.
Pyruvate Binding Site-Unique positions for the pyruvate atoms were obtained from the 2F o Ϫ F c electron density map of PFLϩpyrϩCoA computed without the pyruvate atoms. Pyruvate binds exclusively in the active site of each protomer ( Fig.  2a) with temperature factors of the same magnitude as those of the surrounding protein atoms. Details of the binding mode of pyruvate are summarized in Fig. 2c. The carbonyl C of pyruvate lies 2.6 Å from Cys-418 S␥, which is less than the van der Waals distance and indicates the formation of a weak bond. However, no significant deviations from planarity at the carbonyl C could be detected. Apparently, pyruvate does not form a thiohemiketal with PFL in the crystal structure.
Analysis of the PFLϩoxaϩCoA data set (Table I) revealed that oxamate binds in the active site of PFL in a mode similar to that of pyruvate (Fig. 2, b and c). Oxamate is an isosteric and chemically inert analog of pyruvate (NH 2 replaces the CH 3 of pyruvate) and was used in a previous study (10) to mimic the pyruvate substrate. Not unexpectedly, we found in our new highly isomorphic structures, PFLϩpyrϩCoA and PFLϩoxaϩCoA, that the differences between equivalent pyruvate and oxamate atomic coordinates are small and amount to a root mean square of 0.46 Å. Allowing for a rigid body rotation and translation (26), oxamate and pyruvate could be superimposed with a root mean square of 0.11 Å. Oxamate is slightly rotated (6°) with respect to pyruvate and translated by 0.4 Å away from Cys-418 so that its carboxamide C atom is found at a This structure is based on previously published data (accession code 3pfl) and was corrected and re-refined in the light of the new high resolution data of this work (see "Experimental Procedures").
b Values in this lable given in brackets refer to the outer shell. c R meas and R mrgd-F as defined by Diederichs and Karplus (30) are quality measures of the individual intensity observations and of the reduced structure factor amplitudes, respectively. d R ϭ ⌺ԽF obs ϪF model Խ⁄ ⌺F obs , where F obs and F model are observed and atomic model structure factor amplitudes, respectively. R free is the R-factor calculated for 5% of randomly chosen reflections, which were excluded from the refinement.
e The numbers refer to atoms or molecules in the asymmetric unit, which contains either one or two PFL protomers for the monoclinic or tetragonal crystals, respectively. a distance of 3.0 Å from Cys-418 S␥. The small differences in the binding distances for pyruvate and oxamate are statistically significant and not artifacts caused by stereochemical restraints of the refinement procedure. This was confirmed by the F o (1) Ϫ F o (2) difference Fourier technique using the refined PFLϩpyrϩCoA structure without pyruvate for phase calculation.
The same small differences between the pyruvate and oxamate binding modes were found by comparison of the PFLϩpyr (Table I) with the re-refined PFLϩoxa structures. Moreover these structures, when compared with their corresponding CoA-containing ones, displayed no significant differences. This excludes the possibility that the binding of CoA caused the observed differences between pyruvate and oxamate interactions with atoms in the active site.
CoA Binding Site-A unique location for each CoA atom was obtained from the 2F o Ϫ F c density map of the PFLϩpyrϩCoA structure computed without the CoA atoms (Fig. 3a). One CoA molecule binds to the surface of each protomer near the subunit-subunit interface of the PFL dimer (Figs. 1b and 3, a and  b). The adenine moiety is located at a distance of 15 Å from the active site with the adenine plane normal pointing toward the protein center. CoA assumes a syn glycosidic torsion angle with the pantetheine chain extending away from the active site toward the opposing monomer where the thiol of CoA is found at a distance of Ϸ30 Å from both active sites. The syn conformation of CoA seen in our structure is rather unusual and has been observed only in the structure of the acyl-CoA-responsive transcription factor FadR (27) (Protein Data Bank entry 1h9g).
The binding site for the adenine of CoA comprises Phe-149, Asn-145, and Gln-146 of a short ␣-helix (Fig. 3, a and b). Phe-149 and Asn-145 are strictly conserved. Remarkably, one side of the phenyl ring of Phe-149 is entirely exposed to solvent in all CoA-free structures. Upon CoA binding, the phenyl ring makes stacking interactions with the imidazole ring of adenine, while Asn-145 and Gln-146 fix the adenine amino group by hydrogen bonds. The interactions of adenine with this site contribute significantly to the binding of CoA. This view is supported by our finding that the adenine base of 5Ј-deoxyadenosine binds in exactly the same mode to this site, while the ribose remains flexible (structural data not shown).
The ribose of CoA adopts the 2Ј-endo conformation. It is fixed by salt bridges between the 3Ј-and 5Ј-phosphates with Lys-161 and a weak hydrogen bond of the ring oxygen with the strictly conserved Arg-160. The pantetheine hydroxyl and the two amide nitrogens form hydrogen bonds to a strongly bound water molecule and the main chain carbonyl oxygens of Phe-149 and Asp-150. The thiol group is located in a predominantly hydrophobic pocket formed by the side chains of residues Phe-200, His-227, Leu-197, and Ala-224 of the opposing monomer.
Analysis of the PFLϩoxaϩCoA data set by the F o (1) Ϫ F o (2) difference Fourier technique revealed that CoA binds in exactly the same way, since no significant difference density near the CoA binding site was found. Replacing pyruvate with oxamate in the active site has no influence on CoA binding. DISCUSSION An earlier proposal for the first half-reaction of the PFL mechanism (14,16) assumed the initial formation of a thiohemiketal between pyruvate and the active site Cys-419. The reaction was thought to proceed by attack of the Cys-418 thiyl radical on the carboxylate C resulting in pyruvate cleavage. The formation of a thiohemiketal was inferred from gel filtration experiments at 4°C which showed a [ 14 C]pyruvate adduct to the non-radical form of PFL (14). Recent structural work on  (Fig. 1a) that precludes thiohemiketal formation and assigns roles to the active site cysteines that differ from the earlier proposal. The new view was based on the assumption that pyruvate would display the same binding mode as the substrate analog oxamate. However, in contrast to pyruvate, oxamate is principally unable to form a thiohemiketal-like compound and for this reason the structural data were insufficient to rule out alternatives to the new view that might involve the formation of a thiohemiketal. The PFLϩpyr and PFLϩpyrϩCoA structures described here are consistent with the new mechanistic proposal. The binding mode of pyruvate is very similar to that of oxamate. Moreover, pyruvate does not form a thiohemiketal, since the molecule does not display any significant deviations from planarity of its carbonyl group despite the close proximity of the carbonyl C to Cys-418 S␥ (Fig. 2a). Apparently, pyruvate is tightly bound in the active site forming a kinetically stable complex with PFL, which could FIG. 2. Structure of the active site with bound pyruvate. a, stereoview of the active site with overlaid electron density for pyruvate contoured at 1.5 (annealed omit map computed without pyruvate and CoA). b, stereoview of the active site with overlaid electron density for oxamate contoured at 1.5 (annealed omit map computed without oxamate and CoA). Oxamate is slightly rotated with respect to pyruvate and translated by 0.4 Å away from Cys-418 so that its carboxamide C atom is found at a distance of 3.0 Å from Cys-418 S␥. c, schematic representation. Distances are between non-hydrogen atoms and given in Å; values in brackets refer to oxamate instead of pyruvate. Strictly conserved amino acid residues are boxed, while conserved residues are enclosed in dashed boxes.
well explain the outcome of the gel filtration experiments (14).
The PFLϩpyr and PFLϩpyrϩCoA structures indicate a weak bond between the C2 of pyruvate and the S␥ of Cys-418 thiol (bond length 2.6 Å), reminiscent of a launched nucleophilic attack on C2. In the radical form of PFL this attack would proceed upon hydrogen abstraction from Cys-418 thiol by Cys-419 thiyl. In the non-radical form the attack cannot proceed. The crystal structures are suggestive of a snapshot of a point on the reaction coordinate when the radical has moved from its storage location at Gly-734 to Cys-419, leaving Gly-734 C␣ as a sp 3 carbon (Fig. 1a, boxed state). At this point the only difference between the radical and non-radical forms of PFL is the absence of a single hydrogen atom at Cys-419 S␥ in the enzyme. Thus, the PFLϩpyr and PFLϩpyrϩCoA crystal structures provide excellent approximations to the corresponding radical enzyme-substrate complexes at the moment prior to generation of the Cys-418 thiyl. The observed structure of the active site, which is identical in the PFLϩpyr and PFLϩpyrϩCoA crystals, is consistent with the radical mechanism of the first half reaction: 1) pyruvate is bound in the active site displaying its planar C2 (sp 2 ) at sub-van der Waals distance to the S␥ of Cys-418 (Fig. 2). The active site is designed to fix pyruvate in a position that prevents the formation of a thiohemiketal (implying a tetrahedral sp 3 configuration at C2), which would render hydrogen abstraction from Cys-418 thiol impossible (2). The thiol of CoA in the PFLϩpyrϩCoA structure is located outside of the active site. Thus, besides the catalytic residues, there is no radical quenching group present in the active site that could interfere with pyruvate cleavage or lead to a loss of the radical.
Consistent with biochemical data, our structures show that the presence of CoA is not required for pyruvate cleavage. If CoA is present during the first half-reaction, it binds in a waiting mode. Upon completion of the first half-reaction PFL must undergo some structural change that allows the thiol of CoA to reach the active site to pick up the acetyl from Cys-418. This requires a new binding mode for the ribose-pantetheine moiety of CoA that differs from that observed in the PFLϩpyrϩCoA structure.
An approximate model for the new CoA binding mode could FIG. 3. CoA binding mode. a, stereoview of the (2F obs Ϫ F calc )-electron density map covering CoA. The map is computed for the PFLϩpyrϩCoA structure and contoured at 1.5 (annealed omit map computed without pyruvate and CoA). b, schematic representation of interactions between CoA and PFL. Distances are between non-hydrogen atoms and given in Å. Strictly conserved amino acid residues are boxed, while conserved residues are enclosed in dashed boxes. Residues of the opposing protomer are marked by asterisks. c, left: arrangement of CoA, pyruvate, and key residues; right: the thiol of CoA could reach Cys-418, Cys-419 within a distance of 5 Å after release of formate to pick up the acetyl if the ribose-pantetheine moiety rotates around the glycosidic bond from the syn to the anti conformation. be obtained if one rotates the ribose-pantetheine moiety around the N-glycosidic bond. This could be visualized as a "fishing model" in which the angler (adenine) stands on a platform (Phe-149 and Asp-150) and moves the fishing rod (ribose-pantetheine moiety rotating from the syn to the anti conformation) through the air (solvent) into the water (protein) so that the hook (CoA thiol) catches the fish (acetyl at Cys-418). Indeed, as shown by model calculations, the ribose-pantetheine moiety can freely rotate through the solvent around the N-glycosidic bond from the unusual syn to the anti conformation so that the thiol reaches the active site (Figs. 1c and 3c). Since 5Ј-Ophosphorylated adenine nucleotides like CoA prefer the anti conformation in solution, the syn/anti transition is favored by a gain in free energy. Presumably, the strictly conserved Arg-160 supports this transition by forming a new salt bridge with the pyrophosphate of CoA in the anti conformation. Moreover, the modeled new binding mode for CoA involves predominantly conserved residues and requires only moderate conformational changes of PFL that do not disrupt secondary structure. We expect that the new CoA binding site formed upon completion of the first half-reaction also allows for binding of a free CoA molecule (anti conformation) that approaches from solution, by-passing the waiting position seen in the crystal structure.
The fixed binding site for the adenine of CoA throughout the reaction cycle proposed here is consistent with biochemical findings (28) that S-ethyl-(8-azido)-CoA and S-ethyl-(8-amino)-CoA, which prefer the syn conformation due to their bulky substituents at the C8 position, are competitive inhibitors (K I values 0.035 and 0.2 mM, respectively) of acetyl-CoA (K m ϭ 0.05 mM (29)). This means that the CoA binding site in the radical form of PFL overlaps with that of the inhibitors.
Taken together our structural data strongly support the proposed radical mechanism (Fig. 1a) for the first half-cycle, provide clear evidence for the adenine binding site of CoA, and suggest how the thiol of CoA could reach the active site, formally replacing formate. It remains undecided by our data whether the presence of the acetyl at Cys-418, the release of formate, or the C␣ configuration of Gly-734 (sp 2 ) triggers formation of the new binding mode for the ribose-pantetheine moiety required for the second half-reaction. Our finding that CoA may adopt a waiting position ready to get involved after pyruvate cleavage is consistent with the indirect group transfer of PFL (ping-pong) mechanism. This could contribute to optimizing the enzyme for the synthesis of acetyl-CoA, a key compound in numerous biochemical pathways.