Crystal Structure and Function of PqqF Protein in the Pyrroloquinoline Quinone Biosynthetic Pathway*

Pyrroloquinoline quinone (PQQ) has received considerable attention due to its numerous important physiological functions. PqqA is a precursor peptide of PQQ with two conserved residues: glutamate and tyrosine. After linkage of the Cγ of glutamate and Cϵ of tyrosine by PqqE, these two residues are hypothesized to be cleaved from PqqA by PqqF. The linked glutamate and tyrosine residues are then used to synthesize PQQ. Here, we demonstrated that the pqqF gene is essential for PQQ biosynthesis as deletion of it eliminated the inhibition of prodigiosin production by glucose. We further determined the crystal structure of PqqF, which has a closed clamshell-like shape. The PqqF consists of two halves composed of an N- and a C-terminal lobe. The PqqF-N and PqqF-C lobes form a chamber with the volume of the cavity of ∼9400 Å3. The PqqF structure conforms to the general structure of inverzincins. Compared with the most thoroughly characterized inverzincin insulin-degrading enzyme, the size of PqqF chamber is markedly smaller, which may define the specificity for its substrate PqqA. Furthermore, the 14-amino acid-residue-long tag formed by the N-terminal tag from expression vector precisely protrudes into the counterpart active site; this N-terminal tag occupies the active site and stabilizes the closed, inactive conformation. His-48, His-52, Glu-129 and His-14 from the N-terminal tag coordinate with the zinc ion. Glu-51 acts as a base catalyst. The observed histidine residue-mediated inhibition may be applicable for the design of a peptide for the inhibition of M16 metalloproteases.

Pyrroloquinoline quinone (PQQ), 4 an aromatic orthoquinone, has been recognized as the third class of redox cofactors in addition to the well known cofactors, nicotinamide (NAD(P) ϩ ) and flavin (FAD, FMN) (1). PQQ was first identified from methanol dehydrogenase in methylotrophic bacteria (2), and several bacterial dehydrogenases, such as glucose dehydrogenases, quinate dehydrogenase, and alcohol dehydrogenase, were later found to be quinoproteins (3)(4)(5). Recently, sugar oxidoreductase in the basidiomycete Coprinopsis cinerea (6), 2-keto-D-glucose dehydrogenase from Pseudomonas aureofaciens (7) and pyranose dehydrogenase from C. cinerea (8) were all characterized as novel PQQ-dependent enzymes. Free PQQ has been identified in a wide variety of foods (9) and milk (10). PQQ is an essential nutrient for proper growth and development in mice (11). The strong-antioxidant capacity of PQQ protects living cells and biomolecules from oxidative stress in vivo and in vitro (12,13). Furthermore, PQQ exerts a protective effect against ultraviolet irradiation-induced human dermal fibroblast senescence in vitro (14) and suppresses the serum low density cholesterol level to prevent various diseases (15). In addition, PQQ may improve skin conditions and slow the progression of osteoarthritis (16).
The chemical structure of PQQ was determined in 1980 (17). Since then gene clusters involved in the synthesis of PQQ from different bacteria that range from 4 genes in Acinetobacter calcoaceticus to 11 genes in Pseudomonas fluorescens B16 have been gradually identified (18 -21). Most of the clusters are organized as six or seven genes (pqqABCDEF/G) in an operon. However, the PQQ biosynthetic pathway was unknown until a putative biosynthetic pathway was proposed based on the function of conserved genes (pqqA-F) in Klebsiella pneumoniae in 2008 ( Fig. 1) (22). In this pathway PqqA was determined to be a precursor peptide with two conserved residues: glutamate and tyrosine (23,24). PqqE is the only enzyme in the pathway with significant sequence similarity (16% sequence identity to molybdenum cofactor biosynthesis protein A) to radical S-adenosylmethionine proteins, which are capable of catalyzing C-C bond formation (25) and may be responsible for the formation of the C-C bond at atoms C9 and C9a of PQQ (26). PqqD, a peptide chaperone, forms a ternary complex with the radical S-adenosylmethionine protein, PqqE, and assists PqqE in the modification of PqqA (27). PqqF shares 15.7% of its sequence identity with human insulin-degrading enzyme (IDE), which is capable of cleaving different peptide bonds. This finding suggests that PqqF cleaves the four peptide bonds at R1 and R2 of glutamate and R2 and R3 of tyrosine (22). After the linkage of C␥ of glutamate and CE of tyrosine in the precursor PqqA by PqqE and proteolytic cleavage of the four peptide bonds by PqqF, the amino group of Glu and the OH group (C5a) of Tyr are primed to spontaneously form a Schiff-base. PqqC catalyzes the final step of the reaction, which involves a ring closure at N1 and the removal of eight electrons and eight pro-tons from 3a-(2-amino-2-carboxyethyl)-4,5-dioxo-4,5,6,7,8,9hexahydroquinoline-7,9-dicarboxylic-acid to form PQQ (28). Finally, the PQQs are transported by PqqB to the periplasm, where the bacterial dehydrogenases reside (24,29,30). Moreover, a bioinformatics analysis also indicated that the oxidation of the tyrosine of PqqA before its cross-linking with glutamate may be performed by PqqB (31).
PqqF, which possesses an inverted zinc binding motif (HXXEH), belongs to the M16 family of the clan ME peptidases. M16 peptidases can be further categorized into three subfamilies (M16A-C) based on their primary structure. The A and C subfamilies comprise tandem pairs of ϳ500-residue domains in a single polypeptide chain (32,33). The N-and C-terminal domains are joined by a short hinge loop in M16A or by a helical hairpin in M16C. The N-terminal domain is the catalytic domain, but the C-terminal domain is also essential for the activity (33). These two domains form a clamshell shape in all structures of known M16 family members. The closed and open conformation of clamshell could be observed in the structures. IDE, a M16A protein, was crystallized in a closed clamshell with substrate locked inside its central cavity (33), whereas another M16A protein, E. coli pitrilysin (PDB ID 1Q2L), was crystallized in an open conformation in the absence of peptide. M16B enzymes are the heterodimers composed of two separate subunits (34,35), and the ␣ subunit of the M16B enzymes consists of a glycine-rich loop in the C-terminal domain that protrudes into the active site (36). An Arg/Tyr pair, which plays an important role in substrate binding, is conserved in the C-terminal domain among the M16 enzymes lacking the glycine-rich loop (32,33).
Although the structures of PqqB, PqqC, and PqqD, which are involved in PQQ biosynthesis, have been solved, the initial steps of the PQQ biosynthesis pathway remain unknown (30,37,38). However, the pqqF gene of K. pneumoniae has been shown to be essential for PQQ biosynthesis in E. coli and E. coli pts (phosphotransferase system) mutants on minimum medium (39).
Prodigiosin is a secondary metabolite produced by Serratia marcescens, actinomycete, and other Gram-negative bacteria (40). Prodigiosin has attracted considerable attention due to its anti-fungal/anti-bacterial, immunosuppressive, and anticancer activities (41,42). However, the biologically active pigment prodigiosin is inhibited by growth in glucose-rich medium (43). Quinoprotein glucose dehydrogenase uses PQQ as a cofactor, and its substrates are D-glucose and ubiquinone, a component of the electron transport chain (44,45). S. marcescens glucose dehydrogenase (GdhS) was previously shown to require PQQ for glucose dehydrogenase activity (46). Because GdhS is necessary for the glucose-mediated inhibition of the prodigiosin phenotype (GIP) and GdhS requires PQQ, it is expected that mutations in the PQQ biosynthetic genes would impair a GIPdefective phenotype. Additionally, previous studies have indicated that this inhibitory effect is pH-dependent; the presence of glucose dehydrogenase and D-glucose would result in the accumulation of gluconic acid and cause a pH drop of the media, and this drop of the pH would then inhibit the production of prodigiosin by Serratia (47).
In this study we constructed a deletion mutant of the pqqF gene and found that deletion of the pqqF gene eliminated the glucose-mediated inhibition of prodigiosin in Serratia sp. FS14. Furthermore, we present the crystal structure of PqqF and propose a catalytic mechanism of PqqF via a structural comparison of PqqF with other M16 family proteins.

Results
Deletion of pqqF Eliminated the Inhibition of Prodigiosin Production via D-Glucose in Serratia sp. Fs14 -The pqqF gene is located in the PQQ gene clusters, and its function in PQQ biosynthesis has not been well established in Serratia. To study the importance of the pqqF gene, a pqqF gene-deletion mutation was constructed, and the pH and amount of prodigiosin in glucose-rich medium were then determined. The prodigiosin and pH measurements revealed that the glucose-mediated inhibition of the prodigiosin production is evident in Serratia sp. FS14. Wild-type FS14 and FS14⌬pqqF produced the same amount of prodigiosin in glycerol peptone broth. However, the wild-type FS14 strain yielded a relatively lower amount of prodigiosin in the presence of glucose, whereas the FS14⌬pqqF strain generated the same amount of prodigiosin regardless of the presence of glucose ( Fig. 2A). The pH of the medium of the FS14⌬pqqF cultures that were grown in glycerol peptone was identical with that of the wild-type FS14 strain. The pH of the wild-type FS14 cultured in glycerol peptone broth supplemented with 2% glucose (w/v) decreased to 3.5 during the first 48 h. However, the pH of FS14⌬pqqF grown in the same glycerol peptone glucose medium slowly decreased during the first 12 h and began to rapidly increase from 12 h to 36 h (Fig. 2B). To assess whether the loss of glucose-mediated inhibition of prodigiosin was caused by the pqqF mutation, the wild-type FS14 that contained an empty vector, FS14⌬pqqF, which contained an empty vector, and FS14⌬pqqF, which contained pqqF-pWDX, were prepared for complementation experiments. The FS14⌬pqqF strain that contained the empty vector produced prodigiosin first. The prodigiosin levels of both the wild-type FS14, containing an empty vector, and FS14⌬pqqF, containing the pqqF-pWDX strains, were undetectable when cultured in glycerol peptone glucose broth (Fig. 2C). During growth in glycerol peptone glucose broth, the pH of the FS14⌬pqqF strain that contained the pqqF-pWDX and wild-type FS14 strain that contained the empty vector decreased to 3.0 gradually. In contrast, the pH trend obtained for FS14⌬pqqF strain, which contained the empty vector, was coincident with the pH trend obtained in the mutant trials; the pH values decreased during the first 12 h and then increased rapidly (Fig. 2D). These results suggested that mutation of the pqqF gene eliminated the glucose-mediated prodigiosin inhibition phenotype, and pqqF is essential for the biosynthesis of PQQ in Serratia sp. FS14.
Overall Structure of PqqF-The crystal structure revealed that PqqF is a dimer of two structurally identical monomers with the backbone root mean square deviation (r.m.s.d.) of 0.42 Å. The overall structure of the PqqF monomer exhibits a closed clamshell-like shape, and the 14-amino acid-residue-long tag formed by the N-terminal affinity tag of the pET28a vector precisely protrudes into the counterpart active site (Fig. 3A). The topology of the PqqF monomer is presented in Fig. 3B. Each monomer includes four structurally similar domains, as is often observed in M16 peptidases. Starting from the ␤ strand (␤1, ␤7, ␤14, and ␤19 for domains 1-4, respectively), each domain is connected to another domain via the contacts between one ␣ helix and one ␤ strand. The N-terminal lobe consists of domains 1 (residues 1-220) and 2 (residues 221-406), and the C-terminal lobe contains domains 3 (residues 439 -604) and 4 (residues 610 -768). These two lobes are then linked by a 34amino acid-long loop between domains 2 and 3. The structure of the loop linking these two functional lobes indicates that PqqF belongs to the M16A subfamily. PqqF-N and PqqF-C, each one-half of the closed clamshell, form a cavity with the volume of ϳ9400 Å 3 . All M16 family members have a similar organization in which each lobe forms one-half of a clam-shell. The helices contribute to the external surface of the PqqF structure. In contrast, the internal surface is lined with ␤ strands.
A protein interface analysis showed that the two monomers form a stable dimer. 228 residues formed the interaction interface, and 50% of them are hydrophobic. The dimer interface buries 8070 Å 2 , accounting for 12% of the total accessible surface area of each dimer. The interaction involves 42 hydrogen bonds and 13 salt bridges per dimer. Because the affinity tag is from the vector, to check whether the dimer is still stable without tags in the structure, we removed the tag from the structure and analyzed the interface. The result showed that 53 residues from each of the two monomers formed the interaction interface and that Ͼ60% of them are hydrophobic. The dimer interface buries 4031 Å 2 or 6.3% of the total accessible surface area of each dimer. The interaction involves 21 hydrogen bonds and 4 salt bridges per dimer resulting in a stable dimer. The above results suggest that the N-terminal tag from the expression vector contributes to the formation of a more stable dimer.
Active-site Structure-The active site of PqqF is located in the zinc binding motif, which is also commonly observed in other M16 family proteases. His-48 and His-52 of the zinc binding motif HXXEH and Glu-129, located in helix5, coordinate with the zinc ion. Glu-51 of the zinc binding motif acts as a base catalyst. Unexpectedly, the electron density suggests that the zinc ion is also coordinated by the free N-terminal nitrogen atom and the imidazole side chain of His-14, which was introduced by the expression vector (Fig. 4A). The carboxyl oxygen atom of Glu-122 is highly conserved in M16 family enzymes and forms a hydrogen bond with the ND1 atom of His-52 at a distance of 2.6 Å. The interaction of Glu-122 with His-52 restricts the protonation state of His-52 such that the lone pair of ND1 is exposed, which is suitable for metal ion binding. The NH1 atom of R56 forms a hydrogen bond with the OE2 atom of Glu-122 to reinforce the interaction of Glu-122 with His-52.
The carbonyl oxygen atom of Ile-160 interacts with the ND1 atom of His-48 in the same manner at a distance of 2.7 Å (Fig.  4B). The active site of PqqF is situated in the negatively charged cavity, and the remaining portions of the PqqF-N and PqqF-C lobes are primarily positively charged (Fig. 5).
Interaction of the N-terminal Peptide with PqqF-In total, 23 residues (MGSSHHHHHHSSGLVPRGSHMAS) in the N terminus of PqqF were introduced during the construction of the recombinant plasmid pET28a-pqqF. However, the electron density showed that only a peptide of 14 residues (HSS-GLVPRGSHMAS) is located at the N terminus of PqqF; thus, the first nine residues were not observed in the structure. Because the PqqF protein was purified by nickel affinity chromatography, the N-terminal 23 residues of PqqF were supposedly present in the purified protein, so the missing 9 residues may have been removed by PqqF during the crystallization process.
The model of the N-terminal peptide suggests several potential interactions with PqqF. The residues Glu-129, Gln-666, Leu-78, Glu-24, Glu-739, Val-665, and Arg-656 are in position to potentially form hydrogen bonds with the N-terminal peptide (Table 1). Hydrogen bonds and salt bridges are present between His-14 and Glu-129, and coordination bonds are noted between His-14 and the zinc ion (Fig. 4C). Additionally, the B-factor of the remaining N-terminal residues was also analyzed, as shown in Table 2. The B-factor of His-14 (56.5 Å 2 ), which is similar to that of its coordinating zinc ion (55.0 Å 2 ), is much lower than that of the remaining 13 residues from the N-terminal tag ( Table 2). These analyses indicated that the binding affinity of His-14 with the zinc ion is obviously stronger than the peptide-peptide interaction.
Structural Comparison-A structural similarity search that was performed with the atomic coordinates of PqqF using the DALI server yielded the human IDE (PDB ID 2G54) (33), E. coli pitrilysin (PDB ID 1Q2L), a putative zinc protease (PDB ID 3EOQ) (34), the mitochondrial processing peptidase ␣ subunit (PDB ID 1HR6) (35), a probable zinc protease (PDB ID 4NNZ), a peptidase M16 inactive domain family protein (PDB ID 3GWB), and ubiquinol-cytochrome-c reductase complex core (PDB ID 3TGU) (48) ( Table 3). The alignment of PqqF with these resolved structures suggests that the homology of PqqF with other M16 family proteins is quite low. The r.m.s.d. of the superimposed main chains ranged from 3.0 to 4.0. For instance, with the highest Z-score, PqqF shares only 16.1% sequence identity, and the r.m.s.d. are 3.93 with E. coli pitrilysin. IDE exhibits the second highest similarity with the PqqF structure. Approximately 86% of the residues of PqqF were aligned with IDE, yielding 15.7% sequence identity. Ͻ400 residues of PqqF, which is just a little bit more than half of the PqqF residues, were aligned with the M16B and M16C proteases.
Interestingly, although different M16 family members exhibit a very low sequence identity and a very high main chain r.m.s.d., these proteins display the same fold. Each monomer contains four structurally similar domains with a similar topological structure. Specifically, a ␤-sheet with four to six strands forms the internal face of each domain; two to four ␣-helices are packed against the other face of the ␤-sheet and between two and four additional ␣-helices are situated at the other end of the sheet. In addition, an ␣ helix is located between the sheets of the two domains in the N-or C-terminal lobes. Both E. coli pitrilysin and IDE belong to the same M16A family and share a slightly higher sequence identity with PqqF. The N-and C-terminal lobes of PqqF form a close clamshell-like structure. A comparison of the PqqF structure with those of E. coli pitrilysin (PDB accession number 1Q2L) and IDE (PDB accession number 2G54) was performed. E. coli pitrilysin adopts an open substrate-free conformation (Fig. 6A). The IDE structure that included substrate binding was closed (Fig. 6B). The structural alignment of the PqqF with IDE and E. coli pitrilysin showed that 667 residues of PqqF aligned with IDE with a backbone r.m.s.d. of 3.5 Å, but only 433 residues aligned with pitrilysin with a backbone r.m.s.d. of 3.9 Å due to its different conformation (pitrilysin is found in an open conformation).
Despite the similarities described above, PqqF and IDE exhibited significant structural differences. PqqF has a smaller cavity than IDE. We compared the PqqF-N and PqqF-C lobes with the IDE-N and IDE-C lobes, as shown in Fig. 6, C and D, respectively. Compared with the structure of IDE, helices 17,19,20,24, and 27 of the PqqF-C lobe and helices 3, 4, 10, 11, 12, 13, 14, 15, and 16 of the PqqF-N lobe moved toward the internal chamber, and helices 11, 13, and 14, particularly, moved by a distance of 7-9 Å toward the center cavity. The distance between helices ␣15 and ␣17 of PqqF and the corresponding helices of IDE was found to be 5-6 Å. All strands of PqqF also rotated toward the center, and ␤8, ␤9, ␤10, and ␤11 of the N-terminal lobe shifted and rotated. ␤16 and ␤17 in the C-terminal lobe of PqqF even rotated at an angle of ϳ90°. Only helices 2, 5, and 24 among all of the 27 helices and 22 strands of PqqF were offset to the outside at a distance of ϳ3 Å as compared with IDE. Thus, compared with that of IDE, the overall structure of PqqF was shrunk, resulting in a markedly smaller catalytic chamber. Helix 2 harbors the zinc binding motif and is situated in domain 1. The superimposed result of domain 1 that primarily bears the catalytic activity of PqqF, pitrilysin, and IDE shows that the fold of domain 1 of PqqF is similar to that of domain 1 in the other two structures (Fig. 6E). Furthermore, the active sites of the three proteins are aligned, as shown in Fig. 6F. The active site of E. coli pitrilysin is the typical catalytic center of the M16 family proteases, which involves the coordination of a zinc ion with two histidines of HXXEH, a glutamate and a water molecule. The Tyr-16 of the substrate insulin B chain of IDE replaced the water molecule. Instead of a water molecule, His-14  JULY 22, 2016 • VOLUME 291 • NUMBER 30 was observed in our structure. In addition, the active site of the PqqF structure was compared with that of E. coli pitrilysin and the IDE, and the results revealed that PqqF also contains the conserved Arg-656/Tyr-663 pair. However, the side chain of Tyr-663 was opposite from the IDE conserved residue Tyr-831.

D-Glucose-mediated Inhibition of Prodigiosin Production-
The entire genome of Serratia sp. FS14 has a pqq cluster that consists of pqqA-F. However, the pqqF gene is absent in some bacteria. Further study is required to determine whether the pqqF gene is essential to PQQ biosynthesis. The D-glucose-mediated inhibition of GIP is commonly observed in Serratia, and GIP is based on pH reduction. No prodigiosin is produced at a pH below 4.0, and the inhibition of prodigiosin production is correlated with the low pH (49). Furthermore, quinoprotein glucose dehydrogenase is the major factor in GIP, and the genes involved in the biosynthesis of PQQ are important in GIP (47). Here, we determined the prodigiosin production and pH of cultures of FS14⌬pqqF and wild-type FS14 strains, and no GIP effect was observed with FS14⌬pqqF grown in glucose-rich medium. FS14⌬pqqF containing the empty vector produced a significant amount of prodigiosin. However, wild-type FS14 strain containing the empty vector and mutant FS14 containing the pWDX-pqqF plasmid did not produce prodigiosin throughout the studied time course. Moreover, the pH of the medium of both wild-type FS14 containing the empty vector and FS14⌬pqqF containing the pWDX-pqqF plasmid decreased to below 3.0. Thus, glucose caused a low pH, and the low pH inhibited cell growth and resulted in a lack of prodigiosin production. The data from the complementary trials are consistent with the mutant study results, indicating that the significant increase in the culture pH is coincident with the prodigiosin production and that the pqqF gene is essential to the PQQ biosynthesis under our experimental conditions (Fig. 2).
Substrate Specificity of PqqF-The length of the remaining N-terminal tag was exactly the distance between the active site and the dimer interface. Thus, His-14 was located in the active site, and the peptide bond between H-15 and His-14 was hydrolyzed by PqqF given the length of the N-terminal peptide. The peptide bond between the two histidines of the N-terminal tag was cleaved by PqqF, suggesting that the cleavage sites of PqqF may be not very specific. This finding is also consistent with the proposed mechanism that all four peptide bonds within PqqA are cleaved by PqqF. PqqF-N and PqqF-C lobes, which comprise one-half of a closed clamshell, form a chamber with the volume of ϳ9400 Å 3 . The cavity volumes of both IDE and AtPreP are similar in size (ϳ13000 Å 3 ), and the substrate can be trapped in the chambers for hydrolysis (32,33). In contrast, the chamber volume (ϳ9400 Å 3 ) of PqqF is markedly smaller compared with that of the IDE, and this smaller chamber of PqqF can only accommodate a relatively small substrate. This finding is consistent with the fact that PqqA, which is a 25-residue polypeptide, is smaller     (32,33). The M16A protein IDE is crystallized in a closed clamshell with the substrate locked inside its central cavity, whereas another M16A protein, E. coli pitrilysin (PDB: 1Q2L), is crystallized in an open conformation in the absence of the substrate peptide. The PqqF structure was closed even though no natural substrates were added to the protein solution during the crystallization trials. However, we observed that the 14-residue tag of each monomer was inserted into the active sites of other monomer in the dimer, which caused the PqqF structure to adopt a closed conformation. We failed to obtain crystals from the crystallization screening of PqqF, which was purified using the vector containing a His tag at the C terminus of PqqF. It is possible that the remaining N-terminal tag facilitated the crystallization of the PqqF protein and that the closed form of PqqF is more stable.

Structure and Function of PqqF
Inhibition Mechanism of the N-terminal Peptide-At the catalytic site, the free N-terminal nitrogen atom and the imidazole side chain of His-14 completed the zinc coordination sphere that was formed by His-48, His-52, and Glu-129 (Fig. 3A). Binding to this site reduced the affinity for the substrate, resulting in the observed one-to-one binding stoichiometry. The substrate cannot adequately bind to the catalytic site, resulting in inactivity. Additionally, the pseudo-substrate interacts with residues from both the N-and C-terminal lobes, maintaining the enzyme in a closed conformation. The additional N-terminal tag not only occupies the active site but also stabilizes the closed, inactive conformation. Therefore, this additional N-terminal tag ''locks" the protease in the closed and inactive conformation. The active site of PqqF that contains the zinc binding motif is typical of the M16 family. Furthermore, histidine may play a main role in the combination of PqqF with the N-terminal tag, as shown by B-factor analysis. The coordination bonds and the dimer interactions combine the N-terminal tag in the cavity. The peptides, which consist of a histidine at the N terminus and its analogs, may effectively inhibit the activity of the M16 proteases. IDE, a M16A member of the Zn 2ϩ -metal-loprotease family, cleaves various substrates, including insulin, amyloid-␤, insulin-growth factor-II (IGF-II), glucagon, somatostatin, ubiquitin, and amylin and transforming growth factor ␣ (50, 51). Moreover, the Ide gene has been associated with type-2 diabetes and Alzheimer disease in humans (52,53). After the discovery of IDE, pharmacological inhibitors of IDE have attracted considerable attention. The first substrate-based zinc binding hydroxamate-based inhibitors of IDE were described by Leissring et al. (54). However, their hydroxamate group combined with an arginine residue limits their use as pharmacological probes. Additionally, the compounds binding to the catalytic site and to a distal exosite in the N-terminal lobe have been reported to inhibit IDE (55,56). Recently, a novel class of IDE inhibitors was discovered, but these compounds do not interact with the catalytic zinc of the protease. Instead, the inhibitors bind to the dual exosite of IDE (57). The co-crystallization of the N-terminal tag acting as a pseudo-substrate with PqqF provides a novel idea for future designs of IDE inhibitors.
According to sequence and function conservation, Arg-656 and Tyr-663 of PqqF were found to be equivalent to Arg-824 and Tyr-831 of IDE, which are important for substrate binding in the C-terminal lobe (33) to complete the active site. However, the structural alignment indicates that the direction of the side chain of Tyr-663 is opposite to that of Tyr-831. This observation may result from the fact that the tag that binds to the PqqF cavity is not its native substrate. The phenolic side chain of Tyr-663 was repelled in the reverse direction by the tag introduced from the expression vector. The conserved Arg/Tyr pair of E. coli pitrilysin was not in the same location because the structure of E. coli pitrilysin was in an open conformation.
The docking of PqqF with PqqA was attempted to analyze the interaction of PqqF with PqqA. However, no favorable conformation was available because the overall structure of PqqA did not match with the cavity of PqqF. PqqA may be further modified by PqqD, a peptide chaperone that is recognized by PqqE.

Experimental Procedures
Construction of Deletion Mutant FS14⌬pqqF and Complementary FS14 pqqF-pWDX-Serratia sp. FS14, which was isolated by our laboratory, belongs to the Serratia genus. The original chromosomal mutant genes of pqqF (⌬985-1875) were produced through digestion of the internal restriction sites of pqqF gene fragment with the restriction enzyme PstI. Suicide plasmid pWDF (a pKNG101 (58) derived kanamycin-resistant suicide plasmid constructed by our laboratory) and the helper plasmid pRK2013 were used for the mutant construction and the pWDX plasmid (which was derived from pET28a and pTrc99a lac promoter constructed by our laboratory) was used for complementation of the mutation in Serratia sp. FS14. The gene fragments obtained after digestion were ligated and then cloned into the suicide vector pWDF, and the resultant plasmids were transferred to the wild-type FS14 through triparental conjugation. Wild-type FS14 and mutant FS14⌬pqqF were grown at 301 K in LB broth. E. coli strains DH5␣ and DH5␣ () were grown at 310 K in LB broth. E. coli cells containing the pWDF or pWDX plasmid or the helper plasmid pRK2013 and recipient Serratia cells (wild-type FS14 or FS14⌬pqqF) were grown to the exponential phase in LB medium. The cells were then mixed at a ratio of 1:1:1 and washed twice with LB medium before being spotted onto an LB plate and incubated at 301 K for 24 h. The conjugants were selected after spreading the spots incubated overnight in the conjugation plates onto LB plates containing 60 g ml Ϫ1 kanamycin and 15 g ml Ϫ1 tetracycline. Mutants were then selected using the method described by Muhl and Filloux (59). The recombinant pWDX-pqqF plasmid and empty pWDX vector were transferred into mutant FS14⌬pqqF and wild-type FS14 cells to serve as com- plement and control, respectively. The constructs were verified by PCR amplification and sequencing. Prodigiosin and pH Measurements-Time course experiments were performed as follows: five overnight cultures from single colonies of wild-type FS14, FS14⌬pqqF, wild-type FS14 containing the empty vector pWDX, FS14⌬pqqF, containing the empty vector pWDX, and FS14⌬pqqF, containing pqqF-pWDX strains were grown in 3 ml of LB broth and subcultured in 50 ml of LB broth; finally they were simultaneously transferred to 50 ml of glycerol peptone broth and glycerol peptone glucose broth (2% glucose (w/v), 1% glycerol(v/v), and 1.5% peptone (w/v)). Wild-type FS14 containing the empty vector pWDX, FS14⌬pqqF, containing the empty vector pWDX, and FS14⌬pqqF, containing pqqF-pWDX were transferred to glycerol peptone glucose broth to a final A 600 of 0.050 for prodigiosin and pH measurements. Kanamycin at a final concentration of 30 g ml Ϫ1 was added into the broth to prevent loss of pWDX plasmid in the complement trials, and 0.03 mM isopropyl-␤-D-thiogalactopyranoside was added to the cultures of FS14⌬pqqF containing pqqF-pWDX to induce the expression of PqqF protein. Three parallel controls were prepared for every strain, and 3-ml cultures were taken out for measurements of the prodigiosin concentration and pH at 6,12,24,36,48,60, and 72 h. Prodigiosin concentration was measured at 534 nm with an extinction coefficient of 7.07 ϫ 10 4 from liquid cultures as previously described (60). The pH was determined from the supernatant using Sartorius pH analyzer (model PB-10).
Cloning, Expression, and Purification-The pqqF gene was amplified by PCR using the genomic DNA of Serratia sp. FS14 as template with the primers 5Ј-aatagctagcATGACACTCGC-CGCTTCGTGG and 5Ј-tatctcgagTCAGCCGGCAACGAA-CAGC. The restriction sites of endonucleases NheI-XhoI of pqqF (bases underlined) were created by the primers. Both the amplified pqqF gene and the expression vector pET28a were digested with NheI-XhoI and then ligated with T4 DNA ligase (Thermo). The constructed expression plasmid, which was verified by restriction enzyme digestion and DNA sequencing, was transformed into E. coli strains C43 (DE3) for protein expression. The resultant protein contained an N-terminal His 6 tag. Protein expression in E. coli C43 (DE3) cells cultured in Luria-Bertani (LB) medium supplemented with kanamycin (30 g ml Ϫ1 ) was induced with 0.1 mM isopropyl-␤-D-thiogalactopyranoside when the A 600 reached 1.0. After 3 h of incubation at 301 K, the cells were harvested by centrifugation at 4425 ϫ g for 6 min. The cell pellet was resuspended and disrupted in lysis buffer consisting of 50 mM Tris-HCl (pH 8.8), 0.3 M NaCl, 5 mM imidazole, 10% glycerol (v/v), and 1 mM PMSF added just before sonication. The insoluble fraction was removed by centrifugation at 24,000 ϫ g for 30 min at 277 K. The recombinant protein was purified by affinity chromatography with Ni-NTA Superflow resin (Qiagen). The column was washed sequentially with 6 bed volumes of lysis buffer containing 5 mM and 20 mM imidazole. The protein was eluted with lysis buffer containing 50 mM imidazole. The eluted protein was concentrated to 2 ml by centrifugation with an Amicon Ultra-15 filter (Millipore) and further purified with a 150-ml Superdex 200 (GE Healthcare) gel filtration column that was pre-equilibrated with 20 mM Tris-HCl (pH 8.8), 300 mM NaCl, and 10% glycerol (v/v). The protein was eluted with the same buffer. The peak pattern illustrated two oligomerization states. Because the first peak eluted in the void volume, the fractions of the second peak were collected and concentrated to 10 mg ml Ϫ1 for crystallization. Selenomethionine-labeled PqqF was produced using the metabolic inhibition method (61). The purification method using for SeMet-PqqF was the same as that used for native PqqF with the exception that 10 mM ␤-mercaptoethanol was added to all of the buffers used to prevent selenium oxidization.
Crystallization-The purified recombinant PqqF was used to perform the crystallization trials by the sitting-drop vapor diffusion method with the protein concentrations of 5 mg ml Ϫ1 and 10 mg ml Ϫ1 at 277 K and 293 K at the same time. Crystallization conditions used were commercial screening kits from Hampton Research (PEGRx TM 1-2, Crystal Screen TM , Crystal Screen 2 TM , Crystal Screen Cryo TM , Crystal Screen Lite TM , Index TM ) and Microlytic (MCSG1-4). The crystal drops containing 1 l of protein and 1 l of reservoir solution were equilibrated against 50 l of reservoir solution. The preliminary crystals were obtained in 35% Tacsimate at pH 7.0 and 277 K. The crystal quality was significantly improved when the crystallization was performed at 289 K with the addition of 10 mM ␤-mercaptoethanol. Then, microseeding method was adopted to generate single crystals suitable for diffraction. Crystal solution was sequentially diluted to the concentrations of 10 Ϫ1 , 10 Ϫ2 , 10 Ϫ3 , 10 Ϫ4 , 10 Ϫ5 , and 10 Ϫ6 to prepare as the seed. At the same time, the protein concentration of 5 mg ml Ϫ1 and 10 mg ml Ϫ1 and precipitant concentration of 30, 31, 32, and 33% were also altered to render the solution lower supersaturated where only growth was supported. Using 10 Ϫ3 and 10 Ϫ4 dilutions as the seeds, single crystals were obtained in 30% Tacsimate (pH 7.0) with a protein concentration of 5 mg ml Ϫ1 via microseeding under the above-mentioned conditions. Because PqqF shares very low sequence identity (Ͻ20%) with solved crystals structures, selenomethionine-labeled PqqF was produced to obtain crystals for determining the phase. A selenomethionine variant of PqqF was crystallized as the recombinant wild-type protein using the sitting-drop vapor diffusion method at 289 K. The crystallization drops contained 1 l of protein (5 mg ml Ϫ1 in 20 mM Tris-HCl buffer (pH 8.8), 300 mM NaCl, 10%(v/v) glycerol, and 10 mM ␤-mercaptoethanol) and 1 l of reservoir solution (30% Tacsimate (pH 7.0)) equilibrated against 50 l of well solution (30% Tacsimate (pH 7.0) and 2 mM DTT). For data collection, the crystals were equilibrated in reservoir buffer containing 20% glycerol and flash-frozen in liquid nitrogen.
X-ray Diffraction Data Collection and Structure Determination-Diffraction data sets were collected at 100 K on the beamline BL19U of Shanghai Synchrotron Radiation Facility, Shanghai, China using an Pilatus 6M detector. The dataset for the selenomethionine crystal was processed using the HKL3000 to the resolutions of 2.8 Å (62). 20 selenium atoms of a total 22 were successfully located by SHELXD (63), and the phenix.autosol in the PHENIX software package (64) was used to refine the locations of substructures, calculate the initial phases, and make density modification, non-crystal symmetry improvement, and model auto-building. After the initial auto model building, 1164 of a total of 1584 residues were success-fully located. The initial model was then refined against a native dataset at a resolution of 2.5 Å. The final model was manually completed by the program Coot (65) and with several rounds of refinement using phenix.refine and refmac5 (66). Statistics for data collection and structure refinement are summarized in Table 4. The figures showing the protein structures were generated using PyMOL (67).
Author Contributions-D. X. and W. W. conceived the study. Q. W., T. R., and C. M. conducted the experiments. All the authors analyzed the data. Q. W., W. W., D. X., and T. R. wrote the manuscript.