We propose a reinforcement learning based approach to query object localization, for which an agent is trained to localize objects of interest specified by a small exemplary set. We learn a transferable reward signal formulated using the exemplary set by ordinal metric learning. Our proposed method enables test-time policy adaptation to new environments where the reward signals are not readily available, and outperforms fine-tuning approaches that are limited to annotated images. In addition, the transferable reward allows repurposing the trained agent from one specific class to another class. Experiments on corrupted MNIST, CU-Birds, and COCO datasets demonstrate the effectiveness of our approach.

ABSTRACTHuman α‐amino‐β‐carboxymuconate‐ε‐semialdehyde decarboxylase determines the fate of tryptophan metabolites in the kynurenine pathway by controlling the quinolinate levels for de novo nicotinamide adenine dinucleotide biosynthesis. The unstable nature of its substrate has made gaining insight into its reaction mechanism difficult. Our electron paramagnetic resonance (EPR) spectroscopic study on the Cu‐substituted human enzyme suggests that the native substrate does not directly ligate to the metal ion. Substrate binding did not result in a change of either the hyperfine structure or the super‐hyperfine structure of the EPR spectrum. We also determined the crystal structure of the human enzyme in its native catalytically active state (at 1.99 Å resolution), a substrate analogue‐bound form (2.50 Å resolution), and a selected active site mutant form with one of the putative substrate binding residues altered (2.32 Å resolution). These structures illustrate that each asymmetric unit contains three pairs of dimers. Consistent with the EPR findings, the ligand‐bound complex structure shows that the substrate analogue does not directly coordinate to the metal ion but is bound to the active site by two arginine residues through noncovalent interactions. Proteins 2015; 83:178–187. © 2014 Wiley Periodicals, Inc.

The previously reported crystal structures of α-amino-β-carboxymuconate-ε-semialdehyde decarboxylase (ACMSD) show a five-coordinate Zn(II)(His)(3)(Asp)(OH(2)) active site. The water ligand is H-bonded to a conserved His228 residue adjacent to the metal center in ACMSD from Pseudomonas fluorescens (PfACMSD). Site-directed mutagenesis of His228 to tyrosine and glycine in this study results in a complete or significant loss of activity. Metal analysis shows that H228Y and H228G contain iron rather than zinc, indicating that this residue plays a role in the metal selectivity of the protein. As-isolated H228Y displays a blue color, which is not seen in wild-type ACMSD. Quinone staining and resonance Raman analyses indicate that the blue color originates from Fe(III)-tyrosinate ligand-to-metal charge transfer. Co(II)-substituted H228Y ACMSD is brown in color and exhibits an electron paramagnetic resonance spectrum showing a high-spin Co(II) center with a well-resolved (59)Co (I = 7/2) eight-line hyperfine splitting pattern. The X-ray crystal structures of as-isolated Fe-H228Y (2.8 Å) and Co-substituted (2.4 Å) and Zn-substituted H228Y (2.0 Å resolution) support the spectroscopic assignment of metal ligation of the Tyr228 residue. The crystal structure of Zn-H228G (2.6 Å) was also determined. These four structures show that the water ligand present in WT Zn-ACMSD is either missing (Fe-H228Y, Co-H228Y, and Zn-H228G) or disrupted (Zn-H228Y) in response to the His228 mutation. Together, these results highlight the importance of His228 for PfACMSD's metal specificity as well as maintaining a water molecule as a ligand of the metal center. His228 is thus proposed to play a role in activating the metal-bound water ligand for subsequent nucleophilic attack on the substrate.

This review examines the mechanisms propelling cofactor-independent, organic cofactor-dependent and metal-dependent decarboxylase chemistry. Decarboxylation, the removal of carbon dioxide from organic acids, is a fundamentally important reaction in biology. Numerous decarboxylase enzymes serve as key components of aerobic and anaerobic carbohydrate metabolism and amino acid conversion. In the past decade, our knowledge of the mechanisms enabling these crucial decarboxylase reactions has continued to expand and inspire. This review focuses on the organic cofactors biotin, flavin, NAD, pyridoxal 5′-phosphate, pyruvoyl, and thiamin pyrophosphate as catalytic centers. Significant attention is also placed on the metal-dependent decarboxylase mechanisms.

The enzymatic activity of Pseudomonas fluorescens alpha-amino-beta-carboxymuconic-epsilon-semialdehyde decarboxylase (ACMSD) is critically dependent on a transition metal ion [Li, T., Walker, A. L., Iwaki, H., Hasegawa, Y., and Liu, A. (2005) J. Am. Chem. Soc. 127, 12282-12290]. Sequence analysis in this study further suggests that ACMSD belongs to the amidohydrolase superfamily, whose structurally characterized members comprise a catalytically essential metal cofactor. To identify ACMSD's metal ligands and assess their functions in catalysis, a site-directed mutagenesis analysis was conducted. Alteration of His-9, His-177, and Asp-294 resulted in a dramatic loss of enzyme activity, substantial reduction of the metal-binding ability, and an altered metallocenter electronic structure. Thus, these residues are confirmed to be the endogenous metal ligands. His-11 is implicated in metal binding because of the strictly conserved HxH motif with His-9. Mutations at the 228 site yielded nearly inactive enzyme variants H228A and H228E. The two His-228 mutant proteins, however, exhibited full metal-binding ability and a metal center similar to that of the wild-type enzyme as shown by EPR spectroscopy. Kinetic analysis on the mutants indicates that His-228 is a critical catalytic residue along with the metal cofactor. Since the identified metal ligands and His-228 are present in all known ACMSD sequences, it is likely that ACMSD proteins from other organisms contain the same cofactor and share similar catalytic mechanisms. ACMSD is therefore the first characterized member in the amidohydrolase superfamily that represents a C-C breaking activity.

The enzymatic activity of Pseudomonas fluorescens alpha-amino-beta-carboxymuconic-epsilon-semialdehyde decarboxylase (ACMSD) is critically dependent on a transition metal ion [Li, T., Walker, A. L., Iwaki, H., Hasegawa, Y., and Liu, A. (2005) J. Am. Chem. Soc. 127, 12282-12290]. Sequence analysis in this study further suggests that ACMSD belongs to the amidohydrolase superfamily, whose structurally characterized members comprise a catalytically essential metal cofactor. To identify ACMSD's metal ligands and assess their functions in catalysis, a site-directed mutagenesis analysis was conducted. Alteration of His-9, His-177, and Asp-294 resulted in a dramatic loss of enzyme activity, substantial reduction of the metal-binding ability, and an altered metallocenter electronic structure. Thus, these residues are confirmed to be the endogenous metal ligands. His-11 is implicated in metal binding because of the strictly conserved HxH motif with His-9. Mutations at the 228 site yielded nearly inactive enzyme variants H228A and H228E. The two His-228 mutant proteins, however, exhibited full metal-binding ability and a metal center similar to that of the wild-type enzyme as shown by EPR spectroscopy. Kinetic analysis on the mutants indicates that His-228 is a critical catalytic residue along with the metal cofactor. Since the identified metal ligands and His-228 are present in all known ACMSD sequences, it is likely that ACMSD proteins from other organisms contain the same cofactor and share similar catalytic mechanisms. ACMSD is therefore the first characterized member in the amidohydrolase superfamily that represents a C-C breaking activity.

Alpha-amino-beta-carboxymuconate-epsilon-semialdehyde decarboxylase (ACMSD) is a widespread enzyme found in many bacterial species and all currently sequenced eukaryotic organisms. It occupies a key position at the branching point of two metabolic pathways: the tryptophan to quinolinate pathway and the bacterial 2-nitrobenzoic acid degradation pathway. The activity of ACMSD determines whether the metabolites in both pathways are converted to quinolinic acid for NAD biosynthesis or to acetyl-CoA for the citric acid cycle. Here we report the first high-resolution crystal structure of ACMSD from Pseudomonas fluorescens which validates our previous predictions that this enzyme is a member of the metal-dependent amidohydrolase superfamily of the (beta/alpha)(8) TIM barrel fold. The structure of the enzyme in its native form, determined at 1.65 A resolution, reveals the precise spatial arrangement of the active site metal center and identifies a potential substrate-binding pocket. The identity of the native active site metal was determined to be Zn. Also determined was the structure of the enzyme complexed with cobalt at 2.50 A resolution. The hydrogen bonding network around the metal center suggests that Arg51 and His228 may play important roles in catalysis. The metal center configuration of PfACMSD is very similar to that of Zn-dependent adenosine deaminase and Fe-dependent cytosine deaminase, suggesting that ACMSD may share certain similarities in its catalytic mechanism with these enzymes. These data enable us to propose possible catalytic mechanisms for ACMSD which appear to be unprecedented among all currently characterized decarboxylases.

3-Hydroxyanthranilate-3,4-dioxygenase (HAD) is a non-heme Fe(II) dependent enzyme that catalyzes the oxidative ring-opening of 3-hydroxyanthranilate to 2-amino-3-carboxymuconic semialdehyde. The enzymatic product subsequently cyclizes to quinolinate, an intermediate in the biosynthesis of nicotinamide adenine dinucleotide. Quinolinate has also been implicated in important neurological disorders. Here, we describe the mechanism by which 4-chloro-3-hydroxyanthranilate inhibits the HAD catalyzed reaction. Using overexpressed and purified bacterial HAD, we demonstrate that 4-chloro-3-hydroxyanthranilate functions as a mechanism-based inactivating agent. The inactivation results in the consumption of 2 +/- 0.8 equiv of oxygen and the production of superoxide. EPR analysis of the inactivation reaction demonstrated that the inhibitor stimulated the oxidation of the active site Fe(II) to the catalytically inactive Fe(III) oxidation state. The inactivated enzyme can be reactivated by treatment with DTT and Fe(II). High resolution ESI-FTMS analysis of the inactivated enzyme demonstrated that the inhibitor did not form an adduct with the enzyme and that four conserved cysteines were oxidized to two disulfides (Cys125-Cys128 and Cys162-Cys165) during the inactivation reaction. These results are consistent with a mechanism in which the enzyme, complexed to the inhibitor and O2, generates superoxide which subsequently dissociates, leaving the inhibitor and the oxidized iron center at the active site.