7,8-Dihydroxyflavone is a direct inhibitor of pyridoxal phosphatase 1

18 Vitamin B6 deficiency has been linked to cognitive impairment in human brain disorders for 19 decades. Still, the molecular mechanisms linking vitamin B6 to these pathologies remain 20 poorly understood, and whether vitamin B6 supplementation improves cognition is unclear as 21 well. Pyridoxal phosphatase (PDXP), an enzyme that controls levels of pyridoxal 5’-22 phosphate (PLP), the co-enzymatically active form of vitamin B6, may represent an 23 alternative therapeutic entry point into vitamin B6-associated pathologies. However, 24 pharmacological PDXP inhibitors to test this concept are lacking. We now identify a PDXP 25 and age-dependent decline of PLP levels in the murine hippocampus that provides a rationale 26 for the development of PDXP inhibitors. Using a combination of small molecule screening, 27 protein crystallography and biolayer interferometry, we discover and analyze 7,8-28 dihydroxyflavone (7,8-DHF) as a direct and potent PDXP inhibito r. 7,8-DHF binds and 29 reversibly inhibits PDXP with low micromolar affinity and sub-micromolar potency. In 30 mouse hippocampal neurons, 7,8-DHF increases PLP in a PDXP-dependent manner. These 31 findings validate PDXP as a druggable target. Of note, 7,8-DHF is a well-studied molecule in 32 brain disorder models, although its mechanism of action is actively debated. Our discovery of 33 7,8-DHF as a PDXP inhibitor offers novel mechanistic insights into the controversy 34 surrounding 7,8-DHF-mediated effects in the brain. 35 36 37 38

Cellular PLP availability in the brain depends on numerous factors, including the intestinal absorption of B6 vitamers, extracellular phosphatases, inter-organ transport and intracellular enzymes and carriers/scavengers involved in PLP formation and homeostasis (2).Specifically, intracellular PLP is formed by the pyridoxal kinase (PDXK)-catalyzed phosphorylation of pyridoxal, or the pyridox(am)ine-5'-phosphate oxidase (PNPO)-catalyzed oxidation of pyridox(am)ine 5'-phosphate to PLP.PLP is highly reactive and can undergo condensation reactions with e.g., primary amino groups or thiol groups in proteins or amino acids.Although the mechanisms of PLP delivery within the cells are still largely unknown, it is clear that the intracellular availability of PLP for co-enzymatic functions depends on PLP carriers/scavengers and on the hydrolytic activity of pyridoxal 5'-phosphate phosphatase (PDXP) (2,(27)(28)(29)(30).
We have previously shown that the genetic knockout of PDXP (PDXP-KO) in mice increases brain PLP levels and improves spatial memory and learning, suggesting that elevated PLP levels can improve cognitive functions in this model (30).We therefore reasoned that a pharmacological inhibition of PDXP may be leveraged to increase intracellular PLP levels and conducted a high-throughput screening campaign to identify small-molecule PDXP modulators.Here, we report the discovery and the structural and cellular validation of 7,8dihydroxyflavone (7,8-DHF) as a selective PDXP inhibitor.7,8-DHF is a well-studied molecule in brain disorder models characterized by impaired cognition, and widely regarded as a tropomyosin receptor kinase B (TrkB) agonist with brain-derived neurotrophic factor (BDNF)-mimetic activity (31).However, a direct TrkB agonistic activity of 7,8-DHF has been called into question (32)(33)(34)(35)(36).Our serendipitous discovery of 7,8-DHF as a direct PDXP inhibitor provides an alternative mechanistic explanation for 7,8-DHF-mediated effects.More potent, efficacious, and selective PDXP inhibitors may be useful future tools to explore a possible benefit of elevated PLP levels in brain disorders.

PDXP activity controls PLP levels in the hippocampus
The hippocampus is important for age-dependent memory consolidation and learning, and impaired memory and learning is associated with PLP deficiency (3).To study a possible contribution of PDXP and/or PDXK to age-related PLP homeostasis in the hippocampus, we performed Western blot analyses in young versus older mice.Unexpectedly, we found that both PDXP and PDXK expression levels were markedly higher in hippocampi of middle-aged than of juvenile animals (Fig. 1a).These data suggest an accelerated hippocampal PLP turnover in older mice, consistent with previous findings in senescent mice (37).
An analysis of total hippocampal PLP levels in PDXP-WT and PDXP-KO mice showed an age-dependent profile.PLP levels appeared to peak around 3 months of age (possibly reflecting PLP-dependent neurotransmitter biosynthesis and metabolism during the postnatal developmental period) and descended back to juvenile levels by 12 months of age in both genotypes.Although total hippocampal PLP levels in PDXP-KO mice also decreased with age, they consistently remained above PLP levels in control mice (Fig. 1b; two-tailed, unpaired t-test of PLP levels in PDXP-WT vs. PDXP-KO hippocampi, all ages combined: p<0.0001).
PLP is protected from hydrolysis by binding to proteins, and PDXP is expected to dephosphorylate only non-protein-bound PLP (38).To test this, we prepared protein-depleted PLP fractions from PDXP-WT and PDXP-KO hippocampal lysates using 3 kDa molecular weight cutoff centrifugal filters.The quantification of PLP in these fractions demonstrated that PDXP loss indeed only increased the pool of protein-depleted PLP, both in young (18-42 days old) and older mice (252-352 days old, corresponding to mature/middle-aged mice), whereas the levels of protein-bound PLP remained unchanged (Fig. 1c).While the hippocampal levels of non-protein-bound PLP dropped by about 60% over this time span in PDXP-WT mice, they remained elevated in PDXP-KO mice ~2-fold higher in younger, and ~5-fold higher in older PDXP-KO compared to the respective PDXP-WT; see Fig. 1 -figure supplement 1 for exact mouse ages).We conclude that hippocampi of older mice are characterized by a specific decrease in the levels of non-protein-bound PLP, and that this agedependent PLP loss is dependent on PDXP activity.These observations establish that PDXP is a critical determinant of PLP levels in the murine hippocampus and suggest that intracellular PLP deficiency may be alleviated by PDXP inhibition.
In vitro activity assays using PLP as a substrate confirmed that 7,8-DHF directly blocks murine and human PDXP activity with submicromolar potency and an apparent efficacy of ~50% (Fig. 2a, b).The efficacy of 7,8-DHF under these conditions may be limited by the poor solubility of this flavone (see also Fig. 2c).This interpretation is supported by Todd et al. who established a solubility profile of 7,8-DHF and demonstrated that only ⁓20% of the compound remains in solution beyond a concentration of 5 µM (35).
We used a biolayer interferometry (BLI) optical biosensing technique to further characterize the binding of 7,8-DHF to PDXP (Fig. 2c).Consistent with a specific interaction, 7,8-DHF binding to PDXP was concentration-dependent and fully reversible.Steady-state analysis of a 7,8-DHF serial dilution series yielded an affinity (K D ) value of 3.1 ± 0.3 µM (data are mean values ± S.E. of n=4 measurements; see Fig. 2 -figure supplement 3 for the three other measurements) using a 1:1 dose-response model.Global analysis of the sensorgrams assuming a 1:1 binding model resulted in an affinity of 2.6 ± 0.5 µM, in line with the steady-state results (Fig. 2c).Due to the aforementioned poor solubility of 7,8-DHF (35), concentrations higher than 12.5 µM could not be evaluated.As expected, 5,7dihydroxyflavone showed no signal in the BLI, in line with previous experiments (see Fig. 2b).With its molecular size of 254 Da and its physicochemical properties, 7,8-DHF is a typical fragment-like molecule (41).Typical association rate constants (k on ) for fragments are limited by the rate of diffusion and are higher than 10 6 M −1 s −1 .Interestingly, 7,8-DHF showed a slow k on of 1050 M −1 s −1 , which is atypical and rarely found for fragment-like molecules (42), and a k off rate of 0.03 s −1 .With the commonly used estimation of ΔG ~ pK D and a heavy atom number of 19, 7,8-DHF shows a high ligand efficiency of 0.39, which makes it an interesting molecule for further medicinal chemistry optimization.Taken together, these data support a direct and reversible physical interaction between 7,8-DHF and PDXP that leads to PDXP inhibition.
HAD phosphatases are Mg 2+ -dependent phospho-aspartate transferases that consist of a Rossman-like catalytic core linked to a cap domain.The insertion site, structure and size of the cap define the substrate selectivity of the respective enzyme.The "capless" C0-type HAD phosphatases contain either a very small or no cap, resulting in an accessible catalytic cleft that enables the dephosphorylation of macromolecular substrates.Larger C1 or C2 caps act as a roof for the entrance to the active site; most C1/C2-capped HAD phosphatases consequently dephosphorylate small molecules that can gain access to the catalytic cleft.Cap domains also contain so-called substrate specificity loops that contribute to substrate coordination.Hence, caps are distinguishing features of HAD phosphatases (38,(43)(44)(45).
To probe the selectivity of 7,8-DHF for PDXP, a C2-capped HAD phosphatase, we tested five additional mammalian HAD phosphatases, including two other C2-and two C1capped phosphatases and one C0-type enzyme.When assayed at a (nominal) concentration of 5, 10 and 40 µM (i.e., up to ~40-fold above the IC 50 value for PDXP-catalyzed PLPdephosphorylation), 7,8-DHF was inactive against four of the tested enzymes.7,8-DHF inhibited PGP, the closest PDXP relative, with an IC 50 value of 4.8 µM (Fig. 2d).This result is consistent with the criteria applied during the initial counter-screen (see above).Together, our data show that 7,8-DHF selectively inhibits PDXP, and that higher 7,8-DHF concentrations can also target the PDXP paralog PGP.

Mode of PDXP inhibition
To probe the mechanism of PDXP inhibition, we assayed the steady state kinetics of PLP dephosphorylation in the presence of increasing 7,8-DHF concentrations (Table 1).Analysis of the derived kinetic constants demonstrated that 7,8-DHF increased the K M up to ~2-fold, and slightly reduced v max values ~0.7-fold.Thus, 7,8-DHF mainly exhibits a mixed mode of PDXP inhibition, which is predominantly competitive.

Co-crystal structure of PDXP bound to 7,8-DHF
To investigate the mechanism of PDXP inhibition in more detail, we co-crystallized homodimeric, full-length murine PDXP in complex with this compound.7,8-DHF-bound PDXP crystallized in the cubic space group I23 with one dimer per asymmetric unit, with protomer A containing the inhibitor and protomer B representing an apo-state (Fig. 3a).The structure was refined following molecular replacement with full-length murine PDXP (here referred to as apo-PDXP; Protein Data Bank /PDB entry 4BX3) to a resolution of 2.0 Å resulting in an R work of 18.4% and an R free of 21.1% (PDB code 8QFW).Data collection and refinement statistics are summarized in Table 2. prevented by a salt bridge between Arg62 and Asp14 of a symmetry-related A-protomer (Fig. 3d).The χ 1 and χ 2 torsion angles of the Arg62 side chain observed in the B-protomer correspond to those observed for this side chain in both protomers of the apo-structure (4BX3).To allow binding of the inhibitor, the side chain of Arg62 needs to adopt a completely extended conformation, which is prevented by the salt bridge.
It is currently unclear whether the PDXP crystal state with only a single inhibitor bound per dimer reflects the state in solution.Due to the limited solubility of 7,8-DHF, we were unable to address the stoichiometry of 7,8-DHF binding to the PDXP dimer with isothermal calorimetry.It is conceivable that the PDXP crystal packing is very stable (indeed, 7,8-DHF-bound PDXP crystallized in the same cubic space group as apo-PDXP, see ref. (46), including the aforementioned salt bridge between Arg62 of the B-subunit and Asp14 of a symmetry-related molecule), and that the free energy generated by the formation of the crystal lattice is higher than the free energy generated upon inhibitor binding.Further inspection of 7,8-DHF-bound PDXP indicated that the crystallographic neighbors of protomer A and B are dissimilar.While the 7,8-DHF binding site is readily accessible in protomer A, the corresponding site in protomer B is shielded by two crystallographic neighbors (Fig. 3

figure supplement 1).
In protomer A, 7,8-DHF binds directly in the PDXP active site, with its phenyl ring at a distance of 8.3 Å from the Asp25 nucleophile, and 3.8 Å from Asp27, which functions as a general acid/base in the two partial reactions of the catalytic cycle (47).The introduction of PLP from PDB code 2CFT (human PDXP in complex with PLP) for visualization purposes revealed a partial overlap between the PLP and 7,8-DHF binding sites (Fig. 3e).While PLP is deeply buried in the catalytic cleft, the bulkier 7,8-DHF molecule obstructs the active site entrance with its 1,4-benzopyrone moiety and its phenyl ring protrudes into the catalytic cleft.
7,8-DHF is embedded in a cavity of the PDXP active site that is exclusively formed by the active site of protomer A, without a contribution of the dimerization interface with protomer B. One side of this cavity is formed by the more polar residues Asp27, Asn60, Ser61 and Arg62, whereas the opposite side is established by the more hydrophobic residues Tyr146, His178, Pro179, and Leu180.Adjacent to this hydrophobic stretch, the polar residue Glu148 is located at the active site entrance, directly opposite of Arg62 on the more polar side of the 7,8-DHF binding channel (Fig. 3f).Inhibitor binding appears to be primarily stabilized by two hydrogen bonds, as well as polar and non-polar interactions.The side chain hydroxyl group of Ser61 forms a direct hydrogen bond with the ketone group of the inhibitor, which is additionally coordinated by the Ser61 backbone nitrogen atom.Furthermore, Glu148 forms a direct hydrogen bond via its carboxylic acid with the 7-hydroxyl group of 7,8-DHF.The side chains of the polar residues Asp27, Asn60, and Arg62 engage in van der Waals interactions with 7,8-DHF.The two hydroxyl groups of the 7,8-DHF benzyl ring engage in van der Waals interactions with the guanidinium group of Arg62 and the carboxylic acid function of Glu148.
On the more hydrophobic side of the binding cavity, Tyr146 forms π-electron stacking interactions with the pyrone ring of 7,8-DHF.In addition, the His178 imidazole group coordinates the 7,8-DHF phenyl ring via π-π stacking.His178, located in the substrate specificity loop, and Asn60 and Arg62 are also important for PLP binding (46,48).All PDXP residues found to engage in 7,8-DHF interactions are identical in murine and human PDXP (Fig. 3

-figure supplement 2).
To verify the putative 7,8-DHF -PDXP interactions, we introduced single mutations into the binding interface.Asn60, Arg62, Tyr146, Glu148 and His178 were each exchanged for Ala (PDXP N60A , PDXP R62A , PDXP Y146A , PDXP E148A , or PDXP H178A , respectively).Since the carboxamide group of Asn60 can form a hydrogen bond with the carboxylate moiety of the Asp27, and a loss of this interaction in the PDXP N60A variant is predicted to alter the PDXP structure, we additionally mutated Asn60 to Ser (PDXP N60S ).PDXP variants were recombinantly expressed and purified from E. coli (see Fig. 3 -figure supplement 3 for protein purity).Figure 3g (left panel) shows that all PDXP variants were enzymatically active.As expected, the phosphatase activities of PDXP N60A and of PDXP H178A were reduced.
The somewhat elevated phosphatase activity of PDXP Y146A and PDXP E148A is currently unexplained.Importantly, all variants except PDXP N60A and PDXP N60S were resistant to 7,8-DHF, supporting the essential role of each of these residues for inhibitor binding (Fig. 3g, right panel).
Based on the inhibitor-bound structure and the predominantly competitive component of PDXP inhibition by 7,8-DHF (increased K M , see Table 1), it seems likely that 7,8-DHF sterically hinders substrate access to the active site, and competes with PLP coordination.In addition, BLI measurements (see Fig. 2c) showed a relatively slow association rate and extended residence time of 7,8-DHF (τ=30.3s).This may indicate a possible reorganization of the amino acids and water molecules within the binding site as a function of slow-onset 7,8-DHF binding.The reduced rate of product formation may account for the observed noncompetitive component of 7,8-DHF-mediated PDXP inhibition (reduction of v max , see Table 1).

7,8-DHF functions as a PDXP inhibitor in hippocampal neurons
To investigate cellular target engagement of 7,8-DHF, we isolated primary hippocampal neurons from PDXP-WT and PDXP-KO embryos.PDXP deficiency increased total PLP levels 2.4-fold compared to PDXP-WT neurons (Fig. 4a).This finding is in good agreement with the PLP increase resulting from PDXP loss in total hippocampal extracts (see Fig. 1).
The larger absolute PLP values in cultured neurons are likely attributable to the high concentration of the PLP precursor pyridoxal (20 µM) in the culture medium.We did not observe PDXP-dependent changes in pyridoxal kinase (PDXK) expression (Fig. 4b) and could not detect pyridox(am)ine-5'-phosphate oxidase (PNPO) in hippocampal neuronal cultures, suggesting that the PLP increase was primarily caused by the constitutive PDXP loss.
To assess the consequences of 7,8-DHF treatment on PLP levels in hippocampal neurons, we chose short-term incubation conditions (45 min, 20 µM) to avoid possible secondary effects of the inhibitor.As expected, the acute effect of 7,8-DHF treatment in WT cells was much more subtle (⁓9% increase in total PLP) than the impact of long-term PDXPdeficiency (441.3 ± 62.6 nmol PLP/g protein in DMSO solvent control-treated cells versus 482.7 ± 130.4 nmol PLP/g protein in 7,8-DHF treated cells; data are mean values ± S.E. of n=4 independent experiments).However, this effect is likely underestimated because only the PDXP-accessible pool of non-protein-bound PLP may be impacted by 7,8-DHF (see Fig. 1c).
Due to the limited number of available hippocampal neurons, we were unfortunately unable to obtain sufficient quantities of protein-depleted PLP pools to address this question.
Acute changes in the PLP/PL ratio may be a more sensitive indicator of PDXP activity than changes in total PLP levels alone, because PDXP inhibition is expected to increase cellular levels of PLP (the PDXP substrate) and to concomitantly decrease the levels of PL (the product of PDXP phosphatase activity).The PLP/PL ratio is also independent of the exact protein concentration in a given extract of hippocampal neurons, thus optimizing comparability between samples.As shown in Fig. 4c, 7,8-DHF significantly increased the PLP/PL ratio in PDXP-WT, but not in PDXP-KO hippocampal neurons (+18% versus +1% compared to the respective DMSO controls).Together, these data indicate that 7,8-DHF can modulate cellular PLP levels in a PDXP-dependent manner and validate PDXP as a 7,8-DHF target in primary hippocampal neurons.

DISCUSSION
PLP deficiency has been associated with human brain disorders for decades (3), yet causal links remain unclear.Aside from vitamin B6 administration, pharmacological strategies to elevate intracellular PLP levels are lacking.Here, we identify 7,8-dihydroxyflavone (7,8-DHF) as a direct PDXP inhibitor that increases PLP levels in hippocampal neurons, validating PDXP as a druggable target to control intracellular PLP levels in the brain.We also present a high-resolution 7,8-DHF/PDXP co-crystal structure that could facilitate the design of more potent, efficacious, and selective PDXP inhibitors in the future.Such molecules might improve the control of intracellular PLP levels and help to elucidate a possible contribution of PLP to the pathophysiology of brain disorders.Our observation that the expression of PDXP is substantially upregulated in hippocampi of middle-aged mice suggests that a therapeutic vitamin B6 supplementation alone may not suffice to elevate intracellular PLP levels under conditions where the PLP-degrading phosphatase is hyperactive.
Although PLP deficiency is thought to contribute to the respective human conditions (3,74,75), PLP-dependent processes have not yet been considered in the context of 7,8-DHFinduced effects.7,8-DHF was initially discovered as a small-molecule TrkB agonist with BDNFmimetic activity (76).BDNF, a high-affinity TrkB ligand, is an important neuropeptide for nervous system function and pathology.Consensus is emerging that BDNF plays a key role in the treatment response to neuropsychiatric drugs (32).Therapeutics that target BDNF/TrkBsignaling are thus of interest as disease-modifying agents in several brain disorders.Since BDNF does not cross the blood-brain barrier, attempts have been made to develop small molecule BDNF mimetics.Several candidates have been reported, including 7,8-DHF (33,76).Nevertheless, the on-target selectivity and efficacy of these compounds is actively debated.Using quantitative and direct assays to measure TrkB dimerization and activation, TrkB downstream signaling pathways, TrkB-dependent gene expression and cytoprotection, 7,8-DHF and other reported small-molecule TrkB agonists failed to activate TrkB in cells (33)(34)(35).An electrophysiological study in acute hippocampal slice preparations demonstrated that 7,8-DHF potentiates hippocampal mossy fiber-CA3 synaptic transmission in a TrkB receptorindependent manner (77).Overall, it appears that the mechanism of action of 7,8-DHF is incompletely understood, but 7,8-DHF targets other than TrkB so far have remained elusive.
The identification of 7,8-DHF as a PDXP inhibitor reported here indicates that this flavone may modulate vitamin B6-dependent processes and suggests that PDXP could be explored as a pharmacological entry point into brain disorders.

PDXP knockout mice
Floxed PDXP mice (Pdxp tm1Goh ) were generated on a C57Bl/6J background, and whole-body Pdxp knockouts were achieved by breeding with B6.FVB-Tg(EIIa-cre)C5379Lmgd/J (EIIa-Cre) transgenic mice, as described (30).All experiments were authorized by the local veterinary authority and committee on the ethics of animal experiments (Regierung von Unterfranken).All analyses were carried out in strict accordance with all German and European Union applicable laws and regulations concerning care and use of laboratory animals.

Preparation of hippocampi and hippocampal neurons and immunocytochemistry
Mice were sacrificed by cervical dislocation, and brains were immediately placed on a precooled metal plate and dissected under a Leica M80 binocular (Leica, Wetzlar, Germany).
Hippocampi were weighed and flash-frozen in liquid nitrogen.The entire procedure was performed in <3 min.Hippocampal lysates were prepared by the addition of ice-cold PBS (200 µL PBS/10 mg hippocampal wet weight) and homogenized for 1 min in a TissueLyser II instrument (Qiagen, Hilden, Germany).One fourth of the obtained volume of each lysate was used for the analysis of total PLP concentrations as described below.To determine proteindepleted PLP (27), the remaining volume of each lysate was centrifuged at 14,000 × g for 15 min at 4°C.The supernatant was applied to 3 kDa MWCO filters (Amicon Ultra-0.5 Centrifugal Filter; Merck Millipore, Darmstadt, Germany), and centrifuged at 14,000 × g for 45 min at 4°C.The flow-through was collected and prepared for HPLC analysis (see below).
Primary hippocampal neuronal cultures were prepared from mouse embryos at embryonic day 17.Hippocampi were incubated with 0.5 mg/mL trypsin, 0.2 mg/mL EDTA and 10 µg/mL DNase I in PBS for 30 min at 37°C.Trypsinization was stopped by adding 10% fetal calf serum.Cells were dissociated by trituration, counted, and seeded at a density of 150,000 cells per 35 mm dish.Dissociated cells were grown in neurobasal medium supplemented with L-glutamine and B27 supplement (A3582801, Life Technologies, Dreieich, Germany) with an exchange of 50% of the medium after 6 days in culture.After 21 days of differentiation (day in vitro 21/DIV21), 7,8-DHF (20 µM) or DMSO (0.02%, v/v) was added to the hippocampal neuronal cultures for 45 min.Cells were rinsed once with PBS (37°C), lysed in 150 µL ice cold H 2 O, and placed at -80°C for at least 30 min.

Determination of PLP and PL by high-performance liquid chromatography (HPLC)
Samples were derivatized as described (78).Briefly, 100 µL of lysate were mixed with 8 µL derivatization agent (containing 250 mg/mL of both semicarbazide and glycine), and incubated on ice for 30 min.Samples were then deproteinized by addition of perchloric acid (8 µL of a 72% (w/v) stock solution), followed by centrifugation at 15,000 × g for 15 min at 4 °C.Supernatants (100 µL) were neutralized with 10 µL NaOH (25% (v/v) stock solution), and 2 µM pyridoxic acid was added as an internal standard.PLP and PL were subjected to the same derivatization protocol to establish a standard curve.Samples were analyzed on a Dionex Ultimate 3000 HPLC (Thermo Fisher Scientific, Dreieich, Germany), using 60 mM Na 2 HPO 4 , 1 mM EDTA, 9.5% (v/v) MeOH; pH 6.5 as mobile phase.PL, PLP and pyridoxic acid were separated on a 3 μm reverse phase column (BDS-HYPERSIL-C18, Thermo Fisher Scientific).Chromatograms were analyzed using Chromeleon 7 software (Thermo Fisher Scientific).

Expression and purification of recombinant proteins
All purification steps of murine PDXP and murine PGP were carried out exactly as described (40).N-terminally His 6 -tagged PDXP variants were expressed exactly as described for PDXP-WT (40), except that the His 6 -tag was not cleaved off.Human GST-PDXP was transformed into E. coli BL21(DE3) (Stratagene Europe/VWR, Darmstadt, Germany).Protein expression was induced with 0.5 mM isopropyl β-d-thiogalactopyranoside for 18 h at 20 °C.All subsequent purification steps were carried out at 4 °C.Cells were harvested by centrifugation for 10 min at 8000 × g and resuspended in lysis buffer (100 mM triethanolamine/TEA, 500 mM NaCl; pH 7.4) supplemented with protease inhibitors (EDTA-free protease inhibitor tablets; Roche, Mannheim, Germany) and 150 U/mL DNase I (Applichem, Renningen, Germany).Cells were lysed using a cell disruptor (Constant Systems, Daventry, UK), and cell debris was removed by centrifugation for 30 min at 30,000 × g.GST-PDXP was batchpurified on a glutathione sepharose 4B resin (GE Healthcare, Uppsala, Sweden).After extensive washing with 25 column volumes of wash buffer (50 mM TEA, 250 mM NaCl; pH 7.4), GST-PDXP was eluted in wash buffer supplemented with 10 mM reduced glutathione, concentrated, and further purified in buffer A (50 mM TEA, 250 mM NaCl, 5 mM MgCl 2 ; pH 7.4).using a HiLoad 16/60 Superdex 200 pg gel filtration column operated on an ÄKTA liquid chromatography system (GE Healthcare).

High-Throughput Screen for PDXP Modulators
The screening campaign (chemical library, screening protocol, concentration-dependent assays, data analysis) was conducted exactly as described previously (40), except that the primary screen was done with PDXP, the counter-screen was done with PGP, and PDXP inhibitor hits were validated using 5'-pyridoxal phosphate (PLP) as a physiological PDXP substrate.

IC 50 determinations, enzyme kinetics, and compound selectivity
Buffer conditions for enzymatic assays were as previously published (40).Purified phosphatases were pre-incubated for 10 min at RT with serial dilutions of flavones.
Dephosphorylation reactions were started by the addition of the indicated substrate; buffer with substrate and the respective flavone but without the enzyme served as a background control.Prior to compound testing, time courses of inorganic phosphate release from the respective phosphatase substrates were conducted to ensure assay linearity.Inorganic phosphate release was detected with malachite green solution (Biomol Green; Enzo Life Sciences, Lörrach, Germany); the absorbance at 620 nm (A 620 ) was measured on an Envision 2104 multilabel reader (Perkin Elmer, Rodgau, Germany).Released phosphate was determined by converting the values to nmol P i with a phosphate standard curve.Data were analyzed with GraphPad Prism version 9.5.1 (GraphPad, Boston/Massachusetts, USA).For IC 50 determinations, log inhibitor versus response was calculated (four parameter).To derive K M and k cat values, data were fitted by nonlinear regression to the Michaelis-Menten equation.

Biolayer interferometry (BLI)
PDXP was biotinylated using the EZ-Link NHS-PEG4-Biotin kit, as recommended by the manufacturer (Thermo Fisher Scientific), and loaded on Super Streptavidin Biosensors (SSA) (Sartorius, Göttingen, Germany) as follows.SSA sensors were equilibrated for 1 h at RT in BLI assay buffer (250 mM triethanolamine, 5 mM MgCl 2 , 250 mM NaCl, 0.005% (v/v) TWEEN-20; pH 7.5), loaded with 200 µg/mL biotinylated PDXP, blocked with 2 µg/mL biocytin, and washed in BLI assay buffer.Reference SSA sensors were blocked with 2 µg/mL biocytin (79).Six point 1:1 serial dilution series of 7,8-DHF and 5,7-DHF were prepared in DMSO, and BLI assay buffer was added to the wells to obtain a 7,8-DHF starting concentration of 25 µM.The final DMSO concentration was 5% (v/v).Buffers for baseline, dissociation, and buffer correction wells were supplemented with the same amount of DMSO for identical buffer conditions.Four measurements were carried out per condition, using one sensor set for two measurements.All measurements were conducted on an Octet K2 device (Sartorius) using 96-well plates.Assay settings were as follows: baseline measurement 45 sec, association time 90 sec, dissociation time 150 sec.The resulting data were processed using the double reference method of the Octet analysis software for removal of drifts and well-towell artefacts.Kinetic analyses were performed using the Octet analysis software.The steady state analysis was carried out with OriginPro 2021b (OriginLab, Northampton/Massachusetts, USA), using a dose-response model for regression.

PDXP crystallization and data collection
For co-crystallization with 7,8-DHF, full-length murine PDXP (10 mg/mL in 50 mM triethanolamine; 250 mM NaCl; 5 mM MgCl 2 ; pH 7.4) was supplemented with a three-fold molar excess of the flavone.Prism-shaped crystals of 7,8-DHF-bound PDXP were grown at 20°C in 0.1 M phosphate citrate (pH 4.2) and 40% (v/v) PEG 300 using the sitting-drop vapor diffusion method.Crystals were cryoprotected for flash-cooling in liquid nitrogen by soaking in mother liquor 25% (v/v) glycerol.Diffraction data were collected from flash-cooled crystals after addition of 25% glycerol at a temperature of 100 K on beamline BL 14.1 at the BESSY (Helmholtz Zentrum Berlin, Germany) synchrotron.The structure of 7,8-DHF-PDXP was solved by molecular replacement with the program Phaser (80) with the structure of the murine PDXP (PDB entry 4BX3) as search model.e R free same as R for 5% of the data randomly omitted from the refinement.The number of reflections includes the R free subset.

FIGURES AND
Numbers in parentheses refer to the highest resolution data shell.
Ramachandran statistics reflect the percentage of residues in favored/allowed/outlier regions.
PDXP-KO hippocampi (all ages combined; two-tailed, unpaired t-test) p<0.0001.Bold table entries indicate those hippocampal extracts that were further separated for an analysis of protein-depleted and protein-bound PLP (see Fig. 1c).A Coomassie Blue-stained gel is shown.
Like apo-PDXP, 7,8-DHF-bound PDXP homodimerizes via its cap domain.The alignment of the structures representing apo-PDXP, the inhibitor-free protomer B, and the 7,8-DHF-bound PDXP protomer A resulted in root mean square deviations (RMSDs) of 0.42 Å (A vs. B in this structure), 0.35 Å (A vs. B in 4BX3), 0.41 Å (A in this structure vs.A in 4BX3, Fig. 3b), 0.49 Å (A in this structure vs. B in 4BX3), 0.45 Å (B in this structure vs.A in 4BX3) and 0.37 Å (B in this structure vs. B in 4BX3) based on Cα atoms, which indicate that binding of the inhibitor did not result in significant changes in the backbone conformations.In particular, we did not discern distinct open and closed active site conformations (40) or a tilting of the substrate specificity loop (46) in 7,8-DHF-bound PDXP compared to apo-PDXP.All catalytic core residues and the Mg 2+ cofactor are correctly oriented in the presence of the inhibitor.Hence, 7,8-DHF binding does not appear to impact the overall fold or the open/closed dynamics of PDXP.7,8-DHF was observed to only bind to one subunit (the A-chain; Fig. 3a) with welldefined density (Fig. 3c) and full occupancy since its average B-factor of 45.8 Å 2 closely matches the B-factors of the surrounding atoms.Binding to the other subunit (B-protomer) is Primers were purchased from Eurofins Genomics (Ebersberg, Germany), and all constructs were verified by sequencing (Microsynth Seqlab, Göttingen, Germany).

Figure 4 .
Figure 4. Effect of 7,8-DHF on the PLP/PL ratio in cultured hippocampal neurons from WT or PDXP-KO mice.
b R pim = Σ hkl 1/(N-1) ½ Σ i | I i (hkl) -I (hkl) | / Σ hkl Σ i I(hkl),where N is the redundancy of the data and I (hkl) the average intensity.c <I/σI> indicates the average of the intensity divided by its standard deviation.d R work = Σ hkl ||F o | -|F c || / Σ hkl |F o | where F o and F c are the observed and calculated structure factor amplitudes.

Table 1 . Kinetic constants of PDXP-catalyzed PLP hydrolysis in the presence of 7,8-
TABLES The data are mean values ± S.E.M. of n=3 independent experiments, except for the solvent control samples (n=6).Curves were fitted and parameters K M (Michaelis-Menten constant); v max , (maximum enzyme velocity); k cat (turnover number) were derived using the Michaelis-Menten model in GraphPad Prism 9.5.1.The k cat values were calculated from the maximum enzyme velocities using a molecular mass of 31,828 Da for PDXP.DMSO concentrations were kept constant (0.1% DMSO under all conditions, including the solvent control samples).

Table 2 . Data Collection and Refinement Statistics.
a R sym = Σ hkl Σ i | I i -<I> | / Σ hkl Σ i I i where I i is the i th measurement and <I> is the weighted mean of all measurements of I.