7,8-Dihydroxyflavone is a direct inhibitor of human and murine pyridoxal phosphatase

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

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 (Wilson et al., 2019).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 eLife digest Vitamin B6 is an important nutrient for optimal brain function, with deficiencies linked to impaired memory, learning and mood in various mental disorders.In older people, vitamin B6 deficiency is also associated with declining memory and dementia.Although this has been known for years, the precise role of vitamin B6 in these disorders and whether supplements can be used to treat or prevent them remained unclear.This is partly because vitamin B6 is actually an umbrella term for a small number of very similar and interchangeable molecules.Only one of these is 'bioactive', meaning it has a biological role in cells.However, therapeutic strategies aimed at increasing only the bioactive form of vitamin B6 are lacking.
Previous work showed that disrupting the gene for an enzyme called pyridoxal phosphatase, which breaks down vitamin B6, improves memory and learning in mice.To investigate whether these effects could be mimicked by drug-like compounds, Brenner, Zink, Witzinger et al. used several biochemical and structural biology approaches to search for molecules that bind to and inhibit pyridoxal phosphatase.
The experiments showed that a molecule called 7,8-dihydroxyflavone -which was previously found to improve memory and learning in laboratory animals with brain disorders -binds to pyridoxal phosphatase and inhibits its activity.This led to increased bioactive vitamin B6 levels in mouse brain cells involved in memory and learning.
The findings of Brenner et al. suggest that inhibiting pyridoxal phosphatase to increase vitamin B6 levels in the brain could be used together with supplements.The identification of 7,8-dihydroxyflavone as a promising candidate drug is a first step in the discovery of more efficient pyridoxal phosphatase inhibitors.These will be useful experimental tools to directly study whether increasing the levels of bioactive vitamin B6 in the brain may help those with mental health conditions associated with impaired memory, learning and mood.
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) (Wilson et al., 2019;Ciapaite et al., 2023;Fux and Sieber, 2020;Jang et al., 2003;Jeanclos et al., 2019).
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 (Jeanclos et al., 2019).We therefore reasoned that a pharmacological inhibition of PDXP may be leveraged to increase intracellular PLP levels and conducted a highthroughput screening campaign to identify small-molecule PDXP modulators.Here, we report the discovery and the structural and cellular validation of 7,8-dihydroxyflavone (7,8-DHF) as a preferential 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 (Liu et al., 2016).However, a direct TrkB agonistic activity of 7,8-DHF has been called into question (Wang et al., 2022a;Boltaev et al., 2017;Pankiewicz et al., 2021;Todd et al., 2014;Chen et al., 2011).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 (di Salvo et al., 2012).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 (Figure 1a).These data suggest an accelerated hippocampal PLP turnover in older mice, consistent with previous findings in senescent mice (Fonda et al., 1980).
An analysis of total hippocampal PLP levels in PDXP-WT and PDXP-KO mice showed an agedependent profile.PLP levels appeared to peak around 3 months of age (possibly reflecting PLPdependent 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 (Figure 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 (Gohla, 2019).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 of age) and in older mice (252-352 days of age, corresponding to mature/middle-aged mice), whereas the levels of proteinbound PLP remained unchanged (Figure 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 Figure 1-source data 3 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 age-dependent 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.
A high-throughput screening campaign identifies 7,8-DHF as a PDXP inhibitor Pharmacological small-molecule PDXP inhibitors are currently lacking.To identify PDXP inhibitor candidates, we screened the FMP small-molecule repository containing 41,182 compounds for molecules able to modulate the phosphatase activity of recombinant, highly purified murine PDXP (see Figure 2figure supplement 1 for a schematic of the screening campaign).Difluoro-4-methylumbelliferyl phosphate (DiFMUP) was used as a fluorogenic phosphatase substrate in a primary screen.Compounds that altered DiFMUP fluorescence by ≥50% (activator candidates) or ≤25% (inhibitor candidates) were subjected to EC 50 /IC 50 value determinations.Of these, 46 inhibitor hits were selected and counterscreened against phosphoglycolate phosphatase (PGP), the closest PDXP relative (Seifried et al., 2014;Jeanclos et al., 2022).Eleven of the PDXP inhibitor hits (with an IC 50 PDXP < 20 µM, and IC 50 PDXP < IC 50 PGP or no activity against PGP) were subsequently validated in a secondary assay, using PLP as a physiological PDXP substrate (see Figure 2-source data 5 for all 11 inhibitor hits).Only one PDXP-selective inhibitor hit (7,8-DHF, a naturally occurring flavone) blocked PDXP-catalyzed PLP dephosphorylation (IC 50 ~ 1 µM).
We used a biolayer interferometry (BLI) optical biosensing technique to further characterize the binding of 7,8-DHF to PDXP (Figure 2c).Consistent with a specific interaction, 7,8-DHF binding to PDXP was concentration-dependent and fully reversible.As a result of the poor solubility of the molecule, a saturation of the binding site was not experimentally accessible.Steady-state analysis of a 7,8-DHF serial dilution series yielded an affinity (K D ) value of 3.1±0.3µM (data are mean values ± SE of n=4 measurements; see Figure 2-figure supplement 2 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 (Figure 2c).As expected, 5,7-dihydroxyflavone showed no signal in the BLI, in line with previous experiments (see Figure 2b).With its molecular size of 254 Da and its physicochemical properties, 7,8-DHF is a typical fragment-like molecule (Congreve et al., 2003).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 1.05•10 4 M −1 s −1 , which is atypical and rarely found for fragment-like molecules (O'Connell et al., 2019), 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 The online version of this article includes the following source data for figure 1: Source data 1.Western blot quantification (to Figure 1a).Source data 2. Quantification of pyridoxal 5'-phosphate (PLP) levels in hippocampi of pyridoxal 5'-phosphate phosphatase (PDXP)-WT and knockout of PDXP (PDXP-KO) mice (to Figure 1b and c).
Source data 3. Analysis of total hippocampal pyridoxal 5'-phosphate (PLP) levels in pyridoxal 5'-phosphate phosphatase (PDXP)-WT and knockout of PDXP (PDXP-KO) mice.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.

Selectivity of 7,8-DHF
PDXP is a member of the large family of haloacid dehalogenase (HAD)-type hydrolases (Seifried et al., 2013).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 the so-called substrate specificity loops that contribute to substrate coordination.Hence, caps are distinguishing features of HAD phosphatases (Gohla, 2019;Seifried et al., 2013;Burroughs et al., 2006;Huang et al., 2015).
The online version of this article includes the following source data and figure supplement(s) for figure 2: Source data 1.Phosphatase activity assays (to Figure 2a).
Source data 4. Effect of 7,8-dihydroxyflavone (7,8-DHF) on the phosphatase activity of different phosphatases (to Figure 2d).Source data 5. Pyridoxal 5'-phosphate phosphatase (PDXP) inhibitor hits.to increase the activity of calcineurin.As expected, 7,8-DHF inhibited PGP, the closest PDXP relative, with an IC 50 value of 4.8 µM.This result is consistent with the criteria applied during the initial counterscreen (see above).In addition to PGP, 7,8-DHF inhibited the C1-capped soluble cytosolic 5'-nucleotidase 1A (NT5C1A) with an IC 50 value of ~10 µM.NT5C1A is an AMP hydrolase expressed in skeletal muscle and heart (Bianchi and Spychala, 2003), which is also sensitive to inhibition by small molecules that target the closest PDXP-relative PGP (Jeanclos et al., 2022).Together, the selectivity analysis of 7,8-DHF on a total of 12 structurally and functionally diverse protein-and non-protein-directed phosphatases show that 7,8-DHF preferentially inhibits PDXP, and that higher 7,8-DHF concentrations can also target the PDXP paralog PGP and the nucleotidase NT5C1A.

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 structures of PDXP bound to 7,8-DHF
To investigate the mechanism of PDXP inhibition in more detail, we co-crystallized homodimeric, fulllength murine and human PDXP (mPDXP, hPDXP) with this compound.7,8-DHF-bound murine PDXP co-crystallized with phosphate in the cubic space group I23, with protomer A containing the inhibitor and protomer B representing an inhibitor-free state (Figure 3-figure supplement 1a).The structure was refined following molecular replacement with full-length murine PDXP (here referred to as apo-mPDXP; 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).We additionally obtained two co-crystal structures of human PDXP with 7,8-DHF: one in a phosphate-containing and one in a phosphate-free form.Both forms crystallized in the tetragonal space group P4 3 2 1 2, and each protomer of both structures contained the inhibitor (Figure 3a).These structures were refined following molecular replacement with full-length human PDXP (PDB entry 2P27, here referred to as apo-hPDXP) to a resolution of 1.5 Å resulting in an R work /R free of 17.0/19.4%(phosphate-bound 7,8-DHF-hPDXP, PDB code 9EM1), and to a resolution of 1.5 Å resulting in an R work /R free of 18.2/20.5%(phosphate-free 7,8-DHF-hPDXP, PDB code 8S8A).Data collection and refinement statistics are summarized in Table 2.
Like their respective apo-forms, 7,8-DHF-bound murine and human PDXP homodimerize via their cap domains (Figure 3a and Figure 3-figure supplement 1a).The Cα atom-based alignment of the Table 1.Kinetic constants of pyridoxal 5'-phosphate phosphatase (PDXP)-catalyzed pyridoxal 5'-phosphate (PLP) hydrolysis in the presence of 7,8-dihydroxyflavone (7,8-DHF).The online version of this article includes the following source data for table 1: Source data 1.Kinetic constants of pyridoxal 5'-phosphate phosphatase (PDXP)-catalyzed pyridoxal 5'-phosphate (PLP) hydrolysis in the presence of 7,8-dihydroxyflavone (7,8-DHF).3).Hence, binding of the inhibitor did not result in significant changes in murine or human PDXP backbone conformations.All catalytic core residues and the Mg 2+ cofactor are correctly oriented in the presence of the inhibitor.We conclude that 7,8-DHF binding does not appear to impact the overall fold of murine or human PDXP.7,8-DHF was observed to only bind to one subunit (the A-chain) of murine PDXP (Figure 3-figure supplement 1a) with well-defined density (Figure 3-figure supplement 1b) 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 prevented by a salt bridge between Arg62 and Asp14 of a symmetryrelated A-protomer in the crystal (Figure 3-figure supplement 1c).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 murine 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.However, preventing mPDXP salt bridge formation by mutating Asp14 to Ala did not alter the efficacy of 7,8-DHF inhibition (Figure 3-figure supplement 1d; see also Figure 3d for the characterization of the PDXP-Arg62Ala variant).It is therefore currently unclear whether the mPDXP 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 mPDXP crystal packing is very stable (indeed, 7,8-DHF-bound mPDXP crystallized in the same cubic space group as apo-mPDXP, see Kestler et al., 2014, 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.
In contrast to murine PDXP, 7,8-DHF bound to human PDXP with a ratio of two inhibitors per homodimer (Figure 3a) and well-defined density (Figure 3-figure supplement 2a).Interestingly, the orientation of the inhibitor was markedly affected by the presence or absence of phosphate (Figure 3b).In the presence of phosphate, the inhibitor moiety that is closest to the Mg 2+ cofactor is the uncharged phenyl ring of 7,8-DHF.In the absence of phosphate, the inhibitor is flipped horizontally, with the in cyan.Mg 2+ ions are shown as deep purple spheres and phosphate ions are shown in sphere representation with the phosphorous atom in orange.(b) Orientation of 7,8-DHF in the active sites of human 7,8-DHF-PDXP in the presence or absence of phosphate.Structural details of bound 7,8-DHF and adjacent residues of the active sites are shown.Left, phosphate-containing 7,8-DHF-PDXP (wheat yellow, cartoon representation).Right, phosphate-free 7,8-DHF-PDXP (light blue, cartoon representation).7,8-DHF is shown in stick representation (cyan C-atoms).The corresponding amino acids in murine PDXP are given in parentheses (see also Figure 3-figure supplement 1e and f).(c) Comparison of the Mg 2+ coordination spheres.From left to right: human apo-PDXP (PDB: 2P27), human PDXP in complex with pyridoxal 5'-phosphate (PLP) (PDB: 2CFT), human PDXP in complex with 7,8-DHF in the presence of phosphate (PDB: 9EM1), human PDXP in complex with 7,8-DHF in the absence of phosphate (PDB: 8S8A).The catalytically essential Mg 2+ is shown as a deep purple sphere.In 2CFT, Mg 2+ was exchanged for Ca 2+ , which is shown here as a light brown-colored sphere.The online version of this article includes the following source data and figure supplement(s) for figure 3: Source data 1.Phosphatase activity and 7,8-dihydroxyflavone (7,8-DHF) sensitivity of pyridoxal 5'-phosphate phosphatase (PDXP) and PDXP variants (to Figure 3g).Irrespective of the orientation of 7,8-DHF in the PDXP active site, the inhibitor is embedded in a cavity that is exclusively formed by the active site of protomer A, without a contribution of the dimerization interface with protomer B. All PDXP residues found to engage in 7,8-DHF interactions are identical in murine and human PDXP (Figure 3-figure supplement 3).One side of this cavity is formed by the more polar residues Asp27, Asn60, Ser61, and Arg62 (identical amino acid residue numbering in mPDXP and hPDXP), whereas the opposite side is established by the more hydrophobic  residues Tyr150, His182, Pro183, and Leu184 (corresponding to Tyr146, His178, Pro179, and Leu180 in mPDXP).Adjacent to this hydrophobic stretch, the polar residue Glu152 (Glu148 in mPDXP) is located at the active site entrance, directly opposite of Arg62 on the more polar side of the 7,8-DHF binding channel (Figure 3b and Figure 3-figure supplement 1e).
Interestingly, Glu152 (or Glu148) and Arg62 can form an intramolecular salt bridge that obstructs the active site entrance (Figure 3-figure supplement 4).This interaction was observed in phosphatefree 7,8-DHF-hPDXP and phosphate-free PLP-hPDXP, as well as in apo-hPDXP and apo-mPDXP.In contrast, the 7,8-DHF binding pose that is dictated by the concomitant binding of phosphate and 7,8-DHF interferes with the Glu152 (Glu148)-Arg62 interaction in both, hPDXP and mPDXP (Figure 3figure supplement 4).Thus, although we did not find evidence for major cap/core or substrate specificity loop movements (Jeanclos et al., 2022;Kestler et al., 2014) in PDXP, the presence or absence of a salt bridge formed between the cap domain residue Glu152 (Glu148) and the core domain residue Arg62 indicates subtle conformational changes in PDXP that may mediate an opening or a closure of the active site entrance.
Inhibitor binding in the presence of phosphate is identical in human and murine PDXP (Figure 3figure supplement 1f) and appears to be primarily stabilized by two hydrogen bonds, as well as polar and non-polar interactions (Figure 3b, left panel).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, Glu152 (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 (Glu152).On the more hydrophobic side of the binding cavity, Tyr146 (Tyr150) forms π-electron stacking interactions with the pyrone ring of 7,8-DHF.In addition, the His178 (His182) imidazole group coordinates the 7,8-DHF phenyl ring via a cation-π interaction.His178 (His182), located in the substrate specificity loop, and Asn60 and Arg62 are also important for PLP binding (Kestler et al., 2014;Knobloch et al., 2015).
Inhibitor binding in the absence of phosphate is primarily stabilized by metal coordination and hydrogen bonds, as well as polar and non-polar interactions (Figure 3b, right panel).The 7-hydroxyl group of 7,8-DHF is involved in an octahedral Mg 2+ coordination, albeit with an elongated oxygen-Mg 2+ distance of 2.7 Å, leading to the displacement of a water molecule (Figure 3c).This inhibitor-based water displacement is not observed in the phosphate-containing murine or human 7,8-DHF-PDXP structures.The 8-hydroxyl group of 7,8-DHF forms a hydrogen bond with Arg239 and His182.The ketone group of the inhibitor participates in a water-bridged hydrogen bond to the carboxyl group of Asp27 and the backbone amine group of Gly33.Like in the phosphate-containing structure, the His182 imidazole group coordinates the 7,8-DHF phenyl ring via π-π stacking.In addition, Tyr150 forms edge to face π-π stacking interactions with the phenyl ring of 7,8-DHF.The side chains of the Table 3.Alignment of murine and human pyridoxal 5'-phosphate phosphatase (PDXP) structures.polar residues Asp27, Asn60, and, to some degree, also of Arg62, engage in van der Waals interactions with 7,8-DHF.
To verify the putative 7,8-DHF-PDXP interactions, we introduced single mutations into the binding interface.Asn60, Arg62, Tyr146, Glu148, and His178 in mPDXP 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 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 Eschericha coli (see Figure 3-figure supplement 5 for protein purity).Figure 3d (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 and the minor contribution of the weak van der Waals interactions between Asn60 and 7,8-DHF during inhibitor binding (Figure 3d, right panel).These data also suggest that Asn61 contributes to the limited efficacy of 7,8-mediated PDXP inhibition in vitro.
Based on the inhibitor-bound structures 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 (Figure 3-figure supplement 2b).In addition, BLI measurements (see Figure 2c) showed a relatively slow association rate and extended residence time of 7,8-DHF (τ=30.3s).This may indicate a reorganization of the Mg 2+ coordination due to inhibitor binding, and a reorientation of 7,8-DHF during the PDXP catalytic cycle.The reduced rate of product formation may account for the apparent mixed mode 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 (Figure 4a).This finding is in good agreement with the PLP increase resulting from PDXP loss in total hippocampal extracts (see Figure 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 PDXK expression (Figure 4b) and could not detect 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 PDXP deficiency (441.3±62.6 nmol PLP/g protein in DMSO solvent control-treated cells versus 482.7±130.4nmol PLP/g protein in 7,8-DHF-treated cells; data are mean values ± SE 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 Figure 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 Figure 4, 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 (di Salvo et al., 2012), yet causal links remain unclear.Aside from vitamin B6 administration, pharmacological strategies to elevate intracellular PLP levels are lacking.Here, we identify 7,8-DHF as a direct PDXP inhibitor that The online version of this article includes the following source data for figure 4: Source data 1.Quantification of pyridoxal 5'-phosphate (PLP) and PLP/PL levels in hippocampal neurons (to Figure 4a and c).
Source data 2. Quantification of western blots (to Figure 4b).
increases PLP levels in hippocampal neurons, validating PDXP as a druggable target to control intracellular PLP levels in the brain.We also present three high-resolution 7,8-DHF/PDXP co-crystal structures that will 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.
7,8-DHF was initially discovered as a small-molecule TrkB agonist with BDNF-mimetic activity (Jang et al., 2010).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 (Wang et al., 2022a).Therapeutics that target BDNF/TrkB signaling are thus of interest as disease-modifying agents in several brain disorders.Since BDNF does not cross the bloodbrain barrier, attempts have been made to develop small-molecule BDNF mimetics.Several candidates have been reported, including 7,8-DHF (Boltaev et al., 2017;Jang et al., 2010).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, TrkBdependent gene expression and cytoprotection, 7,8-DHF and other reported small-molecule TrkB agonists failed to activate TrkB in cells (Boltaev et al., 2017;Pankiewicz et al., 2021;Todd et al., 2014).An electrophysiological study in acute hippocampal slice preparations demonstrated that 7,8-DHF potentiates hippocampal mossy fiber-CA3 synaptic transmission in a TrkB receptor-independent manner (Kobayashi and Suzuki, 2018).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 (Jeanclos et al., 2019).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 pre-cooled 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 protein-depleted PLP (Ciapaite et al., 2023), 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%, vol/vol) 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 HPLC
Samples were derivatized as described (Talwar et al., 2003).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 the addition of perchloric acid (8 µL of a 72% [wt/vol] 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% [vol/vol] stock solution), and 2 µM pyridoxic acid (PA) 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% (vol/vol) MeOH; pH 6.5 as mobile phase.PL, PLP, and pyridoxic acid (PA) 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).

Phosphatase plasmids and cloning
N-terminally GST-tagged, human PDXP was in pGEX-4T-1 (Amersham Biosciences, Amersham, UK).N-terminally His 6 -SUMO-tagged human PDXP was cloned into pET-SUMO (coding for human SUMO; a kind gift of Dr. Pedro Friedmann Angeli, Rudolf-Virchow-Center, University of Würzburg, Germany); His 6 -SenP2 (EMBL Heidelberg) was in pET-M11.All other phosphatases were of murine origin and were subcloned into pET-M11 (EMBL), as described (Jeanclos et al., 2022).Murine Pdxp point mutants (generated by nested PCR) were subcloned into the NcoI (PciI for Psph) and EcoRI restriction sites of pET-M11, using Q5 Hot Start High-Fidelity DNA Polymerase (New England Biolabs, Frankfurt/Main, Germany).Pdxp-D14N was generated with the Platinum SuperFi II DNA Polymerase Mastermix according to mutagenesis protocol A provided by the manufacturer (Thermo Fisher Scientific) and cloned into pET-M11 as described above.Primers were purchased from Eurofins Genomics (Ebersberg, Germany), and all constructs were verified by sequencing (Microsynth Seqlab, Göttingen, Germany).

Expression and purification of recombinant proteins
Human His 6 -SUMO-tagged PDXP was grown in ZYP-5052 autoinduction medium for 7 hr at 37°C, followed by 48 hr at 21°C (Studier, 2005).All purification steps of murine PDXP and murine PGP were carried out exactly as described (Jeanclos et al., 2022).The purification of the His 6 -SUMO-tagged human PDXP was carried out exactly as described for murine His 6 -tagged PDXP (Jeanclos et al., 2022), except that human SenP2 protease was used to cleave the His 6 -SUMO-tag.N-terminally His 6 -tagged PDXP variants and His 6 -SenP2 were expressed as described for PDXP-WT (Jeanclos et al., 2022).With the exception of PDXP-D14, the His 6 -tag was not cleaved off.Human GST-PDXP was transformed into E. coli BL21(DE3) pLysS (Stratagene Europe/VWR, Darmstadt, Germany).Protein expression was induced with 0.5 mM isopropyl β-d-thiogalactopyranoside for 18 hr 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 batch-purified 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 (Jeanclos et al., 2022), except that the primary screen was done with PDXP, the counter-screen with PGP, and PDXP inhibitor hits were validated using PLP as a physiological PDXP substrate.
Flavone stocks were prepared at 10 mM in 100% DMSO.A constant final DMSO concentration of 0.4% was maintained under all conditions, and solvent control samples contained 0.4% DMSO without compounds.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 a 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, MA, 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
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 hr at RT in BLI assay buffer (250 mM TEA, 5 mM MgCl 2 , 250 mM NaCl, 0.005% [vol/vol] 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 (Wartchow et al., 2011).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% (vol/vol).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 s, association time 90 s, dissociation time 150 s.The resulting data were processed using the double reference method of the Octet analysis software for removal of drifts and well-to-well artifacts.Kinetic analyses were performed using the Octet analysis software.The steady-state analysis was carried out with OriginPro 2021b (OriginLab, Northampton, MA, USA), using a dose-response model for regression.Due to the poor solubility of 7,8-DHF, the highest concentration of 25 µM was not included in the analysis.

Figure 1 .
Figure 1.Role of pyridoxal 5'-phosphate phosphatase (PDXP) in hippocampal pyridoxal 5'-phosphate (PLP) homeostasis.(a) Age-dependent expression of pyridoxal kinase (PDXK) and PDXP in murine hippocampi.Left panels, representative western blots of three hippocampi for each genotype.The same blots were reprobed with α-actin antibodies as a loading control.The age of the investigated mice is indicated above the blots.Right panel, densitometric quantification of hippocampal PDXP and PDXK western blot signals, corrected by the corresponding actin signals.Young mice were Figure 1 continued on next page

Figure 2
Figure 2 continued on next page

Figure 3 .
Figure 3. X-ray crystal structures of human pyridoxal 5'-phosphate phosphatase (PDXP) in complex with 7,8-dihydroxyflavone (7,8-DHF).(a) The models were refined to a resolution of 1.5 Å for full-length human 7,8-DHF-PDXP with phosphate (PDB code 9EM1, colored in wheat yellow, left panel) and 1.5 Å for full-length human 7,8-DHF-PDXP without phosphate (PDB code 8S8A, colored in light blue, right panel).One protomer of each homodimeric PDXP is shown in cartoon representation and the other protomer in surface representation.7,8-DHF is displayed in sphere representation with its C-atoms Figure 3 continued on next page Water molecules are shown as blue spheres.(d) Verification of 7,8-DHF-PDXP interactions.Left panel, phosphatase activity of purified PDXP or the indicated PDXP variants.Data are mean values ± SD of n=3 biologically independent experiments.Right panel, determination of the IC 50 values of 7,8-DHF for purified PDXP or the indicated PDXP variants.Data are mean values ± SD of n=3 biologically independent experiments.Apparently missing error bars are hidden by the symbols.

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.‡ <I/σI> indicates the average of the intensity divided by its standard deviation.§ 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. ¶ R free same as R for 5% of the data randomly omitted from the refinement.The number of reflections includes the R free subset.

Figure 4 .
Figure 4. Effect of 7,8-dihydroxyflavone (7,8-DHF) on the pyridoxal 5'-phosphate (PLP)/PL ratio in cultured hippocampal neurons from WT or knockout of pyridoxal 5'-phosphate phosphatase (PDXP-KO) mice.(a) Effect of long-term PDXP deficiency on total PLP levels in hippocampal neurons.Data are mean values ± SE of n=4 biologically independent experiments.Statistical significance was assessed with a two-tailed, unpaired t-test.A representative image of primary hippocampal neurons stained for the neuronal marker protein MAP2 is shown in the insert (pixel intensities were color-inverted for better visualization).Scale bar, 100 µm.(b) Western blot analysis of PDXP and pyridoxal kinase (PDXK) expression in hippocampal neuron samples shown in (a).The same blots were reprobed with α-actin antibodies as a loading control.The densitometric quantification of PDXK signals is shown on the right; data are mean values ± SE of n=4 biologically independent experiments.(c) Effect of 7,8-DHF (20 µM, 45 min) or the DMSO solvent control (0.02% vol/vol, 45 min) on the PLP/PL ratio in hippocampal neurons of PDXP-WT or PDXP-KO mice.Source data are available for this figure.
18-42 days of age, older mice were 252-351 days of age; n=7 individual hippocampi were analyzed per group.Data are mean values ± SD.Statistical analysis was performed with unpaired, two-sided t-tests; p-values are indicated.(b)Age-dependent,total PLP concentrations in isolated hippocampi of PDXP-WT and knockout of PDXP (PDXP-KO) mice.PLP was derivatized with semicarbazide and analyzed by HPLC.Each symbol represents the result of the PLP determination in an individual hippocampus.Data were fitted by Gaussian least-squares analyses.(c)Determination of protein-bound and protein-depleted PLP in PDXP-WT and PDXP-KO hippocampal lysates of young (18-42 days of age) and older mice (252-352 days of age).The number of analyzed hippocampi is indicated in the bars.Data are mean values ± SD.Statistical analysis was performed with two-way ANOVA and Tukey's multiple comparisons test.Significant differences (adjusted p-values) in protein-depleted PLP levels are indicated.The exact age of analyzed mice is listed in Figure1-source data 3. Source data are available for this figure.
The data are mean values ± SEM of n=3 technically 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).Source data are available for this table.