Introduction

Vitamin B6 is an essential micronutrient that plays an important role in the nervous system (1, 2), with the vitamin B6 status affecting cognitive function at any age (3, 4). Population studies indicate that low vitamin B6 levels are common among older people (5), and suggest that vitamin B6 deficiency may influence memory performance and may contribute to age-related cognitive decline (6–9). Vitamin B6 deficiency is also associated with other conditions characterized by impaired learning and memory, including neuropsychiatric disorders (10–12), Alzheimer’s disease (13) and inflammation (14, 15). Nevertheless, the exact molecular mechanisms linking vitamin B6 to these pathologies are often insufficiently understood, and whether vitamin B6 supplementation improves cognition is unclear (4, 5, 16–22).

The term vitamin B6 encompasses the enzymatically interconvertible compounds pyridoxine, pyridoxamine, pyridoxal (collectively referred to as B6 vitamers) and their phosphorylated forms. Among these, only pyridoxal 5′-phosphate (PLP) is co-enzymatically active. In humans, PLP is known to be required for 44 distinct biochemical reactions, including the biosynthesis and/or metabolism of neurotransmitters, amino acids, lipids, and glucose. In addition, B6 vitamers display antioxidant and anti-inflammatory functions (23–26).

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, 27–30).

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 7,8-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 (31). However, a direct TrkB agonistic activity of 7,8-DHF has been called into question (32–36). 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.

Results

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).

Figure 1.

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 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-dihydroxyflavone 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 Fig. 2 – figure 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 EC₅₀/IC₅₀ value determinations. Of these, 46 inhibitor hits were selected and counter-screened against phosphoglycolate phosphatase (PGP), the closest PDXP relative (39, 40). Eleven of the PDXP inhibitor hits (with an IC₅₀ PDXP <20 µM, and IC₅₀ PDXP < IC₅₀ PGP or no activity against PGP) were subsequently validated in a secondary assay, using PLP as a physiological PDXP substrate (see Fig. 2 – figure supplement 2 for all 11 inhibitor hits). Only one PDXP-selective inhibitor hit (7,8-dihydroxyflavone/7,8-DHF, a naturally occurring flavone) blocked PDXP-catalyzed PLP dephosphorylation (IC₅₀ ∼1 µM).

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). We next examined whether commercially available 7,8-DHF analogs might be more potent or efficacious PDXP inhibitors. We tested flavone, 3,7-dihydroxyflavone, 5,7-dihydroxyflavone (also known as chrysin), 3,5,7-trihydroxyflavon (galangin), 5,6,7-trihydroxyflavone (baicalein) and 3,7,8,4’-tetrahydroxyflavone. Figure 2b shows that of the tested 7,8-DHF analogs, only 3,7,8,4’-tetrahydroxyflavone was able to inhibit PDXP, albeit with an IC₅₀ of 2.5 µM and thus slightly less potently than 7,8-DHF. These results suggest that hydroxyl groups in positions 7 and 8 of the flavone scaffold are required for PDXP inhibition. The efficacy of PDXP inhibition by 3,7,8,4’-tetrahydroxyflavone was not substantially increased at concentrations >40 µM (relative PDXP activity at 40 µM: 0.46 ± 0.05; at 70 µM: 0.38 ± 0.15; at 100 µM: 0.37 ± 0.09; data are mean values ± S.D. of n=6 experiments). Concentrations >100 µM could not be assessed due to impaired PDXP activity at the DMSO concentrations required for solubilizing the flavone.

Figure 2:

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. 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 ± 0.3 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). As expected, 5,7-dihydroxyflavone 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 10⁶·M⁻¹s⁻¹. Interestingly, 7,8-DHF showed a slow k_on of 1.05·10⁴ M⁻¹s⁻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.

Selectivity of 7,8-DHF

PDXP is a member of the large family of haloacid dehalogenase (HAD)-type hydrolases (43). HAD phosphatases are Mg²⁺-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, 43–45).

To probe the selectivity of 7,8-DHF for PDXP, a C2-capped HAD phosphatase, we tested eight other mammalian HAD phosphatases, including two other C2-, four C1- and two C0-type enzymes. In addition, we analyzed the activity of 7,8-DHF towards the prototypical tyrosine phosphatase PTP1B (which is known to be sensitive to specific flavonoids, ref. (46)); the serine/threonine protein phosphatase calcineurin (PP2B); and a DNA/RNA-directed alkaline phosphatase (calf intestinal phosphatase, CIP) (Fig. 2d). When assayed at nominal concentrations of 5, 10 and 40 µM (i.e., up to ∼40-fold above the IC₅₀ value for PDXP-catalyzed PLP-dephosphorylation), 7,8-DHF was completely inactive against six of the tested enzymes. At the highest tested concentration of 40 µM, 7,8-DHF weakly inhibited PTP1B and the polynucleotide kinase-3’-phosphatase PNKP and appeared to increase the activity of calcineurin. As expected, 7,8-DHF inhibited PGP, the closest PDXP relative, with an IC₅₀ value of 4.8 µM. This result is consistent with the criteria applied during the initial counter-screen (see above). In addition to PGP, 7,8-DHF inhibited the C1-capped soluble cytosolic 5’-nucleotidase 1A (NT5C1A) with an IC₅₀ value of ∼10 µM. NT5C1A is an AMP hydrolase expressed in skeletal muscle and heart (47), which is also sensitive to inhibition by small molecules that target the closest PDXP-relative PGP (40). 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.

Table 1.

Co-crystal structures of PDXP bound to 7,8-DHF

To investigate the mechanism of PDXP inhibition in more detail, we co-crystallized homo-dimeric, full-length 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 (Fig. 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 P 4₃2₁2, and each protomer of both structures contained the inhibitor (Fig. 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.

Figure 3.

Table 2.

Like their respective apo-forms, 7,8-DHF-bound murine and human PDXP homodimerize via their cap domains (Fig. 3a and Fig. 3 – figure supplement 1a). The Cα atom-based alignment of the structures representing murine apo-PDXP and murine 7,8-DHF-bound PDXP resulted in root mean square (RMS) deviations in the range of 0.43-0.71 Å. Even smaller values were obtained when human apo-PDXP, human PLP-bound PDXP and human 7,8-DHF-bound PDXP were superimposed with RMS deviations in the range from 0.29-0.54 Å (Table 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 Mg²⁺ 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.

Table 3.

7,8-DHF was observed to only bind to one subunit (the A-chain) of murine PDXP (Fig. 3 – figure supplement 1a) with well-defined density (Fig. 3 – figure supplement 1b) and full occupancy since its average B-factor of 45.8 Ų 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 symmetry-related A-protomer in the crystal (Fig. 3 – figure supplement 1c). The χ₁ and χ₂ 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 (Fig. 3 – figure supplement 1d; see also Fig. 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 ref. (48), 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 (Fig. 3a) and well-defined density (Fig. 3 – figure supplement 2a). Interestingly, the orientation of the inhibitor was markedly affected by the presence or absence of phosphate (Fig. 3b). In the presence of phosphate, the inhibitor moiety that is closest to the Mg²⁺ cofactor is the uncharged phenyl ring of 7,8-DHF. In the absence of phosphate, the inhibitor is flipped horizontally, with the hydroxylated chromone substructure of 7,8-DHF now located closest to the Mg²⁺ ion (Fig. 3b, compare left and right panels). The inhibitor localization in the presence of phosphate was identical in human and murine PDXP (Fig. 3 – figure supplement 2e). The localization of the phosphate ion that co-crystallized with 7,8-DHF-bound human or murine PDXP overlaps exactly with the localization of the PLP phosphate moiety introduced from PDB code 2CFT (human PDXP in complex with PLP) for visualization purposes (Fig. 3c and Fig. 3 – figure supplement 2b), indicating that the phosphate ion is bound in a catalytically relevant position.

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 (Fig. 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 (Fig. 3b and Fig. 3 – figure supplement 1e).

Interestingly, Glu152 (or Glu148) and Arg62 can form an intramolecular salt bridge that obstructs the active site entrance (Fig. 3 – figure supplement 4). This interaction was observed in phosphate-free 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 (Fig. 3 – figure supplement 4). Thus, although we did not find evidence for major cap/core or substrate-specificity loop movements (40, 48) 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 (Fig. 3 – figure supplement 1f) and appears to be primarily stabilized by two hydrogen bonds, as well as polar and non-polar interactions (Fig. 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 (48, 49).

Inhibitor binding in the absence of phosphate is primarily stabilized by metal coordination and hydrogen bonds, as well as polar and non-polar interactions (Fig. 3b, right panel). The 7-hydroxyl group of 7,8-DHF is involved in an octahedral Mg²⁺ coordination, albeit with an elongated oxygen-Mg²⁺ distance of 2.7 Å, leading to the displacement of a water molecule (Fig. 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 Asp239 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 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 E. coli (see Fig. 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 (Fig. 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 (Fig. 3 – figure supplement 2b). In addition, BLI measurements (see Fig. 2c) showed a relatively slow association rate and extended residence time of 7,8-DHF (τ=30.3 s). This may indicate a reorganization of the Mg²⁺-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 (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.

Figure 4.

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 ± 130.4 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 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.

The discovery of 7,8-DHF as a direct PDXP inhibitor was unexpected. Interestingly, numerous in vivo studies have reported the effectiveness of 7,8-DHF in brain disorder models, including rodent models of Alzheimer’s disease (50–57), depression (58–63), schizophrenia (64–68), epilepsy (69, 70) and autism (71–74). Although PLP deficiency is thought to contribute to the respective human conditions (3, 75, 76), PLP-dependent processes have not yet been considered in the context of 7,8-DHF-induced effects.

7,8-DHF was initially discovered as a small-molecule TrkB agonist with BDNF-mimetic activity (77). 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/TrkB-signaling 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, 77). 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 (33–35). 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 (78). 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.

Materials and methods

Materials

Unless otherwise specified, all reagents were of the highest available purity and purchased from Sigma Aldrich (Schnelldorf, Germany). 3,7,8,4’-Tetrahydroxyflavone was obtained from Ambinter (Orléans, France), all other flavones were from Sigma Aldrich.

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 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 (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 H₂O, and placed at −80°C for at least 30 min.

For immunocytochemistry, DIV21 primary hippocampal neurons were fixed with 4% (w/v) paraformaldehyde in phosphate-buffered saline (PBS) for 15 min at RT. After washing twice with PBS, 50 mM NH₄Cl was added for 10 min. Cells were then permeabilized with 0.1% (v/v) Triton X-100 and blocked with 5% (v/v) goat serum in PBS for 30 min at 22°C. Cells were incubated with mouse monoclonal anti-MAP2 antibodies (1:500 dilution, clone MAB3418, Millipore, Darmstadt, Germany) for 1 h in 5% goat serum/PBS at 22°C. Alexa-488-labeled secondary goat anti-mouse antibodies (1:500 dilution; Dianova, Hamburg, Germany) were applied for 1 h. Nuclei were counter-stained with 4’,6-diamino-2-phenylindole (DAPI), and slides were mounted with Mowiol. Images were acquired using an inverted IX81 microscope equipped with an Olympus UPLSAPO 60× oil objective (numerical aperture: 1.35) on an Olympus FV1000 confocal laser scanning system, using a FVD10 SPD spectral detector and diode lasers of 405 nm (DAPI) and 495 nm (Alexa488).

Determination of PLP and PL by high-performance liquid chromatography (HPLC)

Samples were derivatized as described (79). 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 Na₂HPO₄, 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).

Western blotting

Tissue or cell homogenates (prepared as detailed above for HPLC analysis) were extracted with 4 × RIPA buffer (final concentration, 50 mM Tris, pH 7.5; 150 mM NaCl, 1% (v/v) Triton X-100, 0.5% (v/v) sodium deoxycholate, 0.1% (w/v) SDS, 1 mM 4-(2-aminoethyl)benzenesulfonyl fluoride (Pefabloc), 5 μg/mL aprotinin, 1 μg/mL leupeptin, 1 μg/mL pepstatin) for 15 min at 4 °C under rotation, and lysates were clarified by centrifugation (20,000 × g, 15 min, 4 °C). Protein concentrations in the supernatants were determined using the Micro BCA Protein Assay Kit (Thermo Fisher Scientific). Proteins were separated by SDS-PAGE and transferred to nitrocellulose membranes by semidry-blotting. Antibodies were purchased from the following providers: Merck Millipore (mouse monoclonal α-actin mAb1501, dilution 1:5000); Cell Signaling (rabbit monoclonal α-PDXP clone C85E3, dilution 1:1000; Cell Signaling, Danvers/Massachusetts, USA), Sigma Aldrich (rabbit polyclonal α-PDXK/AB1, #AV53615, dilution 1:1000), and Thermo Fisher Scientific (rabbit polyclonal α-PNPO, #PA5-26400, dilution 1:1000, as used in ref. (30)). Western blots were densitometrically quantified with NIH ImageJ, version 1.45i.

Phosphatase plasmids and cloning

N-terminally GST-tagged, human PDXP was in pGEX-4T-1 (Amersham Biosciences, Amersham, UK). N-terminally His₆-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); His₆-SenP2 (EMBL Heidelberg) was in pET-M11. All other phosphatases were of murine origin and were subcloned into pET-M11 (EMBL), as described (40). 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. The following primers (oligonucleotide sequence 5’-3’; fwd, forward; rev, reverse) were used:

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 His₆-SUMO-tagged PDXP was grown in ZYP-5052 autoinduction medium for 7 h at 37° C, followed by 48 h at 21°C (80). All purification steps of murine PDXP and murine PGP were carried out exactly as described (40). The purification of the His₆-SUMO-tagged human PDXP was carried out exactly as described for murine His₆-tagged PDXP (40), except that human SenP2 protease was used to cleave the His₆-SUMO-tag. N-terminally His₆-tagged PDXP variants and His₆-SenP2 were expressed as described for PDXP-WT (40).With the exception of PDXP-D14, the His₆-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 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 MgCl₂; 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 with PGP, and PDXP inhibitor hits were validated using 5’-pyridoxal phosphate (PLP) as a physiological PDXP substrate.

IC₅₀ determinations, enzyme kinetics, and compound selectivity

Conditions for enzymatic assays were as previously published (40), with the following modifications. Bovine brain calcineurin (PP2B, Sigma Aldrich #C1907) activity against the PKA regulatory subunit type II (phosphopeptide DLDVPIPGRFDRRVpSVAAE; Sigma Aldrich #207008) was assayed at 37°C in 100 mM NaCl, 50 mM Tris, 6 mM MgCl₂, 0.5 mM CaCl₂, 0.5 mM DTT, 0.025% (v/v) NP40; pH 7.5. Recombinant human PTP1B (amino acids 1-321, Cayman Chemical, Ann Arbor, Michigan, USA) activity against the Tyr⁹⁹² autophosphorylation site of EGFR (DADEpYLIPQQG; Santa Cruz Biotechnology, Heidelberg, Germany) was assayed at 30°C in 150 mM NaCl, 50 mM 2-(N-morpholino)ethanesulfonic acid, 1 mM EDTA; pH 7.2. Murine PDXP-D14A was assayed exactly like PDXP-WT in 30 mM triethanolamine, 5 mM MgCl₂, 30 mM NaCl; pH 7.5, supplemented with 0.01% (v/v) Triton X-100.

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 (A₆₂₀) 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 IC₅₀ 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 MgCl₂, 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 (81). 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-to-well 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. Due to the poor solubility of 7,8-DHF, the highest concentration of 25 µM was not included in the analysis.

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 MgCl₂; pH 7.4) was supplemented with a three-fold molar excess of the flavone. Prism-shaped crystals of 7,8-DHF-bound murine 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. Human PDXP crystals were grown at 20 °C in 0.1 M Tris (pH 8.5) and 1 M diammonium hydrogen phosphate, or in 0.1 M HEPES (pH 7.0), 15% (v/v) Tacsimat pH 7.0 (Hampton Research, Aliso Viejo, USA) and 2% (w/v) PEG 3350 using the sitting-drop vapor diffusion method. Crystals were cryoprotected for flash-cooling in liquid nitrogen by soaking in mother liquor containing 25% (v/v) glycerol. Diffraction data of murine PDXP in complex with 7,8-DHF were collected from flash-cooled crystals at a temperature of 100 K on beamline BL 14.1 at the BESSY synchrotron (Helmholtz Zentrum Berlin, Germany). Diffraction data of 7,8-DHF bound to human PDXP were collected on beamline ID23-2 at the ESRF (Grenoble, France) [https://data.esrf.fr/doi/10.15151/ESRF-ES-1409594895]. Diffraction data were processed using XDS (82) and further analysed with Aimless (83) of the CCP4 suite (84). The structures of 7,8-DHF-PDXP were solved by molecular replacement with the program Phaser (85) with the structure of the murine PDXP (PDB entry 4BX3) or human PDXP (PDB entry 2P27) as search models, and refined with Phenix (86). Model building was carried out in COOT (87). Structural illustrations were prepared with PyMOL 2.5.1 (88).

Acknowledgements

We thank Carola Seyffarth and Nicole Bader for excellent technical assistance, Dr. Jochen Kuper for collecting the murine PDXP diffraction data, the staff at beamline BL14.1 of the BESSY synchrotron, and the staff at beamline ID23-2 of the ESFR synchrotron for technical support. A part of this work was initially funded by the DFG Collaborative Research Center SFB688 (TP A11 to A.G.).

Data availability statement

The previously published PDB entry 4BX3 of murine apo-PDXP [http://doi.org/10.2210/pdb4BX3/pdb], 2P27 of human apo-PDXP [http://doi.org/10.2210/pdb2P27/pdb] and 2CFT of PLP-bound human PDXP [http://doi.org/10.2210/pdb2CFT/pdb] are used in this manuscript. X-ray crystallographic data of 7,8-DHF-bound murine PDXP generated in this study have been deposited in the PDB and can be accessed under the PDB entry 8QFW [http://doi.org/10.2210/pdb8QFW/pdb]. X-ray crystallographic data of 7,8-DHF-bound human PDXP generated in this study can be accessed under the PDB entries 8S8A (without phosphate) [http://doi.org/10.2210/pdb8S8A/pdb] and 9EM1 (with phosphate) [http://doi.org/10.2210/pdb9EM1/pdb]. The corresponding raw diffraction images have been deposited in the Xtal Raw Data Archive and can be accessed under the XRDA entries 8QFW [https://xrda.pdbj.org/entry/8qfw], 8S8A [https://xrda.pdbj.org/entry/8s8a], and 9EM1 [https://xrda.pdbj.org/entry/9em1].

Author contributions

Conceptualization, A.G.; validation, E.J.; formal analysis, M.B., C.Z., L.W., S.B., M.N., H.S., E.J., A.G.; investigation, M.B., C.Z., L.W., A.K., K.H., S.B., M.N., E.J.; resources, C.V., M.N., J.P.v.K., H.S.; writing – original draft preparation, A.G. with input from E.J. and H.S.; writing – revised manuscript: A.G. and M.B. with input from H.S. and S.B.; visualization, M.B., C.Z., L.W., S.B., E.J., A.G.; supervision, E.J., A.G.; project administration, A.G.; funding acquisition, A.G.

Competing interests

A.G. is a recipient of a research project grant from Boehringer Ingelheim International GmbH. This project funding is independent of and has no overlap with the work described in this manuscript. The other authors declare no competing interests.

Supplementary information

Figure 1 – figure supplement 1.

Figure 2 – figure supplement 1.

Figure 2 – figure supplement 2.

Figure 2 – figure supplement 3.

Figure 3 – figure supplement 1.

Figure 3 – figure supplement 2.

Figure 3 – figure supplement 3.

Figure 3 – figure supplement 4.

Figure 3 – figure supplement 5.