The expression of four pyridoxal kinase (PDXK) human variants in Drosophila impacts on genome integrity

In eukaryotes, pyridoxal kinase (PDXK) acts in vitamin B6 salvage pathway to produce pyridoxal 5′-phosphate (PLP), the active form of the vitamin, which is implicated in numerous crucial metabolic reactions. In Drosophila, mutations in the dPdxk gene cause chromosome aberrations (CABs) and increase glucose content in larval hemolymph. Both phenotypes are rescued by the expression of the wild type human PDXK counterpart. Here we expressed, in dPdxk1 mutant flies, four PDXK human variants: three (D87H, V128I and H246Q) listed in databases, and one (A243G) found in a genetic screening in patients with diabetes. Differently from human wild type PDXK, none of the variants was able to completely rescue CABs and glucose content elicited by dPdxk1 mutation. Biochemical analysis of D87H, V128I, H246Q and A243G proteins revealed reduced catalytic activity and/or reduced affinity for PLP precursors which justify this behavior. Although these variants are rare in population and carried in heterozygous condition, our findings suggest that in certain metabolic contexts and diseases in which PLP levels are reduced, the presence of these PDXK variants could threaten genome integrity and increase cancer risk.

the hypothesis that vitamin B 6 would promote antiproliferative effects on cancer cells 10 . In addition, epidemiological studies reported that elevated circulating amounts of B 6 vitamers as well as an intense consumption of vitamin B 6 -containing food correlate with a reduced incidence of several distinct tumors including colorectal cancer 11 . Furthermore, it has been reported that high expression levels of PDXK positively correlate with survival of non-small cell lung cancer (NSCLC) patients 12 . Although many studies converge towards a protective role of B 6 in cancer, underlying molecular mechanisms are not completely understood. Recently, several lines of evidence are accumulating which pinpoint to DNA damage as a possible link between metabolism and cancer. In particular, it has been proposed that under metabolic stress conditions or in the case of reduced availability of necessary nutrients some cellular processes such as DNA acetylation/methylation, synthesis of DNA precursors and ROS production can be altered causing DNA damage which can drive cells toward cancer 13 . Vitamin B 6 is an antioxidant molecule and plays an important role in one-carbon metabolism, a set of reactions involved in the transfer of one-carbon groups which are at the basis of amino acid and nucleotide metabolism. As a consequence, vitamin B 6 could represent a possible candidate to mediate the cross talk between metabolism and DNA damage. In line with this hypothesis, using Drosophila as model system, we demonstrated that mutations in the dPdxk gene cause chromosome aberrations (CABs) rescued by PLP supplementation. The same effect is produced by treating wild type flies with PLP analogues such as 4-deoxypyridoxine (4-DP) or inhibitors of PLP-dependent enzymes like cycloserine and penicyllamine 14 . Besides eliciting CAB formation, dPdxk 1 mutations increase the glucose content in larval hemolymph. We demonstrated that in dPdxk 1 mutants hyperglycemia and CABs are interconnected by a cause-effect relationship, in which high glucose is largely responsible for CABs. High glucose triggers AGE formation, which through ROS production leads to CABs 14 . The role of PDXK in chromosome integrity maintenance has also been demonstrated in yeast showing that mutations in the BUD16 gene, the PDXK ortholog, cause gross chromosome rearrangements largely mediated by altered DNA synthesis 15 . The impact of low PLP levels on genome integrity has also been tested on human cells. HeLa cells deprived of PLP by RNA interference directed against the PDXK gene showed chromosome aberrations (38.5% vs 2.0% in mock cells). Moreover, the treatment of mock HeLa cells with 4-DP causes chromosome aberrations 14 , p53 binding protein 1 (53BP1) and γ-H2AX repair foci accumulation 15 . Confirming the evolutionarily conserved role of PDXK gene in DNA integrity maintenance, we have previously demonstrated that the expression in Drosophila of the wild type human PDXK gene in a dPdxk 1 background fully rescues CABs 14 .
In this work, we investigated whether human PDXK variants present in the population can impact on DNA integrity and be considered predictive of an increased cancer risk. For this purpose, we expressed four human PDXK variants (carrying missense mutations) into dPdxk 1 mutant flies and tested them for CABs showing that none of them was able to completely rescue the CAB phenotype. Biochemical analysis of all these variants reveled a compromised catalytic activity and/or affinity for their substrates, which explained their "loss of function" behavior. These results translated to humans suggest that mutations in PDXK gene can impact on genome integrity and predispose to cancer.

Results
Genetic and biochemical basis of the study. We previously demonstrated that mutations in the Drosophila dPdxk gene cause lethality at the third larval instar and formation of CABs which can be rescued by PLP supplementation 14 . siRNAs directed against human PDXK gene induce the formation of CABs in fibroblasts and HeLa cells 14 and, in addition, the PLP analogue 4-deoxypyridoxine (4-DP) increases the formation of 53BP1 repair foci in HeLa cells 15 . These findings prompted us to speculate that humans carrying mutations in the PDXK encoding gene could have an increased propensity to accumulate chromosome aberrations and a consequent increased risk to develop malignances. To evaluate this hypothesis, we decided to express human PDXK loss-of-function variants in flies homozygous for the dPdxk 1 mutation 14 and test them for their effects on CABs. The rationale of this strategy comes from our previous data 14 showing that a wild type copy of the human PDXK gene can rescue CABs when expressed in dPdxk 1 mutants.
We analyzed four human PDXK variants: Asp87His (D87H), Val128Ile (V128I), His246Gln (H246Q) and Ala243Gly (A243G). The first three have been picked up from the Exome variant server (Exome variant server, http://evs.gs.washington.edu/EVS/) which contains numerous human PDXK variants not yet associated to any disease. The variants have been chosen considering their putative damaging effects predicted in silico by the PolyPhen-2 software and their evolutionarily conservation of the mutated residues in Drosophila. In particular D87H displays the highest damaging score (1.0) and concerns a conserved position in Drosophila and human PDXKs; V128I and H246Q carry mutations in invariant positions (http://www.flyrnai.org/cgi-bin/DRSC_orthologs.pl) and in addition display high damaging scores (0.99 and 0.98 respectively) (Fig. 1A,C). These variants are very rare in the population (their frequency ranging from 2.84e-5 to 7.97e-6; https://gnomad.broadinstitute.org/) and are carried in heterozygous state. A243G (Fig. 1A,C) is absent in major databases, displays a damaging score of 0.72 ( Supplementary Fig. S1) and has been found in a genetic screening in patients with gestational diabetes. The rationale for the aforementioned investigation was the previously reported beneficial effect of vitamin B 6 on gestational diabetes and its possible effect on insulin secretion in a murine model [16][17][18] .
None of the chosen variants concerns amino acid residues that are directly involved in substrate binding or catalysis; however, these residues are located in, or connected to, pivotal structural regions of the enzyme. PDXK is a dimer of two identical monomers. Each subunit is made by 9 α-helices and 11 β-strands that form a central β-sheet flanked by helices on both sides. Asp87 is placed on the C-terminal end of an active site loop connecting α3 helix and strand β4, which plays a crucial role in substrate binding. Tyr84, which is part of this loop, directly interacts with the B 6 vitamer substrate stacking to its pyridine ring ( Supplementary Fig. S2A) (PDB code: 3KEU and Musayev et al. 19 ). The Val128 residue is part of another active site loop, located between strands β6 and β7 ( Supplementary Fig. S2B), that closes on the active site as ATP binds to the enzyme. This loop, which is referred www.nature.com/scientificreports www.nature.com/scientificreports/ to as a flap in pyridoxal kinases, provides hydrogen bond interactions to the ATP βand γ-phosphates (Tyr127, adjacent to Val128, binds to the γ-phosphate of ATP), and is believed to sequester ATP for catalysis, preventing its unproductive hydrolysis in the absence of a bound B 6 vitamer [20][21][22] . Thus, variants of residues V128 and D87, which are spatially quite close to each other, are expected to affect substrate binding. On the other hand, A243 and H246, which are also close together, are located at the C-terminal end of helix α7, formed by residues 231-234, on the opposite side of the monomer with respect to the active site and therefore at a distance from it ( Supplementary  Fig. S2B). However, the N-terminus of this helix, which is positioned at the active site, contributes with its positive charge to binding of the PLP product of the reaction catalyzed by PDXK, stabilizing its phosphate moiety 23 . A243G and H246Q variants are predicted to have milder effects than V128I and D87H on the enzyme catalytic properties.
Human pDXK variants fail to rescue cABs in dPdxk 1 mutant flies. By site-directed mutagenesis, we generated four constructs of human HA-tagged PDXK cDNA, each containing a variant, and introduced them in flies by germline mediated transformation. We validated the expression of these constructs by western blot analysis using an anti HA antibody (Fig 1B and Supplementary Fig. S3). By performing suitable crosses (described in An antibody anti-HA was used to detect PDXK proteins. α-tubulin served as loading control (LC). (C) Clustal Omega pairwise sequence alignment of Drosophila melanogaster Pdxk (NP_996031.1) with that of Homo sapiens (NP_003672.1). Dash indicates the identical residues, colon indicates the conserved residues. Mutated residues carried by variants examined in this study are boxed.
Methods), we introduced these variants in dPdxk 1 mutant flies and tested them for CABs in DAPI stained brain preparations from third instar larvae. As reported in Fig. 2A,B, the D87H, V128I, H246Q and A243G variants were unable to completely rescue CABs unless the larvae were reared in a medium containing PLP (1 mM). The expression of either wild type or variant PDXK forms in a dPdxk 1 /+ background did not produce CABs, allowing us to exclude any dominant negative effect (Fig. 1B).
Taken together, these findings suggest that all tested variants behave as loss of function alleles which impact on genome integrity. pDXK human variants expressed in dPdxk 1 flies impact on glucose metabolism. Mutations in the dPdxk gene, besides promoting CAB formation, increase glucose content in the hemolymph, generating "diabetic" flies 14 . We have previously proposed a model in which high glucose levels in the hemolymph resulting from the dPdxk 1 mutation is largely responsible for CABs because it induces an overproduction of Advanced Glycation End products (AGEs). In turn, AGEs produce oxygen reactive species (ROS) that are well recognized for damaging DNA and leading to CAB formation 14 . To verify whether the expression of D87H, V128I, H246Q and A243G variants influenced glucose homeostasis, we tested glucose levels and found that none of the variants was able to significantly reduce hyperglycemia caused by dPdxk 1 mutation, which was instead decreased by PLP treatment (Fig. 3A). In addition, immunostaining experiments showed that the dPdxk 1 cells expressing D87H, V128I, H246Q and A243G variants accumulated AGEs, which were further increased by 1% glucose treatment (Fig. 3B,C), according to the hypothesis that AGEs are largely responsible for CABs in PLP depleted cells 9,14,24,25 . H246Q, D87H, V128I and A243G variants respond differently to PLP precursors. To investigate why PDXK variants failed to rescue CABs in the dPdxk 1 background, we examined the effect of PLP precursors (PL, PM and PN) on CAB frequency in brains from larvae expressing D87H, V128I, H246Q and A243G variants. As reported in Fig. 4A-D we tested PLP precursors at two different concentrations (0.5 and 1 mM) and we found that each variant behaved differently. In brains expressing D87H, PL treatment did not rescue CABs. A feeble rescue effect was observed with PM (1 mM). In contrast, PN reduced significantly CAB frequency but only at 1 mM concentration. In brains expressing V128I, neither at 0.5 nor at 1 mM concentration PL was able to rescue CAB frequency, whereas both PM and PN rescued CABs at 1 mM concentration. In H246Q expressing cells PL and PM reduced CAB frequency only at 1 mM concentration, whereas PN reduced CAB frequency also at 0.5 mM concentration. In A243G expressing neuroblasts all precursors rescued CABs but only at 1 mM concentration. Note that PLP treatment rescued CABs at both concentrations in all tested variants.
These data suggest that the replacement of H246, V128, D87 and A243 amino acids with those found in the respective variants can impact on PDXK functionality by altering in different ways its capability to phosphorylate the substrates.
Biochemical characterization of H246Q, D87H, V128I and A243G variants. To better understand why the four PDXK mutant proteins responded differently to PLP precursors, variant PDXK enzyme forms were recombinantly expressed in Escherichia coli, purified to homogeneity and characterized with respect to www.nature.com/scientificreports www.nature.com/scientificreports/ their catalytic properties with PL, PN and PM as substrates. As shown in Table 1, all the four mutant enzymes showed altered kinetic parameters. In particular, the D87H mutation strongly increased K M for PL, and to a lesser extent also increased K M for PN and PM, leaving K M for ATP and k cat almost unaltered. These observations are well-matched with the location of the Asp87 residue on an active site loop that plays a crucial role in the B 6 vitamer substrate binding, and in particular with the direct interaction between the adjacent Tyr84 with the B 6 vitamer pyridine ring (Supplementary Fig. S2A). The V128I mutation drastically increased K M for PL and K M for ATP with this vitamer, whereas it did not affect k cat . This was expected, considering the role of the active site loop containing Val128 in substrate binding and specifically in the interaction with ATP ( Supplementary Fig. S2B). K M for PN and PM were also increased, although with these vitamers K M for ATP and k cat were unaltered. This latter observation is difficult to explain on the basis of the available observed results; however, it suggests that binding of B 6 vitamers different from PL may affect the modality of ATP binding by the loop containing Val128. The H246Q mutation somewhat reduced the affinity for ATP when using PL as substrate and the affinity for PM, while it did not affect K M for PL and PN. In addition, it had the effect to halve k cat with PL. Finally, A243G mutation displayed a behavior very similar to that showed by the H246Q mutation. Given the distance of Ala243 and His246 from the active site, the observed alteration of the kinetic parameters, although relatively mild, testifies that the mutation of these residues is somehow transmitted to the active site of the enzyme.
Taken together, these findings indicate that the specific changes introduced in the four examined variants reduce the PDXK functionality threatening genome integrity and impairing glucose homeostasis. www.nature.com/scientificreports www.nature.com/scientificreports/

Discussion
Limitations imposed by human subject research can be overcome by generating models of human diseases in experimental organisms. Due to widely conserved pathways which govern metabolism, Drosophila has been retained a precious organism for the study of metabolic human genetic diseases, either as a means of validating  the causative nature of candidate genetic variants found in patients, or as a means of obtaining functional information about novel disease-linked genes.
Here we used Drosophila to validate the effects of four human variants of the pyridoxal kinase encoding gene on genome integrity and glucose metabolism. We previously showed that in Drosophila the depletion of this enzyme results in CABs which represent the consequence of hyperglycemia, another phenotype also elicited by Pdxk depletion. Also in human cells PDXK silencing produces CABs suggesting functional conservation 14 .
Low PDXK expression levels have been correlated to lung cancer 12 and more recently also to insulin resistance which leads to type 2 diabetes 26 . However in literature, no robust data exist that link specific PDXK variants to any specific disease, except for a recent work in which two biallelic mutations in PDXK have been associated to polyneuropathy 27 . Conversely, numerous variants (also in homozygous condition) of the PNPO enzyme, which acts downstream of PDXK in the salvage pathway, have been associated to epilepsy 28 . The reason of this difference could be that cells with an impaired PNPO function can anyway produce some PLP throughout the conversion of PL into PLP mediated by PDXK; in contrast in cells lacking PDXK, phosphorylation of B 6 vitamers does not occur and PLP is not produced at all. Thus, it is reasonable to expect, also by considering the wide spectrum of functions covered by PLP, that severe mutations of the PDXK gene seriously compromise early developmental stages causing lethality in the homozygous condition.
Databases contain PDXK variants (from heterozygous carriers), some of which carry changes in amino acid positions that are well conserved in Drosophila. Here we used Drosophila to validate the effects of four human variants (D87H, V128I, H246Q reported in the Exome variant server and the novel A243G variant) of the pyridoxal kinase encoding gene on genome integrity and glucose metabolism.
We found that none of the human variants expressed in dPdxk 1 mutant flies could rescue CABs, differently from what observed with the wild type copy of the PDXK human gene 14 . This finding reinforces the notion (suggested by in silico analysis) that the examined variants are loss-of-function alleles. The expression of all these variants did not rescue hyperglycemia caused by dPdxk 1 mutation nor the accumulation of AGEs that in dPdxk 1 flies is largely responsible for CABs. The impaired rescue of hyperglycemia displayed by the A243G variant is particularly interesting because it could be considered as a preliminary indication of the association of PDXK gene with diabetes that will be investigated in future studies. Though the role of PDXK mutations in diabetes mellitus is purely speculative at this stage of our research, nevertheless, based on the effects of PDXK variants on AGEs, we favor the idea that pathophysiology of hyperglycemia might be linked to a combination of impaired insulin action on target tissues (i.e. insulin resistance) and reduced beta cell function, as it has been hypothesized for mutations of APPL1 gene associated to monogenic diabetes 29 .
The kinetic characterization of the variant enzymes showed that all mutations affected the catalytic activity of PDXK, although with different modalities (Table 1). In general, the effect of D87H and V128I mutations are more drastic than those of H246Q and A243G, according to the location of the former couple of residues in a more critical region of the enzyme. Also, it is worth noting that D87H and V128I behave similarly, and also variants H246Q and A243G display similar biochemical defects, in agreement with the relative proximity of these residues. Interestingly, PLP precursors (PL, PM and PN) have different effects on CAB frequency observed upon expression of different variants. The results obtained from the in vitro characterization of the enzymes parallel this observation, showing that the kinetic parameters of PLP precursors are differently affected by the mutations. In particular, the higher K M for PL (about 10-fold than wild type) displayed by D87H explains why this mutant protein did not respond to PL. Differently, K M values for PM and PN about 5 times higher than wild type explain the rescue observed only at 1 mM concentration.
The very high K M for both ATP and PL found in the V128 mutant protein explains why PL failed to rescue CABs, whereas PN and PM reduced CAB frequency but only at the higher concentration. The H246Q variant, whose mutant enzyme form has normal kinetic parameters with PN, responds to both concentrations of this vitamer. The same mutant, displaying slightly altered kinetic parameters for ATP (when PL is used as substrate) and PM, responded positively to these precursors but only at the higher concentration. The A243G mutant enzyme displayed kinetic parameters very similar to those showed by H246Q. Similarly to H246Q, this variant responded to PL and PM; however, 0.5 mM PN was unable to reduce CAB frequency. Such different effects of PDXK mutations on the catalytic properties of PDXK are very interesting, since they are related to structure-function relationships of the enzyme. However, their full understanding is not possible on the basis of the available data and is postponed to future investigations. PLP treatment of brains from larvae expressing each of the three variants drastically reduced CABs suggesting that this molecule can enter brain cells. A similar behavior had been previously observed in brains from either dPdxk 1 14 and sgll RNAi larvae (sgll is the Drosophila ortholog of PNPO) 25 . In contrast, literature reports that in humans whereas dephosphorylated B 6 vitamers can enter cells and pass the blood-brain barrier, PLP needs to be dephosphorylated to PL before entering cells 3,30 . However, there are no data confirming that such a mechanism also occurs in flies. Thus, we can assume that at least under our experimental conditions (e.g. an excess of PLP that may force the system) PLP can enter cells without being dephosphorylated to PL.
In Drosophila dPdxk 1 is a recessive mutation and heterozygotes do not show CABs. We could expect that the same is also true in humans; thus, heterozygous carriers of the examined variants should not exhibit CAB phenotype. However, there are certain contexts in which PLP levels are low, and reduced functionality of PDXK could be particularly critical. For example, it is known that during pregnancy an increased PLP demand to support fetal development causes a drastic reduction of PLP levels that in some cases can contribute to gestational diabetes onset. Thus, we can envisage that in pregnant women PDXK variants also carried in heterozygous condition could be dangerous. Analogously PDXK variants in heterozygous condition could also impact on genome integrity in either patients treated with drugs that reduce PLP levels or also in patients affected by pathologies such as celiac disease and diabetes which per se decrease PLP levels [31][32][33] . Thus, being able to detect PDXK mutations in such contexts could preserve genome integrity and, in the future, may pave the way for personalized cares based on B 6 administration. www.nature.com/scientificreports www.nature.com/scientificreports/ Methods Materials. All reagents and buffers used for protein purification and enzyme assays were from Sigma Aldrich.
The enzyme PNPOx from E. coli was purified as previously described in Di Salvo et al. 34 .
Drosophila stocks and crosses. dPdxk 1 mutation was previously described in Marzio et al. 14  The progeny of this cross, PDXK VAR /CyGFP; dPdxk 1 /TM6B, was crossed inter se to obtain a stable stock. From this stock larvae PDXK VAR ; dPdxk 1 / dPdxk 1 selected for their non-Tubby phenotype have been analyzed. To test PDXK VAR on a dPdxk 1 /+ background we analyzed larvae PDXK VAR /Cy; dPdxk 1 /TM6B from the same stock.
The Oregon R strain was used as wild-type control. All stocks were maintained and crosses were made at 25 °C on standard Drosophila medium (prepared from cornmeal, sucrose, brewer's yeast, agar, water and treated with propionic acid) or on a supplemented medium (see below). The balancers and the genetic markers used in these crosses are described in detail in FlyBase (http://flybase.bio.indiana.edu/). PDXK genes carrying the mutations were then cloned into a pCaSpeR-tubulin vector 35 . The correct generation of the variants was verified by Sanger sequencing and recombinant plasmids were introduced in flies by germline transformation (Best Gene Inc. Service, USA). chromosome cytology. Colchicine-treated Drosophila metaphase chromosome preparations for CAB scoring were obtained as previously described in Gatti and Goldberg and in Merigliano et al. 24,36,37 . Anti-AGEs immunostaining of brain preparations from third instar larvae was carried out according to Bonaccorsi et al. 38 . Preparations were rinsed in PBS 0.1% Triton (PBST), incubated overnight at 4 °C with rabbit anti-human AGE antibody (1:200 in PBST; ab23722, Abcam,UK), rinsed in PBST, and then incubated for 1 hr at room temperature with the secondary antibody (Alexa-Fluor-555-conjugated anti-rabbit antibody 1:300 in PBST; Molecular Probes, Eugene, OR). All fixed preparations were mounted in Vectashield H-1200 with 4,6 diamidino-2-phenylindole (DAPI) (Vector Laboratories, Burlingame, CA) to stain the DNA. To quantify cells positive to AGE immunostaining at least 1000 cells were analyzed for each genotype. All cytological preparations were examined with a Carl Zeiss (Thornwood, NY) Axioplan fluorescence microscope, equipped with an HBO100W mercury lamp and a cooled charged-coupled device (CCD camera; Photometrics CoolSnap HQ).

treatments of larvae and isolated brains.
To test the effects of PLP and PLP precursors on CABs, brains were dissected from third instar larvae and incubated for 4 hours in 2 ml of saline supplemented with 10% fetal bovine serum (FBS, Corning) with addition of 0.5 or 1 mM PLP, PM, PL or PN. 1 h before fixation brains were treated with colchicine 10 −2 M (final concentration). To test the effects of glucose on AGE frequency brains were dissected from third instar larvae and incubated in 2 ml of saline supplemented with 10% fetal bovine serum (FBS, Corning) for 4 hours with addition of 1% glucose. Then, brains were treated according to the above described procedure for immunostaining. To test the effect of PLP on glucose content we reared flies in standard medium supplemented with PLP 1 mM. About 5 days later larvae were dissected to extract hemolymph and to measure glucose content (see paragraph below).
Glucose measurement. Glucose concentration in hemolymph from third instar larvae was measured using the Infinity Glucose Hexokinase reagent (Thermo scientific). Hemolymph collection and glucose measurement were done as described in Marzio et al. 14 . Western blotting. Extracts for Western blotting were prepared by lysing samples of 20 brains in 150 mM NaCl, 50 mM Tris-HCl, pH 7.5, 30 mM NaF, 25 mM b-glycerophosphate, 0.2 mM Na3VO4, Triton X-100 1%, and complete Protease Inhibitor Cocktail (Roche). Extracts were immunoblotted according to Somma et al. 39 ; blotted proteins were detected using an antibody against HA tag (Anti-HA-Peroxidase 12013819001 Roche). Anti-alpha tubulin (SIGMA) was used as loading control. Primary antibodies were detected using HRP conjugated anti-mouse and anti-rabbit IgGs and the ECL detection kit (all from GE Healthcare). Chemiluminescent blots were imaged with the ChemiDoc MP imager (Bio-Rad). Band intensities were quantified by densitometric analysis with Image Lab software (Bio-Rad).