ARTICLE https://doi.org/10.1038/s42003-022-03311-1 OPEN Structural and biochemical evidence that ATP inhibits the cancer biomarker human aldehyde dehydrogenase 1A3 Albert Castellví1,2,3, Raquel Pequerul1,3, Vito Barracco1, Judith Juanhuix 2, Xavier Parés1 & Jaume Farrés 1✉ Human aldehyde dehydrogenase (ALDH) participates in the oxidative stress response and retinoid metabolism, being involved in several diseases, including cancer, diabetes and obesity. The ALDH1A3 isoform has recently elicited wide interest because of its potential use 1234567890():,; as a cancer stem cell biomarker and drug target. We report high-resolution three-dimensional ALDH1A3 structures for the apo-enzyme, the NAD+ complex and a binary complex with ATP. Each subunit of the ALDH1A3-ATP complex contains one ATP molecule bound to the adenosine-binding pocket of the cofactor-binding site. The ATP complex also shows a molecule, putatively identified as a polyethylene glycol aldehyde, covalently bound to the active-site cysteine. This mimics the thioacyl-enzyme catalytic intermediate, which is trapped in a dead enzyme lacking an active cofactor. At physiological concentrations, ATP inhibits the dehydrogenase activity of ALDH1A3 and other isoforms, with a Ki value of 0.48 mM for ALDH1A3, showing a mixed inhibition type against NAD+. ATP also inhibits esterase activity in a concentration-dependent manner. The current ALDH1A3 structures at higher resolution will facilitate the rational design of potent and selective inhibitors. ATP binding to ALDH1A3 enables activity modulation by the energy status of the cell and metabolic reprogramming, which may be relevant in several disease conditions. 1 Department of Biochemistry and Molecular Biology, Faculty of Biosciences, Universitat Autònoma de Barcelona, 08193 Bellaterra, Cerdanyola del Vallès, Barcelona, Spain. 2 Alba Synchrotron, carrer de la Llum 2-26, 08290 Cerdanyola del Vallès, Barcelona, Spain. 3These authors contributed equally: Albert Castellví, Raquel Pequerul. ✉email: jaume.farres@uab.cat COMMUNICATIONS BIOLOGY | (2022)5:354 | https://doi.org/10.1038/s42003-022-03311-1 | www.nature.com/commsbio 1 ARTICLE COMMUNICATIONS BIOLOGY | https://doi.org/10.1038/s42003-022-03311-1 T he human aldehyde dehydrogenase (ALDH) superfamily accumulation of toxic aldehydes with a concomitant increase in comprises 19 enzyme isoforms catalyzing the NAD(P)- cell death and inhibition of tumor growth2,3. dependent irreversible oxidation of aldehydes to their More than 300 ALDH crystal structures are deposited in the corresponding carboxylic acids. ALDHs act on a wide variety of Protein Data Bank, many of them corresponding to human aldehyde substrates, some of them derived from lipids, such as ALDH1A1 and ALDH2. Most initial structure-function studies retinoids, steroids and lipid peroxidation products. They play were performed with ALDH2 (in fact an ALDH1 family isoform, important roles in multiple biological processes, including oxi- based on sequence identity) and residue numbering is usually dative stress response, lipid hormone metabolism, and cell dif- referred to this isoform. ALDH1A1, ALDH1A2, ALDH1A3 and ferentiation. They also participate in oxidative phase I drug ALDH2 isoforms are tetrameric enzymes with 55-kDa subunits metabolism, contributing to chemotherapeutic drug resistance1. sharing 70% amino acid sequence identity. So far, only six Recently, ALDHs have been involved in diabetes, obesity and ALDH1A3 crystal structures at a low resolution of 2.80–3.25 Å metabolic reprogramming of tumor cells. Cytosolic ALDH1A have been deposited, some in complex with inhibitors (Supple- enzymes, and specifically the ALDH1A3 isoform, are good mar- mentary Table 1). These structures consistently show the ALDH kers of cancer stem cells and have been implicated in carcino- monomers formed by three domains: NAD+-binding, catalytic genesis, chemoresistance and relapse. The role of the increased and oligomerization domains. At the intersection of the substrate- ALDH activity in cancer stem cells is not well understood. and cofactor-binding pockets sits a strictly conserved cysteine Nevertheless, the stemness of this subpopulation of cells has been residue (Cys302, ALDH2 numbering), which is essential for related to the participation of ALDHs in retinoic acid-mediated catalysis, acting as the active-site nucleophile. It is well accepted signaling pathways, and to the high antioxidant activity of these that the reaction starts with the nucleophilic attack of the alde- enzymes, specifically their ability to counteract the effect of hyde carbonyl group by the Cys thiol, and proceeds through the reactive oxygen species. ALDH1A inhibition results in formation of a thiohemiacetal intermediate, hydride transfer to NAD+, deacylation and NADH release. The deacylation step Table 1 Data collection and model refinement statistics. requires the conserved Glu268, which acts as a general base to activate a water molecule that can make a nucleophilic attack on the thioacyl intermediate. This catalytic mechanism also explains PDB code ALDH1A3 apo ALDH1A3- ALDH1A3-ATP the esterase activity of ALDH using the same active-site residues. NAD+ complex complex The cofactor-binding domain in ALDH deviates from the canonical Rossmann fold4–7, which consists of two βαβαβ motifs 7QK7 7QK8 7QK9 for dinucleotide binding, found in other NAD(P)+-dependent Data collection statistics: enzymes. In ALDH, the topological order of β strands and α Unit cell 82.96, 81.58, 82.11, helices is the same but the Rossmann fold contains five instead of (a, b, c) (Å) 90.13, 159.19 89.52, 158.64 90.01, 158.71 the usual six β-strands and the mode of interaction of the cofactor Space group P 21 2 21 P 21 2 21 P 21 2 21 is changed. In ALDH, the GXGXXG motif, usually found asso- Resolution 48.44–2.29 48.00–1.89 48.19–1.78 range (Å) (2.35–2.29) (1.93–1.89) (1.81–1.78) ciated with the β1-αA loop in the canonical Rossmann fold, is Unique reflections 53,899 (4253) 92,123 (4461) 111,992 (4978) now found to be in the β4-αD loop (as GXTXXG) and involved in Multiplicity 6.6 (6.1) 4.4 (4.5) 6.6 (6.1) interactions with the nicotinamide ring. In contrast to most other Completeness (%) 98.9 (95.7) 99.1 (98.6) 99.4 (90.9) dehydrogenases, there are no positively charged residues sur- < I/σI> 11.0 (1.7) 15.5 (1.9) 15.7 (0.9) rounding the pyrophosphate moiety of NAD+, only the helical Wilson B-factor 40.77 33.06 31.88 dipole provided by an α-helix. The adenosine ring rests in a cleft (Å2) between two α-helices, while the phosphodiester bridge is solvent R-merge 0.124 (1.391) 0.051 (0.803) 0.062 (2.369) exposed making possible its rotational motion. This particular Rmeas 0.134 (1.524) 0.057 (0.908) 0.067 (2.592) binding of the cofactor has functional consequences, as it allows Rpim 0.052 (0.613) 0.026 (0.417) 0.026 (1.034) for the nicotinamide ring to sample between different con- CC1/2 0.99 (0.73) 0.99 (0.74) 0.99 (0.57) formers, whereas the adenine portion remains relatively immo- Model refinement statistics: bile. Two conformations, one extended (for hydride transfer) and R-work / R-free 0.1903 0.1799 0.1926 / 0.2270 / 0.2039 / 0.2150 one contracted (for hydrolysis) have been identified and are Number of non- compatible with the two-step catalytic mechanism. The first H atoms: conformation is associated with NAD+, as it brings the C4 atom macromolecules 7399 7400 7457 of the nicotinamide at the right distance from the Cys302-Sγ, ligands 54 193 139 while the second conformation is suitable for NADH and permits solvent 358 600 617 Glu268 to act as a general base in the activation and orientation of RMSD (bonds) (Å) 0.012 0.012 0.012 the attacking water molecule. The conformational flexibility of RMSD (angles) (°) 1.56 1.55 1.60 the nicotinamide moiety of the cofactor is typically associated in Ramachandran 95.28 96.41 96.62 crystal structures with increased thermal B-factor values. Thus, favored (%) cofactor isomerization is needed for catalysis, differently from Ramachandran 4.41 3.08 3.28 what is observed in other NAD(P)-dependent dehydrogenases, allowed (%) where changes in protein conformation and domain closure Clashscore, 2.43 3.32 2.72 all atoms occur upon coenzyme binding and release6. MolProbity score 1.57 1.51 1.50 Mg2+ ions have been shown to stimulate the activity of Average B-factor 56.55 40.91 45.39 ALDH1A3 and ALDH2 but impair the activity of ALDH1A1 and (Å2): ALDH1A28. This effect has been related to the different rate- macromolecules 56.27 39.94 44.48 limiting step for these isoforms: deacylation and cofactor dis- cofactor − 41.41 47.43 sociation, respectively. Mg2+ ions would induce a change in the solvent 59.88 48.91 53.58 relative population of the local nicotinamide conformation of the cofactor, by stabilizing one of the conformations of the Values in parenthesis refer to the highest resolution shell. 2 COMMUNICATIONS BIOLOGY | (2022)5:354 | https://doi.org/10.1038/s42003-022-03311-1 | www.nature.com/commsbio COMMUNICATIONS BIOLOGY | https://doi.org/10.1038/s42003-022-03311-1 ARTICLE Fig. 1 Structural description of ALDH1A3. a Half ALDH1A3 homotetramer contained in the asymmetric unit (left). The homotetrameric biological form appears in the crystallographic unit cell (right subfigure) after applying the symmetries of the space group. b Binding pockets in the cofactor (magenta) and catalytic (orange) domains. c Ribbon representation of the three described ALDH1A3 structures (PDB codes 7QK7, 7QK8 and 7QK9, respectively). The ribbon radius is scaled and normalized to the maximum B-factor reached in the ALDH1A3 apo structure. Residues in the main chain marked in red show the highest B-factor values, while those marked in blue show the lowest values. phosphodiester region and thus the orientation of the nicotina- (Fig. 1b). The volumes of the cofactor- and the substrate- mide ring relative to its ribose6,7. binding pockets are very similar in all the structures, as calculated Recently, we carried out the enzymatic characterization of the using the POVME algorithm. The volume of the pocket in the three human ALDH1A enzymes with non-retinoid substrates as cofactor-binding domain is 444, 439 and 411 Å3 for the apo form, well as retinaldehyde isomers8, and reported the inhibition the ALDH1A3-NAD+ and the ALDH1A3-ATP complexes, kinetics with some novel compounds9. In view of the increasing whereas the volumes of the substrate-binding pockets are 231, importance of ALDH1A3 in various disease states and the need of 220 and 208 Å3 for the same forms, respectively (Fig. 1b). The structures at higher resolution for structure-based drug design, we cofactor- and catalytic-binding domains are defined by two sets of focused here on contributing higher-resolution structures for parallel beta sheets and alpha helices each, whereas the ALDH1A3. In the current study, three previously unreported oligomerization domain is a set of three parallel beta sheets ALDH1A3 structures are presented: the apo-enzyme, the complex (Fig. 1b). The cofactor-binding domain contains a non-canonical with NAD+, and with ATP bound to the adenosine-binding Rossmann-fold motif, slightly different from that found in other pocket of the cofactor-binding site. ATP inhibited ALDH activity NAD(P)+-dependent enzymes6. The three crystal structures at physiological concentrations, opening the possibility of ALDH described in this work do not present remarkable conformational modulation by the energy status of the cell. changes among them and share the same disordered areas (Fig. 1c). The ALDH1A3 apo structure at 2.29-Å resolution shows a few Results electron density blobs inside the substrate-binding pocket near High-resolution structures of apo ALDH1A3 and of binary the catalytic Cys314 in both monomers (Fig. 2). These electron complexes with NAD+ and ATP. The three studied crystal forms densities could not be modeled with any of the molecules present of ALDH1A3, the apo form and the two complexes of the protein in the crystallization cocktail. A PEG molecule can also be seen with NAD+ and ATP, were obtained under the same crystal- within the substrate-binding pocket in both subunits of the lization conditions. Only the additives in the crystallization crystal ALDH1A3 apo structure (Fig. 2). This PEG molecule solution were different in each case as described under Materials remains conserved in the other structures obtained in this work. and methods. Data collection and model refinement statistics are Each monomer of the ALDH1A3-NAD+ complex at 1.89-Å shown in Table 1. resolution contains one NAD+ molecule positioned in each of the The current structures show the highest resolution cofactor-binding pockets (Fig. 3a). As expected, the NAD+ (1.78–2.29 Å) among all ALDH1A3 structures deposited in the molecule in both subunits is accommodated showing its extended PDB (Supplementary Table 1). All crystal structures contain a conformation (suitable for hydride transfer) and interacting homodimer in the asymmetric unit (Fig. 1a, left). The relevantly directly with residues Thr178, Trp180, Lys204, Glu207, Ser258, biological assembly of the enzyme, the homotetrameric form, was and Glu411 through hydrogen bonds, and to other 20 residues via confirmed by size-exclusion chromatography, and is obtained in hydrophobic interactions (Fig. 3c, left). The adenosine ribose is the protein crystal when the space group symmetry is applied held in position by hydrogen bonds between the 2′ and 3′ (Fig. 1a, right subfigure). As previously observed in other hydroxyl oxygen atoms and Lys204. In addition, the 2′ hydroxyl members of the enzyme superfamily, each ALDH1A3 subunit is oxygen forms a hydrogen bond with the sidechain of Glu207, composed of three structural domains: a catalytic domain, a which is part of the same loop. Thr178, forming a hydrogen bond cofactor-binding domain and an oligomerization domain with the 3’ hydroxyl oxygen through its backbone oxygen, is COMMUNICATIONS BIOLOGY | (2022)5:354 | https://doi.org/10.1038/s42003-022-03311-1 | www.nature.com/commsbio 3 ARTICLE COMMUNICATIONS BIOLOGY | https://doi.org/10.1038/s42003-022-03311-1 Fig. 2 The apo ALDH1A3 crystal structure (PDB code 7QK7). 2Fo–Fc electron density map at σ = 1 and fitted crystal structure of the apo form of monomer A (a) and B (b) of ALDH1A3 close to the NAD+-binding and substrate-binding pockets. unique to ALDH1A3 in comparison to other ALDH1A and site of ALDH1A3. As anticipated, the average B-factor increased ALDH2 enzymes, which have an Ile residue at this position for from α- to γ-phosphate groups of ATP (66–72 Å2), being highest this purpose. Only Glu411 makes two hydrogen bonds to the 2′ for the latter, indicating more flexibility at this end of the ATP and 3′ hydroxyl oxygen atoms of the nicotinamide ribose. The molecule (Supplementary Fig. 1). position of the nicotinamide ribose is further stabilized by van der The ALDH1A3 structure complexed with NAD+ presented Waals contacts with the sidechain of Phe413. The extended here has been compared to two previously published ALDH1A3 conformation of NAD+ allows hydride transfer between the structures (Supplementary Table 1, PDB codes 5FHZ13 and substrate and the C4 atom of the nicotinamide ring of the 6S6W14). The low resolution (2.90 and 3.25 Å, respectively) of cofactor, a necessary step in the catalytic mechanism which would these two previously reported structures, which include a not be possible with the NAD+ molecule in contracted form. The homotetramer in the asymmetric unit, did not show the canonical catalytic Cys is modeled on both subunits as a single conforma- extended or contracted conformation for NAD+ (Fig. 4a). In fact, tion pointing toward the substrate-binding pocket region (Fig. 3a). the nicotinamide ring of the modeled cofactors in monomers C However, as suggested by the continuous electron density and D of the homotetramer is pointing away from active-site between Cys314 and the NAD+ molecule in monomer A, it is Cys314. In contrast, the NAD+ molecules in the complex possible that the Cys residue was in a double conformation presented here show an extended conformation (suitable for (Fig. 3a), as noticed in other ALDH structures, although its hydride transfer), well accommodated in the cofactor-binding functional significance is uncertain10–12. This double conforma- pocket, as seen in the highest resolution electron density maps tion would give rise to some steric hindrances between the Cys (polder maps, Fig. 3a). Likewise, our ALDH1A3-NAD+ structure and the cofactor and would explain the fractional occupancies of is consistent with all the ALDH-NAD+ structures having the the latter molecule, having values equal to 1 in the adenosine cofactor in the extended form, as it is the case of the ALDH1A2 moiety region but 0.74 and 0.64 in the nicotinamide part in (PDB code 6B5G15, Fig. 4b). The superimposition between our monomers A and B, respectively. ALDH1A3-NAD+ complex and the ALDH1A1-NADH complex The ALDH1A3-ATP structure at 1.78-Å resolution contains (PDB code 4WB916) evidences the movement of the cofactor to a one ATP molecule in each subunit (Fig. 3b). The ATP molecule contracted form (suitable for the hydrolysis step) and the catalytic in monomers A and B is located in the adenosine-binding moiety Cys303/Cys314 and Glu280/Glu269 residues in ALDH1A1 of the cofactor-binding pocket and is accommodated via direct (Fig. 4c). hydrogen bonding with Thr178, Trp180, Lys204, Glu207, and The ALDH1A3-NAD+ complex contains a PEG molecule Ser258, and via hydrophobic interactions with ten additional (PEG1) close to the Cys314 residue in the substrate-binding residues (Fig. 3c, right). Interestingly, all the close contacts are pocket, with 6 and 4 oxygen atoms in monomers A and B, exactly the same for the common part between NAD+ and ATP, respectively (Fig. 5a). Also, in the substrate-binding pocket there including the hydrogen bonds, through water molecules, with is a small PEG molecule (PEG2) in the same place in all subunits Asn181, Gln208 and Ser245. Trp180 is hydrogen bound both to (Fig. 5a). None of the two PEG1 molecules were covalently bound NAD+ and ATP pyrophosphate groups. The α- and β-phosphate to the Cys314 residues, as it can be deduced from the distances groups of ATP have identical interactions to those of the NAD+ between the Cys314-Sγ and the PEG1 atoms and also from the pyrophosphate, including four water molecules, while the γ- electron density shown in Fig. 5b. Next to Cys314 there is a water phosphate group is not showing any specific interaction. Eight molecule that, as it will be shown later, will play an important role hydrophobic interactions are also conserved. The highest B-factor in the position of the PEG1 molecule in the ALDH1A3-ATP and lower occupancy values in the NAD+ molecule are detected complex (Fig. 5b). Finally, Cys313 appears in a double in the nicotinamide moiety in both subunits A and B, whereas the conformation in both monomers without forming a disulphide adenosine moiety displays the best ordered part of the whole bond with Cys314 in any structure (Fig. 5a). molecule. The average B-factor and the occupancy values of the The ALDH1A3-ATP structure also contains two PEG adenosine moiety in the NAD+ complex are very similar to those molecules (PEG1 and PEG2) inside the substrate-binding pocket of the whole ATP molecule in the ALDH1A3-ATP complex. This in monomers A and B (Fig. 6a). However, in this case, the PEG1 illustrates the good anchoring of ATP in the adenosine-binding molecule is covalently bound to Cy314 in both subunits with a 4 COMMUNICATIONS BIOLOGY | (2022)5:354 | https://doi.org/10.1038/s42003-022-03311-1 | www.nature.com/commsbio COMMUNICATIONS BIOLOGY | https://doi.org/10.1038/s42003-022-03311-1 ARTICLE Fig. 3 NAD+ and ATP interactions with ALDH1A3 (PDB codes 7QK8 and 7QK9, respectively). a Polder OMIT maps contoured at σ = 4 with fitted structures over the NAD+ molecules for each monomer of the ALDH1A3-NAD+ complex. b Polder OMIT maps contoured at σ = 6 around the ATP position in each monomer in the ALDH1A3-ATP complex, with the fitted ATP molecules. c LIGPLOT describing interactions between NAD+ and ALDH1A3 (left), and ATP and ALDH1A3 (right). Red circles show the structural elements placed in equivalent positions in ALDH1A3-NAD+ and ALDH1A3-ATP complexes. distance between Sγ-Cys314 and C1-PEG1 atoms of 1.82 and has been described in solutions of PEG 40017 and its presence was 1.83 Å for monomers A and B, respectively (Fig. 6b), which is a previously reported in a human ALDH2 structure, where the typical distance for a covalent C − S bond. In addition, the authors confirmed the presence of aldehydes derived from PEG6,18, covalent union between Cys314 and the PEG1 molecule is as well as in the tomato ALDH10 crystal structure (PDB code observed in the continuous electron density shown in the polder 4I9B19). As in the ALDH1A3-ATP complex described here, the C1- maps. The O1-PEG1 atom is hydrogen bonded to the N-Cys314 PEG-aldehyde atom in the ALDH10 structure is covalently bound and Nδ2-Asn181 atoms in both monomers and occupies the to the Sγ-Cys314 atom at a 1.86-Å distance (Supplementary Fig. 2). position of the water molecule near Cys314 in the ALDH1A3- Therefore, we cannot rule out the possibility that PEG1 molecules NAD+ structure (Fig. 6b). Additional water molecules are fitted covalently bound to Cys314 residues in the ALDH1A3-ATP in both monomers at the position of the nicotinamide ring of complex could be PEG-aldehydes. The genuine interest of the NAD+ in the ALDH1A3-NAD+ structure (Fig. 6a). ALDH1A3-ATP structure is that it can be interpreted as a true PEG molecules are frequently found as ligands in crystal thioacyl-enzyme intermediate in the ALDH catalytic mechanism, structures and, particularly, in ALDH. PEG aldehyde formation which is trapped in a dead enzyme lacking an active cofactor. COMMUNICATIONS BIOLOGY | (2022)5:354 | https://doi.org/10.1038/s42003-022-03311-1 | www.nature.com/commsbio 5 ARTICLE COMMUNICATIONS BIOLOGY | https://doi.org/10.1038/s42003-022-03311-1 Fig. 4 Structural alignment of present ALDH1A3-NAD+ structure with those of ALDH1A3, ALDH1A2, and ALDH1A1. a Alignment of current ALDH1A3- NAD+ (in cyan) with monomers C (in pink) and D (in purple) of the PDB 5FHZ structure (ALDH1A3 complexed with NAD+ and retinoic acid). b Alignment between ALDH1A3-NAD+ (in cyan) and PDB 6B5G structure (ALDH1A2 with NAD+ and 3-ethoxythiophen-2-yl){4-[4-nitro-3-(pyrrolidin-1-yl)phenyl] piperazin-1-yl}methanone, in orange). The conformation of all conserved residues remains unchanged except for the catalytic Cys320/Cys314 and the non-catalytic Met192/Met186 residues. c Alignment between the ALDH1A3-NAD+ (in cyan) complex and PDB 4WB9 structure (ALDH1A1 complexed with NADH, in green). Cofactor molecules and catalytic Cys and Glu residues are depicted with sticks while other residues interacting with the cofactor are shown with thin lines. Table 2 Inhibition of ALDH activity in the presence of suggested by our results, is consistent with similar bonds observed 2 mM ATP. in other ALDH ternary complexes19–21. Remaining activity (%) Inhibition of ALDH1A3 activity by ATP. The binding of ATP ALDH1A1 ALDH1A2 ALDH1A3 ALDH2 ALDH3A1 to a member of the ALDH superfamily is unprecedented in the 43 ± 1 68 ± 2 22 ± 1 89 ± 1 25 ± 1 literature. Since the ATP molecule is able to bind ALDH1A3 in The inhibitory effect is expressed as the percentage of remaining activity (%) ± standard the cofactor-binding pocket, the next step was to check if the ATP deviation, after 5-min pre-incubation in the absence of cofactor. n = 2, technical replicates for molecule could act as an enzyme inhibitor. A concentration of each isoform. Standard activity was measured at saturating concentrations of hexanal in the corresponding reaction buffer, containing 2 mM ATP and concentrations of the corresponding 2 mM ATP was chosen since it was that of the crystallization cofactor near Km values. solution and is near its physiological concentration22. The inhi- bitory effect of ATP against several ALDH isoforms is shown in Table 2. ATP, at physiological concentrations, is inhibiting the Although thioacyl-enzyme intermediates have been observed in dehydrogenase activity of ALDH, likely by interfering with the other ternary complexes of the ALDH superfamily19–21, this is, binding of cofactor. All ALDH isoforms tested were inhibited to along with human ALDH7A1, the only reported crystal structure of various degrees. ALDH1A3 and ALDH3A1 were the most sen- a human tetrameric ALDH showing the thioacyl-enzyme inter- sitive isoforms, while ALDH2 was affected the least. These results mediate. Our results are in good agreement with the thiohemiacetal are consistent with the observed binding of ATP to the intermediate observed in tomato ALDH10 (PDB code 4I9B19, adenosine-binding site of ALDH1A3. The effect of ATP on Supplementary Fig. 2) and in the non-phosphorylating glyceralde- ALDH enzymes is not dependent on the cofactor presence in the hyde 3-phosphate dehydrogenase from Streptococcus mutants20. pre-incubation step with the reaction mixture. Thus, we observed The role of Asn181 in stabilizing the reaction intermediate through the same effect by ATP when the reaction was started with NAD+ hydrogen bonding with the O atom of the carbonyl substrate, as or hexanal as a substrate. 6 COMMUNICATIONS BIOLOGY | (2022)5:354 | https://doi.org/10.1038/s42003-022-03311-1 | www.nature.com/commsbio COMMUNICATIONS BIOLOGY | https://doi.org/10.1038/s42003-022-03311-1 ARTICLE Fig. 5 Cofactor and substrate structures in the ALDH1A3-NAD+ complex (PDB code 7QK8). a 2Fo–Fc electron density maps contoured at σ = 1 over Cys313, Cys314, Glu280, NAD+, PEG1, PEG2 and water molecules inside the cofactor and substrate-binding pockets. b For monomers A and B, hydrogen bonds of Cys314, Asn469 and NAD+ with PEG1, and polder OMIT maps at σ = 3.5 over Cys314 and PEG1. A highly conserved water molecule near Cys314 is depicted in red. The parts of the structure depicted with thin lines in monomer B represent those no longer interacting with the PEG1 molecule, with respect to monomer A. It is well known that the ATP-Mg complex, in the range of the dissociation constant (Kd) of ATP and ATP-Mg2+ could millimolar concentration, is the largest pool of Mg2+ ions within have been determined by a non-enzymatic method. None- the cell. The complex follows a 1:1 stoichiometry, with a Kd value theless, the fact that, in the absence of Mg2+, the Km value for of 0.15 mM23. Since ALDH1A3 activity is activated by free Mg2+, NAD+ decreases (from 130 to 24 µM) is consistent with the it is conceivable that it also is finely modulated by the interplay idea that Mg2+ is weakening NAD+ binding affinity. In that between ATP and Mg2+. In order to check the effect of Mg2+ event, we could surmise that Mg2+ is also weakening ATP ions on the NAD+/ATP binding to ALDH1A3, the IC50 value for binding. ATP was determined in the presence or absence of MgCl2 Then we determined the inhibition constant (Ki) value and the (Fig. 7a). The IC50 value of ALDH1A3 for ATP was type of inhibition for ATP against ALDH1A3, using various 2.05 ± 0.23 mM, in the presence of 30 mM MgCl2 (under concentrations of NAD+, without adding Mg2+ ions. Data were standard assay conditions), and 0.43 ± 0.06 mM, in absence of plotted (Fig. 7b) and could be best fitted to the equation for mixed Mg2+ ions. In the latter case, a lower concentration of NAD+, inhibition (Eq. 1), where Ki and K’i are the constants for the near the Km value (24 µM) was used because the lack of Mg2+ competitive and uncompetitive parts of the inhibition: also affects NAD+ kinetics, lowering the Km value for NAD+. For comparison, under standard assay conditions (130 µM NAD+), V max  ½S 1 þ K½I0 without adding Mg2+ ions, the IC50 value was 0.75 ± 0.05 mM.  i  v¼ Overall, this supports the notion that Mg2+ ions activate 1 þ K½I  ð1Þ ALDH1A3 also in the presence of ATP. Km    þ ½S i Thus, the effect of Mg2+ ions is to weaken ATP binding (IC50 1 þ K½I0 value of 2.05 versus 0.43 mM). Since both ATP and NAD+ i interact with Mg2+, the former effect could be interpreted as a The Ki value is 0.48 ± 0.08 mM, which is similar to the IC50 result of either the destabilization of ATP binding or the value (0.43 ± 0.06 mM). The mixed inhibition type is not strengthening of NAD+ binding. In order to unravel the incompatible with ATP competing for binding to the same site confounding effect of Mg2+ on ATP in the presence of NAD+, as the adenosine moiety of the cofactor, as seen from the crystal COMMUNICATIONS BIOLOGY | (2022)5:354 | https://doi.org/10.1038/s42003-022-03311-1 | www.nature.com/commsbio 7 ARTICLE COMMUNICATIONS BIOLOGY | https://doi.org/10.1038/s42003-022-03311-1 Fig. 6 Cofactor and substrate structures in the ALDH1A3-ATP complex (PDB code 7QK9). a 2Fo−Fc electron density maps contoured at σ = 1 over Cys314, Cys313, Glu280, NAD+, PEG1, PEG2 and water molecules inside the cofactor and substrate-binding pockets. b Polder OMIT maps at σ = 5 over the Cys314 and PEG1, and crystallographic models showing hydrogen bonds between Cys314 and Asn181. The parts of the structure depicted with thin lines in monomer B represent those no longer interacting with the PEG1 molecule, with respect to monomer A. Fig. 7 Inhibition kinetics of ATP against ALDH1A3. a Sigmoidal plots to determine IC50 values. The ALDH1A3 activity was assayed in 50 mM HEPES, 5 mM DTT, pH 8.0, without MgCl2 (blue curve) or including 30 mM MgCl2 (red curve). Hexanal was used as a substrate at saturating concentration (250 µM). NAD+ concentration was 24 µM in the absence of MgCl2, and 130 µM with 30 mM MgCl2. Experimental values with all replicate points are shown. n = 2, technical replicates for each ATP concentration (0.01–100 mM). b Michaelis-Menten plots fitted to the equation for mixed inhibition (Eq. 1). ATP concentrations (mM): 0.0 (empty circles), 2.0 (solid circles), 5.0 (empty squares) and 7.5 (solid squares). Hexanal was used as a substrate at a saturating concentration (250 µM). The values of kinetic parameters from this fit were: Vmax = 0.18 ± 0.01 U/mg; Km = 24 ± 2 µM; Ki = 0.48 ± 0.08 mM; Ki’ = 1.68 ± 0.12 mM. Experimental values with all replicate points are shown. n = 3, technical replicates for each NAD+ concentration (5–1000 µM). Parameter values are expressed as the calculated value ± standard error. structure. Various mechanisms could account for non- following a compulsory ordered substrate addition or product competitive inhibition patterns of inhibitors binding in the active release, as it is the case for ALDH. The rationale behind is that, site24. For instance, this behavior is feasible in enzymes that while ATP and NAD+ bind to the same site, they may not always undergo multiple isomerization steps or bisubstrate kinetics bind to the same exact enzyme species or conformational state. 8 COMMUNICATIONS BIOLOGY | (2022)5:354 | https://doi.org/10.1038/s42003-022-03311-1 | www.nature.com/commsbio COMMUNICATIONS BIOLOGY | https://doi.org/10.1038/s42003-022-03311-1 ARTICLE Table 3 Inhibitory effect of ATP, ADP or AMP on the implications, since ALDH supplies the mitochondrial respiratory ALDH1A3 activity measured with hexanal. chain with NADH, which is the source of ATP. This could provide a regulatory mechanism for metabolic shifting to readjust the flux of ATP production. [Compound] (mM) ATP (%) ADP (%) AMP (%) The homeostatic level of intracellular ATP ranges from 0.5 to 0.5 47 ± 5 56 ± 4 94 ± 2 5 mM22 and may change in tumor cells and upon chemother- 5.0 9.3 ± 3.5 5.1 ± 1.1 79 ± 2 apeutic treatment, meaning that ALDH1A3 activity could be The inhibitory effect is expressed as the percentage of remaining ALDH1A3 activity (%) ± physiologically modulated by changes in the energy status of the standard deviation, using 250 µM hexanal as a substrate after the addition of 0.5 and 5.0 mM cell. It has been reported that intracellular ATP levels in cancer ATP, ADP or AMP. n = 2, technical replicates. The enzyme and ATP/ADP/AMP mixtures were incubated for 5 min at room temperature, in the presence of 24 µM NAD+ cofactor, in 50 mM cells are higher than those in normal cells of the same tissue HEPES, 5 mM DTT, pH 8.0. The reaction was initiated by the addition of hexanal after pre- origin, and that drug-resistant cancer cells have even higher ATP incubation. The effect of ATP and ADP was similar when 0.5 and 5.0 mM of each compound were added to the reaction. A much lower effect of AMP on ALDH1A3 activity was observed levels than drug-sensitive cancer cells from which the resistant when these concentrations were used. cells were derived34,35. ALDH activity, upregulated in some cancer types, may contribute to ATP production in cancer cells We then investigated whether ADP was also an inhibitor of through the supply of NADH36. In this regard, ALDH1A3 ALDH1A3. From inhibition kinetics (using hexanal or all-trans- expression has been linked to glucose consumption37,38. Recently, retinaldehyde as a substrate), we observed that the ADP it has been shown that ALDH1A3 stimulates metastasis by inhibitory effect was similar to that of ATP. This finding is in altering glucose catabolism39 and that silencing of ALDH1A3 line with the small contribution of the γ-phosphate group to ATP leads to significantly decreased glucose consumption, reduced binding. In contrast, AMP provided a much weaker inhibition if lactate and ATP production in cancer cells40. There is some any (Table 3 and Supplementary Table 2). evidence that ALDH1A3 promotes glycolysis through the acti- We also checked the inhibition of ALDH1A3 activity by ATP vation of the PI3K/AKT/mTOR signaling pathway39. On the in the presence of the physiological substrate all-trans-retinalde- other hand, it is conceivable that cellular ATP levels may mod- hyde, using an HPLC-based methodology (Supplementary Fig. 3). ulate ALDH1A3 activity by a negative feedback mechanism, We observed that the ALDH1A3 activity decreased from 63 to similarly to some enzymes participating in the glycolytic pathway, 16% when ATP concentration increased from 2.0 to 7.5 mM, which may be relevant in other disease states such as obesity and confirming the inhibitory effect of ATP. The effect of ADP and diabetes. Overall, ALDH1A3 and possibly other ALDH isoforms AMP also mimicked what was found using hexanal as a substrate are emerging as targets of choice for the control of cell pro- (Table 3 and Supplementary Table 2). liferation and metabolic disorders. It is well established that ALDH isoforms possess esterase In summary, three high-resolution structures have been elu- activity towards p-nitrophenyl esters, being the active-site Cys cidated for ALDH1A3, one of them revealing a previously the essential nucleophile for both esterase and dehydrogenase unreported ATP-binding site in the adenosine-binding moiety of reactions25,26. Thus, the hydrolysis of p-nitrophenyl acetate takes the cofactor-binding site. Enzymatic assays revealed that ATP is a place at the same catalytic center which is responsible for the good inhibitor of both the dehydrogenase and esterase activities dehydrogenase activity27. Although NAD+ is not needed for of ALDH1A3. In fact, ATP at physiological levels effectively esterase activity, it has been reported that it may enhance the modulates ALDH1A3 activity. This allows us to speculate that the acylation rate by promoting the attack of the active-site energy status of the cell is associated with ALDH1A3 and its nucleophile on the carbonyl group of the substrate28. potential role in metabolic reprogramming in disease conditions, Since ALDH1A3 also displays esterase activity in the absence of such as cancer, diabetes and obesity. The availability of high- NAD+, we tested the effect of ATP against ALDH1A3 esterase resolution 3D structures paves the way for structure-based inhi- activity. The p-nitrophenol production was followed in the bitor design and discovery of drugs targeting ALDH1A3. presence or absence of NAD+, and combining with various concentrations of ATP (Supplementary Table 3). NAD+ Methods enhanced esterase activity of ALDH1A3, which is consistent ALDH isoform purification and ALDH1A3 crystallization. Each of ALDH1A1, with deacylation being the rate-liming step8. The activity dropped ALDH1A2, ALDH1A3, ALDH2 and ALDH3A1 cDNA was cloned into the pET-30 Xa/LIC expression vector. The expression of the His6-tagged protein in E. coli down to 40% in the absence of cofactor. ATP could not substitute strain BL21(DE3) pLysS was induced at 20 °C by 1 mM IPTG at OD600 0.8–1 and for NAD+ to enhance esterase activity, meaning that the incubated overnight in 1 L 2xYT culture media. Cells were subsequently pelleted by adenosine moiety is not sufficient to activate the Cys nucleophilic centrifugation at 4 °C and then resuspended in 30 mL of lysis buffer (20 mM Tris- attack. Remarkably, ATP alone inhibited esterase activity in a HCl pH 8.0, 500 mM NaCl, 10 mM DTT and 20 mM imidazole). The resuspended concentration-dependent manner. Likewise, when NAD+ was cells were disrupted by sonication and the pellet was discarded. The supernatant was loaded in 1 mL nickel ion affinity chromatography column previously equili- present, ATP also decreased esterase activity. brated with 20 mM Tris-HCl, pH 8.0, 500 mM NaCl and 20 mM imidazole. Highly pure fractions of each ALDH isoform were eluted with 200 mM imidazole and buffer was changed to 5 mM Tris-HCl, pH 8.0, 150 mM NaCl and 10 mM DTT. Discussion ALDH1A3 crystals were grown by using the hanging drop method. Five µL of ATP binding sites are not uncommon in NAD(P)-dependent protein solution were mixed with 5 µL of crystallization solution and the resulting drop was equilibrated for a week against a reservoir containing 500 µL of dehydrogenases and a few have been reported in members of the crystallization solution at room temperature. Protein solution consisted of ALDH superfamily, but never before in human ALDHs. ATP- 5–10 mg/mL of ALDH1A3 in storage buffer supplemented with 10 mM MnCl2 for binding sites were described in the crystal structures of human the apo form, 2 mM NAD+ for the NAD+ complex and 2 mM ATP for the ATP mitochondrial NAD-dependent isocitrate dehydrogenase and complex. The crystallization solution consisted of 100 mM BIS-Tris, pH 5.5, NAD(P)-dependent malic enzyme, which are inhibited by ATP. 2–2.2 M ammonium sulfate and 5% PEG 400. The seeding method allowed growing the crystals reproducibly reaching their maximum size (approximately A non-phosphorylating glyceraldehyde-3-phosphate dehy- 0.5 mm) in 3–4 days. The crystals were then transferred into a cryoprotecting drogenase from Thermoproteus tenax participates in glycolysis solution consisting in 100 mM BIS-Tris, pH 5.5, 2.5 M ammonium sulfate, 5% PEG and is regulated by ATP, and ALDH5A1 from Arabidopsis 400 and 15% glycerol prior to be loop-mounted and stored in liquid nitrogen. thaliana binds ATP at its NAD+-binding site29–33. Here, ATP functions as an active-site inhibitor of ALDH. The modulation of X-ray data collection, data processing and structure refinement. Diffraction ALDH activity by ATP might have interesting physiological data collection at 100 K was performed in the BL13-XALOC beamline41 at the COMMUNICATIONS BIOLOGY | (2022)5:354 | https://doi.org/10.1038/s42003-022-03311-1 | www.nature.com/commsbio 9 ARTICLE COMMUNICATIONS BIOLOGY | https://doi.org/10.1038/s42003-022-03311-1 ALBA synchrotron using a Pilatus 6 M photon-counting detector (DECTRIS, reaction was carried out for 15 min at 37 °C in a final volume of 500 µL, after 5-min Baden, Switzerland). The beam was set at 12.661 keV photon energy (λ = 0.979 Å). preincubation with 2 mM, 5 mM and 7.5 mM of each ATP, ADP or AMP, and using Each complete data set was indexed and integrated using XDS42 in the space group 50 mM HEPES, 5 mM DTT, pH 8.0, as the reaction buffer and a concentration of P 21 2 21. Data sets were scaled and the structure factors were generated using cofactor near the Km value (24 µM NAD+). The reaction was stopped with 1 mL of AIMLESS43. Cell content analysis indicated two molecules in the asymmetric unit cold hexane:dioxane:isopropanol (50:5:1, v/v/v), and retinoids were extracted by a two- with a solvent content of 50.75%. The initial phases to solve the structure were step procedure with the same solvent mixture54,55. The aqueous phase was removed, calculated by molecular replacement with Phaser44 using the PDB entry 5FHZ as a and the organic phase was evaporated under a N2 stream. Retinoids were dissolved in model. Structure refinement was carried out with Buster45 and the examination of 200 µL hexane and analyzed by a modification of a published method56, as follows. the electron density maps and manual rebuilding were performed with Coot46. Retinoids were separated by HPLC on a Novapak® Silica 4 µm, 3.9 × 150 mm in Polder maps were calculated with phenix.polder47. The POVME48 algorithm hexane:methyl-tert-butyl ether (96:4, v/v) mobile phase, at a flow rate of 2 mL/min, was used to measure the volume of the substrate and cofactor-binding pockets in using a Waters Alliance 2695 HPLC. Elution was monitored at 370 nm and 350 nm for all refined ALDH1A3 structures. LigPlot+49 was used to point the interactions all-trans-retinaldehyde and all-trans-retinoic acid, respectively, using a Waters 2996 between the enzyme and the solvent with the ligand. All figures in this work were photodiode array detector. Quantification of retinoids was performed by interpolation performed with Pymol50. of HPLC peak areas into a calibration curve of known retinoid concentration. All compound manipulations were performed under red light to prevent photoisomerization. Fluorimetric assay for inhibition screening of ATP against ALDH1A isoforms, ALDH2 and ALDH3A1. Enzymatic activity was measured fluorimetrically using a Cary Eclipse (Varian) fluorimeter to follow the purification procedure and to check Statistics and reproducibility. GraFit 5.0 (Erithacus Software) was used for the enzyme concentration before each kinetic experiment8. Standard activity was kinetic data analysis. For most analyses, all technical replicates (n = 2 or 3) were measured at saturating concentrations of substrate using 30 µM hexanal for shown, and parameter values were expressed as the calculated value ± ALDH1A1, 250 µM hexanal for ALDH1A2, ALDH1A3 and ALDH2, and 750 µM standard error. hexanal for ALDH3A1. Regarding cofactor concentrations, all assays were per- formed at NAD+ concentrations near Km values (25 µM NAD+ for ALDH1A1, 20 µM NAD+ for ALDH1A2, 130 µM NAD+ for ALDH1A3 and ALDH2, and Reporting summary. Further information on research design is available in the Nature 100 µM NADP+ for ALDH3A1)8. The inhibition screening was performed in Research Reporting Summary linked to this article. 50 mM HEPES, 0.5 mM EDTA, 0.5 mM DTT, pH 8.0, for ALDH1A1, ALDH1A2 and ALDH2; 50 mM HEPES, 5 mM DTT, pH 8.0, for ALDH1A3; and 50 mM Tris/ HCl, 5 mM DTT, pH 8.0, for ALDH3A1. ATP was dissolved in the appropriate Data availability reaction buffer, and the corresponding solution was added to the reaction mixture Structure deposition: The atomic coordinates and structural factors of the three human in a final concentration of 2 mM. ATP, hexanal as a substrate and enzyme were ALDH1A3 structures, i.e., the apo form, the ALDH1A3-NAD+ complex and the incubated for 5 min at 25 °C and the reaction was started by the addition of ALDH1A3-ATP complex, have been deposited in the Protein Data Bank (PDB, cofactor NAD+. The ATP effect was also tested when the 5-min pre-incubation www.rcsb.org) with the accession codes 7QK7, 7QK8, and 7QK9, respectively. The was carried out in the presence of cofactor and the reaction was started by the source data underlying Fig. 7 are provided as Supplementary Data 1. Any remaining addition of hexanal. information related to the data generated or analyzed in this study is available from the corresponding author upon reasonable request. Determination of kinetic constants and type of inhibition of ATP against ALDH1A3. The IC50 value of ALDH1A3 for ATP was determined using the Received: 27 May 2021; Accepted: 24 March 2022; experimental conditions described above, except for a concentration of cofactor near the Km value in the absence of Mg2+ ions (24 µM NAD+). The effect of magnesium ions was also checked using a reaction buffer with MgCl2 (50 mM HEPES, 30 mM MgCl2, 5 mM DTT, pH 8.0). In this case, 130 µM NAD+ was used. The IC50 parameters were calculated by nonlinear regression analysis, fitting the initial rates to the appropriate equation using GraFit 5.0 (Erithacus Software). Parameter values were given as the calculated value ± standard error of ten dif- References ferent ATP concentrations (0–100 mM) run in duplicate, and the standard error 1. Marchitti, S. A., Brocker, C., Stagos, D. & Vasiliou, V. Non-P450 aldehyde value was less than 20% of the mean value. Activity assays to determine the type of oxidizing enzymes: the aldehyde dehydrogenase superfamily. Expert Opin. inhibition and Ki value of ATP against ALDH1A3 were carried out in the absence Drug Metab. Toxicol. 4, 697–720 (2008). of Mg2+ ions, using various concentrations of NAD+ cofactor at fixed substrate 2. Grimley, E. et al. Aldehyde dehydrogenase inhibitors promote DNA damage and inhibitor concentrations. Specifically, the concentration of hexanal was 250 µM in ovarian cancer and synergize with ATM/ATR inhibitors. Theranostics 11, and the concentrations of ATP were 2.0 mM, 5.0 mM and 7.5 mM, keeping the 3540–3551 (2021). same conditions as those of the inhibition screening and IC50 experiments and 3. Pérez-Alea, M. et al. ALDH1A3 is epigenetically regulated during melanocyte using GraFit 5.0 for data analysis. The data obtained at different concentrations of transformation and is a target for melanoma treatment. Oncogene 36, ATP were fitted to the Michaelis-Menten equation to determine the values of 5695–5708 (2017). kinetic parameters, Vmax and Km. Then, results were fitted to the competitive, non- 4. Steinmetz, C. G., Xie, P., Weiner, H. & Hurley, T. D. Structure of competitive, mixed and uncompetitive inhibition equations and the kinetic con- mitochondrial aldehyde dehydrogenase: the genetic component of ethanol stant values were determined from the type of inhibition showing the most aversion. Structure 5, 701–711 (1997). accurate fit. Parameter values were expressed as the calculated value ± 5. Moore, S. A. et al. Sheep liver cytosolic aldehyde dehydrogenase: the structure standard error. reveals the basis for the retinal specificity of class I aldehyde dehydrogenases. Structure 6, 1541–1551 (1998). Spectrophotometric assay to determine the effect of ATP on ALDH1A3 6. Perez-Miller, S. J. & Hurley, T. D. Coenzyme isomerization is integral to esterase activity. ALDH1A3 was assayed for esterase activity in a Cary Bio catalysis in aldehyde dehydrogenase. Biochemistry 42, 7100–7109 (2003). (Varian) spectrophotometer. Stock solutions of 10 mM p-nitrophenyl acetate were 7. Hammen, P. K., Allali-Hassani, A., Hallenga, K., Hurley, T. D. & Weiner, H. prepared with ethanol as a solvent to minimize the spontaneous hydrolysis, always Multiple conformations of NAD and NADH when bound to human cytosolic ensuring that the final ethanol concentration in the assay did not exceed 1% (v/v)51. and mitochondrial aldehyde dehydrogenase. Biochemistry 41, 7156–7168 Reactions were performed at 25 °C in 50 mM HEPES, 0.5 mM DTT, pH 7.5, with (2002). 500 µM NAD+ as a cofactor, or in the absence of cofactor. In both cases, ATP was 8. Pequerul, R. et al. Structural and kinetic features of aldehyde dehydrogenase added at 500 µM (an equimolar concentration of cofactor) or 2 mM (IC50 con- 1A (ALDH1A) subfamily members, cancer stem cell markers active in retinoic centration) in order to check the ATP effect in presence or absence of NAD+. The acid biosynthesis. Arch. Biochem. Biophys. 681, 108256 (2020). reactions were initiated by the addition of 50 µM p-nitrophenyl acetate to the 9. Jiménez, R. et al. Inhibitors of aldehyde dehydrogenases of the 1A subfamily as enzyme solutions in 1-mL cuvettes. The formation of p-nitrophenol was followed putative anticancer agents: Kinetic characterization and effect on human at 400 nm (ε400 = 18,200 M−1 · cm−1)52 to monitor esterase activity, which was cancer cells. Chem. Biol. Interact. 306, 123–130 (2019). corrected for the non-enzymatic hydrolysis of p-nitrophenyl acetate. 10. Khanna, M. et al. Discovery of a novel class of covalent inhibitor for aldehyde dehydrogenases. J. Biol. Chem. 286, 43486–43494 (2011). 11. Tsybovsky, Y. & Krupenko, S. A. Conserved catalytic residues of the HPLC-based assay for the inhibitory effect of ATP on ALDH1A3 using all-trans- ALDH1L1 aldehyde dehydrogenase domain control binding and discharging retinaldehyde as a substrate. An HPLC method was used to evaluate the reti- of the coenzyme. J. Biol. Chem. 286, 23357–23367 (2011). naldehyde dehydrogenase activity of human ALDH1A3. Working solutions of retinoids 12. Hayes, K. et al. The quaternary structure of Thermus thermophilus aldehyde were prepared by 10-min sonication in the presence of equimolar delipidated BSA53. The actual concentration of solubilized retinoid was determined based on the corre- dehydrogenase is stabilized by an evolutionary distinct C-terminal arm sponding molar absorption coefficient (ε370 = 29,500 M−1·cm−1). The enzymatic extension. Sci. Rep. 8, 1–14 (2018). 10 COMMUNICATIONS BIOLOGY | (2022)5:354 | https://doi.org/10.1038/s42003-022-03311-1 | www.nature.com/commsbio COMMUNICATIONS BIOLOGY | https://doi.org/10.1038/s42003-022-03311-1 ARTICLE 13. Moretti, A. et al. Crystal structure of human aldehyde dehydrogenase 1A3 40. Ni, W. et al. High expression of ALDH1A3 might independently influence complexed with NAD+ and retinoic acid. Sci. Rep. 6, 1–12 (2016). poor progression-free and overall survival in patients with glioma via 14. Quattrini, L. et al. Imidazo[1,2-a]pyridine derivatives as aldehyde maintaining glucose uptake and lactate production. Cell Biol. Int. 44, 569–582 dehydrogenase inhibitors: novel chemotypes to target glioblastoma stem cells. (2020). J. Med. Chem. 63, 4603–4616 (2020). 41. Juanhuix, J. et al. Developments in optics and performance at BL13-XALOC, 15. Chen, Y. et al. Structural basis of ALDH1A2 inhibition by irreversible and the macromolecular crystallography beamline at the Alba Synchrotron. J. reversible small molecule inhibitors. ACS Chem. Biol. 13, 582–590 (2018). Synchrotron Radiat. 21, 679–689 (2014). 16. Morgan, C. A. & Hurley, T. D. Development of a high-throughput in vitro 42. Kabsch, W. XDS. Acta Crystallogr. Sect. D. Biol. Crystallogr. 66, 125–132 assay to identify selective inhibitors for human ALDH1A1. Chem. Biol. (2010). Interact. 234, 29–37 (2015). 43. Evans, P. R. & Murshudov, G. N. How good are my data and what is the 17. Hemenway, J. N. et al. Formation of reactive impurities in aqueous and neat resolution? Acta Crystallogr. Sect. D. Biol. Crystallogr. 69, 1204–1214 (2013). polyethylene glycol 400 and effects of antioxidants and oxidation inducers. J. 44. McCoy, A. J. et al. Phaser crystallographic software. J. Appl. Crystallogr. 40, Pharm. Sci. 101, 2271–2280 (2012). 658–674 (2007). 18. Larson, H. N., Weiner, H. & Hurley, T. D. Disruption of the coenzyme 45. Bricogne G., et al. BUSTER version 2.10.3. (2017). binding site and dimer interface revealed in the crystal structure of 46. Emsley, P., Lohkamp, B., Scott, W. G. & Cowtan, K. Features and development mitochondrial aldehyde dehydrogenase ‘Asian’ variant. J. Biol. Chem. 280, of Coot. Acta Crystallogr. Sect. D. Biol. Crystallogr. 66, 486–501 (2010). 30550–30556 (2005). 47. Liebschner, D. et al. Polder maps: improving OMIT maps by excluding bulk 19. Kopečný, D. et al. Plant ALDH10 family identifying critical residues for solvent. Acta Crystallogr. Sect. D. Struct. Biol. 73, 148–157 (2017). substrate specificity and trapping a thiohemiacetal intermediate. J. Biol. Chem. 48. Wagner, J. R. et al. POVME 3.0: Software for Mapping Binding Pocket 288, 9491–9507 (2013). Flexibility. J. Chem. Theory Comput. 13, 4584–4592 (2017). 20. D’Ambrosio, K. et al. The first crystal structure of a thioacylenzyme 49. Laskowski, R. A. & Swindells, M. B. LigPlot+: multiple ligand-protein intermediate in the ALDH family: new coenzyme conformation and relevance interaction diagrams for drug discovery. J. Chem. Inf. Model. 51, 2778–2786 to catalysis. Biochemistry 45, 2978–2986 (2006). (2011). 21. Luo, M., Gates, K. S., Henzl, M. T. & Tanner, J. J. Diethylaminobenzaldehyde 50. Schrödinger, L. ThePyMOL Molecular Graphics System, Version 2.0. is a covalent, irreversible inactivator of ALDH7A1. ACS Chem. Biol. 10, 51. Feldman, R. I. & Weiner, H. Horse liver aldehyde dehydrogenase. II. Kinetics 693–697 (2015). and mechanistic implications of the dehydrogenase and esterase activity. J. 22. Huang, H. et al. Physiological levels of ATP negatively regulate proteasome Biol. Chem. 247, 267–272 (1972). function. Cell Res. 20, 1372–1385 (2010). 52. Weiner, H., Hu, J. H. J. & Sanny, C. G. Rate limiting steps for the esterase and 23. Salin, K. et al. Simultaneous measurement of mitochondrial respiration and dehydrogenase reaction catalyzed by horse liver aldehyde dehydrogenase. J. ATP production in tissue homogenates and calculation of effective P/O ratios. Biol. Chem. 251, 3853–3855 (1976). Physiol. Rep. 4, 1–9 (2016). 53. Gallego, O. et al. Comparative functional analysis of human medium-chain 24. Blat, Y. Non-competitive inhibition by active site binders. Chem. Biol. Drug dehydrogenases, short-chain dehydrogenases / reductases and aldo-keto Des. 75, 535–540 (2010). reductases with retinoids. Biochem. J. 399, 101–109 (2006). 25. Wang, X. & Weiner, H. Involvement of glutamate 268 in the active site of 54. Evans, J. E. & McCaffery, P. HPLC/MS analysis of retinoids. In Retinoids: human liver mitochondrial (Class 2) aldehyde dehydrogenase as probed by Methods and Protocols; Chapter 8 (eds. Sun, H. & Travis, G. H.) 149–162 site-directed mutagenesis. Biochemistry 34, 237–243 (1995). (2010). 26. Farrés, J., Wang, T. T. Y., Cunningham, S. J. & Weiner, H. Investigation of the 55. McCaffery, P. et al. Retinoid quantification by HPLC/MSn. J. Lipid Res. 43, active site cysteine residue of rat liver mitochondrial aldehyde dehydrogenase 1143–1149 (2002). by site-directed mutagenesis. Biochemistry 34, 2592–2598 (1995). 56. Kane, M. A., Chen, N., Sparks, S. & Napoli, J. L. Quantification of endogenous 27. Sidhu, R. S. & Blair, A. H. Human liver aldehyde dehydrogenase. Kinetics of retinoic acid in limited biological samples by LC/MS/MS. Biochem. J. 388, aldehyde oxidation. J. Biol. Chem. 250, 7899–7904 (1975). 363–369 (2005). 28. Takahashi, K. & Weiner, H. Nicotinamide adenine dinucleotide activation of the esterase reaction of horse liver aldehyde dehydrogenase. Biochemistry 20, 2720–2726 (1981). Acknowledgements 29. Hung, H. C. et al. Functional roles of ATP-binding residues in the catalytic site This research was funded by the Spanish Ministerio de Ciencia e Innovación (Agencia of human mitochondrial NAD(P)+-dependent malic enzyme. Biochemistry Estatal de Investigación, grant number PID2020‐119424RB‐I00 / AEI / 10.13039/ 44, 12737–12745 (2005). 501100011033). V.B. obtained financial support from the Tuscany Region Ph.D. School 30. Sun, P., Bai, T., Ma, T. & Ding, J. Molecular mechanism of the dual regulatory in Biochemistry and Molecular Biology. The authors thank the staff of beamline BL13- roles of ATP on the αγ heterodimer of human NAD-dependent isocitrate XALOC at ALBA Synchrotron. dehydrogenase. Sci. Rep. 10, 1–11 (2020). 31. Busch, K., Piehler, J. & Fromm, H. Plant succinic semialdehyde Author contributions dehydrogenase: dissection of nucleotide binding by surface plasmon resonance The study was designed by A.C. and J.F. Protein expression and purification was per- and fluorescence spectroscopy. Biochemistry 39, 10110–10117 (2000). formed by A.C., R.P. and V.B., who also were involved in crystallization trials and X-ray 32. Cao, R., Zeidan, A. A., Rådström, P. & Van Niel, E. W. J. Inhibition kinetics of data acquisition. Structural analysis and image rendering were performed by A.C., with catabolic dehydrogenases by elevated moieties of ATP and ADP - Implication insightful contributions from J.J. and J.F. Inhibition kinetics was carried out by R.P. The for a new regulation mechanism in Lactococcus lactis. FEBS J. 277, 1843–1852 manuscript was written by A.C., R.P. and J.F., with input from J.J. and X.P. Funding (2010). acquisition by X.P. and J.F. 33. Brunner, N. A., Brinkmann, H., Siebers, B. & Hensel, R. NAD+-dependent glyceraldehyde-3-phosphate dehydrogenase from Thermoproteus tenax. J. Biol. Chem. 273, 6149–6156 (1998). Competing interests 34. Qian, Y. et al. Extracellular ATP is internalized by macropinocytosis and The authors declare no competing interests. induces intracellular ATP increase and drug resistance in cancer cells. Cancer Lett. 351, 242–251 (2014). 35. Wang, X. et al. Extracellular ATP, as an energy and phosphorylating molecule, Additional information Supplementary information The online version contains supplementary material induces different types of drug resistances in cancer cells through ATP available at https://doi.org/10.1038/s42003-022-03311-1. internalization and intracellular ATP level increase. Oncotarget 8, 87860–87877 (2017). Correspondence and requests for materials should be addressed to Jaume Farrés. 36. Kang, J. H. et al. Aldehyde dehydrogenase is used by cancer cells for energy metabolism. Exp. Mol. Med. 48, e272 (2016). Peer review information Communications Biology thanks Ernst Schonbrunn and the 37. Mao, P. et al. Mesenchymal glioma stem cells are maintained by activated other, anonymous, reviewer(s) for their contribution to the peer review of this work. glycolytic metabolism involving aldehyde dehydrogenase 1A3. Proc. Natl Primary Handling Editors: Eve Rogers. Acad. Sci. USA 110, 8644–8649 (2013). 38. Duan, J. J., Cai, J., Guo, Y. F., Bian, X. W. & Yu, S. C. ALDH1A3, a metabolic Reprints and permission information is available at http://www.nature.com/reprints target for cancer diagnosis and therapy. Int. J. Cancer 139, 965–975 (2016). 39. Nie, S. et al. ALDH1A3 accelerates pancreatic cancer metastasis by promoting Publisher’s note Springer Nature remains neutral with regard to jurisdictional claims in glucose metabolism. Front. Oncol. 10, 1–14 (2020). published maps and institutional affiliations. COMMUNICATIONS BIOLOGY | (2022)5:354 | https://doi.org/10.1038/s42003-022-03311-1 | www.nature.com/commsbio 11 ARTICLE COMMUNICATIONS BIOLOGY | https://doi.org/10.1038/s42003-022-03311-1 Open Access This article is licensed under a Creative Commons Attribution 4.0 International License, which permits use, sharing, adaptation, distribution and reproduction in any medium or format, as long as you give appropriate credit to the original author(s) and the source, provide a link to the Creative Commons license, and indicate if changes were made. The images or other third party material in this article are included in the article’s Creative Commons license, unless indicated otherwise in a credit line to the material. If material is not included in the article’s Creative Commons license and your intended use is not permitted by statutory regulation or exceeds the permitted use, you will need to obtain permission directly from the copyright holder. To view a copy of this license, visit http://creativecommons.org/ licenses/by/4.0/. © The Author(s) 2022 12 COMMUNICATIONS BIOLOGY | (2022)5:354 | https://doi.org/10.1038/s42003-022-03311-1 | www.nature.com/commsbio