General features of the mADK structures
Crystal structures of Δ19 mADK complexed with adenosine (mADK:ADO) and full- length-mADK complexed with adenosine and adenosine-5′-diphosphate (mADK:ADO:ADP) were determined at 1.2 and 1.8 Å resolution, respectively (Table 1). The final models include the residues 19 to 361 of mADK:ADO (Supplementary Fig. 1a) and 20 to 360 of mADK:ADO:ADP (Supplementary Fig. 1b). Crystals from mADK:ADO grew in the monoclinic space group P1,21,1, while crystals from mADK:ADO:ADP grew in the orthorhombic space group P21,21,21. The enzyme consists of two unequally sized domains (i.e., a large three-layer (αβα) sandwich domain and a small lid α/β domain), closely matched to the previously reported human ADK16 (Supplementary Fig. 2a) The large domain possesses two binding cavities that are the sites for ATP (BS1) and ADO (BS2). The structures shown here feature the BS1 and BS2 binding sites, the catalytic base Asp316 and an anion hole motif (DTNGAG) in the large domain, whereas the small or lid domain contains the residue Arg148 that is involved in binding to ATP (Fig. 1). In addition to the substrates of each complex, we observed functionally relevant ions in the ternary complex including one potassium, one chloride and two magnesium ions.
Interactions involved in ADO and ADP binding
In the mADK:ADO complex, electronic density for ADO was identified in the BS1 site, which is a large hydrophobic cluster involving residues Ile308, Val300, Ala314, Phe318, Ala343, Ile347 and polar residues Thr281, Arg284, Asp302, Gln305, His340, Gln303, Thr287, Thr311 and Gln282 (Supplementary Fig. 3a). In this site, ADO assumes a standard anti-conformation with a glycosidic torsion angle of −165.4°, while the sugar ring adopts the common C3′-endo conformation. The adenine group is hydrogen bonded to Gln305, involving atoms N1-NE2 and N6-OE1. The ribose ring makes no direct contact with the enzyme but instead establishes water-mediated interactions with the main chain atoms of residues Thr281, Gly283, and His340. Considering the peculiar occupancy of BS1 by ADO in both mADK (PDB: 5KB6) and HsADK (PDB:1BX4) binary complexes, and the potential importance of such an occupancy to the mechanism of auto-inhibition of ADK by the substrate, we next performed a detailed analysis of the superposed structures of both mADK and HsADK binary complexes (Supplementary Fig. 2a). The superposition of these binary complexes showed marked differences between the structures (overall r.m.s.d. 3.14 Å, 9.74 Å to the lid and 0.77 Å to the large domain). Distinct from the mADK, the HsADK binary structure assumes a closed conformation (~34.6° bending of the lid over the large domain), which agrees with the notion17,18 that the occupancy of the BS2 site by the substrate is a major factor in determining the closure of the lid over the large domain in the active form of the enzyme. Moreover, a detailed analysis of the BS1 in the superposed structures of mADK and HsADK binary complexes showed that ADO occupies a similar pocket of BS1 in both structures (Supplementary Fig. 2a). This configuration suggests that ADO may compete with ATP for the binding to BS1, which could impact on the auto-inhibitory influence of ADO on ADK. Thus, we next sought to explore the potential mechanism of ADO auto-inhibitory influence on ADK. The inhibition of ADK by ADO and the Michaelis-Menten kinetics for ATP are shown in the Fig. 2a,b, and Supplementary Table 1. In addition, we also show that stepwise increases of ADO reduce the affinity of mADK to ATP, as indicated by the parallel increases of the Km (Fig. 2c). The mechanism of ADO inhibition was assessed by double reciprocal Lineweaver-Burk plots data (Fig. 2d). The apparent Km values observed for the ATP are higher, the greater the concentrations of ADO in the assay (Supplementary Table 2), indicating that ADO is a competitive inhibitor for ATP, which corroborates with data from previous kinetic studies19. Overall the combination of the structural and kinetic data of the present study help to clarify some discrepancies in the literature about the kinetic mechanism and the structural models available for ADK of different species in its apo and complexed forms2. Although our present data do not completely discard the possibility of the existence of an additional regulatory adenosine-binding site in a different region of the enzyme12,13,14,17, it seems that, at least for the ADK from mammalian species, this is not the case.
In the mADK:ADO:ADP complex, the electronic density for ADO occupies the BS2 site formed by residues Asn30, Leu32, Leu56, Asp34, Gly79, Gly80, Ser81, Asn84, Cys139, Ala152, Leu150, Leu154, Phe186, Gly313, Asn312 and Asp316 (Supplementary Fig. 3b). In this complex, ADO shows a glycosidic torsion angle of −123.3° and a O4′-endo sugar pucker. The adenine ring displays a stacking interaction with Phe186, a hydrogen bond with Ser81 (N3-N), and interactions through water molecules with Phe186 (O) and Asn30 (OD1). Several hydrogen bonds anchor the ribose moiety: the O2′ and O3′ atoms bind to the side chain of residues Asp34, Gly80, and Asn84, while the O5′ atom interacts with the side chain of the catalytic residue Asp316. In this complex, ADP occupies BS1. This site arises from the edges of β strands β11, β12, β13, β14 and α helices α12 and α13 on the enzyme surface. The adenine group of ADP binds to Gln305 through a hydrogen bond and to Asp302 through a water-mediated interaction. Val300, Ile308, Ile347, Ala343 and Ala314 complete a hydrophobic cluster around the adenine ring (Supplementary Fig. 3c). The sugar ring does not interact directly with the enzyme, but it establishes hydrogen bonds with the solvent molecules through its O2′ and O3′ atoms. The phosphate tail of ADP interacts with the enzyme through an extensive network of direct and water-mediated hydrogen bonds. The O2A atom of the α-phosphate interacts with Thr281 via Oγ1, Gly283 via N and Mg2+. Water molecules and Mg2+ mediate additional contacts with the α-phosphate and β-phosphate groups. The β-phosphate also interacts directly with mADK through hydrogen bonds with Arg148 via NH1-O1B from the lid domain, with Asn239 via ND2-O1B, and Gly315 via N-O3B from the large domain. Additional interactions between β-phosphate, Mg2+ and water molecules further confer stability to ADP binding to mADK.
Conformational differences between the binary and the ternary mADK complexes
Crystal structures of ADK from different species have been determined and demonstrate a large extent of structural conservation, despite the considerable sequence diversity16,17,20,21,22,23,24. From these data arises a model in that the binding of ADO to the BS2 site induces a large-scale bending motion of the lid domain toward the large domain, resulting in a “closed” conformation in the active state. Subsequent formation of an anion hole, induced by the binding of ATP, completes the structural requirements for catalysis23,25. The superposition of the binary and ternary mADK structures highlights the large-scale conformational changes expected for the transition of a fully open to a closed form of the enzyme (Fig. 3a). The least-square fitting over all α-carbon atoms resulted in an r.m.s.d deviation between the structures of 3.32 Å. The ternary complex is in a closed conformation with the two domains moving by up to 39.4° compared to the binary complex, which assumes a conformation that is reminiscent of the apo form of ADK from T. brucei18 (PDB: 4N08) (Fig. 3b). Besides, we found localized differences between the binary and the ternary complex, which show how binding to the substrates and the changes in closed conformation prepare the enzyme for catalysis (Fig. 3c). In the binary complex, Asn312 (which is an integral part of the anion hole) appears in two alternate conformers as a function of the torsion angle Ψ1 (gauche+ and trans conformations). However, in the closed ternary complex, it only adopts the trans conformation, which results from steric impositions by the Cβ, Cγ, and Cα of Arg148 of the lid subdomain. This residue is brought to interact with the β-phosphate of ADP in the closed conformation, which perturbs the segment comprising the Gln303-Ile309 sequence adjacent to the anion hole segment (Fig. 3d). Interestingly, these changes appear to coincide with the presence of K+ between the α11-12 and the C-terminal loops..
Atomic detail of the potassium-binding site in mADK
In this study, the high-resolution data along with coordination geometry analysis of the ternary mADK crystal structure enabled the identification of an atomic density corresponding to a K+ in the ADK structure (Fig. 4a). The K+ binding site locates in the large subdomain surrounded by two loops that span from residues Ile309 to Asn312 (α11 and α12 loop) and residues Arg348 to Cys352 (C-terminus), adjacent to the BS1 pocket (Fig. 4b). The ion adopts an octahedral geometry with the backbone carbonyl oxygen of residues Asp310, Asn312, Ile346, Arg349 and Gly351 (Fig. 4c). Besides, the critical residue for K+ coordination, Asp310, contacts the ion via main and side chains and presents a distinct rotameric conformation in the K+ -bound form. In the absence of K+ (mADK binary complex) this residue is rotated away from the ion (Fig. 4d). Of note, this is the last residue in the conserved anion hole sequence, which is postulated to deprotonate the alcohol function during catalysis. Similar monovalent cation binding sites have been reported for members of the ribokinase family26, suggesting a conserved influence of cation interaction structure on the formation of the anion hole in ADK.
Potassium modulates mADK activity
The role of K+ in the mADK catalytic activity was determined by measuring the initial rate of the reaction as a function of chloride salt concentration in the assays (i.e., KCl, LiCl, NaCl) (Fig. 5a). Basal enzyme activity of mADK was still detected under conditions of zero K+ concentration in the buffer assay, suggesting that the K+ ion is not needed for the basal activity of this enzyme, similarly to what was previously suggested by studies performed in extracts of human placenta27. However, given the difficulties to completely eliminate the K+ ion in the biological samples (either from E. coli or other sources), one cannot be sure that K+ is not necessary for the basal activity of ADK. In the samples of purified E. coli recombinant mADK, the K+ ion increased the initial rate of the reaction of mADK by 2.5-fold with respect to a buffer without K+. The dependence of mADK activation on K+ is saturable, with maximal kinase activity observed at 40 mM KCl. These data are well described by an equation derived from a one-site binding model. The estimated Kd of 10.4 mM suggests that at the physiological concentration (140 to 150 mEq/L), K+ fully activates the enzyme. The effects of Na+ and Li+ were also measured to ensure that the effects of the monovalent cations on the mADK activity were limited to K+ and were not dependent on an ionic strength effect. We observed no significant change in the mADK activity with increasing concentrations of NaCl or LiCl. To address the mechanism behind the activation by K+, we characterized the enzymatic activity of mADK as a function of the ATP concentration in the presence or absence of 100 mM KCl (Supplementary Table 3). There was a two-fold increase in the apparent Kcat values (Kcat0 = 0.25 s−1 versus KcatK+ = 0.46 s−1), while the apparent Km remained almost the same with and without K+ (Km0 = 2.4 µM versus KmK+ = 2.1 µM). In fact, the addition of K+ to the reaction duplicated the catalytic efficiency (Kcat/Km) from 0.1 to 0.22 s−1/µmol (Fig. 5b). Next, we examined if the structural integrity of the K+ binding site would be critical for the influence of this ion on mADK activity. Asp310 was targeted for mutation because structural analysis indicated that this residue is a crucial constituent of the loops that coordinate K+ binding to mADK (Fig. 4c). We performed mutations D310A, which substitute Asp310 by a neutral residue, and D310P because the Pro residue would be expected to break two interactions and yet affect the volume of the K+ ion coordination sphere cavity. Of note, as shown in the Supplementary Fig. 4a to d, all biophysical data supported that the D310A and D310P are well folded and still proficient at the catalytic point of view (Supplementary Table 1), but with reduced affinity (Supplementary Fig. 4e to h) to ATP (Km). As expected, both mutations completely abolished the sensitivity of the enzyme to the K+ (Fig. 5c,d), which is likely associated to the inability of the enzyme to be activated through conformational transitions induced by the binding to the K+ ion, similarly to other members of ribokinase family26,28.
Wild-type mADK and D310P mutant ATP interaction by NMR investigation
To further investigate the impact of the D310P mutation on the enzyme interaction with ATP, we next used two- and three-dimensional NMR spectroscopy to compare the wild-type and the D310P mutant of mADK. 3D NMR experiments used U-labeled mADK for backbone sequential resonance assignments. From the expected 331 peaks, we observed 209 peaks in the 15N-1H HSQC, from which we attributed 200 peaks to ADK residues (Supplementary Fig. 5a). As shown in the Supplementary Fig. 5b, the missing signals are located at multiple stretches of residues. Residues represented in gray were those that were not found in NMR assays. During the titration steps in the 15N-1H HSQC protocol saturation of mADK by the ligand some signals appear in the spectra, possibly due to conformational changes, but the overlapped signals turn the spectra crowded and more difficult to attribute them. Moreover, the protein becomes unstable.
Individual 2D 15N-1H HSQC spectra from samples of 15N-labeled wild-type and D310P mutant mADK were recorded in titration series with adenosine triphosphate-gamma-S (ATPγS) for 8 hours, using a 600-MHz spectrometer with a cryoprobe. The proteins were stable under the temperature and buffer conditions of the experiment, as revealed by the analysis of the spectra from the reference samples. We focused on residues around the BS1 site, which showed similar spectra in the reference samples of the wild-type and D310P mutant mADK. Comparing the signals from both wild-type (Fig. 6a) and the mutant D310P mADK (Fig. 6b), we found distinct changes in the 2D 15N-1H HSQC spectra during the initial steps of the interaction with ATPγS. Analysis of the spectra obtained from the titration experiments of the wild-type mADK indicated exchange broadening of residues Gly146, Arg148, Leu150, Asn212, Asn239, Thr281, Thr287, Ile288, Asn304, Gly313, Gly321, and Gly339 (Fig. 6c). However, the spectra of the D310P mutant showed changes restricted to the residues Arg148, Leu150, Asn212, Asn239, and Thr287. Besides, signals from many other residues such as Gly79, Lys98, Gly104, Asn132, Gly137, A140, Ile143, Lys164, Asp167, Leu213, and Phe298 underwent a reduction in intensity (Fig. 6d). The results showed that ATPγS caused significant chemical shift changes in the wild-type mADK, especially in the residues around the BS1 site. However, the D310P mutant did not present substantial exchange broadening in the residues adjacent to the BS1 site (Gly146, Asn304, Gly321, Gly313 and Gly339). It appears, therefore, that the integrity of the K+ binding site is crucial to stabilizing the BS1 site, favoring the nucleotide interaction. Considering that D310P mutation might cause ATP binding to BS2, we show in Supplementary Fig. 6a a rotated pose to highlight the residues of the mutant mADK that are affected during ATPγS titration. As noted, residues of the BS2 (shown in orange) (Supplementary Fig. 6b) remained unchanged during ATP titration, indicating that the D310P mutation does not predispose the interaction of ATP to the BS2.
Also, 1D 1H STD NMR spectra were obtained for the wild-type mADK (Fig. 6e) and for the D310P mutant (Fig. 6f) with ATPγS (1:400) to assess the modifications in the substrate that result from the interaction with mADK. Signals from ATPγS H-2 were the ones that underwent the most substantial increase in intensity in the presence of either the wild-type or the mutant mADK. However, the enhancement of the signal (%STD) for the other protons was higher for the wild-type than those for the mutant mADK. The signal-to-noise ratio to the acquisition of the 1H STD NMR experiments demonstrated a value of approximately 40% lower for the mutant. In the case of 15N-1H HSQC titration, a greater amount of substrate was needed to see an initial backbone amide group cross-peak modification. Thus, 1D 1H STD NMR indicate that the D310P mutant has an unfavorable interaction with ATPγS relative to the wild-type mADK, which agrees with the kinetic data showing an increased Km for ATP in the D310P mADK.
Depletion of cellular K+ is accompanied by ADK inhibition
To test whether fluctuations in the intracellular levels of K+ could influence the activity of ADK in vivo, we performed experiments in HEK297 cells subjected to a previously reported experimental protocol of depletion29, followed by restoration of intracellular K+. In agreement with the in vitro data, ADK activity was reversibly inhibited when the K+ concentration in the culture medium was reduced, while the activity was restored when the K+ levels were gradually increased to normal levels (Fig. 7).