Structure and Dynamics of ATP and the ATP–Zn²⁺ Complex in Solution

Abstract

Despite the crucial role of ATP in life and artificial life-like applications, fundamental aspects relevant to its function, such as its conformational properties and its interaction with water and ions, remain unclear. Here, by integrating linear and two-dimensional infrared spectroscopy with ab initio molecular dynamics, we provide a detailed characterization of the vibrational spectra of the phosphate groups in ATP and in its complex with Zn²⁺ in water. Our study highlights the role of conformational disorder and solvation dynamics, beyond the harmonic normal-mode analysis, and reveals a complex scenario in which electronic and environmental effects tune the coupling between phosphate vibrations. We identify βγ-bidentate and αβγ-tridentate modes as the preferential coordination modes of Zn²⁺, as was proposed in the literature for Mg²⁺, although this conclusion is reached by a different spectral interpretation.

Adenosine 5′-triphosphate (ATP) is a ubiquitous substrate for many biological reactions and plays a crucial role in life as the primary energy source in cells. Recently, other possible, and to date controversial, roles of this molecule were suggested, as a cosolute governing biomolecular folding stability,¹ phase separation, and aggregation.^2,3 Furthermore, its fueling and templating ability makes ATP a key element for the design of artificial life-like materials.^4,5 Most of the special properties of ATP hinge on its triphosphate chain, which can easily store and release energy by bond breaking and reforming. These reactivity aspects are strongly influenced by the significant charge density of the chain and its ability to bind divalent ions. The combination of a highly charged moiety with the apolar nucleobase is likely crucial for imparting distinct properties to the hydration network of ATP⁶ and to its interaction with other molecules. It comes as no surprise that the chemical and structural features of ATP have been widely investigated since the middle of the past century.⁷ Since the early days, ³¹P nuclear magnetic resonance (NMR) spectroscopy was employed to investigate the binding of metal to the triphosphate chain. A downfield shift of the three phosphorus signals was observed upon metal binding: large for P_β, smaller for P_γ, and very small for P_α. This was at first taken as an indication of preferential monodentate or bidentate binding to P_β and P_γ, without direct P_α contact.⁷ It was subsequently pointed out that, though purely electronic factors cannot be discarded,⁸ the experimental differences are due mainly to changes in the conformation of the polyphosphate chain.⁹⁻¹¹ Thus, chemical-shift data are inconclusive with regard to the metal binding site. In addition to the geometry and energetics of its binding to divalent ions, uncertainty about other fundamental aspects that are relevant to the ATP function, such as its conformational and hydration properties, remains.

The stretching vibrational bands of phosphates in the range of 1000–1300 cm^–1 were early identified as very sensitive probes of environmental effects.^12,13 In both dimethyl phosphate (DMP)¹⁴ and DNA,^15,16 these were found to exhibit a red-shift with an increase in water content, which was ascribed to the decrease in the P–O stretching force constants upon formation of hydrogen bonds. More recently, much deeper insight into the origin of these effects has been reached by combining linear and two-dimensional infrared (2D-IR) spectroscopy and ab initio calculations.^17,18 Spectroscopic features were related quantitatively to the nature and geometry of ion coordination and to the structure and dynamics of local hydration of polynucleotides. In particular, the role of direct Mg²⁺–phosphate contact ion pairs in stabilizing the folded tertiary structure of tRNA was unveiled.¹⁹ In ATP, the presence of three phosphate groups adds complexity to the scenario. IR spectroscopy was applied early to this molecule,^20,21 in studies generally aimed at identifying the coordination sites of divalent ions. A major achievement was reached in 1988 by Takeuchi and co-workers.²² Using IR and Raman spectroscopy with selective isotopic ¹⁸O substitution of the phosphate oxygens, they assigned several IR bands to vibrations predominantly involving single phosphate groups and then, from the spectral changes upon Mg²⁺ binding, they proposed that at pH 7.5 ATP–Mg²⁺ is present as a mixture of βγ-bidentate and αβγ-tridentate complexes. This interpretation was the basis for subsequent vibrational spectroscopy investigations of analogous systems.^23,24

In this Letter, we aim to unveil the interplay between the solvation and conformation of aqueous ATP and its complex with Zn²⁺, in a combined IR spectroscopy and ab initio simulation approach. The choice of zinc is motivated by the fact that this is one of the most abundant metal ions in biological materials, where its complexes with ATP play a crucial role in intra- and extracellular metabolic pathways.^25,26 The ATP–Zn²⁺ complex is even preferred over the more common ATP–Mg²⁺ by a few kinases,²⁷⁻²⁹ such as the NAD kinase.³⁰ Zn²⁺ exhibits versatile chemistry due to its ability to interact with nitrogen, oxygen, sulfur, and halogen atoms, which is attributed to its borderline hard/soft character.³¹ It can coordinate to polyamine ligands^32,33 to form stable complexes capable of binding to phosphate, which is a desirable feature for the synthetic design of artificial nucleases^32,33 and of ATP-fueled self-assembling systems.⁴ Although similar in size to the more widely investigated Mg²⁺,³⁴ Zn²⁺ presents a different coordination chemistry. With respect to previously investigated monophosphate complexes,^17−19,35 ATP and its metal complexes pose further spectroscopic and computational challenges, which originate from the conformational variability and from the possible couplings between vibrational modes.

We present IR absorption spectra of phosphate stretching vibrations of ATP with Zn²⁺ ions at different concentrations and study vibrational line shapes and couplings by 2D-IR spectroscopy. To account for the fluctuations of the triphosphate chain, the interactions between water and nucleotide, and temperature and anharmonic effects,³⁶ we use ab initio molecular dynamics simulations (AIMD). IR spectra are computed via the autocorrelation function (ACF) of the time derivative of the total dipole moment³⁷ (Berry phase approach^38,39). Studies of relatively simple systems³⁷ have demonstrated the advantages of this method, but there have been only very few applications to flexible solutes in an interacting environment.⁴⁰ We disentangle and assign the vibrational bands through the vibrational density of states (VDOS)³⁷ and discuss the different spectral contributions in terms of electronic and hydration effects involving the three phosphates. We show that the IR spectra of ATP in water can be explained only by invoking the contribution of different conformers and, in the presence of Zn²⁺, of different binding modes, and we identify βγ and αβγ as the prevalent ones. Although these same modes were identified in previous IR spectroscopy studies of ATP–Mg²⁺ and GTP–Mg²⁺ complexes,^22,23 our spectral interpretation is different.

For the experiments, adenosine 5′-triphosphate disodium hydrate was dissolved in pure water at concentrations of 0.1 and 0.2 M, for the linear IR and the 2D-IR spectra, respectively. The solutions had a pH of 7.8, where the hydrate is fully dissociated and the triphosphate chain has a charge of −4e (e is the elementary charge). ZnCl₂ was then added to this solution in varying concentrations (0.025–0.1 M) while the pH of 7.8 was maintained by adding appropriate amounts of NaOH. Upon addition of Zn²⁺, we did not observe frequency changes of fingerprint modes of adenine in the range of 1280–1625 cm^–1 (see Figure S1). In particular, the IR band at ∼1605 cm^–1, which is due to a combination of a pyrimidine ring and NH₂ scissor vibration,⁴¹ displays an unchanged spectral position. We conclude that the adenine moiety plays a minor role in binding Zn²⁺ ions.

Figure 1A shows linear IR absorption spectra in the range of 1000–1300 cm^–1, recorded at increasing Zn²⁺ concentrations. The absorption background from the solvent water was subtracted. The spectrum of ATP exhibits a strong band at 1120 cm^–1 with a shoulder at 1080 cm^–1 and a high-frequency band at 1230 cm^–1. In ref (22), the group of bands between 1060 and 1160 cm^–1 was assigned to symmetric PO₂ stretching vibrations, ν_S(PO₂), and to the degenerate PO₃ stretching vibration (at 1120 cm^–1), whereas the band at 1230 cm^–1 was ascribed to asymmetric PO₂ stretching vibrations, ν_A(PO₂). In our experiments, with an increasing Zn²⁺ concentration, the peak absorbance of both bands decreases, and a pronounced reshaping of the spectral envelopes arises. The high-frequency band blue-shifts by some 15 cm^–1 and reveals a two-band substructure. The changes are qualitatively similar to those reported for ATP in the presence of Mg²⁺ ions.²²

Figure 1.

(A) Linear infrared absorption spectra of the ATP–Zn²⁺ complex for 0.1 M ATP and Zn²⁺ concentrations ranging from 0.025 to 0.1 M (sample thickness of 25 μm). (B) Weighted differential absorption, obtained by subtracting the weighted spectrum of hydrated ATP from the total absorption spectrum. Assuming that all Zn²⁺ ions are bound to ATP, the different curves represent the absorption of ATP–Zn²⁺ complexes at various Zn²⁺ concentrations of ≤0.1 M (black line).

To separate the IR spectrum A_complex of ATP–Zn²⁺ complexes from the spectrum A_ATP of uncomplexed ATP, the differential absorbance A_complex = A_ATP/Zn – (1 – c_Zn)A_ATP was calculated. Here, A_ATP/Zn is the spectrum of the samples containing Zn²⁺ ions at concentration c_Zn. This treatment implicitly assumes that all Zn²⁺ ions form 1:1 complexes with ATP. In Figure 1B, such differential spectra A_complex are plotted for Zn²⁺ concentrations between 0.025 and 0.1 M. The spectra reveal an essentially linear increase in absorbance and a minor spectral reshaping with an increase in c_Zn.

2D-IR spectra of ATP and the ATP–Zn²⁺ complex in water were derived from three-pulse heterodyne-detected photon echo data recorded with femtosecond mid-infrared pulses. Experimental details are available in the Supporting Information. The 2D-IR spectra for a waiting time T of 300 fs are presented in Figure 2, where panel B shows the spectrum of ATP and panel C the spectrum of the ATP–Zn²⁺ complex. The absorptive 2D signal, given by the real part of the sum of the rephasing and non-rephasing 2D signals, is plotted as a function of excitation frequency ν₁ (ordinate) and detection frequency ν₃ (abscissa). Yellow–red contours (positive signal) along the diagonal ν₁ = ν₃ are due to v = 0–1 excitations of the different vibrations, while blue contours (negative signal) originate from the red-shifted v = 1–2 transitions. From the separation of such two components along the ν₃ axis, one estimates a diagonal anharmonicity Δ ≃ 15–20 cm^–1 for all modes.

Figure 2.

(A) Cuts of the 2D-IR spectra of ATP (black line) and the ATP–Zn²⁺ complex (blue line) in water along the frequency diagonal ν₁ = ν₃. The normalized absorptive 2D-IR signal is plotted as a function of detection frequency ν₃. (B and C) Two-dimensional infrared (2D-IR) spectra of ATP and the ATP–Zn²⁺ complex in water. The absorptive 2D-IR signal is plotted as a function of excitation frequency ν₁ and detection frequency ν₃. Positive signals are shown as yellow–red contours, and negative signals as blue contours. The signal change between neighboring contour lines is 6.5%. The spectra were recorded at a waiting time T of 300 fs.

Figure 2A shows diagonal cuts of the 2D-IR spectra along the frequency diagonal ν₁ = ν₃. The diagonal peaks of the 2D-IR spectra occur at detection frequencies ν₃ close to the frequency positions of the corresponding absorption bands in the linear IR spectra (cf. Figure 1). As the diagonal cuts make evident, the relative strengths and spectral width of the yellow–red diagonal peaks are different from the linear spectrum due to their different 2D line shapes and vibrational population lifetimes. The line shapes of such peaks are elongated along the diagonal, reflecting substantial inhomogeneous broadening of the underlying vibrational transitions. This finding points to structural disorder in the sample, e.g., a distribution of solvation geometries, conformational structures, and, in the case of the ATP–Zn²⁺ complex, coordination modes. The 2D-IR spectrum of ATP exhibits a pronounced cross peak around ν₁ = 1090 cm^–1 with components at ν₃ = 1200 and 1240 cm^–1, i.e., a frequency splitting Δν₃ = 40 cm^–1. This peak is due to anharmonic couplings between symmetric and asymmetric PO₂ stretching vibrations ν_S and ν_A. Describing the modes in an excitonic picture of two weakly coupled oscillators at fundamental frequencies ν_I = 1120 cm^–1 and ν_II = 1240 cm^–1, for which Δν₃ < (ν_II – ν_I), one derives an intermode coupling strength β = (ν_II – ν_I)[Δν₃/(4Δ)^1/2] ≈ 85 cm^–1 with Δ = 20 cm^–1.⁴²

Following the experimental suggestion that the nucleobase is not involved in the direct binding of the cation, we assumed methyl triphosphate (MTP) as a model for ATP in AIMD simulations. This permits reduction of the computational cost in favor of a more extensive sampling. In the calculation, fully deprotonated MTP, carrying a charge equal to −4e, was assumed, in a box containing 123 water molecules. An exhaustive sampling of the conformational space of the triphosphate chain is still beyond the reach of AIMD.⁴³ Therefore, to gain insight into the role of the conformational diversity, we considered three independent trajectories starting from different MTP initial structures suggested by previous molecular dynamics simulations, for a total of 126 ps and five different conformations (details in the Supporting Information). Depending on the chain conformation, the three phosphate groups experience different intra- and intermolecular interactions, due to the different local environment, and this strongly influences their vibrational frequencies. As a consequence, the spectra of individual conformers (Figure 3A) are significantly different from each other (dashed lines in Figure 3B and extended results in the Supporting Information). The final spectrum is obtained as a linear combination of single conformer contributions (black line in Figure 3B) by fitting the experimental line shape (red line in Figure 3B). Although the spectra of single conformers seem to be quite far from the experimental spectrum, the fitted spectrum reproduces the main experimental features. This points to the important role of the conformational distribution, a behavior in line with the pronounced inhomogeneous broadening present in the 2D-IR spectrum (Figure 2B). Higher weights in the fitting function are found to correspond to triphosphate chains with larger distances between the negatively charged oxygens and with a higher number of water molecules bridging close O–O pairs (details in the Supporting Informatioon). In particular, conformer 3, with an extended geometry (Figure S2C), i.e., close O–O pairs, and without waters bridging neighboring oxygens on average, does not contribute at all.

Figure 3.

(A) Snapshot of conformers 1, 2, 4, and 5 of MTP (red for oxygen, brown for phosphorus, light blue for carbon, and white for hydrogen) in water (blue lines in the background) from the AIMD simulation. Labels of phosphate groups shown for MTP 1. (B) Experimental IR spectrum of a 0.1 M water solution of ATP (red) and simulated spectrum of MTP (black), as a function of frequency. Spectra of single MTP conformers (dashed), with intensities weighted by their specific contribution to the calculated spectrum (1, green, 12.97%; 2, violet, 13.75%; 4, light blue, 39.83%; 5, ocra, 33.45%; root-mean-square deviation of 0.0007). The shaded areas correspond to and (yellow), (light blue), and (green).

Deeper insight into the origin of the spectral features was reached by the VDOS analysis, whose results for two MTP conformers are shown in Figure 4 (extended analysis in the Supporting Information). Each line in the plots is calculated from the ACF of the velocities of the three (four) atoms in a PO₂ (PO₃) moiety. For a single PO₂ group, two bands are expected, one at higher frequency for the asymmetric stretching (ν_A) and one at lower frequency for the symmetric one (ν_S).¹² In panels A and B of Figure 4, the presence, instead, of four bands in the same frequency range for each PO₂ group indicates that vibrations are delocalized over the two groups. The two bands at a lower frequency result from the combination of the symmetric (PO₂)_α and (PO₂)_β vibrations, , whereas those at a higher frequency result from the combination of the asymmetric vibrations, . A difference is noticed between the symmetric and asymmetric bands. While the two symmetric bands are separated by an average splitting of 36 cm^–1, which does not strongly depend on the conformation, the splitting between the asymmetric bands is generally smaller and exhibits important changes from one conformer to another. Analysis of the trajectories shows that the splitting in the asymmetric bands is correlated to the relative hydration level of the two phosphate groups, which changes with the chain conformation. In particular, the splitting increases with the level of hydration of (PO₂)_α relative to (PO₂)_β (Figure S15). The ν_A bands of the (PO₃)_γ group overlap with the (ν_S)_αβ band, in agreement with previous assignments.²²⁻²⁴ The larger frequency splitting within the ν_S band, compared to the splitting within the ν_A band, clearly appears in both the band widths of the linear IR spectrum (Figure 1A) and the diagonal cut of the 2D-IR spectrum (Figure 2A, black line).

Figure 4.

VDOS of (PO₂)_α (red), (PO₂)_β (blue), and (PO₃)_γ (green) for conformers (A) 1 and (B) 4 of MTP (see the Supporting Information for details about the calculation of the splittings).

Changes in MTP–Zn²⁺ coordination are characterized by high energy barriers, and even in classical molecular dynamics simulations, enhanced sampling techniques are needed to explore all possible geometries.^43,44 To investigate the effect of Zn²⁺ binding on the IR spectra of ATP, we performed AIMD simulations of one MTP–Zn²⁺ complex in 122 water molecules, starting from complexes in different coordination modes: αβγ-tridentate, αγ-and βγ-bidentate, and γ-monodentate (Figure 5A). For each of them, different triphosphate conformations were considered [three for αβγ, four for αγ, three for βγ, and two for γ (details and extended analysis in the Supporting Information)]. Different spectra were obtained, for not only different coordination modes but also for the various conformations of a given coordination (dashed lines in Figure 5B). Similar to unbound MTP, the main features of the experimental spectrum could be recovered by fitting with a linear combination of the spectra calculated for single structures. Only bidentate and tridentate structures are found to contribute to the total spectrum: βγ and αγ by ∼50% and ∼10%, respectively, and αβγ by ∼40% (Figure 5).

Figure 5.

(A) Snapshot of αβγ, αγ, βγ, and γ coordination modes of the MTP–Zn²⁺ complex in water from MD simulations (MTP and water as in Figure 3A, Zn²⁺ in green). (B) Experimental IR spectrum of a 0.1 M water solution of the ATP–Zn²⁺ complex (red) and calculated spectrum of the MTP–Zn²⁺ complex (black). Spectra of single conformers of MTP–Zn²⁺ complexes in different coordination modes (dashed), weighted according to their contribution to the calculated spectrum (αβγ₁, lime, 28.88%; αβγ₃, olive green, 10.00%; αγ₃, pink, 11.35%; βγ₁, light blue, 20.71%; βγ₂, blue, 19.95%; βγ₃, cyan, 9.11%; root-mean-square deviation of 0.0003). Shaded areas colored as in Figure 3B.

The flexible coordination chemistry of Zn²⁺ allows it to adapt to different coordination geometries. The metal is pentacoordinated in αβγ and αγ complexes, where the coordination shell is completed by two and three water molecules, respectively. In βγ complexes, instead an octahedral geometry is always adopted, which includes four water molecules. The conformations of the most strongly contributing complexes are characterized by the P_α–O–P_β–O and O–P_β–O–P_γ torsional angles falling in the gauche+, gauche– domain, as generally found in crystals of ATP–divalent metal complexes (Table 3 of ref (45)). The VDOS analysis shows that the average splitting of (ν_S)_αβ is nearly unchanged upon Zn²⁺ binding, whereas the average splitting of (ν_A)_αβ increases from 17 to 31 cm^–1. This larger VDOS splitting is associated with the appearance of a two-band substructure in the (ν_A)_αβ band of the computed spectrum and to the blue-shift with respect to the unbound MTP spectrum. The extent of frequency splitting of the (ν_A)_αβ band is related to the difference in the local chemical environment of the two phosphate groups in the presence of the metal (Figure S15). The linear and the 2D-IR spectra display the change in the frequency splitting of the ν_A bands (Figure 1) and related 2D-IR diagonal peaks (Figure 2A, blue line). Moreover, there is a pronounced blue-shifted component at ∼1170 cm^–1 in the 2D-IR spectrum of ATP–Zn²⁺ complexes (Figure 2A,C), which we assign to a (PO₃)_γ vibration.

To gain insight into the electronic effect of binding to Zn²⁺, we calculated the electron density map of MTP in the presence and absence of coordinated metal. The difference map (Figure S14) shows that the metal subtracts electron density from the P–O bond, thus weakening it. This is consistent with the results of additional simulations we performed for DMP and the DMP–Zn²⁺ complex in water. The PO₂ VDOS exhibits a red-shift of the asymmetric band upon metal binding (see the Supporting Information), which is in agreement with experimental findings for the diethyl phosphate–Zn²⁺ complex.⁴⁶ We conclude that the blue-shift of the rightmost IR band of ATP, experimentally observed upon Zn²⁺ binding, does not have a simple and direct connection with an electronic effect on a single phosphate, but it results from a change in the frequency splitting in delocalized (ν_A)_αβ vibrations, which is induced by the metal and tuned by conformational and environmental degrees of freedom. A blue-shift of the asymmetric stretching band was also observed in the IR spectra of ATP–Mg²⁺²² and GTP–Mg²⁺ complexes.²³ However, a different interpretation, in terms of vibrations localized on single phosphate groups, was given, which was supported by QM/MM simulations and harmonic normal-mode analysis (NMA) in ref (24). Interestingly, binding to Mg²⁺ causes an upshift¹⁸ of the asymmetric stretching vibration of DMP, which is the opposite of the effect experimentally observed and predicted for Zn²⁺ (see the Supporting Information). The two metals have the same charge and very close ionic radii but different electronic properties, Mg²⁺ being generally classified as a hard metal and Zn²⁺ as a soft metal.

In conclusion, the integration of IR spectroscopy and AIMD simulations is found to be a viable strategy for the structural characterization of ATP and its complexes with metal ions. A multifaceted scenario emerges in which electronic effects are superimposed on conformational fluctuations and on couplings with the local chemical environment, which cannot be easily grasped by static approaches, such as harmonic NMA at zero temperature. We have explained the origin of the distinct spectral effects of binding of Zn²⁺ to ATP and to single monophosphates, and we have identified the βγ and αβγ modes as the preferential coordination modes of the ATP–Zn²⁺ complex. These are the same coordination proposed in the literature for the ATP–Mg²⁺ complex²² considering the effect of the metal on the frequencies of asymmetric stretching vibrations localized on individual phosphate groups. In the future, the extension of the IR–AIMD strategy to ATP–Mg²⁺ complexes would help to identify the microscopic mechanisms behind the analogies with and the differences from the ATP–Zn²⁺ complex.

Supporting Information Available

The Supporting Information is available free of charge at https://pubs.acs.org/doi/10.1021/acs.jpclett.4c02296.

• Linear and two-dimensional infrared spectroscopy, computational methods, computational results for MTP and for the MTP–Zn²⁺ complex, AIMD simulations of dimethyl phosphate (DMP) and its complexes with Zn²⁺ and Mg²⁺, Figures S1–S17, and Tables S1–S11 (PDF)

The authors declare no competing financial interest.