Article

pubs.acs.org/IC

Predicting the Stability Constants of Metal-Ion Complexes from First

Principles
Ondrej Gutten and Lubomír Rulíšek*
Institute of Organic Chemistry and Biochemistry, Gilead Sciences Research Center & IOCB, Academy of Sciences of the Czech
Republic, Flemingovo nám. 2, 166 10 Praha 6, Czech Republic
*
S Supporting Information

ABSTRACT: The most important experimental quantity

See https://pubs.acs.org/sharingguidelines for options on how to legitimately share published articles.

describing the thermodynamics of metal-ion binding with

various (in)organic ligands, or biomolecules, is the stability
constant of the complex (β). In principle, it can be calculated
as the free-energy change associated with the metal-ion
complexation, i.e., its uptake from the solution under standard
Downloaded via NEXEON on November 29, 2024 at 07:48:49 (UTC).

conditions. Because this process is associated with the inter-

actions of charged species, large values of interaction and
solvation energies are in general involved. Using the standard
thermodynamic cycle (in vacuo complexation and solvation/
desolvation of the reference state and of the resulting
complexes), one usually subtracts values of several hundreds
of kilocalories per mole to obtain final results on the order of
units or tens of kilocalories per mole. In this work, we use density functional theory and Møller−Plesset second-order
perturbation theory calculations together with the conductor-like screening model for realistic solvation to calculate the
stability constants of selected complexes[M(NH3)4]2+, [M(NH3)4(H2O)2]2+, [M(Imi)(H2O)5]2+, [M(H2O)3(His)]+,
[M(H2O)4(Cys)], [M(H2O)3(Cys)], [M(CH3COO)(H2O)3]+, [M(CH3COO)(H2O)5]+, [M(SCH2COO)2]2−with eight
divalent metal ions (Mn2+, Fe2+, Co2+, Ni2+, Cu2+, Zn2+, Cd2+, and Hg2+). Using the currently available computational protocols,
we show that it is possible to achieve a relative accuracy of 2−4 kcal·mol−1 (1−3 orders of magnitude in β). However, because
most of the computed values are affected by metal- and ligand-dependent systematic shifts, the accuracy of the “absolute”
(uncorrected) values is generally lower. For metal-dependent systematic shifts, we propose the specific values to be used for the
given metal ion and current protocol. At the same time, we argue that ligand-dependent shifts (which cannot be easily removed)
do not influence the metal-ion selectivity of the particular site, and therefore it can be computed to within 2 kcal·mol−1 average
accuracy. Finally, a critical discussion is presented that aims at potential caveats that one may encounter in theoretical predictions
of the stability constants and highlights the perspective that theoretical calculations may become both competitive and
complementary tools to experimental measurements.

1. INTRODUCTION applies a standard thermodynamic cycle consisting of the studied

Recent developments in both density functional theory (DFT) process in the gas phase (in this case, complexation of the ions
and ab initio wave function theory domains of computational with ligands) and (de)solvation of all of the species involved (i.e.,
chemistry, together with advances in the modeling of solvation of the complexed molecules vs free ligands and hydrated metal
effects,1,2 resulted in the situation that theoretical calculations ions).5 For ionic species, these elementary processes are usually
nowadays represent an integral part of many chemical and bio- associated with large energies of several hundreds of kilocalories
chemical studies.3 An appropriate selection and benchmarking of per mole (the gas-phase association of the ion···neutral or ion···
all components and methods necessary for accurate predictions ion species and their solvation/desolvation energies) that almost
of free-energy values for studied chemical processes, together cancel each other out to yield the final ΔG values of several
with a careful setup of the model system (which is not always kilocalories per mole (corresponding to dissociation or stability
trivial), leads to high-quality computational data that comple- constants in the typical millimolar to femtomolar range).
ment and rival the experimental counterparts. However, what seems to be a small difference from the com-
One of the challenges in computational (bio)chemistry is putational point of view and from the perspective of the large
related to the quantitative treatment of metal-ion coordination energy changes associated with the elementary processes is
in biomolecules, experimentally quantified by the stability a markedly large value in chemical and biological systems.
constant (β) that is the equilibrium constant for the formation
of a complex in solution.4 In order to calculate these observable Received: April 25, 2013
thermodynamic quantities from first principles, one usually Published: September 3, 2013

© 2013 American Chemical Society 10347 dx.doi.org/10.1021/ic401037x | Inorg. Chem. 2013, 52, 10347−10355
Inorganic Chemistry Article

Differences on the order of ∼5 kcal·mol−1 govern many is of limited accuracy, and it is always advisable to compare it with
fundamental phenomena, such as metal-ion selectivity in more accurate methods.32
biomolecules.6−10 Thus, a deeper theoretical understanding of The aim of this study is a careful analysis and inspection of all
the metal-ion uptake by (bio)molecules may help us to answer of the aforementioned caveats in the ab initio calculations of the
fundamental questions, such as why nature selected various metal stability constants performed on a much broader set of
ions for performing specific functions.11 Theoretical calculations complexes with experimentally determined stability constants,
can be viewed (and used) as a unique and complementary tool to using the set of eight biologically relevant divalent metal ions:
well-established experimental methods12−14 to correlate the Mn2+, Fe2+, Co2+, Ni2+, Cu2+, Zn2+, Cd2+, and Hg2+. Various
calculated or experimental thermodynamics with the structural details in adopted computational protocols are carefully analyzed
details.15−18 Once, and only once, a satisfactory agreement and discussed, and it is concluded that while the accuracy of
between the computed and experimental data is obtained, “absolute” values of binding free energies is still out of reach, the
decomposition of the total free-energy change, energy/structure average accuracy of relative affinities of 2 kcal·mol−1 might be, in
mapping, or analysis of the electronic structure of the studied principle, achievable.
system can be done and provides us with the insights and
concepts.19,20 2. METHODS
Many systematic efforts to quantitatively assess the selectivity 2.1. Computational Details. All calculations reported in this work
of metal binding by theoretical methods were described in the were performed using the TURBOMOLE 6.3 program. The quantum-
literature over the past 1.5 decades.21−29 Recently, we presented chemical calculations were mostly performed using DFT, while in a
a computational study5 in which we critically evaluated the few cases, the MP2 method has been employed. The geometry
performance of the ab initio and DFT electronic structure optimizations were carried out at the DFT level, employing the density-
methods in calculations of the energetics associated with metal- fitted (vide infra) BP86 functional (RI-BP86)33a,34 and the def-TZVP
ion complexation on a set of five model [MXn]c+ complexes basis set on all of the atoms.35,36 All gas-phase structures represent true
(M = Fe2+, Cu2+, Zn2+, and Cd2+; n = 2 and 4−6) spanning four minima, based on a frequency calculation. The single-point energies
were then calculated using the RI-BP86, RI-PBE,37 BH-LYP,33a−c and
common coordination geometries. Reference values for the gas- B3LYP33a−d,38 functionals or at the RI-MP2 level of theory. All DFT
phase complexation energies were obtained using the CCSD(T)/ calculations using nonhybrid functionals (as well as MP2 calculations)
aug-cc-pVTZ method and compared with cheaper methods, such were expedited by expanding the Coulomb integrals in an auxiliary basis
as DFT and RI-MP2. In the same study,5 we applied two popular set, using the resolution-of-identity (RI) approximation (density
and presumably accurate solvation methodsconductor-like fitting).39 A multipole-accelerated version of the RI algorithm was
screening model for realistic solvation (COSMO-RS) and used for MP2 calculations. In most of the cases, the def2-TZVP basis
universal solvation model employing the full solute density set35 was employed for all of the atoms, with two exceptions represented
(SMD)to find out whether the calculated free-energy (ΔG) by the RI-PBE and RI-BP86 calculations, where def-TZVP was used.
Grimme’s D3 dispersion correction was used for all DFT calculations.40
changes associated with metal-ion complexation in solution
For Cd and Hg, small-core Stuttgart/Dresden pseudopotentials were
match (or are at least in the range of) the experimental stability applied to account for scalar relativistic effects.41
constants. The computational data highlighted several intricacies All metal ions were considered in their 2+ oxidation state. In all cases,
in theoretical predictions of the stability constants that may result high-spin states are assumed (with a couple of exceptions discussed in
in errors of several tens of kilocalories per mole in the final ΔG the text). The choice is justified by calculating energies of low-spin
(−RT ln β) values: (a) the covalent character of some metal− alternatives of selected systems in which one may expect the low-spin
ligand bonds [e.g., copper(II) thiolate]; (b) various definitions states to be closer in energy (Table S9 presented in the Supporting
of the reference state of some systems (e.g., Jahn−Teller Information, SI). The exhaustive treatment of all of the spin-state
unstable [Cu(H2O)6]2+ vs [Cu(H2O)4]2+); (c) the presence of splittings for all of the studied complexes is beyond the scope of this
study; for iron(II) systems, the reader is referred, for example, to the
the negatively charged ligand in the metal coordination sphere. above-mentioned work of Droghetti et al.31
A similar approach has been evaluated in a recent study by Solvation (free) energies of all studied species were calculated using
Delgmann and Schenk.30 The investigation has confirmed that the COSMO-RS method,42,43 as implemented in the COSMOtherm
conventional solvation treatment by methods like the polariz- program,44 using the BP_TZVP_C30_1201.ctd parametrization file.
able continuum model (PCM) or COSMO is insufficient. The The geometries in the solvent (water) were optimized using the
application of more advanced methods (specifically COSMO-RS COSMO method,45 with COSMO radii of 2.0 Å for Mn−Zn, 2.2 Å for
examined therein) in combination with appropriate quantum- Cd, and 2.4 Å for Hg and εr = 80.0. Quite some effort has been exerted
chemical methods is essential to obtain good quantitative agree- to find the true minima; however, because of higher computational
demands, this was not achieved in all of the cases. Single-point
ment with experimental data. Although a number of problematic
calculations used for the preparation of COSMO-RS calculations were
points concerning COSMO-RS have been highlighted, in done according to the recommended protocol, which includes
combination with careful analysis it stands as a very powerful RI-BP86/def-TZVP calculations with Grimme’s D3 dispersion
tool for dealing with solvation effects. correction and uses ε = ∞ (ideal conductor) or 1 (vacuum) with the
The choice of a proper quantum-chemical method is very same radii as those described previously, vide supra. The Gibbs free
problematic. Although DFT is a popular choice, it is clear that energy (sometimes denoted free enthalpy) of a system with metal ion M
none of the current functionals can present a final and universal and set of ligands {Li} ≡ L (introduced in Figure 1) is defined as
answer for a wide range of properties. This is especially true for calc, μ
transition metals, which exhibit very diverse chemistry. For GM,L = Eel + Gsolv + EZPVE − RT ln(qtransqrotqvib) (1a)
example, local-density approximation and generalized-gradient
where Eel is the in vacuo energy of the system, Gsolv is the solvation free
approximation functionals overstabilize low-spin states, although energy, EZPVE is the zero-point vibrational energy, and −RT ln(qtrans qrot
they can perform reasonably well in describing certain properties, qvib) accounts for the entropic terms and the thermal correction to the
e.g., geometries.31 BP86 is of special interest to this study because enthalpy, obtainable from a frequency calculation and utilizing the ideal-
of its involvement in the COSMO-RS protocol. Although it is gas approximation (T = 298 K and p = 2.48 MPa, which correspond to
considered to have a decent price/performance ratio, energetics 1 M concentration).46 As is discussed in more detail below, these latter

10348 dx.doi.org/10.1021/ic401037x | Inorg. Chem. 2013, 52, 10347−10355

Inorganic Chemistry Article

Figure 1. Model complexes: (a) [M(NH3)4]2+; (b) [M(NH3)4(H2O)2]2+; (c) [M(Imi)(H2O)5]2+; (d) [M(His)(H2O)3]+; (e) [M(His)(H2O)4]+; (f)
[M(Cys)(H2O)4]; (g) [M(Cys)(H2O)3]; (h) [M(CH3COO)(H2O)3]+; (i) [M(CH3COO)(H2O)5]+; (j) [M(SCH2COO)2]2−.

two terms may sometimes present a nontrivial technical challenge. For some of the systems, we try two different values for this variable,
Therefore, we preferred to use the following equation as the practical which gives rise to the 10 systems that were studied, as depicted in
(albeit not theoretically pure) solution and approximation to the free Figure 1.
energy of the complex, using only a single, COSMO-optimized Including more H2O molecules than are needed to saturate the first
geometry: coordination sphere (i.e., filling the second coordination sphere) might
calc raise a concern about artificial hydrogen bonds being formed. This issue
GM,L = Eel + Gsolv (1b) is very difficult to address because the exact hydrogen-bonding network
Throughout the work, the values of Gcalc around the complex is not known. Nevertheless, we carried out model
M,L (based on eq 1b) and Eel
obtained via the RI-BP86 method are reported, whereas Eel obtained via calculations for the cysteinate complex (MD sampling). By comparing
the RI-MP2 method and the values of Gcalc,μ its complexes with 18 and 4 H2O molecules, we observed that similar
M,L (based on eq 1a) can be
found in the SI (Tables S1−S6). hydrogen-bonding patterns are present. Probably the weakest cor-
Experimental stability constants, βM,L, were converted to differences respondence has been found for HgII. However, it is not clear whether
in the Gibbs free energy, using the formula 18 H2O molecules are sufficient to prevent potential artifacts or whether
a missing hydrogen-bonding pattern is due to insufficient sampling
exp [MLn]θ (10 systems with 18 H2O molecules for each metal ion). These
ΔGM,L = − RT ln KM,L ≈ − RT ln βM,L = − RT ln structures can be found in the SI.
[M]θ [L]θ n
The concentration constants for most (often for all) metal ions in the
(2)
studied series were available and cover a wide range of values with
where K stands for the thermodynamical stability constant, β for the L /βL ) = 1−35, where βL and βL are the maximum (mostly
log(βmax min max min
experimental (concentration) stability constant, and [X]θ for the in the Hg2+ complex) and minimum (mostly Mn2+) values of the
standard concentration. concentration constants for a given set of ligands L.
This is an approximation because the experimental values were obtained
at generally nonzero ionic strengths and the measured concentration
constants differ somewhat from the thermodynamical stability constants. 3. RESULTS AND DISCUSSION
The difference between the two, i.e., the dependence of the concentration 3.1. Experimental Free Energies of Complexation. In
constants on the ionic strength, was generally rather low in comparison to Table 1, we summarized the available experimental information
errors in the calculated values. Rigorously, the values could be extrapolated
to I = 0, for example, with the help of Debye−Hückel theory to estimate the on free-energy changes associated with the complexation of
activity coefficients (or some of its extension, such as the Davies equation) metal ions in the studied systems. The values of complexation
using at least three experimentally determined values at varying ionic free energies for a given metal M and set of ligands L, ΔGexp M,L, are
strength. Such data are not available in all cases, whereas in some other derived from the experimental values of log βM,L (via eq 2)
cases, multiple values originating in different sources can be found, and this obtained from Martell’s tables.47 We may observe the known
questions the justifiability of such extrapolations to thermodynamical general trends for 3d metal ions, conforming to Irving−Williams
stability constants. Therefore, we prefer to use the experimental values series of stability constants with Ni2+ and Cu2+ as the strongest
pertinent to specific conditions in our comparison with the calculated binders, whereas the order of the other metal ions does vary
values, thus assuming the activity coefficient to be equal to 1.
In silico, ΔGcalc somewhat (mainly Co, Zn, and Cd). However, the magnitude
M,L was calculated as the difference of the Gibbs free
energies, Gcalc
M,L, as defined in eq 1, of the products and reactants in the
of these differences, |ΔGexp,max
L − ΔGexp,min
L |(Mn2+↔Cd2+), varies
general complexation reaction: significantly among the studied systems (from 1.1 kcal·mol−1 for
the [M(CH3COO)]+ system to 15 kcal·mol−1 for [M(NH3)4]2+).
[M(H 2O)6 ]2 + + {n L}c → [MLn(H 2O)m ]c + 2 + (6 − m)H 2O These magnitudes are even more pronounced if we include Hg2+
(3) in the series, which has in all cases the highest binding affinity.
where c is the total formal charge of the ligands. The numbers of ligands, 3.2. Calculated Values of the Free Energies of Metal-
n, and H2O molecules, m, are specified in the corresponding tables. Ion Complexation. In Table 2, we summarize the calculated
2.2. Model Systems. The set of model systems comprised values of complexation free energies, ΔGcalc
M,L, calculated according
six complexes ([M(NH3)4]2+, [M(Imi)]2+, [M(His)]+, [M(Cys)], to eqs 1b and 3 using RI-BP86 for gas-phase electronic energies
[M(CH3COO)]+, and [M(SCH2COO)2]2−). These systems are simple
enough to avoid serious problems with the correct geometry description
(the corresponding values of ΔGcalc M,L obtained using RI-MP2 for
with only a few binding modes to be tested. We have no structural the gas-phase electronic energies and ΔGcalc,μ M,L obtained using
information about the model systems apart from the number of ligands eqs 1a and 3 using various methodsRI-BP86, RI-PBE, B3-LYP,
and their protonation state. This leaves an open question of how BH-LYP, and RI-MP2are listed in Tables S6 and S1−S5 in the
many H2O molecules should be explicitly included in the calculation. SI, respectively). Representative equilibrium geometries of the
10349 dx.doi.org/10.1021/ic401037x | Inorg. Chem. 2013, 52, 10347−10355
Inorganic Chemistry Article

Table 1. Estimated Free Energies of Complexation (in kcal·mol−1), ΔGexp

M,L, of the Studied Metal Ions As Calculated from
Experimental Stability Constantsa
−1
ΔGexp
M,L [kcal·mol ]
b
complex I Mn Fe Co Ni Cu Zn Cd Hg
[M(NH3)4]2+ 2 −2.3 −4.5 −7.6 −11.1 −17.6 −13.2 −10.1 −26.1
[M(Imi)]2+ 0.1 −1.7 −3.3 −4.1 −5.7 −3.5 −3.7 −12.5h
[M(His)]+ 0.1 −4.5 −8.0c −9.4 −11.8 −13.9 −8.9 −7.7
[M(Cys)] 0.1 −6.5 −9.0d −11.1 −13.4 −12.4 −13.8g −19.7
[M(CH3COO)]+ 0 −1.9 −1.9 −1.9 −1.9 −3.0 −2.2 −2.6 −5.9
[M(SCH2COO)2]2− 0.1 −10.3 −14.9e −16.6 −17.8f −20.5 −59.8i
a
It is assumed that these represent thermodynamical equilibrium constants. Unless stated otherwise, the values are for T = 298.15 K. bIonic strength
in mol·dm−3. cI = 3. dT = 293.15 K. eI = 0. fT = 293.15 K. gT = 310.15 K; I = 0.15. hI = 3. iI = 1.
−1 a
Table 2. Calculated Values of Complexation Free Energies, ΔGcalc
M,L (kcal·mol )

metal ion
system Mn Fe Co Ni Cu Zn Cd Hg
[M(NH3)4]2+ 13.1 11.0 2.2 −1.8j −8.0 −0.6 1.4 −11.6
[M(NH3)4(H2O)2]2+ −8.4b −11.3b −14.4b −17.9b −28.5c −14.9d −17.3d −27.7d
−8.1b −11.3b −14.6b −17.3b −24.8b −12.1b −14.2b −29.9d
[M(Imi)(H2O)5]2+ −3.7 −4.6k −5.8 −6.1 −6.4 −4.4 −4.0 −11.2
[M(H2O)3(His)]+ 5.6 4.3 0.0 −0.9 −5.1 3.9 4.0 −5.2k
[M(H2O)4(His)]+ −11.8e −12.9e −11.0e −17.9e −23.9e −13.4e −13.2e −21.4k
−13.0f −14.8f −15.8f −19.6f −20.5f −14.4f −14.4f −20.6k
[M(H2O)3(Cys)] 13.8 12.1 6.4 9.1 −1.6 12.9 6.4 −13.9
[M(H2O)4(Cys)] 4.1g 1.4g 0.6g −0.4g −14.4g,k 1.9g −4.3g −24.0g,j
1.3e 1.7e −2.9e −0.6e −4.5e,k −2.6g −6.9g −22.1g,j
1.3e 2.9e −4.9e −3.3e −12.4e,k 0.4e −6.7g −22.6j
6.1g 3.8g −0.3g 3.5g −12.5g,k 1.6g −2.1g −22.9j
[M(CH3COO)(H2O)3]+ 30.1 31.9 27.5 35.8 25.0 28.7 27.3 22.7
[M(CH3COO)(H2O)5]+ 3.7h 4.5h 2.8h 5.0h 0.0h 4.5h 2.8h −1.0h
4.6i 6.1i 6.4i 7.7i −0.2i 7.4i 5.1i 0.3i
[M(SCH2COO)2]2− 35.6 31.0 25.7 40.7j 8.9k 28.0 21.5k −8.0
a
Calculated using the RI-BP86/COSMO-RS protocol utilizing eqs 1b and 3. bOctahedral with H2O molecules in mutual cis (upper row) or trans
(lower row) position. cSquare pyramidal with H2O in the axial position; the other H2O is in the second sphere. dTetrahedral; H2O molecules in the
second sphere. eTridentate binding mode. f(N,NImi) bidentate binding mode. g(N,S) bidentate binding mode. hSyn binding mode of acetate. iAnti
binding mode of acetate. jWe think that these values of ΔGcalc M,L might be incorrect and are excluded from further analysis for various reasons: (i)
[Ni(NH3)4]2+ and [Ni(SCH2COO)2]2− systems in square-planar geometry are low-spin complexes, as opposed to all other systems.
[Co(SCH2COO)2]2− is potentially low-spin as well but is included in the analysis and assumed to have a high-spin ground state; (ii) the equilibrium
geometry of [Hg(H2O)m(Cys)] is entirely different from the geometries of other systems because of the preference of Hg for linear geometry.
Although this trait is not unique to these systems, it still makes a direct comparison using the current protocol problematic. kExperimental stability
constants are not available for these complexes.

studied complexes are depicted in Figure 2, whereas the “noise”, and the ensemble is dominated by contribution from a
complete set of all equilibrium geometries of studied complexes single conformer. On the other hand, all conformers in a single
is deposited in the SI. coordination sphere ([M(H2O)3(Cys)] system) are very similar,
In an ideal case, each system would be represented by an and the weights, as well as contributions, are almost identical.
ensemble of structures. However, a satisfactory sampling would For these reasons, we opt to represent the systems by a single
require tens, and possibly hundreds, of conformations with at structure, bearing fully in mind that systems with nonnegligible
least two coordination spheres, which reaches way beyond differences in conformational entropy will suffer systematic
affordable limits. Dealing with less sizable statistics is not mistreatment. However, we do not expect this to be an issue for
reliable and, because of “steep” Boltzmann weights, is likely to be our chosen set of simple ligands. For some complexes, multiple
dominated by the most negative (i.e., the most stable) entry. We conformations were considered as potential candidates. How-
illustrate this by comparing Boltzmann-weighted averages of the ever, even a single binding mode is not easy to characterize by
overall free-energy change (i.e., calculated stability constant) for a single value of ΔGcalc
M,L (as can be demonstrated by a range of
one of the systemsthe [M(H2O)x(Cys)] complexwith the values obtained for, e.g., [M(H2O)4(Cys)] system; Table 2).
same value obtained using the protocol utilized in this work This is, in part, due to the lack of implementation of structure
(i.e., considering only one optimized structure). An ensemble of optimization with respect to our definition of Gcalc
M,L (eqs 1a and 1b),
10 structures with two (Table S7 in the SI) or one (Table S8 in which includes two largely opposing terms (COSMO energy and a
the SI) coordination sphere of H2O molecules has been used COSMO-RS correction). Only optimization with respect to the
in this comparison. Providing a larger bulk of water (18 H2O COSMO energy was available, whereas a rigorous optimization
molecules in [M(H2O)3(Cys)]·15H2O system) introduces large may provide different preferences of binding modes for more
10350 dx.doi.org/10.1021/ic401037x | Inorg. Chem. 2013, 52, 10347−10355
Inorganic Chemistry Article

Figure 2. Representative equilibrium geometries for selected complexes studied in this work: (a) [Mn(His)(H2O)4]+, (N,N,O) binding mode, one H2O
in the second coordination sphere; (b) [Mn(His)(H2O)4]+, (N,N) binding mode; (c) [Mn(Cys)(H2O)4], (N,S,O) binding mode, one H2O in the
second coordination sphere; (d) [Mn(Cys)(H2O)3], (N,S) binding mode, trigonal-bipyramidal geometry, one H2O in the second coordination sphere;
(e) [Mn(SCH2COO)2]2−.

complex systems. Even if the global minimum with respect to available. For a given method and a given set of ligands L, there
M,L could be found, it may not necessarily represent the “true”
Gcalc is a single value of LSSL and it represents how far, on average,
structure of the real complex in solution. The reason for this is the calculated result is from the experimental one, i.e., an average
discussed in detail in section 3.3. In subsequent sections, the error. The values of LSSL are listed in Tables 4 (last column) and
conformations with the lowest Gcalc
M,L are used as representatives of a S1−S6 in the SI.
given system. The ligand-specific shifts, LSSL, are significant, ranging from
The values of ΔGcalcM,L for some of the systems([M(NH3)4- less than +8 to −50 kcal·mol−1. The somewhat good cor-
(H2O)2]2+, [M(Imi)(H2O)5]2+, [M(H2O)4(His)]+, and [M- respondence between the experimental and calculated values of
(CH3COO)(H2O)5]+)are in the vicinity of their experimental free energies of complexation for the four systems ([M(NH3)4-
counterparts, but even for these systems, the error commonly (H2O)2]2+, [M(Imi)(H2O)5]2+, [M(H2O)4(His)]+, and [M-
exceeds 6 kcal·mol−1 in either direction which is not satisfactory. (CH3COO)(H2O)5]+) mentioned in section 3.2 is, in part, due
For other systems, the deviations between theory and experi- to lower values of their ligand-specific shifts (+6.3, +1.0, +7.4,
ment are greater still. and −5.4 kcal·mol−1, respectively). LSSL consists of two major
It turns out that compelling information can be unveiled if we contributions. One originates in the ZPVE − RT ln Q term that is
examine the differences between the experimental and calculated neglected in eq 1b (included in eq 1a) and is especially notable
values, ΔΔGM,L, defined as in cases where there is a change in the number of species upon
exp
ΔΔGM,L = (ΔGM,L calc
− ΔGM,L ) complexation (e.g., [M(NH3)4]2+, [M(CH3COO)(H2O)3]+,
(4)
[M(H2O)3(Cys)], and [M(SCH 2COO)2]2−). Comparing
where ΔGexp
M,L is defined by eq 2 and listed in Table 1 and ΔGM,L is
calc Table 2 (based on eq 1b) with Table S1 in the SI (values obtained
obtained from calculation via eqs 1b and 3 and summarized in by using eq 1a) shows that if this contribution is included, the
Table 2. Next, we show that a large part of this discrepancy can magnitude of the overall error significantly decreases.
be identified and qualitatively predicted. To this end, we split The ZPVE − RT ln Q term is commonly estimated by invoking
ΔΔGM,L into two contributions: the ligand-specific shift, LSSL, the ideal gas and rigid-rotor/harmonic-oscillator approximation,
and the metal-specific shift, MSSM,L. which requires a gas-phase structure of the system that needs to
3.3. Analysis of Ligand-Specific Shifts. The first of these be obtained in addition to the COSMO-optimized structure.
contributions, the ligand-specific shift, is constructed as an average However, this may introduce an error that increases the more the
of all values of ΔΔGM,L for a given set of ligands, L, and denoted two structures diverge from each other. An alternative approach
as LSSL: is to utilize these approximations for a COSMO-optimized structure.
def However, besides being accompanied by technical complications,
ΔΔGM,L it is not rigorous48 and, again, introduces a systematic error that is
LSSL = ∑ difficult to control.
M = {Mn 2 +,...,Hg 2 +}
n (5)
The other source of error is a certain bias of the adopted
“def” indicates that the only allowed values of M are those protocol for solvating the charged ligands in their unbound and
for which ΔΔGM,L is defined; i.e., both ΔGexp
M,L and ΔGM,L are
calc
bound states. The magnitude of this error largely corresponds to
10351 dx.doi.org/10.1021/ic401037x | Inorg. Chem. 2013, 52, 10347−10355
Inorganic Chemistry Article

Table 3. Contribution of Protocol Bias to Ligand-Specific Shifts, LSSL (kcal·mol−1)

acid−conjugate base ΔGexp ΔGcalc ΔΔGexp/calc complexation LSSLa
NH3 → NH4+ −12.6 −17.3 4.7 [M(H2O)6] + 4NH3 → [M(NH3)4(H2O)2] + 2H2O
2+ 2+
2.8
Imi → ImiH
+
−9.5 −12.6 3.1 [M(H2O)6]2+ + Imi → [M(Imi)(H2O)5]2+ + H2O 0.3
His− → HisH22+ −23.0 27.1 4.1 [M(H2O)6]2+ + His− → [M(His)(H2O)3]2+ + 3H2O 3.0
Cys2− → Cys+ −27.8 −23.2 −4.6 [M(H2O)6]2+ + Cys2− → [M(Cys)(H2O)3] + 3H2O −8.0
CH3COO− → CH3COOH −6.5 −4.6 −1.9 [M(H2O)6]2+ + Ac− → [M(Ac)4(H2O)5]+ + H2O −8.6
SCH2COO2− → HSCH2COOH −19.3 −9.7 −9.6 [M(H2O)6]2+ + 2tg2− → [M(tg)2]2− + 6H2Ob −13.5
a
Calculated using the RI-BP86/COSMO-RS protocol based on eqs 1a and 3; i.e., the terms ZPVE and RT ln Q are included. Full data, from which
the LSSL value has been obtained, can be found in Table S1 in the SI. btg stands for the thioglycolate ligand, (SCH2COO)2−.

the type of group (carboxylic acid, thiolate, amine nitrogen, etc.). metal ions are all available, they certainly are not available for the
Apparently, complexation energies for acetate, thioglycolate, resulting complex.
and cysteine are not negative enough (even if the ZPVE − RT ln 3.4. Analysis of Metal-Specific Shifts. If one is, however,
Q term is included), implying too favorable solvation of the focused on the selectivity of a particular site for a given metal ion,
unbound ligands. For the other systems (ammonia, imidazole, the ligand-specific shifts, LSSL, can be disregarded because they
and histidine), the values are, on the contrary, too negative, do not affect the relative affinities of a series of metal ions for the
indicative of overstabilization of bound imidazole and amine particular site. We define the metal-specific shifts as
N atoms over neutral unbound states. In the case of histidine, MSSM,L = ΔΔGM,L − LSSL (6)
which experiences the opposing effects of carboxyl O and N
atoms, the latter prevails. In other words, it is a measure of how the predicted values
Because the complexation process can be viewed as the reverse deviate from the actual relative affinities. In order to quantify how
of proton exchange with H2O molecules (i.e., a proton rather systematic the shifts are, we use two types of root mean squares
than a metal ion acts as the Lewis acid), we tried to justify the (RMSs) of these RMSM,L values: denoted as RMSL and RMSM.
validity of the above hypothesis by calculating the pKa values (or, The former, RMSL, is calculated from the values of RMSM,L for
more precisely, ΔGprotonation values, which are also applicable in one specific set of ligands L and all possible metal ions:
the case of multiple protonation sites) of studied ligands.
As can be seen from the results presented in Table 3, the
def
(MSSM,L )2
ligand-dependent shifts, LSSL, qualitatively mimic the deviations
RMSL = ∑
M = {Mn 2 +,...,Hg 2 +}
n
between theoretically predicted and experimental pKa’s (i.e., (7)
ΔΔGexp/calc), and they may serve as the “zero-order” estimates of A low value of RMSL implies that the relative affinities of metal
the expected LSSL values. It must be emphasized that this ions for a given set of ligands L are reproduced well. The latter,
correlation does not include the number of ligands and uses only RMSM, is analogously defined for one specific metal ion and all
an average of values (viz. definition of LSSL in eq 5) over a series possible sets of ligands L:
of metal ions. It is, by no means, meant to be quantitative, but
it does demonstrate the major contributions to LSSL, i.e., the def
(MSSM,L )2
protocol bias and the ZPVE − RT ln Q term. RMSM = ∑
L = {[M(NH3)4 ]2 + ,...,[M(SCH 2COO)2 ]2 ‐ }
n
The nonnegligible magnitudes of these systematic deviations
(Table 3) warn that extra care needs to be taken when comparing (8)
two conformations for which the difference between their A low value of RMSM implies that the affinity of a given metal ion
M,L values is in the direction of the “protocol bias”.
respective ΔGcalc M is reproduced with a similar error for various systems.
For example, in complexes with histidine, the (N,NImi) binding First, we focus our attention on the values of RMSL in Table 4.
mode seems to be preferred over the tridentate mode, but Admittedly, some of the systems ([M(Imi)]2+ and [M(CH3COO)]+)
whether this might be due to the bias of the protocol used or it is have a smaller range of binding free energies (cf., ΔGexp M,L in
indeed the preferred binding mode found in solution is difficult Table 1), which probably also contributes to the lower variance of
to conclude unambiguously. On the other hand, for cysteine, the MSSM,L and, hence, lower RMSL. However, this is not the sole
(N,S) mode is always preferred over the (N,O) mode (data not reason for the lower values of RMSL because these also remain
shown) in spite of the understabilization of the bound thiolate quite low for more selective systems ([M(Cys)] and [M(His)]+).
group, which leaves more confidence in concluding that this is Additionally, a large part of RMSL is often due to one or two
indeed the preferred binding mode. outlying values, while the other values are much smaller.
The description and analysis of both components of LSSL Systems that differ in the number of water ligands do possess
remain largely qualitative. Still, we feel that comprehending some similarity in the values of individual metal-specific shifts
the nature of LSSL (or, in general, understanding systematic but are not entirely equivalent in this respect. While this can be
deviations in protocols used for calculations of solvation Gibbs indicative of one of the systems being a better representation of
free energies) is important because it has significant implications an actual species in solvent, it has to be borne in mind that the
in various calculations of the thermodynamic properties of search for local minima is not consistent across the metal-ion
molecules [stability constants, reduction potentials, or acidity series, and this can easily be a more influential factor than the
constants (pKa’s)]. geometry of ligands around a metal ion.
A better quantitative insight into the LSSL values might be The values in Table 2 were calculated using RI-BP86 and
obtained by comparing the calculated solvation energies with eqs 1b and 3, although in our previous work,5 we mildly advocated
experimental data. This approach, however, is only partially for use of the RI-MP2 method for calculating the gas-phase
applicable because even if the solvation values for ligands and interaction (complexation) energies for the complexes of divalent
10352 dx.doi.org/10.1021/ic401037x | Inorg. Chem. 2013, 52, 10347−10355
Inorganic Chemistry Article

Table 4. Calculated Values of Metal-Specific Shifts and Related Statistics (kcal·mol−1)a

MSSM,L
system Mn Fe Co Ni Cu Zn Cd Hg RMSL LSSL
[M(NH3)4] 2+
−2.8 −2.8 2.9 excl. b
3.1 0.1 1.2 −1.8 2.5 −12.7
[M(NH3)4(H2O)2]2+ −0.3 0.6 0.7 0.5 4.6 −4.5 0.9 −2.5 2.7 6.3
[M(Imi)(H2O)5]2+ 1.0 1.5 1.0 −0.3 −0.1 −0.7 −2.3 1.3 1.0
[M(H2O)3(His)]+ 0.8 −1.4 1.5 −0.1 2.0 −1.9 −0.9 1.5 −10.9
[M(H2O)4(His)]+ 1.1 −0.6 −1.0 0.5 2.7 −1.9 −0.7 1.5 7.4
[M(H2O)3(Cys)] 0.9 0.0 3.6 −1.3 −4.2 1.0 excl.b 2.6 −21.1
[M(H2O)4(Cys)] 0.8 −1.9 2.3 −1.5 −1.3 1.6 excl.b 1.8 −8.5
[M(CH3COO)(H2O)3]+ −0.7 −2.5 1.9 −6.4 3.3 0.4 1.3 2.7 3.2 −31.3
[M(CH3COO)(H2O)5]+ −0.2 −0.9 0.8 −1.5 2.4 −1.2 0.0 0.5 1.3 −5.4
[M(SCH2COO)2]2− 1.5 1.3 4.9 excl.b −1.4 −4.7 3.6 −46.9
MSSM,avgc 0.2 −0.9 1.9 −1.1 2.5 −1.6 0.4 −1.3
RMSM 1.3 1.4 1.6 2.3 1.5 1.7 1.0 2.6
a
The protocol used was RI-BP86/COSMO-RS. Only the most negative value of ΔGcalc M,L for each of the systems is listed, and only those for which
experimental data are available are used for calculation of the LSS and RMS quantities. The empty fields are due to missing experimental data.
b
“excl.” denotes results that were excluded from analysis; Table 2. cArithmetic average of MSSM,L values over all systems (over all values of “L”).

metal ions and ligands with N, S, and O donor atoms [considering computational cost, it has another valuable property, as discussed
the CCSD(T) calculations as the reference]. In contrast, the in the following paragraph. Similar conclusions concerning a
RI-BP86 functional belonged to the least satisfying methods good price/performance ratio for BP86 for transition-metal
investigated. Surprisingly, the composite RI-MP2/COSMO-RS complexes were also formulated by Furche and Perdew.32
protocol (i.e., RI-MP2 values used for the gas-phase complexation 3.5. Elimination of Bias for Individual Metal Ions. We
energies and the RI-BP86/def-TZVP method used for calculation turn our attention to the values of RMSM presented in Table 4
of the solvation energies within the COSMO-RS framework) is (i.e., root mean square of metal-specific shifts for a given metal, as
comparable to the presented RI-BP86/COSMO-RS values (data defined by eq 8). These can be loosely interpreted as a bias of the
can be found in the SI, Tables S2 and S6, for values obtained using adopted protocol for a given metal ion. An encouraging finding
eqs 1a, 1b, and 3). The same essentially holds true for the three is that, in the case of RI-BP86, by calibration of the described
other functionals (RI-PBE, BHLYP, and B3LYP; data in the SI, protocol a large part of this bias could be eliminated. Thus, the
Tables S3−S5) that were shown to be superior (BH-LYP, new value is calculated as follows:
B3-LYP) or on par (RI-PBE) to RI-BP86 in the gas phase [with
C calc
respect to the CCSD(T) reference]. ΔGM,L = ΔGM,L + MSSM,avg + LSSL (9)
In order to provide some arguments in favor of the observed
performance of RI-BP86, we may recall that the presented where ΔGcalcM,L is the free energy of binding calculated as before,
computational scheme of the calculation of G (see eqs 1a and 1b) MSSM,avg’s are calibration values obtained in a fashion described
includes terms of which the gas-phase complexation energies, Eel, below, i.e., one value for each metal ion. The last term, LSSL, is a
and solvation energies, Gsolv, are the most important ones. They ligand-specific shift, which is unknown for reasons discussed in
almost cancel out to yield the final values of G on the order of section 3.3. We use the exact values here (obtained from
units or tens of kilocalories per mole. It is worth mentioning that experimental values) to highlight the increased precision of the
neither of these two contributions alone contains accurate obtained relative values, which can be seen from a comparison of
information about the absolute or relative complexation energies. Tables 1 and 5.
In the COSMO-RS protocol, the solvation energy is obtained This calibration is doable thanks to a relatively small variation
from the gas-phase energy and COSMO single-point (ε = ∞) of the metal-specific shift, RMSM,L, across different systems, i.e.,
calculations using the functional for which COSMO-RS has been low values of RMSM. Interestingly, these values are lowest for
parametrized, i.e., the RI-BP86 functional. If we use RI-BP86 RI-BP86, whereas the RI-MP2, BHLYP, B3LYP, and PBE
(with the same basis set) for the gas-phase energies as well, methods are trailing behind in this sense in almost all cases.
this value cancels out and is eliminated from the final G of a Although calibration can be done in a number of ways, the
given species. Hence, only the RI-BP86 energy of a system in a quality of which will always depend on the set of systems chosen,
conductor-like environment (ε = ∞), as described by COSMO the results should not vary fundamentally. This statement is
theory and a COSMO-RS correction to the nonideal-conductor based on relatively low values of RMSM for all M for a diverse
behavior of the solvent, is present in the final value of G. group of ligands presented, and these are assumed to remain low
It is possible that the COSMO(-RS) RI-BP86 energy is free of even if we included other systems. To minimize the influence of
the taint present in the gas-phase RI-BP86 calculations or that any one of the systems on the calibration, we use average values
this is compensated for in the COSMO-RS scheme or it is at least of MSS (listed in Table 4), denoted as RMSM,avg.
not too variable across the studied metal ions. It should also be Values of ΔGCM,L are presented in Table 5 and can be directly
remembered that the evaluation can be skewed when the systems compared to experimental values from Table 1. The contribution
studied are not represented well. One of the obvious candidates of the calibration can be assessed by comparing the RMSL values
is the [M(CH3COO)(H2O)3]+ system, which has a consistently of calibrated (RMSCL ) and uncalibrated (RMSorigL ) protocols. The
low reproducibility of relative affinities. Either way, RI-BP86 calibration improves the prediction of selectivity in almost all
appears as a competitive choice to the investigated alternatives cases, with a single exception of the [M(Imi)(H2O)5]2+ system.
(see the SI, Tables S1−S6). Apart from its considerably lower Although this specific calibration is certainly not optimal, it is
10353 dx.doi.org/10.1021/ic401037x | Inorg. Chem. 2013, 52, 10347−10355
Inorganic Chemistry Article

Table 5. Calculated Values of Complexation Free Energies after Removal of Metal-Specific Bias ΔGCM,L (eq 9) and the
Corresponding RMSL (kcal·mol−1)a
ΔGCM,L
complex Mn Fe Co Ni Cu Zn Cd Hg RMSCL RMSorig
L
2+
[M(NH3)4] 0.6 −2.6 −8.6 excl. b
−18.1 −14.9 −10.9 −25.7 1.7 2.5
[M(NH3)4(H2O)2]2+ −1.9 −6.0 −6.4 −12.7 −19.7 −10.3 −10.6 −25.0 1.7 2.7
[M(Imi)(H2O)5]2+ −2.5 −4.5c −2.9 −6.2 −2.9 −5.1 −2.6 −11.6 1.9 1.3
[M(H2O)3(His)]+ −5.1 −7.5 −9.0 −12.9 −13.4 −8.6 −6.4 −17.5c 0.8 1.5
[M(H2O)4(His)]+ −5.5 −8.4 −6.5 −13.4 −14.0 −8.6 −6.6 −15.4c 1.5 1.5
[M(H2O)3(Cys)] −7.2 −10.0 −12.8 −13.2 −20.2c −9.9 −14.4 excl.b 1.5 2.6
[M(H2O)4(Cys)] −7.1 −8.1 −11.6 −13.0 −20.4c −12.8 −15.0 excl.b 0.7 1.8
[M(CH3COO)(H2O)3]+ −1.1 −0.4 −1.9 3.4 −3.8 −4.2 −3.5 −9.9 2.6 3.2
[M(CH3COO)(H2O)5]+ −1.6 −1.9 −0.8 −1.5 −2.9 −2.6 −2.2 −7.8 0.8 1.3
[M(SCH2COO)2]2− −11.3 −16.8 −19.3 excl.b −35.4c −20.5 −25.0c −56.2 2.4 3.3
a
The protocol used was RI-BP86/COSMO-RS. b“excl.” denotes results that were excluded from analysis; Table 2. cExperimental stability constants
were not available for these complexes.

conceptually simple, avoids parametrization, and as such is a The fitness of the method used for electronic structure
significant improvement that gets the majority of the absolute calculation is significantly altered when it is to be combined with
values of precision of the relative metal-ion affinity below the solvation energies calculated using the COSMO-RS protocol.
threshold of 2 kcal·mol−1, or even 1 kcal·mol−1. A seemingly inappropriate (as judged by the accuracy of the
gas-phase interaction energies) RI-BP86 functional performs,
4. CONCLUSIONS in conjunction with COSMO-RS, better than other methods
Throughout this work, we tried to give a complete account of not only in the accuracy of the relative affinities but also in the
our efforts aiming at quantitatively accurate predictions of the consistency of behavior in a wide range of systems of diverse
stability constants of metal ions in (bio)inorganic systems, using character.
the modern methods of computational chemistry. Together with Admittedly, the protocol has a notable weak point. The
a careful benchmarking of quantum-chemical methods to obtain question of structural representation of the system (e.g., the
accurate gas-phase complexation energies carried out in our coordination of H2O molecules) is not easily addressed because
previous study,5 this leads us to think that the computational the protocol is unable to reliably predict a correct conformation
protocol used in this study represents the current state-of-the-art because of their inconsistent treatment. Partly, at least, this is
of computational chemistry to treat the problem at hand due to the nonzero charge of the systems. However, there is
(ab initio predictions of the “absolute values” of the stability no simple dependence of the magnitude of error, nor obvious
constants). The only ingredient missing might be the extensive remedy, available, leaving a direct comparison of the affinity of
conformational sampling of all of the studied complexes, as metal ions for different ligands still elusive to computational
has been mentioned in the discussion. However, the increase treatment, whereas the metal-ion selectivity for the particular site
can be addressed reasonably well.

■
of computational demands to address this problem would be
formidable, and we are not aware of a standardized protocol that
would enable one to treat large sets of various complexes on ASSOCIATED CONTENT
equal footing. *
S Supporting Information
Looking at the results presented in Table 2, one may conclude Equilibrium geometries of all of the molecules studied. This
that straightforward application of the presented protocol leads material is available free of charge via the Internet at http://pubs.
only to a mediocre agreement between the experimental and acs.org.

■
theoretical stability constants at best and that this problem
cannot be handled properly by contemporary theoretical AUTHOR INFORMATION
chemistry. However, careful analysis of the trends in computed Corresponding Author
stability constants and systematic errors present therein enabled
*E-mail: [email protected].
us to suggest a computationally sound and robust protocol for
estimating the relative affinities of metal ions for the formation of Notes
complexes with ligands to within ∼2 kcal·mol−1 average accuracy The authors declare no competing financial interest.
(after removal of the systematic metal-specific shifts). This
requires a single geometry that represents the structure in the
water (solvent) environment. Using these, only a single RI-BP86
■ ACKNOWLEDGMENTS
The project was supported by the Institute of Organic Chemistry
calculation with COSMO of product complex is required, as long and Biochemistry, Academy of Sciences of the Czech Republic
as decomposition of the free energy into individual steps of the (Project RVO: 61388963).

■
thermodynamic cycle is not desired; COSMO-RS solvation
energy calculation is also required but comes at practically no REFERENCES
computational cost. A large part of the protocol’s taint is (1) Cramer, C. J.; Truhlar, D. G. Acc. Chem. Res. 2008, 41, 760−768.
eliminated by simple calibration. The choice of the particular (2) Klamt, A.; Mennucci, B.; Tomasi, J.; Barone, V.; Curutchet, C.;
calibration values for removal of the metal-dependent shifts is Orozco, M.; Luque, F. J. Acc. Chem. Res. 2009, 42, 489−492.
based on a set of structurally simple systems, and its details do not (3) Rokob, T. A.; Srnec, M.; Rulíšek, L. Dalton Trans. 2012, 41, 5754−
fundamentally influence the results obtained. 5768.

10354 dx.doi.org/10.1021/ic401037x | Inorg. Chem. 2013, 52, 10347−10355

Inorganic Chemistry Article

(4) Beck, M. T.; Nagypal, I. Complex Equilibria: Stability Constants; (47) Martell, A. E.; Smith, R. M. Critical Stability Constants; Plenum
Halsted Press: New York, 1989. Press, New York, 1974−1989.
(5) Gutten, O.; Bešsě ová, I.; Rulíšek, L. J. Phys. Chem. A 2011, 115, (48) Klamt, A.; Mennucci, B.; Tomasi, J.; Barone, V.; Curutchet, C.;
11394−11402. Orozco, M.; Luque, F. J. Acc. Chem. Res. 2009, 42, 489−492.
(6) Sigel, R. K. O.; Sigel, H. Acc. Chem. Res. 2010, 43, 974−984.
(7) Dudev, T.; Lim, C. Chem. Rev. 2003, 103, 773−787.
(8) Glusker, J. P. Adv. Protein Chem. 1991, 42, 1−76.
(9) Rulíšek, L.; Vondrásě k, J. J. Inorg. Biochem. 1998, 71, 115−127.
(10) Lee, K. H.; Matzapetakis, M.; Mitra, S.; Marsh, E. N. G.; Pecoraro,
V. L. J. Am. Chem. Soc. 2004, 126, 9178−9179.
(11) Williams, R. J. P. Biometals 2007, 20, 107−112.
(12) Handbook of Metal−Ligand Interactions in Biological Fluids;
Berthon, G., Ed.; Marcel Dekker: New York, 1995; Vol. 2.
(13) Peacock, A. F. A.; Hemmingsen, L.; Pecoraro, V. L. Proc. Natl.
Acad. Sci. U. S. A. 2008, 105, 16566−16571.
(14) Sudhir, P.-R.; Wu, H.-F.; Zhou, Z.-C. Rapid Commun. Mass
Spectrom. 2005, 19, 1517−1521.
(15) Dudev, T.; Lim, C. Annu. Rev. Biophys. 2008, 37, 97−116.
(16) Kuppuraj, G.; Dudev, M.; Lim, C. J. Phys. Chem. B 2009, 113,
2952−2960.
(17) Senn, H. M.; Thiel, W. Angew. Chem., Int. Ed. 2009, 48, 1198−
1229.
(18) Ryde, U. Curr. Opin. Chem. Biol. 2003, 7, 136−142.
(19) Kamerlin, S. C. L.; Haranczyk, M.; Warshel, A. J. Phys. Chem. B
2009, 113, 1253−1272.
(20) Warshel, A. Computer Modeling of Chemical Reactions in Enzymes
and Solutions; John Wiley & Sons, Inc.: New York, 1997.
(21) Rulíšek, L.; Havlas, Z. J. Phys. Chem. A 1999, 103, 1634−1639.
(22) Rulíšek, L.; Havlas, Z. J. Chem. Phys. 2000, 112, 149−157.
(23) Rulíšek, L.; Havlas, Z. J. Am. Chem. Soc. 2000, 122, 10428−10439.
(24) Rulíšek, L.; Havlas, Z. J. Phys. Chem. A 2002, 106, 3855−3866.
(25) Rulíšek, L.; Havlas, Z. Int. J. Quantum Chem. 2003, 91, 504−510.
(26) Rulíšek, L.; Havlas, Z. J. Phys. Chem. B 2003, 107, 2376−2385.
(27) Kožíšek, M.; Svatoš, A.; Buděsí̌ nský, M.; Muck, A.; Bauer, M. C.;
Kotrba, P.; Ruml, T.; Havlas, Z.; Linse, S.; Rulíšek, L. Chem.Eur. J.
2008, 14, 7836−7846.
(28) Dudev, T.; Lim, C. J. Phys. Chem. B 2001, 105, 10709−10714.
(29) Dudev, T.; Lim, C. Acc. Chem. Res. 2007, 40, 85−93.
(30) Delgmann, P.; Schenk, S. J. Comput. Chem. 2012, 33, 1304−1320.
(31) Droghetti, A.; Alfé, D.; Sanvito, S. J. Chem. Phys. 2012, 137,
124303.
(32) Furche, F.; Perdew, J. P. J. Chem. Phys. 2006, 124, 044103.
(33) (a) Becke, A. D. Phys. Rev. A 1988, 38, 3098−3100. (b) Lee, C.;
Yang, W.; Parr, R. G. Phys. Rev. B 1988, 37, 785−789. (c) Becke, A. D. J.
Chem. Phys. 1993, 98, 5648−5652. (d) Stephens, P. J.; Devlin, F. J.;
Frisch, M. J.; Chabalowski, C. F. J. Phys. Chem. 1994, 98, 11623−11627.
(34) Perdew, J. P. Phys. Rev. B 1986, 33, 8822−8824.
(35) Weigend, F.; Ahlrichs, R. Phys. Chem. Chem. Phys. 2005, 7, 3297−
3305.
(36) Schäfer, A.; Huber, C.; Ahlrichs, R. J. Chem. Phys. 1994, 100,
5829−5835.
(37) Perdew, J. P.; Burke, K.; Ernzerhof, M. Phys. Rev. Lett. 1996, 77,
3865−3868.
(38) Hertwig, R. H.; Koch, W. Chem. Phys. Lett. 1997, 268, 345−351.
(39) Eichkorn, K.; Treutler, O.; Ö hm, H.; Häser, M.; Ahlrichs, R.
Chem. Phys. Lett. 1995, 240, 283−290.
(40) Grimme, S.; Antony, J.; Ehrlich, S.; Krieg, H. J. Chem. Phys. 2010,
132, 154104.
(41) Andrae, D.; Haussermann, U.; Dolg, M.; Stoll, H.; Preuss, H.
Theor. Chim. Acta 1990, 77, 123−141.
(42) Klamt, A. J. Phys. Chem. 1995, 99, 2224−2235.
(43) Klamt, A.; Jonas, V.; Buerger, T.; Lohrenz, J. C. W. J. Phys. Chem.
1998, 102, 5074−5085.
(44) Eckert, F.; Klamt, A. COSMOtherm, version C3.0, release 12.01;
COSMOlogic GmbH & Co. KG: Leverkusen Germany, 2011.
(45) Klamt, A.; Schuurmann, G. J. Chem. Soc., Perkin Trans. 2 1993,
799−805.
(46) Jensen, F. Introduction to Computational Chemistry; John Wiley &
Sons: New York, 1999.

10355 dx.doi.org/10.1021/ic401037x | Inorg. Chem. 2013, 52, 10347−10355