Perspective pubs.acs.org/jmc

Ligand Binding Thermodynamics in Drug Discovery: Still a Hot Tip? Stefan Geschwindner,*,† Johan Ulander,*,‡ and Patrik Johansson*,† †

Discovery Sciences, AstraZeneca R&D Mölndal, S-43183 Mölndal, Sweden CVMD Innovative Medicines, AstraZeneca R&D Mölndal, S-43183 Mölndal, Sweden

‡

S Supporting Information *

ABSTRACT: The use of ligand binding thermodynamics has been proposed as a potential success factor to accelerate drug discovery. However, despite the intuitive appeal of optimizing binding enthalpy, a number of factors complicate routine use of thermodynamic data. On a macroscopic level, a range of experimental parameters including temperature and buffer choice significantly influence the observed thermodynamic signatures. On a microscopic level, solute effects, structural flexibility, and cooperativity lead to nonlinear changes in enthalpy. This multifactorial character hides essential enthalpy contributions of intermolecular contacts, making them experimentally nonobservable. In this perspective, we present three case studies, reflect on some key factors affecting thermodynamic signatures, and investigate their relation to the hydrophobic effect, enthalpy−entropy compensation, lipophilic ligand efficiency, and promiscuity. The studies highlight that enthalpy and entropy cannot be used as direct end points but can together with calculations increase our understanding of ligand binding and identify interesting outliers that do not behave as expected.

■

The use of thermodynamic signatures or fingerprints is proposed as a way to classify the binding of compounds. Typically thermodynamic plots of the measured free binding energy ΔGobs, the measured binding enthalpy ΔHobs, and the calculated entropic contribution −TΔSobs are used to illustrate the driving forces of the binding process (Figures 3a and 7). The sum of ΔHobs and −TΔSobs will result in the overall free energy of binding ΔGobs, enabling a quick assessment of their individual contributions and thus the affinity. A large negative value for ΔHobs is typically interpreted as being advantageous, as this could reflect a positive change in the number and/or strength of noncovalent bonds on going from the free to the bound state. A large negative value for −TΔSobs (indicating a favorable increase in binding entropy) is often attributed to nonspecific hydrophobic effects and often characterized as disadvantageous. Ultimately, the hypothesis is to generate molecules that show a balanced thermodynamic profile in terms of larger enthalpic and smaller entropic contributions. An additional proposal for using thermodynamic data is to select high-throughput screening (HTS) or fragment hits that

INTRODUCTION Within the past decade, the possibility of using the thermodynamics of compound binding to guide medicinal chemistry efforts has gained much attention.1−5 Triggered by impressive technology developments in the area of label-free technologies and spurred by some land-marking publications of key opinion leaders,6−8 expectations were quickly raised that the use of binding thermodynamics could improve the outcome of early drug discovery. Several authors suggested that the optimization of ligand binding enthalpy ΔHobs might result in more selective leads with less ADMET issues, arguing that enthalpy-driven binding is primarily based on specific target−ligand interactions like hydrogen bonds, salt bridges, or directed van-der-Waals contacts as opposed to entropy-driven binding based on nonspecific lipophilic interactions.6−9 Following the initial excitement and fuelled by a range of apparent success stories built around HIV-protease and HMG-CoA reductase inhibitors,10,11 it became apparent that most examples were rationalized retrospectively with the help of structural information, giving little guidance on how to prospectively use enthalpy and entropy data in lead generation and optimization programs in the absence of such information. © 2015 American Chemical Society

Received: October 1, 2014 Published: April 27, 2015 6321

DOI: 10.1021/jm501511f J. Med. Chem. 2015, 58, 6321−6335

Perspective

Journal of Medicinal Chemistry

Figure 1. Schematic representation of an isothermal titration calorimeter (left) and the results of a characteristic titration experiment (upper right) with the associated data analysis (lower right). Controlled movement of the piston, which is driven by the injection motor, enables the exact injection of sub-μL amounts of a ligand solution into the sample cell, which usually contains a solution of the macromolecule. The raw data obtained through multiple injections, as shown in the upper right, will be integrated to generate the binding isotherm shown underneath, from which the binding affinity KD (determined by the slope of the binding isotherm) and the observed binding enthalpy ΔHobs (maximum amplitude) can be extracted.

computational approaches and the omnipresent phenomenon of enthalpy−entropy compensation. With these foundations, we highlight and discuss some key challenges that currently limit the application of thermodynamic data in lead generation and optimization. We aim to raise the awareness about the wealth of microscopic changes that accompany ligand binding, masking the individual contribution of intermolecular bond formation or breakage. Our ambition is to balance the level of expectations on the impact of enthalpy and entropy data for compound series selection and rational design as well as to indicate the potential of ligand binding thermodynamics combined with computational methods as a way to identify and exploit unexpected ligand binding modes.

primarily display a favorable binding enthalpy and then improve potency by a balance between entropic and enthalpic contributions.12 Interestingly, classification of compounds based on their thermodynamic profiles was already performed more than three decades ago by observing thermodynamic differences of antagonists and agonists to the β-adrenergic receptor.13 However, new studies have shown that such a thermodynamic discrimination might require a more critical assessment. For example the classification of agonists and antagonists acting on the histamine H(1) receptor revealed that a definite discrimination between antagonism and agonism based on thermodynamic parameters is not possible.14 A fundamental complication for thermodynamics-guided drug design arises due to the fact that the experimentally observed ΔHobs and the calculated ΔSobs are global parameters, containing a mix of many contributions. The rationale for enthalpy optimization is primarily based on the assumption that the measured ΔHobs is dominated by direct ligand−protein interactions. This simplified view neglects the complexity and dynamics of ligand binding events, including factors like protein and/or ligand flexibility as well as solvation and desolvation effects that significantly affect the obtained thermodynamic signatures.15−17 In particular, the nonlocal nature of hydrogen bond networks leads to a nonlinear response to perturbations and significantly complicates the analysis of thermodynamic data.18−20 As a consequence of these issues, the concept of applying binding enthalpy in the drug design process has been challenged by several authors,21−23 and the strategy has not yielded the proposed impact on lead generation success. This article will introduce some general concepts relating to practical and theoretical approaches and considerations when attempting to measure and understand compound−target interactions. This includes measurement of thermodynamic data, the concept of lipophilic ligand efficiency, the dominant role of water in ligand binding processes, as well as

■

DETERMINATION OF LIGAND BINDING THERMODYNAMICS The concept ligand binding thermodynamics is built around the consideration and directed manipulation of the two constituents of the standard Gibbs free energy of binding, the standard enthalpy ΔH°, and standard entropy ΔS°. The standard Gibbs free energy of binding ΔG° is defined through eq 1, with the equilibrium binding constant KD as well as the temperature as variable parameters. ΔG° = −RT ln(KD)

(1)

The experimentally observed ΔG° is the difference of the standard free energy for the process of mixing two solutions with different concentrations of constituents. ΔH° and ΔS° are state functions; each can be expressed in terms of the other two as given by the following equations:

6322

ΔG° = ΔH ° − T ΔS°

(2)

ΔH ° = ΔG° + T ΔS°

(3)

ΔS° = (ΔH ° − ΔG°)/T

(4) DOI: 10.1021/jm501511f J. Med. Chem. 2015, 58, 6321−6335

Perspective

Journal of Medicinal Chemistry Both ΔH° and ΔS° are typically sums over an enormous amount of degrees of freedom, many contributions are opposing each other and comparable in amplitude, and in the end sums up to a comparably smaller number, ΔG°. These linear relationships imply two things: First, the effect on binding affinity from, e.g., enhancing the enthalpic binding contribution may be counteracted from increases in the entropic component. The difficulty in increasing affinity has sometimes been attributed to a phenomenon called entropy−enthalpy compensation (EEC).24 Some degree of EEC appears quite intuitive, as the formation of any complexation, e.g., intermolecular bonds, will be associated with a loss of degrees of freedom and decrease in entropy for the individual partners. Several mechanisms for additional EEC have been proposed, including correlations that arise only due to systematic experimental errors.23,25 Second, the temperature during the measurements of either affinity or enthalpy will have smaller or larger influence on the obtained values, dictated by the molar heat capacity change at constant pressure ΔCp of the system under investigation. The molar heat capacity change can be expressed by several equations, of which eq 5 is synonymous with its name: ΔCp = (ΔΔH /ΔT )p

determination of the ligand and the protein concentrations.25 The accuracy of the estimates for affinity and enthalpy is totally reliant on the accurate concentration of the titrant, and errors on the scale of 15−20% are quite common. These errors will directly impact the estimated binding enthalpy but to a much lesser extent the free binding energy due to the logarithmic dependence of ΔG on KD (see eq 1). The errors in affinity determination are usually lower in comparison with traditional biochemical assays used for example in screening campaigns (0.2−0.3 log units variation in pKD based on unpublished inhouse data), thus making ITC the golden standard for affinity determination. Importantly, owing to the intimate relationship of enthalpy and entropy, the experimental errors in ΔHobs will directly influence the calculation of the binding entropy and will result in apparent enthalpy−entropy compensation. This will consequently lead to distorted thermodynamic profiles and subsequent flaws in the interpretation. Thus, great care needs to be taken when setting up an ITC experiment and any thermodynamic data needs to be scrutinized to allow proper data comparison. This becomes even more important when comparing data from larger data sets that have been produced by multiple laboratories like PDBcal,31 as the experimental setups will show large within between a set of diverse users. The choice of buffer in an ITC experiment can also have a large impact on the observed enthalpy if there are contributions from the heat of ionization of the buffer due to proton movements during binding.30 This was extensively discussed in a recent publication,5 nicely detailing the experimental rigor required to draw sound conclusions from ITC data. Briefly, the assessment of different buffer systems with varying ionization enthalpies ΔHion can uncover proton movements between protein, ligand, and buffer, which is of particular importance if the original buffer system has a high ionization enthalpy like, e.g., Tris (ΔHion = 11.34 kcal/mol). This assessment allows a correction of ΔHobs according to the ionization of the used buffer and the number of protons that move during the binding process. Such analysis might actually help to uncover important mechanistic information on the ligand binding process. In addition, buffer salts that influence the hydration of ligands have also been implied in changing the thermodynamic signatures and masking other effects.32 Any ITC experiments that are performed without controlling such effects could potentially be meaningless and might lead to improper conclusions. An alternative way of producing thermodynamic data for compound binding is done by calculating the so-called van’t Hoff enthalpy ΔHvan’t Hoff by measuring the changes in the binding affinity depending on the temperature.14,33 This relationship between binding enthalpy and binding affinity is described through the van’t Hoff equation:

(5)

The equation implies that differences in ΔCp can be interpreted as the changed ability to absorb heat upon temperature increase. A change in temperature usually has only a small influence on the measured affinity, which is contrasted by a much stronger impact on the measured enthalpy. This is a general phenomenon and can have remarkable consequences when interpreting thermodynamic data, in particular if the associated heat capacity changes are unknown. The majority of protein−ligand interactions experience a nonzero heat capacity change that can display a large degree of variation due to direct and indirect changes in solvation during complex formation.26 This will be further discussed in one of the experimental examples below. A more comprehensive analysis has additionally shown that heat capacity changes are not only connected to changes in solvation but also to variations in molecular flexibility and conformational fluctuations induced by ligand binding.27,28 This adds another level of complexity for the interpretation of ΔHobs and ΔSobs data. Appreciating the large impact of temperature on the measured binding enthalpy is of particular importance when such data is produced by isothermal titration calorimetry (ITC, see Figure 1). ITC enables the determination of the affinity, the binding enthalpy, and the stoichiometry in a single titration experiment at constant temperature.29,30 It involves the monitoring of the produced (exothermic binding event) or absorbed (endothermic binding event) heat during protein− ligand binding. During the titration process, which is typically done by injecting a concentrated ligand solution from a syringe into a protein solution that is contained in the calorimeter cell, the protein becomes increasingly saturated until all binding sites are occupied. As a consequence, the observed heat change will vary during the titration and will upon appropriate choice of the experimental conditions produce a binding isotherm that allows precise estimates of the affinity, the enthalpy, and the stoichiometry of the binding reaction (Figure 1). A factor that has attracted less attention are the experimental errors during an ITC measurement.23 This is highlighted in a recent study showing the impact of errors made in the

ΔH ° = −R(δ ln(KD)/(δ1/T ))

(6)

where R is the gas constant and T is the absolute temperature. The instantaneous slope of a plot of ln KD vs 1/T, multiplied by −R, can be used to determine ΔH° under the assumptions that the binding process is a two-state reaction and that ΔH° is temperature-independent. Although one might intuitively expect a good correlation between calorimetric and van’t Hoff enthalpies, there are a number of reports that highlight discrepancies between the two approaches, indicating that the van’t Hoff analysis is fraught with error and difficulty.34−36 Measurement errors that are usually accompanied by affinity determinations together with the frequent observation of 6323

DOI: 10.1021/jm501511f J. Med. Chem. 2015, 58, 6321−6335

Perspective

Journal of Medicinal Chemistry

If we express the lipophilic ligand efficiency (eq 7) in terms of chemical potentials, we get

nonzero heat capacity changes could lead to inaccurate estimation of the binding enthalpy. Thus, calorimetric enthalpies are currently considered the preferred choice to obtain thermodynamic data in drug discovery.

LLE = μLT − μL + μL − μLoct = μLT − μLoct

■

LLE thus subtracts μL, but we are left with μLoct, switching the reference state from the aqueous solution to the ligand in octanol. The usefulness of LLE hinges on the fact that differences in Log D between molecules are dominated by variations in the chemical potential in water rather than in octanol. The suitability of this should be best for similarly sized ligands. However, the choice of octanol is arbitrary and is far from perfect because we can have relatively large effects arising from compound−octanol interactions. It is also worth noticing that some studies have also used competing proteins as reference systems, as exploited by Ladbury.7 To examine the difference between enthalpy- and LLEdriven optimization, we will decompose ΔGBind into a thermodynamic “half cycle” (Figure 2) which can be used to compare affinities between ligands toward the same target.

ENTHALPIC OPTIMIZATION VS LIPOPHILIC LIGAND EFFICIENCY A major incentive for the interest in thermodynamic profiling is to get a quantitative measure on how specific the potency toward the desired target is. A simple way of stating this is that we search for compounds that have a strong affinity toward the target rather than being pushed into a binding pocket by solvophobic interactions. The assumption that highly specific binding has large enthalpic contributions and that, e.g., hydrophobically driven binding is dominated by entropic forces indicates to some extent a link between enthalpy and the lipophilic ligand efficiency (LLE).37 LLE is a common metric often used to estimate the specificity of binding and can generally be written as −Log(potency)−Log(partition coefficient). If we use KD as potency and Log D as partition coefficient, we get LLE = pK D − Log D

(10)

ΔGBind = G[LP(aq)] − G[L(aq) + P(aq)]

(11)

(7)

Several authors have argued that optimizing ligand binding enthalpy is equivalent to optimizing lipophilic ligand efficiency.37,38 This may be true under some circumstances but is not always the case as demonstrated in one of the experimental examples below (Figure 6). To illustrate the implications of this assumption, we need to look at the individual components of protein−ligand binding. The free energy of binding is determined by the difference in chemical potential between the complex and the isolated components in solution. The free energy difference for binding of a single molecule toward its target can be written as ΔGBind = μLT − μL − μT

(8)

where μ is the chemical potential of the target/ligand complex, and μT and μL as μLT that of the solvated target and ligand, respectively. The chemical potential of molecular species “I” is the derivative of the Gibbs free energy with respect to number of particles “i”: LT

⎛ ∂G ⎞ μi = ⎜ ⎟ ⎝ ∂ni ⎠T , P , Nj ≠ i

Figure 2. Thermodynamic cycle describing the driving forces in ligand binding and the relation to LLE based on the assumption that the water phase is similar in the Log D and the binding assay. The dissociation constant pKD is determined by the difference in chemical potential between the solvated aqueous phase and the local chemical potential of the ligand−protein complex. The chemical potential for the isolated solvated protein is a constant and does not depend on the ligand, i.e., pKD* = pKD + constant. Similarly, Log D is determined by the difference in chemical potential of the ligand in the aqueous and octanol phase.

(9)

Here the subscript T,P,Nj≠i indicates that the temperature and pressure as well as the number of all other chemical species is kept constant. The term μT is the same for all ligands and gives that the relative affinity between ligands is determined by the difference in μLT and μL. The absolute value of the chemical potential depends on the choice of reference state, but knowing the relative chemical potential in solution between compounds could serve as an optimal reference point to assess general promiscuity. To understand the strength of the actual interactions in a protein− ligand complex, it would be of interest to subtract the chemical potential in water or at least the relative μL between ligands. Relative chemical potentials in solution are, however, not readily available, and we have to use other parameters as substitutes. The (octanol/water) partition coefficient Log D is routinely measured and is one such potential choice.

If we compare ligands “i” and “j”, we have the difference ΔΔGBind i→j which, using a thermodynamic cycle, can be written as (cf. Supporting Information): ΔΔGiBind → j = G[LjP(aq)] − G[Li P(aq)] + G[Lj(aq)] − G[Li(aq)]

(12)

Thus, G[L(aq) + P(aq)] does not have to be taken into account when comparing binding free energies between different ligands. If the system is dilute, the free energy of the free ligand and free protein can be assumed to be independent of each other and we can write 6324

DOI: 10.1021/jm501511f J. Med. Chem. 2015, 58, 6321−6335

Perspective

Journal of Medicinal Chemistry ΔGBind = G[LP(aq)] − G[L(aq)] − G[P(aq)]

captured with high accuracy. The effect of polar groups (often favors water, but not always) is more challenging to predict, and as we introduce more polar groups, the accuracy of predictive models goes down.47 For log D (pH/ionization dependent), there is a strong signal from ionization of the molecule which introduces a large uncertainty in lipophilicity predictions, in particular for compounds with pKas close to the pH.48 The entropic signal from cavity desolvation should to a first approximation be a function of the number of water molecules displaced and thus the ligand size. However, when we compare ligands of near equal size, it is much less apparent what entropy signature to expect as we introduce or remove polar groups. The hydration of ions have for instance two large but almost perfectly canceling entropic contributions, a negative ion−water interaction entropy and a positive water reorganization entropy. Consequently, rare gas atoms and alkali−halide ions have nearly identical hydration entropies.49 Water molecules can form highly ordered networks and are also thought to adapt highly ordered states near biological surfaces.43 This is supported by molecular dynamics studies that indicate that the formation of well-ordered water clusters in the vicinity of hydrophobic surfaces would give an entropy gain upon desolvation.50,51 However, several recent studies suggest that especially protein solvation might be more complex than originally thought. Terahertz spectroscopy data indicate that the waters of the two first solvation shells of a macromolecular surface behave very differently compared to bulk water.52,53 This is supported by nuclear magnetic resonance experiments54,55 as well as spectroscopic data,56 showing that water molecules in hydrophobic cavities may be significantly disordered. The cooperativity of hydrogen bonds results in hydrogen bond energies being about 250% stronger in liquid water solutions compared with H2O monomers.18 Pairwise additive potentials have been estimated to have errors exceeding 20% already for clusters larger than pentamers.19 Water is also expected to form a closer packing than the enthalpically favored tetrahedral packing. The significantly distorted hydrogen bonds allow the quantized vibrational modes to be excited and contribute to the entropy.18,20 These complications make the surface tension dependence on cavity curvature nontrivial to calculate. A number of theoretical studies indicate that the thermodynamics of individual solute waters are determined by the surface distribution of H-bond acceptors and donors. Olano and Rick performed free energy calculations of transferring a water molecule into two types of protein cavities, a polar cavity represented by the bovine pancreatic trypsin inhibitor (BPTI) and a more hydrophobic cavity represented by the I76A mutant of barnase.57 They found that the entropy of hydration varied significantly between the two proteins. The entropy of hydration for the BPTI cavity was negative, whereas it was positive for the barnase cavity. This implies that desolvation of the hydrophobic barnase pocket would be entropically unfavorable. Similar results were found by Setny and co-workers using a hydrophobic receptor model system, indicating that the release of water from a hydrophobic pocket would be enthalpically favorable and entropically unfavorable.58 This is in line with NMR measurements that show that the displacement of hydrophobic waters can be connected to a significant entropy loss.55 Thus, the thermodynamic properties of water might span the entire range from enthalpically favorable/entropically unfavorable to bulk-like to enthalpically

(13)

Again, the term G[P(aq)] is the same for all ligands and need not be considered when comparing affinities between ligands. Thus, we can write a thermodynamic cycle according to Figure 2 that illustrates how LLE shifts the reference state of the ligand in water L(w) to ligand in octanol L(o). Using the thermodynamic cycle, we can write an expression for the requirements for LLE and enthalpy to equal each other (cf. Supporting Information): ΔH[L(w) → L(o)] ≈ T ΔS[L(w) → LP]

(14)

This equation states that any enthalpy difference originating from the ligand transfer from water to octanol is balanced by the entropy difference for transferring the same ligand from octanol to the binding site. This relation should not be expected to hold in general, implying that LLE and enthalpy signatures do not necessarily measure the same thing even when the binding mode is conserved and the protein conformational ensemble is unaffected. One of the differences between LLE and enthalpy can be highlighted by the following example: The strength of electrostatic interactions depends highly on the polarizability of the surroundings. The protein interior and in particular hydrophobic pockets will therefore accentuate Coulomb effects. Thus, to a first approximation, it is natural to assume that introducing a hydrogen bond with near optimal geometry in a hydrophobic pocket with an “unsatisfied” hydrogen bond interaction site would give a strong enthalpic signal in a thermodynamic sense. On the other hand, we may not distinguish this thermodynamic signal from an intramolecular hydrogen bond. However, the ability to make intramolecular hydrogen bonds is likely to make a molecule more lipophilic than a similar molecule without this possibility. Thus, the ligand with intramolecular hydrogen bonds should have lower LLE than if it instead exploited protein interaction points but could have the same thermodynamic profile, also indicating a case when thermodynamic measurements and lipophilic ligand efficiency can be complementary.

■

MOLECULAR DETERMINANTS OF ENTHALPY AND ENTROPY AND THE ROLE OF WATER Solvation and Desolvation. The concept of enthalpy optimization is connected to the classical interpretation of the hydrophobic effect, whereby the aggregation of hydrophobic ligands decreases the disruption of bulk water and increases the entropy of the system.39−42 A net increase in hydrogen bonds upon ligand binding is usually thought to be associated with a favorable enthalpy, ΔHobs. However, a hydrogen bond between a ligand and the protein often competes with bound waters, leading to a nonobvious change in number and characteristics of the combined hydrogen interactions.43 Desolvation is also traditionally expected to lead to favorable entropy due to the liberation of solvent molecules. This is to a large extent due to the difference in size of the solvent molecules and the ligand and is captured, e.g., in classical Flory−Huggins polymer theory,44,45 which describes the entropy penalty upon polymerization. Conceptually, it states that as we grow a polymer, the entropy penalty for restraining the individual monomers in a particular cavity decreases. In general, the size is the dominant factor for hydrophobicity;46 it captures the free energy of creating a cavity in the solvent. This term typically stands for over 90% of the variation for lipophilicity of alkanes and is 6325

DOI: 10.1021/jm501511f J. Med. Chem. 2015, 58, 6321−6335

Perspective

Journal of Medicinal Chemistry

incorporated ad hoc in, e.g., quantum density functional applications. For biological macromolecules, quantum chemical calculations for large conformational ensembles are not feasible and simpler force fields are used. The development of force fields for classical mechanics simulations are constantly evolving and may now be able to reproduce the individual thermodynamic behavior for many small molecules, including peptides.75 Modeling of explicit water consumes a lot of computational power, and water can adopt incalculable number of relevant states for a given protein conformation. Although recent technology advances have enabled millisecond simulations76 of medium-sized proteins in solution, actual calculations of true free energy differences are in practice intractable and one is forced to use other approaches and techniques to circumvent this. Perturbation theory tells us that if two systems are similar enough, one can expect simulations to be able to capture free energy differences between different states.77 This can allow investigations of small perturbations of, e.g., ligands or amino acids mutations by putting restraints on the system. A rough categorization of strategies for calculating binding free energies are end-point methods, reaction coordinate calculations, or alchemical transformations.78,79 The strategies are not mutually exclusive, and there exists many variants.79−83 The challenge lies in that we do not beforehand know the applicability domain of a particular system, i.e., that we have sampled the thermodynamically relevant states.78 An example of large difference in thermodynamic behavior between closely related states is highlighted in the analysis of Fenley et al.84 A key success factor in general is thus the identification of a suitable reaction coordinate85,86 and to identify the domain of applicability. For both of these components, high-quality experimental information is crucial. Enthalpy−Entropy Compensation (EEC). The creation of a rigid protein−ligand complex results in a favorable enthalpy gain that is compensated to some extent by an entropy loss caused by the decrease of the available states for the system. This is the basis for EEC.24,86,87 The existence of additional EEC has been questioned by a number of studies due to a small range of observed ΔGobs values from ITC studies and the linear relationship between ΔHobs and ΔSobs that can cause an apparent compensation.23,88,89 It is also clear that many of the published measurements contain large errors.25 However, the observed correlation between ITC and van’t Hoff data,33,90,91 several experimental and computational studies,92 as well as rigorous statistical modeling93 support the notion that EEC is a real physical phenomenon. This effect is also connected to the behavior of water as has been highlighted by some recent work involving a series of benzothiazole sulfonamide ligands with different patterns of fluorination binding to human carbonic anhydrase (HCA).94 Differences in the structure and thermodynamic properties of the water network surrounding the bound ligands were found to be an important contributor to the observed EEC. These results also support the hypothesis that water molecules can be an integral part of the binding site and are as important as the direct contact interactions between the protein and the ligand in defining the thermodynamics of binding.92 A recent simulation of the bovine pancreatic trypsin inhibitor BPTI, describing a millisecond dynamics trajectory, enabled the analysis of several folding/unfolding events.76 In another recent paper,95 Fenley et al. analyzed this trajectory and identified two folded states of BPTI where one was dominated by enthalpy

unfavorable/entropically favorable in a hydrophilic or a hydrophobic surrounding.59 Effects like these have been studied in detail using the pheromone-binding mouse major urinary protein (MUP).60−62 MUP promiscuously binds a number of different hydrophobic ligands like primary aliphatic alcohols and pyrazine derivatives in a deep hydrophobic binding pocket. Interestingly, lipophilic ligand binding in this hydrophobic pocket was found to be associated with a strong enthalpic signature and a negative change in heat capacity. This unexpected behavior was explained by the suboptimal hydration of the MUP pocket.63,64 A comparison of ITC measurements of ligand−MUP binding in deuterium and in protium indicated that the solvation contribution to binding enthalpy was very small.65 In addition, long time scale MD simulations suggested a significant “dewetting” of the hydrophobic cavity. The few remaining waters were found to be significantly disordered, resulting in a favorable contribution to the observed binding enthalpy upon desolvation and thus canceling the enthalpic solvation penalty of the hydrophobic ligands. The solvent was unable to compensate for the gain in dispersive protein−ligand interactions, resulting in a strong enthalpic profile.66 Calculating Thermodynamic Parameters. To characterize a free energy of binding requires that we define a bound and a nonbound state. Molecules are dynamic entities, and a state is not characterized by a single conformation but a manifold of microstates that forms a free-energy landscape.67−71 Thus, in the case of noncovalent binding where the ligand binds reversibly to the target protein, we are forced to group these states into macrostates, either in terms of function or a (conformational/spatial) reaction coordinate.72 Experimental examples of this are binding constants, KD or IC50s. Regardless of the definition used, it always artificially separates conformations in the overall ensemble as belonging to one or another state. The experimental signal from ITC is the observed enthalpy ΔHobs and a calculated KD that give us a value for the entropy for the whole process of binding. The entropy can be further split into, e.g., conformational, vibrational, or rotational changes. For most conceptual purposes, classical mechanics can be used to explain the thermodynamics of binding. As a result of this, the momentum and hence the molecular weight plays no part in binding entropy. As opposed to enthalpy, entropy will differ depending on the choice of reference state. Usually the concentration 1 M is chosen, but this is arbitrary.73 Because macromolecules can adopt a large number of alternative conformations, any calculations of the relevant free energy landscape of binding will be computationally expensive even for very modest size systems using explicit atomic descriptions. The many degrees of freedom are in stark contrast with typical chemical reactions which often are determined by well-defined energy barriers. Thus, it is simply not enough to calculate the energies for a few conformations. The free energy estimation is thus a sum of many terms, and there is a risk of error propagation which increases the demand for precision and accuracy. Enthalpy calculations for a given conformation can for many small systems be calculated with high precision using quantum chemistry if the interaction is dominated by chemical bonds.74 How large systems that are feasible depends highly on the type of atoms involved. If there are systematic errors in accuracy, e.g., between different functional groups, also high precision calculations may lead to spurious results. Dispersion forces are still difficult to address ab initio and are often 6326

DOI: 10.1021/jm501511f J. Med. Chem. 2015, 58, 6321−6335

Perspective

Journal of Medicinal Chemistry and the other state by entropy. Interestingly these two states are only separated by a small conformational switch in a few residues. Hence restricting the degrees of freedom for a single residue may shift the entire protein conformational ensemble to be dominated by either entropy or enthalpy. The authors introduce a concept called entropy/enthalpy transduction to describe these phenomena. Such a strong correlation between local and global protein conformation and entropy is also supported by NMR measurements that indicate that some proteins increase their mobility upon ligand association on the pico- to nanosecond time scale to compensate the entropic costs of binding,16,66 illustrating the concept of entropy− entropy compensation.

■

EXAMPLES HIGHLIGHTING CURRENT CHALLENGES IN THE ANALYSIS OF THERMODYNAMIC DATA The nonlinear connection between protein, ligand, and water, conformation, enthalpy, and entropy significantly complicates the analysis of measured thermodynamic data. This is highlighted by a few examples from three drug discovery projects that we have accumulated over the past years. Entropy−Entropy Compensation with Cyclin-Dependent Kinase 5 (CDK5) and CDK5/p25. CDK5 is a proline-directed kinase and involved in several sensory pathways.96 CDK5 has been implicated as an interesting target for treating Alzheimer’s disease by preventing hyperphosphorylation of tau and neurofibrillary tangle formation.97 The activation of CDK5 is controlled by the p35 and p39 proteins through the formation of a specific complex. The membranebound p35 can be proteolytically cleaved by Calpain to generate the soluble product p25, retaining its kinase activation potential. The crystal structure of the CDK5/p25 protein complex98 provided an insight into the activation mechanism and revealed that the activation loop of CDK5 is tethered in an extended conformation that closely resembles that observed in the active kinase. There are no major structural changes observed in the ATP-binding pocket upon complex formation. The binding of roscovitine, which is a specific cyclin-dependent kinase inhibitor that binds competitively in the ATP-binding site of CDK5,99 has been studied with ITC using both CDK5 alone as well as in the full complex. Interestingly, one can observe different thermodynamic profiles for the binding of roscovitine to CDK5 and to CDK5/p25 (Figure 3a), which can only be explained by different thermodynamic states of the two species, as the binding mode and the molecular interaction pattern is unaltered. While the binding enthalpies are not significantly different, a larger discrepancy can be observed between the entropic binding terms. The entropic penalty that has to be paid during the binding process to the noncomplexed CDK5 can potentially be rationalized by an unfavorable decrease in conformational entropy of CDK5 by adopting of the active CDK5 conformation seen in the X-ray structure (Figure 3b). As p25 induces a conformation resembling that of the active protein, this entropic penalty is reduced in the case of the binding to the CDK5/p25 complex, resulting in a significant affinity increase. The observed difference in the free energy states of the binding partners on the thermodynamic signature highlights that changing profiles might not necessarily reflect alterations in the ligand interaction or binding mode per se. Differences in protein flexibility even in remote regions of a protein might change the overall thermodynamic fingerprint as

Figure 3. Thermodynamic profiles of roscovitine-binding to CDK5 and the CDK5/p25 complex obtained by ITC at 25 °C. The error bars represent the error from the fitting of the raw data. The thermodynamic profiles in (a) indicate a significant change in the entropic contributions leading to an affinity increase for binding to the CDK5/p25 complex due to a reduced entropic penalty. (b) The structure of CDK5/p25100 shows that roscovitine (orange) binds on the opposite side of p25 (green) and that the CDK5 (magenta) complex adopts an active kinase conformation. Experimental details can be found in the Supporting Information.

exemplified by the entropy−enthalpy transduction phenomenon caused by conformational shifts.95 Keeping in mind that most target proteins are usually part of much larger multiprotein complexes as in this example, the required experimental reductionism as reflected by the use of specifically engineered and purified protein constructs can lead to a disconnect between a thermodynamic profile experienced under physiological conditions. Thermodynamic Profiling of β-Site APP Cleaving Enzyme 1 (BACE1) Inhibitors. BACE1 is a key player in the production of toxic amyloid beta clusters in the brain.101 Cerebral amyloid beta depositions is a strong hallmark of Alzheimer’s disease,102,103 and a large number of drug disovery projects have been targeting BACE1 to halt progression of the disease. We have also developed a number of promising BACE1 leads,104−106 one of which recently passed phase 1 clinical trials.107 During the course of optimization of these compounds, we measured the thermodynamic signatures of about 60 BACE1 leads using the same experimental setup and temperature. The identical experimental conditions of the measurements and the rigorous measurement of active compound concentrations make it a large consistent ITC data set that can be used for comparisons of binding enthalpy, binding entropy, and their correlation to other parameters, physicochemical properties, and drug discovery metrics. BACE1 features a shallow active site pocket containing the two catalytic residues Asp32 and Asp228, located between the BACE1 N- and C-terminal lobes.108 The pocket is sheltered by a long flexible β hairpin flap that partially covers the catalytic side chains. A second flexible loop located on the other side of the active site near Ser10 forms a deep water-filled crevice called the S3 subpocket. The Ser10 loop has been seen to adopt two distinct ligand-induced conformations connected to the intricate water network of the subpocket. Compounds that do not disturb the water network keep the pocket in its open conformation, while even a slight displacement of the top water 6327

DOI: 10.1021/jm501511f J. Med. Chem. 2015, 58, 6321−6335

Perspective

Journal of Medicinal Chemistry (A) will cause the pocket to collapse, releasing four well-defined waters. In addition, the size of the pocket indicates space for at least two more water molecules (Figure 4).106,109

Figure 4. Examples of the two BACE1 S3 subpocket conformations. (a) Open form: PDB IDs 4B78 1.5 Å, 4B72 1.6 Å, 4B77 1.8 Å,109 4B1D 1.95 Å, 4B1E 1.95 Å,106 and 4AZY 1.79 Å.110 (b) Closed form: PDB IDs 4B1C 1.95 Å,106 4B70 1.6 Å,109 4ACX 2.0 Å, 4ACU 1.75 Å,111 and 4B00 1.83 Å.110

The measured BACE1 ITC data spans a KD range from about 10 μM to 10 nM, with ΔHobs values from −11 to +1 kcal/mol and ΔSobs values from −8 to +3 kcal/mol, typical for a hit identification−early lead optimization campaign (Supporting Information, Table S2). As seen in many cases,112−114 the data set exhibits strong apparent enthalpy entropy compensation (Figure 5a). Part of this correlation might be attributed to the linear relationship between errors in ΔHobs and ΔSobs and the limited ΔGobs window from −11 to −7 kcal/mol. As discussed above, the consistent experimental conditions of all measurements, the determination of ligand concentrations by NMR, and good control over the ligand binding competency of the used BACE1 material are likely to limit the experimental errors in this study. However, plots of the experimentally determined ΔHobs and ΔGobs as well as of ΔS and ΔG values will be intrinsically less prone to introduce correlation artifacts (Figure 5b,c).115,116 The BACE1 data set consists of 28 compounds that do not enter the S3 subpocket and 30 compounds that do, releasing the contained waters. Dunitz estimated that the displacement of a single water molecule at a binding interface would give about 2 kcal/mol in entropy.117 These calculations did not consider the details of the interface or the possible interactions but gives an indication of the potential size of such an effect. Interestingly, the closure of the S3 subpocket and the displacement of the four plus two waters did not give rise to a strong entropic shift in any of the enthalpy vs entropy or entropy vs free energy plots (Figure 5a,c), and the open/closed compound does not show any clear clustering. As the shape of the closed cavity is very well-defined, it is tempting to speculate that the conformational thermodynamic shift when going from the open to the closed form might be similar for all compounds that close the pocket. This cannot be assumed for the different interactions of the ligands in the pocket, potentially masking an entropy-based desolvation signal. To further study the properties of the displaced waters, the open S3 subpocket was investigated through semicontinuum solvent analysis using SZMAP.118 Two of the displaced waters (B, C) (Figure 4) are located in the negative displaced neutral difference free energy areas and are coordinated by three hydrogen bonds each while two are close to zero (A, D) and are coordinated by two

Figure 5. BACE ITC data set. (a) ΔHobs vs −TΔSobs colored by open (red)/closed (blue) S3 subpocket. (b) ΔHobs vs ΔGobs colored by open/closed S3 subpocket. (c) −TΔSobs vs ΔGobs colored by open/ closed S3 subpocket. The experimental values of the ITC measurements can be found in Supplorting Information, Table 2. 6328

DOI: 10.1021/jm501511f J. Med. Chem. 2015, 58, 6321−6335

Perspective

Journal of Medicinal Chemistry

potentially be rationalized by the large conformational change of the Ser10 loop and the mixed enthalpic entropic profile of the displaced waters. However, if all closed pockets are equal, we would expect both these two effects to give a similar thermodynamic shift for all compounds, preserving a potential connection between ΔHobs and LLE. The lack of correlation of the closed compounds indicates that the individual interactions, the flexibility of the loop, and the various S3 sp phenyl meta substitutions cannot be captured by a crude single-parameter ΔH or LLE model. The different behavior thus indicates the change in binding mode, suggesting that a plot of enthalpy vs LLE might potentially be informative to highlight compounds that do not share the same mode of action. Thermodynamics of Thrombin−Inhibitor Binding. Thrombin (EC 3.4.21.5) is an interesting target for anticoagulant treatment due to its key role in the blood clotting cascade.119 Effective inhibitors of this enzyme have a tendency to form a very rapid enzyme−inhibitor complex, and the well-known inhibitor Melagatran is no exception.120 A kinetic and thermodynamic study including Melagatran and a series of five close analogues were recently published, serving as a representative case for a typical lead optimization program.91 To relate the changes in the structural features of the ligand to the observed differences in the thermodynamic profiles, structural information on all of the enzyme−inhibitor complexes were generated. The structures were used to gain detailed understanding of the specific interactions that are either formed or broken during the complex formation, including changes in the position/conformation of the ligands, the protein binding site, as well as in the organization of the surrounding solvent. The thermodynamic profiles for Melagatran and some close analogues were determined via a van’t Hoff analysis using affinity data from surface plasmon resonance (SPR) as well as from ITC. Despite the large heat capacity changes that have been observed in the ITC experiments, there was a striking similarity between the van’t Hoff enthalpy and the calorimetric enthalpy for all the tested inhibitors at 25 °C, indicating that good control over the experimental setup together with statistical rigor can lead to similar estimates. However, the large negative change in heat capacity led to interesting consequences when visualizing the thermodynamic data at different temperatures, as exemplified in Figure 7. Compound B appears to have a favorable entropic contribution to the binding at 20 °C (Figure 7 and Figure 8d). There is a much smaller enthalpic binding contribution, and such a thermodynamic profile might not have been considered as an attractive starting point given the original interpretation of thermodynamic signatures.6,8 Due to the large negative heat capacity change (ΔCp = −0.42 kcal/mol), this picture changes dramatically at 35 °C, potentially leading to a different assessment of the data. The profile is now transformed from an entropically to an enthalpically dominated binding profile with only a small entropic contribution. Such a profile might potentially appear as a more attractive starting point for lead optimization. However, given the frequent occurrence of nonzero heat capacity changes during ligand binding, enthalpydominated binding might be more the rule than an exception at physiological temperatures. When looking at the relationship between the crystal structures and the observed binding enthalpy ΔHobs for some more thrombin compounds at 35 °C, the complexity of data interpretation increases even further (Figure 8). As seen from

potential hydrogen bonds each (Supporting Information, Figure 1). This indicates that several of the waters will give significantly less entropic contribution upon desolvation than the classical 2 kcal/mol and some waters might even give an entropic penalty and an enthalpic gain, complicating the interpretation of the observed data. As mentioned earlier, several authors have argued for the existence of a correlation between LLE and the binding enthalpy.37,38 However, these correlations have either been based on very few data points or by combining the experimental data from many different sources and databases on different targets with unknown or poorly described experimental conditions. By comparing the ITC data of the compounds with an open BACE1 S3 subpocket, we find a decent correlation between LLE and the measured ΔHobs (Figure 6a), in line with previous studies. However, if we instead plot ΔHobs versus LLE of the closed S3 sp compounds, we find a strikingly poor correlation (Figure 6b). This could

Figure 6. Correlation between enthalpy and lipophilic ligand efficiency of BACE1 compounds in the open and closed S3 subpocket. (a) ΔHobs vs LLE(pKD) for the open S3 subpocket. (b) ΔHobs vs LLE(pKD) for the closed S3 subpocket. The experimental details for the ITC measurements can be found in Supporting Information, Table 2. 6329

DOI: 10.1021/jm501511f J. Med. Chem. 2015, 58, 6321−6335

Perspective

Journal of Medicinal Chemistry

differences around the carboxy- or amide-end of the P3 groups of the ligands induce changes in the thrombin binding site via different conformations of a Glu192 side chain. This has interesting consequences for the observed thermodynamics. When comparing Melagatran (Figure 8b), which has a carboxygroup at the P3 position, with compound A (Figure 8c), which contains a carboxy-amide, a significant reduction in the binding enthalpy can be observed. However, the introduction of the amide moiety enables the formation of a new hydrogen bond with Glu192 not present in the Melagatran structure, while maintaining the hydrogen bond to the backbone amide of Gly219. The additional hydrogen bond is achieved via a significant shift in the side chain position of Glu192, with the consequence of losing some solvent positions and thus a notable change in the water network. The changes lead to a decrease in the binding enthalpy and would thus leave the formation of the additional hydrogen bond undetected in the absence of structural information. Switching from the primary amide to the tertiary dimethyl-amide as in compound B results in the lost ability to form a hydrogen bond with Glu192 (Figure 8d). The bulky moiety forces Glu192 into one of the dual side chain conformations seen in the Melagatran complex and an additional water molecule can be observed at the position that was previously occupied by the second conformation seen in the Melagatran complex. As with the previous example, the totality of changes in the solvent structure as well as the structural changes within the binding site masks the breaking of this hydrogen bond and results in a slight increase in the binding enthalpy. The study emphasizes that interpreting thermodynamic measurements requires detailed information on both structure and solvent rearrangements as well as subtle changes in ligand orientation even within a series of very similar ligands.

Figure 7. Effect of a large negative heat capacity change on thermodynamic signatures for thrombin inhibitors. The data is taken from Winquist et al.91 and depicts the determined thermodynamic signature for compound B (see Figure 8d) at 20 and 35 °C. Error bars represent the mean standard error of duplicate measurements.

■

SUMMARY AND OUTLOOK It is often assumed that the net enthalpy contribution of a protein−ligand binding event is dominated by polar interactions, whereas the entropy of binding is dominated by desolvation effects. However, this assumption does not hold in general. The ordered distributed water networks are susceptible to small changes in ligand or protein makeup, leading to large differences in the observed thermodynamic behavior upon perturbation. Recent experiments and simulations have shown that solvent displacement and binding in hydrophobic cavities can be strongly enthalpically driven,54,58 contrary to the traditional picture of hydrophobic association. This could be attributed to the curvature of the binding pocket as well as the distribution of hydrogen bonding partners that restrict water from forming stable networks making it more vapor-like, as suggested in the study of McCammon et al.58 Water will shift its thermodynamic properties depending on its location from bulk or near macromolecular surfaces with large dependence on the polarity or shape of the binding cavities. Both the enthalpic and entropic character and strength of hydrogen bonds are more sensitive to the local environment than previously thought,18,20 making hydrophobicity less dominated by the classical clathrate picture.39 This suggests that trends in thermodynamic signatures across different protein target classes are generally less correlated with cavity desolvation than previously assumed.7 From a theoretical point of view, one can easily envisage the application of binding thermodynamics during lead selection and optimization in a rational drug design process. However,

Figure 8. Relationships between structural data and the observed binding enthalpy ΔHobs for Melagatran and close analogues in complex with thrombin. (a) Overlay of the six Melagatran analogues used showing only subtle changes in ligand orientation.91 (b) Structure of Melagatran, PDB ID 4BAH 1.94 Å. Please note the dual conformation of Glu192. (c) Structure of compound A, PDB ID 4BAO 1.87 Å. (d) Structure of compound B, PDB ID 4BAM 1.88 Å. Changes in the solvent inventory, the position of amino acid side chains in the binding site, as well as formation and breakage of hydrogen bonds result in unpredictable changes in the observed binding enthalpy. See also Supporting Information, Figure 2 for a 2-dimensional projection of the thrombin−ligand interactions for Melagatran, compound A, and compound B. All data is taken from ref 91 and represent the observed enthalpy data at 35 °C determined via duplicate measurements.

the structures, even small changes in the chemistry of a ligand can lead to significant changes in the interactions, even though the overall binding mode might appear conserved. The 6330

DOI: 10.1021/jm501511f J. Med. Chem. 2015, 58, 6321−6335

Perspective

Journal of Medicinal Chemistry

using thermodynamic data. Comparison of experimental ITC data with, e.g., LLE, simple solvent calculations or more rigorous free energy perturbations can enable the identification of compounds that do not behave as expected. Identifying and scrutinizing those outliers appears currently to be the most impactful use of thermodynamic profiling. The outliers could, as in the case of the enthalpy/entropy transduction argument used in the BPTI study, help to identify compound series that shift their binding mode, induce different motions in the target protein, or distinguish intramolecular hydrogen bonds from those between protein and ligand. A combination of calculations and experimental thermodynamic data might thus be more suitable to further guide optimization and has through the identification and exploitation of unexpected binding signatures the potential to significantly increase our understanding of the detailed parameters that govern protein−ligand binding.

due to the multifactorial character of the measured binding enthalpy, the individual contributions of intermolecular interactions like the formation of hydrogen bonds or ionic interactions are not direct experimental observables. This leads to the requirement of using structural information to postrationalize the observed binding enthalpies and thus puts the generation of thermodynamic data in question. However, this does not imply that ITC measurements have no applications in drug discovery. Target-based drug discovery relies heavily on high-quality protein reagents, and ITC offers an effective way to quickly assess protein quality, functionality, and functional concentration. As such, it is used to validate other biochemical assays, to screen smaller compound decks, or to establish structure−activity relationships. Finally it enables label-free affinity measurements and delivers important mechanistic information, as exemplified by the ability to characterize the binding of compounds to nonactivated kinases with little or no enzymatic activity.121,122 The inability to correctly interpret ITC data23 was nicely highlighted by a study on the extremely simple interaction of Ca2+-binding to EDTA.123 The addition of flexible chelating arms onto EDTA was surprisingly found to lead to an increase in affinity due to favorable entropic contributions. As a result, it was impossible to construct a simple, self-consistent thermodynamic binding model even for this very basic system. The recent studies of BPTI and MUP also show how painfully susceptible the energy landscape of folding or binding can be to small perturbations. Due to these complications, raw thermodynamic data as well as thermodynamically derived metrics like enthalpic efficiency and size-independent enthalpic efficiency (SIHE)3 should be used with great care. The multifactorial character of the enthalpy also questions some of the conclusions of the enthalpic size limit3,124 as well as a direct connection between observed enthalpy and the probabilistic complexity model.125 So can ligand binding thermodynamics still be regarded as a hot tip in drug discovery? No, not in a routine setting or with enthalpy and entropy regarded as isolated end points. Experimentally obtained thermodynamic data and crude derived parameters thereof are simply not well-suited to be used for direct red/green decision-making. It is not unlikely that some of the underlying parameters of the measured enthalpy and entropy might correlate with other interesting and relevant compound parameters. However, no such correlation has been convincingly shown thus far. From a thermodynamic point, LLE is more directly related to protein−ligand specificity than a predefined enthalpy/entropy balance. Hence LLE may be a more relevant parameter for risk estimation of general promiscuity and as such a more attractive target for optimization than other common metrics. One should bear in mind that LLE is not an end point in itself but needs to be related to the therapeutic dose and pharmacokinetic profile of a compound. A more optimal end point from a general promiscuity point of view would be to combine the predicted therapeutic dose with relative chemical potentials of the free drug in plasma. Although the potential for thermodynamic signatures as simple end points appear limited, it is likely that the combination of computer simulations, crystallography, and experimental thermodynamic data can contribute to the understanding of the forces that govern protein−ligand binding. The CDK5, thrombin, and BACE examples illustrate both some of the complications and a hint of the potential of

■

ASSOCIATED CONTENT

* Supporting Information S

Thermodynamic cycle and derivation of eq 14, ITC analysis of roscovitine-binding to CDK5 and CDK5/p25, ITC analysis of BACE1−inhibitor binding, SZMAP semicontinuum solvent analysis of the BACE1 S3 subpocket, two-dimensional projections of thrombin−ligand interactions for PDB structures 4BAH, 4BAO, and 4BAM. The Supporting Information is available free of charge on the ACS Publications website at DOI: 10.1021/jm501511f.

■

AUTHOR INFORMATION

Corresponding Authors

*For S.G.: phone, +46 31 7762197; E-mail, stefan. [email protected]. *For J.U.: phone, +46 31 7065207; E-mail, johan.ulander@ astrazeneca.com. *For P.J.: phone, +46 31 7064570; E-mail, patrik.johansson@ astrazeneca.com. Author Contributions

The manuscript was written through contributions of all authors. All authors have given approval to the final version of the manuscript. Notes

The authors declare no competing financial interest. Biographies Stefan Geschwindner is a Principal Scientist in Biophysics currently working within the Discovery Sciences department at AstraZeneca R&D Mölndal, Sweden. Prior to this, he received his Ph.D. in Biochemistry at the University of Frankfurt, Germany. His interests and current focus include drug discovery, lead generation, and optimization as well as label-free, affinity-based screening applications with a particular focus on biophysical descriptors of protein−ligand interactions. Johan Ulander works as Associate Principal Scientist in the computational chemistry section at Cardiovascular and Metabolic Diseases (CVMD) at AstraZeneca R&D Mölndal, Sweden. Prior to joining AstraZeneca, he did postdoctoral research at the University of California, San Diego (USCD), and University of Houston. He received his Ph.D. in theoretical physical chemistry from Gothenburg University and has a B.Sc. in Molecular Biology from the University of Umeå. He has 10 years of experience in drug discovery from early stage hit and target identification to late stage drug optimization. His 6331

DOI: 10.1021/jm501511f J. Med. Chem. 2015, 58, 6321−6335

Perspective

Journal of Medicinal Chemistry

(16) MacRaild, C. A.; Daranas, A. H.; Bronowska, A.; Homans, S. W. Global changes in local protein dynamics reduce the entropic cost of carbohydrate binding in the arabinose-binding protein. J. Mol. Biol. 2007, 368, 822−832. (17) Syme, N. R.; Dennis, C.; Bronowska, A.; Paesen, G. C.; Homans, S. W. Comparison of entropic contributions to binding in a “hydrophilic” versus “hydrophobic” ligand−protein interaction. J. Am. Chem. Soc. 2010, 132, 8682−8689. (18) Luck, W. A. P. The importance of cooperativity for the properties of liquid water. J. Mol. Struct. 1998, 448, 131−142. (19) Xantheas, S. S. Cooperativity and hydrogen bonding network in water clusters. Chem. Phys. 2000, 258, 225−231. (20) Nilsson, A.; Pettersson, L. G. M. Perspective on the structure of liquid water. Chem. Phys. 2011, 389, 1−34. (21) Whitesides, G. M.; Krishnamurthy, V. M. Designing ligands to bind proteins. Q. Rev. Biophys. 2005, 38, 385−395. (22) Schneider, G. Virtual screening: an endless staircase? Nature Rev. Drug Discovery 2010, 9, 273−276. (23) Chodera, J. D.; Mobley, D. L. Entropy−enthalpy compensation: role and ramifications in biomolecular ligand recognition and design. Annu. Rev. Biophys. 2013, 42, 121−142. (24) Dunitz, J. D. Win some, lose some: enthalpy−entropy compensation in weak intermolecular interactions. Chem. Biol. 1995, 2, 709−712. (25) Tellinghuisen, J.; Chodera, J. D. Systematic errors in isothermal titration calorimetry: concentrations and baselines. Anal. Biochem. 2011, 414, 297−299. (26) Cooper, A.; Johnson, C. M.; Lakey, J. H.; Nollmann, M. Heat does not come in different colours: entropy−enthalpy compensation, free energy windows, quantum confinement, pressure perturbation calorimetry, solvation and the multiple causes of heat capacity effects in biomolecular interactions. Biophys. Chem. 2001, 93, 215−230. (27) Cameron, D. L.; Jakus, J.; Pauleta, S. R.; Pettigrew, G. W.; Cooper, A. Pressure perturbation calorimetry and the thermodynamics of noncovalent interactions in water: comparison of protein−protein, protein−ligand, and cyclodextrin−adamantane complexes. J. Phys. Chem. B 2010, 114, 16228−16235. (28) Cooper, A. Protein heat capacity: an anomaly that maybe never was. J. Phys. Chem. Lett. 2010, 1, 3298−3304. (29) Ward, W. H.; Holdgate, G. A. Isothermal titration calorimetry in drug discovery. Prog. Med. Chem. 2001, 38, 309−376. (30) Holdgate, G. A. Making cool drugs hot: isothermal titration calorimetry as a tool to study binding energetics. BioTechniques 2001, 31, 164−166 168, 170 passim.. (31) Li, L.; Dantzer, J. J.; Nowacki, J.; O’Callaghan, B. J.; Meroueh, S. O. PDBcal: a comprehensive dataset for receptor−ligand interactions with three-dimensional structures and binding thermodynamics from isothermal titration calorimetry. Chem. Biol. Drug Des. 2008, 71, 529− 532. (32) Harper, E. A.; Black, J. W. Histamine H3-receptor agonists and imidazole-based H3-receptor antagonists can be thermodynamically discriminated. Br. J. Pharmacol. 2007, 151, 504−517. (33) Horn, J. R.; Russell, D.; Lewis, E. A.; Murphy, K. P. Van’t Hoff and calorimetric enthalpies from isothermal titration calorimetry: are there significant discrepancies? Biochemistry 2001, 40, 1774−1778. (34) Mizoue, L. S.; Tellinghuisen, J. Calorimetric vs. van’t Hoff binding enthalpies from isothermal titration calorimetry: Ba2+−crown ether complexation. Biophys. Chem. 2004, 110, 15−24. (35) Naghibi, H.; Tamura, A.; Sturtevant, J. M. Significant discrepancies between van’t Hoff and calorimetric enthalpies. Proc. Natl. Acad. Sci. U. S. A. 1995, 92, 5597−5599. (36) Liu, Y.; Sturtevant, J. M. Significant discrepancies between van’t Hoff and calorimetric enthalpies. III. Biophys. Chem. 1997, 64, 121− 126. (37) Hopkins, A. L.; Keseru, G. M.; Leeson, P. D.; Rees, D. C.; Reynolds, C. H. The role of ligand efficiency metrics in drug discovery. Nature Rev. Drug Discovery 2014, 13, 105−121.

interests include theoretical biophysics and statistical mechanics with applications in drug design, pharmacokinetics, and dynamics. Patrik Johansson holds an MSc in computational physics and a Ph.D. in molecular biology from Uppsala University and joined the Structure & Biophysics group at AstraZeneca R&D Mölndal, Sweden, as a protein crystallographer. Prior to this, he was a postdoctoral fellow at the Max-Planck Institute for Biochemistry in Munich, working on structure−function studies of the FAS type I multisynthase in the group of Dieter Oesterhelt. At AstraZeneca, he has focused on early stage hit finding, lead generation, and structure-based drug design and has a keen interest in the combination of experimental data and molecular modeling in drug discovery.

■

ABBREVIATIONS USED HTS, high-throughput screening; SPR, surface plasmon resonance; ITC, isothermal titration calorimetry; EEC, enthalpy−entropy compensation; MUP, mouse major urinary protein; CDK5, cyclin-dependent kinase 5; LI, lead identification; LO, lead optimization; LLE, lipophilic ligand efficiency; HCA, human carbonic anhydrase; BACE1, β-site APP cleaving enzyme 1; BPTI, bovine pancreatic trypsin inhibitor

■

REFERENCES

(1) Holdgate, G. A. Thermodynamics of binding interactions in the rational drug design process. Expert Opin. Drug Discovery 2007, 2, 1103−1114. (2) Freire, E. A thermodynamic approach to the affinity optimization of drug candidates. Chem. Biol. Drug Des. 2009, 74, 468−472. (3) Ferenczy, G. G.; Keseru, G. M. Enthalpic efficiency of ligand binding. J. Chem. Inf. Model. 2010, 50, 1536−1541. (4) Garbett, N. C.; Chaires, J. B. Thermodynamic studies for drug design and screening. Expert Opin. Drug Discovery 2012, 7, 299−314. (5) Klebe, G. Applying thermodynamic profiling in lead finding and optimization. Nature Rev. Drug Discovery 2015, 14, 95−110. (6) Freire, E. Do enthalpy and entropy distinguish first in class from best in class? Drug Discovery Today 2008, 13, 869−874. (7) Olsson, T. S.; Williams, M. A.; Pitt, W. R.; Ladbury, J. E. The thermodynamics of protein−ligand interaction and solvation: insights for ligand design. J. Mol. Biol. 2008, 384, 1002−1017. (8) Ladbury, J. E.; Klebe, G.; Freire, E. Adding calorimetric data to decision making in lead discovery: a hot tip. Nature Rev. Drug Discovery 2010, 9, 23−27. (9) Kawasaki, Y.; Freire, E. Finding a better path to drug selectivity. Drug Discovery Today 2011, 16, 985−990. (10) Muzammil, S.; Armstrong, A. A.; Kang, L. W.; Jakalian, A.; Bonneau, P. R.; Schmelmer, V.; Amzel, L. M.; Freire, E. Unique thermodynamic response of tipranavir to human immunodeficiency virus type 1 protease drug resistance mutations. J. Virol. 2007, 81, 5144−5154. (11) Carbonell, T.; Freire, E. Binding thermodynamics of statins to HMG-CoA reductase. Biochemistry 2005, 44, 11741−11748. (12) Ferenczy, G. G.; Keseru, G. M. Thermodynamics guided lead discovery and optimization. Drug Discovery Today 2010, 15, 919−932. (13) Weiland, G. A.; Minneman, K. P.; Molinoff, P. B. Fundamental difference between the molecular interactions of agonists and antagonists with the beta-adrenergic receptor. Nature 1979, 281, 114−117. (14) Wittmann, H. J.; Seifert, R.; Strasser, A. Contribution of binding enthalpy and entropy to affinity of antagonist and agonist binding at human and guinea pig histamine H(1)-receptor. Mol. Pharmacol. 2009, 76, 25−37. (15) Daranas, A. H.; Shimizu, H.; Homans, S. W. Thermodynamics of binding of D-galactose and deoxy derivatives thereof to the Larabinose-binding protein. J. Am. Chem. Soc. 2004, 126, 11870−11876. 6332

DOI: 10.1021/jm501511f J. Med. Chem. 2015, 58, 6321−6335

Perspective

Journal of Medicinal Chemistry (38) Shultz, M. D. The thermodynamic basis for the use of lipophilic efficiency (LipE) in enthalpic optimizations. Bioorg. Med. Chem. Lett. 2013, 23, 5992−6000. (39) Kauzmann, W. Some factors in the interpretation of protein denaturation. Adv. Protein Chem. 1959, 14, 1−63. (40) Ruelle, P.; Kesselring, U. W. The hydrophobic effect. 1. A consequence of the mobile order in H-bonded liquids. J. Pharm. Sci. 1998, 87, 987−997. (41) Jelesarov, I.; Bosshard, H. R. Isothermal titration calorimetry and differential scanning calorimetry as complementary tools to investigate the energetics of biomolecular recognition. J. Mol. Recognit. 1999, 12, 3−18. (42) Solomonov, B. N.; Sedov, I. A. Quantitative description of the hydrophobic effect: the enthalpic contribution. J. Phys. Chem. B 2006, 110, 9298−9303. (43) Ladbury, J. E. Just add water! The effect of water on the specificity of protein−ligand binding sites and its potential application to drug design. Chem. Biol. 1996, 3, 973−980. (44) Huggins, M. L. Thermodynamic properties of solutions of longchain compounds. Ann. N. Y. Acad. Sci. 1942, 43, 1−32. (45) Flory, P. J.; Krigbaum, W. R. Thermodynamics of high polymer solutions. Annu. Rev. Phys. Chem. 1951, 2, 383−402. (46) Mannhold, R.; Van De Waterbeemd, H. Substructure and whole molecule approaches for calculating log P. J. Comput.-Aided Mol. Des. 2001, 15, 337−354. (47) Mannhold, R.; Poda, G. I.; Ostermann, C.; Tetko, I. V. Calculation of molecular lipophilicity: state-of-the-art and comparison of log P methods on more than 96000 compounds. J. Pharm. Sci. 2009, 98, 861−893. (48) Wan, H.; Ulander, J. High-throughput pKa screening and prediction amenable for ADME profiling. Expert Opin. Drug Metab. Toxicol. 2006, 2, 139−155. (49) Ben-Amotz, D.; Underwood, R. Unraveling water’s entropic mysteries: a unified view of nonpolar, polar, and ionic hydration. Acc. Chem. Res. 2008, 41, 957−967. (50) Vaitheeswaran, S.; Yin, H.; Rasaiah, J. C.; Hummer, G. Water clusters in nonpolar cavities. Proc. Natl. Acad. Sci. U. S. A. 2004, 101, 17002−17005. (51) Young, T.; Abel, R.; Kim, B.; Berne, B. J.; Friesner, R. A. Motifs for molecular recognition exploiting hydrophobic enclosure in protein−ligand binding. Proc. Natl. Acad. Sci. U. S. A. 2007, 104, 808−813. (52) Heugen, U.; Schwaab, G.; Brundermann, E.; Heyden, M.; Yu, X.; Leitner, D. M.; Havenith, M. Solute-induced retardation of water dynamics probed directly by terahertz spectroscopy. Proc. Natl. Acad. Sci. U. S. A. 2006, 103, 12301−12306. (53) Ebbinghaus, S.; Kim, S. J.; Heyden, M.; Yu, X.; Heugen, U.; Gruebele, M.; Leitner, D. M.; Havenith, M. An extended dynamical hydration shell around proteins. Proc. Natl. Acad. Sci. U. S. A. 2007, 104, 20749−20752. (54) Ernst, J. A.; Clubb, R. T.; Zhou, H. X.; Gronenborn, A. M.; Clore, G. M. Demonstration of positionally disordered water within a protein hydrophobic cavity by NMR. Science 1995, 267, 1813−1817. (55) Denisov, V. P.; Carlstrom, G.; Venu, K.; Halle, B. Kinetics of DNA hydration. J. Mol. Biol. 1997, 268, 118−136. (56) Soper, A. K.; Finney, J. L. Hydration of methanol in aqueous solution. Phys. Rev. Lett. 1993, 71, 4346−4349. (57) Olano, L. R.; Rick, S. W. Hydration free energies and entropies for water in protein interiors. J. Am. Chem. Soc. 2004, 126, 7991−8000. (58) Setny, P.; Baron, R.; McCammon, J. A. How can hydrophobic association be enthalpy driven? J. Chem. Theory Comput. 2010, 6, 2866−2871. (59) Denisov, V. P.; Venu, K.; Peters, J.; Hörlein, H. D.; Halle, B. Orientational disorder and entropy of water in protein cavities. J. Phys. Chem. B 1997, 101, 9380−9389. (60) Sharrow, S. D.; Novotny, M. V.; Stone, M. J. Thermodynamic analysis of binding between mouse major urinary protein-I and the pheromone 2-sec-butyl-4,5-dihydrothiazole. Biochemistry (N. Y.) 2003, 42, 6302−6309.

(61) Bingham, R. J.; Findlay, J. B.; Hsieh, S. Y.; Kalverda, A. P.; Kjellberg, A.; Perazzolo, C.; Phillips, S. E.; Seshadri, K.; Trinh, C. H.; Turnbull, W. B.; Bodenhausen, G.; Homans, S. W. Thermodynamics of binding of 2-methoxy-3-isopropylpyrazine and 2-methoxy-3isobutylpyrazine to the major urinary protein. J. Am. Chem. Soc. 2004, 126, 1675−1681. (62) Malham, R.; Johnstone, S.; Bingham, R. J.; Barratt, E.; Phillips, S. E.; Laughton, C. A.; Homans, S. W. Strong solute−solute dispersive interactions in a protein−ligand complex. J. Am. Chem. Soc. 2005, 127, 17061−17067. (63) Barratt, E.; Bingham, R. J.; Warner, D. J.; Laughton, C. A.; Phillips, S. E.; Homans, S. W. Van der Waals interactions dominate ligand−protein association in a protein binding site occluded from solvent water. J. Am. Chem. Soc. 2005, 127, 11827−11834. (64) Barratt, E.; Bronowska, A.; Vondrasek, J.; Cerny, J.; Bingham, R.; Phillips, S.; Homans, S. W. Thermodynamic penalty arising from burial of a ligand polar group within a hydrophobic pocket of a protein receptor. J. Mol. Biol. 2006, 362, 994−1003. (65) Chervenak, M. C.; Toone, E. J. A direct measure of the contribution of solvent reorganization to the enthalpy of ligand binding. J. Am. Chem. Soc. 1994, 116, 10533−10539. (66) Roy, J.; Laughton, C. A. Long-timescale molecular-dynamics simulations of the major urinary protein provide atomistic interpretations of the unusual thermodynamics of ligand binding. Biophys. J. 2010, 99, 218−226. (67) Wolynes, P. G.; Eaton, W. A. The physics of protein folding. Phys. World 1999, 12, 39−44. (68) Onuchic, J. N.; Socci, N. D.; Luthey-Schulten, Z.; Wolynes, P. G. Protein folding funnels: the nature of the transition state ensemble. Folding Des. 1996, 1, 441−450. (69) Nussinov, R.; Wolynes, P. G. A second molecular biology revolution? The energy landscapes of biomolecular function. Phys. Chem. Chem. Phys. 2014, 16, 6321−6322. (70) Zhuravlev, P. I.; Papoian, G. A. Protein functional landscapes, dynamics, allostery: a tortuous path towards a universal theoretical framework. Q. Rev. Biophys. 2010, 43, 295−332. (71) Zhuravlev, P. I.; Papoian, G. A. Functional versus folding landscapes: the same yet different. Curr. Opin. Struct. Biol. 2010, 20, 16−22. (72) Gilson, M. K.; Zhou, H. Calculation of Protein−Ligand Binding Affinities. Annu. Rev. Biophys. Biomol. Struct. 2007, 36, 21−42. (73) Zhou, H.; Gilson, M. K. Theory of free energy and entropy in noncovalent binding. Chem. Rev. 2009, 109, 4092−4107. (74) Phillip Bowen, J.; Güner, O. F. A perspective on quantum mechanics calculations in ADMET predictions. Curr. Top. Med. Chem. 2013, 13, 1257−1272. (75) Piana, S.; Klepeis, J. L.; Shaw, D. E. Assessing the accuracy of physical models used in protein-folding simulations: quantitative evidence from long molecular dynamics simulations. Curr. Opin. Struct. Biol. 2014, 24, 98−105. (76) Shaw, D. E.; Maragakis, P.; Lindorff-Larsen, K.; Piana, S.; Dror, R. O.; Eastwood, M. P.; Bank, J. A.; Jumper, J. M.; Salmon, J. K.; Shan, Y.; Wriggers, W. Atomic-level characterization of the structural dynamics of proteins. Science 2010, 330, 341−346. (77) Zwanzig, R. W. High-temperature equation of state by a perturbation method. I. Nonpolar gases. J. Chem. Phys. 1954, 1420− 1426. (78) Hansen, N.; Van Gunsteren, W. F. Practical aspects of freeenergy calculations: a review. J. Chem. Theory Comput. 2014, 10, 2632−2647. (79) Barducci, A.; Bonomi, M.; Parrinello, M. Metadynamics. Wiley Interdiscip. Rev.: Comput. Mol. Sci. 2011, 1, 826−843. (80) Gallicchio, E.; Levy, R. M. Recent theoretical and computational advances for modeling protein−ligand binding affinities. Adv. Protein Chem. Struct. Biol. 2011, 85, 27−80. (81) Gumbart, J. C.; Roux, B.; Chipot, C. Standard binding free energies from computer simulations: what is the best strategy? J. Chem. Theory Comput. 2013, 9, 794−802. 6333

DOI: 10.1021/jm501511f J. Med. Chem. 2015, 58, 6321−6335

Perspective

Journal of Medicinal Chemistry (82) Rosta, E.; Hummer, G. Free energies from dynamic weighted histogram analysis using unbiased Markov state model. J. Chem. Theory Comput. 2015, 11, 276−285. (83) Chipot, C. Frontiers in free-energy calculations of biological systems. Wiley Interdiscip. Rev.: Comput. Mol. Sci. 2014, 4, 71−89. (84) Gilson, M. K.; Fenley, A. T.; Muddana, H. Bridging simulations and calorimetry: computational studies of binding thermodynamics and entropy−enthalpy transduction. Biophys. J. 2014, 106, 658a. (85) Zhuravlev, P. I.; Wu, S.; Potoyan, D. A.; Rubinstein, M.; Papoian, G. A. Computing free energies of protein conformations from explicit solvent simulations. Methods 2010, 52, 115−121. (86) Potoyan, D. A.; Zhuravlev, P. I.; Papoian, G. A. Computing free energy of a large-scale allosteric transition in adenylate kinase using all atom explicit solvent simulations. J. Phys. Chem. B 2012, 116, 1709− 1715. (87) Lumry, R.; Rajender, S. Enthalpy−entropy compensation phenomena in water solutions of proteins and small molecules: a ubiquitous property of water. Biopolymers 1970, 9, 1125−1227. (88) Krug, R. R.; Hunter, W. G.; Grieger, R. A. Enthalpy−entropy compensation. 2. Separation of the chemical from the statistical effect. J. Phys. Chem. 1976, 80, 2341−2351. (89) Sharp, K. Entropy−enthalpy compensation: fact or artifact? Protein Sci. 2001, 10, 661−667. (90) Horn, J. R.; Brandts, J. F.; Murphy, K. P. Van’t Hoff and calorimetric enthalpies II: effects of linked equilibria. Biochemistry (N. Y.) 2002, 41, 7501−7507. (91) Winquist, J.; Geschwindner, S.; Xue, Y.; Gustavsson, L.; Musil, D.; Deinum, J.; Danielson, U. H. Identification of structural−kinetic and structural−thermodynamic relationships for thrombin inhibitors. Biochemistry 2013, 52, 613−626. (92) Breiten, B.; Lockett, M. R.; Sherman, W.; Fujita, S.; Al-Sayah, M.; Lange, H.; Bowers, C. M.; Heroux, A.; Krilov, G.; Whitesides, G. M. Water networks contribute to enthalpy/entropy compensation in protein−ligand binding. J. Am. Chem. Soc. 2013, 135, 15579−15584. (93) Olsson, T. S. G.; Ladbury, J. E.; Pitt, W. R.; Williams, M. A. Extent of enthalpy−entropy compensation in protein−ligand interactions. Protein Sci. 2011, 20, 1607−1618. (94) Lockett, M. R.; Lange, H.; Breiten, B.; Heroux, A.; Sherman, W.; Rappoport, D.; Yau, P. O.; Snyder, P. W.; Whitesides, G. M. The binding of benzoarylsulfonamide ligands to human carbonic anhydrase is insensitive to formal fluorination of the ligand. Angew. Chem., Int. Ed. Engl. 2013, 52, 7714−7717. (95) Fenley, A. T.; Muddana, H. S.; Gilson, M. K. Entropy−enthalpy transduction caused by conformational shifts can obscure the forces driving protein−ligand binding. Proc. Natl. Acad. Sci. U. S. A. 2012, 109, 20006−20011. (96) Cheung, Z. H.; Fu, A. K.; Ip, N. Y. Synaptic roles of Cdk5: implications in higher cognitive functions and neurodegenerative diseases. Neuron 2006, 50, 13−18. (97) Monaco, E. A., III. Recent evidence regarding a role for Cdk5 dysregulation in Alzheimer’s disease. Curr. Alzheimer Res. 2004, 1, 33− 38. (98) Tarricone, C.; Dhavan, R.; Peng, J.; Areces, L. B.; Tsai, L. H.; Musacchio, A. Structure and regulation of the CDK5−p25(nck5a) complex. Mol. Cell 2001, 8, 657−669. (99) Ahn, J. S.; Radhakrishnan, M. L.; Mapelli, M.; Choi, S.; Tidor, B.; Cuny, G. D.; Musacchio, A.; Yeh, L. A.; Kosik, K. S. Defining Cdk5 ligand chemical space with small molecule inhibitors of tau phosphorylation. Chem. Biol. 2005, 12, 811−823. (100) Mapelli, M.; Massimiliano, L.; Crovace, C.; Seeliger, M. A.; Tsai, L. H.; Meijer, L.; Musacchio, A. Mechanism of CDK5/p25 binding by CDK inhibitors. J. Med. Chem. 2005, 48, 671−679. (101) Hardy, J. A.; Higgins, G. A. Alzheimer’s disease: the amyloid cascade hypothesis. Science 1992, 256, 184−185. (102) Haass, C.; Selkoe, D. J. Soluble protein oligomers in neurodegeneration: lessons from the Alzheimer’s amyloid betapeptide. Nature Rev. Mol. Cell Biol. 2007, 8, 101−112. (103) Walsh, D. M.; Selkoe, D. J. A beta oligomersa decade of discovery. J. Neurochem. 2007, 101, 1172−1184.

(104) Edwards, P. D.; Albert, J. S.; Sylvester, M.; Aharony, D.; Andisik, D.; Callaghan, O.; Campbell, J. B.; Carr, R. A.; Chessari, G.; Congreve, M.; Frederickson, M.; Folmer, R. H.; Geschwindner, S.; Koether, G.; Kolmodin, K.; Krumrine, J.; Mauger, R. C.; Murray, C. W.; Olsson, L. L.; Patel, S.; Spear, N.; Tian, G. Application of fragment-based lead generation to the discovery of novel, cyclic amidine beta-secretase inhibitors with nanomolar potency, cellular activity, and high ligand efficiency. J. Med. Chem. 2007, 50, 5912− 5925. (105) Geschwindner, S.; Olsson, L. L.; Albert, J. S.; Deinum, J.; Edwards, P. D.; de Beer, T.; Folmer, R. H. Discovery of a novel warhead against beta-secretase through fragment-based lead generation. J. Med. Chem. 2007, 50, 5903−5911. (106) Gravenfors, Y.; Viklund, J.; Blid, J.; Ginman, T.; Karlstrom, S.; Kihlstrom, J.; Kolmodin, K.; Lindstrom, J.; von Berg, S.; von Kieseritzky, F.; Bogar, K.; Slivo, C.; Swahn, B. M.; Olsson, L. L.; Johansson, P.; Eketjall, S.; Falting, J.; Jeppsson, F.; Stromberg, K.; Janson, J.; Rahm, F. New aminoimidazoles as beta-secretase (BACE-1) inhibitors showing amyloid-beta (Abeta) lowering in brain. J. Med. Chem. 2012, 55, 9297−9311. (107) Alexander, R.; Budd, S.; Russell, M.; Kugler, A.; Cebers, G.; Ye, N.; Olsson, T.; Burdette, D.; Maltby, J.; Paraskos, J.; Elsby, K.; Han, D.; Goldwater, R.; Ereshefsky, L. AZD3293 A novel BACE1 inhibitor: safety, tolerability, and effects on plasma and CSF Aβ peptides following single- and multiple-dose administration. Neurobiol. Aging 2014, 35, S2. (108) Hong, L.; Koelsch, G.; Lin, X.; Wu, S.; Terzyan, S.; Ghosh, A. K.; Zhang, X. C.; Tang, J. Structure of the protease domain of memapsin 2 (β-secretase) complexed with inhibitor. Science 2000, 290, 150−153. (109) Ginman, T.; Viklund, J.; Malmström, J.; Blid, J.; Emond, R.; Forsblom, R.; Johansson, A.; Kers, A.; Lake, F.; Sehgelmeble, F.; Sterky, K. J.; Bergh, M.; Lindgren, A.; Johansson, P.; Jeppsson, F.; Fälting, J.; Gravenfors, Y.; Rahm, F. Core refinement toward permeable β-secretase (BACE-1) inhibitors with low hERG activity. J. Med. Chem. 2013, 56, 4181−4205. (110) Swahn, B. M.; Kolmodin, K.; Karlstrom, S.; von Berg, S.; Soderman, P.; Holenz, J.; Berg, S.; Lindstrom, J.; Sundstrom, M.; Turek, D.; Kihlstrom, J.; Slivo, C.; Andersson, L.; Pyring, D.; Rotticci, D.; Ohberg, L.; Kers, A.; Bogar, K.; von Kieseritzky, F.; Bergh, M.; Olsson, L. L.; Janson, J.; Eketjall, S.; Georgievska, B.; Jeppsson, F.; Falting, J. Design and synthesis of beta-site amyloid precursor protein cleaving enzyme (BACE1) inhibitors with in vivo brain reduction of beta-amyloid peptides. J. Med. Chem. 2012, 55, 9346−9361. (111) Swahn, B. M.; Holenz, J.; Kihlstrom, J.; Kolmodin, K.; Lindstrom, J.; Plobeck, N.; Rotticci, D.; Sehgelmeble, F.; Sundstrom, M.; Berg, S.; Falting, J.; Georgievska, B.; Gustavsson, S.; Neelissen, J.; Ek, M.; Olsson, L. L.; Berg, S. Aminoimidazoles as BACE-1 inhibitors: the challenge to achieve in vivo brain efficacy. Bioorg. Med. Chem. Lett. 2012, 22, 1854−1859. (112) Krishnamurthy, V. M.; Bohall, B. R.; Semetey, V.; Whitesides, G. M. The paradoxical thermodynamic basis for the interaction of ethylene glycol, glycine, and sarcosine chains with bovine carbonic anhydrase II: an unexpected manifestation of enthalpy/entropy compensation. J. Am. Chem. Soc. 2006, 128, 5802−5812. (113) Reynolds, C. H.; Holloway, M. K. Thermodynamics of ligand binding and efficiency. ACS Med. Chem. Lett. 2011, 2, 433−437. (114) Brandt, T.; Holzmann, N.; Muley, L.; Khayat, M.; WegscheidGerlach, C.; Baum, B.; Heine, A.; Hangauer, D.; Klebe, G. Congeneric but still distinct: how closely related trypsin ligands exhibit different thermodynamic and structural properties. J. Mol. Biol. 2011, 405, 1170−1187. (115) Exner, O. Entropy−enthalpy compensation and anticompensation: solvation and ligand binding. Chem. Commun. 2000, 1655− 1656. (116) Beasley, J. R.; Doyle, D. F.; Chen, L.; Cohen, D. S.; Fine, B. R.; Pielak, G. J. Searching for quantitative entropy−enthalpy compensation among protein variants. Proteins: Struct., Funct., Genet. 2002, 49, 398−402. 6334

DOI: 10.1021/jm501511f J. Med. Chem. 2015, 58, 6321−6335

Perspective

Journal of Medicinal Chemistry (117) Dunitz, J. D. The entropic cost of bound water in crystals and biomolecules. Science 1994, 264, 670. (118) Grant, J. A. A smooth permittivity function for Poisson− Boltzmann solvation methods. J. Comput. Chem. 2001, 22, 608−640. (119) Nutescu, E. A.; Wittkowsky, A. K. Direct thrombin inhibitors for anticoagulation. Ann. Pharmacother. 2004, 38, 99−109. (120) Elg, M.; Gustafsson, D.; Deinum, J. The importance of enzyme inhibition kinetics for the effect of thrombin inhibitors in a rat model of arterial thrombosis. Thromb. Haemostasis 1997, 78, 1286−1292. (121) Sullivan, J. E.; Holdgate, G. A.; Campbell, D.; Timms, D.; Gerhardt, S.; Breed, J.; Breeze, A. L.; Bermingham, A.; Pauptit, R. A.; Norman, R. A.; Embrey, K. J.; Read, J.; VanScyoc, W. S.; Ward, W. H. Prevention of MKK6-dependent activation by binding to p38alpha MAP kinase. Biochemistry 2005, 44, 16475−16490. (122) VanScyoc, W. S.; Holdgate, G. A.; Sullivan, J. E.; Ward, W. H. Enzyme kinetics and binding studies on inhibitors of MEK protein kinase. Biochemistry 2008, 47, 5017−5027. (123) Christensen, T.; Gooden, D. M.; Kung, J. E.; Toone, E. J. Additivity and the physical basis of multivalency effects: a thermodynamic investigation of the calcium EDTA interaction. J. Am. Chem. Soc. 2003, 125, 7357−7366. (124) Hann, M. M.; Keseru, G. M. Finding the sweet spot: the role of nature and nurture in medicinal chemistry. Nature Rev. Drug Discovery 2012, 11, 355−365. (125) Hann, M. M.; Leach, A. R.; Harper, G. Molecular complexity and its impact on the probability of finding leads for drug discovery. J. Chem. Inf. Comput. Sci. 2001, 41, 856−864.

6335

DOI: 10.1021/jm501511f J. Med. Chem. 2015, 58, 6321−6335