Thermodynamic profiling for fragment-based lead discovery and optimization

ABSTRACT

Introduction: The enthalpic and entropic components of the ligand–protein binding free energy reflect the type and quality of the interactions and relate to the physicochemical properties of the ligands. These findings have significance in medicinal chemistry optimizations since they suggest that the thermodynamic profiling of the binding may help monitor and control the unfavorable size and hydrophobicity increase typically accompanying affinity improvements and leading to suboptimal pharmacokinetic properties.

Areas covered: This review describes the ligand–protein binding event in terms of elementary steps, their associated interactions, and their enthalpic and entropic consequences. The relationships among the breaking and forming interactions, the binding thermodynamic profile, and the physicochemical properties of the ligands are also discussed.

Expert opinion: Analysis of the size dependence of available affinity and favorable enthalpy highlights the limitation of the simultaneous optimization of these quantities. Indeed, moderate, rather than very high affinities can be conciliated with favorable physicochemical and pharmacokinetic profiles as it is supported by the affinity range of historical oral drugs. Although thermodynamic quantities are not suitable endpoints for medicinal chemistry optimizations owing to the complexity of the binding thermodynamics, thermodynamic profiling together with structural studies can be advantageously used to understand the details of the binding process and to optimize it.

KEYWORDS:

• Binding thermodynamics
• enthalpy driven optimization
• entropy driven optimization
• fragment binding
• fragment optimization

1. Introduction

A principal objective of early phase medicinal chemistry optimizations in drug discovery programs is to increase the binding affinity of ligands toward their protein targets. Binding affinity is directly related to the free energy change accompanying ligand–protein binding. The relationship between affinity and the enthalpy and entropy content of binding free energy can be conveniently expressed in terms of the quantities in Eq. (1) [1]

(1) $p{K}_d=p{K}_H+p{K}_S$(1)

where $p{K}_d=-log{K}_d$ with $K_d$ being the dissociation constant of the ligand–protein complex (more precisely it has to be replaced by $\frac{K_d}{1M}$), and $p{K}_H=\frac{-\mathrm{\Delta }H}{2.303RT}$ and $p{K}_S=\frac{\mathrm{\Delta }S}{2.303R}$. $p{K}_H$ measures the enthalpic and $p{K}_S$ the entropic contribution to the binding affinity. The transformation of enthalpy ($\mathrm{\Delta }H$) and entropy ($\mathrm{\Delta }S$) changes into $p{K}_H$ and $p{K}_S$, respectively, has the advantage that affinity ($p{K}_d$) and the thermodynamic quantities ($p{K}_H$, $p{K}_S$) are expressed at comparable scales and higher values corresponds to more favorable contributions.

The recognition that binding thermodynamics relates to the optimization and to the properties of drugs and drug candidates first appears in a series of papers by Freire and coworkers published from the turn of the millennium [2–5]. The enthalpic and entropic contributions to ligand–protein binding were determined for HIV inhibitors, and later for HMG-CoA reductase inhibitors, and it was shown that compounds that come later into the market tend to have increasing enthalpic contributions. This finding can be attributed to the challenge and longer timeframe of optimizing ligand affinity toward its target protein by increasing the favorable enthalpic ($p{K}_H$) rather than the entropic ($p{K}_S$) contribution of binding. It was also recognized that compounds with enthalpy dominated binding tend to be more drug like by having favorable physicochemical and pharmacokinetic properties [5–7].

The thermodynamic characterization of ligand–protein binding is most often performed with isothermal titration calorimetry (ITC) measurements [8,9]. They directly provide binding enthalpy and dissociation constants that can be used to obtain free energy and entropy, and thus full thermodynamic profile of the binding. Measuring the temperature dependence of the dissociation constant and using the van’t Hoff equation offers another way to determine thermodynamic quantities [10]. However, ITC is considered to be the preferred method as it gives thermodynamic data from a single titration, it is a label-free technique, and it can be applied at a wide range of binding affinities. Its principal limitation is high protein requirement, however, significant progress has been achieved to decrease protein consumption, increase throughput [11,12] and to make ITC applicable to membrane proteins [13]. It should be noted that measured enthalpies depend on conditions like temperature, pH and buffer, and they may need corrections when ligand–protein binding is linked to protonation or ion binding [14].

Thermodynamic profiling of ligand–protein binding has generated a large amount of data that are useful for recognizing trends in compound property changes in drug discovery programs and to directly support medicinal chemistry optimizations.

2. Thermodynamics of the ligand–protein binding process

In principle, affinity ($p{K}_d$) optimization can be achieved by increasing $p{K}_H$, $p{K}_S$, or both. Here, we have to mention enthalpy–entropy compensation [15] that is a general phenomenon and tends to compensate enthalpy and entropy changes resulting in a modest change in $p{K}_d$. The basic challenge in affinity optimizations can be formulated as to overcome enthalpy–entropy compensation to achieve increased $p{K}_d$. Ligand–protein binding in water is a complex process with various competing enthalpic and entropic contributions [5–7]. Binding requires at least a partial desolvation of both the ligand and the protein, potential conformational changes of both partners, and the formation of ligand–protein interactions. The desolvation of polar groups has a large enthalpic penalty [16] that can only be compensated by good quality ligand–protein interactions. The emphasis here is on the good quality, since especially polar interactions are highly sensitive to the relative geometry of the interacting partners. Therefore, the exchange of ligand–water and protein–water hydrogen bonds to ligand–protein and water–water hydrogen-bonds may eventually be unfavorable and may not yield $p{K}_H$ increase. In contrast, the desolvation of apolar ligand and protein surfaces that become buried in the ligand–protein complex is typically favorable owing to the $p{K}_S$ increase attributed to the disappearance of the polar water–apolar solute interactions. Therefore, it is usually a feasible strategy to increase $p{K}_S$ by building into the ligand nonpolar groups with good shape complementarity to the protein binding pocket. Another optimization technique is to fix the ligand conformation in the geometry of the bound ligand and thus moderating configuration entropy decrease upon ligand binding. These considerations provide rationale for the ease of entropic optimization compared to the enthalpic optimization.

The quantification of the above effects is not straightforward. The association of free energy changes of solvation with surfaces is an approximation that does not take into account atomic details. This is the reason why apolar and polar surface burial give modest correlation with binding free energy and with its components [17]. Moreover, the large variation of free energy changes associated with surface burial as obtained with various models [17–21] suggests model dependence.

A part of conformational entropy decrease upon ligand binding originates from the more restricted variation of torsion angles in the complex compared to the solvated ligand and protein. The loss of conformational entropy is estimated to be around 3 kJ/mol at 300 K for each rotatable bond fixed in the complex [22]. The association of two species, the protein and the ligand, results in a loss of rigid body translational and rotational entropy and its estimated magnitude is around 15–20 kJ/mol at 300 K [23]. These values, however, are crude estimations and the actual magnitude of the entropy loss is affected by the flexibility of the complex; entropy may be associated with the new vibrational modes of the complex.

Ligand–protein binding free energies can be calculated with molecular dynamics based methods with an accuracy of 5–10 kJ/mol in favorable cases [24,25]. Atomic level simulations are best suited for the calculation of binding free energy differences between ligands. The calculation of enthalpic and entropic components, however, faces serious convergence difficulties. Particular attention has been devoted to the calculation of the thermodynamic quantities associated with the perturbation of the binding site water network upon ligand binding. The estimation of the enthalpy and entropy assigned to individual water molecules by the WaterMap method [26] is based on molecular dynamics simulations and the application of the inhomogeneous solvation theory [27]. The complementary WaterFLAP approach [28] assigns enthalpy and entropy by empirical scoring of the water network built using GRID [29] water hotspots and optimized by short molecular dynamics simulations. Computations may support the interpretation of binding thermodynamic data and the design of ligands with improved properties [30–32].

Valuable insight into the binding process can be obtained by combined structural and thermodynamic analyses of a set of related complexes. These analyses are able to reveal how various effects like the water mediated ligand–protein binding [33], the displacement or trapping of water molecules upon ligand binding [34–36], and the interplay between hydrogen-bonding and hydrophobic contacts [37] influence the thermodynamic balance of binding. Ligand binding may be accompanied with protonation/deprotonation processes whose thermodynamic consequences are pH and buffer dependent, and their dissection leads to intrinsic thermodynamic parameters most relevant in drug design [38,39]. These studies highlight the complexity of ligand binding thermodynamics and show that it is often difficult to pinpoint what the origins of the changes in entropy and enthalpy are [40,41]. Owing to their multifactorial character and non-additivity, thermodynamic quantities cannot be used as direct endpoints, but they are useful in analyzing and supporting the optimization of specific systems [42–44].

In spite of the above difficulties in attributing accurate thermodynamic values to the various effects accompanying ligand–protein complex formation, the dissection of the process into elementary steps, and the association of approximate enthalpy and entropy values to these steps supports the identification of useful relationships between thermodynamics and the quality of compounds in the drug discovery process, on one hand, and lead identification and optimization, on the other hand. Both aspects will be discussed in the forthcoming sections.

3. Thermodynamics and drug-like properties

Achieving affinity improvement is a major objective of early phase medicinal chemistry optimizations; however, the optimization of other properties becomes increasingly important at later stages. Physicochemical properties and the related pharmacokinetic profile are important characteristics of drug candidates, and their simultaneous optimization with affinity is the main challenge in drug discovery programs. Following the early recognition of the relation among affinity, drug-like properties and binding thermodynamics [5–7], significant effort has been devoted to explore, understand and take advantage of this relationship in drug discovery projects.

It is well documented that medicinal chemistry optimizations are typically accompanied with unfavorable changes of physicochemical properties [45–47]. Compounds tend to be larger and increasingly hydrophobic with improving affinity. This is, however, not beneficial, since large and hydrophobic compounds have unfavorable physicochemical properties and are more prone to suboptimal ADME profile as it was shown by the comparison of compounds from various stages of the discovery process [45,48–51]. It was also shown that property inflation is a general phenomenon and is basically independent from the hit identification strategy, rather it could be traced back to optimization practices [7]. It is common to achieve affinity improvements by adding lipophilic groups to facilitate desolvation, and by limiting ligand flexibility using chain to ring replacements typically with the introduction of predominantly apolar groups. Both tactics yield molecular weight and lipophilicity increase, and both contribute to the affinity improvement by increasing binding entropy. Therefore, entropic optimization is often associated with unfavorable ADME properties. In contrast, enthalpic optimization attempts to improve affinity by building into the ligand functional groups with optimal directional interactions to the protein target. This approach can avoid the excessive increase of lipophilicity and the deterioration of ADME properties; therefore, it is considered to be preferable.

Affinity optimization by extending compounds with new groups of atoms is in line with the observation that maximal available affinity increases with the number of heavy atoms. It was shown by Kuntz et al. that for small molecules a sharp increase of available affinity can be achieved with the number of heavy atoms, while the increase is modest for larger molecules [52]. Similar trend was found for compounds with binding thermodynamic data [1]. The maximal $p{K}_D$values as a function of heavy atom count tend to increase as it is shown in Figure 1(a).

Figure 1. (a) Logarithmic plot of affinity (pK_d[max]) of the most potent ligands versus the number of heavy atoms (HAC). (b) Logarithmic plot of maximal favorable enthalpy (pK_H[max]; black round markers and straight line), and enthalpy of the most potent ligands (pK_H[pK_dmax]; red square markers and dashed line) versus the number of heavy atoms (HAC). Data were binned by ΔHAC = 3 and maximal values in each bin are shown. Based on data from Ref. [1].

A similar plot of the maximal available binding enthalpy as a function of molecular size expressed as heavy atom count shows the opposite trend in Figure 1(b). This plot was obtained with over 800 compounds [1] including natural biological ligands and synthetic compounds from medicinal chemistry programs with a large variety of protein targets. The plot in Figure 1(b) implies that there is a limit for available favorable binding enthalpy, and this limit decreases with ligand size. Moreover, the binding enthalpies of maximal affinity ligands ($p{K}_H\left[p{K}_{dmax}\right]$) also show a decreasing trend with ligand size, and they are consistently lower than the available maximal favorable enthalpy ($p{K}_H\left[max\right]$) values (Figure 1(b)). These findings suggest that this is not only the medicinal chemistry optimization practice that hampers enthalpic optimization, but there is an inherent limit for the simultaneous optimization of both affinity and enthalpy. The reconciliation of the dilemma between affinity and enthalpy optimization is suggested by the observation that historical oral drugs show median pXC50 (negative logarithm of IC₅₀ or EC₅₀) value of ˜7.7 (˜50 nM) [53,54] showing that moderate, rather than very high affinity can be combined with favorable ADME properties to produce compounds with high therapeutic value.

4. Thermodynamics and fragment-based drug discovery

Fragment-based drug discovery (FBDD) is an approach for identifying chemical starting points for drug discovery programs [55–57]. FBDD starts with the screening of fragment-sized compounds, often defined as molecules with less than 23 heavy atoms or with molecular weight less than 300 Da. Fragment hits typically bind with high ligand efficiency (LE; affinity per heavy atom) and can be optimized into leads and drug candidates with large operational freedom owing to their small size and advantageous physicochemical properties. A further advantage of fragment screening is the better sampling of the chemical space compared to the high throughput screening of lead-like compounds.

The small size and limited number of pharmacophore elements of fragments allow binding with few optimal interactions, albeit with low affinity. A rationale for the success of the fragment approach is given by the complexity model [47,58] claiming that the chance of forming useful interactions falls with increasing ligand complexity. Analyses of fragment interactions in experimental protein complexes show that binding occurs at the hot spot of the protein that is a polar site buried in a lipophilic environment [59–61], where fragments typically bind with few polar interaction [60], frequently with optimal geometry hydrogen-bonds [59]. In contrast, the contribution of apolar desolvation and associated free energy gain with predominantly entropic origin is limited owing to the small size of fragments [59,61]. Therefore, the formation of polar interactions provides favorable enthalpy that is able to compensate for the unfavorable rigid body translational and rotational entropy loss upon binding [23] and results in enthalpy dominated fragment binding as it was shown by the thermodynamic analysis of a large number of fragment–protein complexes [61–63].

The few dominantly polar interactions formed by fragments can be assigned to pharmacophore patterns, and the screening of ultra-low-molecular-weight compounds [64] and minimal pharmacophores [55] has the promise of improved exploration of the binding site and better coverage of the chemical space. Although we are not aware of thermodynamic data for these very small fragments, we anticipate that their binding is enthalpy dominated and they are well suited for enthalpy driven optimization. It has to be noted, however, that in spite of the beneficial properties of fragments, their successful optimization into advantageous leads and drug candidates is not guarantied. Historical data show that fragment and HTS derived lead properties are not necessarily different [7], and the inherent difficulty of enthalpy driven optimization of larger compounds as it is illustrated in Figure 1(b) prevents late stage enthalpic optimization. Nevertheless, early optimizations of fragment hits offer the opportunity for the simultaneous improvement of affinity and binding enthalpy as it was shown by the analysis of the thermodynamic data of Astex drug discovery programs [61]. The combined application of structural and thermodynamic information primarily by X-ray and ITC, respectively, was found to be the key for successful early optimizations [61,65]. These considerations make the use of thermodynamic data particularly fruitful in fragment-based approaches.

5. Case studies for thermodynamics supported fragment optimizations

Table 1 lists publicly available thermodynamic data of fragment–protein binding. These data provide the basis for analyzing how thermodynamic quantities depend on the properties of the binding partners [61,62]. These data combined with structural information for a set of related complexes are extremely useful in understanding the details of the binding process as it is exemplified with the case studies presented in the forthcoming sections.

Table 1. Reported experimental thermodynamic data for fragment–protein binding.

Target	Reference
Aldose Reductase	[66]
Beta-Glucosidase	[67]
Carbonic Anhydrase II	[36,42,68–70]
Enoyl-ACP Reductase	[71]
Farnesyl Diphosphate Synthase	[72]
Fatty Acid-Binding Protein	[73]
HSP90	[95]
Major Urinary Protein	[74–77]
MMP12	[93]
p38a	[78,79]
Pantothenate Synthetase	[80,81]
PIM-1 Kinase	[82]
PLP-dependent Transaminase	[83]
Polyandrocarpa Lectin	[84]
Protein Kinase A	[40]
Quinone Reductase 2	[85]
Ribonuclease A (Bovine)	[86]
Streptavidin	[87]
Thrombin	[37,88]
Thymidine Kinase	[89]
Transcriptional Regulator TtgR	[90]
Trypsin	[91,92]

5.1. Matrix metalloproteinase (MMP12) inhibitors

One of the first studies on the impact of binding thermodynamics on fragment optimization has been published by Bertini et al. [93]. The authors investigated the optimization of low affinity fragments to nanomolar MMP12 inhibitors by evaluating the key compounds along the optimization path by biochemical and ITC measurements, and X-ray crystallography. Matrix metalloproteinases are zinc containing proteases. Most of the high affinity MMP inhibitors form direct interactions with the zinc ion located at the substrate binding groove. Due to the chelating properties of hydroxamic acids, the identification of the low affinity and nonspecific acetohydroxamic acid (1) was not unexpected. In fact, 1 binds to MMP12 with a K_d of 6.18 mM, its binding mode has been confirmed by X-ray crystallography and the subsequent ITC analysis revealed this fragment as a high enthalpy binder. The first optimization step was realized by a linking strategy connecting the acetyl group of 1 to the terminal amino group of the 4-methoxy-benzenesulfonamide (2) that binds to the S1’ pocket of MMP12 (Figure 2). This modification resulted in a dramatic improvement in the potency that was increased by five orders of magnitude, reaching the K_d of 61 nM. Analyzing the thermodynamic profiles of the linked fragments, the observed improvement in K_d can be traced back to the much improved binding entropy of 3.

Figure 2. (a) Optimization of MMP12 starting fragments using the fragment linking strategy. (b) Thermodynamic profiles of the starting fragments and the linked compound.

Linking the fragments, they lose their translational and rotational degree of freedom and the related entropy penalty of the linked compound contributes to its increased binding affinity. In this particular case, the improvement in the binding affinity (ΔΔG = −15.2 kJ/mol) was basically due to this effect (−ΔTΔS = −17.4 kJ/mol) while enthalpic components did not contribute significantly (ΔΔH = 2.8 kJ/mol). Based on its binding affinity and thermodynamic profile, compound 3 was selected for further optimization. In the first round the 4-methoxy group of 3 has been replaced to 4-fluoro (4) and 4-phenyl (5) substituents (Figure 3). The introduction of the 4-fluoro group did not improved the affinity (K_d = 65 nM); however, the 4-phenyl derivative (5) was more potent (K_d = 2 nM) than 3. Comparative analysis of the corresponding X-ray structures revealed that the 4-methoxyphenyl group dives deeper into the S1’ pocket as compared to the 4-fluorophenyl moiety that probably could displace less water molecules at the binding site. On the other hand, the additional phenyl ring of 5 provides new sites for hydrophobic interactions. The thermodynamic profiles of compounds 3–5 are in line with these structural findings (Figure 3).

Figure 3. Optimization of the aromatic substituent in 3. The diagrams show the change of the thermodynamic quantities (ΔΔG: change in the binding free energy, ΔΔH: change in the binding enthalpy and −ΔTΔS: change in the binding entropy) along the optimization step.

Replacement of the 4-methoxy substituent to 4-fluoro was associated with some improvement in the binding entropy that was compensated by the enthalpic penalty in line with decreased occupancy of the S1’ site. Introducing the apolar phenyl group to position 4; however, resulted in improvements in both the enthalpic and the entropic components. Owing to its larger size, the phenyl group fills the S1’ site better that contributes to the slight improvement to the binding enthalpy. The most significant factor, however, is the increase of entropic reward that may be attributed, at least partly, to improved desolvation.

In the second round of the optimization, the authors investigated the effect of substituents introduced to the sulfonamide nitrogen atom. Having the best balance between its potency and the physico-chemical and thermodynamic profiles, the sulfonamide nitrogen of the 4-methoxy analogue (3) was substituted by an isobutyl (6), a 2-hydroxyethyl (7), and a 2,3-dihydroxypropyl (8) groups (Figure 4). All of the N-substituents resulted in improvements in the binding affinity, K_d values reached the single digit nanomolar range (K_d = 4 nM, 8 nM, and 8 nM, respectively). The highest affinity has been achieved by alkyl group (iBu in compound 6) while alcohols (7 and 8) showed modest improvement. Analysis of the X-ray structures suggests that except for the isobutyl group, the terminal hydroxyl group of the alcohols forms interactions with water molecules located at the entrance of the substrate binding site. Thermodynamic profiling of compounds 6–8 provided more details on the energetics of ligand binding. ITC measurements showed that the N-alkyl derivative (6) gains binding free energy from the entropic reward potentially originate from its improved desolvation. Contrary, the introduction of hydroxyl substituents provided more polar interactions and therefore improved the affinity via the corresponding enthalpic reward. Comparing, however, the thermodynamic profiles of the alcohol (7) and the diol (8), it seems that the enthalpic reward achieved in 7 has been largely compensated by an entropic penalty. Interestingly, the profile of the diol (8) suggests much less gain in the binding enthalpy and some reward in the binding entropy. The decreased enthalpy contribution can be rationalized by the less favored desolvation of the diol.

Figure 4. Optimization of the sulfonamide nitrogen substituent in 3. The diagrams show the change of the thermodynamic quantities (ΔΔG: change in the binding free energy, ΔΔH: change in the binding enthalpy and −ΔTΔS: change in the binding entropy) along the optimization step.

In summary, the optimization of MMP12 inhibitors provided a good example for analyzing the thermodynamic character of the two most important fragment optimization strategies, linking and growing. While linking is mostly entropically driven, there are multiple opportunities for fragment growing. Structural studies and the parallel analysis of the thermodynamic profiles allow understanding the changes in the binding mode and energetics along the optimization path. These data together might help optimizing protein–ligand interactions while keeping the physico-chemical properties under control.

5.2. Carbonic anhydrase (CA) inhibitors

Thermodynamic characterization is now a routine label-free methodology for the investigation of protein ligand interactions in industry settings. One of the first studies of this type has been published by a Pfizer team on carbonic anhydrase ligands [42]. Profiling 17 substituted benzene sulfonamides (Figure 5) the authors identified compounds with better affinity than the unsubstituted core. In addition to the 4-chloro (K_d = 141 nM) and 4-cyano (K_d = 142 nM) derivatives they found that both the 2-fluoro (K_d = 314 nM) and 3-fluoro (K_d = 118 nM) analogues showed submicromolar affinity.

Figure 5. Thermodynamic profiling of benzene sulfonamides on carbonic anhydrase.

Comparing the thermodynamic profiles of the most active compounds it has been suggested that the significantly improved potency observed for the 4-Cl and 3-F compounds is driven by a favorable entropic component. Interestingly, the thermodynamic profile of the 2-fluoro-benzenesulfonamide showed significant improvement in the binding enthalpy. Considering only potency data, the 2-fluoro derivative is inferior; however, the more enthalpic character of this compound suggests improved polar interactions with the protein binding site. In fact, X-ray crystallography of the 2-fluoro analogue cocrystallized with CA revealed a specific interaction formed between the 2-fluoro group and the backbone NH of Thr200. In contrast, 3-fluoro, 4-chloro and 4-cyano substituents are all located close to the hydrophobic part of the binding site (Figure 6).

Figure 6. Experimental binding modes of the 2-fluoro and 3-fluoro benzenesulfonamides at the binding site of human carbonic anhydrase. The 2-fluoro group points toward the hydrophilic wall (green) of the binding site. Contrary, the 3-fluoro group is located in a dominantly hydrophobic subpocket that is preferred by all the other halogens and also the nitriles used at different positions. Copyright: Royal Society of Chemistry [94] https://pubs.rsc.org/en/content/chapter/bk9781849733533-00023/978-1-84973-353-3.

Based on the crystallographic and thermodynamic analysis, the 2-fluoro derivative forms more directed interactions with the target that provides a stable binding mode necessary for further optimization of the initial fragment. Growing the 2-fluoro and 3-fluoro derivatives with a 4-benzylamide moiety (Figure 7) resulted in a significant increase in the binding affinity of the former fragment (55 fold, K_d = 5.7 nM). The affinity improvement was much less for the 3-fluoro analogue (6 fold, K_d = 20 nM).

Figure 7. Fragment growing applied for the optimization of 2-fluoro and 3-fluoro benzenesulfonamides against human carbonic anhydrase. The thermodynamic profile of the optimized 2-fluoro fragment preserved its enthalpic character.

Focusing to the thermodynamic profiles of the optimized fragments, both bindings are enthalpy driven. The addition of the p-benzylamide group increased the enthalpy content of the 2-fluoro derivative more significantly than that observed for the corresponding 3-fluoro analogue. Furthermore, the same optimization strategy turned the originally less active 2 fluoro fragment into a more potent lead than that obtained from the 3-fluoro-benzenesulfonamide fragment.

In summary, this study shows that evaluation of the thermodynamic profiles of prioritized fragment sets would provide useful information to pick up the most promising fragment hits for further optimizations. Structural and thermodynamic characterization of the fragment ligand complex might help to understand the driving forces behind the binding that points farther than the affinity alone. Highly enthalpic fragments form good quality polar interactions at the binding site including H-bonds with the target and structural waters. Thus, the stable binding mode might provide better starting point for further optimization.

5.3. Heat shock protein 90 (HSP90) inhibitors

One of the first Astex program entered into the clinical phase aimed the discovery of HSP90 inhibitors for oncology indications. This program was initiated by the X-ray and NMR screening of the corporate fragment library that provided two series of hits. One of these series was aminopyrimidines exemplified by 9 that was picked up as a potential starting point [95]. Despite of its moderate affinity (K_d = 240 μM), 9 has a stable binding mode in a deep pocket of HSP90 and forming good quality H-bonds with Asp93 and four structural waters located at the binding site (Figure 8). Comparing the binding mode of 9 to that of ADP revealed the overlapping position of the aminopyrimidine and the adenosine rings and suggested to extend 9 toward the small hydrophobic pockets available at the close proximity. Another important feature observed in the binding mode of 9 was that the pyridine ring is twisted around the biaryl axis. Replacing the pyridine ring with an ortho methoxyphenyl substituent preserved this twisted conformation. Furthermore, the methyl group introduced at position 2 of the pyrimidine ring fits well to one of the hydrophobic pockets (Figure 8). These changes led to compound 10 showing 60 fold improvement in the binding affinity (K_d = 4 μM).

Figure 8. Binding modes of compounds 9 and 10 obtained by X-ray crystallography. Structural waters are indicated by red spots.

In the next step of the optimization, the 2-methyl substituent at the aminopyperidine core has been replaced by a chlorine (Figure 9, compound 11) that improved the affinity further 12-fold reaching the submicromolar range (K_d = 0.35 μM).

Figure 9. Thermodynamic profiling of the fragments optimized against HSP90 at Astex.

Figure 10. Thermodynamic profiling of protein kinase A ligands obtained by systematic growing of the starting fragment (13).

Comparing the thermodynamic profiles [96] of 10 and 11 there is a significant improvement in the binding enthalpy (ΔΔH = −7.1 kJ/mol) that suggests improved polar interactions to the target. In fact, the X-ray structure of the HSP90-11 complex revealed the formation of a new halogen bond to Gly97 NH that was kept after the introduction of another methoxy group at position 6 of its 2-methoxyphenyl ring (Figure 9). This step yielded compound 12 with a K_d value of 0.07 μM that represents a 5-fold improvement in the binding affinity relative to that of 11. Analysis of the corresponding thermodynamic profiles suggests this transformation being entropy driven (−ΔTΔS = −10.9 kJ/mol). The entropic character of this step could be readily interpreted by the limited rotational freedom of the 2,6-dimethoxyphenyl ring in solution that results in more beneficial changes in the conformational entropy of 12.

This study demonstrates that careful optimization of the initial fragment hit might provide new polar interactions with the target that improves the binding affinity significantly and the meantime might stabilize the binding mode of the fragment. Structural studies with the parallel evaluation of binding thermodynamics could help identifying the key contacts. This information is specifically useful for late stage optimizations focusing other properties but keeping the beneficial interaction profile to maintain the binding affinity. This optimization also demonstrates the role of restricted ligand flexibility, a well-known optimization strategy applied in many medicinal chemistry programs. Again, structural and thermodynamic analyses help identifying the prioritized conformation of the ligand and the reward realized in the conformational component of the binding entropy.

5.4. Protein kinase A (PKA) ligands

In a recent publication, Klebe and coworkers investigated the thermodynamics of fragment growing using protein kinase A as a model target [40]. Their starting point was isoquinoline-5-sulfonamide (13) that binds to PKA with a K_d of 6.2 μM. Fragment growing was realized in three subsequent steps adding an N-methyl (14), an N-aminoethyl (15), and an N-propylaminoethyl (16) substituent to the sulfonamide group.

The thermodynamic profile of this small series of compounds was recorded and each of the ligands was cocrystallized with the target. Since all the ligands of this dataset are derived by systematic construction of the N-substituent of the sulfonamide moiety, it is therefore specifically useful to study the structural and thermodynamic consequences of fragment growing. However, it should be noted that in real life fragment optimizations growing strategies involve multiple often parallel and sometimes dependent growing vectors.

Analyzing the thermodynamic profiles, it is interesting to see that binding enthalpies become less favorable while the entropy rewards are increasing with the increasing size of the ligands (Figure 10). The increasing number of polar atoms and the increasing number of rotatable bonds from 13 to 16, however, suggest inverse trends for both the enthalpy and the entropy of binding, respectively. Based on the comparative analysis of the corresponding X-ray structures the authors provided reasonable explanations for these observations (Figure 11).

Figure 11. Superposition of the crystal structures obtained for ligands 13–16. Orientations of sulfonamide oxygens indicate the different binding modes of 13 and 14 vs. 15 and 16. As a consequence, the Gly-rich loop was found in closed and open conformation, respectively.

Analyzing the ligand conformations obtained in the binding pocket of PKA, it turned out that the binding modes of 13 and 14 are almost identical. Interestingly, however, the poses of the larger compounds, 15 and 16, obtained by the growing strategy were found to be significantly different. The sulfonamide group was rotated by 80 degrees relative to the conformation found in 13 and 14 that resulted in a completely different interaction pattern for the smaller (13 and 14) and the larger ligands (15 and 16). In the case of 13 and 14, the NH of the sulfonamide forms a H-bond to Glu127. In contrast, the larger N-substituents of 15 and 16 prevent the formation of this H-bond and the sulfonamide NH instead forms interactions with Thr51 of the Gly-rich loop. This H-bond reorganizes the Gly-rich loop that adopts a closed conformation as opposed by the open conformation found in the complexes of 13 and 14. The largely changed interaction patterns might thus explain the decreased enthalpic reward realized in the complex formation of the larger ligands. Entropic effects were rationalized by the analysis of ligand interactions formed with binding site water molecules. Comparative analysis of structural waters identified at the binding pocket revealed that the terminal propyl group of 16 displaced a water molecule present in the crystal structure of 13, 14, and 15 complexes since releasing this water to the bulk improves the binding entropy. Although the analysis of water networks was limited by the quality of the structures molecular dynamics simulations suggested similar mechanisms behind the entropic preference of 15 over 14 binding.

Despite that fragment growing follows different and more rational ways in drug discovery programs, this study shed light to the complexity of structure–thermodynamics relationship. ITC measures the net values of binding thermodynamics that is impacted by the interaction partners, including the protein target, the ligand, and the water molecules at the binding site. In addition to direct protein–ligand contacts, differences in the solvation patterns influence the thermodynamic profile significantly. Together with induced fit effects and conformational rearrangements occurred upon binding, high resolution X-ray crystallography and protein observed NMR measurements might help understanding and interpreting the binding thermodynamics of protein ligands.

6. Conclusion

The analysis of ligand–protein binding thermodynamics revealed links between the thermodynamic profiles and the physicochemical and pharmacokinetic properties of ligands. The opposite trends in available affinity and favorable enthalpy change with ligand size have been recognized and interpreted. They are in line with the demonstrated enthalpy dominated binding of fragment-sized compounds. Moreover, fragments are well suited for enthalpy driven early optimizations using binding thermodynamic and structural data that help understanding ligand binding and support medicinal chemistry optimizations to provide drug candidates with favorable pharmacokinetic profiles.

7. Expert opinion

Ligand–protein binding is associated with free energy change whose enthalpic and entropic components reflect the type and quality of interactions. As a general principle, the formation of polar, directional interactions between the ligand and the protein contributes to favorable enthalpy, while the desolvation of apolar surfaces of the interacting partners is the major source of favorable entropy. This implies that ligands with appropriate polar groups have the chance of enthalpy driven binding. However, optimal polar interactions with substantial enthalpy gain are restricted to a small range of the relative positions of the interacting partners. In contrast, the entropy gain achieved with apolar interactions are less sensitive to the geometrical parameters and the introduction of apolar groups in the ligand offers a more facile way to improve the binding free energy. This has the consequence that affinity increase is easier to achieve via entropic, rather than enthalpic optimizations [5–7]. This, however, has an unfavorable impact on the drug like properties of ligands, since large and apolar compounds tend to have suboptimal pharmacokinetic parameters [45–47]. The recognized relationship between the binding thermodynamic profile and the physicochemical and pharmacokinetic parameters suggests that monitoring and controlling binding thermodynamics may help improving the quality of drug candidates.

Analysis of a large amount of binding thermodynamic data revealed that maximal ligand affinity increases with ligand size, however, the favorable enthalpy tends to decrease with ligand size. Moreover, it has been shown that compounds with optimal affinity tend to have lower than maximal favorable binding enthalpy [1]. These findings point out the limitation of the simultaneous optimization of affinity and enthalpy and suggest that moderate, rather than very high affinities can be conciliated with favorable physicochemical and pharmacokinetic profiles. This is also reflected by the observation that historical oral drugs tend to have affinity in the double digit nanomolar range [53,54].

The thermodynamic profiling of ligand binding in medicinal chemistry optimizations is most often used together with X-ray crystallography based atomic resolution structural analysis. Studies of congeneric series reveal important details like the role of water molecules [33–36] and the interplay between hydrogen-bonding and hydrophobic contacts [37]. They highlight the complexity of ligand binding thermodynamics, and point out that thermodynamic quantities cannot be used as direct endpoints, but they are useful in analyzing and supporting the optimization of specific systems [42,43].

Theoretical considerations suggest and experimental data confirm that fragment-sized compounds exhibit enthalpy driven binding to proteins [59,61,62]. Moreover, fragments are well suited for the optimization of polar interactions when supported by thermodynamic profiling and structural studies [61]. These considerations represent the thermodynamic rationale for FBDD. Monitoring the thermodynamic profile is therefore beneficial in fragment-based optimization programs. Published case studies confirmed that thermodynamic characterization is useful analyzing the key interactions and rationalizing the binding mode. These pieces of information were found to be crucial in successful fragment optimization programs.

Article Highlights

• Ligand–protein binding thermodynamics is related to the type and quality of interactions

• Optimal geometry polar interactions provide favorable binding enthalpy

• The desolvation of apolar groups is the major source of favorable binding entropy

• Enthalpy, rather than entropy driven optimizations tend to provide compounds with favorable pharmacokinetic profiles

• Enthalpic optimizations are more challenging than entropic optimizations

• Binding thermodynamic profiling with structural studies supports medicinal chemistry optimizations

• Available ligand affinity shows increasing, and favorable enthalpy shows decreasing trend with ligand size

• Enthalpy dominated binding is a general feature of fragment-sized compounds

• Fragments are well suited for enthalpy driven early optimizations

This box summarizes key points contained in the article.

Declaration of interest

The authors have no other relevant affiliations or financial involvement with any organization or entity with a financial interest in or financial conflict with the subject matter or materials discussed in the manuscript apart from those disclosed.

Reviewer Disclosures

Peer reviewers on this manuscript have no relevant financial or other relationships to disclose.

Additional information

Funding

This work was supported by the Hungarian Brain Research Program under Grant 2017-1.2.1-NKP-2017-00002.