The binding of two biomolecules viewed from the atomic level is highly complex. It involves the formation or removal of many individual non-covalent bonds both between the interacting molecules as well as with solvent. Currently, our understanding of the thermodynamic quantification of biomolecular interactions is somewhat naïve. ITC (isothermal titration calorimetry) provides a rapid route to a full thermodynamic characterization of a biomolecular interaction. Armed with these data, what are we really able to understand about complex formation and can any of this information provide a useful tool to aid drug development? Correlations between thermodynamic data and structural detail have been investigated, allowing insight into ways in which these can be used to understand protein–ligand interactions and provide input into the decision-making process in drug development.
- biomolecular interaction
- drug design
- enthalpic efficiency
- isothermal titration calorimetry (ITC)
- lead optimization
Measurement of thermodynamic parameters has become a common addition to the arsenal of biophysical data used in characterization of biomolecular interactions. However, in many cases, this additional information does not really add to our understanding of the biological system or enhance our knowledge of strategies for potential pharmacological intervention. This is largely down to our naïvity in understanding how the global thermodynamic parameters measured in an experiment can be attributed to the different structural events which occur on going from the free to the bound state. In the present overview, the value of this ‘coarse grained’ thermodynamic parameterization is discussed with respect to understanding biological systems and its adoption in drug discovery.
Biology is based on the study of defined interactions between macromolecules such as proteins, DNA and lipids. Therefore, to understand biology, we must have a way of quantifying the driving forces which make such interactions occur. Thermodynamic parameterization provides one form of quantification. By assuming the interaction occurs under equilibrium conditions, we can provide a measure of the ‘tightness’ or ‘affinity’ of the interaction. A further level of information can be gleaned by determining the underlying changes in enthalpy and entropy (ΔH and ΔS respectively) for the interaction. The instrument of choice which provides opportunity for determination of these parameters in one experiment is the isothermal titration calorimeter (for reviews on the method, see [1–7]).
ITC (isothermal titration calorimetry) provides a direct measurement of the heat associated with forming or breaking non-covalent bonds as molecules form a complex. The direct measurement of this heat (or change in enthalpy, ΔHobs) is unique to calorimetric methods. Other methods require indirect calculation of this term via the van 't Hoff relationship. Since the ΔHobs obtained is proportional to the number of molecules going from the free to the bound state, this term also provides a probe of the extent of interaction as one component is added to another throughout the course of a titration. From this, the equilibrium binding constant, KB (=1/Kd, where Kd is the dissociation constant) can be determined. Armed with the ΔHobs and the KB, the ΔGobs and ΔSobs can be calculated from the following relationships and hence a complete thermodynamic characterization of the interaction can be obtained: (1) (2)It should be noted that the calorimetric experiment does not just include measurement of the heat associated with the formation of a complex, but also the heat associated with other potential accompanying events such as solvent rearrangement, protonation/deprotonation and conformational changes in the interacting molecules. To highlight this fact, the subscript ‘obs’ is commonly used.
ITC experiments cover an optimum range of dissociation constants of 100 μM to 10 nM; however, modification of experimental design using competition methods can be used to extend this range significantly to both weaker and stronger affinities. Furthermore, ITC is not limited by requirements of immobilization of interacting components or chemical modification.
Having determined the thermodynamic parameters, how can they inform us about an interaction? The ΔGobs reports on the affinity (being derived from the equilibrium binding constant). The ΔHobs is the heat associated with the making and breaking of non-covalent bonds on going from the free to the bound state. Thus, for example, an interaction in which a net increase in the number of hydrogen bonds of the system occurs would be accompanied by a favourable (i.e. negative) ΔHobs. One has to bear in mind that, often when a hydrogen bond is made between biomolecules, it is replacing bonds that were made previously with solvent water and hence the net enthalpic effect of the bond on complex formation is reduced due to desolvation.
The ΔSobs reports on the overall change in the degrees of freedom of a system. For the most part, this can be largely associated with the effect of liberating water molecules from the surfaces of the binding biomolecules as they are buried in the complex. Interaction with a biomolecular surface causes water molecules to adopt an ordered state [8,9]. Thus burial of such surfaces will result in an increase in degrees of freedom of the system as these water molecules are released from the surface into bulk solvent.
Why determine thermodynamic parameters of biological systems?
Having obtained the thermodynamic parameters from the ITC experiment, what value do they have in interpretation of biologically relevant interactions and as input to drug development? Clearly, one major reason for quantification of biomolecular interactions is to ascertain how tightly they interact. The KB,obs, and hence the derived ΔGobs, give a direct evaluation of this. A good example of the use of the affinity measurement in understanding biology is in understanding the issue of specificity and selectivity in protein interactions in tyrosine kinase-mediated signal transduction [10–12].
Protein interactions involved in signalling pathways emanating from RTKs (receptor tyrosine kinases) in eukaryotic cells are generally mediated through well-defined domains. Commonly, SH2 (Src homology 2) domains form the recognition site for interactions with phosphorylated tyrosine residues (pYs) which have been post-translationally modified by tyrosine kinases. Assuming that signals are processed via a linear pathway, for signal transduction without corruption, these interactions have to be specific such that a signal from one type of receptor does not interfere with one from another. The level of specificity would be expected to be high, since a given cell can express in excess of 100 different proteins containing these highly sequence-conserved, approx. 100-amino-acid-containing, domains. Thus the interaction of a particular tyrosylphosphoprotein in a given pathway with another protein with the appropriate SH2 domain has to be mutually exclusive of any other SH2 domain interactions. For specific interactions, it is estimated, on the basis of potential concentrations of other competing ligands, that the affinity of a specific ligand should be approximately three orders of magnitude tighter than a non-specific interaction. One way to assess the specificity of an interaction is to compare the ΔGobs of a given molecular complex formation with that expected for a non-specific interaction. SH2 domains were generally shown to recognize amino acid sequences proximal and C-terminal to the pY residue. The SH2 domain from the protein Src preferentially recognizes the sequence EEI (Glu-Glu-Ile). Thus exposing this to tyrosylphosphopeptides which contain the recognition sequences for another SH2 domain should provide a significant difference in the ΔGobs. Table 1 shows the thermodynamic parameters for the binding of the SH2 domains from Src to a series of tyrosylphosphopeptides. Changing the residue in the pY+3 position to one of different size and polarity leads to approx. 4 kJ·mol−1 difference in ΔGobs (or only about one order of magnitude difference in affinity). In addition, the ‘specific’ sequence binds in with an affinity only approx. 60-fold tighter than a sequence which is purported to be specific for a different SH2 domain (pYVPM). Furthermore, a sequence which has no similarity to the ‘specific’ sequence whatsoever (pYQPG) binds only two orders of magnitude weaker than the pYEEI sequence.
This apparent lack of specificity in the binding of ligands to SH2 domains can be exemplified further by looking at the affinities of a series of lead compounds developed as potential inhibitors of the binding of tyrosylphosphoproteins to the Src SH2 domain. Six compounds shown in Figure 1 were based on a pY moiety and a rigidified peptidomimetic backbone, but were designed to have different functional groups to explore the chemical space in the pY+3 pocket (i.e. the pocket on the surface which accommodates the isoleucine residue from the ‘specific’ sequence). The data (Table 2) show that, despite the fundamental differences in the chemical composition of the functional groups, the change in free energies of binding are all very similar. Thus the pY+3 pocket shows a high level of promiscuity to ligand binding, which is not something expected from a site required to have high levels of specificity in signalling .
The thermodynamic data derived from ITC, which demonstrate the lack of specificity in ligands designed to probe the pY+3 pocket of an SH2 domain, also reveal a clear advantage of ITC measurement. On perusal of the binding data, it is apparent that, despite similar ΔG values, the underlying changes in enthalpy and entropy are quite different for each interaction. Although these terms compensate to give similar affinities across the compound series, the differences in ΔHobs and TΔSobs between compounds are striking. These terms represent a secondary level of information which is clearly important in defining an interaction. The significance of these terms in understanding biology and ligand development is discussed below.
The thermodynamic–structure correlation challenge
One of the major challenges in molecular biophysics now is to try to improve our understanding of the enthalpic and entropic contributions to binding. The major hurdle in this endeavour involves the correlation of thermodynamic data with structural detail. Provided with a high-resolution structure which enables precise definition of the positions and lengths of non-covalent bonds and how these bonds vary, it would be a major achievement to be able to predict the affinity of an interaction and the underlying thermodynamic parameters associated with binding. This would be of potentially huge financial and temporal benefit to the pharmaceutical industry, since it could circumvent the synthesis and assaying of numerous compounds designed to inhibit a specific binding site. Alternatively, it also would be of great value to be able to make some estimate of the structural nature of a binding site (or protein conformational change) simply from ascertaining the thermodynamic binding data on going from the free to the complexed form of a protein and ligand. The emphasis of this effort has to be in providing a tool capable of making simple correlations that can be determined easily, as opposed to more sophisticated tools such as complex energy calculations from molecular dynamics simulations. With this in mind, there have been numerous attempts to find correlations between binding data and changes in structural detail on going from the free to the bound state (e.g. [14–16]). In many of these investigations, the structural component of the correlation has been represented by molecular surface area burial. This is easily measured using standard techniques (such as that described by Lee and Richards ) based on the change in water-accessible surface on going from the free to the bound state.
A recent study on the correlation between high-resolution structure and thermodynamics in protein complexes highlights the complex nature of protein–ligand interactions and the difficulty in finding a correlation between these parameters . In that study, data from the SCORPIO (Structure-Calorimetry Of Reported Protein Interactions Online) database (http://scorpio.biophysics.ismb.lon.ac.uk/scorpio.html) derived from over 450 literature-reported ITC experiments on proteins interacting with ligands ranging from biologically relevant to synthetic was assessed against structural information based on surface area burial. As with previous studies, the parameter of change in accessible surface area was adopted as a suitable representation of structure perturbation. The study revealed some surprising and somewhat disappointing features generally portraying a lack of any real correlation. For example, it was assumed that there would be some correlation between the enthalpy change, ΔHobs (which reports on the net change in heat of formation or breaking of non-covalent bonds on going from the free to the bound state) and the change in polar surface area. This was because the main component of the heat of binding would be derived from hydrogen bonds which are made between polar/charged groups . No detectable correlation was observed. Another lack of an expected correlation was found between the change in entropy, ΔSobs, and the burial of hydrophobic surface area. This was surprising, since it has generally been accepted that the exposure of non-polar atoms on a protein or ligand to bulk water results in an ordering of the water molecules. Thus burial of this surface would be expected to liberate these ordered water molecules, resulting in an increase in the overall degrees of freedom of the system and hence the entropy.
Some interesting observations were made from the analysis of the database of ITC data. The favourable ΔSobs contributions for naturally occurring protein–ligand interactions were found to be significantly less than those for synthetic compounds (Figure 2). This probably reports on the fact that the synthetic compounds, which are generally drugs or lead compounds, are designed to have increased affinity, and because it is very hard to add in a specific non-covalent bond from the novel ligand to the protein. The easiest way for a synthetic chemist to improve the ΔGobs is to add atoms to the molecule which can get buried on complex formation. This increases the favourable entropy term based on the removal of ‘ordered’ solvent from the protein and ligand surfaces [8,9].
So what value does thermodynamic parameterization have?
The lack of clear concise correlations between the ITC-generated thermodynamic data and the structural detail on going from the free to the bound state perhaps reveals our naïve understanding of all the contributors to the thermodynamic parameters concomitant with a protein–ligand interaction. Further endeavour in this area is important if useful tools are going to be generated on the basis of simple measurement of thermodynamics [20–23].
In the absence of clear correlations, the thermodynamic information derived from an interaction can be used as an input for decision making in drug development. As described above, the affinities and hence the ΔGobs for a series of lead compounds can often be very similar. Making a decision on which compound(s) to take forward can be difficult on the basis of affinity data alone (see above), thus other properties of the interaction are useful to consider. The data shown for the compounds in Table 2 reveal that the next level of thermodynamic data for the interactions (i.e. the ΔHobs and TΔSobs) provide further quantification of the interaction. The enthalpic and entropic data reveal the components of the overall energy of the system. These must be of value in defining the interaction and how it can be modified to increase the affinity. One way in which this has been proposed is in using the ΔHobs as a decision-making tool when attempting to assess which compound to advance when presented with a series of lead compounds [23–25]. The general idea here is that, as stated above, the ΔHobs reports directly on the net formation and/or breaking of bonds on forming a complex. Thus it follows that the more favourable the ΔHobs term for an interaction is, the better the complement of non-covalent bonds formed on going from the free to the bound state. The design of novel non-covalent bonds into a protein–lead compound interface is extremely difficult (see the above mention of entropic contributions in natural compared with synthetic ligands). Thus, in the case where a decision has to be made on a series of similar compounds which are aimed to inhibit a specific target protein-binding site, the compound with the most favourable enthalpy term would be preferred because it can reasonably be surmised that that molecule is forming the best complement of interactions, and improvements in affinity through rationally designed bonds would be less of a requirement. Of course, the use of this tool for decision making in drug development would be taken along with other commonly adopted meters such as those delineated by Lipinski's rules.
Since this approach is seen as a tool to compare potential ligands for a given binding site, quantification can be provided in the form of an enthalpic efficiency (EE) for the interaction : (3)where Q is non-hydrogen atoms or molecular mass. This provides medicinal chemists with an index, much like the ‘ligand efficiency’ which is currently widely adopted , and allows the assessment of improvement in compound design.
Perhaps unsurprisingly, the use of EE is not a panacea to the pharmaceutical industry. The values have to be used concomitantly with further information available on the target molecule-binding site. A good example of this is presented by considering the compounds shown in Table 2 and Figure 1. The binding site on the Src SH2 domain accommodates an aliphatic hydrophobic isoleucine side chain in the ‘specific’ peptide (see above). Accordingly, the protein presents a complementary largely hydrophobic surface. The molecule with the highest EE is compound 1. The aromatic group can clearly be accommodated in the hydrophobic binding site and can form non-covalent π–π stacking interactions with one (or both) pocket tyrosine residues, which are likely to contribute to the ΔHobs [27,28]. Interestingly, compounds 3 and 4 jointly have the second highest EE. Both of these are very different from 1, but are similar to each other in having acidic groups. The relatively high EE values and more favourable ΔGobs for these compounds suggest that inclusion of the hydrogen-bonding acid group might be a sensible design strategy.
ITC instrumentation has become widely adopted in the biological science community as a standard tool for assay of biomolecular interactions. Despite this widespread usage, there is little real understanding of how the thermodynamic parameters determined can be attributed to the manifold changes that a binding event incurs. The measurement of a single binding event in a calorimeter can also invoke thermodynamic contributions from solvent rearrangement, biomolecular conformational change, protonation of atoms of the interacting molecules, binding of salts and other solvent constituents, etc. All of these potential contributions are included in the ‘observed’ thermodynamic parameters reported . As a result, it is difficult to characterize which events dominate the thermodynamics of an interaction without careful parsing of the individual terms. Thus the thermodynamics of an interaction in isolation may not provide useful information in describing complex formation. However, the strength of the data comes from comparison of the binding of similar molecules to a given macromolecule or measurement of the interaction over a range of different conditions. Comparison in this way enables the assessment of the thermodynamic effects of subtle changes to the ligand, whereas measurement over changing conditions can provide added information, e.g. changing the pH can allow the identification of protonation events which occur on binding and contribute to the ITC evaluation .
In addition to this, the determination of the thermodynamic signature of an interaction could have a role in decision making in drug-development programmes. As described above, the ΔHobs can provide a measure of the formation of non-covalent bonds on complex formation. Thus, in comparing a series of compounds, the selection of the compound with the most favourable enthalpy term seems to be sensible, since it should have the best complement of bonds.
As our understanding of the thermodynamic contributions to binding improves with the gathering of more biophysical data, and our way of representing the structural perturbation associated with binding becomes less coarse, our capability to use thermodynamic parameterization as a tool for understanding biologically relevant interactions as well as assisting drug design will make the use of the ITC even more powerful.
J.E.L. is funded through a grant from the G. Harold and Leila Y. Mathers Foundation and is a Fellow of the UT Trust.
I thank Dr T.S.G. Olsson and Dr M.A. Williams for their input into Figure 2.
Experimental Approaches to Protein–Protein Interactions: A Biochemical Society Focused Meeting held at University of Sheffield, Sheffield, U.K., 11–12 January 2010. Organized and Edited by Michael Sutcliffe (Manchester, U.K.) and Mike Williamson (Sheffield, U.K.).
Abbreviations: EE, enthalpic efficiency; ITC, isothermal titration calorimetry; SH2, Src homology 2
- © The Authors Journal compilation © 2010 Biochemical Society