Graeme L. Conn
Department of Biochemistry, Emory University School of Medicine, Atlanta, GA, USA
Proteins, nucleic acids (DNA and RNA) and other biomolecules do not work in isolation. Rather, the many biological functions of these molecules are defined by their often complex networks of intermolecular exchanges with a diverse array of other biomolecules and bioactive ligands. Enzymes and macromolecular machines like the ribosome, for example, must specifically assemble all required subunits, essential co‐enzymes and/or co‐substrates in order to efficiently catalyse turnover of their bound substrates. Further, enzyme activity may be controlled or localised through interaction with other regulatory binding partners. Similarly, biomolecules that play structural or transport roles in the cell must precisely associate to form their complex macromolecular assemblies. More broadly, regulation of cellular processes from genome replication and stability to each step of gene expression involves a complex and precisely coordinated collection of biomolecular interactions. As a result, understanding what drives biomolecular interactions is at the basis of understanding biology itself.
Many approaches are available to study biomolecular interactions, each with its associated strengths and limitations. These approaches range from qualitative identification of potential binding partners to detailed, quantitative analyses of binding affinities and kinetics. The choice of an optimal approach is often dictated by the specific system under study, including availability or otherwise of purified biomolecules in a suitable form, as well as the ultimate goal of the experiment. A yeast two‐hybrid screen, phage display or co‐immunoprecipitation (e.g. from complex mixtures or cell lysate) may be a good starting point to identify potential interactions that can be further validated by other direct binding methods. Co‐elution of purified binding partners from a size exclusion column might also serve as a simple qualitative assay for binding and the impact of changes (such as amino acid substitution or nucleic acid sequence changes) in one or both binding partners.
Classical biophysical approaches for quantitatively analysing binding often require a source of purified components, though a number of methods are equally applicable to complex mixtures, such as cell lysates, or even to interactions within whole cells. All methods also require a means of detecting one of the binding partners or the binding process itself. For protein–nucleic acid interactions, approaches such as filter binding and electromobility shift assay (EMSA) have a long‐standing record of successful use. Such experiments typically use radioactive phosphorus (32P)‐labelling for its high sensitivity though fluorescent labelling of the nucleic acid is also possible. A limitation of these methods is that detection of binding requires separation of bound and free ligand: this process can perturb the binding equilibrium, potentially leading to inaccuracies in the measured affinities and also making the approach less well suited for low affinity measurements.
Where fluorescently labelled proteins/polypeptides, nucleic acids or small molecule ligands are available, a variety of other methods becomes possible, including fluorescence anisotropy/polarisation (FP), Förster resonance energy transfer (FRET), fluorescence correlation spectroscopy (FCS), analytical ultracentrifugation (AU) and, more recently, microscale thermophoresis (MST). While each of these approaches has advantages and limitations, a key benefit of fluorescence‐based approaches, and many other classical biophysical methods (including isothermal titration calorimetry, ITC), is that they are performed in free solution at thermodynamic equilibrium. It is also notable that many fluorescence‐based approaches are equally applicable to studies of purified components as well as interactions within complex mixtures or whole cells.
For in vitro studies of purified binding partners, a key advantage of approaches like FP is that they do not require separation of bound and free ligand allowing binding to be quantified without perturbing the equilibrium and making them suitable for measurement of low affinity interactions. A further specific advantage of FP is the ability to set up the experiment in a straightforward manner as a competition assay in which a single fluorescently labelled ligand is pre‐bound to the target molecule and this interaction is disrupted by titration of unlabelled competitor ligands. This approach can be used, for example, to relatively rapidly determine apparent affinities for libraries of protein/nucleic acid variants or small molecules for a given target. More recently, MST has also gained significantly in popularity as a versatile approach to study interactions of purified components and within complex mixtures. In MST, a solution or mixture containing the binding partners is placed with specialised thin capillaries and an infrared laser used to induce a microscopic temperature gradient that perturbs two parameters that both typically change upon binding: thermophoresis, or the movement of the molecule in the temperature gradient, and a temperature‐related fluorescence intensity change that depends on the fluorophore's microenvironment. Key advantages of MST are that it is broadly applicable, performed in free solution, without the need for immobilisation, and requires only that one molecule is fluorescently labelled.
Finally, amongst the most widely used approaches for detailed, quantitative measurements of binding are surface plasmon resonance (SPR) and the conceptually similar ‘dip and read’ method, biolayer interferometry (BLI), though, as noted above, MST is also rapidly gaining in popularity. The power of these approaches lies in their capacity to directly measure kinetics of binding, i.e. rates of association (k on ) and dissociation (k off ), from which the equilibrium binding affinity (K d ) can be calculated. The main disadvantage of these approaches is the need to immobilise one of the binding partners on a sensor or tip surface (typically the smaller binding partner as the signal depends on changes in the immobilised mass on the surface). Nonetheless, the ability to define binding kinetics can provide important insights into the nature of binding interactions that are not usually possible with other methods, which only measure or approximate equilibrium binding affinities. However, as a widely applicable and quantitative method for analysis of biomolecular interactions, ITC has firmly established its place at the forefront of these approaches since its initial development several decades ago [1,2]. Indeed, ITC is often referred to as the ‘gold standard’ for analysing intermolecular interactions.
ITC has two key advantages: it is performed entirely in solution, meaning that there is no need to immobilise one of the interacting partners (as is the case for SPR or BLI), and it detects a universal signal, meaning no label is required (as is the case for EMSA, FP or MST). However, probably the foremost benefit of using ITC is that it allows the user to determine a complete thermodynamic profile of binding from a single experiment. Specifically, ITC can be used to directly determine binding affinity (K a = 1/K d ), stoichiometry (n) and enthalpy (ΔH). From these, Gibbs free energy (ΔG) and entropy (ΔS) of binding can also be calculated (see Section 12.2). Measuring the enthalpy of binding provides insight into how favourable interactions contribute to binding, as it reflects the total contributions of hydrogen bonding as well as hydrophobic and electrostatic interactions. While generally more complex to interpret, entropy of binding, on the other hand, reports on the collective contributions of binding‐induced changes in molecular freedom of motion and reorganisation of bound or free water molecules, ions, etc. Such information may not be readily discernible, even from high resolution structures, making ITC a powerful and often essential component of detailed structure–function studies of biomolecules. The insights ITC can provide on thermodynamics and kinetics of binding can also be particularly valuable in the process of drug development in identifying the best candidate(s) for further optimisation of their physicochemical properties from a group of leads that might otherwise appear equivalent on their binding affinities alone [3–5].
Following the introduction of the first commercial microcalorimeters over three decades ago, these instruments are now routinely found both in individual research labs as well as in shared equipment facilities in academic and industrial settings. More recent improvements in instrumentation have increased sensitivity while reducing the amount of sample required, opening the method to the study of harder to produce or purify biomolecules. In practical terms, while not every biomolecular interaction can be analysed by ITC, the method has without question found exceptionally wide application in the study of discrete biological systems and is applicable to measurement of many different types of binding interaction. The following sections will describe the basic principles of ITC theory, current instrumentation typically available at most research institutions and the application of ITC to the study of biological macromolecular interactions. Next, detailed protocols and a trouble‐shooting guide will be provided, which are exemplified by two case studies with one example each of macromolecule–macromolecule and macromolecule–small molecule interaction.
Finally, before delving into the theory and practice of ITC, a brief note on nomenclature. For simplicity of the descriptions that follow, the molecule (titrand) that is placed in the ITC sample cell will be referred to as the ‘macromolecule’ and its binding partner (titrant) that is placed in the automated pipette used for injection (titration) during the experiment will be referred to as the ‘ligand’. However, while often the most practical arrangement, in principle there is no requirement for a larger protein or nucleic acid macromolecule to be the binding partner placed in the sample cell, and the ‘ligand’ may equally be a second protein, DNA/RNA or small molecule. While this chapter will focus on the use of ITC for analysis of biomolecular binding interactions, the power of the approach and instrumentation extends to advanced methods for global ITC data analysis and measuring binding and measuring enzyme kinetics [6–8].
ITC is a non‐destructive, label‐free analytical technique that is performed entirely in solution. In principle, ITC can be applied to the study of any intermolecular interaction provided heat is either evolved (exothermic) or taken up (endothermic) upon binding. This ‘universal signal’ detected in ITC means that there is no requirement for labelling with radioisotope, chromophore or fluorophore, or for immobilisation of one binding partner to a surface. While the nature of the signal detected is one of ITC's principal advantages, it also results in an important limitation: all other processes occurring in the ITC cell upon titration of ligand will contribute to the signal. Although small background signals (such as heats arising from mechanical mixing of the ligand into the cell solution) can be readily accounted for, other more significant changes due to mismatches in solution conditions between macromolecule and ligand samples can cause significant problems. Complex binding interactions, e.g. where protein–ligand interaction leads to conformational change on a different timescale, or binding‐induced aggregation, will also complicate ITC analysis and interpretation and may require other approaches for a complete resolution.
While some limitations exist, another key advantage of ITC is that experiments can be performed under a wide variety of different experimental conditions. Such flexibility may be useful to tailor solution conditions such that they are optimal for the specific biomolecules under study, which may require specific ionic strength or other additives like glycerol for increased solubility and stability. Flexibility in terms of solution and temperature parameters also allows the potential to adjust experimental conditions to maximise the signal detected. Finally, variations in factors like pH, ionic strength or temperature can be used to directly learn more about the binding interaction. For example, experiments performed under different conditions of ionic strength can provide insight into the role of electrostatic interactions in binding, while variations in solution pH can provide information on the role of ionisation state and the mechanism of binding. In short, time invested in optimising the conditions used for a particular binding interaction is likely to be worthwhile and potentially highly informative on the nature of the interaction. With the speed and reduced sample requirement of the most recent versions of modern ITC instruments (see Section 12.2.2 below), as well as the ability in some cases to automate many aspects of data collection and initial analysis, such optimisations can be performed relatively quickly. Additionally, as ITC is performed in solution and is non‐destructive, precious samples can in principle be recovered and used for additional ITC experiments (if the two binding partners are readily separable) or for other purposes.
Despite its flexibility and broad applicability, there are unfortunately still some instances where ITC may not be a useful approach. As noted above, the signal measured in ITC is proportional to binding enthalpy, meaning that interactions with very low ΔH may not be detectable. In such cases, the signal might be increased by altering the temperature at which titrations are performed or increasing the concentrations of both macromolecule and ligand to proportionally increase the signal generated. However, in practical terms (e.g. protein solubility) it may not always be possible to adjust the necessary parameters sufficiently. Additionally, due to the nature of the relationship between K d and reactant concentrations in ITC, simply increasing concentrations may lead to other complications in data analysis. Nonetheless, with highly purified and carefully prepared samples, ITC is generally applicable to most biological interactions with binding affinities in the low nM to high μM range (K d ∼ 10−8 to 10−4 M). Weaker binding affinities can also be measured reliably, though a full thermodynamic profile of binding may not be obtainable [9,10]. Other approaches involving displacement of a weaker binding ligand can be used to extend this useful range to much tighter interactions (K d ∼ 10−9 to 10−12 M) [11].
Commercial ITC microcalorimeters have been available for several decades and are produced today by a number of different manufacturers. Those made by MicroCal (currently part of Malvern Panalytical) and TA Instruments are among those most commonly found in most academic or industrial laboratories. The MicroCal/Malvern Panalytical VP‐ITC and Auto‐iTC200 will be specifically referred to in descriptions later in this chapter, as these are the instruments with which the author is most familiar. However, the sample preparation and requirements, general principles of ITC experiment design and many practical aspects of performing a binding analysis will be more broadly applicable to newer instruments from MicroCal/Malvern Panalytical (such as the PEAQ‐ITC range) or those from other manufacturers.
The basic principles of ITC instruments have not changed significantly in the last decade or so but two notable advancements in the technology have arisen during this time. First, the volumes of sample required to run an ITC titration have been substantially reduced compared to early instruments with the development of small cell (∼200 μl) instruments such as the PEAQ‐ITC and iTC200 (MicroCal/Malvern Panalytical) and Affinity ITC and Nano ITC (TA Instruments). While these instruments have the same limited throughput and requirement for substantial user intervention in an experimental setup as older instruments, the reduced volumes of sample needed (both macromolecule and ligand) have opened the method to use greater effort to produce biomolecules. Importantly, despite the reduced sample requirements, these instruments can detect comparable or smaller heats of binding and operate with similar low levels of noise as instruments with larger cells. The second major recent advance has been the development of automation built around these same small‐cell ITC instruments. Automation has the advantages of eliminating some of the steps that typically cause problems for novice users, such as proper sample loading, as well as allowing for unattended operation for multiple ITC experiments.
Manual instruments such as the iTC200 or the older workhorse VP‐ITC are most likely to be found in individual labs and may be shared by a relatively small number of users. Significant expertise in sample handling and instrument use, particularly sample loading, are critical for obtaining high quality data with these instruments. Automation brings additional cost for purchase and maintenance, meaning that instruments like the Auto‐iTC200 are more often found in managed shared facilities (e.g. [12]). Such ITC facilities can accommodate a much larger group of users with a wide range of expertise, supported by an instrument that is very easy to use. Whatever the specific type of instrument available, the principle of its operation and ultimate experimental output will be largely the same.
A typical ITC instrument (Figure 12.1a) consists of two identical coin‐shaped or cylindrical cells made of a chemically inert and highly thermally conducting metal (such as gold or Hasteloy alloy). Each cell has a long, narrow neck through which it is filled with either the macromolecule sample or a reference solution (typically water). In the case of the sample cell, this long neck is also where the automated titration pipette is placed for injection of the ligand, the second component of the binding reaction. Within the instrument, the two cells are held under conditions of constant pressure (isobaric) and temperature (isothermal) within an adiabatic jacket, meaning that no heat is transferred into or out of this enclosed system. Under application of constant cooling, power is applied to heaters on each cell to maintain them at constant and equal temperature.
When a sample of the ligand is injected from the automated pipette, a binding reaction occurs in the sample cell and heat is either evolved (for an exothermic reaction) or taken up (for an endothermic reaction). In the more common case of an exothermic binding reaction, the heat increase in the cell due to binding is sensed by the instrument and a proportional reduction made to the power applied to the sample cell. In other words, the total heating power from the binding reaction and sample cell heater are maintained at a constant total level, with any heat input from binding compensated by a drop in the heat input from the cell heater. The effect of this power compensation is to maintain the sample and reference cells at an equal temperature (ΔT = 0). Similarly, for an endothermic binding reaction, increased power will be applied to the cell heater to maintain the constant total heating power with the same final result.
The difference in power that needs to be applied to the sample cell heater is the direct output measurement in ITC. Further, the power difference is directly related to the molar enthalpy change associated with the binding process, allowing determination of binding thermodynamics. The specific measurement recorded by an ITC instrument is the time‐dependent sample cell power input (i.e. power per unit of time, such as μcal/s or μJ/s) needed to keep the temperature of the sample and reference cells equal. This output is usually plotted in real‐time during the course of the ITC experiment and appears as a series of sharp changes (peaks) in applied power, each corresponding to one injection of the ligand. In the case of an exothermic binding reaction, less power needs to be applied and the peaks therefore appear as negative deflections from the baseline. The example titration shown in Figure 12.1b corresponds to an exothermic binding reaction.
At the end of the ITC experiment the peaks corresponding to heater power are integrated with respect to time, from the beginning of the ligand injection to the point where the applied power returns to the pre‐injection baseline value. This process of peak integration gives the total heat exchanged per injection and these heat effects can then be analysed as a function of the molar ratio of the binding partners to give the thermodynamic parameters of the interaction under study (Figure 12.1c). The entire ITC experiment and recording of sample cell power take place under computer control, and software associated with the instrument will usually generate an initial automated analysis of the raw data. As described further below, it is usually necessary to manually check and adjust the peak integrations and to perform additional analysis steps before the most accurate fit of the interaction thermodynamic parameters is obtained.
The power of ITC comes from its ability to provide a complete thermodynamic analysis of a binding reaction from a single experiment. Specifically, the user can directly determine the affinity (K a ), stoichiometry (n) and enthalpy (ΔH) of binding. From these directly determined parameters, the Gibbs free energy (ΔG) and entropy (ΔS) of binding can also be calculated. The following descriptions present a simplified view of the theory behind ITC analysis.
In an ITC experiment, the heat evolved (q) in the cell of volume V o upon binding of the macromolecule (M) and titrated ligand (L) is proportional to the concentration of the macromolecule–ligand complex (ML) and related to ΔH of binding by Eq. 12.1. This can be expressed as the ratio of concentrations of complex and total macromolecule as in Eq. 12.2:
In the simplest case of a macromolecule with a single binding site for one ligand, the extent of reversible binding, i.e. concentration of the ML complex, is given by the equilibrium binding constant (Eq. 12.3). Note that some ITC fitting software (e.g. Origin® supplied by MicroCal/Malvern Panalytical) generates the association equilibrium binding constant, K a (units per molar, M−1), whereas biologists typically discuss binding in terms of the dissociation binding constant, K d (units of molar, M). Free energy of binding can be calculated from the K a (or K d ) measured by ITC using Eq. 12.4, where R is the gas constant and T is temperature (in K):
In terms of binding affinities and binding energies, it follows from these relationships that a larger value of K a (or smaller value of K d ) refers to tighter binding between the macromolecule and ligand, as well as a more negative free energy of binding, ΔG. Finally, ΔG can be partitioned out into its enthalpic (directly determined by ITC) and entropic (ΔS) components in Eq. 12.5, allowing determination of each of these thermodynamic parameters:
Having made the decision to study a binding interaction using ITC, the novice user (and even experienced ITC practitioner) is faced with a plethora of potential experimental variables when considering an analysis of a new system. For example, in sample preparation, which specific protein, nucleic acid or other biomolecule(s)/ligand(s) are the optimal ones to begin with? How much is needed and how should each component be purified? What experimental conditions, instrument settings and approaches for data analysis should be used?
Before beginning any new analysis, the most important first consideration is how much of each sample is needed and whether can these be feasibly produced at the level of purity needed for ITC. The answer to this question depends on both the instrument available and the system itself. For the MicroCal/Malvern Panalytical VP‐ITC (∼1.4 ml cell volume) or standard cell configurations of TA Instruments' ITC microcalorimeters (1 ml cell volume), up to 2 ml of macromolecule sample and 0.5 ml of ligand sample is needed for each titration experiment. For automated small‐cell (∼200 μl cell volume) instruments such as the Auto‐iTC200, these volumes are considerably less at 400 and 120 μl for macromolecule and ligand samples, respectively. For manually loaded small cell instruments, such as the MicroCal/Malvern iTC200 or TA Instruments' small‐cell configuration of the Nano ITC, there may up to a further 50% reduction in the volumes needed.
As discussed further below, macromolecule and ligand concentrations are important variables in optimising ITC experiments that also directly impact the quantity of material needed for each titration. Typically, the concentration of the macromolecule should be >10 times the expected K d with the ligand sample at least ∼10 times more concentrated, such that binding approaches saturation by the midpoint of the titration. However, for a new system with no prior knowledge of binding affinity, a first experiment with macromolecule and ligand concentrations of 10 and 100 μM, respectively, is a reasonable starting point (for a 1 : 1 binding stoichiometry). For a 50 kDa protein, this cell concentration corresponds to 0.5 mg/ml, or ∼0.2 to 1 mg total protein depending on the sample cell size in the instrument to be used. For the same protein placed in the pipette, approximately two to three times more is required. For this reason, the harder to produce, more costly and/or most challenging to concentrate component is more often placed in the instrument sample cell. Of course, no matter which instrument is available or the optimal concentrations for the sample, significantly more material will be needed to complete a rigorous experiment complete with appropriate controls and titration replicates. However, if generating sufficient material for a pilot experiment is readily feasible, one should be able to quickly determine whether ITC will ultimately produce useful data for the system under study, thus making it worthwhile to further the investment in time and effort to produce the samples.
Most ITC instruments are supplied with software capable of simulating ITC experiments from a set of parameters (n, ΔH and K d ) provided by the user, allowing some optimisation of starting concentrations to be performed before even beginning a first experiment. Simulating ITC experiments can be particularly useful if some information is already known about the system in question, such as the approximate K d from another binding measurement method or reasonably guessed from experiments with closely related molecules such as protein homologues. The simulation software is also a useful and simple way for the novice user to begin to get a feel for how sample concentration (both absolute and the ratio of macromolecule and ligand concentrations), K d and ΔH influence the shape of the final binding isotherm obtained.
The preferred starting concentrations for an ITC experiment are commonly determined by consideration of a unitless parameter c, which is the product of the predicted affinity of the system and the total macromolecule concentration:
Optimally, experiments should be performed in the range 5 < c < 100. However, in most cases all three parameters measured by ITC (i.e. K a , n and ΔH) can be determined with confidence for 1 < c < 1000 in a single experiment. The macromolecule concentration should therefore be determined with these ranges of c values in mind. For example, for a binding reaction with a K d of 10−6 M (or 1 μM), ideally a concentration of 5–500 μM should be used. For a given binding affinity, the curvature binding isotherm is dependent on the value of c (Figure 12.2). This makes it possible to manipulate the shape of the binding isotherm by adjusting the concentrations of the cell sample to optimise the quality of the data obtained. Of course, obtaining the fully optimal concentration may not always be possible: very low reactant concentrations might result in too small an experimental signal (which depends on ΔH of binding) whereas very high concentrations might result in heats that are too high or may simply be impractical or impossible for biological macromolecules (e.g. if a protein has a tendency to aggregate at a high concentration).
What happens outside of this optimal range of c values? For very tight binding reactions (where conditions of c > 1000 are likely) only ΔH and n can be accurately determined directly from an ITC experiment. Additional approaches, such as using competition experiments with a pre‐bound competing ligand, may however still allow the user to determine a K d [11]. In contrast, for weak binding events it may still be possible to obtain useful data working in the ‘low c range’, by using a very high concentration of ligand to drive binding to near saturation at the end of the titration [ 9, 10]. In this scenario, only the binding affinity can be measured with confidence. ΔH will not be well defined by the experiment and ΔH and n will be correlated in the fit of the data to a binding model. However, in cases where n is known with confidence and can be fixed in the fitting procedure if necessary, such experiments can still be very informative.
In summary, the availability of some prior knowledge (or reasonable guesses) of the system to be studied, use of ITC experiment simulation software and consideration of the c parameter can facilitate the early identification of suitable sample concentrations for an optimal ITC experiment. However, as noted above, in the absence of knowledge about a new system, a first guess of 10 and 100 μM concentrations in cell and pipette, respectively, is a reasonable starting point for an anticipated 1 : 1 binding stoichiometry for a typical macromolecular interaction. From initial estimates of n, K d and ΔH from this first pilot experiment, the same simulation tools can then be used to rapidly identify optimal concentrations for subsequent titrations.
The importance of sample preparation and planning for an ITC experiment cannot be overstated. The quality of data produced, and thus reliability of insights into binding, are only ever as good as the reagents used to generate them. The following sections delve deeper into many of the necessary practical considerations before embarking on an ITC experiment, but it is worth noting three particularly critical points from the outset:
While no one set of protocols can cover all possible scenarios for a given ITC experiment, the following sections aim to serve as general guides for best practice in accomplishing each of the above goals. An important early consideration in experiment design is which molecule to place in the cell and which in the automated injection pipette. While in principle the same binding parameters should be determined regardless, practical reasons often dictate how the experiment is set up. For example, since the macromolecule sample requires considerably lower concentration, less total amount of material and uncertainties in concentration have less impact on the binding parameters determined, protein or other biomolecules prone to aggregation/precipitation may be best placed in the sample cell. Those molecules for which concentration can be very accurately measured may be best considered as the ligand and placed in the pipette.
The quality of data produced by an ITC experiment is critically dependent on the quality of the samples used to generate them. Accordingly, all macromolecules, ligands, etc. should be as purified as possible, free of degradation products and other contaminants, and ideally conformationally homogeneous. While each individual macromolecule or ligand will require its own specific protocol for purification, the following general considerations should be kept in mind when optimising sample preparation for ITC.
Many factors must be considered when deciding on the ‘optimal’ solution conditions for a given ITC experiment. Of course, a primary requirement is to use solution conditions – buffer/pH, ionic strength, reducing agent, organic solvent, or other additives, cofactors, etc. – that are best suited for the long‐term (> several hours to days) solubility and stability of the macromolecule(s) or ligand to be studied. In the absence of homogeneous, well‐behaved samples, no ITC experiment is going to succeed! Keeping this in mind, however, it is also important to note that some limitations on solution conditions do exist and, most critical for performing a successful ITC experiment, the ability is needed to precisely match the solution conditions for the macromolecule and ligand.
Even carefully following these suggested guidelines, slight differences in buffer composition between the ligand and macromolecule are possible due to co‐solvents, salts or pH, particularly where samples are prepared independently (i.e. not co‐dialysed), dissolved from lyophilised material or have components like organic solvents that need to be added immediately prior to experiments. In such cases, a test titration(s) of ligand into the macromolecule sample buffer and/or the ligand sample buffer into the macromolecule before performing the binding experiment may be advisable to check for large heats of dilution that could mask the binding signal. Finally, it is important to keep in mind that many of these steps used for buffer exchange and matching can result in potentially large changes in sample concentration. As a result, concentration measurements should be made following these steps, ideally on the final samples to be used in the ITC experiment.
The accuracy of data obtained from an ITC experiment is dependent on precise determination of sample concentrations. In particular, error in determining the concentration of the ligand in the pipette is directly proportional to the error of measurements of n, K a and ΔH, and therefore the calculated free energy (ΔG) and entropic (ΔS) contribution to binding. In contrast, errors in cell component concentration only affect the accuracy of the determined stoichiometry. In cases where the concentration of one component is more reliably known, this may be an important consideration for experiment design: placing the component with the most accurate concentration in the pipette will result in the highest accuracy of the data produced in the ITC experiment.
Concentrations of each sample should be determined using an applicable method such as absorbance measurement, where a precise measured or calculated extinction coefficient is known, colorimetric assays such as BCA for proteins or HPLC with appropriate standards for small molecules. Ideally the concentrations of both binding partners should be accurately determined after buffer matching and final sample preparation. Two additional steps, described in the experimental protocols below, can also modestly impact the final sample concentrations. First, it is commonly recommended to degas the samples to avoid signal artefacts due to air bubbles or release of dissolved gases during the titration, particularly at higher temperatures. However, this also has the potential to reduce the sample volume via evaporation and thus increase concentration. Of course, since typically a little more sample is prepared than needed, the concentration can simply be rechecked to ensure there is no significant change. Second, rinsing of the sample cell will leave some residual buffer in the cell which will dilute the sample (by ∼1% to 2%). While this is harder to accurately account for, as noted above, this change will only impact the determined binding stoichiometry and can likely be ignored in most cases.
In each ITC experiment, a high concentration solution of ligand is injected (titrated) from an automated pipette (or syringe) in a series of equal injections into a more dilute sample of the macromolecule in the instrument cell. The final task, with final purified samples in hand, is therefore placing these two sample solutions in their respective components of the ITC instrument. Correctly loading the samples, for example without introducing any air bubbles, is essential for a successful experiment. How this is accomplished depends on the type of instrument available and is described below for both a manual VP‐ITC and the fully automated Auto‐iTC200 MicroCal/Malvern Panalytical instruments. Newer manual instruments than the VP‐ITC are likely to have partial automation that simplifies the sample loading process. Additionally, while the same general goals and principles apply, the practical procedures for setting up instruments from other manufacturers, such as the Nano ITC from TA Instruments, may also vary considerably (links to additional web resources for Nano ITC cell and syringe loading are provided under Website Resources at the end of this chapter).
Prior to beginning any experiment with precious samples, the user should make sure the ITC instrument is in good working order. Check any instrument logs for the date of last use and any notes from previous users to make sure the instrument was cleaned following manufacturer recommended or local instructions for cell and pipette cleaning. For automated instruments, visually inspect the fine tubing that carries samples and other reagents during the filling process for signs of precipitated materials or blockages. Running a test titration, which could simply be a small number of water‐into‐water injections, is a good idea in either type of instrument to confirm the instrument is operating as expected with a stable baseline close to the set reference power at the beginning of the test titration. Watching the pipette and cell loading processes on an automated instrument during this test titration is a good way to identify any potential issues before placing precious samples on the instrument.
Sample loading on the VP‐ITC involves the use of a custom air‐tight Hamilton syringe with a protective plastic sheath around the metal needle. Check that this sheath extends very slightly (∼1 mm) beyond the tip of the metal needle to avoid scratching the surface of the ITC cell. Approximately 2 ml of the macromolecule and 0.5 ml of the ligand solutions are required for the filling processes for the cell and pipette, respectively. Also keep aside some of the final macromolecule dialysis buffer (10–20 ml) for control titrations and rinsing the cell during sample loading.
Identifying the optimal experimental parameters for a given system is likely to take some trial and error. However, the following guidelines and suggestions should provide a suitable starting point for most experiments.
At the end of the titration, the macromolecule–ligand mixture can be retrieved from the sample cell using the Hamilton syringe. Clean out the ITC cell with water and repeat the procedure for preparing the cell for sample loading to perform additional experiments. Ideally, at least one or two repeats of each macromolecule–ligand titration should be performed for reproducibility, in addition to any control experiments of ligand solution into buffer. At the end of a set of experiments (or between each experiment if the sample is heavily precipitated), clean the VP‐ITC by manually filling and emptying the cell several times with detergent solution (e.g. 5% Contrad‐70) and then rinsing using distilled water. Alternatively, the ThermoVac system, if available, can be used to flush larger volumes of these solutions through the cell. For more vigorous cleaning the cell can be filled with detergent of a higher concentration (up to 20% Contrad‐70), for an extended time (60 minutes) and/or at an elevated temperature (65 °C).
The Auto‐iTC200 instrument (and other semi‐ or fully automated systems) greatly improves the consistency of ITC experiments by automating sample loading, which is one of the most challenging aspects of performing an experiment for novice users. As with the VP‐ITC, the software supplied with the instrument is used to define the experimental parameters and control the running of each ITC titration, but an additional benefit is the ability to queue many experiments for unattended operation. Samples are supplied in up to four 96‐well deep well blocks and the cell can also be filled using a macromolecule solution from five 30 ml tubes. While this means that the Auto‐iTC200 could, in principle, run hundreds of experiments without user intervention, more commonly experiments run in smaller groups with periodic vigorous cleaning (which can also be automated) and manual checking of data quality.
Towards the end of the ITC experiment, the final few additions of ligand should result in relatively small peaks of equal height corresponding to the ‘heat of dilution’. This heat arises from the mechanical mixing of the ligand into a sample containing the macromolecule already fully saturated by ligand binding. As discussed in more detail above, close matching of the macromolecule and ligand solutions is essential to avoid large heats of dilution that could mask the actual binding signal. These non‐binding related heats need to be subtracted from each of the points in the macromolecule–ligand titration before further data analysis. This can be accomplished in several ways:
After heats of dilution have been accounted for by one of these approaches, the enthalpy of binding at saturation should approach zero and the binding isotherm can be fitted using an appropriate model to determine binding parameters (see Section 12.3.4 below).
Several problems with ITC experiments are possible that can directly impact data quality. Some of these may be more common in one type of instrument than another, e.g. through the introduction of air bubbles into the sample cell due to a poor cell filling technique on a manual instrument that is less likely with an automated system (though these can have their own sample loading issues if not well maintained). Other problems with data quality may have their origin in problems with the samples themselves. Fortunately, many of these common problems give rise to identifiable signatures in the resulting titrations (Figure 12.4) that can point the user in the right direction to resolve them.
An ideal titration will begin a series of approximately equal, large peaks that become reduced as binding is saturated with further injections of ligand during the experiment. The final peaks, corresponding to the heats of dilution of ligand after binding is fully saturated, should be close to zero (Figure 12.4a). Underfilling of the cell or pipette will lead to erratic peaks and a shifting baseline that will likely make data unusable. A strongly downward stepping or shifting baseline may be indicative of an underfilled pipette, while upward changes may point to an underfilled or dirty cell. Such instrument issues are not likely to be common on a well‐maintained and clean instrument but can happen on an Auto‐iTC200 instrument if some of the components or fine tubing needed to move the samples around become dirty or clogged. After ensuring that the instrument has sufficient supplies of water, detergent cleaning solution, methanol and N2 gas, watching the processes of cell/pipette loading and cleaning should pin‐point any problem areas.
Wildly oscillating or moving baselines, long times for equilibration of baseline or baseline values far from the set reference power may be resolved by thorough cleaning of the sample cell but could also be signs that the instrument has more serious issues that require expert service.
Other issues with sample quality or experiment design may also manifest in the following ways:
Having performed a complete titration of ligand into macromolecule, along with any associated control experiments, the next steps are to perform initial adjustments of the raw data (e.g. baseline) and then to fit the measured heats to an appropriate model for the binding reaction, e.g. a single set of identical sites, two sets of independent sites, sequential binding, etc. A more detailed description of this process and the binding models for fitting ITC data can be found elsewhere (e.g. [16]). This fitting procedure allows the direct determination of binding affinity (K d ), stoichiometry (n) and enthalpy (ΔH) of binding with the goal of modelling the experimental data with the least error and using the simplest model that is in accordance with any prior information on the system under study. Free energy and entropy of binding (ΔG and ΔS, respectively) can then be calculated according to the equations of Section 12.2.3. Such data analysis and fitting of binding to one of several possible models is typically done in specially modified software supplied with the ITC instrument. With the VP‐ITC and Auto‐iTC200 this analysis is done with a customised version of Origin software.
Interaction of transcription factors (TFs) with their DNA binding sites is a key regulatory process in gene expression and an essential activity in living cells [17]. TFs directly control RNA transcription from the DNA genetic material by promoting (activator TF) or blocking (repressor TF) recruitment of RNA polymerase to the DNA [18]. Through these activities, TFs ensure both the correct level of expression in the correct cell type in multicellular organisms. TFs can act as either homodimeric complexes or as monomers, but always contain at least one DNA binding domain (DBD) that is responsible for directing specific interaction of the TF with its associated target DNA.
TF interaction with DNA can be driven by so‐called ‘direct read‐out’ through formation of direct hydrogen bonds and hydrophobic interactions between protein and DNA, as well as ‘indirect read‐out’ of DNA shape and electrostatics. These contacts can also be made directly between the TF and DNA or mediated by water molecules, and each can therefore make distinct contributions to the enthalpy and entropy of binding. The ability of TFs to bind to multiple related sequences with biologically relevant affinities has been rationalised based on ‘enthalpy–entropy compensation’ arising from relatively subtle changes in the contributions of these different types of interaction. However, it is less clearly understood how some monomeric TFs are apparently able to bind more than one distinct DNA sequence, which would require more significant adjustment in the number and type of interactions necessary to maintain binding. To unravel the basis of this phenomenon in molecular detail, Morgunova et al. [19] performed a detailed mechanistic study of DNA binding by the posterior homeodomain protein HOXB13 and several other monomeric TFs, using X‐ray crystallography, molecular dynamics (MD) simulations and ITC. These studies used two distinct high affinity DNA sequences CTCGTAAA (DNATCG) and CCAATAAA (DNACAA), which differ by the underlined nucleotides and to which HOXB13 binds with similar affinity despite their significant sequence difference.
X‐ray crystal structures of HOXB13 bound to each DNA sequence revealed little difference in the HOXB13 protein structure in each complex, while the two DNAs differed in the bending of their backbone at the location of the distinguishing nucleotides (Figure 12.5a). Additionally, there were few differences observed in HOXB13's interactions with each DNA (Figure 12.5b). Together with extensive mutagenesis experiments, these structures suggested that the differences in protein–DNA interactions or commonalities in DNA structure cannot fully explain the dual sequence preferences of HOXB13 and other TFs. Rather, the major difference between the structures appeared to be the role of bridging water molecules in the interaction of HOXB13 with each DNA. Specifically, water molecules in the DNACAA complex appeared more stably bound and making interactions that were likely to correspond to stronger enthalpic contributions. In contrast, water molecules in the HOXB13‐DNATCG complex appeared significantly more disordered, suggesting a high entropy state. These ideas were further supported by MD simulations and free energy perturbation calculations. However, to directly test their hypothesis of distinct enthalpic and entropic minima driving TF binding to distinct DNA sequences, the authors turned to ITC analysis with its power to provide a complete thermodynamic profile of HOXB13 binding.
For use in ITC experiments, HOXB13 was expressed in Rosetta 2(DE3)‐pLysS Escherichia coli cells using an auto‐induction protocol and purified by Ni2+‐affinity chromatography [ 19,20]. DNAs were chemically synthesised by a commercial supplier as single strands and the complementary sequences mixed and annealed to form each double‐stranded DNA. Both HOXB13 and each DNA were prepared in a solution containing 20 mM HEPES buffer pH 7.5, 300 mM NaCl, 10% glycerol and 2 mM TCEP, and titration experiments were performed on an iTC200 microcalorimeter at the Protein Science Facility at Karolinska Institute, Sweden. Each experiment was performed at 25 °C and involved 23 injections of HOXB13 ‘ligand’ (150 μM) into DNA solution (12–16 μM) in the sample cell. All data were analysed using the customised Origin software supplied with the instrument and binding parameters (n, K d , ΔH) determined by a non‐linear least square fit of the data to the one binding site model as implemented in the package. The TΔS and ΔG of binding were obtained using Eq. 12.5. Each titration was performed in triplicate and average values for the determined binding parameters were reported and reproduced here.
ITC titrations of HOXB13 into each DNA revealed a very similar affinity and free energy of binding for both sequences, as anticipated (Figure 12.5c and d). The data also revealed, however, the strongly differing contributions of enthalpy and entropy in each case. For DNACAA, ΔH was significantly increased whereas ΔS was strongly reduced compared to DNATCG (Figure 12.5d). The authors were thus able to draw the conclusion from these direct experimental measurements that HOXB13 binding to one DNA sequence (DNACAA) represents an enthalpically driven minimum in binding free energy while the other (DNATCG) an entropically driven minimum (Figure 12.5e). This novel finding also demonstrates the power of ITC to provide a complete picture of binding thermodynamics to complement other biophysical approaches like X‐ray crystallography in understanding the molecular basis of biomolecular function.
RNA post‐translational modifications such as nucleobase or ribose sugar methylations occur in many different types of RNA in both prokaryotes and eukaryotes. In ribosomal RNA (rRNA), for example, such modifications are critical for correct RNA folding and ribosome subunit assembly, as well as ribosome function in protein synthesis [21,22]. Additionally, rRNA methylation can directly block the activity of ribosome‐targeting antibiotics and is an important contributor to the escalating problem of bacterial resistance to antibiotics such as macrolides and aminoglycosides [23–25]. In some other rarer cases, specific rRNA modifications can also be required for proper antibiotic binding and therefore antimicrobial activity. One example of this scenario is the ribosome‐targeting antibiotic capreomycin, which requires the ribose 2′‐O‐methylations incorporated by the 16S/23S rRNA methyltransferase TlyA for its antimicrobial activity [26]. Given the ubiquity of RNA modifications in biology and their important contributions to current biomedical problems, such as antibiotic resistance, identifying the enzymes responsible and defining their mechanisms of action, is an active area of investigation. An important aspect of such mechanistic understanding of rRNA methyltransferase enzyme activity is defining how they bind their obligatory co‐substrate for the methylation reaction, S‐adenosyl‐L‐methionine (SAM). In a recent study from our lab on the capreomycin resistance‐related rRNA methyltransferase TlyA from Mycobacterium tuberculosis, we used ITC to define the role of a new protein motif critical for SAM binding by this enzyme [27].
At the outset of our study on TlyA, a key goal was to determine a high resolution crystal structure of the intact enzyme. However, when crystals of the full‐length protein could not be obtained, we used a collection of proteases to identify stable TlyA fragments than might be more amenable to crystallisation. Limited proteolysis with the endopeptidase GluC, which selectively cleaves the polypeptide backbone to the C‐terminal side of accessible glutamic acid residues, identified one such stable fragment. Further analyses suggested that GluC cleavage occurred primarily at glutamic acid 59 (E59), immediately adjacent to a short tetrapeptide linker between the TlyA N‐terminal domains (NTD) and C‐terminal domains (CTD), …59E↓RAWVS64…, where the site of cleavage is denoted by the arrow and the underlined sequence is the interdomain linker. A new construct was designed for expression of the isolated TlyA CTD (residues 64‐268) beginning with the first residue of α‐helix 1 (α1) in the CTD. Using this new construct, a crystal structure of the TlyA CTD was determined at 1.7 Å resolution. The TlyA CTD adopts a canonical Class I methyltransferase fold [28], including all the expected conserved motifs required for SAM binding. Despite this, we were unable to obtain a TlyA‐CTD:SAM complex structure by co‐crystallisation or soaking of preformed TlyA CTD crystals. We therefore used ITC to gain complementary insights into SAM binding by TlyA.
For these ITC analyses, 6× His‐tagged wild‐type TlyA and TlyA CTD proteins were expressed in E. coli BL21 (DE3) and purified using His‐Trap HP and Superdex 75 16/60 columns on an ÄKTA Purifier FPLC for Ni2+‐affinity and gel filtration column chromatography, respectively. TlyA proteins were exhaustively dialysed against 50 mM Tris buffer (pH 7.5) containing 120 mM NaCl and 10% glycerol and concentrated using a centrifugal concentration device to 60–100 μM. The final dialysis buffer solution was used to resuspend SAM at 1.5 mM final concentration. As binding was expected to be relatively weak, ITC experiments were performed at the highest practically achievable protein concentration and much higher ligand (SAM) concentration to drive binding towards saturation at the end of the titration (i.e. a ‘low c’ ITC experimental design). All ITC experiments were performed at 25 °C and used 16 injections of 2.4 μl of SAM into the cell containing the protein. After subtraction of residual heats derived from the average values of the final two or three injections, ITC data were fit to the model for one‐binding site. The equilibrium dissociation constants (K d ) reproduced here (Table 12.1) are the averages of at least three experiments with their associated standard deviation.
Table 12.1 SAM binding by TlyA proteins. a
Binding affinity | ||
Protein | K d (μM) | Fold‐change b |
TlyA | 23.4 ± 2.9 | — |
CTD | No binding | — |
RAWVCTD | 20 ± 1.1 | — |
TlyA‐R60A | 87.2 ± 8.5 | 3.6 |
TlyA‐R60E | 62.8 ± 23 | 2.7 |
TlyA‐A61V | 21.1 ± 4.3 | 1.0 |
TlyA‐W62A | 234 ± 56 | 10 |
TlyA‐W62F | 98.5 ± 27 | 4.3 |
TlyA‐V63A | 470 ± 19 | 20 |
a Data are from Witek et al., J. Biol. Chem., 2017 [27].
b Relative to wild‐type TlyA, indicated for RAWV linker substitutions only.
Wild‐type TlyA bound SAM with low micromolar affinity as expected (Figure 12.6a and Table 12.1), comparable to many other rRNA methyltransferase enzymes. In contrast, the isolated TlyA CTD protein failed to show any binding to SAM (Figure 12.6b), despite its crystal structure showing all expected conserved SAM binding motifs being present. While this result was surprising, it was consistent with our inability to obtain a TlyA CTD‐SAM co‐crystal structure. TlyA cleaved by GluC was examined next, both as the cleaved NTD/CTD mixture and the isolated CTD purified by gel filtration chromatography. Both samples were shown by ITC to retain the wild‐type SAM binding affinity (data not shown). Together, these results suggested that some element outside the CTD is critical for SAM binding and a new protein expression construct was created to express the CTD protein including the four amino acids of the interdomain linker at its N‐terminus (residues 60‐268; RAWVCTD). Remarkably, addition of the RAWV tetrapeptide sequence fully restored SAM binding to the wild‐type TlyA affinity (Figure 12.6c and Table 12.1).
These results suggested that the RAWV sequence of the TlyA interdomain linker might form a critical but previously unappreciated motif necessary for SAM binding by the TlyA family of rRNA methyltransferases. Site‐directed mutagenesis was used to make one or two substitutions of each of the RAWV amino acids and the impact of these changes were tested by measuring SAM binding by ITC (Table 12.1). Consistent with sequence analysis of the RAWV tetrapeptide, which showed W62 and V63 to be most strongly conserved, changes at these two residues had the greatest impact on SAM binding affinity. X‐ray crystallographic studies of the TlyA RAWVCTD protein revealed that the RAWV tetrapeptide is able to adopt two distinct conformations: a ‘helix’ form that extends α1 at the beginning of the CTD or a ‘loop’ structure that closely corresponded to the conformation of the equivalent sequence in several hemolysin proteins that had previously been used for homology modelling of TlyA [29].
Interactions made by V63 and other changes in the SAM binding pocket, which were only observed in the RAWVCTD ‘helix’ structure, also offered insight into a potential molecular basis for the role of the new RAWV motif in SAM binding (Figure 12.6d and e). The newly identified ability of the RAWV interdomain linker to adopt two distinct structural conformations, as well as its critical importance for SAM binding as revealed by ITC, suggested that this short sequence may play a critical functional role in substrate recognition by TlyA. Specifically, the question of how TlyA can specifically recognise and modify the 2′‐OH of two cytosine residues within distinct structural contexts, 23S rRNA C1920 in the 50S subunit and 16S rRNA C1409 in the 30S subunit, may have its answer in a common substrate binding‐induced interdomain linker conformational change that is linked to SAM binding affinity and thus TlyA activity. Fully delineating the molecular mechanisms of TlyA and other rRNA methyltransferase activities will require further investigations, of which ITC will likely be a critical component.
ITC is firmly established as one of the best choices of approach to study many different types of biomolecular interaction. With its power to fully define the thermodynamics of binding in a single experiment, ITC has found broad application in studies of basic biological processes as well as in drug discovery. Further advances in instrument technology and the associated software have also simultaneously reduced the previously high sample quantity demands and simplified the processes of performing and analysing ITC data. As such, ITC is an approach that should be among the first considered for any study that aims to understand the nature and role of a biomolecular interaction.
I thank Dr Debayan Dey and Mr Zane Laughlin (Emory University) and the anonymous reviewers for their helpful comments and suggestions on this chapter. Our research is currently supported by NIH/NIAID grant R01‐AI088025.