Background

Fluorescence is a form of luminescence used in many fields ranging from fluorescent labeling in biology and medical diagnostics to fluorescence spectroscopy in chemistry. In fluorescence spectroscopy, incident radiation can promote a molecule to adopt one of many higher vibrational energy states, including those in an excited state. Upon collision with other molecules, some of the vibrational energy is lost until it returns to the lowest vibrational level of an excited state. Upon transition from the excited to the ground state, photons are then emitted at wavelengths that depend on the vibrational state the electron re-occupies in the ground state. Therefore, typically the intensity of photon emission varies with wavelengths and the emitted light has a longer wavelength than the incident radiation (Lakowicz, 1999).

In protein biochemistry, fluorescence spectroscopy is routinely used for monitoring protein (un)folding, and investigating protein conformational changes, binding and interactions. One common fluorescence assay relies on the intrinsic fluorescence of proteins that arises from excitation of aromatic amino acids, mainly tryptophan (Teale and Weber, 1957). Phenylalanine, despite being excitable, has a low quantum yield while tyrosine, despite having high quantum yield, is often quenched naturally (Möller and Denicola, 2002). Tryptophan, being a large hydrophobic amino acid is usually partially or fully buried in hydrophobic sites within proteins, or bound to ligands through hydrophobic interactions such as π-π stacking. Tryptophan can be selectively excited at 295 nm as there is little absorption by other residues at this wavelength. Upon excitation, tryptophan gives rise to an emission spectrum that peaks at 355 nm. Structural changes in the vicinity of tryptophan residues induced by ligand/partner interactions, protein conformational changes, self-association or protein folding/denaturation, can alter the intensity of fluorescence as well as introduce a wavelength shift in the emission spectrum (Möller and Denicola, 2002). In most binding assays, a decrease in fluorescence and a blue shift to lower wavelengths are observed upon binding which is attributed to the increase in hydrophobicity around the tryptophan sites. This is known as tryptophan fluorescence quenching and under the right conditions, can be used to measure the equilibrium binding constant, also known as the association constant K_a.

If P is the protein being studied and L is the ligand (we take this to mean any binding partner which can include protein, peptide and small molecule ligands) that P binds, then at equilibrium:

and the association constant K_a is defined as () while its inverse (i.e., ) is the dissociation constant, K_d.

To assess whether tryptophan fluorescence quenching may be a suitable technique for measuring the binding constant of a protein for its partner or ligand, several questions that should be considered are listed below. “No’s” in Questions 1 and 2 would suggest tryptophan fluorescence quenching is unlikely to be the method of choice and other techniques should be explored for measuring the binding affinity of interest. A “Yes” to Question 3 would indicate further investigations are required to check the feasibility of the experiment and in favorable cases, the effect may be corrected for during data analysis, as shown in the HusA:haem example. Note that in the case where the interaction of interest is a protein:protein or protein:peptide interaction, then either component can be considered the “protein” and the “ligand”.

Does the protein have one or more tryptophan(s) that are likely to be in different environments when the protein is free versus bound?
Is there an absence of tryptophan(s) in the ligand?
Does the ligand absorb at or near the excitation or emission wavelengths of tryptophan at the concentrations required for the titration study? If so, is the inner filter effect significant? In some cases, the inner filter effect can be minimized by using lower ligand concentrations or its contribution accounted for with “control” titrations (see later sections).

The two following questions which apply equally to other techniques used for measuring binding affinity also need to be considered. To a large degree, “Yes’s” are required but the answer may depend on the techniques and set up used which can alter detection limits and dynamic range, as well as the solution conditions required.

Are the protein, ligand and the complex soluble at the concentrations required in the solution condition used for the titration? This will depend on the sensitivity of the instrument, K_d and solution properties of the components. Point 5 below further explains concentration considerations.
Can you work with a protein concentration that is not too high relative to the expected K_d? As a rough rule of thumb, a concentration that ranges from as low as practical to 10x that of the estimated K_d is a good starting point. Can you titrate in the ligand such that its final concentrations in the sample can span 0.1x to at least 5x (and preferably 10x) that of the estimated K_d. A titration may need to be repeated with additional titration points if the measured K_a is significantly different from the range estimated initially to get an accurate measurement of K_a.

The relationship between protein:ligand complex, protein and ligand concentrations, K_a and appearance of the binding curve.
To illustrate the importance of Question 5, an interactive Excel spreadsheet (binding.xls) is available to download. The following section explains why binding affinity measurements are only meaningful when “sensible” protein and ligand concentrations are chosen relative to the binding affinity. In the spreadsheet, only the yellow boxes (K_a, P_total–total protein concentration, L_total–total ligand concentration at the end of the titration) are modifiable. The graphs on the left-hand side in the Excel spreadsheet are obtained by solving the quadratic equation as derived from equation [1] while the graph on the right-hand side shows the comparison between the quadratic solution and using the L_free (Ligand free) can be approximated by L_total assumption. This assumption is valid when the protein concentrations used and/or the affinity (or K_a) is relatively low. This is often the case for enzyme-linked immunosorbent assay (ELISA) type reactions unless the binding affinity is very high (K_a > ~10⁹ M^-1) and hence this model is often included as the default for 1:1 binding in many curve fitting software. However, this does not always apply especially in biophysical studies where the protein concentrations need to be high due to detection limits.

The following set of graphs (Figures 1A and 1B) illustrates the case when the ligand concentration chosen is too low relative to the protein concentration in an example titration where L is being titrated into P resulting in a straight line for the concentration of PL which has a very similar appearance despite a 10-fold difference in the K_d.

Figure 1. Calculated binding curves showing the relationship between [PL] and [L_total] at a constant [P_totoal]. Here [P_total] is chosen to be 10 times that of the final [L_tot]. A. K_d = 0.1 μM, P_total = 10 μM, L_total =1 μM. B. K_d = 1 μM, P_total = 10 μM, L_total =1 μM.

If this titration has continued to a ligand concentration that is 10x that of the protein concentration, the two PL concentration profiles can now be easily distinguished (Figures 2A and 2B).

Figure 2. Calculated binding curves showing the relationship between [PL] and [L_tot] at a constant [P_tot]. Here [P_tot] is chosen to be 10 times less than that of the final [L_tot]. A. K_d = 0.1 μM, P_total = 10 μM, L_total =100 μM. B. K_d = 1 μM, P_total = 10 μM, L_total =100 μM.

However, no matter how high the final L_total concentration is, the PL concentration profiles remain very similar once the affinity is “too” high. Shown in Figure 3 is the comparison between a K_d of 0.01 μM and 0.001 μM with P_total and L_total otherwise unchanged.

Figure 3. Calculated binding curves showing the relationship between [PL] and [L_tot] at a constant [P_tot]. Here [P_tot] is chosen to be far higher than the K_d’s. A. K_d = 0.01 μM, P_total = 10 μM, L_total =100 μM. B. K_d = 0.001 μM, P_total = 10 μM, L_total =100 μM.

The only way to “fix” this is we can drop the protein concentration. As seen in Figure 4, we can recover the difference in PL concentration profiles by dropping the protein concentration (and corresponding ligand concentration) to 0.1 μM so it is comparable to the K_d’s. However, in the case of tryptophan fluorescent quenching or other techniques that require protein signals to be measured, this may not be feasible given the detection limit of the technique.

Figure 4. Calculated binding curves showing the relationship between [PL] and [L_tot] at a constant [P_tot]. Here [P_tot] is chosen to be higher than the K_d’s but still in a reasonable range. A. K_d = 0.01 μM, P_total = 0.1 μM, L_total = 1 μM. B. K_d = 0.001 μM, P_total = 0.1 μM, L_total =1 μM.

As such, it may be necessary to use more than one technique to “nail” a binding affinity. It is also advisable to predict what the binding curve looks like given what one knows about the system before choosing a technique and the concentrations of protein and ligand to use.
What other techniques are commonly used for measuring binding affinity?
In addition to the tryptophan fluorescent quenching assay described, several other assays can be used to measure protein binding affinities. Generally, techniques with lower sensitivities (e.g., NMR spectroscopy) are more suited for measuring weaker binding and vice versa. That is, the typical concentrations need for signal detection should match the expected K_d’s. Common assays include labeled ligand-binding assays, label-free binding assays and thermodynamic binding assays. An example of a labeled ligand-binding assay is a fluorescent ligand assay in which the binding partner is fluorescently labeled and monitored for a change in fluorescent intensity or anisotropy once the complex forms (Breen et al., 2016). This assay is useful as it offers a range of wavelengths depending on the fluorophore, however attachment of the fluorophore to the ligand may lead to conformational changes that can interfere with the binding reaction. An example of a label-free binding assay is UV-visible (UV-vis) absorption spectroscopy. It is a powerful technique to study the binding reaction of a chromophore that is sensitive to changes induced by ligation (Nienhaus and Nienhaus, 2005). The binding affinity can be measured by observing absorption profiles of a ligand upon titration of its binding partner. Despite being useful for haem and other “colored” ligands, this assay is not suitable for most ligands. This technique also fails to explore the binding reaction from the protein’s point of view. Thermodynamic binding assay, such as Isothermal Titration Calorimetry (ITC) uses enthalpy changes in the ligand binding reaction to measure the binding affinity and thermodynamics of an interaction (Krainer and Keller, 2015). It is very useful for binding reactions that produce significant enthalpy changes but not otherwise. Other commonly used biophysical techniques for measuring protein:ligand binding affinities include surface plasmon resonance (SPR), microscale thermophoresis (MST), and Bio-Layer Interferometry (BLI) with SPR and BLI having the additional capability to follow binding kinetics.

HusA:haem titration as an example
Our previous work demonstrated that the tryptophan fluorescence of a haem-binding protein from Porphyromonas gingivalis, termed Haem uptake system protein A (HusA), can be quenched by its ligand haem (Gao et al., 2010). HusA has three tryptophan residues however only one is located at the haem-binding site (Protein Data Bank accession code 6BQS; Figure 5A) and binds a range of porphyrins including haem (Figure 5B). UV-visible (UV-vis) spectroscopy was also used to explore haem binding by HusA and produced a similar K_a estimate. ITC was also attempted, however, the heat changes were too small even at the highest concentrations of protein and haem used so was not pursued.

Figure 5. Structure of HusA (A) and haem (B). The structure of HusA is displayed in ribbon with its three tryptophan residues shown as sticks. The tryptophan (W130) at the haem binding pocket is highlighted in cyan.

Despite being an efficient static fluorescent quencher of HusA, haem also displays background quenching known as the inner filter effect. This inner filter effect occurs when haem absorbs excitation light or fluorescence emitted by tryptophan, known as the primary or secondary inner filter effects, respectively (Ghisaidoobe and Chung, 2014). This creates an apparent continual quenching and hence binding. To correct for this, control titrations were carried out by titrating haem into N-Acetyl-L-tryptophanamide (NATA) (Zelent et al., 1998; Fonin et al., 2014), a fluorophore that does not bind haem. This data was used to correct for the inner filter effect.