Protein-ligand affinity determinations using covalent labeling-mass spectrometry

Abstract

Determining binding affinity is an important aspect of charactering protein-ligand complexes. Here, we describe an approach based on covalent labeling (CL)-mass spectrometry (MS) that can accurately provide protein-ligand dissociation constants (K_d values), using diethylpyrocarbonate (DEPC) as the labeling reagent. Even though DEPC labeling reactions occur on a timescale that is similar to the dissociation/re-association rates of many protein-ligand complexes, we demonstrate that relatively accurate binding constants can still be obtained as long as the extent of protein labeling is kept below 30%. Using two well-established model systems and one insufficiently characterized system, we find that K_d values can be determined that are close to values obtained in previous measurements. The CL-MS based strategy that is described here should serve as an alternative for characterizing protein-ligand complexes that are challenging to measure by other methods. Moreover, this method has the potential to provide simultaneously affinity and binding site information.

Graphical Abstract

Covalent labeling/mass spectrometry along with ligand titrations can be used to determine protein-ligand dissociation constants while at the same time identifying protein-ligand binding sites.

Introduction

Understanding the details of protein-ligand interactions has implications for drug discovery, design, and development.^1,2 As a part of characterizing protein-ligand complexes, binding affinity determinations are important. To acquire the binding affinity of a ligand for a protein, mass spectrometry (MS) based methods have been explored as an option because of the advantages of the technique, which include high specificity, small sample consumption, and simplicity (e.g. no protein or ligand immobilization is needed).³ Native electrospray ionization mass spectrometry (ESI-MS) allows noncovalent protein-ligand complexes to be transferred intact from solution to the gas phase,^4-6 which makes thermochemical^7,8 and ESI-MS-based titration experiments possible.^9,10 While such titration experiments have been used successfully for many different protein-ligand systems, some challenges can exist, such as differences in protein and protein-ligand complex ionization efficiencies, in-source dissociation of the complex, or false positives caused by non-specific ligand binding.³

Other MS-based methods that encode protein-ligand binding information into the mass of the protein have also been developed. Hydrogen deuterium exchange (HDX)-based methods have been used to measure the dissociation constants (K_d) for protein-ligand complexes. One of the first examples was the method known as SUPREX (or stability of unpurified proteins from rates of H/D exchange),¹¹ which utilizes denaturants to monitor the relative stabilities of a protein and its ligand complex to obtain a K_d value.¹⁰ Another HDX-MS method to determine K_d values is PLIMSTEX,^13,14 which relies on labeling in the context of ligand titration. PLIMSTEX also offers information about protein-ligand stoichiometries and dynamics at the same time. When peptide-level information is obtained during these experiments, the binding site can also often be acquired, although sometimes the obtained binding site information can be ambiguous due to changes in protein dynamics that can occur distant from the binding site.

For HDX-based methods, back exchange is a potential concern, and the need to measure multiple exchange time points can make the measurements time consuming. Covalent labeling (CL) methods can avoid these challenges, as the covalent bond that is formed limits label loss, multiple time-point measurements are not needed, and residue-level binding site information can be obtained.^15-19 An early example of a CL-MS approach that was used to determine protein-ligand affinities is SPROX (or stability of proteins from rates of oxidation), which uses hydrogen peroxide in the presence of increasing concentrations of denaturant to oxidize proteins with and without ligand present as a means of determining K_d.²⁰ More recently, a method based on fast photochemical oxidation of proteins (FPOP) called LITPOMS (ligand titration, fast photochemical oxidation of proteins and mass spectrometry) was reported.^21,22 LITPOMS uses hydroxyl radicals to modify the protein of interest, and then relies on decreased protein oxidative modifications at the ligand-binding site as the ligand concentration is increased. The method yields information about ligand binding affinities, binding sites, and even ligand-induced structural changes when combined with bottom-up sequencing. As compared to HDX-based methods, CL methods like LITPOMS require little sample dilution, making it easier to study weakly binding protein-ligand complexes and identify binding sites.

Given the potential advantages of CL for determining K_d values, we were interested in assessing if other CL methods and reagents could provide reliable protein-ligand binding constants. In particular, we were interested in investigating CL methods that do not require sophisticated equipment such as lasers or flow cells. It is known that CL-MS, including non-radical CL methods, are able to identify the ligand binding sites of proteins by identifying residues that undergo decreases in labeling due to ligand-induced decreases in solvent accessibility of the residues involved in binding.^15,23-26 If such CL methods are applied in a ligand titration context, K_d values should be accessible.

A key concern with non-radical based labeling reagents is their relatively slow reaction kinetics, which are on the order of milliseconds to minutes^16,27 as compared to radical-based reagents that react on the sub-millisecond timescale.^28,29 The slower reaction timescale might mean that the ligand dissociation and re-association processes that are a normal part of the equilibrium could occur on comparable timescales, yielding incorrect K_d information because the labeling could distort the original binding equilibrium. Such a concern is presumably not present with radical-based methods such as LITPOMS, because labeling happens on a faster timescale than ligand dissociation/re-association. We predict, though, that relatively accurate binding constants can still be obtained using slower labeling reagents as long as the extent of protein labeling at the binding site is minimized enough to prevent significant label-induced changes to the equilibrium. In this work, we use several model protein-ligand complexes to demonstrate this idea. We find that reasonably accurate K_d values can be determined when the extent of protein labeling is kept relatively low.

Materials and Methods

Materials.

The maltose binding protein (MBP) was obtained from MyBioSource (San Diego, CA). Human full-length β-2-microglobulin (β2m) was purchased from Lee Biosolutions (Maryland Heights, MO). Lysozyme from chicken egg white and the following chemicals were obtained from MilliporeSigma (St.Louis, MO): DL-dithiothreitol (DTT), diethylpyrocarbonate (DEPC), dimethyl(2-hydroxy-5-nitrobenzyl)sulfonium bromide (HNSB), epigallocatechin gallate (EGCG), guanidine hydrochloride (GuHCl), imidazole, iodoacetamide, maltose monohydrate, MOPS, MOPS sodium salt, N,N’,N” triacetylchitotriose (NAG₃), L-tryptophan, and urea. Acetonitrile, copper sulfate (CuSO₄), formic acid, potassium acetate, sodium phosphate, sodium phosphate monobasic monohydrate, HPLC grade water, and a 1 M Tris buffer (pH 8) stock solution were all purchased from Fisher Scientific (Fair Lawn, NJ). Centricon molecular weight cutoff (MWCO) filters were obtained from Millipore (Burlington, MA). Sequencing grade modified trypsin and sequencing grade chymotrypsin were obtained from Promega (Madison, WI).

Sample Preparation.

Samples containing 30 μM lysozyme were prepared in a 20 mM MOPS (pH 7.4) buffer, and stock solutions of NAG₃, which binds lysozyme, were prepared in HPLC grade water. Samples containing 24 μM MBP were prepared in 20 mM sodium phosphate buffer (pH 7.5), and stock solutions of maltose were prepared in the same buffer. Samples containing 30 μM β2m were prepared in 25 mM MOPS buffer (pH 7.4), 60 μM Cu(II) (CuSO₄), 500 mM urea, and 200 mM potassium acetate, which are conditions under which the protein can aggregate to form amyloid fibrils. EGCG stock solutions were prepared in water. Different concentrations of the small molecule ligands were mixed with the corresponding protein to conduct the titration experiments. All samples were equilibrated at room temperature (22 °C) for 1 to 5 min after the protein was mixed with the ligand and before performing the CL reactions.

HNSB Labeling.

HNSB stock solutions were prepared in water immediately before use. The protein labeling reactions were initiated by adding an aliquot of the HNSB stock solution to the buffered protein-ligand sample. The reactions were allowed to proceed at room temperature for a defined time period before quenching them via the addition of tryptophan to a final concentration of 5 mM. The concentrations of HNSB and reaction times were varied to examine the effect of different labeling extents on the measured K_d values

DEPC Labeling.

DEPC stock solutions were freshly prepared in acetonitrile for each experiment. The protein labeling reactions were initiated by adding an aliquot of the DEPC stock solution to the buffered protein-ligand sample, with the final DEPC concentration depending on the protein and the experiment. The final volume percentage of acetonitrile was less than 1% in all experiments. The reactions were allowed to proceed at 37 °C for 1 min, before being quenched by the addition of imidazole at a final concentration of 10 mM. These labeling conditions were similar to our group’s previous studies with this CL reagent.^27,30

Proteolytic Digestion.

CL-modified proteins that were subjected to proteolytic digestion and LC/MS/MS analysis were prepared by the following procedure. Before denaturing the protein, samples were diluted with water to a volume of 400 μL and then were concentrated using a 10,000 MWCO filter to a final volume of 40 μL. Samples containing MBP were reconstituted with 0.25 M Tris (pH 8.0) and 6 M GuHCl to a final volume of 100 μL. After incubating at 55 °C for 1 h to denature the protein, the samples were allowed to cool before being diluted again by 300 μL of water. Then, the samples were concentrated by a 10,000 MWCO filter to a final volume of 40 μL. The dilute-concentrate procedures were repeated one more time to reduce the concentration of GuHCl and increase the concentration of the protein. The 40 μL samples were then reconstituted in a 0.1 M Tris (pH 8.0) buffer and digested by trypsin for 2 h at 37 °C, if the protein was labeled by HNSB, or by chymotrypsin for 2 h at 25 °C, if the protein was labeled by DEPC.

The samples containing β2m were reconstituted to a final concentration of 0.1 M Tris, 13% (v/v) acetonitrile, and 10 mM DTT. After incubating at 55 °C for 1 h, the sample was allowed to cool, and the reduced disulfide bond was alkylated by 14 mM iodoacetamide (prepared in a Tris buffer, pH 8.0) in the dark at room temperature for 15 min. The denatured and reduced protein was then digested by chymotrypsin for 2 h at 25 °C.

Liquid Chromatography–Mass Spectrometry (LC/MS).

For intact protein analyses, measurements were performed on a Bruker (Billerica, MA) AmaZon quadrupole ion trap mass spectrometer equipped with an electrospray ionization source. Typically, the electrospray needle voltage was kept at ~4 kV, and the capillary temperature was set at 220 °C. HPLC separations were performed using a Thermo Scientific (Waltham, MA) Ultimate 3000 HPLC with an OPTI-TRAP MICRO column (1 mm x 12 mm, Optimize Technologies Inc., Oregon City, OR). The proteins were eluted using a gradient of acetonitrile containing 0.1% formic acid that increased from 5 to 99% for 10 min at the flow rate of 50 μL/min.

For peptide mixture analyses, measurements were performed by LC-MS on a Thermo Scientific Orbitrap Fusion mass spectrometer equipped with a nano-electrospray ionization source. The needle voltage was set at 2.1 kV, and the ion transfer tube temperature was set at 325 °C. The resolution of the Orbitrap was set to 60,000, the MSI AGC target was set at 4 × 10⁵ ions with a maximum injection time of 50 ms. Collision-induced dissociation (CID) with a normalized collision energy of 35% was used for tandem mass spectrometry (MS/MS). Data-dependent selection for precursor ions with ion abundances above 5,000 was used, and a dynamic exclusion of 30 s was activated after 3 spectra were acquired for any given precursor ion within 5 s. The MS/MS AGC target and maximum injection time were set to 5 × 10⁴ ions and 100 ms, respectively. HPLC separations were conducted by a Thermo Scientific Easy-NanoLC 1000 system with a Thermo Scientific Acclaim PepMap C18 nanocolumn (15 cm x 75 μm ID, 2 μm, 100 Å). Peptides were eluted using a gradient of acetonitrile containing 0.1% formic acid that increased from 0 to 50% for 60 min at the flow rate of 0.3 μL/min.

Determination of Covalent Labeling Modification Percentages.

Residue level CL modification percentages (% labeling) were calculated by integrating the peak area of eluting peptides (i.e., modified or unmodified) using extracted ion chromatograms, as described in our previous work.^26,30 The % labeling was calculated based on equation 1.

$$\%\text{labeling}=\frac{{\sum }_{i=1}^n{\sum }_{z=1}^m{A}_{i,z}^{modified}}{{\sum }_{i=1}^n{\sum }_{z=1}^m{A}_{i,z}^{modified}+{\sum }_{i=1}^n{\sum }_{z=1}^m{A}_{i,z}^{unmodified}}\times 100$$ (1)

A_i,z represents the peak area of the peptide of interest. Peak areas from all of the measurable peptides (i) that contain the residue of interest, labeled or not, and all detectable charge states (z) are included. The determined modification percentages are relative rather than absolute values because the modified and unmodified peptides have different ionization efficiencies and LC elution times. It should be noted that the % labeling in equation 1 was only used for peptides in which all the charge states that were summed were present in all samples.

To determine intact protein CL modification percentages, the calculation is similar but simplified. Averaged ion abundances during protein elution were used instead (equation 2), and the ion abundance of multiple charge states were summed.

$$\%\text{labeling}=\frac{{\sum }_z{I}^{modified}}{{\sum }_z{I}^{modified}+{\sum }_z{I}^{unmodified}}\times 100$$ (2)

In determining whether two labeling results were different, an unpaired student t-test was applied with a 99% confidence interval.

Calculation of K_d.

In this work, we focus on the most frequently encountered case where only one ligand (L) binds to one protein (P), and only one kind of protein-ligand complex (PL) is formed. The dissociation constant (K_d) in such a system can be described by equation 3.

$$K_d=\frac{[P][L]}{[PL]}$$ (3)

where [P], [L], and [PL] are the equilibrium concentrations of P, L, and the complex PL, respectively. The total concentration of ligand ([L]₀) and protein ([P]₀) has the following concentration relationship with related fractions in the system, respectively (equations 4 and 5):

$$[L{]}_0=[L]+[PL]$$ (4)

$$[P{]}_0=[P]+[PL]$$ (5)

Because ligand binding decreases the extent of labeling of the residues at the binding interface, we assume there is a relationship between the covalent labeling result (represented by % labeling) and the concentration of ligand-free protein in the system ([P]). This relationship can be normalized to the extent of labeling when the protein is 100% bound to the ligand (i.e. maximum ligand) and when no ligand is present. The resulting relationship is represented by equation 6:

$$\frac{[P]}{[P{]}_0}=\frac{\Delta \%labeling}{\%labelin{g}_{maximunligand}-\%labelin{g}_{noligand}}$$ (6)

By algebraically combining equations 3-6, the following relationship that relates the relative change in labeling to K_d in terms that include the total ligand and protein concentrations (equation 7) can be derived.

$$\begin{gathered} relativelabelingfractionalchange=\frac{\Delta \%labeling}{\%labelin{g}_{maximunligand}-\%labelin{g}_{noligand}} \\ =(([L{]}_0+[P{]}_0+K_d)-(([L{]}_0+[P{]}_0+K_d{)}^2-4\times [P{]}_0\times [[L{]}_0]{)}^{0.5})∕2[P{]}_0 \end{gathered}$$ (7)

A plot of the relative labeling fractional change, based on labeling percentage measurements, as a function of total ligand concentration used during the titration experiments can be used to determine K_d. Fitting of equation 7 was performed using a customized non-linear curve fit in Origin 8 (OriginLab Corporation, Northampton, MA). The results from each experimental replicate were fit individually, and then an averaged K_d value from at least three experimental replicates is reported.

Solvent accessible surface area (SASA) calculations.

Solvent accessible surface areas (SASA) for individual residues were calculated from a given protein’s PDB structure using GETAREA.³¹ The probe radius was set to 1.4 Å with no gradient calculation applied. The reported SASA percentage is the ratio of side-chain surface area in the protein to the random coil value for that same residue in the tripeptide Gly-X-Gly.

Results and Discussion

Effect of Labeling Extent on the Apparent K_d Value.

We predict that CL-MS based on slow labeling reactions is capable of determining the K_d values of a protein-ligand complex as long as the overall fraction of labeled protein is kept at low levels such that the perturbation to the equilibrium is limited. To illustrate this idea, consider a hypothetical protein-ligand system with a real K_d value of 10 μM (K_d1) that upon CL of the free protein results in a significantly weaker interaction with the ligand and a new K_d value of 1000 μM (K_d2). The concentration relationship between each fraction in the system can be described by equations 8-11, where [L] is the free ligand concentration at equilibrium, [P] is the unlabeled protein concentration, [P’] is the labeled protein concentration, [PL] is the concentration of the unlabeled protein-ligand complex, [P’L] is the concentration of the labeled protein-ligand complex, [L₀] is total ligand concentration, and [P₀] is the initial total protein concentration.

$$[L]+[PL]+[P^{\prime }L]=[L_0]$$ (8)

$$[P]+[P^{\prime }]+[PL]+[P^{\prime }L]=[P_0]$$ (9)

$$K_{d1}=\frac{[P][L]}{[PL]}$$ (10)

$$K_{d2}=\frac{[P^{{}_{\prime }}][L]}{[P\prime L]}$$ (11)

If one considers the percentage of the labeled protein as described in equation 12, then the apparent K_d will be defined by equation 13.

$$\%\text{labeling}=\frac{[P^{\prime }]+[P^{{}_{\prime }}L]}{[P_0]}$$ (12)

$$App{K}_d=\frac{[P+P^{{}_{\prime }}]\times [L]}{[P^{\prime }L+PL]}$$ (13)

The relationship between the apparent K_d and the percent labeling can then be plotted for any labeling extent between 0 and 100% (Figure 1). Not surprisingly, when there is no labeling, the apparent K_d is equal to the real K_d (i.e. 10 μM in this hypothetical situation), whereas at 100% labeling the apparent K_d is equal to the new K_d of the more weakly binding labeled protein (i.e. 1000 μM in this hypothetical situation). More interesting is the observation that when the percent labeling is kept low (e.g. below 30%), the apparent K_d differs from the real K_d by less than a factor of 2, meaning that as long as the extent of labeling is kept low, reasonably accurate K_d values can be obtained. It should be pointed out that often the variation in measured K_d values for protein-ligand complexes differ by more than a factor of 2 when different measurement methods are compared.³² Of course, if the percent labeling is much higher (e.g. above 70%), the apparent K_d value is more than an order of magnitude higher than the real K_d value, which is typically unacceptably inaccurate.

Figure 1:

(Top) Relationship between the apparent K_d and the real K_d for a hypothetical model in which the real K_d is 10 μM and the K_d upon 100% labeling is 1000 μM, upon considering equations 8-13 in the text. (Bottom) Expanded view of the theoretical plot, illustrating that when the percent labeling is about below 60%, the apparent K_d is within 5-fold of the real K_d, and is within 2-fold of the real K_d when the labeling percent is below 30%.

Lysozyme-NAG₃ Binding.

To test our idea, the first model system that we selected was the NAG₃-lysozyme complex. Because the NAG₃ binding site includes two Trp residues, Trp62 and Trp63 (PDB 1HEW),^33,34 we used HNSB, which is a Trp-specific labeling reagent. Reactions of lysozyme with HNSB in the presence and absence of NAG₃ result in only Trp62 being labeled (Figure 2a) as indicated by LC/MS/MS data (Figure S1). The solvent accessible surface area (SASA) of Trp63 is very low, hindering its reaction with HNSB. Because only Trp62 is labeled, the labeling ratio of the intact protein could be used during the ligand titration experiments.

Figure 2.

HNSB covalent labeling of intact lysozyme (30 μM) in the presence of (NAG)₃ using a 1.5 mM concentration of HNSB and 10 s reaction time. (a) Lysozyme surface structure with the (NAG)₃ binding site shown from PDB 1HEW. Trp62 is indicated in blue, and (NAG)₃ is in orange. (b) HNSB covalent labeling percentage of intact lysozyme as a function of (NAG)₃ concentrations. (c) HNSB labeling results fit using equation 7, resulting in a lysozyme-(NAG)₃ K_d of 14 ± 2 μM. These experiments were done in triplicate. Fitting statistics for are found in Table S1 in the Supporting Information.

To determine the K_d value using CL-MS, a titration curve was generated in which the NAG₃ concentration was varied between 0 and 600 μM at a fixed lysozyme concentration of 30 μM. HNSB was added at a concentration of 1.5 mM and allowed to react with the protein for 10 s at each NAG₃ concentration. This titration experiment resulted in a decrease in the percent labeling with increased total ligand concentration (Figure 2b). As expected, increasing NAG₃ concentrations lead to more protein molecules in which Trp62 is protected from labeling, leading to a decrease in the percent labeling. The HNSB labeling data were then plotted and fit using equation 7 (Figure 2c), and a K_d of 14 ± 2 μM is obtained. Previous measurements of the lysozyme-NAG₃ K_d by methods such as fluorescence, UV, and ESI-MS provide values that range from 6 to 60 μM.^32,35-39 Our solution conditions are most similar to those reported by Zenobi and co-workers³² in which they obtained a K_d value of 20 ± 4 μM. Clearly, our CL approach is able to afford a value that is consistent with prior studies.

To investigate if the CL modification extent significantly perturbs the ligand-binding equilibration (leading to variations in the apparent K_d), additional HNSB labeling conditions were explored in the context of the same titration experiment. Labeling with 1.5 mM HNSB for 30 s and 0.75 mM HNSB for 10 s resulted in more and less extensive labeling, respectively, than the conditions used to generate Figure 2b. These resulting titration curves are shown in Figure 3a, and the corresponding fits to equation 7 are shown in Figures 3b and 3c. Comparing the K_d values from the three labeling conditions, the results are not significantly different from each other at 95% confidence interval. Moreover, there is no clear trend in the K_d value with the extent of labeling, indicating that the labeling extent does not have a strong influence on the resulting K_d value at these low labeling extents. The reason for this is likely the fact that the extent of the protein labeling at higher ligand concentrations is very low (< 10% for almost all ligand concentrations), resulting in very minimal perturbation to the protein-ligand binding equilibrium. Indeed, HNSB concentrations 10 times higher and reaction times twice as long do not lead to labeling levels beyond 20% when the ligand is present at high concentrations (data not shown).

Figure 3.

HNSB covalent labeling of intact lysozyme (30 μM) in the presence of (NAG)₃ at different extents of labeling. (a) HNSB labeling extent for intact lysozyme at three different labeling conditions as a function of (NAG)₃ concentration. (b) HNSB labeling at 1.5 mM HNSB and a 30 s reaction time fit to equation 7, resulting in a lysozyme-(NAG)₃ K_d of 19 ± 2 μM. (c) HNSB labeling at 0.75 mM HNSB and a 10 s reaction time fit to equation 7, resulting in a lysozyme-(NAG)₃ K_d of 17 ± 4 μM. The experiments in each case were done in triplicate. Fitting statistics for are found in Table S1 in the Supporting Information.

Maltose-Binding Protein-Maltose Binding.

The second model system that we investigated was the maltose binding protein (MBP) bound to maltose (Figure 4a). This system allowed us to investigate if accurate K_d values can be acquired by using the extent of labeling at multiple residues. HNSB (1.0 mM) and DEPC (300 μM) were separately used to label the protein to an average labeling extent of 1 to 1.2 labels per protein in the absence of the ligand. Upon comparing MBP labeling in the presence and absence of maltose, we find that Trp340 from the HNSB labeling and Lys297 from the DEPC labeling show statistically significant decreases upon ligand binding (Figure S2a). These residues are expected to decrease in labeling because based on the crystal structures for the protein and its complex with maltose, their SASA values all decrease upon ligand binding.⁴⁰ The SASAs of Trp230 and Trp232 also decrease upon ligand binding (Table S2) and are labeled by HNSB, but it was difficult to get reliable labeling extents for these residue because the peptides containing these residue had significant day-to-day variations in the degree of oxidation at Met224, which is present in the same peptide as these residues. The reasons for these variations in Met224 oxidation are unclear.

Figure 4.

(a) Surface structure of MBP in the ligand-free “open” state (PDB 1OMP) and the ligand-bound “closed” state (PDB 1ANF). (b) HNSB labeling extent of Trp340 in MBP as a function of maltose concentration. (c) DEPC labeling extent of Lys297 in MBP as a function of maltose concentration. (d) HNSB labeling results for Trp340 fit using equation 7, resulting in a MBP-maltose K_d of 7 ± 4 μM. (e) DEPC labeling results for Lys297 fit using equation 7, resulting in a MBP-maltose K_d of 4 ± 2 μM. The concentration of MBP was 24 μM in each experiment, and the experiments were done in triplicate. Fitting statistics for are found in Table S1 in the Supporting Information.

Separate titrations were conducted with HNSB labeling (Figure 4b) and DEPC labeling (Figure 4c) with a fixed MBP concentration of 24 μM, while the concentration of maltose was varied from 0 to 285 μM. As was observed for the lysozyme-NAG₃ system, the extents of labeling at Trp340 from HNSB labeling and Lys297 from DEPC labeling decrease as the maltose concentration is increased. Although, because the MBP measurements are from peptide fragments after digestion and LC/MS, the errors are larger than observed for the lysozyme system, particularly for Lys297, which undergoes a very low extent (< 1%) of DEPC labeling.

From the HNSB labeling data of Trp340, a K_d value of 7 ± 4 μM is obtained (Figure 4b and d). This value is comparable to a K_d value of 1 to 3 μM that was measured previously by fluorescence.^41-44 Interestingly, other Trp residues (e.g. Trp6, Trp10, Trp94, Trp129, and Trp158) do not change in SASA upon ligand binding (Table S2) and therefore do not undergo any significant changes in labeling extent. As examples, Trp10 and Trp158 are modified by HNSB, but because these residues are distant from the ligand-binding site, their extents of labeling do not change significantly as the maltose concentration is increased (Figure S2c). Residues like Trp10 and Trp158 serve as useful controls for residues like Trp340 that undergo changes in labeling upon ligand binding.

We also explored titrations with DEPC because it can modify up to six different types of amino acids, including Lys, His, Tyr, Ser, Thr, and Cys residues.¹⁶ Even though DEPC can label so many residues, only Lys297 undergoes a significant decrease in labeling extent in the presence of maltose. Indeed, of the 25 residues in MBP that can be labeled by DEPC, Lys297 undergoes the greatest percent change in SASA value (from 38% to 23%) upon maltose binding (Table S2). Upon plotting the labeling data for Lys297, a K_d value of 4 ± 2 μM is obtained (Figure 4e). This value is similar to the value obtained from HNSB labeling of Trp340 and is comparable to the literature value for the MBP-maltose complex.^41-44 It is worth noting that the data for Lys297 provides a reasonable measure of the K_d even though it is not a residue directly interacting with maltose. This observation demonstrates that residues in regions that undergo a structural change upon ligand binding can also be used for K_d determination. Several other residues, such as Lys26, Lys239, and Lys295 (Figure S2d), undergo no change in labeling upon maltose binding, and thus serve as useful controls.

Like the HNSB labeling experiments of the NAG₃-lysozyme complex, the low extent of labeling observed at MBP residues that undergo changes in SASA (i.e. Trp340 and Lys297) upon ligand binding facilitates determination of a K_d value that is close to the literature value. In the case of MBP, the extents labeling of Trp340 and Lys297 are below 25% and 0.5%, respectively. Because labeling by both HNSB and DEPC is limited to about one label on average per protein molecule, which is spread across multiple modifiable residues all over the protein, the extent of modification at any given residue, including binding residues, is low (see Figure S2a). Thus, the number of modified protein molecules that might perturb the protein-ligand equilibrium is expected to be low.

β-2-microglobulin-EGCG Binding.

We next applied CL-MS to determine the K_d for epigallocatechin gallate (EGCG) bound to β-2-microglobulin (β2m). β2m forms amyloid fibrils in patients who undergo long term dialysis due to kidney failure,⁴⁵ and EGCG (Figure 5a) has recently been found by our group to prevent Cu(II)-induced β2m amyloid formation in vitro by redirecting β2m aggregation toward amorphous, re-dissolvable aggregates.⁴⁶ Thus, we decided to apply a CL-MS based method to determine the K_d of this complex.

Figure 5.

EGCG binding to Cu(II)-β2m results in decreases and increases in DEPC labeling at several residues. (a) Structure of EGCG. (b) Residues undergoing covalent labeling with DEPC mapped onto the Cu(II)-free structure of β2m (PDB 1JNJ). Note that there is no structure is available for Cu(II)-β2m. DEPC modified residues that undergo no significant change upon ligand binding are shown in cyan. Those residues that decrease or increase in labeling extent are shown in blue and red, respectively. The proposed EGCG binding site is indicated by the black circle.

EGCG concentrations ranging from 0 to 115 μM, a β2m concentration of 30 μM, and a Cu(II) concentration of 60 μM were used in a titration experiment along with DEPC labeling, using 120 μM of the labeling reagent. DEPC labeling at high EGCG concentrations indicate that several residues undergo both increased and decreased extents of labeling upon ligand binding, including the N-terminal amine, Thr4, Lys6, His13, His31, and Lys91 (Figure S3). Most of these residues are near the N-terminus region of the protein (Figure 5b), suggesting EGCG’s binding site is between one β sheet that includes residues from Lys6 to Ser11 and another that includes residues from Lys91 to Asp96.

Interestingly, some of the residues that undergo increased labeling, specifically the N-terminal amine and His31, are residues that comprise the Cu(II) binding site in the Cu(II)-β2m complex.^47,48 The increased reactivity of these residues with DEPC as the EGCG concentration is increased suggests that EGCG is disrupting the Cu(II) binding site to some extent. Indeed, the K_d for Cu(II) bound to β2m was found to increase from 3 μM to 49 μM in the presence of EGCG,⁴⁶ which decreases the concentration of the β2m-Cu(II) complex. The reduced concentration of the metal-bound protein means that when EGCG is present the N-terminus and His31 are free in more protein molecules to react with DEPC.

DEPC labeling as a function of EGCG concentration for residues that both increase and decrease in labeling in the presence of EGCG are shown in Figure 6. The K_d values that are acquired from the labeling results of the N-terminal amine, Lys6, and Lys91 are 12 ± 5 μM, 3 ± 2 μM, and 11 ± 5 μM, respectively. These values are reasonably similar, considering the error bars, and they indicate an EGCG-β2m K_d value of between 3 and 12 μM. It is interesting to note that although the increase in the labeling of the N-terminal amine is likely due to the release of Cu(II) caused by ligand binding, the resulting value is consistent with the K_d value obtained from the residues that are protected from DEPC labeling. The K_d values acquired by CL-MS are also consistent with the K_d value of 6 μM that was obtained from size exclusion chromatography data in previous work.⁴⁶

Figure 6.

(a) DEPC labeling extent of N-terminus, Lys6, and Lys91 in Cu(II)-β2m as a function of EGCG concentration. (b) DEPC labeling results for the N-terminus, fit using equation 7, resulting in a Cu(II)-β2m and EGCG K_d of 12 ± 5 μM. (c) DEPC labeling results for Lys6, fit using equation 7, resulting in a Cu(II)-β2m and EGCG K_d of 3 ± 2 μM. The concentration of β2m was 24 μM in each experiment, and the experiments were done in triplicate. Fitting statistics for are found in Table S1 in the SI.

Conclusions

We have demonstrated that CL-MS can be used to obtain reasonably accurate K_d values for protein-ligand complexes. In particular, we find that slow-reacting labeling reagents like DEPC and HNSB can provide this information even when labeling from these reagents might be expected to perturb the equilibrium of the protein-ligand complex. We find that accurate K_d values can be obtained when the extent of protein labeling is kept low. The experimental results from two model systems and one insufficiently characterized protein-ligand system support this conclusion and show that our CL-MS based strategy is able to determine protein-ligand binding affinity with reasonable accuracy. An intriguing prospect of using CL-MS to determine K_d values is the ability to simultaneously identify the ligand binding site, if the residue level labeling results contain sufficiently detailed information. This CL-MS based ligand titration strategy might serve as an alternative for characterizing protein-ligand complexes that are difficult to measure by other methods, such as fluorescence spectroscopy, surface plasmon resonance spectroscopy, or nuclear magnetic resonance spectroscopy.