Introduction

Despite the rapid development of protein-based therapeutic biologics, small molecules are still remarkably dominant in development pipelines of the biopharmaceutical industry, making up over 90% of the therapeutics in use [1]. Small-molecule drugs have relatively stable chemical properties and are mostly non-immunogenic. Their size and chemical composition often allow them to penetrate cell membranes and reach desired delivery destinations. Generally, small-molecule drugs are designed to inhibit or modulate the function of specific biological macromolecules (e.g., protein, DNA). Thus, characterization of the interactions between a small molecule inhibitor/modulator and its target macromolecule is important for drug development, as knowledge of these interactions is essential for detailed understanding of the molecular mechanism of action.

MS-based protein footprinting is a valuable tool to characterize protein structure, dynamics, and interactions with small molecules. Hydrogen-deuterium exchange (HDX) and hydroxyl radical footprinting are the two most commonly applied protein footprinting methods. HDX reports the stability and structural protection of the protein by measuring the exchange of amide hydrogens with deuterium on the protein backbone [2]. Its utility has been established by extensive applications in studying protein-protein, antibody-antigen, protein-DNA, and protein-membrane interactions [3,4,5], showing successes in probing large interfaces in protein-ligand complexes. The use of HDX to probe protein-small molecule interfaces has also been widely reported [6,7,8,9].

The relatively small protection afforded to the protein by the small molecule, however, poses potential challenges to the HDX sensitivity. Dai et al. [10] utilized HDX to probe the structural dynamics of the estrogen receptor in complex with various small molecule modulators, and used this information to classify the modulators to correlate with their pharmacological profile. Hernychova et al. [11] applied HDX to the interaction of the protein MDM2 with a small molecule, Nutlin3, and observed reduced HDX kinetics upon ligand binding in regions surrounding the pocket only at relatively high ligand-to-protein ratio (4:1). Wang et al. [12] used HDX with ligand titration to obtain the affinities of a small molecule drug with Apolipoprotein E3 at the peptide level, observing convincing changes in HDX at the binding sites. Despite these successes, direct mapping of a small molecule interaction when the binding is hydrophobic may be particularly difficult because the binding interface does not involve hydrogen bonding of the protein backbone, explaining a lack of sensitivity to binding for HDX.

By contrast, hydroxyl radical footprinting reports changes in solvent accessibility of amino acid side chains via covalent and irreversible oxidative modification. Here, differences in bound vs. unbound states are less dependent on hydrogen bonding. Furthermore, hydrophobic side chains (Phe, Leu, Ile, Val) are reactive with ·OH, leading to potentially sensitive indications of binding [13].

Here, we describe an implementation of fast photochemical oxidation of protein (FPOP), which uses laser-induced hydrolysis of hydrogen peroxide to generate hydroxyl radicals [14, 15]. Using a radical scavenger, we varied the timescale of labeling under the assumption that the radical footprinting is faster than protein conformational change or unfolding induced by modification [16]. Although FPOP has been applied to protein folding, protein aggregation [17,18,19], and protein-ligand interactions including epitope/paratope mapping [20,21,22,23,24], to our knowledge, FPOP and hydroxyl radical footprinting in general have not yet been applied to protein/small-molecule interactions. Herein, we test the capability of the method for this application.

We chose to compare HDX with hydroxyl-radical-based FPOP for the interactions of a benzodiazepine inhibitor (compound 1, Figure 1(a)) and human bromodomain-containing protein 4 (BRD4). BRDs are protein interaction modules that specifically recognize acetylation motifs, a key event in the reading process of epigenetic marks [25]. Inhibitors targeting BRD have therapeutic efficacy as anti-inflammatory, antiviral, and anticancer agents [26]. Compound 1 is a potent (nM) inhibitor that disrupts the function of the bromodomain family (BRD2, BRD3, and BRD4). A cocrystal structure of the bromodomain 1 of BRD4 and compound 1 shows that the inhibitor occupies the central hydrophobic cavity of the protein used for acetyl-lysine recognition, thus directly antagonizing the interaction between the bromodomain and the acetylated histone peptides by steric competition [26]. The outcome of our comparative study not only is valuable for understanding the principles of HDX and FPOP but also provides insight into the structural capabilities of the methods, guiding the choice of suitable MS-based footprinting methods for investigating protein structure and interactions.

Experimental

Materials

The bromodomain 1 (residue 41–168) of human BRD4 (referred to as BRD4 hereafter) and compound 1 was provided by Bristol Myers Squibb. l-Glutamine, l-methionine, catalase, hydrogen peroxide (H2O2), trifluoroacetic acid (TFA), formic acid (FA), phosphate buffer saline (PBS, 10 mM phosphate, 138 mM NaCl, 2.7 mM KCl), urea, dithiothreitol, and iodoacetamide were from Sigma Aldrich (St. Louis, MO). Trypsin and chymotrypsin were from Promega (Madison, WI).

HDX

To prepare compound 1-bound BRD4, the protein solution and compound 1 were combined to form a 40 μM:48 μM concentration ratio and incubated at 25 °C for 30 min to allow formation of the complex. Aliquots of 3 μL of 40 μM BRD4 with and without compound 1 were then continuously labeled in 37 μL of D2O in 1 × PBS (pH was adjusted to 7.0 to give a pD of 7.4) at 25 °C at various time intervals (10 s, 30 s, 1 min, 2 min, 15 min, 1 h, 2 h, 4 h, and 6 h). The reaction was quenched by adding 60 μL of 3 M urea and TFA (final pH = 3.0). Samples were then flash frozen in liquid nitrogen and stored at - 80 °C. Control samples were prepared in like fashion by using 1× PBS in water in place of deuterium oxide. Each sample was thawed immediately prior to LC-MS analysis. The protein was digested with pepsin by using a custom-packed pepsin column (2 mm × 20 mm) at 25 °C at a flow rate of 100 μL/min. The resulting peptic peptides were then trapped on a Zorbax Eclipse XDB-C-8 trap column (2.1 × 15 mm, 3.5 μm) and desalted for 3 min, followed by separation of the peptides on a Hypersil Gold C-18 analytical column (2.1 × 50 mm, 2.5 μm) by using a gradient of 4–80% acetonitrile with 0.1% formic acid at a flow rate of 100 μL/min. The trap and analytical columns, as well as the tubing for protein digestion and separation, were placed at 4 °C in the ice bath to minimize back-exchange of the digested peptides. Peptides were detected using a LTQ XL Orbitrap mass spectrometer (Thermo Fisher Scientific), at a mass resolving power of 100,000 at m/z 400. The capillary temperature and voltage for ESI were 275 °C and 49 V, respectively. All experiments were performed in duplicate.

HDX Data Analysis

To identify the peptic peptides generated from the protein and to provide a list of peptides to be followed during HDX, peptide mapping with a non-deuterated protein sample was conducted. The product-ion mass spectra were acquired in a data-dependent mode, with six most abundant ions from each scan selected for MS/MS. Peptides, as identified using Mascot (Matrix Science, London, UK), were used to guide the HDX setup. The HDX mass spectra were analyzed with HDX Workbench (Scripps, Jupiter, FL). The *.raw files for all HDX runs were imported into HDX Workbench to calculate the centroid mass of the isotopic envelope (m) of the peptide. The percentage of the deuterium level was calculated by deuterium (%) = {[m(P) − m(N)]/[m(F) − m(N)]} × 100, where m(P), m(N), and m(F) are the centroid values of the partially deuterated peptide, non-deuterated peptide, and fully deuterated peptide, respectively. All HDX were performed in duplicate. The error of the deuteration (%) at each time point was represented by the standard deviation from the duplicate runs.

FPOP

A complex of bound BRD4 with compound 1 was prepared by incubating 20 μM BRD4 with compound 1 in a 1:1 M ratio for 30 min at 25 °C in PBS. Unbound BRD4 was prepared as a 20 μM stock solution in PBS. The FPOP protocol used here was based on a previous report [14] with minor changes. The 248-nm KrF excimer laser (GAM Laser Inc., Orlando, FL) used to irradiate the sample solution was adjusted to 32 mJ/pulse. The laser bean was focused onto 150-μm i.d. fused silica tubing (Polymicro Technologies, Pheonix, AZ) to give a 2.2-mm irradiation window (see ref. 15 for detailed setup of the laser and the optical bench). The flow rate and pulse frequency were adjusted to ensure 20% of the sample solution volume was not irradiated between laser shots. The laser pulse frequency was controlled by an external pulse generator (B&K Precision, Yorba Linda, CA). H2O2 was added to each sample to a final concentration of 20 mM prior to injection into the FPOP capillary. The final concentration of BRD4 for FPOP labeling was 10 μM. In the FPOP labeling, equal amount of time (45 s) was allowed to elapse between mixing of the protein with H2O2 and starting the syringe pump. This amount of delay was applied to all samples submitted to FPOP to ensure consistent exposure time of the protein to H2O2 prior to laser illumination. Samples of 50 μL were injected through the FPOP apparatus and collected in tubes containing 10 μL of 10 mM catalase and 20 mM methionine to react with any remaining H2O2 and prevent oxidation artifacts. In addition, FPOP control of BRD4 mixed with all the FPOP reagents, including H2O2, was handled in the same manner as the samples, but not laser-irradiated. FPOP labeling and control was conducted in triplicate at four different concentrations of His (0, 0.3, 3, 30 mM) in a sequential manner on the same day under the same conditions. The labeled sample was stored in  -80 °C for future use.

LC-MS/MS

For intact-level BRD4 characterization, 10 μL of the FPOP-labeled sample was loaded onto a Zorbax Eclipse XDB-C-8 trap column (2.1 × 15 mm, 3.5 μm) and eluted at a flow rate of 100 μL by using an Agilent 1100/1200 separation system. Solvent A was water with 0.1% (v/v) FA, and solvent B was 80% (v/v) acetonitrile with 0.1% (v/v) FA. The gradient started from 5% B, increased to 15% B in 0.3 min, ramped to 50% B in 5.2 min, increased to 100% B in 0.5 min, held at 100% B for 0.5 min, returned to 5% B in 0.1 min, and equilibrated at 5% B for 2 min. The eluted protein was submitted to a MaXis 4G quadrupole-time-of-flight mass spectrometer (Bruker Daltonics, Billerica, MA). The mass spectrometer was operated in the positive-ion electrospray ionization mode at a mass resolving power of 30,000 (m/z 400).

For peptide and residue-level characterization, urea was added into each 20 μL aliquot of the FPOP sample to a final concentration of 6 M to denature the protein. Disulfide bonds of the protein were reduced by adding dithiothreitol to a final concentration of 5 mM, and samples were incubated at 37 °C for 30 min. Iodoacetamide was subsequently added to a concentration of 15 mM to alkylate the free cysteines. Samples were incubated in the dark at room temperature for 30 min. PBS was added to dilute the urea to 1 M. Samples were digested overnight with trypsin or chymotrypsin at 37 °C with a protease-to-protein ratio of 1:25 (w/w). The digestion was terminated by acidifying the sample with 1% trifluoroacetic acid. The sample was then vacuum-dried on a speedvac, and each peptide digest was reconstituted with 60 μL 0.1% FA in water for LC-MS analysis. An aliquot of 5 μL of protein digest was submitted to LC-MS/MS analysis. Solvent A and B were the same as above. Peptides were pre-concentrated on an Acclaim PepMap C18 column (Thermo Scientific, 100 μm × 2 cm, 5 μm, 100 Å) and desalted for 10 min at a flow rate of 4.5 μL/min by using solvent A. Separation was performed on a custom-packed C18 column (CSH, 75 μm × 15 cm, 3.5 μm, 130 Å) at a flow rate of 400 nL/min with a gradient of 2–65% solvent B in 90 min using a Nano UltiMate 3000 Rapid Separation system (Dionex, Co.). The solution flow was directed with a Nanospray Flex source coupled for online detection using a Q Exactive Plus orbitrap mass spectrometer (Thermo Scientific) with a spray voltage of 2.5 kV and a capillary temperature of 250 °C. In the data-dependent mode, the 15 most abundant ions were submitted to higher energy collision dissociation (HCD) with an isolation window of m/z 1.5 and a normalized collisional energy of 30%. Resolving power was 70,000 (at m/z 400) for MS1 and 17,500 (at m/z 400) for product ions in MS/MS.

FPOP Data Analysis

Identification of the unmodified peptides and assigned modifications were done by using Byologic (Protein Metrics) and further validated by manual inspection. Modification sites were identified based on MS/MS. Signal intensities of the unmodified peptide (Iu) and its modified species (Iox) were integrated using Byologic (Protein Metrics) from the extracted ion chromatograms (XICs). The FPOP modification level of a peptide was calculated using the following equation: % modified = Iox/(Iox + Iu) × 100. Quantification of the modified species was based on the dominant FPOP products with oxygen addition/substitution(s) (+ 15.9949 Da, + 31.9898 Da, etc.). All the FPOP oxidation levels reported in Table 1 (for the reporter peptide), Figure 4 and SI Fig. 3 (for BRD4 peptides) were corrected by subtracting the background oxidation determined from the un-illuminated FPOP controls. The error of FPOP modification was represented by the standard deviation from the triplicate experiments.

Table 1 FPOP Modification Levels of the Reporter Peptide. Extent of modification of the reporter in the unbound and bound BRD4 sample. Scavenger concentrations of His are 0, 0.3, 3, and 30 mM, respectively. FPOP labeling was done in duplicate.
Full size table

Results and Discussion

Mapping by HDX

In our model system, benzyl (1-methyl-6-phenyl-4H-benzo[f][1,2,4]triazolo[4,3-a][1,4]diazepin-4-yl)carbamate (compound 1) (Figure 1A) is a competitive inhibitor of BRD4, which possesses a binding cavity for acetyl-lysine recognition (Figure 1B) [26]. The structure of bromodomain adopts a left-handed, four-helix bundle topology. Superimposing the X-ray crystal structure of compound 1-bound BRD4 (bound) on that of the ligand-free protein (unbound) shows the two structures are nearly identical (Figure 1C), indicating that no significant structural rearrangement of the protein is caused by compound 1 binding.

Figure 1
figure 1

Binding of BRD4 to compound 1. (a) Molecular structure of compound 1. (b) Crystal structure of BRD4 in complex with compound 1 (PDB 2YEL), referred to as “bound BRD4” hereafter. Secondary structures and surface presentation of the protein are shown in cyan and white, respectively. (c) Bound BRD4 in cyan is overlaid with the ligand-free BRD4 in orange (PDB 4LYI), referred to as “unbound BRD4” hereafter. Images generated with the PyMOL molecular graphics system, version 2.0 Schrödinger, LLC.

Full size image

In HDX, the hydrophobic core of BRD4 is expected to show a very slow rate of deuterium uptake owing to solvent inaccessibility. Thus, we extended the time course of D2O exposure to 6 h to allow exchange to occur in the hydrophobic cavity (SI Fig. 1A). The sequence coverage for the peptic peptides in HDX is 100%. In accord with the nearly identical structures of BRD4 in the unbound and bound states, most regions in BRD4 show no change in HDX kinetics upon compound 1 binding, as represented by peptides in Figure 2A–D. By way of contrast, we found some regions exhibit slightly reduced rates and extents of exchange in the bound state (Figure 2E–J). However, the differences are too modest compared with what we would typically expect for a binding interface [27,28,29], and statistically they are inconclusive to determine the binding site (SI Figure 1B). Instead, HDX shows that BRD4 only undergoes a slight decrease in hydrogen bonding and flexibility upon compound 1 binding, and we conclude that the overall structures of the unbound and bound BRD4 proteins are very similar. Indeed, we are seeking a binding scenario such as this to test whether protein footprinting is sensitive to small-molecule binding.

Figure 2
figure 2

HDX kinetics of BRD4 upon binding of compound 1. Comparison between HDX of unbound (blue) and bound (orange) BRD4 reveals representative peptide regions (a–d) that do not show changes in HDX upon binding, and regions (e–j) showing small changes upon binding to compound 1. Amino acid numbering and charge state of the peptide from pepsin digestion of BRD4 are shown above each plot. See experimental section for the calculation of the deuterium percentage and error.

Full size image

Given the small changes in HDX resulting from binding, we turned to FPOP to probe the interaction of BRD4 with compound 1. Compared with HDX, which interrogates the structure and dynamics of backbone amides, we expected FPOP to show higher sensitivity in mapping the binding cavity of BRD4 with compound 1, because the binding interface may involve sidechain interactions and, moreover, many of the hydrophobic residues are reactive with the OH radicals used for footprinting.

Mapping by FPOP using a Reporter Peptide Strategy

To obtain FPOP kinetic curves of the unbound and bound BRD4, we performed the labeling experiment with various concentrations (0, 0.3, 3, 30 mM) of histidine as the radical scavenger to give different timescales of labeling. In the previous FPOP experiment with a fixed scavenger concentration, we usually applied Student’s t test to access significance of the difference. In this work, instead of relying on modification under a single FPOP labeling to make the judgment, we chose to modify the protein under four different timescales. The exposure time to the hydroxyl radical is tuned by using different scavenger concentrations and reflected by the % modification of the reporter. This strategy permits a time-dependent FPOP measurement and comparison. The fitted curves give a relative trend of change as a function of the labeling time, offering higher sensitivity and credibility in identifying changed regions compared with a single-point FPOP experiment. Mass spectra of the intact BRD4 show that the extent of FPOP modification gradually attenuates in response to increased amounts of the scavenger (Figure 3). To track the lifetime of the hydroxyl radical in the sample, we used Leu-enkephalin (YGGFL) as a peptide reporter for the hydroxyl radical [30] by mixing each protein sample with a fixed amount of Leu-enkephalin prior to FPOP. Upon laser irradiation, the reporter is labeled competitively with BRD4 by the hydroxyl radical under the same conditions, and thus the reporter-% modified can be used to follow the hydroxyl radical lifetime in the sample.

Figure 3
figure 3

Global FPOP modification of BRD4 as a function of the scavenger concentration. Mass spectra of intact unbound BRD4 (+14 charge). FPOP was done without addition of the scavenger in the protein sample (a), and with 0.3, 3, 30 mM his used as scavenger (b–d). (e) Control sample of BRD4 (with addition of H2O2) submitted to the same FPOP workflow as the experimental group except the laser was turned off. (f) Starting state of BRD4 in PBS buffer.

Full size image

Compound 1, a polyphenolic molecule, is likely to be reactive with the hydroxyl radical, and if so, this ligand will quench some of the radicals. Thus, there may be a different overall oxidation potential for the bound sample solution compared with that containing the unbound protein. The modification of compound 1 will not compromise our efforts to locate the binding site provided the count of radicals available to the protein can be normalized for the unbound and bound states. In fact, the “unintended” quenching effect of compound 1 is negligible under high concentrations of scavenger (3 and 30 mM His) as indicated by the equal modification extents of the reporter for the unbound and unbound samples when submitted to FPOP (Table 1). In absence of the scavenger (His = 0 mM), however, the reporter for the bound sample is significantly less modified than for the unbound BRD4 (relative difference 32%). In the presence of 0.3 mM His, a small difference in the reporter-% modified is observed. Those results demonstrate that in FPOP, the modification extent is well controlled because the dominating chemistry that controls the lifetime of the hydroxyl radical is the reaction of the radical with the large excess of scavenger rather than that with the protein or ligand.

To obtain comprehensive coverage across the BRD4 sequence in the bottom-up proteomics strategy used for analysis, we performed digestion with trypsin and chymotrypsin separately. The sequence coverages from trypsin and chymotrypsin digestion are 47% and 64%, respectively (SI Fig. 2). The combined coverage from the two sets of digestion is 75%. The protein N and C-termini are not covered owing to the presence of multiple prolines on the N-terminus and possibly the loss of the hydrophilic C-terminal peptides in the chromatographic separation. Plotting the BRD4 peptide-% modified as a function of the reporter-% modified provides a “time-course” for the FPOP modification. The curves are similar to the HDX kinetic curves characterized by deuteration extents at different deuterium exposure times (Figure 4) that are commonly viewed in the field.

Figure 4
figure 4

FPOP kinetic curves of BRD4. The x-axis of each curve is the % modified of the reporter peptide, which correlates with the lifetime of the hydroxyl radical. The y-axis is the % modified of the peptide from the FPOP-modified BRD4 in unbound (blue) and bound (orange) states, respectively. Numbering and sequence of the peptides from digested BRD4 are shown above each plot. Identified residues with FPOP modification are highlighted in red on the peptide sequence. See experimental section for calculation of the FPOP modification and error.

Full size image

In FPOP quantification, we assumed equal efficiency and rate for proteolytic digestion of the unmodified and modified sites under denaturing conditions. For peptide regions not containing Met, the % modified for the protein shows a linear correlation with that of the reporter (Figure 4A–F, I, and K). Among those regions, 76–83 clearly exhibit reduced modification in the bound state. Therefore, this region is likely part of the binding site.

In contrast, FPOP modifications of peptide regions that contain one or more Met (e.g., 99–114, 100–112, 121–133, and 149–156) saturate and decrease as the lifetime of the hydroxyl radical becomes longer (Figure 4G, H, J, and L). The modifications occur nearly exclusively on Met residue(s). The turnover in the kinetic curves suggests sequential modifications, but we were unable to find certain higher oxidized products (e.g., + 32 Da, + 48 Da) by manual search even at long labeling times (peptides with multiple oxidation are usually rare, so they are not considered in the algorithm used here for peptide-level quantification). Indeed, the oxidation of Met is known to be extremely complicated [13] and produces a number of products, some likely unidentified. Xie and coworkers [31] reported similar difficulties with methionine and with cysteine residues when using dose-dependent FPOP to measure absolute solvent accessibility in two protein systems. In addition, the decreased Met modification at lower scavenger concentrations is also likely due to altered protein-ligand interaction. Because the polyphenolic compound 1 ligand is very prone to oxidation, it can be extensively modified under long exposure to the hydroxyl radical and result in altered structure and interaction with BRD4.

In the FPOP quantification, the intact- and peptide-level modifications reported are corrected by subtraction of the background oxidation from that of the un-illuminated control. Some of the regions (e.g., 76–83 and 100–112) showed “non-zero intercepts” at t = 0 (i.e., at 0% modification of reporter) if one simply extrapolates the curve linearly. In the time-dependent FPOP experiment, peptides showing “non-zero intercepts” are not unusual, especially for residues that have large intrinsic reaction rate constants for the hydroxyl radical (e.g., Met, Try, and Phe) [30, 32]. This is because the correlation between modification of reporter (x) and modification of BRD4 (y) is not necessarily linear, depending on the ratio between rate constants of sample peptide (ks) and the reporter (kr) towards the hydroxyl radical [30]. When ks is close to kr, the relationship between y and x is nearly linear. Although most curves can be well fitted as linear curves within the range of FPOP timescale (controlled by the scavenger concentrations) used in this work, a “burst” phase in modification on highly reactive residues in some regions could occur at shorter labeling times [30] and give an apparent “non-zero intercept”.

Correlating HDX and FPOP Mapping with BRD4 Structures

A cocrystal structure of the complex indicates that compound 1 binds to the acetyl-lysine recognition pocket of BRD4. Using proximity of BRD4 residues with compound 1 as a criterion, we considered that residues of the protein within 4 Å from compound 1 could be key contacts. Using this criterion, we identified W81 (3.6 Å), P82 (3.9 Å), F83 (3.7 Å), L92 (3.7 Å), and N140 (3.3 Å) as key residues that directly interact with compound 1 from the crystal structure (Figure 5A).

Figure 5
figure 5

Interaction of BRD4 with compound 1. (a) Key contacts (magenta) of the protein (cyan) in contact with compound 1 (blue) based on the X-ray structure. The distance criterion for a contact was chosen to be 4 Å. (b) Peptide regions that show reduced kinetics of HDX in the bound state (Figure 2E–J) mapped onto the BRD4 structure (PDB 2YEL). For those regions, ones showing low and intermediate rates in the deuterium uptake are colored in green and salmon, respectively. (c) Peptide regions showing reduced FPOP modification for the bound state (Figure 4D) are mapped onto the BRD4 structure and colored in magenta. Images were generated with PyMOL molecular graphics system, version 2.0 Schrödinger, LLC.

Full size image

HDX

Mapping the peptide regions that show reduced HDX in the ligand-bound protein (Figure 2E–J) onto the BRD4-compound 1 complex structure indicates that those regions span the binding pockets and the surrounding sequence (Figure 5B). As discussed earlier, the differences in HDX, however, are subtle and do not provide high confidence that a binding pocket has been identified. This may be because (1) the protein only undergoes subtle structural change upon compound 1 binding, as indicated by the nearly identical crystal structures of the unbound and bound protein, (2) the size of the ligand is too small to provide efficient shielding from the solvent at the binding interface, and (3) the binding is not significantly involved with peptide bonds but rather with side-chain interactions.

Despite these uncertainties, the kinetics of HDX provides information on the secondary structure elements of the region based on their effects in HDX. For example, regions 124–136 and 149–153 (colored in green in Figure 5B) that are comprised of two α-helix bundles are structurally rigid and resist solvent contact; therefore, they show low extents of HDX (Figure 2G, H, and J). By contrast, region 119–124 is a loop located on the surface of the protein. Its flexible structure and solvent exposure allow the region to take up deuterium rapidly and undergo extensive HDX even at short times (Figure 2C). Regions 72–88 and 137–145 (colored in salmon in Figure 5B) cover both fast and slow exchanges and show intermediate rates of HDX (Figure 2E and I).

FPOP

Differential labeling by FPOP shows that region 76–83, containing key contacts W81, P82, and F83, is decreased for the bound state (Figure 4D), clearly indicating a binding interface (Figure 5C). Unlike the large FPOP change usually observed in protein-protein interactions [20, 21], the difference at the putative binding region (76–83) with compound 1 is relatively small owing to the weak protection from a small molecular ligand. Despite the relatively modest change, the difference in region 76–83 is significant because the data with four reporter concentrations are consistent in term of change, and the fitted curve provides confidence in the relative trend of change. Furthermore, an overlapping peptide 77–91 (from separate trypsin digestion) containing the putative binding interface also shows an unambiguous shift in FPOP modification between the two states (SI Fig. 3). The consistent information from the overlapping peptides serves as a cross-validation of the difference seen by FPOP and adds more credibility in assigning the binding interface. Further residue-specific analysis of this region indicates the FPOP modification occurs only on W81, which is at the entrance of the binding pocket and accessible to the solvent (and H2O2) (Figure 2A). No modifications were detected on P82 and F83 presumably because the two residues are buried deep in the hydrophobic pocket and have low solvent accessibilities (Figure 5A). The three residues W81-P82-F83 constitute the conserved WPF shell motif in the bromodomain family; this domain plays an important role in positioning inhibitors through hydrophobic van der Waals interactions [33, 34].

In addition, L92 and N140 are in contact with compound 1 based on the crystal structure. Regions covering the two residues (84–92 and 138–148) show overlapping FPOP kinetic curves in unbound and bound BRD4 (Figure 4E and K). Although modification was detected on L92, we found no difference in its modification levels between unbound and bound protein. This is likely due to the insufficient solvent protection from compound 1 and the structural flexibility of L92 in a dynamic loop (Figure 5A). Owing to the low intrinsic reactivity of Asn with the hydroxyl radical, N140 did not undergo modification in FPOP. Instead, modification was detected on K141 and P142 nearby (Figure 4K).

Although Met tends to exhibit large changes in FPOP modification in response to minor structural perturbations (SI Fig. 4), presumably due to its high intrinsic reactivity with the hydroxyl radical [21, 35], caution is needed when interpreting changes in modification extents on Met and associating them with major structural effects. Furthermore, FPOP may result in a “false positive” in suggesting a conformational change associated with Met, but this problem can be avoided by performing the experiment under a high concentration of the scavenger (e.g., ~ mM His), because at short labeling times the modification extents on the Met for unbound and bound are more modest and, importantly, are identical (Figure 4G, H, J, and L). More research is required to understand this effect and the mechanism of Met on FPOP yields.

Conclusions

Results from HDX show very similar kinetics for the unbound and compound-1 bound BRD4, indicating the sensitivity of the method is not sufficient in this case to map directly the intermolecular interface because the shielding of solvent by the small molecule binding in a hydrophobic pocket is relatively insignificant. In FPOP, however, the region containing the conserved WPF shell and residues nearby show a clear decrease in FPOP modification upon compound 1 binding, consistent with the predicted interactions at the binding interface determined by crystallography. Although there may be other residues in the vicinity of the binding, they do not vary in their solvent accessibility reported by FPOP.

In addition, regions containing Met residues display distinct chemical kinetics in FPOP, including saturation in the modification at extended labeling time and a relatively large difference in modification resulting from a modest change in protein dynamics. Those differences seem to be false positives and require further evaluation.

Although the specific interactions between BRD4 and a small molecule (compound 1) are difficult to discern based on FPOP or HDX data, the small differences nevertheless provide valuable insights regarding the location of the small molecule binding site on the protein. These insights can be used to inform molecular modeling studies directed towards binding site identification and protein-ligand binding model generation. This strategy of molecular modeling guided by experimental data is particularly useful when a high-resolution structure of the protein-ligand complex cannot be obtained via X-ray crystallography or NMR methods.