Automated specific amino acid footprinting mass spectrometry: repurposing an HDX platform for determining reagent feasibility

Abstract

Protein footprinting is a mass spectrometry (MS)-based approach to measure protein conformational changes. One approach, amino acid specific labeling, imparts an irreversible modification to protein side chains but requires careful selection of the reactive reagent and often time-consuming optimization of experimental parameters prior to submission to bottom-up analysis. In this work, we repurpose a hydrogen-deuterium exchange MS (HDX-MS) LEAP HDX system for automated specific amino acid footprinting MS, demonstrating its efficacy in reaction optimization and monitoring applicability to specific ligand binding systems. We screened reagent conditions for two model ligand-binding systems and demonstrate the method’s efficacy for measuring differences induced by ligand binding. Our proof-of-concept experiments provide a platform for rapidly screening specific amino acid reagents and reaction conditions for particular protein systems.

Protein footprinting mass spectrometry (MS) is an analytical tool used for higher order structure (HOS) determination.^{1, 2} Footprinting reagents impart covalent modifications, thereby increasing protein mass, as a function of solvent accessible surface area (SASA). Amino acid specific reagents used for footprinting are generally irreversible and modify side chains, allowing for specificity for certain amino acids (e.g., methylglyoxal for Arg^{3, 4} and glycine ethyl ester for acidic residues Asp/Glu⁵) or more general functional groups (e.g., benzoyl fluoride⁶ or diethyl pyrocarbonate⁷ for nucleophiles, such as the amine, thiol, hydroxyl, and carboxylate groups of amino acid side-chains). This technique is also commonly referred to as covalent labeling MS (CL-MS). While the study of principles governing amino acid specific footprinting reagents is ongoing, the choice of appropriate reagents and reaction conditions for the protein system poses a challenge and may require investment in method optimization. For example, significant effort may be required to determine reaction kinetics, to assess label-induced perturbations of higher order structure, and to establish reagent concentration to achieve desired sensitivity and dynamic range.^{8, 9}

Hydrogen-deuterium exchange MS (HDX-MS) is a category of footprinting wherein backbone amide hydrogens exchange with deuterium of deuterated water. In contrast to amino acid-specific footprinting, the exchange kinetics¹⁰ and the choice of reaction considerations are well-established and generally accepted.¹¹ The greatest challenges for HDX-MS analyses are the occurrence of deuterium back-exchange during sample handling, digestion, and LC separation and the requirement for proteolysis at low pH which precludes the use of trypsin or other sequence-specific proteases. Deuterium uptake is sensitive to changes in pH, ionic strength, temperature, among others¹²; furthermore, reliable HDX data requires high precision and reproducibility, both of which are challenged by human error while performing many acquisitions.^{10, 11, 13} To solve these problems, Trajan’s LEAP HDX automated workflow has been developed to reproducibly prepare many (currently, up to 100) HDX samples.¹⁴ The general workflow and sample handling involves dispensing the protein, diluting with D₂O, incubating for a specified time, quenching the HDX reaction, and denaturing the protein with chaotropes prior to online digestion, LC separation and MS analysis (Figure 1a).

Figure 1.

Comparison of workflows for HDX, 1-step footprinting, and 2-step footprinting methods (left to right, respectively). All three consist of a protein dispense step (green), reagent addition (blue), wait for incubation (colorless, grey dotted line), quench step (orange), and LC-MS analysis (purple).

Motivated by the significant time needed to characterize specific amino acid footprinting reagents, here we describe repurposing the HDX-MS automated workflow to screen reaction conditions and reagent choice to footprint two-state protein systems. Myoglobin (Mb) and calmodulin (CaM) are common model proteins for which changes in protein structure upon ligand binding are established.^15-18 Histidine (His) residues in Mb are involved in direct binding with the heme (Figure 2a).¹⁹ In contrast, acidic residues (aspartic acid (Asp) and glutamic acid (Glu)) directly coordinate calcium ions (Figure 2b).²⁰ To monitor changes in side chain solvent accessibility upon ligand binding residues of Mb (His) and CaM (Asp/Glu), we selected the established reagents DEPC and glycine ethyl ester (GEE) for Mb and CaM, respectively. We adjusted the automated methods to accommodate the special requirements for these footprinting reagents (e.g., DEPC is minimally soluble in water, requiring dilution in acetonitrile prior to labeling). While DEPC is a one-step footprinting method that allows for minimal modification of the HDX method (Figure 1b, Figure S1a), GEE requires the addition of an initiating reagent 1-ethyl-3-(3-dimethylaminopropyl)carbodiimide (EDC) (i.e., a 2-step reaction, Figure 1c, Figure S1b).⁵ Using these automated footprinting methods and intact MS analysis, we screened the concentrations of each reagent to determine appropriate conditions for specific amino acid modification and to confirm differential modification for the ligand-free and ligand-bound states of Mb and CaM.

Figure 2.

(a) Myoglobin structure (PDB 1WLA) with DEPC-reactive side-chains (Lys, His, Tyr, Ser, Thr) shown as sticks. The heme is shown in green, and the heme-binding His residues are in blue. (b) Calmodulin structure (PDB 1CLL) with GEE-reactive side chains (Asp, Glu) shown as sticks. The Ca²⁺ ions are shown in green, Ca²⁺-binding Asp and Glu residues are shown in red.

EXPERIMENTAL METHODS

Materials.

All chemicals, solvents, and proteins were purchased from Millipore (St. Louis, MO), unless otherwise indicated, and used without further purification. Bovine brain calmodulin was purchased from Biovision Inc (Waltham, MA). Ethylene glycol-bis(β-aminoethyl ether)-N,N,N′,N′-tetraacetic acid (EGTA) was purchased from Fluka (Charlotte, NC). 1-Ethyl-3-(3-dimethylaminopropyl)carbodiimide (EDC) was purchased from Thermo Scientific (Waltham, MA).

Sample preparation.

Protein apo- and holo-myoglobin samples were reconstituted in 1 x phosphate buffered saline solution (PBS, pH 7.4). Bovine brain calmodulin was reconstituted in 1 x PBS and incubated with 1 mM EGTA and 1 mM CaCl, respectively, overnight at 25 °C as previously described.^{5, 21} DEPC was diluted to 80 mM with acetonitrile; a series of DEPC concentrations were prepared using 50% serial dilutions with acetonitrile. 100 mM imidazole quench was prepared in water. EDC and GEE were dissolved in water and diluted to 10 mM and 200 mM, respectively; a series of EDC and GEE solutions were prepared using 90% serial dilutions for concentration screening or 50% dilutions of EDC for the apo-/holo-calmodulin measurements. A 50 mM ammonium acetate quench was prepared in water.

All other covalent labelling steps were performed on Trajan’s LEAP HDX system (Trajan Scientific and Medical, Ringwood, Victoria, Australia).

DEPC labeling method

A solution of 150 μL of 100 mM imidazole was predispensed into quench vials. Volumes of 5 μL of differing concentrations of DEPC were dispensed into separate reaction vials. To the reaction vial with DEPC, 95 μL of 0.5 μM protein sample was added to start the labelling reaction, which was allowed to proceed for 3 min at 22 °C, with syringe mixing. After the incubation, the reaction mixture was transferred to the quench vial to stop the reaction. The total mixture was then injected into the LC-MS system.

GEE labeling method

A solution of 150 μL of 50 mM ammonium acetate was predispensed into quench vials. Volumes (5 μL) of differing concentrations of GEE were dispensed into separate reaction vials. To the reaction vial with GEE, 5 μL of 10 μM protein sample was added, followed by 25 μL of differing concentrations of EDC to start the covalent labelling reaction. The reaction was allowed to proceed for 5 min with syringe mixing at 22 °C. After 4 min, the reaction solution was diluted to 100 μL with 1 x PBS. After incubation, the reaction mixture was transferred to the quench vial to stop the reaction, and the total mixture was injected into the LC-MS system.

LC-MS analysis

The protein samples were trapped on a ZORBAX 300SB-C3 column (2.1 x 12.5 mm, 5 μm, Agilent, Santa Clara, CA) trap, desalted for 3 min with 200 μL/min water/0.1% formic acid, then eluted using a 3.5 min gradient of 100 μL/min 4%-80% acetonitrile/0.1% formic acid for direct infusion into a MaXis II ETD Q-TOF MS (Bruker Daltonics, Billerica, MA) for MS analysis. All samples were prepared in at least triplicate and analyzed at 0 °C.

Data processing

All mass spectra were deconvoluted using Byos (v. 4.3, Protein Metrics Inc., Cupertino, CA). Fraction modified was calculated as in equation (1)

$$1-\frac{A_0}{\Sigma A_n}$$ (1)

where A₀ and A_n are the abundances of the unmodified protein and abundances of protein signals with n number of modifications (n ≥ 0). The weighted number of modifications was calculated as in equation (2).

$$\frac{\Sigma (A_n\cdot n)}{\Sigma A_n}$$ (2)

RESULTS AND DISCUSSION

Automated DEPC labeling is sensitive to changes in Mb ligand-binding.

The workflow for DEPC modification is similar to that of HDX. The reaction is initiated by adding a molar excess of DEPC (dissolved in acetonitrile) to a protein sample solution and quenched by adding excess imidazole⁷ (Figure 3A). That similarity expedites the development of an automated footprinting workflow. The most significant difference between the two methods is the dilution of DEPC in acetonitrile. To keep the concentration of acetonitrile low in the reaction (~5% V/V), the automated DEPC method reverses the order and volume of additions. We first add a low volume of concentrated DEPC reagent in acetonitrile to the vial and follow that with a 95% dilution with dilute protein sample in buffer (Figure 3A). We also modify the low volume DEPC transfer steps to improve reproducibility: an air gap in the syringe is added before drawing up the solution to prevent mixing residual solution in the syringe with the intended concentration of DEPC; four mixing cycles with the syringe are added during the reaction incubation wait time to prevent/minimize the impact of diffusion-limited reactivity; and the syringe is washed with acetonitrile both before and after DEPC solution transfer.

Figure 3.

(a) Detailed workflow for automated DEPC labelling showing protein dispense step (green), reagent addition (blue), wait for incubation (colorless, grey dotted line), quench (orange), and LC-MS analysis (purple). (b-d) Automated DEPC footprinting of apo- and holo-myoglobin (black and gray, respectively). (e-g) Automated DEPC footprinting of apo- and holo-calmodulin (black and gray, respectively). Note that the deconvoluted mass spectra and modification extents for apo- and holo-calmodulin overlap and are thus not distinguishable. (b, e) Representative deconvoluted mass spectra for intact protein using 0, 10, and 40 molar equivalents of DEPC-to-protein. DEPC modifications indicated in blue. (c, f) Fraction DEPC-modified myoglobin and calmodulin, respectively. (d, g) Abundance weighted number of DEPC modifications for myoglobin and calmodulin, respectively.

DEPC reacts with nucleophilic groups of amino acid side-chains (cysteine (Cys), histidine (His), lysine (Lys), tyrosine (Tyr), serine (Ser), and threonine (Thr)) to produce carbethoxylated residues (i.e., a mass shift of +72 Da) (Figure S1a).⁷ Incubation of apo-myoglobin with zero to 40 molar equivalents of DEPC produced up to five additions of +72 Da, consistent with DEPC modification (Figure 3b). We report the dependence of DEPC modification of apo- and holo-Mb with respect to reagent concentration as the fraction of protein modified by DEPC (Figure 3c) and the abundance-weighted number of DEPC modifications (Figure 3d). With increasing DEPC concentration, the fraction of apo-myoglobin modified by DEPC approaches one, signaling the disappearance of the unmodified protein. The weighted number of DEPC modifications on apo-myoglobin increases to approximately three with additional DEPC, illustrating the increasing preference for more than one modification. Overall, the error associated with each datapoint is low, as expected for the improved precision from automated sample preparation.

DEPC modification of holo-Mb is measurably decreased relative to that of the apo state (Figure 3b-d), consistent with heme coordination lowering the reactivity of directly coordinating His residues, as well as other residues within and proximal to the binding site (Figure 2a).^{19, 22, 23} In contrast, CaM Ca²⁺ binding sidechains (Asp/Glu) (Figure 2b) are DEPC reactive, but the modification readily reversible, and only low levels of CaM modification are observed limiting the dynamic range for differential measurements (Figure 3e)²⁰. No differences are observed in DEPC modification between the two states (Figure 3f, 3g), suggesting that DEPC is not sensitive to Ca^2* binding to CaM and the consequential conformational changes, at least at the protein level. Indeed, two-state Mb and CaM serve as appropriate positive and negative controls, respectively, for demonstrating the efficacy of our repurposed automated footprinting platform.

Matrix of EDC and GEE concentrations identifies appropriate labeling conditions for CaM.

Despite that GEE labeling is a two-step reaction, implementation of GEE labeling using the automated DEPC labeling method requires only minor modifications. To initiate the GEE reaction, the protein is incubated with an excess of EDC and a larger excess of GEE before quenching with excess ammonium acetate.⁵ GEE and EDC are water soluble, reducing concerns about dilution with organic solvents. Starting from the DEPC method described above, we add low volumes of GEE and protein solutions to the reaction vial. Following that mixing, we initiate the reaction by dilution with a larger volume of EDC, in contrast to the DEPC method initiation using dilution with the protein solution. In the last minute of the incubation, we dilute the reaction mixture with PBS buffer and transfer the reaction mixture to the predispensed ammonium acetate quench (Figure 4a).

Figure 4.

(a) Detailed workflow for automated GEE footprinting, showing protein dispense step (green), reagent addition (blue), wait for incubation (colorless, grey dotted line), quench steps (orange), and LC-MS analysis (purple). (b-e) Screening the effect of EDC and GEE molar equivalents on modification extent. The fraction GEE-modified (b, d) and abundance weighted number of GEE modifications (c, e) of apo-calmodulin. (f) Representative deconvoluted spectra for intact calmodulin using 0, 1250, and 5000 molar equivalents of EDC-to-protein and 5000 equivalents of GEE-to-protein. GEE modifications indicated in blue. For full identification of GEE and EDC modifications as a function of concentration, see Figure S2. (g) Fraction of the protein population GEE-modified for calmodulin. (h) Abundance weighted number of DEPC modifications for calmodulin.

Both GEE and EDC are reagents that can produce covalent modifications of proteins. EDC is widely used as a zero-length Glu/Asp-to-Lys crosslinking reagent, inducing an overall mass difference of −18 Da (Figure S1b).^{1, 5} For GEE footprinting, modification of side-chain carboxyl groups with EDC produces an O-acylisourea (i.e., a “dead-end” cross link,+154.5 Da). ²⁴ GEE then reacts with the O-acylisourea-modified residue to produce an amide with the NH₂ group of glycine (+85 Da) and release of urea. The ester portion of this amide/ester product is also prone to hydrolysis (decreasing the mass shift to +57 Da).⁵ The need to add a second reagent introduces an additional variable affecting modification extent—the concentration of the second reagent. We screened a matrix of both EDC concentrations (5000 to 50 molar equivalents) and GEE concentrations (20000 to 200 molar equivalents) to optimize the GEE modification of apo-calmodulin (Figure S2, Figure 4b-e). Incubation of CaM with 20000 molar equivalents of GEE and no EDC produces no modifications, but incubation of CaM with no GEE and 5000 molar equivalents yields a mixture of one-to-two crosslinks and one-to-two EDC dead-end crosslinks (Figure S2). For the mixtures of GEE and EDC, we observe a mixture of modified products consistent with crosslinking, EDC dead-end crosslinks, and GEE amide modification, but no hydrolyzed GEE products. We hypothesize that the decreased sample handling time for the automated method may reduce GEE ester hydrolysis relative to manual sample preparation and bottom-up analysis, where we observe this phenomenon more regularly. The mixture of modifications complicates interpretation. For example, CaM with one EDC dead-end crosslink (+154.5 Da) is within 2.5 Da of CaM with two GEE modifications and one crosslink (+152 Da); nevertheless, GEE modification (+85 Da) shows no conflicting mass differences. Therefore, the reported GEE modification extents include only GEE and not crosslinking or dead-end EDC. Using the fraction of GEE modified products and weighted number of GEE modifications as a function of reagent concentrations (Figure 4b-e), we observe that the limiting reagent for GEE modification is EDC. Varying GEE concentrations at constant EDC concentration produced little difference in GEE modification extent, whereas we observe a clear relationship between the concentration of EDC and GEE modification extent. For differential measurements, we chose to use a constant concentration of GEE and vary the concentration of EDC.

Automated GEE labeling is sensitive to changes in CaM ligand-binding.

With a constant concentration of GEE, the addition of increasing concentrations of EDC produced increasing numbers of GEE modifications on CaM. Incubation of apo-CaM with 5000 molar equivalents GEE and zero to 5000 molar equivalents of EDC produced up to six additions of GEE (Figure 4f). We report the dependence of GEE modification of apo- and holo-CaM with respect to EDC concentration as the fraction of protein modified by GEE (Figure 4g) and the abundance weighted number of GEE modifications (Figure 4h). With increasing EDC concentration, the fraction of apo-CaM modified by GEE approaches one, indicating the disappearance of the unmodified protein. The weighted number of GEE modifications on apo-myoglobin increases to approximately two with additional EDC, illustrating the increasing preference for more than one modification. As seen with our 1-step DEPC experiment, the error associated with each datapoint is low, indicating high precision and repeatability are preserved despite the additional step required by GEE labeling.

GEE modification of holo-CaM is measurably decreased relative to that of the apo state (Figure 4f-h). These differential EDC/GEE footprinting results for CaM are consistent with the lowered reactivity of the Asp and Glu carboxylates upon coordination with Ca²⁺ (Figure 2b).²⁰ In contrast, modification of Mb-heme binding sidechains (His) (Figure 2a) is not expected with GEE. Using the same methods and concentrations as used for CaM, we find only mass shifts consistent with crosslinking for holo Mb, but no mass shifts consistent with GEE or EDC dead-end crosslinks (Figure S3). The presence of crosslinks but no GEE modifications suggests that the acidic residues in Mb are either forming salt-bridges (observed in PDB 1WLA) with lysine residues or otherwise not solvent accessible.¹⁹ The observation of both crosslinking and footprinting has promising implications for new application of the reaction. The GEE reaction provides a measure of solvent accessibility, and the proximity-based crosslinks reflect intra- or intermolecular Asp/Glu-to-Lys interactions, in one experiment.

CONCLUSION

Automated labeling of model proteins with established specific amino acid labeling reagents proves to be efficacious, with promise to improve precision and streamline footprinting experimental design. Our design repurposes Trajan’s LEAP HDX automated workflow to label model protein-ligand binding systems. The outcomes demonstrate high precision, rapid screening of reagent conditions before time- and material-consuming bottom-up proteomics. This setup is also amenable to screen additional reaction conditions (e.g., temperature, pH, labeling time). With the application of ETD/ECD or the addition of sample digestion steps, the samples prepared using the automated platform may also be useful for top-down or bottom-up proteomics footprinting experiments, respectively. Furthermore, the sample remaining after optimization may be removed from the robot sample tray and submitted to offline bottom-up sample processing and analysis. A fully automated bottom-up footprinting system, however, poses several obstacles requiring further development. Finally, this platform may also be used for rapid assessment for novel reagent discovery and validation by using model proteins or cyclic peptides to monitor modification site specificity, reaction kinetics, and quench efficacy.