Mass Spectrometry-Based Protein Footprinting for High Order Structure Analysis: Fundamentals and Applications

Abstract

Proteins adopt different higher order structures (HOS) to enable their unique biological functions. Understanding the complexities of protein higher order structures and dynamics requires integrated approaches, where mass spectrometry (MS) is now positioned to play a key role. One of those approaches is protein footprinting. Although the initial demonstration of footprinting was for the HOS determination of protein/nucleic acid bonding, the concept was later adapted to MS-based protein HOS analysis, through which different covalent labeling approaches “mark” the solvent accessible surface area (SASA) of proteins to reflect protein HOS. Hydrogen deuterium exchange (HDX), where deuterium in D₂O replaces hydrogen of the backbone amides, is the most common example of footprinting. Its advantage is the footprint reflects SASA and hydrogen bonding, whereas one method drawback is the labeling is reversible. Another is slow irreversible labeling of functional groups on amino-acid side chains by targeted reagents with high specificity, probing structural changes at selected sites. A third footprinting approach is by reactions with fast, irreversible labeling species that are highly reactive and footprint broadly several amino-acid chains on the time scale of sub-milliseconds. All these covalent labeling approaches combine to constitute a problem-solving toolbox that takes mass spectrometry as the measurement tool for HOS elucidation. As there has been a growing need for MS-based protein footprinting in both academia and industry owing to its high throughput capability, prompt availability, and high spatial resolution, we present a summary of the history, descriptions, principles, mechanisms, and applications of these covalent labeling approaches. Moreover, their applications are highlighted according to the biological questions they can answer. This review is intended as a tutorial for MS-based protein HOS elucidation and as a reference for investigators seeking a MS-based tool to address questions in protein science.

Graphical Abstract

1. Introduction

Proteins carry out the programmed activities encoded by genes. Although constructed by the polymerization of only twenty distinct amino acids, their numerous biological functions require the high diversity arising from sequence variations and post-translational modifications. As living organisms evolve, proteins acquire specialized abilities that can be organized into different functional classes¹: enzymes catalyze different intracellular and extracellular reactions; structural proteins provide support and maintain the structural rigidity of the cell; transport proteins facilitate trans-membrane flow of certain materials; regulatory proteins work as sensors and switches to regulate gene expression and protein activities; motor proteins facilitate macroscopic movements; and signaling proteins transduce messages to facilitate the communication of different cellular components. Despite high functionality and diversity, a protein’s biological function and corresponding mechanisms of action are determined by their three-dimensional (3D) structure that is encoded in its primary sequence as demonstrated through the famous ribonuclease refolding experiment.² The exploration of protein structure-function relationship at the molecular level has developed into an independent subject named structural biology, for which characterization of protein high order structure (HOS) is the goal.

1.1. Protein Higher Order Structures

The understanding of protein HOS is a key topic in biology because the functional mechanisms of proteins are encoded in their 3D structures. Protein structure can be viewed as having four distinct orders (Figure 1).^{1, 4} Primary structure refers to a linear combination of amino acids into a polymeric chain, which was first proposed in 1902.⁵ Although originally heavily debated, its importance was settled after Sanger and coworkers^6-7 first sequenced insulin. The consensus is that protein primary sequence is determined by genetic information encoded in nucleic acids; that genetic information is translated into the order or sequence of the 20 common amino acids in a protein. The principal means of determining primary structure of proteins has become MS-based sequencing or proteomics analysis.

Figure 1.

Four orders of protein structure exemplified by human deoxyhemoglobin (PDB ID 2HHB³).

Upon forming linear polymeric chain, proteins fold to yield local structures including α-helices, β-sheets, β-turns, Ω-loop, etc. with the first two being most common. Such local ordering represents the secondary structure of proteins.

Tertiary structures of proteins are their overall 3D structures that result from the folding of the secondary structural components and other unstructured motifs. Although many proteins function individually, there are others that are not biologically active until they interact with other proteins or ligands, and these interactions give rise to quaternary structure. An overall 3D structure of a protein complex that contains multiple protein subunits comprises tertiary structure. The term “protein high order structure” often refers to the secondary, tertiary and quaternary structures of a protein. This is the subject of “structural proteomics”, and protein footprinting is a key component of the subject.

Besides the covalent peptide bonds that assemble amino acid building blocks into primary sequence, non-covalent interactions including hydrogen bonding,^8-9 charge-charge interactions (salt bridge),^10-11 hydrophobic interactions,^12-13 aromatic-aromatic interactions (π-π stacking),¹⁴ cation-π interactions,¹⁵ and Van der Waals forces¹⁶ stabilize protein HOS. There is also a covalent contributor that stabilizes protein HOS, namely disulfide bonds.^17-18 All these forces work together to overcome the conformational entropy of protein folding (decrease in entropy from random coil to folded protein) and to stabilize a protein in its folded state.^19-21

1.2. Biophysical Approaches for Characterizing Protein Higher Order Structure

In 1958, Sir John Kendrew²² first reported the high-resolution structure of sperm whale myoglobin by X-ray crystallography, for which he shared the 1962 Nobel Prize in Chemistry with Max Perutz²³ for their studies of the structures of globular proteins. Their contributions are a milestone in the field of structural biology, opening the field to pursuits of protein HOS.

After decades of development, X-ray crystallography and nuclear magnetic resonance (NMR) became the “gold standard” for determining protein HOS. The interaction of an ordered protein crystal with X-ray radiation causes the incident beam to diffract. By measuring their diffraction angles and intensities, it is possible to obtain a 3D electron density map of a crystal, from which the protein 3D structure is constructed at atomic resolution.²⁵ X-ray crystallography provides atomic-level resolution of various protein sizes, promoting it to be the most widely used spectroscopic approach in determining protein HOS to date. Its disadvantage is obvious as well; a crystalized protein sample is a must, and protein crystallization has always been challenging.^26-27 Moreover, X-ray crystallography provides a solid-state structure of proteins, posing concerns about whether proteins alter their structures upon crystallization as compared with their structures in solution or in vivo.^28-29

NMR resolves protein structure by determining chemical shifts and structural restrains.^30-32 The development of multi-dimensional NMR (pioneered by Richard Ernst, who was awarded the 1991 Nobel Prize in Chemistry for his contributions to high resolution NMR) allowed protein structures to be determined with near-atomic resolution by employing structural restraints obtained from homo-nuclear and hetero-nuclear couplings.^33-34 Solution NMR for determining protein 3D structures was pioneered by Kurt Wüthrich,^35-36 who was later recognized with the 2002 Nobel Prize in Chemistry. Solution NMR determines the liquid-state protein structures, which should resemble native states in the absence of interactions with proteins, ligands, and salts. Although mainly applicable to small proteins, recent work demonstrates that in special cases, the approach can accommodate protein molecular weights up to 1 MDa.³⁷

Another unique feature of solution NMR is its ability to measure protein dynamics in solution, which is not easily accessible at high resolution by other techniques.³⁸ For those proteins that are not readily soluble, membrane³⁹ and fibril⁴⁰ proteins for example, solid-state NMR contributes significantly.^41-42 The “dark side” of NMR-based protein structural elucidation is the sample amount, which is generally on the order of milligrams. The data acquisition and analysis are also time-consuming and labor intense. Some proteins cannot maintain their structure for long data-acquisitions and signal averaging.

X-ray crystallography and NMR have distinct advantages and drawbacks, but they complement each other.^43-44 As two of the most important pillars in protein HOS elucidation, the combination has contributed over 97 % of high resolution protein structures in Protein DataBank (PDB).⁴⁵

Another emerging technique in high-resolution protein HOS elucidation is cryogenic electron microscopy (Cryo-EM), whose name is self-explanatory (EM at cryogenic temperature).^46-47 Upon snap-freezing the sample at cryogenic temperature, water molecules in the sample remain amorphous, thus minimizing distortion of the protein structure.⁴⁸ Damage by the electron beam to the sample is minimized by the cryogenic temperature.⁴⁹ The ability to average multiple randomly-aligned EM pictures accelerated the development of Cyro-EM.⁵⁰ These factors combined to enable structural determination of proteins with high-resolution. An early electron crystallographic study by Henderson and coworkers⁵¹ demonstrates the capability of cryo-EM for determining membrane protein structure with near-atomic resolution (< 4 Å). Continued technical advances allow single-particle cryo-EM to characterize non-crystalline samples^52-53 and determine protein structures with atomic resolution (to 1.8 Å)^{47, 54}. Cryo-EM is also capable of characterizing protein complexes^55-56 and membrane proteins⁵⁷. As cryo-EM generally favors proteins with high molecular weights,⁵⁸ recent advances made possible the characterization of proteins with sizes < 100 kDa⁵⁴. Its requirements of low sample amount, straightforward sample preparation, and high structural resolution make it the “rising star” in structural proteomics.^46-47 With X-ray crystallography and NMR, these three methods are the basis for determining protein structure with the highest resolution.

In addition to the high-resolution approaches, there are others including mass spectrometry (MS) that can characterize the proteins with lower resolution. Circular dichroism (CD)^59-60, Ultra-violet (UV) resonance Raman⁶¹ and Fourier transform infrared spectroscopy (FT-IR)^62-63 can provide a general or overall characterization of protein secondary structure without providing atomic coordinates. UV at 280 nm is most commonly used in fast quantification of protein concentrations but contributes little to structure.⁶⁴ Fluorescence and Förster Resonance Energy Transfer (FRET) specialize in probing protein-protein interactions and protein conformational changes for regions of a protein.^65-67 There are other approaches that characterizes the protein from a global perspective, including dynamic light scattering^68-69 and its coupling with size exclusion chromatography⁷⁰ to focus on the hydrodynamic radius of proteins, negative staining electron microscopy^71-72 and atomic force microscopy, which examine the morphologies of proteins,⁷³ isothermal titration calorimetry, which discloses the thermodynamic properties of proteins,^74-75 surface plasmon resonance, which measures the kinetic properties of proteins upon interacting with others,^76-77 and many more. All these techniques combine to provide a biophysical toolbox for protein HOS elucidations; they are summarized schematically in Figure 2.

Figure 2.

A summary of commonly used biophysical tools for characterizing protein HOS. Protein structure exemplified by calcium-free bovine calmodulin (PDB ID 1CFD²⁴)

In addition to these biophysical approaches, computer modeling also plays a significant role in understanding protein HOS.^78-80 Early ideas of structural prediction were developed in the 1960s,^81-82 but it was not until 1974 that first structural prediction algorithm was actually developed.⁸³ These demonstrations were restricted to secondary structure predictions owing to limited computing power and mechanistic understanding of protein folding. Later on, with higher resolution, protein 3D structures could be resolved, template-based modeling (or homology modeling) became possible where available high-resolution structures serve as scaffolds when predicting HOS of unknown proteins.^{79, 84-85} Free modeling (ab initio), on the other hand, predicts protein HOS from scratch,^86-87 and feasibility was first demonstrated in 2005.⁸⁸ Thanks to the development of computational approaches and bioinformatics, protein HOS prediction has developed into an independent research subject and is contributing more and more to the structural biology.^89-90

1.3. Mass Spectrometry-Based Protein Structure Analysis

1.3.1. Historical Overview of Protein Mass Spectrometry

1.3.1.1. Ionization

To analyze a protein molecule in a mass spectrometer, ionization is the first step. The MS-based ionization methods were extensively developed in the 1960s to 1980s, key developments include chemical ionization,⁹¹ electrospray ionization (ESI),⁹² field desorption,⁹³ laser desorption,^94-95 Californium-252 plasma desorption,⁹⁶ secondary ion mass spectrometry,⁹⁷ fast atom bombardment,^98-99 matrix-assisted laser desorption ionization (MALDI)^100-101 and many more. Several of these techniques involved sample desorption; that is, MALDI and its precursor laser desorption, plasma desorption (PD),^102-103 and fast atom bombardment (FAB).¹⁰⁴ The latter was extensively used for over nearly two decades (from 1980 to 2000) for studies of peptides and small proteins. Many of these approaches are vital steps in the long evolution of ionization methods, but none work as well as ESI or MALDI. PD preferentially ionizes the molecules on the surface of support, and its efficiency is low and affected by the homogeneity of the surface layer. FAB is a relatively hard ionization method with low ionization efficiency and a relatively high “chemical-noise” background and a propensity to fragment proteins upon desorption, limiting its use to small proteins.

Although FAB and PD convincingly demonstrated that proteins are amenable to MS analysis, the ionization scene improved significantly with two separate but complementary breakthroughs. In 1988, Tanaka and coworkers¹⁰⁵ improved MALDI by incorporating 30 nm cobalt particles to ionize successfully lysozyme and chymotrypsinogen and to obtain their masses while Hillenkamp and Karas¹⁰⁶ were simultaneously developing the MALDI protocol that is used today, moving systematically from peptides to proteins. Meanwhile, Fenn^107-108 developed ESI and successfully ionized various proteins, showing that multiple charging brings the m/z of the ions into a mass range of mass spectrometers where analysis was possible.

These two developments, particularly ESI, allow MS to become the enabling tool in the new field of proteomics and subsequently in protein HOS analysis, which further grow into one of the most important biophysical tools in structural biology nowadays. Their efforts led to awards of the 2002 Nobel Prize in Chemistry to Koichi Tanaka and John Fenn, shared with Kurt Wüthrich (in the field of protein NMR), for their contribution in “the development of methods for identification and structure analyses of biological macromolecules”. Among these two methods, ESI (and later nano-ESI^109-110) has become more extensively used in proteomics and HOS structure analysis owing to its capability to install multiple charges in the analyte, its ease in setup, and most importantly its compatibility with liquid chromatography (LC), uniting in a concatenated way separation and analysis.

1.3.1.2. Instrumentation

Besides development of ionization methods, development of MS instrumentation also greatly elevated the detection limit and resolving power, empowering MS to work with larger proteins and deliver protein structural information with high mass resolving power. The advances in mass analysis show an evolutionary thread that included quadrupole mass filters,¹¹¹ double-focusing sector mass spectrometers^112-113 that evolved into three¹¹⁴ and four¹¹⁵ (and even more¹¹⁶) sectors for tandem-MS (MS/MS). Time-of-flight (TOF) instruments profited developments of straight,¹¹⁷ reflectron,¹¹⁸ and orthogonal¹¹⁹ TOF mass analyzers. Complementing and possibly exceeding them are quadrupole ion traps,¹²⁰ Fourier-transform ion cyclotron resonance (FTICR) instruments,¹²¹ and more recently orbitrap mass analyzers^122-123. Modern MS instruments are almost exclusively hybrid instruments that contain two or more different mass analyzers (represented by quadrupole-TOF¹²⁴ and quadrupole-orbitraps¹²⁵), allowing rapid, accurate, and precise mass determination of both precursor and product ions, a long sought-after goal in MS.

1.3.1.3. Fragmentation

The use of MS in protein HOS analysis depends not only on the accurate measurement of mass to charge ratio (m/z) but also on fragmentation methods that cleave the peptide backbone to provide sequence information that can locate sites of modification introduced as a protein footprint.¹²⁶ MS-based ion fragmentation required the introduction of collision induced dissociation (CID) and then tandem instruments, establishing the area of MS/MS.^126-128

Owing to its rich history and high popularity at the time that peptides and proteins were first ionized, it is not surprising that CID was immediately developed as a method for to sequence peptides and to apply these ionization advances to unknowns.^129-131

Over the years, different fragmentation methods including surface induced dissociation (SID),^132-133 electron transfer dissociation (ETD),¹³⁴ electron capture dissociation (ECD),¹³⁵ electron detachment dissociation (EDD),¹³⁶ ultraviolet photodissociation (UVPD),^137-138 infrared multiphoton dissociation (IRMPD)¹³⁹ were developed for studies of peptides and proteins. Their unique characteristics, advantages and limitations were reviewed in detail elsewhere.^{126, 140}

1.3.2. Labeling-Based Protein Footprinting

MS contributes to peptide and protein determination at all four levels of structure.¹⁴¹ The primary structure of a protein can be determined through MS/MS based de novo sequencing or partial sequencing and database searching.¹⁴² Although protein sequencing was expedited mainly by sequencing its corresponding genome, there are still needs to sequence a protein when its corresponding genome is unknown or the sequence requires verification. MS/MS-based sequencing has become the nearly exclusive route to primary structure, replacing Edman degradation owing to its accuracy, certainty, and speed.¹⁴³ In addition to sequencing, the MS/MS approach also enables profiling of post translational modifications by following shifts in m/z that pinpoint the modification type.^144-148 Indeed, this capability is the basis for protein footprinting where finding the location of a chemical modification introduced purposely is the basis of determining protein footprints that are indicative of HOS.

Footprinting probes secondary and tertiary structure by mapping solvent-accessible surface area (SASA) through various covalent-labeling approaches including hydrogen/deuterium exchange (HDX). Quaternary structures are often characterized by differential footprinting that locates the contacting region by responding to changes in SASA for two different states (e.g., bound and unbound). Native spray MS, a gentle version of ESI that uses aqueous media, can maintain the non-covalent interactions in the gas phase to yield information on topology and stoichiometry but not on interface.^149-153 Ion mobility MS reports protein shape, a component of HOS, in the gas phase.^154-156 Chemical crosslinking, an in-solution approach, is now extensively used to provide some information on protein quaternary structures and to locate interfaces.^157-159 All these approaches are the analytical basis for MS-based structural proteomics and provide high order structure information, although not as detailed as that from NMR, X-ray crystallography, and Cryo-EM.

In this review, we have adopted a focus on the emerging role of MS-based protein HOS determination by different covalent labeling approaches, which is becoming one of the most informative MS-based structural proteomics tools. Covalent labeling with different reagents that react at solvent accessible surfaces of the protein leaves a “footprint”, serving as a valuable probe for SASA that further reflects several aspects of protein HOS; thus, we term this workflow “protein footprinting”, a term that was initially proposed and demonstrated with HOS analysis of nucleic acids in 1978¹⁶⁰ and later adopted to analyze proteins¹⁶¹. We recommend wide adoption of “protein footprinting” as a key term to facilitate online literature searching. Early protein footprinting approaches were mostly cleavage-based, where protein SASA is revealed by limited proteolysis occurring at sites of the highest SASA^162-163 or by chemical cleavage by reactions with reactive radical species¹⁶⁴. Because these approaches do not afford high spatial resolution, cleavage-based footprinting was soon replaced by labeling-based approaches.

Over the years two distinct categories of covalent labeling approaches have been developed and applied: reversible and irreversible. Reversible labeling is represented by HDX, which takes advantage of the hydrogen exchange between active hydrogens in the protein molecule and the hydrogen in the water solvent.¹⁶⁵ Upon dissolving protein in deuterium oxide (D₂O), solvent-accessible and weakly H-bonded hydrogen atoms in the protein will exchange with deuterium atoms in the solution, where mass differences between hydrogen (1.0078 Da) and deuterium (2.0136 Da) atoms serve as useful indicator for a mass spectrometer to report such exchanges. Although the exchanged H or D in backbone are covalently bonded in the protein, the exchange process is highly dynamic and reversible. Thus, for ex situ methods like MS, protein HDX needs to be performed under constrained conditions to prevent back exchange, which would quickly “erase” the footprint, and preserve the structural information from the labeling. Details about protein HDX will be covered in Section 2.

Another category is irreversible protein labeling. Labeling reagents generally react with solvent accessible amino acid side chains, leaving a chemical “mark” that can be identified in an upcoming analysis. Nature already does elegant protein labeling; that is, by post translational modifications (PTM).^{144, 166-167} Mainly enzymatic processes introduce certain functional groups to incorporate proteins to modify them, often by adding or subtracting charges, so that new biological functions can be enabled by changing the conformation and the binding opportunities of a protein.¹⁶⁸

Inspired by nature’s use of PTMs, investigators have developed different chemical reagents that irreversibly label solvent accessible surface areas of proteins. One sub-class of reagents are readily available today are targeted reagents that react with only certain amino acid residues with high specificity but low reaction rates.¹⁶⁹ An example is acetylation of lysine or esterification of aspartic acid. The second category includes fast labeling reagents, mostly radical species, which are highly reactive but have lower specificity.¹⁷⁰ The advantage of irreversible labeling is that sample handling and post labeling are simpler and can be slower than that of HDX, allowing to take full advantage of all the separation advances made in proteomics. The different labeling timescales allow irreversible labeling to address several biological questions that are not amenable to HDX.

1.3.3. Bottom-Up and Top-Down

With MS/MS, locating labels and determining the protein footprint that reflects HOS can be performed with either bottom-up or top-down approaches.^171-172 Bottom-up protein analysis requires proteolytic digestion of the proteins prior to MS. Most approaches involving MS utilize a bottom-up approach as it has higher throughput, better compatibility with LC systems, greater sensitivity, and requires less advanced fragmentation technologies than top-down approaches. The key challenge for structural proteomics is to obtain high sequence coverage of the targeted protein, which is not a requirement for proteomics strategies that target proteins can be identified by using partial sequences determined for peptides that are “proxies” for the protein. Furthermore, PTMs are sometimes not accurately located owing to biased digestion efficiencies, as protease digestion efficiency is lower when its cleavage site is post translationally modified, as it may be in a footprint.¹⁷³

Top-down MS has been extensively developed in the past 20 years thanks to advances in MS fragmentation methods.^{172, 174} As the intact protein is submitted directly to the spectrometer, extensive fragmentation of the intact protein molecule must be induced to yield high spatial resolution. The major advantages of top-down is the capability of recovering different protein isoforms that are combinations of post-translational modifications and sequence variants. High fragmentation coverage is needed for footprinting, but today this is achieved mainly for small proteins. The approach is especially attractive when enzymatic cleavage sites are limited.¹⁷⁵ All these advantages make top-down a promising choice in proteomics.¹⁷⁶ Its drawbacks are primarily low throughput, limited sensitivity, and a lack of residue-level sequence coverage.¹⁷⁷ Reduced propensity for fragmentation in large proteins is due to the many modes or degrees of freedom for delocalizing energy in a big molecule, making it challenging to investigate proteins with high molecular weights.¹⁷⁸

In the field of protein HOS elucidation by MS, there are fewer examples that couple protein footprinting with top-down MS,^179-181 than with bottom-up. The primary reason is that sequence coverage at residue-level is critical for high spatial resolution, for which top-down currently fails to deliver. Combined developments of instrumentation, fragmentation methods, and sample handling strategies may afford a brighter future in protein footprinting.

There has been a growing need for MS-based protein footprinting in both academia and industry owing to its high throughput, prompt availability, and high spatial resolution. In the coming sections, we will cover the history, principles, designs, and mechanisms of covalent labeling-based protein footprinting approaches. We will largely confine our attention to reports over the last ten years, emphasizing those that report advances in methodology, rather than routine adoptions, to highlight the biological questions that can be addressed. Our goal is to provide a review that will serve as both tutorial for MS-based protein HOS analysis and a reference for investigators seeking a MS-based tool to address their question in protein science.

2. “Unbiased” Protein Footprinting – Hydrogen Deuterium Exchange

2.1. An Historical Perspective

HDX is one of the most important and used covalent labeling protein footprinting approaches to date despite its reversible nature. Although HDX is usually distinguished from other irreversible footprinting approach by classifying the latter as covalent labeling or covalent footprinting, HDX labeling also involves covalent bonds. The distinction is not covalent labeling as the exchange but rather that the exchange of N-H for N-D in HDX is a reversible process. We argue that the distinction between HDX and other footprinting is “reversible vs. irreversible” not covalent vs. an implied non-covalent bonding.

Protein HDX was first demonstrated in 1954 by Kaj Linderstrøm-Lang,¹⁸² who was later recognized as “father of HDX”¹⁸³. In the very first HDX experiment, Hvidt and Linderstrøm-Lang¹⁸² incubated dried pork insulin into D₂O to allow exchange, followed by snap-freezing and lyophilization. The protein powder was then suspended in H₂O. The deuterium content was measured by the density of the newly obtained protein-H₂O solution in a gradient tube. Later, they studied the pH and temperature dependence of H/D exchange rates,^184-185 made a connection between H/D exchange rate and the dynamics of protein molecules,^186-187 and formulated equations that underpin the basis for protein HDX^{185, 187-188}. Efforts by Linderstrøm-Lang and coworkers^{183, 189} are part of the early history of protein HDX, but in a display of scientific insight, these investigators demonstrated an approach of tremendous potential, opening a new era of covalent labeling-based protein HOS elucidation.

To enhance spatial resolution, HDX was later coupled with other characterization approaches (e.g., UV spectroscopy,¹⁹⁰ IR spectroscopy,^191-192 neutron diffraction¹⁹³), among which NMR was the most successful in early applications. The NMR applications began with a series of studies in late 1950s when Saunders and Wishnia^194-196 first demonstrated that the exchange of proteins and D₂O solvent can be measured by NMR with high spatial resolving power to give residue-level information at least for small proteins.¹⁹⁷ In 1976, these efforts were extended to the determination of individual amide proton exchange rates, as demonstrated by Wüthrich and coworkers.³⁰ Later, Wagner and Wüthrich¹⁹⁸ coupled two dimensional correlation spectroscopy (2D-COSY) NMR with H/D exchange rate measurements whereby exchange rates for every amino protons in a protein can be obtained in a single experiment. Recent development of fast 2D ¹H-¹⁵N correlation NMR demonstrates measurement of fast HDX for amide bonds within a few seconds of acquisition time.¹⁹⁹ Along with these developments, NMR measurements also contributed significantly to the mechanistic understanding of protein HDX phenomena.^200-201

The use of HDX-NMR in studying protein-protein interactions was first demonstrated by Roder and coworkers²⁰² in 1990. Although powerful and promising, HDX-based protein-protein interaction studies are not often made in NMR. Early protein NMR approaches suffered from limited spatial resolution, making it even more challenging to characterize large proteins by NMR.

As a complement to NMR, MS began to “pick up the baton” to extend HDX further to larger proteins at lower and lower concentrations. The demonstration of HDX followed by ESI MS was first by Katta, Chait, and Carr²⁰³ in 1991 who showed that the mass spectrum of a small protein (ubiquitin) changed significantly when it was sprayed in deuterated solvent. As the mass of the deuteron (2.014) differs measurably from that of hydrogen (1.008), MS monitors HDX by tracking the centroid mass of the isotopic envelop of a specific protein or peptide. As exchange-in continues, the centroid shifts to higher m/z and reports the average number of Hs that have exchanged with Ds.

The first example of off-line HDX and proteolytic digestion to reveal the regions of protein that exchanges was by Zhang and Smith.²⁰⁴ They termed their approach as “the protein fragmentation method”, and they used fast atom bombardment (FAB) as their ionization method. This approach is the precedent for what is done today (except FAB is replaced by ESI).

2.2. Mechanism of Exchange

To provide a basic understanding of mechanism of HDX of the amide backbone, we will briefly discuss some fundamentals.^{165, 200, 205} Among all hydrogens in a protein molecule, those that are part of O-H, S-H and N-H bonds exchange most rapidly with solvent water (D₂O) molecules. In a commonly executed “exchange-in” scenario, the protein is first solubilized in H₂O. Upon diluting with D₂O, labile Hs on the protein exchange with the surrounding solvent Ds, leading to an increase in mass that can be measured by a mass spectrometer. There is also an “exchange-out” mode where the protein is first incubated in D₂O, followed by addition of H₂O to cause a D-to-H exchange. Exchange-out is not as often used as “exchange-in,” and it will not be discussed further. Although HDX happens for both protein backbones and amino acid sidechains containing exchangeable Hs (e.g., NH₂, OH, SH), most experiments focus on backbone hydrogens. The hydrogens on the sidechains are weakly involved in hydrogen bonding, making their exchange fast and not easily followed with most experimental setups. These exchanges do not confuse the experimental outcome because, as fast exchangers, they return to an -XH state during workup and proteolysis that uses H₂O, as expected in an ex situ measurement.

Because the HDX of protein side chains is fast and not readily measurable, the total exchange is nearly only that of the amide N-H in peptide bonds, making HDX an “unbiased” labeling method because every amino acid residue (except Pro) has an N-H hydrogen that can be exchanged. The rate of exchange is fastest for amides that are solvent exposed and not involved in hydrogen bonding, for example, unstructured regions in a protein. HDX rates for these N-H hydrogens are represented by k_ch, which is a function of pD, the temperature of the solvent, and the nature of the side chains on neighboring amino acid residues. Previous studies reveal that HDX is both acid and base-catalyzed, and that k_ch minimizes at a pH around 2.5,^206-207 the pH that is chosen for quenching the reaction in a step that is essential for any ex situ measurement.

HDX for HOS determination is successful because the rate of HDX for backbone N-H hydrogen atoms is determined by local conformation. Protein local orders are stabilized by intramolecular N-H⋯O=C hydrogen bonding.²⁰⁸ The breathing motion of these hydrogen bonds will alter the conformational states of the targeted hydrogens that comprise the N-H bonds from closed to open states with a rate constant of k_op. As the protein is highly dynamic, the open N-H bonds can also close with a rate constant of k_cl. A local view of HDX process for a backbone N-H, thus, can be represented by Eq. 1.

$${\text{N-H}}_{\text{closed}}\underset{k_{cl}}{\leftarrow }\stackrel{k_{op}}{\to }{\text{N-H}}_{\text{open}}\stackrel{D_{2}O,k_{ch}}{\to }{\text{N-D}}_{\text{open}}\underset{k_{op}}{\leftarrow }\stackrel{k_{cl}}{\to }{\text{N-D}}_{\text{closed}}$$ (1)

In principle, every N-H possesses a unique combination of k_op, k_ch and k_cl. The observed rate constant for HDX, k_HDX, is determined mainly by two factors: the solvent accessibility and the local hydrogen bonding in the region of the amide bond. For regions that are highly flexible, the effect of these two factors is minimized so that k_HDX = k_ch. For regions that are structured, the HDX occurs under two limiting regimes, EX1 and EX2.^{205, 209} EX1 is characterized by k_ch >> k_cl, for which k_HDX = k_op. In EX2, k_cl >> k_ch, so that k_HDX = K_opk_ch, in which K_op = k_op/k_cl. Practically, EX2 is more prevalent than EX1.^{187, 201} There are also intermediate exchange regimes that are combinations of EX1 and EX2.²⁰⁹

2.3. Experimental Approach

Most HDX measurements compare two states of a protein (e.g., bound vs. unbound, wildtype vs. mutant); thus, this approach in footprinting is termed “differential”. In a differential experiment, HDX for a ligand-unbound protein is compared to the ligand-bound state, and the effects of local hydrogen bonding and back exchange cancel, leaving the differences in HDX to represent changes in solvent accessibilities and H-bonding for the two different states as the only factors determining the relative differences in HDX.²¹⁰

Experimentally, HDX is most commonly executed in a “bottom-up” fashion,^{207, 211} which originated from early experiments of tritium exchange.^212-213 Bottom-up HDX was first introduced by Zhang and Smith²⁰⁴ in 1993. In the peptide mapping mode, the protein or peptide (in H₂O) is denatured and, following quenching at pH ~ 2.5 (in H₂O), is digested and made ready for LC-MS/MS analysis for peptide identification and m/z determination to establish the average number of D taken up in the exchange (Figure 3). One goal is to optimize the quenching condition to permit high sequence coverage to make available a peptide list comprised of peptide sequences, precursor masses, and their elution times under optimized quenching conditions. The quenching solution is usually composed of denaturant and an agent for disulfide bond reduction (if there are any) and is at pH = 2.5, where the rates of back-exchange (back exchange of deuterium in the protein to hydrogen) are minimized. The quenching conditions, once optimized, are then used in further HDX experiments.

Figure 3.

Schematic illustration of a bottom-up peptide mapping workflow, a necessary step prior to HDX. Green dots in the protein structure indicate hydrogen atoms in the peptide bonds.

In the actual HDX experiment, the protein of interest is initially dissolved in H₂O, and the solution diluted with D₂O and incubated for several different times to allow measurement of the HDX kinetics (Figure 4). Upon exchanging, solvent-accessible amide hydrogens become replaced by D. The exchange is then quenched by decreasing the solution pH to 2.5. Quenching is followed by fast protease digestion and then MS analysis. The digestion is usually carried out online by flowing the protein solution through a pepsin column. The solvents used in HPLC are at acidic pH. Good experimental design shortens the time prior to when the deuterated peptides are analyzed by MS to minimize back-exchange. The obtained data are analyzed by fitting the isotopic distribution for each peptide, from which a mass centroid is calculated, and the deuterium uptake is determined by the difference in peak centroids between the exchanged protein (or the peptides) and the unexchanged or control.

Figure 4.

Schematic illustration of a bottom-up HDX workflow. Green and red dots in the protein structure indicate hydrogen and deuterium atoms in the peptide bonds, respectively.

Measurement of HDX at the residue level with MS is challenging. Low-energy collisional activation found on orbitraps, ion traps, or Q-Tofs cannot be used because that activation induces deuterium scrambling,^214-215 erasing all the forward exchange that was scrupulous introduced. Novel fragmentation methods that minimize scrambling include ECD^{135, 216} and ETD,²¹⁷ making possible, in some cases, measurement of residue-level deuterium uptake or of HDX in a “top-down” fashion^218-219.

2.4. Recent Advances and Applications

In this section, we will review several technical advances that increased the capabilities of HDX. One important goal is to increase its spatial resolution, ultimately to a single amino acid. Several approaches were developed to achieve a single-residue resolution, including vigorous proteolytic fragmentations to give overlapping peptides,^220-221 high-pressure on-line digestion to improve digestion efficiency,^222-223 incorporation of ETD and bottom-up HDX for enhanced spatial resolution,²²⁴ and computation methods that facilitate processing the data from aforementioned methods^{220, 225-226}. Novel digestion protocols give shorter peptides and more overlapping peptides. Single-residue HDX information can be obtained by considering the absolute deuterium uptake levels of two overlapping peptides that differ by a single amino acid residue.^220-223 This spatial resolution should also be achievable through fragmentation to give smaller peptides formed by ETD or ECD fragmentation that minimizes deuterium scrambling. Quantification of the resulting fragments will give data that resemble the deuterium uptakes for those fragments in solution.²²⁴ Novel computation methods either facilitate the automatic processing of the overlapping peptides²²⁵ or utilize novel algorithms to deconvolute the residue-level HDX behavior through a peptide-level HDX curve (by either close examination of isotopic envelope shape information²²⁰ or a Bayesian approach for deconvolution²²⁶).

To shorten further the exchange time, investigators developed theta capillary spray²²⁷ and gas-phase HDX^228-230 that allow exchange times as short as 20 μs to be followed. With further optimization, it should be possible to track deuterium uptake on the amino acid residue sidechains, which was not possible in conventional HDX (even with rapid mixing or stopped-flow) because the back-exchange of the active protons is rapid. It also allows HDX to capture fast dynamic processes that was not possible before.

To expand the capabilities of HDX to binding affinities, investigators developed titration-based HDX workflows including stability of unpurified proteins from rates of H/D exchange (SUPREX)²³¹ and protein–ligand interactions by mass spectrometry, titration, and H/D exchange (PLIMSTEX).²³² These approaches characterize ligand binding sites, binding orders (if there are multiple ligands that bind the protein of interest), and, most importantly, site-specific binding affinities. Both of these workflows were demonstrated at the peptide-levels,²³³ showing that it is possible to carry out single-residue level with the combination of the efforts mentioned above (ETD or multiple proteolytic digestions).

To expand the ability of HDX to characterize membrane proteins, effective removal of detergent²³⁴ and lipid,²³⁵ optimization of HDX conditions for enhanced transmembrane domain sequence coverage,²³⁶ and post-HDX deglycosylation²³⁷ were developed. Post exchange detergent and lipid removal are based on either chromatographic separation and organic solvent extraction²³⁴ or on the interaction between zirconium(IV) oxide and phospholipids²³⁵. These advances make possible the characterization of HDX for membrane proteins without incurring extensive back-exchange. Post-HDX deglycosylation removes highly heterogenous glycans without compromising the deuterium labeling on the peptides, allowing an investigator to obtain a well-resolved, high coverage HDX analysis.²³⁷ All these advances broaden the horizon of HDX.

Other than the experimental advances, we witnessed a burst of HDX data processing software in the past decade, some of the examples include HDExaminer,²³⁸ DynamX (formerly HX-Express),²³⁹ HDX Workbench,²⁴⁰ AUTOHD,²⁴¹ Mass Analyzer, ^242-243 HD Desktop,²⁴⁴ HeXicon,²⁴⁵ ExMS,²⁴⁶ HDX-Analyzer,²⁴⁷ HDX Finder,²⁴⁸ and Mass Spec Studio²⁴⁹. There are also new HDX data visualization software include MSTools,²⁵⁰ MEMHDX,²⁵¹ Deuteros,²⁵² and HDX-Viewer²⁵³. The development of these software tools contributes greatly to the broad application of HDX.

Undoubtedly, HDS-MS is now become relatively routine and is currently used in many applications in biochemistry and biophysics. The large number of these applications makes it challenging to cover them in this review. Instead, we will stick with our theme of protein footprinting and emphasize HDX as a footprinting method to provide HOS information. We will not review applications of HDX-MS because there are many specialized and comprehensive^{165, 254-255} reviews that cover most applications including mapping epitopes and characterizing biotherapeutics,^256-259 monitoring protein folding dynamics,^260-261 locating protein binding sites,^{210, 262-263} examining conformations of individual proteins or of large complexes,^264-266 probing allosteric effects,²⁶⁷ monitoring protein-membrane interactions,²⁶⁸ and developing methods to minimize deuterium scrambling during MS fragmentations²⁶⁹. A collaborative effort from 2018 summarizes the key aspects of performing HDX experiments, providing valuable recommendations on how to carry out an effective HDX study properly.²¹¹ Lastly, for more details on HDX theory, the reader should consult early reviews by Englander et al.²⁰⁰ and Smith et al.²⁰⁷.

2.5. Conclusion

The union of HDX and MS launched MS into the field of protein HOS determination. Because HDX causes a minimal perturbation on protein structure, has a relatively simple setup and analysis, aided by a number of data processing packages, and requires no design or synthesis of reagents, the adoption by the biochemistry community has been rapid. HDX covalently “labels” all protein solvent-accessible backbone amide hydrogens (only amino acid without an exchangeable H is Pro). Unlike targeted- and the fast-labeling reagents whose labeling efficiencies are hugely variable and determined by residue-specific reactivities, HDX occurs across the whole protein, and its labeling efficiencies depend on local structure (solvent accessibility and hydrogen bonding). Thus, HDX footprints proteins in a relatively “unbiased” way. The advantages of HDX, as envisioned by early pioneers, have been realized, and new applications are being implemented, making HDX an important technology for HOS nearly three decades after its introduction.

3. Targeted-Labeling Reagents

The 20 amino acids have different kinds of functional groups on their side chains; most have functional groups (COOH, SH, NH₂, OH, CONH₂, aromatic ring) that can be labeled with chemical reagents that react specifically, usually with one or two amino acids. These modifying reactions can be used for footprinting provided the reactivity of these groups depends on SASA of the protein and the modifications in the early stages do not affect the protein structure, minimizing the biased report during the footprinting itself. Reagents react directly and usually slowly with specific solvent-accessible side chains in contrast to free radicals, which react rapidly. The product contains a characteristic mass tag that can be detected by MS analysis. Although numerous reagents have been developed to react with the amino acid side chains, a qualified protein footprinting reagent needs to label the protein under physiological conditions with reasonable efficiency and speed. The size and hydrophilicity of the reagent should be close to that of water to ensure its reactivity is an indicator of SASA. Most of these reagents developed to date are highly specific, targeting one or two side chains or functional groups although some can react with more than two.

In this section, we will review commonly used targeted protein footprinting reagents, starting with their history in protein labeling and then discussing their chemistry and the products resulting from labeling. Many of the protein labeling reactions were discovered and implemented some time ago but not as footprinting reagents. That had to await the development of MS methods for sensitive, specific, and effective analysis to locate the modified sites. Following the early work, MALDI was chosen for analysis, but more recently the combination of ESI and LC-MS/MS has become the mainstay analytical tool. Our purpose in this section is to organize the reagents and discuss original development. We will not cover all the applications of these reagents.^{169, 270-274}

Because there are many reagents, this section is organized according to the target residues and the types of functional groups they contain. Cysteine is highly reactive owing to its highly nucleophilic thiol group. Tryptophan and tyrosine contain hydrophobic/aromatic side chains that are electron rich. Aspartic and glutamic acids are acidic residues for which many derivatization classic reactions are available. Arginine, histidine and lysine contain basic residues, and they are more difficult to footprint. Although some of the residues are preferentially charged under physiological conditions, their charge state depends on pH. All schemes in this section depict the charge neutral forms for simplicity.

Chemical crosslinkers, although bifunctional derivatives of targeted labeling reagents, are designed to afford information on protein/protein interfaces. They can be viewed as bifunctional footprinters, and, thus, we included them in a general discussion later in this section. Both footprinting and cross-linking have much in common and viewing them together provides an opportunity to compare and contrast.

A unique function for targeted protein labeling is covalent inhibition, where a bulky labeling reagent may inhibit a biochemical reaction. The covalent attachment of the labeling reagent may occupy a binding pocket and block a functional site, inhibiting measurably protein function. Although operationally similar to footprinting, this form of protein labeling will not be covered here because this field was reviewed.^275-277 The covalent attachment of functional probes was also developed extensively during the past decade, as reviewed in a perspective by Tamura and Hamachi.²⁷⁸

3.1. Cysteine

Cysteine (Cys) is one of the most reactive amino acids owing to the high intrinsic reactivity of the nucleophilic thiol side chain. Cys fulfills a variety of protein biological functions including forming disulfide bonds to stabilize structure, binding metal ions, transferring hydrogen and oxygen atoms, electrons, and hydrides, catalyzing redox processes, and mediating hydrolysis.²⁷⁹ Modifying Cys in a native protein, therefore, should be an informative footprinting goal as specific Cys groups throughout a protein are often important in biochemical processes.²⁸⁰

Cys labeling can be achieved through a large number of reagents, and many of them were developed as catalytic functional inhibitors (e.g., organomercurial compounds)²⁸¹ or affinity tags (e.g., functionalized biotin tag)²⁸². The use of these reagents is not the primary goal of this review and, therefore, the reagents will not be discussed further. Interested readers should refer to a recent book by Lundblad.²⁷¹

The thiol group in the Cys side chain becomes an even stronger nucleophile when it deprotonates to a sulfide anion. Therefore, most of the Cys labeling reactions take advantage of the nucleophilicity of -SH, or better, -S⁻.

The first class of Cys-labeling reagents, α-haloketo compounds, represented by iodoacetamide (IAM),^283-284 label Cys by S_N2 nucleophilic substitution (Scheme 1, top). Other popular derivatizing reagents include iodoacetic acid,²⁸⁵ iodoacetanilide²⁸⁶ and iodoacetate²⁸⁷. Other than labeling Cys, IAM and its derivatives also modify Met, His, Lys, Tyr and Glu but with lower reactivity.^288-291 The reactivity of IAM not only depends on the solvent accessibilities of targeted residues but also on the local environments, a property that relates to most targeted labeling reagents.²⁹² The chloro- and bromo α-haloketo compounds are less reactive towards thiol groups (the leaving-group propensities for the halides are I > Br > Cl^293-294) and, thus, chloro and bromo α-haloketo compounds are not commonly used in protein HOS analysis. These derivatizing agents, however, are more useful in affinity labeling as the reaction is facilitated because the reagent binds nearby, placing even an modestly reactive group in an appropriate position for reaction.²⁹⁵

Scheme 1.

Cys labeling pathways with different labeling reagents

Maleimide derivatives are more suitable footprinters by the introduction of bromomaleimides, which label Cys residues reversibly (Scheme 2).^296-298 In comparison to NEM, labeling by bromomaleimides involves nucleophilic substitution that preserves the double bond in the maleimide ring. Preservation of double bond allows a reaction with a second Cys residue (through Michael addition), which converts a bromomaleimide into a chemical cross-linker.^297-298 The reaction product, thiomaleimide, can be hydrolyzed to dehydroalanine or converted back to Cys by chemical reduction. Such versatility enables novel workflows in labeling or footprinting free Cys residues.

Scheme 2.

Cys labeling pathways with bromomaleimides

The other class of well-established Cys labeling reagents is N-alkylmaleimide that labels Cys through Michael addition (Scheme 1, bottom); a common example is N-ethylmaleimide (NEM).^299-301 Although NEM labels Cys residues with considerably higher reactivity and specificity than does IAM,²⁹² other nucleophiles including His and Lys can also be modified by NEM.^302-303 In the LC-MS analysis, however, the N-alkylmaleimide is prone to hydrolysis, forming an isomeric mixture of maleamic acid adducts.²⁹¹ The molar absorptivity of NEM is 620 M⁻¹cm⁻¹ at 302 nm,²⁹⁹ enabling UV-Vis spectroscopy for analysis in NEM kinetic studies. The alkyl chain length in N-alkylmaleimide should be considered when the analysis is by MALDI as that group can be responsible a non-uniform effect on signal intensity.³⁰⁴

IAM and NEM are widely used not only in MS-based footprinting that takes advantage of their reactivity with Cys but also for protecting free Cys residues, a standard practice in almost all MS-based proteomics experiments. In a pioneering study that used IAM as a Cys footprinting reagent, Wu and coworkers³⁰⁵ investigated the active sites of human type II inosine 5’-monophosphate dehydrogenase (IMPDH). 6-Chloropurine riboside 5'-monophosphate (6-Cl-IMP) is a structural analog of the IMPDH substrate and was used as an IMPDH inhibitor by modifying one of eight Cys residues in IMPDH. The results of IAM footprinting of IMPDH in its ligand-free and 6-Cl-IMP-bound states demonstrate that seven Cys residues show comparable IAM labeling, whereas Cys 331 was not labeled by IAM but by 6-Cl-IMP in the bound state, strongly suggesting that Cys 331 is critical in facilitating the enzymatic activity of IMPDH. IAM derivatives were also modified to serve as fluorescent probes for monitoring protein conformational changes in vivo with relatively good spatial resolution.^306-308 Chumsae et al.³⁰⁹ incorporated Cys fluorescence labeling with MS-based peptide footprinting to study the unpaired Cys residues in human immunoglobulin G subclass 1 antibody; details will be discussed in section 4.5.

NEM labeling coupled with MS-based protein footprinting was pioneered by Reich and coworkers³¹⁰, who applied NEM to footprint the critical Cys residue in DNA methyltransferase. In the ligand-free state, Cys 25, 116 and 223 were modified by NEM, but upon incubation with DNA and sinefungin (a structural analog of cofactor S-adenosylmethionine), Cys 223 became essentially unmodified, demonstrating its critical role in preserving the enzymatic function of DNA methyltransferase. Notably, all the site-specific Cys labeling products were characterized by tandem MS with a magnet sector instrument, a demanding task at the time and now more readily accomplished with modern Q-ToF and orbitrap instruments.

Thiol groups in Cys residues can also be nitrosylated by active nitrogen compounds such as nitric oxide and tetranitromethane.^311-312 Reagents for Cys include cyanogen bromide³¹³ selenium reagents,³¹⁴ N-(phenylseleno) phthalimide,³¹⁵ 5,5'-dithiobis-(2-nitrobenzoic acid) (Ellman’s reagent or DTNB) and its derivatives,³¹⁶ 2,2-dipyridyl disulfide and 4,4′-dipyridyl disulfide,³¹⁷ vinyl pyridine^318-319 and acrylamide.³²⁰ These reagents will not be discussed further as their utility in MS-based Cys footprinting remains to be established.

The oxidized form of cysteine, cystine, exists with a disulfide bond. Cystine footprinting, interestingly, originates from the analysis of wool and its oxidation.^321-322 To label disulfide bonds, cystine residues need to be reduced to cysteine. Many reducing reagents are available including cyanide,³²³ phosphorothioate,³²⁴ (2S)-2-amino-1,4-dimercaptobutane,³²⁵ bis-(2-mercaptoethyl)sulfone and N,N′-dimethyl-N,N′-bis(mercaptoacetyl)hydrazine,³²⁶ sodium borohydride,³²⁷ and 2,3-dimercaptopropanol³²⁸ that can deliver effective disulfide reduction. Three reagents, however, are commonly used to reduce disulfide bonds in proteins. Dithiothreitol (DTT), originally synthesized by Evans et al.³²⁹ in 1949, is able to reduce disulfide bonds as was first demonstrated by Cleland³³⁰ in 1964 (DTT is also known as Cleland’s reagent). DTT is prone to oxidation when using in air, with a half-life of 10 h in neutral solution (pH = 7.5, 20 °C), decreasing to 1.4 h when the pH = 8.5,³³¹ making it critical to prepare a fresh DTT reagent solution. The oxidized form of DTT has an absorbance maximum at 283 nm, which can be utilized to determine the extent of disulfide bond reduction.³³²

Unlike DTT that only functions under near-neutral pH, tris(2-carboxyethyl)phosphine (TCEP)³³³ reduces disulfide bonds rapidly under acidic conditions. TCEP is reasonably stable under acidic conditions, but its oxidation is significant at pHs greater than 7.³³⁴ All these features led to broad acceptance and application of TCEP in HDX workflows.

The last reagent is 2-mercaptoethanol (β-mercaptoethanol, BME), which is stable and functions under alkaline pH.³³⁵ With the development of DTT and TCEP, BME is less used in disulfide bond reduction, but it is also used in protein structural analysis.^{271, 336} This and the other reducing reagents are usually used under denaturing conditions for better disulfide reduction, which is not appropriate for footprinting.

In addition to conventional reducing reagents, a novel approach is to utilize electrochemistry to reduce disulfide bonds.³³⁷ This development may be particularly useful in HDX experiments, as it enables fast and effective on-line disulfide reduction.³³⁸ The reduction converts cystine residues to cysteine residues that are ready for labeling or “capping” to avoid disulfide bond re-formation. MS-based disulfide mapping (cystine labeling) will be discussed in section 4.5.

3.2. Tryptophan

Tryptophan (Trp) is an electron-rich amino acid with a heterocyclic aromatic side chain; Trp is the lowest abundance amino acid in proteins.³³⁹ As Trp is hydrophobic and rarely found on protein surfaces, its footprinting is not common. In the case of protein-protein and protein-membrane interactions, however, Trp contributes significantly,^340-343 motivating application of footprinting.

Trp and Tyr are responsible for most of the characteristic protein UV absorbance at 280 nm.⁶⁴ Trp is also responsible for most intrinsic protein fluorescence of proteins.^344-345 These unique features allow Trp residues to be photo-activated under UV irradiation.^346-351 In brief, Trp under irradiation by UV light results in a tryptophan radical and a hydrated electron.^{346-347, 349} The formation of tryptophan radical-mediated reactive oxidative species leads to hydrogen peroxide as a product,^{348, 352} which further oxidizes nearby residues³⁵¹. A recent book by Lundblad²⁷¹ provides a detailed overview of Trp photoactivation.

The indole nitrogen atom in the side chain of Trp has a pK_a greater than 15,²⁷¹ making it difficult to activate for chemical modification. In limited demonstrations, Previero and Cavadore³⁵³ developed formylation of Trp with hydrochloric acid and acetic acid, leading to N-formyl tryptophan as product. In another example, Linder and coworkers³⁵⁴ employed 1,1,3,3-tetramethoxypropane to react with Trp residues under acidic conditions, producing a substituted acrolein with a free aldehyde. This chemistry was further developed by bringing together another immobilized hydrazide that reacts with the free aldehyde to afford chemical cross-linking.³⁵⁵

A better Trp labeling strategy is to target the C-H bond adjacent to the indole nitrogen, which is done by the three reagents as shown in the Scheme 3. 2-Hydroxy-5-nitrobenzyl-bromide (HNB) was first demonstrated to be an effective Trp labeling reagent by Koshland and coworkers^356-358; HNB, therefore, is also known as Koshland’s reagent (Scheme 3, top). Over the intervening years, many HNB derivatives were developed to label Trp, including dimethyl(2-hydroxy-5-nitrobenzyl)sulfonium bromide (HNSB),³⁵⁹ 2-methoxy-5-nitrobenzyl bromide,³⁶⁰ and 2-acetoxy-5-nitrobenzyl chloride.³⁶¹ The applicability of HNB in protein HOS analysis was first demonstrated by Strohalm et al.^362-363, where the investigators followed effective HNB labeling of Trp residues for the model proteins lysozyme, cytochrome c, and myoglobin by using MALDI-MS. Only solvent-accessible Trp residues were labeled under physiological conditions whereas all Trp residues did react in a denaturing environment, suggesting HNB to be a valuable Trp footprinter. Another widely used derivative of HNB is HNSB.^364-365 Unlike HNB, which must be prepared in an organic solvent (usually dry acetone), HNSB is a water-soluble sulfonium salt, which is an important factor that grants relevance in footprinting under physiological-like conditions.

Scheme 3.

Trp labeling pathways with various reagents

The second Trp labeling reagent is N-bromosuccinimide (NBS),^366-367 which converts the indole ring to an oxindole (Scheme 3, middle). Among many successful demonstrations of NBS in footprinting Trp residues, the pioneering study by Takahashi et al.³⁶⁸ adopted NBS labeling to footprint Trp residues of an α-amylase inhibitor PHA-I of Phaseolus vulgaris. PHA-I adopts a tetrameric conformation of (αβ)₂, and two of eight Trp residues (Trp 188 on each β subunit) show significant oxidation by NBS. When complexed with porcine pancreatic α-amylase, however, none of the Trp residues in the PHA-I were modified, suggesting that Trp 188 in the β subunit is critical for its inhibitory activity. This is a fine example that not only illustrates the effectiveness of NBS in assessing protein HOS but also shows the importance of Trp residues in assisting protein-ligand interactions.

The derivative of NBS, N-chlorosuccinimide (NCS), was also reported as an Trp labeling reagent.³⁶⁹ In addition to Trp footprinting, all N-halosuccinimide reagents including NBS,³⁶⁶ NCS³⁷⁰ and N-iodosuccinimide³⁷⁰ are capable of cleaving peptide bonds that occur upon further oxidation of oxindole. NCS is the most adopted reagent for this purpose.³⁷¹ This peptide bond cleavage served as an early Trp footprinting approach but has not been developed further because the development of proteomics capability with mass spectrometers has enabled other analytical approaches.

The third reagent that labels Trp residues is 2-nitrophenylsulfenyl chloride (Scheme 3, bottom), which was first reported by Scoffone and Rocchi³⁷² in 1968. The reagent, 2,4-dinitrophenylsulfenyl chloride, was used together with 2-mercaptoethanol to attach covalently, on a Trp residue, a thiol group that can be further utilized for subsequent reactions.³⁷³ The application of 2-nitrophenylsulfenyl chloride for MS-based Trp footprinting was first demonstrated by Simpson and coworkers³⁷⁴ in 1993, where the investigators evaluated the effect of Trp derivatization on the biological activity and conformational stability of murine interleukin-6. Upon incubating with 2-nitrophenylsulfenyl chloride, Trp 36 and 160 showed significant labeling as identified by ESI-MS. The labeled interleukin-6 underwent a decrease in the biological activity in the murine-hybridoma growth-factor assay using 7TD1 cells and in binding affinity with its corresponding receptor. The overall conformation of labeled interleukin-6, however, remained unperturbed by CD measurements. They concluded that Trp 36 and 160 are critical in maintaining biological activity but not the structural integrity of interleukin-6. The derivative of 2-nitrophenylsulfenyl chloride, 2,4-dinitrophenylsulfenyl chloride has also been used as a TrP footprinting reagent as demonstrated by Zhang et al.³⁷⁵ in 1997.

In summary, Trp labeling by targeted labeling reagents is less developed and used compared with footprinting of other residues, which is, in part, because all these labeling reagents favor acidic environments.¹⁶⁹ Under physiological pH, these reagents are either less specific (HNB³⁵⁷ and NBS³⁶⁷) or less reactive (NBS^376-378). Other than the reagents introduced above, other Trp labeling reagents including acetyl chloride,³⁷⁹ hypothiocyanous acid,³⁸⁰ and chloroform (under photo-activation)³⁸¹. These reagents require harsher reaction conditions, and, thus, are less suitable as reagents for protein HOS analysis.

3.3. Tyrosine

Tyrosine (Tyr) is a phenol-containing hydrophobic amino acid that is often found on the protein surfaces.^382-383 Tyrosine is a critical target of several post translational modifications, including nitration,^384-387 sulfation^388-389 and phosphorylation^390-393. These PTMs, especially phosphorylation, play an important role in cell signaling,³⁹³ making Tyr an important target in protein footprinting.³⁹¹

Among all the Tyr modification approaches, nitration has the richest history. Early demonstration of Tyr nitration dates back to 1910s when Johnson and Kohmann³⁹⁴ determined the structure of nitrotyrosine and characterized Tyr nitration with nitric acid. Research over the years evolved to establish tetranitromethane (TNM) as the preferred Tyr nitration reagent, as was first reported by Wormall³⁹⁵ in 1930 but not well established until the 1960s.^396-397 Tyr nitration by TNM is a representative of electrophilic aromatic substitution,³⁹⁸ after which a nitro group is added to the electron-rich Tyr ring (Scheme 4, top). With excess amounts of TNM, a Tyr residue can be di-substituted as well. Although reaction between TNM and Tyr is reasonably specific, noticeable labeling of His,³⁹⁷ Met,^{397, 399} Trp,^{397, 400} and Cys⁴⁰¹ at alkaline pH can also occur. Reactions between proteins and TNM will also induce intra- and inter-molecular chemical cross-linking via Tyr residues under acidic pH,^{398, 402-403} providing a representative example of zero-length cross-linking⁴⁰⁴ (see sections 3.9 and 4.3 for more information). A good illustration for TNM-facilitated zero-length chemical cross-linking is the dimerization of insulin, which was reported independently by several research groups.^{403, 405}

Scheme 4.

Tyr labeling pathways with several reagents

TNM labeling coupled with MS analysis, as a footprinting platform has been used for protein HOS analysis. A pioneering study is by Ploug et al.⁴⁰⁶, where they characterized the interactions between urokinase-type plasminogen activator and its glycolipid-anchored receptor. Using MALDI-MS, they identified Tyr 57 in the receptor and Tyr 24 in the isolated growth factor-like module of the activator as the key interacting residues. In another study, Šantrůček et al.⁴⁰⁷ evaluated the correlation between TNM labeling extents and the SASA of the Tyr residues by using horse heart cytochrome c, hen egg-white lysozyme, and human serum albumin as model systems and MALDI-MS as the characterization tool. Surprisingly, they found only weak correlations between the TNM labeling extents and the SASA calculated from the crystal structure. For example, Tyr 148 and 341 in human serum albumin are highly reactive towards TNM but are poorly solvent-accessible. This study not only indicates that users be aware of other effects on footprinting by targeted labeling reagents and understand the importance of differential experiments when the factors that govern apparent reactivities are not fully understood. Method development studies are highly recommended.

Another Tyr nitration reagent is peroxynitrile, and it has become significant in the field of nitric oxide physiology and biological oxidations.^408-409 Its application in MS-based protein HOS analysis, however, is limited.⁴¹⁰ One possible reason is the low stability of peroxynitrile as compared with TNM. There are also reports demonstrating electrochemical⁴¹¹ and copper-catalyzed⁴¹² tyrosine nitration. Although promising, its application in protein HOS analysis remains to be established.

The second approach to label Tyr is iodination. As depicted in Scheme 4 (middle), iodination can lead to mono-iodotyrosine or di-iodotyrosine depending on reaction conditions. Although seldom used in protein footprinting, Tyr iodination is still valuable in protein radiolabeling (¹³¹I has a radioactive decay half-life of 8 days⁴¹³) and in pharmacokinetic studies.⁴¹⁴ To date there are several methods for Tyr iodination, including sodium iodide (NaI) coupled with N-chloro-4-methyl-benzenesulfonamide (chloramine T),^415-417 I₂ with iodine monochloride (ICl),^418-419 1,3,4,6-tetra-chloro-3a,6a-diphenyl-glycouril (iodogen),⁴²⁰ lactoperoxidase⁴²¹ and N-iodosuccinimide⁴²². Iodination of Tyr by these reagents occurs as electrophilic aromatic substitutions, with iodonium ion or hypoiodous acid as the reactive species.⁴¹⁴ Other than Tyr iodination, oxidation of Met, Cys and Trp by these iodinating reagents should not be ignored.⁴²³

To utilize Tyr iodination for protein footprinting, Hobba et al.⁴²⁴ in 1996 first demonstrated its efficacy by probing the insulin-like growth factor binding sites of bovine insulin-like growth factor protein-2. When the latter protein was footprinted by chloramine T-mediated iodination, Tyr 60 showed protection upon interacting with its binding partner. Confirmatory evidence is that iodinated bovine insulin-like growth factor protein-2 has a significantly lower binding affinity with its binding partner, consistent with the importance of Tyr 60 whose iodination perturbs binding. Note that Tyr iodination can also be achieved by a radical-mediated pathway,¹⁸¹ which will be discussed in section 5.6.

The last Tyr labeling reagent for discussion is N-acetylimidazole (NAI), which was first demonstrated to be a protein labeling reagent in 1963.⁴²⁵ Subsequent study by the same group showed the applicability of NAI as a protein footprinting reagent by analyzing solvent accessible and buried Tyr residues in several proteins.⁴²⁶ NAI acetylates phenolic hydroxyl groups through nucleophilic substitution, after which the Tyr residue becomes O-acetylated (Scheme 4, bottom). Acetylation of Ser and Lys also occurs occasionally.^427-428 Tyr labeling by NAI is optimized at a pH between 7.0 and 8.0 to minimize the hydrolysis of NAI and the product.^428-430 The application of NAI in mapping solvent accessible Tyr residues was nicely demonstrated by Pucci and coworkers⁴³¹ in 1997, where the surface topology of a small de novo designed β-protein minibody was mapped by MS-based peptide footprinting. The NAI footprint revealed that Tyr 35, 39 and 59 are exposed for NAI labeling whereas Tyr 15, 24 and 47 are silent. These results were further confirmed by TNM footprinting. Together with footprinting Lys (by acetic anhydride) and Arg (by 1,2-cyclohexanedione), the investigators obtained a comprehensive understanding of the minibody. Because NAI targets the hydroxyl group whereas iodination and TNM activate the C-H bond on the aromatic ring, this combination of Tyr footprinting will presumably provide different conclusions when a specific Tyr residue is involved in hydrogen bonding.⁴³¹ When Tyr is involved in hydrogen bonding, the hydroxyl group is not available for NAI footprinting whereas TNM is still capable of labeling the aromatic C-H bonds.

There have been two other acetylation reagents that are less specific as compared with NAI, 3-acetoxy-1-acetyl-5-methylpyrazole⁴³² and N-(2,2,5,5,-tetramethyl-3-carbonylpyrrolidine-1-oxyl)imidazole⁴³³. Both label aliphatic hydroxyl groups (Ser and Thr) as well.

Apart from the three major reagents introduced above, some other reagents for Tyr footprinting are acid anhydrides and acyl chlorides (target phenolic hydroxyl groups),^{428, 434} cyanuric fluoride (targets phenolic hydroxyl groups),^435-436 various diazonium salts (target phenolic C-H bonds),^437-438 polyhalogenated quinones (target phenolic hydroxyl groups),⁴³⁹ p-nitrobenzenesulfonyl fluoride (NBSF, targets phenolic hydroxyl group),^440-441 diisopropylphosphorofluoridate (targets phenolic hydroxyl groups),^442-443 p-nitrophenyl acetate (targets phenolic hydroxyl groups)⁴⁴⁴ and hemin-activated luminol derivatives (target phenolic C-H bonds)⁴⁴⁵. Although this wide variety of reactive reagents demonstrates the importance of functional groups in footprinting, most await establishment in MS-based protein HOS elucidation.

3.4. Aspartic Acid and Glutamic Acid

Carboxylic acid functional groups in aspartic (Asp) and glutamic acid (Glu) residues are the targeted functional groups in Asp and Glu labeling. By incubating with methanol under acidic conditions, carboxyl acids can be esterified as first demonstrated for lysozyme in 1945.⁴⁴⁶ Later on, IAM,^447-448 N-ethylphenylisoxazolium tetrafluoroborate⁴⁴⁹ and various diazo derivatives^450-454 were also found to label Asp and Glu. These esterification reactions, however, are highly reversible and require methanol solvent and acidic conditions, making them inappropriate for footprinting under physiological conditions. Diazo reagents, although reactive under milder conditions, are not stable in aqueous solutions. For these reasons, the above reagents are less commonly used today in protein footprinting. Another class of reagent that was applied in Asp and Glu labeling is Woodward’s reagent K (WRK).^455-457 Despite its success in testing covalent inhibition,^458-459 it has not been used protein HOS interrogation because it may be too bulky and subject to steric hindrance to probe protein SASA effectively.

The emergence of carbodiimide-based reagents offers a welcome solution for footprinting acidic residues. The reagents are water-soluble and capable of labeling Asp and Glu residues under physiological conditions. Two early promising reagents are 1-cyclohexyl-3-(2-morpholinyl-(4)-ethyl) carbodiimide^460-463 and 1-ethyl-3-(3-dimethyl-aminopropyl) carbodiimide.⁴⁶² On the other hand, their reaction products have been poorly characterized, owing in part to limited characterization techniques at the time of study and to low yields.⁴⁶³

The breakthrough of carbodiimide Asp and Glu labeling was by Hoare and Koshland⁴⁶⁴ in 1966, where carbodiimide serves as activating reagent, and a second reagent is incorporated to accomplish the labeling, affording an improved labeling efficiency and higher confidence in identification. In the original demonstration, N-benzyl-N'-3-dimethylaminopropylcarbodiimide (BDC) was the activating carbodiimide, and glycine methyl ester was the labeling reagent. Subsequent studies^465-466 revealed that the basis for quantification of Asp and Glu in proteins by carbodiimide-activated Asp and Glu labeling is activation of the carboxylic acid by carbodiimide to form an O-acylisourea intermediate (see Scheme 5). The intermediate then reacts rapidly with the nucleophilic labeling reagent to give a stable final product. In principle, any combination of water-soluble carbodiimide and a highly reactive nucleophile can facilitate such reactions. Currently, the activating carbodiimide used in Asp and Glu labeling is almost exclusively 1-ethyl-3-(3-(dimethylamino)propyl)carbodiimide (EDC), thanks to its high solubility and stability. The labeling in the early work, measured by amino acid analysis of the protein constituents, was motivated by analytical needs to “count” the number of Asp and Glu in a protein by quantifying by amino acid analysis the increase in glycine that occurs as a consequence of labeling.

Scheme 5.

Asp and Glu labeling pathways with EDC-activated labeling reagents as demonstrated by Asp

The evolution of Asp and Glu footprinting from basic chemistry to application is typical for development of footprinting other amino acid residues. The extension of the labeling to footprinting and the study of protein HOS was not imagined in the early work, probably because complete analysis of the labeling was difficult in the absence of modern MS. Now, applications are not only facilitated by improvements in analysis technology but also extended by reagent development of several nucleophiles that couple with EDC, including glycinamide (GA), glycine ethyl ester (GEE) and benzhydrazide (BHD), providing for the field a new birth (pathways for these reagents are in Scheme 5).

Asp and Glu labeling by GA (and possibly by isotopic encoded GA for more reliable identification and quantification) and glycine methyl ester⁴⁶⁶ was not applied to study protein HOS until decades later. The applicability of GA as a MS-based Asp and Glu footprinting reagent was first demonstrated by Akashi et al.⁴⁶⁷ in 1993. Using a differential bottom-up footprinting workflow, the investigators identified Asp 101 as a key interacting residue in hen egg-white lysozyme when binding with its inhibitor, tri-N-acetylglucosamine. Moreover, the investigators utilized multiple MS-based techniques, where they measured intact mass of the protein by ESI-MS, the digested peptides by LC separation, UV detection and Frit-FAB MS, and conducted tandem-MS for selected peptides to identify the modified residues by a sector mass spectrometer. Taken together, these systematic characterization steps prefigured the modern protein footprinting workflow. Although multiple mass spectrometers were used at that time, the analysis can now be accomplished by a single LC-MS/MS system.

In perhaps the first EDC-facilitated GEE footprinting of Asp and Glu, Sanderson and Mosbaugh⁴⁶⁸ investigated how PBS2 uracil-DNA glycosylase inhibitor (Ugi) protein inactivates uracil-DNA glycosylase (Ung) by forming a protein-protein complex in which Ugi mimics duplex DNA bacteriophage PBS2 uracil-DNA glycosylase inhibitor protein. Electrophoresis was used to separate forms of the protein that differ in the amounts GEE adduction. Subsequent bio-functional assays determined their activities whereas protein cleavage by cyanogen bromide and analysis by MALDI MS and amino-acid sequencing localized the GEE modification sites. Because species that are heavily modified are biologically inactive, they were able to identify in systematic studies the key Asp and Glu residues that retain the protein’s activity to be Glu-28 and Glu-31 in α2-helix.

As mentioned above, modern approaches utilize on-line LC in combination with MS and MS/MS for better separation and quantification with a single analytical system. In an example from 2009, Wen et al.⁴⁶⁹ utilized GEE to footprint the Fenna-Matthews-Olson (FMO) antenna protein from Chlorobaculum tepidum. By LC separation and MS and MS/MS identification, locations and extents of labeling of FMO protein located in a native membrane as well as on a chlorosome-depleted membrane were compared. The differential footprinting reveals for the first time the orientation of the protein associated with the cytoplastic membrane as the Bchl a #3 side. Later on, isotopic encoded GEE was implemented for a better and more convenient identification.⁴⁷⁰

Although effective, GEE has its disadvantages. The ester group in the GEE molecule will undergo hydrolysis under both acidic and basic conditions. As the LC solvent usually contains acid to provide better ionization efficiency in MS, the hydrolysis problem can be significant.⁴⁶⁹ Practically both the GEE modified species and its hydrolyzed species were quantified, which not only complicates the data analysis but also increases the errors in the quantification. To overcome these disadvantages, Guo et al.⁴⁷¹ developed BHD, which reacts with Asp and Glu under EDC activation (Scheme 5, bottom). Moreover, BHD labels the Asp and Glu residues with higher efficiencies than does GEE. BHD also requires lower reagent amounts, resembling better the native-like environment of the target protein molecule.

3.5. Arginine

Arginine (Arg) labeling draws significant attention owing to its participation in salt bridges.^472-473 Moreover, Arg residues can form up to five hydrogen bonds with other partners, facilitate hydrophobic interactions through its three methylene carbons, and interact with other aromatic functional groups via its guanidinium group.³⁸³ All these interactions make Arg critical in the protein folding that gives functional structures for protein and protein complexes. Post translational modification of arginine residues is also critical in regulating protein-nucleic acid⁴⁷⁴ and protein-protein interactions.⁴⁷⁵ Its chemical labeling, however, is challenging as the large pK_a (high basicity) of the guanidino group (greater than 12)⁴⁷⁶ keeps Arg protonated at ~ pH = 7 and decreases its nucleophilicity.

Arginine residues are mostly labeled by vicinal dicarbonyl reagents, resulting in a cyclic product (summarized in Scheme 6). Arg labeling by phenylglyoxal and its derivatives was first demonstrated by Takahashi⁴⁷⁷ in 1968, where the reactivities of phenylglyoxal and the Arg residues in peptides and bovine pancreatic ribonuclease A were systematically studied. The reaction between phenylglyoxal and an Arg residue is a two-step process, as shown in Scheme 6, and both steps are reversible. The final product to which two phenylglyoxal molecules are added, however, is stable under acidic conditions but readily decomposes at neutral and basic pH.^477-479 Other phenylglyoxal derivatives including p-hydroxyphenylglyoxal,⁴⁸⁰ 4-hydroxy-3-nitrophenylglyoxal,⁴⁸¹ p-nitrophenylglyoxal,⁴⁸² p-azidophenylglyoxal⁴⁸³ were also demonstrated to be valuable Arg labeling reagents. Given a deep understanding of the reactivity and appropriate reaction conditions, phenylglyoxal and its derivatives are the most extensively used Arg labeling reagents.²⁷¹

Scheme 6.

Arg labeling pathways with several reagents

As modern HPLC systems when coupled with MS analysis use acidic mobile phases (if working in the positive-ion mode of the mass spectrometer), stability of labeling product at acidic pH allow phenylglyoxal to be a MS-friendly Arg footprinter. The application of phenylglyoxal as an Arg footprinting reagent was pioneered in 1995 by Coggins and coworkers⁴⁸⁴, who used ESI-MS to identify Arg 19 and Arg 23 as critical functioning residues for type II dehydroquinases in Streptomyces coelicolor and Aspergillus nidulans, respectively. Because MS instrumentation with ESI was beginning to emerge at the time of that study, key Arg residues had to be located by bottom-up MS analysis and confirmed by comparing the masses of resulting peptides with theoretical values instead of MS/MS. In a more sophisticated example later by McLafferty and coworkers,⁴⁸⁵ phenylglyoxal Arg footprinting was coupled with tandem MS to allow confident identification of three key Arg residues that facilitate the function of rabbit muscle creatine kinase.

Another popular vicinal dicarbonyl reagent that labels Arg residues is 2,3-butanedione, which was first reported by Yankeelov and Crawford⁴⁸⁶ in 1968. Upon reacting with Arg, a five-membered ring is the primary product (Scheme 6). The reaction between an Arg residue and 2,3-butanedione is highly reversible, and the labeling proceeds significantly slower than that of phenylglyoxal, limiting the application of 2,3-butanedione. In 1973, Riordan⁴⁸⁷ discovered that borate has a dramatic effect on the reactivity of 2,3-butanedione with Arg residues, as borate stabilizes the initial diol product, thus pushing the reaction to a higher yield (Scheme 6). Similarly, Leitner and Linder^488-491 demonstrated that phenylboronic acid is also capable of stabilizing the initial adduct. This series of studies also represents the early applications of 2,3-butanedione as an Arg footprinting reagent in MS-based HOS analysis.^488-491 Currently, 2,3-butanedione labeling is mainly executed in borate buffers in the dark to minimize the photoactivation of 2,3-butanedione and to increase the labeling specificity and efficiency. 2,3-Butanedione is also capable of reacting with citrulline (the product of Arg residue upon reacting with peptidylarginine deiminase), as demonstrated in a few MS-based footprinting studies.^492-493

The last vicinal dicarbonyl reagent for Arg labeling is 1,2-cyclohexanedione, which was first shown to modify Arg residues under basic conditions (pH > 12 for effective deprotonation) in 1967.⁴⁹⁴ This pH is not biologically relevant and greatly limits its application in protein labeling and footprinting. In 1975, Patthy and Smith^495-496 discovered that borate can stabilize the diol product between 1,2-cyclohexanedione and an Arg residue under physiological pH, a similar phenomenon as Riordan⁴⁸⁷ observed for 2,3-butanedione.

The application of 1,2-cyclohexanedione in MS-based Arg footprinting was first reported by Przybylski and coworkers⁴⁹⁷ in 1992, when they found that four out of 11 Arg residues were modified in egg white lysozyme. By comparing the labeled Arg residues with those in a crystal structure, the investigators found an inverse correlation between labeling efficiency and solvent accessibilities. Upon closely examining the local environments of the labeled Arg residues, they were able to show that the proton acceptor groups are in close proximity. These neighboring proton acceptors assist deprotonation of Arg residues (first step in 1,2-cyclohexanedione labeling) and facilitate Arg labeling in an intramolecular catalytic fashion. All in all, this work is not only one of the pioneering studies that demonstrate protein footprinting by targeted labeling reagents but also is an early example that addresses effects of local environment on the labeling efficiencies of targeted labeling reagents.

An Arg labeling reagent that is not a vicinal dicarbonyl compound is represented by 3-ethoxy-1,1-dihydroxy-2-butanone (kethoxal), which is a popular RNA footprinting reagent that reacts with the guanine group.^498-499 Kethoxal was later demonstrated to react with Lys residues in proteins.^500-502 MS-based Arg footprinting by kethoxal was first demonstrated by Fabris and coworkers⁵⁰² in 2005, where they adopted Arg-containing protected amino acids, di- and tri-peptides, and model proteins to investigate the reaction product between kethoxal (structures depicted in Scheme 6) and an Arg residue and to demonstrate its capability as a Arg footprinting reagent. As the initial addition of kethoxal forms a diol product, the reaction can also include borate for higher yield.

Other than the reagents introduced above, there are other Arg footprinters including diphenylethanedione,⁵⁰³ malondialdehyde,^504-506 methylglyoxal^507-508, ninhydrin^509-510. These reagents, although they show potential for reacting with Arg, have found little application thus far in Arg footprinting.

3.6. Histidine

Targeted labeling of histidine (His) has drawn considerable attention owing to its high abundance in many enzyme active sites.⁵¹¹ The two nitrogen atoms on the imidazole ring impart a unique character as an acid-base catalyst,⁵¹² which further facilitates various enzymatic transformations. In general, His residues are not critical for structural stability of proteins,⁵¹³ and surface-accessible His residues are not involved in catalytic domains.⁵¹⁴ These general observations make His footprinting important for assessing structures of enzymes because buried His residues often play critical roles in bio-catalytic processes.

Early His modifications were achieved by using bromoacetic acid and bromoacetate.^{288, 293, 515} Two nitrogen atoms on the imidazole are potential nucleophilic sites for modification. As halogen atoms are good leaving groups, many α-halo carboxylic acids and amides were developed and utilized in His labeling, including but not limited to bromoacetate,^{288, 516-521} chloroacetate,^{288, 293, 517} iodoacetate,^{517, 519, 522} bromoacetamide⁵²³ and IAM⁵²⁴. Although effective, these reagents soon fell into disuse, primarily due to their lack in specificity. These reagents were later found to be reactive with other amino acid residues, most notably cysteine, which can be modified even more rapidly than histidine.²⁸⁹ IAM today is one of the most widely used Cys alkylating reagent as mentioned in section 3.1. His labeling by bromoacetate, a reagent that can be used for Cys-free proteins, however, requires days to complete, greatly limiting broad application^{516, 521} and causing concerns about structural change during the footprinting and misleading over-labeling. A similar situation applies to methyl-p-nitrobenzenesulfonate, which can label His residues but now principally serves as a alkylation reagent for Cys.^525-526

Another reagent that is related to the α-halo carboxylic acid is 2-bromo-1-(4-bromophenyl)ethenone (p-bromophenacyl bromide), which irreversibly labels His residues; this reagent is similar to other α-halo carboxylic acids. p-Bromophenacyl bromide was first reported to react with carboxyl groups in Asp and Glu at the active sites of pepsin.⁴⁴⁷ Subsequent studies demonstrated its capability for labeling His residues.^527-528 Although still being used today, p-bromophenacyl bromide has been almost exclusively used in labeling phospholipase A₂.^{271, 529-530}

The most extensively used His labeling reagent to date is diethylpyrocarbonate (DEPC), whose reaction pathway with His is described in Scheme 7. Development of His labeling by DEPC reactions with intact proteins was pioneered by Fedorcsák and coworkers^531-533 in the 1960s. DEPC has advantages over the other reagents because it has relatively high specificity²⁷¹ and reactivity⁵³⁴. In the pH range between 5.5 to 7.5, DEPC primarily labels His residues, but it is also reactive with Arg, Cys, Lys, Ser, Thr, and Tyr residues.^535-538 As illustrated in Scheme 7, mono-modification of His by DEPC is reversible in the presence of other nucleophiles⁵³⁸ as well as under both acidic and basic conditions (half-life for 3-carboethoxyhistidine is 55 h at pH of 7, 2 h at pH of 2, and 18 min at pH of 10⁵³⁹).

Scheme 7.

His labeling pathways with DEPC

With excess DEPC, the His residue can be di-modified irreversibly. The reactivity between His and DEPC increases in acidic buffer lower than pH 6 to 7.⁵⁴⁰ The most significant disadvantage of DEPC in His labeling is that DEPC is insoluble in water, making it essential to add water-miscible organic solvents (e.g., methanol, acetonitrile). The amount of organic solvent needs to be carefully tuned to prevent protein denaturation, which can be measured in preliminary experiments using, for example, CD. A recent study also suggests that the presence of hydrophobic residues increases the local concentration of DEPC at the protein surface, increasing the reactivity.⁵⁴¹

The potential of DEPC in MS-based protein footprinting was explored through a demonstration study that used FAB-MS to characterize the reaction product between DEPC and five model peptides.⁵⁴² Taking advantage of fragmentation accompanying FAB ionization, the investigators identified major reaction products between DEPC and His residues as well as their fragmentation pathways. In one of the pioneering MS-based His footprinting studies that uses DEPC as labeling reagent, Glocker et al.⁵⁴³ in 1996 mapped solvent-accessible His residues in recombinant human macrophage colony-stimulating factor β by using MALDI-MS for analysis. Among four DECP-modified His residues, His 9 and His 15 are directly correlated with biological activity whereas the activity is not affected by labeling of His 176 and His 210. In another study, Halsall and coworkers⁵⁴⁴ used LC ESI-MS to determine the location and modification extent of solvent-accessible His residues in α₁-acid glycoprotein. In addition to its biological significance, this work nicely illustrates a modern MS-based protein footprinting workflow, where the digested peptides are separated by HPLC, followed by MS/MS analysis to determine the modification sites, and quantification by integrating extracted ion chromatograms. Recent advances of DEPC labeling were mostly by Vachet and coworkers,^{169, 535, 541, 545-548} and some of their contributions will be reviewed in Section 4.

Other than the reagents introduced above, epoxides can also label His, but these reagents, are not specific towards His. Very recently, Joshi and Rai⁵⁴⁹ scanned a series of nucleophiles and found 2-cyclohexenone reacts with the His residue in different proteins with high specificity. Like DEPC, 2-cyclohexenone is poorly soluble in water, necessitating the undesirable addition of organic solvents during labeling. Although not without drawbacks, this study nicely demonstrates the potential of 2-cyclohexenone and its derivatives in His footprinting.

3.7. Lysine

Lysine (Lys) residue side chains and the N-terminus are the sites of primary amine groups. As these primary amines are most commonly found on protein surfaces,¹⁶⁹ labeling Lys residues becomes valuable for probing protein surface structure and its changes with respect to ligand binding or environmental change.⁴⁹⁷

Several different reagents were developed to label the primary amine group in Lys residues (summarized in Scheme 8). The most widely adopted approach is acylation by acid anhydrides, as exemplified by acetic anhydride in Scheme 8. Reaction between acetic anhydride and a Lys residue (Scheme 8, top-left) is highly pH-dependent, as the acetic anhydride readily hydrolyzes under acidic pH. Acetic anhydride can also react with hydroxyl groups found on Ser and Thr sites and aromatic Tyr sites, and with the imidazole ring (His). Formation of O-acetyl tyrosine can be minimized under alkaline pH or in the presence of acetate.^550-551 The other side reactions (with Ser, Thr and His) are also reversible and can be minimized by optimizing the conditions.⁵⁵² Although acetic anhydride is one of the first Lys labeling reagents,⁵⁵³ it still is used today in MS-based protein HOS analysis.^{497, 554-562} Lys footprinting by acetic anhydride was first demonstrated by Fenselau, Vestling and coworkers in 1991⁵⁶³, where they studied the melittin binding sites in calcium-bound calmodulin. By comparing Lys modifications of calcium-bound calmodulin under both melittin bound and unbound states, they identified Lys 21, 75 and 148 as interacting Lys residues by FAB-MS analysis. In another pioneering demonstration, Przybylski and coworkers⁴⁹⁷ analyzed the surface topology of egg white lysozyme. Footprinting Lys residues in egg white lysozyme and quantifying by ²⁵²Cf plasma desorption MS, they obtained reactivities for six Lys residues and showed the reactivities correlate well with the SASA values derived from a crystal structure. The results highlight early examples of the potential of targeted amino acid labeling in combination of MS analysis, primitive by today’s opportunities, for analyzing protein HOS.

Scheme 8.

Lys labeling pathways with several reagents

Another popular anhydride for Lys modification is succinic anhydride (Scheme 8, middle-left).⁵⁶⁴ Under physiological pH, Lys is usually positively charged as the primary amino group is protonated. When Lys reacts with succinic anhydride, the product is a carboxylic acid that deprotonates under neutral pH (Scheme 8). In other words, Lys modification by succinic anhydride results in charge reversal, which usually leads to dissociation of multimeric protein aggregates and has been widely utilized to solubilize insoluble proteins.^565-568 Another unique feature of succinic anhydride is that it can efficiently modify Lys residues at acidic pH,¹⁶⁹ which is significantly different than those for acetic anhydride. Other than amino groups, succinic anhydride is also found to be reactive towards Tyr residues.⁵⁶⁹ In the earliest example of Lys footprinting with acetic anhydride and MS-based protein HOS analysis (1996), Przybylski et al.⁵⁷⁰ studied the Rhodobacter capsulatus general diffusion porins. MS together with X-ray crystallography were used to identify the N-terminus and three other modified solvent-accessible Lys residues (Lys 46, 298 and 300). Charge reversal upon succinylation induced an increase in cation selectivity and single-channel conductance of the porin.

Other than acetic and succinic anhydrides, the acylation of Lys can also be done with other anhydrides, including citraconic anhydride,^571-572 maleic anhydride,^573-574 trimellitic anhydrides,⁵⁷⁵ phthalic anhydride,⁵⁷⁶ 3-hydroxyl phthalic anhydride,⁵⁷⁵ hexahydrophthalic anhydride,^577-578 methyltetrahydrophthalic anhydride,^{577, 579} succinic anhydride,^580-581 cis-aconitic anhydride^{580, 582} and fatty acid anhydrides⁵⁸³.

Unlike succinic anhydride that reverses the charge of Lys upon labeling, methyl acetimidate labeling retains the positive charge on the Lys residue at physiologically relevant pH (Scheme 8, bottom-left).⁵⁸⁴ Methyl acetimidate is a representative reagent in the imidoester family that targets α-amino groups. Other popular imidoester reagents include ethyl acetimidate,⁵⁸⁵ methyl isonicotinimidate,⁵⁸⁶ S-methylthioacetimidate,⁵⁸⁷ isethionyl acetimidate,⁵⁸⁸ and S-sulfethylthioacetimidate⁵⁸⁹. The latter two are sulfonic salts, are membrane-impermeable, and can be utilized to study soluble domains of cell membrane and membrane-bound proteins, providing spatial specificity (inner and outer membrane sides).⁵⁹⁰ S-Methylthioacetimidate also labels Cys, resulting a methyl disulfide with the sulfhydryl group.⁵⁹¹ The most significant drawback for imidoesters is their limited half-life in water owing to the rapid hydrolysis of the esters.^{584, 592} Therefore, imidoesters have not been often used in protein HOS analysis. The bis-imidoesters, however, are a popular cross-linking reagent, and details on this application are in section 3.9.1.

N-Hydroxysuccinimide ester (NHS-ester, Scheme 8, top-right) derivatives are another class of reagents in Lys labeling. NHS-ester was adopted in protein modification by Blumberg and Vallee⁵⁹³ in 1975. Its sulfonic salt, sulfo-NHS ester, was developed by Staros^594-595 in 1982 and was widely adopted owing to its improved solubility.

Other than Lys, NHS-esters also react with Ser, Thr, and Tyr residues.⁵⁹⁶ The NHS-ester is also frequently used for biotinylation of proteins, a process that is of great importance in protein purification because the modified protein binds tightly with streptavidin (and also avidin).^597-598 The biotin derivative of the NHS-ester has been widely used in MS-based Lys footprinting and in locating binding sites.^599-604 In pioneering work, Knock et al.⁶⁰⁰ in 1991 footprinted Aplysia egg-laying hormone with a biotin-functionalized NHS ester. Time-dependent labeling of the hormone revealed that among three amino groups (Lys 36, Lys 8 and N-terminal NH₂) that were labeled by NHS-ester, Lys 36 is the most solvent-accessible whereas the N-terminal amino group is the most protected. In combination with data from a bioactivity assay, they concluded that Lys 8 and the N-terminal amino group are critical for preserving bioactivity.

Another biologically significant reagent that is based on NHS-ester chemistry is the Bolton-Hunter reagent, succinimidyl-3-(3[¹²⁵I],4-hydroxyphenyl)propionate, which plays an important role in protein radiolabeling (¹²⁵I is radioactive).⁶⁰⁵ The fatty acid of sulfo-NHS is also used for footprinting because it can modify membrane proteins.⁶⁰⁶ NHS-esters are also widely used in chemical cross-linking of proteins, as discussed in section 3.9.1.

2,4,6-Trinitrobenzenesulfonic acid (TNBS) reacts with primary amines with reasonable specificity (Scheme 8, middle-right), and the incorporation can be easily measured by utilizing its optical absorbance at 420 nm, a characterization strategy that is readily adoptable.^607-609 These unique features to footprint free amino groups (N-terminus α-amine and ε-amine in Lys residues) in proteins at the time (1960s) were utilized when routine spray or desorption ionization coupled with high resolution MS analysis was not available. Moreover, protein HOS analysis and conformational changes can also be followed with TNBS coupled with absorbance spectroscopy in a protocol that is similar to protein HOS analysis with fluorescence spectroscopy (absorption versus emission).^610-612 TNBS footprinting of the N-terminal α-amino group induces a strong hydrophobic shift in the LC separation, and this feature was utilized in peptide separations of complex proteomic samples.⁶¹³ The combination between TNBS labeling and MS detection in protein HOS interrogation seems promising but is not yet well established.

Similar to TNBS, another potentially favorable yet poorly demonstrated footprinting reagent is methyl acetyl phosphate (Scheme 8, bottom-right).⁶¹⁴ As the phosphate group carries negative charges under physiological conditions, methyl acetyl phosphate becomes a useful probe for characterizing anion binding sites.^615-617 A similar phosphate-containing Lys labeling reagent is pyridoxal-5′-phosphate (PLP), which also preferentially binds to anion binding sites prior to Lys labeling.^618-619 Its fluorescence⁶²⁰ and UV absorbance⁶²¹ properties were utilized to study protein conformational changes. The application of PLP in MS-based protein footprinting, however, is not well tested. One problem is that the phosphate group will cleave to lose phosphoric acid during the CID process, preempting useful sequence-specific fragmentation. This problem is resolvable by using ETD in the MS/MS.

Other than the reagents discussed above, dinitrofluorobenzene,⁶²² 4-chloro-3,5-dinitrobenzoic acid,⁶²³ cyanate,⁶²⁴ aldehydes (formaldehyde, glycolaldehyde) coupled with reducing reagents (sodium borohydride, sodium cyanoborohydride or amino boranes)^625-626 can also label Lys residues. Reducing sugars are also capable of reacting with the ε-amino groups in Lys.⁶²⁷ These reagents, however, were not designed or used for protein HOS analysis and, thus, will not be covered further.

3.8. Other residues

Methionine (Met) is an attractive footprinting target, yet its labeling by targeted reagents is challenging. Despite being easily oxidized under various conditions, Met is only reactive with targeted labeling reagents under acidic conditions, where Met functions as a nucleophile.⁶²⁸ Alkylating reagents including iodoacetate,²⁹⁰ methyl iodide,⁶²⁹ iodoacetamide,⁶³⁰ and iodoacetic acid,⁶³¹ all of which react with Met in a pH range of 2 – 4. Furthermore, there are only a few demonstrations where these reactions occur under gentle conditions (iodoacetate with Met residues in pig kidney general acyl-CoA dehydrogenase pH = 6.6).⁶³² Thus, the likely disruption of protein native structure by placing it in an acidic environment make this reagent too risky as a footprinter.

Site-specific backbone cleavages of Met-containing peptides by cyanogen bromide (CNBr) was extensively used in the 1980s.^{271, 633-634} Upon reacting with CNBr, Met forms a five-membered ring with methyl thiocyanate as the leaving group. The five-membered ring subsequently cleaves to yield a peptide homoserine lactone and subsequently a peptide homoserine after hydrolysis. This reaction was usually coupled with top-down MS for analysis,^{172, 635} but top-down analysis becomes less necessary owing to the increasing effectiveness of current methods for inducing fragmentations via MS/MS.

Another study by Reid et al.⁶³⁶ utilized ω-bromoacetophenone as a Met alkylating reagent, and subsequent CID activation in MS/MS analysis caused a neutral loss that leaves a single characteristic product ion for Met. All in all, Met footprinting by targeted labeling reagents seems to be ineffective in contrast to Met oxidation^637-639 and the fast-labeling by hydroxyl radicals in the synchrotron and FPOP platforms (see section 5 and 6).

Serine (Ser) and threonine (Thr) both contain hydroxyl groups in their side chains, making these two residues as likely targets for different enzymatically introduced PTMs including O-linked glycosylation^640-642 and phosphorylation⁶⁴³. Introducing chemical modifications, however, is quite challenging. Both Ser and Thr can be acylated at lower reaction rates by p-nitrophenyl acetate,⁶⁴⁴ a reagent that primarily reacts with the stronger nucleophilic Tyr⁶⁴⁵ and Lys⁶⁴⁶. When located in N-terminus, Ser and Thr can be oxidized by periodic acid to form an aldehyde.⁶⁴⁷ Formic acid will esterify Ser and Thr residues at high concentrations in a reaction known as O-formylation.⁶⁴⁸ NHS-esters also react with Ser and Thr as side reactions.⁵⁹⁶ These reactions, however, are not of sufficient yield to footprint Ser and Thr residues in proteins under physiological conditions.

Asparagine (Asn) and glutamine (Glu) are the amide derivatives of Asp and Glu, respectively. The primary modification they undergo is deamidation, which occurs naturally as a PTM⁶⁴⁹ and is also observed during cyanogen bromide/formic acid cleavage.⁶⁵⁰

Glycine (Gly), alanine (Ala) and valine (Val), leucine (Leu), isoleucine (Ile) and proline (Pro) are generally inert to most chemical modifications that target functional groups as their residue side chains consist of unreactive H, alkyl, and cycloalkyl groups. Although the phenyl group of phenylalanine (Phe) can undergo electrophilic substitution, its chemical activation in a protein is also challenging under psychological conditions because the phenyl group is not reactive. Although Leu, Ile, Pro, and Phe are not activated by normal chemical processes, remarkably they can be modified by free-radical reactions, as will be covered in section 5.

3.9. Chemical Cross-linkers

Chemical cross-linking occurs via the formation of covalent bonds between interacting proteins and within a protein on adjoining regions. This structural proteomics tool often uses a bifunctional chemical reagent and can be viewed as bifunctional footprinting. Many reagents have been developed, promoting more and more applications of chemical cross-linking. The character of a cross-linking reagent is determined by the nature of chemistry, spacer length, and built-in functional groups. Cross-linking can be an effective complement to monofunctional reagents that serve as footprinters, and therefore the subject is briefly covered in this review particularly to make that point. The emphasis is on the commonly used reagents, which fits the theme of footprinting. Indeed, the location of monolinks can be viewed as a footprint. Our discussion is organized around the cross-linking reagents as follows.

3.9.1. Amine-reactive Cross-linkers: NHS-Ester and Imidoester

N-hydroxysuccinimide Ester

NHS-esters are the most widely used cross-linkers in field. The ester group undergoes attack by nearby nucleophilic sites (e.g., primary amines on Lys side chains or at the N-terminus of a protein/peptides, hydroxyl groups on Ser or Thr, and even sulfhydryl groups on Met) to form C-X bonds (X = N, O, S). The differing reactivities with various nucleophilic groups, however, depends on reaction conditions and can introduce bias. Primary amines possess the highest reactivity at physiological pH, and their reactivity can be further enhanced when the pH is > 7.^{595, 651} The hydrolysis of NHS-esters has a half-life of 4-5 hours at pH 7 and 0 °C⁶⁵²; the hydrolysis, however, is accelerated under alkaline conditions and at higher temperature.^653-654 Other possible products (i.e., esters and thioesters formed with Ser/Thr or Met) are less stable and undergo more rapid hydrolysis.⁶⁵⁴ Although the reaction is favored under acidic conditions (pH = 6.0), cross-linking at pH = 7 still occurs. The current consensus is to consider all possible residues as sites for reactions in simple protein systems, whereas only Lys residues and the N-terminus are recommended for cross-linking a large complex or in a whole proteome study to minimize dispersion of cross links and maximize probability for detecting the cross-links.⁶⁵⁵

The first NHS-ester originated in late 1970s as a homo-bifunctional cross-linker.⁶⁵⁶ Since then, many NHS-esters possessing useful physical and chemical properties have made them suitable for answering a range of biological questions. Designs include cross-liners with high hydrophobicity and zero charge, being lipophilic and membrane-permeable and ideal for intramembrane cross-linking (e.g., disuccinimidyl suberate (DSS) and disuccinimidyl glutarate (DSG), Scheme 9). Other cross-linkers incorporate a sulfonate group, which imparts water-solubility and avoids steps of pre-dissolution in organic solvents that could perturb aqueous conditions and cause some protein denaturation.

Scheme 9.

Common NHS-ester cross-linkers

Currently, sulfo-NHS esters have become the dominant cross-linkers in characterizing soluble proteins and their interactions. bis(Sulfosuccinimidyl)suberate (BS3, Scheme 9), the most extensively used sulfo-NHS cross-linker, can be encoded with deuterium to facilitate better cross-link assignment. The spacer length is 11.4 Å, allowing cross-linking for residues separated by 27 Å measured between two α carbons in an amino acid residues that are linked (termed Cα).⁶⁵⁷

New developments in NHS-esters resonate well with the rapid growth in proteomics research enabled by advanced MS. More complicated systems with larger-scale protein candidates require better sequencing in MS/MS for a confident cross-linking identification. The disadvantage of using conventional cross-linkers is that the user must sequence specie made up of two cross-linked peptides and the cross-linker, introducing complexity into fragmentation and challenging the acquisition of high-quality, readily interpretable MS/MS data. Cleavable cross-linkers are, therefore, designed to address that problem.⁶⁵⁸ NHS-esters have incorporated chemical-cleavable motifs (e.g., S-S bond in 3,3'-dithiobis(sulfosuccinimidylpropionate) (DTSSP)⁶⁵¹) and CID-cleavable functional groups that include C-S bonds (e.g., disuccinimidyl sulfoxide (DSSO)⁶⁵⁹ and cyanurbiotindimercaptopropionyl succinimide (CBDPS)⁶⁶⁰) and C-N bonds (e.g., disuccinimidyl dibutyric urea (DSBU)⁶⁶¹ and N-hydroxyphthalamide ester of biotin aspartate proline (BDP-NHP)⁶⁶²). Chemical structures of these cross-linkers are shown in Scheme 9. In addition, some NHS-crosslinkers contain biotin tags (e.g., CBDPS and BDP-NHP), to provide a means of enriching the crosslinked species by affinity purification.

Imidoester

Imidoesters, introduced in 1966 as one of the oldest reagents for protein cross-linking, are water-soluble reagents that react specifically with primary amines (see Lys footprinting section above).⁶⁶³ The functional imidate group reacts to form an amide bond via a few intermediates at an optimized pH range of 8-10.⁶⁵³ One major advantages of using imidoester cross-linkers is that the reaction product, an amidine, carries one positive charge. Charge removal that occurs with most lysine-targeting cross-linkers may disrupt intramolecular and intermolecular interactions and distort protein conformation, giving a biased result. The lifetime of imidoesters, however, is limited by rapid hydrolysis, being less than 30 min.^664-665 The most commonly used cross-linkers in this class are dimethyl pimelimidate (DMP), dimethyl suberimidate (DMS) and a cleavable analog, dimethyl 3,3′-dithiobispropionimidate (DTBP), as depicted in Scheme 10.

Scheme 10.

Common imidoester cross-linkers

3.9.2. Carboxylic Acid-reactive Cross-linkers: Carbodiimide and Dihydrazides

The most widely used carbomiimide is EDC (Scheme 11), also known as a “zero-length” cross-linker.^{404, 666} When this reagent facilitates cross-linking of carboxylate groups and primary amines, there is no spacer chain inserted between the targeted proteins, but rather an amide bond (~ 3 Å) between COOH and NH₂-containing sidechains.^{404, 666} One critical reaction intermediate, O-acylisourea, is not stable in aqueous condition (see section 3.4 and Scheme 5), and it continues to degrade back to the carboxyl group. Therefore, sulfo-NHS is often incorporated in the cross-linking protocol to transform the O-acylisourea into a stabilized NHS-ester for more efficient conjugation. EDC cross-linking shows the highest reactivity at pH 4.5 and still a moderate reaction efficiency at neutral pH.⁶⁶⁷ 2-(N-Morpholino) ethanesulfonic acid (MES) and phosphate buffer are compatible with carbodiimide reagents; however, more concentrated reagents should be used in the latter buffer owing to the reduced reactivity.

Scheme 11.

Common carbodiimide and dihydrazide cross-linkers

Another commonly used cross-linker for Asp and Glu is dihydrazides. In 2008, Kruppa and Novak⁶⁶⁸ first reported its use in combination with EDC activation. The reaction was carried out in acidic conditions (pH = 5.5), which is not physiologically friendly, to give a low cross-linking yield. Later in 2014, Aebersold and coworkers⁶⁶⁹ reported a new coupling reagent (other than EDC), 4-(4,6-dimethoxy-1,3,5-triazin-2-yl)-4-methyl-morpholinium chloride (DMTMM) and achieved significantly increased reaction yields and better biocompatibility at neutral pH. Recently, Lei and coworkers⁶⁷⁰ developed a coupling reagent-free crosslinker, bis(trimethylsilyldiazomethyl)dioxaoctane (Scheme 11), that offers selectivity and efficiency under physiological conditions. Isotopic encoded dihydrazides are now commercially available with different spacer lengths (e.g., adipic acid dihydrazide (ADH) and pimelic acid dihydrazide (PDH), Scheme 11).

3.9.3. Sulfhydryl-reactive Cross-linkers: Maleimide

Maleimide cross-linkers react specifically with sulfhydryl groups at pH 6.5-7.5 to form stable thioethers.^671-672 Under alkaline conditions (pH > 8.5), primary amines are also possible targets but not Tyr and His.⁶⁷³ One major concern in maleimide cross-linking is to conjugate free sulfhydryl groups, requiring prior reduction of existing disulfide bonds; breaking the -S-S- bond has high potential to distort protein native structure. A more common way of employing maleimide chemistry is in combination with NHS-esters, where both functional groups are coupled onto a heterobifunctional cross-linker. These cross-linkers can be incubated with a protein sample in a stepwise manner, reducing disulfide bonds after reaction with primary amines.⁶⁵³ In addition, to ensure the best cross-linking performance, many thiol-containing compounds (e.g., DTT and BME) should be eliminated in the reaction buffer.

3.9.4. Carbonyl-reactive Cross-linkers: Hydrazide

Hydrazides are carbonyl-reactive reagents that exhibit highest reactivity at pH 5-7. Aldehydes and ketones are major targets that can be found in glycoproteins introduced by oxidation of the polysaccharide.⁶⁵³ The hydrazone bonds thus formed are moderately stable in aqueous solution and can be further secured by reducing the double bond to a secondary amine.

3.9.5. Photoreactive Cross-linkers: Aryl Azide and Diazirine

Photoreactive cross-linkers are generally nonspecific owing to the high reactivity of the intermediate species, a nitrene, carbene, or free radical. The most adopted chemistry utilizes nitrenes and carbenes, whose precursors are azides or diazirines, respectively. Among the three major categories, phenyl azides play a dominant role in current applications. Different substituents on the aromatic rings shift their UV absorption dramatically; therefore, they are selected with a biological question in mind.⁶⁵³ Nitrophenyl azides, which can be activated at 300-460 nm, are compatible with most studies because this long wavelength causes minimal damage to protein molecules. Upon UV activation, the nitrene diradical can insert into most chemical bonds but with a preference for active C-H and N-H sites.⁶⁷⁴ Diazirines are a relatively new class, first reported in the 1990s, and they have better photostability than phenyl azides.⁶⁷⁵

Carbene diradicals are usually generated photochemically at 355 nm, and they show reactivity with both single and double bonds. Like the nitrene diradical, heteroatom-H bonds undergo easier insertion. Some diazirine reagents are designed as analogs of amino acids (e.g., photo-Leu and photo-Met⁶⁷⁶) that can be incorporated into the protein sequence during translation for in-situ radical generation. High reactivity of carbenes and nitrenes can also cause problems in cross-link identification. The use of homobifunctional photoreactive cross-linkers tends to make product analysis complex; therefore, one radical precursor is usually combined with an amine- or sulfhydryl-reactive motif. The commonly used cross-linkers are N-5-azido-2-nitrobenzoyloxysuccinimide (ANB-NOS)⁶⁷⁷ and sulfosuccinimidyl 4,4'-azipentanoate (sulfo-NHS-diazirine, sulfo-SDA), as seen in Scheme 12.

Scheme 12.

Chemical structure of common photoreactive cross-linkers, ANB-NOS and sulfo-SDA

3.9.6. Summary

We have seen extensive development of chemical cross-linking during the past two decades, enabled by evolving MS instrumentations, new data processing software, and novel cross-linkers. Chemical cross-linkers are bifunctional protein footprinters, which are governed by similar principles as protein footprinting by targeted reagents. The bi-functional nature of the reagent allows generating distance restraints between two cross-linked residues, and such restraints can be further utilized to locate protein/protein binding interfaces, to assess protein topologies, to characterize the protein interactome in a large complex, and to facilitate protein docking and modeling. Some applications of chemical cross-linking are discussed in section 4.3.

3.10. Conclusion and Perspective

Protein footprinting takes advantage of early research that provided many effective chemical reactions for protein labeling. These reagents can now serve as probes to characterize protein SASA and as a basis to reason about protein HOS. The sites of modification can now be efficiently measured by MS, owing to the rapid development of LC separations and hybrid mass spectrometers with MS/MS capabilities. The chemical cross-linker, which is viewed as a bifunctional protein footprinter, when connected to two proteins offers valuable distance restraints between the proteins. These targeted footprinting reagents coupled with MS detection now are utilized to answer several biological questions and are becoming increasingly significant in structural biology, as will be discussed in Section 4.

A limitation of targeted footprinting is that reagents primarily react with one or a few residues that contain functional groups including bases, acids, nucleophiles and aromatic rings, leaving a sizable fraction of the amino acid residues to be “silent”. Moreover, the footprinting reactions are relatively slow, making most targeted footprinters unable to characterize fast protein dynamics (e.g., protein folding and protein aggregation). To overcome these drawbacks requires a new approach that has broader residue coverage and higher reaction rate. Fast footprinting reagents represented by reactive radical species meet these requirements as discussed in section 5.

4. Applications of Targeted-Labeling Reagents

Targeted-labeling has become a mature field in structural proteomics. Research in targeted-labeling has evolved from developing reagents and methods to addressing biological questions or, in many cases, developing a method and using it to solve a biological problem. The maturation of the field runs parallel with MS instrumentation and method development. The key enabling progresses are (1) developing chemical reactions that are suitable for protein labeling and (2) utilizing high sensitivity, specificity, and speed now routinely available with modern mass spectrometer systems for analysis. Both steps are mature and permit careful application where perturbation of structure by the inserted reagent can be minimized or even eliminated.

A general workflow for bottom-up protein HOS analysis through irreversible protein labeling (through either targeted labeling reagents or reactive radical species) is shown in Figure 5 below.

Figure 5.

Schematic illustration of bottom-up protein HOS analysis through irreversible labeling approaches

In the general workflow, a protein of interest is first labeled in a native or native-like environment, during which the labeling reagent reacts with the solvent-accessible amino acid sidechains. Proteolytic digestion cleaves the protein molecules into peptide fragments that are then submitted to LC separation and MS analysis. By comparing the mass-to-charge (m/z) ratios of the precursor and its corresponding fragments, the investigator can assign chromatographic peaks to either the unmodified or the modified protein via its “proxy” peptides. Quantifying the modification extent can be achieved by integrating the corresponding chromatographic peaks (via extracted ion chromatograms) and comparing areas assigned to “modified” and “unmodified” peptides. When LC is insufficient to separate modified peptide isomers, fragmentation by ETD to allow comparison of the resulting product ion intensities may be an alternative approach for accurate quantification.⁶⁷⁸ Assignment of the precise location of the modification can often be ascertained by MS/MS, on-line with the chromatography. The underlying assumption is that a residue side chain that is more solvent-accessible will be modified more heavily as compared with a residue that is buried inside the 3D structure of a protein. Coarse aspects of protein HOS can be elucidated by analyzing the modification fractions of the intact protein, “proxy” peptides, and even single-residue levels, taking advantage of the specificity of the labeling chemistry.

Conventionally, protein footprinting is almost exclusively executed in a differential manner, for which the labeling fractions for a specific region or residue are compared for different protein states. The difference in states can be achieved by incorporating a binding partner, changing the temperature, varying the denaturant concentrations or pH, or modifying the protein in metagenesis. The changes in protein conformations that occur with these perturbations can be elucidated with spatial resolution limited by the number and spacing of the targeted residues. Recent developments also forecast that elucidation and prediction of protein HOS protein footprinting data for a single state can be done, as will be covered in later sections.

Generally, the labeling is irreversible, allowing a wide range of post-labeling sample handing procedures (i.e., sample enrichment, deglycosylation, and longer LC gradients) as compared to that used for HDX, where labeling is reversible and precautions are necessary to minimize back exchange. The irreversibility feature also allows addressing problems that HDX cannot.

Two disadvantages can limit the approach. First, the coverage is limited to the targeted residue(s), and the number of them in the protein of interest can be low. Second, because the labeling chemistry is generally slow (minutes), there is always a question whether the targeted protein undergoes a conformational change in the early stages of labeling and then continues to be modified to yield a composite and misleading footprint.

Given the maturity of the field and the large number of applications in the literature, we will select several applications of protein footprinting by targeted labeling reagents. The applications will be grouped by the biological questions that are addressed. We will limit our review to publications since 2009 and primarily highlight demonstrations that either embody novel developments or implement improved workflows. For work prior to 2009, we recommend a comprehensive review by Mendoza and Vachet¹⁶⁹ for a detailed historical perspective and early applications. For a discussion of reagent and method development, the reader is directed to section 3 where pioneering work is discussed.

4.1. Analyzing Metal-Ion Binding

Metal ion-protein interactions facilitate many cellular functions, especially signal transduction, as 25-50% of all proteins found in organisms contain metal ions.^679-682 Metal ions interact with proteins by coordination whereby electron pairs from the amino acid side chains bond with the vacant orbitals of the metal ions.⁶⁸² We will classify the protein-binding metal ions into two distinct categories based on their coordination properties. Soft metal ions represented by Zn(II), Cu(II), Cd(II) and Fe(II) bind with His and Cys whereas hard metal ions (e.g., Ca(II) and Mg(II)) usually coordinate with oxygen-containing functional groups in Asp and Glu.^683-684

Many biophysical approaches have been applied to characterize protein-metal ion interactions, including MS,⁶⁸⁵ NMR,⁶⁸⁶ X-ray crystallography,⁶⁸¹ cryo-EM,⁶⁸⁷ and fluorescence⁶⁸⁸. One approach is native MS whereby complexes are directly observed, but this begs the question of whether the gas-phase structures are secure and trustworthy reporters of those in the liquid phase. Footprinting avoids this question by using the mass spectrometer only as a measurement approach, not a reaction vessel. Footprinting can be conducted by HDX,^{232, 689} fast irreversible labeling,^690-691 or slow irreversible labeling. All can reveal metal-ion binding sites and the corresponding protein conformational changes induced by the metal binding. HDX probes the binding-induced backbone conformational changes, whereas Asp and Glu, commonly present in a binding site of a hard metal ion, do not react well with most of the fast-labeling free radicals in that version of footprinting. Although capable, HDX and fast free-radical footprinting (next section) report the consequences of metal-ion binding by modifying regions of the protein (i.e., nearby peptide bonds in HDX) that may not be directly involved in binding. A direct footprinting experiment seems better suited for probing metal-ion binding because they probe directly the metal ion binding unlike indirect methods that report on conformational changes occurring nearby. All of this assumes that the structural perturbation caused by the footprinting is minimal.

As examples of a direct, targeted footprinting reagents, DEPC affords modification of His, carbodiimide-activated reagents (e.g., GEE) label Asp and Glu, and succinimide derivatives, iodoacetic acid, and N-alkylmaleimides, for example, label Cys.¹⁶⁹

4.1.1. Protein Binding with Soft Metal Ions

Protein binding of soft metal ions, Cu(II) and Zn(II), is the most extensively studied. In a pioneering assessment of Cu(II) binding sites in a prion protein (PrP, from human), Qin and Westaway⁶⁹² in 2002 applied DEPC to footprint histidine-dependent Cu(II) binding sites and used MALDI for the analysis. No histidines are protected upon Ca(II), Mn(II), or Mg(II) binding, one or two residues are protected by binding of Zn(II) or Ni(II) ions, and five histidines are protected upon Cu(II) binding. Through a classical bottom-up approach, the investigators pinpointed the five histidine residues in mouse PrP23–231 that show reduced DEPC footprinting upon chelating with Cu(II), revealing the Cu(II) binding sites, which were further confirmed by a separate electron paramagnetic resonance (EPR) study.⁶⁹³ Subsequently, the same group successfully extended the method to identify the Cu(II) binding sites in doppel (a glycosylphosphatidylinositol-anchored protein).⁶⁹⁴

In 2007, Sarkar and coworkers⁶⁹⁵ applied DEPC footprinting in combination with EPR, UV and fluorescence spectroscopies to pinpoint the critical Cu(II) binding residues of human copper metabolism gene MURR1 (mouse U2af1-rs1 region1) domain (61-154) as His101, Met110 and His134. By adopting similar strategies, Zhao and Waite⁶⁹⁶ investigated the metal-binding properties of mcfp-4 matrix protein in load-bearing junctions, Binolf et al.⁶⁹⁷ characterized the Cu(II) anchoring sites in α and β-synuclein, Vachet and coworkers^545-546 located the Cu(II) binding sites in β-2-microglobulin and its pre-amyloid oligomers, and Karmakar and Das⁶⁹⁸ identified the copper binding sites in αA crystallin.

Another relevant yet distinctive demonstration in 2009 is by Ginotra and Kulkarni⁶⁹⁹, who characterized the solution structure of Cu²⁺(His)₂ under physiological conditions. The coordination between Cu(II) and His in proteins occurs through a Cu²⁺(His)₂ complex, whose coordination regime is either N4O4 or N3O2 donor atoms. The former was supported by various spectroscopic techniques in solution whereas the latter was favored from X-ray crystallography at a pH of 3.7. By DEPC labeling of Cu²⁺(His)₂ complex under physiological conditions and MS identification, the investigators found the most abundant species was a three DEPC adduct, indicating that the Cu²⁺(His)₂ complex exists in solution as a neutral five-coordinate structure with N3O2 donor atoms.

Protein binding with Zn(II) usually involves the canonical Zn(II) binding motif, termed a zinc finger, whereby a Cys thiolate is the chelating functional group.⁷⁰⁰ Binding with Zn(II) greatly reduces the solvent accessibility of the cysteine side chains and slows the footprinting. Forest and coworkers⁷⁰¹ in 1999 investigated Zn(II) binding of the ferric uptake regulation protein from E. coli by using IAM footprinting. Among three reactive Cys residues, Cys92 and 95 are involved in the Zn(II) binding. The remaining Cys132 alkylates with the fastest reaction rate among the three, implying that it is surface-exposed and unbound.

In 2001, Apuy el al.⁷⁰² used cleverly designed NEM-based Cys labeling in a pulsed fashion. Their experiment established the relative cysteine thiolate reactivities of six zinc fingers in metal-response element-binding transcription factor-1 as F5 > F6 ≫ F1 > F2 ≈ F3 ≈ F4, indicating lower binding affinities for zinc fingers F5 and F6. This study is also an early example of the use of isotopic encoding in protein footprinting whereby the investigators used h₅-NEM and d₅-NEM to allow a better quantification.

Atsriku et al.⁷⁰³ in 2005 investigated the relative reactivities of Cys residues in the zinc finger of estrogen receptor by iodoacetic acid labeling and found that Cys 240 is more reactive than Cys 237, although both are located in the second zinc finger of the estrogen receptor. The differences in reactivities are explained by the unequal positioning towards basic amino acids that affects the thiol pK_a. Hanas and coworkers^704-705 in 2005 used a Cys₂His₂ zinc finger peptide, Sp1-3, as a model system to investigate the interaction between Zn(II), Au(I) and the zinc finger protein. In addition to identifying the binding site through Cys footprinting by IAM, the investigators determined the relative binding affinities between the zinc finger and Zn(II) and Au(I). Upon lowering the IAM concentration (100 μM), Zn(II) binding becomes distorted whereas the Au(I) binding was not affected at 100 μM IAM and was diminished at 500 μM IAM, a concentration that eliminates Zn(II) binding. These results indicate that the zinc finger in Sp1-3 binds tighter to Au(I) than to Zn(II).

Millhauser and coworkers⁷⁰⁶ in 2007 characterized the Zn(II) binding properties of PrP (from Syrian Hamster) by DEPC footprinting. Syrian hamster PrP combines Zn(II) and Cu(II) binding in one protein; the Zn(II) binding, however, can change the overall Cu(II) redox properties by altering the distribution of copper ions among various binding modes. More importantly, DEPC footprinting coupled with a Zn(II) titration revealed the zinc binding affinity of prion protein 60-91, demonstrating new opportunities for targeted labeling reagents in the quantitative understanding of protein metal-ion interactions. Upon binding with Zn(II), the number of DEPC-reactive sites decreases, leading to a decreased DEPC footprinting as measured by MS. The system composition as a function of Zn(II) concentration (changed via a titration) can thus be monitored by quantifying the extent of DEPC footprinting and can be fitted by modeling to extract binding affinity.

Karmakar and Das⁷⁰⁷ in 2012 footprinted α-crystallin by DEPC followed by MALDI-ToF analysis and identified His79, 107 and 115 as key Zn(II) binding residues in αA-cystallin whereas His104, 111 and 119 are critical for Zn(II) binding of αB-crystallin. Their method is limited proteolysis; regions in which Zn(II) is bound show missed cleavages in trypsin proteolysis because the Zn(II) binding stabilizes the region in which it binds and causes missed cleavages. In 2013, Russell and coworkers⁷⁰⁸ used NEM labeling to map non-binding sites and to detect directly peptides containing the metal to locate the Zn(II) and Cd(II) binding to human metallothionein-2A. By combining bottom-up, top-down, and ion mobility approaches, the investigators not only identified the metal binding sites and binding orders but also provided insights on the structural transitions during the metal binding and demetallation in the gas phase. The idea of footprinting the metal chelating groups with targeted labeling reagents was also adopted to label specific cysteine residues by using reversible metal protection, which is important in protein engineering.⁷⁰⁹

In another study, Ramakrishnan et al.⁷¹⁰ characterized the Zn(II) binding sites of hepatitis B virus X protein (HBx) through HDX and NEM footprinting. HDX locates four peptides that exhibit decrease in deuterium uptake upon binding with Zn(II) whereas NEM footprinting pinpoints the critical Cys residues that are responsible for the Zn(II) binding. Moreover, the NEM footprinting in this study was executed as a kinetic study, where the use of multiple NEM labeling times increased the confidence for identification of sites as is found with HDX kinetic curves and can provide some assurance that the protein structure is not perturbed (perturbation would give a change in kinetics). Time-dependent footprinting by targeted labeling reagents is discussed in section 4.3.

4.1.2. Protein Binding with Hard Metal Ions

In the case of hard ions, Ca(II) is one of the most characterized owing to its biological significance in signaling.^711-712 The canonical Ca(II) motif, an EF-hand, utilizes the negatively charged carboxyl groups in Asp and Glu inter alia to chelate Ca(II) ions.^{679, 713} In a study by Zhang et al.⁷¹⁴ in 2012, the Asp and Glu residues in calmodulin (CaM) were footprinted by GEE in a method evaluation experiment. The Asp and Glu located on the EF-hands of CaM show reduced GEE footprinting upon binding with Ca(II), which is consistent with a SASA calculation of these residues. This study successfully demonstrates the efficacy of targeted labeling reagents in identifying hard-metal ion binding sites with high spatial resolution. By similar approach in the same year, Arata and coworkers⁷¹⁵ identified the Ca(II) binding sites in sarcoplasmic reticulum Ca²⁺-pump ATPase by DEPC footprinting.

Very recently, Guo et al.⁴⁷¹ examined the Ca(II) and Mg(II) binding properties of CaM by BHD footprinting and MS analysis. Besides demonstrating the efficacy of BHD, a new footprinter, in identifying metal binding sites, the investigators used comparison of the modification extents between Ca(II)-bound CaM and Mg(II)-bound CaM to determine that the conformational changes of CaM induced by Mg(II) binding are less significant than those by Ca(II). Because the high resolution structure of Mg(II)-bound, full-length CaM is not currently available,⁷¹⁶ this study shows the opportunities for footprinting to assist in HOS studies where X-ray crystallography is difficult. Successful probing of the low affinity interactions between Mg(II) and CaM also shows the potential of using targeted labeling reagents in protein-metal ion analysis and illustrates again opportunities afforded by isotope encoding.

Although powerful, the uses of targeted labeling reagents in probing protein-metal interactions are not numerous. An important reason is that the metal-binding pocket is usually conserved, making the size of the labeling reagent critical. Another reason is that reaction rates between targeted labeling reagents and the residues are usually slow, making it challenging to characterize metal-binding systems with low binding affinities (whose metal ion off-rates are usually fast). Footprinting by using fast-labeling reagents, although an indirect method, can also overcome the above disadvantages to characterize protein-metal ion interactions, a subject to be covered in section 6.4.

4.2. Mapping Cys and Disulfide Bonds

Disulfide bonds are critical in protein folding and protein structural stabilization, making localization of disulfide bonds an important task. MS-based approaches contribute significantly to the determination of disulfide bonds.⁷¹⁷ Although many methods including plasma-induced oxidative cleavage,⁷¹⁸ ETD-based disulfide fragmentation,⁷¹⁹ electron-transfer higher energy dissociation (EThcD)-based disulfide fragmentation,⁷²⁰ UVPD-based disulfide fragmentation,^721-722 on-line electrolytic cleavage,⁷²³ and ion mobility⁷²⁴ were applied to characterize disulfide bonds, they either require advanced instrumentation or suffer from disulfide scrambling. Researchers also developed novel algorithms that identify disulfide bonds by examining MS/MS fragmentation patterns of disulfide-linked di-peptides.^725-726 Its application in unknown systems, however, is limited because the fragmentations are poor and the enrichment is challenging, problems that also confront cross-linking. Although classical, disulfide bond identification by targeted Cys labeling may be an effective and reliable approach, with potential for mapping disulfide bonds. Therefore, to illustrate opportunities in this area, we will review a few examples that are either pioneering studies or show novel workflows

The potential of MS in disulfide mapping was first demonstrated by Morris and Pucci⁷²⁷ in 1985 remarkably with FAB-MS, where they identified disulfide-linked dipeptides by aligning the peptide masses with the protein sequence. Some of the early FAB-MS based efforts were reviewed by Smith and Zhou⁷²⁸ in 1990. These methods evolved into quantitative assessments of free Cys and disulfide bonded by MALDI-MS, where a “count” of the free Cys residues in a protein can be made by analyzing the mass shift upon modifying the free Cys residues.^729-730

To locate disulfide bonds is challenging and cannot be done universally with MS/MS because fragmentation is not always cooperative. In a pioneering study in 1996, Watson and coworkers⁷³¹ utilized 2-nitro-5-thiocyanobenzoic acid (NTCB) to footprint free Cys and cystine. NTCB is a cyanylation reagent developed by Vanaman and coworkers⁷³² in 1973, and it labels free Cys by replacing the hydrogen of the thiol with a cyanide. The cyanylated Cys undergoes backbone cleavage under alkaline pH (Scheme 13). The method is utilized by choosing specific peptide cleavages, guided by location of free Cys residues and determining the location of free Cys residues by analyzing the resulting peptides, here by MALDI-MS. Its application in disulfide mapping was by a differential approach, where the protein is footprinted by NTCB under both native and reduced conditions (TCEP was used for disulfide bond reduction). The Cys residues that are not cleaved under native condition but cleaved under reducing condition can be assigned to disulfide bonds.

Scheme 13.

Free Cys footprinting by NTCB-mediated cleavage

Modern disulfide-bond footprinting makes use of the silence of Cys residues towards labeling reagents when they are covalently bound to form a disulfide bond. To distinguish disulfide-bonded and solvent-inaccessible Cys requires careful design and suitable control experiments. In one of the early applications in 2000, Glocker and coworkers⁷³³ developed a general workflow to map disulfide bonds with targeted labeling reagents. They applied their approach to heat shock protein Hsp33, a redox-regulated molecular chaperone with six Cys residues. In its reduced state, all six Cys are in the thiol form, and the protein is not biologically active. Exposing the protein to oxidizing conditions promotes conversion of Hsp33 into its active form, during which the protein folds, assisted by formation of disulfide bonds. Upon IAM footprinting followed by MS analysis, all six Cys in reduced Hsp33 can be alkylated. In the oxidized state, however, Cys 232, 234, 265 and 268 are silent in IAM footprinting. Peptide-level MS analysis suggests that two disulfide bonds are Cys232-S-S-Cys234 and Cys265-S-S-Cys268. Cys141 was found to be highly reactive in both protein states, indicating that it is not involved in disulfide formation. In contrast, the modification extent of Cys 239 decreased significantly but did not disappear for the oxidized state, suggesting that Cys 239 is not involved disulfide bonding but rather part of a significant conformational change, burying it upon Hsp33 activation.

In a 2009 study, Chumsae et al.³⁰⁹ incorporated Cys fluorescence labeling with MS-based peptide mapping to study the unpaired Cys residues in human Immunoglobulin G subclass 1 antibody (IgG1). Although all 32 Cys in IgG1 can be involved in disulfide bonds, the presence of small amounts of free Cys is a common feature in recombinant and wildtype IgG1 antibodies. To differentiate the free and disulfide-bonded Cys residues, the IgG1 was first labeled with 5-idoacetamidofluorescein (5-IAF), a fluorescent reagent that reacts with free Cys residues. Subsequently, the 5-IAF labeled IgG1 was submitted to disulfide bond reduction and iodoacetic acid labeling. As a result, the free Cys residues are labeled with a fluorescence probe whereas the cystine residues are labeled with iodoacetic acid. Upon digestion, the peptides were submitted to LC separation, fluorescence detection, and MS analysis. As the free Cys residues are less abundant in IgG1, the fluorescence labeling improves the detection of free Cys-containing peptides. This study also is a good example of an integrated approach that addresses an important biological question (especially in pharma and biotechnology).

One year later in 2010, Weerapana and coworkers⁷³⁴ developed an isotopically-encoded IAM derivative with an attached phenyl ring. The carbons of the phenyl ring can be either ¹²C, the light form, or ¹³C, the heavy form, to allow isotope encoding. A subject protein was first labeled to give the light form, tagging the free Cys resides. TCEP was then added to reduce the disulfide bonds, followed by a second stage of labeling where any Cys formerly in a disulfide bond is are now labeled by the heavy form. Although originally demonstrated for quantitative profiling of Cys reactivities in primary structure proteomics,^734-735 the idea of isotopic-encoded reagents is readily applicable in footprinting disulfide bonds. It avoids the problem of using two different reagents (one to label free Cys and another to label the disulfide-bonded Cys after reduction). The reactivities of two different reagents are seldom the same, leading to ambiguities in quantification. Isotopic-encoded alkylating reagents nicely overcome this drawback and add precision and efficiency.

More examples of targeted Cys footprinting in combination with MS peptide mapping with discussions of experimental protocols and issues of disulfide scrambling can be found elsewhere.^{169, 736-737}

4.3. Characterizing Protein-Ligand and Protein-Protein Binding Interfaces

Among all biological questions that MS-based protein HOS elucidation can address, locating protein-protein binding interfaces is the most widely asked. Upon interacting with a binding partner, the solvent accessibility of the binding interface usually decreases, leading to a decreased footprinting. Here we present three examples of targeted protein footprinting where improved methodology was developed. Each example shows unique features from an analytical perspective.

During the past decade, carboxyl group footprinting by GEE has been adopted in characterizing the structure of antibodies as well as their interaction with antigens and drugs.^738-743 In a particular example, Li et al.⁷³⁸ applied orthogonal MS-based footprinting approaches to characterize the interaction between the extracellular region of human interleukin-6 receptor α-chain (IL-6R) and two types of adnectin (adnectin 1 and 2) that bind with picomolar and nanomolar affinity, respectively. Different from labeling in many other studies, GEE footprinting was executed in a time-dependent manner (Figure 6a-c) as commonly seen in HDX studies. Region 135-138 becomes protected upon ligand binding (Figure 6a), and the differences are made more apparent by obtaining kinetic curves rather than relying on single time points. Subsequent MS/MS analysis indicates the GEE labeling is exclusively on residue Glu 140.

Figure 6.

GEE labeling kinetics for selected IL-6R peptides in the ligand-free (gridded circle), adnectin 1-bound (triangle) and adnectin 2-bound state (diamond) state. (a) Region 135–148 shows decreased GEE incorporation upon adnectin 1/adnectin2 binding, whereas (b) region 274–284 shows increased GEE modification upon adnectin1 binding. (c) Representative peptide region without differentiable GEE modification extent between bound and unbound as a control. Dashed trend curves in (a), (b), and (c) are generated by linear or 2^nd-degree polynomial fitting. Figure adopted with permission from Ref. ⁷³⁸. Copyright 2017 American Chemistry Society.

In comparison to region 135-148, region 274-284 becomes deprotected upon binding with adnectin 1 but does not change significantly when bound to adnectin 2 (Figure 6b). Region 27-41 was not involved in binding as the GEE labeling extents remain comparable with and without IL-6R binding. The investigators concluded that the adnectin 1 and 2 binding region is a flexible loop between two β-strands in the cytokine-binding domain of IL-6R, which was verified by other MS-based footprinting in this study. Besides demonstrating the advantages of applying orthogonal methods in characterizing a single protein system, the protocol executed the targeted labeling in a kinetic way, significantly enhancing the confidence in the data interpretation as compared with the commonly used single time point labeling.

In examples in 2008 and 2011, Kiessling and coworkers^744-745 developed isotope-coded affinity tag (ICAT) coupled targeted labeling reagent for protein footprinting. The bromoacetamide ICAT reagent, often used to alkylate free Cys residues, was repurposed for footprinting (Figure 7a). The idea of ICAT originates from proteomics,^746-747 where isotopic encoding enables absolute quantification, and the affinity tag facilitates post-labeling enrichment. Here the goal is to footprint the binding interface of adaptor protein CheW and the multidomain histidine kinase CheA (and its Cys-depleted form CheA*) in the Escherichia coli (E. Coli) chemotaxis signaling system. As protein-protein interactions in signaling systems are usually transient and low in affinities, it is necessary to add excess binding partners to the system to shift the binding equilibrium to the bound state. When the binding partner is a large protein, as exemplified by CheA, excess protein will create an intense protein/peptide background and potentially overwhelm the MS detection. The ICAT footprinting reagent enables post-labeling enrichment, making it possible to label the CheW under high CheA concentration without concern for the high CheA background during MS analysis.

Figure 7.

Overview of Cys footprinting with an ICAT reagent. (a) Structure of bromoacetamide ICAT reagent. Dots represent ¹²C for light reagent and ¹³C for heavy reagent. (b) Workflow for alkylation rate determination by ICAT isotope pairs. (c) – (e) Representative alkylation time courses of CheW Cys variants footprinted in the presence (red) or absence (black) of CheA*. Figure adopted with permission from Ref. ⁷⁴⁴. Copyright 2011 Elsevier.

The isotopic encoding enables accurate quantification. As depicted in Figure 7b, the complex is first footprinted by the heavy reagent, where all the solvent accessible Cys residues can be labeled. After a certain labeling time, the protein complex is denatured, and the light-reagent is added, during which those Cys residues that were originally protected can now be labeled by the light reagent. The sample is then digested and submitted to affinity enrichment. In this study, the Cys footprinting by heavy reagent was executed under multiple time points (t_Alk), and the relative fraction of a specific Cys residue under solvent-accessible and protected states was quantified by comparing the signal intensities from a single injection of the heavy and light-labeled species. Alkylation rates of specific Cys residues can be obtained by plotting the fraction of labeling by the light reagent as a function of t_Alk, where a more abundant light-labeled fraction indicates a decrease in solvent accessibility upon binding with CheA.

In the experiment, CheA* is used instead of native CheA to minimize the effect of free Cys from CheA on the alkylation kinetics of CheW Cys residues. As there is no Cys in the CheW sequence, several mutants were expressed and purified, each with a single-site Cys mutation to facilitate the alkylation footprinting. Cys residues in CheW that show decreased alkylation rates upon mixing with CheA* are the binding sites, as exemplified in Figure 7d. Among nine mutants that were analyzed, four showed significant decreases in alkylation rates upon binding with CheA* (46, 48, 59, 60), indicating that these four positions of CheW are likely involved in CheA binding. The enrichment by ICAT not only offers a better quantification, but also enables the footprinting of transient and weak-binding partners. Taken together with the two earlier examples by Ramakrishnan et al.⁷¹⁰ and Li et al.⁷³⁸, footprinting of targeted residues that were executed at multiple labeling times will significantly enhance the confidence of the data interpretation.

In a third study, Vachet and coworkers^{546, 748} studied the pre-amyloid dimer and tetramer of β-2-microglobulin (β2m) by footprinting with different reagents. This study nicely demonstrates a canonical workflow and data presentation of protein HOS analysis by multiple targeted labeling reagents, where the focus is the differences between ligand-bound and ligand-unbound states, and the results are mapped onto existing structural models or used to construct a new model based on the footprinting results.

Like many other amyloid-forming proteins, β2m self-associates into fibrillar amyloid in the presence of Cu(II) and can be deposited in the musculoskeletal system. Because targeted labeling reagents are usually reactive with specific amino acid residues, a strategy of using several labeling reagents is effective to increase the amino acid coverage. By combining sulfo-N-hydroxysuccinimide acetate, DEPC, and 2,3-butanedione labeling, the investigators interrogated the N-terminus and Asn, Arg, His, Lys, Ser, Thr and Tyr residues in β2m. Comparison between the footprinting of β2m in monomeric and dimeric states suggests that the dimer interface is formed by antiparallel stacking of ABED β-sheets by two β2m monomers. Further, in-solution footprinting data rule out the interaction of D-D strands in the dimer, which is not consistent with a previously resolved X-ray crystal structure, emphasizing the need to use complementary HOS structural characterizations. In a subsequent study with similar experimental procedures, the interface of β2m tetramer was identified as D strands from one dimer unit and G strands from another dimer unit.⁷⁴⁸

Because β2m tends to aggregate, the same group also studied the binding sites of its amyloid inhibitors (e.g., rifamycin SV, structure depicted in Figure 8a), and the results are in Figure 8.⁷⁴⁹ The changes in modification fractions upon binding with rifamycin SV (Figure 8a) show that four residues become protected (K6, K91, K94 and R97), whereas the N-terminus becomes more exposed. These results were mapped onto the β2m surface structure (Figure 8c) and compared with protein-ligand docking. The comparison strongly suggests a binding pocket between Lys75 and Arg 97 (Figure 8d). Considering all the results, the authors concluded that rifamycin SV likely binds near Lys 94 and Arg 97 and perhaps near Lys 91, which is along the G-strand of β2m (Figure 8b). The mismatch between docking and footprinting is likely due to the structural perturbation of β2m induced by rifamycin SV; however, the similarity is encouraging. Two other ligands were also evaluated in a similar fashion.

Figure 8.

Covalent labeling results of β2m binding with rifamycin SV. (a) Changes in labeling modification percentages with rifamycin SV bound to the Cu(II)–protein complex. BD is 2,3-butanedione. Error bars represent standard deviations from triplicate measurements. Asterisks above the bars represent the residues that undergo a significant change in modification level at 95% confidence as determined by a two-sample unpaired Student’s t test. The arrows at the top of the graph indicate the locations and directions of the seven β strands in β2m. (b) Ribbon structure of β2m, showing the seven β strands labeled A through G. (c) β2m surface structure illustrating the sites undergoing significant changes in covalent labeling induced by rifamycin SV. Sites that increase in labeling are colored red, whereas those that decrease in labeling are colored blue. (d) Protein–ligand docking results for verification. Figure adopted with permission from Ref. ⁷⁴⁹. Copyright 2017 American Chemistry Society.

Lastly, targeted labeling reagents were also adopted to footprint protein-nanomaterial interfaces to provide critical insights in the toxicity and potential biomedical applications of nanomaterials.^750-752 Determining protein-protein and related interfaces with fast labeling reagents is also attracting considerable attention, and this subject will be covered in section 5 and 6.

4.4. Assessing Topology and Stoichiometry through Chemical Cross-linking

Chemical cross-linking has developed as a complementary area of research owing to its capability to probe HOS and to locate and define protein/protein interfaces. Our intention, however, is not to review this topic comprehensively, but to describe the workflow, show how XL can be viewed as a means of footprinting (i.e., “double footprinting”), discuss its role in integrated approaches, and highlight some recent developments to unify the methodology described in this work.

Cross-linking mass spectrometry (XL-MS) is effective at assessing protein complex topologies and elucidating protein structures besides capturing protein-protein interactions. Conceptually, XL utilizes bifunctional labeling reagents to cross-link (footprint) the constituent proteins forming an interface, thus providing information on the interacting species and their interfaces. When dealing with large proteins, XL can also report on the overall protein conformation.

To date, most of the MS-based XL studies have been by a bottom-up approach, whose workflow is presented in Figure 9. Briefly, proteins are first incubated with chosen cross-linkers under the optimized conditions that are established usually by monitoring with techniques simpler than MS (e.g., gel electrophoresis). The tethered proteins are then submitted to enzymatic digestion, enrichment of informative peptides, and LC-MS/MS for separation and detection. The MS data are further analyzed by using search engines to identify cross-linked peptides with an uncertainty specified by mass tolerance and false discovery rate (FDR). Significant advances since 2008 in data analysis have facilitated this approach; new software includes but is not limited to pLink,⁷⁵³ xQuest,⁷⁵⁴ XlinkX⁷⁵⁵ and StavroX⁷⁵⁶. Generated cross-link maps not only identify the connectivity of the adjacent protein subunits but also provide distance restraints as given roughly by the molecular separation between the two reactive functional groups of the cross-linker.

Figure 9.

Workflow of bottom-up cross-linking mass spectrometry

XL-MS alone is a middle-to-low spatial resolution approach because usually there are limited number of reactive residues at or near an interface of two proteins, and few crosslinks form. Young et al.⁷⁵⁷ in 2000 showed that by combining XL distance restraints and computational modeling is a compelling way to improve the resolution for elucidating protein structures. The workflows can be adapted for integrative modeling^758-759 and de novo structure prediction^760-761. The major challenges of computational studies are the sampling and scoring of the generated models⁷⁶². Larger data sets derived from XL-MS effectively decrease the size of the sampling space and benefit the scoring function, thus promoting efforts to find high-confidence models. Given there can be various reactive groups and spacer lengths on cross-linkers, a strategy involving multiple reagents should be taken to insure comprehensive information.

In 2014, Chait and coworkers⁷⁵⁹ demonstrated the use of DSS and a zero-length EDC in characterization of the nuclear pore sub complex of Nup84. Two data sets generated from use of two reagents delivered different but complementary structural information that significantly benefited the subsequent model construction. Many other studies adopted similar ideas by combining amine-targeting reagents and carboxyl-targeting reagents (e.g., dihydrazides^{669, 763}) or nonselective cross-linkers^764-765. In addition, the use of reagent combinations with different spacer lengths is recommended because short cross-linkers yield fewer cross-links but with narrower distance restraints, whereas long cross-linkers afford more cross-links but less structural definition because the distance assignments are over a broader range.⁷⁶²

New sample preparation methods also have developed rapidly (e.g., on-bead cross-linking⁷⁶¹), improving the cross-linking chemistry and providing better analysis sensitivity. Many examples are discussed in recent reviews.^{658, 766-767}

By way of contrast, footprinting by targeted and free-radical reagents and/or by HDX maps solvent accessible regions and reflects differences in protein dynamics or binding events. This information provides deeper understanding of the entire protein structure, not just interfaces, and can adjudicate constructed 3D models from XL for in-solution proten.^768-770

XL is not limited to structural characterization of single protein complex, but rather it can promote protein-protein interactions (PPIs) studies in proteome-wide investigations.^771-774 To meet the need for understanding the roles, functionalities and mechanistic behavior of many protein complexes through their PPIs maps, cleavable XL reagents have been developed,⁷⁷⁵ simplifying and increasing the accuracy for identification of cross-linked peptides. Incorporation of affinity groups and sophisticated enrichment procedure have led to success; examples are studies performed on E. coli⁷⁷³ and Caenorhabditis elegans (C. elegans)⁷⁷⁴. For example, the Heck group⁷⁷⁶ reported a XL-MS study of whole human cell lysates in which they identified 2179 unique cross-links. These protein complexes (e.g., 80S ribosomal core complex) reveal novel interactions and provide new structural insights.

To summarize, the increasing number of publications based on XL-MS continue to demonstrate its utility in MS-based biophysics, particularly in combination of other techniques. Integrating XL with other methods enables characterization of dynamic biological systems even of heterogeneous systems. The continuing development of cross-linkers will increase applications. Reagents that react more rapidly and target more and more amino acids, and instrumentation advances that provide new fragmentation methods in MS/MS (e.g., CID, ETD, EThCD and UVPD), are expected for the future. Furthermore, improved separation and enrichment procedures will accommodate the increasing mixture complexity following XL and digestion and allow detection of low abundant yet informative cross-linked species.

4.5. Footprinting Fast Kinetic Processes

Given the high reactivity of the Cys thiol group, specific labeling reagents were also developed to follow protein folding in addition to footprinting a static state.⁷⁷⁷ An early application addressed slow events in the folding of recombinant human macrophage-colony stimulating-factor β, which takes tens of hours to complete.⁷⁷⁸ Melarsen oxide (p-(4,6-diamino-1,3,5-triazin-2-yl)aminophenylarsonous acid) used in this study footprints Cys in 10 min, which is sufficiently fast for this slow folding system but far too slow to footprint most protein folding.

To shorten the footprinting time scale and make Cys footprinting capable of monitoring protein folding pathways in general, other reagents, methyl methanethiosulfonate (MMTS)⁷⁷⁹ and DTNB,³¹⁶ were applied because they react with Cys on the time scale of millisecs. Furthermore, the steric sizes of these reagents can provide insight on the structural changes during conformational opening/closing.⁷⁸⁰ These reagents were successfully applied in a pulse-labeling approach to follow unfolding of apo-myoglobin.^781-782

In an elegant demonstration of specific amino acid footprinting, Jha and Udgaonkar⁷⁸³ in 2007 applied pulse Cys labeling by MMTS to study the fast folding reaction of barstar. Tracking the labeling of ten Cys residues as a function of refolding times showed that four remain solvent-accessible throughout the folding. The other six Cys residues become protected upon folding, and their rates of reaction vary by 3-fold and are dependent on their locations in the protein. The investigators showed the general workflow of MS-based protein folding analysis and demonstrated footprinting with a single residue can provide sound information. Unfortunately, Cys is the only residue that can be labeled on the millisecond time scale, probably because both the -SH group and the reagent are reactive. A workaround may be to express the protein with more Cys residues located strategically to inform on the roles of various regions. The downside is that the proteins may respond to these substitutions to adopt a non-native state. Thus, a general approach to study folding by specific amino acid labeling will be difficult to achieve. Moreover, labeling times are still too long for probing short lived intermediates and fast folding. Fast free radical labeling reagents can overcome these drawbacks, as will be discussed in section 6.2.

In an unusual approach to fast reactions, McLuckey and workers^784-787 starting in 2009 demonstrated the covalent labeling of peptides in the gas phase. They illustrated the reactions between peptides and 4-formyl-1,3-benzenedisulfonic acid (reacts with primary amines),⁷⁸⁴ NHS-ester (reacts with primary amine and carboxylates),^785-786 Woodward’s Reagent K (reacts with carboxylic acids)⁷⁸⁷. Although not yet applied in gas-phase protein footprinting or folding, this approach is potentially advantageous because gas-phase ion reactions are rapid, sometimes even collision controlled. A disadvantage is that reactions in the gas phase may not be relevant to in intro and in vivo biochemistry.

4.6. Footprinting in vivo

It can be questioned whether protein biophysical properties and their HOS obtained in vitro are truly reflective of those in vivo as the native environment in which a protein functions are challenging to reproduce outside a living cell. Thus, protein HOS analysis in vivo is a long-sought goal of protein biophysics. Macromolecular crowding within the cell limits the diffusion of the protein molecules, and that further affects the biophysical properties of proteins including their interactions with nucleic acids and their reaction kinetics and association rates with their binding partners.^788-789 Protein footprinting in vivo may be the solution. When one recognizes that the expanding capabilities of MS-based proteomics analysis permit deeper and deeper analysis of an entire proteome, the goal of in vivo footprinting comes closer to realization.

Early developments of protein in vivo labeling primarily focus on the localization and quantification of proteins in the cellular environment. For localization purposes, a common practice is to attach fluorophores covalently to the target proteins.^790-791 The fluorophores can be either small organic molecules⁷⁹² or proteins⁷⁹³, which then illuminate under a fluorescence probe. The labeling can be non-covalent and mediated by an engineered ligand that has a fluorophore attached.^794-795 Here the labeling efficiencies are generally low but relatively specific. Recent demonstrations couple FRET with protein in vivo labeling (semisynthetic protein switches) to allow characterization of PPIs and protein conformational changes.⁷⁹⁶ Despite success in protein engineering and controlling the functions of proteins, these approaches have low structural resolution and sometimes low sensitivity that limit applications in protein HOS analysis.⁷⁹⁰

Protein labeling in vivo coupled with MS are used for quantification in proteomics and metabolomics. These approaches almost always involve isotope encoding for increasing accuracy and speed. Although these methods are used more and more, their goal is not structural information. For this reason, and because these approaches were reviewed elsewhere,^797-799 they will not be discussed further.

High resolution protein HOS analysis in vivo is challenging. Among all three high-resolution approaches, NMR has the richest history of characterizing proteins in vivo.⁸⁰⁰ Because a protein’s X-ray structure cannot be acquired under native conditions, that approach holds little promise. Macromolecular crowding makes it almost impossible to obtain cryo-EM images for a specific protein of interest under in-cell conditions.⁸⁰¹ Protein footprinting in vivo was first done by cross-linking, which can elucidate topological features of protein and protein complexes and capture PPIs to reveal their biological functions. Implementation of cross-linking chemistry in the native environments are challenged by the heterogeneity and complexity of the cross-linked products.^802-803 The Bruce lab⁸⁰⁴ in pioneering efforts starting in 2009 developed a series of MS-cleavable cross-linkers (i.e., protein interaction reporter (PIR)) that undergo characteristic fragmentation during CID, thus providing more precise cross-links identification. To optimize the acquisition of fragments for MS3 sequencing, the investigators developed a real-time analysis for cross-linked peptides technology (ReACT).^{662, 805} In addition to biotin-based enrichment, ReACT has been successfully applied to XL in many organisms (e.g., Shewanella oneidensis,⁸⁰⁴ E. coli,⁸⁰⁶ Pseudomonas aeruginosa,⁸⁰⁷ Acinetobacter baumannii,⁸⁰⁸ and human cells⁸⁰⁹). Other cleavable cross-linkers are also available (e.g., bis(succinimidyl)-3-azidomethylglutarate (BAMG)⁸¹⁰ and azide-A-DSBSO⁸¹¹).

Recently, Liu et al.⁸¹² characterized the interactome of native mouse-heart mitochondria with DSSO and identified 3322 unique physical contacts, the largest survey to date of mitochondrial proteins interactions. In vivo cross-linking experiments reveal localization of many subunits and provide evidence for coexisting respiratory super-complex assembly.

In other demonstrations, Blankenship and coworkers^813-814 combined in vivo XL-MS with time-resolved spectroscopy and characterized a fully functional megacomplex composed of a phycobilisome antenna complex and photosystems I and II from the cyanobacterium Synechocystis PCC 6803. More applications and progress have been reviewed lately.^{803, 815} Although XL-MS has shown its effectiveness, there are still many challenges raised by the complexity of biological systems. More and better methodologies in the sample preparation, enrichment and MS detection will allow more comprehensive characterization of interactomes, providing another dimension of information to understand the molecular machinery that impacts living systems.

Another approach to in cell footprinting is illustrated by the investigation of the membrane protein human vitamin K epoxide reductase (hVKOR) by Li, Gross, and coworkers.⁸¹⁶ Membrane proteins are appropriate targets for in vivo footprinting, because retaining their native structures in vitro can be challenging, especially for complexes with multiple subunits.⁸¹⁷ Although native MS allows probing non-covalent interactions associated with membrane proteins, structural information is hard to obtain with this approach.^{151, 818-819} In the study of hVKOR, Li and coworkers⁸¹⁶ targeted an intramolecular disulfide bond (Cys51-Cys132) that stabilizes a key intermediate redox state, a state that can perturbed by binding with warfarin, a highly prescribed drug to prevent blood clotting. In brief, h₅-NEM was added to penetrate live cells and footprint free Cys in the native environment. After lysing the cells, the remaining disulfide bonds were reduced and labeled with d₅-NEM to footprint the newly reduced Cys (Figure 10a) in a methodology discussed in Section 4.2. The sample was submitted to LC-MS and MS/MS for identification (Figure 10c-d). The partition of a specific Cys residue in its free and disulfide-bounded states can be obtained by quantifying the ratio of its h₅-NEM and d₅-NEM labeled species (peak intensities in the MS spectra), defined as “apparent oxidized fraction” (Figure 10b).

Figure 10.

Schematic illustration of in vivo Cys footprinting workflow. (a) Scheme of MS analysis. The redox status of cysteines was quantified by differential isotope footprinting with NEM before and after a reduction step (right). Intact disulfide bonds were detected under nonreducing conditions (bottom). (b) Intracellular cysteine status was determined by quantitative MS before (red) and after DTT reduction (green). Bar graphs show means and errors of multiple peptides from three independent experiments **P < 0.001; ***P < 0.0001; #P > 0.05 by two-tailed Student’s t test. (c) Extracted-ion chromatograms showing a Cys51-Cys132 disulfide-linked peptide (top), which separates into two individual peptides (bottom) after reduction of the disulfide. A reference peptide is used as the standard (Std) for peak comparison. (d) MS/MS spectrum of the Cys51-Cys132 linked peptide (from c). Adopted with permission from Ref. ⁸¹⁶. Copyright 2017 Springer Nature.

To facilitate a differential footprinting to illustrate whether a specific Cys residue prefers the free or disulfide-bounded state in the native cellular environment, the experiments were executed in two different conditions, illustrated in Figure 10b as “native” and “DTT”. The “native” state denotes the footprinting under native cellular environment as mentioned above. The “DTT” state, however, was executed under reducing conditions. Upon admitting DTT to the cell, the disulfide-bound Cys residues under native state were reduced and footprinted by h₅-NEM. As a result, apparent oxidized fractions for the disulfide-bounded Cys residues decreases under “DTT” state as compared with “native” state, as seen for Cys43, 51, 132 and 135. The remaining three Cys residues (Cys16, 85 and 96) do not show decreased apparent oxidized fractions, indicating that these three Cys residues are in reduced state (free Cys) under native cellular environment. By such approach, the investigators were able to elucidate the native architecture of hVKOR, reveal the electron-transfer pathway, and propose an hVKOR catalysis inhibition mechanism by the ligand warfarin. A subsequent paper summarizes the precautions, technical difficulties, and advantages of in vivo membrane protein footprinting.⁸²⁰

In another demonstration, Li, Gross, and coworkers⁸²¹ applied in vivo carboxyl group footprinting to probe formation and breakage of the salt bridges that are hypothesized to transport carbohydrates of human glucose transporters (GLUTs). By footprinting with isotope-encoded reagent GEE, they identified key salt bridges in GLUT1 on both its intracellular and extracellular sides. Subsequent mutation of the residues involved in salt bridges lead to a significant substrate uptake deficiency of the cell, providing confirmation that salt bridges serve as molecular switches. The investigators conducted footprinting under different substrate conditions to reveal the gating mechanism of the transporter as well.

MS-based in vivo footprinting, as a probe of protein solvent accessibility, offers structural evidence that can be directly related to functional states. A challenge is the complexity introduced by the high crowding of intracellular components, requiring use of large quantities of reagent that must be admitted to the cell. The dynamic range of cell proteins is high, and following low-abundance proteins also offers challenges. Nevertheless, the success in answering focused questions in a few examples demonstrate that protein footprinting in vivo by targeted labeling reagents has significant promise. Like in vivo cross-linking, better strategies in post-labeling sample enrichment will enable broader adoption of this method.

4.7. Integrating Footprinting with other Approaches

Each of the methods reviewed here has limitations that may be overcome by combining with other methods. Integrating footprinting with other approaches is most seen with XL, owing to the limited structural resolution of XL but also its distance restraints that can be readily used as inputs in modeling.

To construct better a 3D model to near-atomic resolution, the XL platform can be coupled with electron microscopy^822-823 and X-ray crystallography.^{759-760, 824} Electron density maps often show areas that diffract poorly because they are flexible or heterogeneous. To localize each subunit accurately, restrained distances can be contributed by XL-MS. One compelling example is reported by Greber et al.⁸²⁵, who determined the architecture of the porcine 39S large proteome in the mammalian mitochondrial ribosome. A previous cryo-EM structure was resolved at 4.9 Å. Many of the protein extensions (a sequence change extending the amino acid sequence at the N- or C-terminal end with one or more amino acids) that do not have homology models cannot be refined at this resolution. In a combined approach, distance restraints from XL can be used in refining structure considerations, leading to a higher spatially resolved model (at 3.4 Å). The better-described protein assembly now reveals many interaction “hotspots” (e.g., the active site of peptidyl transferase and the path of its idiosyncratic exit tunnel). Integration of native MS and XL-MS can lead to remarkable structure elucidations^826-829 of macromolecular assemblies that are usually difficult to purify and crystalize by traditional methods. Native MS can determine topology and stoichiometry of protein complexes. The approach is even more effective when combined with ion mobility MS, although direct determination of the critical binding residues is nearly impossible. On the other hand, XL-MS reveals the topology and connectivity of interacting regions by using solution chemistry and gas-phase analysis (via MS). The two strategies complement each other by providing architectural information of multiprotein assemblies.

Because cross-linking occurs via strong, covalent bonding, integrating it with other approaches may allow detection of transient protein-protein interactions that are hard to investigate otherwise. For example, the yeast initiation factor 3 (eIF3) protein, consisting of six subunits, regulates translational processes as a scaffold for many initiation factors. The molecular architecture, however, is elusive owing to its dynamic nature. Politis et al.⁸³⁰ in 2015 determined the composition of the complex by sequentially dissociating subcomplexes, monitoring them by native MS, and coupling that information with physical contacts and approximate distances taken from cross-linking. The integrated information was encoded as connectivity restraints for scoring purposes in the model generation. The outcome is a high-resolution, 3D topological model of eIF3 assembly in complex with eIF5. To preserve the transient and dynamic interactions, a two-step cross-linking protocol was utilized, in which formaldehyde fixation is implemented at sub-stoichiometry levels prior to initiating the cross-linking chemistry.⁸³¹ Structural stabilization of this type greatly facilitates cross-link formation and can be reversed at high temperature before MS identification.

Other MS-based footprinting methods have also been used in conjunction with XL-MS (e.g., footprinting by targeted labeling reagents⁸³² and HDX^{768, 833-835}). In one example, Zhang et al.⁷⁶⁸ evaluated the combination of HDX, XL-MS, and molecular docking for characterization of the binding interface of interleukin 7 and its α-receptor. HDX reports widespread protection in the receptor. There is, however, no differential evidence of binding-induced protection or remote conformational change. By employing different reagents in XL, the investigators significantly increased the spatial resolution of binding site assignment. To generalize the integrated approach, protein-protein docking was executed with different number of crosslinks as distance restraints. With model clustering and HDX adjudication, a high confidence model was obtained with root mean square deviation (RMSD) below 2.0 Å (with respect to the known crystal structure) with only two crosslinks.

Integrated methods can now play a role in protein structural modeling. By utilizing the structural restraints from DEPC labeling, cross-linking, and native MS for protein structural modeling, Politis and coworkers⁸³² in 2017 successfully constructed structural models of several protein complexes, tryptophan synthase, carbamoyl phosphate synthetase, and the double heterohexameric ring RvB1/2 with high-resolution subunit structures. A comparison of a representative model for tetrameric tryptophan synthase derived from all available structural restraints and its corresponding crystal structure (Figure 11) show agreement with a RMSD of 9.6 Å. Systematic evaluation of the modeling show that restraints from DEPC labeling and chemical cross-linking markedly increase the predictive power.

Figure 11.

Representative model of the tetrameric tryptophan synthase and its corresponding crystal structure. Inter-residue proximities (XL-MS) and residue solvent accessibilities (covalent labeling MS, CL-MS) are highlighted. Figure adopted with permission from Ref. ⁸³². Copyright 2017 American Chemistry Society.

The workflow was tested by employing three model systems and then applied to a study of the F-type ATP synthase from spinach chloroplasts, a protein complex without a high-resolution structure. By integrating these experimental measurements with molecular dynamic simulations, the conformational states of the peripheral stalk as well as the flexible regions within the complex were revealed. This work demonstrates the potential of MS-based footprinting data to aid computer modeling.

In other work in 2019, Aprahamian and Lindert⁸³⁶ describe in detail a Rosetta script that utilizes DEPC footprinting data to provide better restraints in protein structure predictions. Restraints from cross-linking can direct the modeling of protein complexes with multiple subunits; however, the short supply of modifiable residues for DEPC labeling constrained the SASA information. Fast labeling reagents (radicals) generally modify multiple residues in a single footprinting experiment that overcomes this drawback. More details about computer-aided protein structural modeling and radical footprinting-aided protein structural predictions can be found in section 6.6.

4.8. Conclusion

To conclude, protein footprinting by targeted labeling reagents has been utilized to address a number of biological questions and is considered as a powerful tool for MS-based protein HOS analysis. As mentioned at the beginning of the section, two major disadvantages limit its broad application. The limited targeting residues can be overcome by developing and adopting novel labeling reagents. Fast radical species can also complement such limitation, as discussed in section 5 and 6.

Another question is whether the targeted protein undergoes a conformational change in the early stages of labeling and then continues to be modified to yield a composite and misleading footprint. In a study of interferon-β1a, Kaltashov and coworkers⁸³⁷ demonstrated that alkylation of Cys17 induces steric clashes within the native structure, as seen by the unfolding of helix D containing a 88-102 segment. On the contrary, there are also numerous studies to support the validity of targeted protein footprinting, either by a combination of multiple footprinting approaches^{738, 838-839} or by comparing the targeted footprinting results with crystal structures^{169, 540, 840}. In a particular example, Li et al.⁷³⁸ used HDX, GEE footprinting and hydroxyl radical footprinting to study the binding region of IL-6R to adnectin 1 and 2. All three footprinting approaches suggest the same binding region of 135-141. In other words, the validity of the slow but targeted GEE labeling is supported by two other protein footprinting approaches that functions at different time scales (20 s to 10 min for HDX, sub-millisecond for hydroxyl radical footprinting).

The contradicting judgements on the effectiveness and validity of protein footprinting by targeted reagents are system dependent. In selected systems, modification of critical residues will induce catastrophic protein conformational changes. It is thus important to characterize a protein system with multiple footprinting approaches. Different labeling time scales of these approaches can also be utilized to provide complementary and synergistic understanding of a protein’s HOS.⁸³⁹

5. Fast Labeling Reagents – Reactive Radical Species

Thus far, we reviewed two means of footprinting: HDX and specific amino acid labeling. Both involve covalent labeling, but HDX footprinting is reversible, imposing constraints on protein and peptide analysis to minimize back exchange. HDX is less disruptive of protein structure than specific amino acid labeling because it occurs in medium in which H₂O has been replaced by D₂O. The concentration of the “reagent” D₂O is 50 M, a concentration that changes negligibly over the time of footprinting. Exchange occurs by pseudo first order kinetics on the microsecond to many hours time scale. The coverage is nearly complete, including all the common amino acids except Pro, and the modifications that are counted occur on the peptide backbone.

In contrast, specific amino acid labeling has low coverage, modifying one or a few amino acid side chains. Some reagents that modify -COO⁻ or -NH₃⁺ groups change the local charging of the protein; other reagents add bulky steric hinderance to the protein (even as large as another protein), both modifications posing a danger of structure disruption. Although a few footprinting reactions occur on the millisecond time scale, most require minutes or more during which the modified protein can change conformation and continue to react and produce a footprint that is composite. The concentrations of reagents are 10² to 10⁴ times greater than that of the protein, unlike HDX where the concentration of D₂O is approximately 10⁷ times greater than that of the protein. Considering that a large protein could have 10³ potentially reactive amino acid residues, the reagent concentration may change, causing the reactions to have mixed kinetics.

The third class of footprinting reagents, fast labeling regents, enjoys features of HDX and specific amino acid labeling but has some important advantage over both. The modifying reagents are usually radical species that react with solvent-accessible amino acid residues to report on protein SASA. This form of footprinting shares characteristics with both HDX and specific amino acid labeling. Given the high reactivity of radical species, the labeling time can be as short as μs,⁸⁴¹ which makes possible the fastest footprinting available to follow protein dynamics, providing a “snapshot” of the protein SASA with residue-level resolution for those residues that react. Fast footprinting can also probe protein folding/unfolding. The theoretical coverage is broader than that of specific amino acid labeling, but it is nearly impossible to realize the coverage of HDX because the reactivity between radicals and the side chains can vary by ~ 10³, and the reactions will emphasize the more reactive side chains. The reagent concentrations are 10² to 10³ times to that of a protein; for large proteins and reactive reagents, the free radicals will become limiting reagents. Both specific amino acid labeling and free radical labeling usually produce irreversible modifications, thereby facilitating effective protein isolation and digestion prior to the MS analysis.

In this section and the next, we will review all available fast labeling reagents and describe their generation, development, biological relevance, and application in protein footprinting. We will also discuss their advantages and limitations.

5.1. Hydroxyl Radical

Hydroxyl radical (^●OH) was initially used for nucleic acid footprinting, where it reacts with solvent accessible hydrogens in deoxyribose (DNA) and ribose (RNA), cleaving their nucleic acid chains.¹⁶⁴ The motivation is to footprint nucleic acids, to reveal their SASA and determine their HOS, and to map protein DNA/RNA interactions. Later, the idea was extended to protein footprinting. For DNA/RNA footprinting, the approach is based on selective cleavages, whereas with protein footprinting, the approach is based on chemical modifications. Today, ^●OH has become the most used fast labeling reagent.

5.1.1. Biological Relevance

The hydroxyl radical is a naturally occurring reactive oxygen species (ROS) that participates in several cycles that regulate several physiological functions of living organisms.^842-844 Early studies identified the important signaling role of ^●OH in regulating several physiological functions of living organisms, especially its contribution to retrograde redox signaling from organelles to the cytosol and nucleus.^{842, 844} The size and hydrophilicity of ^●OH are similar to those of the water molecule,⁸⁴⁵ allowing it to probe the SASA of biomacromolecules (to travel where water can travel). The oxidative damage in many pathologies caused by ^●OH naturally and in radiation damage attracted early attention, motivating determination of its rate constants^{170, 846} and mechanisms⁸⁴⁷ in reactions with amino acids. The reaction rate constants between ^●OH and amino acids range from 1.7 × 10⁷ to 3.5 × 10¹⁰ M⁻¹s⁻¹ with preference toward aromatic, heterocyclic and sulfur-containing sidechains (Table 1). The broad residue coverage in ^●OH reactions contrast significantly with the targeted or specific labeling reagents covered above. The early fundamental studies, likely driven by radiation concerns, form a solid foundation for the use of ^●OH in protein footprinting, contributing to its growing applications.

Table 1.

Rate constants for reactions between amino acids and ^●OH^{170, 846}

Substrate	Rate Constant (M−1s−1)	pH	Substrate	Rate Constant (M−1s−1)	pH
Cys	3.5 × 1010	7.0	Pro	6.5 × 108	6.8
Trp	1.3 × 1010	6.5 – 8.5	Gln	5.4 × 108	6.0
Tyr	1.3 × 1010	7.0	Thr	5.1 × 108	6.6
Met	8.5 × 109	6 – 7	Lys	3.5 × 108	6.6
Phe	6.9 × 109	7 – 8	Ser	3.2 × 108	~6
His	4.8 × 109	7.5	Glu	2.3 × 108	6.5
Arg	3.5 × 109	6.5 – 7.5	Ala	7.7 × 107	5.8
cystine	2.1 × 109	6.5	Asp	7.5 × 107	6.9
Ile	1.8 × 109	6.6	Asn	4.9 × 107	6.6
Leu	1.7 × 109	~6	Gly	1.7 × 107	5.9
Val	8.5 × 108	6.9

5.1.2. Generating ^●OH in Solution

The generation of ^●OH has a long history, during which many different methods were developed. In this section, only methods that are useful for footprinting biomacromolecules will be reviewed. Early approaches made use of ^●OH-induced protein backbone cleavages,¹⁶⁴ similar to those in nucleic acid footprinting. Upon cleavage, gel electrophoresis was used to determine the cleavage sites. Low cleavage efficiency, lack of accurate mass determination, and low precision greatly limits the further development of such an approach; thus, it will not be covered in detail.

5.1.2.1. Fenton and Fenton-like Chemistry

The catalytic property of Fe(II) in promoting the oxidation of tartaric acid by hydrogen peroxide (H₂O₂) was first reported by Fenton in 1894.⁸⁴⁸ Later on, Cu(I), Ti(III), Co(II) and Cr(III) were found to behave similarly.^849-852 Such processes were later called Fenton reactions. Although it was supposed that ^●OH is the key reactive species in Fenton chemistry, it was not well established until Haber and Weiss^853-854 first proposed a chain reaction mechanism in 1932. The understanding of the mechanism was later expanded by Barb and coworkers^855-856, and it is now referred to as “classical Fenton pathway”, as summarized in Eq. 2-8 below.

$${\text{Fe}}^{2+}+{\mathrm{H}}_2{\mathrm{O}}_2\to {\text{Fe}}^{3+}+{}^{•}\text{OH}+{\text{OH}}^{\text{-}}$$ (2)

$${\text{Fe}}^{3+}+{\mathrm{H}}_2{\mathrm{O}}_2\to {\text{Fe}}^{2+}+{\text{HO}}_2^{•}+{\mathrm{H}}^{+}$$ (3)

$${\mathrm{H}}_2{\mathrm{O}}_2+{}^{•}\text{OH}\to {\text{HO}}_2^{•}+{\mathrm{H}}_2\mathrm{O}$$ (4)

$${\text{HO}}_2^{•}\leftrightarrow {\mathrm{O}}_2^{\text{-}•}+{\mathrm{H}}^{+}$$ (5)

$${\text{Fe}}^{3+}+{\text{HO}}_2^{•}\to {\text{Fe}}^{2+}+{\mathrm{O}}_2+{\mathrm{H}}^{+}$$ (6)

$${\text{Fe}}^{3+}+{\mathrm{O}}_2^{\text{-}•}\to {\text{Fe}}^{2+}+{\mathrm{O}}_2$$ (7)

$${\text{Fe}}^{2+}+{}^{•}\text{OH}+{\mathrm{H}}^{+}\to {\text{Fe}}^{3+}+{\mathrm{H}}_2\mathrm{O}$$ (8)

In Fenton chemistry, ^●OH is generated by oxidation of Fe(II) and Fe(III) with H₂O₂ (Eq. 2 and 3). This chemistry is not optimum for footprinting biomacromolecules. For example, the reaction has an optimal rate at pH of 3 – 4,^855-856 which is clearly denaturing for many biomolecules. Under physiological conditions, however, Fe(III) readily precipitates.

In 1985, Tullius and Dombroski^857-859 proposed an elegant system that utilizes Fenton chemistry to map DNA-protein binding sites. By incubating Fe(II)-ethylenediaminetetraacetic acid (Fe(II)-EDTA), H₂O₂, and ascorbate with DNA, they achieved an effective footprint after a few minutes. The approach exploits Fenton chemistry in two ways. First, with EDTA as chelators for Fe(II) and Fe(III), the solubility of these ions under physiological pH increases significantly. EDTA also minimizes the binding of these ions to biomacromolecules and increases their catalytic efficiency.⁸⁶⁰ Second, ascorbate can reduce Fe(III)-EDTA back to Fe(II)-EDTA, fulfilling a cycle. Thus, the Fe(II-EDTA)/H₂O₂/ascorbate Fenton system had become the usual approach for footprinting, and it was applied in many subsequent DNA footprinting studies.⁸⁶¹

For protein footprinting, the system was first applied to footprint cAMP receptor protein under backbone-cleavage conditions by Heyduk and coworkers⁸⁶² in 1994, and then by Sharp and Hettich⁸⁶³ in 2003, who extended oxidative labeling to apo-myoglobin by using ^●OH to map amino acid side chains as a measure of their SASA. A more recent variation on this theme by Monroe and Heien⁸⁶⁴ combines electrochemistry with this classical reaction in which an electrical flow cell was used to reduce Fe(III) to Fe(II), thus completing a cycle. Footprinting of ubiquitin demonstrated the efficacy of such method in probing protein SASA.

Another powerful approach using Fenton chemistry was first demonstrated by Rana and Meares^865-867 in 1990. As a replacement of Fe(II)-EDTA in a classical Fenton system, they synthesized a novel class of reagents represented by Fe-(S)-1-(p-bromoacetimidobenzyl)-EDTA (Fe-BABE). Fe-BABE is composed by three structural units (Scheme 14), a metal-chelating unit that binds with reactive metal centers to facilitate ^●OH generation, a sulfhydryl reactive unit that reacts with cysteine residues on proteins, and a linker separating them. The design localizes the Fe-BABE in a specific position of a protein. Upon incubating with H₂O₂, ^●OH is only generated in close proximation with the localized metal center. Fe-BABE has the advantage of high conjugation rate with proteins, high anchoring yield, and compatibility with neutral pH, making such site-directed ^●OH footprinting effective in probing spatial relationship in protein-nucleic acid^868-870 and protein-protein⁸⁷¹ complexes. Very recently, this approach was adopted for characterizing protein-carbohydrate interactions,⁸⁷² whose details will be discussed in Section 6.7.

Scheme 14.

Structure of Fe-BABE

Fenton chemistry is probably the most easily accessible method for ^●OH protein footprinting, as it can be achieved without the help of sophisticated instruments. The biological relevance of iron makes it “friendly” with most biological systems, as demonstrated by in vivo footprinting of the membrane protein porin OmpF.⁸⁷³ Moreover, given that multiple reactive metal centers can initiate Fenton-like reactions, successful ^●OH protein footprinting can be achieved by using Co(II),⁸⁷⁴ Cr(III),⁸⁷⁴ Ni(II),⁸⁷⁵ Cu(II)^876-877, and Mn(II)/Mn(III) for generating ^●OH upon reaction with H₂O₂.⁸⁵¹ This general property makes Fenton-like chemistry worth considering to footprint proteins that bind to these active metal centers.

On the other hand, ^●OH generation by Fenton chemistry usually takes minutes, during which the protein may get over labeled and alter its conformation during the footprinting owing to the changes in hydrophilicity or perhaps to some modification of amino acid residues with low SASA.⁸⁷⁸ Once the protected region becomes exposed, additional ^●OH can further label the newly-exposed sites, generating a misleading readout. Newer ^●OH generation approaches usually label proteins on the time scale of milliseconds or less to minimize over labeling. Although a subsequent study employing Fenton chemistry achieves a steady ^●OH concentration at 2 ms, the amount of H₂O₂ (30%) required is certainly stressful to most biomacromolecules.⁸⁷⁹

5.1.2.2. Synchrotron Water Radiolysis

A synchrotron X-ray source is capable of delivering a continuous spectrum of 10¹⁴ – 10¹⁵ photons per second with energies ranging from 5 to 30 kV.⁸⁸⁰ Implemented by Chance and coworkers^881-883, synchrotron-based ^●OH footprinting was first demonstrated for nucleic acids and later for proteins⁸⁸⁴. The latter initiates the modern era for protein footprinting by fast-labeling reagents because, for the first time, the labeling timescale was reduced to milliseconds.⁸⁴⁵ Mechanistically, synchrotron X-rays utilize water as the ^●OH precursor. High energy photons ionize water to produce hydrated electrons (e_aq⁻) and activated water (H₂O*), and these initiate the subsequent reactions as summarized in Eq. 9-11 below.¹⁷⁰

$$2{\mathrm{H}}_2\mathrm{O}+{\mathrm{e}}^{\text{-}}(\text{ionizing irradiation})\to {\mathrm{H}}_2{\mathrm{O}}^{+•}+{\mathrm{H}}_2{\mathrm{O}}^{\ast }+{{\mathrm{e}}_{\text{aq}}}^{\text{-}}$$ (9)

$${\mathrm{H}}_2{\mathrm{O}}^{+•}+{\mathrm{H}}_2\mathrm{O}\to {}^{•}\text{OH}+{\mathrm{H}}_3{\mathrm{O}}^{+}$$ (10)

$${\mathrm{H}}_2{\mathrm{O}}^{\ast }\to {}^{•}\text{OH}+{\mathrm{H}}^{•}$$ (11)

Besides labeling solvent-accessible amino acid side chains in the protein, ^●OH undergoes quenching reactions due to its high reactivity. Under anaerobic conditions, primary quenching is a self-recombination at a diffusion rate limit (Eq. 12 - 13).⁸⁸⁵

$${}^{•}\text{OH}+{{\mathrm{e}}_{\text{aq}}}^{\text{-}}\to {\text{OH}}^{\text{-}}$$ (12)

$$2^{•}\text{OH}\to {\mathrm{H}}_2{\mathrm{O}}_2$$ (13)

Under aerobic conditions, oxygen is involved in the quenching process, leading to other ROS (Eq. 14-15).

$${{\mathrm{e}}_{\text{aq}}}^{\text{-}}+{\mathrm{O}}_2\to {\mathrm{O}}_2^{\text{-}•}$$ (14)

$${\mathrm{H}}^{•}+{\mathrm{O}}_2\to {\text{HO}}_2^{•}$$ (15)

In subsequent studies, ^●OH dosimetry, the effects of different buffers, different beam currents and other additives in the system were thoroughly investigated.⁸⁸⁶ These efforts were comprehensively reviewed elsewhere.¹⁷⁰

Synchrotron-based water radiolysis has significant advantages in protein footprinting, as it utilizes water as ^●OH precursor. Water concentration is ~55 M, and it is not limiting. Although water interacts with proteins, there are many more molecules that constitute bulk, and this may not be the case for other radical sources. More importantly, water is the medium for biological systems, and its use simplifies the experimental procedure. The fast labeling speed, easy dose control (by controlling exposure time with a shutter⁸⁸⁶) and high reproducibility all contribute to its utility. An obvious downside is there are limited synchrotron sources. A recent development of a new beamline should make the method more powerful and more accessible than in the past to the general research community.⁸⁸⁷

5.1.2.3. Laser Photolysis of Hydrogen Peroxide

Upon exposure to UV around 250 nm, H₂O₂ homolytically cleaves into two ^●OH with a primary quantum yield of 0.4 – 0.5.^888-889 In the absence of reactive substances, the resulting ^●OH reacts with the H₂O₂ that did not undergo photolysis as shown in Eq. 16-18.

$${\mathrm{H}}_2{\mathrm{O}}_2\to 2{}^{•}\text{OH}$$ (16)

$${}^{•}\text{OH}+{\mathrm{H}}_2{\mathrm{O}}_2\to {\mathrm{H}}_2\mathrm{O}+{\text{HO}}_2^{•}$$ (17)

$${\text{HO}}_2^{•}+{\mathrm{H}}_2{\mathrm{O}}_2\to {\mathrm{H}}_2\mathrm{O}+{\mathrm{O}}_2+{}^{•}\text{OH}$$ (18)

The rate constants for the latter two reactions are 2.7 × 10⁷ and 7 × 10⁹ M⁻¹s⁻¹, respectively.⁸⁸⁹ Large reaction rate constants guarantee fast labeling.

In 2004, Sharp et al.⁸⁹⁰ first footprinted lysozyme and β-lactoglobulin by ^●OH generated from H₂O₂ photodissociation. For this protocol, the protein sample contains 15% H₂O₂, and the Irradiation (by a UV lamp) takes up to 5 min. This high H₂O₂ concentration is stressful to many proteins. Moreover, the long labeling time is a risk that the system will be over labeled, as discussed above.

To reduce the labeling time and lower the H₂O₂ concentration, Aye and Sze⁸⁹¹ replaced the UV lamp with a pulsed Nd:YAG laser that operates at 266 nm (frequency quadrupled). They used a light pulse energy of 2 mJ/pulse and a pulse width is 3-5 ns in a static system. H₂O₂ was added to 0.3% just prior laser irradiation to minimize exposure of the protein to a mild oxidizing agent. The residue H₂O₂ after irradiation was removed by snap-freezing and lyophilizing the aliquot (possibly a source of error as protein oxidation by H₂O₂ can occur in the solid state at low T as reported later⁸⁹²). These improvements over Fenton chemistry combine to achieve a moderate oxidation of ubiquitin with a single laser shot. As compared with Fenton chemistry, the H₂O₂ concentration and labeling time were both reduced significantly, allowing a higher confidence of the result.

At the same time, Hambly and Gross⁸⁴¹ reported a method that couples H₂O₂ laser photolysis and a flow system to footprint proteins oxidatively, which was later named fast photochemical oxidation of proteins (FPOP). As compared with using a Nd:YAG laser, the KrF excimer laser they used has an output of 248 nm, which is closer the λ_max of H₂O₂. The laser pulse width was 17 ns, and the power was 50 mJ/pulse, both readily achieved with such a laser. These efforts combine to afford measurable oxidation levels of proteins with 0.04% H₂O₂. The flow system, as illustrated in Figure 12, ensures that the proteins in the aliquot are only irradiated once. Flow rate was calculated in correspondence with laser pulse width and frequency to allow a 20 – 25% exclusion volume, which minimizes any “double shots” of the protein solution. Left-over H₂O₂ was decomposed by catalase in the collection tube. Moreover, the reaction duration was controlled and became tunable by adding a radical scavenger (Glu⁸⁴¹ and later His¹⁸¹), which not only limits the labeling time to 0.5 μs for minimizing over-labeling but also allows better control of the labeling process⁸⁷⁸ and makes possible probes of fast kinetic processes including protein folding/unfolding^893-894. These advantages make FPOP the most adopted platform for protein footprinting by fast labeling reagents.

Figure 12.

Schematic illustration of FPOP setup

As compared to synchrotron water radiolysis, H₂O₂ photolysis retains most of its advantages in generating ^●OH. Although H₂O₂ is a necessary precursor, low concentrations minimize any oxidation interference to protein native states. Most importantly, H₂O₂ photolysis makes the ultra-fast ^●OH protein footprinting more accessible to the general research community although efforts are underway, principally by a commercial developer, to incorporate a discharge lamp to replace the laser, which imposes safety requirements and some expertise for handling. Nevertheless, subsequent developments take advantage of the strengths of FPOP, and a description of several applications will be given in Section 6.

On the other hand, H₂O₂ is a relatively strong oxidant, which may make this approach troublesome for proteins that are prone to oxidation. A modified FPOP apparatus with a mixer located just before laser window largely minimizes the H₂O₂-induced protein oxidation.⁸⁹⁵ A recent report suggests that H₂O₂ can interact with the protein, giving local high concentrations of the H₂O₂ and the radicals on regions of the protein surface.⁸⁹⁶

In response to the original claim that the radicals have been scavenged in ~ 1 μs, Konermann and coworkers⁸⁹⁷ employed a reporter dye to follow the FPOP kinetics and showed that the free radical oxidation of the dye extends over tens of ms, indicating a longer radical lifetimes than the original prediction. This phenomenon likely applies to the secondary radicals formed as ^●OH reacts. It remains to be seen if these less reactive radicals react with the protein. The assertion about μs lifetime is probably correct for the primary radicals (i.e., ^●OH).

Another consideration for this platform is the potential migration of initially formed protein radicals to other reactive residues that are buried below the surface. The investigators found that two solvent-inaccessible residues were modified by FPOP, possibly in accord with a potential long-distance radical transfer (radical jumping) to buried residues.⁸⁹⁸ This and other considerations need further attention to make FPOP more robust and applicable to more biological systems.

5.1.2.4. High Voltage Electrical Discharge

High voltage electrical discharge takes advantage of the requirement to use mass spectrometry in the analysis by a simple adaption of a typical ESI source to increase the spray voltage to 6-8 kV.⁸⁹⁹ To make the radical, Downard and coworkers^899-900 increased the potential difference between the emitter tip and the grounded collection plate (either the MS inlet or a collection vessel) to induce a corona discharge, during which oxygen in the nebulizer gas will be activated to give a plasma at the tip of the emitter (Eq. 19-20).

$${\mathrm{O}}_2\to {\mathrm{O}}_2^{+•}+{\mathrm{e}}^{\text{-}}$$ (19)

$${\mathrm{O}}_2^{+•}+{\mathrm{H}}_2\mathrm{O}\to {\mathrm{H}}_3{\mathrm{O}}^{+}+{}^{•}\text{OH}$$ (20)

Meanwhile, water from the ESI spray can be activated as well (Eq. 21).

$$2{\mathrm{H}}_2\mathrm{O}\to {\mathrm{H}}_2{\mathrm{O}}^{+•}+{\mathrm{e}}_{\text{aq}}^{\text{-}}\to {\mathrm{H}}_3{\mathrm{O}}^{+}+{}^{•}\text{OH}+{\mathrm{e}}^{\text{-}}$$ (21)

By switching the nebulizer gas from air to pure O₂, the oxidation efficiency can be doubled.⁹⁰¹ The approach is also compatible with both positive and negative polarities.⁹⁰² The oxidatively labeled protein is then submitted directly to the mass spectrometer for top-down analysis (proteins are dissolved in ammonium acetate to facilitate sufficient ionization), or it can be accumulated in a collection vessel for further treatment and proteolysis. In subsequent studies, an O₂-assisted electrical discharge has been successfully utilized to study protein complexes between ribonuclease S-protein—S-peptide⁹⁰³ and calmodulin—melittin⁹⁰⁴. In a more recent twist, Maleknia and Downard⁹⁰⁵ joined MALDI and electrical discharge protein footprinting to afford an even higher throughput.

Given the popularity of ESI in protein analysis, high voltage electrical discharge makes ^●OH protein footprinting readily available to almost every MS lab. The fast-labeling timeframe and high labeling efficiency seem to be promising. Given that charge-induced protein unfolding occurs in an ESI source,⁹⁰⁶ a question arises whether the protein of interest can maintain its high-order structural integrity during the oxidative footprinting. It is always preferred to perform protein footprinting under native or near-native states to minimize these ambiguities, where a high-voltage ESI spray is not optimal. Moreover, it is challenging to footprint binding systems with weak interactions through this method, as weak interactions are less likely to be preserved in an ESI droplet than in neutral solution. The labeling time frames and the radical lifetimes also remain to be determined.

The concern of charge-induced unfolding by electrical discharge-based ^●OH labeling was recently addressed in 2017 by Minkoff and Sussman, who developed a novel experimental approach named plasma induced modification of biomolecules (PLIMB, Figure 13).⁹⁰⁷ Instead of delivering the protein through an ESI source, the protein aliquot was placed in a grounded Eppendorf tube with a charged needle on top of the liquid. A plasma discharge is induced when supplying 1 – 31 kV of potential to the needle with frequencies of 0 to 15 kHz. Over a period of 60 s, denatured and native bovine serum albumin were oxidatively labeled to achieve modification fractions up to 15 % at peptide level. As part of the initial demonstration, PLIMB was adopted to characterize the epidermal growth factor-induced structural changes in the extracellular domain of its receptor. These examples demonstrate efficacy in facilitating ^●OH-based protein footprinting.

Figure 13.

Schematic illustration of PLIMB setup. Adapted with permission from ref.⁹⁰⁷. Copyright 2017 Springer Nature.

PLIMB offers a benchtop solution for radical protein footprinting by overcoming the major drawback for conventional electrical discharge methods, as the protein of interest is now footprinted in solution under near-native conditions. On the other hand, the plasma discharge in PLIMB only happens at the surface of the aliquot. Given the short lifetime of ^●OH,¹⁷⁰ most of the ^●OH labeling occurs in limited regions of the solution. Means to uniformly label the sample aliquot and to minimize the over labeling are likely to be future developments of PLIMB.

5.1.2.5. Gamma Ray Water Radiolysis

Gamma ray (γ-ray) is an electromagnetic radiation that composed of high energy photons. Similar to synchrotron X-ray water radiolysis, γ-rays excite water molecules and trigger reactions that produce ^●OH (Eq. 2-8).⁹⁰⁸ γ-ray-based ^●OH footprinting, first applied to study nucleic acids,⁹⁰⁹ was implemented by Nukuna et al.⁹¹⁰ to footprint cytochrome C in 2004. This approach was adopted by a few research groups in subsequent years,^911-916 but more recently, the approach is seldom used due to safety concerns of the highly-penetrable nature of γ-rays and the emergence of other approaches.

5.1.2.6. Other Hydroxyl Radical Generation Methods

Besides the methods introduced above, there are other methods that for ^●OH footprinting, including ultrasound sonolysis,^917-918 fast neutrons,^919-921 peroxynitrous acid decomposition,^922-923 pulsed electron beam,⁹²⁴ photolysis of N-hydroxypyridine-2(1H)-thione,^925-926 boron-doped diamond electrochemical surface mapping,⁹²⁷ and ozonolysis⁹⁰¹. Because these methods were only used in nucleic acid footprinting or not developed for protein footprinting, they will not be covered further in this review.

5.1.3. Residue Specificity and Proposed Reaction Pathways

One of the unique advantages of ^●OH in protein footprinting is its broad reactivity. Xu and Chance^{170, 912, 928-930} conducted systematic studies and found that the reactivity between ^●OH and amino acid residues rank as Cys > Met > Trp > Tyr > Phe > cystine > His > Leu, Ile > Arg, Lys, Val > Ser, Thr, Pro > Gln, Glu > Asp, Asn > Ala > Gly. Among all these residues, Gly, Ala, Asp and Asn are unlikely to serve as useful probes owing to their low reactivity (also see Table 1). Although the reactivities for Ser and Thr are higher than that of Pro, which has been found to be a reactive and useful substrate, their oxidation products are not easily detectable and may be dispersed among several pathways (e.g., oxidation of -OH to =O). From rate constant determinations and reports of experience with ^●OH and protein modification, the general conclusion is that 14 out of 20 amino acid residue sidechains are active in ^●OH-based protein footprinting experiments. In our experience, the presence of highly reactive residues (Cys, Met, Trp, Tyr, Phe, His) may siphon most of the reagent radicals to their modification, reducing the coverage. The products resulting from oxidative modifications are residue-dependent, and they have been covered in detail in the Xu and Chance review¹⁷⁰; most are +16 Da oxidations as summarized in Table 2 Below.

Table 2.

Primary oxidation produces (upon ^●OH footprinting) and the corresponding mass changes (in reactivity order) for various amino acid residue sidechains.⁹¹²

Sidechain	Modification and Mass Changes
Cys	sulfonic acid (+48), sulfinic acid (+32), hydroxy (−16)
Met	sulfoxide (+16), sulfone (+32), aldehyde (−32)
Trp	hydroxy- (+16, +32, +48, etc.), pyrrol ring-open (+32)
Tyr	hydroxy- (+16, +32, etc.)
Phe	hydroxy- (+16, +32, etc.)
cystine	sulfonic acid (+48+H), sulfinic acid (+32+H)
His	oxo-(+16), ring-open (−22, −10, +5)
Leu	hydroxy- (+16), carbonyl (+14)
Ile	hydroxy- (+16), carbonyl (+14)
Arg	deguanidination (−43), hydroxy- (+16), carbonyl (+14)
Lys	hydroxy- (+16), carbonyl (+14)
Val	hydroxy- (+16), carbonyl (+14)
Ser	hydroxy- (+16), carbonyl (−2, or +16-H2O)
Thr	hydroxy- (+16), carbonyl (−2, or +16-H2O)
Pro	hydroxy- (+16), carbonyl (+14)
Gln	hydroxy- (+16), carbonyl (+14)
Glu	decarboxylation (−30), hydroxy- (+16), carbonyl (+14)
Asp	decarboxylation (−30), hydroxy- (+16)
Asn	hydroxy- (+16)
Ala	hydroxy- (+16)
Gly	N/A

^aFor aliphatic side chains, +14 Da products are normally much less than +16 Da products.

^bFor Ser and Thr, only trivial amount of +16 and −2 Da products were found.

Mechanistically, ^●OH activates the solvent accessible sidechains by removing H^● from an activated site, adding a ^●OH to a double bond, resulting in a protein radical that is subsequently quenched differently depending on the origin of the radical (H₂O or H₂O₂).

For synchrotron radiolysis, Xu and Chance,^{170, 928-930} in a comprehensive review, summarized the modification pathways of ^●OH-active amino acid residue sidechains.¹⁷⁰ Leu, Ile, Val, Pro, Arg, Lys, Glu and Gln sidechains are activated through H^● abstraction whereas His, Phe, Tyr, Trp and Met are primarily by ^●OH addition. The resulting protein radicals are subsequently quenched by dissolved O₂ in aerobic conditions. Reaction pathways between ^●OH and the extremely reactive Cys and cystine residues are complicated and not well understood. Alternatively, a likely reaction pathway that covers major oxidative products was proposed for Cys.⁹³⁰ Similar pathways should apply to the electrical discharge-based ^●OH labeling, as the precursor for ^●OH is also water.

In contrast to the two systems mentioned above, where uniformly distributed water is the ^●OH precursor, any H₂O₂ photolysis approach, as represented by FPOP, follows slightly different reaction pathways. A recent quantum calculation study revealed the hydrogen bonding between H₂O₂ and amino acids including His, Arg, Tyr, Cys, Thr, Gln, Asp, Lys, Met and Trp.⁹³¹ Such hydrogen bonding will induce a pre-formed H₂O₂ – amino acid residue complex, which will result in a local fluctuation in ^●OH concentrations upon laser photolysis. Liu, Gross, and coworkers⁸⁹⁶ studied the reaction pathways between ^●OH and 13 different amino acid residues on a FPOP platform. By using ¹⁸O-enrichment of all three available oxygen sources (H₂O, H₂O₂ and dissolved O₂) on an FPOP platform, one at a time, they differentiated three classes of residues based on their oxygen uptake preferences. His, Arg, Tyr, and Phe preferentially take oxygen from H₂O₂. Met competitively take oxygen from H₂O₂ and dissolved O₂, whereas Leu, Ile, Val, Pro, Lys, Asp, Glu and Gln take oxygen exclusively from O₂. Given that the ^●OH activation pathway is very similar to that in water radiolysis, such differences in oxygen uptake preferences reveal a different pathway for some activated residues than described in the Xu and Chance¹⁷⁰ review. Besides quenching by dissolved O₂, selected activated residues can also be favorably modified by reaction with another ^●OH in the FPOP setup, owing to hydrogen bonding with H₂O₂ and preferential localization of the reagent and, of course, the radicals.

As mentioned earlier, ^●OH is also capable of cleaving protein backbones, whose pathways is reviewed elsewhere.^{170, 847}

Summarizing the role of ^●OH in footprinting, we recognize unprecedented advantages as a fast protein footprinter. Biological relevance, similar size and hydrophilicity as water, fast labeling time, high labeling efficiency, multiple generation methods, broad residue coverage, well established platforms, and mature understanding of the labeling pathways make it the most utilized fast-labeling reagent. Its demonstrated utility forecast many more promising applications (see section 6). On the other hand, the +16 Da from oxidative labeling can be problematic, as oxidation is common in native biological systems, and it is challenging to distinguish the oxidations that exist prior to labeling and those that are from ^●OH reactions. Other oxidations besides those producing +16, can add background of many low abundance products to complicate the analysis. Furthermore, even though 14 amino acid residues are reactive with ^●OH, those that are most reactive (S-containing, aromatics) can be too competitive, reducing the overall coverage and spatial resolution achieved for the protein. These drawbacks motivate the development of other fast labeling reagents that provide complementary coverage to ^●OH labeling and employ an MS tag other than +16 Da.

5.2. Sulfate Radical Anion and other Sulfur Containing Radicals

5.2.1. Biological Relevance

Sulfate radical anion, SO₄^−•, is a potent oxidant with a reduction potential of 2.43 V at neutral pH.⁹³² Its strong oxidation capability can cause considerable damage to different cellular components including lipids, carbohydrates, proteins and DNA/RNA.⁹³³ It is an exogenous factor that promotes other oxidant formation in biological systems,⁹³⁴ and it has been applied to inactive pathogenic microorganisms in disinfection⁹³³. Many investigators hold that the oxidation pathways of biomolecules by SO₄^−• are like those of the ^●OH; however, detailed mechanistic studies are still needed.

5.2.2. Radical Generation in Solution and Applications in Biology

The production of the sulfate radical anion in vitro primarily utilizes two precursors, persulfate (S₂O₈²⁻) and peroxymonosulfate (HSO₅⁻). Activation methods are needed to cleave the O-O peroxide bond and give SO₄^−• generation. UV irradiation⁶, thermolysis, radiation² and transition-metal catalysis^935-936 are possibilities. As a promising oxidant, SO₄^−• may be useful for answering biological questions. In 1983, the Werbin group⁹³⁷ cross-linked DNA with lysosome to study reactive mechanism of a carcinogen, N-acetoxy-N-acetyl-2-aminofluorene (N-AcO-AAF). The Kodadek⁹³⁵ and Jovin groups⁹³⁶ both demonstrated efficient SO₄^−•-generation from a ruthenium complex, Ru(bipyridine)₃, for crosslinking initiation. The latter study further shows that SO₄^−• induces tyrosyl radical formation, allowing inter cross-links between alpha-synuclein (αSyn) aggregates. The covalently bound large oligomers exhibit distinct biophysical properties and increased toxicity.

5.2.3. SO₄^−•-based Protein Footprinting: Residue Specificity and Proposed Reaction Pathways

Bridgewater and Vachet ^938-940 in 2005-2006 first used sulfate radical anion to determine binding and map the SASA of proteins. Their generation of SO₄^−• utilizes a metal-catalyzed oxidation (MCO), a process that enables radical formation at the site of a redox-active metal when bonded to a peptide/protein (Scheme 15). Cu/Zn superoxide dismutase was chosen as a model system in the presence of ascorbate and persulfate. High concentrations of ascorbate (i.e., 100 mM) rapidly initiate the redox cycle through Cu(II) reduction to provide scavenger-generated SO₄^−•, which reacts by pseudo-first order kinetics. The study showed that at an optimized SO₄^−• concentration (i.e., one generated from 1 mM persulfate) allows the oxidation to occur only on the nearby Cu-bound amino acids. Increasing the persulfate concentration allows modification of non-Cu-bound residues within 10 Å radius of the Cu center. UV-Vis absorption spectroscopy and oxidation kinetics further confirmed the survival of the intact protein structure throughout the analysis, showing promise for future applications under their conditions. Although the steric effects of a specific residue may alter its oxidation extent, this tunable SO₄^−•-based approach provides another approach for three-dimensional mapping of the protein environment around a reactive site. This approach may be particularly important to map metal-binding residues that have low solvent accessibility, and it may be incorporated into an FPOP platform.

Scheme 15.

General scheme that produces reactive oxygen species during metal-catalyzed oxidation reactions.

In 2010, the Gross group⁹⁴¹ generated sulfate radical anion on the FPOP platform by photolysis of ⁻OSO₂-O-O-O₂SO⁻ at a quantum yield of 0.55. As a stronger oxidant than hydroxyl radical, the sulfate radical anion gives comparable extents of modification on the protein-level with 3-5 times lower amounts of precursor (Figure 14a). Incubation with persulfate anion without laser irradiation doesn’t introduce observable oxidation on the protein, showing negligible background from the precursor reagent. In addition, the investigators showed that the reactivity and specificity of SO₄^−• is similar to those of ^●OH for residues Met, Trp Glu and Ser; the reactant radical favors His and Tyr (Figure 14b and 14c). The overall reactivity ranking of SO₄^−• is:

$$\text{Met}>\text{Tyr}=\text{Trp}>\text{Phe}=\text{Glu}=\text{His}>\text{Ser}>\text{Pro}>\text{Asp}=\text{Thr}>\text{Lys}=\text{Gln}>\text{Leu}=\text{Val}=\text{le}$$

Figure 14.

(a) The ESI-quadrupole time-of-flight mass spectra of the 15^th charge state of β-lactoglobulin submitted to different labeling conditions as shown in the figure. (b) The relative difference in fraction modified between persulfate and the H₂O₂ approaches for CaM, aMb, bradykinin, and angiotensin II residues, are averaged per amino acid type. The error bars denote the average pairwise-comparison standard error. (c) The maximum fraction modified among all the same amino acid residues is plotted. Black bars denote H₂O₂ FPOP. Reprinted with permission from Ref.⁹⁴¹ Copyright 2010 American Chemical Society.

Oxidation sulfate-radical oxidation can happen through multiple pathways. Aromatic and Met side chains are the main targets for the strong electrophile SO₄^−• and form a radical cation intermediate, which subsequently undergoes either fragmentation or hydration.^942-943 SO₄^−• can also oxidize a carboxyl anion (R-COO⁻) to give acyloxyl radicals (R-COO^●) followed by decarboxylation to generate carbon-centered radicals.⁹⁴³ Moreover, hydrogen abstraction by SO₄^−• is another pathway to oxidative modification, likely occurring on Leu, Ser and Phe.⁹⁴⁴

Surprisingly, although sharing some similarities, persulfate FPOP and hydroxyl FPOP also have different chemical outcomes. For systems where there is need to maintain more “physiologic-like” conditions, the lower amount of reagent for persulfate anion makes it a more desirable choice. Different physical properties of the two radicals also determine their preferred circumstances for surface mapping. Negatively charged SO₄^−• is prone to form electrostatic interactions with other positively charged groups, introducing a possible bias. One example is shown by Gau et al.⁹⁴¹, who compared per-residue fractional modification for six histidines in apomyoglobin to their SASA values calculated from an X-ray and NMR study. A good correlation (R² = 0.83 in Figure 15b) shows the potential of SO₄^−• footprinting to be an effective experimental measure of SASA. The interaction of SO₄^−• and H64, an axial ligand of the heme iron, however, contributes to the enhanced reactivity of H64 (Figure 15a). On this matter, the hydroxyl radical, which bears no charge and has a comparable size to water, may be a better choice as footprinter that responds to SASA. In addition, the hydroxyl radical is more membrane-permeable, suitable for studies on micelles, liposomes and nanodics.⁹⁴⁵ The sulfate radical anion, on the other hand, reacts more efficiently on extracellular-accessible residues. The combination of the two in a tandem format may be effective in addressing complicated protein systems consisting of phospholipid bilayers imbedded with protein analyte.

Figure 15.

The modification yields of apomyoglobin His residues, where (a) all histidines were included and (b) His64 was omitted, are plotted against their calculated SASA, shown with least-squares straight-line fits. The SASA values were calculated by using a 1.4 Å probe. Reprinted with permission from Ref.⁹⁴¹ Copyright 2010 American Chemical Society.

5.3. Carbonate Radical Anion

5.3.1. Biological Relevance

The carbonate radical anion, CO₃^−•, is formed primarily from the bicarbonate-carbon dioxide pair that exists in cells to maintain physiological pH.⁹⁴⁶ Superoxide dimustase (SOD),⁹⁴⁷ xanthine oxidase (XO)⁹⁴⁸ and hemoprotein in an iron (III) state can adopt conditions whereby the buffer components are sources of CO₃^−•, causing oxidative damage on biomolecules.⁹⁴⁹ Owing to its longer half-life compared to those of other radicals (e.g., ^●OH), CO₃^−• can diffuse farther and cause oxidation distant from its site of formation. One of the well-recognized roles of CO₃^−• is to modulate peroxynitrite activity by abstracting hydrogen from amino acids, primarily tyrosine. The sequential NO₂ addition will lead to nitro-substituted residues, which may be involved in many chronic diseases.⁹⁵⁰

5.3.2. Radical Generation in Solution and Applications in Biology

Carbonate radical anion can be produced in vitro in many ways including UV photolysis⁹⁵¹ of a cobalt complex, [Co(NH₃)₄CO₃]ClO₄, in phosphate buffer, enzymatic production from SOD1 in a bicarbonate buffer⁹⁵¹, pulse radiolysis⁹⁵² of an N₂O saturated NaHCO₃ solution, and laser flash photolysis of persulfate anion in bicarbonate buffer.⁹⁵³ To monitor the production of CO₃^−•, EPR spin trapping is often engaged.⁹⁴⁶ In 2001, Geacintov and coworkers⁹⁵³ generated CO₃^−• as a secondary radical from the sulfate radical anion and, for the first time, characterized its site-selective oxidation of guanine in double-stranded oligonucleotides. Subsequently, many groups focused on the consequences of protein oxidation in biology. One of the foundation studies was done in 1973,⁹⁵⁴ where the rate constants of the carbonate-radical reactions towards some biochemical compounds, including many free amino acids, were measured. Tyrosine, tryptophan and methionine are the most reactive with rate constants on the order of 10⁸ M⁻¹s⁻¹. Gebicki and coworkers⁹⁵² showed that horseradish peroxidase (HRP) and other proteins that are rich in Tyr, Trp, and Met residues experience a 20-30% loss in their activity after exposed to 100 μM CO₃^−•. Paviani et al.⁹⁵¹ characterized a ditryptophan cross-linked product of lysozyme under carbonate radical anion mediated reactions; the study delineated the origin of a previously detected ditryptophan in the non-amyloid aggregation of human SOD,⁹⁵⁵ a process that may be pathogenic for amyotrophic lateral sclerosis⁹⁵⁶.

5.3.3. CO₃^−•-based Protein Footprinting: Residue Specificity and Proposed Reaction Pathways

The carbonate radical anion as a protein footprinter was first evaluated by Zhang et al.⁹⁵⁷ and demonstrated on a FPOP platform. The hydroxyl radical generated from hydrogen peroxide photodissociation reacts with carbonate/bicarbonate (CO₃^2-/HCO₃⁻) buffer to form CO₃^−•. To ensure its dominant presence in solution, highly concentrated CO₃^2-/HCO₃⁻ (e.g., 700 mM) is necessary, supported by a numerical simulation based on second-order kinetics. Experimentally, Zhang et al.⁹⁵⁷ showed no modification in a no-laser control and a similar oxidation pattern as that produced by ^●OH FPOP upon laser irradiation (Figure 16a). The last thorough study of CO₃^−• prior to this work took free amino acids as substrates,⁹⁵⁴ and that may not be relevant for proteins in solution particularly because the N-terminal amines of the free amino acids may complicate the measurements of side-chain reactivity. Therefore, a systematic study of model peptides and proteins was conducted, taking into consideration the residue context, solvent accessibility, and local environment. From the results, the reactivity of the carbonate radical anion can be assigned as follows:

$$\text{Met}\approx \text{Trp}>\text{Tyr}>\text{His}\approx \text{Phe}$$

Figure 16.

(a) The quadrupole time-of-flight (QTOF) mass spectra of the + 20 charge state of apo-myoglobin (aMb) under different labeling conditions as given in the figure. (b) Comparison of the residue-level fraction modified (in percentage) of a ‘Peptide Cocktail’ by CO₃^−• and ^●OH. Different colors represent different peptides, and the corresponding sequences are shown. Solid bars denote the ^●OH oxidation extent, and patterned bars denote the CO₃^−• oxidation. The data were corrected with respect to a negative control. Error bars are the standard deviations of three independent runs. The inset is an enlarged portion of the figure. For bradykinin, only F undergoes oxidation with CO₃^−•, whereas F and P are both oxidized by ^●OH but not distinguishable because the chromatograms of the two modified peptides overlap. Reprinted with permission from Ref. ⁹⁵⁷ Copyright 2019 Elsevier.

The outcome is similar to that reported previously.⁹⁵⁸

Given that CO₃^−• (E° = 1.58 V at pH = 7) ⁹⁵⁹ is a weaker oxidant than ^●OH, the former showed more specificity towards residues with electron-rich side chains (i.e., Met, Trp and Tyr), which underwent even more oxidation than with ^●OH (Figure 16b). Phe and His are also reactive but only modestly. Other aliphatic amino acids are basically inert. The pH also has considerable effect on the reactivity of a specific residue (e.g., His where the imidazole ring carries more charges at pH ~ 6 than under neutral or basic conditions). The electrostatic interaction between the positive side chain and the negative charged CO₃^−• may contribute to a higher local radical concentration that promotes the oxidative chemistry. The oxidation pathway is proposed in the Summary and Perspective (Figure 22), where the CO₃^−• forms a protein-centered radical that undergoes addition of hydroperoxyl radical. Water hydrolysis finally leads to an oxidized product.

The carbonate radical anion can be a good candidate to characterize structural dynamics and binding interfaces of protein complexes, especially when the involved residues are reactive with it. The more specific modifications not only enable higher oxidation levels to reveal subtle differences but also allow faster data analysis and higher throughput. A limitation of its use is also obvious; compared to ^●OH, CO₃^−• has a narrower range of reactivity, and the larger size of CO₃^−• compared to ^●OH would likely afford lower spatial resolution footprints. In addition, the pH of the relatively high concentration of the CO₃^2-/HCO₃⁻ buffer used in the FPOP experiments is basic, possibly not friendly for maintaining a protein’s native state. Furthermore, its generation is not straightforward, diminishing its convenience as a footprinter. It may be that the other production methods can relieve the concerns. Nevertheless, the outcomes show the value of the FPOP platform in fundamental studies of radicals and radical ions that are relevant in biology.

5.4. Carbenes

5.4.1. Biological Relevance and Radical Generation in Solution

Diazirines, common precursor of carbene diradicals, were synthesized in 1960 ⁹⁶⁰ and emerged as a versatile photoaffinity labeling (PAL) agents in the late 1970s.⁹⁶¹ Investigators typically activate diazirines with a UV laser at approximately 350 nm to cause release of N₂ molecules and give an equimolar amount of carbene diradicals. Highly reactive carbenes form irreversible covalent bonds with proteins, allowing stringent downstream affinity purification and target identification. The use of diazirine-based PAL is an effective strategy to understand protein-drug or other small molecule interactions and identify new drug binding sites. To address specific questions, investigators have designed and synthesized several diazirine analogs to adapt the affinity agent to the protein sample environment. For example, adamantylidene, a lipophilic reagent and an analog of adamantane, was used to obtain topological information of Na/K ATPase⁹⁶² and Ca-ATPase⁹⁶³ in membranes as early as 1983. In addition, H-diaziflurane, an analog of halothane, allowed the examination of binding sites of inhaled anesthetics and their action mechanism.⁹⁶⁴

More recently, reagents containing diazirines were adopted as a new class in photo-activatable cross-linkers, and several were described.^{667, 965} The ability to react with many bond types or amino acid residues makes carbenes powerful reagents to capture protein dynamics and intermolecular interactions. Obtaining multiple cross-links magnifies the information by providing more distance restraints on an interacting protein system, and thereby furnishing vital data for molecular simulation or docking to give a more complete description of the system.

5.4.2. Residue Specificity and Proposed Reaction Pathway

Richards et al.⁹⁶⁶ first described the carbene diradical as a footprinting reagent in 2000. Methylene was generated from diazirine gas (CH₂N₂) upon UV irradiation and allowed to footprint α-lactalbumin. The labeling yield, however, was low, owing to the limited solubility of gaseous CH₂N₂ in aqueous media. Furthermore, the explosive gas requires conscientious preparation, storage, and safe handling, limiting wide application.

Later in 2011, Schriemer and coworkers⁹⁶⁷ reported a new diazirine-based reagent, photoleucine (Reagent 1 in Scheme 16) that has higher water solubility and stability than CH₂N₂. The reaction platform incorporates a Nd:YAG pulsed laser (355 nm, 1000 Hz) for radical initiation and a 96-well plate cover with lids with slits, allowing the laser beam to enter the protein solution. With this experimental setup, the investigators obtained a maximum conversion of the diazirine to the carbene diradical by using an irradiation time of 2 min in the absence of other competitive chromophores. Furthermore, they found that photoleucine does not react with targeted proteins without laser activation. Upon photolysis, irreversibly labeled products exhibit a characteristic +115.03 Da mass shift (Figure 17a and 17b).

Scheme 16.

Diazirine-based footprinting reagents

Figure 17.

(a) Deconvoluted mass spectra of myoglobin after footprinting with carbenes. (b) Deconvoluted mass spectra of labeled and unlabeled holo-CaM. (c) Ca²⁺-binding induces conformational change on calmodulin where apo-calmodulin (closed circles) and holo-calmodulin (open circles) were labeled with 100 mM photoleucine in phosphate buffer and monitored as function of time. (d) Free holo-calmodulin (filled circles) is referenced to M13-bound holo-calmodulin (open circles). Reprinted with permission from Ref. ⁹⁶⁷. Copyright 2011 American Chemical Society.

$$\Delta mas{s}_{\text{av}}={\left[\frac{\sum m_i{I}_i}{\sum I_i}\right]}_{\text{labeled}}-M_{\text{unlabeled}}$$ (22)

To examine the labeling sensitivity of carbene diradicals for different accessible surfaces of proteins, two calmodulin systems (i.e., with/without Ca⁺ and bound/unbound to the peptide M13) were investigated. The average number of labels on each protein molecule can be estimated by determining the difference in the centroid masses between the labeled and unlabeled proteins as seen in a deconvolved mass spectrum of the intact protein (Eq. 22 and Figure 17c, d), where m_i and I_i represent mass and signal intensity, respectively. A reduction in protein surface area resulting from Ca⁺ binding or M13 binding is reflected clearly as the measured labeling extent over 2-10 min of irradiation; namely holo-CaM is 45 ± 7% less labeled and M13-CaM is 39 ± 5% less than apo-CaM.

Residue-level quantification of carbene-induced modification needed attention, and that was discussed in a sequel study.⁹⁶⁸ Theoretically, carbene diradicals can insert into X-H bonds and C=C bonds.⁹⁷⁰ Insertion into carboxylic acid functional groups will form labile esters that can be lost in CID fragmentation, thus complicating data interpretation and even losing information. The Shriemer group⁹⁶⁸ chose ETD fragmentation as an alternative to CID and compared the two fragmentation methods by reporting the fraction modified for each y/z ion. Although most fragments shared the same trend, the y₁₀ ion gave poor precision because it undergoes a neutral loss (Figure 18a). Notably, reducing the collision energy reduces the loss, but the abundance of the peptide fragments (product ion in MS/MS) is also reduced. ETD is a superior fragmentation method for retaining labile modifications to afford more comprehensive information than CID. ETD, however, performs poorly when peptides are low in charge and small in size. The best approach might be to use a combination of the two modes of MS/MS.

Figure 18.

(a) Fractional distributions of carbene label derived from reagent 1 as determined by ETD (z-ions, filled circles) and HCD (y-ions, open circles) for peptide LTDEEVDEMIR (117–127) in CaM. (b) Residue-level reagent incorporation for select residues, based on ETD fragmentation of MKDTDSEEEIR (77–87), VFDKDGNGYISAAELR (92–107) and LTDEEVDEMIR (117–127) with Reagent 1 (gray bars) and Reagent 2 (black bars). Error bars are ± 1 standard deviation. Reprinted with permission from Ref.⁹⁶⁸ Copyright 2012 American Chemical Society. (c) Average frequency of carbene insertion at each residue generated from the photolysis of reagent 2, 3, and 4 in the presence of protein digests (777 peptides). Site of the label insertion was located with MS/MS with a Fusion Lumos with EThcD fragmentation and analyzed with Mass Spec Studio software. Reprinted with permission from Ref.⁹⁶⁹. Copyright 2017 Springer Nature.

Another issue is the electrostatic interaction between the carbene precursor and various amino acid side chains; the interaction concentrates the precursor molecule around the site, promoting more modification and a biased residue preference. Switching to another carbene precursor reagent, 4,4-azipentanoic acid (2 in Scheme 16), that contains no positively charged amine group, shows the effect. Although both reagents give similar results for Tyr, Lys, Glu, and Asp, positive-charged photoleucine shows higher reactivity with the negatively charged aspartic acid (Figure 18b). In addition, the ionic strength of the buffer and solution temperature also affect the electrostatic interaction, where increasing ionic strength decreases electrostatic interactions⁹⁷¹ and higher temperatures will lower the dielectric constant⁹⁷².

A related approach was also reported by the Gross group⁹⁷³ in 2015, when they adapted carbene generation in solution on the FPOP flow system. Careful control over the exclusion volume guarantees that photoleucine and the protein CaM are mainly irradiated once, thus diminishing concerns of perturbing the solution equilibrium by the generated nitrogen gas and causing conformational change with excess labeling. The outcome is less modification for holo-CaM, consistent with its more compact conformation and significant labeling on Try, Asp and Glu, possibly owing to interactions of the protein with photoleucine, concentrating the reagent on the protein surface.

In 2017, Schriemer and coworkers⁹⁶⁹ refined the carbene platform to employ a single-shot laser with higher energy (i.e., 150 mJ) to avoid nitrogen perturbation and protein conformational change at the induced air-water interface. Prior to irradiation, the sample solution was snap-frozen by liquid nitrogen before laser irradiation to restrict radical diffusion, to maintain protein HOS, and to minimize quenching of carbene radical and increase modification. To establish residue specificity with carbene chemistry, the investigators, in a Herculean study, footprinted 777 peptides by using three different precursors, a negatively charged precursor 2 (Scheme 16), a neutral reagent 3 (3,3'-azibutan-1-ol, Scheme 16), and a positively charged reagent 4 (3,3'-azibutyl-1-ammonium, Scheme 16). The labeling trends are similar for the three reagents, and the bond insertion propensities generally are a function of side-chain polarity and size. Reagents 2 and 3 both favor Arg, Glu and Asp, whereas, the neutral reagent 4 shows higher reactivity with His (aromatic and neutral) in addition to the three residues, Arg, Glu, and Asp. Remarkably, many hydrophobic amino acids show noticeable modification (Figure 18c), supporting the high reactivity of carbenes, even with aliphatic groups. The nature of the carbene precursor also has an impact on the residue selectivity from participating in complex molecular interactions to increasing the local concentration of the precursor ion at the protein surface.

Manzi et al.⁹⁷⁴ tested a carbene precursor, 4-(3-(trifluoromethyl)-3H-diazirin-3-yl)benzoate (reagent 5, Scheme 16) that is more reactive than those from reagents 1 – 4. The investigators’ design involved installing an adjoining trifluoromethyl group (Figure 19a), leading to the use of less reagent and less irradiation time (i.e., 10 mM for 4 s irradiation). Further, the labeling efficiency improved in comparison to the use of photoleucine at 100 mM and 16 s irradiation. The improved reactivity is due to increased stabilization of the carbene radical by the added trifluoromethyl group and increased hydrophobicity but little change in zwitterionic character. They tested the new reagent on an unknown protein complex (i.e., the deubiquitinating enzyme ubiquitin specific protease 5 (USP5) upon binding with di-ubiquitin (di-Ub)). USP5 is a multi-domain cysteine protease including two ubiquitin associated domains (UBA) and a Zn-finger ubiquitin-binding domain (ZnF-UBP). Previous studies showed the binding stoichiometry between USP5 and di-Ub to be 1:1, suggesting additional binding sites than those in Znf-UBP. These sites were not identified. The investigators footprinted the C335A mutant of USP5 with carbenes in the presence and absence of one equivalent di-Ub and observed distinct binding regions (Figure 19b), which were mapped onto the X-ray structure of USP5 (Figure 19c). The catalytic domain was shown to be the other binding site and, additionally, a remote conformational change for the region represented by peptide G606-K630 was found. The design of new, successful radical precursors indicates that there are more opportunities for improvement and application, emphasizing the potential of carbene footprinting as an effective and accurate structural probe for HOS of proteins.

Figure 19.

(a) Extent of labelling of a range of proteins with reagent 1 (100 mM, 16 s irradiation), and reagent 5 (10 mM, 4 s irradiation (*1 s in the case of CaM)). (b) Fractional modification by reagent 5 of USP5 peptides in the presence (black bars) and absence (white bars) of di-ubiquitin. Error bars are ± standard deviations (n = 3) and significant differences (Student’s t-test, p < 0.05) are highlighted with a red dot. (c) Model of USP5 (based on PDB 3IHP) showing the locations of the five peptides (red) that are masked from labelling by di-ubiquitin binding and their relative locations to the ZnFUBP and catalytic domains. Reprinted with permission from Ref.⁹⁷⁴ Copyright 2016 Springer Nature.

The pathways for carbene chemistry may involve several radical intermediates whose structures and reactivities can be tailored by using different precursors. For example, it is possible to generate by photolysis not only singlet⁹⁷⁵ and triplet carbenes⁹⁷⁶ but also diazo isomers that further decompose into carbocations.^977,978 Although a singlet carbene preferentially inserts into O-H, N-H and S-H bonds,⁹⁷⁷ it is challenging to pinpoint the dominant pathway just from the nature of the modified products. Insertion into Thr can be done by singlet carbene through O-H bond insertion or by triplet insertion into the methylene group located on the amino acid side chain. In addition, the small energy difference (i.e., ~ 2 kcal/mol⁹⁷⁹) between the two states add complications because mixtures of products can form. Because heteroatom-containing residues are usually the favored sites of reaction (e.g., Glu, Asp, Tyr and Arg), the singlet state may be favored; however, blended pathways are more likely. A general route is shown in Scheme 17.

Scheme 17.

Proposed carbene reaction pathway

Carbene footprinting has high potential in structural biology. The labeling time is shorter than 10 ns,⁹⁸⁰ faster than most protein folding. Irradiation on a flow-system or excitation after snap freezing eliminates deceptive modifications originating from carbene insertions in a protein that has undergone a protein conformational change. Compared to the hydroxyl radical, the short survival time of carbenes owing to reaction with solvent water obviates the need for a scavenger. Carbene generation by diazirines is at a less damaging wavelength to proteins (i.e., ~ 350 nm). In addition, most carbene precursors do not react with proteins prior to laser irradiation (unlike H₂O₂ for ^●OH), and this lack of reactivity minimizes background interference. Furthermore, the resultant mass shift for carbene modification can be adjusted to be bio-orthogonal by tailoring the precursor design. The physical properties of reagents, however, can favor preconcentration on the surface of the protein, possibly delivering biased residue preference or even information loss. Those properties can also be chosen advantageously to promote binding in lipid membranes, permitting footprinting of transmembrane proteins. As this field develops and more diverse carbene reagents are implemented, a better understanding of interactions will emerge to permit rational design of new carbenes. New footprinting reagents that can target specific residues or provide comprehensive coverage are expected in the future.

5.5. Trifluoromethyl Radical

5.5.1. Biological Relevance and Applications

Fluorine-containing compounds are extremely rare in biology; only five entities containing F have ever been identified.⁹⁸¹ The most common molecule is fluoroacetate, which is found in many tropical plants as a toxin. Footprinting reactions that insert either fluorine or fluorine-containing substrates may be advantageous because fluorine is the most electronegative⁹⁸² (Pauling Electronegativity = 4.0) common substance and has a small radius (1.33 Å), not so dissimilar to that of H,⁹⁸³ allowing F to be a surrogate for H.

Introduction of fluorine alters existing compounds to be more metabolically and thermally stable, more lipophilic, and possibly more interactive with a targeted protein⁹⁸⁴ through contact with hydrophobic patches. Therefore, anthropomorphic fluorinated compounds already play a role in pharmaceuticals, up to 20-25%⁹⁸¹. One indispensable integrant is the trifluoromethyl group (CF₃) found in several drugs (e.g., Celebrex (Pfizer) and Sarafem (Eli Lilly)).

The CF₃ group was utilized early in several applications in structural proteomics. In 1962, Singer⁹⁸⁵ developed affinity labeling, in which a targeted protein non-covalently interacts with a probe molecule (e.g., a drug). The probe is comprised of the drug plus a reactive motif installed by chemical synthesis. The probe is activated so that it covalently binds with the protein in a similar way as the drug to “mark” the binding site.

The Brunner group⁹⁸⁶ advanced the approach by introducing a CF₃ group in a diazirine-containing reagent, 3-trifluoromethyl-3-phenyldiazirines (TPDs). TPDs are chemically stable and exhibit high quantum yields for carbene formation upon UV irradiation. Even though the linear diazo rearrangement is inevitably formed upon UV irradiation, CF₃ and phenyl stabilize the linear diazo and prevent it from undergoing intramolecular rearrangements (e.g., to an arylalkyl diazo derivative that competitively diffuses away from the active site before decomposing to the carbene, leading to a complex labeling patterns) and diminished side-chain reactions.⁹⁷⁰

5.5.2. Radical Generation in Solution

Trifluoromethyl radicals were generated in late 1940s from the reaction of homogenized CF₃I⁹⁸⁷ in the presence of ethene. Because gaseous CF₃I is not easy to control and store, other radical precursors including Te(CF₃)₂⁹⁸⁸ and trifluoroacetic anhydride⁹⁸⁹ were later developed. High temperature and low radical yield limit their application, however. Alternatively, Togni’s⁹⁹⁰ and Umemoto’s⁹⁹¹ reagent are better choices, and they require activation by various transition metals (e.g., Ru(bipyridine)₃Cl₂^992-993 and CuI⁹⁹⁴). The increased efficiency of generation and the reactivity of the radical depend largely on the metal-containing compounds that are used.

Whereas, organic solvents are usually necessary to retain the activity of the catalysts. A more biological compatible ^●CF₃ precursor is the Langlois reagent, first reported in 1991⁹⁹⁵ and promoted by Baran and coworkers^996-997. The system consists of sodium trifluoromethanesulfinate (CF₃SO₂Na) and t-butylhydroperoxide (TBHP) as an oxidant. In the absence of transition metals, ^●CF₃ can be initiated from the oxidant and generated in an aqueous buffer. Trifluoromethylation on various heteroarenes including pyridines, uracils and xanthines occurs with moderate to good yields (33%-96%). In addition, a protein system, β-lactamase, could be successfully labeled in Tris buffer while maintaining the protein’s functionality. More recently, Fennewald et al.⁹⁹⁸ incorporated an addition catalyst, TPGS-750-M, to achieve a better yield under milder reaction conditions.

5.5.3. ^●CF₃-based Protein Footprinting: Residue Specificity and Proposed Reaction Pathway

Trifluoromethyl radical as a footprinting reagent, was implemented by Cheng et al.⁹⁹⁹ in 2017 to be used on the FPOP platform. The radical precursor is the water-soluble salt, NaSO₂CF₃ (Langlois reagent). The ^●CF₃ formation is initiated by the ^●OH, from photolysis of hydrogen peroxide. A likely mechanism is the ^●OH displaces ^●CF₃ by attack on the S=O bond to form HOSO₂⁻, the conjugate base of sulfurous acid, a mechanism different than that suggested in the original paper. The subsequent trifluoromethylation of proteins likely occurs by capping a protein radical (generated through H^● abstraction by either ^●OH or ^●CF₃ (Figure 22).

The investigators tested the ability of the ^●CF₃ to footprint neuropeptide Y (18-36), apo-/holo-myoglobin (aMb/hMb), and VKOR, a transmembrane protein, on the FPOP platform. After laser irradiation, multiple CF₃-adducts were formed for the intact protein as evidenced by mass shifts of +67.987 Da for each CF₃ addition (Figure 20a). Notably, the trifluoromethylation of VKOR occurs selectively on the solvent-exposed region rather than in the transmembrane regions. Though ^●CF₃ is hydrophobic, it is generated from a water-soluble precursor that is not membrane permeable. Given the hydrophobicity of ^●CF₃, its generation in the “unfriendly” aqueous environment explains, in part, its high reactivity. The addition of a CF₃ group to the protein increases the hydrophobicity of the peptides formed by digestion, shifting their reversed-phase retention times to longer values, a phenomenon that is different for ^●OH footprinting, where the peptides have higher hydrophilicity and usually elute earlier than their corresponding unmodified peptides (Figure 20b).

Figure 20.

(a) Deconvoluted mass spectrum of CF₃-modified aMb. (b) EIC of unmodified and CF₃-modified peptide 32–45 of aMb. (c) Comparison of the extent of modification at the residue level for hMb and aMb. (d) A key mass shift of +68 Da is consistent with an unusual Gly labeling, here CF₃ modification on Gly-276 of peptide 267–83, a glycine with large SASA. Reprinted with permission from Ref.⁹⁹⁹ Copyright 2017 John Wiley & Sons, Inc.

The ^●CF₃ can react with 18 out of 20 different amino-acids residues, except Met and Cys, showing its complementary nature with ^●OH, which reacts rapidly with Met and Cys. There is some evidence that residues containing aliphatic side chains, which are inert with ^●OH (e.g., Gly and Ala), are reactive with ^●CF₃ (Figure 20d). The reactivities with various residues, however, are not identical. ^●CF₃ is highly electron deficient and, thus, preferentially reacts with aromatic side chains (e.g., Trp, Tyr, His and Phe). Furthermore, ^●CF₃ successfully reports the structural difference in aMb/hMb, as expected for a good footprinter. As found in previous studies, region 80-96 undergoes significant conformational changes in the conversion between the two states, and that change is confirmed by ^●CF₃ footprinting on His 81/82/93 (Figure 20c). Similarly, His 119 also undergoes more modification in the apo-state whereas other residues react to comparable extents in both the apo and holo states.

Trifluoromethyl radical footprinting appears to have interesting advantages. The broader and complementary residue coverage compared to ^●OH makes it a useful candidate in the structural biology “tool-box”. Combination of ^●CF₃ and ^●OH in tandem allows more comprehensive characterization of the residues on a targeted protein than either radical alone, therefore providing a better opportunity to capture subtle structural changes. In addition, ^●CF₃ can survive in the presence of typical radical quenching reagents^{996, 999} like dimethyl sulfoxide and retain its reactivity. In many membrane protein systems where a detergent is necessary for solubilization, ^●CF₃ footprinting has potential provided a precursor can be redesigned that will partition from water to the membrane (detergent). One limitation of using the Langlois reagent as precursor is the need for initiation by ^●OH. Reagents that quench ^●OH will also lower the trifluoromethylation yield. Alternative fluorination reagents are expected in the future.

5.6. Iodine Radical

5.6.1. Biological Relevance

In physiology, iodine plays essential roles in metabolic regulation of thyroid function, especially hormone production.^1000,1001 Nearly all ingested iodine is carried and circulated as iodide,¹⁰⁰² which in the presence of ROS or thyroid peroxidase (TPO)¹⁰⁰³, can be oxidized into molecular iodine, which oxidizes tyrosine residues on thyroglobulin to form thyroxine and tri-iodothyronine hormones.¹⁰⁰⁴ The reactive specie causing iodination, however, are not settled, although a likely one is the iodine radical.¹⁰⁰⁵

Iodination strategies have been widely adopted in radioimmunology.¹⁰⁰⁶ Iodide radical appears to react also with histidine residues¹⁰⁰⁷ under more alkaline conditions but giving a lower yield than with Tyr.

5.6.2. Radical Generation and Applications in Biology

Generation of iodine radical in a bio-relevant condition is usually achieved by TPO^1008,1009 or UV photolysis¹⁰¹⁰ of iodine-containing compounds. When iodine is directly incorporated into DNA and proteins, the photo-labile C-I bond can produce useful radicals to address mechanistic and structural questions concerning the region surrounding the C-I bond. The generated iodine radical, on the other hand, seems to serve as only as a leaving group.

In 2005, Hiroshi and co-workers¹⁰¹¹ utilized 5-iodouracil as a substitute reagent for DNA synthesis. UV photolysis of the modified DNA causes a strand break to give products that serve to probe local structure.¹⁰¹² A subsequent study further elaborated the method to include interactions of DNA and proteins.¹⁰¹³ A photochemical reaction forms cross-links that report on interactions between proteins and the targeted DNA both in vivo and in vitro.

Ly and Julian^1014-1015 implemented starting in 2008 a similar strategy for footprinting proteins. Starting with electrophilic iodination of a protein, the investigators isolated and photolyzed the intact protein at 266 nm in a linear ion trap mass spectrometer to generate odd-electron species that modify nearby Tyr and His resides and direct subsequent CID fragmentation. Radical-directed dissociation dominates the cleavages via characteristic pathways, primarily backbone fragmentations at modified Tyr residues. Secondary backbone cleavages were also observed in proximity of modified Tyr, especially if Pro is present. Similar chemistry occurs via His iodination in the absence of Tyr. The informative fragments can report on the presence of D-amino acids,¹⁰¹⁶ elucidate protein tertiary structure, and assist with identification of proteins in proteomics experiments. Later on, Ranka et al.¹⁰¹⁷ characterized the free iodo-tyrosine rearrangement pathways under UVPD to show that loss of iodide radical leads to a high-energy radical in the aromatic ring that engages in hydrogen/proton rearrangements. This study provides fundamental understanding of UVPD-based top-down analysis of iodinated proteins. More applications and detailed analysis are expected for these novel approaches.

5.6.3. I^●-based Protein Footprinting: Residue Specificity and Proposed Reaction Pathways

The iodine radical has potential to be an effective fooprinting reagent given its specific reactivity towards tyrosine and histidine. In footprinting, one precursor for the free radicals is 4-iodobenzoic acid, an organic iodide with some water solubility. The I^● is formed presumably in concert with a ^●C₆H₄COOH (represented by R^● in Figure 22) by photolysis of I-C₆H₄-COOH at 248 nm. The carboxylphenyl radical likely abstracts an H^● from the OH of Tyr or from the NH of the imidazole ring to give a stabilized radical that is subsequently “capped” by reaction with I^● to give an iodinated protein (Figure 22), although there may be other mechanisms. This chemistry can be initiated on an FPOP platform, as shown by Chen et al.¹⁸¹ in the presence of a histidine scavenger to control the I^● lifetime and thereby reduce labeling-induced conformational changes.

After laser irradiation, the hMb and aMb were analyzed as intact proteins (Figure 21), where a single addition of I^● leads to 125.90 Da mass shift. Modifications occur to give more mono-, bi- and tri-iodinated aMb than for hMb, consistent with a more exposed conformation for aMb resulting from heme removal. In contrast to ^●OH oxidation, the nearly exclusive iodination on Tyr and His simplifies the interpretation and data analysis, making possible a coupling with top-down MS for residue-level information. Top down ECD fragmentation on a 12 T FT-ICR gave a *z₈⁺ ion that showed that iodination of Tyr146 was 26% for aMb but not detectable for hMb. On the contrary, iodination of Tyr103 (difference between *z₅₅⁺ and *z₄₄⁺) was similar for both proteins. The results can show that Tyr146 must be involved in heme-binding.

Figure 21.

(a) Full ESI mass spectra of iodinated hMb. (b) Full ESI mass spectrum of iodinated aMb. The unmodified, mono-, di-, and tri-iodinated species of the 15^th charge state are indicated by the number of stars. Reprinted with permission from Ref. ¹⁸¹. Copyright 2012 Springer Nature.

Owing to the limited sequencing efficiency of ECD technology at the time of the research, modified residues located in the middle region of a protein would be better analyzed with traditional bottom-up methods. His residues, although 30-100 times less reactive than Tyr,¹⁰¹⁸ can also be addressed via a bottom-up strategy. Iodine radical footprinting may become a sensitive structural probe as is seen in an application to two other protein systems (i.e., apo-/holo-carbonic anhydrase II and EDTA-/zinc-insulin). The capability of this approach to yield structural information is seen by the abundant labeling of the region that carries the incipient f helix in apo state and the remarkably low to non-detected footprinting of that region of the holo state.

Although the coverage afforded by the iodide radical is limited, footprinting two targeted residues can answer specific questions with easier data analysis. Compared to ^●OH footprinting, the unique and larger mass shift increases the confidence of assigning the modification sites. Moreover, the precursor of the iodine radical doesn’t react with proteins, which minimizes the background modification and allows simplified post-label sample handling. Unlike specific amino acid labeling, the modifications on the FPOP platform are fast, alleviating concerns about labeling-induced conformational changes.

5.7. Summary and Perspective

Radical footprinting is an effective means to acquire HOS information for proteins. Radical reactions are much faster (nano to milliseconds) than reactions of a conventional chemical reagent that modifies one or a few amino acid residues. Therefore, free radical footprinting will deliver fast, broad, and less biased information by largely avoiding questions of labeling-induced perturbation of protein structure. In addition, the short timescale can allow a sensitive report of subtle structural and even dynamical changes. An advantage of the labeling speed is the successful temperature jump, two-laser experiment on the FPOP platform where ^●OH footprinting can successfully track the folding of barstar at times as short as a few tenths of millisecond.¹⁰¹⁹ An example of versatility is the tracking of multiple intermediate oligomeric states during amyloid beta aggregation.¹⁰²⁰

Although different radicals react via distinct pathways, they share similar features (Figure 22a). Many radicals (e.g., ^●CF₃, I^●, SO₄^−•, CO₃^−•, ^●OH) permit modification by forming a protein-centered radical that is subsequently quenched, whereas others (e.g, ^●OH and carbene diradicals) can directly add to a protein molecule. Whatever the process, the informative modifications are usually by an irreversible covalent bond to an amino acid side chain, facilitating downstream sample handling and analysis comparing to that of reversible labeling. A protein footprinting-based “toolbox”, that is being built for the future, will contain radicals that show different residue specificity to address specific biological questions (Figure 22b). For example, when protein systems are rich in Glu and Asp (e.g., as for calcium-binding proteins), GEE or carbene footprinting may be a good choice with quick and simple data analysis and high throughput. More importantly, investigators can customize reagent combinations for targeting specific residues or for obtaining high coverage. Other radical species (e.g., nitric oxide radical) will be tested, and new radical precursors will emerge soon to footprint both soluble and transmembrane proteins. A “toolbox” with diverse reagents and associated modeling software will aid investigations of complex biological questions with effectiveness, sensitivity, and accuracy.

Figure 22.

Summary of radical-based footprinting reagents of (a) proposed pathways and (b) residue specificity.

6. Applications that Utilize Fast Labeling Approaches

Since the initial introduction of protein footprinting by free radicals in 1999,⁸⁸⁴ combinations of different reactive species and several experimental designs have successfully demonstrated the reliability of fast-labeling approaches and its ability to address various biological questions. In this section, we will discuss recent advances since the Chance review in 2007¹⁷⁰ and highlight key developments.

6.1. Mapping Epitopes

Monoclonal antibodies (mAb)¹⁰²¹ are important biomacromolecules owing to their broad applications in analytical assays^1022-1024 and their profound potential in therapeutics.^1025-1026 Antibodies bind to antigens with high specificity and strong affinity, and this property has been utilized for some time in immuno-histochemistry¹⁰²² and in the enzyme linked immunosorbent assay (ELISA),¹⁰²³ assays that are now standard practice around the world. The highly specific recognition also makes it possible to develop targeted therapeutics.^1025-1026 These advances rely on a deep understanding of antibodies from both a structure and functionality perspective.

Interacting surfaces in antibodies and antigens are termed as paratopes and epitopes, respectively. Paratopes mainly consist of loop regions of 10–15 amino acid residues and are termed “complementarity determining regions” (CDRs). Paratopes of a typical mammalian antibody are found in six CDRs, three from the heavy chain and the remainder from the light chain.¹⁰²⁷ Although the combinations of amino acid residues that comprise the six CRDs are nearly unlimited,¹⁰²⁸ the structure and position of the paratopes are usually well-defined in accord with the rigid scaffold of the antibodies.

Antigens, however, differ dramatically from antibodies and from each other. Epitopes in antigens do not exist in a specific structure. The binding site for the antigen (epitope) depends on the location of the actual binding interface that interacts with antibodies. In other words, epitopes can be located anywhere on the antigen surfaces.¹⁰²⁹ Moreover, an epitope can be either linear (consecutive) or conformational (assembled).¹⁰³⁰ A consecutive epitope is composed of a single continuous structural motif, whereas an assembled epitope consists of multiple and non-adjoining binding motifs. As the majority of reported epitopes are assembled,^1029-1030 a complex binding scheme usually pertains and requires responsive methods for mapping epitopes.

X-ray crystallography,¹⁰³¹ NMR¹⁰³² and alanine-scanning mutagenesis¹⁰³³ are the most commonly used high-resolution biophysical approaches in mapping epitopes. X-ray crystallography and NMR reveal the binding regions, allowing assignment of epitopes as surfaces on the basis of proximity of the antibody and antigen. Such contact, however, may not always lead to structural recognition. Mutagenesis maps the epitope from a functional aspect, but that approach may not be informative for high order structure and the amino acid substitutions may affect the protein HOS. Alanine scanning, whereby amino acid residues are replaced one at a time or in a group (called “shave” analysis) coupled with binding assays to judge the effect of the mutation, locating functional epitopes with high confidence. Other functional assays including but not limited to studies of synthetic peptides representing the antigen¹⁰³⁴ or represented by antigen truncation¹⁰³⁵ were also developed as high throughput but low spatial resolution approaches as compared to the alanine scanning.

To complement these approaches, MS-based epitope mapping methods were developed substantially and now are regarded as reliable approaches.^1036-1038 Two different schemes were originally used; that is, epitope excision¹⁰³⁶ and epitope extraction.¹⁰³⁹ Both approaches utilize the specific binding between antigen and immobilized antibodies. In an epitope excision workflow, an immunocomplex is formed in solution, after which a protease is added to digest the antigens. The peptides from the antigen that contain the epitope remain bound with an immobilized antibody, and they are then extracted and characterized by MS. Epitope extraction, however, starts by digesting antigens and utilizes the resulting peptides in a peptide-immobilized antibody complex. In this case, those peptides that bind the antibody represent epitopes.

In both schemes, the investigator identifies epitope peptides by using MS and then sequences them by MS/MS. The amount of antigens required can be as low as sub-pmol¹⁰⁴⁰. MS-based approaches significantly lower the sample amount and reduce the need for sophisticated sample preparation and data analysis as compared with X-ray crystallography and NMR, while preserving mid-to-high spatial resolution from a structural view. MS-based approaches locate potential epitopes with significant less effort, providing valuable guidance for designing proceeding mutagenesis and other biological functional assays to locate functional epitopes confidently.

More recently, MS-based epitope mapping approaches utilize HDX to label the protein chemically and monitor changes in SASA that result when antigens bind with antibodies.¹⁰⁴¹ Thanks to the high binding affinities for many immunocomplexes, off-rates do not heavily distort the HDX kinetics. Both the antigen and the immunocomplex can be footprinted under native-like conditions, minimizing potential perturbations of structure. Although the HDX rates are significantly slowed (quenched) at a pH of 2.5 and 0 °C,^184-185back-exchange cannot be avoided. Proteases that function under acidic environments are generally limited to pepsin and Fungal protease XIII,¹⁰⁴² both of which are non-specific, and the resulting peptide mixture is complicated and challenges the data analysis. Moreover, H/D scrambling during MS/MS analysis limits the spatial resolution to the peptide level; that resolution can be increased possibly to the residue level by using ECD¹³⁵ and ETD²¹⁷, which presumably do not scramble the D labels.^216-217

Epitope mapping by fast labeling methods was first demonstrated by Jones and Gross¹⁰⁴⁴ in 2011, where the epitope of serine protease thrombin was footprinted with hydroxyl radicals on the FPOP platform. Two regions show binding-induced protection as evidenced by the decrease in extent of modifications (representative results are in Figure 23a), where five distinct peptides exhibit increased protection upon binding (darker bar represents unbound and lighter bar represents bound). The protection spans 42 amino acid residues, but such protection seems too extensive considering the size of thrombin. Taking advantage of the irreversible labeling of FPOP chemistry, the investigators could further assign protection at the residue level (results in Figure 23b). Among 14 resolvable residues, protection in the antibody-bound form occurs between residue D133 and Y150; those results are mapped in Figure 23c. Proposed epitopes agree well with those from an earlier HDX study.¹⁰⁴¹

Figure 23.

(a) Extent of modification of thrombin alone (darker bars) and antibody-bound thrombin (lighter bars) for the five peptides that span the 130-171 region. (b) Extent of modification of thrombin alone (darker bars) compared to antibody-bound thrombin (lighter bars) at the residue level. (c) Structural model of thrombin (PDB file 2AFQ¹⁰⁴³) with the proposed epitope colored in red and the loop regions colored in blue. The individual residues that show increased modification for antibody-bound thrombin are specified. (d) Extent of modification of peptides from thrombin alone (darker bars) and antibody-bound thrombin (lighter bars) show increased solvent accessibility in the antibody-bound form. Reproduced with permission from ref ¹⁰⁴⁴. Copyright 2011 American Chemical Society.

Moreover, four residues (Figure 23b) and three peptides (Figure 23d) show deprotection upon forming the immunocomplex. This deprotection, not observed with HDX, is assigned as a remote conformational change. HDX monitors both backbone amide hydrogen bonding and solvent accessibility whereas FPOP emphasizes the latter. If allosteric or remote conformational changes do not involve changes in hydrogen bonding but occur as reorientation of amino acid residue side chains, they will not be seen by HDX.

Subsequent studies of similar design reveal the epitope of human epidermal growth factor receptor (EGFR)-Andectin 1,^1045-1046 IL-6R-Adnectin 1&2,⁷³⁸ and human immunodeficiency virus 1 gp120 envelope glycoprotein,¹⁰⁴⁷ demonstrating even more the efficacy of fast labeling approaches and making a case for generality.

As mentioned earlier, changes in SASA/hydrogen bonding may not directly associated with biological recognition. Binding-induced protection can occur either from binding at the epitope region or from remote conformational change sites, which can be differentiated by site-specific mutagenesis and functional assays or possibly by crosslinking.

A 2017 study by Li et al.¹⁰⁴⁸ applied orthogonal methods including FPOP, HDX, and alanine shave mutagenesis and determined energetic epitopes of an antibody/Interleukin-23 interaction. By examining the system with FPOP footprinting, they found that five regions of IL-23 p19 domain show protection (results mapped onto the crystal structure in Figure 24a). Similarly, HDX reveals four slightly different protected regions, three of which overlap with those reported by FPOP (Figure 24b). Imperfect overlap motivates subsequent alanine shave analysis, where a series of mutants can be designed based on MS-analysis, enabling confirmation of the residues/peptides identified by the two MS methods. Four distinct mutants show significant lower binding affinities as compared with the wildtype protein. Note that all four mutants retain near wildtype-like conformations as confirmed by optical and NMR approaches. Thus, this comprehensive approach points to the first two regions of IL-23 p19 domain as epitopes, whereas protection for regions 124-139 (by FPOP) and 145-153 (by HDX) are from remote binding-induced conformational change. This study is a good example of the workflow of fast radical labeling and HDX MS to guide mutagenesis and functional assay for epitope mapping.

Figure 24.

(a) Epitope regions determined by FPOP mapped on the crystal structure of IL-23 p19 domain. Color code: no significant difference (gray), minor epitope region (cyan), and major epitope region (blue). The p40 subunit is colored in purple. (b) Epitope regions determined by FPOP, HDX, and Ala shave energetics are mapped on the linear sequence of the IL-23 p19 domain. M3, D3, E9, L9 stands for M35A, D36A, E93A, L97A, respectively. Adapted with permission from ref ¹⁰⁴⁸. Copyright 2011 American Chemical Society.

In addition to the advantages of low sample amount, mid-to-high resolution spatial information and high-throughput, fast labeling approaches can bring amino-acid-residue level spatial resolution to epitope mapping without the need of specialized MS instrumentation of ECD or ETD fragmentations in HDX or the effort needed for alanine-scanning mutagenesis. Irreversible labeling is not subject to the back-exchange of the label as applies to HDX, and the sample can be examined with any fragmentation method without concern for scrambling of the label. Moreover, the labeled protein can be digested with a variety of enzymes at their preferred pH’s, whose cleavages can be highly specific (e.g., trypsin), simplifying the data analysis. Fast labeling approaches are sensitive to subtle changes in the SASA of amino acid side chains, which may not be accessible to HDX.

FPOP, which generates ^●OH through laser photolysis of H₂O₂, controls the lifetime of the radicals to avoid perturbation of native protein conformation. The concentration of ^●OH needed for comprehensive footprinting is not large and is readily generated by the photolysis (~ 1 mM of radicals can be formed). In a FPOP-based epitope mapping approach, protection is evaluated through changes in the extent of modification, requiring that addition of the antibody to the antigen solution does not increase reactive sites so extensively as to quench the radicals and lower the modification extent of the antigen. This is particularly a problem when the antigen is small compared to the large antibody. This potential bias can be overcome by either introducing small molecule radical dosimeter¹⁰⁴⁹/reporter peptide¹⁰⁵⁰ to calibrate the effective ^●OH concentration or by keeping the mass of total protein nearly the same in the control and the test experiments. An excellent way of doing this would be to include a nonbinding antibody when footprinting the antigen as control. To generate ^●OH through other approaches like synchrotron water radiolysis may also be a good alternative.⁸⁸⁴

6.2. Tracking Protein Folding/Unfolding

Dating back to 1961, the ribonuclease refolding experiment by Christian B. Anfinsen² showed that small globular proteins can fold, without assistance from other biomolecules, to their free energy minimum state. Soon after, Cyrus Levinthal¹⁰⁵¹ argued that such protein folding would be too slow because the protein has too many possible conformations to search before finding its energy minimum, known today as Levinthal Paradox. Following the ribonuclease refolding experiment, Anfinsen later proposed a thermodynamic hypothesis wherein the protein native structure is only determined by the protein’s amino acid sequence.¹⁰⁵² Decades of structure determination resulted in the accumulation of over 140,000 high resolution structures in the Protein Data Bank (PDB). It is now accepted that protein folding is driven primarily by interactions including hydrogen bonding, van der Waals interactions, backbone dihedral angle preferences, hydrophobic interactions, and electrostatic interactions.¹⁰⁵³ Understanding protein folding pathways and their corresponding intermediates, however, remains a huge problem that now has become a field of research onto itself.

Modern views of protein folding take an energy landscape as a foundation for the folding problem. Rather than a defined folding pathway, multiple routes facilitate protein folding from an unstructured chain of amino acids to a native conformation; that is, a protein negotiates a folding funnel.^1054-1055 Modern structural prediction algorithms pioneered by Critical Assessment of protein Structure Prediction (CASP) also contribute greatly to addressing this problem.¹⁰⁵⁶ All these developments were reviewed previously.^1057-1059

Experimentally, folding or unfolding is followed, after introducing a perturbation as a function of time, through characterization by several methods;¹⁰⁶⁰ perturbations that can be used are a sudden change typically in temperature,^{893-894, 1061-1062} denaturant concentration,^1063-1064 pH,¹⁰⁶⁵ or pressure¹⁰⁶⁶. As secondary structural motifs including α-helices and β-sheets usually form in 0.1 – 10 μs,¹⁰⁶⁷ and some small proteins can fold as fast as tens of microseconds¹⁰⁶⁸, it is important that the characterization/labeling method is fast enough to follow the process and capture any intermediates. For decades, several approaches including circular dichroism,¹⁰⁶⁹ fluorescence,⁶⁶ FI-IR,¹⁰⁷⁰ and NMR^1071-1072 have been most extensively used. Although lower in spatial resolution, the detection time limit for optical approaches can be as low as 10 fs,¹⁰⁷³ a time that is ideal for tracking rapid structural transitions. On the other hand, using rapid mixing and multidimensional NMR coupled with HDX has been successful in characterizing folding processes on the time scale of milliseconds with high spatial resolution.¹⁰⁷² Recent developments of NMR instrumentation allow observation of folding intermediates on the timescale of 100 μs,¹⁰⁶⁶ greatly elevating the spatial resolution over that achieved by optical approaches. Modern single molecule measurements¹⁰⁷⁴ and mutational approaches¹⁰⁷⁵ also shed light on folding.

Following protein folding/unfolding by MS was first demonstrated by Chait and coworkers¹⁰⁷⁶ in 1990, who electrosprayed cytochrome C at different pHs. They observed different charge-state distributions when the protein was introduced in different conformations; an unfolded form favors higher charge states. This observation enables MS to follow, in a simple and global way, the protein conformational changes upon perturbation, a representative one being myoglobin reconstitution coupled with online continuous-flow labeling.¹⁰⁷⁷ In the same decade, HDX was married to MS,^{203, 1078} and the combination was used soon after to address lysozyme folding. ¹⁰⁷⁹ Further development of a “rapid mixing” apparatus made possible the measurement of folding kinetics by HDX-MS,^{1071-1072, 1080} and, in combination with protease digestion, allowed peptide-level spatial resolution.⁷⁰ All these efforts empower MS in protein-folding analysis and were reviewed elsewhere.^{260, 777, 1059}

The application of free-radical footprinting in folding studies was first demonstrated by Chance, Woodson and coworkers^880-881 for RNA, where a “stopped-flow” apparatus combined with synchrotron-generated ^●OH radicals probed ribozyme folding dynamics. Proteolysis to form peptides and MS to locate the footprinting provided some regional specificity.

This approach was later applied to protein unfolding. By varying urea concentration, Maleknia and Downard¹⁰⁸¹ followed apomyoglobin denaturation through a series of ^●OH footprinting experiments. Upon protease digestion, they showed that helices A and B/C unfold in a cooperative fashion whereas helix G unfolds locally at lower denaturant concentration. Thermodynamic parameters were obtained, and they agree with those from fluorescence-based measurements. Although measured under equilibrium conditions, this study for the first time demonstrated the possibility for radical labeling in addressing protein folding/unfolding. Later studies by Poor et al.¹⁰⁸² utilized this idea and demonstrated successfully the refolding the paramyxovirus fusion (f) protein by changing equilibrium temperatures and determining the free energy landscape of bacterial immunity protein (Im7) folding by combining mutational analysis with free radical labeling¹⁰⁸³.

Although the approaches in equilibrium measurements are useful, it is challenging and likely more informative, to examine protein folding kinetically. Taking advantage of the Hambly and Gross⁸⁴¹ demonstration of ^●OH protein footprinting through laser photolysis of H₂O₂ in a flow system, Konermann and coworkers¹⁰⁸⁴ integrated rapid mixing with ^●OH labeling to study the protein folding/unfolding in a kinetics experiment. In 2009, Stocks and Konermann¹⁰⁸⁴ first built a continuous-flow rapid mixing device for probing protein conformational changes during unfolding, shown in Figure 25a. The rapid mixing apparatus has two mixing “tees”. Using three syringes, the investigators mixed native protein (hMb), H₂O₂, and denaturant HCl in a continuous fashion. By tuning the flow rate, pulsed laser frequency, the distance between the second mixing tee M2 and the window for laser irradiation, the delay time between protein denaturation and ^●OH labeling can be manipulated. In this study, four different time points, 50 ms, 500 ms, 10 s and 5 min, were investigated. Using tryptic digestion, the investigators followed oxidative modifications as a function of denaturing/unfolding time of hMb regional specificity. Peptide T7 represents the region that unfolds initially upon adding denaturant (Figure 25b). Peptides T2, T10 and T13, on the other hand, stay relatively protected up to 500 ms after mixing with HCl. Peptide T16 represents the region that unfolds between 50 to 500 ms. The hMb becomes completely unfolded after mixing with HCl for 5 min. On the other hand, by starting with denatured protein and replacing HCl with renaturing buffer, the investigators could study the folding of cytochrome c,¹⁰⁸⁵ the folding and dimerization of S100A11,¹⁰⁸⁶ the folding of an integral membrane protein bacteriorhodopsin,¹⁰⁸⁷ and even sub-millisecond folding of apomyoglobin.¹⁰⁸⁸

Figure 25.

(a) Schematic illustration of the three-syringe, continuous-flow, rapid mixing setup for oxidative ^●OH labeling. (b) Normalized oxidation levels of tryptic peptides plotted as a function of unfolding time t. Blue boxes highlight three peptides that retain considerable protection at 50 and 500 ms whereas the peptide highlighted by a red box is protected at 50 ms but not at 500 ms. Adapted with permission from ref ¹⁰⁸⁴. Copyright 2009 American Chemical Society.

Recently, the delay time was further shortened to the microsecond timescale by a newly designed microfluidic mixer, and the unfolding was followed by FPOP. The data reveal two kinetic phases of egg lysozyme that occur before 1 ms.¹⁰⁸⁹ The requirement for low sample amounts and the achievement of regional specificity, coupled with kinetic capability open new possibilities for protein folding studies even on the sub millisec time frame.

In 2010, Chen, Rempel and Gross⁸⁹³ reported a design that combined FPOP with a laser-induced temperature-jump to probe sub millisecond folding. As demonstrated in Figure 26a⁸⁹³, two lasers were controlled by a signal generator and a delay circuit, and their beams were aligned to intersect sequentially at the same transparent window on the flow capillary. A Nd:YAG laser with a Raman shifter produced a “heating pulse” at 1900 nm, a wavelength that is absorbed by water to induce a temperature jump of ~ 20 °C. A fraction of second later, an excimer laser at 248 nm generated ^●OH that footprints the protein as a function of time. By controlling the delay time between firing the two lasers, it is possible to track the protein folding/unfolding kinetics with high spatial and time resolution. The delay time of this two-laser approach is no longer limited by fluid dynamics, as in rapid mixing, but by the time separation between the T jump laser and the ^●OH labeling laser. By tuning the reaction conditions, ^●OH should be able to footprint the protein irreversibly in times as short as 1 μs^{878, 893} or less, making it possible to follow folding dynamics of fast folding proteins.

Figure 26.

(a) Schematic illustration of the flow system intersected by two laser beams at a transparent window, as part of a temperature-jump apparatus. Adapted with permission from ref ⁸⁹³. Copyright 2009 American Chemical Society. (b) Proposed intermediate for barstar early folding. Residues colored in red, pink and gray represent residues that are closely associated with hydrophobic core, weakly associated with hydrophobic core and not involved in early folding intermediate, respectively. Adapted with permission from ref ⁸⁹⁴. Copyright 2009 American Chemical Society.

The investigators successfully footprinted the first intermediate state of barstar folding, which happens within 2 ms of the temperature jump.⁸⁹⁴ Changes of the modification fractions at the amino acid residue level revealed the key residues that serve as a nucleus for barstar early folding as seen in Figure 26b,⁸⁹⁴ where residues H17, L20 and L24 can be assigned as closely associated with the hydrophobic core around which barstar folds; residues I5 and F74 are weakly associated with the hydrophobic core. Residues W53, L88 and many others, however, are not involved in early folding, because their fractional modification does not change with delay time, indicating their solvent exposure does not change up to 2 ms after the temperature jump.

Radical species have the advantage of short reaction time, which is ideal for following fast processes. The ^●OH FPOP-based two laser approach brings high resolution to submillisecond protein folding studies,^893-894 which was not easily accessible prior to this work. Secondary structural motifs for some fast-folding proteins, however, can form as fast as 0.1 μs¹⁰⁶⁷ thus requiring an even faster probe. Carbene diradicals label protein at the nanosecond timescale,⁹⁷³ suggesting that those footprinters may be appropriate to address such problems.

Despite these advantages, the two-laser apparatus is only applicable if the temperature jump is sufficient to cause the protein to fold or unfold. Folding is more restrictive as few proteins are denatured at low temperature and become renatured upon heating. The change in temperature that the laser can deliver is also limited. From this perspective, although relatively slow,¹⁰⁸⁹ with certain exceptions, rapid mixing is more universal as it can introduce various perturbations. Recent development of theta-capillary emitter for the ESI source greatly shortens the mixing time and pushes the limit down to 270 ns – 27 μs.¹⁰⁹⁰ It was demonstrated that this novel development can be used in both protein folding studies¹⁰⁹¹ and HDX²²⁷ of proteins at μs timescale. These exciting developments open new possibilities for protein folding studies. By spraying denatured protein and deuterated renaturing buffer through a theta-capillary emitter and analyzing the resulting protein in an ETD-based top-down fashion, it may be possible to analyze protein folding with high spatial resolution. A disadvantage may be that the protein conformational change occurring upon ESI is not physiologically relevant.

6.3. Assaying Protein Aggregation

Studies of protein aggregation and its relationship to neurodegenerative diseases date back to 1910, when Fritz Heinrich Lewy first observed unusual protein aggregates in the brains of patients with Parkinson’s disease.^1092-1093 These aggregates were later termed as Lewy bodies¹⁰⁹⁴ and became the diagnostic for Parkinson’s disease. Nowadays, it is well understood that a Lewy body is composed of misfolded α-Syn. Alzheimer’s disease, as first described by Alois Alzheimer,¹⁰⁹⁵ is another major neurodegenerative disease induced by protein misfolding and aggregation of amyloid beta (Aβ)¹⁰⁹⁶. A 2016 report indicates there are approximately 6.2 million people globally who were affected by Parkinson’s disease.¹⁰⁹⁷ The number is 29.8 million for Alzheimer’s disease.¹⁰⁹⁷ Unfortunately, neither of these diseases have cures, and their pathologies are not well understood.^{1093, 1096, 1098-1099} Thus, it is important to develop novel approaches to locate and quantify protein aggregates, to study the aggregation mechanism, and to develop therapies.

Characterization of protein aggregates is challenging, especially at high spatial resolution. The final state of these aggregates, solid Aβ fibrils, for example, can be characterized with high resolution approaches including solution (labeled by D₂O and solubilized in dimethyl sulfoxide)¹¹⁰⁰ and solid-state NMR¹¹⁰¹, x-ray crystallography¹¹⁰² and modern cryo-EM^1103-1104. The low molecular weight soluble oligomeric intermediates that “come and go” during aggregation, although physiologically important, are not amenable to those techniques owing to low solubility, high heterogeneity, intrinsic disorder, instability, and high aggregation propensity. Some of these aggregates can be investigated by atomic force microscopy (AFM) to give a high-resolution morphological picture of early aggregation intermediates.¹¹⁰⁵ On the contrary, no site-specificity or spatial resolution are available when using such approaches.

Among all characterization approaches, fluorescence is the most widely adopted to follow protein aggregation. With the discovery of thioflavin T (ThT),¹¹⁰⁶ a dye that when it binds β sheet-rich structures exhibits an enhancement in fluorescence, has become the classic way of following aggregation.¹¹⁰⁷ The Aβ aggregation process can then be tracked and quantified both in vivo and in vitro. Given its broad applicability for characterizing different aggregation stages, ThT fluorescence is regarded as a characteristic of amyloid fibril formation.¹¹⁰⁸ Nevertheless, limited spatial resolution and possible distortion of the aggregation process by adding the dye have always been major concerns.¹¹⁰⁹

MS-based approaches can contribute to this field from two distinct points of view. MALDI-imaging enables label-free quantification of Aβ aggregates.¹¹¹⁰ Ion mobility MS is capable of separating different Aβ oligomers based on their collision cross sections in the gas phase.¹¹¹¹ HDX coupled with MS also reveals structural information of Aβ fibrils,¹¹¹² protofibrils,¹¹¹³ and even Aβ aggregation kinetics¹¹¹⁴. With top-down sequencing and ECD fragmentation, H/D scrambling is minimized, allowing residue-level information of Aβ oligomers.^1115-1116 These efforts were recently reviewed.^1117-1119

The ability to study Aβ aggregates by radical footprinting was first demonstrated in 2014 when Klinger, Chance, Axelsen and coworkers¹¹²⁰ utilized synchrotron-based HRF to examine the fibrils and prefibrillar forms of the 40-residue Aβ (Aβ_1-40). By comparing the protection factors for selected residues in prefibrillar and fibril Aβ_1-40, they mapped the footprinting data onto high-resolution solid-state NMR models and showed that the solution information obtained from HRF-MS is consistent with several core filament structural models elucidated in solid state, but not with other models that they rejected. Moreover, protection factor analysis supported the linear heterogeneity of the Aβ_1-40 fibrils; that is, two- and three-filament assemblies alternating along the length of the fibril.¹¹²⁰ Measured protection factors of the residues on the flexible loop regions of Aβ_1-40 are also consistent with the structural constraints from solid-state NMR studies.

Later on, Li et al.¹⁰²⁰ applied FPOP to study the aggregation process of 42-residue Aβ (Aβ_1-42). By following the Aβ_1-42 as a function of incubation time with FPOP, the investigators found a clear decrease in modification fraction shown in Figure 27a, consistent with a decrease in SASA owing to aggregation. By modeling the results with a two-nucleation/two-autocatalytic mechanism, they successfully fit the data (solid line in Figure 27a) and constructed a system composition plot as a function of incubation time, shown in Figure 27b. Plotting relative fractions for four components, the investigators proposed an Aβ_1-42 early aggregation mechanism, where Aβ_1-42 monomer rapidly assembles into paranuclei and further accumulates until a certain threshold. Upon passing that limit, the paranuclei self-catalyze a structural reorganization to deplete the monomers and form readily the mature fibrillar aggregates.¹⁰²⁰

Figure 27.

FPOP and kinetic modeling results for time-dependent aggregation of 42-residue Aβ (Aβ_1-42) at (a, b) global, peptide and residue levels. (a) Global level results for Aβ_1-42 aggregation. Solid curve is from a fitting model based on two autocatalytic reactions. (b) Concentrations for representative species (M-monomer, D-paranuclei, D*-protofibrils, D**-fibrils) as a function of incubation time based on kinetic simulation. (c, d, e) Peptide-level results for Lys-N digested Aβ_1-42, N-terminal region 1-15, middle region 16-27 and C-terminal region 28-42, respectively. (f, g, h) Residue level results for three representative residues, H6, F19/F20 and M35, respectively. Points in each plot represent experimental data (10 μM, pH 7.4, no agitation) and error bars are standard deviations from three independent trials. Solid and dashed lines in (c – h) are model fits independent of or constrained by the global rates, respectively. Adapted with permission from ref ¹⁰²⁰. Copyright 2009 American Chemical Society.

Upon digesting the Aβ_1-42, spatial resolution is elevated to regional and some residue levels. The N-terminal region remains solvent accessible throughout aggregation (Figure 27c) as seen by the minimal change in modification in that region during aggregation. The central (Figure 27d) and C-terminal regions (Figure 27e), however, actively participate in aggregation, as evidenced by significant decreases in modification fraction. Similar logic applies to residue-level interpretation, where SASA of H6 (Figure 27f) changes little during aggregation. Residue F19/F20 (Figure 27g) and M35 (Figure 27h) are actively involved in aggregation, and the investigators proposed that F19/F20 contributes a driving force for Aβ_1-42 aggregation by serving as a hydrophobic nucleation interface.¹⁰²⁰ Although some residue-level resolution was obtained, not all residues could be studied, confirming the need for alternative free-radical footprinters that prefer to react with other residues than those reactive with ^●OH and give complementary information.

MS-based approaches make possible the characterization of early aggregation states, especially transiently existed intermediates, which was not possible by the earlier low-resolution optical approaches. MS coupled with radical footprinting deepens the understanding of Aβ folding mechanisms with some residue-level resolution, and this approach can be further extended to other aggregating systems. Although the footprint cannot be assigned to a specific oligomer, the approach is sensitive to the broad range of states and their changes during aggregation. Nevertheless, the approach can further evolve for aggregation studies of proteins alone or with small organic molecules or metal ions that either inhibit or promote aggregation.

From a broader perspective, the two MS-based footprinting approaches, pulsed-HDX and radical labeling, complement each other. Both methods report changes in SASA; HDX is sensitive to backbone hydrogen bonding and solvent accessibility whereas radical footprinting monitors changes in sidechain orientations. Moreover, radical footprinting can shorten the detection timescale to milliseconds. By combining pulsed HDX and fast radical labeling, one can obtain an understanding of protein aggregation intermediates and mechanisms, and these insights should prove invaluable in developing novel therapeutics.

6.4. Probing Ligand Binding, Affinity and Dynamics

$$Protein(P)+Ligand(L)\stackrel{k_{on}}{\to }\underset{k_{off}}{\leftarrow }Prrotein-LigandComplex(PL)$$ (23)

$$K_a=\frac{k_{on}}{k_{off}}=\frac{[PL]}{[P][L]}$$ (24)

$$K_d=\frac{k_{off}}{k_{on}}=\frac{[P][L]}{[PL]}$$ (25)

Numerous proteins interact with ligands to facilitate biological processes for which quantitative understanding is vital.¹¹²¹ Protein-ligand binding affinity measurements are a quantitative measure of these processes; by definition, affinity is the equilibrium constant for the reaction of ligand binding to a protein. It can be expressed as a K_a or K_d (see equations 23 - 25). To date, three general approaches have been used to characterize such interactions. Measurements include circular dichroism,¹¹²¹ fluorescence and fluorescence polarization,^1122-1123 FT-IR,¹¹²⁴ and NMR,^{1032, 1125} that provide read-out of the system composition at given ligand concentrations. These readouts are further utilized, via equations 24 and 25, to derive macroscopic binding affinities. Surface plasmon resonance (SPR) measures the ligand on and off-rates, thus providing binding affinity through a kinetic approach.^{76, 1126} Isothermal titration calorimetry (ITC) measures heat flow during a ligand titration, and binding affinities are obtained employing the van’t Hoff equation.⁷⁴ All these methods are considered to be standard in binding affinity determination, and ITC and SPR are commonly used for their speed and ease of operation.¹¹²⁷ On the other hand, these methods are limited by spatial resolution, often giving none, large sample amount requirement, and sometimes by special sample preparation.

MS-based approaches can advance this field by providing another direct approach that significantly lowers the sample amount. A caveat is whether direct MS can preserve the biological-relevant environment.¹¹²⁷ Under optimal conditions, a MS-based binding assay only requires high-picomole amounts of samples.^1128-1129 Similar with concentration-measurement approaches, MS examines the binding system by measuring its composition (i.e., the concentrations at equilibrium) either directly or indirectly. The direct approach takes advantage of native spray, a gentle form of ESI whereby the protein can be sprayed in a native or near-native state, retaining non-covalent interactions^{151, 1130-1131} and allowing direct measurements of the various equilibrium concentrations;^{1127, 1132} this was first demonstrated by Loo et al.¹¹³³, where they determined the affinities for binding between ribonuclease S-protein and S-peptide. Early applications were in several disciplines including protein-protein,^1134-1135 protein-peptide,¹¹³⁶ protein-oligonucleotide,¹¹³⁷ protein-small molecule,¹¹³⁸ peptide-antibiotic¹¹³⁹ and small molecule-RNA¹¹⁴⁰ complexes.

This direct method has been carefully developed for quantifying protein-glycan interactions, as pioneered by Klassen and coworkers.^{1127, 1141-1143} In 2019, Nguyen and Donald¹¹⁴⁴ measured small molecule-protein interactions through nanoscale ion emitters coupled with native spray, which was not possible previously owing to the high salt concentration needed to stabilize such weak interactions. Although the method is convenient, there is always a question of whether the measured gas-phase concentrations truly represent those in solution; further, the ionization mechanism is complex and not yet well understood.¹¹²⁷ The glycan/protein interactions agree well with those determined by other solution approaches, possibly because the ligands are similar.

Indirect approaches, on the other hand, requires pre-detection labeling. Using this form of labeling in MS was first done by using HDX, where two separate titration methods were demonstrated in the early 2000s, namely, SUPREX^{231, 1145} and PLIMSTEX²³². Both utilize the differences in deuterium uptake between ligand-free and ligand-bound states at a selected exchange time. SUPREX uses denaturant as titrant, and the titration mid-point is extracted by modeling the data; those data are further extrapolated to give folding free energies and binding affinities.^{231, 1146} PLIMSTEX, on the other hand, uses the ligand as titrant, and binding constants are derived by fitting the titration curve.²³² Both approaches have been extended to the peptide level, providing regional specificity and improving spatial resolution.²³³

Owing to the relatively long times used for HDX, neither of these approaches are compatible with systems with high ligand-off rates. They also suffer from post-labeling back exchange. Footprinting by radical species overcomes these drawbacks owing to the irreversible and fast labeling that pertain. In 2019, Liu et al.¹¹²⁹ proposed a method named LITPOMS (ligand titration, fast photochemical oxidation of proteins and mass spectrometry) that for the first time marries fast labeling species to binding affinity determination. As shown in Figure 28, a series of aliquots with different ligand concentrations comprise the titration process. Each aliquot is FPOP footprinted with ^●OH, and further analyzed by MS. LITPOMS can be executed under two different concentrations, where the high-concentration experiment reveals binding stoichiometry and the low-concentration experiment can be modeled to extract binding affinity, just as with PLIMSTEX.

Figure 28.

Schematic illustration LITPOMS workflow. Reprinted with permission from ref. ¹¹²⁹. Copyright 2019 Springer.

A proof-of-concept study demonstrated that LITPOMS successfully yields binding stoichiometry, binding sites, and binding affinity of a tight-binding system, namely calcium-bound calmodulin (holo-CaM) and melittin. Trypsin digestion in combination with LC-MS/MS extends the spatial resolution to peptide and even amino acid residue level. Using this platform, the investigators found three peptide-level binding sites and six critical binding residues, two of which are in loop regions that do not show binding at the peptide level.¹¹²⁹

LITPOMS was quickly adopted, also in 2019, to study a more complex binding system, Ca²⁺-CaM, in which CaM binds to four Ca²⁺ ions in a cooperative fashion.¹¹⁴⁷ As demonstrated in Figure 29, LITPOMS successfully identifies four classes of behavior, all of which are reasonably explained. Other than the simple binding behavior shown in Figure 29b, composite binding curves in Figure 29d, f and h show a combination of binding and conformational changes induced remotely. For example, Ca²⁺ binding at the region represented by peptide 127-148 (Figure 29h) triggers a structural opening at another region represented by peptide 38-74 (Figure 29d), preparing this site to take its own Ca²⁺ and revealing allosteric behavior that was not well understood before. Such composite LITPOMS curves can be further dissected by using residue-level analysis together with several different protease digestions to increase spatial resolution. The investigators could distinguish clearly the binding and conformational changes, thus demonstrating that a detailed picture can be obtained of calmodulin during binding with calcium by using a single approach. Moreover, binding order can be assigned through ranking the onset point of LITPOMS decrease as EF4 > EF-3 > EF-2 > EF-1,¹¹⁴⁷ and was confirmed by a subsequent simulation study.¹¹⁴⁸ Site-specific binding affinities can also be obtained by modeling of eight LITPOMS curves together. This measurement strategy may be particularly effective for signaling proteins that undergo numerous complex conformational changes in executing their functions.

Figure 29.

LITPOMS response of Ca²⁺ - CaM titration at peptide level, where modification fractions were plotted as a function of calcium:calmodulin concentration ratio. Four different classes of behaviors are shown in black (a and c), magenta (b), blue (d, f and h) and olive (e and g). Red solid lines in (b), (d), (e), (f), (g) and (h) are from fitting by a model used in PLIMSTEX. Reprinted with permission from ref ¹¹⁴⁷. Copyright 2019 American Chemical Society.

Binding affinity measurements through radical labeling and MS is promising, as it provides most of the key information about the protein-ligand interactions via a single approach that reveals not only the binding sites and binding affinities but also the binding orders, allosteric behavior, and protein binding dynamics at high spatial resolution.¹¹⁴⁷ To obtain all this information previously required a combination of several sophisticated methods. LITPOMS should have utility in studying complex binding systems, signaling proteins as mentioned above, as it fills the gap that exists between methods offering no spatial resolution to high-resolution x-ray or NMR structures with an approach that offers mid spatial resolution for ligand-free and ligand-bound states.

Challenges remain, however, as there is not yet a radical species that can react with all 20 amino acid residues in a single experiment. For example, ^●OH reacts with all 20 amino acids, but the range of reactivity is too large to allow the least reactive residue to compete with the most reactive in a single experiment, as was discussed earlier in this review. Thus, development of novel and complementary labeling reagents is vital, whereby one footprinter can react with the residues with competitive reactivity and several footprinters can cover all the amino acids in multiple experiments, or even in a single experiment utilizing multiple reagents. In that way, all residues involved in binding can report binding-induced SASA changes. Moreover, modeling of titration data can be challenging, especially for proteins that bind multiple ligands and yield titration curves whose fitting requires construction of complex mathematical models.

6.5. Labeling in vivo and in Animals

As discussed in section 4.5, labeling in vivo has unique advantages, which forecasts a bright future. In vivo labeling by radical species was first demonstrated with nucleotides, where Ottinger and Tullius¹¹⁴⁹ used hydroxyl radicals to footprint the lambda repressor-DNA complex in live E Coli cells. Hydroxyl radicals in this study were generated by ionizing water molecules with a ¹³⁷Cs gamma ray source, which required up to 15 min to produce a sufficient concentration of radicals. To minimize heat damage to the cells during the long irradiation, Woodson and coworkers^1150-1151 utilized synchrotron X-rays to reduce the exposure time to as short as 100 ms and froze their samples (-34 °C to −38 °C) to minimize damage¹¹⁵². Recently, similar experiments were done with a flow system to minimize double labeling and further reduce the exposure time to 10 – 20 ms.¹¹⁵³ Better X-ray dose control also improves the accuracy and reproducibility.

In vivo protein footprinting by radical species, however, was not demonstrated until 2009 when Zhu and Sze⁸⁷³ first performed in vivo protein footprinting with hydroxyl radicals generated by Fenton chemistry. The protein of interest was OmpF porin from E. Coli, an outer membrane channeling protein that facilitates aqueous passive transport in gram negative bacteria. Their approach began with incubating live E. coli with Fe(II)-EDTA and H₂O₂ and then allowing reactions of ^●OH from Fenton chemistry to label the solvent-accessible residues in vivo. As demonstrated in Figure 30c, solvent accessibilities of two loop regions, L1 and L3, do not change significantly upon opening and closing the channel, suggesting that the loops are not responsible for manipulating the channel.⁸⁷³ In other words, entrance of the channel is not blocked by the extracellular loops. Peptides from regions inside the channel (Figure 30d), however, are heavily oxidized when the channel is opened and barely oxidized when the channel is closed. Data support the hypothesis that the three resolved β-sheets are in the channel pore, as demonstrated by Figure 30a and b. Although voltage gating of OmpF was shown previously in vitro, this study for the first time reports the observation of gating phenomena in a native cellular environment.⁸⁷³ Moreover, the novel footprinting also shows the capability of in vivo protein footprinting by radical species in addressing real biological questions.

Figure 30.

In vivo labeling of OmpF porin by hydroxyl radicals. 3D structure of OmpF (PDB ID 1OPF) from (a) side view and (b) top view. Structural components of loop 1, 3 and β-sheet 4, 7, 15 are highlighted in blue, green, red, purple and orange, respectively. Oxidation efficiency for (c) loops and (d) β-sheets under different experimental conditions are plotted. Error bars are standard deviations from duplicates. (c) and (d) are reprinted with permission from ref ⁸⁷³. Copyright 2009 American Society for Biochemistry and Molecular Biology.

In subsequent work, Shcherbakova et al.⁸⁷⁹ shortened the timescale of ^●OH generation. By increasing the concentration of Fe(II)-EDTA to 2 mM, a 40% loss of fluorescence intensity was observed after 2 ms, suggesting that ^●OH can be rapidly generated and that this approach can label other short-lived intermediates.

Another widely used ^●OH generation method, FPOP, was also adopted for in vivo protein labeling. As first demonstrated by Jones and coworkers¹¹⁵⁴, African green monkey kidney cells were mixed with H₂O₂ and submitted to KrF laser irradiation for ^●OH generation and protein labeling. H₂O₂ crosses the membrane through both passive diffusion and via channeling proteins such as aquaporin, allowing sufficient H₂O₂ inside the cell to give a good yield of ^●OH upon laser irradiation. Although H₂O₂ is toxic to live cells, the time required to execute the experiment may be short enough to avoid serious toxicity. Indeed, viability tests prior to labeling suggest that over 70% of cells remain alive under the experimental conditions used for the footprinting. The investigators observed 105 proteins that are oxidized by ^●OH. These different proteins were from different subcellular compartments as shown in Figure 31a, indicating an outstanding dynamic range for protein detection.¹¹⁵⁴

Figure 31.

(a) Pie chart of the subcellular compartment location (African green monkey kidney cell) of oxidized proteins that were identified by LC-MS/MS. Proteins that are present in multiple compartments are represented multiple times. Reprinted with permission from ref ¹¹⁵⁴. Copyright 2015 American Chemical Society. (b) Correlation of residue-level FPOP modifications with SASA in the open (circles and solid line) and tight (diamond and dashed line) states of actin. Reprinted with permission from ref¹¹⁵⁴. Copyright 2015 American Chemical Society. (c) Schematic illustration of an improved flow system for in vivo FPOP labeling. Blue arrows indicate flow and colored lines represent tubing. Reprinted with permission from ref¹¹⁵⁵. Copyright 2016 American Chemical Society. (d) Pie chart of the oxidatively modified proteins within different body systems of c. elegans. Reprinted with permission from ref¹¹⁵⁶. Copyright 2019 American Chemical Society.

When zooming into a specific protein (e.g., actin), FPOP modification fractions can be correlated with SASA of the two different states of actin; namely, open and tight. Results in Figure 31b show a better correlation with the open state (R² = 0.89), suggesting that majority of actin molecules in the native Vero cells are in their open state.¹¹⁵⁴ Later the same group improved the experimental apparatus by introducing a sheath buffer (Figure 31c) to reduce cell aggregation greatly and tube clogging.¹¹⁵⁵ Their design also ensures that radiation exposure of each cell in the flow system remains comparable. As a result, the number of identified oxidized proteins increased by 13-fold.¹¹⁵⁵

In 2019, Jones and coworkers¹¹⁵⁶ introduced small, live worms, C. elegans into the flow system, and they were able to identify oxidatively labeled proteins from different body systems as shown in Figure 31d. A closer look at the myosin chaperone protein UNC-45 suggests that the modification fractions obtained in this in vivo study correlate well with the SASA calculated from the crystal structure.

Owing to incomplete understanding of the absolute extent of modification, most MS-based protein footprinting approaches have been executed in a differential way, where the extent of labeling for a peptide/residue in different states (e.g., ligand-bound versus ligand-unbound) can be compared in valid way. Although promising, work by Jones et al. fails to reveal such differences up until now.

In one of the pioneering demonstration of in vivo radical footprinting through a differential approach, Zhu and Sze¹¹⁵⁷ footprinted the structural changes of epidermal growth factor receptor upon binding with epidermal growth factor (EGF) in live E. Coli cells. Hydroxyl radicals were generated by an FPOP approach and the oxidation extents of EGF-free and EGF-bound states were compared. The results are consistent with crystal structures of these two states. Their study marries conventional bottom-up structural proteomics with in vivo free radical labeling, probing differential structural changes in a real cellular environment. Moreover, their workflow including enrichment by immunoprecipitation and in-gel digestion is applicable to other studies.

Another important class of in vivo protein footprinting by radical species is proximity labeling, pioneered by Ting and coworkers starting in 2013.^1158-1159 Unlike all other labeling approaches introduced prior to this study, proximity labeling is usually facilitated by enzymes. The first enzyme used to facilitate the labeling is ascorbate peroxidase (APEX), and the method was appropriately termed APEX labeling. In brief, APEX is genetically tagged onto the protein of interest, whose location in the cell is generally known. Upon adding H₂O₂ and biotin-phenol into the cell, APEX catalyzes the transformation of biotin-phenol into biotin-phenoxyl radicals, which then footprint nearby proteins (Figure 32). The biotin tag facilitates the post-labeling protein enrichment. As a result, the investigators were able to identify 495 proteins within the human mitochondrial matrix, 31 of which were not previously linked to mitochondria. APEX labeling is highly specific, as the radicals are short-lived (< 1 ms)¹¹⁶⁰ and the labeling radius is within 20 nm.¹¹⁶¹ APEX labeling was optimized in 2015 by introducing different enzymes¹¹⁶² and labeling reagents,^1163-1164 making APEX labeling widely applicable to various cellular systems. On the other hand, APEX labeling is mainly used for proteomics purposes (i.e., determining primary structure). Its potential in protein footprinting for HOS elucidation remains to be explored. The idea of proximity labeling inspired another investigation of protein-carbohydrate interactions,⁸⁷² which is covered in section 6.8.

Figure 32.

Schematic illustration of APEX labeling workflow. Reprinted with permission from ref. ¹¹⁵⁸. Copyright 2013 American Association for the Advancement of Science.

All in all, in cell and in vivo labeling appears to have unprecedented advantages in structural proteomics to complement in vitro approaches that usually do not accurately reproduce the cellular environment. Another advantage for these approaches is their application in studying membrane proteins and their complexes, which have poor solubility and instability in vitro. The short timescale of ^●OH labeling enables the detection of many transient structural intermediates, especially with the FPOP approach. Modern MS instruments also allow the detection of thousands of unique proteins in a single run. All these advantages forecast a strong future for in vivo labeling.

On the other hand, current demonstrations of in vivo labeling fail to reveal as much structural information as conventional in vitro bottom-up approaches. As there are thousands of proteins in a single cell, most current studies must still emphasize protein identification. It is still challenging to track structural changes of a single protein during a biological event. This requires better protein separation and enrichment, further development of MS instrumentation, and improved data processing software. Previous efforts have demonstrated the correlation between absolute modification fractions and SASA. By using computer-based protein structure modeling, footprinting can produce constraints that can be used to predict protein structure. In this way, we may be able to bypass the differential approach without sacrificing structural resolution. Several studies show promising results on this topic, and they will be covered in the next section.

6.6. Footprinting in Supporting Computer-based Structural Predictions

So far, all applications described in this section have been based on differential experiments, where the modification fractions for specific peptides and residues were tracked as a function of protein states. Different protein states are achieved, for example, as ligand-bound vs. unbound in epitope mapping, as native vs. mutant, after different folding times, at different aggregation states, at various ligand concentrations in affinity determination, and after conformational changes (e.g., opening and closing a channel). Although straightforward, the differential approach has some limitations. Although various systems can be characterized, comparisons require relatively pure systems where different protein states can be clearly differentiated.

Ideally, HOS would be determined directly by footprinting a single protein state. Radical labeling and even other MS-based approaches, however, are not capable of providing enough restraints to construct a high-resolution protein structure as can NMR and X-ray crystallography. Computer-based protein HOS prediction is powerful and has great potential, yet the incomplete understanding of protein folding does not allow accurate prediction of protein HOS.^{79, 1058} Data from NMR,¹¹⁶⁵ small angle X-ray scattering,¹¹⁶⁶ cryo-EM^1167-1168 and site-directed spin labeling EPR¹¹⁶⁹ overcome such limitations because they are based on structural restrictions or restraints.

An MS-based approach that is particularly amenable to modeling is chemical crosslinking.^{761, 1170} Most of HOS elucidated from this workflow, however, is of protein complexes, because only with large and complex systems is it possible so far to obtain sufficient crosslinks for modeling. Further, the principal HOS outcome is a protein-protein interface and not the full structure although more detailed structure determinations are becoming possible. It may be that footprinting combined with crosslinking will lead to more reliable models than from either method itself.⁷⁶⁸

Structural prediction based on radical footprinting data was not possible until it can be properly quantized and reasonably correlated to the absolute SASA. Gerega and Downard¹¹⁷¹ in 2006 developed a docking algorithm named PROXIMO that adopts directly modification fractions for each resolvable residue as restraints. They were able to dock properly the calmodulin-melittin and Ribonuclease S-protein-peptide complexes. Although straightforward, their approach failed to consider the different reactivities between ^●OH and different amino acid residues, and that failure may produce bias in the footprinting data. This pioneering work expands the possibilities of radical footprinting and motivates more precise data quantification.

To date, three groups have demonstrated the efficacy of modeling footprinting results. In 2015, Huang and Chance¹¹⁷² first proposed a protection factor (PF) that successfully correlates single-state footprinting data with absolute SASA. The observed reaction rate from a dose-response curve from synchrotron-based ^●OH footprinting (modification fraction versus X-ray irradiation time) is a function of both intrinsic reactivity of ^●OH and the nature of the amino acid residue and its solvent accessibility.¹⁶⁰ Teasing out a structural contribution can be done by normalizing the observed reaction rate constant with respect to the residue-specific intrinsic reactivity. As a result, solvent accessibilities (calculated from X-ray structures) for three model systems and local structural contacts exhibit a quantitative agreement with calculated PFs, motivating future developments.

Recently, the investigators adopted this idea for characterizing the human estrogen receptor alpha (hERα), which contains a DNA-binding domain and a ligand-binding domain.¹¹⁷³ Structures of these two domains are known individually but not in the complex. Using results from ^●OH radical footprinting, PF calculations of resolved residues, and data from SAXS, the investigators generated sufficient structural restraints to guide successfully molecular docking. The outcome is a successful structure determination of an asymmetric L-shaped structure of the multidomain hERα, revealing key mechanisms that facilitate allosteric function.

In another elegant experiment, Xie and Sharp¹¹⁷⁴ introduced a method for assessment of sidechain absolute SASA values by using an FPOP-based ^●OH footprinting platform. By incorporating adenine as ^●OH dosimeter, they obtained reaction rates for each residue through multi-point FPOP experiments (different ^●OH doses). These data were further normalized with respect to intrinsic reaction rates of ^●OH and free amino acids. Normalized protection factors obtained through these efforts were compared with fractional SASA, and a good linear regression was obtained.¹¹⁷⁴ Further, two SASA prediction models were constructed for the soluble proteins myoglobin and lysozyme, and the predicted SASA’s were compared with a calculated SASA from crystal structures by using a protein-unfolding MD simulation.¹¹⁷⁴ Although the incorporation of radical protein footprinting data in some structural modeling is successful, use of the footprinting data directly in protein structure prediction is yet to be established.

In 2018, Lindert and coworkers¹¹⁷⁵ developed a new Rosetta score term named hrf_ms_labeling that utilizes residue-level PFs from ^●OH footprinting data as constrains to predict protein structures. PFs were calculated based on the formula proposed by Huang and Chance¹¹⁷². With Rosetta, they generated 20,000 structural models with an ab initio method for four different proteins (i.e., calmodulin, cytochrome C, myoglobin and lysozyme). The Rosetta score for each model was plotted against their RMSD with respect to their native crystal structures as shown in Figure 33a. The 20,000 models for each of the proteins covers a broad range of RMSDs. Aligning the top-scoring model with the native structure allows better visualization of the differences, as there are even topology mismatches for calmodulin and lysozyme (Figure 33b). Although there are models that have near-atomic resolution (RMSD ≤ 2Å), they were not recognized as the top scoring models by Rosetta. When rescoring the same 20,000 models with restraints from experimentally determined PFs, similar distributions were identified (Figure 33c). With the help of new scoring function hrf_ms_labeling, the RMSDs for the top-scoring models improve significantly for all four proteins. A visualization of the outcome (Figure 33d) shows that all models identify the correct protein topology.¹¹⁷⁵ Structure models for cytochrome C and myoglobin come with near-atomic resolution, indicating the efficacy and applicability of the newly developed scoring function. This work demonstrates for the first time that incorporation of ^●OH protein footprinting data can greatly enhance model quality in protein structural prediction, elevating the use of MS-based footprinting from a qualitative description to a quantitative evaluation. If these efforts continue to be successful, protein footprinting may no longer limited to differential experiments, and the results for a single protein state may be transformed into a high-resolution structure. The script of this newly developed scoring function and the corresponding instructions are freely available with Rosetta.

Figure 33.

(a) Rosetta score versus RMSD (with respect to the native structure) plots for 20000 models generated from Rosetta ab initio for each of four proteins. Top scoring model is highlighted by a star in each plot. (b) Top scoring models from the Rosetta score versus RMSD distributions in (a) (color) superimposed on the respective native model (gray). PDB ID for these native models are depicted at the top of the figure. (c) Rosetta score + hrf_ms_labeling versus RMSD (with respect to the native structure) plots for each of the four proteins after rescoring with the new score term. The top-scoring model is highlighted by a star in each plot. (d) Top scoring models from the Rosetta score + hrf_ms_labeling rescoring distributions in (c) (color) superimposed on the respective native model (gray). Reprinted with permission from ref ¹¹⁷⁵. Copyright 2018 American Chemical Society.

In another example, the same group¹¹⁷⁶ developed novel Rosetta scoring function that utilizes SID MS data and demonstrated that SID data significantly enhance the confidence of structural predictions of protein complexes.

Although still in its early stages and requiring further work, structural prediction utilizing footprinting data shows significant promise. It will be necessary to generalize these workflows and make them compatible with other footprinting approaches. Further, it will be even more significant to develop a new structural prediction algorithm that is guided by footprinting data rather than by rescoring the existing models, as hinted in a subsequent study by Lindert and coworkers⁸³⁶ in 2019.

6.7. Revealing Pathways in Biological Systems

ROS are those that form upon incomplete reduction of oxygen; they include superoxide anion (O₂^−•), H₂O₂, singlet oxygen (¹O₂) and ^●OH.^{844, 1177} These species play essential roles in regulating various functions in biological systems.⁸⁴⁴ On the other hand, ROS are highly reactive, and are likely to induce oxidative damage at unfavorable locations. An interesting 2017 study takes a lead from free-radical damage and goes on to reveal successfully the water channels in photosystem II (PSII).¹¹⁷⁸

PSII is a membrane-bound oxidoreductase that catalyzes the conversion of water oxidation to molecular oxygen and simultaneously the reduction of plastoquinone. Sunlight is converted into chemical energy, and electron transfer in photosynthesis is thus initiated. A high resolution structure of PSII identified the catalytic center as Mn₄CaO₅, where substrate water is oxidized to O₂.¹¹⁷⁹ The pathway for the substrate water molecule into the catalytic center, however, is poorly understood. As ^●OH is the side-product of such conversion, and the size and hydrophilicity of ^●OH are similar to those of H₂O⁸⁴⁵, as discussed earlier, it is possible to map the H₂O channel by tracking the movement of ^●OH and its oxidative damage. In other words, the ^●OH modified residues leave a trail or “footprint” of natural oxidative damage that line the wall of H₂O channel, revealing the structural aspect of the PSII. Using an approach that includes analysis used for protein footprinting, the investigators identified three distinct water channels, two of which aligned well those identified in previous studies. This work resembles that of Ting,^1158-1159 described in Section 6.5, except the oxidation occurs naturally.

This simple experimental design utilizes the naturally occurring ^●OH and related species as labeling reagents to footprint PSII in vivo. The structural information obtained in this work increases the mechanistic understanding of the PSII system. Moreover, the outcome motivates further development of radical protein footprinting and ROS localization to understand the chemistry of ^●OH and to augment, even supplant, other low-resolution approaches for detection of ^●OH.¹¹⁷⁷ MS-based in situ radical footprinting offers an elegant way to localize the ^●OH by tracking its damage trails on the surrounding amino acid side chains. The utilization of naturally occurring radical species in protein footprinting may also provide insights into “natural” oxidative modifications and whether they introduce perturbations to native biological systems.

Inspired by this idea, a recent study focused on CO₃^−•, a naturally occurring radical, and used the FPOP platform to characterize the reactions of CO₃^−• that may occur in vivo.⁹⁵⁷ Although a reactive species, the CO₃^−• radical anion does not qualify as a good footprinter owing to ambiguity in its generation from ^●OH or other strong oxidizing radicals, underscoring that novel mythologies and other radical species are necessary in the footprinting field.

6.8. Other Applications

In addition to the applications covered above, there are other promising studies showing the capability of free radical protein footprinting to address other structural concerns.

The first is the identification of bound waters in the proteins. Water plays an essential role in protein folding, structure and stability.^1180-1181 X-ray diffraction¹¹⁸² and NMR¹¹⁸³ can detect highly conserved water molecules that are located bound with polar or charged residues of proteins. Few experimental approaches, however, are able to differentiate waters that are bound to the surface and the interior of proteins. Synchrotron-based ^●OH protein footprinting ionizes water to give ^●OH.^{170, 1184} Using a careful experimental design, Gupta and Chance¹¹⁸⁴ footprinted cytochrome C under two different temperatures of 25 °C and −35 °C.

Freezing the sample during synchrotron irradiation limits the diffusion of ^●OH. A comparison of modification fractions of a certain peptide or residue under two different temperatures enables the differentiation of types of water that locate in these regions. A significant decrease (13 to 200-fold) suggest that the modification is from bulk water that gives ^●OH and labels the protein through diffusion. Sites that suffer a moderate decrease (3 to 10-fold) are located on the protein surface, and these sites have bound water molecules. Residues that have minimum changes (decreases less than 2-fold) upon footprinting at two different temperatures must have internal, closely located water to give ^●OH radicals that react with adjoining residues without significant diffusional motion. Results from ^●OH footprinting are consistent with those seen in an X-ray structure.¹¹⁸⁴ Further experiments utilized a time-resolved ¹⁸O/¹⁶O exchange apparatus and probed the water dynamics on the submillisecond timescale,¹¹⁸⁴ offering a unique view of protein-water interacting dynamics that was only possible with NMR before.¹¹⁸³ Synchrotron-based ^●OH footprinting has unique advantages in studying protein-water interactions, as water serves as the precursor of ^●OH, and it is capable of reacting with proteins that are in different sites, even within the lipid bilayer of membrane proteins.^1185-1186

The second application focuses on protein-carbohydrate interactions. Cell membranes are covered with heterogenous glycans that create an interactive environment and govern many cellular functions.¹¹⁸⁷ Many cellular functions are based on protein-glycan interactions and glycan-mediated protein-protein binding, yet characterization of the protein-glycan interactions is mainly limited to glycan arrays.¹¹⁸⁸ Very recently, Li and Lebrilla⁸⁷² reported a novel workflow termed protein oxidation of sialic acid environment (POSE) that footprints those proteins in close proximity to sialic acids, an idea that shares features with APEX labeling introduced earlier. As sialic acid usually terminates the glycan chains in the glycoproteins, the first step is to functionalize the sialic acid with an azido group (Figure 34a) and to introduce the active iron species through click chemistry, the glycan chain now serving as a probe (Figure 34b). Upon adding H₂O₂, ^●OH is generated by Fenton chemistry, thus oxidatively labeling the surrounding proteins (Figure 34c). POSE allows in situ labeling of proteins that are closely located to glycans, providing an effective and efficient way to screen possible glycan-binding targets.

Figure 34.

Schematic illustration of POSE workflow. (a) The incorporation of ManNAz into cell surface glycoproteins as SiaNAz. (b) The bio-orthogonal reaction of the DBCO-FeBABE probe with the SiaNAc group conducted via click-chemistry. (c) The oxidation of proteins in the sialic acid environment by ^●OH generated by Fenton Chemistry. Reprinted with permission from ref ⁸⁷². Copyright 2019 Royal Society of Chemistry.

Finally, there are other promising applications of fast labeling approaches including footprinting membrane proteins,^{945, 1189-1191} following the early onset of ROS oxidative damage,¹¹⁹² probing protein hidden conformations,¹¹⁹³ mapping protein-DNA interactions¹¹⁹⁴ and many more. These efforts all provide valuable information in their respective fields, but they will not be covered in detail in this review.

7. Conclusions and Perspective

MS-based protein HOS analysis has increased the throughput and decreased the sample amount requirements of protein HOS determination while becoming an effective complement to traditional biophysical characterization methods. More importantly, different protein footprinting approaches allow MS-based approaches to view protein HOS from different perspectives. For example, HDX usually measures the kinetics of exchange and infers HOS from the “labeling” of protein backbone amide bonds. Targeted or specific labeling reagents exploit various organic reactions to report on side-chain solvent accessibilities. Fast labeling reagents react with the residue side chains on the time scale of ns to ms, affording a “snapshot” of protein structure and dynamics. Taken together, MS-based protein labeling approaches “paint” the protein solvent accessible surfaces over a time frame from ns to days to afford a comprehensive understanding of the protein of interest. Moreover, modern proteomics digestion (bottom-up) workflows, top-down fragmentation techniques together with ultra-sensitive MS instruments yield not only HOS information at mid-to-high (single amino acid residue level) spatial resolution, but also dynamics of the protein under various conditions that are hardly accessible by a single technique. This is important because a footprint should be done with high coverage, unlike primary structure identification in traditional proteomics. Over the past decades, MS-based structural proteomics approaches have grown extensively and now can answer important and challenging biological questions. None of these developments were possible without technical innovations in the field of MS.

7.1. Hydrogen Deuterium Exchange

During the past decade, HDX has grown to be a mature tool for MS-based protein HOS characterization in both academia and in the biopharmaceutical/biotechnology industry. Despite its growth and success, HDX methods still face multiple challenges. One challenge is the tedious, complex measurement, which is being met with robotic automation of the HDX measurement, making bottom-up HDX more readily applicable to most simple protein systems.¹¹⁹⁵ Titration-based HDX workflows including SUPREX and PLIMSTEX provide site-specific binding affinities in addition to locating the binding sites.^231-232 Incorporation of ion-mobility offers an orthogonal dimension in the separation of constituent peptides prior to MS measurements, facilitating the deconvolution of overlapping isotopic patterns for co-eluting peptides.¹¹⁹⁶ Novel fragmentation methods including ETD and ECD together with multi-enzyme digestions allow investigators to reach out for residue-level HDX information.²²¹ Exchange in theta capillaries²²⁷ and in the gas phase^228-230 may allow HDX to be measured on the μs time scale. Innovations in MS instruments permit HDX to be conducted in a top-down fashion. Although promising, these novel advances still require additional demonstration to spur even broader adoption.

Most current applications of HDX focus on soluble protein systems. As HDX is reversible, its measurement by MS is ex situ, setting experimental constraints that minimize back exchange. These restraints limit application of HDX in complex protein systems; for example, glycoproteins that require post-labeling deglycosylation,²³⁷ membrane proteins need post-labeling lipid or detergent removal,²³⁶ and structurally rigid proteins that are challenging to denature and digest. Although pioneering studies demonstrate feasibility, routine and robust workflows still need to be developed and demonstrated before generalization and widespread acceptance will occur.

Another important aspect of HDX-based protein HOS analysis is the proper utilization of absolute deuterium uptake for structural modeling. Although most current HDX measurements are conducted in a differential manner to locate binding sites or determine the effects of mutation, absolute uptake reports on backbone dynamics and structural features of the protein. Fast exchange is typical of protein loops or intrinsically disordered regions whereas very slow exchange is characteristic of either structurally rigid or deeply buried regions. Although the concept of absolute protection estimation via HDX measurement is well appreciated, it is poorly demonstrated when combined with computer modeling to predict protein HOS.^1197-1199 The quantitative assignment of protection not only requires high-quality data acquisitions but also reliable data processing. As it is challenging, the outcoming is also rewarding. Incorporating protection extent from HDX measurements into protein structure prediction will greatly elevate the contributions of HDX to structure determination, as is seen in recent radical labeling demonstrations.¹¹⁷⁵

7.2. Irreversible Labeling

As compared with HDX, protein HOS analysis by irreversible labeling and MS has a shorter history even though other opportunities of specific amino acid labeling originated 60 y ago. Many applications both for specific amino acid labeling and free radical footprinting, however, have been developing during the past decade. One advantage is the irreversibility of the labeling, making it compatible with sophisticated post-labeling sample treatment such as long gradients in HPLC or extensive signal averaging as in top-down studies. These features enable approaches that answer broad biological questions about many types of proteins.

Despite its broad compatibility, a missing piece of the picture is membrane protein footprinting. Early studies incorporating nanodiscs¹²⁰⁰ and other membrane mimetic systems¹²⁰¹ show the feasibility of footprinting membrane proteins. Although these demonstration studies used both targeted labeling reagents^{816, 820-821} and fast radical species^{873, 945, 1190}, the primary focus was the soluble domain. An effective footprinting approach for the transmembrane domains of membrane protein remains to be established although there are hints that productive footprinters exist. Unlike HDX, whose labeling reagent, D₂O, cannot be extensively partitioned into bio-membranes, the targeted labeling reagents and the radical precursors can be chemically tailored to favor the membrane where they can footprint the transmembrane regions and yield information on those regions that are buried and those that adjoin the membrane.¹²⁰²

Another fundamental aspect of targeted and radical protein labeling that remains to be addressed is the effect of microenvironment; that is, the effect of adjacent residues on the reactivities between targeted residue and the labeling reagents. Such effects are mainly explained as fluctuations of local reagent concentrations, local hydrogen-bonding schemes, or the electronic or steric environments determined by the nearby functional groups. The effect of adjacent residues on the exchange rates in HDX was thoroughly studied and is now well established.²⁰⁶ For footprinting by targeted labeling reagents, a few studies hint that adjacent residues play a role.^{497, 541, 1203-1207} In the case of radical footprinting reagents, there are two recent reports that discuss local concentration fluctuations of labeling reagents.^{896, 1208}

Understanding the effect of the microenvironment on labeling efficiencies is crucial in determining the intrinsic structural factors that are involved in interpreting footprinting data, and these effects should not be overlooked. Only from a deep understanding can we take full advantage of footprinting results and their incorporation into protein structural modeling. The needed studies for irreversible labeling, however, are more challenging than those for HDX, whose labeling reagent is high-concentration solvent, D₂O, itself. One can reasonably assume that the effect of microenvironment will be reagent-dependent, further complicating efforts to unravel the effects but also offering an opportunity to tailor reagents.

From a practical perspective, several issues need to be addressed in upcoming research in this field. First is the implementation of novel labeling reagents. As mentioned earlier, a good footprinter needs to provide a distinct mass tag that can be properly differentiated from naturally occurring PTMs. Because targeted labeling reagents generally introduce mass tags that do not occur naturally, protein footprinting by radical species is mostly carried out with hydroxyl radicals, which label protein by oxidation, producing a +16 Da mass shift inter alia. As oxidation often occurs not only naturally but also during post labeling sample handling and even during LC separations, it is sometimes ambiguous to distinguish the oxidations from footprinting and those from other processes. This small limitation can be overcome by development of novel labeling reagents that not only retain the advantages of widely adopted hydroxyl radicals but also “paint” the protein surfaces, affording another distinctive mass tag.

Another possible way to address this issue is to develop cleavable or isotopically encoded radical precursors as labeling reagents, an idea that has been extensively utilized in protein chemical cross-linking and in protein footprinting. Using this approach, investigators employ targeted labeling reagents that can readily accommodate isotopic encoding to shift the mass of the mass tag. These efforts will surely shorten the analysis time and increase the confidence of identification and quantification in radical-based protein footprinting. This is difficult with hydroxyl radical footprinting because at best the reagent mass can only be shifted by 2 Da, which is insufficient to resolve for large peptides and proteins.

The second is the development of novel labeling reagents (examples were discussed in section 5) especially for residues that are activated with difficulty and, therefore, react sparingly. Although it is feasible to activate selective C-H bonds from an organic chemistry perspective, to achieve it under physiological conditions is considerably more challenging because the modification must be done rapidly. Labeling of Ala, Gly and Pro whose side chains are functionalized with difficulty is likely only with free radicals, but even then, the reactivity will be low considering that proteins contain many other reactive residues. Although specific labeling of Gly residues at the protein N-terminus by targeted labeling reagents was demonstrated recently,¹²⁰⁹ general modification of these residues is difficult and been realized only in one study to our knowledge.¹²¹⁰ Moreover, bridging the fields of protein chemical modification and protein footprinting offers the opportunity to “reinvent the old”. Protein chemical modifications have been extensively developed for many years to satisfy several biological and analytical needs. Some of the reactions and reagents that are adopted in those studies can be used to footprint proteins as well, as has been done in the past and continues today (consider DEPC, GEE, IAM, NHB, TNM footprinting, for example).

Last is a technical comment to emphasize the need to conduct thorough evaluations when applying a labeling reagent to study a specific protein system. Technical issues including the compatibility of the labeling reagent and the protein, the optimum reagent dose and labeling time to minimize double-labeling, to achieve the proper normalization of labeling extents in differential experiments, and to invoke the statistical justifications of differences in labeling extents, and many more. These points should be considered as standard operating procedures, yet few studies to date report evaluations prior to presenting the results, perhaps not surprising given that the field is young. Many of these issues can be addressed by, for example, optical evaluation of protein integrity (by ion mobility MS, CD, infrared spectroscopy or other low-resolution biophysical approaches) and the use of a reporter small molecule or peptide dosimeters to enable normalization between different states of the protein. These efforts and results should be reported in future studies to convince the user community that the methods are robust.

7.3. Broad Perspective

From a broad view, each MS-based labeling approach will contribute to the understanding of protein HOS from a different perspective, motivating the thoughtful and creative choice of different reagents. As mentioned above, the methods of HDX, targeted labeling reagents, and radical species inform on protein HOS and their changes on different time scales. Their targeted amino acid residues on the protein are also different. Applying combinations of different approaches can provide a deeper understanding of the protein and a higher resolution structure. Recent demonstrations have shown the power of such integrated strategies.^{738, 768, 838-839, 1048, 1211-1215} Its broad application, however, remains to be explored.

Platform integration not only expands the protein footprinting approaches but also requires clever application of the various sample handling and MS techniques. For example, one can couple native MS and protein footprinting in a preliminary analysis to assist in the design of a study of protein ligand interactions, to determine mixing ratios that push the binding equilibriums to the product complex, and to explore the binding regions. Ion mobility MS coupled with protein footprinting will illuminate protein conformational changes during and after labeling, providing a view of protein conformational changes induced by chemical labeling. Multiple ion dissociation techniques coupled with different enzymatic digestions will deliver better sequence coverage, even to the residue level, which is one goal in MS-based protein HOS analysis.

Improved reversed-phase chromatography and adoption of other separation strategies (e.g., normal phase, capillary electrophoresis) will allow separation of isomeric peptides formed in labeling (e.g., chemically modified peptides that bear the same modification on multiple residues) and their quantification from extracted ion chromatograms and other approaches. Improved digestion strategies (e.g., mixed and immobilized enzymes) will also improve spatial resolution. Better separation will also minimize ion suppression during ionization, separate structural isomers that have the same m/z, and ultimately improve the precision of quantification and the spatial resolution of the protein HOS analysis. Enhanced separation capability, possibly even with 2D approaches, is particularly important for footprinting complex mixtures of proteins or large proteins that digest to provide a large number of peptides. Modern protein and peptide separation approaches, including multi-dimensional LC,^1216-1217 size exclusion chromatography,^1218-1219 ion-exchange chromatography,¹²²⁰ have all been coupled with MS analysis. Although most of the current demonstrations are for primary-structure proteomics and native MS, they can surly be adopted for protein HOS analysis soon.

Upon obtaining the MS data, database searching and spectra identifications by software is also critical. Over the past decade, we witnessed a burst of MS software for MS/MS identification, de novo sequencing, HDX, irreversible labeling, chemical cross-linking, native MS, ion mobility MS, for proteomics, glycomics, lipidomics, and metabolomics. Although these developments largely contribute to data processing and visualization, they will continue to be important components in the “engine” of MS-based structural proteomics and ultimately lead to more automated data processing.

Lastly, to increase the spatial resolution of MS-based protein HOS studies, computer modeling is also critical. As discussed in section 6.6, pioneering efforts demonstrate early efforts pointing to a promising future of this area. Incorporating footprinting data into structural predictions in a straightforward way breaks the resolution limit of MS approaches, making MS-based protein HOS analysis a biophysical method that delivers high resolution results as does X-ray crystallography, cryo EM, and NMR or that, at least, complements them, even for protein complexes¹¹⁷⁶. A simple example is the use of footprinting to recover the information loss for flexible protein regions in an X-ray crystal structure. Thus, footprinting data can increase the confidence of protein structural predictions both on its own and in combination with other approaches.

To achieve these goals, deeper understanding of footprinting fundamentals is crucial, because only from a solid understanding of fundamentals can we to evaluate the footprinting data properly, especially for footprinting a single protein state rather than employing a differential experiment. Development of user-friendly software that can take footprinting data as input for structural modeling is also needed.

7.4. Concluding Remarks

The combination of advanced MS instrumentation, novel protein labeling workflows, wide ranging labeling reagents, careful experimental design, and precise data processing will continue to advance MS footprinting as a tool well beyond protein molecular weight and primary structure determination. MS as a robust and informative protein structural characterization tool has been established during the past 30 years, and we predict that MS-based approaches will continue to contribute significantly to the field of structural proteomics and the broader field of structural biology.

Abbreviations

¹O₂

singlet oxygen

3D

three dimensional

5-IAF

5-idoacetamidofluorescein

6-Cl-IMP

6-Chloropurine riboside 5'-monophosphate

Aβ

amyloid beta

ADH

adipic acid dihydrazide

AFM

atomic force microscopy

Ala

alanine

aMB

apo-myoglobin

ANB-NOS

N-5-azido-2-nitrobenzoyloxysuccinimide

APEX

ascorbate peroxidase

Arg

arginine

Asn

asparagine

Asp

aspartic acid

αSyn

alpha-synuclein

β2m

β-2-microglobulin

BAMG

bis(succinimidyl)-3-azidomethylglutarate

BDC

N-benzyl-N'-3-dimethylaminopropylcarbodiimide

BDP-NHP

N-hydroxyphthalamide ester of biotin aspartate proline

BHD

benzhydrazide

BME

b-mercaptoethanol

BS3

Bis(Sulfosuccinimidyl)suberate

C. elegans

Caenorhabditis elegans

CaM

calmodulin

CASP

critical assesment of protein structure presiction

CBDPS

cyanurbiotindimercaptopropionyl succinimide

CD

circular dichroism

CDR

complementarity determining region

chloramine T

N-chloro-4-methyl-benzenesulfonamide

CID

collision induced dissociation

CL-MS

covalent labeling MS

Cryo-EM

cryogenic electron microscopy

Cys

cysteine

DEPC

diethylpyrocarbonate

di-UB

di-ubiquitin

DMP

dimethyl pimelimidate

DMS

dimethyl suberimidate

DMTMM

4-(4,6-dimethoxy-1,3,5-triazin-2-yl)-4-methyl-morpholinium chloride

DSBU

disuccinimidyl dibutyric urea

DSG

disuccinimidyl glutarate

DSS

disuccinimidyl suberate

DSSO

disuccinimidyl sulfoxide

DTBP

dimethyl 3,3′-dithiobispropionimidate

DTNB

5,5'-dithiobis-(2-nitrobenzoic acid)

DTSSP

3,3'-dithiobis(sulfosuccinimidylpropionate)

DTT

Dithiothreitol

E. Coli

Escherichia coli

eaq^-

hydrated electrons

ECD

electron capture dissociation

EDC

1-ethyl-3-(3-(dimethylamino)propyl)carbodiimide

EDD

electron detachment dissociation

EDTA

ethylenediaminetetraacetic acid

EGF

epidermal growth factor

EGFR

human epidermal growth factor receptor

eIF3

yeast initiation factor 3

ELISA

enzyme linked immunosorbent assay

EPR

electron paramagnetic resonance

ESI

electrospray ionization

ETD

electron transfer dissociation

EThcD

electron transfer higher energy dissociation

FAB

fast atom bombardment

FDR

false discoverer rate

Fe-BABE

Fe-(S)-1-(p-bromoacetimidobenzyl)-EDTA

FMO

Fenna-Matthews-Olson

FPOP

fast photochemical oxidation of proteins

FRET

Förster Resonance Energy Transfer

FTICR

Fourier-transform ion cyclotron resonance

FT-IR

fourier transform infrared spectroscopy

GA

glycinamide

GEE

glycine ethyl ester

Glu

glutamic acid

Glu

glutamine

GLUTs

human glucose transporters

Gly

glycine

g-ray

gamma ray

H₂O*

activated water

HBx

hepatitis B virus X protein

HDX

hydrogen deuterium exchange

hERα

human estrogen receptor alpha

His

histidine

hMB

holo-myoglobin

HNB

2-Hydroxy-5-nitrobenzyl-bromide

HNSB

dimethyl(2-hydroxy-5-nitrobenzyl)sulfonium bromide

holo-CaM

calcium-bound calmodulin

HOS

higher order structure

HRP

horseradish peroxidase

hVKOR

human vitamin K epoxide reductase

IAM

iodoacetamide

ICAT

isotope-coded affinity tag

IgG1

human Immunoglobulin G subclass 1 antibody

IL-6R

human interleukin-6 receptor α-chain

Ile

isoleucine

IMPDH

human type II inosine 5’-monophosphate dehydrogenase

iodogen

1,3,4,6-tetra-chloro-3a,6a-diphenyl-glycouril

IRMPD

infrared multiphoton dissociation

ITC

isothermal titration calorimetry

kethoxal

3-ethoxy-1,1-dihydroxy-2-butanone

LC

liquid chromatography

Leu

leucine

LITPOMS

ligand titration, fast photochemical oxidation of proteins and mass spectrometry

Lys

lysine

m/z

mass to charge ratio

mAb

Monoclonal antibodies

MALDI

matrix-assisted laser desorption ionization

MCO

metal-catalyzed oxidation

Melarsen oxide

p-(4,6-diamino-1,3,5-triazin-2-yl)aminophenylarsonous acid

MES

2-(N-Morpholino) ethanesulfonic acid

Met

methionine

MMTS

methyl methanethiosulfonate

MS

mass spectrometry

MS/MS

tandem MS

MURR1

mouse U2af1-rs1 region1

N-AcO-AAF

N-acetoxy-N-acetyl-2-aminofluorene

NAI

N-acetylimidazole

NBS

N-bromosuccinimide

NCS

N-chlorosuccinimide

NEM

N-ethylmaleimide

NHS-ester

N-Hydroxysuccinimide ester

NMR

nuclear magnetic resonance

NTCB

2-nitro-5-thiocyanobenzoic acid

PAL

photoaffinity labeling

p-bromophenacyl bromide

2-bromo-1-(4-bromophenyl)ethenone

PD

plasma desorption

PDB

protein data bank

PDH

pimelic acid dihydrazide

PF

protection factor

Phe

phenylalanine

PIR

protein interaction reporter

PLIMB

plasma induced modification of biomolecules

PLIMSTEX

protein–ligand interactions by mass spectrometry, titration, and H/D exchange

PLP

pyridoxal-5’-phosphate

POSE

protein oxidation of sialic acid environment

PPIs

protein-protein interactions

Pro

proline

PrP

prion protein

PSII

photosystem II

PTM

post translational modification

ReACT

real-time analysis for cross-linked peptides technology

RMSD

root mean square deviation

ROS

reactive oxygen species

SASA

solvent accessible surface area

Ser

serine

SID

surface induced dissociation

SOD

Superoxide dimustase

SPR

surface plasmon resonance

sulfo-SDA

sulfosuccinimidyl 4,4'-azipentanoate

SUPREX

stability of unpurified proteins from rates of H/D exchange

TBHP

t-butylhydroperoxide

TECP

tris(2-carboxyethyl)phosphine

Thr

theronine

ThT

thioflaven T

TNBS

2,4,6-Trinitrobenzenesulfonic acid

TNM

tetranitromethane

TOF

time-of-flight

TPDs

3-trifluoromethyl-3-phenyldiazirines

TPO

thyroid peroxidase

Trp

tryptophan

Tyr

tyrosine

UBA

ubiquitin associated domain of USP5

Ugi

PBS2 uracil-DNA glycosylase inhibitor

Ung

uracil-DNA glycosylase

UPS5

ubiquitin specific protease 5

UV

ultra-violet

UVPD

ultraviolet photodissociation

Val

valine

WRK

Woodward's reagent K

XL-MS

Cross-linking mass spectrometry

XO

xanthine oxidase

ZnF-UBP

Zn-finger ubiquitin-binding domain of USP5

Biography

Xiaoran Roger Liu received his B.E. in Applied Chemistry from Each China University of Science and Technology in 2014. He then moved to The University of Akron. During his M.S. work with Dr. Toshikazu Miyoshi, he focused on structural characterization of carbon fiber precursors during thermal stabilization with solid-state NMR. He is currently working towards his Ph.D. in Washington University in St. Louis with Dr. Michael L. Gross. His research interest is to develop and apply novel mass spectrometry-based approaches in characterizing protein high order structures.

Mengru Mira Zhang was born in Daqing, China and received her B.E. degree in 2014 at East China University of Science and Technology. In 2013, she moved to Ohio and joined the Department of Polymer Science at the University of Akron under the supervision of Dr. Li Jia, focusing on organometallic synthesis and its catalytic application for novel polymers. After obtaining her M.S. degree in 2015, she decided to pursue her Ph.D. degree in Dr. Michael L. Gross lab at Washington University in St. Louis with interests in developing and utilizing multiple biophysical methods to address different biological questions.

Michael L. Gross is a chemist who has worked independently in mass spectrometry since 1968. He began his career at the University of Nebraska - Lincoln, where he was distinguished professor of chemistry and director of an NSF Center for Mass Spectrometry until 1994. He then moved to Washington University in St. Louis as professor of chemistry, immunology, medicine and principal investigator of the WU NIH NIGMS Mass Spectrometry Research Resource. During his career of 50+ years, he has authored and coauthored over 650 publications and book chapters. His primary research interests have evolved to structural proteomics and MS-based biophysics.