Sub-residue Resolution Footprinting of Ligand-Protein Interactions by Carbene Chemistry and Ion Mobility-Mass Spectrometry

Abstract

The knowledge of ligand-protein interactions is essential for understanding fundamental biological processes and for the rational design of drugs that target such processes. Carbene footprinting efficiently labels proteinaceous residues, and has been used with mass spectrometry (MS) to map ligand-protein interactions. Nevertheless, previous footprinting studies are typically performed at residue-level, and therefore the resolution may not be high enough to couple with conventional crystallography techniques. Herein we developed a sub-residue footprinting strategy based on the discovery that carbene labeling produces sub-residue peptide isomers and the intensity changes of these isomers in response to ligand binding can be exploited to delineate ligand-protein topography at sub-residue level. The established workflow combines carbene footprinting, extended liquid chromatographic separation and ion mobility (IM)-MS for efficient separation and identification of sub-residue isomers. Analysis of representative sub-residue isomers located within the binding cleft of lysozyme and those produced from an amyloid-beta segment have both uncovered structural information heretofore unavailable by residue-level footprinting. Lastly, a “real-world” application shows that the reactivity changes of sub-residue isomers at Phe399 can identify the interactive nuances between estrogen-related receptor α, a potential drug target for cancer and metabolic diseases, with its three ligands. These findings have significant implications for drug design. Taken together, we envision the sub-residue level resolution enabled by IM-MS-coupled carbene footprinting can bridge the gap between structural MS and the more-established biophysical tools, and ultimately facilitate diverse applications for fundamental research and pharmaceutical development.

Table of Contens

INTRODUCTION

The interactions between proteins and ligands play key roles in regulating, sensing and signaling a multitude of biochemical processes.¹ Mapping the binding sites between proteins and ligands at molecular level is thus vital for developing drugs that target such processes. Traditional structural analysis approaches including X-ray crystallography and nuclear magnetic resonance (NMR) spectroscopy provide high resolution information but present certain limitations, including time-consuming procedures and requirement for a relatively large amount of crystallized proteins.² With the development of mass spectrometry (MS), structural MS has emerged, and diverse workflows to study high-order structures of protein-ligand interactions have been established.³ Methods including hydrogen/deuterium exchange (HDX), covalent labeling, crosslinking and rapid footprinting deliver structural information regarding protein, protein assemblies and protein-ligand interactions in native systems.⁴

Nevertheless, one of the most frequently used approach, HDX, suffers from H/D scrambling in gas phase upon collision-induced dissociation (CID) and variable levels of amide H/D back-exchange due to the labile nature of deuterium uptake.^5–6 Covalent labeling and crosslinking irreversibly modify reactive amino acid residue sidechains, offering an alternative to HDX. However, these chemical labeling-based approaches are biased towards analysis of protein topography and ligand-protein binding when reactive amino acids are in proximity and thus deliver structural information with relatively low resolution.⁷ Moreover, the temporal resolution afforded by traditional chemical labeling is not amenable to the analysis of dynamic interactions encountered in biological systems.

To this end, rapid footprinting has emerged as a promising approach that addresses some of these limitations. One of such rapid footprinting strategies, fast photochemical oxidation of proteins (FPOP), uses a pulsed laser to generate hydroxyl radical (·OH) that reacts with oxidizable residues and thus can rapidly and stably probe protein surface on a microsecond time scale.^8–9 Although it relies on H₂O₂ and the middle-ultraviolet (UV) laser irradiation for photolysis, which can impair the native structures of certain proteins, radical footprinting features unique advantages in protein topographical analysis.^10–12

Among various footprinting strategies, an appealing alternative to ·OH footprinting is diazirine-based footprinting. This method relies on the gentle¹³, near-UV laser to irradiate diazirine reagents and then efficiently generate highly reactive carbene radicals. Resultantly, the entire protein surface can be effectively labeled on a few nanosecond time scale.¹⁴ Notably, carbene radicals can insert into X-H (X=C, O, N, S) bonds of accessible amino acids on the protein surface. The insertion produces certain mass shift at residue-level, which can be identified by MS with high throughput and sensitivity. Specifically, a variety of aliphatic diazirines including photoleucine, 4,4-azipentanoic acid, 3,3’-azibutan-1-ol, 3,3’-azibutyl-1-ammonium and methylene diazirine have been developed and applied to map protein surface and interaction regions.^15–20 Meanwhile, a recently reported reagent, trifluoromethylaryl diazirine (TFMAD), was reported to provide superior performance with improved labeling efficiency.^{13, 21–22}

Quantitative analysis of the carbene-labeled peptides can measure labeling kinetics and stoichiometry (rates and labeling ratios) at residue-specific level for target proteins, which are dictated by the exposure of amino acids at surface to carbene radicals. Therefore, when protein topography is changed in response to ligand binding or formation of protein complexes, residue-level changes of surface exposure can be examined by altered labeling rates or levels. This carbene-based footprinting method has been successfully applied to identify the substrate-binding domain for a large deubiquitinating protease, ubiquitin-specific protease 5 (USP5),¹³ and map the interfacial sites of an integral membrane protein ompF.²¹

Nevertheless, there is still room for improvement. Notably, one of the current bottlenecks lies in the limit of measurable resolution. To bridge the gap between the residue-level structural MS analysis and the more-established biophysical tools such as X-ray crystallography and NMR for protein structural characterizations, we sought to address this challenging task. Based on a surprising discovery that carbene insertion occurs at different atoms of the same residues and thereby produces peptide isomers at sub-residue level using TFMAD, we hypothesize that, if the carbene footprinting-generated sub-residue level peptide isomers can be differentiated and quantified, it will be attainable to map protein topography and to identify ligand-protein binding interfaces at a sub-residue resolution.

Previous studies have demonstrated differentiation of peptide isomers using ion mobility (IM)-MS. IM-MS separates gaseous ions based on their differences in mass, charge and shape at millisecond time scale, and possesses the unique potential in resolving structural isomers.²³ Thus, it has been applied to resolve peptide isomers carrying post-translational modifications at different sites and stereoisomers with D/L-amino acids.^24–26 Moreover, recent advancements in IM-based gas phase separation such as traveling-wave ion mobility spectrometry (TWIMS) together with collision cross section (CCS) prediction software that enables accurate matching of experimental CCS with predicted values hold potential in assignment of sub-residue level peptide isomers.²⁷

Inspired by these recent technical advancements, herein we developed an effective workflow to enable sub-residue level carbene footprinting for higher-order structural analysis of proteins and ligand-protein interactions. We allowed carbene radicals to attach to different atoms in identical amino acid residues, and the resultant sub-residue isomers of radical reactions are subjected to liquid chromatography (LC) in conjunction with IM-MS for an improved structural separation and quantitative analysis. The quantitative changes of peptide isomers brought about by ligand binding or conformational switching are subsequently translated to protein topography at a sub-residue resolution. The power of this strategy has been demonstrated by its capability of accurately probing the binding cleft of lysozyme and mapping the subtle conformational changes of an amyloid-beta peptide aggregate with improved resolution over traditional residue-level footprinting. Finally, we applied the sub-residue footprinting approach to explore the binding modes for an orphan receptor estrogen-related receptor α (ERRα) with its three ligands. Analysis of the sub-residue isomers of Phe399 in ERRα allows us to gain structural insights regarding how the newly reported agonists interact with ERRα and modulate its activity in a different manner from its inverse agonist when crystallography data of agonists-target is lacking. Therefore, we anticipate that structural MS techniques with improved resolution at sub-residue level in probing ligand-target and protein-protein interactions will contribute to deeper understanding of ligand-protein interactions for both fundamental research and drug development in the future.

EXPERIMENTAL SECTION

Materials

Commercially obtained reagents were purchased from Sigma-Aldrich (St. Louis, MO, USA) unless otherwise noted. TFMAD was synthesized by following the approach described by Manzi et al.¹³ The structure of TFMAD is shown in Figure S1. Hen egg white lysozyme (HEWL) and tetra-N-Acetylglucosamine (NAG4) were purchased from J&K Scientific Ltd. (Beijing, China). Amyloid peptide ¹⁶KLVFFA²¹ was purchased from Guoping Pharmaceutical Company (Anhui, China). The 6xHis-ERRα-LBD plasmid was obtained as described by Greshik et al.²⁸ Compound 1a and DK3 were synthesized as previously described.^29–30 The 6xHis-ERRα-LBD was expressed in BL21 (DE3) competent E. coli cells. Cells were grown at 37 °C until the optical density (OD) reaches 0.4, and then grown at 18 °C for 12 hours with 0.1 mM isopropyl β-D-thiogalactoside. Nickel affinity chromatography was then used to purify harvested 6xHis-ERRα-LBD plasmids. Lastly, purified 6xHis-ERRα-LBD was buffer-exchanged with 20 mM Tris/150mM NaCl buffer by microcon 10 kDa centrifugal filter unit with ultracel-10 membrane (Millipore, Milford, MA, USA).

Carbene Labeling

HEWL and TFMAD were prepared in 20 mM Tris/150mM NaCl buffer with the final concentrations at 100 μM and 10 mM, respectively. The mixture was allowed to equilibrate at room temperature for 15 min. For the ligand-treated group, NAG4 was added to a final concentration of 100 μM before irradiation. Aliquots (10 μL) of the mixtures were then placed in vials and snap-frozen by liquid nitrogen (77K) for diffusion control.^{13, 16} Irradiation was performed by a 349 nm Nd:YLF laser (Laserwave OptoElectronic Technology Co., Beijing, China) with repetition frequency at 1 kHz and pulse energy at 120 μJ for 20s. The laser irradiation time was optimized for labeling efficiency in our case. Nevertheless, the irradiation duration should be shortened and cautiously evaluated for probing dynamic and transient ligand-protein interactions. For ERRα footprinting, purified ERRα, ligands (20 μM, dissolved in 20 mM Tris/150 mM NaCl buffer) and TFMAD (10 mM, dissolved in 20mM Tris/150mM NaCl buffer) were mixed and snap-frozen followed by irradiation at 349 nm as described for HEWL. Influence of laser irradiation on protein structures was evaluated by comparing the enzymatic activity of the model protein lysozyme with and without laser irradiation and shown in Figure S2.

Carbene-labeled proteins were then denatured by 8 M urea for 30 min followed by reduction by 5 mM dithiothreitol for 25 min at 56 °C and alkylation by 15 mM iodoacetamide (avoid light) for 20 min. Samples were then digested by trypsin (1:20, protease/protein ratio) for 12 h at 37 °C. The digested samples were desalted by C₁₈ ZipTips (Millipore) and subjected to MS analysis.

Lyophilized Aβ_16–21 KLVFFA peptide was solubilized in phosphate buffer saline (PBS) at 2.5 mM and then incubated for 3 days with agitation to induce aggregation.³¹ Its turbidity at 450 nm was measured on a Synergy H1 Hybrid Multi-Mode Microplate Reader (BioTek, USA). For the control group, ¹⁶KLVFFA²¹ was solubilized without incubation. TFMAD prepared at 20 mM in PBS was added to the control and aggregated peptide samples separately at 1:1 ratio (v/v). The mixtures were irradiated immediately followed by desalting as described above.

LC-MS/MS Coupled to Ion Mobility Spectrometry

Samples were analyzed by a Waters Synapt G2-Si mass spectrometer coupled with a Waters nanoAcquity ultra-performance LC system (Milford, MA, USA). A Waters Symmetry C₁₈ trapping column (180 μm × 20 mm, 5 μm) and a Waters HSS T3 column (150 mm × 75 μm, 1.8 μm) were used for LC separation. Mobile phase A and B consisted of 0.1% formic acid in water and 0.1% formic acid in acetonitrile (ACN), respectively. The LC gradient was set as following: 0–1 min, 0–1% phase B; 1–72 min, 1%– 35% phase B; 72–74 min, 35–85% phase B; 74–76 min, 85% phase B; 76–78 min, 85%−1% phase B followed by 12 min re-equilibration with a steady flow rate kept at 0.3 μL/min. ESI parameters were set as following: 3 kV capillary voltage, 30 V sampling cone and 80 °C source temperature. Spectra were acquired over a mass range from m/z 350–2000 in data-dependent acquisition (DDA) mode. Ten most intense ions for each scan were subjected to collision induced dissociation (CID) using a Collision Energy (CE) Ramp setting.

For ion mobility separation, TWIMS was calibrated with polyalanine as described by Bush et al.³² TWIMS parameters were as following: nitrogen pressure of 3 mbar, wave height of 40 V, wave velocity of 550 m/s. Targeted MS/MS acquisition coupled with IM separation was used to measure the arrival time for fragment ions of carbene-labeled peptide isomers within a mass range of m/z 50–1200. The trap CE for targeted MS/MS acquisition was optimized in a range of 10–40 V. For quantitative analysis of peptide isomers, continuous MS spectra in a mass range of m/z 50–2000 were acquired with 0.2 s scan time/spectrum.

Data Analysis

Data acquisition was carried out in MassLynx 4.1 (Waters). Raw files collected in DDA mode were searched against HEWL and ERRα sequences using PEAKS Studio 8.5 (Bioinformatics Solutions Inc., Waterloo, Ontario, Canada). A mass tolerance of 10 ppm was allowed for precursor ions and 0.02 Da for fragment ions. Carbamidomethylation (+57.0215) of cysteine was set as fixed modification. Variable modification included oxidation (+15.9949) on methionine and carbene modification (+202.0242) on any residue. Quantlynx 4.1 was used to quantify peptides using .raw files collected in MS scan mode. For peptide-level footprinting, the fraction of carbene labeling (label ratio) was determined by integrating the chromatographic peak areas (A) for unlabeled and labeled peptides^33–34, respectively (eq 1).

$$\mathrm{Label}{\mathrm{Ratio}}_{peptide}=\frac{A_{labeledpeptide}}{A_{unlabeledpeptide}+A_{labeledpeptide}}$$ (1)

For residue-level footprinting, an integral strategy was used. For the carbene-labeled isomeric peptides that were not fully resolved by LC, the label ratio for each identified residue/sub-residue isomer was determined using tandem MS as previously described by Jumper et al16 and Manzi et al¹³ (Supporting Methods). Notably, measuring residue-level carbene labeling by ergodic fragmentation such as collision-induced dissociation (CID) and high-energy CID (HCD) is prone to mis-quantification by acquiring negative label ratios for certain residues due to the site of carbene labeling altering fragmentation pathways, as well as CID/HCD being more prone to neutral mass losses. Electron-transfer dissociation (ETD) is a better alternative in MS/MS ions-based residue-level quantitation when such instrumentation setup is available.¹⁶ Meanwhile, for the baseline-resolved carbene-labeled isomeric peptides, the labeling proportion per residue was determined by integrating the chromatographic peak area of the labeled peptide carrying carbene modification on a given residue (A_{labeled residue}), that of the unlabeled peptide (A_{unlabeled peptide}), and that of the labeled peptide (A_{labeled peptide}) (eq 2). For A_{labeled residue}, the peak area was calculated by combining across the set of peaks corresponding to sub-residue level peptide isomers for a given residue. For A_{labeled peptide}, peptide forms that carry carbene modification on all the residues for a given peptide were included for quantification.

$$\mathrm{Label}{\mathrm{Ratio}}_{residue}=\frac{A_{labeledresidue}}{A_{unlabeledpeptide}+A_{labeledpeptide}}$$ (2)

For sub-residue level footprinting, sub-residue level isomers produced by carbene labeling at different atoms of an identical residue were assigned and differentiated by MS/MS and IM-MS. The arrival time of fragment ions from distinct sub-residue isomers were recorded with DriftScope 2.8 (Waters). The label ratio of each sub-residue isomer was calculated by integrating the chromatographic peak area of a given carbene-labeled sub-residue isomer (A_{labeled isomer}), that of the unlabeled peptide (A_{unlabeled peptide}), and that of the labeled peptide (A_{labeled peptide}) (eq 3).

$$\mathrm{Label}{\mathrm{Ratio}}_{isomer(i)}=\frac{A_{isomer(i)}}{A_{unlabeledpeptide}+A_{labeledpeptide}}$$ (3)

Two-way analysis of variance (ANOVA) followed by Fisher’s least significant difference (LSD) test was used for studying the label ratio changes (indicative of reactivity alterations due to ligand binding) among the sub-residue positional isomers with p<0.05 considered as significant difference. For the analysis of sub-residue isomers in ¹⁶KLVFFA²¹, the sub-residue level isomeric peptides modified at Lys16 were identified, and the corresponding label ratios in monomeric and aggregation states were compared for each isomer using Student’s t-test. For the analysis of ligand-bound ERRα, one-way ANOVA followed by Fisher’s LSD test was used for studying whether any of the three ligands binding leads to reactivity changes for each sub-residue isomer at Phe399.

AutoDock Vina was used for docking simulations.³⁵ DeepCCS software was used to predict CCS values with deep neural network.²⁷ Structures were displayed with PYMOL (Schrödinger LLC). Further information about CCS measurement and prediction can be found in Supporting Information.

RESULTS AND DISCUSSION

Sub-residue resolution carbene footprinting enabled by ion mobility mass spectrometry

The carbene footprinting approach has been demonstrated as a powerful tool in probing ligand binding sites and mapping high-order structures of proteins.^{13, 16, 19, 21} Upon laser irradiation, carbene radicals can attack X-H bonds and thus are inserted into the surface of proteins. Therefore, amino acid residues located in the region involved in ligand-protein interaction would become less labeled due to the steric hindrance induced by ligand binding. Quantitative analysis of the carbene-labeled peptides thus enables localization of specific residues whose reactivities exhibiting pronounced changes as putative ligand-binding sites. Nevertheless, previous carbene footprinting strategies often employed a short LC gradient (<40 min).^{13, 16, 18–19, 21} This acquisition approach sacrifices chromatographic resolution. In this study, we employed an extended LC gradient (~70 min) for improved separation capability. Structural isomers generated by carbene inserted to different atoms of identical residue were thereby separated and analyzed by IM-MS. The recorded arrival time and CCS values can subsequently be used to differentiate and assign sub-residue peptide isomers. This IM-MS-based footprinting strategy improves the carbene footprinting capability from residue to sub-residue level as illustrated in Figure 1. Therefore, we anticipate that this approach would open up a new window to map the higher-order structures of proteins, protein-protein assemblies and ligand-protein interactions with greater structural details, making structural MS a complementary tool of high resolution to couple with the well-established biophysical techniques.

Figure 1.

The workflow of sub-residue level carbene footprinting approach enabled by LC-IM-MS. With the aid of extended LC gradient and IM-based gas phase separation, sub-residue isomers generated by carbene labeled on different atoms from an identical amino acid residue can be differentiated and used to map protein conformational changes in response to ligand binding with higher resolution compared to conventional residue-level footprinting strategies.

Quantitative analysis of sub-residue isomers provides conformational footprints with improved resolution

We first verified the proposed strategy using a model protein, HEWL and its ligand NAG4. Carbene footprints showed that the intensity of the carbene-labeled peptide ⁶²WWCNDGR⁶⁸ decreased significantly in response to NAG4 incubation (Figure 2A). This finding agrees with the location of this peptide within the binding cleft of lysozyme (Figure 2B, Figure S3).¹³ Among all the residues of this protein, we noted that Trp62 exhibited the most significantly reduced reactivity due to NAG4 binding (Figure S3), corroborating its previously reported essential role in carbohydrate recognition and catalytic function.^36–37 Specifically, both accurate mass and MS/MS fragmentation analysis suggest that three peptides carrying carbene modification on residue Trp62 yet sharing identical sequence were detected with retention time at 57.8 min, 58.2 min and 58.7 min, respectively (Figure 2C, Figure S4). More interestingly, they displayed nearly indistinguishable MS/MS fragmentation patterns (Figure 2D, Figure S4). This phenomenon is reminiscent of a previous study of FPOP that reported peptide isomers generated by ·OH labeling at different atoms of the same residue.^38–39 Therefore, we hypothesized that these three isomers may be resultant products of carbene radicals inserted to different X-H bonds of the same residue Trp62. To validate the postulation, we employed IM-MS to resolve these isomers. We first measured the arrival time of MS/MS fragment ions produced by ⁶²W^carbeneWCNDGR⁶⁸. Figure 2E and Figure S5 showed that the serial y ions from isomers conferred identical arrival time, whereas the carbene-labeled a₁ ion exhibited different arrival time values, validating the existence of multiple sub-residue isomers differing in Trp62 (Table S1).

Figure 2.

Quantitative analysis of sub-residue isomers provides conformational footprints of higher resolution than residue-level footprinting using HEWL and NAG4 as model system. (A) Label ratio of peptide ⁶²WWCNDGR⁶⁸ in the absence (blue) and presence of NAG4 (red). (B) The molecular interactions between HEWL and NAG4 (based on PDB 1LZC). The residue Trp62 was highlighted in red and ⁶³WCNDGR⁶⁸ was highlighted in yellow. (C) The extracted ion chromatogram (XIC) of carbene labeled peptide ⁶²WWCNDGR⁶⁸. (D) Representative MS/MS spectra of two sub-residue level isomers that carry carbene modification at different atoms in Trp62. (E) The representative extracted arrival time distributions (ATDs) of fragment ions from the three sub-residue peptide isomers of Trp62. The shifted arrival time (statistical analysis results specified in Table S1) was marked with tick, whereas no shift observed was marked with cross. The raw extracted ATDs were summarized in Figure S5 in Supporting Information. (F) Label ratios of Trp62-labeled peptide isomers quantified by sub-residue level footprinting analysis. Bars are shown as mean ± s.d. (standard deviation), and significant differences are marked (n = 3, two-way ANOVA, p < 0.05 as *). The statistical details are provided in Table S2, SI. (G) Label ratios of Trp62-labeled peptides quantified by residue-level footprinting analysis. Bars are shown as mean ± s.d. and significant differences are marked (n = 3, Student’s t-test, p < 0.05 as *).

Next, we asked whether the sub-residue structural isomers are capable of mapping protein topography at higher spatial resolution than the conventional footprinting approaches. We thus quantified the labeling ratios for the three major sub-residue Trp62 isomers in the control and NAG4-treated groups, and found that they all displayed decreased label ratios in response to NAG4 binding (Figure 2F). Moreover, the label ratio changes among the three sub-residue isomers were significantly different (by two-way ANOVA, Table S2).This finding reveals that subtle reactivity changes at distinct atoms of identical residues induced by ligand binding can be attained by sub-residue footprinting, which can thus deliver more detailed structural information regarding NAG4 binding-induced accessibility alterations than the residue-level analysis (Figure 2G). To our knowledge, the discovery reported in this study represents the first attempt to exploit sub-residue peptide isomers in mapping protein topography and identifying ligand-protein binding interfaces for improved spatial resolution.

In the case of HEWL footprinting, we found 17 types of amino acids residues were labeled by TFMAD reagents. This accords to previous knowledge that singlet carbene, the radical generated by photoactivated TFMAD, is prone to insert into O-H, N-H and S-H bonds compared to C-H bonds.¹⁸ Meanwhile, the label ratio, indicative of carbene reactivity, of the same amino acid/sub-residue position is also affected by the microenvironment. For instance, we found the label ratios of Trp residue located at different domains of lysozyme varied, and the values showed good correlation with solvent accessible surface area (SASA) values (correlation factor ~0.89, Figure S6). The influence of exposure/orientation on residue/sub-residue reactivity to probes has also been reported in not only rapid radical footprinting methods including carbene footpinting and Fast Photochemical Oxidation of Proteins,^{16, 40} but also other chemical modification approaches using reagents such as benzhydrazide and diethylpyrocarbonate.^41–42 Therefore, by comparing the label ratios of identical residues/sub-residue with and without ligand incubation, we can infer the ligand binding sites whose reactivities are significantly affected by ligand-protein interactions.

Differentiation and assignment of sub-residue carbene-labeled isomers for amyloid peptides

After demonstrating that carbene footprinting produces sub-residue peptide isomers and enables structural analysis of proteins at improved spatial resolution, we wondered whether the identities of the sub-residue level isomers can be assigned. We chose to investigate the conformational changes of an Aβ segment ¹⁶KLVFFA²¹.³¹ These segment peptides are known to be capable of self-assembling to well-organized fibers that have recapitulated the essential properties of those generated from the full-length parent protein.⁴³ Therefore, they are viewed as an Aβ pharmacophore, and show great potential in high-throughput screening of small molecules that inhibit Aβ fibrillation for Alzheimer’s disease (AD) treatment.^{31, 43} Besides the region ¹⁷LVFFA²¹ that makes up the central hydrophobic cluster (CHC) during Aβ fibrilogenesis,³¹ increasing evidences points to the possibility that the nucleophilic Lys16 is also of paramount importance for modulating Aβ assembly equilibrium.⁴⁴ Specifically, a previous study that designed molecular tweezers targeting Lys16 managed to inhibit the aberrant Aβ aggregation and toxicity.⁴⁵ Consequently, more knowledge regarding the conformational changes of Lys16 when the ¹⁶KLVFFA²¹ peptides switch from monomeric to aggregation state gained via sub-residue footprinting is expected to shed light on the rational design of chemicals targeting this region.

As shown in Figure 3A, we found that three isomers carrying carbene modification at Lys16 were identified. The corresponding MS/MS spectra of the acquired sub-residue isomers are different. For instance, the low mass ion of lysine, originally at m/z 84.08 and shifted to m/z 286.10 due to carbene labeling, was only present in the MS/MS spectra of sub-residue isomer #1 and isomer #2 (arbitrarily assigned by retention time). Additional MS/MS acquisition experiments conducted in a range of ramped collision energy values (5–50 V) showed isomer #3 cannot yield this fragment ion (Figure 3B). Based on the production pathway of this low mass ion (Figure S7), we inferred that isomer #3 carries carbene modification at the N-terminal.

Figure 3.

Differentiation and assignment of sub-residue carbene-labeled peptide isomers for an Aβ segment ¹⁶KLVFFA²¹. (A) The XIC of carbene-labeled ¹⁶KLVFFA²¹ with three sub-residue isomers of Lys16 being highlighted. (B) The XIC of carbene labeled lysine low mass ion at m/z 286.10. (C) The mass shift of fractionated carbene-labeled ¹⁶KLVFFA²¹ isomers induced by reductive dimethylation using excess CD₂O and NaBH₃CN. The modification at amino groups was recorded. (D) Label ratios of the whole peptide, of the region of ¹⁷LVFFA²¹ and of the residue Lys16 when the peptide is in monomer and aggregation states. Bars are shown as mean ± s.d. and significant differences are marked (n = 4, Student’s t-test, p < 0.05 as *). (E) Label ratios of sub-residue isomers of Lys16 in monomer and aggregation state. Bars are shown as mean ± s.d. and significant differences are marked (n = 4, Student’s t-test, p < 0.05 as *). (F) ¹⁶KLVFFA²¹ segments are stacked as pairs with the hydrophobic β-sheets bound together, forming a basic unit of fiber. The structure is shown based on PDB file 3OW9.

To further accurately assign these isomers, we fractionated the peptide isomers by high performance liquid chromatography (HPLC) and employed reductive dimethylation reaction to modify the available primary amine groups using deuterated formaldehyde (CD₂O). We found peptides with carbene labeled at residues in ¹⁷LVFFA²¹ all showed a mass shift corresponding to two dimethylation modifications since these peptides have two free amine groups, the N-terminal and ε-NH₂ in lysine (Figure 3C). In contrast, sub-residue isomer #1 showed a mass shift corresponding to single dimethylation + methylation, whereas isomer #2 and #3 conferred a mass shift corresponding to dimethylation. We reasoned that carbene labeling at the side chain -CH₂- of Lys16 may thwart the reductive dimethylation of amine due to steric hindrance brought by the relatively bulky carbene group, leading to incomplete dimethylation of the ε-NH₂ in Lys16. Therefore, we assigned isomer #1 as the form that has carbene labeling occurred at the Lys16 side chain -CH₂- groups. The single dimethylation of isomer #2 is identified as the form that has carbene-labeling on ε-NH₂ of Lys16. This can be explained by the electrophilic nature of the trifluoromethylaryl group that results in diminished reactivity of ε-NH₂ and the N-terminal to CD₂O (Figure S8).⁴⁶ Such assignment is cross-validated by the different charge states among the isomers (Figure S9). Furthermore, it is also supported by the presence of carbene diagnostic ions in MS/MS spectra of isomer #2 and #3, since these two forms are prone to lose carbene modification during CID due to the lower bond energy of C-N bond compared to C-C bond (Figure S9).

After assigning the sub-residue isomers, the next intuitive question is whether the reactivity of these atoms experience changes during Aβ aggregation. We used an aggregation model to mimic an early-stage onset of ¹⁶KLVFFA²¹ aggregation (Figure S10), and examined the concomitant reactivity change of Lys16. We first found the overall label ratio of the peptide ¹⁶KLVFFA²¹ and that of the CHC region (¹⁷LVFFA²¹) almost held constant, whereas the peptide carrying carbene labeling specifically at Lys16 (¹⁶K^carbeneLVFFA²¹) showed an increase in intensity upon aggregation compared with the monomer state (Figure 3D). This suggests increased accessibility of Lys16 in aggregated peptides. Then, we asked how the sub-residue isomers at Lys16 changed between the two states. Quantitative analysis revealed that the ε-NH₂ and N-terminal of ¹⁶K^carbeneLVFFA²¹ both conferred higher reactivity to carbene labeling in the aggregation state (Figure 3E), indicative of their increased exposure to solvents. In contrast, the isomers carrying labels at side chair -CH₂- did not show significant reactivity differences between the two states.

The increased exposure of ε-NH₂ and the constant exposure of CH₂ in Lys16 upon aggregation is in line with previous X-ray crystallization results that reported, once in aggregation state, ¹⁶KLVFFA²¹ peptides form steric zippers with the hydrophobic region ¹⁷LVFFA²¹ stacking closely yet exposing the ε-NH₂ of Lys16 to solvents (Figure 3F, Figure S11).^47–48 Nonetheless, it is noteworthy that none of the previous investigations have clarified the role of N-terminal in the amyloid-forming process when using this self-complementary segment ¹⁶KLVFFA²¹ for drug screening. For instance, a study reported the design of multiple compounds named Binders of Amyloid Fibers (BAFs) based on the hydrogen bonds formed between the hydroxyl, carbonyl and anion groups in BAFs and the polar ε-NH₂ of Lys16.³¹ Our sub-residue footprinting indicates that the exposed N-terminal upon aggregation may compete with ε-NH₂ for available BAFs, potentially leading to disturbed ligand interactions and compromised accuracies regarding the inhibitory effect of BAFs on Aβ aggregation. Based on these results, we proposed that capping of the active N-terminal by modification such as acetylation and N-Hydroxysuccinimide (NHS) ester will improve the screening accuracy for Aβ aggregation inhibitors if the steric zipper peptide ¹⁶KLVFFA²¹ is employed as an Aβ model. It should be noted that the structures of monomeric ¹⁶KLVFFA²¹ is lacking in this case, since current structural examination of Aβ segments is based on crystallography accomplished by X-ray crystallography and NMR crystallography, both of which require the formation of Aβ segment crystals. Nevertheless, this state cannot be obtained from monomeric Aβ segment peptides. This dilemma again reiterates the usefulness of carbene footprinting in probing topography when no crystallography data is available. It is also anticipated that the development of other biophysical methods may cross-validate the structural information generated by sub-residue level carbene footprinting in the future.

Consequently, the importance in probing higher-order structures of protein assembly and ligand-protein complexes at sub-residue resolution is well-demonstrated using this Aβ segment, since such structural details of Lys16 would be missing if a conventional residue-level footprinting strategy is used. Meanwhile, compared to the laborious and time-consuming X-ray crystallization, our approach accomplishes sub-residue level structural analysis for aggregated amyloid peptides within one day, making footprinting a promising tool complementary with current biophysical techniques.

Sub-residue level footprinting of drug-target interactions

Lastly, we applied the sub-residue level footprinting approach to precisely probe the drug-target interactions for a “real world” application in drug research and development (R&D). ERRα is chosen as a representative drug target in this study. As one of the oldest nuclear orphan receptor, its functions and therapeutic values have been extensively studied.⁴⁹ For example, the activation of ERRα influences lipid metabolism and energy homeostasis, and can be used to combat metabolic diseases such as diabetes and obesity.⁵⁰ Moreover, increased activity of ERRα has been closely associated with promoted osteoclastogenesis and myogenesis, and suppressed macrophage cytokine production.⁵¹ Nevertheless, coordinated upregulation of ERRα and its coactivator peroxisome proliferator-activated receptor coactivator-1α (PGC-1α) has also been observed in breast cancer models,⁵² suggesting the potential of inhibiting ERRα for effective anti-cancer therapy. Taken together, the complexity of disease-specific ERRα regulatory consequences thus necessitates the development of both inverse agonists and agonists for finely modulating the activity of ERRα.⁵³

Until now, inverse agonists of ERRα such as compound 1a and XCT-790 have been discovered by high-throughput drug screening.⁵³ The interaction between compound 1a and the ERRα-ligand binding domain (LBD) has been further elucidated by crystallography (PDB: 2PJL).³⁰ Nevertheless, no experimental data is available regarding the interaction between ERRα and its agonists, which has hindered the understanding of structure-driven activation mechanism and rational design of ERRα agonists with greater potency. Although molecular docking analysis has provided some clues for ligand-ERRα interactions, we aimed to experimentally investigate the binding modes of ERRα with the three ligands including an inverse agonist compound 1a and two agonists DK3 and cholesterol. The structural insights are expected to facilitate future development of ERRα modulators with distinct regulatory functions.

Using the peptide-level footprinting approach, the peptide ³⁹⁵VLAHFYGVK⁴⁰³ located in the ligand binding region was chosen for further study due to its significantly decreased intensity upon incubation with ligands (Figure S12). According to previous crystal structure and dynamic simulation, Phe399 in peptide ³⁹⁵VLAHFYGVK⁴⁰³ plays a key role in forming ligand binding pocket. Meanwhile, sub-residue level footprinting analysis suggests three sub-residue level isomers of Phe399 residue were detected for this peptide (Figure 4A, Figure S13-S14). The arrival time of serial y ions combinatorially differentiated the three peptide isomers possessing identical sequence ³⁹⁵VLAHF^carbeneYGVK⁴⁰³ (Figure 4B, Figure S15, Table S3). Subsequently, due to the lack of modifiable group for Phe399 that differs from Lys16 in KLVFFA, we used the experimental CCS values measured for the carbene-labeled fragment ions and matched those with theoretical ones that are calculated by deep neural network-based software DeepCCS. DeepCCS predicts the gas-phase CCS values of ions based on their simplified molecular-input line-entry system (SMILES) notation with high accuracy (median relative error of 2.7%).²⁷ Meanwhile, this requires only a small amount of processing power compared with quantum mechanics (QM)-based molecular modelling, making DeepCCS suitable for large-scale assignment of carbene-labeled sub-residue isomers. Nevertheless, further use of QM-based modelling for CCS value prediction is also highly valued as a complementary method to DeepCCS.

Figure 4.

Sub-residue level footprinting differentiates binding interactions between ERRα and its three ligands. (A) The XIC of carbene labeled peptide ³⁹⁵VLAHFYGVK⁴⁰³. (B) Representative extracted ATDs of fragment ions from the three sub-residue peptide isomers of Phe399. The shifted arrival time (statistical analysis results specified in Table S3) was marked with tick, whereas no shift observed between the isomeric ions was marked with cross. The raw extracted ATDs were summarized in Figure S15. (C) Label ratios of Phe399 residue in the absence and presence of the three ligands. Bars are shown as mean ± s.d., and significant differences are marked (n = 4, one-way ANOVA, p < 0.05 as *). (D) Label ratios of each sub-residue isomer at Phe399 in the absence and the presence of the three ligands. Bars are shown as mean ± s.d., and significant differences are marked (n = 4, one-way ANOVA, p < 0.05 as *). (E) Molecular docking of the region in proximity to Phe399 when ERRα is in a ligand-bound state. The structure was based on the PDB file 2PJL. Color scheme: red = Phe399 residue, blue = compound 1a, magentas = DK3, yellow = cholesterol.

By CCS value comparison, we assigned the three sub-residue level peptide isomers as carbene labeled-Phe399 at meta, ortho and para position (numbered as isomer #1, #2 and #3, respectively) (Figure S16, Table S4). To further examine our assignment of positional isomers tentatively made by CCS matching, we have additionally synthesized a shorter Phe-containing peptide VLAHFVK comparing to ERRα peptide ³⁹⁵VLAHFYGVK⁴⁰³, and found the retention order of the Phe isomer-containing peptides is the same as the full-length peptide with great consistency between the predicted CCS values and experimentally measured ones (Table S4). Experimental assignment of the detected sub-residue isomers can be accomplished by fractionating and collecting each positional isomer followed by NMR analysis. Nevertheless, this has been proven as a highly challenging task due to isomeric peptide co-elution. Alternatively, we proposed to synthesize the carbene-labeled amino acid isomers and use these isomeric amino acids for peptide standard synthesis. This ongoing work requires tremendous efforts in chemical synthesis, analytical separation and spectroscopic analysis, yet is expected to definitively discern and assign peptides carrying sub-residue positional isomers generated by carbene labeling.

Next, we were interested to explore whether the high-resolution structural information obtained by sub-residue footprinting can distinguish the binding modes to ERRα for the three ligands. We first calculated the residue-level reactivity changes and noted the decreased reactivity of Phe399 after ligand incubation according to one-way ANOVA analysis (Figure 4C). However, no significant difference is observed when comparing reactivity changes of Phe399 among the compound 1a, DK3 and cholesterol-treated groups. Then, we quantified the intensity differences of the three sub-residue isomers of Phe399 among the ligand-treated groups and compared with the control group, respectively (Figure 4D). For para site-modified Phe399, we found the carbene label ratios of this specific peptide form all decreased significantly in the compound 1a, DK3 and cholesterol-incubated groups when compared with the vehicle group. In contrast, the label ratio of carbene modification at ortho-Phe399 was not affected by any of the tested ligands (Figure 4D). This agrees with the docking results showing that the ortho site of Phe399 is in greater proximity to the protein backbone (Figure 4E), which probably lowers its sensitivity to ligand binding events. Additionally, the intensity of the peptide carrying carbene modification at the meta site was not significantly affected by compound 1a and DK3-treatment, whereas it pronouncedly decreased due to cholesterol binding. This observation can be explained by the proximity of the long and flexible hydrocarbon chain of cholesterol to the meta site according to the molecular docking results (Figure 4E). Taken together, the stronger interaction of cholesterol at meta site compared to the inhibitor compound 1a may contribute to the stabilization of Phe399 in Helix 11 and subsequently adjacent Helix 12, which is normally considered as the coactivator binding site of PGC-1α for inducing ERRα activation.^54–55 This stabilizing effect differs from the agonist DK3, and opposes to the inverse agonist compound 1a, which is known to displace Helix 12 and leads to interrupted ERRα-PGC-1α binding and decreased ERRα activity.³⁰

In summary, our experiment directly probed the binding sites of ERRα and its inverse agonist/agonists, which have never been conducted with experimental means. The sub-residue footprinting strategy unravels interaction nuances among the meta, ortho and para site-modified isomers at sub-residue level for Phe399. Such knowledge is expected to contribute to uncover the precise structure-based mechanisms regarding how small molecules activate ERRα, which has remained in dark and hindered the design and development of more effective agonists of therapeutic value. Beside drug design, the structural details of molecular interactions between ligand-proteins and protein aggregates/complexes also show potential in promoting the development of target engagement-based lead screening assays.

CONCLUSIONS

Herein we proposed a footprinting strategy with improved resolution (sub-residue level) by using extended LC in conjunction with IM-MS. This method allows us to finely and accurately map the protein-ligand binding sites and probe protein/peptide conformational changes at sub-residue resolution. The wide applicability of this strategy in drug screening and development has been initially demonstrated with an amyloid peptide ¹⁶KLVFFA²¹ and a drug target ERRα. The sub-residue footprinting approach is also applicable to other radical-based footprinting approaches. With further advancement in chromatography and IM-based gas phase separation, we anticipate that the sub-residue level footprinting will be complementary with biophysical approaches by promoting diverse applications including structural characterization and target engagement-based screening for both fundamental research and pharmaceutical industries due to its speed, robustness, simplicity and amenability to native systems.