Taming Conformational Heterogeneity on Ion Racetrack to Unveil Principles that Drive Membrane Permeation of Cyclosporines

Abstract

Our study reveals the underlying principles governing the passive membrane permeability in three large N-methylated macrocyclic peptides (N-MeMPs): cyclosporine A (CycA), Alisporivir (ALI), and cyclosporine H (CycH). We determine a series of conformers required for robust passive membrane diffusion and those relevant to other functions, such as binding to protein targets or intermediates, in the presence of solvent additives. We investigate the conformational interconversions and establish correlations with the membrane permeability. Nuclear magnetic resonance (NMR) and cyclic ion-mobility spectrometry-mass spectrometry (cIMS-MS) are employed to characterize conformational heterogeneity and identify cis-amides relevant for good membrane permeability. In addition, ion mobility selected cIMS-MS and infrared (IR) multiple-photon dissociation (IRMPD) spectroscopy experiments are conducted to evaluate the energy barriers between conformations. We observe that CycA and ALI, both cyclosporines with favorable membrane permeabilities, display multiple stable and well-defined conformers. In contrast, CycH, an epimer of CycA with limited permeability, exhibits fewer and fewer stable conformers. We demonstrate the essential role of the conformational shift from the aqueous cis MeVal11–MeBmt1 state (A1) to the closed conformation featuring cis MeLeu9–MeLeu10 (C1) in facilitating membrane permeation. Additionally, we highlight that the transition from A1 to the all-trans open conformation (O1) is specifically triggered by the presence of CaCl₂. We also capture a set of conformers with cis Sar3–MeLeu4, MeLeu9–MeLeu10, denoted as I. Conformationally selected cIMS-MS and IRMPD data of [CycA+Ca]²⁺ show immediate repopulation of the original population distribution, suggesting that CaCl₂ smooths out the energy barriers. Finally, our work presents an improved sampling molecular dynamics approach based on a refined force field that not only consistently and accurately captures established conformers of cyclosporines but also exhibits strong predictive capabilities for novel conformers.

Introduction

The precise regulation of interactions between peptides and cellular membranes holds significant potential for unlocking a broad range of innovative applications including cargo delivery, tissue labeling, cellular imaging, and therapeutic intervention. Within this realm, N-methylated macrocyclic peptides (N-MeMPs) have emerged as a class of cyclic compounds exhibiting promising medicinal properties. Since their molecular weights (MW) are 600 Da or higher, their chemical actions are not confined by Lipinski’s rule of 5, which describes traditional therapeutics with a molecular mass of less than 500 Da, no more than 5 hydrogen bond donors, no more than 10 hydrogen bond acceptors, and an octanol–water partition coefficient not greater than 5.¹⁻³ Their biochemical and therapeutic attributes are expected to fill the gap in chemical space between small molecules and linear peptides.⁴⁻⁸

Approximately 75% of disease-relevant proteins involved in intracellular protein–protein interfaces (PPIs) are considered “undruggable” by small molecules.⁹⁻¹¹ Cyclosporine A (CycA), an 11-residue cyclic peptide that binds at the interface between cyclophilin A (CypA) and two calcineurins, illustrates a clear example that macrocyclic compounds, including N-MeMPs, can potentially address this disparity.

A significant hurdle remains: most existing, and notably the most exciting, N-MeMP drugs (e.g., CycA) originate from natural products, and we appear to be unable to design new drugs that work similarly to CycA. The relatively limited deployment of cyclic peptide medicinal agents thus far has resulted from the lack of strategies to design membrane-permeable cyclic peptides. Membrane permeability is a critical property that is essential for any drug to target intracellular proteins. Consequently, over the past two decades, only 18 cyclic peptides have been approved for clinical use, with only two that target intracellular proteins (Romidepsin and Voclosporin).¹²⁻¹⁴ This starkly contrasts the approval of 32 small-molecule drugs in 2021¹³ and 16 in 2022.¹⁵ None of the recently approved small-molecule drugs target PPIs or could potentially replace cyclic peptide drugs such as CycA and Voclosporin (a synthetic cyclosporine). Therefore, a comprehensive understanding of the factors that control the membrane permeability of N-MeMPs is crucial to address the bottleneck in the drug design and will unlock their unique pharmaceutical novelty.

CycA, with a primary sequence of cyclo(MeBmt1—Abu2—Sar3—MeLeu4—Val5—MeLeu6—Ala7—d-Ala8—MeLeu9—MeLeu10—MeVal11), revolutionized the field of organ transplants.¹⁶⁻¹⁸ CycA contains seven N-methylated residues, several consecutive, implying that up to seven amide bonds can adopt cis or trans configurations. Mathematically, this results in up to 2⁷ = 128 possible conformers solely classified based on backbone cis/trans configurations. This number is substantial, especially when contrasted with fewer than a handful of available high-resolution structures. Within the cyclosporine family, one structural modification, such as (de)methylation or stereoinversion, can drastically alter the structures, biological target(s), and membrane permeability.¹⁹⁻²³

The same cyclosporine can bind to distinct protein targets with unique bound states. For example, CycA binds at the interface between CypA and calcineurins to exert its immunosuppressant property²⁴⁻²⁶ with a conformer with all-trans amides (O1) and also binds to the Regulator Gene of Glucosyltransferase (RGG) in Streptococcus to exhibit its inhibitory effect with a conformer that has three cis-amides.²⁷ This is intriguing because it goes against the traditional model in protein–ligand theory, which assumes rigid ligands and flexible protein active sites. In contrast, cyclosporines exhibit enough flexibility to adjust their conformation to fit the groove-shaped binding sites.

Before the report of the aqueous conformer of CycA,²⁸ the closed (C1) form²⁹ with a cis MeLeu9–MeLeu10 in apolar solvents and the open (O1) form in the protein complex²⁴⁻²⁶ formed the basis for the passive membrane permeability hypothesis, where C1 is thought to exist in the membrane environment, and O1 exists in aqueous solution.^19,30 This hypothesis and limitations in high-resolution structural data have resulted in contradictory structure–function data. For instance, cyclosporines exhibiting comparable biological activities, such as CycA and cyclosporine G (CycG),^31,32 were erroneously believed to assume different conformations.²⁰ It is noteworthy that both CycA and CycG primarily adopt the A1 conformation in solution, but while A1 of CycG was documented in 2003,³³ that of CycA²⁸ was only reported two decades later (in 2022).

The presence of diverse conformations is the enabling factor for the unique and diverse functions of cyclosporines. However, we have yet to understand the extent of this conformational space and the potential links between various conformation states (or structural dynamics, the interconversion among these conformers) and the functional roles.

This work establishes the key principles that drive membrane permeation in the cyclosporine family. The relationship between conformational heterogeneity, isomerism, and passive membrane diffusion of cyclic peptides has not yet been fully understood. Evidently, analogues stabilizing either O1 or C1 failed to achieve the desired membrane permeability and affinity to CypA.^19,21,34 The membrane permeability of cyclosporines is likely dictated by interconversions among a set of conformers (aqueous, membrane-bound, and intermediates in between) rather than individual conformers.³⁰ The problem at hand is to identify precisely what all possible conformers are and how the environments can stabilize each of them and thus their permeabilities. To validate this hypothesis, this work (a) investigates cyclosporine analogues, chosen due to their differences in membrane permeability, and (b) synergizes a collection of bioanalytical techniques for a robust sampling of cyclosporine conformational space.

In particular, we chose to work with CycA, Alisporivir (ALI), and cyclosporine H (CycH). ALI is a synthetic nonimmunosuppressant analogue, developed by Debiopharm, that has comparable membrane permeability to CycA and also binds strongly to CypA.^35,36 ALI was strategically designed for differing from CycA at only two residues, d-MeAla3 and EtVal4, so it does not bind calcineurin.^34,37,38 ALI is currently being researched for potential use in treating hepatitis C^35,39,40 and SARS-CoV-2.⁴¹ The second analogue, CycH, is a CycA epimer with a d-MeVal11. The biochemical properties of CycA and CycH are drastically different. The most notable ones are CycH’s poor membrane permeability and weak affinity to CypA.³⁴ A summary of the macrocycle residue alterations with respect to CycA and their physiochemical properties is provided in Figure 1.

Figure 1.

Chemical structure of CycA. A summary of residue alterations with respect to CycA is provided in the table as well as illustrated in blue for ALI and red for CycH. The seven N-methylated amides are also highlighted in yellow. The cellular permeability and relative binding affinity to CypA are also provided.

This work uses a combination of solution two-dimensional nuclear magnetic resonance (2D-NMR), three ion-mobility platforms; drift-tube, cyclic, and trapped ion mobility spectrometry (DT-IMS, cIMS, and TIMS, respectively), and ion spectroscopy (IR multiple-photon dissociation; IRMPD) coupled with TIMS and Fourier-transform ion cyclotron resonance (FT-ICR). The ultimate goal is to comprehensively capture major and minor cyclosporine conformers in aqueous environments, their interconversion, and their responses to external perturbation (temperature or ionic salts). With 2D-NMR, conformers are classified based on their cis/trans amide configurations and chemical shifts. Direct exchange between conformers with different cis/trans configurations has not been observed in 2D-NMR because the typical activation energy for cis/trans isomerization in linear molecules is approximately 20 kcal/mol,^42,43 which is equivalent to transition times on the order of minutes. IMS offers a longer time scale than NMR, and ion activation on mobility-selected conformers provides unique information about interconversion.⁴⁴ IRMPD provides spectroscopic signatures of the molecule and its conformers in the mass spectrometer. Finally, we will present a molecular dynamics (MD) approach based on a reparameterized force field, specific for cyclosporines, to enhance conformational sampling and improve predictive power.

Notably, pairing multiple complementary techniques in analyzing biomolecules has not been the standard practice for the structural analysis of cyclic peptides in general and cyclosporines in particular. Our work herein describes a thorough sampling of cyclosporine conformers, illuminating the critical roles of “redundant” N-methylation in cyclosporines in their biochemical functions. We also highlight new approaches to predict the membrane permeability of cyclosporine-based analogues.

Materials and Methods

NMR Spectroscopy

Sample preparation, instrumentation, and software details can be found in the Materials and Methods section of the Supporting Information. 2D-NMR experiments are vital for atomically resolving the conformers and obtaining chemical exchange information between states. 2D through-bond NMR experiments zTOCSY (Total Correlation Spectroscopy), ¹³C-gHSQC (Heteronuclear Single Quantum Coherence) with adiabatic pulses, and ¹³C-gHMBC (Heteronuclear Multiple Bond Correlation) with adiabatic pulses are used to identify each amino acid residue and assign the protons and carbons within them. Then, ROESY (Rotating Frame Overhauser Enhancement Spectroscopy) with adiabatic pulses (ROESYAD, also known as easy ROESY) is a through-space experiment that indicates protons that are in close spatial proximity (≤5 Å). This experiment is used to connect the residues in sequence and identify the cis/trans amide connections. The distance restraints from ROESY experiments can be used to construct a structural model in NMR molecular dynamics simulations. Additionally, the ROESY cross peaks (ROEs) in the same phase as the diagonal indicate nuclei in chemical exchange on the NMR time scale (approximately 300 ms).

Ion Mobility Spectrometry-Mass Spectrometry

All sample preparation, instrumentation, and software details can be found in the Materials and Methods section of the Supporting Information. DT-IMS-MS experiments were conducted to evaluate the mass spectra, collisional cross sections (CCS), and arrival time distributions (ATDs) of the conformers. DT-IMS instrument parameters are listed in Table S1.

cIMS-MS experiments were conducted to obtain high-resolution ion mobility chromatographic information. The multipass capabilities of cIMS-MS allow for increased separation of conformers compared to DT-IMS. Conformations can then be mobility-selected and reintroduced into the cyclic ion racetrack for additional passes, which provide further separation and conformational interconversion insights. Instrument parameters are listed in Table S2.

IR Multiple-Photon Dissociation Spectroscopy

Sample preparation, instrumentation, and software details can be found in the Materials and Methods section of the Supporting Information and Table S3. IRMPD spectroscopy is a powerful technique used to probe an ion’s vibrational modes and, by extension, the structure.⁴⁵⁻⁴⁷ We verify the gas-phase conformational structures selected by IMS-MS. The setup is based on Fourier-transform ion cyclotron resonance (FTICR)-MS (Bruker SolariX) with an electrospray ionization (ESI) source and a trapped ion mobility spectrometry (TIMS) stage. In the ICR cell, the m/z- and IMS-selected ions are irradiated with the pulsed, tunable laser light from FELIX so that IRMPD spectra are obtained. The experimental IR spectra are compared with DFT-predicted IR spectra to make conformation assignments.

Computational Modeling

Temperature-based replica exchange molecular dynamics (T-REMD) simulations were performed in an explicit methanol solvent box for CycA, ALI, and CycH. The parameters for the macrocyclic solute molecules were generated from a modified ff14SB AMBER force field.⁴⁸ For charge derivation, we have employed the PyRED server⁴⁹ to perform a restrained electrostatic potential fit for the atomic charge values using four separate conformations of CycA, two of which are from the RCSB Protein Data Bank and the other from our NMR/X-ray/neutron crystallography data.⁵⁰⁻⁵³ Four dipeptide fragments (Abu2—Sar3, Val5—MeLeu6, MeLeu9—MeLeu10, and MeVal11—MeBmt1) were used for charge derivation in addition to three single amino acid fragments (MeLeu4, Ala7, and d-Ala8). The four amino acid pairs in CycA’s X-ray structures show significant experimental cis propensity. Therefore, we accordingly estimated the contribution from torsion of the peptide bond as 0 kcal/mol. In doing so, we assume that the energetic contribution from other sources is more relevant in conformer selection. The simulations are 2.5 μs/replica. Simulation convergence was assessed by block analysis and correlation matrices (Figures S1 and S2). The conformers were further geometry optimized for energy calculations with Density Functional Theory (DFT). Additional details on the REMD simulation parameters and DFT calculations (Figure S3) can be found in the Supporting Information.

Results

NMR Spectroscopy of CycA, ALI, and CycH

An illustration of the NMR-observed cis-amide backbone connections in CycA, ALI, and CycH is shown in Scheme 1. The circles indicate the residues connected by a methylated amide bond and are filled in with black if a cis configuration has been observed via our NMR data. The red and blue dashed lines denote additional structures containing both cis-amides. The conformational heterogeneity of the three cyclosporines is investigated by 2D-NMR spectroscopy (Tables S4 and S15 for ¹H and ¹³C assigned chemical shifts). Cis-amide bonds are followed and assigned in the Hα-Hα region of the ROESY spectra. Conformers are defined as structures with unique cis/trans amide backbones. States are denoted as the conformer letter followed by an integer, with “1” being the most abundant state, e.g., A1 and A2 refer to two states of the “aqueous” conformer, both containing the cis MeVal11—MeBmt1. We recall that “states” are defined as various forms of the same conformer, which have identical cis/trans amide backbone conformations and are assumed to have similar structures. States are often observed in slow exchange on the NMR time scale. This nomenclature was used in our previous work.²⁸

Scheme 1. Peptide Backbone Connection Scheme of CycA, ALI, and CycH Based on NMR Data.

Circles indicate residues that are connected by a methylated amide. Black circles denote an NMR-observed cis-amide. Each color represents a unique conformer. Cis-amides that are connected are found in the same conformer. The all-trans conformers were observed for all three macrocycles. d-Ala8—MeLeu9 and MeLeu10—MeVal11 are always trans. D3 is partial.

All three compounds (CycA, ALI, and CycH) adopt C1. We previously showed that CycA populated A1 and C1 as the major conformers. In addition, there are minor conformers of CycA with cis-amides Abu2—Sar3 and Val5—MeLeu6 that we termed D1–3.²⁸ Here, our NMR data reveal that O1 and A1 are prominent conformers for ALI, while C1 and B1 (a cis Val5-MeLeu6 conformer) are minor conformers (partially assigned). CycH was found to adopt C1 and O1 as major conformers, which is in agreement with CycH’s X-ray crystal structures.³³ CycH also adopts B1 as a major conformer and a partially assigned cis Abu2—Sar3 as a minor conformer. Intriguingly, while both CycA and ALI populate A1, CycH does not (Figure 2A). Of note, we verify that the conformers observed in the NMR experiments have the same cis-amide bonds and the same hydrogen bonds as the X-ray crystal structures (see Tables S16–S21).

Figure 2.

(A) MeVal11(Hα)– MeBmt1(Hα) ROESY cross peaks (ROEs) of A1 and A2, indicating the cis-amides. The ROEs are found for CycA and ALI, but they are absent for CycH. (B) ROESY spectrum showing exchange peaks (in red) between states. (C) The NH-NH region of the ¹H NMR spectrum of CycA, ALI, and CycH is shown in panel C. All spectra were taken at 25 °C. The samples are 5 mM in CD₃OD with 10% H₂O.

Additionally, the chemical exchange between states can be assessed by signals in the HN-HN region of the ROESY spectra with the same phase as that on the diagonal (Figure 2B). Counting the exchange peaks, CycA and ALI exhibit at least 15 exchanges among states, while CycH exhibits at least 21. The CycA and ALI exchange signals are strong, while the CycH spectrum is more congested with poorly defined peaks, resulting in overlapping signals and making atomic assignment difficult (Figure 2C).

NMR Spectroscopy of CycA+Ca²⁺

We and others previously performed 2D-NMR experiments of CycA in the presence of CaCl₂, where we found that the population of O1 significantly increased.⁵⁴ However, previous studies did not thoroughly investigate other changes in solution, especially the abundance of A1 and the formation of any new conformer. Here, we repeat the experiments and carefully monitor the changes in the samples over 2 weeks. An equilibrated ensemble is attained after 2 weeks (see Figure S4). Further investigation of the equilibrated sample reveals two main observations, as elaborated below.

First, the population of C1 is relatively unaffected, but A1 is decreased, and O1 is increased (Figure 3A). This is consistent with our previous report.⁵⁴ Second, the evolution of a new signal becomes apparent at temperatures greater than 25 °C (Figure 3B). Analysis of the TOCSY spectrum identifies this signal as the Hα of a MeLeu6. This signal was traced in the ROESY spectrum to reveal that it belongs to a conformer with a cis Sar3—MeLeu4 and MeLeu9—MeLeu10 backbone configuration (Figure 3C), which we have denoted as I2. We referred to this conformer as I2 because Bernardi and co-workers previously reported another conformer with the same cis-amides.⁵⁵ Henceforth, we designate the conformer identified by Bernardi as I*, the conformer anticipated by T-REMD (without salt) as I1, and the novel conformer observed in the presence of CaCl₂ as I2 (refer to Table S19 for hydrogen bond calculations related to I2).

Figure 3.

(A) Hα-Hα ¹H NMR region of 5 mM CycA samples in CD₃CN with approximately 0.7% H₂O with and without 5 mM calcium chloride at 25 °C. Proton signals of MeBmt1 are annotated for each conformer. (B) Partial ¹H NMR spectra of CycA:CaCl₂ sample at various temperatures. The signal highlighted in light purple is the Hα of MeLeu6 of I2. (B) Hα-Hα ROEs of the 2 cis-amides found of I2. Two MeLeu4-Sar3 ROEs were observed because Sar3 has two Hα’s.

IMS-MS of CycA, ALI, and CycH

The conformational ensembles of the macrocycles are further investigated with DT-IMS (Figure S5) and multipass cIMS-MS. The mass-selected IMS mobiligram of [CycA+H]⁺ in Figure 4 shows four major features (similar to the data of ALI as seen in Figure S6) after 10 passes on the cyclic ion racetrack. CycH, on the other hand, shows only two features after 9 passes.

Figure 4.

cIMS-MS data for singly protonated (A) CycA and (B) CycH. The IM profile of CycA after 10 passes contains four major features, whereas that of CycH after 9 passes contains only two features. Mobility-selecting conformers with either the shortest or the longest drift times (green or purple features of [CycA+H]⁺; red or brown features of [CycH+H]⁺) yields partial distributions of the initial ensembles without the addition of ion activation. With ion activation energy, the two distinct conformers of CycH can be re-established from either conformer, while this is not the case for CycA. (C) cIMS-MS data of the doubly charged calciated adduct of CycA, [CycA+Ca]²⁺. The conformational ensemble without ion mobility selection is shown in panel a. cIMS-MS data on mobility-selected conformers (A1-cyan, C1-orange, O1-red) are shown in panels b and c. Initial distribution is re-established starting from any of the three conformers without ion activation. I1 was tentatively assigned based on the theoretical CCS calculations, which indicates that I1 is slightly larger than C1, and is significantly smaller than O1.

For CycA, mobility-selecting the most compact or extended conformers and reintroducing them into the cyclic path does not re-establish the initial conformational ensemble. Partial re-equilibration is observed at 60 eV (lab frame) when ion activation is used. Nonetheless, even at 80 eV, the initial distributions of four IM features are not established, indicating a high energy barrier between the compact and extended conformers. A similar result is obtained for the ALI (Figure S6).

On the other hand, when each IM feature of [CycH+H]⁺ was separately introduced to the ion racetrack, the original conformers were repopulating after only about four passes without ion activation (Figure 4B). This is evident by the almost immediate overlapping of the mobility-selected conformer baselines. The repopulation of the two distinct conformers can be seen after 8 additional passes. Population distributions do not change much after increasing the energy to 80 eV.

IMS-MS of CycA in the Presence of Ca²⁺

When comparing the DT-IMS-MS data of [CycA+H]⁺ and [CycA+Ca]²⁺ and making assignments based on the relative differences in CCS among the major conformers (A1 < C1 < O1; see Table S23 for experimental and theoretical CCS), the data suggest that the population of O1 increases significantly, while that of A1 decreases in the presence of CaCl₂. With cIMS-MS, we can reproduce this result with a better resolution (Figure 4C; and see Figure S8 for [CycH+Ca]²⁺ and [ALI+Ca]²⁺ data). Lastly, the IM feature assigned to C1 is broad and tailing toward the longer drift time, indicating some minor conformers between C1 and O1. Given this, one could argue that Ca²⁺ binds initially to A1, facilitating A1 to O1 conversion.

However, if that mechanism were correct, we would expect that Ca²⁺ selectively perturbed the transition from A1 to O1 but not from A1 to C1 or from C1 to O1. Ion activation experiments with cIMS-MS indicate that the conformational transitions among all CycA conformers are more fluid in the presence of Ca²⁺. Recall that the same experiments on singly protonated CycA revealed that it was impossible to re-establish the conformational ensemble starting with either the most compact or most extended conformers, even with an 80 eV activation voltage. On the other hand, with [CycA+Ca]²⁺, the introduction of any of the Ca²⁺ bound conformers to the ion racetrack quickly establishes the original ensemble without the addition of activation voltage (Figure 4C, panels b and c).

Computational Modeling of CycA, ALI, and CycH

With the modified force field, our T-REMD data capture all of the experimentally detected CycA, ALI, and CycH conformers reported. Namely, with CycA, these include the X-ray crystal structures of A1,²⁸C1,⁵⁶O1,²⁴ and also RGG (the conformer with three cis-amides that binds to the RGG protein²⁷), the C1(57) and O1(33) of the free CycH, and O1 of ALI,³⁷ see Figure 5.

Figure 5.

Data from ∼2 μs of T-REMD simulations. (A) Backbone RMSD comparison of T-REMD (blue) with X-ray structures (red) for CycA (i-iv are A1, C1, O1, and RGG of CycA, respectively), ALI (O1), and CycH (i and ii are O1 and C1 of CycH, respectively), indicating the accuracy of our T-REMD to sample each experimentally observed conformation. (B) The cis/trans isomer distribution of CycA, ALI, and CycH sampled by the simulation. Convergence of the simulation was assessed by using the relative entropy of blocked conformer distributions.

For CycA, conformational heterogeneity is much higher in the simulation data (no conformer occupies more than 25% of the total population) that those obtained by Ono et al. (in methanol, Ono et al. predicted the all-trans conformer to be ∼50%).⁵⁸ This result also shows a better agreement with our NMR data than that of Ono et al.⁵⁸ For ALI, the only reported X-ray crystal structure is its all-trans open form in the CypA:ALI complex, which the simulations captured. More importantly, the T-REMD results of ALI and CycH are not biased toward the known structures of CycA, for which the X-ray structures were used to parametrize the force field. This indicates that our force field provides a fair assessment of the cyclosporines.

T-REMD data validate our NMR and IMS-MS findings. Specifically, the data highlight that CycH is much less heterogeneous than CycA and ALI. CycH’s O1 is the predominant conformer, whereas in CycA and ALI, the two competing conformers are A1 and C1, or A1 and O1, respectively.

The predictive capabilities inherent in our T-REMD data are strong. Notably, while there is a preference toward conformers with no or one cis-amides, the computational results also indicate the presence of conformers with more than one cis bond. For example, the Pearson correlation matrixes of cis/trans methylated amides (Figure S2) indicate the strong propensity for the coexistence of Abu2—Sar3, Val5—MeLeu6, and MeVal11—MeBmt1 (found in the RGG-bound conformer), or Sar3—MeLeu4 and MeLeu9—MeLeu10. Interestingly, the conformer with cis Sar3—MeLeu4 and MeLeu9—MeLeu10 (i.e., “I1”). Conformers with the same cis-amides have previously been reported in the NMR experiment of CycA in the presence of Mg(ClO₄)₂ by Bernardi and co-workers;⁵⁵ however, there might be variations in backbone structures). Our T-REMD experiments predict that the I conformers are among the relevant conformers of CycA. We will discuss I1 and its variants (I* and I2) in more detail in the next section.

Regarding ALI, T-REMD data reinforce that the chemical modification stabilizes O1 but does not impair A1. The change of Sar3 to d-MeAla3 lowers the cis propensity of the amide between residues 2 and 3, while the change of MeLeu4 to EtVal4 has the same effect on the amide between residues 3 and 4. In parallel, T-REMD predicts that in addition to A1 and O1, ALI should populate a conformer with two cis-amides Val5—MeLeu6 and MeVal11—MeBmt1. This result is reasonable because the chemical modifications at residues 3 and 4 in ALI strategically stabilize the trans configurations at the amides among residues 2, 3, and 4. Thus, this conformer shares a similar cis/trans configuration to CycA’s RGG, except for the lack of cis-amide between Abu2 and Sar3. A cis Val5—MeLeu6 conformer was previously observed in NMR experiments for CycA (denoted as D3 in Limbach et al.²⁸) and presently observed in ALI, but the conformer was partially resolved.

In summary, the T-REMD simulations capture the experimental conformers observed in NMR and X-ray crystallography and predict the existence of conformers with more than one cis bond that may play essential roles in conformational transitions among the major conformers.

Ion Mobility-Selected IR Multiple-Photon Dissociation Spectroscopy

To verify the identity of the gas-phase [CycA+Ca]²⁺ conformers beyond the CCS from the IMS-MS data, IRMPD spectroscopy is employed with and without TIMS selection. The TIMS mobilogram of [CycA+Ca]²⁺ (Figure 6A) is similar to the ATD of the same species obtained using cIMS-MS (Figure 4) and DT-IMS.⁵⁴ A notable difference is the intensity of peak I1, which becomes intense in the TIMS mobilogram.

Figure 6.

(A) TIMS mobiligram of [CycA+Ca]²⁺. Four features are observed and are annotated as A1, C1, I1, and O1. (B) DFT-calculated IR spectra of [CycA+Ca]²⁺’s O1, C1, A1, and I1 with Ca²⁺ obtained from DFT calculations and compared to experimental IRMPD spectra of [CycA+Ca]²⁺ when mobility-selecting the peaks corresponding to A1, C1, I1, and O1 (orange, dark blue, green, purple, respectively) and without TIMS (black). The C–H stretching frequency (2500–3000 cm^–1) and the O–H/N-H stretching frequency range (3000–3700 cm^–1) are shown. A scaling factor of 0.950 was applied to the computed harmonic frequencies to account for the anharmonicity.

DFT-calculated IR spectra of CycA’s O1, C1, A1, and I1 in the presence of Ca²⁺ obtained from DFT calculations are compared with the experimental IRMPD spectrum of [CycA+Ca]²⁺ (Figure 6B). The IRMPD spectrum shows a shouldered band in the C–H stretching region and two distinct bands in the O–H and N–H stretching ranges, which are attributed to MeBmt1 and to nonmethylated amides, respectively.

We also recorded TIMS-mobility selected IRMPD spectra for each peak observed in the TIMS mobilogram.⁵⁹ The IRMPD spectra of peaks assigned to C1 and O1, respectively are almost identical, while that of A1 has a small difference in the band intensities. We posit that when a calciated conformer is mobility-selected in the TIMS, it quickly re-equilibrates into a conformational ensemble, either upon extraction from the TIMS and storage in the ICR cell or upon IR excitation. The rationale is consistent with our cIMS-MS data on mobility-selected ions shown in Figure 4C and discussed above. Note that the observation that there are discernible structures in the cIMS and TIMS data is not in conflict with our statement that re-equilibration may happen upon extraction from the TIMS, residence in the ICR cell, or upon IR multiple-photon excitation (MPE).

Discussion

Conformational Heterogeneity, Especially the Presence of Closed and Aqueous Conformers, Is Critical for Good Membrane Permeability

Conformational heterogeneity is commonly regarded as an unfavorable characteristic of a ligand in protein–ligand interactions. However, it is an inherent outcome of N-methylation in cyclic peptides. The presence of conformational heterogeneity and flexibility contradicts the notion of cyclization, which aims to reduce the conformational space in the molecule. Here, our work helps rationalize the ability of cyclosporines to cross membranes and rearrange to bind protein targets.

All three compounds (CycA, ALI, and CycH) adopt C1. Among them, the hydrogen bonding patterns differ slightly in each macrocycle. According to our MD simulations, ALI cannot form the Abu2-NH···Val5-CO hydrogen bond, possibly because of the backbone strain from the d-amino acid at position 3, with the added methyl group replacing a hydrogen that normally eclipses the adjacent N-methylation. CycH does not form the two hydrogen bonds involving residues 7 and 8, presumably due to epimer-induced backbone strain. Of note, despite these differences, each macrocycle forms the β-turn between residues 3 and 4 and successfully shields their amide hydrogens from solvent. This is clear evidence that a closed conformer alone does not indicate good membrane permeability. This is consistent with early work on cyclosporine O (CycO), which stabilizes the closed form but has poor membrane permeability.²¹

Our previous NMR study showed that O1 is a minor population of CycA that becomes populated in the presence of CaCl₂ (more discussion to follow).⁵⁴ Here, our NMR data reveal that O1 is a prominent conformer in ALI. Since both CycA and ALI are good membrane-permeable peptides,^35,37,41 we conclude that the abundance of O1 (or generally, a conformer with all-trans amides) does not predict membrane permeability. Regarding O1, while this conformer of CycA and ALI are very similar (as both molecules bind tightly to CypA^24,25,37), that of CycH (which does not bind strongly to CypA³⁴) is different. The characteristic transannular hydrogen bond of O1 in CycA is that between the side chain hydroxyl of MeBmt1 and MeLeu4’s carbonyl oxygen. According to X-ray structures and DFT and MD calculations, this hydrogen bond persists in ALI despite the substitution on residue 4. However, CycH’s epimerization introduces enough ring strain to entirely prohibit this hydrogen bond, preferring a hydrogen bond between MeBmt1’s side chain oxygen and Abu2’s carbonyl oxygen instead.

Lastly, both CycA and ALI, the two macrocycles with good membrane permeability, populate A1, while CycH, which has poor membrane permeability, does not (Figure 2A). The l to d isomerization of MeVal11 disrupts the ability of the MeVal11—MeBmt1 amide bond to adopt the cis configuration. The A1 conformer in CycA was previously probed via NMR and crystallography techniques, and its relevance to CycA biochemistry was discussed in detail.²⁸ Cyclosporines that eliminate cis MeVal11—MeBmt1 via demethylation of MeVal11, such as cyclosporine E (CycE)²⁰ and some analogues of CycO,⁶⁰ also have poor membrane permeability. Collectively, the presence of A1 in membrane-permeable CycA, ALI, and CycG (see ref.³³) suggests that this conformer is critical for the macrocycles to diffuse across the lipid bilayer passively.

Beyond individual conformers, we envisage, based on the NMR data alone, that the number of exchanges among states is inversely proportional to conformational heterogeneity. CycA and ALI’s NMR spectra show fewer exchanges but strong peaks, indicating fewer exchanges among states with the same cis–trans configurations. On the other hand, CycH’s spectra contain more exchanges but weaker peaks (Figure 2B, C), indicating more exchanges among states with the same cis–trans amide backbone. Collectively, we conclude that CycA and ALI populate more conformers with unique cis–trans amides in comparison to CycH, which could be important for membrane permeation. More importantly, the results suggest that exchanges among states captured by 2D-NMR can infer the conformational heterogeneity of cyclosporines. To illustrate the assumption that a cyclosporine characterized by low conformational heterogeneity (few conformers with distinct cis–trans amides) can exhibit numerous states (with the same cis–trans amides but differing backbone shapes or side chain orientations) and transitions between these states (as observed in the NMR data, Figure 2C), we calculated the RMSDs for the C1 and O1 conformers of CycA, ALI, and CycH. We chose these two conformers because all three macrocycles populate them. The RMSDs to the X-ray crystal structure for CycH’s C1 and O1 display higher values and significantly broader ranges than those for CycA and ALI (Figure S5). This highlights that the conformers of CycA and ALI are more stable than that of CycH, exhibiting fewer variations in backbone structures and side chain orientations.

The IMS-MS data also support the conclusion from NMR (Figure 4). Of note, CycA and CycH were previously studied by Hyung et al. using traveling-wave IMS (TW-IMS), in which the ATD of CycA shows two features while that of CycH only shows one.⁶¹ Our DT-IMS-MS data also support that CycH adopts fewer conformers than CycA, as the width of its IM profile is more narrow (Figure S6). Here, using cIMS-MS, which offers higher IM resolution than DT-IMS and TW-IMS, the difference in conformational heterogeneity of CycA and CycH is validated. The results from conformationally selected ion activation experiments validate that the CycH conformers are much less stable than CycA and ALI conformers.

Another line of compelling evidence is from our computational modeling data, where O1 was found to be dominant in the conformation space of CycH (Figure 5). CycA and ALI, which have good membrane permeabilities, populated A1 and C1. Also, from the computational data, the population of A1 is significant for both CycA and ALI, indicating that it is crucial for membrane permeability. Intermediate conformers, some serendipitously captured in previous NMR experiments with salts,^55,62 are also detected in our simulations and experiments, as elaborated in the previous sections. While it is not yet possible to elucidate the specific role of these “rare” conformers, we particularly focused on the I conformer, which has two cis-amides Sar3—MeLeu4 and MeLeu9—MeLeu10. This conformer becomes visible in the presence of CaCl₂.

Calcium Chloride Promotes Low-Abundance Conformers

Because metal cations such as Na⁺, K⁺, Ca²⁺, Mg²⁺, or Zn²⁺ are found in biology at relatively high concentrations, they are essential and may contribute to the biochemistry of cyclosporines in vivo. However, their possible roles in the conformational isomerism of cyclosporines have not been fully investigated. The use of salts in NMR and X-ray crystallography experiments is precedented. For example, Mg(ClO₄)₂ was used to obtain X-ray crystal structures of free CycH and CycG.³³ KCl and NaCl can be used to crystallize the aqueous conformer (A1) of CycA.²⁸ In these crystallography experiments, no metal cations or counterions were found in the refined crystal structures, indicating that they do not bind strongly to the macrocycles and alter their structures. Their role was hypothesized to be involved in stabilizing flexible side chains such as MeBmt1.³³

In parallel, a more diverse set of salts (and counterions) were used in early NMR experiments where uncommon cis-amide bonds of CycA were observed.^{43,54,55,62,63} We previously showed that CaCl₂ greatly increased the population of CycA’s O1 in acetonitrile (that is otherwise a minor conformer), and Ca²⁺ was predicted to bind to CycA near d-Ala8 of O1.^54,62

Our NMR and cIMS-MS data on CycA in the presence of CaCl₂ show the presence of the I conformer. Its presence is positively correlated with the increase in O1 and the decrease in A1, suggesting that they could be important “catalytically” for the transition from A1 (the aqueous conformer) to O1 (the conformer required for binding the protein targets). However, as the mobility-selected cIMS-MS experiments (Figure 4C) indicate, CaCl₂ lowers the energy barriers for interconversion between A1, O1, and C1.

IRMPD spectroscopy of [CycA+Ca]²⁺ shows that the presence of some CycA conformers is validated through the bands in the O–H/N-H stretching frequency region (Figure 6). The O–H/N-H stretching region varies slightly between conformers, as indicated by the DFT-calculated IR spectra in Figure 6B. Comparing this region of the calculated data to the experimental IRMPD data, it can be deduced that the [CycA+Ca]²⁺ ion consists of O1, as indicated by a strong band at 3450 cm^–1, and possibly a mixture of A1, C1, and I1 as indicated by a medium band at ∼3350 cm^–1 and a very weak band near 3700 cm^–1 (which is not predicted for O1). Notably, the IRMPD spectrum of peak A1 (Figure 6B) closely resembles the calculated IR spectrum of A1. We hypothesize that some collisional activation occurs upon exiting the TIMS (i.e., MPE) and note the relatively long storage time in the ICR cell; the additional energy or prolonged storage time would allow for the conformational interconversion of the mobility-selected conformers to the original (no TIMS) conformational distribution. This is reminiscent of the re-established conformational ensembles observed in the cIMS data after the addition of activation voltage (Figure 4A). Nonetheless, the IRMPD data support the idea that O1 is promoted by CaCl₂.

Here, we offer an explanation of the role of CaCl₂: The presence of salt can alter the dielectric properties of the solvent, causing a disturbance in the conformational ensemble. In simpler terms, CaCl₂ does not necessarily create a new conformation, but it renders otherwise less prevalent conformations detectable. Additionally, each salt has the potential to influence the peptide structure, introducing novel states.

It has been shown theoretically and experimentally that the addition of alkali and alkaline-earth halides to high dielectric solutions results in a nonlinear decrease in the dielectric constant due to ion-solvation effects.⁶⁴⁻⁶⁸ Barthel and co-workers demonstrated the effects of various electrolytes on the dielectric constant and relaxation of acetonitrile by analyzing the complex permittivity spectra in the frequency range 0.95 ≤ ν (GHz) ≤ 89 at 25 °C.⁶⁹ The dielectric constants of the acetonitrile:electrolyte solutions, up to molarities of 1 M, were found to decrease regardless of the electrolyte identity.

Previous NMR data indicate that the closed conformer (C1) exists at 55% in deuterated acetonitrile compared to 19% in deuterated methanol/water.⁵⁴C1 is the most dominant conformer in a low-dielectric environment, while O1 favors a high-dielectric, aqueous environment. Perhaps the presence of CaCl₂ at 5 mM lowers the dielectric constant of the acetonitrile (pure acetonitrile; ε_s = 35.84⁶⁹) to slightly mimic the properties of methanol (pure methanol; ε_s = 33.70⁷⁰) enough to alter the stability of A1 and O1, but not C1. Note that LiClO₄ has been shown to increase the originally low dielectric constant of THF.⁷¹ This further highlights the effects that metal cations have on the dielectric constant of the solvent, which is important for the solubility (stability) of the CycA conformers. What conformers would become visible may depend on the ability of the salt to alter solvent properties.

Of note, Köck et al. performed NMR on CycA in the presence of LiCl in deuterated tetrahydrofuran (THF).⁶³ They showed that CycA adopts O1, compared to only C1 in the absence of LiCl. This means that the Li⁺ and the Ca²⁺ ions have similar effects on the cis/trans amide preference of CycA. Gray et al.⁵⁴ compared the NMR chemical shifts of the all-trans conformers (O) observed in samples containing CaCl₂, LiCl, or CypA.⁶³ Variations in chemical shifts were noted, indicating that the salts may slightly modulate the structures of the conformer. An average difference of 0.52 ppm was observed for the backbone ¹H chemical shifts (Hα’s and HN/MeN’s).

This rationale could also provide support for earlier observations of atypical cis-amides when salts are present. For example, cis Sar3—MeLeu4 was observed in the presence of CaCl₂ (this work), Mg(ClO₄)₂,⁵⁵ and Ca(ClO₄)₂.⁶²Cis Val5—MeLeu6 was observed in pure MeOD/H₂O,²⁸ or in the presence of Ce(ClO₄)₃.⁵⁵Cis Abu2—Sar3 was also observed in pure MeOD/H₂O,²⁸ or in bound state of CycA to RGG.²⁷

Furthermore, employing the cIMS platform, we conducted tandem MS/MS on mobility-selected [CycA+Ca]²⁺ ions (Figure S8). Regrettably, the MS/MS spectra exhibited considerable similarity. This result can likely be attributed to the selected conformers undergoing equilibration into an ensemble during the collision-induced dissociation process.

Nonetheless, to further support our rationale that different salts may produce different states of the same conformer, we discuss the various states of I under different solvent conditions. With CaCl₂, we observed I2, which was visible at temperatures greater than 25 °C (Figure 3). I2 has distinct chemical shifts from C1, and thus its chemical shifts can be fully assigned (Table S4). I2 is also characterized by two hydrogen bonds between Abu2 and Val5 and between Ala7 and MeVal11.

Regarding I* observed in the NMR studies by Bernardi et al., who worked with Mg(ClO₄)₂,⁵⁵ that study reported very limited chemical shifts (mainly some Hαs). Therefore, we conducted NMR experiments on CycA with Mg(ClO₄)₂ to elucidate I*. We found that the NMR chemical shifts of I* are different compared to the chemical shifts of I2; see Table S4. On average, a difference of 0.42 ppm was observed for the backbone ¹H chemical shifts (Hα’s and HN/MeN’s).

Regarding the atomistic structures, the structure obtained through the Mg²⁺-restrained MD (I*) by Bernardi coordinates Mg²⁺ inside the ring.⁵⁵ Our I1 structure predicted by T-REMD (no salt) adopts a configuration comparable to C1,⁵⁶ with a small change in secondary structure at the novel cis-amide linkage. Thus, we expect the NMR chemical shifts of I1 to largely overlap those of C1. In C1, the geometry around the Sar3-MeLeu4 peptide linkage adopts a standard type-II’ β turn, characterized by an Abu2-NH···Val5-CO hydrogen bond. The predicted I1 structure utilizes the same hydrogen bond to adopt instead a Type-VIb β turn (φ–ψ angles given in Table S23), usually reserved for cis-proline bonds.⁷² Comparing C1 and I1, Abu2, Val5, and MeLeu6 differ only by rotamers likely to be in fast exchange at room temperature (Figure 7A, B). The backbone structure is identical aside from the β-turn region, and consequently, the hydrogen bonding of amide hydrogens distal to Sar3 and MeLeu4 is also identical to C1. The cis-amide linkage causes a deviation from the original Type II′ secondary structure characteristic where the carbonyl appears on the other side of the plane of the macrocycle, giving a mirrored appearance relative to the C1 closed form (Figure 7C, D). This finding is further validated by comparing I* and C1¹H NMR chemical shifts. The chemical shifts of I* and C1 are almost identical. Therefore, based on the consistent hydrogen bonding, cis-amide patterns, and chemical shift comparisons, we conclude that the I* conformer detected in the NMR experiments with Mg(ClO₄)₂ has the same structure as the one captured by the T-REMD simulation (i.e., I* = I1).

Figure 7.

Comparison of backbone structures of (A) C1 by Loosli et al.²⁹ and (B) I1 obtained from T-REMD and optimized with B3LYP/def2svp, the D3 version of Grimme’s empirical dispersion model, and using a self-consistent reaction field methanol solvent model. Side chain carbons are colored black to emphasize slight rotamer variation. The difference in the secondary structure around the Sar3—MeLeu4 peptide linkage is emphasized by both (C) front and (D) back views.

From T-REMD data and energy calculations, I1 is one of the most favored structures of CycA, as it has the second lowest relative energy among all of the conformers of CycA, slightly less stable than C1 but more stable than both A1 and O1. We suspect that I1 is always present at very low concentrations, but its detection by NMR requires higher temperatures in combination with the stability effects of the metal salts. I2 is a result of I1 being perturbed by CaCl₂, whereas Mg(ClO₄)₂ promoted and also retained I1. These salts can interact weakly with the macrocycle, change the ion–solvent pairing, and increase the dielectric constant of the environment. It would be valuable for future research to explore the effects of the counterions (the Hofmeister series) on the population of cyclosporine conformers.

Moreover, in the cIMS-MS data (Figure 4C), I2 likely contributes to the broad tailing feature assigned to C1. Additionally, due to its β-turn structure, I1/I2 are likely to be more abundant in low dielectric environments, suggesting its potential contribution to membrane permeation. This speculation finds support in the detection of cis Sar3—MeLeu4 in [CS–NH^MeBmt]CycA, which is a thiocyclosporin derivative of CycA where the C=O of MeBmt is replaced with C=S.⁷³

Nonetheless, despite its higher energetic stability, the lower abundance of I1 compared to that of A1 and O1 presents intriguing questions. The energy difference of 1.4 kcal/mol corresponds to a ratio of approximately 10:1. Considering that I1 is approximately 5 kcal/mol higher in energy than C1 (see Figure S3), it suggests that the abundance of I1 (or A1 and O1) should be negligible in comparison to that of C1. The role of I1/I2 and whether the high cis propensity of Sar3 and MeLeu4 impacts the activity of CycA remains unclear. In membrane-permeable ALI, I1 is reduced and replaced by a conformer with cis Val5—MeLeu6 and MeVal11—MeBmt1. In future research, it would be valuable to investigate the effects of enforcing trans configurations for Sar3—MeLeu4 and Val5—MeLeu6 to determine whether the activity of cyclosporines is compromised.

Then why are O1 and A1 observed? The all-transO1 conformer represents a transitional state between conformers with distinct cis-amides. In the presence of metal salts, the conformational population distribution can be altered, changing the relative abundance of the conformers that are already in solution, such as leading to a higher occurrence of O1 despite its higher energy level. Lastly, A1 demonstrates the capability to engage in hydrogen bonding with water,²⁸ which significantly contributes to its stability and may account for its higher abundance compared to I1.

Conclusion

Our study reveals that access to the A1 conformer, not the O1 conformer, facilitates membrane permeation of cyclosporines. When the C1 conformer is excessively stable, it may become trapped or immobilized within the membrane, hindering its functionality. While C1 is more populated in low-dielectric environments (such as a membrane), it is not enough to make the macrocycle membrane-permeable. Membrane permeability is likely driven by the transitions between C1 and A1. Further studies will investigate into the intermediates between the two conformers, especially at the interface between aqueous and membrane. Moreover, O1 is populated only in the presence of CaCl₂.

Metal salts such as CaCl₂ perturb conformational population distributions, leading to increased visibility of low-abundance, high-energy conformers. We suggest that these conformers are responsive to the dielectric properties of the environment, potentially influencing their solubility (or stability). Our investigation uncovered up to four major conformers of CycA in solution, highlighting the intricate interplay between conformers with known and unknown functions.

Significantly, the impact of metal cations primarily involves modification of solvent dielectric properties and conformer stability rather than specific chelation to particular residues. Although these nuanced effects have been discerned through crystallography, their thorough investigation necessitates the integration of various techniques, such as gas-phase ion mobility spectrometry, spectroscopy, solution-phase NMR, and computational modeling, as exemplified in the present study.

Moreover, this work addresses the incomplete passive membrane permeability hypothesis, which previously focused solely on the C1 and O1 conformers.^19,74,75 It also helps reconcile conflicting data regarding the structure–function relationship of cyclosporines, putting more spotlight on the aqueous conformers of cyclosporines. The positive correlation between the presence of “rare” conformers with unique cis-amides, such as I1 and I2, and the transition from A1 to O1 may indicate the roles of “redundant” N-methylation found in cyclosporines. The access to their high-energy conformers increases the chance for the macrocycles to adapt to the untraditional protein binding sites. It also allows for a more fluid transition among chemical environments, such as membranes and aqueous. Our study provides the fundamental principles that will aid the design of future large cyclic peptides for intracellular targets.

Finally, through the concurrent application of solution-phase (NMR) and gas-phase (IMS-MS and IRMPD) analyses on cyclosporines, we demonstrated that gas-phase methods are effective at preserving the conformers observed in the solution phase and exhibit good predictive power in identifying novel conformers. This comprehensive approach allows us to capture not only the lowest-energy conformers (e.g., C1 of CycA) but also higher-energy conformers (A1 and O1) and minor conformers (I2) that bear relevance to the biochemical properties of the macrocycle. Our investigation also underscored challenges in acquiring structural data of macrocyclic peptide conformers, considering that energy barriers for interconversion can often be modulated through ion activation, trapping, and laser irradiation. Additionally, we suggest that softer methods, such as cryogenic MS, may be able to address these challenges.

Supporting Information Available

The Supporting Information is available free of charge at https://pubs.acs.org/doi/10.1021/jacsau.4c00011.

• Detailed experimental methods and instrument parameters (Tables S1–S3), NMR chemical shift assignments (Tables S4–S15), NMR hydrogen bond calculations (Tables S16–S21), AMBER parameters (Table 22), theoretical CCS values (Table S23), angles of β-turns in CycA conformers (Table S24), block analysis and convergence test for REMD simulations (Figure S1), correlation matrices obtained from REMD data (Figure S2), DFT data (Figure S3), and additional IMS-MS, REMD, and NMR data (Figures S4–S9) (PDF)

Author Contributions

T.D.D. conceived and managed the project. M.N.L. collected the DT-IMS and 2D-NMR data. E.T.L. performed and analyzed REMD simulations. L.v.T., J.M., and J.O. performed the IRMPD experiments. H.J.O. collected the cIMS-MS data. T.D.D., M.L.N., E.T.L., C.A.S., V.A.N., and J.O. designed the experiments. The manuscript was written through the contributions of all authors. All authors have given approval to the final version of the manuscript. CRediT:Miranda N. Limbach formal analysis, investigation, methodology, writing-original draft, writing-review & editing; Edward Lindberg formal analysis, investigation, writing-original draft, writing-review & editing; Hernando J. Olivos data curation, investigation, methodology; Lara van Tetering formal analysis, investigation, writing-review & editing; Carlos A. Steren methodology, writing-review & editing; Jonathan Martens formal analysis, investigation; Van A. Ngo methodology, resources, writing-review & editing; Jos Oomens data curation, formal analysis, investigation, methodology, project administration, resources, supervision, writing-review & editing; Thanh D Do conceptualization, formal analysis, funding acquisition, investigation, methodology, project administration, resources, supervision, writing-original draft, writing-review & editing.

Eli Lily and Co. Young Investigator Award (to T.D.D.).

The authors declare no competing financial interest.