Collision induced unfolding and molecular dynamics simulations of norovirus capsid dimers reveal strain-specific stability profiles

Abstract

Collision induced unfolding (CIU) is a method used with ion mobility mass spectrometry to examine protein structures and their stability. Such experiments yield information about higher order protein structures, yet are unable to provide details about the underlying processes. That information can however be provided using molecular dynamics simulations. Here, we investigate the gas-phase unfolding of norovirus capsid dimers from the Norwalk and Kawasaki strains by employing molecular dynamics simulations over a range of temperatures, representing different levels of activation, together with CIU experiments. The dimers have highly similar structures, but their CIU reveals different stability that can be explained by the different dynamics that arises in response to the activation seen in the simulations, including a part of the sequence with previously observed strain-specific dynamics in solution. Our findings show how similar protein variants can be examined using mass spectrometric techniques in conjunction with atomistic molecular dynamics simulations to reveal differences in stability as well as differences in how and where unfolding takes place upon activation.

---

Introduction

In nature, the majority of proteins exist not as monomeric species but self-assemble into multimeric complexes,¹ either as a result of selection pressure associated with protein functions, physical properties such as stability, or as a consequence of non-adaptive evolutionary processes, where interfaces can become entrenched without apparent functional advantages.² Virus capsids are an example of the former, where rapid self-assembly of monomers results in an encapsulating structure that holds both the genome and essential proteins, and also protects them until their necessary release. This makes virus capsids excellent models for research on protein assembly and stability.³ Noroviruses are the leading cause of viral gastroenteritis,^4,5 an acute inflammation of the gastrointestinal tract that can cause severe complications particularly for infants, the elderly or people with compromised immune systems.⁶ The norovirus capsid has been subject to several studies on the aforementioned assembly and stability,^7–10 as they are an essential part of understanding viral mechanisms of infection.¹¹ This is especially true for a foodborne pathogen that has demonstrated extraordinary environmental stability.^12,13

The viral genome of noroviruses encodes for two structural proteins, the major viral protein 1 (VP1) and minor VP2 capsid protein.^12,14,15 The icosahedral capsid, enclosing the single-stranded RNA genome, consists of 90 dimers of VP1, and the exact VP1 sequence is linked to the virus's infection rate.^16,17 For that reason, several genogroups and genotypes were categorised based on the VP1 sequence, including the prototypic GI.1 Norwalk strain, and novel emerging strains, such as the GII.17 Kawasaki strain responsible for recent outbreaks in Asia.^16,18,19 The general structure of VP1 is subdivided into two domains; the shell (S) domain, facing the inside of the capsid, and the protruding (P) domain, which is outward-facing and further divided into two subdomains, P1 and P2.^12,14,15 A comparison of the VP1 dimer of Norwalk and Kawasaki is shown in Fig. 1.

Fig. 1 Structural comparison of the norovirus dimers. The structures of the Norwalk and Kawasaki VP1 dimers look similar at first glance, but have specific differences when the structures are overlaid. The structure of Norwalk was obtained from the protein data bank (PDB) entry 1IHM,²⁰ and Kawasaki model structures were generated using AlphaFold.²¹ Norwalk consists of 530 amino acid residues per protein chain, whereas Kawasaki comprises a total of 540 residues per chain. The sequence identity between the two VP1 is 46.60% as calculated by using the USCF Chimera tool.²²

Native mass spectrometry (MS), which preserves non-covalent interactions between biomolecules during the experiment, is a powerful method for interrogating protein complexes in general, but virus capsid assembly in particular due to its ability to separate a large range of coexisting proteoforms and assembly intermediates and quantify each one.²³ Native MS has been applied to study the stability of norovirus structures and intermediates during its assembly, revealing structural information about the viral capsid and its proteins.^7,8 Using ion mobility (IM) spectrometry, where ions interact with an inert buffer gas and are separated as a consequence of their collisions with the gas, proteins with identical or similar mass-to-charge (m/z) ratio, including different conformational states of the same protein, can often be quantified separately, and the different drift times can be converted to collision cross sections (CCSs), which are effective sizes of the ions.^24–26 In fact, a CCS depends on the 3D structure of an ion and can therefore be used for modelling purposes or to test structural hypotheses.^27–31 By combining these two methods, valuable insights of the intensities for different m/z ratios from MS experiments are complemented with CCSs of the proteins in the gas phase. In collision induced unfolding (CIU), the velocity of the ions are increased prior to the IM by applying an acceleration voltage and making them collide with a gas inside the instrument results in an increased internal energy in the ions, which if high enough, causes the proteins to unfold.³² This is manifested as changes of the CCS at specific voltages, which reveal the transition voltages for the different unfolding intermediates, and provide information about a protein's conformation and stability.^33,34 CIU can be used as a fingerprint technique to detect differences between protein variants, but also quantitatively to assess the stabilising effects of different ligands.^35,36

Despite their many advantages, and despite potentially high mass- or CCS resolution, CIU and other IM-MS based methods cannot on their own readily give high-resolution information of the spatial structure of proteins, simply because MS data is not closely connected to the exact positions of atoms in a molecule, and a specific CCS is not uniquely given by a single structure. A computational approach to investigate and provide detailed information about the dynamics of gas-phase biomolecules is given by molecular dynamics (MD) simulations. MD allows for the observation of the dynamics of proteins on an atomistic level, thus making it an ideal tool to complement native MS with high-resolution structural information.³⁷ Moreover, MD can be employed to simulate the unfolding of a protein. During a CIU experiment, the unfolding of a protein is triggered by the applied collision energy.³³ A common approach to simulate the CIU of a protein using MD is to run sets of simulations over a range of different temperatures, where higher temperatures are used to mimic the activation in the experiments.^34,38–40

Here we present CIU experiments complemented with MD simulations using temperature to emulate activation, where VP1 dimer structures of Norwalk and Kawasaki were unfolded in the gas phase in order to elucidate similarities and dissimilarities of the two norovirus strains, and to what extent differences can be probed with CIU experiments. The data was analysed not only to understand at what levels of activation unfolding takes place, but also to see how we can use MD to unravel the structural changes that take place. We find that the dimers of the two strains display different resilience to the activation and also that they differ in how their structures actually break.

Methods

Protein mass spectrometry analysis

Noro VLP production. To produce noroVLPs (norovirus virus-like particles), full-length VP1 genes were cloned and expressed in insect cells using a baculovirus system as described previously.^41,42 GenBank accession numbers are AY602016.1, AF504671.2, and LC037415.1 for variant GI.1 Norwalk, GII.10 Vietnam 026, and GII.17 Kawasaki 308, respectively. In brief, recombinant VP1 bacmid was transfected into Sf9 insect cells and incubated for 5–7 days at 27 °C. After incubation, the culture medium was collected and centrifuged for 10 min at 3000 rpm at 4 °C. Then, the supernatant containing recovered baculovirus was used to infect Hi5 insect cells and incubated for 5 days at 27 °C. After this incubation, secreted noroVLPs were acquired from the culture medium by first centrifuging for 10 min at 3000 rpm at 4 °C and then 1 hour at 6500 rpm at 4 °C. Subsequently, the obtained supernatant was concentrated by ultracentrifugation (Ti45 rotor, Beckman) at 35 000 rpm for 2 hours at 4 °C to get concentrated noroVLPs. Next, the concentrated noroVLPs were purified using CsCl equilibrium gradient ultracentrifugation at 35 000 rpm (SW56 rotor, Beckman) for 18 hours at 4 °C. Finally, noroVLPs were pelleted for 2 hours at 40 000 rpm (TLA55 rotor, Beckman) at 4 °C and dissolved in PBS buffer at pH 7.4.

Sample preparation for MS. NoroVLP samples were exchanged to 50 mM ammonium acetate (99 percent purity, Sigma-Aldrich) at pH 10 using Vivaspin500 centrifugal concentrators (10 000 MWCO, Sartorius, Göttingen, Germany). These buffer conditions were chosen to dissociate dimers reliably from the intact capsids. Calibrants for CCS calculations from traveling wave ion mobility mass spectrometry (TWIMS) were alcohol dehydrogenase (ADH) from yeast (Sigma Life Sciences), pyruvate kinase (PK) from rabbit muscle (Sigma Life Sciences) concanavalin A (Conc A) from jack bean (Sigma Life Sciences) and bovine serum albumin (BSA, Carl Roth). These were exchanged using Micro Bio-spin 6 Columns (6000 MWCO; Bio-Rad) at 1000 × g and 4 °C into 100 mM AmAc pH 7.4. Measurements were performed at approximately 14–26 μM VP1 and 20 μM calibrants. Native MS and travelling wave IMS (TWIMS) measurements were performed on a G1 Synapt HDMS (Waters, Manchester, UK and MS Vision, Almere, the Netherlands). Capillaries were made by pulling borosilicate glass tubes (inner diameter 0.68 mm, outer diameter 1.2 mm with filament, World Precision Instruments, Sarasota, FL, USA) using a two-step program in a micropipette puller (Sutter Instruments, Novato, CA, USA) with a squared box filament (2.5 × 2.5 mm). Subsequently, capillaries were gold-coated using a sputter coater (Safematic, Zizers, Switzerland, process pressure 5 × 10⁻² mbar, process current 30.0 mA, coating time 100 s, 3 runs to vacuum limit 3 × 10⁻² mbar argon).

Ion mobility mass spectrometry. Ions were introduced into the vacuum of the G1 Synapt HDMS mass spectrometer at a source pressure of 10 mbar using a nanoESI source and positive ion mode. Voltages of 1.3 kV and 100 V were typically applied to the capillary and cone, respectively. Argon was used as collision gas and nitrogen as drift gas. The trap region was kept at approximately 1.8⁻² mbar and the ion mobility cell at 0.375 mbar. Transfer voltage was kept at 50 V. A wave height of 15 V and wave velocity of 400 ms⁻¹ was chosen for the ion mobility separation. Wave height for the source and trapping region was 0.2 V and velocity 300 ms⁻¹. For the transfer region this was 3 V and 248 ms⁻¹, respectively. Trap region bias was set to 10 V and the pusher interval was 128 μs with a width of 5. Collision voltage, ranging from 5 to 100 V, was constant for each separate measurement, which consisted of 50–100 scans. Calibrants were measured with 20 V collisional energy and for 50 scans as well. A caesium iodide spectrum was recorded the same day for instrument calibration.

Analysis of ion mobility data. CIU plots were created using CIUSuite 2,⁴³ unidec,⁴⁴ MassLynx V4.1 SCN 566 and Driftscope V2.8 (Waters Coorp., Wilmslow, UK). MassLynx was used to analyse native mass spectra and determine suitable m/z for analysis for each charge state. Triplicates of ion mobility data sets were averaged and cropped using the respective functionalities in CIUSuite 2. For CCS determination, calibrant drift times were extracted with MassLynx, used to create a linear fit and subsequently applied to calculate CCS.^45,46 All reported CCSs should be viewed as TWCCS_N2, according to recommended nomenclature.⁴⁷ CCS values were then transposed onto the mobility raw.csv files to create CIU fingerprints. Feature finding was applied with a feature width of 100 and gap length of 25 steps. CIU50 were then calculated accordingly.

Molecular dynamics simulations

Preparing virus structures. Structures for the Norwalk VP1 dimers were obtained from the PDB entry 1IHM,²⁰ and missing atoms were added using the UCSF Chimera tool (version 1.14).²² Since the dimer structure for Kawasaki was not available in the PDB, we employed AlphaFold in order to predict it using GenBank entry LC037415 as the input sequence.^21,48

Solution simulations. In order to acquire stable and equilibrated protein structures for the CIU simulations, the dimers were initially simulated in solution. The simulations were performed on the Rackham cluster of the Uppsala Multidisciplinary Center for Advanced Computational Science (UPPMAX) using the Gromacs software package.⁴⁹ Gromacs version 4.5.7 has well-optimised kernels for simulating without cut-offs, which is suitable for gas-phase simulations where the absence of solvent greatly increases the effective range of electrostatics. We used this version throughout the investigation for reasons of consistency. As the choice of the force field has been shown to have only a minimal influence on the stability of the proteins,⁵⁰ we employed the OPLS-AA force field⁵¹ and virtual sites⁵² as presented in our earlier studies.^50,53–57 Initially, the structures were placed in a simulation box of dodecahedral geometry with periodic boundary conditions (PBC). The structures were solvated using TIP4P water,⁵⁸ in line with earlier similar studies,^50,53–56 and the saline concentration was adjusted to a physiological level of 154 mM, emulating in vivo conditions and maintaining an ionic strength that is approximately comparable to the ammonium acetate buffer in the experiments. The net charge of the dimers in solution was set corresponding to a pH of 7. Afterwards, the energy of both systems were minimised using the steepest descent algorithm, and a short MD simulation of 50 ps with applied position restraints was conducted for each of them. The temperature of the systems was adjusted to 300 K over a 5-ns long simulation, using the velocity rescaling thermostat⁵⁹ with a coupling constant of τ equal to 0.2 ps and the LINCS constraint algorithm⁶⁰ to constrain all bonds. The Berendsen barostat was used to modulate the pressure of each system over a subsequent 5 ns simulation with a coupling constant of τ equal to 0.1 ps.⁶¹ The systems were allowed to relax in solution and simulated over 100 ns at a time step of 2.5 fs in an isobaric, isothermal ensemble, where the electrostatic interactions were captured with the particle mesh Ewald algorithm⁶² with a real-space cut-off of 1 nm. The relax simulation provided solution-equilibrated starting structures for subsequent simulations. A total of seven structures was extracted, from which another set of 100 ns of production simulations with a 4 fs time step were started, whilst the remaining parameters were kept the same as for the relax simulation. The final structures of the production simulations were then taken as starting structures for the gas-phase simulations.

Simulated collision induced unfolding. Before any in vacuo computation, the net charge of the proteins was adjusted to +22e in order to reflect the protonation the dimers acquire during electrospraying in MS experiments.⁸ The protocol for setting the protein protonation is the same as we present in our recent research, based on the gas-phase basicity of the residues and their availability to the solvent, which in the experiment evaporates and leaves excess charge on the protein.⁵⁵

The structures extracted from the solution simulations were not placed under PBC, as to mimic perfect vacuum conditions, and the steepest descent algorithm was used to minimise the energy of the system in vacuo. In order to emulate activation and potential unfolding of the protein structures, we simulated each of the seven replicas at multiple temperatures: {300, 400, …, 900} K. The specific temperatures were adjusted with the Berendsen thermostat over a short 10 ps MD simulation with a time step of 0.5 fs and a coupling constant of τ equal to 0.1 ps, where all bonds were constrained with the LINCS algorithm.⁶⁰ Eventually, 30 ns of production simulations with a 2 fs time step were started to capture the dynamics of the virus dimers at the specified temperatures.

Data analysis. Investigating the potential unfolding of both VP1 dimers, the accumulated MD data was analysed first by computing the theoretical CCS. We used the ion mobility projection approximation calculation tool (IMPACT) software to compute the theoretical CCS from the MD simulations.²⁵ To better understand the unfolding, each trajectory was split into six time slots, each accounting for 5 ns of the simulation length. For each of those time slots, a single representative structure was selected by clustering all structures in the time slot. We decided to use clustering as it allows to find a single representative structure for each time slot that should reflect potential states of the unfolding process, rather than finding the representative structures based on averaging the coordinates of all structures in a time slot, which can give rise to physically unrealistic conformations. The time-evolution of the structures was quantified by calculating of the root mean square deviation (RMSD). This was done for each representative structure using the corresponding structures from time slot 1 as reference. Per-residue root mean square fluctuations (RMSF) were then calculated to track the mobility of the structures in each time slot. Unfolding is directly linked to a change of the intramolecular distances, and therefore also of the contacts, between residues. We thus employed the MDAnalysis python package in order to calculate the distances between each pair of residues during the MD simulations per time slot.⁶³ Here, contacts were defined if and only if the minimum distance between a pair of residues was within 3.5 Å. This was calculated and integrated over time for each residue pair, then normalised by the number of frames in order to get the fraction of time each pair is in contact. Thus, a value of 0 indicates a residue pair not ever being in contact, whereas a value of 1 means that the contact between the two residues persists uninterrupted throughout the time slot.

Results and discussion

Experimental unfolding patterns

The unfolding processes of both VP1 dimers were characterised using TWIMS. While the full capsid and the P dimer have both been subjected to ion mobility,^8,64 this is the first time the VP1 homodimer has been in the focus. The unfolding pattern of VP1 of GI. 1 Norwalk and GII. 17 Kawasaki can be seen in the CIU fingerprints (Fig. 2).

Fig. 2 Ion mobility of Norwalk and Kawasaki VP1 dimers. A CIU fingerprints of 22+ charge states of GI.1 Norwalk and GII.17 Kawasaki were observed from 5 to 100 V. See Fig. S3 (ESI†) for other charge states. Norwalk and Kawasaki display distinct unfolding patterns, where Norwalk shows multiple conformational changes, suggesting higher susceptibility to collisional unfolding. Dashed lines indicate CIU50 for conformational transitions as identified with CIUSuite 2. B Representative mass spectra of both dimers with selected charge states annotated. CIU fingerprints and CIU50 for these charge states can be found in Fig. S2 and Table S1 (ESI†).

Here, both dimers start with similar CCSs slightly above 6000 Ų. Both show an equal behaviour up until 40–60 V collision energy. Here, the CCS of Norwalk increases abruptly, suggesting a conformational transition. Using the CIUSuite 2 software⁴³ to analyse the mobility data opened the possibility to calculate the CIU50 (Table S1, ESI†). This is done by assigning features to the data sets, representing distinct conformations (Fig. S1, ESI†) and subsequently determining the collisional voltage where 50 percent of transition to a larger conformation has occurred (Fig. S2, ESI†).⁴³ For Norwalk the software calculates the first CIU50 at 51 V, which is in agreement with the CIU fingerprint (Fig. 2). The additional CIU50 for Norwalk are 71 V and 90 V. In comparison, Kawasaki has a transition around 70 V, which shows fine structure resulting in two transitions in the program. Norwalk going through multiple conformational changes, compared to a single one for Kawasaki, suggests a higher susceptibility of Norwalk to collisional unfolding and consequently less stability.

The experimental data here indicate strain-specific stability differences, which has been shown in previous works for the full capsids of the respective noroVLP. Kawasaki was observed to be more resistant to harsh environmental conditions than Norwalk.^7,12 This may in part explain some differences between the unfolding patterns in the IM experiments, as harsh buffer conditions were used to disassemble the capsids. Interestingly, this also agrees with a recent study on the mechanical properties of noroVLP. Atomic force microscopy unveiled differences in the elasticity of the capsids, also demonstrating Kawasaki to be more stable than Norwalk.¹⁰ A caveat here is of course the difference from gas phase, where we test unfolding, and in-solution testing. In the gas phase, hydrophobic contacts are weakened and hydrophilic ones are strengthened. Nevertheless, our observation of strain-specific dimer stability differences is consistent with the notion that the stability of the full capsid originates in the stability of VP1, which is the protomer of the norovirus virion and the dominating protein in the natural capsid.

CCSs indicate unfolding at higher temperatures. Using MD, we simulated the evolution of both Kawasaki and Norwalk VP1 structures in their +22e charge state under different levels of activation implemented via increased temperature. While MD is a powerful complement to MS, it is important to be aware of its limitations. The accuracy of the force fields might be reduced in the gas phase, and the difficulty of treating proton transfer reactions adequately during the simulations preclude prediction of some processes in MS.^65,66 The latter can be a problem when mimicking CIU, because protons might move between acids and bases during the unfolding steps, which enables other unfolding pathways as a consequence of changes in the electrostatic interactions within the protein. Another challenge is the nanosecond timescales in gas-phase MD, which is very short compared to the micro- to millisecond timescales in the experiments. The activation in our simulations are most likely overly strong in order to achieve structural changes on short timescales and is therefore not fully comparable to the activation in the experiments on a quantitative level. We take those difficulties into account when interpreting our simulations, especially for more activated systems where the structures change considerably and are less likely to have a fully realistic proton configuration. Consequently, we expect the simulations to be better able to reveal the relative stability of the two protein variants and the point where their structures break, than to give the specifics of the unfolded species and the structures' resilience to activation in absolute terms.

Theoretical CCS can be computed directly from MD trajectories, allowing for a connection between the high level of structural detail in MD and IM-MS experiments. To this end, we calculated the time evolution of the CCS for each temperature, averaged over the replicas, both for Norwalk and Kawasaki (Fig. 3). Temperatures up to 600 K did not result in any large increase of the CCS. In fact, initially, the CCSs decreased, most likely related to the structures being exposed to vacuum, resulting in side chains collapsing onto the protein surface.⁶⁷ We have seen a similar trend in our recent study and in published literature, linked to vacuum-compaction being the reason for an initial decrease of the CCS.^55,65,68,69 After 30 ns of MD simulation, both Norwalk and Kawasaki structures from 300 to 500 K arrive at similar CCSs (Fig. 3). These values are somewhat lower than what we see in the CIU experiments at low activation voltages (Fig. 2). The discrepancy might be explained by the tendency of the projection approximation, which IMPACT relies on, to slightly underestimate CCSs.⁷⁰ We therefore assume that the structures are indeed largely intact in the CIU up to the first unfolding step, apart from minor changes mostly on the protein surface arising from the adaptation to a solvent-free environment.^65,69

Fig. 3 Time-evolution of the CCS. The top panels shows the CCSs over time, averaged over replicas, for the Norwalk and Kawasaki at different temperatures, and the bottom panels show the CCSs averaged for each time slot and temperature together with the respective standard error of the mean. The individual CCS values shown in the bottom panels are shown in Table S2 (ESI†). Up to 600 K, the dimers resist unfolding. At 700 K Kawasaki retains a low CCS, but for Norwalk the CCS indicates unfolding. At temperatures above 700 K both protein variants unfold, but Norwalk to a higher degree than Kawasaki. Interestingly, Kawasaki indicates an initial drop of CCS, even at 900 K, whereas Norwalk shows an immediate increase of CCS at 900 K.

At 600 K Kawasaki is still at a CCS similar to those at lower temperatures, whereas the CCS is notably higher and increasing for Norwalk at that temperature, indicating differences in stability for the two dimers. This becomes even more evident when comparing the trends for Norwalk and Kawasaki at 700 K. Whilst we can observe an elevated and increasing CCS-trend for Kawasaki, Norwalk displays a more pronounced CCS increase from the start and over the 30 ns of simulation. At 800 and 900 K, both dimers appear to unfold, again with Norwalk reaching higher CCSs than Kawasaki. The strain-specificity shown for the experimental CCS values described above are further in agreement with the MD simulations. The experimental CCS values show a similar pattern to the time-evolution of theoretical CCS calculations (Fig. 3), especially for 700 K and Norwalk.

Defining time slots to represent states of unfolding. To investigate the simulated CIU in more detail, the MD trajectories were split into six time slots, each spanning 5 ns of simulation duration, with the purpose of identifying and describing an underlying unfolding mechanism of the VP1 dimers. The CCSs were averaged within each time slot, as shown in Fig. 3 and displayed in Table S1 (ESI†) together with their respective standard errors. Whilst the boundaries of the time windows are arbitrary, this structuring of the data enables us to calculate averages and error for different parts of the trajectories in a straightforward manner, and facilitates comparisons within our data as well as with values from literature. We acknowledge however that the different processes behind the experimental and simulated data (structural response to increasing activation vis-a-vis time evolution of individual structures) in conjunction with the computational challenges outlined earlier complicates a quantitative comparison of all states observed in the experiments with the simulations. Consequently, we make mostly qualitative interpretations of the observations from the MD in relation with the CIU.

For Norwalk and Kawasaki at temperatures up to 500 K, initially larger CCS values around 5200 Ų are seen for time slot 1, decreasing over time for the remaining time slots and matching those published for MS experiments of the Norwalk capsid proteins⁸ and our experimental data (Fig. 2). At 700 K, Kawasaki demonstrates a slight CCS increase of approximately 1%, whereas the CCS of Norwalk increases from time slot 1 to 6 by roughly 10% at the same temperature. At temperatures 800 and 900 K, both Norwalk and Kawasaki seem to undergo unfolding based on the rapid increase of the average CCS over the time slots. Interestingly, Norwalk increases by approximately 17% from time slot 1 to 6, both at 800 and 900 K, with 900 K data showing overall larger CCS values as during 800 K simulations. The Kawasaki dimer displays a 17% increase of CCS at 800 K as well, but from a lower starting value in time slot 1, and by approximately 23% when comparing the CCS values at 900 K for time slot 1 to time slot 6.

Overall, splitting the MD simulation data into time slots, and comparing the averaged CCS values for those time slots between each other could funnel the information of the underlying dynamics in a more comprehensive way. Moreover, this allows us to view the structures in the time slots as series of states the protein structures pass through as they unfold, with the potential to reveal an unfolding mechanism to compare to experiments.

Representative structures and comparison of unfolding time slots

With the simulation data split into six time slots, we clustered all structures that would fall within such a time slot in order to obtain a single representative structure for each slot, which we use to track and visualise the unfolding process. The structures for each temperature are presented in Fig. S4–S17 (ESI†). The RMSDs for the representative structures were calculated (Table 1) using the structure for time slot 1 as reference, which allows for an estimation of the deviation between the structures over the course of the simulations.

Table 1 Structural deviations within time slots. Comparing to the representative structures for time slot 1 within each temperature allows the calculation of deviations for all representative structures. Up to 600 K, Norwalk and Kawasaki structures demonstrate relatively low deviations from time slot 1. Whilst Kawasaki deviates at 700 K, the data suggests a progressively increasing deviation from the norm at 700 K for Norwalk, and even more for higher temperatures for both dimers

Time slots	RMSD per time slot for Norwalk relative to time slot 1 [Å]
	300 K	400 K	500 K	600 K	700 K	800 K	900 K
Time slot 1	0	0	0	0	0	0	0
Time slot 2	1.24	3.14	4.43	6.09	9.56	28.64	43.39
Time slot 3	2.60	2.32	4.58	6.30	13.60	32.40	51.09
Time slot 4	1.47	3.17	4.83	6.55	14.30	37.84	51.39
Time slot 5	2.54	3.19	4.74	6.74	16.65	41.54	55.76
Time slot 6	2.60	3.23	4.75	7.09	22.97	44.96	64.66
Time slot 1	0	0	0	0	0	0	0
Time slot 2	2.05	3.00	2.21	4.47	4.29	10.10	20.21
Time slot 3	1.26	2.94	2.28	4.71	5.84	14.91	31.90
Time slot 4	2.10	3.02	4.14	4.87	7.38	16.99	33.57
Time slot 5	2.10	3.05	4.05	4.79	7.53	20.47	37.56
Time slot 6	2.06	3.11	4.68	4.89	7.73	20.96	59.29

---

From the CCS calculations for 300 to 500 K, Norwalk dimers are seemingly stable and do not change their CCS much. This is confirmed by looking at the representative structures for these temperatures, shown in Fig. 4 and Fig. S4–S6 (ESI†). The RMSD calculations further support these observations, where the deviations increase with higher temperatures (Table 1). Especially the RMSD values for 500 K show that the structures corresponding to time slot 2 to 6 are deviating by at least 4 Å from the representative structure for time slot 1, indicating that more and more differences between these structures exist. Whilst the conformation of the Norwalk structure for time slot 1 at 600 K (Fig. S7, ESI†) resembles those at lower temperatures, a clear difference can be seen compared to time slot 2, where the structure of chain A is seemingly different as that of chain B. With an RMSD of 6.09 Å between these time slots, the deviation is most likely associated with parts of chain A starting to protrude from the rest of the structure. However, unfolding seems to not take place here on these time scales, as the deviations of the remaining time slots do not reflect large parts of the structure to change, which can also be seen looking at the respective structures in Fig. S7 (ESI†), and supported by the calculated CCSs. Yet again, 700 K seems to mark a limit for when the Norwalk dimer succumbs to the unfolding forces. Inspecting the structures in Fig. S8 (ESI†), already time slot 1 shows signs of unfolding, restricted to the same extruded area of chain A seen at 600 K (Fig. S7, ESI†). Starting from the termini, the structures show a gradual unfolding of the Norwalk dimer, explaining the steep increase of the RMSD from time slot 1, where after 10 ns of simulation, chain B began to unfold as well. Overall, the unfolding of the dimer is shown here to occur evenly over the simulation time. At 800 K (Fig. S9, ESI†), the Norwalk dimers start with an overall shape that resembles for the most part that of the native dimer, but from time slot 2 and onwards the structures are unfolded with no similarity of the original Norwalk dimer. With the RMSD ranging up to above 40 Å, the unfolding process seems to progress rapidly, which explains the observed CCS trends. At 900 K, unfolding occurs from the beginning of the simulations, where the structures are seemingly being ‘pulled’ apart as time progresses (Fig. S10, ESI†). From the large deviations presented in Table 1, together with the CCS trends at 900 K, the unfolding of Norwalk is shown to be happening promptly and chaotically, with the only clearly discernible mechanism is an unraveling starting at the termini.

Fig. 4 RMSF per residue projected on selected structures. The RMSF calculations were projected onto the selected representative structures, using the ribbon thickness and colour to show RMSF values. At 300 K, Kawasaki shows almost identical representative structures at 300 K for the first and last 5 ns of MD simulations, whereas Norwalk is showing slight distortions for time slot 6. At temperatures 700+ K, unfolding can be observed for Norwalk, whilst Kawasaki still shows a globular-like shape. Moreover, the representative structures suggest different regions within the proteins the unfolding to occur from. Norwalk displays unfolding to be most likely originating from the termini, with the N-terminus starting to unfold at 700 K. For the other dimer, Kawasaki, 800 K is required to induce unfolding, starting from residues within the P2-domain of the protein. For more details on all representative structures, please see Fig. S4–S17 (ESI†).

The CCSs for the Kawasaki dimer structures are similar between 300 and 600 K (Fig. 3 and Table S2, ESI†). A similar observation can be made by looking at the RMSDs per time slot (Table 1). We can rationalise this by comparing the representative structures at those temperatures (Fig. S11 and S14, ESI†). Here, whilst a small RMSD increase is seen from 300 to 500 K, the relatively small values most likely indicate minor structural deviations. Whilst the structures for 600 K demonstrate similar shapes and sizes, the RMSD values show deviations from the time slot 1 structure of 4.47 Å or more, indicating increasingly larger differences than for the lower temperatures. Larger structural differences are eventually seen for the representative structures at 700 K (Fig. S15, ESI†), which is further reflected by the RMSD increasing to 7.73 Å. As such, 700 K appears to be the threshold for unfolding in our simulations. Looking in more detail at the structures at 800 K (Fig. S16, ESI†), where time slot 1 displays a structure that resembles the native dimer structure the most, unfolding of the Kawasaki dimer happens from time slot 2 and onwards, seemingly starting from chain A in time slot 2, and for chain B in later time slots. The RMSD increases throughout the time slots at 800 K, up to approximately 21 Å. Overall, the structures for 800 K suggest unfolding over 30 ns of MD simulation, and moreover, that the unfolding is seemingly starting from specific areas of the protein, whilst the core of the protein retains a more globular and compact, yet disorganised structure. At the highest simulated temperature, 900 K, the representative structures and the RMSDs between them suggest severe structural rearrangements (Fig. S17, ESI†). Whilst initially the dimers still seem to have a globular-like structure, for time slot 2 to 5, not only is the unfolding visualised, but one can observe large perturbations of the structure stepping from one time slot to the next. With the calculated deviations between the representative structures starting from 20.21 Å between time slot 1 and 2, and ranging up to almost 60 Å at the end of the MD simulations, showing drastic unfolding of the dimer at 900 K.

Interrogating the unfolding pattern tracking residue dynamics and distances

Assigning a representative structure per time slot for each temperature data set, together with the initial CCS calculation, allowed us to present the dynamics of the dimers at the specified temperatures as a series of potential unfolding states. Here, whilst the dimers are shown to be rather stable at temperatures ranging from 300 to 600 K, the higher temperatures, especially 800 and 900 K, are seemingly inducing enough energy to cause the structural unfolding of VP1. The propagation of the overall dimer shape could be inferred from looking at the representative structures across the time slots and temperatures, but the dynamics at a finer level is harder to discern. To further pinpoint such details, the RMSF was calculated for each protein chain (Fig. S18–S31, ESI†), and compared together with distance and contact maps, all calculated as averages for each time slot (Fig. S32–S45, ESI†).

Looking at the RMSFs for Norwalk at 300 to 500 K (Fig. S18–S20, ESI†), one can observe similarities between the time slots. Whilst the RMSFs seem to increase with higher temperature, the values likely reflect an overall increase in mobility, especially around the termini, rather than unfolding. This is further supported by the almost identical patterns of the distance and contact maps for all time slots (Fig. S32–S34, ESI†). At 600 K, the majority of the protein is still displaying a similar RMSF trend over all time slots, both for chain A and B (Fig. S21, ESI†). Noteworthy are the high RMSF values for the N-termini of both chains, whereas the C-terminus of chain A is seemingly more mobile at 600 K than for chain B, displayed as well in the distance maps (Fig. S35, ESI†). The representative structure for Norwalk time slot 2 at 600 K (Fig. S7, ESI†) shows an extruded part of the protein chain, which is given by the N-terminus, and therefore could indicate that area to be a potential starting point of the unfolding process. As the N-terminus of chain B shows basically an identical RMSF over all time slots, we could conclude that chain B, whilst being highly mobile, is still stable enough and does not start to unfold at 600 K. At 700 K, the RMSF trends for the time slots are starting to differentiate from each other (Fig. S22, ESI†), furthermore confirming, as suggested by the CCS and RMSD values at that temperature, that 700 K is the threshold temperature to induce unfolding for Norwalk. In detail, whilst both termini of chain A are still displaying large RMSF values, we can see other parts of the protein becoming more mobile as well. A prominent peak is given for residues between approximately Ala₁₂₀ and Ile₁₃₉, comprising a loop between β-strand_E and α-helix₃ within the S-domain of the protein chain.²⁰ This area is increased in mobility the most over time, as compared to the rest of the protein (excluding termini), which is further shown in the distance and contact maps (Fig. S36, ESI†). Towards 800 and 900 K, disturbances in the Norwalk dimers become more evident. The RMSFs in Fig. S23 (ESI†) exhibit, next to the prominent peaks as observed already at 700 K, further differences from one time slot to the next. High RMSF values are seen throughout each chain, where additional peaks, for example shown around residue Asn₃₀₀ within the P2-domain, are becoming more prominent as the simulation progresses. The distance maps in Fig. S37 (ESI†) provide further information of the unfolding, where the distances increase whilst the contact patterns become progressively more diffuse. Moreover, differences in the unfolding dynamics of the chains can be seen, where for example the distances between residues of chain B to each other are not increasing as much as for chain A, potentially indicating a more compact conformation of chain B at 800 K than chain A, whilst residues within chain A are increasingly departing from each other. At 900 K, the RMSFs suggest large distortions and mobility throughout the protein chains (Fig. S24, ESI†), with chain A and B progressively being more “chaotic” and dissimilar. High fluctuations are recorded for several areas of the protein chains, where interestingly the N-terminus of chain A actually shows smaller values than for 800 K, whilst the same residues in chain B are reaching RMSF values above 70 Å. Moreover, as shown in Fig. S38 (ESI†), where unfolding is already present for time slot 1, the distances and contacts for time slot 3 to 6 are very similar, which could indicate that 900 K is inducing unfolding too rapidly and highly incoherently for it to provide useful information to track the unfolding process with any confidence.

The Kawasaki data shows similar RMSFs from 300 K to 500 K (Fig. S25–S27, ESI†), as was observed for Norwalk as well. Together with the distance and contact maps (Fig. S39–S41, ESI†), this suggests similar underlying dynamics across these temperatures. Whilst a slight increase in RMSF can be seen in at 600 K, the distances indicate no obvious differences of the dynamics between the six time slots, and as such complement what was seen for the average CCS and RMSD at 600 K. Moreover, both chains are still exhibiting similar dynamics throughout, even at 600 K. At 700 K, we can see that the dynamics of the residues are increasing over the MD simulations (Fig. S29, ESI†). Especially at later simulation times, for time slot 5 to 6, peaks throughout the protein chains reach up to 20 Å. Interestingly, the termini show relatively low RMSF values, a very different observation as compared to Norwalk, where the termini, especially the N-termini, are displaying high mobility. Turning to the contact maps for Kawasaki at 700 K in (Fig. S43, ESI†), we can further see that the structures, whilst being highly mobile, are seemingly able to withstand unfolding, as no obvious increase in distance maps can be seen. Moreover, the contact patterns for the time slots at 700 K are displaying similar patterns, indicating the absence of induced unfolding. At 800 K however, the CCS of Kawasaki increases, accompanied by RMSF increases for that temperature (Fig. S30, ESI†). The termini show high RMSFs that increase over time, with new parts of the protein becoming progressively more dynamic. Especially parts of the P2-domain, around residue Ala₃₅₀, present a prominent, increasing peak over time. That area is seen to unfold in the representative Kawasaki structures for time slot 2 and above at 800 K, which is reflected by the increasing distances involving the corresponding residues (Fig. S44, ESI†). We described a similar observation in a recent study, where we experimentally investigated different P dimers with hydrogen–deuterium exchange together with MD simulations.⁷¹ Amongst the investigated proteins was the P dimer of the Kawasaki strain, and the results suggested increased dynamics both during the experiments as well as during the simulations for the same area we observe the unfolding to start from in this investigation. Interestingly, Kawasaki is seemingly unfolding from a completely different part of the protein than Norwalk, which appears to start unfolding from the N-termini already at 700 K. The Kawasaki contact patterns fade for higher temperatures, suggesting gradual unfolding, with structural differences within or between replicas, and a more subtle trend than Norwalk. Eventually, at the highest temperature of 900 K, two areas of the structure have large RMSFs (Fig. S31, ESI†). They comprise largely residues of the S- and P2-domains, spanning residues from approximately Ala₁₀₀ to Cys₂₀₀, as well as Trp₃₀₀ till Phe₄₀₀, which were mobile also for Norwalk. Interestingly, the latter part includes a short loop for which strain-specific dynamics has been observed, and where glycan binding is known to take place as part of the infection process.⁷¹ The RMSF increases over time, per time slot, and reaches values above 40 Å, which is evidently linked to large distances being observed in the corresponding contact maps as a consequence of the unfolding (Fig. S45, ESI†). However, the distance maps of Kawasaki appear to suggest a more gradual unfolding process, as the distances increase steadily, yet for the same specific parts of the protein, rather than all over the structure.

Conclusions

In this study we investigated how we can use CIU and MD to find and explain differences in dynamics and stability between different variants of the same protein. Specifically, we interrogated the VP1 dimers from two norovirus strains with minor structural differences, but with decidedly distinct physical properties on the capsid level. The CIU experiments revealed differences in the unfolding dynamics, with the Kawasaki variant not only requiring more activation to start unfolding, but also going through fewer conformational transitions overall. MD simulations of the same proteins, where elevated temperatures mimicked the activation in the experiments, gave insights into the conformational changes the protein underwent during the unfolding. We employed a decomposition of our MD simulations into time slots, which we think can be developed further, to gain useful information for and from CIU experiments, utilising the mutually complementary nature of MD and IM-MS. First, we note that a high temperature is necessary to induce unfolding on relative short timescales of 30 ns. Moreover, the temperature thresholds at which the first instances of unfolding were observed were different for the two dimers, consistent with the different profiles obtained from CIU. For example, for Norwalk, we have seen that 700 K is enough to induce unfolding, and higher temperatures make it progressively more chaotic, which makes it difficult to pinpoint and describe the full pathway. Kawasaki requires 800 K and above to undergo unfolding, as is suggested by the representative structures and RMSF calculation (Fig. 4), whereas at 700 K, Kawasaki is stable enough to withstand unfolding. In addition to the caveats outlined earlier, it is important to stress that classical force fields might not represent high (or low) temperatures accurately, which entails that the exact dependence on temperature can have poor correspondence to actual temperatures in an experiment. Compounding this, we use very high temperatures to compensate for the short time scales compared to the millisecond-timescales in IM-MS, which risk affecting the apparent stability of unfolding intermediates, and the unfolding pathways thus risk becoming artificially “smooth”. We also remind ourselves that as proteins unfold in experiment there is the possibility of scrambling of protons across the structure, which is not accommodated in our simulations, which also limits how confidently we can draw conclusions about the most unfolded structures. This is particularly true for multimeric proteins such as our dimers, where charges are known to partition asymmetrically when the proteins dissociate,⁷² typically leading to an asymmetric dissociation where one highly charged subunit unfolds completely, leaving the rest of the complex more intact and at a lower m/z ratio. This complicates the correspondence between simulations and experiments further. It is challenging to incorporate such dynamics in MD. Important advances have been made recently,^73,74 but it remains a difficult endeavour and to our knowledge there is to date no established means for accomplishing it that is thermodynamically let alone kinetically accurate. Notwithstanding such obstacles, the simulations should provide a qualitatively correct view of the unfolding, at least for moderately unfolded states, especially with respect to the weakest points of the proteins and the relative stability between the variants.

With the two dimers having similar sizes and shapes, we see that their structural responses to the activation required to induce unfolding are different. The starting point of the unfolding process, for example, differs between Norwalk and Kawasaki; Norwalk is starting to unfold starting from the N-termini, whereas Kawasaki is unfolding from parts of the P2-domain. In a recent study, we have shown that residues in that part of Kawasaki, around residue Ala₃₅₀, are displaying large dynamics,⁷¹ a similar observation we make during the unfolding simulated here. It is interesting to see how features of those in vitro dynamics carry over to the gas phase. The importance of the N-termini for capsid stability has been previously investigated.^7,9 A higher instability of the critical N-terminus may also influence the overall decreased stability of the Norwalk capsid compared to Kawasaki as shown in other studies.^7,10,12 In light of the new data presented here together with earlier results, the VP1 dimer is becoming the likely origin of the strain-dependent distinction in physical properties of the fully assembled capsids. To our knowledge, we present here the first study of combining both experimental and theoretical research on the unfolding of norovirus dimers, providing valuable insights into their stability and details about their behaviour and thus adding to the compounding evidence on previously identified structural linchpins. And as such we also demonstrate how CIU and MD can be combined to shed more light on protein systems than either technique in isolation. We firmly believe that studies of this kind will gradually advance our understanding of how proteins respond to the experimental conditions and how they can be experimentally interrogated, and that future advances in computational techniques and algorithms as well as in experimental methods and instrumentation will accelerate that process.

Author contributions

Maxim N. Brodmerkel: conceptualization, methodology, formal analysis, investigation, validation, writing – original draft, visualization. Lars Thiede: methodology, formal analysis, investigation, writing – editing. Emiliano De Santis: methodology, writing – review & editing. Charlotte Uetrecht: writing – review & editing, supervision, funding acquisition. Carl Caleman: conceptualization, writing – review & editing, supervision, funding acquisition. Erik G. Marklund: conceptualization, resources, writing – review & editing, supervision, funding acquisition.

Conflicts of interest

There are no conflicts to declare.

Acknowledgements

The authors would like to acknowledge the Uppsala Multidisciplinary Center for Advanced Computational Science (UPPMAX), the National Academic Infrastructure for Supercomputing in Sweden (NAISS), former Swedish National Infrastructure for Computing (SNIC), and the provided computational resources under projects SNIC 2022-22-854, 2022-22-925, 2022-22-947, 2022-5-415 and 2022-23-57. EGM and CC are supported by Project grants from the Swedish Research Council (2020-04825 and 2018-00740). EDS, CU, CC and EGM acknowledge support from a Röntgen Ångström Cluster grant provided by the Swedish Research Council and the Bundesministerium für Bildung und Forschung (2021-05988, 05K22PSA). MNB, EDS, CU, CC and EGM acknowledge support from the MS SPIDOC consortium, funded by the European Horizon 2020 research and innovation programme under grant agreement 801406. CU and LT acknowledge funding by EIC Pathfinder Open Ariadne-Vibe (964553). CC further acknowledges the Helmholtz Association through the Center of Free-electron Laser Science at DESY. Lastly, we thank Kristina Lorenzen and the XBI BioLab for instrument access.

Notes and references

1. D. S. Goodsell and A. J. Olson, Annu. Rev. Biophys. Biomol. Struct., 2000, 29, 105–153.
2. G. K. A. Hochberg, Y. Liu, E. G. Marklund, B. P. H. Metzger, A. Laganowsky and J. W. Thornton, Nature, 2020, 588, 503–508.
3. N. F. Steinmetz, T. Lin, G. Lomonossoff and J. Johnson, Structure-based engineering of an icosahedral virus for nanomedicine and nanotechnology, Springer, 2009.
4. E. J. Elliott, BMJ, 2007, 334, 35–40.
5. R. I. Glass, U. D. Parashar and M. K. Estes, N. Engl. J. Med., 2009, 361, 1776–1785.
6. N. Zachos, in Viral Gastroenteritis, ed. L. Svensson, U. Desselberger, H. B. Greenberg and M. K. Estes, Academic Press, Boston, 2016, 1.1, pp. 1–21.
7. G. K. Shoemaker, E. van Duijn, S. E. Crawford, C. Uetrecht, M. Baclayon, W. H. Roos, G. J. Wuite, M. K. Estes, B. V. Prasad and A. J. Heck, Mol. Cell. Proteomics, 2010, 9, 1742–1751.
8. C. Uetrecht, I. M. Barbu, G. K. Shoemaker, E. Van Duijn and A. J. Heck, Nat. Chem., 2011, 3, 126–132.
9. R. Pogan, V. U. Weiss, K. Bond, J. Dülfer, C. Krisp, N. Lyktey, J. Müller-Guhl, S. Zoratto, G. Allmaier and M. F. Jarrold, et al. , Vaccines, 2020, 9, 8.
10. Y. Feng, R. Pogan, L. Thiede, J. Müller-Guhl, C. Uetrecht and W. H. Roos, Viruses, 2023, 15, 1482.
11. C. Scheller, F. Krebs, R. Minkner, I. Astner, M. Gil-Moles and H. Wätzig, Electrophoresis, 2020, 41, 1137–1151.
12. R. Pogan, J. Dülfer and C. Uetrecht, Curr. Opin. Virol., 2018, 31, 59–65.
13. L. Barclay, G. Park, E. Vega, A. Hall, U. Parashar, J. Vinjé and B. Lopman, Clin. Microbiol. Infect., 2014, 20, 731–740.
14. M. E. Hardy, FEMS Microbiol. Lett., 2005, 253, 1–8.
15. G. I. Parra, Virus Evol., 2019, 5, vez048.
16. E. Robilotti, S. Deresinski and B. A. Pinsky, Clin. Microbiol. Rev., 2015, 28, 134–164.
17. C. Preeti, D. G. Miranda, I. P. Gabriel, C.-W. C. Martin, G. Kim, M. Vito, W. Qiuhong, A. W. Peter, K. Kazuhiko and V. Harry, et al. , J. Gen. Virol., 2019, 100, 1393–1406.
18. K. Bányai, M. K. Estes, V. Martella and U. D. Parashar, The Lancet, 2018, 392, 175–186.
19. Y. Liao, L. Xue, J. Gao, Y. Zuo, Y. Liang, Y. Jiang, W. Cai, J. Yang, J. Zhang and Y. Ding, et al. , Gut Pathog., 2022, 14, 31.
20. B. V. Prasad, M. E. Hardy, T. Dokland, J. Bella, M. G. Rossmann and M. K. Estes, Science, 1999, 286, 287–290.
21. J. Jumper, R. Evans, A. Pritzel, T. Green, M. Figurnov, O. Ronneberger, K. Tunyasuvunakool, R. Bates, A. Ždek and A. Potapenko, et al. , Nature, 2021, 596, 583–589.
22. E. F. Pettersen, T. D. Goddard, C. C. Huang, G. S. Couch, D. M. Greenblatt, E. C. Meng and T. E. Ferrin, J. Comput. Chem., 2004, 25, 1605–1612.
23. C. Uetrecht and A. J. Heck, Angew. Chem., Int. Ed., 2011, 50, 8248–8262.
24. F. Lanucara, S. W. Holman, C. J. Gray and C. E. Eyers, Nat. Chem., 2014, 6, 281–294.
25. E. G. Marklund, M. T. Degiacomi, C. V. Robinson, A. J. Baldwin and J. L. Benesch, Structure, 2015, 23, 791–799.
26. V. Gabélica and E. Marklund, Curr. Opin. Chem. Biol., 2018, 42, 51–59.
27. E. Mack, J. Am. Chem. Soc., 1925, 47, 2468–2482.
28. B. T. Ruotolo, K. Giles, I. Campuzano, A. M. Sandercock, R. H. Bateman and C. V. Robinson, Science, 2005, 310, 1658–1661.
29. M. Kaldmäe, C. Sahin, M. Saluri, E. G. Marklund and M. Landreh, Prot. Sci., 2019, 28, 1024–1030.
30. M. Landreh, C. Sahin, J. Gault, S. Sadeghi, C. L. Drum, P. Uzdavinys, D. Drew, T. M. Allison, M. T. Degiacomi and E. G. Marklund, Anal. Chem., 2020, 92, 12297–12303.
31. S. B. A. Turzo, J. T. Seffernick, A. D. Rolland, M. T. Donor, S. Heinze, J. S. Prell, V. H. Wysocki and S. Lindert, Nat. Commun., 2022, 13, 4377.
32. K. B. Shelimov, D. E. Clemmer, R. R. Hudgins and M. F. Jarrold, J. Am. Chem. Soc., 1997, 119, 2240–2248.
33. S. M. Dixit, D. A. Polasky and B. T. Ruotolo, Curr. Opin. Chem. Biol., 2018, 42, 93–100.
34. C. Eldrid, T. Cragnolini, A. Ben-Younis, J. Zou, D. P. Raleigh and K. Thalassinos, Anal. Chem., 2022, 94, 16113–16121.
35. A. Laganowsky, E. Reading, T. M. Allison, M. B. Ulmschneider, M. T. Degiacomi, A. J. Baldwin and C. V. Robinson, Nature, 2014, 510, 172–175.
36. D. D. Vallejo, C. K. Jeon, K. F. Parson, H. R. Herderschee, J. D. Eschweiler, D. I. Filoti and B. T. Ruotolo, Anal. Chem., 2022, 94, 6745–6753.
37. E. G. Marklund and J. L. Benesch, Curr. Opin. Struct. Biol., 2019, 54, 50–58.
38. Z. Hall, A. Politis, M. F. Bush, L. J. Smith and C. V. Robinson, J. Am. Chem. Soc., 2012, 134, 3429–3438.
39. S.-H. Chen and D. H. Russell, J. Am. Soc. Mass Spectrom., 2015, 26, 1433–1443.
40. C. E. Bartman, H. Metwally and L. Konermann, Anal. Chem., 2016, 88, 6905–6913.
41. G. Hansman, K. Natori, T. Oka, S. Ogawa, K. Tanaka, N. Nagata, H. Ushijima, N. Takeda and K. Katayama, Arch. Virol., 2005, 150, 21–36.
42. G. S. Hansman, H. Saito, C. Shibata, S. Ishizuka, M. Oseto, T. Oka and N. Takeda, J. Clin. Microbiol., 2007, 45, 1347–1349.
43. D. A. Polasky, S. M. Dixit, S. M. Fantin and B. T. Ruotolo, Anal. Chem., 2019, 91, 3147–3155.
44. M. T. Marty, A. J. Baldwin, E. G. Marklund, G. K. A. Hochberg, J. L. P. Benesch and C. V. Robinson, Anal. Chem., 2015, 87, 4370–4376.
45. K. Thalassinos, M. Grabenauer, S. E. Slade, G. R. Hilton, M. T. Bowers and J. H. Scrivens, Anal. Chem., 2009, 81, 248–254.
46. M. F. Bush, Z. Hall, K. Giles, J. Hoyes, C. V. Robinson and B. T. Ruotolo, Anal. Chem., 2010, 82, 9557–9565.
47. V. Gabelica, A. A. Shvartsburg, C. Afonso, P. Barran, J. L. Benesch, C. Bleiholder, M. T. Bowers, A. Bilbao, M. F. Bush and J. L. Campbell, et al. , Mass Spectrom. Rev., 2019, 38, 291–320.
48. Y. Matsushima, M. Ishikawa, T. Shimizu, A. Komane, S. Kasuo, M. Shinohara, K. Nagasawa, H. Kimura, A. Ryo and N. Okabe, et al. , Eurosurveillance, 2015, 20, 1–12.
49. B. Hess, C. Kutzner, D. Van Der Spoel and E. Lindahl, J. Chem. Theory Comput., 2008, 4, 435–447.
50. E. G. Marklund, D. S. Larsson, D. van der Spoel, A. Patriksson and C. Caleman, Phys. Chem. Chem. Phys., 2009, 11, 8069–8078.
51. G. A. Kaminski, R. A. Friesner, J. Tirado-Rives and W. L. Jorgensen, J. Phys. Chem. B, 2001, 105, 6474–6487.
52. K. A. Feenstra, B. Hess and H. J. Berendsen, J. Comput. Chem., 1999, 20, 786–798.
53. A. Patriksson, E. Marklund and D. van der Spoel, Biochemistry, 2007, 46, 933–945.
54. T. Mandl, C. Östlin, I. E. Dawod, M. N. Brodmerkel, E. G. Marklund, A. V. Martin, N. Timneanu and C. Caleman, J. Phys. Chem. Lett., 2020, 11, 6077–6083.
55. M. N. Brodmerkel, E. De Santis, C. Uetrecht, C. Caleman and E. G. Marklund, Curr. Res. Struct. Biol., 2022, 4, 338–348.
56. M. N. Brodmerkel, E. De Santis, C. Caleman and E. G. Marklund, Protein J., 2023, 1–14.
57. A. Sinelnikova, T. Mandl, H. Agelii, O. GrÅnäs, E. G. Marklund, C. Caleman and E. De Santis, Biophys. J., 2021, 120, 3709–3717.
58. W. L. Jorgensen, J. Chandrasekhar, J. D. Madura, R. W. Impey and M. L. Klein, J. Chem. Phys., 1983, 79, 926–935.
59. G. Bussi, D. Donadio and M. Parrinello, J. Chem. Phys., 2007, 126, 014101.
60. B. Hess, H. Bekker, H. J. Berendsen and J. G. Fraaije, J. Comput. Chem., 1997, 18, 1463–1472.
61. H. J. Berendsen, J. v Postma, W. F. Van Gunsteren, A. DiNola and J. R. Haak, J. Chem. Phys., 1984, 81, 3684–3690.
62. M. Parrinello and A. Rahman, J. Appl. Phys., 1981, 52, 7182–7190.
63. R. J. Gowers, M. Linke, J. Barnoud, T. J. E. Reddy, M. N. Melo, S. L. Seyler, J. Domanski, D. L. Dotson, S. Buchoux, I. M. Kenney and O. Beckstein, Proceedings of the 15th Python in Science Conference, 2016, pp. 98–105.
64. J. Z. Bereszczak, I. M. Barbu, M. Tan, M. Xia, X. Jiang, E. van Duijn and A. J. Heck, J. Struct. Biol., 2012, 177, 273–282.
65. D. van der Spoel, E. G. Marklund, D. S. D. Larsson and C. Caleman, Macromol. Biosci., 2011, 11, 50–59.
66. T. M. Allison, P. Barran, S. Cianférani, M. T. Degiacomi, V. Gabelica, R. Grandori, E. G. Marklund, T. Menneteau, L. G. Migas, A. Politis, M. Sharon, F. Sobott, K. Thalassinos and J. L. P. Benesch, Anal. Chem., 2020, 92, 10872–10880.
67. K. Breuker and F. W. McLafferty, Proc. Natl. Acad. Sci. U. S. A., 2008, 105, 18145–18152.
68. M. Landreh, E. G. Marklund, P. Uzdavinys, M. T. Degiacomi, M. Coincon, J. Gault, K. Gupta, I. Liko, J. L. P. Benesch, D. Drew and C. V. Robinson, Nat. Commun., 2017, 8, 13993.
69. A. D. Rolland and J. S. Prell, Trends Anal. Chem., 2019, 116, 282–291.
70. J. L. P. Benesch and B. T. Ruotolo, Curr. Opin. Struct. Biol., 2011, 21, 641–649.
71. J. Dülfer, H. Yan, M. N. Brodmerkel, R. Creutznacher, A. Mallagaray, T. Peters, C. Caleman, E. G. Marklund and C. Uetrecht, Molecules, 2021, 26, 2125.
72. J. C. Jurchen and E. R. Williams, J. Am. Chem. Soc., 2003, 125, 2817–2826.
73. L. Konermann, J. Phys. Chem. B, 2017, 121, 8102–8112.
74. R. Marchese, R. Grandori, P. Carloni and S. Raugei, J. Am. Soc. Mass Spectrom., 2012, 23, 1903–1910.

---

Footnote

† Electronic supplementary information (ESI) available. See DOI: https://doi.org/10.1039/d3cp06344e

---