International Journal of Biological Macromolecules 164 (2020) 2671–2680

Contents lists available at ScienceDirect

International Journal of Biological Macromolecules

journal homepage: http://www.elsevier.com/locate/ijbiomac

Unravelling the mechanism of pH-regulation in dinoflagellate

Natasha Kamerlin a,⁎, Mickaël G. Delcey b, Roland Lindh c a Science for Life Laboratory, Department of Cell and Molecular Biology, Uppsala University, SE-751 05 Uppsala, Sweden b Department of Chemistry – Ångström Laboratory, Uppsala University, SE-751 02 Uppsala, Sweden c Department of Chemistry – BMC, Uppsala University, P.O. Box 576, SE-751 23 Uppsala, Sweden article info abstract

Article history: Dinoflagellates are the dominant source of in coastal waters. The luminescence reaction in- Received 26 June 2020 volves the oxidation of luciferin by a luciferase , which only takes place at low pH. The pH-dependence Received in revised form 7 August 2020 has previously been linked to four conserved histidines. It has been suggested that their protonation might in- Accepted 7 August 2020 duce a conformational change in the enzyme, thereby allowing substrate access to the binding pocket. Yet, the Available online 19 August 2020 precise mechanism of luciferase activation has remained elusive. Here, we use computational tools to predict the open structure of the luciferase in Lingulodinium polyedra and to decipher the nature of the opening mecha- Keywords: Bioluminescence nism. Through accelerated molecular dynamics simulations, we demonstrate that the closed-open conforma- Dinoflagellates tional change likely takes place via a tilt of the pH-regulatory helix-loop-helix domain. Moreover, we propose Luciferase that the molecular basis for the transition is electrostatic repulsion between histidine-cation pairs, which desta- pH regulation bilizes the closed conformation at low pH. Finally, by simulating truncated mutants, we show that eliminating the Molecular dynamics C-terminus alters the shape of the active site, effectively inactivating the luciferase. © 2020 The Author(s). Published by Elsevier B.V. This is an open access article under the CC BY license (http:// creativecommons.org/licenses/by/4.0/).

1. Introduction favourable or unfavourable, the pH-regulation in this luciferase is unique [5]. The luciferase activity is namely regulated by instead regu- A large number of ocean animals are bioluminescent, meaning that lating substrate access to the active site. they are able to produce through a chemical reaction. A diverse The L. polyedra luciferase polypeptide chain contains three homolo- range of species rely on this phenomena as a means to, e. g., communi- gous domains and each individual domain is known to be enzymatically cate, lure prey, ward off predators, and to find mating partners. The im- active, participating in the bioluminescence reaction with pH-activity portance of bioluminescence is demonstrated by the fact that this trait profiles similar to that of the entire chain [5,6]. The crystal structure of has evolved independently several times in the history of life [1,2]. one of the domains (D3) in its inactive form was solved by Schultz In coastal regions, unicellular protists called dinoflagellates are the et al. [7]in2005(seeFig. 1). With this it was revealed that D3 is com- most commonly encountered bioluminescent organism. In particular, posed of two major subdomains: a β-barrel, thought to contain the cat- Lingulodinium polyedra (L. polyedra), a species of dinoflagellates usually alytic site, and a 3-helix bundle situated above the barrel, which in the studied as a model system in circadian biology, produces a blue light in crystal structure restricts access to the binding pocket. Although the response to water turbulence, such as breaking waves. The biochemical open luciferase structure has remained unknown, a drop in pH presum- reaction responsible for luminescence is the oxidation of a luciferin sub- ably causes the 3-helix bundle to undergo a conformational transition strate, catalyzed by the dinoflagellate luciferase. Pioneering work by which allows the luciferin substrate to bind to the active site. The activ- John Woodland Hastings over several decades laid the foundation for ity is optimal at around pH 6 and almost nil at pH 8 [8,9]. much of our current understanding of dinoflagellate bioluminescence. The pH-regulated closed-open conformational transition of lucifer- For example, Hastings et al. found that the light flashes are emitted ase is a necessary first step in the bioluminescence reaction to enable from organelles, which they named scintillons, located in the cytoplasm substrate binding, yet the mechanism has remained largely unknown. [3]. Physical agitation of the organism triggers a signal cascade resulting One mechanism was suggested by Schultz et al. which involves the pro- inadropofpHwithinthescintillons,andthisdropinpHinturntriggers tonation of four highly-conserved histidines (HIS899, HIS909, HIS924, the bioluminescence reaction [4]. While a pH-dependent mechanism and HIS930) located on the N-terminus (see Fig. 1)[7]. Based on the for activity regulation in often involves active site residues as- crystal structure, it was postulated that the histidines take part in a suming different protonation states, thereby making a reaction -bond network which stabilizes the closed conformation at pH 8. Protonation of these histidines through acidification could be re- ⁎ Corresponding author. sponsible for disrupting the hydrogen bonds, allowing the helices to E-mail address: [email protected] (N. Kamerlin). move apart. Support for this hypothesis was found in experiments

https://doi.org/10.1016/j.ijbiomac.2020.08.071 0141-8130/© 2020 The Author(s). Published by Elsevier B.V. This is an open access article under the CC BY license (http://creativecommons.org/licenses/by/4.0/). 2672 N. Kamerlin et al. / International Journal of Biological Macromolecules 164 (2020) 2671–2680

Fig. 1. 3D structure of the D3 domain of L. polyedra luciferase. (a) The α-helices, including the pH-regulatory 3-helix bundle (α2, α5, and α6), are shown in red and the catalytic β-barrel shown in blue. (b) The four conserved N-terminal histidines (HIS899, HIS909, HIS924, and HIS930) are annotated.

where replacing the histidines by alanine [5] or removing them by trun- our results, we propose a novel opening mechanism that is consistent cating the N-terminus [5,10] produced mutant luciferase with an in- with experimental findings. creased luminescence activity under alkaline conditions. This hypothesis, however, does not explain why these mutants also 2. Methods displayed a significant loss of activity at pH 6 relative to the full length D3. In other words, the proposed mechanism of disruption of a hydro- The molecular dynamics simulations were carried out using gen bond network through histidine protonation is either incorrect or Amber18 [12], with a SHAKE algorithm to constrain all bonds involving only part of the whole picture. hydrogen, thereby allowing a larger time step of 2 fs. The non-bonded An understanding of the link between structure and dynamics of the cutoff was set to 9 Å and the Particle Mesh Ewald method was used luciferase can be gained by employing molecular dynamics simulations. for treating electrostatic interactions. For better performance, the pro- In a recent communication, Donnan et al. used a coupled approach of duction runs used the CUDA version of Amber18 (pmemd.cuda). constant pH molecular dynamics with accelerated molecular dynamics and an implicit solvent model to investigate the effect of pH change 2.1. Protein preparation and preprocessing on the luciferase D3 domain [11]. The authors observed a conforma- tional change at pH 6 and correlated it to the protonation of histidine The crystal structure of L. polyedra luciferase D3 was downloaded residues. They described this change as a reorganization of the N- from the Protein Data Bank (PDB code 1VPR) and refined using terminal domain containing the conserved histidine residues, which Chimera's Dock Prep [13]. As in ref. [11], the 23 missing C-terminal res- are part of the pH-regulatory 3-helix bundle. However, the study did idues were generated using the I-TASSER On-line Server [14–16] and not provide a discussion on how the protonation state change induces PyMol [17] was used to align and extract the modelled coordinates. the observed change in conformation. Interestingly, the authors noted Based on this full length D3 structure, two mutants were created by re- some unexpected structural differences at pH 8 compared with the moving 65 (corresponding to removing α2) and 50 amino acid residues crystal structure, including a highly mobile N-terminus and 3-helix from the N- and C-terminus, respectively. These values were chosen in bundle. order to be able to compare with experimental data, see ref. [10]. While much focus has been placed on understanding the pH- From now on, these mutants will be referred to as N65 and C50. regulation mechanism of the 3-helix bundle and the role of the N- The H++ web server [18–20] was used to find the most probable terminus, the C-terminus has received little attention, although studies protonation states and net charge. Default values were kept for the ex- by Suzuki-Ogoh et al. would indicate that this terminus is important for ternal dielectric and salinity while a comparatively low value of 5 was the luminescence reaction [10]. The disordered 23-aa C-terminus was chosen for the internal dielectric parameter. The latter was motivated unresolved in the crystal structure. Yet, despite not being part of the ac- by the large estimated desolvation penalty of the four conserved tive site, C-terminal truncated mutants were found to have almost no histidines. activity at pH 6. It was speculated that the C-terminus might cap the ac- The protein structures were prepared for Amber simulations using tive site, maintaining a hydrophobic micro-environment and enhancing the tleap module of Amber Tools 18 and parametrised with the Amber fi the ef ciency of the reaction [10]. ff14SB force field. We note that preliminary results using the ff99SB To solve the discrepancy between the prevailing hypothesis for the force field produced unstable helical structures over relatively short fi pH-regulated opening mechanism and experimental ndings, the pres- simulation runs, whereas the updated torsion terms and improved di- ent study uses computational tools to explore in atomistic detail the pH- hedral sampling of ff14SB [21] produced stable structures. dependent opening mechanism of luciferase. Expanding on the work by All existing water molecules in the crystal structure were retained, Donnan et al. [11], constant pH accelerated molecular dynamics simula- given that these participate in a hydrogen-bond network in the interior tions are performed with an explicit solvent at pH 8 and pH 6, to follow cavity of the β-barrel. The systems were solvated using the TIP3P water the transition of the closed luciferase to an open form and to analyze the model in a periodically repeating truncated octahedral box, with a dynamics and interactions of the 3-helix bundle. Furthermore, the role buffer layer of 17 Å between solute atoms and the box boundary. The of the N- and C-termini in modulating the luminescence activity are sys- relatively large buffer value was chosen to accommodate for potentially tematically investigated by also simulating truncated mutants. Based on substantial structural changes during the simulation. The salt N. Kamerlin et al. / International Journal of Biological Macromolecules 164 (2020) 2671–2680 2673 concentration within the scintillon system containing the luciferase is states had changed, 200 fs of solvent relaxation dynamics were per- unknown. Early studies by Hastings et al. found that the soluble lumi- formed before starting the next cycle. nescent system from L. polyedra would not emit light unless salt is added, although certain salts had an inhibitory effect on the activity 2.3. Accelerated molecular dynamics [22,23]. Salt ions were introduced, in addition to net neutralising coun- ter ions, by randomly replacing water molecules to achieve a salt con- Preliminary results with 1 μs classical MD did not reveal any sub- centration of 0.15 M NaCl. stantial large-scale conformational changes, thus motivating the use of an enhanced sampling method, namely accelerated molecular dynam- 2.2. Classical molecular dynamics ics (aMD) [26], to be able to access longer time scale molecular events. The notion behind aMD is to reduce the height of local barriers 2.2.1. Equilibration through the addition of a boost potential whenever the system potential The solvent and ions in each system were energy minimized using energy falls below a predefined threshold level, while leaving those steepest descent and conjugate gradient minimization for 5000 steps above the threshold unmodified. In doing so, aMD accelerates the tran- with a harmonic restraint of 25 kcal/mol per Å2 applied on the protein. sition between different metastable states while conserving the under- A second round of minimization was performed with, instead, a reduced lying topology of the potential energy landscape, i.e., minima are restraint of 5 kcal/mol per Å2 on the protein backbone. L. polyedra typi- maintained as minima. Since its publication, the method has been suc- cally occur in temperate and tropical waters, although the sea surface cessful in capturing millisecond timescale events in a number of biolog- water temperature can vary seasonally from just below zero up to ical systems [27–36]. In constant pH aMD (CpHaMD) simulations, the 25∘C. Thermal destruction of scintillons takes place at temperatures sections of classical MD in each cycle are replaced by aMD steps [37]. above 30∘C and complete inactivation at 40∘C[23]. Following the mini- Here, a dual-boosting approach was employed to concurrently en- mization phase, a round of heating was performed to linearly raise the hance the sampling of torsional degrees of freedom and diffusive mo- temperature from 0 K to 300 K. During heating, a Langevin thermostat tions. For a detailed description of how the boost parameters were was used with a collision frequency of 5.0 ps−1, while gradually releas- selected, see the Supplementary Information (SI). Production runs ing the position restraints on the protein backbone: 0–50 K over 100 ps were performed for a duration of 1.5 μs and an extension was run for with a restraint of 2.5 kcal/mol per Å2,50–100 K over 100 ps with a re- pH 6 up to 2 μs. straint of 1.0 kcal/mol per Å2,and100–300 K over 200 ps with restraints switched off. The first part of heating was performed under constant 2.4. Analysis of results volume (NVT ensemble). The final unrestrained heating was performed under constant pressure (NPT ensemble) using the Berendsen barostat. Standard tools were employed for analysing the results, using Heating was followed by 10 ns of equilibration at the target temperature CPPTRAJ [38], however, in some cases additional tools were developed of 300 K to find a good estimate of the box volume, in this way ensuring as deemed necessary. an average pressure close to the standard pressure. For the remainder of the simulations, the NVT ensemble was used as there is a risk with NPT 2.4.1. Atomic fluctuations that a fluctuating net charge, due to altering protonation states during For our simulation data, a best fit of coordinates to the average struc- the course of a constant pH molecular dynamics (CpHMD) simulation, ture was performed, to remove global rotations and translations, and may cause artefacts in the computed pressure, although this is mostly the atomic positional fluctuations were computed. Simulation-derived an issue for smaller unit cells [24]. For a largely incompressible solvent, atomic fluctuations tend to be larger than those derived from experi- such as water, and an equilibrated system, there should be no discern- mental B-factors, in particular in loop regions [39]. ible statistical difference between NVT and NPT. Experimental B-factors and simulated root-mean-square fluctua- The progress of equilibration was monitored by plotting the poten- tions were mapped onto the luciferase domain using VMD [40]. tial energy, density, pressure, temperature, and root-mean-square devi- ation of the protein backbone. Equilibration was considered complete 2.4.2. Protonation states of titratable residues once these values had stabilized. Using the constant pH output files produced by Amber, the fraction of aMD time during which each titratable residue spent protonated dur- 2.2.2. Constant pH molecular dynamics ing the course of the CpHMD simulation was calculated through the In conventional molecular dynamics (MD) simulations, the proton- cphstats program from Amber Tools 18. This was calculated as a run- ation states are predetermined and assumed to be fixed during the sim- ning average over a window of 60 ns. ulation. To allow an observation of the pH-triggered conformational transition from the inactive to active form, the protonation states of ti- 2.4.3. Hydrogen-bond network tratable residues were allowed to vary by using CpHMD simulations, The constant pH method implemented in Amber is based on a similar to the approach used in ref. [11]. The input files for the constant charge-change only model, meaning that protonation states are altered pH simulations were generated using cpinutil, with the starting pro- by changing the partial charges on atoms of titratable residues. Since the tonation states assigned based on the output from H++. Since histidine topology and trajectory file, by necessity, have a hydrogen designated at and cysteine have reference pKa values near the pH range of interest, every possible point of protonation, an accurate hydrogen bond analysis these were allowed to vary protonation state during the simulation. In needs to explicitly take into account the current protonation state of the addition, Swiss-PDBViewer [25] was used to identify buried aspartate, titratable residue. We implemented our own Python script to analyze glutamate, lysine, and tyrosine residues with less than 15% solvent ac- potential hydrogen bonds using several of the current tools in CPPTRAJ. cessible surface area, which were subsequently also set as titratable. The method is described in detail in the SI. The equilibrated structures were used as a starting point for CpHMD simulations at pH 6 and pH 8. In CpHMD, a hybrid MD/MC technique is 2.4.4. Pocket volume prediction and solvent accessibility used to change the protonation states of titratable residues, by changing The luciferin binding site of D3 is thought to be located within the β- the partial charges on the atoms [24]. Here, each cycle started with 100 barrel. In the closed form, however, the interior pocket is neither large steps of MD with an explicit solvent followed by a trial Monte Carlo step enough to accommodate the luciferin ligand, nor does it have access using a Generalised-Born implicit solvent model. After these steps, an to the exterior. In order to validate that the luciferase in our simulations attempt was made to change the protonation state for each titratable adopts an open form at low pH, it is necessary to inspect changes in both residue and the solvent was subsequently restored. If any protonation the pocket volume and mouth opening. 2674 N. Kamerlin et al. / International Journal of Biological Macromolecules 164 (2020) 2671–2680

While buried cavities are straightforward to define in terms of extent changes of the HLH axis relative to the crystal structure (see Figs. S5 and volume, pockets that are open to the solvent are inherently prob- and S6). lematic due to the lack of clear border. Several algorithms have been de- The dominant motion of the HLH at pH 6, as shown in Fig. S6, ap- veloped to identify and characterize protein pockets [41–44]. However, pears to involve the motif tilting away from the β-barrel. The corre- since we need to efficiently process hundreds of thousands of frames in sponding plot for α2 also shows a tilt in a similar direction, but each trajectory to determine both the volume and mouth opening of the significantly weaker, as previously indicated by its lower RMSF values β-barrel pocket, we developed our own grid-based script (available in Fig. 2. The resulting tilting motion is consistent with the distances be- upon request), specifically tailored and optimized for our case. The tween the helices displayed in Fig. S3. method sets an even-spaced grid in the simulation box and then assigns each grid point as being part of the protein, pocket, or the exterior. Open 3.2. Open structures structures are those in which the interior pocket is in direct contact with the exterior. The algorithm is described in more detail in the SI. The To validate the opening of the full length luciferase D3 structure at quality of the results were cross-validated by visualizing the pocket in pH 6, we estimated the volume of the binding pocket and solvent acces- several snapshots along the trajectory and by using the CASTp 3.0 web sibility, see Fig. 3, using our own script. server [45]. There appear to be a few open structures at pH 8 that are relatively short-lived. Since we are using an enhanced sampling method, such in- 3. Results stances may be statistically rare in reality, but these structures still war- rant more analysis. We performed clustering of these open frames to 3.1. Dynamics of the pH-regulatory 3-helix bundle obtain the most representative structures (see SI, section 7), corre- sponding to 75% of all open frames at pH 8. In these, the 3-helix bundle As an assessment of the atomic displacements, we used the crystal- has not significantly deviated from the crystal structure, however, an lographic B-factors of the luciferase D3 crystal structure to shed some opening to the solvent has formed beneath the bundle near the ridge light on regions of potentially high and low flexibility. The B-factor of the barrel. An open structure is not necessarily synonymous with an values represent the atomic fluctuations of atoms about their average active enzyme. As will be shown next, this pathway to the binding positions. High mobility regions would thus have higher B-factors, how- pocket is quite different from the one explored at pH 6, meaning that ever, the same is true for low structure resolutions [46]. In our case, the these structures would not expose the same residues to the luciferin crystal structure (PDB ID 1VPR) has a reasonably high resolution of and are most likely still inactive. 1.8 Å, meaning that the B-factors are more likely to be a sound reflection Looking instead at pH 6, there is a period during the second half of of the protein flexibility. the simulation where the domain appears to assume a closed conforma- Fig. 2 (a) shows the D3 domain coloured according to the B-factors. tion. A continuation run was performed to confirm that the luciferase The β-barrel, which is expected to contain the active site, is associated returns to an open conformation. with relatively low B-factors, indicating a more rigid and stable struc- Overall, it is evident that the luciferase at pH 6 spends a significantly ture. Interestingly, the top of the helix-loop-helix (HLH) motif of the higher proportion of the simulation in an open conformation, consistent pH-regulatory domain, composed of α5andα6, shows a higher than av- with the experimental activity. A snapshot of such an open structure is erage flexibility. Our simulation results at pH 8 (Fig. 2 (b)) show good shown in Fig. 4, superimposed with the crystal structure, illustrating qualitative agreement, with a Pearson and Spearman's rank correlation how the HLH tilt enables access to the binding site. We note that this coefficient between the experimental B-factors and simulated root- open conformation is different from the one reported by Donnan mean-square fluctuations (RMSF) over all crystal structure residues of et al., in which the 3-helix bundle has tipped over and the HLH is stabi- 0.54 and 0.67, respectively. lized by a structural reorganization of the N-terminal domain containing The mobility of the 3-helix bundle, and in particular the HLH motif, is α2[11]. In our case, the structure of the N-terminus is well preserved notably amplified at the lower pH, in-line with the postulated large- and the HLH motif effectively stays near α2 also in the open form. scale conformational change of this domain. To determine whether some of the fluctuations are due to a deformation of chain segments, the secondary structure propensities of the 3-helix bundle were ana- 3.3. Protonation states of histidine lyzed over time. The α2 helix is structurally stable at both pH, over the entire duration of the simulations, see SI, Fig. S1. The secondary struc- The protonation states of the conserved histidines are shown in tures of α5andα6 are mostly maintained (see Fig. S2), although the ter- Fig. 5. It can be seen that the histidines generally reside in a singly pro- minal residues of α6 become temporarily strained around roughly 1.5 μs tonated state in the alkaline region. However, when the pH is lowered, at pH 6. Based on these results, the HLH motif appears to be moving as a the imidazole side chain becomes positively charged. In addition, rigid body relative to α2. there are two histidine residues located on the lower loop of α6, In short MD simulations lasting 10 ns, Schultz et al. found that mu- HIS1064/1065, which are more readily protonated at pH 6 (see Fig. 6). tant D3 structures, where the four conserved histidines had been re- These results are qualitatively in agreement with those found in ref. placed by alanines, showed an increase in the α2 − α5 distance from [11]. 10 Å to 21 Å [7]. It was proposed that a similar motion at pH 6 for the Considering the remaining titratable residues, all cysteines have a wild type luciferase might lead to the formation of a solvent channel protonated thiol group at either pH. Interestingly, among the buried ti- within the bundle, giving access to the β-barrel. To investigate whether tratable residues, one which shows a distinct pH-dependent behaviour this separation also takes place during our long-timescale simulations, and which has not previously received attention is a lysine residue, the distance between α2 and the respective helices of the HLH motif LYS1089, located at the interface of the β-barrel and the 3-helix bundle. was calculated (see Fig. S3). We do not see the same dramatic increase This residue favours the neutral form at pH 8, but becomes positively in distance as in ref. [7]. Thus, our findings are not necessarily suggestive charged at pH 6 (see Fig. 6) for the entire duration of the simulation. of a channel opening. While the normal pKa value for lysine in water is 10.4, large shifts Interestingly, the most striking feature is that the plots of the dis- have been observed for buried lysine residues, with values as low as tance pairs (Fig. S3) are mirror-images. In other words, an increase in 5.3 [47]. A BLAST [48,49] search (see SI) reveals that this lysine is highly the α2 − α5 distance is correlated with a decrease in the distance be- conserved among bioluminescent dinoflagellate species. An additional tween α2andα6. To be able to understand this correlation, the domain lysine residue at the bottom of the β-barrel, LYS1125, also shows a motion of the HLH motif was visualised by evaluating the angular pH-dependent behaviour, as was previously observed in ref. [11]. No N. Kamerlin et al. / International Journal of Biological Macromolecules 164 (2020) 2671–2680 2675

Fig. 2. Luciferase D3 coloured according to the experimental B-factor values and the simulated RMSF values, respectively. (a) The experimental B-factor values are shown, restricted for clarity between 25 and 50 Å2. Note that the actual lower and upper limits are 1.1 and 53.4 Å2, respectively. (b) The RMSF (in units Å) are shown, measured during simulations at pH 8 and (c) pH 6. The lower- and upper bound have for clarity been set to 1.0 and 8.0, respectively. The RMSF values of the core D3 structure are predominantly below the upper limit, except for the mobile C-terminal residues which show RMSF values as high as 20 Å.

other residue displays significant protonation differences between pH 6 (SER1086 and MET1070) as well as residues on the HLH (GLN1037 and pH 8. and ASP1063). Overall, these results confirm that the conserved histidines play a Our findings indicate that the histidines do form hydrogen bonds central role in the pH-regulated opening mechanism by adopting differ- which could collectively stabilize the closed conformation at pH 8. ent protonation states. Furthermore, the hitherto unexplored LYS1089 Thus, removal of these histidines, e. g., through site-directed mutagene- may similarly have an important function in activating the luciferase. sis to alanine, could facilitate a transition to an open conformation under alkaline conditions. A significant finding, however, is that the interac- 3.4. Hydrogen bonding tions listed above overwhelmingly involve the histidine residues acting as a proton donor or act between main-chain groups, neither of which In the crystal structure, we could identify a number of hydrogen would be disrupted by protonation of the side-chain. Among the few in- bonds involving the histidines, most of which were also noted in ref. teractions that could potentially be affected by a pH drop, HIS899 inter- [7], namely HIS899(HE2)-TYR925(OH) or HIS899(HD1)-VAL1087(O), acts with TYR925 in approximately 46% of the simulation frames and HIS899(H)-ALA1121(O) (missed in ref. [7]), HIS909(HD1)-LEU1050 acts as a proton acceptor only 32% of this time. HIS930 interacts with (O), and HIS930(HE2)-GLN1037(OE1). Among these, only HIS909- GLN1037 9% of the simulation and as a proton acceptor only 35% of LEU1050 and HIS930-GLU1037 connect α2 and the HLH motif, while this time. the interactions of HIS899 may tether the bundle to the β-barrel. If disruption of hydrogen bonds is not responsible for destabilizing The evolution of hydrogen bond interactions as the luciferase ex- the 3-helix bundle at pH 6, the question then arises if new hydrogen plores different metastable states at pH 8 is shown in Fig. S8. HIS899 ini- bonding possibilities with other parts of the protein through proton- tially interacts with a number of β-barrel residues (VAL1087, PHE1090, ation may stabilize a different conformation. The corresponding time and ALA1091), however, these links are eventually lost and the histidine series measured at pH 6 are shown in Fig. S8. The interactions between interacts predominantly with TYR925 (located on α2). HIS909 intermit- HIS899 and the β-barrel are still present in the domain structures indi- tently bonds to various residues located on the loop of the HLH motif, cated to be open (see Fig. 3), while the interactions with TYR925 have notably with GLY1048. The solvated HIS924 does not appear to hydro- been entirely lost. HIS909 has lost all of its hydrogen bonds with the gen bond with any other residues, while HIS930, which points towards HLH motif from the outset. Only HIS930 has gained a few interactions, the core of the bundle, hydrogen bonds with residues on the β-barrel notably with ASN1056 and ASP1061 located on α6, which could be a 2676 N. Kamerlin et al. / International Journal of Biological Macromolecules 164 (2020) 2671–2680 a 1000

950

900

850 )

3 800

750

700 Volume (Å Volume

650

600

550

500 0 0.2 0.4 0.6 0.8 1 1.2 1.4 b Time (µs)

1000

950

900

850 )

3 800

750

700 Volume (Å Volume

650

600

550

500 0 0.5 1 1.5 2 Time (µs)

Fig. 3. Volume of the β-barrel pocket. The volume is measured as a function of aMD time at (a) pH 8 and (b) pH 6. Results are only shown for frames where the pocket volume is >500 Å3 and the pocket mouth opening is >30 Å2.

Fig. 4. Representative open conformation of Luciferase D3. (a) An open structure of full length D3 is shown from the side at pH 6, with the crystal structure shown in gray. (b) The same structure shown from above, with a molecular surface added and the buried pocket of the crystal structure shown in yellow. N. Kamerlin et al. / International Journal of Biological Macromolecules 164 (2020) 2671–2680 2677

Fig. 5. Protonation states of the four conserved histidines. Results are shown both at pH 6 and pH 8, where a value of 1 indicates doubly protonated (positively charged) and 0 indicates singly protonated (neutral) side-chains.

result of the tilting of the HLH. On the whole, it would thus appear that presence of a number of contacts between the C-terminus and the β- hydrogen bonds are lost in the open form rather than gained. barrel, as well as the C- and N-termini, which together help stabilize the tertiary structure. However, a number of contacts between C- terminal residues and the barrel that are preserved at both pH for the 3.5. Truncated mutants full length D3 are entirely lost in the C50 mutant. These are likely signif- icant for the structural stability of the barrel. The roles of α2 and the C-terminus were separately investigated by simulating truncated structures of N65 and C50, based on the full length D3 structure. 4. Discussion In the N-terminal truncated mutant (N65), in the absence of α2and its associated hydrogen bonds and hydrophobic interactions, the Our results suggest that the closed–open conformational transition α5 − α6 motif is no longer maintained upright, as indicated by the of luciferase at low pH takes place through a tilt of the HLH motif time evolution of the angle and by the snapshots in Fig. S9. Removal of away from the β-barrel. Rather than a solvent channel opening or a re- the N-terminus thus has a destabilizing effect on the HLH, which could organization of the N-terminal domain, however, the HLH motif effec- explain the relative increase in luminescence activity at pH 8 observed tively stays near α2 as part of a 3-helix bundle also in the open in experiments. luciferase conformation. We note that both α5 and α6 are connected Interestingly, at either pH, the HLH motif sinks into the β-barrel and to the β-barrel through loops containing glycine residues (GLY1035- displays unique interactions with various residues in the interior of the GLY1036 and GLY1068-GLY1069, respectively). Being the smallest barrel. Specifically, GLN1066 interacts extensively with GLN1087, amino acid with minimal steric hindrance to bond rotations, glycine is SER1096, and THR1089 by forming both side-chain and main-chain hy- likely to be found in protein hinges [50]. In this case, the glycine residues drogen bonds, in this way restraining the motif above the β-barrel. could serve as a putative hinge region, allowing domain motions. This These interactions are entirely absent for the full-length structure. A conformational change modulates the enzymatic activity by exposing volume analysis confirmed that the N65 mutant is more closed under the active site to the solvent at low pH (Fig. 4), enabling the luciferin acidic conditions compared to the full length structure (see Fig. S7), al- to bind. though we do not find an increase in the frequency of open frames at In resolving the crystal structure of the D3 domain, Schultz et al. pro- pH 8. posed that the conserved histidines in their unprotonated form take In C50 at pH 8, the β-barrel remains fairly stable and α2 stays near part in a network of hydrogen bonds which stabilizes the 3-helix bundle the HLH motif during the simulation. In contrast to the full length and [7]. Disruption of such bonds, e. g., upon protonation at low pH or N65 mutant structures, however, the β-barrel domain of C50 displays through substitution of histidines with alanine, could therefore enable substantial atomic fluctuations at pH 6 (see Fig. 7)aswellaslossofsec- the helices to move apart and allow access to the active site. An issue ondary structure, notably of β8. Interestingly, α2 in the absence of the C- with this hypothesis is that it does not account for the loss activity at terminus simply slips down as the histidines become protonated (see pH 6 of HIS→ALA mutant D3, which show a 75% reduction in lumines- Fig. S10). cence activity compared to wild type D3 [5]. Among the four conserved In order to understand the cause of the conformational instability of histidines, only specific substitution of HIS909 resulted in a higher activ- C50, residue-residue contact maps were employed to probe changes in ity at both low and high pH [5], consistent with the fact that in our sim- intramolecular interactions over time. At pH 8, these maps display the ulations, this residue forms a hydrogen bond between the HLH motif 2678 N. Kamerlin et al. / International Journal of Biological Macromolecules 164 (2020) 2671–2680

1 Histidines can assume many different roles in protein architectures, pH6 through different types of molecular interactions with other amino 0.8 pH8 acids [51]. Notably, electrostatic interactions are often critically impor- 0.6 tant for the function and behaviour of proteins [52,53], and histidines

HA play an important role by their pH-dependent ability to exist as neutral f 0.4 or charged conformers. Among the four conserved histidines, mutation of HIS899 and HIS930 showed the greatest impact on the activity at 0.2 pH 6, with a reduction of almost 50% for HIS930 [5]. We note that the 0 core of the 3-helix bundle shelters hydrophobic residues, such as ala- 0 0.5 1 1.5 2 nine, leucine, and phenylalanine. Meanwhile, the mouth opening of Time (µs) the barrel pocket hosts buried ionizable amino acids, such as HIS899, HIS930, HIS1064 (also highly conserved), and LYS1089, which could be- 1 pH6 come positively charged depending on pH and micro-environment. 0.8 pH8 Fig. 8 shows the electrostatic surface potential of D3 for protonated his- tidines and lysine, during the closed-open transition. This shows a 0.6 strong localization of positive charges at the interface of the barrel and

HA the 3-helix bundle. f 0.4 The introduction of positively charged side-chains in the interior of a hydrophobic core could have a destabilizing effect. Specifically, electro- 0.2 static repulsion between these His-cation pairs could create an incen- 0 tive for the helices to move apart. A manifestation of the electrostatic 0 0.5 1 1.5 2 repulsion can be found in the distance between HIS930 and LYS1089 Time (µs) (see Fig. S4), illustrating that these residues in their charged form prefer to be farther apart. Based on our observations, we propose that the tran- 1 sition of luciferase from inactive to active form is caused by electrostatic interactions, driven by the side-chain protonation of histidines at low 0.8 pH. A similar mechanism has been observed, for example, in viral fusion 0.6 proteins, where side-chain protonation and subsequent repulsive inter-

HA actions between histidine-cation pairs located at the interface of do- f 0.4 LYS1098mains induces a conformational HIS1065 change during HIS1064 membrane fusion [54,55]. 0.2 pH6 Considering the role of the N- and C-terminus, it has been observed pH8 0 in deletion experiments that N-terminal truncated mutants which lack 0 500 1000 1500 2000 α2 display a different pH-profile and luminescence activity than the Time (µs) full length D3 [5,10]. These mutants not only show an increase in ratio of activity at pH 8 relative to pH 6, meaning that the pH-regulatory mechanism has somehow been disturbed, but also a decreased activity Fig. 6. Protonation states of HIS1064, HIS1065, and LYS1089. The results are shown for the full length luciferase D3 at pH 6 and pH 8, where 1 indicates doubly protonated (positively at pH 6, down to approximately 20% of the full length structure. This charged) and 0 indicates singly protonated (neutral). could be rationalised by our proposed mechanism in which the absence of α2 removes the stabilizing effect which maintains the 3-helix bundle at pH 8, while reducing the electrostatic repulsion and thus the incen- tive to open at pH 6. Additionally, our simulations suggest that this ef- and α2. Since the histidines in our simulations predominantly act as fect is enhanced by new interaction possibilities of the HLH motif to proton donors, protonation of these residues at pH 6 is unlikely to be interior β-barrel residues, which tether the motif above the binding the destabilizing effect promoting the conformational change. pocket and reduce its mobility. To confirm this effect, we compared

Fig. 7. C50 colour coded according to the RMSF. The values are shown at (a) pH 8 and (b) pH 6. N. Kamerlin et al. / International Journal of Biological Macromolecules 164 (2020) 2671–2680 2679

residing within a hydrophobic core, along with protonation of a highly conserved lysine residue, likely trigger the conformational change through electrostatic repulsion between histidine-cation pairs. Removal of these histidines would facilitate the opening of the pocket at pH 8, while simultaneously reducing the electrostatic incentive to do so under acidic conditions, in agreement with experimental observations. In addition, for the N-terminal truncated mutants, we could show that the decreased luminescence activity at low pH likely also originates from the new hydrogen bonding possibilities of the helix-loop-helix motif with residues within the β-barrel. Finally, truncation of the C- terminus was found to destabilize the β-barrel at low pH. The structural deterioration of the β-strands and thus alteration of the shape of the binding site could be the culprit for the lack of activity observed in ex- periments. In the future, the availability of an open enzyme structure Fig. 8. Molecular surface of D3 coloured by the electrostatic potential. The values range from −10 kcal/(mol⋅e) (red) to 10 kcal/(mol⋅e) (blue). also makes it possible to perform ligand docking and to study the mech- anism of the bioluminescence reaction in dinoflagellate luciferase. the volume of N65 with the full length structure at pH 6 (Fig. S7). How- Funding ever, we note that our simulation does not show a significant increase in the frequency of pocket opening at pH 8, which would be suggestive of a N.K. acknowledges funding from eSSENCE - The e-Science Collabora- higher activity as observed in experiments. One possible explanation tion (Uppsala-Lund-Umeå, Sweden). M.D. acknowledges support from could be statistics. In the full length D3 structure, α2 restricts the the Foundation Olle Engkvist Byggmastare (183-0403). R.L. acknowl- range of motion of the HLH motif. In N65, this motif tilts in a different edges funding from the Swedish Research Council (grant no. 2016- direction at pH 6 and pH 8 (Fig. S9) and essentially remains in this con- 03398). figuration for the duration of the simulation. The lack of support from the rest of the protein means that this domain in N65 is quite flexible Declaration of competing interest and it is possible that the direction of the tilt in these mutants may be random. In this case, our simulation merely sampled one possibility The authors declare that they have no competing interests. for each pH and another simulation at pH 8 might explore a different trajectory in which the tilt would be more similar to the corresponding Acknowledgements one at low pH. This new tilt could indeed result in more open N65 struc- tures than the full length structure at pH 8 (cf. Fig. S7). To confirm this, it The computations were enabled by resources provided by the Swed- would be necessary to simulate a number of independent trajectories to ish National Infrastructure for Computing (SNIC) at HPC2N partially study the propensity of the HLH motif of N65. funded by the Swedish Research Council through grant agreement no. Surprisingly, it has been experimentally observed that truncating 2016-07213. the C-terminus of D3 results in a drastic loss of activity to below 1% of the wild type D3, although this region is not part of the pH-regulatory Appendix A. Supplementary data domain or active site [10]. Our results suggest that this loss of activity may be related to the loss of support provided by the C-terminus, Supplementary data to this article can be found online at https://doi. which results in a destabilization and deformation of the β-strands, spe- org/10.1016/j.ijbiomac.2020.08.071. cifically β8. We note that the up-and-down strands of the β-barrel are structurally maintained by hydrogen-bonding between neighbouring References strands, with the exception of β8andβ9 that are too far apart to make [1] S. Martini, S.H.D. Haddock, Quantification of bioluminescence from the surface to any inter-strand hydrogen bonds. Instead, this region of the barrel ap- the deep sea demonstrates its predominance as an ecological trait, Sci. Rep. 7 pears to be stabilized by the C-terminus spanning across the gap, see (2017), 45750https://doi.org/10.1038/srep45750. Fig. S10. At pH 6, this terminus appears to have moved away and is [2] M.P. Davis, J.S. Sparks, W.L. Smith, Repeated and widespread evolution of biolumi- fi interacting with α2. Our results suggest that the loss of function at nescence in marine shes, PLoS One 11 (2016), e0155154https://doi.org/10.1371/ journal.pone.0155154. pH 6 for C-terminal truncated mutants is an outcome of an altered [3] M.-T. Nicolas, G. Nicolas, C.H. Johnson, J.-M.B.J.W. Hastings, Characterization of the shape of the binding pocket, possibly originating from the loss of con- bioluminescent organelles in Gonyaulax polyedra (dinoflagellates) after fast- tacts between the C-terminus and the β-barrel. This structural instabil- freeze fixation and antiluciferase immunogold staining, J. Cell Biol. 105 (1987) – ity and deformation of the barrel could make luciferin binding 725 735, https://doi.org/10.1083/jcb.105.2.723. [4] N. Krieger, J.W. Hastings, Bioluminescence: pH activity profiles of related luciferase unfavourable or perhaps make important catalytic residues less accessi- fractions, Science 161 (1968) 586–589, https://doi.org/10.1126/science.161.3841. ble, effectively inactivating the luciferase. 586. [5] L. Li, L. Liyun, R. Hong, D. Robertson, J. Hastings, N-terminal intramolecularly con- served histidines of three domains in Gonylaulax luciferase are responsible for 5. Conclusion loss of activity in the alkaline region, Biochemistry 40 (2001) 1844–1849, https:// doi.org/10.1021/bi002094v. [6] L. Li, R. Hong, J. Hastings, Three functional luciferase domains in a single polypeptide We showed that the closed–open conformational transition of lucif- chain, Proc. Natl. Acad. Sci. U. S. A. 94 (1997) 8954–8958, https://doi.org/10.1073/ erase D3 at low pH involves a tilt of the helix-loop-helix motif relative to pnas.94.17.8954. α2 and away from the barrel, with glycine residues likely acting as [7] L.W. Schultz, L. Liu, M. Cegielski, J.W. Hastings, Crystal structure of a pH-regulated luciferase catalyzing the bioluminescent oxidation of an open , PNAS hinges. Our analysis suggested that the conserved histidines partake in 102 (2005) 1378–1383, https://doi.org/10.1073/pnas.0409335102. hydrogen bonding which stabilize the 3-helix bundle and maintain [8] M. Fogel, J. Hastings, A substrate-binding protein in the Gonyaulax bioluminescence the closed conformation in an alkaline environment. However, contrary reaction, Arch. Biochem. Biophys. 142 (1971) 310–321, https://doi.org/10.1016/ to what has been previously thought, disruption of the hydrogen bond 0003-9861(71)90289-X. [9] M. Fogel, J. Hastings, Bioluminescence: mechanism and mode of control of Scintillon network at low pH is likely not the cause but the consequence of the activity, Proc. Natl. Acad. Sci. U. S. A. 69 (1972) 690–693, https://doi.org/10.1073/ conformational change. Instead, side-chain protonation of histidines pnas.69.3.690. 2680 N. Kamerlin et al. / International Journal of Biological Macromolecules 164 (2020) 2671–2680

[10] C. Suzuki. Ogoh, C. Wu, Y. Ohmiya, C-terminal region of the active domain enhances [33] P. Markwick, C. Cervantes, B. Abel, E. Komives, M. Blackledge, J. McCammon, En- enzymatic activity in dinoflagellate luciferase, Photochem. Photobiol. Sci. 7 (2008) hanced conformational space sampling improves the prediction of chemical shifts 208–211, https://doi.org/10.1039/B713157G. in proteins, J. Am. Chem. Soc. 132 (2010) 1220–1221, https://doi.org/10.1021/ [11] P.H. Donnan, P.D. Ngo, S.O. Mansoorabadi, Constant pH accelerated molecular dy- ja9093692. namics investigation of the pH regulation mechanism of Dinoflagellate luciferase, [34] D. Bucher, B.J. Grant, P.R. Markwick, J.A. McCammon, Accessing a hidden conforma- Biochemistry 57 (2018) 295–299, https://doi.org/10.1021/acs.biochem.7b00873. tion of the maltose binding protein using accelerated molecular dynamics, PLoS [12] D. Case, I. Ben-Shalom, S. Brozell, D. Cerutti, T.C. III, V. Cruzeiro, T. Darden, R. Duke, D. Comput. Biol. 7 (2011), e1002034https://doi.org/10.1371/journal.pcbi.1002034. Ghoreishi, M. Gilson, H. Gohlke, A. Goetz, D. Greene, R. Harris, N. Homeyer, Y. Huang, [35] P. Markwick, J. McCammon, Studying functional dynamics in bio-molecules using S. Izadi, A. Kovalenko, T. Kurtzman, T. Lee, S. LeGrand, P. Li, C. Lin, J. Liu, T. Luchko, R. accelerated molecular dynamics, Phys. Chem. Chem. Phys. 13 (2011) Luo, D. Mermelstein, K. Merz, Y. Miao, G. Monard, C. Nguyen, H. Nguyen, I. Omelyan, 20053–20065, https://doi.org/10.1039/c1cp22100k. A. Onufriev, F. Pan, R. Qi, D. Roe, A. Roitberg, C. Sagui, S. Schott-Verdugo, J. Shen, C. [36] L. Pierce, R. Salomon-Ferrer, C. de Oliveira, J. McCammon, R. Walker, Routine access Simmerling, J. Smith, R. Salomon Ferrer, J. Swails, R. Walker, J. Wang, H. Wei, R. to millisecond time scale events with accelerated molecular dynamics, J. Chem. The- Wolf, X. Wu, L. Xiao, D. York, P. Kollman, AMBER 2018, 2018. ory Comput. 8 (2012) 2997–3002, https://doi.org/10.1021/ct300284c. [13] E. F. Pettersen, T. D. Goddard, C. C. Huang, G. S. Couch, D. M. Greenblatt, E. C. Meng, T. [37] S. Williams, C. de Oliveira, J. McCammon, Coupling constant pH molecular dynamics E. Ferrin, UCSF chimera—a visualization system for exploratory research and analy- with accelerated molecular dynamics, J. Chem. Theory Comput. 6 (2010) 560–568, sis, J. Comput. Chem. 25 (13) (2004) 1605–1612, doi:https://doi.org/10.1002/jcc. https://doi.org/10.1021/ct9005294. 20084, URL https://onlinelibrary.wiley.com/doi/abs/10.1002/jcc.20084. [38] D. Roe, T.C. III, PTRAJ and CPPTRAJ: software for processing and analysis of molecular [14] A. Roy, A. Kucukural, Y. Zhang, I-TASSER: a unified platform for automated protein dynamics trajectory data, Theory Comput 9 (2013) 3084–3095, https://doi.org/10. structure and function prediction, Nat. Protoc. 5 (2010) 725–738, https://doi.org/ 1021/ct400341p. 10.1038/nprot.2010.5. [39] L. Meinhold, J.C. Smith, Fluctuations and correlations in crystalline protein dynam- [15] J. Yang, R. Yan, A. Roy, D. Xu, J. Poisson, Y. Zhang, The I-TASSER suite: protein struc- ics: a simulation analysis of staphylococcal nuclease, Biophys. J. 88 (2005) ture and function prediction, Nat. Methods 12 (2010) 7–8, https://doi.org/10.1038/ 2554–2563, https://doi.org/10.1529/biophysj.104.056101. nmeth.3213. [40] W. Humphrey, A. Dalke, K. Schulten, VMD-visual molecular dynamics, J. Molec. – [16] Y. Zhang, I-TASSER server for protein 3D structure prediction, BMC Bioinformatics 9 Graphics 14 (1996) 33 38, https://doi.org/10.1016/0263-7855(96)00018-5. (40), doi:https://doi.org/10.1186/1471-2105-9-40. [41] J. Wagner, J. Sørensen, N. Hensley, C. Wong, C. Zhu, T. Perison, R. Amaro, POVME 3.0: fl [17] L.L.C. Schrödinger, The PyMOL Molecular Graphics System, Version 2.2, 2018. software for mapping binding pocket exibility, J. Chem. Theory Comput. 13 (2017) – [18] R. Anandakrishnan, B. Aguilar, A.V. Onufriev, H++ 3.0: automating pK prediction 4584 4592, https://doi.org/10.1021/acs.jctc.7b00500. and the preparation of biomolecular structures for atomistic molecular modeling [42] P. Schmidtke, V. Le Guilloux, J. Maupetit, P. Tufféry, fpocket: online tools for protein – and simulation, Nucleic Acids Res. 40 (W1) (2012) W537–W541, https://doi.org/ ensemble pocket detection and tracking, Nucleic Acids Res. 38 (2010) W582 W589, 10.1093/nar/gks375. ISSN 0305-1048, doi:https://doi.org/10.1093/nar/gkq383, URL doi:10.1093/nar/ gkq383. [19] J. Myers, G. Grothaus, S. Narayanan, A. Onufriev, A simple clustering algorithm can [43] D.B. Kokh, S. Richter, S. Henrich, P. Czodrowski, F. Rippmann, R.C. Wade, TRAPP: a be accurate enough for use in calculations of pKs in macromolecules, Proteins 63 tool for analysis of transient binding pockets in proteins, J. Chem. Inf. Model. 53 (2006) 928–938, https://doi.org/10.1002/prot.20922. (5) (2013) 1235–1252, https://doi.org/10.1021/ci4000294, URL 10.1021/ci4000294. [20] J. Gordon, J. Myers, T. Folta, V. Shoja, L. Heath, A. Onufriev, H++: a server for esti- [44] T. Paramo, A. East, D. Garzón, M.B. Ulmschneider, P.J. Bond, Efficient characterization mating pKas and adding missing to macromolecules, Nucleic Acids Res. of protein cavities within molecular simulation trajectories: trj_cavity, J. Chem. The- 33 (2005) W368–W371, https://doi.org/10.1093/nar/gki464. ory Comput. 10 (5) (2014) 2151–2164, https://doi.org/10.1021/ct401098b, URL 10. [21] J.A. Maier, C. Martinez, K. Kasavajhala, L. Wickstrom, K.E. Hauser, C. Simmerling, 1021/ct401098b. ff14SB: improving the accuracy of protein side chain and backbone parameters [45] W. Tian, C. Chen, X. Lei, J. Zhao, J. Liang, CASTp 3.0: computed atlas of surface topog- from ff99SB, J. Chem. Theory Comput. 11 (2015) 3696–3713, https://doi.org/10. raphy of proteins, Nucleic Acids Res. 46 (2018) W363–W367, https://doi.org/10. 1021/acs.jctc.5b00255. 1093/nar/gky473. [22] J. Hastings, V. Bode, Light and Life, The Johns Hopkins University Press, Baltimore, [46] Z. Sun, Q. Liu, G. Qu, Y. Feng, M. Reetz, Utility of B-factors in protein science: 1961. interpreting rigidity, flexibility, and internal motion and engineering thermostabil- [23] R. Desa, J.W. Hastings, The characterization of Scintillons bioluminescent particles ity, Chem. Rev. 119 (2009) 1626–1665, https://doi.org/10.1021/acs.chemrev. from the marine dinoflagellate, Gonyaulax polyedra, The Journal of General Physiol- 8b00290. ogy 51 (1968) 105–122, https://doi.org/10.1085/jgp.51.1.105. [47] D.G. Isom, C.A. Castañeda, B.R. Cannon, B.G.-M. E., Large shifts in pKa values of lysine [24] J. Swails, D. York, A. Roitberg, Constant pH replica exchange molecular dynamics in residues buried inside a protein, PNAS 108 (2011) 5260–5265, https://doi.org/10. explicit solvent using discrete protonation states: implementation, testing, and val- 1073/pnas.1010750108. – idation, J. Chem. Theory Comput. 10 (2014) 1341 1352, https://doi.org/10.1021/ [48] S.F. Altschul, T.L. Madden, A.A. Schäffer, J. Zhang, Z. Zhang, W. Miller, D.J. Lipman, ct401042b. Gapped BLAST and PSI-BLAST: a new generation of protein database search pro- [25] N. Guex, M. Peitsch, SWISS-MODEL and the Swiss-PdbViewer: an environment for gram, Nucleic Acids Res. 25 (1997)https://doi.org/10.1093/nar/25.17.3389 – comparative protein modeling, Electrophoresis 18 (1997) 2714 2723, https://doi. 3389–340. org/10.1002/elps.1150181505. [49] S.F. Altschul, J.C. Wootton, E.M. Gertz, R. Agarwala, A. Morgulis, A.A. Schäffer, Y.-K. [26] D. Hamelberg, J. Mongan, J. McCammon, Accelerated molecular dynamics: a prom- Yu, Protein database searches using compositionally adjusted substitution matrices, fi ising and ef cient simulation method for biomolecules, J. Chem. Phys. 120 (24) FEBS J. 272 (2005) 5101–5109, https://doi.org/10.1111/j.1742-4658.2005.04945.x. – (2004) 11919 11929, https://doi.org/10.1063/1.1755656. [50] S.C. Flores, L.J. Lu, J. Yang, N. Carriero, M.B. Gerstein, Hinge Atlas: relating protein se- [27] D. Hamelberg, J. McCammon, Fast Peptidyl cis-trans isomerization within the flexi- quence to sites of structural flexibility, BMC Bioinformatics 8 (2007) 167, https:// ble Gly-rich flaps of HIV-1 protease, J. Am. Chem. Soc. 127 (2005) 13778–13779, doi.org/10.1186/1471-2105-8-167. https://doi.org/10.1021/ja054338a. [51] S.-M. Liao, Q.-S. Du, J.-Z. Meng, Z.-W. Pang, R.-B. Huang, The multiple roles of histi- [28] P. Markwick, G. Bouvignies, M. Blackledge, Exploring multiple timescale motions in dine in protein interactions, Chem. Cent. J. 7 (2013)https://doi.org/10.1186/1752- protein GB3 using accelerated molecular dynamics and NMR spectroscopy, J. Am. 153X-7-44. Chem. Soc. 129 (2007) 4724–4730, https://doi.org/10.1021/ja0687668. [52] M. Perutz, Electrostatic effects in proteins, Science 201 (1978) 1187–1191, https:// [29] D. Hamelberg, C. de Oliveira, J. McCammon, Sampling of slow diffusive conforma- doi.org/10.1126/science.694508. tional transitions with accelerated molecular dynamics, J. Chem. Phys. 127 (2007), [53] A. Warshel, Electrostatic basis of structure-function correlation in proteins, Acc. 155102https://doi.org/10.1063/1.2789432. Chem. Res. 14 (1981) 284–290, https://doi.org/10.1021/ar00069a004. [30] D. Hamelberg, J.A. McCammon, Mechanistic insight into the role of transition-state [54] J.S. Harrison, C.D. Higgins, M.J. O’Meara, J.F. Koellhoffer, B.A. Kuhlman, J.R. Lai, Role of stabilization in cyclophilin a, J. Am. Chem. Soc. 131 (2009) 147–152, https://doi. electrostatic repulsion in controlling pH-dependent conformational changes of viral org/10.1021/ja806146g. fusion proteins, Structure 21 (2013) 1085–1096, https://doi.org/10.1016/j.str.2013. [31] P. Markwick, G. Bouvignies, L. Salmon, J. McCammon, M. Nilges, M. Blackledge, To- 05.009. ward a unified representation of protein structural dynamics in solution, J. Am. [55] H.-W. Chang, C.-H. Yang, Y.-C. Luo, B.-G. Su, H.-Y. Cheng, S.-Y. Tung, K.J.D. Carillo, Y.- Chem. Soc. 131 (2009) 16968–16975, https://doi.org/10.1021/ja907476w. T. Liao, D.-L.M. Tzou, H.-C. Wang, W. Chang, Vaccinia viral A26 protein is a fusion [32] B. Grant, A. Gorfe, J. McCammon, Ras conformational switching: simulating suppressor of mature virus and triggers membrane fusion through conformational nucleotide-dependent conformational transitions with accelerated molecular dy- change at low pH, PLoS Pathog. 15 (2019) e1007826, https://doi.org/10.1371/ namics, PLoS Comput. Biol. 5 (2009), e1000325https://doi.org/10.1371/journal. journal.ppat.1007826. pcbi.1000325.