GS-SMD server for steered molecular dynamics of peptide substrates in the active site of the γ-secretase complex

Abstract Despite recent advances in research, the mechanism of Alzheimer's disease is not fully understood yet. Understanding the process of cleavage and then trimming of peptide substrates, can help selectively block γ-secretase (GS) to stop overproduction of the amyloidogenic products. Our GS-SMD server (https://gs-smd.biomodellab.eu/) allows cleaving and unfolding of all currently known GS substrates (more than 170 peptide substrates). The substrate structure is obtained by threading of the substrate sequence into the known structure of GS complex. The simulations are performed in an implicit water-membrane environment so they are performed rather quickly, 2–6 h per job, depending on the mode of calculations (part of GS complex or the whole structure). It is also possible to introduce mutations to the substrate and GS and pull any part of the substrate in any direction using the steered molecular dynamics (SMD) simulations with constant velocity. The obtained trajectories are visualized and analyzed in the interactive way. One can also compare multiple simulations using the interaction frequency analysis. GS-SMD server can be useful for revealing mechanisms of substrate unfolding and role of mutations in this process.


INTRODUCTION
The ␥ -secretase (GS), which is a membrane protease complex, is well known for its role in Alzheimer's disease (AD), as it produces the pathogenic amyloid ␤ (A ␤) peptide. AD is the leading cause of dementia and is characterized by a presence of amyloid plaques, composed mostly of A ␤ peptides, in the brains of AD patients. The extracellular presence of A ␤ plaques as well as soluble oligomeric forms of A ␤ ar e consider ed the major cause of AD ( 1 , 2 ). Howe v er, it has been also demonstrated that GS plays a role in embryonic de v elopment (via Notch signaling ( 3 )), adult tissue homeostasis, signal transduction and protein degradation ( 4 ). Among these substrates, the amyloid precursor protein (APP) has been the most studied since its cleavage and then trimming by GS leads to generation of aggregation-prone A ␤ peptides. To explore the potential of GS as a therapeutic target, with the aim to de v elop inhibitors or modulators of GS activity, the studies on substrate-based peptides and In total, GS cleaves over 170 single-pass membrane proteins of type-I (with its N-terminus on the extracellular side of the membrane). Comparing the transmembrane (TM) domain sequences of the substrates one can find interesting dependencies. Dendrogram of their TM sequences is divided into two parts (Figure 2 A). Analyzing them separately one can find that the substrates from upper part of dendro gram mostl y have valine residues in their TM domain (Figure 2 B), while the substrates from lower part of dendro gram predominantl y have l ysine r esidues (Figur e 2 C). Such comparisons indica te tha t the interactions between the substrates and GS are different for different substrates, TM domain is not only an anchor, and one can effecti v ely study its properties and interactions by unfolding the substrate in the acti v e site of GS. The interacti v e dendrogram in GS-SMD server allows selecting the substrate and includes links to the protein sequence UniProt database ( 21 ) and the whole length protein structure Al-phaFold database ( 22 , 23 ).
Spontaneous unfolding of GS substrates in the acti v e site takes a very long time since the enzymatic activity of GS is extr emely low. Ther efor e, in our server we employ the steered molecular dynamics (SMD) simulations to unfold the cleaved C-terminal part of the substrate. The method of protein unfolding, known as a single molecule force spectroscopy, is widely used (24)(25)(26) to re v eal the internal forces responsible for protein stability and for studying dynamics of protein folding. Here, we employ the process of unfolding to mimic the natural e v ent of substrate unfolding in the acti v e site of GS during preparation for the next cut. In GS-SMD server it is also possible to combine unfolding e v ents using differ ent pulling dir ections, velocities and spring constants in subsequent SMD simulations, by taking the structure of the GS-substrate comple x deri v ed from the previous simulation to the next one. One can also run MD simulation without pulling on the server to relax the extended / unfolded structure of the substrate. Knowledge of unfolding e v ents and interactions that can be obtained by our server can facilitate development of substrate-specific inhibitors or modulators.
The GS-SMD server can be useful for a broad range of scientific w ork ers including molecular modelers and experimentalists, as well as students. The server is friendly to be used by inexperienced people, and the e xtensi v e tutorial e xplains the procedures and analyses in detail. According to our knowledge there are no similar w e b servers for conducting SMD simulations for proteins or w e b servers for molecular dynamics of GS.

Web interface
The w e b interface is built in Bootstrap 4.4 while the visualization of structures is performed by NGL Viewer ( 27 , 28 ). Circular plots of molecular interactions are created in flareplot (trajectory data is processed by time-flare script) and the interacti v e linear plots are implemented in Google Charts. Visualization of a pulling direction by an arrow is implemented in the SMD form on the MD&mut input page and also on Analysis page. The progress bar showing a status of the job and the remaining simulation time is displayed after starting the SMD / MD simulation.

Backend of the server
The GS-SMD server is processing jobs using the Celery message passing queue. Web interface, database queries and server logic are implemented with the Django framework. The SMD simulations are performed in an implicit water and membrane environment using the IMM1 ( 29 ) method implemented by us in NAMD ( 30 ) v.2.14 running on CUDA-enabled GPUs server.

The implicit environments
The SMD / MD simulations performed at GS-SMD server are employing the implicit heterogeneous environment which includes both solvent and membrane media. The approach is based on the implicit membrane methodology IMM1 ( 29 ) being the extension of EEF1 method ( 31 ). IMM1 is parameterized for CHARMM19 force field combined with a Gaussian model for the solvation free energy. CHARMM19 is the united-atom force field without alipha tic hydrogen a toms w hich additionall y diminishes the number of atoms in the system. The solvation model calculates how the neighboring atoms affect the solvation energy of a gi v en atom by excluding solvent from the surrounding space. The membrane is implemented as a slab, parallel to the xy plane centered at z = 0, with a smooth transition of solvation parameters for each atom at the interface using the continuous switching function. The transition region between the hydrophobic core and the water environment is about 0.6 nm wide, which is in agreement with experimental data of lipid bilayers. All the necessary details of the underlying methodology of implicit environments are described in our earlier paper on GPCRsignal server ( 32 ) which was constructed for studying dynamical interactions between GPCRs (G protein-coupled receptors) and their effector proteins (G proteins and arrestins). In SMD / MD simulations, in the implicit water-membrane environments, ther e ar e no explicit water molecules and no lipids. It makes simulations faster but there are some drawbacks, for instance, there is no possibility of bridging hydrogen bonds by water molecules, howe v er, a formation of direct hydrogen bonds is feasible.

Curation of GS complex and threading
For SMD / MD simulations we have employed cryo-EM structure of GS with APP substrate (PDB id: 6IYC) ( 9 ). The N-terminus of the substrate is not visible except for residues 1-6 interacting with NCT. The missing unfolded fiv e residues 7-11 were reconstructed. The side chains that were onl y partiall y visib le in the GS comple x wer e also r e-constructed. The hydrogen atoms were added to the whole complex at pH 7 and the hydrogen bond network was optimized. The mutation of one of the catalytic residues, D385A, introduced to pre v ent substrate cleav age b y GS, was re v ersed. Then, the r estor ed r esidue D385 was protonated to create the proper catalytic environment. Only one catalytic residue was protonated to comply with the cleavage mechanism of the aspartyl protease. The lacking in cryo-EM structure N-terminal part of PS-1 (residues 1-72) as well as the long loop between TM6 and TM7 helices (residues 292-375) of PS-1 were not reconstructed as they are not necessary for studying unfolding e v ents in the acti v e site of GS. The GS-APP structure without a loop between TM6 and TM7 helices of PS-1 was recently used for GaMD (Gaussian accelerated molecular dynamics) simulations. That structure proved to be stable during the simulation of 2000 ns ( 11 ).
The currently available experimental structures of GS (PDB id: 6IYC, 6IDF, 6LQG, 6LR4, 7C9I, 7D8X) are nearly identical including the same positions of side chains of GS subunits. Ther efor e, we decided to use only one structure of GS in our server. If new, and different, experimental structures of GS will be available, they will be included in the server as input structures. We do not allow to use input structures generated by users, as there is no verification if the input structure is really GS, and also the threading procedure for such structure would be problematic. In order to study unfolding of APP after the cleavage, we cut out the residues behind the cleavage site, receiving the substrate consistent with the A ␤ 49 fragment, howe v er, its Nterminus is shortened, because 33 residues of the substrate, and not 49, are visible in the cryo-EM structure 6IYC. As it was found by Bhattarai et al. ( 11 ), AICD (APP intracellular domain) is necessary for initial pulling the C-terminus Nucleic Acids Research, 2023, Vol. 51, Web Server issue W255 Figure 3. The scheme of SMD simulations with constant velocity. In the GS-SMD server the C ␣ atom of the C-terminal residue of the substrate is defined as SMD atom. This atom is pulled by a virtual atom attached to the SMD atom by a virtual spring with a spring constant k.
of the substrate to set it up for the next cleavage step --in our server this is achieved by direct pulling of C-terminus of the substrate. To study other A ␤ fragments of APP and also other GS substrates we employ the threading procedure with the sequence of a gi v en substrate applied to the structure of GS-APP (PDB id:6IYC). Threading is conducted in open-source PyMOL by mutating all residues of initial APP substrate to the selected substrate / mutant --the sequences are aligned to impose their cleavage sites at C-terminal ends. During mutagenesis PyMOL selects rotamers to maximally avoid steric clashes. Other short contacts ar e r emoved during the optimization procedure: 3 × (1000 steps of conjugated gradient minimization of energy and 2 ps MD simulation).

SMD and MD simulations
For SMD / MD simulations in the implicit environments the Langevin dynamics is used with a damping constant of 40 ps −1 and with 2 fs time step whereas all bond lengths are constrained using SHAKE ( 33 ) algorithm. The simula tions run a t tempera ture 298 K with nonbonded interactions cutoff at 14 Å and switching at 12 Å . Originally, the IMM1 method was implemented in CHARMM program ( 34 ), howe v er, we ported this method to NAMD ( 30 ) to take advantage of GPGPU and parallelization. To unfold the substrate in the binding site of GS a force is applied to C-terminal part of the substrate (so called SMD atom) by a virtual spring, and its magnitude depends on the virtual spring force constant ( k ) and the velocity of the virtual atom ( v ), while the force vector is always parallel to the pulling vector of the virtual atom. The force vector coordinates, as well as coordinates of SMD atom are monitor ed and r ecorded throughout the simulation. The r esistance experienced by the molecule is expressed as a force r equir ed to over come it (user specifies the pulling velocity and the spring constant), and it is also possible to calculate the work / energy associated with each unfolding e v ent. In case of selecting se v eral residues for pulling the SMD atom r epr esents center of mass of C ␣ atoms of selected residues. A virtual atom moving with a constant velocity is connected to the real atom (SMD atom -in our simulations it is by default C ␣ atom of C-terminus, but it can also be the center of mass of selected residues) of the studied molecule by a virtual spring (Figure 3 ).
To compare SMD simulations in implicit and explicit systems we conducted SMD simulations of the GS-APP complex (shown as an example system in GS-SMD --it corre-sponds to the A ␤ 49 fragment of APP) in full membrane and wa ter (tempera tur e 298 K, pr essur e 1 bar). For explicit solvent simulations, the GS-APP complex was embedded in the membrane using CHARMM-GUI ( 35 , 36 ). The membrane composition used reflects the membranes of brain tissues affected by AD ( 37 ) including glycerophospholipids (POPC), cholesterol, sphingolipids (DSM) and diacylglycerol (DAGL) in a ratio 35:40:15:1. To the proteinmembrane system the solvent w as added: w ater type TIP3P and neutralizing sodium and chloride ions at concentration 0.15 M. The 25 ns long SMD simulations were conducted in NAMD ( 30 ) program with four repetitions. The pulling speed v = 0.1 m / s was applied to the C ␣ atom of C-terminus of the substrate and the spring constant k = 1.0 kcal / mol / Å 2 was employed both for implicit and explicit solvent simulations. The default pulling direction was used -it is based on the expected final position of the substrate C-terminus that forms a ␤-sheet with PS-1 as it is observed in 6IYC cryo-EM structure.

Description of input
The input data for SMD simulations of the GS-substrate complex is pr epar ed on the MD&mut page. It begins with a selection of one of the three curated structures: (i) the whole GS complex, (ii) the membrane part of GS and (iii) PS-1. Then, the user selects the substrate to bind to GS from a list of substrates with defined cleavage sites ( 4 ). The main, or initial, cleavage site is marked with an asterisk. In the next line, showing the membranous sequence of the selected substrate, user can modify the cleavage site by moving the W256 Nucleic Acids Research, 2023, Vol. 51, Web Server issue asterisk to another position and introduce optional mutations, deletions and insertions in the sequence. Only capital letters are allowed. The final sequence fragment that will be used for SMD simulation is highlighted below the input line. The fragment must be minimum 15 residues long and maximum 33 residues long because no more substrate residue is visible in the cryo-EM structure 6IYC ( 9 ) taking into account residues from the cleavage site to the N-terminus. Pressing the [Example input] button loads the whole GS complex with APP substrate and the initial cleavage sitethat fragment is known as A ␤ 49 , howe v er, its N-terminus is truncated by 16 residues non-visible in the cryo-EM structure 6IYC. There is a possibility to introduce mutations into any subunit of GS complex by specifying mutations in each chain -available ranges of residues are specified. The [Default Mutations] button introduces three mutations in chain B (PS-1) which are the early onset AD mutations ( 18 ). Parameters for SMD simulation, specified by user using sliders or input fields, include: pulling velocity (range from 0.01 to 0.5 m / s, default 0.1 m / s), spring constant (range from 0.1 to 10 kcal / mol / Å 2 , default 1.0 kcal / mol / Å 2 ), and a pulling direction by changing the position ( x , y , z ) of the end of the arrow. The start of the arrow is loca ted a t the C ␣ atom of C-terminus of the substrate (or center of mass of C ␣ atoms of selected residues). The arrow is visible at the adjacent interacti v e windo w sho wing the structure of the GS-substrate complex. The four final parameters for SMD simulations are: the total length (range 5-25 ns, default 15 ns), number of frames (range 10-200, default 50), the membrane thickness (20-40 Å , default 31 Å , which is a value from OPM database ( 38 ) for GS complex), and number of tasks to run (1-4, default 1). Since SMD and MD simulations are stochastic processes it is better to conduct more simulations to obtain statistically valid results. After confirming the entered data, the user will see a progress bar of the currently running simulation (or simulations in case of parallel tasks).
The extent of substrate unfolding depends on many factors: length of simulation, pulling velocity, spring constant (a measure of the stiffness of the spring), direction of pulling, and obstacles on the way of pulling. For APP, the maximal length of simulation available on the server, 25 ns (default is 10 ns), is far enough to obtain a conformation for next cut. Such simulation is shown in menu as Example 1. User can select any residue (its C ␣ atom) for pulling in any direction, and it is also possible to select more than one residue for pulling. This could allow to simulate e.g. tilting of the substrate -such tilting can be seen in Example 2 when residues from N-terminus of the substrate are selected. In Example 3, user can do a comparison of four SMD trajectories and conduct the interaction frequency analysis to see similarities and differences in substrate interactions.

Browsing and analyzing the single trajectory
After completing the SMD simulation, the user is r edir ected to the Analysis page. This page includes visualization of GS-substrate contacts via interacti v e flareplot (Figure 4 A) which can display various types of interactions (all, salt bridges , hydrogen bonds , aromatic, van der Waals and hydrophobic). In flareplot, the substrate internal contacts are also visualized to show changes in its structure during unfolding. The adjacent window contains an NGL viewer for visualization of the obtained trajectory (Figure 4 B). The visualization window control panel provides access to a selection of r epr esentations, coloring modes and viewing modes (Figure 4 C). In the NGL viewer, the user can see the obtained trajectory with additional informa tion: visualiza tion of membrane with adjustable opacity and visualization of pulling by an arrow that r epr esents the pulling vector. The arrow is connected by a spring to selected C ␣ atom (or center of mass of C ␣ atoms of residues selected for pulling). In the second part of the Analysis of single trajectory page, the graphs showing force (Figure 4 D) and work / energy (Figure 4 E) as a function of time, as well as changes in the secondary structure of the stretched substrate (Figure 4 F), the heatmap of internal hydrogen bonds of the substrate during its unfolding (Figure 4 G), and information about the task along with the form for the Continue mode (Figure 4 H) are displayed. The user interested in continuing the simulation can do it by selecting the trajectory frame at the bottom of the Analysis page and pressing the [Go to continue form] b utton. The Continue pa ge appears with similar selectable parameters as on the MD&mut pa ge b ut without changing the GS-substrate pair and mutations (the structure from the selected frame must be preserved). In this way, one can explor e differ ent wa ys to unf old the substrate.
Optionally, the user can download the result files packa ge. Packa ge includes: trajectory file (dcd), structure files (pdb and psf), contacts list, contacts frequencies files and flareplot data. It also contains data files for force and work graphs and the starting position of the SMD atom. Job results are to be available on the server at least two weeks after completion of the job. The user can see the results and analyze them using the link (token) copied from the progress bar page, copied from the top of the Analysis page, sent by mail, or included in the job description file in the results package. It is also possible to share the obtained trajectory with other users. This can be done using the [Make this job available to public] button on the results page (available immediately after the simulation is finished or later with the simulation token). The published simulations are visible under the 'Shared jobs' menu item. They will not be removed after two weeks.

Comparison and analysis of multiple trajectories
When se v eral simulations ar e completed, one can compar e them using 'Compare jobs' page to select required trajectories by their tokens. Se v eral r esidue-r esidue interaction types can be analyzed (hydrogen bonds, salt bridges, aromatic) for studies of the substrate interactions (internal interactions as well as to GS subunits). The user can select multiple trajectories into groups A and B, to be compared with one another (Figure 5 A). It is also possible to review each trajectory in a 'Single Trajectory' mode by clicking [View single] button next to trajectory number.
The HeatMap (Figure 5 B) visualizes the frequencies of r esidue-r esidue interactions (contacts) involving the substra te (chain E) calcula ted for each of the trajectories included in the comparison. The contact frequency values span between 0 and 1. Frequency 1.0 means that a certain  interaction is observed in 100% of the trajectory frames, frequency 0.5 is observed in 50% of the tr ajectory fr ames, frequency 0% means that this interaction was not found in particular trajectory but this interaction is collected because it exists in other compared trajectories and is equal or larger than the 'HeatMap threshold' value.
The Interaction frequency difference analysis (so called Sticks - Figure 5 C), identifies only those interactions for which the frequency difference (between trajectory sets A and B) are equal or greater than the 'Sticks threshold'. The stick color r epr esents the value of a fr equency differ ence for that interaction between trajectory sets A and B. Sticks threshold (its default value is 0.4) represents a minimal difference between frequencies of a certain interaction in sets A and B r equir ed to show this interaction as a stick and include the frequency difference of this interaction in a table. The lower the threshold the more interactions will be included.
Due to stochastic nature of conducted simulations several of them are required to obtain reliable results. The more trajectories each set contains, the more likely it is that a frequency difference of a certain interaction between those two sets is not random, but is rather caused by the change of simula tion conditions: dif fer ent dir ection of pulling, spring constant, membrane thickness, GS mode (whole system or part of it), mutation, etc. For comparison of mutated substr ates, the inter actions formed by mutated and unmutated residue will be interpreted as two distinct interactions, yielding two separate rows in heatmap and two separate sticks.

Tutorial and timeline
The GS-SMD w e b serv er includes an e xtended tutorial that guides the user step by step on how to perform SMD simulations. The tutorial starts with a brief explanation of the SMD method, which is the main method of GS-SMD Nucleic Acids Research, 2023, Vol. 51, Web Server issue W259 server. Description of SMD is accompanied by a schematic illustration of the method and the sample graphs (force and work / energy as a function of time) obtained from the server. Then, all menu pages and the features present in the GS-SMD server are described in detail, and each of them is illustra ted with appropria te screenshots. The input parameters are explained and the job submission process is illustrated with exemplary input data. A table of all GS substrates with defined cleavage sites ( 4 ) is provided with links to the Uniprot database ( 21 ). In addition, all the tools for analyses and visualizations available in GS-SMD are presented with exemplary results. There is also a Timeline page with all available GS structures in apo form, with substrates, and inhibitors / modulators.
To facilita te interpreta tion of SMD simula tions by inexperienced users, we have added new data in the tutorial showing the changes in the hydrogen bond network in APP. This was done as an example of single SMD trajectory analysis. By analyzing the SMD trajectory, the user can identify the hydrogen bonds that form / break during the simulation. As such analysis is highly dependent on the GSsubstrate / mutant pair, it is not part of the server and should be performed individually.

Performance and comparison to all-atom SMD simulations in explicit environments
The GS-SMD server allows to perform the peptide substrate unfolding in the GS acti v e site. Depending on the protease structure used (all GS complex, membrane part or only PS-1) the SMD simulation of 25 ns is completed in a pproximatel y 6, 4 and 2 h, respecti v el y. The same sim ulation in explicit membrane and water is 5-6 times longer, not counting system setup time. Considering that multiple SMD simulations of the same system are needed to get a more complete unfolding picture, the reduction in time required is significant. Comparing the force and work / energy versus time graphs obtained in explicit ( Figure 6A, B) and implicit ( Figure 6C, D) environments for GS-APP complex, it can be seen that the graphs are similar in the same force and work / energy ranges. We also observe breaking of hydrogen bonds in the substrate helix and formation of 3 10 helix at the C-terminus of the substrate in both implicit and explicit environments. This indica tes tha t the results obtained on the server ar e r eliable and can be used to investigate unfolding e v ents.

Comparison to experimental results
To compare SMD simulations with the experimental data, we analyzed F / t plots from SMD simulations for wild type APP and its mutants, I45T and T48P --these mutants were studied by Bhattarai et al. ( 11 ), and found to hinder unfolding. The F / t plots ( Figure 7AB) demonstra te tha t APP mutants r equir e longer pull time to reach the first e v ent of unfolding as indicated by the first significant decrease in force. This is especially true for the T48P mutant (T32P on GS-SMD server because of shorter substrate) (Figure 7 B) compared to wild type APP (Figure 6 C). As a result, the SMD method implemented on the server is sensitive enough to obtain r eliable r esults consistent with the experimental data.
Howe v er, one should remember that the simulations are stochastic processes and e v en under the same conditions, different trajectories could be obtained, therefore se v eral SMD simulations ar e r equir ed in each case. Also, comparing different substrates can gi v e erroneous r esults. Ther efor e, we r ecommend making comparisons only for the same substrates with various mutations in the substrate or / and in GS.
In GS-SMD server it is also possible to turn off SMD mode and run MD simulation without substrate pulling.
In this case all SMD parameters are disabled. We verified for APP substra te, tha t after selecting the proper frame from SMD simulation, the relaxation of GS-APP structure in MD simulation in Continue mode can bring the scissile bond (carbonyl group of V30 residue) to the catalytic residues. As seen in Figure 7 C, the violin plots indicate that short distances between the scissile bond and one of catalytic residue (D257) are reachable in all MD simulations. The most frequent distance is in range 4-5 Å indicating that such interaction can be bridged by water molecule although no explicit water molecules ar e pr esent in the simula tions. Bridging by wa ter molecules was found by Bhattarai et al. ( 11 ) in their GaMD (Gaussian-accelerated MD) simulations of GS-APP in explicit environments. Since the water molecule is r equir ed for the cleavage our r esults ar e in agr eement with those from much more sophisticated approaches.
Recently, Suzuki et al. ( 15 ) investigated specific mutations near the APP cleavage site that influence A ␤ production. Most of those mutations were located after the first (epsilon) cleavage site and ther efor e wer e not suitable for GS-SMD. Ther efor e, we decided to study APP mutation T714I (T27 using GS-SMD numbering) and PS-1 mutation K380E and compare to WT. From each SMD trajectory, we chose a frame where a distance of the next cleavage site to Nucleic Acids Research, 2023, Vol. 51, Web Server issue W261 the catalytic residues was the shortest. We considered trimming of A ␤ 49 peptide to A ␤ 46 , that r equir es a cleavage at V717 (V30 in GS-SMD). We compared the SMD work required to pull the substrate into a conforma tion tha t facilitate cleavage (Figure 8 ). In all cases it was possible to bring V30 to the catalytic residues but the r equir ed work was larger for APP T714I: 50 ( ±20) kJ / mol for WT, and 94 ( ±52) kJ / mol for the mutant. The difference in the mean values of work is statistically significant. Our results are consistent with experimental data showing that this mutation considerably reduces the cleavage activity of GS. For PS-1 mutation K380E the mean work was 98 ( ±81) kJ / mol. A high standar d de via tion may suggest tha t the muta tion causes certain destabilization of the acti v e site --it is also not statistically different from mean work for WT. Howe v er, this is also in agreement with experiment since PS-1 K380E mutant does not change GS activity unless associated with APP mutations. Details of simulations are shown in Supplementary material. The SMD works r equir ed to pull the substrate into a conformation that facilitate cleavage are shown in Supplementary Table S1, while the force and work charts for particular SMD simulations are shown in Supplementary Figures S1-S6.
In conclusion, the GS-SMD server allows to quickly obtain the results of unfolding of any substrate of GS in the acti v e site of this protease. The system under study is built by selecting the r equir ed building blocks and there is no need to add membrane, water and equilibrate the system. The server can be used to explore mechanisms of trimming and unfolding the substrate in order to obtain quick but reliable results for later verification in MD / SMD simulations in the explicit environments or with experimental methods. The GS-SMD server can also be used as a database of currently available GS structures and as a useful database of GS substrates divided into subgroups based on their TM sequences.

DA T A A V AILABILITY
GS-SMD is free for all users without logging in. E-mail is optional (for receiving notification of job completion) and is not stored on server after sending notifications. All con-ducted simulations described in this paper are available on GS-SMD server as Shared Jobs or Example Jobs.