How arginine inhibits substrate‐binding domain 2 elucidated using molecular dynamics simulations

Abstract

The substrate‐binding domain 2 (SBD2) is an important part of the bacterial glutamine (GLN) transporter and mediates binding and delivery of GLN to the transporter translocation subunit. The SBD2 consists of two domains, D1 and D2, that bind GLN in the space between domains in a closed structure. In the absence of ligand, the SBD2 adopts an open conformation with larger space between domains. The GLN binding and closing are essential for the subsequent transport into the cell. Arginine (ARG) can also bind to SBD2 but does not induce closing and inhibits GLN transport. We use atomistic molecular dynamics (MD) simulations in explicit solvent to study ARG binding in the presence of the open SBD2 structure and observed reversible binding to the native GLN binding site with similar contacts but no transition to a closed SBD2 state. Absolute binding free energy simulations predict a considerable binding affinity of ARG and GLN to the binding site on the D1 domain. Free energy simulations to induce subsequent closing revealed a strong free energy penalty in case of ARG binding in contrast to GLN binding that favors the closed SBD2 state but still retains a free energy barrier for closing. The simulations allowed the identification of the molecular origin of the closing penalty in case of bound ARG and suggested a mutation of lysine at position 373 to alanine that strongly reduced the penalty and allowed closing even in the presence of bound ARG. The study offers an explanation of the molecular mechanism of how ARG competitively inhibits GLN from binding to SBD2 and from triggering the transition to a closed conformation. The proposed Lys373Ala mutation shows promise as a potential tool to validate whether a conformational mismatch between open SBD2 and the translocator is responsible for preventing ARG uptake to the cell.

1. INTRODUCTION

Cell viability relies on the selective uptake of nutrients from the extracellular environment. ATP‐binding cassette (ABC) transporters are a class of integral membrane proteins that mediate the transport of cellular building blocks such as amino acids and other nutrients across the cell membrane. Besides two transmembrane domains and two cytoplasmic nucleotide‐binding domains, ABC transporters contain substrate‐binding domains (SBDs). The SBDs play a central role in the uptake processes as they mediate the initial substrate binding and the delivery to the translocation subunit of the transporter (Berntsson et al., 2010; Maqbool et al., ²⁰¹⁵). Their capability to bind a diverse array of molecules, encompassing ions, vitamins, sugars, peptides, amino acids, and inorganic anions, underscores their pivotal role in cellular function (Scheepers et al., 2016). Structurally, SBDs are characterized by two rigid subdomains (D1 and D2) bridged by a flexible $\beta$‐sheet hinge region (Fulyani et al., 2013). Classification of these proteins into different structural types can be achieved by the core topology and the composition of this $\beta$‐sheet region (Chandravanshi et al., 2021). For instance, type 1 substrate‐binding proteins (SBPs) feature three strands interconnecting the domains, while type 2 SBPs exhibit only two beta strands. Further categorization into distinct clusters is based on their structural folds and substrate affinities (Chandravanshi et al., 2021). Numerous crystal structures suggest the existence of many SBDs in either a ligand‐free (apo) open conformation or a ligand‐bound (holo) closed state. The conformational states and their associated dynamics are pivotal determinants in the overall transport process (Fulyani et al., 2016; Wolters et al., ²⁰¹⁰). Various biophysical techniques have been employed to unravel mechanistic insights into ligand binding and domain movement of SBDs (Chen et al., 2010; Gouridis et al., ²⁰¹⁵; Manjeet et al., ²⁰¹³; Pittelkow et al., ²⁰¹¹).

However, establishing a precise correlation between the mechanism of ligand binding and the conformational movement of domains can be challenging for SBDs. For example, certain SBDs undergo a conformational shift to the closed form upon binding of different ligands, or even in the absence of any ligand (Feng et al., 2016; Tang et al., ²⁰⁰⁷). Conversely, in SBDs like MalE, the binding of non‐cognate substrates appears to impede the transition to the closed state (Sharff et al., 1993; van den Noort et al., ²⁰²¹). Single‐molecule Förster resonance energy transfer (smFRET) experiments have suggested extensive plasticity in some of these proteins, indicating the presence of multiple semi‐closed conformations that are not easily categorized as open or closed states (de Boer et al., ²⁰¹⁹). The extent to which ligand binding induces domain closure (induced fit model) or whether conformational changes can occur independently of prior ligand binding (conformational selection model) has been studied for several SBDs and SBPs using smFRET (Gouridis et al., 2015; Peter et al., ²⁰²¹; Ploetz et al., ²⁰²¹; Tassis et al., ²⁰²¹) implying an induced‐fit mechanism. However, conflicting findings regarding the coupling between ligand binding and overall adaptation in the l‐glutamine (GLN) binding protein GlnBP from Escherichia coli have been related to the challenge of isolating one mechanism from another through experimental approaches (Han et al., 2024). Hence, formulating a universal model might be difficult due to subtle sequence disparities resulting in distinct and unique structural and dynamic behaviors (Chandravanshi et al., 2021).

Interestingly, substrate binding and transport can be perturbed or even inhibited by competing binders. This can involve non‐cognate ligands, for example, zinc can act as an inhibitor for some bacterial transporters (Chandravanshi et al., 2021) but also other amino acids. For example, the SBD2 of GlnBP binds GLN which is subsequently efficiently transported, but the process can be inhibited by arginine (ARG). Experiments have demonstrated that ARG not only competitively inhibits GLN binding and associated conformational changes induced by GLN but also impedes the uptake of GLN into the cell (de Boer et al., ²⁰¹⁹). This observation suggests that the non‐cognate ARG also binds to SBD2 but fails to induce a conformational change. Understanding the inhibition by ARG in molecular detail could give valuable insights into the mechanism of substrate binding and associated conformational changes necessary for substrate transport (de Boer et al., ²⁰¹⁹).

In this study, we employ molecular dynamics (MD) simulations focusing on the substrate‐binding domain 2 (SBD2) of the GlnPQ importer derived from the Gram‐positive bacterium Lactococcus lactis (see Figure 1).

FIGURE 1.

Crystal structures of the open unliganded state and GLN‐bound closed conformation of SBD2 of GlnPQ from Lactococcus lactis. SBD2 consists of two continuous subdomains: D1 (residues 255–343, 444–484) and D2 (residues 349–438), connected by two anti‐parallel $\beta$‐sheets (residues 344–348, 439–443). In the open conformation, the C‐terminal region (C‐tail) spanning residues 471–484 interacts with D2 and contacts residues within a D2 $\alpha$‐helix (residues 418–427).

Our aim is to deepen our understanding of the intricate relationship between ligand binding and the overall adaptation mechanism for this type 2 SBD. Note, in order to limit the size of the simulation system we restrict the simulations to the SBD2 and separate it from the other parts of the GlnPQ importer. The SBD2, which specifically targets GLN uptake, has already been extensively investigated through structural studies (Maqbool et al., 2015), smFRET experiments (de Boer et al., ²⁰¹⁹), and MD simulations (Kienlein & Zacharias, 2020). The high temporal resolution of MD simulations enabled the separation of GLN binding from domain motion in this system, serving as an asset in complementing experimental work. In our previous MD study, a two‐step mechanism for GLN binding and subsequent closing was proposed (Kienlein & Zacharias, 2020): Extensive MD‐simulation on the unbound open SBD2 structure in the presence of free GLN molecules resulted in reversible binding and dissociation of GLN to the native site on the larger subdomain D1 of SBD2 (but not to the D2 domain) without inducing closing. Binding of GLN to the D1 domain altered the energy landscape for closing (slight favorization) but still involves a free energy barrier for transition to the closed SBD2 conformation. In contrast, the apo SBD2 closing was found to involve a free energy penalty for closing. Hence, the existence of a barrier for closing indicates that initial GLN binding and closing are two separate events. The presence of GLN bound at the D1 just increases the likelihood of closing which once it has happened leads to a stable closed conformation (Kienlein & Zacharias, 2020). Hence, the mechanism differs from a classical induced fit conformational adaptation of the SBD2 upon ligand binding as suggested by smFRET (Gouridis et al., 2015). The simulations are nevertheless compatible with smFRET results that detected closing in the presence of GLN (de Boer et al., ²⁰¹⁹) but cannot distinguish between open bound or unbound SBD2 structures. However, the smFRET studies also indicate infrequent short‐lived closing in the absence of GLN compatible with a free energy penalty for closing in the apo‐SBD2 form. Additionally, our MD simulations predicted Leu480 in a C‐terminal helix to be crucial for controlling the transition barrier of the open‐closed transition of SBD2 (Kienlein & Zacharias, 2020) and suggested that an L480A point mutation would strongly reduce the transition barrier in favor of the closed form. Indeed, this prediction was subsequently validated through smFRET measurements (Gouridis et al., 2021) and supported the conclusions derived from MD and free energy simulations. In the present study, we use extensive continuous (c)MD simulations as well as advanced sampling protocols to decipher the binding mechanism of ARG to SBD2 in atomistic detail and compare it to its cognate substrate, GLN. We find in cMD simulations starting from the open apo SBD2 reversible transient binding of ARG basically to the same site on the D1 domain that also binds GLN without closing on the MD time scale. We characterize the binding strength using free energy simulations and compare it to GLN binding. Interestingly, free energy simulations to induce closing indicate a significant penalty when ARG occupies the native binding site in contrast to GLN that favors closing. This result explains both the competition effect of ARG and also offers an explanation why transport is inhibited by ARG that prevents closing. The analysis also identifies the underlying interactions preventing the adaptation of the closed state in the presence of arginine within the binding pocket. Based on the simulations, we also studied a K373A mutation which we found to allow closure even in the presence of bound ARG. This discovery holds significance as it may help to clarify whether a conformational mismatch between the open SBD2–ARG complex and the transmembrane domains (TMDs) of the GlnBP accounts for the inability to transport ARG into the cell. Note that similar coupling mechanisms have been suggested for other SBDs. For instance, in the maltose importer MaFGK2, which is one of the most thoroughly studied importers, a “concerted” model of transport has been proposed. According to this model, the maltose‐bound MalE (SBD of MaFGK2) and ATP synergistically initiate the catalytic cycle by binding to and reducing the energy of the intermediate conformation (Shilton, 2015). This process facilitates partial rotations in the TMDs, aligning critical catalytic residues at the NBD dimer interface, thereby facilitating ATP binding and subsequent transport into the cell (Chen et al., 2001). Moreover, it has been demonstrated that only the substrate‐bound closed form of MalE efficiently stimulates ATPase activity (Orelle et al., 2008).

This being said, a conformational mismatch might not be the only mechanism hampering ARG uptake; for example, the binding to SBD2 might be too stable to allow the efficient release of the substrate to the translocator. In other importers, substrate binding pockets inside the TMDs have been identified, which might also be linked to transport regulation (Oldham et al., 2013; Yu et al., ²⁰¹⁵). The proposed SBD2(K373A) mutation could potentially aid in deconstructing the intricate translocation process into discrete components for the SBD2 protein in the GlnPQ complex.

2. METHODS

For all MD simulations, the AMBER20 software suite and the ff14SB force field were used (Case et al., 2005; Maier et al., ²⁰¹⁵). Water was treated explicitly with the TIP3P water model (Jorgensen et al., 1983).

2.1. Unrestrained continuous MD simulations

Unrestrained cMD simulations were started from the crystal structure of the open SBD2 conformation (PDB:4KR5) from L. lactis (Fulyani et al., 2013). The protein was solvated in an octahedral box with a minimum distance of $10$ Å to the box boundaries. The protonation states of the titratable amino acids were predicted via Poisson‐Boltzmann calculations, using PROPKA3 (Olsson et al., 2011). Six ARG molecules were placed randomly into the simulation box, resulting in a ligand concentration of ~25 mM. Sodium and chloride ions were added to neutralize the system and reach an ion concentration of 100 mM. After an energy minimization of 2000 steps of steepest descent, the systems were gradually heated up to 300K while keeping positional restraints (force constant of 6 kcal mol⁻¹ Å⁻²) on the protein‐non‐hydrogen atoms during 1 ns simulation time. The restraints were gradually removed for another 1 ns, followed by unrestrained production simulations of more than 6 μs. All bonds involving hydrogen were kept at an optimal length using SHAKE (Ryckaert et al., 1977). The hydrogen mass repartition scheme was used allowing time steps of 4 fs (Hopkins et al., 2015). Trajectory analysis was performed using cpptraj and pytraj (Roe & Cheatham, 2013). Visualization of structures and trajectories was performed using visual molecular dynamics (Humphrey et al., 1996).

2.2. MMPBSA trajectory evaluation

Average interaction energy calculations on the unbiased trajectories have been conducted using the MMPBSA (Molecular Mechanics Poisson‐Boltzmann Surface Area) trajectory post‐processing method, utilizing the MMPBSA.py program of the AMBER package (Genheden & Ryde, 2015; Miller et al., ²⁰¹²). The purpose of the calculations was to obtain an estimate of the contribution of individual residues of protein on the binding of the ARG ligand. Here, the “single trajectory approach” was used, assuming that there are no significant conformational changes upon binding and yielding the mean interaction between substrate and receptor decomposed for each residue at the binding region. For this, 3000 frames in the ligand‐bound state of SBD2 sampled during free unrestrained MD simulation in the range from 100 to 1800 ns have been evaluated. The ion concentration was set to 100 mM (same as in the explicit solvent simulations). The harmonic conformational entropy contribution was included for calculating the total binding energy.

2.3. Absolute binding free energy calculations

To quantitatively compare the binding affinities of both GLN and ARG to the open form of SBD2, a well‐established, advanced sampling protocol, developed by Woo and Roux (2005) was applied (Figure 2). This approach effectively addresses convergence issues typically encountered in standard umbrella sampling (US) while calculating binding free energies, substantially reducing the overall simulation time.

FIGURE 2.

Schematic illustration of the contributions to absolute binding free energy calculations. In addition to the PMF along the radial separation coordinate $\xi$ with applied axial, orientational, and conformational restraints, the contributions associated with the release of each of these contributions have to be calculated in the bound and unbound state. The C_α atoms used to calculate the conformational RMSD restraint are indicated as red dots for illustration of the inclusion of conformational restraints. PMF, potential of mean force; RMSD, root‐mean‐square deviation.

A one‐dimensional potential of mean force (PMF) along the reaction coordinate $\xi$, defining a radial center‐of‐mass (COM) distance between ligand and receptor, is calculated. This calculation involves the imposition of restraints on several orthogonal degrees of freedom, along with maintaining conformational restraints on the binding partners. The restraints were defined between three virtual sites in the open conformation of SBD2 and the according ligand of interest (Gln, Arg) (for exact definitions of the sites, see SI, Figure S1). Due to the small sizes of both ligands, conformational restraints were only applied on the SBD2 protein. The restraint binding free energy $\mathrm{\Delta }{G}_{\mathit{\text{bind}},0,\mathit{\text{restr}}}$ needs to be corrected to obtain a physically meaningful standard binding free energy $\mathrm{\Delta }{G}_{\mathit{\text{bind}},0}$. First, the introduction of axial and orientational restraints is corrected for via free energy perturbation at the binding site ($\mathrm{\Delta }{G}_{a,\mathit{\text{site}}}$, $\mathrm{\Delta }{G}_{o,\mathit{\text{site}}}$) and an analytical calculation in the bulk ($\mathrm{\Delta }{G}_{o,\mathit{\text{bulk}}}$). Note that $\mathrm{\Delta }{G}_{a,\mathit{\text{bulk}}}$ is accounted for implicitely in the context of $\mathrm{\Delta }{G}_{\mathit{\text{bind}},0,\mathit{\text{restr}}}$. Finally, the free energy contributions associated with the conformational restraints are obtained by Boltzmann‐weighted integration of a PMF calculated along a root‐mean‐square deviation (RMSD) coordinate. For precise mathematical derivations, please refer to the original work of Woo and Roux (2005). The standard binding free energy is the sum of the individual contributions:

$$\mathrm{\Delta }{G}_{\mathit{\text{bind,}}0}=\begin{array}{l}\mathrm{\Delta }{G}_{\mathit{\text{bind,}}0,\mathit{\text{restr}}}+\left(\mathrm{\Delta }{G}_{o,\mathit{\text{bulk}}}+\mathrm{\Delta }{G}_{c,\mathit{\text{bulk}}}\right)\\ -\left(\mathrm{\Delta }{G}_{a,\mathit{\text{site}}}+\mathrm{\Delta }{G}_{o,\mathit{\text{site}}}+\mathrm{\Delta }{G}_{c,\mathit{\text{site}}}\right)\end{array}$$ (1)

The restraint PMF along the center of mass distance coordinate was calculated via US by pulling the ligand molecule out of a binding position from the SBD2 binding site. The biasing potentials cover a distance range from $3$ to $29.5$ Å with a window spacing of $0.5$ Å. Each window was simulated for 10 ns with a restraint force constant k of 15 kcal mol⁻¹ Å⁻² while keeping a force constant of 100 kcal mol⁻¹ Å⁻² for axial and orientational restraints. The conformational space of SBD2 was restrained by keeping harmonic restraints with a force constant of 5 kcal mol⁻¹ Å⁻² on all of its $C_{\alpha }$ atoms. The PMF was calculated with the weighted histogram analysis method (WHAM) (Kumar et al., 1992). The free energy contribution $\mathrm{\Delta }{G}_{a,\mathit{\text{site}}}$ was calculated via free energy perturbation and the Zwanzig formula (Zwanzig, 1954). Hereto, the axial constraints were released during eight short simulations (every 10 ns, force constants: 50, 25, 10, 5, 2.5, 1, 0.5, and 0 kcal mol⁻¹ Å⁻²). In a similar way, the $\mathrm{\Delta }{G}_{o,\mathit{\text{site}}}$ term was derived. Because of bulk isotropy, the orientational correction term in the bulk $\mathrm{\Delta }{G}_{o,\mathit{\text{bulk}}}$ can be evaluated analytically. The $\mathrm{\Delta }{G}_{c,\mathit{\text{site}}}$ terms were calculated by re‐evaluation of PMF's along a RMSD reaction coordinate. Here, we employed US in a range from $0.5$ to $5.7$ Å in steps of $0.2$ Å with a force constant of 15 kcal mol⁻¹ Å⁻². Similarly, the $\mathrm{\Delta }{G}_{c,\mathit{\text{bulk}}}$ contributions were obtained.

2.4. Replica‐exchange US simulations

US simulations coupled with Hamiltonian Replica exchange (HREUS) (Curuksu et al., 2009; Sugita & Okamoto, ¹⁹⁹⁹) between neighboring US intervals were performed to obtain the PMF for the opening/closing transition of SBD2 with ARG residing in the binding pocket (same procedure for K373A mutation). Here, the ligand had to be restrained to the binding pocket to prevent its escape in the intermediate umbrella windows. This was done via two harmonic distance restraints between the C atom in the carboxyl group of ARG and the C atom in ARG333's guanidino group as well as the C atom in ARG's guanidino group and the C atom in ASP267's carboxyl group with a force constant of 1 kcal mol⁻¹ Å⁻². The COM of backbone atoms in the larger domain D1 (residues 306–308) and backbone atoms in the D2 region (residues 396 and 397) served as a reaction coordinate $\xi$ (see Figure 5). A set of 16 US windows biased by harmonic potentials with force constants of 1 kcal mol⁻¹ Å⁻² were generated with equidistant spacing of $1$ Å covering distances of $6$ Å up to $21$ Å. During the simulations, exchanges were attempted every 1 ps reaching an acceptance rate of 20%–48%. The total simulation time amounted to 500 ns per window. The simulated distributions were analyzed via WHAM, yielding the corresponding free energy profile. Here, the implementation by Alan Grossfield was used which allows Monte Carlo Bootstrapping analysis (Grossfield, n.d.).

FIGURE 5.

Mechanism of ARG blocking SBD2 closing. (a) Influence of different ligands on the potential of mean force (PMF) upon closing. While GLN (blue line) stabilizes the bound state compared to the unliganded SBD2 (in purple), (red line) ARG hampers closing with a steep free energy increase for $\xi$ distances below 10 Å. Note that ARG was weakly restrained to the binding pocket during umbrella sampling (US) simulations to avoid dissociation (b) Cartoon representation of the simulation setup. The open form SBD2 (PDB: 4KR5, in green) with arginine (stick representation) bound to D1. Two centers of mass (red spheres) were defined to pull SBD2 into the closed state (blue cartoon representation) via US simulations. (c) Local side chain deformations compared to the crystal structure of closed SBD2 (PDB: 4KQP). RMSD of residue side chains to the crystal structure in the close proximity of arginine after best fit to the backbone. Lys373 constantly samples high deviation configurations. (d) Close‐up view of a frame sampled in the closed SBD2‐ARG bound state. The positively charged group of Lys373 cannot adopt the crystal structure (in blue transparent sticks) conformation because of repelling Coulomb interactions with the also positively charged ARG. RMSD, root‐mean‐square deviation.

3. RESULTS

3.1. Dynamics of the open SBD2 in the presence of ARG

Extensive unrestrained MD simulations were started from the open SBD2 state (PDB: 4KR5) with six ARG molecules randomly positioned in the vicinity of the protein (Figure 3a). Over the course of nearly 6 μs of unconstrained simulation, the ARG ligands extensively explored the surface of the open SBD2 conformation (sampled $C_{\alpha }$ positions of every 100th frame shown in Figure 3b). A distinct density peak was identified at the binding site in the D1 subdomain, consistent with the location previously observed for GLN (Kienlein & Zacharias, 2020). In Figure 3c, the global density maximum is illustrated on the left side by showing only positions above the threshold of half of the maximum value (see also SI, Figure S2). Conversely, other surface areas were visited infrequently, indicated by low‐density sampling (additionally, this is reflected in the highly fluctuating RMSD values of the ARG ligands shown in Figure 3a, upper panel).

FIGURE 3.

Arginine (ARG)‐binding mechanism to the open‐protein form of SBD2. (a) Upper panel: Root‐mean‐square deviation (RMSD; original data and running mean) of each ARG ligand (indicated by different colors) from the native‐like binding position of Gln at the D1 domain (PDB: 4KQP). Lower panel: RMSD of the overall SBD2 structure (after alignment of the D1 domains) to the crystal structures of the open (PDB:4KR5) and closed (PDB:4KQP) conformations. Ligand binding does not induce protein closing in the simulated timescales. (b) Illustration of the extensive ARG sampling around the open SBD2 conformation. The sampled ARG's C_α atom positions are shown for every 100th frame of the simulation. An increased point density near the ligand binding site of the SBD2 (green cartoon) indicates ARG accumulation at the site. (c) On the left side, the same data as in (b) are presented, with a density threshold set to half of the maximum value. SBD2 is illustrated in a green cartoon representation. On the right side, upper panel, a detailed view of ARG binding to the D1 domain of open SBD2 is shown, aligned with the geometry found in the crystal structure of the closed GLN‐bound form of SBD2. (GLN is depicted in orange stick representation, and neighboring residues of the crystal structure are in blue stick representation.) For the sake of easy comparison, a close‐up view of the same data obtained from simulations with GLN is shown in the bottom panel, with the crystal structure also in blue stick representation (Data taken from Kienlein and Zacharias (2020)). (d) MMPBSA interaction energy contributions for both ligands (GLN results have been taken from a previous publication; Kienlein & Zacharias, ²⁰²⁰). The charged ARG is capable of forming two stable salt bridges to Asp267 and Arg333. MMPBSA, Molecular Mechanics Poisson‐Boltzmann Surface Area.

Two binding events to the binding pocket, closely resembling the geometry observed in the crystal structure of the closed GLN‐bound form of SBD2 (PDB: 4KQP), were identified (RMSD < $2.5$ Å, with respect to the heavy atoms of ligands subsequent to alignment of D1's heavy atoms to the crystal structure of the closed‐liganded SBD2). Note that binding to the open SBD2 configuration exclusively took place within the binding pocket located on the larger domain D1 (residues 255–343 and 444–484). These binding events exhibited lifetimes of 2000 and >2900 ns, respectively. Notably, these lifetimes are longer than those recorded for GLN (with lifetimes ranging from 100 ns up to 1000 ns, as observed in our previous study (Kienlein & Zacharias, 2020)). The Boltzmann inversion of the probabilities of ARG‐bound versus unbound states, derived from this simulation, resulted in a standard ARG‐binding free energy of $\mathit{\Delta G}\approx -3\text{kcal}/\mathrm{mol}$. Note this can only be considered as an estimate, much longer simulation times are required for observing multiple binding/unbinding events for an accurate calculation.

Ligand‐binding however did not trigger significant domain closing (Figure 3a, bottom panel), with RMSD > $5$ Å to the closed SBD2 (PDB: 4KQP) structure. Several short‐lived transitions to a semi‐open configuration were observed (Figure 3a, bottom panel at times 150, 1450, 4100, and 5140 ns. A snapshot of the semi‐open state is shown in SI Figure S2b). The lifetimes of this conformation ranged from just 20 ps to 80 ns. These findings are in qualitative agreement with FRET measurements, which did not detect any transitions to the closed conformation when Arg was presented to the receptor (de Boer et al., ²⁰¹⁹).

Next, we investigated the molecular differences of ARG‐binding to SBD2 in comparison to its cognate substrate GLN. Analyzing the energetic contributions of individual side chains to binding in both scenarios allows for the elucidation of the molecular origins of target binding and the extended lifetimes observed in the ARG–SBD2 complexes. For the time intervals with site‐specifically bound ligands, MMPBSA trajectory analysis was performed to estimate the mean interaction contributions of individual residues in SBD2, stabilizing both GLN and ARG binding (Figure 3d).

Both ligands establish stable salt bridges between the terminal carboxyl group and the guanidinium group of Arg333 ($\mathit{\Delta G}=-6.34\pm 0.8\text{kcal}/\mathrm{mol}$ for GLN and $\mathit{\Delta G}=-4.6\pm 1.1\text{kcal}/\mathrm{mol}$ for ARG). Other residues involved in the binding process exhibit only slight preferences for ARG‐binding with $\mathit{\Delta \Delta G}=-1.0\text{kcal}/\mathrm{mol}$ for Ser328, $\mathit{\Delta \Delta G}=-1.6\text{kcal}/\mathrm{mol}$ for Ser325, $\mathit{\Delta \Delta G}=-1.1\text{kcal}/\mathrm{mol}$ for Phe270, and $\mathit{\Delta \Delta G}=-0.4\text{kcal}/\mathrm{mol}$ for Gly326. Phe308 and Met327 display a negligible distinction between the two ligands with, $\mathit{\Delta \Delta G}=-0.04\text{kcal}/\mathrm{mol}$ and $\mathit{\Delta \Delta G}=-0.16\text{kcal}/\mathrm{mol}$, respectively.

However, the critical distinction arises from the charged nature of ARG, enabling the formation of an additional salt bridge with Asp267 in close proximity to the binding site (close‐up views for both ligands shown in Figure 3c on the right side). This leads to a notably stronger binding energy in the case of ARG, with $\mathit{\Delta \Delta G}=-4.5\text{kcal}/\mathrm{mol}$. Including all energetic contributions, MMPBSA calculations estimate a slight preference for l‐arginine with the following binding energies for both ligands: $\Delta G_{\mathit{Arg}}=-10.9\text{kcal}/\mathrm{mol}$ and $\Delta G_{\mathit{Gln}}=-9.4\text{kcal}/\mathrm{mol}$ (including a quasi‐harmonic entropy term, value for GLN from Kienlein and Zacharias (2020)). However, it is essential to note that this method does not rigorously calculate absolute binding free energies; instead, it computes a mean interaction energy between the ligand and the receptor. Consequently, more robust derivations of equilibrium binding constants based on configurational ensemble averages were carried out in the subsequent section.

3.2. Absolute binding free energy calculations of ARG and GLN binding

The quantification of the standard binding free energy $\mathrm{\Delta }{G}_{\mathit{\text{bind}},0}$, including all entropic contributions, was performed using a restrained US approach (refer to details in Section 2). Unlike the previous MMPBSA method, this approach explicitly includes solvent, thereby accurately considering both solvent energetic and entropic contributions. The method allows for a more precise comparison between the two competing ligands, as it enables the robust calculation of absolute binding free energies. The fundamental concept of this approach involves the incorporation of geometrical and orientational restraints while computing a PMF along a radial distance coordinate $\xi$ separating receptor and ligand. This inclusion facilitates rapid convergence. The energetic contributions associated with the introduction or relaxation of these restraints can be individually calculated in the respective endstates (bound vs. unbound). These contributions are then considered and factored in to determine the absolute binding free energy.

The calculated absolute binding free energies amount to $\Delta G_{\mathit{\text{bind}},0}=-9.8\text{kcal}/\mathrm{mol}$ ($K_{\mathrm{d}}=0.07\mathrm{\mu M}$) for ARG and $\Delta G_{\mathit{\text{bind}},0}=-5.0\text{kcal}/\mathrm{mol}$ ($K_{\mathrm{d}}=0.25\mathrm{mM}$) for GLN (Figure 4). Therefore, the overarching trend of a higher binding affinity of ARG to the open SBD2 form, as obtained through the preceding MMPBSA analysis, is further reinforced by an even greater difference between the binding free energies observed here with $\mathit{\Delta \Delta }{G}_{0,\mathit{\text{bind}}}=-4.8\text{kcal}/\mathrm{mol}$. The binding affinities for both ligands are substantially higher than the values derived from a straightforward Boltzmann inversion of the binding probabilities observed in the free simulations (see Section 3.1). This disparity can be attributed to the substantial sampling effort required to attain sufficient statistics when simulating ligand binding to a receptor in an unrestrained manner. The challenges associated with free simulations, particularly in capturing a significant number of uncorrelated ligand binding events, are well‐recognized. As more binding events are sampled in free simulations, a convergence of the values obtained through Boltzmann inversion towards those obtained by the advanced sampling approach is expected.

FIGURE 4.

Absolute binding free energy comparison. (a) Schematic illustration of the simulations employed to calculate the potential of mean force (PMF) of binding with multiple restraints acting on the system. The C_α‐atoms (shown as red dots) of SBD2 are restrained to the binding pose by harmonic potentials. The ligands (ARG and GLN) are axially restrained by a polar and an azimuthal angle. The orientation of the ligand is fixed by restraining three Euler angles. These restraining contributions are corrected to get the physically meaningful standard binding free energy $\mathrm{\Delta }{G}_{0,\mathit{\text{bind}}}$ (details in Section 2). (b, c) Calculated PMF along the restrained SBD2‐glutamine separation distance $\xi$. The uncertainty is estimated by splitting the data set into 10 subintervals. (d, e) Energetic contributions for applying/releasing orientational, translational, and configurational restraints for both ligands.

Experimental measurements via isothermal titration calorimetry indicate a dissociation constant of $K_d=0.9\pm 0.2\mathrm{\mu M}$ for GLN corresponding to a $\Delta G_{\mathit{\text{bind}},0}=-8.3\text{kcal}/\mathrm{mol}$ (Gouridis et al., 2015). It is important to note that the calculations presented here exclusively cover the binding of ligands to the open form of SBD2. Given that experimental approaches typically incorporate contributions from domain closure, discrepancies in binding affinity are expected.

The apparent binding affinity of l‐arginine appears to be exaggerated in our simulations. Experimental findings indicate that higher concentrations of ARG hinder the uptake of GLN into the cell, with a K _i value in the millimolar range (de Boer et al., ²⁰¹⁹; Schuurman‐Wolters & Poolman, ²⁰⁰⁵). It indicates that the current calculations overestimate the ARG binding free energy. Binding of charged side chains involves a balance between charge–charge interaction and changes in solvation. Due to the very favorable solvation free energy of the guanidinium group of the ARG side chain ($\approx -80\text{kcal}/\mathrm{mol}$; Dixit et al., ¹⁹⁹⁷; Houriez et al., ²⁰¹⁷), this balance may not be sufficiently accurately represented resulting in an overestimation of electrostatic attraction versus (de‐)solvation. Indeed, previous studies have demonstrated an overestimation in electric field strengths within the utilized ff14SB force field (Bradshaw et al., 2020) that may overstabilize salt‐bridge formation. We anticipate that this overestimation effect has a more pronounced impact in the case of the substrate ARG due to its charged nature.

Stronger binding of ARG largely originates from deviations in the $G_{0,\mathit{\text{bind}},\mathit{\text{restr}}}$ terms ($\mathit{\Delta \Delta }{G}_{0,\mathit{\text{bind}},\mathit{\text{restr}}}=-5.93\text{kcal}/\mathrm{mol}$). This term is derived from the PMF shown in Figure 4b,c and corresponds to the interactions gained when the ligands transition from the bulk solution into the binding pocket. Notably, this observation corresponds well with the MMPBSA analysis that highlights the favorable electrostatic interactions of ARG within the D1 binding site of the open SBD2. The total cost of free energy associated with the loss of conformational freedom $\mathrm{\Delta }{G}_{\mathit{\text{conf}},\mathit{\text{bulk}}}-\mathrm{\Delta }{G}_{\mathit{\text{conf}},\mathit{\text{site}}}$ amounts to $1.12\pm 0.42\text{kcal}/\mathrm{mol}$ for Arg and $1.05\pm 0.46\text{kcal}/\mathrm{mol}$ for Gln. This indicates that the restriction of the conformational flexibility of SBD2 upon ligand binding plays a negligible role in discriminating between Arg and Gln.

The overall loss of orientational freedom upon binding $\mathrm{\Delta }{G}_{\mathit{ori},\mathit{\text{bulk}}}-\mathrm{\Delta }{G}_{\mathit{ori},\mathit{\text{site}}}$ leads to a free energy cost of $4.92\pm 0.13\text{kcal}/\mathrm{mol}$ in the case of ARG and $4.15\pm 0.38\text{kcal}/\mathrm{mol}$ for GLN. That means that the orientational fluctuations of ARG within the binding pocket are more constrained compared to those of GLN (favoring GLN vs. ARG). From a physical point of view, this is reasonable since both ends of ARG are stabilized by stable salt bridges to Arg333 and Asp267 of SBD2 (see Figure 3c, upper panel on the right side).

3.3. Bound ARG blocks the transition to the closed SBD2 conformation

smFRET experiments unveiled that transitions from the open SBD2 to the closed conformation of the protein are largely suppressed in the presence of ARG (de Boer et al., ²⁰¹⁹). Although the unrestrained simulations showed significant binding of ARG to the D1 domain of SBD2 in the open conformation, no transitions to the closed state were observed within the microsecond timescale (see Figure 3a, bottom panel). This suggests the existence of a substantial transition barrier for the opening/closing of the protein. In an attempt to capture the transition during free simulations, we modified SBD2 to SBD2(L480A). This specific point mutation has been proven to notably decrease the transition barrier, as evidenced in both computer simulations (Kienlein & Zacharias, 2020) and FRET experiments (Gouridis et al., 2021). However, while the L480A mutation for GLN was adequate to capture transitions to the closed state in cMD simulations, the introduction of this mutation with ARG bound still did not enable the sampling of significant transitions to the closed SBD2 state. The mutation however led to a more pronounced sampling of the aforementioned semi‐open state (see SI, Figure S3). It indicates, in addition to a high transition barrier, a considerable energetic penalty on the closed state when ARG is bound to the open SBD2.

Next, extensive HREUS simulations were conducted on the wild‐type SBD2 protein. These simulations utilized a reaction coordinate $\xi$ based on a COM distance between the two subdomains D1 and D2, capable of a clear distinction between the open and closed states (Figure 5b). Through the application of harmonic biasing potentials, SBD2 was steered toward the closed state. The resulting PMF along $\xi$ illustrates the free energy expenses required for this manipulation. Note that during these simulations, ARG had to be weakly restrained to the D1 binding site (details in Section 2).

The calculated PMF shows the considerable free energy cost of closing SBD2‐ARG with $\mathrm{\Delta }$ G values exceeding 6 kcal/mol (Figure 5a, in red). Previous MD simulations in contrast have shown a slight preference of the closed state for the SBD2‐GLN system with $\mathit{\Delta G}\sim -0.3\text{kcal}/\mathrm{mol}$ (Figure 5a, in blue). The unfavorable and high‐free‐energy conformations from these pulling simulations, where ARG occupies the binding pocket, were scrutinized for local deformations in comparison to the crystal structure of the closed GLN‐bound SBD2 complex. In this analysis, the backbone of residues in close proximity to the ligand (within $<4$ Å) were individually superimposed onto the crystal reference (PDP: 4KQP), and the RMSD for the side chains was calculated for each frame (Figure 5c).

Among these, Ser325, Met327, and Lys373 displayed the most significant deviations (mean RMSDs: $1.45$ Å for Ser325, $1.38$ Å for Met327, and $1.58$ Å for Lys373). Notably, especially Lys373 consistently exhibited substantial average deviations from the crystal structure (minimal RMDS: $0.1$ Å for Ser325, $0.15$ Å for Met327, and $0.7$ Å for Lys373). Upon inspection of the adopted configurations, a repulsion of the positively charged $ϵ$‐amino group of Lys373 from the ARG amino group, which also carries a positive charge, throughout all frames can be observed. However, the strong deviations and fluctuations in the conformation of the other residues (Ser325 and Met327) also demonstrate increased sterical repulsion due to the presence of ARG that interferes with domain closing. In Figure 5d, an exemplary conformation is depicted where the side‐chain amino group of Lys373 is significantly tilted away from ARG, illustrating the source of the substantial local deviation.

3.4. Mutation K373A in SBD2 allows closing in the presence of bound ARG

To evaluate the hypothesized significant role of Lys373 in impeding the closure of SBD2 when ARG is bound to D1, the impact of a K373A mutation on the overall dynamic behavior of the open SBD2–ARG complex was assessed. First, extensive unrestrained simulations of the SBD2 (L480A/K373A)‐ARG bound complex were performed. Similar to the preceding simulation, where closure of SBD2 (L490A) was prohibited, the mutation of the C‐terminal L480A point mutation was included to decrease the transition times between the open and closed states. In the course of 1.2 μs of free simulation, several short transitions to the semi‐open conformation were observed at times 95, 274, 533, 877, and 1196 ns. Interestingly, a complete closure of SBD2(L480A/K373A) in the presence of ARG was captured, indicated by an RMSD of less than $1.8$ Å to the closed crystal reference 4KQP (see Figure 6a). The closed state was maintained for a duration of 88 ns. Subsequently, a brief adaptation to the closed state occurred between 1151 ns and 1169 ns, with slightly increased deviations from the closed conformation ($\text{RMSD}\sim 2$ Å). However, this configuration quickly transitioned back to the open state, resulting in a lifetime of 18 ns.

FIGURE 6.

Point mutation K373A in SBD2(L480A) drastically lowers the free energy of the closed SBD2 configuration. (a) RMSD to the crystal structures of the closed (PDB:4KQP, in red) and the open (PDB: 4KR5, in blue) conformation. (b) PMF calculated for SBD2 K373A. Note that Leu480 was not mutated to Ala in these simulations. (c) Snapshots of an exemplary closed conformation. Binding site residues have been aligned to the crystal structure (in blue stick representation). PMF, potential of mean force; RMSD, root‐mean‐square deviation.

Closer inspection of the binding site of the fully closed SBD2(L480A/K373A) complex showed minimal structural deviations from the crystallized structure with bound GLN. Figure 6c depicts an exemplary frame from the simulation in superposition to the spatial side‐chain arrangement in the crystal structure (in blue stick representation).

Next, the PMF for the opening/closing transition of SBD2 K373A was calculated following the same procedure as in the unmutated case. Figure 6b illustrates the impact of the K373A mutations on the free energy profile. The mutation drastically decreases the free energy of the closed state by $\mathit{\Delta G}=4.58\text{kcal}/\mathrm{mol}$. The resulting PMF minimum aligns with the stable closing observed in the free simulations.

However, despite this mutation, the open SBD2(L480A/K373A) conformation remains slightly energetically favored. The calculated free energetic cost associated with closure amounts to $\mathit{\Delta G}=0.48\text{kcal}/\mathrm{mol}$.

4. DISCUSSION

SBPs play a vital role in facilitating the cellular uptake of essential nutrients (Berntsson et al., 2010; Fischer et al., ²⁰¹⁰; Maqbool et al., ²⁰¹⁵; Shilton et al., ¹⁹⁹⁶), yet the intricate relationship between ligand binding mechanisms and subsequent conformational changes within SBPs is not fully understood. The conformational states of these proteins hold central importance in governing the overall transportation process (Locher, 2016; Saleh & Kalodimos, ²⁰¹⁷), raising intriguing questions that demand exploration before achieving a comprehensive understanding, for example:

Is a conformational transition a prerequisite for substrate translocation? Unraveling this aspect can provide insights into the regulatory role of protein dynamics in the transport process. How does the binding of ligands relate to conformational transitions? Clarifying this intricate interplay could hold significant importance in understanding substrate specificity. What are the time scales of the various processes preceding the efficient translocation of nutrients into the cell?

Our study addresses aspects of this complex issue, particularly in the context of the SBD2 of the GlnPQ importer from L. lactis, shedding light on its role in l‐glutamine uptake and the inhibitory effects of l‐arginine. smFRET measurements provided already key insights into the competitive interaction between GLN and ARG (de Boer et al., ²⁰¹⁹). These measurements revealed that ARG not only competitively inhibits the GLN‐triggered closed conformation of SBD2 but also impedes GLN translocation into the cell, with an inhibition constant K _i in the millimolar range (de Boer et al., ²⁰¹⁹; Poolman et al., ¹⁹⁸⁷; Schuurman‐Wolters & Poolman, ²⁰⁰⁵).

To explore this phenomenon in molecule detail, we conducted extensive MD simulations alongside advanced sampling methods. Our MD simulations showed that l‐arginine occupies the same binding site at the D1 domain observed in prior simulations (Kienlein & Zacharias, 2020) for its cognate substrate, GLN. We calculated the binding affinities to SBD2 for both GLN and ARG. The computed binding free energy of GLN of ~−5.3 kcal/mol (including a favorable transition to the closed SBD2 form) is in qualitative agreement with an experimental dissociation constant, that translates to a binding free energy of approximately −8 kcal/mol.

ARG exhibited a higher calculated binding affinity (to D1 in the open state) compared to GLN in both free simulations, using MMPBSA and when employing more sophisticated absolute binding free energy simulations. A stable salt bridge formation between the positively charged ARG and Asp267 emerged as the pivotal interaction responsible for the enhanced binding affinity. Nevertheless, experimental evidence indicates a significantly lower binding affinity for ARG although a direct experimental measurement is not available. Experimental data are obtained from the ARG concentrations necessary to inhibit GLN transport and may involve other contributions then just competitive binding. However, from a biological view point stronger binding of ARG is unlikely because it would prevent GLN transport. We attribute this discrepancy to a possible overestimation of charge–charge interactions in salt‐bridge formation, a common issue associated with empirical additive force fields (Deng & Cui, 2022; Kamenik et al., ²⁰²⁰; Petrov & Zagrovic, ²⁰¹⁴; Yoo & Aksimentiev, ²⁰¹⁶; Yoo & Aksimentiev, ²⁰¹⁸). Particularly, attractive solute–solute interactions are notorious for yielding overestimated results (Miller et al., 2017; Yoo & Aksimentiev, ²⁰¹⁸). Indeed, the employed ff14SB force field has exhibited a possible overestimation of such electrostatics interactions, especially within binding sites containing arginine (Bradshaw et al., 2020). We speculate that these artifacts may have a minor impact on the uncharged glutamine. Nevertheless, our results give qualitative evidence for ARG binding at the same site as GLN and offer an explanation for the inhibition of GLN binding by ARG. Our studies furthermore revealed, that when ARG binds to the native binding site at the D1 domain, a transition to the closed SBD2 state is strongly suppressed. Using US simulations combined with HREUS resulted in a substantial penalty, exceeding 6 kcal/mol, for the transition to the closed state under these conditions. The presence of a positively charged Lys residue at position 373 in the D2 subdomain was identified as a main cause, leading to electrostatic repulsion, but sterical repulsion with respect to K373 and neighboring residues may also play a role. To test the importance of K373, we conducted simulations using the SBD2‐K373A variant and observed a significant impact on the free energy profile of closure, with only minimal preference for the open state (~0.48 kcal/mol) for the mutation (in the presence of ARG).

In summary, our simulations suggest the following underlying mechanism for GLN inhibition by ARG, as illustrated in Figure 7. ARG, when present in sufficient concentration, inhibits GLN‐binding by occupation of the same native SBD2 binding site in the D1 subdomain and prevents spontaneous transition to the closed state. We also identified residues in SBD2 that are perturbed by ARG binding and contribute to prevent closing. Introducing an Ala mutation at position 373 (L373A) significantly mitigated this effect and facilitated the adoption of the closed state in computer simulations.

FIGURE 7.

Proposed mechanism of l‐arginine blocking Gln‐transport. Asp267 plays a pivotal role in conferring binding ability to the substrate arginine. Subsequent transition to the closed state of SBD2 is prevented when arginine is bound. Notably, sterical repulsion between the ARG and residues at the D1–D2 interface including charge repulsion between bound ARG and Lys373 emerged as a key interaction that obstructs the conformational change. A conformational mismatch between the open SBD2‐Arg complex and the TMDs might prevent successful passage into the cytoplasm. TMD, transmembrane domain.

Consequently, subsequent substrate transport might be hindered since the open project number SBD2‐ARG complex fails to initiate allosteric interactions with the translocator. Docking of the closed conformation to the transmembrane domains of the importer is an essential step for substrate delivery (Machtel et al., 2019), and the “conformational mismatch” hypothesis has been previously proposed for SBD2 (de Boer et al., ²⁰¹⁹) and other SBDs (Hall et al., 1997; Skrynnikov et al., ²⁰⁰⁰; Tang et al., ²⁰⁰⁷).

However, our simulations suggest that the SBD2(K373A) variant may enable l‐arginine uptake. The mutation represents a promising candidate for experimental validation of this hypothesis, aimed at confirming whether a conformational mismatch between open SBD2 and the translocator indeed explains the observed inhibition of arginine uptake into the cell. Our work not only advances our understanding of the molecular mechanism and dynamics governing SBD2 but also underscores the importance of considering conformational changes and associated free energy landscapes in elucidating the functional aspects of SBPs.

AUTHOR CONTRIBUTIONS

Maximilian Kienlein: Methodology; investigation; writing – original draft; visualization; validation; software; formal analysis; data curation. Martin Zacharias: Conceptualization; funding acquisition; writing – review and editing; validation; project administration; supervision; resources.

Supporting information

Data S1: Supplementary Figures.

Kienlein M, Zacharias M. How arginine inhibits substrate‐binding domain 2 elucidated using molecular dynamics simulations. Protein Science. 2024;33(7):e5077. 10.1002/pro.5077