Insights into autoregulation of a membrane protein complex by its cytoplasmic domains

Abstract

Membrane transporters mediate the passage of molecules across membranes and are essential for cellular function. While the transmembrane region of these proteins is responsible for substrate transport, often the cytoplasmic regions are required for modulating their activity. However, it can be difficult to obtain atomic-resolution descriptions of these autoregulatory domains by classical structural biology techniques, especially if they lack a single, defined structure. The betaine permease, BetP, a homotrimer, is a prominent and well-studied example of a membrane protein whose autoregulation depends on cytoplasmic N- and C-terminal segments. These domains sense and transduce changes in K⁺ concentration and in lipid bilayer properties caused by osmotic stress. However, structural data for these terminal domains is incomplete, which hinders a clear description of the molecular mechanism of autoregulation. Here we used microsecond-scale molecular simulations of the BetP trimer to compare reported conformations of the 45-amino-acid long C-terminal tails. The simulations provide support for the idea that the conformation derived from electron microscopy (EM) data represents a more stable global orientation of the C-terminal segment under downregulating conditions while also providing a detailed molecular description of its dynamics and highlighting specific interactions with lipids, ions, and neighboring transporter subunits. A missing piece of the molecular puzzle is the N-terminal segment, whose dynamic nature has prevented structural characterization. Using Rosetta to generate ensembles of de novo conformations in the context of the EM-derived structure robustly identifies two features of the N-terminal tail, namely 1) short helical elements and 2) an orientation that would confine potential interactions to the protomer in the counterclockwise direction (viewed from the cytoplasm). Since each C-terminal tail only contacts the protomer in the clockwise direction, these results indicate an intricate interplay between the three protomers of BetP in the downregulated protein and a multidirectionality that may facilitate autoregulation of transport.

Significance

The terminal ends of membrane proteins often contain regulatory elements that are poorly characterized from a structural perspective. Despite containing predicted secondary structure elements, such segments may be connected by linkers, interact with lipids, and/or be inherently dynamic, and so are often modified for structural studies or simply not resolvable. This work describes the likely interactions of the terminal segments of the trimeric osmotic stress sensing protein, BetP, by a comparative simulation study of alternative reported conformations of the C-terminal tails and by prediction of an ensemble of conformations of the N-terminal tails. This work lays the groundwork for understanding the complex inter-subunit interplay in BetP autoregulation and provides an approach applicable to extramembranous domains in other membrane proteins.

Introduction

The ability of cells to move molecules across membranes is critical for accumulation of nutrients, extrusion of toxins, and responses to external stresses. Diverse transmembrane proteins dedicated to this process, called transporters, are abundant in all cells. A common stress factor for both human cells and bacteria is a change in osmotic conditions. Hyperosmotic stress, for example, triggers cell shrinkage, oxidative stress, and apoptosis. One strategy to counteract such an increase in external osmolality is to accumulate solutes such as betaine that are compatible with normal physiological functions. These “compatible solutes” can be taken up from the environment to increase the total internal osmolality, but uptake must be energized, e.g., by coupling to an electrochemical gradient. Moreover, such transport activity must be regulated so that uptake is limited in the absence of osmotic stress. Regulation therefore must include first, a mechanism to detect a stimulus, e.g., cell turgor, membrane deformation, or internal ion concentration, and second, a mechanism to transduce this signal to the solute transport machinery. One of the best studied regulated uptake systems is the glycine betaine transporter, BetP, from the soil bacterium Corynebacterium glutamicum. Impressively, BetP combines the sensor, the signal transducer, and the transporter in a single macromolecule (1), constituting an elegant prototype for transporter autoregulation in response to environmental change.

BetP is a homotrimer (2,3,4) in which the 12 transmembrane (TM) helices of each protomer adopt the so-called LeuT fold and constitute the unit responsible for sodium-driven betaine uptake (5). In the absence of activation, this core domain mediates basal levels of transport (6,7). The cytoplasmic N- and C-terminal tails function as osmosensors and osmoregulators (6,8), in part by detecting increases in the cytoplasmic concentration of monovalent cations, such as K⁺, to reach higher turnover rates. While the deletion of the ≈45-residue-long C-terminal segment abolishes activation by osmotic stress, protein variants with N-terminal tail deletions retain some capacity for regulation, albeit requiring higher osmolality to achieve the same degree of activation (4,6).

Structural studies of BetP indicate that the C-terminal segments form inter-subunit contacts. In all available x-ray structures, the helical C-terminal segment of one protomer protrudes into the cytoplasmic space (Fig. 1 a) and interacts with the cytoplasmic loops of an adjacent protomer (4), including with the segment called loop2, which connects TM segment 2 (TM2) and TM3. Since TM3 undergoes some of the most consequential conformational changes during the transport cycle (9), being equivalent to TM1 in LeuT fold, the presence of this interaction implies functional cross talk between the protomers during BetP regulation, as suggested previously based on low-resolution electron microscopy (EM) maps of two-dimensional (2D) crystals (10). However, in the x-ray structures, one of the three C-terminal tails per trimer is also involved in forming the crystal lattice through interactions with the periplasmic surface of another trimer in the neighboring unit cell, a surface that is unavailable to the cytoplasmic C-terminal tails in cells. The presence of this plainly nonphysiological interaction raises the possibility that this orientation of the C-terminal tail is a crystallographic artifact. More recently, a structure of BetP was determined using single-particle cryo-EM in the presence of amphipols to mimic the negative charges abundant in C. glutamicum membranes (11) unlike the dodecylmaltoside detergent micelles present in the crystals. In this EM structure, each C-terminal tail also forms a long helix that interacts with a neighboring subunit, but it adopts a different orientation than in the crystals, lying parallel to the membrane plane along the outer rim of the protein (Fig. 1 b). In this orientation, the C-terminal helix has the potential to interact more closely with lipids as well as with loop6, which connects TM6 to TM7, and with a different region of loop2. Adding to the enigma of why the C-terminal tail adopts different orientations, both crystal and EM structures were obtained under nonactivating conditions (i.e., low K⁺ concentrations and osmolality).

Figure 1.

Arrangement of the C-terminal tails in the BetP trimer and position relative to the cytoplasmic loops. The relative positions of the C-terminal tails (dark orange) to the transporter domain and lipids differs in available structures of BetP. The protein is shown as cartoon helices and colored to highlight TM12 (yellow), loop2a (light green), loop2b (dark green), loop4 (gray), loop6 (blue), and loop10 (pink). POPG lipids resolved in the structure are shown in dark blue and as lines, with the phosphate atom as a sphere. Top panels show the view from the cytoplasm. Lower panels show the view from along the plane of the membrane (brown region) with one protomer in the forefront and the cytoplasmic regions at the bottom. (a) Structure derived from an x-ray crystal structure (PDB: 4C7R) with C-terminal domains inclined with respect to the membrane. (b) Structure derived from an EM map obtained in the presence of amphipols (11) with the C-terminal domains lying flat relative to the membrane surface. To see this figure in color, go online.

Structural information relating to the ≈55-residue-long N-terminal tail is significantly more limited. Secondary structure prediction methods indicate that this tail contains small segments of helix among regions that lack structural definition (8,12). Accordingly, in the protein construct required for crystallographic studies, the first 29 amino acids were truncated and three glutamates were mutated to alanine to increase the helical content (4). Despite aiding crystallization, only a few amino acids of this modified segment were resolved. The recent EM studies were performed with the wild-type sequence of BetP, but unfortunately the N-terminal segment was still not detected, suggesting that this portion is internally unstructured or, perhaps more likely, adopts an ensemble of different orientations with respect to the rest of the protein under the conditions used.

Looking beyond the protein to its environment, structural studies have also revealed densities for individual lipid molecules that suggest specific contacts with the C-terminal helix and other protein regions, including loop2 (11,13).

A complete understanding of how cytoplasmic tails modulate a transporter domain cycling through its outward-open, occluded, and inward-open states requires a molecular-level description of the two cytoplasmic tails and their interplay with stimuli and the rest of the protein. For the C-terminal tail of BetP, the most essential segment for regulation, it is not clear how the two observed conformations are related or how they behave in the absence of crystal contacts and in the presence of a lipid bilayer. In this work we extend the structural studies by investigating the nature of these two possible conformations (which we refer to as either inclined or flat, Figs. 1 and S1 a) using microsecond-scale atomistic molecular dynamics (MD) simulations of the complete BetP trimer. Analysis of these simulations provides a detailed description of the conformational flexibility and interactions of the C-terminal tails in those states and the necessary context for predicting the structural features of the N-terminal tail. Using ab initio modeling, we therefore construct a wide range of possible models of full-length BetP, which indicate that autoregulation involves the cytoplasmic domains of both adjacent protomers.

Materials and methods

MD simulations based on the crystal structure, with inclined C-terminal tails

To explore the crystal structure orientation of the C-terminal tails, the 2.7-Å-resolution symmetric inward-open BetP trimer (PDB: 4C7R (13), UniProt: P54582) was selected. Side-chain rotamers of asparagine, glutamine, and histidine residues were examined using the MolProbity webserver (November 2017) (14) and, after visual inspection, 7 of the 16 proposed were modified: Asn59(B), Gln215(C), Gln346(A), Gln439(B), Asn517(A), and Asn517(C), where the letter in parentheses indicates the chain identifier. Histidine protonation was determined by inspecting possible hydrogen bond networks with other residues in the protein. All ionizable side chains were treated with default ionization states, except for Glu161 (in TM1′ in the center of the four-helix bundle) and Asp439 (in a short intracellular loop connecting TM2′ and TM3′), which are both buried in the low dielectric of the protein (15). We retained the eight complete 1-palmitoyl-2-oleoyl-sn-glycero-3-phosphoglycerol (POPG) lipids present in the crystal structure and added water in protein cavities using DOWSER (16). To allow the crystallographically resolved lipids to exchange with those in their environment without introducing artifacts due to the initial starting configuration, we opted to surround the trimer with a POPG-only bilayer. Embedding the protein-lipid complex into this ≈140-Ų hydrated lipid patch was carried out using GRId-based Force Field Input (GRIFFIN) (17). Sodium and chloride ions were added to a concentration of 200 mM Na⁺, 100 mM K⁺, plus Na⁺ as counterion to neutralize the system. The protein, crystallographic lipids, and ions were kept in place with harmonic restraints while the lipid bilayer was equilibrated using a 20-ns-long MD simulation. As only one of the C-terminal helices (in protomer A) in the crystal structure is fully resolved, we extended the other two protomers (residue 586) by aligning the C-terminal helix of protomer A onto the resolved regions of the other two protomers (residues 549–552 in protomer B and residues 549–561 in protomer C), removing overlapping water molecules. This system is shown in Fig. S1 a.

To assess the effect of potassium ions we prepared a second, equivalent system but with 200 mM Na⁺, 300 mM K⁺, and using Cl^– as counterion to neutralize the charges.

Both the 100 mM K⁺ and 300 mM K⁺ systems were equilibrated by gradually releasing restraints on the protein over 23 ns: 8 ns with harmonic restraints on the atom positions with gradually weaker force constants, then 5 ns with 4 kcal/mol/Ų harmonic restraints on the root-mean-squared deviation (RMSD) of the protomer backbone, using the minimized structure as reference. We then applied harmonic restraints of 4 kcal/mol/Ų and then 1 kcal/mol/Ų during two additional 5-ns-long simulations, on the RMSD of all protomers simultaneously. Restraints with the same force constants were applied to the φ and ψ dihedral angles in the C-terminal tails (residues 549–586) to retain their helicity. Finally, unrestrained simulations of 100 ns were used to obtain snapshots at 100 ns, 20 ns, 10 ns, and 50 ns for use as initial frames for Anton2 simulation replicates 1, 2, 3, and 4, respectively. See Table S1 for simulation timescales.

MD simulations based on the EM data, with flat C-terminal tails

Because of the moderate resolution of the available EM structural data, which at the time of setup was ≈6.8 Å, we did not use the corresponding structural model as the initial configuration for our simulations. Instead, a BetP protomer from the crystal structure was superposed onto the EM structural model using the backbone atoms of the TM regions, defined as residues 59–78, 82–124, 137–167, 178–207, 235–265, 277–294, 301–323, 326–348, 359–388, 394–426, 450–479, 488–510, and 515–544. The C-terminal helix from protomer A of the crystal structure was then fitted onto each of the C-terminal helices in the EM model. As above, the protein was embedded in a hydrated POPG bilayer (Fig. S1 b) and two sets of systems were generated, with either 100 mM K⁺ or 300 mM K⁺ in the aqueous solution. Equilibrations were carried out as above, using protein harmonic restraints. In addition, a restraint was applied to keep the protein inside the EM map using MD flexible fitting (MDFF) (18), reducing the force constant gradually over 68 ns. While using MDFF, restraints were applied to the peptide bond and backbone φ and ψ dihedral angles and chirality to prevent distortion by MDFF forces. For a final 100-ns unconstrained equilibration, the Rosetta EM score (19) of the C-terminal tails was computed as a measure of consistency with the experimental EM map for frame selection. The first simulation (run 1) was initiated with the last frame of that equilibration, while repeat simulations were started using frames with the best agreement (EM score of less than −400 a.u. for run 2), the worst agreement (EM score approximately −370 a.u. for run 3), and intermediate agreement (EM scores in the range −380 to −400 a.u. for runs 4 and 5) to the 6.8-Å EM map. Using these configurations, Anton2 simulations were carried out for up to ≈7.9 μs (see Table S1).

MD simulation parameters

All steps prior to the Anton2 runs were performed with NAMD v2.12 (20). The CHARMM36 force field (21,22,23) was used to describe the protein, lipid, and ions, while TIP3P was used for the water molecules (24). Simulations were performed using a semi-isotropic NPT ensemble (T = 298 K and p = 1 atm, constant x/y ratio) with periodic boundary conditions in all directions. The time step was 2 fs. Electrostatic interactions in the NAMD simulations were calculated using particle mesh Ewald (25) with a real-space cutoff of 12 Å. The cutoff for van der Waals interactions was set to 12 Å, with a switching function starting at 10 Å applied to both electrostatic and van der Waals interactions. The Anton2 simulations used the (1,1,3)-RESPA multigrator (26) (the update interval for bonded, near, and far interactions, respectively), a 9-Å van der Waals cutoff, no long-range dispersion correction, and U-series electrostatics (27). The latter parameters were those selected by the Anton2 “chooser” code, which resulted in a pure hydrated POPG bilayer with properties similar to those obtained during parameterization of the CHARMM36 force field (see supporting results section 1.1 and Fig. S2).

MD simulation analysis

The sampled conformations of the C-terminal tail were analyzed using several metrics. Stability was assessed by aligning on backbone atoms of the TM regions of the corresponding protomer and measuring the RMSD of each C-terminal segment (residues 548–586). The secondary structure was computed using Definition of Secondary Structure of Proteins (DSSP) (28) using MDtraj (29). The orientation of the C-terminal tail was measured using: 1) a crossing angle between TM12 and the C-terminal helix; and 2) a torsion angle defining the pivot of the C-terminal helix on the cytoplasmic surface of the trimer. The crossing angle was defined as the angle between two vectors connecting the centers of mass of the α-carbons of two sets of residues, namely residues 539–541 and 543–545 in TM12 and residues 551–553 and 562–564 in the C-terminal tail. Two pivot angles were measured, reflecting the movement of either the first half or the whole of the C-terminal tail, defined by measuring the torsion angle between four points at the center of mass of the α-carbons of a group of residues. The first point is embedded in the core of the transporter domain (residues 107–110 of TM2). The second point is the projection of point 1 along the axis of TM12, defined as a vector between the centers of mass of residues 532–535 and residues 540–543. The fourth point differs between the two torsion angles and is defined as either the center of mass of residues 554–557 or of residues 575–578, reflecting shorter or longer segments of the C-terminal tail, respectively. The third point is a projection of the fourth point onto the axis of TM12.

The Rosetta EM score (19) was used to measure the overlap between sampled configurations of the C-terminal tail and an independently determined EM map of the symmetric inward-facing BetP resolved at 3.7 Å and with the same features (11), which was obtained after the Anton2 simulations were completed.

AlphaFold modeling of full-length BetP

The full-length BetP sequence (UniProt: P54582) was used as input to AlphaFold-multimer v2.2.0 (30), using the default sequence database selection, but limiting template selection to all structures in the PDB prior to the publication of the first BetP structure 2009-05-26 (4). Twenty predictions were carried out for each neural network “model,” resulting in a total of 100 structural models.

De novo modeling of cytoplasmic N- and C-terminal tails

The N-terminal tail was modeled in the context of the BetP trimer complete with the C-terminal tail orientation identified in the simulations. To benchmark our modeling strategy, we also remodel the structure of the C-terminal tail within the same BetP trimer protein context. Modeling used the FloppyTail protocol (31) in Rosetta v3.9, which was specifically developed for exploring the accessible conformational space of a portion of a protein. To avoid sampling conformations in which the highly charged terminal tails enter the hydrophobic core of the lipid bilayer, we created a slab of pseudo-atoms from POPG atoms (P, C13, C2, C23, C27, C211, C215, C32, C36, C310, and C314) in the cytoplasmic leaflet of the bilayer, and added a repulsion term between the newly modeled tail and these pseudo-atoms. The three-residue-long fragments used to sample the backbone rotamers of the terminal segments were obtained from the ROBETTA server (http://robetta.bakerlab.org/fragmentsubmit.jsp). We excluded sequences homologous to BetP to prevent fragments from x-ray structures of BetP from being used. For the C-terminal tail benchmark, this choice avoids bias toward the known structure. For the N-terminal tail models this choice prevents inclusion of the structures obtained for the crystal construct, which is a modified version with a higher propensity to form helical structures (EEE44/45/46AAA).

Several sets of 100,000 cytoplasmic tail models were generated in the context of the trimer. Instead of identifying a single structure as the “true” terminal tail structure, global metrics were used to assess plausible secondary structure (with DSSP) and directionality of interactions with the rest of the trimer. Models were clustered in structural space using Calibur (32), allowing the protocol to select the optimal threshold. To assess the correlation between the interaction of the N-terminal with the C-terminal tail and intra-N-terminal interactions, two coordination numbers (C) were defined:

$$C=\sum _i\sum _j\frac{1-{\left(r_{ij}/R\right)}^6}{1-{\left(r_{ij}/R\right)}^{12}}\text{,}$$

where i and j refer to each non-hydrogen atom in each residue pair considered, r_ij is the distance between these atoms in the given model, and R is 5 Å. The more contacts formed between the two groups of atoms, the larger the value of the coordination number. Two sets of interactions were analyzed: 1) between residues 548–586 and 1–55, i.e., between the C- and N-terminal tails; or 2) between the two helical portions (residues 10–31 and 36–49) of the N-terminal tail. Coordination numbers were computed using the Collective Variable (COLVAR) Dashboard plugin in VMD (33). The models were clustered based on the 2D space defined by these two coordination numbers using the METAGUI analysis interface (34,35).

Results

The primary question that arises from comparing the structures of BetP is the preferred orientation of the C-terminal helix under nonactivating conditions (low osmolality), i.e., whether this segment is more likely to be inclined with respect to the surface of the membrane, as in the crystal structures, or to lie flat alongside the surface of the protein as in the cryo-EM data. To address this question, we carried out microsecond-scale MD simulations of BetP to examine the behavior of the two alternative orientations in the context of a hydrated POPG lipid bilayer, starting with the inclined orientation. Anionic lipids, including phosphatidylglycerol (PG), predominate in the native membranes of BetP (36). The use of a bilayer with a single lipid type reduces the chances of artifacts due to the specific choice of initial simulation frame. PG lipids were selected because they are present at high levels natively, they associate tightly with BetP even during purification and detergent solubilization (13,37), and they are resolved in many of the available crystal structures (see materials and methods).

Simulations of the inclined C-terminal conformation derived from a crystal structure

The C-terminal helix does not remain inclined in the absence of crystal contacts

The inclined orientation of the C-terminal helix is observed in all reported crystal structures, which include many different conformations of the membrane-spanning transport domain. Of these conformations, the inward-open state is the preferred conformation in resting conditions in the presence of saturating amounts of sodium but in the absence of betaine substrate (15). Therefore, here we simulated the trimer with all three transport domains in inward-open conformations (PDB: 4C7R), and with all three C-terminal helices in an inclined orientation. In the initial simulations, the aqueous solution contained a subactivating (100 mM) K⁺ concentration. The initial x-ray-derived conformation was lost after a few hundred nanoseconds, due not only to changes in the orientation of the helix relative to the membrane domain—as measured using the backbone RMSD (Fig. 2 a)—but also to loss of helicity (Fig. 2 c). The extended orientation was not revisited during a 4.8-μs-long simulation. Given the initial asymmetry of the x-ray-derived conformation (Fig. 1 a), we assessed the dependence of this instability on the choice of initial structure by performing additional 2.4-μs-long simulations starting from different frames of the equilibration. In all these simulations, at least one of the three C-terminal segments diverged from the inclined structure (Fig. S3, a–d) and its helicity was broken around residues 565–570 (Fig. S4, a–d), although the breakages occurred in different subunits, suggesting that the instability is not due to the initial asymmetry per se. The location of the breakpoint is notable, given that the A564P substitution abolishes regulation by BetP (38).

Figure 2.

Structural stability of the C-terminal tails in the conformation derived from the crystal structure (all-inclined). (a and b) RMSD of each of the three C-terminal tails in BetP after aligning the trimer using the transmembrane regions (as defined in materials and methods). The reference structure is the initial structure shown as inset in (a) (top left), while the structure of the final frame in each simulation is shown at the bottom right of the corresponding panel. (c and d) The secondary structure as a function of simulation time computed using DSSP and plotted for every residue in each C-terminal tail using the indicated color scheme. Results are shown for the two longest simulations with 100 mM K⁺ (a and c) and 300 mM K⁺ (b and d) and are also compared with data from repeated trajectories in Figs. S3 and S4. To see this figure in color, go online.

Potassium ions do not stabilize the C-terminal helix in the inclined orientation

The K⁺ concentration used in the aforementioned simulations was below that required for full activation (K_M ≈ 220 mM), implying that the fully extended conformation does not reflect the downregulated state. Given the role of the C-terminal tail in regulation, we considered the possibility that the inclined orientation might require high concentrations of K⁺ to be stabilized, i.e., reflecting an activated conformation. Hence, we simulated the trimer with the inclined helices in the presence of 300 mM K⁺. Again, in each of four repeat simulations of up to 4.8 μs length, one or more of the C-terminal helices (from different subunits) deviated from the inclined orientation (Figs. 2 b and S3 e–h) and irreversibly lost helicity (Fig. 2 d), suggesting that a high K⁺ concentration alone is insufficient to stabilize this highly tilted orientation.

The C-terminal region not involved in crystal contacts samples interactions observed in the crystal structure

The crystal lattice contacts formed by the extended C-terminal helix involve the end of the C-terminal segment, after the observed breaks in the helix (around Ala564). However, the segment prior to the breakage point (residues Val548 to Ala564) also contains significant secondary structure, and thus it is plausible that the first half of the C-terminal tail is conformationally stable and physiologically relevant. Indeed, during our simulations this first segment of the C-terminal tail did not deviate substantially from the initial conformation derived from the x-ray structure (Fig. S5, a–d).

To identify the interactions underpinning the orientation of the first helical segment, we computed the most persistent contacts during the simulations. Several intraprotomer contacts found in the crystal structure (Table S2) were also sampled in the simulations (Table S3 and Fig. S6 a, c, and d; supporting results section 2.1), involving residues in TM12, loop2a, and loop4 and reflecting the orientation of the first part of the C-terminal tail toward the center of its own protomer (Fig. 1 a). In addition, this first segment of the C-terminal tail interacted with both adjacent protomers due to the crisscrossing of the three tails (Tables S2 and S3; Fig. 1 b; Fig. S6, a and d).

Taken together, these analyses indicate that the two segments of the C-terminal tail should be considered separately, since the segment following Ala564 is highly dynamic whereas the prior segment is substantially less mobile and samples intra- and interprotomer contacts also observed in the experimental structure. Thus, the portion of the C-terminal tail not involved in lattice contacts may in fact be adopting a physiologically relevant orientation in the x-ray structure, one that may be supported by intraprotomeric interactions with the cytoplasmic loops.

The conformation of the first part of the C-terminal tail does not change with the addition of potassium ions

Although 300 mM K⁺ did not stabilize the entire length of the C-terminal tail (Fig. 2), we considered whether K⁺ could interact with the more stable components of the transporter in this conformation at this high concentration. In the simulations containing 300 mM K⁺, the first part of the helix is also relatively stable (residues 548–564; Fig. S5, e–h) and forms similar contacts as in simulations with 100 mM K⁺ (Table S3 and Fig. S6 b). Notably, the K⁺ ions themselves form essentially no contacts with the C-terminal tail in this orientation (Fig. S7, a and b), even at a high concentration of K⁺, and therefore do not directly compete with the interactions between the C-terminal tail and the rest of the trimer. The lack of C-terminal-K⁺ interactions is perhaps surprising given the sites recently observed in crystals obtained at high Rb⁺ (a mimic of K⁺) concentrations (11). These sites involve Q557 and Y553 in one C-terminal helix, and T124, Q557, and E552 at the intersection of two C-terminal helices. In the case of the latter region, it is possible that the x-ray conformation being simulated, which was obtained with low K⁺ concentrations, is in a low-energy minimum that cannot be escaped on microsecond timescales, preventing the axial rotation of the C-terminal helices required to form these sites. However, in the case of the Rb⁺ ion interacting with Q557 and Y553 (only two side-chain oxygen atoms from the protein), it is likely that the force field cannot recapitulate such a hydrated site with so few protein atoms. Although in this work we focus on comparing structures obtained in low-K⁺ conditions, it could be of interest to investigate the structure obtained at high Rb⁺ to explore these issues in future work.

Simulations of the flat C-terminal conformation derived from cryo-EM data

Recently determined EM structures suggest that, in the absence of K⁺ and high osmolality, i.e., under nonactivating conditions, the C-terminal tail forms a long, relatively straight helix that lies flat along the trimer cytoplasmic surface (Figs. 1 b and S1) (11). Importantly, the EM structure lacks the issues associated with the crystal contacts discussed in the previous section, although its resolution hinders clear identification of many of the specific interactions that favor the flattened orientation of the helix. Moreover, the EM data were obtained for BetP trimers in amphipols, which mimic the high density of negative charges present in the C. glutamicum membrane but raise the question of whether a lipid bilayer environment might lead to a different orientation of the C-terminal tail. Thus, here we evaluate the stability and dynamics of this flat conformation in microscopic detail and in the presence of an anionic lipid bilayer. To build the simulation system, we fitted the longest x-ray-resolved C-terminal helix into each C-terminal segment in the available EM map and then restrained the protein into the map using MDFF (18) during equilibration (see materials and methods). Multiple unconstrained simulations were performed for up to 7.9 μs (Table S1).

The EM-derived conformation remains stable

In the simulations of BetP carried out with downregulating concentrations of K⁺, the C-terminal helix in the EM-derived conformation did not deviate greatly from the initial flat structure, even over 7.9 μs (Figs. 3 a and S8 a–e). The RMSD values for the C-terminal tail flatten off at ≈4–7 Å, in stark contrast to the rapid and large (15–20 Å RMSD) structural changes observed in the simulations of the x-ray-derived state of BetP (Figs. 2 a and S3), suggesting that the C-terminal helix retains the flat orientation in these simulations. Notably, the helicity was preserved throughout aside from the last seven residues (580–586, Figs. 3 c and S9 a–e), which are also poorly resolved in the EM map.

Figure 3.

Structural stability of the C-terminal tails in the conformation derived from the EM structure (flat). (a and b) RMSD of each of the three C-terminal tails in BetP after aligning the trimer on the transmembrane regions. The reference structure is the initial model, shown as inset in (a) (top left), while the structure of the final frame in each simulation is shown at the top right of the corresponding panel. (c and d) The secondary structure computed using DSSP, plotted for every residue in each C-terminal tail as a function of simulation time, using the indicated color scheme. Results are shown for the two longest simulations with 100 mM K⁺ (a and c) and 300 mM K⁺ (b and d) and are compared with RMSDs in repeated trajectories in Fig. S9. To see this figure in color, go online.

Despite being significantly smaller than for the x-ray structure, these RMSD values are still relatively large and suggest significant dynamics of the helix. To assess whether these fluctuations faithfully reproduce the configurational space observed experimentally, we computed the overlap between the EM map and a computed map from each simulation frame, using the Rosetta EM score (19) (see materials and methods). Since an independently obtained, higher-resolution EM map (3.7 Å) was available at this point, this new map was used for analyzing the simulations (11). According to this metric, most of the simulation snapshots overlap with the experimental map to a similar degree (Fig. 4) and fluctuate around the mean value (Fig. S10), suggesting that the dynamics are physiologically meaningful. Visualizing snapshots with mean values of the EM score (Fig. 5, c and d) reveals how the C-terminal helix maintains a flat configuration during the simulations while accessing alternative orientations that have a similar degree of overlap with the experimental map.

Figure 4.

C-terminal tail dynamics in simulations of the BetP trimer compared with the EM data. (a and b) The Rosetta Electron Microscopy (19) score, computed for each of the three C-terminal tails in each independent simulation (colored lines) or summed over each set of simulations (black lines), initiated with EM-derived structures, is shown as a distribution over the simulations carried out with either 100 mM K⁺ (a) or 300 mM K⁺ (b). The combined, accumulated simulation time is >52.2 μs and 43.2 μs, respectively. (c and d) Simulation snapshot, shown as cartoon helices, compared with the EM map (yellow), and viewed from (c) the cytoplasm or (d) the plane of the membrane, along the direction marked with an arrow in (c). The Rosetta EM score of all three C-terminal tails is near the median value (approximately −127 a.u.). Fluctuations in the C-terminal helix positioning can occur along its full length (blue), at the end of the C-terminal tail (orange), or close to the connection with TM12 (green). In (d), the approximate boundaries of the membrane are marked with orange lines. To see this figure in color, go online.

Figure 5.

Pivoting of the C-terminal helix of BetP in simulations derived from the EM data (all-flat). The orientation of the C-terminal helix is plotted as a distribution over each simulation trajectory (colored lines) and averaged over all five runs (solid black lines). (a and b) Torsion angle defined using the full length of the helical segment of the C-terminal tail, with the endpoint at the center of residues 575–578. (c and d) Torsion angle defined for the first helical segment of the C-terminal tail, with the endpoint at the center of residues 554–557. Simulations were carried out with either 100 mM K⁺ (a and c) or 300 mM K⁺ (b and d). The angles obtained for the corresponding experimental structure are shown as vertical black lines in (a) and (c). Insets illustrate the position of the vectors defining the pivot angles for each C-terminal tail (green, orange, and blue lines), in 32,909 frames from the 7.9-μs-long trajectory carried out in the presence of 100 mM K⁺. The protein is viewed from the cytoplasm. Schematics illustrate how the orientation of the C-terminal tail toward either the lipid bilayer or the center of the neighboring protomer is reflected by larger or smaller pivot angles, respectively. To see this figure in color, go online.

The C-terminal helix derived from the EM maps pivots on the trimer surface

The aforementioned structural analyses indicate that the C-terminal helix of BetP can take substantial excursions around its average position during the simulations. To assess whether the helix truly remains flat, i.e., parallel to the membrane plane, we measured the crossing angle between TM12 and the first C-terminal segment (up to Ala564; see materials and methods). In the simulations based on the EM data, the value of this angle remains around 90° (Table S4), substantially smaller than that obtained for simulations based on the crystal structure (≈135°; Table S4), indicating that the C-terminal helix indeed lies flatter in the former simulations.

In a flatter orientation, the C-terminal helix has a greater potential to interact with surrounding lipids. We therefore next analyzed the extent of pivoting motions in the plane of the membrane, defining a pivot angle where values <90° reflect movements toward the protein surface and values >90° represent fluctuations toward the POPG bilayer (see materials and methods). When measured over the entire length of the C-terminal segment, the distribution of the pivot angle is narrow and centered around the values obtained for the initial structure (≈90°, Fig. 5 a). Visualizing these vectors as lines illustrates the degree of fluctuation that can occur over the length of the C-terminal tail on a multi-microsecond timescale (see inset to Fig. 5 a).

Although it is tempting to consider the C-terminal helix as a single rod-like structure, we noticed a disconnect between the movements of the first and second halves of the C-terminal tail, consistent with previous proposals based on the effects of mutagenesis (8). Defining a shorter version of the helix vector separated out the behavior of these two segments. The resultant distribution is broader, and the initial structures have different values depending on the subunit (Fig. 5 c), indicating local variations in the segment of the C-terminal tail close to TM12 (see green helix in Fig. 4 c and inset to Fig. 5 c). Such repositioning may be facilitated by changes in helicity at the very beginning of the C-terminal tail (residues 548–550; Figs. 6 c and S9 a–e) that affect the length of the linker or simply by repositioning of the helix with respect to TM12 (e.g., Fig 2 c, orange C-terminal helix; Fig. S6 c), even while maintaining contacts at the far end of the domain. Together, these analyses illustrate that although the C-terminal segment is primarily helical, it exhibits complex behaviors beyond those that might be imagined for a rigid helical structure.

Figure 6.

Cation interaction site predicted by MD simulations of the BetP trimer. (a–d) Potassium ion occupancy map (yellow surface), showing for clarity only densities close to the cytoplasmic surface and present in all three protomers. Ion-protein contacts are defined as any K⁺ ion within 3.5 Å of any non-hydrogen atom of residue Asp470. The protein is shown in gray surface representation and as cartoon helices for TM6, loop6, TM7, TM10, and the C-terminal tail, colored by protomer (orange, blue, and green). Asp470 is shown as spheres, and Ser293 is shown as sticks. Simulations were initiated with either (a and b) the inclined orientation of the C-terminal tails found in x-ray crystallographic structures or (c and d) the flat orientation observed in EM maps. The bulk concentration of K⁺ was either 100 mM (a and c) or 300 mM (b and d). (e and f) Close-up of the potassium ion when coordinated by Asp470 in simulations in which the C-terminal tail is inclined (e) or in a flat orientation (f). Representative snapshots with high coordination numbers are shown with the ion (sphere) and nearby amino acids (sticks) from TM6, loop6, TM7, TM10, loop10, and TM11 (cartoon helices). In (e), the C-terminal tail of the adjacent protomer contacts residues in loop6. Individual residues are labeled. Segments from an adjacent protomer are marked with a prime (′). To see this figure in color, go online.

The C-terminal helix in the flat orientation does not contact the counterclockwise protomer

It is notable, given the presence of a bilayer composed entirely of negatively charged lipids, that the highly positively charged C-terminal helix does not simply swing outward to interact with those negatively charged headgroups (Fig. 5). The implication is that such a reorientation is less favorable than existing interactions between the transport domain and the C-terminal tail, analogous to those observed in the inclined conformation. Indeed, some of the intraprotomer contacts are common to both inclined and flat orientations, particularly those involving TM12 and loop2 (Table S3 and Fig. S6; supporting results section 2.1). However, there are also unique interactions—involving those same structural elements—and additionally, the flat orientation does not allow contacts with loop4 (Fig. S11 a, upper panel). These differences in intraprotomer contacts reflect that the start of the C-terminal helix is rotated outward toward the periphery of the trimer compared with a more “tucked in” position in the inclined conformation (Fig. 1).

Interprotomer contacts are abundant in the flat orientation of the C-terminal helix, but these differ substantially from those seen in the inclined orientation. Importantly, we do not observe contacts between any C-terminal segment and residues in the counterclockwise protomer, indicating a strong directionality of the interactions in the flat C-terminal orientation. In the clockwise direction, the C-terminal helix interacts with different residues in loop2 than in the inclined conformation and shows novel interactions with loop6 (Table S3; Figs. 7 c and S11 a [lower panel]; supporting results section 2.1). Both loop2 and loop6 are directly connected to membrane-spanning helices that undergo major conformational changes upon closure of the cytoplasmic pathway (TM3 and TM7, respectively) (9). Preserving these contacts in the flat orientation of the C-terminal helix therefore likely favors the observed inward-open conformation of the transport domain. Release of the C-terminal tail from those interactions may increase the probability of the inward-to-outward-open conformational transition that is required during the betaine transport cycle. Accordingly, these interactions are good candidates for hallmarks of a downregulated state.

Figure 7.

Contacts between the C-terminal tail of BetP and the rest of the trimer. (a and b) Contacts sampled during simulations of the flat orientation within either 100 mM K⁺ (a) or 300 mM K⁺ (b). Interactions involve three distinct segments of the C-terminal tail. Boxes indicate the three regions in the close-ups in (c). The protein is shown as cartoon helices and colored according to Fig. 1. Ions and key interacting residues are shown as spheres. (c) In the close-up figures, pairs of residues forming contacts for >20% of the simulations are shown as sticks. The bubble plot illustrates each contact in these segments and whether it is formed more frequently (black) or less frequently (gray) in the simulation with a higher concentration of K⁺. The size of each circle reflects the magnitude of the change. To see this figure in color, go online.

The flat C-terminal conformation is stable in the presence of a high potassium concentration

Regulation of the transporter is thought to involve changes in the conformation of the C-terminal tail (39). Although we do not expect to observe the complete activation conformational change on these timescales, given the relatively hydrated nature of the C-terminal helix we considered it possible that K⁺ ions could locally disrupt, or otherwise influence, the contacts that keep the C-terminal helix flat against the BetP transporter domain, and so we repeated the simulations of this conformation in the presence of 300 mM K⁺. However, throughout these five simulations, which amounted to 43.2 μs, considering each C-terminal helix as an independent entity, the C-terminal tail remained relatively stable according to the RMSD (Figs. 3 b and S8 f–j), the EM score (Figs. 4 b and S10 f–j), and the helicity (Figs. 3 d and S9 f–j). In addition, in the presence of a high concentration of K⁺, the helix remained parallel to the membrane (Table S4). Similar to the simulations with 100 mM K⁺ (Figs. 5 a and 7 c) the C-terminal helix pivots somewhat on the surface of the trimer (Figs. 5 b and 7 d). We note that the distribution of pivot angles appears broader in the presence of 300 mM K⁺ (Fig. 5 d), especially in the direction of the lipid bilayer; however, the most extreme values were only observed in individual simulation trajectories, limiting our ability to make robust conclusions about the effect of higher K⁺ concentration on tail orientation. Thus, increasing K⁺ has no detectable effect on the C-terminal tail of the EM-based conformation on a multi-microsecond timescale.

Potassium ions do not bind directly to the C-terminal segment, even at a high concentration

The fact that the global orientation of the C-terminal tail is similar in the simulations with both K⁺ concentrations suggests that K⁺ ions are not directly competing with its contacts to the transport domain. Contact analysis confirms that none of the C-terminal tail residues bound K⁺ ions for a substantial proportion of the trajectories with the flat orientation (Fig. S12).

To conclude, our simulations support the notion that the flat orientation is adopted under nonactivating conditions, even in the absence of the N-terminal tail. This flat orientation is best described as a conformational ensemble encompassing various pivoting and rotating movements of the helix over the surface of the trimer and toward the lipid bilayer. Despite these dynamics, however, the C-terminal tail maintains many hydrophobic and charged contacts that together may be important for downregulation of the transport domain.

Interactions of K⁺ with the membrane domain suggest a regulatory binding site

A major question about BetP regulation is how the orientation of the C-terminal tail and the environmental conditions, such as cytoplasmic K⁺ concentration, are transduced to the transporter domain to accelerate the transport rate. We took advantage of the high K⁺ concentrations in our simulations to examine other areas of the transporter that might favor K⁺ binding. This analysis revealed a reoccurring interaction site where K⁺ ions accumulated more than in any other region, resulting in a distinct signal in the occupancy maps on the cytoplasm side of the transporter domain (Fig. 6). The density occurs in a crevice between TM10 and the flexible loop6 connecting TM6 and TM7 (Fig. 6). Specifically, Asp470 and Ser293 contact K⁺ for 27%–57% and 16%–53% of the high-K⁺ simulations, respectively (Figs. S7 and S12), with the occupancy increasing at least threefold with K⁺ concentration. Since TM7 lines the cytoplasmic pathway, this site may affect the (likely rate-limiting) step of resetting the apo transporter to the outward-facing conformation.

Given the observed solvent accessibility and moderate occupancy of the Asp470 crevice, we considered the possibility that this site can nonspecifically bind all monovalent cations. The aqueous solution used in the simulations contained 200 mM Na⁺ (required to counteract the negative charges of the POPG lipids), so we also analyzed the occupancy of Na⁺ ions in this region. At this concentration, Na⁺ ions also interact frequently with Asp470 (Fig. S7, a and b; Fig. S13, a, b, and e), albeit to a lesser extent than K⁺ ions, for 5.4% of the simulations with 100 mM K⁺ (Fig. S13 a). In physiological conditions, with low intracellular Na⁺, K⁺ ions would therefore likely outcompete Na⁺ for this site.

Interestingly, the crevice occupancy is somewhat lower for simulations with the flat orientation than with the inclined orientation of the C-terminal tail (Fig. 6 b). A detailed analysis of the changes in contacts between the C-terminal helix and the cytoplasmic loops (supporting results section 2.2) suggests a set of interaction changes that could occur upon binding of K⁺ to Asp470 to disconnect the C-terminal tail from loop6 and thereby reduce the barriers to closure of the cytoplasmic pathway.

We note that the above findings appear inconsistent with the fact that no density for Rb⁺ was reported to occupy this region in the crystallographic data (11), nor have any cation densities been attributed to this region in other structures. Unfortunately, loop6 is highly dynamic and poorly resolved in BetP structures, presenting a challenge for experimental density assignment. Moreover, crystals of BetP only form at pH 5.5, when Asp470 would likely be protonated (15) and have a decreased affinity for cations. We therefore suggest that this site may be occupied by K⁺ at physiological pH in C. glutamicum, which would favor activation in a manner that is complementary to the Rb⁺ sites observed in an x-ray structure (11).

Modeling of cytoplasmic tails on the full-length BetP trimer

The primary goal of this work was to better understand the structural and dynamic nature of the two cytoplasmic tails of BetP and their interplay in regulation. For the C-terminal helix, considering both the EM structure and the simulation data shown above, it seems clear that the flat arrangement is preferred under nonactivating conditions in a lipid bilayer. This finding provides the foundation for exploring the conformational space of the remaining missing piece of the BetP protein, namely the N-terminal tail. Since the cytoplasmic tails of BetP are not conserved evolutionarily, de novo modeling is required to examine the structural features of the N-terminal tail. We first tested the machine-learning approach AlphaFold-multimer (30), which aims to predict structures of protein assemblies based on training on all the available data in the PDB. To minimize input from known structures, we excluded template structures deposited prior to the release of the first BetP structure (see materials and methods). Nevertheless, the C-terminal helical elements were modeled in the inclined orientation (Fig. S14 a), calling into question the predicted orientation of the N-terminal segments. These results suggest that during training of the neural networks, which included representatives of BetP, AlphaFold learned the orientation of the C-terminal tails present in the x-ray structures.

Given the lack of structural data for the N-terminal tail obtained using either crystallography or EM, it seems likely that there is a substantial proportion of disorder in this region. We therefore set out to understand whether there are predictable features of the N-terminal conformational space when modeled in the context of the EM-based BetP structure derived from our simulations. To this end, we used the de novo FloppyTail procedure (31) of the Rosetta package, which was designed precisely to explore the conformational space of flexible protein regions rather than to predict a single “true” structure.

Remodeling the C-terminal helix with the de novo protocol generates native-like structure

Before modeling the N-terminal tail, we benchmarked the ability of this protocol to describe a BetP terminal tail by remodeling the C-terminal helix and comparing it against the EM structure. Specifically, we generated 100,000 de novo models of the C-terminal tail in the context of the trimer.

Most of the C-terminal tail models obtained with FloppyTail are helical (Fig. S15 b), indicating that this protocol can reliably create secondary structure where needed. Even though most of these models adopt orientations that differ substantially from the EM structure (Fig. S15 a), a few models have RMSD values <5 Å, thus reproducing the flat, extended C-terminal helix conformation. Unfortunately, given the limited degree of packing between the terminal tails and the rest of the protein, there is no suitable energy function for reliably identifying the correct orientations from among the other possibilities. Accordingly, when modeling either tail, we do not expect to be able to identify a single “true” conformation from those present in the sampled models. Instead, to understand whether this de novo approach can be used to gain molecular insights into BetP regulation, we compared the predicted C-terminal tail contacts with those observed during our tens of microseconds of simulations.

As expected, the total count of contacts over the 100,000 models (Fig. S16) is lower than in the MD simulations (Fig. S11), since not all the C-terminal tail models are oriented horizontally on the surface of the trimer. Nevertheless, both in the models and in the simulations the C-terminal tail interacts consistently with loop2 of the same protomer and of the adjacent protomer in the clockwise position (Figs. S11 and S16). However, some contacts formed by the C-terminal tail in the simulations (especially with loop6 in the clockwise protomer, Fig. S11), are not found in any of the models. Vice versa, a few contacts were found in the models—between the C-terminal tail and the C-terminal helix of the counterclockwise protomer—that were not observed in the simulations. This discrepancy is likely due to the larger (although probably discontinuous) conformational space sampled in the models.

In summary, with this de novo strategy, we can reliably predict the secondary structure of a BetP terminal tail, and the full set of models is likely to include native states. In addition, some, but not all, of the predicted contacts are likely to represent native interactions. Thus, we focus on the aggregate features of the N-terminal tail models, which are likely to be most informative.

N-terminal models are diverse, but they share common secondary structure

After producing 100,000 models of the 55-residue-long N-terminal tail, we clustered them on the basis of structural similarity to assess whether one conformation was preferentially sampled. However, the clustering analysis showed that the conformational space is very wide; there is no clear set of consistently sampled structures (Fig. 8 a). Despite this conformational diversity, the secondary structure of the N-terminal tail models is similar across the whole set (Fig. 8 b), with two well-defined α-helices covering residues 21–27 and 42–49. Allowing for other types of helices, these structured segments could be considered to encompass residues 5–31 and 33–50, or around 64% of the tail.

Figure 8.

Structural features of the de novo predicted N-terminal tail of BetP. (a) Clusters of 100,000 models of residues 1–55 of the N-terminal tail, based on their Cα RMSD. For the four most populated clusters, the structure at the center of that cluster is shown as cartoon helices, viewed from the plane of the membrane. The transmembrane segments are colored three shades of gray, the C-terminal segments are orange, blue, and green, and the predicted N-terminal segment is colored yellow. (b) The secondary structure for all 100,000 N-terminal tail models, considering residues 1–55, computed using DSSP, plotted for every residue using the indicated color scheme. (c) Contacts of the modeled N-terminal tail residues 1–55 with individual residues in each of the three protomers of BetP, either intraprotomer (red bars) or interprotomer in the clockwise (blue bars) or counterclockwise (green bars) direction, assuming a view of the trimer from the cytoplasm. Contacts are defined as any two non-hydrogen atoms within 4.2 Å. Bottom panel: available contact surface in the cytoplasmic regions of the BetP trimer, excluding the N-terminal segment. Data are plotted as the mean (black) and standard deviation (green error bars) over the three protomers of the solvent-accessible surface area, relative to the maximum possible for each amino acid type in a GXG tripeptide. (d) Contacts between any residue of the modeled N-terminal tail (residues 1–55) and residues in the counterclockwise protomer. Points are colored according to the number of models in which that contact was observed as indicated by the legend. To see this figure in color, go online.

The N-terminal tail models do not interact with the adjacent clockwise protomer

The N-terminal tails are connected to TM1, which is located on the extreme periphery of the trimer and relatively far from the trimer interface (Figs. 1 and 7, a and b), raising the question of whether the N-terminal segments ever interact with the adjacent protomers. We therefore measured the contacts formed by any N-terminal tail model with the rest of the trimer. Despite the presence of several solvent-accessible regions on the cytoplasmic surface of the trimer (Fig. 8 c, bottom panel), only a few of those regions are contacted by the modeled N-terminal tails (Fig. 8 c, upper panels). Some intraprotomer contacts were observed with residues in loop2 (around residue 130) and loop10 (around residue 480), but most contacts are formed between the N-terminal tail and the C-terminal tail of the adjacent protomer in the counterclockwise direction (Fig. 8, c and d). Surprisingly, none of the models contains contacts between the N-terminal segments and the clockwise protomer, even though, as indicated by the C-terminal tail benchmark, the de novo technique used here is likely to sample a conformational space wider than the true accessible space.

Increasing the length of the modeled segment produced models with similar features

According to our benchmark in the C-terminal tail, the contact predictions may not be as reliable as the secondary structure of the models. We therefore wondered whether the lack of interaction of the N-terminal tail with the clockwise protomer reflects the choice of anchor point from which the N-terminal portion was modeled, i.e., residue 56, which was selected because it is present in most crystal structures of BetP. To test this possibility, we remodeled the N-terminal segment using the entire 60-residue segment up to the first helical residue of TM1. These models are also very diverse (Fig. S17 a) but nevertheless exhibit similar secondary structure (Figs. 8 b and S17 b, c, and f). Remarkably, also in this case, a negligible number of models contacts the clockwise protomer (Fig. S17, c and d), supporting the notion of directionality in the N-terminal tail interaction with the adjacent C-terminal helix.

Unstructured models of the N-terminal tail mainly contact the C-terminal helix of the counterclockwise protomer

The contacts adopted by the N-terminal tail models are governed by two factors: 1) the energetics of those interactions and 2) the steric and geometric constraints of the internal structure of the tail. In particular, the presence of helical segments shortens the available conformational space of the N-terminal tails. To further assess the robustness of the predicted interactions, we built another set of models in which we did not use fragments derived from known crystal structures to guide the backbone conformational space. This approach reduced the formation of secondary structure elements, and as expected the sampled conformations are very diverse (Fig. S18 a), with few and short secondary structure elements (Fig. S18 b). However, even using this extreme approach, most of the N-terminal tail contacts involve the C-terminal tail in the counterclockwise protomer, and almost none contact the clockwise protomer (Fig. S18, c and d).

In summary, the results of the modeling and the EM-derived simulations provide strong support for the possibility of cross talk between the N- and C-terminal regulatory tails of BetP, and, moreover, that this cross talk has bidirectionality, i.e., involves the tails from both protomers on either side of a given subunit. As mentioned previously, the N-terminal tail is probably disordered to some extent, precluding its complete resolution in the EM structure. However, the EM data do reveal a distinct density connected to TM1 that points toward the counterclockwise C-terminal tail only (see Fig. S9 d in (11)), supporting the interaction framework proposed by the de novo modeling. In this framework, each C-terminal helix contacts several different important regulatory segments, namely loop2 of its own protomer, as well as cytoplasmic loops (loop2 and loop6) and the N-terminal tail of the clockwise protomer (Figs. 7, 9, S17, and S18).

Figure 9.

Interactions of the modeled N-terminal tail and a hypothesis for their role in activation. (a) Contacts within the N-terminal tail (residues 1–55) and between the N-terminal tail and the C-terminal tail (residues 548–586) plotted as the proportion of the 100,000 de novo models (blue scale bar). Contacts are defined using a coordination number (see materials and methods) for non-hydrogen atoms of the C- and N-terminal tails versus a coordination number between N-terminal helix 1 (residues 10–31) and helix 2 (residues 36–49). Clusters based on these two variables are mapped onto the 2D plot, with the size of each cluster indicated by the size and purple shading of the circles. (b) Structures of the four most populated clusters in the space of these two variables, viewed along the plane of the membrane. The variance of the models is indicated using surface representation (yellow), while the N-terminal tail model with the greatest degree of helicity and with the corresponding coordination numbers is shown as yellow cartoon helices. The transmembrane segment of the protomer containing the modeled N-terminal tail is shown as a dark-gray transparent surface, with helices in white (transmembrane segments) and orange (C-terminal tail). The other two protomers are shown as white surfaces, with the C-terminal tail of the counterclockwise protomer shown as a green cartoon. (c) Schematic of likely intra- and intertail interactions predicted by the modeling. The fully downregulated state (top left) is expected to resemble that observed in EM measurements in amphipols and low K⁺ concentrations, which is stable in MD simulations. The lack of resolution of the N-terminal tail under these conditions suggests that it is either fully unfolded or connected by a flexible linker, as in clusters 1 and 4. Activation may involve a shift in the conformation of one or more C-terminal helices away from loop6 and loop2b, perhaps toward the trimer midpoint, similar to its orientation in x-ray crystal structures; in this conformation, the helical segment beyond the breakage point (C2) is likely to be more dynamic (bottom panels). Additionally, an extended conformation of the N-terminal helix is predicted to form contacts with the C-terminal domain, as in clusters 2 and 3, suggesting activated state configurations such as those on the right. To see this figure in color, go online.

Two major modes of interaction predicted for the N-terminal tails

The consistency of the predicted contacts across different modeling protocols suggests that some of the residues involved may be important for regulation. In all three sets of models, the most likely contacts involve residues 42–50 of the N-terminal tail, a Glu-rich region, which frequently forms salt-bridge interactions with the basic residues in the C-terminal helix (Figs. 8 d, S17 d, and S18 d). For example, Glu50 contacts Arg582 in 16% of the 55-residue-long N-terminal models. The predominant role of residues 42–50 is perhaps puzzling, since deletions of the first 20 residues of the N-terminal tail showed the same regulatory effect as the truncation of the whole domain (6), implying that only residues 1–20 are essential for the N-terminal modulation of BetP activity. We therefore considered the possibility that the cross talk between the cytoplasmic tails is not occurring via a few strongly interacting pairs of residues but is instead an accumulation of interactions along the two tails. As previously mentioned, the models of the N-terminal tail consistently predicted two helical regions, separated by an unstructured, potentially hinge-like region. Thus, these helices might act as two independent entities that either contact each other, if folded together, or contact the C-terminal tail, if extended. To test this hypothesis, we plotted the contacts formed between the N- and C-terminal tails against the interactions between the two N-terminal helices for each model (Fig. 9 a). Most of the N-terminal tail models either form contacts with the C-terminal helix or contain contacts between the two N-terminal helices, or neither of these two options. Only rarely can the N-terminal helices contact both the C-terminal tail and each other.

To obtain a molecular-level picture of this effect, we clustered the models in the space of these two contact variables. The most populated cluster (cluster 1 in Fig. 9, a and b) represents extended N-terminal tails occupying spaces away from the rest of the BetP trimer. We consider this an unlikely possibility, since truncations alter regulation profiles (4). In the second and third most populated clusters, the structures lack contacts between the N-terminal helices but have many interactions with the C-terminal tails; here, the N-terminal tails are more extended and lie closer to the C-terminal tail. Lastly, the N-terminal models of the fourth most populated cluster are more compact, as if forming a separate domain involving intratail contacts and lacking interactions with the C-terminal tail.

Discussion

In this work we set out to gain insights into the structural features of the regulatory N- and C-terminal domains of a LeuT-fold transporter protein, BetP, by a comparative simulation study of alternative reported conformations of the C-terminal tails and by de novo modeling of an ensemble of conformations of the N-terminal tails, focusing on the downregulated state, which appears to be less dynamic (11) and therefore more accessible computationally than the activated state(s).

Of the two domains, the C-terminal segment is more essential for activity regulation. Hence, we first used MD simulations to assess the stability of two orientations of this segment observed under downregulating conditions: one derived from x-ray crystallographic data and another from EM data. The C-terminal domain derived from the EM data obtained in low K⁺ concentrations is stable for up to 8 μs in a PG lipid bilayer, suggesting that it corresponds to a downregulated conformation of the transporter. By contrast, the simulations of the conformation observed in the x-ray structure gave a more mixed picture: the most C-terminal portion of this helix, which is involved in contacts with other trimers in the crystal lattice, was not stable in any of our simulations. No unique conformation, nor even a distinct conformational ensemble, was obtained for this segment. However, the portion of the C-terminal segment directly linked to the TM region partially retains the conformation observed in the crystal structure. In this orientation, this segment spans the interior cytoplasmic surface of the trimer far from the lipid bilayer and has been suggested to reflect a (pre-)activated conformation, because the trimer arrangement is both asymmetric and contains contacts involving residues known to be essential for regulation (11).

Simulations here were carried out using an anionic bilayer composed of a single lipid type to avoid artifacts due to lipid mixing times. Of the three major C. glutamicum lipid components—all anionic—only POPG lipids were resolved at multiple positions in the x-ray structures, consistent with POPG enrichment despite protein purification and crystallization (13). These crystallographic lipids were incorporated in our simulations to preserve those interactions. The lack of cardiolipin or 1-palmitoyl-2-oleoylphosphatidylethanolamine (POPE) may have consequences for the interactions observed, especially since cardiolipin could be assigned to the higher-resolution cryo-EM map for BetP (11). However, we note that cardiolipin, also known as diphosphatidylglycine, can be partially mimicked by POPG, which can therefore occupy much of the same sites as cardiolipin. Future work should examine such interactions as well as any possible consequences of lipid mixing on the global biophysical characteristics of the bilayer.

Structures of BetP reveal relatively few contact points for the C-terminal tails, many of which are electrostatic in a hydrophilic environment, suggesting that simulations with 300 mM K⁺ might suffice to reveal binding to, or initial destabilization of, one of the two conformations. The timescale of the full effect of K⁺ seems to be significantly slower than these multi-microsecond simulations, but it is also possible that, in the absence of the N-terminal domain, BetP may require an even higher concentration of K⁺ for activation (4,6). Furthermore, the PG lipids used to construct the bilayer are associated with a shift toward higher values of the osmotic stress profile, potentially strongly downregulating the transporter compared with E. coli cells or a C. glutamicum lipid mixture (40). Unexpectedly, the simulations suggest some occupancy of K⁺ ions in a small pocket adjacent to the cytoplasmic pathway comprising Asp470 in TM10 on one side and loop6 (between TM6 and TM7) on the other side. Loop6 is notable for the numerous contacts it forms with the C-terminal helix in the downregulated conformation of the protein (Table S3 and Fig. 7). Asp470 was previously predicted to experience a shift in protonation state depending on the conformation of the transport domain, and on the presence of a Na⁺ ion bound at the so-called Na2 site in the core of the transporter (15). In the inward-facing conformation used here, Asp470 was predicted to be deprotonated at pH 7; therefore, in our simulations of the trimer, Asp470 was a priori assigned to be deprotonated, based on which one might expect that cations will inevitably bind there at the high concentrations of K⁺ used (300 mM). Nevertheless, no other negatively charged residues on the surface of the protein experienced a similar frequency of cation interactions. We note that the cation interactions with Asp470 were not specific to K⁺ ions, as Na⁺ ions were also found to bind in this pocket, albeit with a lower rate, perhaps consistent with its lower concentration (200 mM). Similarly, the aforementioned shifts in predicted protonation states indicate that protons might also compete for this site. Indeed, at pH 5.5, which is required for crystallization, we expect a significant population of the protonated form of Asp470, which may explain why no density for Rb⁺ was observed at this region in crystals obtained with 300 mM RbCl (11). Taken together with the high basal levels of K⁺, low levels of Na⁺, and a neutral pH in the cytoplasm of C. glutamicum (41,42), Asp470 may play a role in activation by K⁺ ions in addition to those observed to involve the C-terminal tail (11).

Building on our understanding of the C-terminal helix orientation in the downregulated state we explored possible conformations of the N-terminal tail, which has a modulatory action on the response of BetP to osmotic stress. The absence of structural data for the N-terminal tail, even in the EM experiment where it is present in the protein construct, could suggest that it forms no consistent structure or contacts with the rest of the protein. However, previous studies revealed that modified BetP proteins with truncations of the first 20, 29, 32, 49, 52, 53, or 60 residues (the latter of which removes the entire N-terminal segment) all require higher osmolalities for activation and have a reduced maximum velocity (6,8), indicating either that the downregulated state(s) is further stabilized by the absence of this domain or that the activated state(s) is more readily formed in its presence. Moreover, peptides with the N-terminal sequence compete with the native attached N-terminal domain to limit the ability of the transporter to reach maximal turnover levels. The nature of its role, from a structural perspective, is therefore puzzling. We can rule out that the N-terminal segment interacts primarily with the membrane surface, based on surface plasmon resonance analysis (8) as well as high density of negative charges in both the lipid headgroups and the N-terminal tail. Therefore, we think it likely that the N-terminal tail does interact with the rest of the protein at some stage during activation, and presumably with the C-terminal tail(s), given its density of positive charges and observed interactions in peptide arrays (8). Here, instead of proposing a unique model of the structure of the N-terminal domain, we identified common features across an ensemble of models, repeatedly challenging our findings by varying the modeling strategy. The models consistently indicated interactions between the N-terminal tail of one protomer and the C-terminal helix of the adjacent protomer in the counterclockwise position (Fig. 8 c), a notion supported by some density assigned to the first portion of the N-terminal tail in the EM structure (11). By contrast, the most stable configuration of the C-terminal tail interacts with the adjacent protomer in the clockwise direction. Therefore, each protomer appears to contact both of its neighbors in the trimer, one through its N-terminal domain and the other via the C-terminal helix, suggesting a strong bidirectional interplay between the three subunits (Fig. 9 c).

That truncations of the first 20, 32, 49, or 60 residues all led to similar shifts in activation curves relative to the wild-type protein (6) might be interpreted as indicating that only the first 20 residues form important contacts with the rest of the protein. Instead, we suggest here, based on the N-terminal tail models, that any complexes between the N- and C-terminal domains involve numerous, weak contacts, including those in the first 20 residues. Consequently, elimination of a few residues in the beginning of the N-terminal segment could disrupt the interaction between the N- and C-terminal domains. A need for many, weak, cooperative interactions between the terminal domains may be an advantage for the regulation mechanism. In this way, disrupting only a few residue-residue contacts (e.g., by competition with cations or changes in local membrane structure) may suffice to completely disengage the N- and C-terminal complex, which in turn may activate BetP transport (Fig. 9).

Taken together, although additional mechanistic details remain to be established especially with respect to the role of the membrane in osmoregulation, these computational studies provide insights into the structure and dynamics of the BetP trimer and its regulatory domains under downregulating conditions.

A modeling strategy similar to that used for the N-terminal domain was used previously for the cytoplasmic N- and C-terminal tails of the human serotonin transporter (hSERT), predicting that they contain conserved secondary structural elements (43). Here, using FloppyTail, we could explicitly consider the possible unstructured nature of the N-terminal segment and its context in the trimer, allowing the prediction of interaction patterns between the modeled segment and the rest of the protein. Such interactions between cytoplasmic segments and the TM core are likely to be of interest for many other membrane proteins whose terminal segments are not resolved in structural studies but play important roles in membrane trafficking, activity regulation, and post-translational modification (44,45,46,47). Therefore, although the proposed regulatory interactions presented here are likely to be specific to BetP, we expect that these computational strategies will be useful for obtaining structural insights into other membrane proteins.

Author contributions

Conceptualization, V.L. and L.R.F.; investigation, V.L., R.T.B., C.K., C.F.-F., and P.S.L.; methodology, data curation, and writing – original draft, V.L.; formal analysis and software programming, V.L., R.T.B., C.K., and P.S.L.; writing – review and editing, all authors; visualization, V.L., C.K., and L.R.F.; supervision, V.L. and L.R.F.; project administration, resources, and funding acquisition, L.R.F.

Editor: Michael Grabe.

Data availability

The input coordinate files and some analysis are available as Zenodo data: https://doi.org/10.5281/zenodo.5600538, https://doi.org/10.5281/zenodo.5600863.