An intracellular pathway controlled by the N-terminus of the pump subunit inhibits the bacterial KdpFABC ion pump in high K⁺ conditions

Abstract

The heterotetrameric bacterial KdpFABC transmembrane protein complex is an ion channel-pump hybrid that consumes ATP to import K⁺ against its transmembrane chemical potential gradient in low external K⁺ environments. The KdpB ion-pump subunit of KdpFABC is a P-type ATPase, and catalyses ATP hydrolysis. Under high external K⁺ conditions, K⁺ can diffuse into the cells through passive ion channels. KdpFABC must therefore be inhibited in high K⁺ conditions to conserve cellular ATP. Inhibition is thought to occur via unusual phosphorylation of residue Ser162 of the TGES motif of the cytoplasmic A domain. It is proposed that phosphorylation most likely traps KdpB in an inactive E1-P like conformation, but the molecular mechanism of phosphorylation-mediated inhibition remains unknown. Here, we employ molecular dynamics (MD) simulations of the dephosphorylated and phosphorylated versions of KdpFABC to demonstrate that phosphorylated KdpB is trapped in a conformation where the ion-binding site is hydrated by an intracellular pathway between transmembrane helices M1 and M2 which opens in response to the rearrangement of cytoplasmic domains resulting from phosphorylation. Cytoplasmic access of water to the ion-binding site is accompanied by a remarkable loss of secondary structure of the KdpB N-terminus and disruption of a key salt bridge between Glu87 in the A domain and Arg212 in the P domain. Our results provide the molecular basis of a unique mechanism of regulation amongst P-type ATPases, and suggest that the N-terminus has a significant role to play in the conformational cycle and regulation of KdpFABC.

1. Introduction

KdpFABC is a unique K⁺ ion transporting complex expressed in bacteria under low external K⁺ concentrations. Bacterial survival under such conditions hinges upon sufficient inward transport of K⁺, and the KdpFABC complex consumes ATP to transport K⁺ against its concentration gradient into the bacterial cytoplasm. The protein thus maintains the requisite high intracellular concentration of K⁺ in low K⁺ growth conditions [1]. Under high external K⁺ conditions, the pump is not required because K⁺ can diffuse into the cell through constitutively expressed K⁺ channels. KdpFABC is a heterotetramer, and the two main functional transport subunits are the ion-channel like subunit KdpA and the pump like subunit KdpB. The KdpA subunit most likely descends from the bacterial KscA family of ion channels, while KdpB clearly belongs to the P-type ATPase family of ion pumps which also include the Na, K ATPase, Ca²⁺-ATPase (SERCA) and the gastric H⁺, K⁺ ATPase [2]. KdpFABC is therefore a one-of-a-kind, unique channel-pump hybrid and an evolutionary outlier. In ion pumps such as SERCA, the ion-binding sites are located in the transmembrane region. The protein undergoes transitions between the so-called E1 and E2 states [3, 4], which face the two different sides of the membrane and have differential affinities for different ions, making the ion-binding site more selective towards the ion present at a lower concentration. The conformational transitions between the E1 and E2 states are driven by the binding of ions to the TM ion-binding sites, and by the ATP-driven phosphorylation and dephosphorylation of the cytoplasmic P domain. ATP binds to the nucleotide-binding N domain which is an internal kinase and phosphorylates the aspartate residue. Finally, the actuator A domain uses a TGES motif to catalyse dephosphorylation of the aspartate residue in the second half of the cycle [5]. The dephosphorylation and phosphorylation of the P domain are accompanied by significant rearrangements in the relative orientations of the cytoplasmic domains, and this rearrangement drives reorientation of the TM helices. The rearrangement of the TM helices opens intracellular or extracellular pathways leading to the ion-binding sites, through which ions bind, or are released. In this way, ions are transported across the membrane through the TM region. For a more detailed review of the conformational changes which occur during the complete transport cycle of P-type ATPases, the reader can consult several outstanding reviews on the subject [6, 7, 5].

In the case of KdpFABC complex, both crystal [8] (Figure 1) and cryo-EM structures [9] show that the Serine residue (Ser162) in the TGES motif is phosphorylated [10]. The functional effect of Serine phosphorylation was shown to be inhibitory, which is a startling contrast to the well established role of the preceding Glutamate residue (Glu161) in catalysing hydrolysis of the phosphorylated aspartate residue. The physiological role of serine phosphorylation has now been traced to the inhibition of KdpFABC when cells are transferred to a high K⁺ medium [11, 12, 10]. Inhibition of the KdpFABC pump under these conditions [11, 12] avoids ATP hydrolysis and the corresponding waste of cellular resources when homeostasis can be achieved by passive ion channels without expending energy. Recent enzymatic studies showed that the Ser162Ala mutation makes the protein constitutively active, while the phosphorylation mimicking Ser162Asp mutation inactivates the protein [10]. Whereas the Glu161Gln mutation traps the pump in the ADP-insensitive E2-P conformation, Ser162 phosphorylation or the Ser162Asp mutation traps the ADP-sensitive E1~P state [10]. Most importantly, modification of Ser162 causes the pump and channel subunits to become uncoupled, suggesting that Ser162 has long-range allosteric effects that influence the conformational flexibility and transitions of the pump during the transport cycle. Here, we employ Molecular Dynamics (MD) simulations of the unphosphorylated (SER) and phosphorylated (PHOS) Ser162 versions of KdpFABC to show that phosphorylation drives the opening of an intracellular water-filled pathway leading to the KdpB ion-binding sites, through TM helices M1, M2 and M4. The access to this pathway is controlled by the interactions of the N-terminus of KdpB with the A domain. We propose that the opening of this previously unknown intracellular pathway is responsible for uncoupling and for trapping KdpFABC in one of the afore-mentioned phosphorylated EP states, thus applying brakes on the pump catalytic cycle and on ATP hydrolysis.

Figure 1:

Starting setup of the simulations performed in this work, showing the crystal structure of the KdpFABC heterotetramer complex, pdb id 5MRW. The channel KdpA channel subunit is coloured green while the KdpC and KdpF subunits are orange. A water molecule (red sphere) is visible in the KdpB pump canonical ion-binding site, and a K⁺ ion (purple) is resolved in the KdpA subunit. Residue Ser162 of the TGES motif in the A domain is phosphoryated. Lipid headgroups are coloured tan. For clarity, lipid tails, hydrating water molecules on both sides of the membrane and bulk ions are not shown.

2. Results and Discussion

In the following discussion, we first discuss and quantify the mobility of the cytoplasmic domains in phosphorylated KdpB. We use Principal Components Analysis (PCA) to show that the cytoplasmic domains occupy several conformations in the PHOS systems, but only one in the SER systems. We then show that the dynamics of the cytoplasmic domains open a cytoplasmic water pathway leading to the ion-binding site of KdpB. Finally, we show that the opening of the water pathway is the result of key conformational changes occurring in the N-terminus and at the interface of the membrane domain and the A-domain of KdpB.

2.1. Cytoplasmic domains are highly mobile in phosphorylated Kdp

The simulations reveal distinct and reproducible differences in the behavior of KdpB that are attributable to serine phosphorylation. The first is the mobility of the cytoplasmic domains of KdpB. In particular, the average root mean squared deviations of each cytoplasmic domain relative to their starting positions are significantly higher in PHOS simulations, which can be seen in the RMSD distributions for the individual domains (Figure 2). The RMSD fluctuates between 0.2 and 0.6 nm for both systems, suggesting an overall stable tertiary structure in the simulations (Figure 2A). The higher deviations from the starting structure reflect that the protein samples more conformations in the PHOS simulations than in the SER simulations. The RMSD of the entire protein is shown in Supplementary Figure 1.

Figure 2:

(A) The root mean squared deviation of the cytoplasmic domains of the protein backbone of the PHOS and the SER simulations over 10 different trajectories of 200 ns each. (B) (C) and (D) A histogram of the RMSD of the A, N and P domains. All three cytoplasmic domains deviate more from the starting structure in the PHOS simulations.

The possible functional relevance of this is discussed in the following sections. A covariance and principal components analysis further elucidates the differing dynamic behavior in PHOS and SER simulations. The first two eigenvectors of the covariance matrix of atomic fluctuations represent the largest deviations in the dataset (Figure 3) and are used to analyze the global collective motions. For this analysis, each protein conformation sampled during the simulations was projected upon the two dimensional space formed by the first two eigenvectors to obtain a 2D-scatter plot (Figure 3A). In addition, a histogram representing the 1D motion along each eigenvector is shown along the margin of each axis. This projection map reveals several clouds which correspond to distinct protein conformations sampled during the simulations. These clouds are likely to represent free energy minima in the protein conformational landscape which are clearly different for PHOS (blue) and SER (orange) simulations. The analysis is similar to what we performed earlier in the context of a different membrane protein [13]. More specifically, this analysis reveals novel conformations, labeled C and D, that are sampled in the PHOS simulations, whereas the SER simulations remain within the larger cloud of conformations labeled B. When a PCA analysis is performed on each of the 10 PHOS simulations independently, conformations similar to C are observed in 8 PHOS simulations, while conformation D is observed in 4 PHOS simulations (Supplementary Figures 2 and 3). A k-means clustering algorithm was used to extract representative conformations from each conformational cluster, which are illustrated in panels B, C and D of Figure 3). In conformations C and D, sampled only in the PHOS simulations, the A-domain (yellow) uncouples from the P-domain (blue), and collapses towards the membrane.

Figure 3:

Principle component analysis of conformations sampled during the simulations. (A) Projection of individual frames from the simulation onto the first two eigenvectors (PC1 and PC2) from the covariance matrix. The prevalent conformation represented by the cloud labeled B is sampled by both PHOS and SER simulations. However, the PHOS simulation has sampled novel conformations represented by the peripheral clouds labeled C and D. Projections onto each individual axis are shown along the top and right margins. (B-D) Structures extracted by k-means analysis of panel A from clouds B, C and D, respectively. Cytoplasmic domains are colored as follows: A-domain yellow, N-domain red, P-domain blue.

Trajectories projected upon the first eigen vectors for the PHOS and SER simulations are shown in Movies S1 and S2.

2.2. Phosphorylation opens an intracellular water pathway leading to the KdpB ion-binding site

The second key difference involves the canonical ion-binding site of P-type ATPases formed near Pro264 on the M4 helix. This site is occupied by a water molecule in the crystal structure of KdpB (Figure 1). In the PHOS simulations, this water molecule escaped almost immediately, whereas it remains bound for the duration of the SER simulations.

This escape is facilitated by the opening of an aqueous channel in KdpB from the cytoplasmic side of the membrane (Figure 4, and supplementary movie S3). As a result, the Pro264-bound water molecule is frequently exchanged with a different water molecule in the aqueous channel. In contrast, water never reaches the canonical binding site and the initial water molecule remained occluded in any of the 10 SER simulations. (Figure 4). To quantify the penetration of water into this site, the radial distribution function g(r), which is a normalised histogram of pair-wise residue distances, is calculated for all water molecules relative to the center of mass defined by Pro264, Thr265, Asp583, and Lys586, which surround the site and coordinate the water molecule in the crystal structure (Figure 4A). These data indicate that this intramembrane site becomes highly hydrated when Ser162 on KdpB is phosphorylated, but not for the unphosphorylated Kdp complex.

Figure 4:

Water accessibility of the canonical ion binding site in KdpB. (A) Radial distribution functions between water and the center of mass of residues Pro264, Thr265, Asp583 and Lys586 from KdpB, which surround the canonical ion binding site. The radial distribution function corresponds to a normalized histogram of distances between two atom selections, averaged over the time of the simulation. (B) Initial configuration of the canonical site for both the SER and PHOS simulations with a water molecule (blue) bound at the location seen in the X-ray structure. (C) and (D) Final frames from PHOS and SER simulations respectively. The canonical ion-binding site is significantly more hydrated in the PHOS simulations with water penetrating the site from the cytoplasmic side of the membrane (top of the panel). Oxygen atoms of water molecules have been shown in red surface representation with a probe radius of 0.15 nm.

The water pathway opens between helices M1 and M2, and in the case of the SER simulations, the pathway is closed by a salt bridge between Glu87 on the A-domain and Arg212 on the P-domain (Figure 5D). The salt bridge is broken in the PHOS simulations as a result of the large scale motions of the cytoplasmic domains, and in this way separating helices M1 and M2, and opening the water channel. An intracellular pathway between helices M1, M2 and M4 is also postulated for the sarcoplasmic reticulum Ca²⁺-ATPase [14], and was observed in the first crystal structure of SERCA [15], as well as in inhibitor-bound structures of SERCA [16]. The M1 helix probably plays a role in the occlusion of the bound Ca²⁺ in SERCA [17, 18]. Palytoxin, which converts the Na+, K⁺ ATPase to a channel-like protein, opens a continuous ion-pathway which also passes between M1, M2, M4 and M6 on the cytoplasmic side [19]. Whether or not the observed pathway leading to the ion-binding sites in our simulations of the KdpFABC comprise of a functional pathway for K⁺ release into the cytoplasm remains to be seen. However, it is clear that similar pathways are linked to functional roles in other P-type ATPases. A cytoplasmic exit pathway was proposed in cryo-EM structures [9], but is different from the pathway observed in the current set of simulations. It is possible that crystal and cryo-EM structures only sample a subset of the large conformational landscape sampled by the phosphorylated pump. So far, no structure of the active dephosphorylated pump has emerged.

Figure 5:

Interactions between M1 and M2 control entry of water into the canonical ion binding site. (A) Snapshot from the PHOS simulations shows that Glu87 and Arg212 are displaced and a water pathway is detected by CAVER (green surface) that leads to the canonical binding site. (B) Snapshot from the SER simulations shows the proximity of Glu87 on M1 and Arg212 on M2. The canonical binding site is below and no water pathway is present. (C) Radial distribution function between the side-chain center of masses of Glu87 and Arg212. Higher peaks for SER simulations indicate stronger interactions between these residues in this system, whereas the lower peaks for PHOS simulations indicate disruption of these interactions. (D) The helical content of the linker between M2 and the A-domain (residues 68-92) is significantly lower for the PHOS simulations indicating unwinding of this structural element.

2.3. Hydration of the ion-binding site is accompanied by changes in the conformations of the N-terminus

The opening of the intracellular pathway leading to the ion-binding sites is accompanied by remarkable changes in the secondary structure of the N-terminus of the KdpB. In the SER simulations, residues 11 through 23 remain helical most of the time during the simulations. In the PHOS simulations, however, the helical content of these residues is nearly completely lost (Figure 5D). The conformations of the N-terminus relative to the ion-binding site are shown in (Figure 1). In the SER simulations, the helical N-terminus is linked to the M1 via a random-coil linker comprising of residues 26 through 34, with 27 and 34 being helix-breaking Proline residues. Part of the N-terminus remains membrane-bound, but the rest projects into the cytoplasm to interact closely with residues 83 through 98 on the M2 helix which also protrude into the cytoplasm (Figure 6C). In the PHOS simulations however, the N-terminus is fully unfolded, and collapses parallel to the membrane. In addition, the M2 helix also unfolds near residues 87 through 92, breaking the salt bridge between Glu87 on M2 and Arg212 on top of helix M4. These data are depicted in Figure 6, where it is clear that the distance between the N-terminus and the residues 87 through 92 fluctuates significantly, and is higher in the PHOS simulations, compared to the SER simulations. These data, combined with the earlier claim that the intracellular pathway between helices M1, M2 and M4 could play a functional role, suggest that the N-terminus of KdpFABC has an important role to play in the conformational cycle of the pump, and may even regulate the release of the K⁺ ion into the cytoplasm. The N-terminus was previously shown to regulate the E1-E2 conformational equilibrium of the Na⁺, K⁺, ATPase by interacting with the cytoplasmic domains [20, 21], and it has since been hypothesized that the N-terminus may play a role in the regulation of the conformations of both the Na⁺, K⁺, ATPase and the H⁺, K⁺, ATPase [22]. The unfolding of the N-terminus is reproducible across all PHOS simulations (Supplementary Figure 4).

Figure 6:

(A) The average distances between the center of mass of the N-terminus (residues 9 through 25) and the cytoplasmic portion of the M2 helix (residus 87 through 92) over all 10 simulations. Except in the first simulation replica, the distance increases and fluctuates more for all PHOS replicas compared to the SER replicas. (B) The histogram of distances shown in A. C and D: The interactions between the N-terminus (green) and residues 83 through 98 (orange) in the A domain in the SER and PHOS simulations respectively. The N-terminus is fully unfolded in the PHOS simulations, but remains helical in the SER simulations. The color coding for the cytoplasmic domains is the same as Figure 1.

The N-terminus lies close to the membrane in the SER simulations. Radial distribution functions (Supplementary Figure 5) indicate that the interactions of the N-terminus are significantly weaker in the PHOS simulations where the N-terminus unfolds. Since the phoshorylation-induced changes allosterically influence the N-terminus via alterations in the conformations of the cytoplasmic domains, we think that lipids only play a passive role in the process, if any.

Taken together, there are several dynamic phenomena that occur in the PHOS simulations, and which are absent in the SER systems, which lead to the formation of the water pathway leading to the ion-binding sites. These changes are: the unfolding of the N-terminus, the unfolding of the M2 helix between residues 86 and 91, the large-scale movement of the cytoplasmic domains and the broken salt bridge between Glu87 and Arg212. We are reluctant to ascribe any of these as a single potential decisive step which lead to hydration of the binding site. Current models of allostery rather emphasize the dynamic nature of many small conformational changes which add up to altered dynamics of a conformational ensemble of a protein upon binding of an allosteric ligand. We advocate this ensemble view of allostery according to which reactions such as phosphorylation can regulate protein function allosterically without gross conformational changes. With regard to the enzymatic effects of serine phosphorylation, the observed uncoupling of aspartyl phosphorylation [10] could be explained by the enhanced accessibility of the canonical ion binding site in KdpB which could provide access for cytoplasmic ions to this site. Since these ions are able to avoid the normal entry pathway through the selectivity filter of KdpA, then selectivity of K⁺ over other monovalent cations (e.g., Na⁺ or H₃O⁺) would be lost.

To conclude, we have carried out MD simulations of phosphorylated and dephosphorylated versions of the bacterial KdpFABC complex to propose that the phosphorylation inhibits the protein by opening an intracellular water pathway leading to the ion-binding sites in the KdpB pump subunit. We also hypothesize that the N-terminal residues guard the entrance to the ion-binding site in the absence of Ser162 phosphorylation, and in this way, the KdpFABC N-terminus may have a significant role to play in the regulation of the pump. Since it can control the entry of water (and ions) to the pump subunit ion-binding site, the N-terminus is also likely to be involved in the regulation and kinetics of the overall conformational cycle of the pump. Our hypotheses can be experimentally validated or refuted by simple mutational experiments such as the measurement of pump activity and ATP turnover in a mutant where the N-terminus is truncated. Our simulations are not very long by current standards. However, all simulation replicas point to the same observation with respect to the hydration of the ion-binding site, lending confidence to our sampling protocols. Although individual simulations are 200 ns each, the total sampling in the simulations is 2 μs for each system. We do not observe a transition from the unfolded N-terminus to a folded N-terminus in any of our simulations, because the transition is likely to occur on a longer time scale. One way to calculate the probability of such a transition would be to calculate kinetic rate constants using Markov state Models, but at least an order of magnitude longer trajectories would be required for this to be accomplished. Simulations using cryo-EM structures as starting models will be implemented in the future, although these structures have a lower resolution than the crystal structure used as a starting model in the current simulations.

3. Materials and Methods

To set up the system for simulation, a bilayer containing 550 lipid molecules was constructed using CHARMM-GUI with POPC and POPE in a 4:1 ratio, as a simple mimic of the bacterial membrane. The published crystal structure for KdpFABC (PDB: 5MRW), including the bound potassium ion as well as the protein, was introduced in the bilayer with the help of the OPM server and the system was hydrated with 75,000 water molecules. We used the crystal structure because it had the highest resolution. 150mM of salt (KCl) was added to the aqueous phase. The protonation states of the titratable amino acids were initially set by PROPKA [23]. Several residues had a pKa higher than 7, and were therefore protonated. However, a short 50 ns simulation run showed that all acidic amino acids had a pKa values well below 7.0. With this in mind, we kept all acidic amino acids deprotonated during the production simulations. Ten independent simulations of 200 ns each were launched for KdpFABC with and without serine phosphorylation, referred to as PHOS and SER simulations respectively.

All-atom MD simulations were performed under periodic boundary conditions using GROMACS version 2016.3 [24, 25, 26, 27, 28]. 10 replicates with different initial velocities were simulated for both the PHOS and the SER simulations. We used the TIP3P water model with Lennard-Jones interactions on hydrogen atoms. A 1.2 nm cutoff was used for non-bonded neighbour list, and was updated every 10 steps. The van der Waals interactions were switched off between 1.0 to 1.2 nm. Electrostatic interactions were treated with the particle mesh Ewald (PME) method [29, 30]. All systems were minimised with 5,000 steps using steepest decent algorithm. For each simulation, the converged system was equilibrated for 5 ns with backbone restraints on the protein followed by a production run of 200 ns. The temperature of the system was kept at 310K with the Nosé-Hoover thermostat [31, 32] after an equilibration run which was performed with the Berendsen thermostat [33]. The pressure was kept at 1 bar with the Parrinello-Rahman barostat [34] along with a semi-isotropic pressure coupling scheme. The Linear Constraint Solver (LINCS) [35] algorithm was used to constrain all bonds containing hydrogen. A 2 fs time step was used and trajectories were sampled every 20 ps. The data analysis was carried out using GROMACS and homemade scripts. The analysis was performed on the concatenated trajectories from the 10 individual simulations. Snapshots were rendered using Visual Molecular Dynamics (VMD)[36]. Data analysis was performed on a single concatenated trajectory for each of the three systems, unless otherwise mentioned. Trajectories for the center of mass of different domains were extracted by MDAnalysis tools [37, 38] and distance and angular data were calculated by homemade scripts.

Covariance and principal components analysis (PCA) was performed using GROMACS. After removing center of mass translation and rotation, the covariance analysis was performed for the protein backbone atoms. First, a covariance matrix C is extracted from atomic fluctuations

$$C=\langle (\mathbf{x}(t)-\langle \mathbf{x}\rangle )(\mathbf{x}(t)-\langle \mathbf{x}\rangle {)}^T\rangle$$ (1)

where x is a 3N-dimensional column vector describing the coordinates of the N protein backbone atoms, and x(t) are the positions at time t. The triangular brackets indicate an ensemble average. C is a 3N x 3N symmetric matrix, and is then diagonalised by a coordinate tranformation R

$$C=R\Lambda R^T$$ (2)

$$\Lambda =R^{T}CR$$ (3)

This orthogonal transformation transforms C into a diagonal matrix Λ = diag(λ₁, λ₂, … λ_3N), which contains the eigenvalues λ_i of the covariance matrix C. The ith column of R contains the ith eigenvector r_i with the corresponding eigenvalue λ_i.

The global collective motions of the trajectory can then be obtained by projecting the trajectory ensemble onto individual eigenvectors, to obtain the principal components p_i, i = 1, 2, …N by taking an internal product between the transpose of the eigenvector and the atomic fluctuation.

$$p_i(t)={\mathbf{r}}_i\cdot (\mathbf{x}(t)-\langle \mathbf{x}\rangle )$$ (4)

Each trajectory snapshot, thus projected on two different eigenvectors, yields one point on the point cloud distributions shown in Figure 3.