Allosteric Priming of E. coli CheY by the Flagellar Motor Protein FliM

Abstract

Phosphorylation of Escherichia coli CheY protein transduces chemoreceptor stimulation to a highly cooperative flagellar motor response. CheY binds to the N-terminal peptide of the FliM motor protein (FliM_N). Constitutively active D13K-Y106W CheY has been an important tool for motor physiology. The crystal structures of CheY and CheY ⋅ FliM_N with and without D13K-Y106W have shown FliM_N-bound CheY contains features of both active and inactive states. We used molecular dynamics (MD) simulations to characterize the CheY conformational landscape accessed by FliM_N and D13K-Y106W. Mutual information measures identified the central features of the long-range CheY allosteric network between D57 phosphorylation site and the FliM_N interface, namely the closure of the α4-β4 hinge and inward rotation of Y- or W106 with W58. We used hydroxy-radical foot printing with mass spectroscopy (XFMS) to track the solvent accessibility of these and other side chains. The solution XFMS oxidation rate correlated with the solvent-accessible area of the crystal structures. The protection of allosteric relay side chains reported by XFMS confirmed the intermediate conformation of the native CheY ⋅ FliM_N complex, the inactive state of free D13K-Y106W CheY, and the MD-based network architecture. We extended the MD analysis to determine temporal coupling and energetics during activation. Coupled aromatic residue rotation was a graded rather than a binary switch, with Y- or W106 side-chain burial correlated with increased FliM_N affinity. Activation entrained CheY fold stabilization to FliM_N affinity. The CheY network could be partitioned into four dynamically coordinated sectors. Residue substitutions mapped to sectors around D57 or the FliM_N interface according to phenotype. FliM_N increased sector size and interactions. These sectors fused between the substituted K13-W106 residues to organize a tightly packed core and novel surfaces that may bind additional sites to explain the cooperative motor response. The community maps provide a more complete description of CheY priming than proposed thus far.

Significance

CheY affinity for the FliM N-terminal peptide (FliM_N), its binding target at the flagellar motor, is increased by phosphorylation to switch rotation sense. Atomistic simulations based on CheY and CheY ⋅ FliM_N crystal structures with and without phosphomimetic double substitution (D13K-Y106W) showed CheY compaction is entrained to increased FliM_N affinity. Burial of exposed aromatic side chains drove compaction, validated by tracking side-chain solvent accessibility with hydroxyl-radical foot printing. The substitutions were localized at the phosphorylation pocket (D13K) and FliM_N interface (Y106W). Mutual information measures revealed these locations were allosterically coupled by a specialized conduit when the conformational landscape of FliM_N-tethered CheY was modified by the substitutions. Novel surfaces stabilized by the conduit may bind additional motor sites, essential for the high cooperativity of the flagellar switch.

Introduction

Escherichia coli CheY is a founding member of a bacterial response regulator superfamily that uses aspartate phosphorylation to regulate diverse signal relays (1,2). The CheY β₅α₅ fold has structural homology with small eukaryotic signal-transducing proteins (3). CheY phosphorylation couples the occupancy of the chemoreceptor patch to the motile response in bacterial chemotaxis. Previous studies of CheY have established it as a model for fundamental design principles in protein allostery (4). Here, we study E. coli CheY binding to the FliM N-terminal peptide (FliM_N) responsible for its initial interaction with the flagellar switch complex.

CheY fused with green fluorescent protein (GFP) is both phosphorylated and dephosphorylated at the polar chemoreceptor patch, generating pulsatile fluctuations in intracellular phosphorylated CheY (CheY∼P) level (5,6). The CheY∼P diffuses to the flagellar motor within the flagellar basal body, interacting with its C-ring (a.k.a. the switch complex), a multi-subunit assembly composed of the proteins FliG, FliM, and FliN. In E. coli, the interaction increases clockwise (CW) rotation (7). Single-cell measurements expressing GFP-CheY under conditions in which CheY∼P is the dominant form have shown that motor rotational bias has a sigmoidal dependence on CheY concentration (Hill coefficient >10.5, K_D = 3 μM) (8), implying highly cooperative action of the captured CheY molecules switching flagellar rotation. More recently, GFP-CheY occupancy was estimated to be ∼1/3 and < 1/10 of the 34 FliM subunits present per motor (9) for single CW and counterclockwise (CCW) rotating motors, respectively (10). The occupancy and rotation state were coupled within the image time resolution (20 ms), whereas GFP-CheY motor dissociation times (70 ms) were faster than the response times to attractant stimuli (11). The single-motor kinetics also imply cooperative CheY-motor interactions.

Biochemical experiments coupled with mutagenesis, motility assays, and x-ray crystal structures have established that CheY is phosphorylated at a single aspartate (D57∼PO₄). The aspartyl phosphate is labile, with a 22.8 s half-life at ambient temperature (12). The affinity for the FliM_N motor binding target of nonphosphorylated E. coli CheY (K_D = ∼450 μM) is 15× weaker than for CheY∼P as measured by fluorescence quenching of CheY residue W58 adjacent to D57 (13,14)). The binding of CheY∼P to isolated, native CCW-locked flagellar switch complexes had K_D stronger than that for FliM_N but was noncooperative (15), in contrast to the in vivo measurements of rotation bias (8) or motor localization (10) that sample both rotation states (Table S1). The conundrum of how cooperative responses arise by CheY∼P binding to FliM_N alone is increased by the fact that FliM_N is separated from the rest of the C-ring by a flexible tether (16). Thus, evidence that CheY interaction with the switch involves two binding sites, initial interaction with FliM_N, followed by a subsequent interaction of the FliM_N-tethered CheY to FliN in E. coli (17), provides a plausible solution. It has remained unclear whether the FliM_N tether facilitates the second-stage binding step only by increasing CheY local concentration or whether structural changes also occur that prime CheY to bind FliN (Fig. 1).

Figure 1.

CheY interactions with the flagellar motor. CheY shade intensity and size denote activation state and FliN binding probability, respectively. Binding of activated CheY (CheY^∗) to isolated switch complexes is not cooperative (H = 1), but the change in flagellar CW or CCW rotation bias is highly cooperative (H > 10) with CheY^∗ concentration. First-stage (1) binding to FliM_N enables-second stage (2) binding to FliN. The increased local concentration and the multiple FliN copies enhance second-stage binding probability. The inactive CheY binds weakly, reducing FliM_N-tethered CheY binding events with FliN below the critical threshold for CW rotation. This study provides evidence for structural changes in CheY that may supplement increased local concentration for second stage binding. To see this figure in color, go oline.

An atomic structure for CheY∼P is not available given the lability of the aspartyl phosphate. Therefore, atomic structures of phosphomimetic CheY proteins obtained by chemical modification (18,19) or mutagenesis (20, 21, 22, 23) have been used to reconstruct the activation mechanism. Although both chemically modified and mutated proteins are used in vitro biochemical assays. only the latter can be studied in vivo (12,20, 21, 22,24, 25, 26). The activating substitutions D13K and Y106W are the most potent modulators reported thus far of FliM_N binding in vitro (13,14,27) and motor rotation bias in vivo (10,28, 29, 30). The comparison of CheYD13KY106W efficacy with CheY∼P from both in vitro and in vivo assays, the substantial knowledge of its effects on motor physiology and the availability of atomic structures with and without FliM_N (Table S1) make CheYD13KY106W the logical first choice for the elucidation of the molecular priming mechanism.

The crystal structures of D13K-Y106W CheY alone and in complex with FliM_N showed bound FliM_N was required for the activated CheY conformation. They established CheY residues K91, Y106, and K119 as part of the FliM_N, binding surface (23). K91 and K119 formed salt bridges with FliM_N. The W106 side chain moved in as FliM_N bound to switch K109 bonding interactions with T87, D57, and, via bound water, D12 (23). The structure of the native CheY ⋅ FliM_N complex exhibited some features of inactive CheY and some features of the active D13K-Y106W CheY ⋅ FliM_N (31). Notably, the orientation of the Y106 side chain matched that for W106 in the D13K-Y106W CheY ⋅ FliM_N complex. The “intermediate” conformation of the native CheY ⋅ FliM_N structure challenged two-state CheY allostery models that coupled the Y- or W106 rotamer state to T87 motions (32). An NMR study on free CheY (33) reached a similar conclusion. CheY has high conformational plasticity, as seen by the discrepancies between crystal structures of activated CheY proteins (19,23,34). The coverage of the conformational landscape by crystal structures is too sparse to resolve the conformational trajectories for activation by phosphorylation or binding targets such as FliM_N. Alteration of low-affinity binding interfaces, a common occurrence in signal-transducing phosphoprotein complexes by crystal packing contacts, is an additional concern (35).

CheY conformational plasticity is not well described by classical protein allostery concepts of “induced fit” (KNF) (36) or “conformational selection” (e.g., MWC (37)) but is accommodated by modern ideas of allostery (38) in which protein-protein interactions between flexible partners have been described in terms of a folding funnel, in which the funnel bottom has a “rugged” landscape with multiple minima (39). Accordingly, molecular dynamics (MD) simulations and solution measurements have supplemented the x-ray crystallography of free CheY structures. MD of free CheY examined the coupling between Y106 rotation and T87 movements triggered by hydrogen bond formation (40), showed that the β4-α4 loop is an important determinant of allosteric signaling affected by lysine acetylation (41), and extracted common design principles between CheY and other response regulators with correlation analyses (42,43).

Here, we detail simulations and solution measurements to better understand the differences between the native and D13K-Y106W CheY crystal structures. We resolved the conformational landscapes by MD simulations with mutual information measures to determine the coupling between protein fragments. Protection experiments with XFMS (x-ray foot printing with mass spectroscopy) (44,45), a technique that probed side-chain solvent accessibility in contrast to deuterium exchange of backbone hydrogen atoms, supported the FliM_N requirement for D13K-Y106W CheY activation reported by the crystal structures and the MD allosteric network model. XFMS has a more straightforward physical rationale than fluorescence quenching for reporting side-chain motions over time-resolved windows and is not limited by the size of the protein assembly. Further analysis of the MD trajectories resolved multiple CheY Y106 rotamer states. Inward orientation was temporally coupled to stabilization of both the CheY fold and the FliM_N interface in the CheY ⋅ FliM_N complex, but not in CheY alone. The coupling increased in D13K-Y106W CheY ⋅ FliM_N. The formation of a distinct module that orchestrates CheY dynamics to stabilize new surface topologies for possible second-stage binding to FliN was the signature of the fully activated D13K-Y106W CheY ⋅ FliM_N state.

Materials and Methods

Structure preparation

Structures of E. coli CheY (Protein Data Bank, PDB: 3CHY; 1.7 Å resolution (46)) and complexes of native (PDB: 2B1J; 2.8 Å resolution (31)) and mutant (13DKY106W) CheY (PDB: 1U8T; 1.5 Å resolution (23)) with FliM_N were downloaded from the PDB. The label CheY^∗ will, henceforth, specifically apply to CheY13DKY106W. The 1U8T unit cell was a tetramer with 2 CheY^∗ and 2 CheY^∗ ⋅ FliM_N complexes. We generated the native CheY ⋅ FliM_N complex structure (1U8T_DY) by in silico mutagenesis (13K → D, 106Y → W) to base the simulations on well-resolved atomic coordinates The reverse mutagenesis and analyses of the crystal structures are detailed in Supporting Materials and Methods, Section B.

Molecular simulations

MD

A set of three replicas of duration 1 μs each was generated for the mutant (1U8T) and native (1U8T_DY) complexes using GROMACS 2016.2 with the Amber ff99sb^∗-ILDNP force field (47). Another set of three replicas of 500 ns duration each was generated for the native CheY (3CHY). Each system was first solvated in an octahedral box with TIP3P water molecules with a minimal distance between protein and box boundaries of 12 Å. The box was then neutralized with Na⁺ ions. Solvation and ion addition were performed with the GROMACS preparation tools (Supporting Materials and Methods, Section B).

Collective motions were identified by PCA of the conformational ensembles. Principal components (PCs) were generated by diagonalization of the covariance matrix of C^α positions. The overlap (cumulative root mean-square inner product) of the PCs between replicas (48)) and the PC dot product matrix was computed with the GROMACS g-anaeig function.

The conformational ensembles were clustered and mean structures representing the major clusters (n > 5) computed with the GROMACS g-cluster function. The energy landscape was computed with PROPKA 3.0 (49). PROPKA calculates the free-energy difference (ΔG) between the folded and unfolded states as the protein charge varies with pH (50). CheY has 37 ionizable groups (9D, 12E, 10K, 4R, 2Y) plus N- and C-termini that determine its net charge. The ΔG is computed from the perturbation of residue pK-values by the protein environment, namely the dielectric-dependent desolvation penalty, backbone and side-chain hydrogen bonds, and interactions with other charged residues. For the complexes, the ΔG was computed for the complex (ΔG_T) as well as CheY alone with FliM_N removed (ΔG_CheY). The ΔG_CheY was the free energy of the CheY fold. The interfacial energy ΔG_interface = ΔG_T − ΔG_CheY.

tCONCOORD

tCONCOORD utilizes distance constraints based on the statistics of residue interactions in a crystal structure library (51,52), to generate conformational ensembles from one crystal structure with solvent modeled as an implicit continuum. tCONCOORD runs compared conformational ensembles for native CheY (3CHY) with double mutant CheY, extracted from the heterogenous 1U8T unit cell that contains structures both with and without FliM_N. Sets of 16⁴ = 65,536 equilibrium conformations with full atom detail were typically generated for each structure. The overlap between ensemble subsets was >99% when the subset size was <0.25 of this value (53). The details are in Supporting Materials and Methods, Section B.

Network analysis

Structural alphabet

Coordinated CheY motions were examined using mutual information. The normalized mutual information (nMI) matrix encodes correlations between conformational states of different parts of the protein backbone (Supporting Materials and Methods, Section B). The states are represented by a structural alphabet (SA), a set of recurring four-residue fragments encoding structural motifs derived from PDB structures (54). Fragments are assigned an SA designation according to backbone dihedral angles, allowing conformation to be specified as a one-dimensional (1D) string (54). The fragments are represented as network nodes, with the connectivity (edges) between them representing their correlated dynamics over the MD trajectory.

Eigenvector analysis

Statistically significant correlations between columns were identified with GSATools (55) and recorded as a correlation matrix. The correlation matrix was used to generate a network model with the residues as nodes and the correlations as edges. In vector notation, the overall connectivity of a given fragment is reported by its eigenvector centrality, E (“centrality”). The contribution of a node to the network was estimated by its E, calculated directly from the correlation adjacency matrix:

$$E\times {\left\{M\right\}}_{corr}=E\times \lambda$$

where the {M}_corr is the correlation matrix. The λ is the eigenvalue.

The nMI contributions of local fragment motions were computed for the top PCs and superimposed on their root mean-square fluctuation (RMSF) profiles to evaluate the mechanical behavior of the network nodes in driving collective motions. Ensemble conformations and MD runs were averaged for computation of the nMI between fragment positions, with >2σ thresholds for selected top couplings. Pearson’s correlations were used for comparison. Significance limits were set in GSATools.

Community analysis

The Girvan-Newman algorithm (56) was used to identify community structure. Then, the network was collapsed into a simplified graph with one node per community, with the node size being proportional to the number of residues. Edge weights represent the number of nMI couplings between communities (57). Community analysis of correlation networks identifies relatively independent communities that behave as semirigid bodies. Graphs were constructed with the igraph library (58) in R (https://cran.r-project.org/web/packages/igraph/) and visualized in Cytoscape (http://www.cytoscape.org/).

Overexpression and purification of CheY proteins

The CheY-pET21b plasmids with E. coli cheY alone and fused with FliM_N (17) were modified to incorporate the double mutation D13K and Y106W. The native and mutated plasmids were expressed in E. coli strain BL21/DE3. The expressed proteins were purified with fast protein liquid chromatography. The FliM_N ⋅ CheY fusion interacts with FliN (17) and is more potent than CheY alone in potentiation of CW rotation (P. Wheatley, unpublished data). Three-dimensional (3D) models of the FliM_N.CheY fusions were obtained with the I-Tasser suite (59). In all top five models, FliM_N was docked in the location seen in the crystal structures of the CheY ⋅ FliM_N complexes. The top model had cs = −1.08, RMSF = 7.2 + 4.2 Å (against CheY, FliM_N crystal structures; Supporting Materials and Methods, Section C).

X-ray foot printing

Protein samples (CheY, FliM_N ⋅ CheY, CheY^∗, and FliM_N, CheY^∗) were prepared in 10 mM potassium phosphate buffer (pH 7.2), 100 mM NaCl, and 10 mM MgCl₂. Exposure range was determined empirically by adding Alexa488 to protein solutions as previously described (60). Sample irradiation was conducted without Alexa488 dye using a microfluidic setup with 100 and 200 mm ID tubing in combination with a syringe pump as previously described (61). After exposure at ALS beamline 3.2.1, samples were immediately quenched with methionine amide to stop the secondary oxidations and stored at −80°C for LCMS analysis.

The oxidized fraction, F, for a single residue modification was given by the equation

$$F=\left\{Xi/(T+\left(\sum Xi)\right)\right\},$$

where Xi is the oxidized residue abundance of one of the monitored residues in a trypsinized peptide and T is the unoxidized peptide.

Best-fit first-order rates were calculated in Sigmaplot version 12. Protection factors (PFs) were calculated as the ratio of the intrinsic residue reactivity over its foot-printing rate (62). Its logarithm (log(PF)) was proportional to the solvent-accessible surface area (SASA). The relation assumes that the foot-printing rate was related to the activation energy associated with the accessibility of the side chain to hydroxy radicals and the initial step of hydrogen abstraction. It empirically gave the best fit for proteolyzed peptides on a model data set, extended here to single residues (62).

Mass spectrometry analysis

X-ray-exposed protein samples were digested by trypsin, and the resulting peptide samples were analyzed in an Agilent 6550 iFunnel Q-TOF mass spectrometer coupled to an Agilent 1290 LC system (Agilent Technologies, Santa Clara, CA). Approximately 10 pmol of peptides was loaded onto the Ascentis Peptides ES-C18 column (2.1 mm × 100 mm, 2.7 μm particle size; Sigma-Aldrich, St. Louis, MO) at 0.400 mL/min flow rate and eluted with the following gradient: initial conditions were 95% solvent A (0.1% formic acid) and 5% solvent B (99.9% acetonitrile, 0.1% formic acid). Solvent B was increased to 35% over 5.5 min and was then increased to 80% over 1 min and held for 3.5 min at a flow rate of 0.6 mL/min, followed by a ramp back down to 5% B over 0.5 min, at which it was held for 2 min to re-equilibrate the column to original conditions. Peptides were introduced to the mass spectrometer from the LC using a Jet Stream source (Agilent Technologies) and spectra acquired with Agilent Mass Hunter Workstation Software B.06.01. The peptide precursor peak intensities were measured in Mass Hunter quantitative analysis software. Further details and data sets are given in Supporting Materials and Methods, Section C.

Results

We analyzed conformational ensembles generated by MD to identify dynamic changes in CheY architecture, using loops and residues implicated in the allosteric relay (see Introduction) as markers. We used XFMS protection experiments to relate the crystal structures to the conformation landscape in solution and test the dynamics predicted by the MD simulations (Fig. 2 A).

Figure 2.

Dynamics of CheY-FliM_N association. (A) The structure of CheY in complex with FliM_N (2B1J-AC) is shown. Colors indicate FliM_N (yellow), tryptic CheY fragments (blue), allosteric relay loops (green), side chains (M (magenta); K (cyan); and Y, W, and F (gold)), D57 C^α (red asterisk), and Mg²⁺ (magenta). (B) MD RMSF (mean +/− standard error) profiles for the combined replica trajectories for the three structures analyzed in this study are given. Bars mark CheY loops α3-β3 (white) and α4-β4 (black). Asterisks mark residues Y- or W106 (black), K109 (cyan), and F111 (yellow). FliM_N residue D12 (red asterisk) forms a salt bridge with CheY K119. (C) Snapshots of CheY (blue) Y106 (red) transitions in 1U8T_DY coupled to internal and interfacial residues are given. FliM_N is yellow. (i) shows T87 (lime); Video S2. (ii) shows K119 (green) and FliM_N D12 (pink); Video S4. To see this figure in color, go online.

CheY-activating residue substitutions D13K and Y106W stabilize FliM_N association

Three MD replica runs each were performed for the native CheY structure (3CHY (46)), the activated D13K-Y106WCheY in complex with FliM_N and alone (1U8T (23)), and a complex of native (nonactivated) CheY with FliM_N engineered in silico from 1U8T (Materials and Methods). The crystal structures showed residue Y106 was in the OUT conformation in CheY (3CHY) but in the IN conformation in CheY ⋅ FliM_N (2B1J) and D13K-Y106WCheY ⋅ FliM_N (1U8T). The Y- or W106 rotamer state was correlated with the orientation of the W58 and F111 side chains. The engineered complex was used instead of the crystal structure (2B1J) (31) because the latter, in addition to the lower resolution, had a systematic bias in its RMSF profile from the N- to C-terminus. The bias may be due to mosaicity in the crystal consistent with increased CheY-FliM_N interfacial dynamics (Fig. S1).

Henceforth, the 1U8T_DY CheY ⋅ FliM_N will be referred to as the “native CheY complex” and CheY^∗ FliM_N as the “mutant CheY complex.” The C^α RMSF profiles for each structure, averaged over three 1 μs runs, are shown in Fig. 2 B. The MD excluded the first three residues (M₁GD₃) of the FliM_N sequence (M₁GDSILSQAEIDALL₁₆) because these were not resolved in the 1U8T structure. The CheY^∗ FliM_N complex had higher RMSF values for the α-helix 1 (residues 22–30) and connected β5-α5 loop (residues 109–114) but lower values for the α₄-β₄ loop (residues 88–96) relative to CheY and CheY ⋅ FliM_N. These flexibility differences were consistent with the altered bond arrangements between residues D12, K13, and K109 (α-helix 1 and β5-α5 loop) and bond formation between K91 (α₄-β₄ loop) and FliM_N D3 seen in the crystal structures (23). The profiles are compared with B-factors for the x-ray structures. The B-factors were high relative to the MD-derived RMSF values, particularly in loop regions, reflecting conformational heterogeneity of these segments in the crystals (Fig. S1).

The 3CHY MD trajectories revealed transitions of Y106 between the OUT and IN states, consistent with electron density observed for both states in the crystal structure. FliM_N secondary structure, the CheYK119-FliM_ND12 salt bridge, and Y106 or W106 rotamer state were conserved between the 2B1J and 1U8T crystal structures. However, the raw MD trajectories of the complexes showed FliM_N had higher mean C^α RMSF values when CheY was wild-type than when it carried the activating substitutions (Videos S1, S2, and S3). This difference was due to association/dissociation of the FliM_N N- and C-termini from native CheY. In CheY^∗ ⋅ FliM_N trajectories, the peptide center was tethered by the CheYK119-FliM_ND12 salt bridge. CheY^∗ W106 was locked IN and part of the segment with the lowest C^α RMSF, together with K109 and F111. In CheY ⋅ FliM_N trajectories, OUT excursions of Y106 cleaved this salt bridge and weakened interfacial attachments (Fig. 2 C; Video S4). Thus, the MD confirmed the suggestion from the CheY ⋅ FliM_N crystal structure that its FliM_N interface was labile.

Video S1. CheY Dynamics

The raw trajectory for 3CHY showing movements of the Y106 side chain (red) and T87 (green)

Video S2. Dynamics of the Inactive CheY ⋅ FliM_N Complex

The raw trajectory for the native complex (1U8T_DY), engineered from 1U8T, showing movements of the Y106 (red) and T87 (green) side chains).

Video S3. Dynamics of the Activated Mutant CheY^∗ ⋅ FliM_N Complex

The raw trajectory for 1U8T, showing movements of the W106 side chain (red) with T87 (green).

Video S4. CheY ⋅ FliM_N Interface Dynamics

Interface extracted from the raw trajectory of the 1U8T_DY complex. Sidechains (CheY 106Y (red) and K119 green), FliMN D12 (pink)).

Two loops control CheY network dynamics

Previous MD simulations focused on the coupling between selected residues implicated from genetic or biochemical data in the long-range allosteric communication within CheY (40,41,43). Here, we develop a CheY network model (Fig. 3 A), constructed in (42) for a quantitative description of allostery within the entire protein from the MD conformational ensembles. The model is based on three key concepts: the SA, the nMI, and the eigenvector centrality (E). The 3D C^α conformation of four-residue peptides is uniquely specified by three bond and torsion angles with distinct conformational clusters (“alphabets”) resolved upon inspection of the PDB protein structure database (54). First, the SA was used to convert the 3D CheY fold to a 1D sequence of four-residue fragments and different conformations in an MD trajectory represented as a sequence of 1D strings. Second, the correlation between the confirmation of different CheY fragments within the sequence was computed as the MI. The nMI was the MI corrected for correlations expected by chance and the estimated uncertainty because of the finite number of conformations in the ensemble. The nMI couplings constituted the “edges” of the CheY network with correlation strength denoted by the edge thickness. The fragment positions within the sequence alignment formed the “nodes” of the network. Third, the connectivity of the network was determined by E, a measure of the influence of individual nodes in the network as reflected by the coupling of their dynamics with other nodes in the nMI correlation matrix. A node E limit value of 0 represented the case in which its dynamics did not affect other nodes in the network. An E limit of one represented the case in which its conformational fluctuations switched the entire network between discrete structural states (see Materials and Methods for formal definitions).

Figure 3.

CheY network dynamics. (A) The global network has nodes (residue fragments) and edges (MI-weighted node interactions). (B) Nodes containing residues that are part of the allosteric relay (W58, K91, Y106, K109, F111) have high scores in the CheY network. In (A) and (B), these residues and control residues (M17, M60, M63) monitored by XFMS are highlighted (yellow circles). (C) Centrality profiles of the complexes are shown (i CheY, FliM_N (green); ii CheY^∗ ⋅ FliM_N (red)) compared with the native CheY profile (mean ± standard error; blue lines). The dotted line (ii, red) plots the mutual information between the local loop fragment dynamics and collective PC1 motions. Complex formation reduced the centrality of the β4-α4 loop that, together with the β3-α3 loop, formed central nodes in the CheY network. Activating mutations eliminated the β4-α4 loop as a node but did not alter the contribution of the β3-α3 loop in CheY^∗ ⋅ FliM_N. Horizontal bars indicate α3-β3 (white) and α4-β4 (black) loops, as in Fig. 2B. To see this figure in color, go online.

First, we identified the central nodes in the CheY global network with the highest connectivity (Fig. 3 B). The central CheY nodes were the loops β3-α3 (D₅₇WNMPNMDG) and β4-α4 (T₈₇AEAKK). A third prominent node just below the 1σ threshold was the short β₅-strand (Y₁₀₆VVKP). Second, we used tCONCOORD, a computationally inexpensive method to generate conformational ensembles, for comparison of the CheY and CheY^∗ conformational landscapes.

Interpretation of differences between CheY and CheY^∗ crystal structures based on isolated landmarks (for example, Y106 rotamer state (IN or OUT in CheY (46)) versus W106 (OUT in CheY^∗ (23)) is complicated by the CheY conformational plasticity. Analysis of the tCONCOORD ensembles showed the central network nodes remained unchanged, with both CheY Y106 and CheY^∗ W106 side chains restricted to a limited OUT orientation range (Fig. S2).

We next examined the CheY and CheY^∗ complexes with FliM_N (Fig. 3 C). We split the CheY ensemble into four subpopulations to assess the significance of differences observed between it and the complexes. The network connectivity, as formalized by centrality plots, showed significant changes in the complexes relative to the CheY protein alone. There was a dramatic reduction in the centrality of loop β4-α4 and associated β-strand ^∗₁₀₆VVKP (^∗ = Y (CheY ⋅ FliM_N) or W (CheY^∗ ⋅ FliM_N)) at the FliM binding surface. Their roles as network nodes were reduced in CheY ⋅ FliM_N and abolished in CheY^∗ ⋅ FliM_N. This trend contrasted with the conservation of these nodes for CheY^∗. The centrality of α-helix 1 increased with its mobility (Fig. 2 C).

Immobilization of the α4-β4 loop modulates CheY collective motions

We used PC analysis to characterize CheY collective motions and their modulation by FliM_N binding and the activating substitutions. The PCs are derived from the atomic coordinate covariance matrix and describe C^α backbone movements, ranked according to the amplitude of the structural variation they explain. The collective motions were described well by the first few PCs, as found for other proteins. The first three principal components (PCI, PC2, and PC3) accounted for >60% of all motions in each case. These three PCs comprise bending and twisting modes organized around the β-sheet core. A core subpopulation of CheY conformations was observed in MD trajectories generated by all three structures. When CheY is in complex with FliM_N, new subpopulations comparable in size to the core were generated. These were distinct in the CheY ⋅ FliM_N and CheY^∗ ⋅ FliM_N complexes. Thus, new conformational ensembles are accessed upon binding of FliM_N, with the potential to produce binding surfaces for additional targets (Fig. S3, A and B).

Loops act as hinge elements for collective motions. Their mechanics give insight into the modules they control (39). We computed loop β3-α3 and β4-α4 hinge flexibility by mapping their RMSFs onto PC1, which accounts for >40% of the total amplitude of the PC motions. Flexibility scaled with the magnitude of the loop RMSFs relative to the mean PC1 RMSF. We computed hinge contribution to PC1 as the nMI between PC1 variance and the local loop fragment dynamics. The long β3-α3 loop partitioned into two segments. The short D₅₇WN and the adjacent M₆₀PNMDG loop segments behaved as rigid (low RMSF) and flexible (high RMSF) hinges, respectively, to control native CheY PC1 dynamics. In the CheY^∗-FliM_N complex, the β3-α3 loop hinge was retained but with inverted flexibility of the two segments. The transition for loop β4-α4 was more dramatic from a flexible hinge in native CheY to a closed hinge that acted as a rigid lever arm in CheY^∗-FliM_N. The reduced flexibility decreased β4-α4 loop centrality and influence on PC1 motions (Fig. S3 C).

Protection experiments support the “intermediate” CheY ⋅ FliM_N structure and the MD allosteric network

We studied homogenous solutions of CheY and FliM_N-CheY fusion proteins (Fig. S4), to measure the changes predicted by the crystal structures and the MD network model. The fusions were critical because the affinity of FliM for active CheY is weak and that for the inactive protein even weaker (Introduction). The crystal structures reported that 1) aromatic side chain internalization in CheY was entrained to FliM_N attachment and 2) the configurations of free CheY with or without the D13K-Y106W substitutions were similar. The MD 3) revealed FliM_N attachment was more labile in the native versus D13K-Y106W complexes and 4) generated a network model to discriminate between CheY fragments that changed upon activation from those that did not. These predictions were assessed by comparing the side-chain solvent accessibility of allosteric relay residues Y- or W106 active, W58, K91, K109, F111, and K119 by hydroxyl-radical foot printing in the native and D13K-Y106W CheY proteins and their FliM_N-fusion constructs. The control residues predicted not to change during activation were the β3-α3 loop residues M60 and M63 in proximity to W58 and the M17 in proximity to D- or K13.

Aromatic residues have high intrinsic side-chain reactivities with hydroxyl radicals, exceeded only by methionine and cysteine (absent from E. coli CheY), followed by the alkaline side chains. Tryptic digestion partitioned CheY into six separated peptides that were distinguished by mass spectroscopy based on their characteristic m/z ratio, allowing oxidation of these residues to be monitored. Dose-response curves were generated for each of the four constructs (CheY, CheY^∗, CheY-FliM_N, and CheY^∗-FliM_N). For each residue examined, the curves from two independent experiments were pooled (Fig. S5).

CheY residues of the allosteric relay at the FliM_N interface and distant from it were designated “interface” and “core” residues, respectively. The oxidation of the interfacial residues (K119, Y- or W106, K91) was reduced in the complexes (Fig. 4 A, Ii–iii). Importantly, oxidation of the core residues also decreased with complex formation (Fig. 4 A, Ci–iii). In contrast, there was no significant difference between oxidation rates for β3-α3 loop control residues M60 or M63 in the fusion proteins versus the free proteins, whereas the oxidation of the control M17 in the fusions was comparable or greater than in the corresponding free CheY proteins (Fig. 4 B).

Figure 4.

XFMS measurements. Dose-response curves for (A) are given. Relay is shown. Interfacial residues (Ii) Y- or 106W are shown. (Iii) K119 is shown. (Iiii) shows K91. Core residues (Ci) K109, (Cii) F111, and (Ciii) W58 are shown. (B) Control residues (i) M60 and M63 and (ii) M17 are shown. Initial rates (dashed lines) were obtained from least-squares linear regression of the decrease in the unoxidized fraction (mean +/− standard error) with dose. To see this figure in color, go online.

PFs were computed from the initial rates from the single-residue dose-response curves, following (62), with intrinsic reactivities mostly determined thus far from measurements on small peptides (63). We first evaluated the agreement between solvent accessibility reported by the XFMS measurements and the crystal structures. PFs read out the SASA, with some caveats (62). The log(PF)s were plotted against the residue SASA in the crystal structures. The overall correlation was comparable to published values for the peptide correlations in model proteins (62), indicating that the changes in the dose-response plots for the monitored residues are due, in large part, to nonpolar bulk solvent accessibility changes (Fig. 5 A). Outliers (M17, K109, F111) were restricted to a small CheY protein volume in the structures (Fig. S6). The crystal structures may not reflect the solution conformation of this local region, but bonding interactions may also contribute (Fig. S6). The correlation improved markedly (0.60 → 0.86), without further correction, if the outliers were excluded.

Figure 5.

Single-residue oxidations related to SASA. (A) Log(PF)s plotted against the side-chain SASA calculated from the crystal structures are given. Pearson correlation coefficients: 0.86 (minus M17 (rose), K109 (cyan); see text). Overall = 0.60 {CheY= (−)0.76; CheY^∗ = (−)0.70; FliM_N ⋅ CheY= (−)0.54; FliM_N ⋅ CheY^∗= (−)0.12}. Best fit (black dashed line), 95% confidence limit (blue lines), and 95% prediction limit (red lines) are shown. (B) Protection of interfacial (K119, Y- or W106, K91), core (F111, K109, W58), and control (M17, M60, M63) residues in CheY^∗, CheY ⋅ FliM_N, CheY∗, FliM_N relative to their protection in CheY is shown. {Protection}_norm = Log {PF/PF_CheY} (mod(error)). Positive values indicate increased protection. To see this figure in color, go online.

The PFs for CheY^∗, CheY ⋅ FliM_N, and CheY^∗ ⋅ FliM_N were then normalized for each residue against the value obtained for CheY (Fig. 5 B). The normalized (log(PF)s) provided a quantitative measure for the increase for both interfacial and core residues in the CheY ⋅ FliM_N and CheY^∗ ⋅ FliM_N fusions relative to the values for CheY. These residues were more protected in CheY^∗ ⋅ FliM_N than CheY ⋅ FliM_N. In contrast, the protection of the control residues in the fusions (CheY ⋅ FliM_N, CheY^∗ ⋅ FliM_N) did not differ significantly from that measured for CheY. The normalized PFs showed no significant difference in protection for interfacial, core, or control residues in CheY^∗ relative to CheY.

The protection profiles showed that solvent accessibility for the allosteric relay residues decreased in the order CheY < CheY ⋅ FliM_N < CheY^∗ ⋅ FliM_N. The control residues either showed no changes or the opposite trend. Changes in the solvent accessibility of CheY^∗ relative to CheY were not significant. Thus, in conclusion, the XFMS experiments validated the main predictions of the crystal structures and the conformational ensembles generated from them.

Energetics of CheY stabilization by FliM_N and D13K and Y106W residue substitutions

The XFMS measurements correlated solution population shifts in selected residue positions to each other and with the crystal structures. The temporal couplings between these shifts could only be studied with MD. We next analyzed the MD trajectories to extract this information.

We examined the coupling between the electrostatic stabilization of the interface and the CheY fold with the rotational states of residue Y106 (106W in CheY^∗ ⋅ FliM_N). The CheY^∗ ⋅ FliM_N 106W side chain was locked IN (Video S3). In contrast, Y106 in CheY ( Video S1) and CheY ⋅ FliM_N ( Video S2) made frequent OUT ↔ IN excursions. Dwell times in the Y106 rotamer states measured from the raw CheY trajectories were 107 ± 34 ns (OUT) and 15 ± 4 ns (IN)). The CheY ⋅ FliM_N Y106 side chain was predominantly in the IN orientation with mean dwell time 239 ± 123 ns, 15-fold greater than for free CheY. The conformational ensembles in the MD trajectories were clustered based on the C^α backbone dynamics (RMSF). The major clusters represented distinct backbone conformational states accessed during the MD runs. The average structures for these clusters were compared with each other and the crystal structures with PROPKA. The mean ΔG-values at pH 7.0 were CheY (−4.8 ± 1.0 (n = 7)) < CheY ⋅ FliM_N (−5.8 ± 1.6 (n = 4)) < CheY^∗ ⋅ FliM_N (−9.9 ± 2.2 (n = 3)). All CheY clusters had Y106 in the OUT orientation (θ = 126.7 ± 3.8°), indicating that CheY Y106 IN states were too short-lived to influence backbone dynamics. CheY^∗ ⋅ FliM_N clusters had W106 in the IN orientation (θ = 54.1 ± 2.3°). The CheY ⋅ FliM_N clusters, in striking contrast, spanned the entire Y106 rotamer range. Thus, the intermediate CheY ⋅ FliM_N Y106 rotamer states were sufficiently stable to affect backbone dynamics (Fig. 6).

Figure 6.

Rotamer Y- or W106 energetics. (A) Interface and CheY fold stabilization are shown: interface, ΔG_int (triangle); CheY fold, ΔG_CheY (circle). Linear regressions (interface (dashed), fold (solid)) are given. (i) CheY ⋅ FliM_N (green) and 2B1J crystal values (lime) are shown. Vertical lines and rectangles show CheY (cyan) and CheY^∗ ⋅ FliM_N (red) θ and ΔG_core ranges, respectively. Correlations: θ − ΔG_interface (R = 0.23, Pearson = 0.63); θ − ΔG_CheY (R = 0.43, Pearson = 0.21). (ii) CheY^∗ ⋅ FliM_N (red) and 1U8T crystal values (purple) are shown. Correlations: θ − ΔG_interface (R = 0.96, Pearson = 0.98); θ − ΔG_CheY (R = 0.85, Pearson = 0.33). (B) CheY conformation and Y106 (green) side-chain rotamer orientation in representatives of the major CheY ⋅ FliM_N clusters are shown. To see this figure in color, go online.

Next, we computed the activation energetics by measurement of ionizable residue electrostatics. There was a weak stabilization of the CheY FliM_N interface and core with the internalization of the Y106 side chain. The buried CheY^∗ ⋅ FliM_N W106 side chain had a substantially more restricted rotation range than the CheY ⋅ FliM_N Y106 side chain. However, the correlation between side-chain orientation and stabilization of CheY^∗ FliM_N interface and CheY^∗ core was stronger, consistent with a more tightly packed CheY^∗ ⋅ FliM_N complex. The stabilization of the interface by the D13K and Y106W residue substitutions was consistent with the different FliM_N binding affinities measured in solution for active versus inactive CheY states. The novel result was the coupled stabilization of the CheY fold for both CheY ⋅ FliM_N and CheY^∗ ⋅ FliM_N.

The energetics computed for the 1U8T crystal structure was in line with results from the MD conformational ensembles. In contrast, the values computed for the 2B1J crystal structure were outliers reporting higher energy states relative to the values obtained from the MD runs, an outcome that may be linked to the increased B-factor values around the 2B1J CheY ⋅ FliM_N interface (Fig. S1) and/or deformation of the local volume around K109, F111, and M17 by crystal packing contacts (Fig. S6).

An emergent sector orchestrates CheY^∗ allosteric communication

We developed the network model for a comprehensive representation of the temporal couplings. The centrality analysis identified network nodes with the dominant couplings, but the non-nodal fragment couplings that constituted >95% of the information available in the nMI matrix were not well represented. We used community analysis, a recently developed tool (64,65), for this purpose. Community networks are collapsed networks that reduce, partition, and map the protein into contiguous, semirigid bodies (“sectors”) that may be schematized for a concise, comprehensive representation. The schematics and their mapping onto the 3D structure will be henceforth referred to as the community “network” and “map,” respectively.

Community analysis of native CheY revealed distinct sectors (n > 5) displaying coordinated dynamics. The β3 strand F₅₃VISD₅₇ occupied a central location in contact with all sectors. Sector A organized around the D57 phosphorylation site coupled to the other sectors, particularly with sector B, which organized around the FliM_N-binding surface. The tCONCOORD CheY^∗ community map, when compared with the corresponding CheY map, showed a small increase in sector A relative to sector C interactions with sector B (Fig. 7 A). This result may indicate limited activation of CheY^∗ relative to CheY detectable with the more sensitive community versus global network but does not challenge the conclusion CheY and CheY^∗ have similar dynamic architecture based on the α₃-β₃ and α₄-β₄ loops as network nodes.

Figure 7.

Changes in community network architecture triggered by D13K and Y106W substitutions and FliM_N peptide. The reduced number of sectors compared to single fragments as nodes gave a concise, quantitative readout of the protein dynamics. (A) CheY and CheY^∗ community maps are given. Networks (boxed insets) from tCONCOORD runs show the reduction in the size of sector C relative to sectors A and B in CheY^∗ versus CheY. (B) CheY, CheY ⋅ FliM_N, and CheY^∗ ⋅ FliM_N community architecture is shown. Networks (top) and maps (bottom) are given. FliM_N = yellow (cartoon representation). The MD detected four dynamic sectors for CheY (A = cyan, B = blue, C = orange, D = red). The sector C from the tCONCOORD runs is resolved into two sectors (C and D) in the MD runs. Node size = sector membership; edge thickness = weighted intersector interactions. Sectors A and B are built around the phosphorylation site (D57 (red asterisks)) and the FliM_N binding surface, respectively. They increase at the expense of sector C upon complex formation. The presence of phosphomimetic substitutions in the CheY^∗ ⋅ FliM_N complex creates an additional sector E from sectors A and B that orchestrates interactions with sectors C and D. (C) CheY^∗ ⋅ FliM_N community map showing sector E surface is given. See Video S5 for 3D perspective. Side chains identify the substituted residues (K13, W106) and FliM_N binding residue K119, a part of sector E. Sectors are colored as in (B). The strength of the top (>+2σ) nMI couplings (lines) couplings are reflected in their thickness and color (low (yellow) → high (red)). D57, red asterisk. To see this figure in color, go online.

The MD resolved the tCONCOORD sector C into two sectors (Fig. 7 B). Importantly, reported residue substitutions partitioned to sectors A and B in the more detailed map according to phenotype (Table S2). Positions, when these are known to affect dephosphorylation kinetics (65), mapped to sector A. Residues known to affect FliM_N binding or rotation bias, such as sites of suppressor substitutions for CW- or CCW-biasing FliM lesions (66), mapped to sector B. Positions yielding mutations that affect interaction with the CheY-phosphatase CheZ (67) were adjacent to sector D, the smallest sector obtained for CheY. Sector C, comparable in size to A, might be influence the overall stability and rigidity of the protein.

Changes in loop dynamics upon complex formation were reflected in the networks (Fig. 7 B). The couplings between sectors A (phosphorylation) and B (FliM_N binding) were strengthened relative to the free protein. Sector B expanded at the expense of sector C and coupled more strongly to sector A in the CheY ⋅ FliM_N network. The mutated residue D13 was part of a loop that flipped from sector A to sector B. A fifth sector (E (K_45-48N₆₂-L_65-68A₁₀₁-S_104-107F_111-114K_119-123)) spanned by the substituted residues (K13, W106) formed in the network of the activated mutant CheY-FliM_N complex (CheY^∗ ⋅ FliM_N). The E-sector fragments were drawn from sector A (K_45, N_62, K₁₁₉), sector B (A_101, S_104, F₁₁₁), or fragments adjacent to these sectors in the free CheY community network. Sector E formed a surface-exposed ridge that connected the FliM_N α-helix, via S_104–107 and K_119–123, to sector C residues E₃₅ and (via K₄₅) E₃₇ (Fig. 7 C; Video S5). The top nMI couplings connected sector E fragments within the central β3-α3 loop to the D57 phosphorylation site. These couplings were unchanged by complex formation.

Video S5. Sector Coupling the FliM_N Interface with the Phosphorylation Site

The CheY^∗·FliMN community map showing the surface profile of the coupling sector E (dark-green, green (adjacent residues)) specific for CheY^∗·FliMN. Top nMI couplings (lines (orange (low) -> red (high)) connect sector E to D57. The β3 strand forms a junction for sectors A (red), B (orange) and C (cyan). Sector D (blue). Sidechains (K13, D57, T87, W107) are colored according to their sector affiliation.

Discussion

The results of this study advance our understanding of CheY conformational plasticity and activation in important ways (Fig. 8).

Figure 8.

Allosteric priming in E. coli CheY. (A) Reaction coordinate (x axis) showing stabilization of the CheY fold coupled with CheY activation is given. The inward rotation of residue Y106 and the increased residue W58 fluorescence quenching due to its internalization, represented by red asterisk size, have been used to measure CheY activation and FliM_N binding, respectively. Horizontal bars indicate multiple local minima. CheY ensembles (blue) have large conformational heterogeneity controlled by a flexible β4-α4 loop. They sample both Y106 IN and OUT rotamer states, but the IN state is too short-lived to influence backbone conformations. FliM_N-bound CheY ensembles (green) sample a conformational landscape with a large ΔG range, with prominent troughs among the local minima that track the progressive stabilization of the CheY fold and concerted internalization of Y106 entrained to tighter FliM_N attachment. FliM_N bound to D13K-Y106W CheY (CheY^∗) confines the CheY fold to conformational space (red) around the global minimum. The β4-α4 loop is immobilized by the CheY K91 - FliM_N salt bridge and W106 plus W58 are locked IN a tightly packed CheY core, with the emergence of a dedicated sector (E) for communication between the phosphorylation site and binding interface. This sector is central to the dynamics of the stabilized CheY core. To see this figure in color, go online.

FliM_N as an allosteric effector

X-ray crystallography, in concert with behavioral and biochemical studies, has built a valuable mechanistic framework based on visual inspection of structural landmarks, guided by chemical intuition. Examination of the native CheY ⋅ FliM_N crystal structure led to the proposal that the complex was an intermediate between active and inactive states, consistent with a flexible β4-α4 loop (31). The structure challenged existing two-state switch models, but puzzlingly, the central element in the models, the Y106 rotamer state, was not in an intermediate conformation but the activated rotamer state, and the decrease in FliM_N affinity relative to the activated complex was difficult to discern. These issues have been resolved by the MD simulations and XFMS measurements reported in this study.

The CheY ⋅ FliM_N conformational landscape generated by MD simulations of the reverse-engineered 1U8T_DY structure had prominent minima that reflected intermediate FliM_N attachment entrained to Y106 rotation states that ranged between the dominant OUT state in free CheY and the W106 IN state in activated CheY^∗ ⋅ FliM_N. XFMS determined solvent accessibility values for the CheY ⋅ FliM_N allosteric relay side chains that were intermediate between values obtained for inactive CheY and active CheY^∗ ⋅ FliM_N. These values were correlated with the protection of the interfacial lysine residues that monitored FliM_N attachment. The D13K-Y106W residue substitutions as seen in the crystal structures did not alter the CheY fold to any significant extent in the absence of FliM_N, a result supported in this study by both simulation and measurement. The MD clarified that FliM_N stabilized CheY and strengthened allosteric communication between its binding interface and the D57 phosphorylation site because of formation, in part, of the CheY K91 - FliM_N D3 salt bridge. The salt bridge decreased the flexibility of the β4-α4 hinge, consistent with earlier studies (31,41).

The dynamics and energetics of activation

This study documents a broad, high-energy CheY conformational landscape with shallow minima consistent with the high conformational plasticity suggested by the CheY crystal structures and early MD studies (Introduction). Network analysis, based on MI between short protein fragments, established that two loops (β3-α3, β4-α4) act as flexible hinges to control the dynamics. The CheY MD trajectories revealed episodes during which the Y106 side chain is buried (IN), but the IN states were too brief to influence backbone dynamics, in contrast to the case for CheY ⋅ FliM_N. The buried states of the Y106 side chain have not been visualized, to our knowledge, in inactive CheY crystal structures.

The CheY conformations of the major CheY ⋅ FliM_N clusters were more stable than the dominant CheY conformations reported by the MD or the conformation in the 2B1J crystal structure. The lifetimes of the CheY Y106 IN states in CheY ⋅ FliM_N were substantially greater than in free CheY and represented in the major clusters. There was a weak correlation between the stability of the CheY fold, the FliM_N interface, and the position of the Y106 side chain. The CheY ΔG-values in the major CheY ⋅ FliM_N clusters overlapped with the values in the inactive CheY and activated CheY^∗ ⋅ FliM_N clusters.

The mean CheY^∗ ⋅ FliM_N ΔG-value was more stable than for CheY ⋅ FliM_N. This was also the case for the interfacial ΔG-values. The position of the W106 side chain was restricted to a narrow range. Nevertheless, the ΔG-values for both the CheY fold and its FliM_N interface, as well as the rotamer position of the W- or Y106 side chain, were similar for the dominant CheY ⋅ FliM_N and CheY^∗ ⋅ FliM_N conformational clusters. The similarity may explain capture during crystallization of the Y106 side chain in 2B1J in a position superimposable with the W106 side chain in the 1U8T complex. The better correlation of W106 side-chain position in the MD clusters and the 1U8T structure with the CheY fold and FliM_N interface ΔG-values reflects the tight packing due to the D13K-Y106W substitutions. The ΔG and W106 rotation angles of the CheY^∗ ⋅ FliM_N clusters did not overlap with values for CheY clusters.

Allosteric communication may range from largely enthalpic, as in lysozyme, to largely entropic, with change in flexibility rather than shape (68). Both energy terms contribute to CheY allosteric activation. The isolation of the global minimum from the multiple minima sampled by the native CheY ⋅ FliM_N conformational ensemble by the D13K-Y106W residue substitutions invoke “conformational selection”. Other aspects, such as the formation of the allosteric relay based on local changes in the loop and side-chain rotamer dynamics triggered by FliM_N attachment, support “induced fit.” Neither description is complete.

Community networks—a new measure for response regulator signal transduction

It has long been recognized that two-state allosteric models have heuristic value but that an analytical description is desirable (32). Many conformational states, as suggested (69), may be essential to explain how subtle changes in CheY sequence trigger diverse motile responses. In Bacillus subtilis, for example, CheY∼P stimulates CCW rather than CW rotation in contrast to E. coli but remains critical for chemotaxis (70). In Thermatoga maritima, the middle domain of FliM could be the second-stage CheY binding target (16), The diverse sensory responses triggered by CheY homologs even within one species (e.g., Caulobacter crescentus (71)), as well as the variable signal transduction strategies employed by response regulators (1), emphasize the need for a more complete description. Community networks have been used previously (65) to identify jointly moving regions governed by side-chain motions that do not track backbone secondary structure. This work is the first application, to our knowledge, of this approach to the response regulator superfamily.

Distinct protein sectors with correlated motions were identified in community networks. The extensive library of CheY residue substitutions was exploited for functional assignment of the sectors. Two sectors, namely the neighborhood of the phosphorylation site (sector A) and the region of FliM_N binding (sector B) had clear functional importance. Two other sectors lacked strong, specific phenotypes and might have broader functions in maintaining the overall CheY fold. The long β3-α3 loop influenced movements of the β3 strand that formed a sector junction, consistent with its central role in the reported PC motions. Similar motions take place in other proteins that utilize β-sheets for signal transduction (72).

FliM_N attachment increased the size of sectors A and B in the CheY community network. The CheY^∗ ⋅ FliM_N community network was distinguished from the CheY and CheY ⋅ FliM_N networks by a fifth sector (E), drawn from sectors A and B, that formed a dedicated conduit between the phosphorylation and FliM_N-binding sites to cement the allosteric linkage, with the substituted residues K13 and W106 at its boundaries. The emergence of sector E was tied to the closure of the β4-α4 hinge by the CheY K91 - FliM_N D3 salt bridge and “freeze out” of W₁₀₆VVKP β-strand dynamics by the burial of aromatic residues for a tightly packed core. This sector connects with all other sectors and has a large surface profile. It may directly or indirectly define a region important for binding to FliN.

The CheY^∗ protein is impaired in its interactions for other chemotaxis proteins, the CheA kinase and CheZ (6), whose CheY binding surfaces that overlap with that for FliM_N (69,73). Sector E may also influence these interactions. An important future goal would be to apply the integrated approach presented here to detect how CheY∼P discriminates between these components of the chemotaxis circuity.

Rotamer reorientation of aromatic side chains is a common theme in phosphoproteins, but diverse strategies for coupling side-chain motions to phosphorylation exist. In eukaryotic protein kinases, activation is controlled by DFG motif loops. These loops take on multiple IN and OUT orientations, with orientation correlated with activation. In Aurora kinase A, phosphorylation triggers transition between distinct IN orientations, rather than between IN and OUT states (74). In calcium calmodulin-dependent kinase, IN and OUT DFG states are loosely coupled to kinase domain phosphorylation (75). In CheY XFMS, reported D₅₇WN₅₉ internalization was coupled to protection at the FliM_N interface. We envisage that XFMS will have applications in other phosphorelays given ongoing developments in mass spectroscopy sensitivity and throughput because most amino acids are modified by hydroxy radicals to a greater or lesser extent.

The sparse sampling by crystal structures may miss high-energy states such as the intermediate states of the CheY 106 side chain that are important for deciphering mechanisms. MD simulations provide a more detailed sampling of the conformational landscape, but the challenge to extract the essential features from the large conformational ensembles obtained is only partially met by standard PCA and RMSF analyses. Community maps have given a concise, comprehensive description based on quantitative criteria for identification of the key features of CheY allosteric activation. They could provide the optimal compromise for mechanistic dissection of signal transduction strategies in the response regulator superfamily.

Editor: Elizabeth Komives.

Supporting Citations

References (76, 77, 78, 79, 80, 81, 82, 83, 84, 85, 86, 87) appear in the Supporting Material.

Author Contributions

P.W.: investigation, visualization, writing—review and editing. S.G.: methodology, investigation. A.P.: conceptualization, software, formal analysis, validation, visualization. Y.C.: methodology, investigation, formal analysis, validation. C.J.P.: methodology, supervision, funding acquisition, writing—review and editing. C.Y.R.: conceptualization, supervision, funding acquisition, writing—review and editing. D.F.B.: conceptualization, supervision, funding acquisition, writing—draft, review, and editing. S.K.: conceptualization; formal analysis; writing—draft, review, and editing; project administration.

Supporting Material

Data S1. Mascot Report

MS/MS run search parameters, and scores for unoxidized and oxidized PSMs.