CryoEM and computer simulations reveal a novel kinase conformational switch in bacterial chemotaxis signaling
Abstract
Chemotactic responses in bacteria require large, highly ordered arrays of sensory proteins to mediate the signal transduction that ultimately controls cell motility. A mechanistic understanding of the molecular events underlying signaling, however, has been hampered by the lack of a high-resolution structural description of the extended array. Here, we report a novel reconstitution of the array, involving the receptor signaling domain, histidine kinase CheA, and adaptor protein CheW, as well as a density map of the core-signaling unit at 11.3 Å resolution, obtained by cryo-electron tomography and sub-tomogram averaging. Extracting key structural constraints from our density map, we computationally construct and refine an atomic model of the core array structure, exposing novel interfaces between the component proteins. Using all-atom molecular dynamics simulations, we further reveal a distinctive conformational change in CheA. Mutagenesis and chemical cross-linking experiments confirm the importance of the conformational dynamics of CheA for chemotactic function.
https://doi.org/10.7554/eLife.08419.001eLife digest
To survive, an organism must be able to collect and interpret information about its environment and behave accordingly. Bacteria are able to do this via a process called “chemotaxis”. Inside the bacteria are sensors that contain a two-dimensional network of proteins called a chemosensory array, which detect chemical changes in the environment and signal to motor proteins to allow the bacterium to move to a more favorable location. Thousands of proteins make up the chemosensory array, and so a key question is how do these proteins interact with each other to work together as a team?
Cassidy, Himes, Alvarez et al. have now used a technique called cryo-electron tomography to determine the three-dimensional structure of the chemosensory array in the bacteria species Escherichia coli in high detail. This revealed the structures of several key components of the array, including some protein regions that are critical for signaling during chemotaxis.
Cassidy et al. were then able to use the electron tomography data to create a model of the array that details all of its individual atoms. Supercomputer simulations of this model revealed that during chemotaxis, a key signaling protein changes shape in a way that is critical for signal processing. This shape change was confirmed to be important for chemotaxis by chemical experiments and tests on mutant E. coli cells.
The next steps are to further improve the structure so that more details of the array organization can be distinguished, as well as to investigate the structure of other signaling states. By assembling structural “snapshots” of these different states, in the long term Cassidy et al. aim to develop models that detail the atoms in every one of the components involved in the chemotaxis signaling pathway.
https://doi.org/10.7554/eLife.08419.002Introduction
Bacterial chemotaxis is a ubiquitous, two-component signal transduction system that allows cells to extract information from environmental chemical gradients and place themselves within the nutrient-optimal portion of their habitat (Wadhams and Armitage, 2004; Capra and Laub, 2012; Eisenbach, 2004). Though the topology and complexity of the protein networks employed in bacterial chemotaxis vary by species, each uses the histidine kinase CheA (component 1) and response regulator CheY (component 2) to set up an intracellular phosphorylation cascade that regulates the motile behavior of the cell (Szurmant and Ordal, 2004). CheA, in particular, is a multi-domain protein, consisting of five separate and functionally distinct domains (P1-P5): P1-phosphoryl transfer domain, P2-substrate binding domain, P3-dimerization domain, P4-kinase domain and P5-regulatory domain. In addition to CheA and CheY, an expanded set of molecules assist in the mechanics of signal reception, transmission, and regulation. Specifically, bacteria utilize dedicated chemoreceptors (also known as methyl-accepting chemotaxis proteins, MCPs) to recognize ambient chemicals and transmit mechanical signals across the cell membrane to affect CheA kinase activity (Ortega et al., 2013; Parkinson et al., 2015). The adaptor protein CheW universally participates in the coupling of conformational changes within receptors to kinase regulation (Szurmant and Ordal, 2004; Liu and Parkinson, 1989). Bacteria, moreover, have evolved the ability to tune or adapt their chemotactic sensitivity to stimulus intensity, giving rise to short-term molecular memory and allowing an appropriate system response over wide ranges of chemical concentrations (Hazelbauer and Lai, 2010; Parkinson et al., 2015). In the case of the model organism, Escherichia coli, the adaptation mechanism involves the use of two enzymes, CheR and CheB, which reversibly modify specific residues in the receptor molecules (Hazelbauer et al., 2008; Hazelbauer and Lai, 2010; Goy et al., 1977; Ortega et al., 2013).
The tunable control of chemotactic activity requires the assembly of collaborative core-signaling units, involving the chemoreceptor trimer of dimers (TOD) (Amin and Hazelbauer, 2010; Li and Hazelbauer, 2011), CheA dimer and CheW monomer (Li and Hazelbauer, 2011; Falke and Piasta, 2014). Through the formation of large, highly organized clusters known as chemosensory arrays, thousands of core-signaling units establish a network of cooperative interactions that dramatically affect signal transmission and regulation and endow the basic two-component chemotaxis infrastructure with heightened information processing and control capabilities (Hazelbauer and Lai, 2010; Falke and Piasta, 2014; Sourjik and Armitage, 2010; Bray et al., 1998; Tu, 2013). Important progress has been made in the characterization of localized portions of array structure using a battery of genetic, biochemical, and biophysical techniques. This progress includes the derivation of atomic structures of the individual core signaling components (Kim and Yokota, 1999; Bilwes et al., 1999; Park et al., 2006; Li et al., 2007; Griswold et al., 2002) and several of their sub-complexes (Park et al., 2006; Li and Bayas, 2013; Briegel et al., 2012) as well as the elucidation of key interactions between the core signaling components in soluble multi-protein complexes (Bhatnagar et al., 2010; Vu et al., 2012; Wang et al., 2012) and in reconstituted, attractant-regulated core complexes (Li and Hazelbauer, 2011; Piasta et al., 2013; Natale et al., 2013; Falke and Piasta, 2014; Li and Hazelbauer, 2014).
Recently, a global view of the extended structural organization of chemosensory arrays has emerged from cryo-electron tomography (cryoET) studies of native bacterial cells (Briegel et al., 2009; 2012; Liu et al., 2012; Zhang et al., 2007). Specifically, chemoreceptor TODs were observed to form hexagonal arrays with a 12 nm lattice spacing conserved across several, distantly related bacterial species including E. coli and T. maritima (Briegel et al., 2009; 2012; 2014a; Liu et al., 2012; Zhang et al., 2007). The conservation of this hexagonal organization has also been demonstrated in non-membrane spanning cytoplasmic chemosensory arrays (Briegel et al., 2014a; 2014b). Additionally, studies using cryoET with sub-tomogram averaging, in tandem with crystallographic structures of portions of the core complex, have reported the extended structure of the array to consist of receptor TODs packed in a two-facing-two fashion about kinase-filled and kinase-empty rings (Briegel et al., 2012; 2014b; Liu et al., 2012). However, due to the thickness of the cells as well as cellular crowding and heterogeneity, past cellular tomography studies have been limited to discerning only the overall arrangement of the core signaling components.
The lack of a high-resolution description of the intact and extended chemosensory array structure has hindered the development of a detailed understanding of molecular events occurring within the array during signaling. To address this problem, we have taken a joint experimental-computational approach. In particular, we have developed a novel reconstitution method yielding ultra-thin monolayer samples of core-signaling complex arrays, from which we derived a three-dimensional density map of the reconstituted core-signaling complex at 11.3 Å resolution using cryoET and sub-tomogram classification and averaging. Through the computational synthesis of existing X-ray crystallography data and our new cryoET data, we have constructed an atomic model of the extended chemosensory array. Our model highlights novel interaction interfaces between the receptor, CheA, and CheW and permits the use of large-scale, all-atom molecular dynamics (MD) simulations (Perilla et al., 2015) to further illuminate the molecular details of a key kinase-signaling event.
Results
Reconstitution of bacterial chemotaxis core-signaling complex arrays
To overcome the limitations imposed by cellular tomography of native chemosensory arrays (Briegel et al., 2009; 2012; Liu et al., 2012; Zhang et al., 2007), we elected to establish an in vitro reconstituted system for high-resolution structural analysis of the signaling complex. Inspired by the template-directed method to assemble functional signaling complexes on lipid vesicles (Montefusco et al., 2007; Shrout et al., 2003), we designed a Ni2+-NTA lipid containing monolayer system (Taylor et al., 2007; Taylor and Taylor, 1999) to reconstitute the two-dimensional (2D) arrays of signaling complexes for structural analysis. To this end, we expressed and purified to high homogeneity E. coli chemotaxis proteins: CheA, CheW, and a His-tagged cytoplasmic signaling domain of the wild-type (wt) Tar receptor (TarCF). His-tagged TarCF can be readily incorporated into the Ni2+-NTA lipid monolayers, seen as homogeneous particles in the EM micrographs of negatively stained specimen (Figure 1A). Only in the presence of all three components (TarCF, CheA, and CheW) were ordered arrays evident (Figure 1B), and even then, these microcrystalline 2D arrays were only formed under strictly constrained input ratios of the three components, a finding that is consistent with previous results indicating that the chemotactic function of the complex is diminished when one of the components is reduced or over-produced (Zhang et al., 2007). The optimal condition for array formation was established to be a mixture of TarCF, CheA and CheW with a molar ratio of 9:18:18 μM for TarCF:CheA:CheW in a lipid monolayer containing 2:1 DOPC:DOGS-NTA-Ni2+ lipids (33% Ni2+-NTA lipid). Notably, the input molar ratio of the reconstitution mixture does not reflect the actual ratio of components incorporated into the monolayer, as illustrated in Figure 1C. The resulting arrays are organized in hexagonal lattices with 12 nm spacing (Figure 1B inset, white arrow), resembling the arrays formed in native cells (Briegel et al., 2012; Liu et al., 2012).

Reconstitution of 2D arrays of the receptor signaling complex on lipid monolayers.
(A&B) Negatively stained electron micrographs of reconstituted lipid monolayers with TarCF only (A) or with TarCF/CheA/CheW (B). Inset, Fourier transform of a region from the monolayer array, indicating a hexagonal lattice with a 12 nm repeat (white arrow). Scale bars, 100 nm. (C) SDS-PAGE gel analysis of the reconstituted monolayer sample (lane 2) and the protein solution (input mixture) used to generate the monolayer arrays (lane 3). Molecular weight markers are indicated (lane 1), and the input mixture contained TarCF:CheA:CheW in a ratio of 9:18:18 μM.
CryoET of the chemotaxis core-signaling complex arrays
Compared to previous cellular tomography studies (Briegel et al., 2012; 2014b; Liu et al., 2012; Zhang et al., 2007), the reconstituted monolayer system is ideal for high resolution structural analysis of chemosensory arrays by cryoET for several reasons: 1) the in vitro reconstituted monolayer array is thin (25 nm) and pseudo-crystalline, compared to cells with thicknesses ranging from 500 nm to 1 µm; 2) the monolayer arrays are reconstituted with purified components, hence the system is well-defined, in contrast to native arrays in the crowded cellular environment; 3) the reconstituted system allows for control over which array components are present as well as manipulation of their signaling state; 4) the in vitro system provides large numbers of sub-tomogram volumes (~3000 core-signaling units/tomogram), thereby improving the noise statistics of the sub-tomogram averaging process central to achieving a high resolution structure. Using cryoET, we collected and reconstructed, correcting for the contrast transfer function (CTF) of the microscope (Fernández et al., 2006), 20 tomograms of monolayers containing reconstituted core-signaling complex arrays. Figure 2A (Video 1) shows a typical raw tomographic slice (without CTF correction) of a reconstituted monolayer, illustrating patches of 2D lattices with information extending beyond 22 Å (inset, arrow). By extracting and classifying CTF-corrected sub-tomograms, centered on each hexagon of receptor TODs (Figure 2—figure supplement 1, yellow circle), we obtained two major classes of the receptor hexagons: one containing a trimer of core-signaling units (CheA2-trimer, Figure 2B and Figure 2—figure supplement 1, cyan boxes) and one containing a hexamer of core-signaling units (CheA2-hexamer, Figure 2C and Figure 2—figure supplement 1, orange box). By mapping the individual sub-tomograms from the above two classes onto the original contributing tomograms, we were able to extract the extended lattice organization of the subunits in the monolayer (Figure 2D), revealing an interlocking of the CheA2-trimer and CheA2-hexamer classes (Figure 2E) consistent with that seen in cellular tomograms (Briegel et al., 2012; 2014b; Liu et al., 2012).

CryoET of monolayer arrays of TarCF/CheA/CheW ternary signaling complex.
(A) A tomographic slice (1.2 nm thick) through the reconstituted monolayer arrays of TarCF/CheA/CheW, without CTF correction. Inset, The Fourier transform of a selected region, displaying Thon rings with information extended to at least 22 Å resolution (arrow). (B&C) Averaged density maps of two sub-volume classes containing receptor hexagons (6 TODs) (red), one with a trimer of CheA dimers (CheA2-trimer) (B) and the other with a hexamer of CheA dimers (CheA2-hexamer) (C). Maps were generated following sub-tomogram volume classification and class-averaging, are contoured at 1.5σ, and are colored according to the height, from the receptor at the top (red) to CheA (blue) below. (D) Spatial arrangement of the CheA2-trimer (cyan) and CheA2-hexamer (orange) in the monolayer lattice array, after mapping the classified sub-volumes back onto the tomogram. The array is formed by interlocking CheA2-trimer and CheA2-hexamer subunits. (E) A schematic lattice model for the chemosensory arrays. Small circles represent receptor dimers; arrows represent CheA dimers (CheA2). Dashed cyan and orange circles highlight a CheA2-trimer and CheA2-hexamer respectively. The lattice unit cell is outlined in black. Related to Figure 2—figure supplement 1 and 2.
Tomographic slices of monolayer arrays.
Related to Figure 2.
To directly compare the lattice organization in our reconstituted monolayer system with that in native E. coli cells, we obtained three CTF-corrected tomograms from wt E. coli cells that were partially lysed, using a phage-gene-induced instant lysis method that we developed recently (Fu et al., 2014), to reduce the sample thickness. Extracting and classifying sub-tomograms containing receptor hexagons from the native E. coli cells revealed the same two classes that were observed in the monolayer system. As with the in vitro monolayer system, the two receptor hexagon classes observed in native E. coli cells also formed an interlocking lattice (Figure 2—figure supplement 2A). Extracting sub-tomograms with an extended unit that contained both the CheA2-trimer and CheA2-hexamer, we obtained an average density map for the in situ native chemosensory arrays that overlapped very well with the map from the monolayer system (Figure 2—figure supplement 2B). Therefore, the in vitro reconstituted monolayer system with purified E. coli proteins faithfully reproduces the lattice organization found in native E. coli cell membranes.
3D density maps of CheA2-trimer and CheA2-hexamer
The 3D classification process further improved the resolution of the class-averaged sub-tomograms of CheA2-trimers (Figure 2B) to 11.3 Å and CheA2-hexamers (Figure 2C) to 17.5 Å resolution, as measured by gold-standard Fourier shell correlation (FSC) (Figure 3—figure supplement 1A). A uniform distribution of in-plane orientations of the sub-tomograms and a relatively well sampled, out-of-plane angle enhanced the quality of the averaged density maps (Figure 3—figure supplement 1B&C). Nevertheless, some resolution anisotropy exists, with 11 Å in X and Y directions and 15.8 Å in Z direction (Figure 3—figure supplement 1A). To take the effect of the anisotropic resolution into account, we low-pass filtered the density map according to the FSC of the Fourier conical shells along various directions (Diebolder et al., 2015). The resulting maps of the CheA2-trimer and CheA2-hexamer clearly delineate the density regions corresponding to the receptor, the CheA-P5/CheW ring at the receptor tip, and CheA kinase domain (Figure 3B&C, Figure 3—figure supplement 2 and further display a number of new features. In particular, the individual receptor dimers are thoroughly resolved, allowing the kinase and CheW/receptor interactions to be isolated to a specific receptor dimer (Figure 3A–C). Moreover, the position of the previously unobserved four-helix bundle of the CheA-P3 dimerization domain is clearly discerned to run parallel to the receptor and is positioned close to CheW-interacting receptor dimers (Figure 3A–C). In addition, our maps dramatically refine the area of density projecting below the CheA-P5 domain, suggesting that the CheA-P4 kinase domain alone occupies this density region (Figure 2B, 3A, and Video 2). The CheA-P1 and CheA-P2 domains, on the other hand, are not resolved, likely due to their conformational flexibility.

CheA2-trimer and CheA2-hexamer density maps with molecular dynamics flexible fitting (MDFF) of computationally constructed T. maritima subunit models.
(A) Overall fitting of the CheA2-trimer density map contoured at 1.5σ. The three core signaling complexes are colored in pink, blue and green. (B) A sectional view of the boxed region in A, rotated 90°. The protein components are labeled at the indicated height of the complex (gray boxes). (C) Sectional views of the gray-boxed regions in B at the receptor level (top), the CheA-P3 and P5/CheW ring region (middle), and CheA-P4 region (bottom). (D) Overall fitting of the CheA2-hexamer density map contoured at 1.5σ. (E) A sectional view at the CheA-P3 and CheW-ring region of the CheA2-hexamer density map. In (A-E), CheA-P3, P4, P5, CheW and receptor are labeled as P3, P4, P5, W and R, respectively, and the CheA-P5/CheW interfaces 1 and 2 are indicated. Related to Figure 3—figure supplement 1.
MDFF model fitting of the CheA2-trimer density map.
Related to Figure 3.
Regarding the CheA-P5/CheW ring, our density map clearly shows a pseudo three-fold symmetry (Figure 2B) in which the density at the CheA-P5/CheW interface between core-signaling units (interface 2) is considerably weaker than the density at the CheA-P5/CheW interface within core-signaling units (interface 1) (Figure 3A&C) (Briegel et al., 2012). This finding is in contrast to the previously described pseudo six-fold symmetry of the CheA-P5/CheW ring (Li and Bayas, 2013). Most importantly, the previously described ‘empty hexagon’ that is surrounded by six CheA-occupied hexagons (Briegel et al., 2012) is not empty, but rather contains a well-ordered continuous ring of densities (Figure 2C) that we were able to unambiguously assign to individual CheW monomers (Figure 3D&E). This ring of CheW, as previously speculated (Liu et al., 2012), provides additional interactions that couple neighboring receptor TODs and strengthens the interlocking baseplate. Hence, our maps confirm the existence of the CheW ring and establish its participation in the structural foundation responsible for the ultra-stability of the chemosensory array (Liu et al., 2012; Erbse and Falke, 2009) and for the high cooperativity and extraordinary sensitivity measured in chemotaxis responses (Goldman et al., 2009).
All-atom model of the T. maritima chemosensory array
The resolution of our cryoET data permitted the unambiguous assignment of distinct regions of density to specific protein components, enabling the construction of all-atom models of the chemosensory array substructures and extended lattice (Figure 4—figure supplement 1). A schematic overview of the modeling procedures carried out in this study is provided in Figure 4—figure supplement 2 with a more detailed discussion of these procedures located in the Methods section. Briefly, we first constructed models of the receptor TOD, CheA-P3P4 dimer, CheA-P5/CheW ring, and CheW-only ring, taking advantage of existing high-resolution X-ray structures from the thermophilic bacterium Thermotoga maritima (Kim and Yokota, 1999; Bilwes et al., 1999; Park et al., 2006; Li and Bayas, 2013). We then heuristically-arranged, using a 12 nm lattice constant, the resulting component models to produce models of the CheA2-trimer and CheA2-hexamer subunits identified by sub-tomogram classification (Figure 4—figure supplement 1D&E). To further refine the key protein-protein interfaces within our atomic models, we adopted a dual MD-based strategy, utilizing both unbiased MD and electron-density-biased molecular dynamics flexible fitting (MDFF) simulations (Trabuco et al., 2008). For the subject of our unbiased refinement simulations, we extracted from the CheA2-hexamer model a portion corresponding to the array unit cell, including six receptor TODs, three CheA dimers, and 12 CheW monomers all together arranged as three coupled core-signaling units (Figure 4—figure supplement 1C, herein the 'unit-cell model'). For our density-biased refinement simulations, we focused our efforts on the CheA2-trimer model, owing to the higher-resolution of its associated density map, and hence, better resolved MDFF biasing forces. Because the CheA-P4 density is not as well defined as the other parts of the complex, likely due to its conformational flexibility, we carried out a rigid-body docking of the CheA-P4 domain, starting from 10,000 random angular orientations and up to 20 Å shifts from the center of the mass. This fitting exercise resulted in 23 classes separated by 3° and 3 Å (Figure 3—figure supplement 3A), generating a metric for the goodness of fit of the P4 domain positioning. In addition to the class of 'best fit'(Figure 3—figure supplement 3B, panel 1), one other class, in which P4 is flipped relative to the best fit, was seemingly structurally possible (Figure 3—figure supplement 3B, panel 5). However, compared to the best-fit class, this alternative class had a lower cross-correlation value, lower occupancy with only a third the number of contributing fits, and the positions of P4-N and C termini are reversed (flipped), making it hard to connect the P3 and P5 termini with short linkers. Thus, we have focused our efforts and resources on the highest ranking class of P4 position. It should be noted, though, that use of the alternative P4 positioning might produce considerably different MD trajectories. Solvation and ionization of the unit cell and CheA2-trimer models produced systems of size 1.25 million and 1.75 million atoms, respectively, which were subsequently energy minimized and equilibrated for 10 ns, as described in the Methods section. The unit-cell model was then subjected to an 80 ns unconstrained production simulation (Video 3), while a 70 ns symmetry-constrained MDFF simulation was used to computationally bias the tertiary structure of the protein components within the CheA2-trimer model according to our 11.3 Å CheA2-trimer density map (Figure 3, Video 2).
The resulting unit-cell and CheA2-trimer models agreed well with previous structural studies, in particular with respect to the residues participating in the CheA-P5/receptor and CheW/receptor interaction interfaces, as defined by NMR (Vu et al., 2012; Wang et al., 2012; Ortega et al., 2013), crystallography (Li and Bayas, 2013), and disulfide mapping studies (Piasta et al., 2013; Natale et al., 2013). For succinctness, specific residues participating in the various protein-protein interfaces within the array have been listed in Supplementary file 1. In addition, the equilibrated model of the T. maritima TOD maintained the conserved trimer-forming contacts observed in the X-ray structure of the E. coli serine receptor (Tsr) (Kim and Yokota, 1999) and revealed two additional trimer-stabilizing salt bridges, namely E387/R389 (conserved as E402/R404 in E. coli Tsr) and E351/R403 (structurally homologous to D363/R415 in E. coli Tsr) (Figure 4—figure supplement 3A). Moreover, in both models, the CheA-P4 kinase domain was seen to stably occupy the region of density directly below the plane defined by the CheW and CheA-P5/CheW rings. Finally, in tandem with the direct visualization of the CheA-P3 dimerization domain in our cryoET density maps, the all-atom model further revealed previously uncharacterized specific interactions between the P3 bundle and adjacent receptors, involving D333/K390 and D345/R379 contact pairs (I304/N405 and D316/R394 in E. coli respectively) (Figure 4B, Supplementary file 1).

CheA dimer conformational switch.
(A) Top and side views of the core-signaling unit, consisting of two receptor TODs (red), a single CheA dimer (blue), and four CheW monomers (green). (B) Two distinct classes, undipped (top) and dipped (bottom), of core-signaling unit structures are present in our MD simulations. Classes differ especially in the orientation of CheA-P4 domain with respect to the rest of the CheA dimer and core signaling unit. Specific contacts that stabilize either conformation are indicated for T. maritima and in parentheses for the corresponding residues in E. coli. CheA-P5 and CheW have been removed for clarity. (C) Time series of CheA dimer conformations extracted from unit cell simulations.. Traces track the projection of the conformations of 27 CheA dimers from the wt (top) and R297A mutant (bottom) unit cell simulations onto the first principal component of the 'dipping' motion. Colored traces track CheA dimers that undergo an extended (>10 ns) 'dipping' motion. Horizontal dashed lines visually demarcate the undipped and dipped CheA dimer classes. Vertical dashed lines separate the initial 80 ns equilibration simulation from nine 450 ns production simulations. Related to Figure 4—figure supplement 1, 2, 3, 4 and 5, and Supplementary file 1.
Molecular dynamics simulation of array unit cell.
Shown here is a 75 ns clip of a wild type unit cell trajectory, illustrating the dynamics of the 1.2 million atom model, including 6 receptor TODs (red), 3 CheA dimers (blue), and 12 CheW monomers (green). Periodic images, shown here with reduced opacity, enforce the boundary conditions of the extended array architecture but are not simulated explicitly. Solvent and ions have been removed for clarity. Related to Figure 4.
A conformational change of the CheA kinase domain
The construction of atomic models of the array unit cell and subunits permitted the use of equilibrium all-atom MD simulations to further investigate the molecular details of dynamic events potentially relevant to signaling. An overview of the key MD simulations conducted in this study is given in Figure 4—figure supplement 4A. The unit cell system contains the minimal set of components needed to represent the full receptor signaling array, which was made possible through the use of periodic boundary conditions to mimic the bulk symmetry of the chemosensory array, preventing the need to interpret potentially problematic effects due to unconstrained boundaries (Figure 4—figure supplement 2C, Video 3). We conducted a series of nine simulations of 450 ns each, using the equilibrated unit-cell model; additionally, we ran ten, 120 ns simulations of the equilibrated CheA2-trimer system for comparison with the unit-cell simulations. Intriguingly, our simulations of both models revealed an ensemble of distinct core-signaling unit conformations (Figure 4 B&C), including structures in which the associated CheA dimer displayed either an undipped conformation (Figure 4B, top) or dipped conformation (Figure 4B, bottom). In the latter case, the P4 domain of one CheA monomer adopted a 'dipped' state through rotations about the P3-P4 and P4-P5 flexible linkers, significantly affecting its contacts with neighboring receptor dimers and the P5 domain (Video 4). As many biochemical, biophysical, and mutational studies have implicated dynamic structural changes within these regions of the core-signaling unit during the propagation of signals (Piasta et al., 2013; Natale et al., 2013; Wang et al., 2014; Briegel et al., 2013), we systematically identified the distinct structural classes of core-signaling unit conformations present in our MD simulations and isolated them for comparative analysis. Specifically, we used the UPGMC hierarchical clustering method (Müllner, 2013) to assign the conformations of the 27 core-signaling units sampled in our unit cell simulations (3 units/unit cell) to groups of similar structure based on their pairwise root-mean-square deviation (RMSD). Cross-examination of structures within the resulting core-signaling unit clusters revealed the formation of two new salt bridges stabilizing the 'dipped' state, namely R297/E397 (R265/E368 in E. coli) between the P3 and P4 domains and E390/R379 (E361/R394 in E. coli) between the P4 domain and nearby receptor tip (Figure 4B, bottom). Moreover, to accommodate the reorientation of the P4 domain, the P3 dimerization bundle was observed to break the receptor contacts (D333/K390 and D345/R379) observed in the ‘undipped’ state (Figure 4B, Video 4), suggesting that the mobility of the P3 bundle plays a key role in the conformational dynamics of the CheA dimer.
Molecular dynamics simulations reveal conformational switch in CheA P4 domain.
Shown here is one of four 'dipping' events observed in the wild type unit cell simulations, leading to modified contacts between the CheA dimer and receptor TODs. Strong contacts between P3 and neighboring receptor dimers (D333/K390 shown here with licorice representation) are disrupted in favor of new contacts between P3/P4 and P4/receptor stabilizing the dipped state (R297/E397 and E390/R379 respectively, shown here with licorice representation). Related to Figure 4.
We next sought to examine the temporal evolution of the dipping motion in each of the CheA dimers present in our simulations. For this purpose, we used Principal Component Analysis (PCA) to systematically derive, from the trajectory of a single dipping CheA dimer, a pseudo reaction coordinate by which to easily monitor the progression towards the 'dipped' conformation. A total of four 'dipping' events were observed in our unit cell simulations, as illustrated by projection of the conformations of the 27 CheA dimer time series onto the first principal component (Figure 4C, top). Importantly, an additional two dipping events were observed in the 30 CheA dimers of the relatively shorter simulations of CheA2-trimer model (Figure 4—figure supplement 5), demonstrating that the ability of the conformational change to occur is not an artifact of the particular choice of CheA P4 positioning during modeling. Interestingly, the three extended 'dipping' events observed in the unit-cell simulations (Figure 4C; red, blue, and green traces) as well as the two events observed in the CheA2-trimer simulations were accompanied by the formation of the R297/E397 contact. Notably, this contact was not formed in the one short dipping event, which returned to the 'undipped' bulk state (Figure 4C; gold trace), suggesting that the R297/E397 contact may play a role in stabilizing the 'dipped' state. To further investigate the significance of the R297/E397 contact for the conformational dynamics of CheA, we launched nine additional unit cell simulations with an R297A mutation to prevent the potential formation of the R297/E397 salt bridge. Indeed, while two CheA dimers exhibited the dipping motion in these simulations, including one dimer that underwent two dips, the mutants quickly return to the bulk (Figure 4C, bottom).
Biochemical validation of CheA conformational change in E. coli cells
To determine if the CheA-P4 dipping motion observed in the MD simulations of the T. maritima chemosensory array is sampled in the native chemotactic response of E. coli, we carried out cysteine disulfide cross-linking experiments. In particular, we tested the interaction interface for contacts existing in the undipped state (I304/N405 and D316/R394) or only in the dipped state (E361/R394) (Figure 5B). Notably, in the simulations, R394 of Tsr switches its contact with D316 of CheA-P3 to E361 of CheA-P4 during the transition of the CheA dimer from 'undipped' to 'dipped' (Video 4).

Biochemical validation of alternative CheA conformations in E. coli.
(A) Swimming ability of E. coli cells with mutations in the CheA-P3 and Tsr interface (I304/N405 and D316/R394) and in the 'dipped' CheA-P4 and Tsr interface (E361/R394). Swimming activities are normalized to the cysless CheA and wt Tsr, ± standard deviation (n=6). Inset, representative images of soft agar plates for swimming ability, with specific constructs labeled in red. (B) Disulphide cross-linking of the CheA-P3 and Tsr interface (I304C/N405C and D316C/R394C) in the undipped CheA dimer conformation (top panel of Figure 5B) and the CheA-P4 and Tsr interface (E361C/R394C) occurring in the dipped CheA-P4 'dipped' conformation (bottom panel of Figure 5B). Non-reducing (top) and reducing (bottom) SDS-PAGE gels were analyzed by immunoblotting for Tsr and CheA. Cross-linked species were indicated with blue arrows. (C) Swimming ability of E. coli cells with mutations at R265 of CheA-P3 domain, normalized to the wt, ± standard deviation (n=8). Related to Figure 5—figure supplement 1.
Using soft-agar assays, it was seen that the chemotactic ability of the I304C/N405C double cysteine mutant is appreciably compromised compared to that of the control (cysless CheA/wt Tsr), any of the single mutants (I304C/wt Tsr, cysless CheA/N405C, cysless CheA/N405S), and when one half of the pair has been mutated to serine (I304C/N405S) (Figure 5A), suggesting that dynamic interaction between CheA-P3 and the receptor is important for chemotactic function. Moreover, in vivo cross-linking and western blot analysis showed a high molecular weight band present only in the double cysteine mutant, suggesting the presence of species formed by cross-linking between CheA-P3 and Tsr (Figure 5B). We also examined cross-linking residue pairs that involve Tsr-R394 interactions with CheA, one in the 'undipped' state (CheA-E316C/Tsr-R394) and the other in the 'dipped' state (CheA-E361C/Tsr-R394). When Tsr-R394 is replaced by a cysteine or serine, either as a single mutant (cysless CheA/R394C, cysless CheA/R394S) or in the context of a double mutant (D316C/R394C, D316C/R394S, E361C/R394C, E361C/R394S), the chemotaxis function of E. coli is partially inhibited. On the other hand, the chemotactic ability of CheA-E361C as a single mutant (E361C/wt Tsr) is also partially inhibited, while CheA-D316C (E316C/wt Tsr) mutation bears no effect on the function (Figure 5A). Furthermore, the cross-linking pattern of both Tsr-R394 mutant pairs showed two high molecular weight bands corresponding to distinct cross-linked species, one with a disulfide formed between Tsr and CheA (Figure 5B, upper band, blue arrows) and the other with a disulfide between two Tsr molecules with the R394C mutation (lower band). Interestingly, the cross-linking of CheA-P4/Tsr (E361C/R394C) in the predicted 'dipped' conformation is much weaker than the cross-linking of CheA-P3/Tsr (D316C/R394C) in the 'undipped' state, though both involve the same R394 residue of Tsr (Video 4). The lower cross-linking efficiency could be due to the relatively infrequent occurrence of the CheA 'dipped' conformation, and/or because the residues are further apart in a dominant conformation, suggesting that the CheA-P4 'dipped' conformation observed in silico may have been sampled within the native chemosensory complex of E. coli.
Our MD simulations of the T. maritima unit cell further indicated that R297 on the CheA-P3 domain is potentially involved in the stabilization of the conformational transition of the CheA-P4 (Figure 4C, Video 4). Indeed, substitution of the corresponding residue in E. coli (R265) with several amino acids of different properties (R265C/S/A/E) were all detrimental to the chemotactic function of E. coli as measured by the soft-agar assay, without affecting the cluster formation (Figure 5C, Figure 5—figure supplement 1). Since this residue is located at the N-terminus of the four-helix P3 dimerization motif, R265 could direct the P2-P3 linker away from the cis subunit and toward the trans subunit, thus anchoring CheA-P1P2 to CheA-P4’ for trans-interaction and phosphorylation (Bilwes et al., 1999). A more complete model of the core-signaling complex for E. coli may be necessary to fully interpret the drastic impact of this single CheA residue on the entire chemotactic machinery.
Discussion
While much effort has been expended in the derivation of models to describe the transduction of ligand-binding events within the receptor proteins, including an established piston mechanism and a hypothesized alternating static–dynamic 'yin-yang' on-off switch model (Falke and Piasta, 2014), how the structure and dynamics of the CheA kinase are affected during signaling remains poorly understood. In this study, we identified, using MD simulations, a dipping motion of the CheA P4 domain, which was functionally characterized using swim assay and cross-linking experiments. While the role of the predicted conformational change in CheA is not immediately clarified in the preliminary biochemical experiments carried out here, our model highlights the importance of CheA dynamics for signaling and suggests that the dynamics of the P4 kinase domain, in particular, warrants special investigation. More importantly, the atomic model presented here, in general, provides improved knowledge of the positioning of the P3 and P4 domains, incorporates the presence of the CheW only ring, and identifies probable novel side-chain contacts within the extended chemosensory architecture. Further improving the resolution of our cryoET data to better than 8 Å using the novel lipid-monolayer system described above would allow generation of an atomic homology model of the E. coli chemosensory array, greatly facilitating the use of the wealth of existing biochemical and biophysical data and providing directly transferrable structural and dynamical predictions. We hope that the findings presented here will inspire further experimental and computational studies towards the elucidation of a complete mechanistic description of signal transduction and amplification within this truly impressive biological sensory apparatus.
Materials and methods
Materials
Plasmids and cell strains used in this study were gifts from Dr. Parkinson, University of Utah, except for plasmid pHTCF (kind gift from Dr. Weis, University of Massachusetts, Amherst). Plasmid pHTCF is an IPTG-inducible expression vector for the N-terminal His6-tagged cytoplasmic fragment of wt aspartate receptor (TarCF) that contains residues 257–553. Plasmids pKJ9 and PPA770 are IPTG-inducible for the expression of CheA and CheW, respectively. Plasmid pRR53 is an IPTG-inducible expression vector (ampR) for the wt serine receptor (Tsr). Plasmid pGP26 is a sodium salicylate (Na-S)-inducible expression vector (camR) for cysteine-less CheA and wt CheW. Plasmids pRR53 and pGP26 were used to generate mutations in Tsr and CheA, respectively.
Protein expression and purification
Request a detailed protocolE. coli strain RP3098, which lacks all Che proteins and chemoreceptors, was transformed with plasmid pKJ9 or PPA770 for CheA or CheW expression, respectively. CheA expression was induced at an OD600 of 0.6–0.8, with 1 mM IPTG, overnight at 15°C. CheA was purified using an Affi-gel Blue column (Bio Rad, Hercules, CA) followed by gel filtration on a Superdex 200 column. Further purification with a Mono Q ion exchange column resulted in >99% homogeneity with an overall yield of 50 mg/L of cells. CheW expression was induced by the addition of IPTG (0.5 mM), at an OD600 of 0.4–0.6, at 37°C. CheW was purified through 20%–40% ammonium sulfate precipitation, a DEAE column followed by a MonoQ ion exchange column and a Superdex 75 size exclusion column. This procedure resulted in highly purified CheW with a yield of 6 mg/L of cells. His6-tagged wt TarCF (His6-TarCFQEQE) was expressed in DH5alpha cells with plasmid pHTCF. TarCF was induced by the addition of IPTG (0.5 mM) at an OD600 of 0.4–0.6 at 37°C and purified with a Ni2+-NTA affinity column followed with a mono Q column for quick removal of imidazole, without dialyzing overnight. The yield for TarCF was excellent (120 mg/L of cells).
Monolayer reconstitution
Request a detailed protocolA Ni2+ lipid containing monolayer system was used to reconstitute the chemotaxis core-signaling complex arrays. A mixture of 9:18:18 µM of TarCF:CheA:CheW in a buffer containing 75 mM Tris-HCl, pH 7.4, 100 mM KCl, 5 mM MgCl2 was applied to a Teflon well, over which we immediately lay a lipid monolayer containing 2:1 DOPC:DOGS-NTA-Ni2+ lipid mixture, at 2 mg/ml concentration. The monolayer set up was left undisturbed in a humidity chamber overnight. The monolayer specimen was picked up with holey carbon grids, stained with 1% uranyl acetate, and examined with an FEI T12 microscope operated at 120 KV.
Cryo-electron tomography
Request a detailed protocolReconstituted monolayers using the best conditions identified by negative staining (Figure 1B), were picked up with perforated R2/2 Quantifoil grids (Quantifoil Micro Tools, Jena, Germany) pre-coated with 10 nm fiducial gold beads on the backside of the grid and plunge-frozen using a manual gravity plunger. This method prevents disruption of the monolayer by supporting single-side blotting which eliminates the contact between the blotting filter paper and the delicate monolayer. The frozen-hydrated EM grids were loaded into FEI Polara cartridges and imaged under low-dose conditions using a Tecnai Polara microscope (FEI Corp., OR.) operating at 200kv. A series of low dose projection images were recorded with tilt angles ranging from 70° to -70° with a Gatan 4K × 4K CCD camera (Gatan, Inc., PA), at a nominal magnification of 39,000×, with a defocus value of 5–8 µm and an accumulated dose of ~60 e-/Å2. A total of 32 tomographic tilt series were collected using an FEI automated tomography software.
3D reconstruction, sub-tomogram classification and averaging
Request a detailed protocolOf the 32 tilt series collected, 20 tilt series with negligible mechanical or physical artifacts were selected for image processing and tomographic volume reconstruction. The monolayer produces an ideal EM specimen: it is thin (25 nm) and also provides strong signals in power spectra, due to near-crystalline packing of the protein components (Figure 2A inset), allowing accurate determination of the Contrast Transfer Function (CTF) using strip-based periodogram averaging in TomoCTF (Fernández et al., 2006). The tilted projection series were roughly aligned using IMOD (Kremer et al., 1996), and the alignment parameters were further refined using fiducial-free Area Matching with Geometry Refinement as implemented in Protomo (Winkler, 2007). Using the refined geometry parameters, the raw projections were centered and rotated so the tilt azimuth was coincident with the Y-axis using the IMOD 'newstack'function. These rotated stacks were corrected for the CTF with phase flipping, and volume reconstructions were made using SIRT as implemented in IMOD. These were calculated using a GPU, thereby removing an additional interpolation in the reconstruction step, by avoiding the use of cosine stretching of the input projections. Reconstructed volumes calculated from 20 SIRT iterations, providing higher contrast, were used for the initial cycles of sub-tomogram extraction and alignment, while those from 60 SIRT iterations were used for the final cycles.
To extract sub-tomograms, initial positions of the receptor complexes, respective to a Cartesian grid defined by each tomogram, were approximated by using a template matching algorithm implemented in Matlab with a reference that emphasized the receptor dimers with little influence from CheA. Both the template and tomograms were low-pass filtered to 4 nm and binned by 3. This resolution, as well as a coarse angular search, were chosen to eliminate any statistical correlation of high resolution information between half data sets in later image processing steps. Following template matching sub-volume extraction, the data were randomly segregated into two groups, which were processed independently for all subsequent steps.
Sub-tomogram alignment and classification were carried out using Protomo's i3 image processing utilities (Winkler, 2007). Using Multivariate Statistical Analysis and Hierarchical Ascendant Classification, eight class averages were produced from each half data set by focusing the analysis on the CheA portion of the complex. Initial references for each half set were generated by choosing averages from eight classes. These references were then used to align class averages chosen to each have ~50 contributing sub-volumes. In the following cycle, the raw sub-tomograms were subject to multi-reference alignment, but only a small in-plane and translational adjustment was allowed. This alignment by classification was repeated five times, while allowing the automatic exclusion of high variance outliers after the second cycle. In addition to the CheA2-trimer and CheA2-hexamer classes (Figure 2B, C), divergent organizations of CheA/receptor complex were also included as references (Figure 2—figure supplement 1). After the final cycle, class averages containing either CheA2-trimer or CheA2-hexamer were manually selected and averaged together for each half data set, and the corresponding gold-standard FSC was calculated to evaluate the reliability of the data. Soft cylindrical masks were used, rather than spherical masks, given the extended slab like nature of the specimen. The final averages of CheA2-trimer or CheA2-hexamer from two half data sets of 3,000 sub-volumes or 300 sub-volumes, respectively, were combined and an empirical correction for the CTF envelope was applied for sharpening, which helped to further clarify the receptor dimers, as well as the P3 dimerization domain.
To access the degree of resolution anisotropy, conical Fourier shell correlations from the two independent half data sets of CheA2-trimer, along each of the principal axes, as well as the 10 axes bisecting them, were calculated (Diebolder et al., 2015). The averaged density map of CheA2-trimer was then low-pass filtered according the conical FSCs along three principle axes by using cones with a 42° half-angle, adjusted for any overlapping regions in reciprocal space.
Computational modeling of T. maritima core components: Receptor trimer-of-dimers (TOD)
Request a detailed protocolA model of the cytoplasmic portion of the T. maritima receptor dimer was taken from the X-ray crystal structure of the TM1143 chemoreceptor (PDB 2CH7) (Park et al., 2006). Using the E. coli receptor TOD (PDB 1QU7) (Kim and Yokota, 1999) as a reference, a T. maritima receptor TOD model (Figure 4—figure supplement 1A, Figure 4—figure supplement 2A) was obtained by arranging individual receptor dimer models from the previous step so that homologous trimer-forming contacts were preserved. CheA-P34: An atomic model of the soluble T. maritima CheA dimer, including the dimerization (P3) and kinase (P4) domains, was based on atomic coordinates from the X-ray crystal structure PDB 1B3Q (Bilwes et al., 1999). CheA-P5/CheW and CheW rings: Atomic models for both the CheA-P5/CheW and CheW rings were based on the X-ray crystal structure of the Receptor/CheA-P5/CheW ternary complex, PDB 4JPB (Li and Bayas, 2013). In the case of the CheW ring model, the P5 domains of the CheA-P5/CheW ring model were exchanged with CheW monomers, using the dual-SH3-like fold shared between by CheA-P5 and CheW, to obtain an appropriate placement and orientation with respect to the neighboring monomers. Figure 4—figure supplement 3 schematically summarizes the modeling procedures described above. All missing loops were added using MODELLER (Sali, 1993). The TOD, CheA-P5/CheW, and CheW ring core component models were subjected to 150 ns of equilibration to ensure their structural integrity.
Construction of T. maritima array subunit models
Request a detailed protocolThe CheA2-trimer and CheA2-hexamer subunits models (Figure 4—figure supplement 1D&E) were constructed heuristically; using as a visual reference the extended organization of kinase-filled and kinase-empty rings evident in our density maps to arrange the components, also assuming an approximate 12 nm lattice constant (Figure 4—figure supplement 3B). Next, we made use of the CheA-P5/receptor interface from the ternary complex structure PDB 4JPB (Li and Bayas, 2013) to model the CheW/receptor interface, assuming a receptor-binding mode homologous to that of CheA-P5. Using the CheA-P5 and CheW monomer/receptor models from the previous step, positional constraints on the receptor TODs were set relative to the height and orientation of the protein rings. Finally, CheA-P3,4 core component models were placed between adjacent TODs in accordance with the patterns observed in our density maps and joined to nearby ring-bound regulatory domains (P5) at the P4-P5 flexible linker. From the CheA2-hexamer model we then extracted a portion corresponding to the array unit cell (Figure 4—figure supplement 2C, Figure 4—figure supplement 3C) for further study with all-atom MD simulations. In addition, symmetry-constrained molecular dynamics flexible fitting (MDFF) simulations (Trabuco et al., 2008) were used to refine the overlap between our experimental densities and heuristically constructed CheA2-trimer and CheA2-hexamer subunit models (Figure 4—figure supplement 3D, Video 2). The Situs modeling package (Wriggers, 2010), was used to rigidly dock the subunit models into their respective cryoET maps to provide the initial overlap for our MDFF simulations.
Molecular dynamics simulations
Request a detailed protocolThe array unit cell model was hydrated with TIP3P water molecules using VMD’s solvate plugin (Humphrey et al., 1996), producing a simulation box defined by hexagonal lattice parameters a=208 Å, b=208 Å, c=334 Å, α=90°, β=90°, γ=120°. Using VMD’s autoionize plugin, the hydrated system was then neutralized and subsequently ionized with sodium and chloride ions to the physiological concentration of 150 mM, resulting in a model containing 1,153,756 atoms. The unit cell model was then subjected to a series of conjugant gradient energy minimizations (300,000 steps in total) and restrained NPT equilibration simulations (10 ns in total). In the same fashion, the CheA2-trimer and CheA2-hexamer subunit models were hydrated and ionized to produce systems of size 1,751,375 atoms (245x245x310 Å) and 4,588,588 atoms (385x405x310 Å) respectively. Each subunit model was then subjected to the same minimization (300,000 steps) and restrained NPT equilibration (10 ns) scheme as the unit cell model. An outline of subsequent equilibration and production simulations is given in Figure 4—figure supplement 4A. Production simulations of the unit cell and MDFF-refined CheA2-trimer models were conducted with weak (spring constant = 0.1 kcal/mol*nm2) harmonic restraints placed on the alpha carbons of the first five membrane-proximal receptor residues to maintain TOD splay in the absence of membrane and crowding agents. In the case of the post-MDFF production simulations of the CheA2-trimer, additional weak harmonic constraints were placed on the outermost CheW and CheA-P5 domains to enforce the bulk array boundary conditions, as the trimer organization does not permit the use of periodic boundary conditions to represent the necessary symmetry.
All molecular dynamics simulations were performed using the parallel molecular dynamics code, NAMD 2.9 (Phillips et al., 2005) and CHARMM22 force field (MacKerell et al., 1998) with CMAP corrections (Mackerell et al., 2004). Equilibrium simulations were conducted in the NPT ensemble with isobaric and isothermal conditions maintained at 1 atm and 330 K for equilibration, or 350 K for production using the Nosé-Hoover Langevin piston, with a period 200 femtoseconds (fs) and relaxation time of 50 fs, and the Langevin thermostat with a temperature coupling of 5 ps-1. The r-RESPA integrator scheme with an integration time step of 2 fs was used (Phillips et al., 2005). SHAKE constraints were applied to all hydrogen atoms. Short-range, non-bonded interactions were calculated every 2 fs with a cutoff of 12 Å and long-range electrostatics were evaluated every 6 fs using the particle-mesh-Ewald (PME) method with a grid size of 1 Å. Periodic boundary conditions with fixed cross-sectional area (x-y plane) were used. MDFF simulations were performed in the NVT ensemble at 330 K using the settings described above with additional restraints applied to prevent loss of secondary structure, chirality errors, and the formation of cis-peptide bonds.
Simulation analysis
Request a detailed protocolVisualization and extraction of raw trajectory data for analysis were performed using VMD. Principal Components Analysis (PCA) was carried out using custom scripts (source code file PCA.py) involving the Numpy, Scipy, and MDAnalysis python packages (Michaud-Agrawal et al., 2011). For the PCA analysis, a single dip-exhibiting CheA dimer was isolated from one of our wild-type unit cell simulations, and each frame (23,331 frames in total) was aligned to the initial CheA dimer model using the P5 domains (residues 543–671). Principal components were computed using the alpha carbons of the P4 domains (residues 352 to 542). The fractional variances accounted for by the top three modes were 41.8%, 31.1%, and 8.1% respectively. Subsequently, the three CheA dimers from each replica of the wild-type unit cell model (27 dimers total), R297A unit cell model (27 dimers total), and CheA2-trimer model (30 dimers total) simulations were extracted, aligned to the P5 domains, and projected on to the top principal component of the wild-type dip-exhibiting CheA dimer. These projections were grouped according to model type to create Figure 4C (top and bottom) and Figure 4—figure supplement 5. Illustrations of the PCA results were produced using the python-plotting package, Matplotlib. Clustering analysis was performed using custom scripts (source code file clustering.py) involving the python packages noted above as well as the implementation of the UPGMC hierarchical, agglomerative clustering algorithm from the fastcluster package (Müllner, 2013). For the clustering analysis, we first extracted the three core-signaling units from each of the nine wild-type unit cell replica simulations, using 1500 frames/core-signaling unit for a total of 40,500 frames. The RMSD distance matrix was then computed using the ‘rms_fit_trj’ function from the MDAnalysis package. Our analysis identified four major clusters of structures within the above distance matrix with relative populations of 80%, 10%, 10% and 2%, representing the undipped and dipped CheA dimer states as well as two intermediate states respectively.
Mutagenesis
Specific mutations on CheA and Tsr were generated by site-directed mutagenesis on the background of cysteine-less CheA (pGP26) (Miller et al., 2006; Zhao and Parkinson, 2006) and wt Tsr (pRR53), respectively. Each mutation was introduced using a pair of primers (Integrated DNA Technologies, Inc., Coralville, Iowa), complementary to the template except for the site of mutation, and PfuUltra II Fusion HS DNA polymerase (Agilent Technologies, Santa Clara, CA ), following the manufacturer’s thermocycling parameters. The presence of the mutations was confirmed by DNA sequencing.
Cross-linking and western blot analysis
Request a detailed protocolStarter cultures were grown in LB broth (10% tryptone/5% yeast extract/10% NaCl), supplemented with appropriate antibiotics, overnight at 37°C with 250-rpm shaking. Subsequently, the overnight cultures were diluted 1:50 into a 5-ml LB broth, supplemented with the appropriate antibiotics and allowed to grow at 37°C with 250-rpm shaking. When the optical density at 600 nm of the cultures reached ~0.8, cells were induced with 100 μM IPTG and 0.6 μM Na-S, in the presence of 100 μM serine, for 1 hr at 37°C. After induction, cells were collected by centrifugation (3000 x g, 4°C for 10 mins) and then re-suspended in cold PBS, in the presence of 100 μM serine. Cross-linking was initiated by addition of 60 μM copper (II) sulfate and 200 μM phenanthroline (1 hr, RT) and stopped by addition of 20 mM iodoacetamide and 3.7 μM neocuproin. Cells were immediately mixed with 4× NuPAGE lithium dodecyl sulfate/PAGE sample buffer (Invitrogen Corp., Carlsbad, CA), with or without reducing agent (dithiothreitol), and then boiled for 5 min before electrophoresis. Samples were analyzed on 4–12% SDS-PAGE gels in MES running buffer ((Invitrogen Corp., Carlsbad, CA)). Gels were transferred to nitrocellulose membranes, blocked, and immunoblotted by using antiserum against Tsr (1:2500) and CheA (1:1250) (gifts from Dr. Subramaniam, NIH), followed by an alkaline phosphatase conjugated anti-rabbit antibody (1:50,000, Sigma). Bands were detected on the membrane using an NBT/BCIP kit (Promega Corporation, Madison, WI)following the manufacturer’s instructions.
Soft agar assays
Request a detailed protocolThe UU2682 strain does not express any of the chemoreceptors, CheA or CheW, rendering it non-chemotactic despite the presence of an intact flagellar system. Presence of both pRR53 (wt-Tsr) and pGP26 (cysless CheA and wt-cheW) is required to rescue the chemotaxis of UU2682, observed as formation of attractant rings on a soft-agar media. To assess the effect of mutations on Tsr and/or CheA on the chemotactic ability of E. coli, the mutant plasmids were introduced into the UU2682 strain and assayed for formation of attractant rings. The soft agar assay protocol used here is adapted from the Parkinson laboratory (University of Utah). Fresh colonies were plated on LB-agar media (10% tryptone/5% yeast extract/10% NaCl/10% agar), supplemented with carbenicillin (100 μg/ml) and chloramphenicol (34 μg/ml), and grown overnight at 37°C. Next day, using a fine-tip toothpick, colonies were picked from the fresh LB-agar plate and stabbed into a soft-agar media (10% Tryptone/5% yeast extract/5%NaCl/0.27% agar) containing antibiotics (carbenicillin 50 μg/ml, chloramphenicol 17 μg/ml), inducers (100 μM IPTG and 0.6 μM Na-S) and 100 μM serine. Plates were then incubated at 32°C for ~8 hr and the diameter of attractant rings immediately measured after incubation.
Data availability
-
Structure of bacterial chemotaxis signaling CheA2-trimer core complex by cryo-electron tomography and subvolume averagingPublicly available at the EBI Protein Data Bank (Accession no: EMD-6319).
-
Structure of bacterial chemotaxis signaling CheA2-hexamer core complex by cryo-electron tomography and subvolume averagingPublicly available at the EBI Protein Data Bank (Accession no: EMD-6320).
-
Cryo-electron Tomography and All-atom Molecular Dynamics Simulations Reveal a Novel Kinase Conformational Switch in Bacterial Chemotaxis SignalingPublicly available at the RCSB Protein Data Bank (Accession no: PDB-3JA6).
-
Tomogram of the reconstituted monolayer of bacterial chemotaxis core signaling complexPublicly available at the Electron Microscopy Data Bank (Accession number EMD-3234).
-
Crystal Structure of the Cytoplasmic Domain of a Bacterial Chemoreceptor from Thermotoga MaritimaPublicly available at the RCSB Protein Data Bank (Accession no: 2CH7).
-
Crystal Structure of CHEA-289, A Signal Transducing Histidine KinasePublicly available at the RCSB Protein Data Bank (Accession no: 1B3Q).
-
The structure of a ternary complex between CheA domains P4 and P5 with CheW and with an unzipped fragment of TM14, a chemoreceptor analog from Thermotoga maritimaPublicly available at the RCSB Protein Data Bank (Accession no: 4JPB).
References
-
The chemoreceptor dimer is the unit of conformational coupling and transmembrane signalingJournal of Bacteriology 192:1193–1200.https://doi.org/10.1128/JB.01391-09
-
Universal architecture of bacterial chemoreceptor arraysProceedings of the National Academy of Sciences of the United States of America 106:17181–17186.https://doi.org/10.1073/pnas.0905181106
-
Bacterial chemoreceptor arrays are hexagonally packed trimers of receptor dimers networked by rings of kinase and coupling proteinsProceedings of the National Academy of Sciences of the United States of America 109:3766–3771.https://doi.org/10.1073/pnas.1115719109
-
Evolution of two-component signal transduction systemsAnnual Review of Microbiology 66:325–347.https://doi.org/10.1146/annurev-micro-092611-150039
-
Conical fourier shell correlation applied to electron tomogramsJournal of Structural Biology 190:215–223.https://doi.org/10.1016/j.jsb.2015.03.010
-
Architecture and signal transduction mechanism of the bacterial chemosensory array: progress, controversies, and challengesCurrent Opinion in Structural Biology 29:85–94.https://doi.org/10.1016/j.sbi.2014.10.001
-
CTF determination and correction in electron cryotomographyUltramicroscopy 106:587–596.https://doi.org/10.1016/j.ultramic.2006.02.004
-
Signal amplification in a lattice of coupled protein kinasesMolecular BioSystems 5:1853–1859.https://doi.org/10.1039/b903397a
-
Sensory transduction in escherichia coli: role of a protein methylation reaction in sensory adaptationProceedings of the National Academy of Sciences of the United States of America 74:4964–4968.https://doi.org/10.1073/pnas.74.11.4964
-
The solution structure and interactions of CheW from thermotoga maritimaNature Structural Biology 9:121–125.https://doi.org/10.1038/nsb753
-
Bacterial chemoreceptors: high-performance signaling in networked arraysTrends in Biochemical Sciences 33:9–19.https://doi.org/10.1016/j.tibs.2007.09.014
-
Bacterial chemoreceptors: providing enhanced features to two-component signalingCurrent Opinion in Microbiology 13:124–132.https://doi.org/10.1016/j.mib.2009.12.014
-
VMD: visual molecular dynamicsJournal of Molecular Graphics 14:33–38.https://doi.org/10.1016/0263-7855(96)00018-5
-
Computer visualization of three-dimensional image data using IMODJournal of Structural Biology 116:71–76.https://doi.org/10.1006/jsbi.1996.0013
-
Solution structure of the bacterial chemotaxis adaptor protein CheW from escherichia coliBiochemical and Biophysical Research Communications 360:863–867.https://doi.org/10.1016/j.bbrc.2007.06.146
-
Core unit of chemotaxis signaling complexesProceedings of the National Academy of Sciences of the United States of America 108:9390–9395.https://doi.org/10.1073/pnas.1104824108
-
Selective allosteric coupling in core chemotaxis signaling complexesProceedings of the National Academy of Sciences of the United States of America 111:15940–15945.https://doi.org/10.1073/pnas.1415184111
-
Role of CheW protein in coupling membrane receptors to the intracellular signaling system of bacterial chemotaxisProceedings of the National Academy of Sciences of the United States of America 86:8703–8707.https://doi.org/10.1073/pnas.86.22.8703
-
Molecular architecture of chemoreceptor arrays revealed by cryoelectron tomography of escherichia coli minicellsProceedings of the National Academy of Sciences of the United States of America 109:E1481.https://doi.org/10.1073/pnas.1200781109
-
All-atom empirical potential for molecular modeling and dynamics studies of proteins †The Journal of Physical Chemistry B 102:3586–3616.https://doi.org/10.1021/jp973084f
-
MDAnalysis: a toolkit for the analysis of molecular dynamics simulationsJournal of Computational Chemistry 32:2319–2327.https://doi.org/10.1002/jcc.21787
-
Fastcluster: fast hierarchical, agglomerative clustering routines for r and pythonJournal of Statistical Software 53:1–18.
-
Reconstruction of the chemotaxis receptor–kinase assemblyNature Structural & Molecular Biology 13:400–407.https://doi.org/10.1038/nsmb1085
-
Signaling and sensory adaptation in escherichia coli chemoreceptors: 2015 updateTrends in Microbiology 23:257–266.https://doi.org/10.1016/j.tim.2015.03.003
-
Molecular dynamics simulations of large macromolecular complexesCurrent Opinion in Structural Biology 31:64–74.https://doi.org/10.1016/j.sbi.2015.03.007
-
Scalable molecular dynamics with NAMDJournal of Computational Chemistry 26:1781–1802.https://doi.org/10.1002/jcc.20289
-
Comparative protein modelling by satisfaction of spatial restraintsJournal of Molecular Biology 234:779–815.https://doi.org/10.1006/jmbi.1993.1626
-
Template-directed assembly of receptor signaling complexes †Biochemistry 42:13379–13385.https://doi.org/10.1021/bi0352769
-
Spatial organization in bacterial chemotaxisThe EMBO Journal 29:2724–2733.https://doi.org/10.1038/emboj.2010.178
-
Diversity in chemotaxis mechanisms among the bacteria and archaeaMicrobiology and Molecular Biology Reviews 68:301–319.https://doi.org/10.1128/MMBR.68.2.301-319.2004
-
Structural studies of cytoskeletal protein arrays formed on lipid monolayersJournal of Structural Biology 128:75–81.https://doi.org/10.1006/jsbi.1999.4167
-
On the freezing and identification of lipid monolayer 2-d arrays for cryoelectron microscopyJournal of Structural Biology 160:305–312.https://doi.org/10.1016/j.jsb.2007.04.011
-
Quantitative modeling of bacterial chemotaxis: signal amplification and accurate adaptationAnnual Review of Biophysics 42:337–359.https://doi.org/10.1146/annurev-biophys-083012-130358
-
The receptor–CheW binding interface in bacterial chemotaxisJournal of Molecular Biology 415:759–767.https://doi.org/10.1016/j.jmb.2011.11.043
-
Making sense of it all: bacterial chemotaxisNature Reviews Molecular Cell Biology 5:1024–1037.https://doi.org/10.1038/nrm1524
-
CheA–receptor interaction sites in bacterial chemotaxisJournal of Molecular Biology 422:282–290.https://doi.org/10.1016/j.jmb.2012.05.023
-
3D reconstruction and processing of volumetric data in cryo-electron tomographyJournal of Structural Biology 157:126–137.https://doi.org/10.1016/j.jsb.2006.07.014
-
Using situs for the integration of multi-resolution structuresBiophysical Reviews 2:21–27.https://doi.org/10.1007/s12551-009-0026-3
-
Direct visualization of escherichia coli chemotaxis receptor arrays using cryo-electron microscopyProceedings of the National Academy of Sciences of the United States of America 104:3777–3781.https://doi.org/10.1073/pnas.0610106104
Decision letter
-
Axel T BrungerReviewing Editor; Stanford University, United States
eLife posts the editorial decision letter and author response on a selection of the published articles (subject to the approval of the authors). An edited version of the letter sent to the authors after peer review is shown, indicating the substantive concerns or comments; minor concerns are not usually shown. Reviewers have the opportunity to discuss the decision before the letter is sent (see review process). Similarly, the author response typically shows only responses to the major concerns raised by the reviewers.
Thank you for submitting your work entitled "CryoEM and Computer Simulations Reveal a Novel Kinase Conformational Switch in Bacterial Chemotaxis Signaling" for peer review at eLife. Your submission has been evaluated by Michael Marletta (Senior editor), a Reviewing editor, and three reviewers. There are several major issues that need to be addressed before a final decision can be made.
The reviewers have discussed the reviews with one another and the Reviewing editor has drafted this decision to help you prepare a revised submission.
Summary:
The manuscript describes the use of subtomogram averaging to determine the structure of a reconstituted chemotactic signaling array at intermediate-resolution.
A complex has been assembled on lipid monolayers between (1) the cytoplasmic domain of Tar chemotaxis receptor, (2) CheA, and (3) CheW. A cryo-tomography density map of the resulting crystalline patches has been obtained, and sub-tomogram averaging has been used to an estimated resolution of the core-signalling unit around 1.3 nm. Flexible-fitting molecular dynamics has been used to dock known crystal structures of these components into the density map. This model and the molecular dynamics simulations suggest a new conformational switch.
Overall, this study defines the reconstituted arrays at an unprecedented resolution and thereby confirms key features of the assembly that have been proposed in the past, but not previously visualized – such as the CheW rings and the P3 and P4 domain positioning. These are important contributions to the understanding of the array structure. However, there are major concerns about the anisotropic character of the EM data and the possible impact on the accuracy of the model. Moreover, the reliability and uniqueness of the fit of the model to the EM data must be assessed in order to judge the quality of the model.
The all-atom MD simulations, which are impressive in their size and duration, suggest a provocative dynamical model of kinase regulation, although the accompanying biochemical data are not highly supportive of the "dipping" mechanism, and further control experiments should be performed. Moreover, it is not entirely clear how these conformational changes are to be interpreted in terms of CheA activation and whether there are other relevant changes.
Reviewer #1:
Essential revisions:
1) The elegant combination of an in vitro reconstitution together with sub-tomogram averaging and molecular modeling is very impressive, but this reviewer has two substantial technical concerns: the data has been obtained by performing tomography on a monolayer sample. All of the unit-cells are equivalently oriented relative to the electron beam and during tomography are tilted through a limited angular range. As is also the case in 2D electron-crystallography experiments, this results in a missing cone of information in the final structure. The resolution of the structure is therefore lower in Z than in X, Y. This situation cannot be avoided for a monolayer sample. This anisotropic resolution has multiple implications: it will lead to apparent smearing of the structure in Z; it will influence the ability to place the subunits into the map, and especially to determine their Z-position; it will affect the MDFF. The authors only quote a single resolution suggesting this effect may not be considered. What is the resolution in Z? Does this allow reliable positioning of the structural models in Z? How reliable is this positioning? How was the anisotropic resolution taken into account in MDFF?
2) Knowing how the available structures have been placed into the EM map, and with what confidence the positions are determined, is essential for assessing the reliability of the final model. It is not clear to me how reliable these positions are. For example, the authors state that the four-helix bundle of the CheA-P3 bundle is clearly discerned to run parallel to the receptor. How clearly? Judging from the available supplementary movie, there is some room for varying this position, as there is for the position of the P5 domain, in particular in the Z direction. What is currently missing is an assessment of how reliable the subunit models are and how accurately the EM map determines their position and arrangement. Are any alternative arrangements possible? With what confidence can this arrangement be selected? How accurately can the domain positions and orientations be determined using an EM map of this resolution and these structural subunit models? How accurate do these initial positions need to be to interpret details after MDFF? One can imagine that with a 30 Å resolution structure, subsequent MDFF would not give a structure reliable to side-chain resolution, while with a 5 Å structure it would – what about at the resolution in this manuscript? In the end the authors go on to determine specific amino acid interactions and conformational changes, so it is important to clearly justify/demonstrate that this level of interpretation is reliable and unique given the input data.
3) Example data should be deposited in the EMDB.
Reviewer #2:
Minor revisions:
1) Various CheA domains are first referred to in the subsection “3D density maps of CheA2-timer and CheA2-hexamer”. It would be helpful if there had been a sentence or two first to introduce the structure of CheA, identifying the domains P1 to P5.
2) The authors rightly point out that the model agrees with previous structural studies, and they go on to validate, in cells, the conformational change that they propose for CheA. This should all be balanced by some discussion of what might be added by further improving the resolution. While docking known structures into EM maps at a resolution of ~1.2 nm is generally accepted as giving useful biological information, it generally is not accepted as giving a definitive final structure.
Reviewer #3:
Essential revisions:
1) The authors should provide more detail with respect to the uniqueness of the P4 positioning in the density maps. This is an important issue because the EM data here alone determines the P4 position and it does not draw on crystal structures, such as most of the other interfaces. The conformation of P4 is also critical to the dynamics simulations that follow. The domain fitting was done reasonably, however; on viewing the density, one wonders if other positions are also nearly as compatible. For example, do the fits converge if different starting conformations are used? Could some metric be given with respect to the goodness of fit? Also on this point, it appears that the nucleotide-free form on the P4 domain was taken, both for the modeling and the dynamics. ATP binding substantially affects the conformation of a large loop on the P4 domain surface. Does ATP-bound P4 give a similar fit as the nucleotide-free form?
2) There is cryoEM data that suggests the P1 and P2 associate quite strongly with the kinase core in the inhibited state (Briegel et al. 2013). If these regions indeed contact P4, they could alter its positioning and dynamics. This point should be noted.
3) This reviewer is concerned that the significance of the R297 salt bridge to stabilizing the dipped conformation is overstated. In the simulations, is 4 versus 2 instances of dipping significant for 27 copies? Also, although 3 of the 4 dwell times for the dipped conformation are indeed longer in the wt, the other one is similar to that seen in the mutant. Is the principle component associated with this motion the only one that shows a difference in occurrence between the wt and mutant trajectories? A more thorough assessment of this data would bring more confidence to the dipping result.
4) The use of "asymmetry" might be misleading to differentiate the conformational states. The term certainly describes the dipped state, however; the asymmetry may simply be a consequence of this being a rare event. The reader gets the impression that there are two states: one symmetric and one asymmetric; however, a dipped symmetric state may also be possible, just unlikely to be seen in the same dimer under the simulation conditions.
5) Little information is given with respect to the PCA in the dynamics trajectories (and there is no mention of it in the Methods section). Also, for a general biological scientist, the significance of Figure 5C will be lost without a better description of what is being represented. It would be useful to know what percent of the eigenvalues are represented by the first component and how many vectors represent a majority of the total motion. Also, the authors focus on this one dipping transition; are there no other significant conformational changes? The arrays are of great interest in part due to their cooperative behavior; one might expect such simulations to reveal recoupling across core particles. There is also data that changes in the P5-receptor interfaces are associated with activation (Piasta et al. 2013). Are any such motions observed? This should be commented on.
6) The cross-linking data supports proximity of the CheA receptor residues, but it does not provide much support for the dipping motion. Control sites would have to be investigated and relative rates of cross-linking determined, ideally for multiple positions. For example, the R394 self cross-link seems to form as readily as CheA-receptor cross-links. R394 self cross-links are not surprising given the symmetry in the trimers-of-dimers, nonetheless, these sites are ~15 Å apart. Furthermore, it is not clear why the N405 self cross-links also do not form, given that these residues are closer to each other than are copies of R394. The statement near the end of the second paragraph of the subsection “Biochemical validation of CheA conformational change in E. coli cells” is not accurate, given that the yields of the 394 self cross-links are quite different in the two experiments. Normalizing to this band, there is not much change between the 394/361 and 394/316 pairs. Even if the cross-linking is less for 316/394, it could mean that those resides are simply further apart in a dominant conformation or show reactivity differences (e.g. 405 vs. 394). There is also the question of what state the cells are in during cross-linking. The attractant serine is present during the experiment, but depending on the timing of the washes, the cells are likely adapted. How much of CheA is activated and how much is inhibited in these experiments? For only a few sites, as investigated here, it would be more powerful to detect an attractant depend effect on relative cross-linking of the reporters. As it stands, the cross-linking data primarily supports the general architecture, but does not make the dipping mechanism more compelling.
7) The cell swimming assays need more description and from Figure 6 alone, it is not clear what parameter is being graphed. "Swimming ability" itself is a misnomer, as the cells can likely swim fine, but can't alternate tumbling properly to migrate in soft agar. This should be explained in the text or figure legend. Do all the mutants form attractant rings at the migration fronts indicative of chemotaxis? Or do some just show spreading, a consequence of some CheA activation, but no regulation (pseudotaxis). Photographs of the swarm plates for key mutants (or those that show differences should be provided as figure supplements). Furthermore, simply measuring the swarm radius does not distinguish a mechanistic defect from an assembly defect. Can an assessment be made (perhaps using the reconstitution system) if the mutants form arrays like wt?
8) In the subsection “3D density maps of CheA2-timer and CheA2-hexamer“, second paragraph, the electron density is less in the CheW/P5 contacts between neighboring core signaling particles than within the core complex, but this does not necessarily mean that the interfaces are weaker; they may instead be more structurally variable, and, hence, averaged out. Consider a change of wording. It is also difficult to see in Figure 3A and 3C that there is less density in the P5/W contacts between core particles. In Figure 3, color P5 and W differently, and demark the interfaces 1 and 2 that are being referred to in the text.
9) This reviewer appreciates that the authors are being cautious with the interpretation of their data, but the reader is largely left to summarize what the findings mean for CheA activation. The mechanistic take home message should be more explicit.
[Editors' note: further revisions were requested prior to acceptance, as described below.]
Thank you for submitting your work entitled "CryoEM and Computer Simulations Reveal a Novel Kinase Conformational Switch in Bacterial Chemotaxis Signaling" for consideration by eLife. Your article has been reviewed by three peer reviewers, and the evaluation has been overseen by a Reviewing Editor and Michael Marletta as the Senior Editor.
The reviewers have discussed the reviews with one another and the Reviewing editor has drafted this decision to help you prepare a revised submission. The following comments need to be addressed before a final decision can be made.
Reviewer #1:
1) Anisotropic resolution of the map.
The authors present the angular distribution of the contributing subunits and argue that because is a good coverage of the tomography tilt range this concern can be disregarded. This is not a valid argument for two reasons. Firstly because the angular coverage is not good – high tilt angles are clearly underrepresented relative to low tilt angles, and this will cause a resolution anisotropy (it is not completeness of coverage, but uniformity of coverage that determines resolution anisotropy). Secondly, at high tilt the monolayer sample is much thicker and the projections have lower signal to noise. High and low tilts do not contribute equivalently to the final structure. This will also cause resolution anisotropy. I reiterate my initial concern – that the resolution of the structure is expected to be lower in Z than in X and Y. In the response to reviewers the authors show some isosurfaces with simulated wedges. It is not possible to sensibly assess the extent of anisotropic resolution from such images, and anyway, fitting is not based on isosurfaces, but on density. I am not persuaded that resolution anisotropy can be largely disregarded as the authors suggest.
This is not an unusual problem and it should not be difficult for the authors to address this issue better. They should be clear in the manuscript that there is resolution anisotropy and make an attempt to measure its extent (previous publications have used for example FSC within cones or 3DSSR to assess anisotropic resolution). The potential effect of this on fitting and MDFF should then be considered. Ideally, this would be taken into account during simulation, but the authors will probably consider this too much work. In that case they should write that this was not taken into account, and discuss what the potential influence might be on the reliability of domain positioning or on the simulations.
2) Reliability and uniqueness of fit.
This question was also asked by reviewer 3. How well does the EM density determine the position of the domains, in particular P4? From the authors response to reviewers it sounds like the position is not well defined, but that different positions still undergo dipping motions. I am still concerned that the orientation of P4 is not well defined by the EM map, that the starting orientation is critical to the MDFF analysis, and that this influences the interpretation of the dipping motion. The authors have access to the raw data and do not seem to be concerned by this. In that case they need to help the reader to assess the reliability of the model. Ideally by providing a metric for the goodness of fit, but minimally by an honest appraisal of the caveats, clearly and openly discussing the reliability of the fit, where alternative positions are possible, and how this might influence the interpretation. This should be in the main text of the manuscript.
Reviewer #2:
The revisions made in response to my comments are fully satisfactory.
Reviewer #3:
In their revised version of the manuscript, the authors have done a nice job of improving the paper and they have largely addressed my concerns. The one exception is the cross-linking studies. I still do not agree with the statement that "Taken together, our cross-linking experiments suggest that the CheA-P4 "dipped" conformation observed in silico is indeed sampled within the native chemosensory complex of E. coli." The inference here is that the cross-linking data supports the dipped conformation as a relevant state of the arrays. It may well be, but, the cross-linking data does not provide strong support for this supposition. Again, the only band to report on the dipped conformation (361/394) is quite weak, much weaker than the self 394-394 band. It's not uncommon to see some cross-linking between residues that are in proximity. In fact, I would be surprised if no cross-linking was observed, even if only the undipped conformation was represented. The authors can lean more on their mutagenesis data to support the dipped conformation, but I don't think the cross-linking makes the case.
https://doi.org/10.7554/eLife.08419.043Author response
[…] Overall, this study defines the reconstituted arrays at an unprecedented resolution and thereby confirms key features of the assembly that have been proposed in the past, but not previously visualized – such as the CheW rings and the P3 and P4 domain positioning. These are important contributions to the understanding of the array structure. However, there are major concerns about the anisotropic character of the EM data and the possible impact on the accuracy of the model.
Because our monolayer samples are not flat like 2D crystals, but are rather wavy, the angular distribution of contributing subunits is very well sampled (Figure 3—figure supplement 1B and C, see Reviewer #1, point 1), thus anisotropic resolution of the density map is not a significant concern. Nonetheless, we tested the effect of missing cone on the density map. As you can see in Author response image 1, there is little density distortion, with a cone semi-angle of ~12°, a conservative estimate for the sampling in this study (Author response image 1, green mesh), thus its impact on the accuracy of the model docking is minimal.

Effect of missing cone on the tomography density map.
Shown in gray solid surface, the ideal density map without missing cone, and maps with different amount of missing cone applied (colored meshes). Left, green mesh 12° cone (this study, Figure 3–figure supplement 1). Right, yellow mesh, 30° cone as would be found in other typical tomography studies. Density maps are all contoured at 1.5σ.
Moreover, the reliability and uniqueness of the fit of the model to the EM data must be assessed in order to judge the quality of the model.
We would like to clarify that the unit cell model used in the bulk of our MD simulation was built heuristically, utilizing existing X-ray structures to best preserve the known interaction interfaces and only indirectly using the density maps to provide loose constraints on portions of the model in which previous data did not exist, in particular the position of the CheA P4 domains. This seems to have been a major source of confusion and concern for the reviewers, and we apologize for not making it clearer. We have revised the Methods section to include detailed steps on how the MD models were generated with a new Figure 4—figure supplement 2. Nevertheless, we carried out additional equilibrium simulations of the MDFF-refined CheA2-trimer model, which exhibit qualitatively the same structural and dynamical features, including the presence of several CheA P4 dips, as shown in the new Figure 4—figure supplement 5 (See Reviewer #1, Point 2 and Reviewer #3, Point 1). Moreover, to test the uniqueness of our model fit, we carried out several additional fitting experiments for comparison, obtaining virtually identical models (See Reviewer #3, Point 1 and Author response image 2).

CheA dimers from multiple initial conformations converge to same fitted model.
Red, blue, and green traces track RMSD of CheA dimers as measured from the initial MDFF-refined model. Though, the dimer conformations diversity during the 120 ns production run, a 20 ns MDFF simulations returns them to their initial MDFF-refined state.
The all-atom MD simulations, which are impressive in their size and duration, suggest a provocative dynamical model of kinase regulation, although the accompanying biochemical data are not highly supportive of the "dipping" mechanism, and further control experiments should be performed.
We designed three cross-linking pairs for the predicted interactions based on the MD model: the interaction only in the dipping state and those only in the undipped state. The cross-linking results agree very well with model prediction, supporting our computational model as well as confirming the existence of a novel dipping state (See Reviewer #3, Point 6). We have included additional control experiments as presented in Figure 5.
We also believe this first plausible model with sufficient details of the basic chemotaxis core-signaling unit will be valuable for researchers in the field to further test and investigate the function and mechanisms of the bacterial chemotaxis signaling.
Moreover, it is not entirely clear how these conformational changes are to be interpreted in terms of CheA activation and whether there are other relevant changes.
We have added discussions on the P4 conformation in terms of CheA activation (Discussion). The dipping motion observed in MD simulation warrants further in depth investigation, such as how these conformational changes are related to CheA activation. This is beyond the scope of this study.
Detailed responses to all points raised by the individual reviewers are provided below.
Reviewer #1:
Essential revisions:1) The elegant combination of an in vitro reconstitution together with sub-tomogram averaging and molecular modeling is very impressive, but this reviewer has two substantial technical concerns: the data has been obtained by performing tomography on a monolayer sample. All of the unit-cells are equivalently oriented relative to the electron beam and during tomography are tilted through a limited angular range. As is also the case in 2D electron-crystallography experiments, this results in a missing cone of information in the final structure. The resolution of the structure is therefore lower in Z than in X, Y. This situation cannot be avoided for a monolayer sample. This anisotropic resolution has multiple implications: it will lead to apparent smearing of the structure in Z; it will influence the ability to place the subunits into the map, and especially to determine their Z-position; it will affect the MDFF. The authors only quote a single resolution suggesting this effect may not be considered. What is the resolution in Z? Does this allow reliable positioning of the structural models in Z? How reliable is this positioning? How was the anisotropic resolution taken into account in MDFF?
We thank the reviewer for the positive remark, and agree with the reviewer that a perfectly oriented sample, such as 2D crystal, has a substantial anisotropic resolution (poorer in Z) with a missing cone of information in the final structure. However, our monolayer samples are not rigid like 2D crystals, but are quite wavy on the cryoEM grids. The angles between the normal of the monolayer and the electron beam can be as much as 15°. On top of this, there is additional specimen pre-tilt ranging 0-10°, in addition to a tomography tilt range ± 70°. The overall angular distribution of all the subunits contributing to the averaged map is presented in a new Figure 3—figure supplement 1B and C for the in-plane rotation and maximum tilt angle, respectively. The plot shows a near uniform distribution of in-plane angles, and a very good coverage of the tomography tilt range, therefore, concerns with missing cone and anisotropic resolution, as well as its impact on the reliability of positioning of MDFF model can be largely disregarded (see overall critique point #1, Author response image 1).
2) Knowing how the available structures have been placed into the EM map, and with what confidence the positions are determined, is essential for assessing the reliability of the final model. It is not clear to me how reliable these positions are. For example, the authors state that the four-helix bundle of the CheA-P3 bundle is clearly discerned to run parallel to the receptor. How clearly? Judging from the available supplementary movie, there is some room for varying this position, as there is for the position of the P5 domain, in particular in the Z direction. What is currently missing is an assessment of how reliable the subunit models are and how accurately the EM map determines their position and arrangement. Are any alternative arrangements possible? With what confidence can this arrangement be selected? How accurately can the domain positions and orientations be determined using an EM map of this resolution and these structural subunit models? How accurate do these initial positions need to be to interpret details after MDFF? One can imagine that with a 30 Å resolution structure, subsequent MDFF would not give a structure reliable to side-chain resolution, while with a 5 Å structure it would – what about at the resolution in this manuscript? In the end the authors go on to determine specific amino acid interactions and conformational changes, so it is important to clearly justify/demonstrate that this level of interpretation is reliable and unique given the input data.
We understand the reviewer’s concerns and have taken a number of steps to better characterize to what degree our models are unique and robust. These steps are discussed in more detail below (also see Reviewer #3, Point 1) and the modeling procedures used in this study, including “how” the available structures were placed in the EM maps, are summarized in a new supplemental figure (Figure 4—figure supplement 3). Here, we will address some of the reviewer’s more general concerns.
Regarding how reliably the EM map determines the positions and arrangement of the protein components involved in the model, we can only strictly state that our MDFF model is reliable up to a resolution equal to that of the lowest resolution piece used in its construction, that is our 12.7 Å density map. Nevertheless, this resolution places tight constraints on the overall positioning and tertiary structure of the proteins involved. In addition, we would like to highlight that the proteins were not docked into the EM map in one-by-one but rather heuristically-constructed, full subunit pieces were docked to best preserve known interfaces (e.g., receptors within receptor TOD from 1QU7, CheA-P5/CheW interfaces from 4JPB, receptor/CheW and receptor/CheA-P5 from NMR, etc.). Assuming the correctness of these previously-characterized interfaces, no other global arrangements are possible. To illustrate this we conducted additional MDFF simulations on the MDFF-refined CheA2-trimer model increasing the coupling of the model to the map by an order of magnitude – a regime in which the density-derived forces overwhelm the electrostatic forces between atoms.
Concerning the interpretation of structural details after MDFF, the reviewer is indeed correct that at the resolutions of the maps present in this study, no side-chain information is present and, hence, MDFF cannot directly bias the structures at the side-chain level. However, we would like to highlight that the strength of the MDFF methodology is the “flexibility” of its fitting. This flexibility arises from the use of the chemically-based MD force field, which adds much additional information to help refine the model’s side chain orientations and interactions during the fitting procedure. Unfortunately, there currently are no robust methods for assessing the “accuracy” (that is, how closely the model matches the “true” structure) of a model derived via the fitting of high-resolution structures into lower-resolution EM maps, especially those with resolution lower than 8 Å. Indeed, this is an area of active research in the field (Xu, X. P., & Volkmann, N. (2015) Arch. Biochem. Biophys.; Schröder, G. F. (2015) C.O.S.B.). Rather, validation of the reliability of a model may still be given in terms of its “precision”, that is the qualitative reproducibility of the model’s key predictions. Importantly, we subsequently carried out extensive, all-atom MD simulations on multiple models and using many independent simulations and demonstrated that the key discovery of this study, namely the dipping conformational dynamics of the CheA kinase domain, is reproduced in multiple simulations of both heuristically- and MDFF-derived models. We then successfully confirmed a number of our model’s predictions in the biochemical experiments, the best test of reliability.
Finally, the authors would also like to note that while the CheA P3 and P5 domains may appear in the supplementary movie to have more room for varying their positions, this appearance is a consequence of the chosen isosurface used to illustrate a rough envelope of the electron density. This envelope can appear tighter or more diffuse depending on the isosurface value chosen. The MDFF biasing force, on the other hand, is based on the actual three-dimensional volumetric density, not the envelope.
3) Example data should be deposited in the EMDB.
The EM maps were deposited in EMDB under accession code EMD-6319 for the CheA2-trimer and EMD-6320 for the CheA2-hexamer, at the time when the manuscript was submitted. The EM map-associated MD model of the T. maritima core-signaling unit was also deposited in PDB database under accession code 3JA6.
Reviewer #2:
Minor revisions:
1) Various CheA domains are first referred to in the subsection “3D density maps of CheA2-timer and CheA2-hexamer”. It would be helpful if there had been a sentence or two first to introduce the structure of CheA, identifying the domains P1 to P5.
We thank the reviewer’s suggestion and have included a sentence to introduce the structure of CheA, identifying the domains P1 to P5 (“CheA is a multi-domain protein, consisting of five separate and functionally distinct domains (P1-P5): P1-phosphoryl transfer domain, P2-substrate binding domain, P3-dimerization domain, P4-kinase domain and P5-regulatory domain”).
2) The authors rightly point out that the model agrees with previous structural studies, and they go on to validate, in cells, the conformational change that they propose for CheA. This should all be balanced by some discussion of what might be added by further improving the resolution. While docking known structures into EM maps at a resolution of ~1.2 nm is generally accepted as giving useful biological information, it generally is not accepted as giving a definitive final structure.
We appreciate the reviewer’s comments and have added a discussion regarding what one might expect from modeling efforts utilizing higher resolution EM maps (“By further improving the resolution of our cryoET to better than 8Å and by using the novel lipid-monolayer system described here, an atomic homology model for E. coli can be derived, which would greatly facilitate the use of the wealth of existing biochemical and biophysical data and permit directly transferrable structural and dynamical predictions”).
Reviewer #3:
Essential revisions:1) The authors should provide more detail with respect to the uniqueness of the P4 positioning in the density maps. This is an important issue because the EM data here alone determines the P4 position and it does not draw on crystal structures, such as most of the other interfaces. The conformation of P4 is also critical to the dynamics simulations that follow. The domain fitting was done reasonably, however; on viewing the density, one wonders if other positions are also nearly as compatible. For example, do the fits converge if different starting conformations are used? Could some metric be given with respect to the goodness of fit? Also on this point, it appears that the nucleotide-free form on the P4 domain was taken, both for the modeling and the dynamics. ATP binding substantially affects the conformation of a large loop on the P4 domain surface. Does ATP-bound P4 give a similar fit as the nucleotide-free form?
We thank the reviewer for the well-thought-out questions regarding the overall quality and uniqueness of our models. We have added additional text and figures, as described below, to address the reviewer’s concerns.
To address the uniqueness of the positioning of the CheA P4 domain and its effect on the observed P4 dynamics, we have carried out an additional series of extensive simulations of the MDFF-refined CheA2-trimer model (Figure 4—figure supplement 3B and Supplementary file 1) for comparison with the dynamics observed in the heuristically-derived unit cell model. As the MDFF refinement procedure affected, in particular, the positioning of the P4 domains relative to the rest of the CheA dimer molecules and receptor tips, and hence their interactions with neighboring proteins, the two models provided different CheA dimer initial conditions from which to seed further simulations. In particular, the MDFF fitting procedure gave rise to a CheA dimer model with an RMSD from the heuristically modeled CheA dimer of ~5.3 Å. Nevertheless, both models exhibited multiple dipping events (Figure 4 and Figure 4—figure supplement 5) demonstrating that the ability of the P4 domain to undergo the dipping motion does not depend on a more precise placement of the P4 domains than our density maps have provided.
In addition, to address the convergence of the MDFF fit, we conducted two additional MDFF simulation on CheA2-trimer model in which the CheA molecules exhibited a mix of P4 conformations. For this purpose we selected as initial conformations two of the ten CheA2-trimer replicas (subsequent to the ~120 ns production simulation) in which one or more of the CheA molecules exhibited dipped or “intermediate” P4 conformations. Author response image 2 depicts, for one of those MDFF simulations, the RMSD of each CheA dimer in the CheA2-trimer (as measured from the initial MDFF-refined CheA dimer model) over the course of the ~120 ns production simulations and 20 ns post-production simulation. Hence, we show that the MDFF fitting of CheA dimers starting from diverse initial configurations all converge on to essentially the same model.
Finally, the reviewer is correct that the nucleotide-free form of CheA was selected for this initial study. In general, maps of resolution less than 1 nm cannot be used to discern differences in secondary structure (Schröder, G. F. (2015). COSB, 31, 20-27). Regarding the structural effects of nucleotide binding on the local structure of the P4 domain because both conformations inhabit essentially the same volume, we cannot, at the current resolution of our maps (>=12.7Å), discriminate with MDFF the difference in conformation of the loop in the nucleotide-free P4 (e.g. PDB 1B3Q) and nucleotide-bound P4 (e.g., PDB 1I58 or 1I5D).
2) There is cryoEM data that suggests the P1 and P2 associate quite strongly with the kinase core in the inhibited state (Briegel et al. 2013). If these regions indeed contact P4, they could alter its positioning and dynamics. This point should be noted.
This was indeed carefully considered when we docked the models. We now have a preliminary density map lacking the P1-P2 domains in the CheA. The map with CheA P345 is very similar to the map with CheA full length (Author response image 3), further suggesting the density corresponds to the P4 domain. The P1-P2 domains are likely very flexible and the densities have been averaged out. The main differences between this study and the previous one (Briegel et al. 2013) are 1) the map resolution is better in this study and number of subunits contributing to the final structure is in a great excess; 2) our density map is from wt receptor, likely in the adapted state, while the other study is in the inhibited state.

Comparison of density maps containing full-length wt CheA and wt CheA-P345.
Densities are contoured at 2σ. Solid blue, with wt CheA-P345; red mesh, with full-length wt CheA. The density with CheA-P345 appears elongated along the Z direction, due to the missing wedge effect and limited amount of data. (A) Side view of the volume slab. (B and C) Top views of volume slabs at the height indicated in A.
3) This reviewer is concerned that the significance of the R297 salt bridge to stabilizing the dipped conformation is overstated. In the simulations, is 4 versus 2 instances of dipping significant for 27 copies? Also, although 3 of the 4 dwell times for the dipped conformation are indeed longer in the wt, the other one is similar to that seen in the mutant. Is the principle component associated with this motion the only one that shows a difference in occurrence between the wt and mutant trajectories? A more thorough assessment of this data would bring more confidence to the dipping result.
We thank the reviewer for this careful observation. After performing additional experiments described below, we agree that while the R297/E397 interaction clearly stabilizes the “dipped” state in silico as signified by an increased dwell time and is also critical for proper signaling in vivo, this does not necessarily mean that it is the stabilization of the “dipped” state that makes the interaction critical for proper signaling. For example, one could imagine the interaction being related to a Class I model given the spatial relationship with this pair and R354/E392 examined in a recent NMR study. On the other hand, it could also be that the rearrangement of P4 is important to bring into register binding sites for P1 in support of a class II model, or some combination of the two. We have altered the main text to clarify that the R297/E397 contact is necessary for signaling and correlated to the “dipped” state, but softened the correlation between the two ideas.
To address this concern experimentally, we extended both our wild-type and R297A mutant unit cell simulation by 50%, from 300 ns each to 450 ns each, in order to improve our sampling. We did not observe any additional stabilized dipping events in the wild-type simulations (Figure 4C, top). Moreover, in the R297A mutant simulations, one of the dipping CheA P4s underwent an additional dip-return event (Figure 4C, bottom), further closing the gap between the sheer number of dipping events observed in the two models. Hence, in light of the new data, we agree with the reviewer that our proposal that the R297/E397 contact may help promote the dipping conformation change is not well supported. We have removed this notion from the main text. On the other hand, our suggestion that the R297/E397 contact may help stabilize the dipping conformation seems to be reinforced by our extended simulations. In particular, in the wild-type simulations, the three (of the four total) dipping CheA dimers in which the R297/E397 contact was formed (Figure 4C, top, red, blue, green traces) maintain this conformation, while the dimer that did not form the R297/E397 contact (yellow trace) returned to the bulk. In addition, both dipping CheA dimers in the R297A mutant simulations, which are of course unable to from the R297/E397 salt bridge, quickly return to the bulk (Figure 4C, bottom, red, blue traces), including the P4 domain, which undergoes two dip-return events (blue trace). Hence, we feel the suggestion of the stabilizing nature of the R297/E397 contact is reasonable in light of the comparative in silicodata. It remains to be determined if the stabilization of the “dipped” state is the primary role of R297 in signaling, or if this is an ancillary effect.
Finally, regarding the differences between the PC projections of the wt and R297A mutant CheA dimer populations, the authors wish to point out that the PCA was constructed to highlight the dip for the purpose of tracking the CheA dimer conformations. Hence, the difference between the CheA dimer dynamics of the two models is most readily observed in the first principal component, which captures a large percent (41%) of this dipping motion. Though the second PC also describes a large portion of the total variance (31%), the two models do not differ significantly from each other in this dimension.
4) The use of "asymmetry" might be misleading to differentiate the conformational states. The term certainly describes the dipped state, however; the asymmetry may simply be a consequence of this being a rare event. The reader gets the impression that there are two states: one symmetric and one asymmetric; however, a dipped symmetric state may also be possible, just unlikely to be seen in the same dimer under the simulation conditions.
We agree with the reviewer and have changed the term "asymmetry" to "dipped".
5) Little information is given with respect to the PCA in the dynamics trajectories (and there is no mention of it in the Methods section). Also, for a general biological scientist, the significance of Figure 5C will be lost without a better description of what is being represented. It would be useful to know what percent of the eigenvalues are represented by the first component and how many vectors represent a majority of the total motion. Also, the authors focus on this one dipping transition; are there no other significant conformational changes? The arrays are of great interest in part due to their cooperative behavior; one might expect such simulations to reveal recoupling across core particles. There is also data that changes in the P5-receptor interfaces are associated with activation (Piasta et al. 2013). Are any such motions observed? This should be commented on.
We appreciate the reviewer’s comments and have added to the Methods section an entry entitled “Simulation Analysis” to describe the PCA procedure in more detail. In that section the fractional variances of the first three modes are given, in particular, the first component is shown to describe roughly 42% of the total motion. In addition, a more detailed main-text discussion of the PCA results was added to better describe their significance (subsection “A conformational change of the CheA kinase domain”, last paragraph). Regarding the observation of “other” significant conformation changes, we agree with the reviewer that our model and methods should be able to provide useful information regarding general motions of the array, including those potentially relevant to its cooperative nature. Indeed, we have sought to develop the first plausible all-atom model for asking these kinds of structural questions and have deposited this model in the PDB database for review and use by other researchers in the field. In this study, however, we have chosen to focus our investigation on a large-scale dipping motion of the CheA-P4 domain. Additional investigation will be necessary to test the full extent of our model’s predictability in other areas of signaling and function, in particular the P5-receptor interface mentioned by the reviewer is of active interest to us.
6) The cross-linking data supports proximity of the CheA receptor residues, but it does not provide much support for the dipping motion. Control sites would have to be investigated and relative rates of cross-linking determined, ideally for multiple positions. For example, the R394 self cross-link seems to form as readily as CheA-receptor cross-links. R394 self cross-links are not surprising given the symmetry in the trimers-of-dimers, nonetheless, these sites are ~15 Å apart. Furthermore, it is not clear why the N405 self cross-links also do not form, given that these residues are closer to each other than are copies of R394. The statement near the end of the second paragraph of the subsection “Biochemical validation of CheA conformational change in E. coli cells” is not accurate, given that the yields of the 394 self cross-links are quite different in the two experiments. Normalizing to this band, there is not much change between the 394/361 and 394/316 pairs. Even if the cross-linking is less for 316/394, it could mean that those resides are simply further apart in a dominant conformation or show reactivity differences (e.g. 405 vs. 394). There is also the question of what state the cells are in during cross-linking. The attractant serine is present during the experiment, but depending on the timing of the washes, the cells are likely adapted. How much of CheA is activated and how much is inhibited in these experiments? For only a few sites, as investigated here, it would be more powerful to detect an attractant depend effect on relative cross-linking of the reporters. As it stands, the cross-linking data primarily supports the general architecture, but does not make the dipping mechanism more compelling.
Although cross-linking experiments cannot catch the dynamic motion of dipping, the data support the appearance of a new interaction interface upon P4 dipping (by close proximity of the CheA – receptor residues), that otherwise would not be formed. Although limited in number, we tested the sites for three scenarios: exists in both states as positive control (I304/N405); only in the undipped state (D316/R394); and only in dipped state (E361/R394), along with necessary negative controls. Regarding the different cross-linking yields (R394) in two different experiments, we have repeated all the cross-linking in a single experimental setting, and updated the Figure 5B (previously Figure 6B) with new data. Additional controls were also added in Figure 5B.
We agree with the reviewer that low cross-linking intensity could mean that the residues are farther apart in a dominant conformation or show reactivity differences, in addition to low frequency of the conformation occurs. We have changed the discussion accordingly to reflect this (subsection “Biochemical validation of CheA conformational change in E. coli cells”, second paragraph).
Regarding the question about the state of the cells during cross-linking, the experiments were carried out in the presence of serine, and the cells are most likely adapted. What we measure is the ensemble of the bacterial population likely containing both activated and inhibited CheA. We don’t know how much CheA is in each state. We would like to point out that one of the major goals of this study is to put outa first plausible model with sufficient details of the basic chemotaxis core-signaling unit for researchers in the field. The cross-linking data support our computational model and the probable novel dipping state. It would be beyond the scope of this study to further detect signaling events with relative cross-linking of the reporters, and define the mechanisms of the dipping conformation. We will certainly follow up the leads and continue our investigation in this respect. Nevertheless, following the reviewer’s suggestions, we carried out the experiments to test the effect of serine on the existing cross-linking pairs under two different conditions: 1) cells were grown in the presence of serine, followed by cross-linking with or without serine washout; 2) cells were grown in the absence or presence of serine, followed by cross-linking. In both cases, the changes in the extent of cross-linking are not substantial or conclusive. A more thorough future investigation is needed.
7) The cell swimming assays need more description and from Figure 6 alone, it is not clear what parameter is being graphed. "Swimming ability" itself is a misnomer, as the cells can likely swim fine, but can't alternate tumbling properly to migrate in soft agar. This should be explained in the text or figure legend. Do all the mutants form attractant rings at the migration fronts indicative of chemotaxis? Or do some just show spreading, a consequence of some CheA activation, but no regulation (pseudotaxis). Photographs of the swarm plates for key mutants (or those that show differences should be provided as figure supplements). Furthermore, simply measuring the swarm radius does not distinguish a mechanistic defect from an assembly defect. Can an assessment be made (perhaps using the reconstitution system) if the mutants form arrays like wt?
We appreciate the reviewer’s comments. We have added detailed description of cell swimming assays in the Methods section (subsection “Soft agar Assays”). We also included the photographs of the representative soft agar plates for key mutants with controls in the revised Figure 5. Furthermore, we carried cryoEM analysis of the E. coli cells carrying these mutants and show that these cells are functional in forming chemotaxis clusters. These results are included in the new Figure 5—figure supplement 1.
8) In the subsection “3D density maps of CheA2-timer and CheA2-hexamer“, second paragraph, the electron density is less in the CheW/P5 contacts between neighboring core signaling particles than within the core complex, but this does not necessarily mean that the interfaces are weaker; they may instead be more structurally variable, and, hence, averaged out. Consider a change of wording. It is also difficult to see in Figure 3A and 3C that there is less density in the P5/W contacts between core particles. In Figure 3, color P5 and W differently, and demark the interfaces 1 and 2 that are being referred to in the text.
We agree with the reviewer’s point and changed the wording in the revised manuscript accordingly. We have clearly marked the interfaces 1 and 2 with arrows accordingly in Figure 3A and 3C, and used a darker surface color to display these interfaces.
9) This reviewer appreciates that the authors are being cautious with the interpretation of their data, but the reader is largely left to summarize what the findings mean for CheA activation. The mechanistic take home message should be more explicit.
We thank reviewer’s suggestion. We have added additional discussion on the implication of our findings (Discussion).
[Editors' note: further revisions were requested prior to acceptance, as described below.]
Reviewer #1:
1) Anisotropic resolution of the map.The authors present the angular distribution of the contributing subunits and argue that because is a good coverage of the tomography tilt range this concern can be disregarded. This is not a valid argument for two reasons. Firstly because the angular coverage is not good – high tilt angles are clearly underrepresented relative to low tilt angles, and this will cause a resolution anisotropy (it is not completeness of coverage, but uniformity of coverage that determines resolution anisotropy). Secondly, at high tilt the monolayer sample is much thicker and the projections have lower signal to noise. High and low tilts do not contribute equivalently to the final structure. This will also cause resolution anisotropy. I reiterate my initial concern – that the resolution of the structure is expected to be lower in Z than in X and Y. In the response to reviewers the authors show some isosurfaces with simulated wedges. It is not possible to sensibly assess the extent of anisotropic resolution from such images, and anyway, fitting is not based on isosurfaces, but on density. I am not persuaded that resolution anisotropy can be largely disregarded as the authors suggest.This is not an unusual problem and it should not be difficult for the authors to address this issue better. They should be clear in the manuscript that there is resolution anisotropy and make an attempt to measure its extent (previous publications have used for example FSC within cones or 3DSSR to assess anisotropic resolution). The potential effect of this on fitting and MDFF should then be considered. Ideally, this would be taken into account during simulation, but the authors will probably consider this too much work. In that case they should write that this was not taken into account, and discuss what the potential influence might be on the reliability of domain positioning or on the simulations.
We agree with the reviewer that the resolution of the structure is poorer in Z than in X and Y. To assess the degree of anisotropy in resolution, we calculated conical Fourier shell correlations as suggested by the reviewer (see Author response image 4). The overall resolution (FSC=0.143) is 11.3 Å (calculated from the same map with an updated mask to include just the CheA2-trimers, while the previous FSC was calculated using an extended mask), and the resolution in Z direction is 15.8 Å. We have revised the manuscript to include this in Figure 3—figure supplement 1A. We further repeated the MDFF exercise, taking anisotropic resolution into account during simulation by using the anisotropic FSC low-pass filtered map. The resulting MDFF model ((Author response image 4B, orange) is similar to the previous one (Author response image 4B, yellow), with an overall RMSD of 2.78Å. The map and MDFF model deposited at PDB and EMDB are being updated.

Resolution anisotropy.
(A) Gold-standard Fourier shell correlation (FSC) of the CheA2-trimer density map. The overall FSC of the map is plotted as a solid black line. The FSC curves for the conical Fourier shells along the X, Y, and Z directions are in solid green, blue and dark-red, respectively, and along the 10 other directions are in dotted lines. The previous FSC (dashed black line) was calculated from the same map but with a more extended mask. (B) MDFF model of the CheA-P4 domain from the anisotropic FSC filtered density map (orange), overlapped with the previous MDFF model (yellow). The pink and light blue spheres indicate the P3-C and P5-N termini, respectively. CheA-P3 and CheA-P5 domains are in gray.
2) Reliability and uniqueness of fit.This question was also asked by reviewer 3. How well does the EM density determine the position of the domains, in particular P4? From the authors response to reviewers it sounds like the position is not well defined, but that different positions still undergo dipping motions. I am still concerned that the orientation of P4 is not well defined by the EM map, that the starting orientation is critical to the MDFF analysis, and that this influences the interpretation of the dipping motion. The authors have access to the raw data and do not seem to be concerned by this. In that case they need to help the reader to assess the reliability of the model. Ideally by providing a metric for the goodness of fit, but minimally by an honest appraisal of the caveats, clearly and openly discussing the reliability of the fit, where alternative positions are possible, and how this might influence the interpretation. This should be in the main text of the manuscript.
We appreciate the reviewer’s comment and carried out additional docking exercises. We fit the CheA-P4 domain via rigid-body transformations starting from 10,000 random angular orientations and up to 20Å shifts from the center of the mass, to generate a metric for the goodness of fit. This fitting resulted in 23 distinct classes (separated by 3° and 3Å), as shown in Author response image 5A. The class of best fit (Author response image 5B, panel 1, with highest cross correlation and highest number of contributing fit) is similar to the MDFF model (Author response image 5B, Panel 10 with overlay). In addition to the best fit, one other possible fit is #5 (Author response image 5B, panel 5), given the proximity of the N and C termini of P4 to the P3 and P5 domain. This fit is less probable compared to the best fit, considering that it has a lower cross-correlation value, a much lower occupancy with ~ 1/3 the number of contributing fits, and the positions of P4-N and C termini are reversed (flipped), making it hard to connect the P3 and P5 termini with short linkers. In the revised manuscript, we have included the fitting metric and a discussion on the possible alternative P4 positions that might influence the interpretation (subsection “All-atom model of the T. maritima chemosensory array”, Figure 3—figure supplement 3).

A metric for the goodness of fit for the docking of the CheA-P4 domain.
(A) Distribution of 23 classes of fits for the P4 domain starting from random orientations. (B) The models from the top 9 highest cross-correlation classes are shown in panels 1-9, with the cross-correlation values and number of contributing fits (%) indicated below. These 9 classes constitute 90.1% of total fits. Panel 10 is an overlay of the #1 fit (blue) with the MDFF model (orange). The red and blue spheres indicate the N and C termini of P4, respectively. The corresponding connecting termini from P3 and P5 are in light blue and pink, respectively. P3 and P5 domains are in gray.
Reviewer #3:
In their revised version of the manuscript, the authors have done a nice job of improving the paper and they have largely addressed my concerns. The one exception is the cross-linking studies. I still do not agree with the statement that "Taken together, our cross-linking experiments suggest that the CheA-P4 "dipped" conformation observed in silico is indeed sampled within the native chemosensory complex of E. coli." The inference here is that the cross-linking data supports the dipped conformation as a relevant state of the arrays. It may well be, but, the cross-linking data does not provide strong support for this supposition. Again, the only band to report on the dipped conformation (361/394) is quite weak, much weaker than the self 394-394 band. It's not uncommon to see some cross-linking between residues that are in proximity. In fact, I would be surprised if no cross-linking was observed, even if only the undipped conformation was represented. The authors can lean more on their mutagenesis data to support the dipped conformation, but I don't think the cross-linking makes the case.
We appreciate the reviewer’s comments. We have revised the manuscript to tone down the inference from the cross-linking experiments and changed the statement to “suggesting that the CheA-P4 "dipped" conformation observed in silico may have been sampled within the native chemosensory complex of E. coli” (subsection “Biochemical validation of CheA conformational change in E. coli cells”, second paragraph).
https://doi.org/10.7554/eLife.08419.044Article and author information
Author details
Funding
National Institute of General Medical Sciences (R01GM085043)
- Peijun Zhang
National Science Foundation (PHY-1430124)
- Klaus Schulten
National Institute of General Medical Sciences (P50GM082251-7518)
- Peijun Zhang
National Institute of General Medical Sciences (9P41GM104601)
- Klaus Schulten
National Institute of General Medical Sciences (5R01GM098243)
- Klaus Schulten
National Science Foundation (OCI-0725070)
- Klaus Schulten
National Science Foundation (ACI-1238993)
- Klaus Schulten
The funders had no role in study design, data collection and interpretation, or the decision to submit the work for publication.
Acknowledgements
We thank Drs. JS Parkinson for bacterial strains and plasmids and S Subramaniam for antisera against chemotaxis components, D Bevan and P Greer for computer technical support and A Makhov for cryoEM instrumental support. In addition, we thank Drs. Wei Han and Davi Ortega for insightful discussions regarding the computational aspects of this study. We also thank Dr. T Brosenitsch for reading the manuscript. This work was supported by the National Institutes of Health NIGMS Grant R01GM085043 (PZ), P50GM082251-7518 (PZ), 9P41GM104601 (KS), and 5R01GM098243 (KS) as well as the National Science Foundation PHY-1430124 (KS). Large-scale molecular dynamics simulations were performed on the Blue Waters supercomputer, financed by the National Science Foundation (awards OCI-0725070 and ACI-1238993).
Reviewing Editor
- Axel T Brunger, Stanford University, United States
Version history
- Received: April 29, 2015
- Accepted: November 18, 2015
- Accepted Manuscript published: November 19, 2015 (version 1)
- Version of Record published: December 31, 2015 (version 2)
Copyright
© 2015, Cassidy et al.
This article is distributed under the terms of the Creative Commons Attribution License, which permits unrestricted use and redistribution provided that the original author and source are credited.
Metrics
-
- 4,715
- Page views
-
- 971
- Downloads
-
- 78
- Citations
Article citation count generated by polling the highest count across the following sources: Scopus, Crossref, PubMed Central.
Download links
Downloads (link to download the article as PDF)
Open citations (links to open the citations from this article in various online reference manager services)
Cite this article (links to download the citations from this article in formats compatible with various reference manager tools)
Further reading
-
- Microbiology and Infectious Disease
- Structural Biology and Molecular Biophysics
The transcriptional regulator SsrB acts as a switch between virulent and biofilm lifestyles of non-typhoidal Salmonella enterica serovar Typhimurium. During infection, phosphorylated SsrB activates genes on Salmonella Pathogenicity Island-2 (SPI-2) essential for survival and replication within the macrophage. Low pH inside the vacuole is a key inducer of expression and SsrB activation. Previous studies demonstrated an increase in SsrB protein levels and DNA-binding affinity at low pH; the molecular basis was unknown (Liew et al., 2019). This study elucidates its underlying mechanism and in vivo significance. Employing single-molecule and transcriptional assays, we report that the SsrB DNA binding domain alone (SsrBc) is insufficient to induce acid pH-sensitivity. Instead, His12, a conserved residue in the receiver domain, confers pH sensitivity to SsrB allosterically. Acid-dependent DNA binding was highly cooperative, suggesting a new configuration of SsrB oligomers at SPI-2-dependent promoters. His12 also plays a role in SsrB phosphorylation; substituting His12 reduced phosphorylation at neutral pH and abolished pH-dependent differences. Failure to flip the switch in SsrB renders Salmonella avirulent and represents a potential means of controlling virulence.
-
- Neuroscience
- Structural Biology and Molecular Biophysics
Ca2+/calmodulin-dependent protein kinase II (CaMKII) is essential for long-term potentiation (LTP) of excitatory synapses that is linked to learning and memory. In this study, we focused on understanding how interactions between CaMKIIα and the actin-crosslinking protein α-actinin-2 underlie long-lasting changes in dendritic spine architecture. We found that association of the two proteins was unexpectedly elevated within 2 minutes of NMDA receptor stimulation that triggers structural LTP in primary hippocampal neurons. Furthermore, disruption of interactions between the two proteins prevented the accumulation of enlarged mushroom-type dendritic spines following NMDA receptor activation. α-Actinin-2 binds to the regulatory segment of CaMKII. Calorimetry experiments, and a crystal structure of α-actinin-2 EF hands 3 and 4 in complex with the CaMKII regulatory segment, indicate that the regulatory segment of autoinhibited CaMKII is not fully accessible to α-actinin-2. Pull-down experiments show that occupation of the CaMKII substrate-binding groove by GluN2B markedly increases α-actinin-2 access to the CaMKII regulatory segment. Furthermore, in situ labelling experiments are consistent with the notion that recruitment of CaMKII to NMDA receptors contributes to elevated interactions between the kinase and α-actinin-2 during structural LTP. Overall, our study provides new mechanistic insight into the molecular basis of structural LTP and reveals an added layer of sophistication to the function of CaMKII.