Main

Computers outperform people at most tasks, but we still excel at a few. The ability to recognize how objects fit together belongs in this category. The inspiration of Shimizu et al.1 was to use this human gift to their advantage to produce an intellectually and aesthetically satisfying, as well as testable, model of a chemotaxis signalling complex. This type of scale modeling is used routinely in various fields of engineering, and it is exciting to see its reemergence in structural biology half a century after Pauling constructed his α-helix and Watson and Crick erected their towering spiral of double-stranded DNA.

Using atomic coordinates from the structures of the core region of the CheA dimer2 and of CheW (provided by F. W. Dahlquist), both from Thermotoga maritima, Shimizu and colleagues programmed a three-dimensional printer to manufacture plastic models on a scale of 15 mm per nm, with a resolution of 0.3 mm (or 0.02 nm). They constructed a similar model for the cytoplasmic tip of the trimer of dimers observed in the crystal structure of the intracellular domain of the Escherichia coli serine receptor, Tsr3. They then manipulated by hand the models of CheW and CheA, and separately of CheW and Tsr, to achieve the best complementary surface fit. They also generated comparable images in modelling applications, and subjected these to energy minimization and other optimization routines to generate the most probable docked structures. In the case of the interction of CheW with Tsr, residues identified in an analysis of allele-specific suppression4 were located at the predicted contact interface.

One could make the criticism that the proposed molecular lattice incorporates too much imagination and not enough data. Another objection might be that the plastic models are rigid, whereas interacting proteins undergo conformational shifts in order to yield the changes in free energy that drive the assembly process. Conceding these limitations, the deduced model is still an extremely valuable contribution. Like all good heuristic models, it ties together existing information, suggests new paradigms, and is open to experimental refutation or verification. Unlike most scientific models, it also exists in the form of brightly coloured plastic pieces that can be assembled into a structure suitable for public display.

The signal-transduction pathway in bacterial chemotaxis is well characterized5. CheA is an autophosphorylating histidine protein kinase that communicates with receptors through the CheW coupling factor. CheA can transfer its phosphoryl group to the response regulator, CheY. Flagellar motors rotate anticlockwise during smooth swimming and clockwise during tumbling; binding of phosphorylated CheY to the motor promotes clockwise rotation.

The intrinsic activity of purified CheA is stimulated 100-fold in a ternary complex of membrane-bound receptors, CheA, and CheW. The response of such a reconstituted system to an attractant ligand establishes the basic paradigm for chemotactic signalling. When an attractant binds to its receptor, CheA activity is inhibited, and phosphotransfer to CheY is correspondingly reduced. As phosphorylated CheY in the cell is rapidly depleted by CheZ phosphatase, the result is a rapid reduction in intracellular levels of phosphorylated CheY and an increased probability of anticlockwise flagellar rotation. Thus, smooth swims are extended, and cells migrate up an attractant gradient.

In principle, this situation does not require any interactions beyond those of a receptor dimer with a CheA dimer and two CheW monomers in a 1:1:1 complex. It is adequate to explain responses to large shifts in the concentration of single attractants. Indeed, it has been calculated that the best strategy for an E. coli cell with a limited number of receptors (1,500–4,500) to detect chemicals in its environment would be to distribute these receptors uniformly (or randomly) over the cell surface6.

So compelling was this argument that E. coli was selected as a negative control by Maddock and Shapiro7 in their search for receptor clusters in the asymmetrically dividing species Caulobacter crescentus , in which only one daughter cell is flagellated. The result was startling. Receptors in C. crescentus indeed cluster at the flagellar pole of the predivisional swarmer cell, but in E. coli, which does not sport a polar flagellum, the receptors are also present in polar patches. Furthermore, polar localization of the receptors in E. coli diminishes when either CheA or CheW is absent.

These observations beg the question of why E. coli chemoreceptors are distributed in a patchy fashion, which is seemingly contrary to sound engineering principles. The answer presumably lies in the nature of the signalling mechanism, as clustering of receptors could account for several unexplained features of chemotaxis. First, a change of less than 1% in receptor occupancy causes a measurable increase in anticlockwise rotational bias8. How can inhibition of the activity of only a few CheA molecules associated with attractant-bound receptors be amplified to give a detectable signal? Second, it is unclear how the low-abundance receptors Tap and Trg mediate strong responses to their attractant ligands when they stimulate CheA activity only weakly9. Finally, the means by which responses to different attractants or repellents in a chemically heterogeneous environment are integrated at the levels of signalling and adaptation is unknown10.

These issues have been dealt with previously by Bray et al. in a conceptual model that invokes interconnected arrays within receptor patches11. Although there is no experimental basis for such extended networks, they are consistent with the existence of receptor patches. However, the identification of reconstituted aggregates of the soluble cytoplasmic domain of a chemoreceptor with CheW and CheA in a stoichiometry of ∼7:1:1 may provide a glimpse of greater structural complexity12.

The model-building exercise reported by Shimizu and colleagues would have been useful even if it just predicted how the individual protein partners interact, but its implications are far greater. The geometry of the proposed receptor–CheW–CheA trigonal complex indicates a straightforward way that it can be extended to form a hexagonal array of indefinite expanse, a clear candidate for the receptor patch. Within such an array, conformational perturbations initiated by the binding of ligand to one receptor dimer could spread in order to amplify or integrate signals from different receptors. The mysterious, but crucially important, linker between the second membrane-spanning segment and the extended cytoplasmic domain of the receptors13 may allow the bending in an otherwise rigid helix that would be required to form the trimer of receptor dimers.

The model also predicts that an `adaptation compartment' may exist between the cell membrane and the hexagonal lattice. The ability of CheR methyltransferase14 and CheB methylesterase15 to bind to the carboxyl-terminal tail of high-abundance receptors, together with their sequestration in this chamber, would restrict their diffusion away from the site at which their activity is needed. Recent studies using proteins fused to green fluorescent protein (GFP) have indicated that CheY and CheZ concentrate at the polar receptor patches as well16. Conveniently, the model predicts that the surface of CheA that interacts with CheY (and probably with CheZ) faces outwards into the bulk cytoplasm, as expected.

The speculative diagram in Fig. 1 represents our attempt to integrate the concept of an extended receptor lattice with other information to present our view of the basic features of chemotactic signalling. This scheme is presented to stimulate thinking and inspire experiments, not as a final explanation. It may also provide a new perspective on how receptor/signalling complexes in eukaryotic cells may be organized.

Figure 1: Schematic view of chemotactic signalling in E. coli.
figure 1

Left panel, the cell is in a balanced signalling state that produces alternating smooth swims (anticlockwise flagellar rotation) and tumbles (clockwise flagellar rotation). Cytoplasmic phosphorylated CheY (CheY–P) remains in the narrow range of concentration at which the motor reverses17 because supply from the receptor patch is balanced by oligomerization and activation of cytoplasmic CheZ phosphatase18 when CheY–P reaches a threshold level. The concentrations of CheY–P that cause motor reversal and CheZ activation are within the same range (∼3 μM). Right panel, an attractant-bound receptor (heavy black arrow) initiates an inhibitory signal that spreads through the patch, shutting off several CheA kinases. CheZ sequestered in the patch, possibly through binding to the short form of CheA19, also becomes activated in response to the inhibitory signal. Less CheY–P is produced, and CheY–P formed elsewhere in the patch is dephosphorylated by CheZ before it diffuses away into the cytoplasm. CheY–P levels in the cytoplasm fall below the level needed to support clockwise flagellar rotation, and smooth swims are extended. As adaptive methylation restores the receptors to a CheA-activating state, more CheY–P is produced, patch-associated CheZ is deactivated, and CheY–P is again supplied in excess, returning the cell to a balanced signalling state. Lateral spread of the inhibitory signal, activation of CheZ, and the high cooperativity (apparent Hill coefficient of 11; ref. 17) of CheY–P in promoting clockwise rotation may all contribute to signal amplification and integration.

Full size image