Bacteria are capable of sensing and responding to changes in their environment. One of the ways they do this is via chemotaxis, regulating swimming behaviour. The chemotaxis pathway senses chemoattractant gradients and uses a feedback loop to change the bacterial swimming pattern; this feedback loop differs in detail between species. In the present article, we summarize the current understanding of the regulatory mechanisms in three species and how these pathways can be viewed and analysed through the ideas of feedback control systems engineering.
- methyl-accepting chemotaxis protein (MCP)
- systems biology
Most bacteria sense changes in their environment and move in response to those changes. Movement is usually the result of flagella that propel the bacteria through the medium. Bacteria can change their swimming direction by either altering the direction of flagella rotation (e.g. Escherichia coli) or by stopping their flagella (e.g. Rhodobacter sphaeroides) with the cell body re-orientating by Brownian motion . The frequency of direction changing is controlled by regulating the frequency of motor switching/stopping, resulting in a biased random walk towards favourable or away from adverse environments. The frequency of flagella switching/stopping is controlled by the chemotaxis pathway. The overall output does not respond to the absolute levels of compounds directly, but instead a signalling pathway with a feedback loop is used to sense changes in the gradient of chemotactic compounds . This feedback mechanism provides bacteria with the ability to sense changes across a wide range of attractant concentrations and thus to work in a range of environmental conditions .
Chemotaxis in E. coli
The paradigm for chemotaxis pathways is that of E. coli (Figure 1) . The pathway itself is relatively isolated from other pathways, that is chemoreceptors sense the external world and only regulate flagellar rotation making it straightforward to manipulate input and measure output behaviours. The reaction parameters and protein concentrations have been measured over a range of conditions and the chemotaxis pathway in E. coli has therefore been extensively modelled , and these models have revealed interesting key properties of the pathway. For example, these models have lead to predictions regarding the localization of proteins within the pathway  which have been verified experimentally. It has also helped rationalize the high gain and sensitivity in the system [7–9].
In the E. coli pathway, changes in the external chemoeffector concentrations are sensed by MCPs (methyl-accepting chemotaxis proteins), which are transmembrane receptors arranged in large quaternary complexes of trimers of MCP dimers at the cell poles. Ligands may bind directly to the receptors or via a periplasmic binding protein. In turn, this binding is signalled across the membrane to a receptor-bound histidine protein kinase, CheA. The receptor controls the CheA autophosphorylation activity, with increasing ligand binding resulting in a decrease in CheA activation. Once phosphorylated, the phosphoryl group is transferred to one of two possible response regulators, CheY and CheB, associated with the P2 domain of CheA [10,11].
Phosphorylated CheY is able to diffuse away from the signalling cluster and interact with the flagella motor causing direction switching of the motor. CheY signalling is terminated by the action of CheZ, which dephosphorylates CheY. This is required to allow the system to respond within a suitable timescale. Because of cell size and the constant buffeting of the environment, bacteria respond to temporal rather than spatial changes, and must respond to a change within approx.1 s to allow a gradient to be sensed .
Phosphorylated CheB, on the other hand, forms a feedback loop back to the receptors, allowing the system to sense gradients over time. This works by altering the methylation state of the receptors, ‘resetting’ them. CheB, when phosphorylated, is a methylesterase, removing methyl groups from specific glutamate residues on the MCPs, changing the charged state and the receptor packing. This works antagonistically to CheR, which is a constitutively active methyltransferase. The ability of the MCP to stimulate CheA is dependent on the methylation state acting antagonistically with ligand. The methylation of the receptors is also dependent on the activation state of the receptor, with active receptors more readily demethylated by CheB and inactive receptors more readily methylated by CheR [13,14]. This simple adaptation system is observed in many other species .
Perfect adaptation and robustness of the E. coli pathway
The methylation feedback loop enables the bacteria to sense a temporal gradient rather than a spatial one and also to be sensitive over a wide range of background concentrations. The E. coli system has also been shown to have intrinsic adaptation due to the feedback system layout. This adaptation was predicted to be robust to large changes in parameter values , something that was also shown experimentally . Thus the pathway will be effective and adapt regardless of variability in gene expression and cell size.
Chemotaxis and adaptation in other species
As more bacterial genomes have been sequenced it has become apparent that E. coli uses quite a simple chemotaxis system. Many other species have multiple homologues of the E. coli system or proteins that do not have a role identifiable in the E. coli pathway . A good example of this is the chemotaxis pathway in Sinorhizobium melioti, which has multiple CheY homologues . In this species, although both CheYs are phosphorylated by a single CheA, only one can interact with the motor to bring about switching, the other CheY acting as a phosphate sink and allowing signal termination, in the absence of CheZ .
The system in Bacillus subtilis differs considerably from that of E. coli; for example, the phosphotransfer has the opposite effect, with the CheA kinase activity increasing as ligand increases . Thus, in this system, phosphorylated CheY promotes smooth swimming of the bacterium. B. subtilis also lacks CheZ (a protein only found in the gamma proteobacteria), with signal termination requiring the action of FliY. FliY is found localized at the flagella motor and shares homology with CheX-like phosphatase proteins found in other species. There are also a number of extra proteins not found in E. coli but involved in forming adaptation feedback loops: CheV, CheC and CheD (Figure 2). These form, along with methylation, three adaptation systems that are not functionally equivalent; at least two of these systems are required for normal chemotaxis . CheV acts as an alternative response regulator and can modulate CheA activity, via a CheW-like domain. CheV is present in a number of other species including Helicobacter pylori . CheC is a CheY phosphatase and CheD a MCP deamidase. CheD interacts with either the receptor or CheC, but only when CheC is bound to CheY-P. The level of CheD receptor interaction controls the activation of the receptor hence forming an adaptive-feedback loop (Figure 2). In addition, unlike in E. coli, where methylation of a glutamate residue results in receptor activation, the effect of methylation in B. subtilis depends on which glutamate residue is methylated, with some having an activating effect on CheA, whereas others have an inhibiting effect . The exact reason for having three adaptation pathways is currently unclear, but it has been suggested that they respond to different levels of concentration change, or gradient steepness, with the methylation system responding to large changes and the CheC/CheD and CheV systems responding to smaller changes in attractant concentration .
Despite the differences and diversity between species they still exhibit the same control strategy .
Chemotaxis and adaptation in R. sphaeroides
Genome analysis of sequenced bacterial species suggests increased complexity in over 50% of species, with multiple chemosensory systems in addition to a larger repertoire of proteins than identified in E. coli . For example, the phototrophic bacterium R. sphaeroides has multiple homologues of the E. coli system organized into three operons. Each operon encodes a complete chemosensory pathway, and two of these are essential for chemotaxis. Fluorescent fusions to proteins encoded in the different operons showed that components of the pathways localized at two discrete parts of the cell, with different receptors and kinases localized to the different clusters . A great deal of in vitro and in vivo work has been completed and yet the exact wiring of the pathway and its control of the single motor remains unclear [24,25], as does the sensory and adaptation role of the different clusters (Figure 3). There is no CheV, CheD or CheC in R. sphaeroides.
The adaptation pathway differs from E. coli as the two different CheB proteins encoded in the two essential R. sphaeroides pathways are found diffuse in the cytoplasm rather than associated with the receptor complex . Measurement of phosphotransfer showed that the CheA associated with the polar cluster can phosphorylate both CheB proteins, whereas the CheA associated with the cytoplasmic cluster can phosphorylate only one of the CheB proteins, the one encoded in the same operon . Modelling the phosphotransfer kinetics of the two chemosensory pathways, protein copy numbers and cellular behaviour suggested reverse phosphotransfer involving diffusible CheB between the cytoplasmic and polar CheAs. It is, however, currently unknown which receptors in which cluster each CheB protein is capable of demethylating. Indeed, methanol release suggests that the system may act more like the B. subtilis system with methanol release (a measure of demethylation) occurring on both addition and removal of an attractant . The ability of the output of one cluster to potentially change the methylation state at the other may result in a more complex methylation system, which is akin to feedback control systems in engineering and would show a different control strategy to the single feedback loops in B. subtilis and E. coli. This would make sense in light of the phosphotransfer network, which is more complex than in E. coli . How this system is connected, and the system properties this would confer, are the subject of current research.
The chemotaxis system provides an example of a simple feedback loop about which there is still much to learn. Why do different species use different feedback adaptation systems? What is the purpose of the two interconnected but independent pathways in R. sphaeroides? How are these two pathways connected to produce a balanced chemosensory response?
Using engineering tools to understand chemotaxis
Control engineering can help us design experiments to probe these questions. Signalling pathways in biology can be viewed as large electronic circuits, with parts of them acting as amplifiers, logic gates etc . Many tools have been developed in engineering to understand circuits from their input–output behaviour [28,29] and we can also use these on biological systems. In order to use some of these tools, the system has to be modelled mathematically and then the outputs of the models compared with experimental output from carefully designed experiments, resulting in either model invalidation or model refinement. .
On the experimental side, tools require the ability to alter the input to the system and measure the output in real time . Experimentally this can be difficult with biological systems, but the chemotaxis system is tractable in this manner via the tethered cell assay . In this assay, cells are attached to a microscope glass cover slip via an anti-flagella antibody: instead of the flagella rotating, the cell body rotates and this rotation can be tracked using computerized motion analysis. This method can be incorporated into a flow cell allowing control over the input and measurement of the output in real time.
Using the data from these experiments, multiple models can be created, each representing one of the possible system connectivities. Nominal model parameters can be obtained using previous knowledge or by fitting them so that each model is able to represent wild-type experimental data. This is particularly important if there are model parameters that are difficult to measure experimentally. Where more than one parameter value is capable of fitting the wild-type data, multiple models could be created for each parameter set, and invalidated using the same methods as for connectivities.
Engineering tools can then be applied to determine whether there is an optimal input to the system that can differentiate between the behaviours of the different models, so that if this input were undertaken in the laboratory it would result in discrimination. Relevant tools include the use of the frequency response (bode plot) of the ‘error’ system, to determine whether there is an optimal frequency which will result in sufficiently different outputs between the models under test . Other methods include looking for a phase shift or using a controller could potentially be used [32,33].
Where the optimal input does not result in successful model invalidation, then changes to the initial conditions can be probed. The advantage of using this approach is that multiple different possibilities can be tested in silico before performing additional experiments. In this way, changes to the initial conditions which would result in the biggest difference between the model outputs can be selected. The ability to choose initial conditions can also be determined by what perturbations are available experimentally in the system under study. In bacterial chemotaxis systems, these include deletion, overexpression of a protein or combinations of these. Robustness and sensitivity analysis can then be performed to ensure that the experiment will provide discrimination, even though some of the parameters may be uncertain .
Once the optimal experiment is selected using this approach, the experiment can be implemented in vivo and the resulting data used for model invalidation . This approach can then be performed as part of an iterative cycle, invalidating models until only one remains. The remaining model represents a signalling pathway and parameter values that can explain all available wild-type and deletion experimental data and hence can give insight into how the pathway is connected. This method also reduces the number of complex experimental procedures needed to get closer to a realistic understanding of sensory signalling.
In the present article, we have highlighted the similarities and differences in bacterial chemotaxis pathways; despite having different compositions they show similar circuit layout. We have also shown that, using a control engineering-inspired method of model generation, parameterization and experimental design, followed by model invalidation, it is possible to learn more about pathway connectivity in complex signalling networks .
This work was funded by the Engineering and Physical Sciences Research Council (EPSRC) [grant number E05708X].
Signalling and Control from a Systems Perspective: A Biochemical Society Focused Meeting held at University of York, U.K., 22–24 March 2010, as part of the Systems Biochemistry Linked Focused Meetings. Organized and Edited by David Fell (Oxford Brookes, U.K.), Hans Westerhoff (Manchester, U.K., and Amsterdam, The Netherlands) and Michael White (Liverpool, U.K.).
Abbreviations: MCP, methyl-accepting chemotaxis protein
- © The Authors Journal compilation © 2010 Biochemical Society