The elegant mutational analysis of HAMP domain function presented by Zhou et al. (2009) in this issue is destined to become an instant classic. Its monumental scope inspires the question, ‘Don’t these people have anything else to do? Don't these people ever sleep?' It certainly stands as a sterling example of how genetic and structural approaches can be wedded to elucidate protein function.
HAMP domains (Butler and Falke, 1998; Aravind and Ponting, 1999; Williams and Stewart, 1999) are structural and functional features of a large family of transmembrane signalling proteins in all manner of prokaryotes. Their ubiquity is attested by the catchy acronym for the rather unwieldy composition histidine kinases, adenylyl cyclases, methyl-accepting chemotaxis proteins and certain phosphatases. Typically, HAMP domains connect extracellular (periplasmic) and membrane-spanning modules to cytoplasmic signalling modules. Their role in sensory transduction is thus a central theme in molecular microbiology.
Proteins containing HAMP domains generally, if not universally, exist as homodimers. The solution NMR structure of the HAMP domain from protein Af1503 of the thermophilic archaeon Archaeoglobus fulgidus (Hulko et al., 2006) is a dimeric four-helix bundle. The two amphipathic helices contributed by each subunit (AS1 and AS2) are aligned in parallel and are joined by a 14-residue flexible connector (Fig. 1A). The helices pack in an x-da configuration, also known as knob-on-knob packing, in which bulky hydrophobic residues face one another at the centre of the bundle. In bacterial chemoreceptors (Fig. 1B), the extended four-helix bundle of the kinase-control module (Kim et al., 1999; Alexander and Zhulin, 2007) has a more typical anti-parallel arrangement of helices packed in an a-d, or knob-in-hole, configuration, in which bulky hydrophobic residues fit into holes formed by small residues in the companion helices.
Hulko et al. (2006) suggested that HAMP domains can exist in two states, the x-da configuration seen in the structure and an a-d packing configuration. A rotation of relatively rigid helices through 26°C would accomplish this transition, as shown in the bottom panel of Fig. 1A. This proposition has been called the ‘gearbox’ model. A cysteine-scanning study of the Tar chemoreceptor (Swain and Falke, 2007) showed patterns of disulphide cross-linking that are consistent with the existence of a four-helix bundle of the type found in the NMR structure.
In the gearbox model, one configuration would presumably impose an ‘on’ state of the transmembrane protein, the other an ‘off’ state. In the case of an Escherichia coli chemoreceptor like the Tsr protein studied by Zhou et al. (2009), these would correspond to the CheA kinase-stimulating state and the CheA kinase-inhibiting state (Fig. 2). These two activities are conveniently monitored by looking at the time the flagella spend in clockwise (CW) rotation. The probability of CW rotation is proportional to the level of phospho-CheY in the cell (Cluzel et al., 2000), and CheY is the response regulator that is a target for phosphorylation by CheA. One can either look at rotation directly, using tethered cells (Silverman and Simon, 1974) or bead-decorated flagella (Chen and Berg, 2000), or indirectly, by monitoring chemotaxis, which requires both CW and counterclockwise flagellar rotation, in semisolid agar (Wolfe and Berg, 1989).
The gearbox has two major problems. First, there is no published structural evidence to indicate that a bona fide HAMP domain takes on a-d packing. Second, the relevant attractant-induced signal in chemoreceptors appears to be a several ångstrom piston-like movement of the second transmembrane helix (TM2) towards the cytoplasm. It is unclear how this movement would be converted into rotation of the helices in the HAMP bundle.
Zhou et al. (2009) undertook a mutational analysis of the Tsr HAMP domain that could test genetic predictions of the gearbox model. The technique chosen could be characterized as carpet bombing; a conscientious attempt was made to replace each of the 19 residues of AS1 and the region immediately upstream, and the 17 residues of AS2 and the region immediately downstream, with each of the other 19 amino acids. The mutational target encompassed all of HAMP except the connector, which was the subject of a previous, and equally thorough, study by the Parkinson laboratory (Ames et al., 2008). The end result was an average of 13 different residues at each of 36 positions. Each of the mutated genes was sequenced in its entirety to eliminate the confusion of secondary mutations, and a receptorless strain harbouring plasmids containing each of the variants was tested for serine chemotaxis in semisolid tryptone agar.
The staggering amount of data gleaned during this epic adventure is shown in one figure. The ∼370 mutations obtained were classified on the basis of the extent of the serine chemotaxis ring formed in receptorless cells harbouring the plasmid-borne mutated gene expressed at physiological levels. The categories were essentially wild type, impaired function or null. Of the 131 null mutations, 51 targeted AS1 and 80 targeted AS2. Six residues in each helix were identified as critical, because many or most of the substitutions at those positions yielded null phenotypes. All but one of the critical residues in each helix was predicted to be part of the x-da packing face expected if the Tsr sequence is threaded onto the Af1503 HAMP structure. The two outliers were literally so: one was located in the N-terminal end of AS1 that is not part of the four-helix bundle, and the other is in the C-terminal end of AS2 that is not part of the bundle. These outlying residues are in the so-called N-terminal and C-terminal caps that connect HAMP to TM2 and the first methylation helix (MH1) of the kinase-control module. Suppression analysis confirmed that most of the null mutation/suppressor mutation pairs of residues might be expected to restore x-da packing. No critical residues lie at positions that would be uniquely important for a-d packing. Surprisingly, only three of the mutant proteins were unstable. This result implies that the HAMP domains most likely had nearly native structures.
This summary does not do justice to the extensive work that was done to determine whether the null mutations were dominant or recessive, or whether they could be rescued by forming mixed trimers with the aspartate chemoreceptor Tar (Ames et al., 2002; Studdert and Parkinson, 2005; Ames and Parkinson, 2006). The suppressible mutations were mostly recessive, and many of them were rescuable, indicating that those perturbations were, not surprisingly, relatively mild. The effect of all of the null mutations can be accounted for by destabilizing the HAMP structure by interfering with x-da packing. The authors present their understated take-home message as: ‘The most parsimonious conclusion . . . is that the x-da packing arrangement is the only functionally important 4-helix HAMP bundle in Tsr’.
The implications of this conclusion are profound. The four-helix bundle may be highly dynamic, so that HAMP can exist in a variety of conformations, ranging from very loosely to exceptionally tightly packed. The extent of the packing would be influenced by input from TM2 that reports on attractant or repellent binding to the periplasmic sensing domain, and on feedback from MH1 that signals the level of adaptive methylation.
Four-helix bundles exhibit heptad repeat patterns. For example, the glutamyl residues that are sites of adaptive methylation in the kinase-control module occur at 7-residue intervals. It is intriguing that there is a break of +4 (or −3) residues between HAMP and the kinase-control module. The authors postulate that this helical mismatch could generate an inverse relationship between packing of HAMP and the kinase-control module. That is, tight x-da packing in HAMP might loosen a-d packing in the kinase-control module, and tight a-d packing in the kinase-control module might loosen x-da packing in HAMP.
The packing of the HAMP domain must also be influenced by the connection between AS1 and TM2. Pulling TM2 away from the cytoplasm might put a strain on AS1 that weakens the packing of the HAMP bundle, and thus promotes kinase-stimulating activity. Pushing TM2 towards the cytoplasm, as upon attractant binding, could release the constraints on HAMP, allowing it to pack more tightly and destabilize the kinase-control module to inhibit CheA activity.
A question that remains unresolved by this study is whether the same mechanism might operate in all transmembrane signalling proteins that have HAMP domains. Although HAMP domains are weakly conserved at the level of amino acid sequence, they are highly conserved at the level of predicted three-dimensional structure. However, the only HAMP domain whose sequence has been determined does not even join input and output modules. It occurs at the C-terminus of an archaeal protein of unknown function. The thermophilic nature of the parent organism makes it likely that the relative stability if Af1503 HAMP is the artificial result of its being locked into one conformation at the non-physiological – for A. fulgidus– temperature at which the NMR studies were done.
Chemoreceptor HAMP domains are notoriously unstable and hard to purify, and chemoreceptor HAMP domains are even more dynamic than those of some sensor kinases, such as that of the osmosensor EnvZ (R. Draheim, pers. comm.). Perhaps there are some proteins in which the gearbox model for HAMP function is relevant. For example, there is no transmembrane signal in the redox sensor Aer. Although its HAMP domain also seems to exist as a four-helix bundle (Watts et al., 2008), signalling apparently occurs via a direct interaction of an FAD-binding PAS domain with HAMP, both of which are within the cytoplasm. Of course, PAS/HAMP interaction could also involve stabilization/destabilization of a dynamic, x-da packed four-helix bundle.
Perhaps the most attractive aspect of the inverse-stability hypothesis proposed by Zhou et al. (2009) is that it is eminently testable. The connections between TM2 and HAMP and between HAMP and MH1 can both be genetically modified. One strongly suspects that there are many more mutations in the pipeline.