Functional and structural characterization of RsbU, a stress signaling protein phosphatase 2C.

RsbU is a positive regulator of the activity of sigmaB, the general stress-response sigma factor of Gram+ microorganisms. The N-terminal domain of this protein has no significant sequence homology with proteins of known function, whereas the C-terminal domain is similar to the catalytic domains of PP2C-type phosphatases. The phosphatase activity of RsbU is stimulated greatly during the response to stress by associating with a kinase, RsbT. This association leads to the induction of sigmaB activity. Here we present data on the activation process and demonstrate in vivo that truncations in the N-terminal region of RsbU are deleterious for the activation of RsbU. This conclusion is supported by comparisons of the phosphatase activities of full-length and a truncated form of RsbU in vitro. Our determination of the crystal structure of the N-terminal domain of RsbU from Bacillus subtilis reveals structural similarities to the regulatory domains from ubiquitous protein phosphatases and a conserved domain of sigma-factors, illuminating the activation processes of phosphatases and the evolution of "partner switching." Finally, the molecular basis of kinase recruitment by the RsbU phosphatase is discussed by comparing RsbU sequences from bacteria that either possess or lack RsbT.

Reversible phosphorylation of proteins is the predominant regulatory mechanism in biology, modulating cellular pro-cesses such as signaling, division, and development. The phosphorylation of regulatory proteins by protein kinases effects a change in their function and structure (1), reversed by the action of protein phosphatases, which restore the regulatory proteins to their original, unphosphorylated state. Hence the cellular response is determined by controlling the enzymatic activities of the mutually antagonistic kinases and phosphatases. Protein phosphatases can be divided into three major groups, defined by their substrate specificity (2): phosphotyrosine (further subdivided into Cdc25 and the conventional and the low molecular weight phosphotyrosine phosphatases), phosphoserine/threonine (further subdivided into protein phosphatase P and M families), and phosphoaspartate phosphatases (e.g. Rap and Spo0E from Bacillus). In addition, there is the dual-specific phosphatase group, which can dephosphorylate phosphotyrosine and phosphoserine/threonine substrates, and it has been recorded that several histidine kinases also have phosphohistidine phosphatase activity (3).
In Bacillus subtilis, five serine/threonine protein phosphatases have been identified that belong to the PP2C subgroup of the protein phosphatase M family, namely PrpC (4), SpoIIE (5), RsbP (6), RsbX and RsbU (7). PrpC plays an important regulatory role in stationary phase (8), and SpoIIE regulates differentiation in Bacillus, a process known as sporulation, by forming a complex with the cell division protein FtsZ (5,9). RsbX, RsbP, and RsbU are involved in the regulation of the alternative factor, B , which controls the general stress response of B. subtilis and other Gram ϩ microorganisms, such as the human pathogens Listeria monocytogenes and Staphylococcus aureus (10,11). Depending on the nature of the stress, the signal is conveyed to B by two separate pathways, which converge on phosphorylated RsbV (RsbV-P), 1 the common substrate for RsbP and RsbU (Fig. 1). A decrease in the intracellular energy level activates RsbP, via RsbQ (12), whereas environmental stresses such as heat or salt shock, or ethanol treatment, activate RsbU, via the kinase RsbT (7,13,14). Dephosphorylated RsbV subsequently liberates B from the transcriptionally inactive B ⅐RsbW complex, by competing with B for binding surfaces on RsbW (15,16), freeing B to bind to core RNA polymerase to activate transcription of the B regulon. The alternative binding of RsbW to B or RsbV is a regulatory mechanism called "partner switching" (17).
RsbT is related to the anti-factors RsbW and SpoIIAB; all are members of the GHKL family of kinases/ATPases, a group that includes two-component histidine kinases, topoisomer-ases, and chaperones. However, RsbT does not bind, as in the other partner-switching mechanisms, to or anti-anti-factors. During exponential growth RsbT is thought to be sequestered in a large supramolecular complex composed of RsbR and RsbS (18). However, during environmental stress RsbT is liberated from the supramolecular complex after phosphorylating its substrates, RsbR and RsbS. RsbT is then free to associate with, and activate, RsbU (7). RsbU is not a substrate for the kinase activity of RsbT (19), and the precise mechanism of the activation of RsbU by RsbT remains unknown. Analysis of the sequence of RsbU shows that it is composed of two domains, a C-terminal domain of ϳ200 amino acids with sequence homology to other PP2C-type phosphatases and a N-terminal domain of ϳ110 amino acids with no significant homology to any non-RsbU sequences. The simplest hypothesis regarding the role of the N-terminal domain of RsbU is that it exerts an inhibitory influence on the C-terminal, catalytic domain, which is relieved by the binding of RsbT to RsbU.
We report here an investigation into the activation process of RsbU that integrates genetics and molecular and structural biology. The phenotype of B. subtilis strains containing deletions in the N-terminal domain of RsbU support the view that this domain plays a critical role in the activation of the phos-phatase by RsbT. These results are discussed in the light of our determination of the crystal structure at 1.6-Å resolution of the N-terminal domain of RsbU, and conclusions are drawn as to the nature of the activation process of RsbU. (Table I) were performed with derivatives of B. subtilis wild type strain 168 (20). For the construction of the various truncated versions of RsbU, there were two approaches. First, the rsbU gene, and truncated versions thereof, were amplified by PCR with chromosomal DNA of B. subtilis strain 168 as the template using appropriate primer pairs and a proofreading Taq polymerase. Subsequently, all these PCR products were digested with HindIII and SalI and ligated into the non-integrative plasmid pDG148 (21), which had also been digested with the same enzymes. The different rsbU alleles were thus placed under the control of the IPTG-regulated promoter P spac . After transformation into Escherichia coli strains TOP10 or TG2, and confirmation of the correct DNA sequence of the rsbU inserts by sequencing, plasmids carrying the wild type rsbU sequence (pGK01) and the truncations ⌬rsbU1-19 (pMB21), ⌬rsbU1-38 (pGK02), ⌬rsbU1-77 (pGK03), ⌬rsbU1-93 (pGK04), and ⌬rsbU1-134 (pGK05) were selected for transformation into B. subtilis BSA140 (Table I). The B. subtilis strain BSA140 carries a ctc::lacZ transcriptional reporter gene fusion, but lacks a functional copy of rsbU because of a deletion of an NdeI fragment internal to the rsbU structural gene. Transformants Pre-stress, the anti-factor RsbW sequesters B in a transcriptionally inactive complex. RsbW also inactivates RsbV by phosphorylation through its kinase activity. During the initial response to an imposed stress (solid arrows) either RsbP (energy) or RsbU (environmental) phosphatases are activated, which dephosphorylate RsbV-P. The product, RsbV, attacks the RsbW⅐ B complex and forms an alternative complex with RsbW by liberating B (dashed arrows). Free B then forms a complex with RNA polymerase, initiating transcription of the ϳ130 general stress response genes. RsbT activates RsbU in the environmental stress response pathway. In the absence of stress, RsbT is held in an inactive form in a large molecular weight complex, the "stressosome" with at least RsbR and RsbS. During stress, RsbR presumably becomes phosphorylated and stimulates the rate of phosphorylation of RsbS; both reactions are catalyzed by RsbT. RsbT is then released from the stressosome to activate RsbU (dashed arrows). The phosphatase for RsbS-P, RsbX, resets the level of B activity to the pre-stress levels. Ringed plus signs indicate which events regulate B activity in a positive fashion, ringed minus signs indicate those that are negative. This model has been adapted from, among others, Refs. 18, 54, and 55.

Bacterial Strains, Media, and Growth Conditions-The experiments conducted in this study
were selected for their resistance to kanamycin (20 g ml Ϫ1 ) creating strains BSG14, BSG15, BSG16, BSG17, BSG18, and BSG19, respectively (Table I).
Bacteria were routinely grown under vigorous agitation in a minimal medium described previously (22) supplemented with 0.2% (w/v) glucose as a carbon source and L-tryptophan (0.78 mM). The cultures were inoculated from overnight cultures propagated in minimal media containing kanamycin to an optical density at 540 nm of 0.05. The expression of the plasmid-encoded rsbU variants was induced by the addition of IPTG to a final concentration of 1 mM. Ethanol stress was imposed on the cells during exponential growth phase (A 540 ϭ 0.3) by the addition of ethanol to a final concentration of 4% (v/v).
In the second approach, B. subtilis strain PB291, previously deleted for rsbU, was transformed with derivatives of plasmid pMLK (23), which directs integration at the amyE locus. A fragment of DNA corresponding to P A -rsbR-rsbS-rsbT-rsbU (where P A represents the A -dependent promoter region of the operon) was excised from plasmid pAW70 (24) by BamHI restriction digest and cloned in pMLK using the BamHI site in amyE. For the cloning of rsbR, rsbS, rsbT, and a fragment of rsbU coding for the C-terminal catalytic domain only (⌬N-rsbU-(1-112), the regions corresponding to P A -rsbR-rsbS-rsbT and ⌬N-rsbU-(1-112) were amplified separately by PCR using primers that allow overlaps between the 3Ј and 5Ј ends, respectively. The annealed PCR products were subsequently used as the template for a further round of PCR to yield a DNA fragment of P A -rsbR-rsbS-rsbT-⌬N-rsbU-(1-112), which was also cloned in pMLK at the BamHI site in amyE. Transformants of these strains, which are diploid for rsbR, rsbS, and rsbT, were selected by loss of amylase activity on starch plates. Bacteria were grown in buffered LB, and subjected to 4% ethanol stress at time 0.
In both cases, the ␤-galactosidase activity of the ctc::lacZ reporter gene fusion was determined by harvesting 1-ml aliquots of cells at appropriate time points by centrifugation at 4°C. ␤-Galactosidase enzyme assays were conducted as previously described (14,25).
Purification of Full-length RsbU and Truncated Proteins-RsbU was overexpressed in E. coli (BL21) and was purified using a procedure slightly modified from that previously published (18). Cells were disrupted by sonication in 30 ml of lysis buffer (50 mM Tris-HCl, pH 8, 1 mM phenylmethylsulfonyl fluoride, 1 mM dithiothreitol, 10 mM MgCl 2 ) and centrifuged for 30 min at 15,000 rpm. The supernatant was then applied to a DEAE-Sepharose (Amersham Biosciences) column preequilibrated with 50 mM Tris-HCl, pH 8, 1 mM dithiothreitol, 10 mM MgCl 2 and the chromatogram was developed with a NaCl gradient from 0 to 400 mM. The RsbU-containing fractions were concentrated by centrifugal filtration (Amicon) and then loaded onto a Superdex-200 gel filtration column (Amersham Biosciences). To remove the few remaining contaminants, RsbU was further purified using high resolution Mono-Q ion exchange chromatography using the same buffer conditions as for the earlier DEAE-Sepharose column. For C-RsbU, which was cloned into pET15b, the overexpression and purification protocols were similar to those for full-length RsbU, except that a Superdex-75 gel filtration column was used instead of Superdex-200. The two N-RsbU constructs (residues 1-84 and 1-112) were purified as described previously (26); RsbT was purified as described previously (18) except that all buffers were supplemented with 100 M ATP to maintain RsbT in a soluble form. Estimates of the molecular sizes of RsbU, N-RsbU, and C-RsbU were obtained by the use of Superdex-200 (Amersham Biosciences) gel filtration chromatography, calibrated against proteins of known molecular sizes.
Measurement of the Phosphatase Activity-RsbV-P, the phosphatase substrate, was prepared as previously described (18). The dephosphorylation reactions were performed in a buffer of 50 mM Tris-HCl, pH 7.5, 50 mM KCl, 10 mM MgCl 2 , 1 mM MnCl 2 , and 1 mM dithiothreitol at 30°C with 30 M RsbV-P, 0.5 M RsbU (or C-RsbU) and, unless specified, 1 M RsbT. The rates of dephosphorylation of RsbV-P were measured at various time intervals by removing 20-l samples, stopping the reaction in each by the addition of 10 l of loading buffer (40% glycerol, 200 mM EDTA, and 0.1% bromphenol blue) and placing the sample on ice until analysis by native gel electrophoresis and Coomassie Blue staining. RsbV-P and RsbV bands are easily separated on a 12% acrylamide gel (16). Eight time points were normally taken per reaction. The gels were scanned and intensities of the bands corresponding to the appearance of RsbV were measured with Scion Image software. The values were then compared with a standard curve of known concentrations of RsbV treated under the same electrophoretic conditions. Crystallographic Methods-The structure of N-RsbU was determined by selenomethionine MAD phasing. To prepare selenomethionyllabeled N-RsbU, the E. coli methionine auxotroph, B834 (DE3), was transformed with pETNRsbU L , a pET15b derivative that directs the The spc gene conferring resistance to spectinomycin was introduced by a Campbell-like recombination into the B. subtilis chromosome and is linked to the rsbU⌬NdeI allele.
IPTG-inducible expression of N-RsbU-(1-112) (26). Transformants were initially grown in LB media, before harvesting, washing, and finally inoculating selenomethionyl media. This culture was grown for a further 4 h before the addition of IPTG to a final concentration of 1 mM and the cells were harvested by centrifugation 4 h later, and the cell pellets were stored at Ϫ80°C overnight. Purification and crystallization were carried out as reported previously (26). The near complete incorporation of selenomethionine was confirmed by mass spectrometry.
MAD diffraction data were collected at beamline BM14 of the ESRF, Grenoble, France, from a single crystal of SeMet N-RsbU at three different wavelengths to ϳ2.1-Å resolution. Each data set was individually integrated and reduced using the HKL suite (27) scaling the Bijvoet pairs separately. Native diffraction data were collected separately on ID14-EH2 to a resolution of 1.6 Å for refinement purposes. Data collection statistics are summarized in Table II. The structure was determined with the program SOLVE in its automatic mode (28). A total of two heavy atom sites were found corresponding to the two internal methionines, with a SOLVE Z-score of 9.1. MAD phasing from these sites produced an electron density map with a clear boundary between protein and solvent, and a mean FOM to 2.3 Å of 0.34. These phases were improved by density modification in RESOLVE (mean FOM of 0.47), producing an electron density map of sufficient quality for RESOLVE to automatically build amino acids 3-34 (helices 1-2) and 44 -58 (helix 3) with the correct sequence. After initial refinement in REFMAC (29), the higher resolution data became available and refinement proceeded against these data, where the R free flags were maintained from the MAD data. Building of the rest of the structure, mostly loops and helix 4, was performed by hand in QUANTA (Acclerys). All other calculations were performed using programs from the CCP4 suite (30). The loop between helices 3 and 4 was built in two stages, commencing only when its path could be traced unequivocally. Successive rounds of rebuilding and refinement were interspersed until refinement converged, at an R work of 19.6% and R free of 22.9%. Greater than 97% of the residues of the final model fall within the core region of the Ramachandran plot, and the overall G factor is 1.6. Statistics for the final model are presented in Table II. Atomic coordinates and structure factor amplitudes have been deposited with the RCSB, with accession code 1w53.

Effect of Deletions of the N-terminal Domain of RsbU in
Vivo-If the N-terminal domain of RsbU indeed acts as an inhibitory element preventing activation of RsbU in the absence of stress, partial or complete deletion of the N-terminal part of RsbU should render the phosphatase constitutively active. To test this hypothesis experimentally, a series of N-terminal truncated RsbU variants were designed and their effect on B activity was tested both in exponentially growing cells and in cells exposed to ethanol, a well known environmental stress stimulus. Specifically, the truncations were of the first 19, 38, 77, 93, and 134 amino acids. A wild-type copy and each of the truncated rsbU genes were cloned by PCR into the self-replicating plasmid pDG148 and thus were placed under the control of the IPTG-inducible promoter, P spac . Plasmids were transformed into a derivative of B. subtilis wild-type strain BSA46, in which the chromosomal copy of rsbU had been inactivated by a deletion of an internal NdeI fragment in rsbU (BSA140) (14). The expression of all truncated versions of RsbU on IPTG induction was verified by Western blot analysis with monoclonal antibodies directed against RsbU (data not shown). B activity was monitored under the same conditions with a ctc-lacZ reporter gene fusion. This fusion is known to be strictly B -dependent (31) and thus the ␤-galactosidase activity of those strains is directly correlated to B activity.
Surprisingly, although all the RsbU variants were expressed, none of the strains harboring the plasmid-encoded truncated rsbU alleles displayed any significant ctc-lacZ reporter gene activity during exponential growth in the presence of the inducer IPTG (data not shown). Only the strain carrying a plasmid-encoded full-length copy of rsbU displayed a modest ␤-galactosidase, and hence B activity, in the presence of IPTG. Therefore, partial or complete deletion of the N-terminal domain of RsbU does not render RsbU constitutively active in vivo. Next, the ability of the truncated RsbU variants to mediate environmental stress-triggered activation of B was tested. Cells grown in the presence or absence of the inducer IPTG were exposed to 4% ethanol, a well known strong inducer of the B regulon (32,33). As expected, wild-type strain BSA46 displayed strong transient induction of the ctc-lacZ fusion, which peaked 20 min after ethanol addition and was not altered by the inclusion of the inducer IPTG (Fig. 2). The internal deletion in rsbU in strain BSA140 completely abrogated any ␤-galactosidase activity. Complementation of the rsbU disruption, by providing a full-length copy of rsbU from plasmid pGK01 to form strain BSG14, resulted in an IPTG-dependent increase in ctc-lacZ-encoded ␤-galactosidase activity that also peaked ϳ20 min after exposure to ethanol, but with a peak height that was roughly half that of BSA46. We do not consider that the difference in ␤-galactosidase activity between strains BSG14 and BSA46 is significant; rsbU is not under the control of P spac in strain BSA46, unlike in strain BSG14, and the relative levels of RsbU in these strains, and hence the activity of B , may differ.
In an independent genetic approach, the promoter and polycistronic coding sequences for rsbR, rsbS, rsbT, and rsbU were cloned into pMLK (23). This plasmid, which permits integration at the amyE locus, was transformed into strain PB291 (24), in which rsbU had previously been deleted in a ctc-lacZ fusion background. This strain responds to ethanol shock in much the same way as BSA46 (data not shown), that is, it is independent of IPTG. A second strain was constructed that contained rsbR-rsbS-rsbT, but here a truncated copy of rsbU was cloned, corresponding just to the catalytic domain (residues 118 -335), instead of the full-length gene. This particular gene product could not induce B activity with or without the imposition of stress (data not shown).
RsbU provided in trans can thus convey environmental stress signals. In contrast, none of the strains carrying Nterminal truncated rsbU genes conferred constitutive activity on RsbU variants, nor could these strains activate B in response to environmental stress. Thus any disruption to the N-terminal domain of RsbU, even a deletion of just the first 19 amino acids (Fig. 2B), is detrimental to its function and has profound effects on B activation.
Comparison of the Phosphatase Activities of RsbU and its C-terminal Domain-The above results show that genetic deletions in the regions of rsbU that encode its N-terminal domain lead to an inability of the cells to activate B in vivo and so to respond to stress. These results suggest that the activation of the phosphatase function of RsbU has been affected by truncations in its N-terminal domain, and that the C-terminal domains created by deletion have no phosphatase activity. We addressed these points in vitro by measuring the phosphatase activity of the recombinant C-terminal domain of RsbU (C-RsbU, residues 118 -335), and comparing it to the activity of RsbU alone. We also compared the phosphatase activity of C-RsbU and of RsbU each in the presence of a 2-fold molar excess of RsbT. The enzymatic assay revealed that while C-RsbU alone was active as a phosphatase toward RsbV-P, the rate of dephosphorylation of RsbV-P was only 4-fold greater than that of full-length RsbU in the absence of RsbT (Fig. 3). In contrast, the rate of dephosphorylation of RsbV-P by RsbU was ϳ40-fold higher when stimulated by the addition of a 2-fold molar excess of RsbT over RsbU (Fig. 3). C-RsbU displayed exactly the same, low phosphatase activity in the presence and absence of RsbT (Fig. 3).
The observation of diminished phosphatase activity of C-RsbU in vitro is consistent with our genetic data, which reveal that N-terminal truncated RsbU has insufficient phosphatase activity to trigger a B -directed response to stress. Therefore, it appears unlikely that the function of N-RsbU is to inhibit C-RsbU, with RsbT relieving this inhibition via an interaction predominantly, but not necessarily wholly, with the N-terminal domain of RsbU. We conclude that the N-terminal domain of RsbU does not exert an inhibitory influence on the C-terminal catalytic domain but rather is absolutely required for the activation of RsbU.
Interactions between RsbU and RsbT-In previous studies, the partner switching behavior of RsbW, RsbV, and B was monitored by the use of non-denaturing polyacrylamide gel electrophoresis (16). We applied the same technique to study the interactions between different combinations of RsbT with RsbU, C-RsbU, and the N-terminal 112 residues of RsbU (N-RsbU-(1-112)), which was previously identified as a proteolytically stable fragment of RsbU (26). RsbT alone did not form a sharp, single band in native gels, and a significant quantity of material was found in the well of the gel (Fig. 4, lane 2). In contrast, RsbU and the isolated N-and C-terminal domains migrated well through the gel, with R f (R f ϭ relative mobility to the dye front) values of 0.41, 0.71, and 0.82 (Fig. 4, lanes 1,  3, and 4).
When RsbT was mixed with an excess of RsbU (Fig. 4, lane  5), an additional species was observed (R f ϭ 0.27) that migrated slower through the gel than RsbU, and which we presume is a complex of RsbT and RsbU. Similarly, an additional band R f value of 0.41, presumably corresponding to the complex of RsbT and N-RsbU-(1-112), was observed when these two proteins were mixed (Fig. 4, lane 6). However, when RsbT and C-RsbU were mixed (Fig. 4, lane 7), no additional species were observed, only those of the individual proteins. RsbT and RsbU are therefore capable of forming a complex stable enough to be observed in the conditions of gel electrophoresis. The data presented in Fig. 4 are consistent with the interaction between these two proteins being mediated predominantly, or solely, through the N-terminal domain of RsbU. The isolated domains of RsbU did not interact with each other under the electrophoresis conditions to reconstitute a "full-length" RsbU (Fig. 4,  lane 8). It is thus unlikely that the two domains of RsbU form a stable intramolecular complex. Other, multidomain proteins whose structures undergo a conformational change as part of a regulatory mechanism cannot always be reconstituted from the purified individual domains. For instance, the two isolated In this study, when RsbT was present in a 2-fold molar excess over RsbU, the phosphatase activity of RsbU was stimulated by ϳ40-fold. This stimulation is greater than that previously reported (7), where the phosphatase activity of RsbU was stimulated ϳ7-fold in the presence of a 5-fold molar excess of RsbT. The difference between these amplification factors might come from the fact that we have discovered that RsbT, which is quite unstable and has a tendency to precipitate, behaves much better when ATP or ADP is added throughout the purification process. We therefore might have used a more "active" RsbT in our experiments. The results shown are the average of four independent experiments and the error bars represent the S.D. Ϯ mean. domains of Spo0A do not interact under native gel electrophoresis conditions. 2 Stoichiometry and Stability of the RsbU⅐RsbT Complex-We have demonstrated that the presence of a 2-fold molar excess of RsbT over RsbU increased the latter's phosphatase activity some 40-fold (Fig. 3). To determine the ratio of RsbU:RsbT that is required for maximum activation of RsbU, the rate of dephosphorylation of RsbV-P by RsbU was measured in the presence of increasing concentrations of RsbT. The phosphatase activity of RsbU increased linearly until the ratio between RsbT and RsbU reached 6:1, and reached its maximum at a ratio of 10:1 (Fig. 5). This maximum activity was ϳ200 times higher than that of RsbU in the absence of RsbT. Either RsbU binds 6 or more molecules of RsbT to gain maximum activation, or the ratio at which maximal activity is observed is a reflection of a low-affinity interaction between RsbT and RsbU.
Because it was possible by native gel electrophoresis to detect a complex between RsbU (or N-RsbU-(1-112)) and RsbT, the band corresponding to the RsbU⅐RsbT complex was excised from the gel, and the proteins within the band were electroeluted and then analyzed by SDS-PAGE. Bands corresponding to RsbU and RsbT were observed, and their relative intensities were measured after digitization of the gel. By comparison to known standards of RsbU and RsbT, we conclude that the stoichiometry of the RsbU⅐RsbT complex under these conditions is 1:1.
To characterize the interactions between RsbU and RsbT further, we used gel filtration chromatography to estimate the size of the purified individual proteins, isolated domains, and complexes. The same technique showed previously that RsbT is a monomer (18). The molecular mass of RsbU was found to be around 80 kDa, a value that corresponds to a dimer of RsbU composed of 38-kDa subunits (data not shown). This result is in close agreement with previous gel filtration experiments performed with a B. subtilis cellular extract from which RsbU eluted as a protein of ϳ90 kDa (34). The apparent molecular mass of the N-terminal 112 residues of RsbU was 34 kDa, suggesting that this molecule is dimeric, or perhaps trimeric. The size of the C-terminal construct of RsbU (residues 118 -335, C-RsbU) predicted from its gene sequence is 25 kDa, which was consistent with the gel filtration elution profile that indicated that C-RsbU is monomeric. Hence it would appear that the dimerization determining motifs of RsbU are restricted to the N-terminal domain. Despite testing a variety of different conditions (pH, NaCl, ATP, and Mg 2ϩ concentration), we could not isolate any complex by size exclusion chromatography, be it between RsbT and RsbU, or between RsbT and N-RsbU- . Given that the measured stoichiometry of the RsbU⅐RsbT complex is 1:1, yet RsbT must saturate RsbU for full activation, and that the RsbU⅐RsbT complex could not be isolated by gel filtration, these observations argue that there is only a weak affinity interaction between RsbU and RsbT.
Crystal Structure of N-RsbU-(1-112)-To understand further the function of the N-terminal domain of RsbU, the structure of N-RsbU-(1-112) was determined by x-ray crystallography, using the MAD technique from crystals of selenomethionyl-labeled N-RsbU-(1-112), and refined against diffraction data to 1.6 Å (Fig. 6A). Statistics of the diffraction data and final refined model are presented in Table II. The electron density map does not reveal the conformation of residues 85-112, and these amino acids are missing from the final structure; we assume them to be disordered. One molecule of N-RsbU-(1-112) comprises four anti-parallel ␣-helices, arranged into a rough "L" shape ( Fig. 6B), forming a four-helical bundle. Such bundles are found commonly in biology in a variety of different circumstances; they are thermodynamically favored in protein folding, as seen in the globin superfamily, and also mediate interactions with themselves in dimer interfaces and with other molecules, such as DNA or proteins, for instance in Spo0B, homeodomains, and focal adhesion kinase. To identify structural neighbors, the structure of N-RsbU was submitted to the DALI (35) server. More than 460 structures were identified as having structural similarity to N-RsbU, with Z-scores of 2.0 or greater. Not surprisingly, the sequence identities between N-RsbU and its structural neighbors were low, mostly less than 10%. No single identifiable biological theme linked the structures, and examples were found from the contexts where four helical bundles are prevalent (e.g. residues from hemoglobin (Z-score 3.1, Protein Data Bank code 1CG5), 2 R. J. Lewis, unpublished observations.    (36). 3 , which harbors residues that are important for the recognition of the Ϫ10 region of the promoter, is also found in contact with SpoIIAB residues in the crystal structure of the F ⅐SpoIIAB complex (37). A recent study of the SpoIIAB homologue and RsbU activator, RsbT, suggests that RsbT residues Arg 19 -Gln 20 , Arg 23 -Asn 24 , and Asp 35 -Gln 36 are involved in RsbT-RsbU interactions (38). These residues are equivalent to those in SpoIIAB that bind F . However, the SpoIIAB-contacting residues in F have spatial equivalents in N-RsbU that are not solvent-accessible and thus can only bind RsbT after significant conformational changes occur in N-RsbU. Therefore, the functional significance of the similarity between N-RsbU and 3 remains to be elucidated.
On examination of the crystal packing, it was evident that N-RsbU-(1-112) had crystallized as a dimer around a crystallographic 2-fold axis. 28% of the available surface area, some 1600 Å 2 , is buried in the dimer interface; 80% of the atoms in the interface are non-polar and no water molecules are observed in the interface at the resolution of 1.6 Å. This extensive dimer interface is composed of residues from all four helices of each protomer, but is dominated by helix 4, which contributes about half of the buried surface in the dimer interface, whereas helices 1 to 3 each contribute ϳ1/6th. Five hydrogen bonds between residues in protomer one and two are duplicated by the 2-fold symmetry of the interface, to make a total of 10 that stabilize the dimer. The residues involved, Gln 31 , Thr 37 , His 51 , Tyr 59 , Asp 65 , Ser 69 , Glu 75 , and Tyr 80 , are conserved across the RsbU family, except for Thr 37 and the interacting pair of Tyr 59 and Asp 65 (Fig. 7). In this latter case, a change at one position in RsbU orthologues is compensated for by an alteration in the other position such that either hydrogen bonding or hydrophobic packing is conserved (e.g. pairings are found of Asp and Asn, or Gly and Phe, at positions 59 and 65 in the orthologues of RsbU). Variation to Leu, Phe, Ala, or Val is permitted at Thr 37 FIG. 6. The structure of N-RsbU. A ribbon diagram color-ramped from blue to red, from the N to C terminus in each protomer is drawn in three, mutually perpendicular views with the middle panel as the reference point (A). Helices 1 (amino acids 2-20) and 2 (24 -40), and helices 3 (44 and 58) and 4 (64 and 84) form two pairs, which are separated by about 45°(C). In panel B, the view is along the 2-fold symmetry axis of the dimer, revealing the L-shaped form of each protomer; in panel C, the relative disposition of helices 1 and 2, relative to helices 3 and 4Ј (where Ј indicates that helix 4 is from the alternative protomer in the dimer) is illustrated. This figure was created with BOBSCRIPT (56) and Raster3D (57).
where F o and F c are the observed and calculated structure factor amplitudes. c R free was calculated with a 5% randomly-selected subset of the reflections not used in refinement.
because residues in this position can interact with Tyr 80 either by forming a hydrogen bond, or by hydrophobic packing. Other residues found at the dimer interface are well conserved across the RsbU orthologues sequenced to date. In a previous study, the mutation P44R was identified in a genetic screen for suppressors of a deletion in rsbX (39). Strains deleted for rsbX exhibit constitutively high B activation, and rsbU44PR suppresses this activity. This mutation also abolishes the activation of B on the imposition of stress. In the structure of N-RsbU, Pro 44 is situated at the dimer interface, sandwiched between conserved hydrophobic residues including Met 77 and Tyr 84 , and in conjunction with Glu 45 it stabilizes the N-terminal end of helix 3. The data of Smirnova et al. (39) indicate that although rsbU44PR accumulates in the cell, it cannot be activated, because the ability of RsbU and RsbT to interact has been destroyed by this mutation. Mutation from proline to arginine at this position is highly likely to disrupt the dimer interface as well as to destabilize helix 3. The inability of rsbU44PR to be activated by RsbT could be explained by the loss of dimerization of RsbU, affecting the binding surface on RsbU for the activator RsbT.
A Potential RsbT Binding Surface on N-RsbU-The P44R mutation in RsbU that affects activation by RsbT led us to consider further the interaction between RsbU and RsbT. An analysis of the completed genomes of Gram ϩ microorganisms that encode a B orthologue reveals that not all of the Rsb proteins are conserved. For instance RsbR, RsbS, RsbT, and RsbX have not been identified in the genomes of S. aureus and Staphylococcus epidermidis. Presumably, in those bacteria where no activator of RsbU has been discovered, RsbU is regulated in a different way to that in B. subtilis. Therefore, a comparison of the amino acid sequence of RsbU orthologues that originates from bacterial species that contain an RsbT activator, and those that do not, may reveal the RsbT-interacting surface in RsbU. The sequence alignment of the N-terminal domains of RsbU orthologues (Fig. 7) provides insight into the structure and function of RsbU. Those residues that are completely conserved in the alignment are found at the dimer interface, participating in either hydrogen bond (e.g. His 51 and Glu 75 Ј, where Ј indicates that this residue is found in the other protomer in the dimer), or van der Waals (e.g. Tyr 11 , Leu 27 , and Leu 73 Ј) contacts. Moreover, a larger amino acid than glycine at position 81Ј could not accommodate the approach of the long carboxylic acid side chain of Glu 45 , which also serves to N-cap helix 3. Other residues that are particularly poorly conserved are located mainly in loops between the secondary structural elements or on the solvent-exposed faces of the four helices.
The residues that are solvent-accessible and conserved in RsbT encoders (Bacilli and Listeriae), and not necessarily so in RsbT non-encoders (Staphylococci), are highlighted in the sequence alignment (Fig. 7). Glu 24 , Tyr 28 , and Arg 35 are found in a discontinuous group on one face of helix 2: the break at position 31 coincides with a 1-residue lysine insertion in the sequences of RsbU from the Staphylococci. In conjunction with the hydrophobic residues Ile 74 Ј, Ile 78 Ј, Met 82 Ј, and Ala 83 Ј from helix 4Ј, these residues form two prominent, parallel surface ridges of edge length ϳ17 Å (Fig. 8A). These surface ridges are only formed as a consequence of the dimerization of N-RsbU and in effect they form a ϳ10-Å wide groove, the floor of which comprises residues Ser 34 , Ile 38 , Val 76 Ј, Gly 79 Ј, and Tyr 80 Ј, which are well conserved across all RsbU orthologues. Overall, the prominent surface features of N-RsbU from B. subtilis are unlikely to be conserved in the Staphylococci, leading us to conclude that these surface patches may represent at least part of the binding surface for RsbT on RsbU. For instance, replacement of Tyr 18 by serine in the Staphylococci would undoubtedly alter the local structure and molecular surface. Calculation of the molecular surface after replacement of the tyrosine at this position by serine reveals that a cavity is formed (Fig. 8B). Fig. 4, we monitored the binding of RsbT by RsbU and N-RsbU-(1-112). However, the structure of N-RsbU-(1-112) did not reveal the conformation of residues 85-112, because this part of the protein was disordered in the crystal. Other residues that are conserved only in those bacteria that code for rsbT include Glu 86 , Arg 91 , Ile 97 , and Ser 99 (Fig.  7), which cannot be modeled in the present structure. These amino acids may also play a role in RsbT binding. To investigate whether residues 85-112 of RsbU contribute significantly to RsbT binding, we constructed a truncated form of N-RsbU that corresponds precisely to the ordered portion of this crystal structure, i.e. residues 1-84. We monitored whether N-RsbU-(1-84) could recruit RsbT in a binding assay, as analyzed by non-denaturing gel electrophoresis. N-RsbU-(1-84) runs a little quicker in electrophoresis than N-RsbU-  in the absence of RsbT, but in its presence an additional electrophoretic species can be observed for both constructs of N-RsbU, which we conclude corresponds to N-RsbU⅐RsbT complexes (Fig. 9a). Note that the band corresponding to the N-RsbU-(1-84)⅐RsbT complex in lane 1 is slightly more diffuse than the band in lane 2 of the N-RsbU-(1-112)⅐RsbT complex. The relative diffuseness may reflect the fact that the interaction of RsbT with N-RsbU-(1-84) is slightly weaker than that between RsbT and N-RsbU-(1-112).

A Comparison of RsbT Binding by N-RsbU-(1-84) and N-RsbU-(1-112)-In
To assess whether there is a significant difference in the binding affinity of the two N-RsbU constructs for RsbT, we designed a competition assay in which a molar equivalent of N-RsbU domains in comparison to RsbU was added to RsbV-P dephosphorylation reactions. The initial rate of RsbV-P dephosphorylation was reduced by 45% for N-RsbU-(1-84) and 40% for N-RsbU-(1-112) (Fig. 9b). The results of these experiments indicate that both N-RsbU-(1-84) and N-RsbU-(1-112) are competent in binding RsbT, and that the major RsbT-binding determinants are found in those amino acids that could be located in the FIG. 7. Sequence alignment of N-RsbU orthologues. The construct used for structure determination, the first 112 amino acids of RsbU, is shown with every 10th amino acid in B. subtilis RsbU numbered and the secondary structural elements drawn above, with ␣-helices represented by blue bars. Residues colored green are totally conserved across the RsbU family, those in red are conserved in those bacteria that encode rsbT, but not necessarily conserved in those bacteria that do not. The residue in black is Pro 44 , mutation of which to Arg suppresses a deletion in rsbX (39). Gaps in the alignment are represented by -; residues that contribute to the dimer interface are highlighted by a * for hydrogen bond donors/acceptors and a^for hydro- crystallographic electron density, residues 1-84. Although we cannot rule out the possibility that residues 85-112 are involved in the binding of RsbT during the activation of RsbU, we can conclude that any interaction must be relatively weak. DISCUSSION PP2C-type phosphatases act in the signaling pathways that regulate the response to stress in eukaryotes (40,41) as well as in B. subtilis and its close relatives (7). The B. subtilis PP2Ctype phosphatase, RsbU, activates the general stress factor, B , following the imposition of environmental stress (see Fig.  1). Induction of the large B -dependent regulon provides the cell with a multiple and pre-emptive stress resistance (42), but also constitutes a considerable burden. Thus, tight control of the activity of B , and therefore RsbU, is crucial. The activities of eukaryotic phosphatases are known to be controlled by a variety of mechanisms, including post-translational modifica-tion and the binding of accessory proteins, but little is known of the regulation of PP2C phosphatases (43). For instance, the eukaryotic cell cycle phosphatases Cdc25A (44) and Cdc25C (45) are activated by phosphorylation. In contrast, the kinase activity of RsbT is not required for activation of RsbU (19).
A simple activation mechanism for RsbU, where the N-terminal domain suppresses the function of the C-terminal domain until a conformational change occurs in RsbU on the binding of the activator, RsbT, would be an elegant solution to the problem of control. Such a mechanism has already been observed for human and rabbit protein phosphatase 5 (PP5), which is stimulated by the interaction of arachidonic acid with the TPR domain of PP5. This interaction drives a conformational change in PP5 that overcomes the TPR-mediated inhibition of the phosphatase domain. In this instance, the enzymatic activity of arachidonic acid-stimulated PP5 approaches FIG. 8. The RsbT-binding surface? A, the molecular surface of the N-RsbU dimer is colored mostly blue: those parts of the surface colored lime correspond to residues that are completely conserved in the sequence alignment in Fig. 7; note that few of them map to the surface, and the ones that do (Leu 27 and Glu 75 Ј; Gly 81 and Glu 45 Ј) are part of the dimer interface, situated between helices 3 and 4Ј. The red patches of the molecular surface relate to those amino acids that are conserved in bacteria that encode RsbT in their genomes, but are not necessarily conserved in bacteria that do not harbor an rsbT gene. B, the inset reveals a section of the surface in the vicinity of Tyr 18 , where the tyrosyl side chain has been replaced with a seryl, with the result that a cavity is formed. In both panels, the same orientation as panel A in Fig. 6 is used, and this figure was prepared with Py-MOL (www.pymol.org). that of the isolated phosphatase domain (46). However, in vitro the isolated catalytic domain of RsbU has a very low phosphatase activity in comparison to RsbT-activated full-length RsbU (Fig. 3). Moreover, N-terminal truncated RsbU proteins cannot induce B activity in vivo, and furthermore, the basal level of B activity in these strains resembles that of wild type (Fig. 2). The N-terminal domain of RsbU is therefore absolutely required for RsbU activation, by recruiting RsbT (Fig. 3).
The results presented here support and extend the earlier conclusions by Kang et al. (19) that RsbT activates RsbU by forming a protein⅐protein complex. In this study, we demonstrate that the association is mediated predominantly, but not necessarily exclusively, by the first 84 amino acids of the Nterminal domain of RsbU (Figs. 4 and 9). The crystal structure of the N-terminal domain of RsbU reveals that only a dimeric form of RsbU can form what we believe to be an RsbT-binding surface. Deletion of the N-terminal part of rsbU (this study), or mutation of Pro 44 in the dimer interface (39), has deleterious effects on the stress response of B. subtilis, presumably because the dimer form of RsbU is destroyed. The residues that would appear to mediate these protein-protein interactions are only conserved in RsbU orthologues from bacteria that include rsbT in their genomes (Fig. 7).
What molecular mechanism is used in the activation of the phosphatase domain of RsbU during the recruitment of RsbT by N-RsbU? RsbU is dependent on manganese, and one method of RsbU activation might involve the modulation of the K m for Mn 2ϩ in the presence of RsbT, leading to a more efficient dephosphorylation of RsbV-P. This model is not without precedent in the PP2C phosphatase family; the catalytic subunit of bovine mitochondrial pyruvate dehydrogenase phosphatase displays a K m for magnesium that increases in the presence of its regulatory subunit (47). However, the K m value for Mn 2ϩ of RsbU, of RsbU in the presence of RsbT, and of the C-terminal domain of RsbU alone were found to be 0.96 mM in all three cases (data not shown) and thus similar to that observed for other PP2C phosphatases (48,49). It would therefore seem unlikely that RsbU is regulated by changes in the K m for its requisite co-factor.
An alternative mechanism might entail the modulation of the oligomeric state, a common means of control. From solution measurements, we have concluded that RsbU, in the absence of RsbT, is a dimer, and the structure of N-RsbU reveals a significant dimer interface (Fig. 6). The thermodynamic barrier of dimer dissociation in this instance is likely to be considerable. Unless the reaction is driven by the free energy of hydrolysis of ATP, for instance, the apparent robustness of the N-RsbU dimer interface should exclude an activation mechanism where the presence of RsbT dissociates the N-RsbU dimer, leading to enzymatic stimulation.
Alternatively, the formation of the RsbT⅐RsbU binary complex might be a prerequisite for phosphatase activation. As in N-RsbU, the non-catalytic domains from phosphatases 2A (50), 2C (51), and 5 (52) are also all ␣-helical (Fig. 10). Despite the fact that there is little meaningful sequence homology between these domains and RsbU, these non-catalytic domains were all found with DALI, with Z-scores of 3.8 (PP2A), 3.2 (PP2C), and 2.2 (PP5). For instance, helices 1, 3, and 4 of N-RsbU were aligned against all three helices in the regulatory domain of PP2C, which has been proposed to play a role in the determination of substrate specificity (51). Similarly, N-RsbU also matched to the TPR motifs from PP5, which are used to mediate macromolecular interactions and the assembly of protein⅐protein complexes. Furthermore, there is structural similarity between N-RsbU and the HEAT repeat units of PP2A, which coordinate the assembly of the phosphatase and regulatory subunits into a functional heterotrimer (50). It appears that the N-terminal domain of RsbU may have evolved to perform a similar scaffolding function, recruiting RsbT, and perhaps it is this binary complex that actually recognizes the RsbV-P substrate.
In the absence of a stable RsbT⅐RsbU complex, determination of the molecular basis of phosphatase activation has proven elusive. The present study reveals that for maximal activation of RsbU, at least a 6-fold molar excess of RsbT must be present (Fig. 5). We have previously demonstrated that it is likely that there are at least six molecules of RsbT bound per stressosome prior to the induction of stress signals (18). It is perhaps not coincidental that the same number of molecules of RsbT that is sequestered by the stressosome is required to activate fully RsbU. The expression of ϳ130 B -dependent genes in response to stress is an energetically demanding process, and therefore has to be kept under strict control. The partnership between the stressosome and RsbU may actually sense the amplitude of the stress signal, and act as a tuning system to regulate the level of B activity in accordance to the level of stress. This partnership could also act as part of a damping mechanism, if a threshold level of RsbU activity is necessary to induce B FIG. 10. Evolution of phosphatase regulatory domains. Ribbon diagrams illustrating the structural similarities between the regulatory domains of protein phosphatases 2A, 2C, and 5, which have been aligned against the equivalent section of the structure of N-RsbU. The aligned parts of structure are colored in gold, the non-equivalent parts are colored semi-transparent blue.