We previously conducted a genetic screen (Ho and Bradshaw, 2021) to identify features of RsbU that are important for phosphatase regulation by isolating gain-of-function variants that are active in the absence of RsbT. Focusing on the phosphatase domain in our initial study, we reported identification of an amino acid substitution in α1 (RsbU^M166V/L) that links the switch element to RsbU activation (Ho and Bradshaw, 2021). From this screen, we isolated a second variant, RsbU^Q94L, located in the linker of RsbU between the N-terminal dimerization domain and the C-terminal phosphatase domain (Figure 1B–D). rsbU^Q94L strains activated σ^B in the absence of rsbT. This activation is comparable to when rsbT is overexpressed to exceed the capacity of the stressosome and is stronger than what we observe in strains expressing rsbU^M166V/L (Figure 1B).

To visualize how the Q94L and M166V/L substitutions could impact RsbU structure and activity, we generated an AlphaFold2 model of an RsbU dimer (Jumper et al., 2021; Mirdita et al., 2022; Figure 1D). The prediction of the complex was of high confidence (Figure 1—figure supplements 1–2) and had a dimer of the N-terminal region consistent with a previous experimental structure (Delumeau et al., 2004). In our model, the linkers form extended alpha-helices (82-112) that cross to connect dimers of the RsbU N-terminal domains and C-terminal phosphatase domains, with Q94 buried in the linker interface and M166 buried in the core of the phosphatase domain (Figure 1D).

The crossing linker helices have similar conformation to homologous linkers observed in crystal structures of Pseudomonas aeruginosa RssB, a phosphatase/adapter protein that also controls the GSR (PDB 3F7A and 3EQ2, Figure 1—figure supplement 3). RssB has an N-terminal response-receiver domain in place of the RsbU four helix bundle dimerization domain, suggesting that linker conformation is a conserved feature of PPM phosphatase dimers with divergent N-terminal regulatory domains (Figure 1—figure supplement 3). The linker helices in the RsbU prediction and the RssB structure splay apart as they enter the PPM domain with similar angle (pitch angles (α) based on Crick parameters for coiled-coils –17.6° and –18.4°, respectively) and pack against α–1 of the PPM fold (Grigoryan and Degrado, 2011). Interestingly, we observed that other bacterial signaling proteins that are regulated by coiled-coil linkers had similar pitch angle for the helices connecting to the catalytic domains (for example the dimeric histidine kinase CheA, 1B3Q, α –16.8°, and the dimeric GGDEF diguanylate cyclase PadC, 5LLW, α –14.3°; Bilwes et al., 1999; Gourinchas et al., 2017; Grigoryan and Degrado, 2011; Figure 1—figure supplement 3). Thus, the linker of RsbU could be a conserved regulatory feature, analogous to linkers of histidine kinases and GGDEF diguanylate cyclases.

Analysis of the RsbU dimer model using Socket2 identifies amino acids 83–97 as a two-turn coiled coil, with Q94 unfavorably packing against leucine 90 (Kumar and Woolfson, 2021; Figure 1—figure supplement 3). To test whether hydrophobicity at position 94 promotes an active state, we tested the activity of RsbU alleles with different amino acid substitutions at position 94 for σ^B in wildtype and rsbT deleted strains. We found that substitutions of Q94 with amino acids that could pack at the intersection of the crossing linker helices (hydrophobic residues isoleucine, valine, and methionine, as well as tyrosine) resulted in mild, RsbT-independent activation of σ^B that was further heightened by RsbT (Figure 1—figure supplement 3). Aromatic hydrophobic residues (phenylalanine and tryptophan) drove constitutive activation of RsbU but did not bypass the requirement for rsbT on the chromosome (Figure 1—figure supplement 1). Substitutions of small or charged residues (alanine, glycine, glutamate, and asparagine) did not lead to activity in either strain (Figure 1—figure supplement 3). These results are consistent with a model in which burial of the amino acid at position 94, such as in the crossed helices of our prediction for RsbU dimer structure, drives RsbU activation.

One prediction of our hypothesis that RsbT stabilizes the crossed alpha helices of the RsbU dimer, is that RsbT should bind more tightly to RsbU^Q94L than to RsbU because the coiled-coil conformation that RsbT binds would be more energetically favorable. To test this, we measured the binding affinity of RsbT for RsbU using fluorescence anisotropy. From these data we obtained a K_D of 4.5±1 µM, which decreased approximately 10-fold for RsbU^Q94L (0.56±0.2 µM; Figure 2A). Thus, RsbU^Q94L stimulates RsbU phosphatase activity by an on-pathway mechanism, supporting the idea that RsbT binding stabilizes the linker in a dimerized conformation to activate RsbU.

We previously found that RsbT activates RsbU by simultaneously decreasing the K_M for manganese cofactor and increasing the k_cat for dephosphorylation of RsbV-P (Ho and Bradshaw, 2021). To test whether RsbU Q94L substitution works by the same kinetic mechanism, we measured dephosphorylation of ³²P-phosphorylated RsbV by RsbU^Q94L (Figure 2B). The k_cat of RsbU^Q94L was nearly the same as for the basal activity of RsbU (2.4±0.1 min^–1 compared to 1.4±0.1 min^–1, and stimulated more than 10-fold by RsbT, 15±0.4 min^–1; Figure 2B; Ho and Bradshaw, 2021). However, RsbU^Q94L had a K_M for MnCl₂ (1.4±0.2 mM) that is more than 50-fold reduced compared to RsbU alone (77±9 mM; Ho and Bradshaw, 2021), and similar to the K_M^MnCl2 of RsbU in the presence of RsbT (0.98±0.07 mM Ho and Bradshaw, 2021; Figure 2B). Addition of RsbT did not change the K_M of RsbU^Q94L for MnCl₂ (1.9±0.6 mM), but the maximum catalytic rate of dephosphorylation increased nearly four-fold to 9.0±0.64 min^–1, approaching the rate of RsbU stimulated by RsbT (Figure 2B). Thus, RsbU^Q94L substantially recapitulates RsbT activation by decreasing the K_M for Mn²⁺, demonstrating that the linker transduces RsbT binding to control the active site of RsbU (Figure 2C). This is consistent with previous findings that the allosteric switch controlling phosphatase activity of SpoIIE couples dimerization to cofactor interaction.

To visualize how RsbT could control the linker to activate RsbU, we generated an AlphaFold2 prediction of a heterotetrameric complex between RsbT and RsbU (Figure 3A). In this model, RsbU adopts a conformation very similar to our prediction of the RsbU dimer (RMSD 1.36 Å), suggesting that our AlphaFold2 RsbU dimer model represents the active state of RsbU even when RsbT was not included in the prediction (a known feature of AlphaFold2 models Chakravarty and Porter, 2022). The predicted structure places RsbT in a hybrid interface that spans the N-terminal dimerization domain of both RsbU monomers and the linker of one monomer (Figure 3A). Supporting the validity of the AlphaFold2 2RsbT/2RsbU model, residues previously implicated in RsbT/U binding (Hardwick et al., 2007; Y28 and E24 from one RsbU monomer, I74, and I78 from the other RsbU monomer, and RsbT R19, R23, and N24) are all buried in the predicted interface (Figure 3—figure supplement 1).

To gain further insight into how RsbT binding is transduced to RsbU linker conformation and phosphatase activity, we performed a genetic screen to isolate suppressors of RsbU^Y28I, which was previously observed to abolish binding to RsbT (Hardwick et al., 2007). After screening a library of PCR mutagenized rsbU^Y28I and rsbT, we identified 11 variants that introduced substitutions to RsbU that suppress rsbU^Y28I: S49G, K53E, T89A, G92E, G92R, Q94L, T110S, V142A, K143E, C165Y, and M173I (Figure 3A–B). We additionally isolated revertants that restored Y28 suggesting that the screen was near saturation, but did not isolate any suppressors in rsbT, which was also mutagenized.

Mapping the RsbU^Y28I suppressor mutations onto our 2RsbU/2RsbT model suggests that they fall into three classes (Figure 3A and B): (1) Substitutions in the RsbU N-terminal dimerization domain that map near the predicted RsbT binding site (S49G and K53E), supporting the predicted interface. (2) Substitutions that map near predicted contacts between RsbT and the RsbU linker (T89A, G92E/R, and Q94L), providing evidence that contacts between RsbT and the linker promote RsbU activation. (3) Substitutions that cluster around the regulatory switch element of the RsbU phosphatase domain (helices α1 and α2 of the PPM fold, T110S, V142A, K143E, C165Y, and M173I). We hypothesize that these substitutions rescue RsbU^Y28I inactivation by reducing the energetic cost (normally provided by RsbT binding) to transition the RsbU phosphatase domain to the active conformation. Together these genetic data provide independent support for the AlphaFold2 model of the 2RsbT/2RsbU complex, and suggest that RsbT activates RsbU by transmitting signals to the switch element in the RsbU phosphatase domain through a coiled-coil linker.

To test the model that RsbT activates RsbU by directly interacting with the linker to dimerize the RsbU phosphatase domains, we introduced a charge swap at position R91 that would abolish a predicted salt-bridge with RsbT D92 (Figure 3C). RsbU^R91E abolished σ^B activity for cells overexpressing rsbT (Figure 3C), and in biochemical assays RsbT stimulation of RsbU^R91E phosphatase activity was abolished (Figure 3D). Finally, we introduced a potentially compensatory charge switch at position D92 of RsbT (RsbT^D92K; Figure 3C). Expression of rsbT^D92K partially restored σ^B activity to rsbU^R91E expressing cells, consistent with direct interaction between these residues (Figure 3C). Together, these results suggest that contacts between RsbT and the linker of RsbU are responsible for phosphatase activation.

We considered two models for why RsbU is inactive in the absence of RsbT. One possibility is that inactive RsbU resembles the dimeric structure predicted by AlphaFold2 and that RsbT induces a subtle conformational change, for example changing the registry or pitch of the crossing linker helices. A second possibility is that inactive RsbU is dimerized by the N-terminal domains but that the linkers of inactive RsbU are flexible and that the phosphatase domains only interact with each other when RsbT orders the linkers into a crossing conformation. To distinguish between these two models, we performed small angle X-ray scattering coupled to size exclusion chromatography with multi-angle light scattering (SEC-MALS-SAXS) to assess the shape and flexibility of RsbU dimers in solution (Table 1 and Figure 4—figure supplements 1–2 S4A). RsbU eluted as a single peak from the size exclusion column with an average molecular weight of 77 kDa (calculated dimeric molecular weight 77.3 kDa) determined by multi-angle light scattering (MALS; Figure 4—figure supplement 1). However, our sample had a minor contaminant of heterodimeric complexes of RsbU in which one monomer is C-terminally truncated that could influence the SAXS analysis (Figure 4—figure supplement 1). We therefore deconvoluted the SAXS profiles using evolving factor analysis (EFA) in RAW to isolate scattering profiles from the individual components (Hopkins et al., 2017; Meisburger et al., 2016; Figure 4—figure supplement 2).The EFA deconvolution identified a major component (MW 75.6 kDa calculated from V_c method), which eluted earlier than a smaller component (MW 48.8 kDa calculated from V_c method; Table 1 Guinier fits, and Figure 4—figure supplement 2). We conclude that the larger component is a RsbU dimer, while the smaller fragment is likely a heterodimeric complex of RsbU and the C-terminal truncation product of RsbU (Figure 4—figure supplement 1).

Indicative of significant deviation between the RsbU structure in solution to the AlphaFold2 model, the scattering intensity profile (I(q) vs. q) was a poor fit (χ² 12.53) to a profile calculated from the AlphaFold2 model of an RsbU dimer using FoXS (Schneidman-Duhovny et al., 2016; Schneidman-Duhovny et al., 2013; Figure 4A). We therefore assessed the SAXS data for the RsbU dimer for features that report on flexibility (Kikhney and Svergun, 2015). First, the scattering intensity data lacked distinct features caused by the multi-domain structure of RsbU from the AlphaFold2 model (Figure 4A). This can be visualized more readily from a dimensionless Kratky plot, which is bell-shaped and peaks above 1.3, indicating that the RsbU dimer is non-globular (Figure 4B). Additionally, the data do not show the two peaks from the FoXS computed model of the AlphaFold2 dimer that reflect the extended two domain structure (Kikhney and Svergun, 2015; Receveur-Brechot and Durand, 2012; Figure 4—figure supplement 2). The scattering data had several additional characteristic features indicating flexibility: the P(r) plot is smooth and has an extended tail (Cordeiro et al., 2017; Kikhney and Svergun, 2015; Receveur-Brechot and Durand, 2012; Figure 4C), R_g and I(0) calculated from P(r) are larger than when calculated from Guinier fits (Cordeiro et al., 2017; Kikhney and Svergun, 2015; Receveur-Brechot and Durand, 2012; Table 1), molecular weight calculated using the MoW2 modified Porod volume method is an overestimate (Rambo and Tainer, 2011; Table 1), and the R_g (39.05±0.16 Å) was larger than the R_g of the AlphaFold2 dimer model (34.05 Å). Prompted by this evidence of flexibility, we performed multi-state modelling using MultiFoXS (Schneidman-Duhovny et al., 2016), which incorporates flexibility and heterogeneity. MultiFoXS produced a two-state solution with χ² of 1.2 (as compared to the one-state model with χ² of 12.5) and low residuals across the entire range of data (Figure 4D, Figure 4—figure supplement 2). We allowed flexibility for the linker residues 82–96 while keeping the N-terminal domain dimerized because a stable dimer of the N-terminal domain is observed in solution, was present in a previous X-ray crystal structure of C-terminally truncated RsbU{Delumeau:2004ep}, and C-terminally truncated RsbU coelutes as a heterodimer with the full-length protein (Figure 4—figure supplement 1). Importantly, the precise region of allowed linker flexibility does not have a substantial impact on the MultiFoXS fit (Figure 4—figure supplement 3), but constraining the C-terminal phosphatase domain dimer does not yield a good fit to the data (χ² of 5.4; Figure 4—figure supplement 3). The structural models predicted by MultiFoXS depict a dimerized N-terminal domain with flexible linkers and PPM domains that do not make contact. Notably, this approach also would include models in which both domains remain dimerized, suggesting that such models are not the best fit of the data. The two conformations that were selected as the best fit to the scattering data by MultiFoXS also have very different R_g values, (34.91 Å [71%] and 45.69 Å [29%]), indicating RsbU samples a range of conformational space that includes a more compact state(s) consistent with the AlphaFold2 model and a significantly extended state(s), emphasizing that the linker is highly flexible. Together, we conclude that the RsbU linker appears to be flexible in the absence of RsbT (Figure 4G).

To test the hypothesis that RsbT rigidifies the linker to produce an active RsbU dimer, we attempted SEC-MALS-SAXS analysis of the RsbT/U complex. RsbT/U eluted with molecular weight (determined by MALS) of 88 kDa compared to 105 kDa for a 2RsbT/2RsbU heterotetramer (Figure 4—figure supplement 1), suggesting that the RsbT/U complex had largely disassembled on the column. We therefore performed SEC-MALS-SAXS on the RsbT/RsbU^Q94L complex, which has a higher affinity (Figure 2A, Figure 4—figure supplement 4). The RsbT/U^Q94L complex eluted as a major peak with a higher molecular weight shoulder (Figure 4—figure supplements 1 and 4; also visible when we ran RsbU^Q94L alone. Figure 4—figure supplement 1), with the average molecular weight of the main peak (determined by MALS) of 125 kDa (Figure 4—figure supplement 1). This is larger than the predicted molecular weight of a heterotetrameric 2RsbT/2RsbU^Q94L complex (105 kDa), suggesting that some higher order complexes form. EFA deconvolution (Meisburger et al., 2016) identifies two components. The first deconvolved component had a SAXS determined molecular weight of 164 kDa, while the second had a SAXS molecular weight of 103 kDa, consistent with a 2RsbT/2RsbU complex (105 kDa). Together with the fact that RsbT co-elutes with RsbU^Q94L throughout the entire peak (Figure 4—figure supplement 1) and the MALS molecular weight, we assess that this second component is an 2RsbT/2RsbU hetero-tetramer. The identity of the higher molecular weight component of the scattering data is ambiguous, but could represent a larger assembly of RsbT/RsbU^Q94L or multimers of RsbU^Q94L that were observed when RsbU^Q94L was analyzed alone (Figure 4—figure supplement 1). Several lines of evidence demonstrate that the 2RsbT/2RsbU^Q94L complex lacks the flexibility of the 2RsbU complex. First, the R_g from the Guiner fit (35.06±0.43 Å) matches well with the model computed from the AlphaFold2 prediction of the heterotetrameric complex of 2RsbT/2RsbU using FoXS (33.68 Å) and is smaller than the R_g we observed for the RsbU dimer (39.05±0.16 Å). Second, the features of the scattering data that suggested flexibility for RsbU dimers all suggest that the RsbT/RsbU complex is more rigid (the quality of the FoXS fit, a lower peak of the dimensionless Kratky plot, and a shorter tail of the P(r) plot; Cordeiro et al., 2017; Kikhney and Svergun, 2015; Receveur-Brechot and Durand, 2012; Figure 4C, E, F, Figure 4—figure supplement 4). Third, fitting the scattering profile (I(q) vs. q) to the AlphaFold2 model yielded a χ² of 1.83 without requirement for flexibility (Figure 4F). As expected, based on the lack of evidence of flexibility in the scattering profile and P(r) function, the model is not improved by the use of multi-state modeling by MultiFoXS, which was run as a control (Schneidman-Duhovny et al., 2016; Figure 4—figure supplement 4). Together, these data support the validity of the AlphaFold2 model of the heterotetrameric 2RsbT/2RsbU complex, and additionally suggest that RsbT rigidifies the otherwise flexible linker of RsbU to dimerize the phosphatase domains (Figure 4G). Together with our biochemical data linking RsbU linker conformation to RsbT binding, metal cofactor interaction, and phosphatase activity, this provides a comprehensive model for how RsbU is activated (Figure 4G).

B. subtilis has a second GSR-initiating PPM phosphatase, RsbP, that has an N-terminal PAS domain and dephosphorylates RsbV to activate σ^B in response to energy stress. Our AlphaFold2 prediction of an RsbP dimer shows an extended alpha-helical linker connecting dimeric PAS domains and phosphatase domains. Like RsbU, the interface between the dimerized phosphatase domains is predicted to be mediated by α1 and α2 of the phosphatase domain. We found that previously identified constitutively activating mutations in RsbP (Brody et al., 2009) map to similar positions as the gain-of-function mutations we identified here for RsbU in their respective predicted structures, namely the interface between α1 and α2 of the phosphatase domains and the linker. RsbP is activated by a partner-protein, RsbQ, that has analogous function to RsbT for RsbU. An AlphaFold2 prediction of the heterotetrameric 2RsbP/2RsbQ complex places the two molecules of RsbQ in a hybrid interface between the PAS domains and linkers, similar to how RsbT contacts RsbU. These predictions suggest a testable hypothesis that RsbP is controlled through an activation mechanism similar to that of RsbU (Figure 5A).

To further probe the generality of the linker-mediated phosphatase dimer we observe for RsbU, we generated AlphaFold2 predictions for dimers of other documented GSR initiating phosphatases (C. difficile RsbZ [N-terminal tandem GAF domains; Figure 5B], S. coelicolor OsaC [N-terminal GAF and PAS domains], and M. tuberculosis Rv1364c [N-terminal PAS domain, C-terminally fused to an anti-σ factor domain and an anti-anti-σ factor domain]), as well as from examples of common domain architectures of PPM phosphatases selected from InterPro (response receiver [Streptomyces coelicolor, MY40396.1], GAF [Rhodococcus aetherivorans, WP_029544106.1], and HAMP [Synechocystis sp., WP_010874143.1] domains; Figure 5—figure supplement 1). In each case, AlphaFold2 produced high-confidence predictions of phosphatase domains dimerized through a shared α0, α1, α2 interface, connected to dimeric N-terminal domains by crossing α-helical linkers (Figure 1—figure supplements 1–2, Figure 5—figure supplement 1). Consistent with a model in which the stability of the linker plays a conserved regulatory role, the AlphaFold2 models for many of the predicted structures have unfavorable polar residues buried in the coiled-coil interface (positions a and d, for which non-polar residues are most favorable; Figure 5—figure supplement 2). From this analysis, we speculate that linker-mediated phosphatase domain dimerization is an evolutionarily conserved, adaptable mechanism to control PPM phosphatase activity. This mechanistic/structural congruence provides a framework for how distantly related phosphatases can be regulated to control diverse biological processes in bacteria (Figure 5E).

Using unbiased genetic screens, biochemical reconstitution, biophysical approaches, and structural prediction, we have discovered that RsbT activates RsbU by rigidifying an otherwise flexible linker region to allosterically activate RsbU as a phosphatase. Our structural prediction, supported by our genetic and structural data suggest that RsbU linkers assume a crossed α-helical conformation that dimerizes the phosphatase domains (Figure 4G). This dimerization activates RsbU through a conserved switch element at the dimer interface, decreasing the K_M for metal cofactor and increasing the catalytic rate. Our structural predictions suggest that this is a widespread mechanism to regulate the activity of bacterial protein serine/threonine phosphatases. While RsbU activation is controlled by a flexible-to-rigid transformation of the linker, it is possible that other conformational changes of the linker could be alter the PPM dimer to control phosphatase activity through the conserved switch element.

Escherichia coli and B. subtilis are the most extensively studied models for GSR activation (Battesti et al., 2011; Hecker et al., 2007). However, key mechanistic features of the activation mechanisms differ between the two systems. The E. coli GSR is initiated by stabilizing the alternative sigma factor σ^S (RpoS) against degradation by ClpXP (Figure 5C). Under non-stressed conditions, degradation of σ^S is mediated by the adapter protein, RssB, which is inhibited by anti-adaptors (IraD, IraM, and IraP) and stimulated by phosphorylation on its N-terminal response-receiver domain during stress. Historically, the differences between these systems were interpreted to mean that they operate by distinct mechanistic principles. However, RssB contains a C-terminally truncated (and therefore inactive) PPM phosphatase domain, suggesting some aspects of its regulatory mechanism could be shared with RsbU (Battesti et al., 2013).

Our findings reveal that E. coli RssB and B. subtilis RsbU use three shared regulatory features to control GSR activation through their distinct downstream mechanisms. First, the positioning of analogous flexible linkers controls GSR activation by RssB and RsbU (Brugger et al., 2023a; Brugger et al., 2023b; Dorich et al., 2019). The C-terminal pseudo-phosphatase domain of RssB is joined to its N-terminal σ^S binding response-receiver domain by a predominantly α-helical linker (termed the segmented helical linker, or SHL) of similar length to the linker of RsbU (Dorich et al., 2019). The SHL folds to occlude the σ^S interacting face of the response-receiver domain when bound to anti-adapters, and is remodeled to promote σ^S interaction when the response-receiver domain is phosphorylated (Brugger et al., 2023a; Brugger et al., 2023b; Dorich et al., 2019). Supporting the hypothesis that the SHL is mechanistically related to the RsbU linker, substitution of W143, the homologous residue to Q94 of B. subtilis RsbU, blocks regulation by IraD and IraP (Battesti et al., 2013). Second, the interface between the linker and α1 of the phosphatase domain is critical for regulation of both RssB and RsbU (Brugger et al., 2023a; Dorich et al., 2019). W143 packs in a hydrophobic pocket formed with the α1 helix of the pseudo-phosphatase domain of phosphorylated RssB and IraD-bound RssB (Brugger et al., 2023a; Dorich et al., 2019). Substitution of residues in α1, including A216 (corresponding to M166 of RsbU), block regulation of RssB by anti-adapters, underscoring the mechanistic similarity (Battesti et al., 2013). Third, conformational change of the switch helices (α1 and α2) of the phosphatase domain is of shared regulatory importance (Battesti et al., 2013). HDX experiments demonstrate that there is decreased exchange of the α1 helix of the pseudo-phosphatase domain of RssB when σ^S is bound, suggesting that the conformational dynamics of this region are reduced by substrate recognition (Brugger et al., 2023a).

These striking mechanistic similarities between RssB from E. coli and RsbU from B. subtilis suggest that key regulatory features were conserved when an ancestral phosphatase lost enzymatic activity and became a dedicated adapter protein in E. coli and related γ-proteobacteria. Interestingly, P. aeruginosa has a homologue of RssB/RsbU that functions as an adapter for σ^S degradation, yet has a catalytically active, full-length phosphatase domain and resides in an operon with an rsbV homologue (rssC; Rodríguez-Martínez et al., 2023). Crystal structures of P. aeruginosa RssB revealed that it forms a dimer with crossing linkers and dimerized phosphatase domains with striking similarity to the dimer predicted for RsbU (Figure 5, Figure 5—figure supplement 2C). Intriguingly, adapter protein function of P. aeruginosa RssB requires RssC (Rodríguez-Martínez et al., 2023), and E. coli RssB has been observed to weakly dimerize (Dorich et al., 2019), further supporting a direct mechanistic link between phosphatase regulation and adapter protein function. While not all GSR regulatory mechanisms involve a phosphatase or adapter protein related to RsbU/RssB, these findings suggest that linker-mediated control of a PPM phosphatase is a broadly conserved mechanism for GSR initiation across bacterial phyla.

Histidine kinases and GGDEF diguanylate cyclases are the two most prevalent families of bacterial signal transduction proteins. A striking feature of their regulatory mechanisms is that they are often active in dimeric complexes that are allosterically controlled by N-terminal regulatory domains that propagate signals through crossing α-helical linkers (Zschiedrich et al., 2016; Randall et al., 2022; Gourinchas et al., 2017). Our findings that the crossing α-helical linkers are structurally similar (Fig. S2C) and that shared regulatory domains (response receiver, HAMP, and PAS domains) connect phosphatase dimers by a coiled-coil linker (Figure 5A-C, Figure 5—figure supplement 1) suggests that this shared allosteric regulatory architecture allows modular exchange of N-terminal domains between PPM phosphatases, histidine kinases, and GGDEF diguanylate cyclases (Figure 5D). In support of this, the RsbU N-terminal domain and RsbT were previously shown to control GGDEF diguanylate cyclase activity in Moorella thermoacetica (Quin et al., 2012), histidine kinases with RsbU N-terminal domains are found in operons with rsbT and stressosome genes in some Proteobacteria species (Heinz et al., 2022), and AlphaFold2 prediction of these dimeric complexes reveals coiled-coil linkers to the effector domains (Figure 5E–F). Similarly, the PAS/α/β hydrolase module that controls RsbP phosphatase activity in B. subtilis is found coupled to histidine kinases and GGDEF diguanylate cyclases across bacterial phyla (Nadezhdin et al., 2011).

While the specific mechanism discovered here for linker control of RsbU activity (flexibility controlled by a binding partner) is distinct from the models for linker control of histidine kinases and GGDEF diguanylate cyclases (piston shift, scissoring, and helical rotation; Zschiedrich et al., 2016; Randall et al., 2022), the shared dimeric architecture suggests that N-terminal regulatory domains could be modularly exchanged across effector domain families (Figure 5G). Allosteric control of PPM switch element conformation could theoretically be achieved by piston shift, scissoring, or helical rotation. Future characterization of PPM phosphatases will be required to determine how widespread linker flexibility is as a regulatory mechanism and how modularly sensory domains can be exchanged across signaling effector families.

All proteins were expressed in E. coli BL21 (DE3) cells grown at 37 °C to an OD₆₀₀ of 0.4 and induced at 16 °C for 14–18 hr with 1 mM isopropyl β-d-1-thiogalactopyranoside (IPTG) unless otherwise specified. Cells were harvested and purified as follows:

Small Angle X-ray Scattering (SAXS) was performed at BioCAT (beamline 18ID at the Advanced Photon Source, Chicago) with in-line size exclusion chromatography (SEC) to separate sample from aggregates and other contaminants and ensure optimal sample quality. In-line multiangle light scattering (MALS), dynamic light scattering (DLS) and differential refractive index (dRI) measurements were recorded for additional biophysical characterization (SEC-MALS-SAXS). The samples were loaded on a Superdex 200 10/300 Increase GL Column (Cytiva) run by a 1260 Infinity II HPLC (Agilent Technologies) at 0.6 mL/min. The flow passed in order through an Agilent UV detector, a MALS detector and a DLS detector (DAWN Helios II, Wyatt Technologies), and an RI detector (Optilab T-rEX, Wyatt) before the SAXS flow cell. The flow cell consists of a 1.0 mm ID quartz capillary with ~20 µm walls. A coflowing buffer sheath is used to separate sample from the capillary walls, helping prevent radiation damage (Kirby et al., 2016). Scattering intensity was recorded using a Pilatus3 X1 M (Dectris) detector that was placed 3.682 m from the sample giving us access to a q-range of 0.0027 Å^–1 to 0.33 Å^–1. 0.5 s exposures were acquired every 1 s during elution and data was reduced using BioXTAS RAW 2.1.4 (Hopkins et al., 2017; Hopkins 2024). Buffer blanks were created by averaging regions flanking the elution peak and subtracted from all measured profiles. Because of the heterogeneity in the samples, EFA was used to deconvolve the scattering profiles of the overlapping components (Meisburger et al., 2016). These scattering profiles were used for subsequent analysis in RAW. The quality of the deconvolution was assessed based on the mean weighted χ² for each deconvoluted component and the χ² of the combined fit to each frame of the dataset (Figure 4—figure supplement 2B–E, Figure 4—figure supplement 4B–E). RAW was used to run GNOM from the ATSAS package (version 3.2.1) (Svergun, 1992; Manalastas-Cantos et al., 2021). Description of each SAXS sample and results of the analysis are presented in (Table 1; Trewhella et al., 2017). The composition of each scattering component was determined based on the SAXS derived molecular weights, the MALS profile, and gels from the purification of the protein or protein complex (Figure 4—figure supplement 1; Table 1). Molecular weights and hydrodynamic radii were calculated from the MALS and DLS data respectively using the ASTRA 7 software (Wyatt). All samples were run in 20 mM HEPES pH 7.5, 100 mM NaCl, 5 mM DTT. Wild-type RsbU was run at a concentration of 3.5 mg/mL, RsbT/RsbU^Q94L was run at a concentration of 2.5 mg/mL, RsbU^Q94L was run at a concentration of 3.5 mg/mL, RsbT/RsbU was run at a concentration of 2.5 mg/mL. FoXS was used to calculated predicted scattering profiles from the model, and was compared to the experimental scattering profiles (Schneidman-Duhovny et al., 2016; Schneidman-Duhovny et al., 2013). Output parameters calculated using FoXS are provided in Table 1. MultiFoXS was used to generate a flexible structure ensemble with multicomponent fits and was compared to the experimental scattering profiles (Schneidman-Duhovny et al., 2016). MultiFoXS input and output parameters are provided in Table 1.

We site-specifically labeled RsbT with tetramethylrhodamine 5-maleimide (TMR) dye at its unique cysteine (position 6). We added increasing concentrations of RsbU and RsbU^Q94L to 200 nM RsbT^TMR and measured the change in anisotropy (Figure 2A). All measurements were done at 25 °C in 50 mM K•HEPES pH 7.5, 100 mM NaCl, and 10% glycerol. All measurements of spectra were conducted with a slit width of 7 nm, and anisotropy measurements were conducted with a slit width of 7 nm. The excitation wavelength was 551 nm and the emission wavelength was 577 nm. Anisotropy measurements were taken 30 s apart for six measurements, and then were averaged and plot against its corresponding concentration of RsbU. Anisotropy data was then fit to a quadratic binding curve using KaleidaGraph, reported errors are the error of the fit.

Phosphatase assays were performed with RsbV-P that was labeled with ³²P by incubating RsbV (40 µM), 6His-RsbW (45 µM), and 100 µCi of γ-³²P ATP overnight at room temperature in 50 mM K•HEPES pH 7.5, 50 mM KCl, 10 mM MgCl₂, and 2 mM DTT. Unincorporated nucleotide was removed by buffer exchange using a Zeba spin column equilibrated in 50 mM K•HEPES pH 8.0, 100 mM NaCl. 6H-RsbW was then removed by Ni-NTA resin equilibrated in 50 mM K•HEPES pH 8.0, 100 mM NaCl, 20 mM imidazole. The flow-through fraction from the Ni-NTA resin containing RsbV-³²P was then exchanged into 50 mM K•HEPES pH 8.0, 100 mM NaCl buffer using 3 subsequent Zeba spin columns to remove all unincorporated nucleotide and free phosphate. Labeled RsbV-³²P was aliquoted and frozen at –80 °C for future use.

All phosphatase assays were performed at room temperature in 50 mM K•HEPES pH 7.5, 100 mM NaCl. The concentrations of enzyme, substrate, MnCl₂, and MgCl₂ were varied as indicated. 10 µM RsbT was additionally added to reactions as indicated. Reactions were stopped with 0.5 M EDTA, pH 8.0 and run on PEI-Cellulose TLC plates developed in 1 M LiCl₂ and 0.8 M acetic acid and imaged on a Typhoon scanner. Phosphatase assays were performed more than three independent times as separate experiments. Multiple turnover reactions were performed with trace RsbV-³²P and an excess of cold RsbV-P. Number of turnovers were calculated for each timepoint from (1 – fraction substrate)*[substrate]/[enzyme]. Data shown in figures is from a single representative experiment, and reported errors are the error from the fit.

B. subtilis cells were grown in Lennox lysogeny broth (LB, Sigma-Aldrich) as liquid medium or as plates supplemented with 15% Bacto Agar (Difco). Antibiotics were used as appropriate for maintenance of plasmids or for selection of transformants: MLS (50 µg/mL erythromycin, 250 µg/mL lincomycin), kanamycin (10 µg/mL). For plates used to visualize σ^B reporter activity, 80 µg/mL of 5-bromo-4-chloro-3-indolyl-β-D-galactopyranoside (X-gal) and 1 mM IPTG were added. All B. subtilis strains were derived from the PY79 strain background and are listed in the table of strains. Plasmids based on pHB201 (Bron et al., 1998) were introduced to B. subtilis strains using natural competence (Harwood and Cutting, 1990). All B. subtilis strains used in this study are listed in Table 2, E. coli strains are listed in Table 3.

Genetic screens were performed as described previously (Ho and Bradshaw, 2021). Plasmids containing rsbU or rsbT/rsbU^Y28I were subjected to PCR mutagenesis using GoTaq DNA polymerase mix (Promega) supplemented with 2 mM additional MgCl₂. Pools of mutagenized inserts were assembled into digested plasmid using isothermal (Gibson) assembly and transformed into E. coli DH5α cells. Mutation rate of approximately one single-nucleotide polymorphism per kilobase of DNA was confirmed by sequencing individual clones. Plasmids were then introduced to B. subtilis using natural competence and cells were plated on selective medium containing IPTG and X-gal for two days at 37 °C. Blue colonies were selected, restruck on plates with and without IPTG, and plasmids recovered from single colonies that retested as σ^B positive were sequenced using Sanger sequencing. For the RsbU^Y28I suppressor screen, two independent pools of PCR mutagenized plasmid were screened. In addition to the reported variants that were selected, we identified simple revertants that restored Y28 in both pools, and mutations in the promoter and RBS that we predict enhanced protein production. These variants were not studied further. Individual amino acid changing mutations were rebuilt in the parental plasmid using PCR (QuickChange) mutagenesis and the phenotypes were validated by replating on indicator plates. All strains shown in the figures are from these rebuilt plasmids.

AlphaFold2 predictions were performed using ColabFold (Mirdita et al., 2022). Alphafold2_multimer_v2 was used in unpaired_paired mode with no templates with 3 recycles, 200 iterations, and greedy pairing strategy. The predicted aligned error plots for all AlphaFold2 structures are shown in Figure 1—figure supplement 1. The predicted local distance difference test scores (pLDDT) mapped onto the structures are shown in Figure 1—figure supplement 2.

Socket2 coiled-coil predictions were performed using Socket2 Colab (Kumar and Woolfson, 2021). Socket2 was used with a packing cutoff of 7.0 Å. Analysis done using Socket2 is shown in Figure 5—figure supplement 2.

All strains, plasmids, and coordinate files reported in this manuscript are available by request to the corresponding author.