Conformation control of the histidine kinase BceS of Bacillus subtilis by its cognate ABC-transporter facilitates need-based activation of antibiotic resistance

Bacteria closely control gene expression to ensure optimal physiological responses to their environment. Such careful gene expression can minimize the fitness cost associated with antibiotic resistance. We previously described a novel regulatory logic in Bacillus subtilis enabling the cell to directly monitor its need for detoxification. This cost-effective strategy is achieved via a two-component regulatory system (BceRS) working in a sensory complex with an ABC-transporter (BceAB), together acting as a flux-sensor where signaling is proportional to transport activity. How this is realized at the molecular level has remained unknown. Using experimentation and computation we here show that the histidine kinase is activated by piston-like displacements in the membrane, which are converted to helical rotations in the catalytic core via an intervening HAMP-like domain. Intriguingly, the transporter was not only required for kinase activation, but also to actively maintain the kinase in its inactive state in the absence of antibiotics. Such coupling of kinase activity to that of the transporter ensures the complete control required for transport flux-dependent signaling. Moreover, we show that the transporter likely conserves energy by signaling with sub-maximal sensitivity. These results provide the first mechanistic insights into transport flux-dependent signaling, a unique strategy for energy-efficient decision making.

enabling the cell to directly monitor its need for detoxification. This cost-effective strategy is 23 achieved via a two-component regulatory system (BceRS) working in a sensory complex 24 with an ABC-transporter (BceAB), together acting as a flux-sensor where signaling is 25 proportional to transport activity. How this is realized at the molecular level has remained 26 unknown. Using experimentation and computation we here show that the histidine kinase is 27 activated by piston-like displacements in the membrane, which are converted to helical 28 rotations in the catalytic core via an intervening HAMP-like domain. Intriguingly, the 29 transporter was not only required for kinase activation, but also to actively maintain the 30 kinase in its inactive state in the absence of antibiotics. Such coupling of kinase activity to 31 that of the transporter ensures the complete control required for transport flux-dependent 32 signaling. Moreover, we show that the transporter likely conserves energy by signaling with 33 sub-maximal sensitivity. These results provide the first mechanistic insights into transport 34 flux-dependent signaling, a unique strategy for energy-efficient decision making. regulated, which requires that the bacterium is able to accurately determine the severity of an 43 antibiotic attack and adjust gene expression accordingly. We previously described a new 44 sensory strategy in Bacillus subtilis that produces an exquisitely fine-tuned response, with 45 three orders of magnitude of output modulation over an input dynamic range of three orders 46 of magnitude (Fritz et al., 2015). This regulatory logic is achieved by a two-component 47 system (BceRS) whose histidine kinase, BceS, works in a sensory complex with an ABC- provides resistance against the peptide antibiotic bacitracin (Ohki et al., 2003). A defining 51 characteristic of this system is that the histidine kinase lacks any apparent sensory domains 52 and instead requires the transporter for activation in response to bacitracin (Rietkötter et al.,53 2008; Hiron et al., 2011).  The BceB permease is in bright green with the 8th transmembrane helix marked in dark green and the extracellular domain indicated by a green crescent; two molecules of the ATPase BceA are shown in pale green associated with the BceB monomer, according to the A2B stoichiometry (Dintner et al., 2014). The left section of the schematic shows the system in its inactive (OFF) state, and the repressing function of the transporter indicated by a red blunt arrow. The right shows the active (ON) state in the presence of antibiotics, exemplified by bacitracin, with the activating function of the transporter shown as a red lightning bolt. The motions of kinase domains upon activation are shown as red block arrows and BceS autophosphorylation by a 'P'. The positioning of the symbols within the sensory complex for repression/activation is arbitrary. ATP hydrolysis reactions and antibiotic removal are indicated by grey arrows; for simplicity ATP hydrolysis is only shown for one ATPase. 2011; Revilla-Guarinos et al., 2014). Based on computational modelling of the response 60 dynamics in the B. subtilis Bce system, we have established that signaling is proportional to 61 transport activity (Fritz et al., 2015). Because the transporter mediates the actual resistance, 62 presumably by removal of the antibiotic from its target in the cell envelope (Gebhard, 2012;63 Fritz et al., 2015;Greene et al., 2018;Kobras et al., 2020), increased transporter production 64 acts a negative feedback on signaling, which creates the characteristic gradual, fine-tuned 65 response (Fritz et al., 2015). This flux-sensing mechanism effectively allows the bacterium to 66 adjust its response according to its need for additional detoxification capacity, implementing 67 a cost-effective 'produce-to-demand' strategy. However, the molecular mechanism of this 68 new signaling strategy, i.e. how the activity of the histidine kinase is controlled within the 69 sensory complex and how this creates the observed analog response behavior, remains 70

unknown. 71
Histidine kinases typically relay extracellular information to the cytoplasm via 72 autophosphorylation at a conserved histidine residue, which then leads to phosphorylation of 73 a cytoplasmic response regulator that elicits changes in gene expression (Gao and Stock, 74 2009). Despite their importance in bacterial signaling, we still have limited understanding of 75 the molecular mechanisms by which histidine kinases transmit information through the 76 cytoplasmic membrane. This is in part due to the modular architecture of the kinases, which 77 allows a plethora of different input domains to be coupled to a conserved catalytic core. The 78 core consists of a DHp (also known as HisKA) domain that facilitates dimerization and 79 histidine phosphotransfer, and a C-terminal ATPase domain (Gao and Stock, 2009 flux-sensing Bce-like systems are unique in that the histidine kinases fully rely on their 95 cognate transporters for activation, and remain inactive when the transporter is deleted 96 (Rietkötter et al., 2008;Gebhard and Mascher, 2011). Moreover, implementing the reported 97 flux-sensing strategy requires kinase activity to be directly proportional to transport activity 98 (Fritz et al., 2015). To our knowledge, no information is available on how accessory proteins 99 may achieve such positive and precise control over a histidine kinase. 100 In the present study, we sought to understand the mechanistic basis of histidine kinase 101 activation in the Bce system of Bacillus subtilis and how this links to the system's analog 102 signaling behavior. Through computational predictions, detailed mutagenesis and cysteine 103 cross-linking and accessibility analyses, we provide evidence that the kinase BceS is 104 activated via helical rotations in the DHp domain, accompanied by a piston movement in the 105 second transmembrane domain. Physical motions are most likely relayed via a poorly 106 conserved intervening HAMP-like domain. We also show that the transporter not only acts as 107 the activator of the kinase, but also controls its inactive state. We propose it is this complete 108 control over kinase conformation that couples its activity to that of the transporter. 109 Surprisingly, mutagenesis of the putative signaling domain of the transporter, BceAB, 110 highlighted that the system has not evolved for maximal signaling sensitivity, which likely 111 represents an additional layer of fine-tuning and energy conservation. of BceS using SMART (Letunic et al., 2012) showed two N-terminal transmembrane helices, 116 predicted to end at residue F55, and a prototypical cytoplasmic kinase core with a DHp 117 domain from residues E115 to I172 and ATPase domain from S216 to N326. The 60 amino 118 acid linker between the transmembrane and DHp domains contained no predicted domains. 119 To investigate the cytoplasmic composition of BceS in more detail, we used I-TASSER 120 (Zhang, 2008) to predict the tertiary structure of the linker region and DHp domain (residues 121 R56 to N174; Fig. 2a). The closest structural analog was a hybrid protein consisting of the 122 HAMP domain of Af1503 A291V from Archaeoglobus fulgidus and the DHp domain of E. coli 123 EnvZ (PDB 3ZRW; see methods for quality scores), which shares 20% sequence identity and 124 40% similarity with the BceS fragment used for structure prediction. The homology model of 125 BceS showed the typical DHp domain fold of two antiparallel helices (DHp1 and DHp2), 126 joined by a short loop. In the functional kinase dimer, this will form the four-helix bundle 127 characteristic for most histidine kinases (Gao and Stock, 2009) (Fig.2c). Interestingly, the 128 linker between the membrane interface and DHp domain adopted a HAMP-like conformation 129 with two short parallel helices (HAMP-like 1 and HAMP-like 2), where HAMP-like 2 130 continues into DHp1 as a single continuous helix (Fig. 2a). The relevance of this domain, 131 The residues that form Cys-crosslinks in panel b are highlighted, E115 (red), E118 (purple) and L119 (green). Residue K167, which sits opposite E115, is shown in orange. b, Percentage of crosslinking for each BceS Cys variant was calculated from relative band intensities between monomer and dimer within each lane using Fiji (Schindelin et al., 2012). Amino acid substitutions were introduced in parent construct bceSWT-His8 (SGB369) and BceS production was controlled by addition of 0.02 % (w/v) xylose. Data are shown as mean ± SD from 2-5 independent repeats. c, d, Schematic diagram of the BceS DHp dimer configuration. Helices labelled DHp1' and DHp2' show the second protomer. Left, top down view of the helices in the structural model; right, helical wheel diagram of the top two helical turns in the DHp domain; residues that form crosslinks are highlighted as before. physical distance from each other and the DHp domain, we did not expect these residues to 145 interfere with our analysis, and indeed no crosslinking product was observed upon oxidation 146 with iodine in wild-type BceS (Fig. 2b). Next, single cysteine substitutions were introduced 147 in the first two helical turns of the helix DHp1 (N114 to M120), and in equivalent positions 148 near the end of the helix DHp2 (Q166 to R168). Disulfide bond formation in these variants 149 should only occur if the substituted residues come into close proximity in the kinase dimer 150 (Pakula and Simon, 1992). The only residues that consistently formed detectable crosslinks 151 were at positions 115, 118 and 119, with the highest degree of crosslinking at position 115 152 (Fig. 2b). This is in good agreement with their positioning at the core of the four-helix bundle 153 and shows that the structural model closely reflects the conformation of BceS in the cell (Fig.  154   2c). 155 To ensure that the introduction of cysteines had not affected the integrity of the protein, we 156 tested the activity of each BceS variant in vivo. This was based on their ability to induce 157 expression from the system's target promoter fused to a reporter gene (PbceA-luxABCDE or 158 PbceA-lacZ), following exposure to the antibiotic bacitracin (Fritz et al., 2015). Only the 159 K116C and K167C substitutions led to a complete loss of activity (Fig. S1a). It is therefore 160 possible that the lack of crosslinking observed for the latter, which also faces the core of the 161 DHp domain (Fig. 2c), could be explained by defects in the protein rather than the distance 162 between DHp2 helices in the kinase dimer. residues at the membrane-proximal end of the four-helix bundle (Fig. 3). In the wild-type 166 kinase, E115 and K167 create a net neutral charge in the bundle core. As described above and  kinase activity could be returned to near wild-type levels by restoring the net neutral charge 181 through combining the E115K and K167D substitutions (Fig. 3a). We therefore concluded 182 that the high constitutive activity in the absence of the inducer was indeed due to the presence 183 of four residues with the same charge in the DHp domain core. DHp domain, we turned to molecular dynamics (MD) simulations. As discussed above, the 206 homology model (Fig. 2) shows excellent agreement with experimental validation, indicating 207 that it is a reasonable starting point for MD simulations. Given the relatively large putative 208 rotation, we applied Gaussian accelerated MD (GaMD) to more completely sample 209 conformational space (Miao et al., 2015). 210 To assess conformational differences in the DHp domain between the wild-type kinase and 211 the constitutively active E115K/K116E variant, GaMD simulations were performed on both 212 for a total of 2 µs each (four replicas of 500 ns; Figure S3a). From our simulations, we 213 calculated the potential of mean force (PMF) for a series of internuclear distances for both 214 wild-type and variant residues, to serve as a proxy for the energetic barrier between 215 conformational sub-states. To best illustrate the relative orientation of helices in the DHp 216 domain bundle, we focused on positions 115 and 167, facing the bundle core (Fig. 3). The 217 PMF plots for the internuclear pairs of these residues between monomeric units of the DHp 218 domain are shown in Figure 3c. The lower the PMF value, the more likely the helices will 219 occupy the conformational state defined by the internuclear distances shown. In the wild-type 220 DHp domain, the PMF plot for distances between monomers at amino acids 115 and 167 221 shows a broad minimum across most of conformational space (Fig. 3c, left). This suggest that 222 there is a relatively high propensity for the domain to explore a diversity of conformational 223 space. Figure 3c (right) shows the same data for the variant DHp, which, instead of the broad 224 minimum, now shows two distinct, more restricted minima. More specifically, the tendency 225 towards a longer K167 distance in the variant would seem consistent with our model for helix 226 rotation ( Figure 3b). To more clearly explore the differences between conformational space 227 accessed, we performed principal components analysis (PCA) of the data from Figure 3c. 228 This indicates that there is a significant difference in the principal component space sampled 229 (Fig. S3b), meaning the variant DHp occupies a conformational space that is not accessible to 230 the wild-type domain. Based on the high constitutive activity of the variant, it is plausible to 231 assume that this new conformational space represents the active state of the kinase. 232 Importantly, our simulations show that the wild-type kinase is unable to adopt this 233 conformation on the simulated time scales. This is consistent with our in vivo data that shows 234 no detectable kinase activity without activation through the transporter. instead. Additionally, the conserved Pro residue in helix 1, and Glu in helix 2, typical for 248 HAMP domains, are not present in BceS (Fig. 4a). 249 Previous investigations of HAMP domains have highlighted one particular residue with key 250 influence on the signaling state, most commonly occupied by small hydrophobic residues that 251 favour bundle packing (Hulko et al., 2006). This residue corresponds to L67 in BceS (Fig. 4a,  252 black arrow). Based on structural work on the HAMP domain of the A. fulgidus protein 253 did not find any substitutions that led to constitutive activation. The only two substitutions 273 that caused a complete loss of function were E60C and F63C ( Fig. 4 and S4). Interestingly, 274 these fall on the first two positions of the packing motif (Fig. 4a, white arrows and 4d), with 275 E60 located one packing layer above the critical L67 discussed earlier, and are therefore 276 likely of structural importance. Both variant proteins are present in the cytoplasmic substitutions. c, BceS signaling activity following L67 amino acid size substitution harbouring PbceA-luxABCDE and Pxyl-bceS. Amino acid substitutions were introduced in the parent construct bceSWT-His8 (SGB369). Signaling was induced with 1µg ml-1 of bacitracin. Data are shown as mean ± SD from 3-18 biological repeats. **, significant difference (p<0.01) in activity compared to WT construct, determined by un-paired t-test; ns, not significant; in the absence of bacitracin, no significant differences were detected. d, Helical wheel diagram of the HAMP-like domain. Individual helices are abbreviated HL-1 and HL-2; the labels HL-1' and HL-2' show the second protomer. Residues essential for BceS activity are E60 (red) and F63 (yellow); the key packing residue L67 is shown in blue; red/blue semicircles are used to indicate positioning of residues in subsequent helical layers. e, BceS signaling activity of the constitutive ON variant (BceSE115K) combined with loss-of-function substitutions at positions 60 and 63. Amino acid substitutions were introduced in parent construct bceSWT-His8 (SGB401). Signaling in cells harbouring PbceA-lacZ and Pxyl-bceS was induced with 1µg ml-1 of bacitracin, BceS production was controlled by addition of 0.02 % (w/v) xylose. Signalling is reported as -galactosidase activity in Miller units, and data are shown as mean ± SD from 3-7 biological repeats. The inset shows an anti-His Western blot analysis to confirm membrane localisation and protein levels of wild-type (WT) BceS and its variants. membrane at the same levels as the wild type protein according to Western blot analysis (Fig.  278 4e inset), showing that the lack of signaling is not caused by general protein misfolding. To 279 test if the defect of the two variants was due to a loss of signal transfer from either the 280 membrane region of the protein or the transporter, we combined each of the LOF mutations 281 with the E115K substitution that constitutively activates BceS. All resulting variants 282 remained inactive (Fig. 4e), suggesting that the substitutions in the HAMP-like domain likely 283 caused structural disruptions, rather than specifically affecting signal transfer. Unden, 2015)) of its second transmembrane helix. This analysis requires a cysteine-free 293 background, and we therefore replaced the three native Cys residues mentioned earlier by 294 serine and confirmed that he C45S/C198S/C259S BceS variant was still fully functional (Fig.  295   S5). Individual cysteine replacements were done from positions 51 to 65, surrounding the 296 predicted membrane interface at W54 or F55. The essential residues E60 and F63 were not 297 replaced. All variants were signaling competent (Fig. S5). 298 The position relative to the membrane was then tested for each Cys residue based on its 299 solvent accessibility, using a modified version of the established E. coli protocol (Monzel and 300 Unden, 2015). In brief, treatment of whole cells with N-ethyl-maleimide (NEM) should block 301 any solvent-exposed Cys residue, whereas in untreated cells the sulfhydryl groups should 302 remain available for later reactions. Following isolation of membranes, any unblocked Cys 303 residues are then reacted with methoxypolyethylene glycol maleimide (PEG-mal), resulting 304 in a 10 kDa increase in size, detectable by Western Blotting. For cytoplasmic Cys residues in 305 BceS, we therefore expected to see a shift in size only in samples that had not been treated 306 with NEM (Fig. 5, blue labels). In contrast to the established E. coli methodology (Monzel 307 and Unden, 2015), we could not observe PEGylation of membrane-embedded Cys residues in 308 B. subtilis, regardless of the presence of SDS, possibly due to differences in membrane 309 composition. We therefore interpreted those residues as being membrane embedded that did 310 not show a size shift in either presence or absence of NEM treatment (Fig. 5, red labels). 311 Our results showed that residues up to position 56 resided in the solvent-inaccessible 312 membrane environment, while residues from Y57 onwards were accessible and therefore 313 cytoplasmic (Fig. 5, 'OFF state'). We then repeated the same analysis in the background of 314 the E115K/K116E substitutions that place the kinase in a constitutively active conformation. 315 Here, residues from F55 onwards became consistently solvent-accessible (Fig. 5, 'ON state'), 316 showing that BceS activation involved an inward piston-like displacement by two amino 317 acids of the second transmembrane helix. An interesting observation was residue Y64, which 318 labelled as inaccessible in the OFF state, despite being surrounded by solvent-exposed 319 positions. It is likely that this residue is involved in protein-protein interactions, e.g. within 320 BceS or potentially with BceB, that in the inactive state prevent its labelling through steric 321 effects, as has been proposed before (Monzel and Unden, 2015). 322 We next sought to test whether BceS could be activated by artificially inducing a shift 323 through introduction of positively charged arginine residues near the membrane interface. 324 This approach has been shown to cause signal-independent activation of the E. coli histidine kinase DcuS (Monzel and Unden, 2015). We therefore substituted three of the residues near 326 the cytoplasmic end of transmembrane helix 2 of BceS (W54, F55 and Y57) with Arg, 327 individually and in pairs. (Fig. S6). Variants containing the W54R substitution had 328 approximately 60% higher activities upon stimulation with bacitracin (Fig. S6)  In cells harbouring the wild-type sequence of BceS, the OFF-state activity was below the 346 detection limit of our assays, and deletion of the transporter therefore had no measurable 347 effect (Fig. 6). However, in the E115K or the E115K/K116E variants, basal kinase activity 348 was high, allowing us to test the effect of bceAB deletion in the absence of bacitracin. Signaling activity of BceS was measured in strains harbouring PbceA-luxABCDE and challenged with 10µg ml-1 of bacitracin (grey) or left untreated (black). Amino acid substitutions were introduced into parent construct Pxyl-bceS-His8, in a strain background carrying a ΔbceS deletion (BceAB +; SGB792), or a ΔbceSAB deletion (BceAB -; SGB818). Data are shown as mean ± SD from 5-12 biological repeats. Results from unpaired t-test are shown by brackets across the strains and conditions compared, analyzing basal and induced activities separately. ****, p<0.0001; ***, p <0.001; ns, p>0.05. Differences with versus without bacitracin were significant (p<0.005) in all strains.

Removing the transporter in this background caused a striking further increase in activity, 350
rendering the kinase fully activated irrespective of the presence of the inducing antibiotic 351 (Fig. 6). Therefore, the inactive transporter can inhibit the activity of the constitutively active 352 DHp domain variants approximately ten-fold. Our data are consistent with a model where 353 inactive BceAB exerts close control over BceS keeping the kinase in its inactive state (Fig. 1,  354 left). This reveals a further layer of kinase control in addition to the transporter's known 355 activating effect, e.g. preventing spurious signaling in the absence of antibiotics.  Signaling assays showed that all variants except G543C reached a similar amplitude of 374 reporter induction as strains containing the wild-type construct (Fig. S7). However, we 375 observed some clear difference in the sensitivity of signaling, i.e. the minimal concentration 376 of antibiotic required to elicit a response. In cells with the wild-type construct, the lowest 377 bacitracin concentration to trigger signaling was 3 µg ml -1 , while a number of variants had a 378 higher threshold of 10 µg ml -1 (Fig. 7, green vs pink dots; Fig. S7). The majority of these fell 379 into the C-terminal part of the scanned region, indicating potential defects in coupling to the 380 ATPase. Surprisingly, we found several variants that showed a marked increase in signaling 381 sensitivity and responded already at 0.3 µg ml -1 or even 0.03 µg ml -1 bacitracin (Fig. 7, dark 382 blue dots; Fig. S7). We had previously seen that a reduction of BceAB transport activity can 383 cause such a shift in signaling threshold, because the negative feedback created by removal of 384 the antibiotic is missing (Fritz et al., 2015). However, as none of these variants showed any 385 change in resistance (Table S2), loss of antibiotic removal cannot explain the increased 386 signaling sensitivity. 387 Residue mapping showed that they occurred with a regular periodicity of every four amino 388 acids (Fig. 7b), suggesting positions along the same face of the transmembrane helix. Figure  389 7c shows a homology model from I-TASSER of the final three transmembrane helices of 390 BceB for illustration. Substitutions with the relatively small amino acid Cys may have 391 enabled tighter helix packing, potentially contributing to more effective signaling. To test 392 this, we replaced two of the positions, L546 and Q550, with a series of other amino acids, 393 including small Ala; charged Glu or Asp; and large Phe. Regardless of the substitution, 394 deviation from the wild-type sequence at these positions consistently led to an increase in 395 signaling sensitivity (Fig. 7b). It is difficult to envisage that all substitutions caused tighter 396 protein interactions. An alternative explanation may be that they are involved in coupling 397 extracellular ligand binding to ATP hydrolysis, and that weakened coupling through the substitutions allowed signaling to occur at lower concentrations. Despite numerous genetic 399 and biochemical attempts, the precise residues with which BceS and BceB interact remain 400 unknown, likely due to an extensive interface between the proteins involving multiple 401

contacts. 402
While it was beyond the scope of this study to pursue a further functional characterization of 403 the transporter, our findings clearly show that BceB contains a series of residues that prevent 404 signaling from occurring at lower bacitracin concentrations. Given that any replacement we 405 tested led to more sensitive signaling, we therefore concluded that evolution of the 406 transporter has actively avoided such hyper-sensitive signaling, constituting a further layer in 407 minimizing the fitness cost of resistance. Under laboratory conditions, we have not yet been 408 able to detect a competitive disadvantage from harbouring hyper-sensitive variants of 409 BceAB. However, fast growth in carefully controlled conditions is not representative of the 410 complex microbial interactions in natural habitats, where energy-conserving regulatory 411 strategies will likely contribute to the competitive success of a microorganism. 412

DISCUSSION 413
In this study, we set out to investigate the molecular mechanisms of the flux-sensing strategy 414 employed by Bce-like antibiotic resistance systems. Their hallmark feature is that histidine 415 kinase activity is directly proportional to transport activity across a large input dynamic range 416 of antibiotic concentrations (Fritz et al., 2015). This careful adaptation of signaling ensures 417 the resistance transporter is produced precisely to the cell's demand for detoxification 418 capacity, a very cost-efficient strategy to control gene expression. The results presented here 419 now provide molecular level understanding of how such close control over kinase activity 420 can be achieved. The unusual behavior of residue Y64 in the Scan-SCAM analyses may provide a first 457 glimpse of such, but this is not currently resolved by our experimental data. 458 To complete signaling and kinase activation, the transmembrane motions will have to be 459 translated into changes within the conserved kinase core to cause autophosphorylation, the predicted coiled-coil motif. BceS therefore appears to be the more typical subtype of this 478 group of kinases than NsaS. Interestingly however, akin to the alternative helix packing 479 mechanism proposed for HAMP domains (Hulko et al., 2006), activation of NsaS was also 480 proposed to involve switching between helical packing states of the coiled-coil (Bhate,481 Molnar, Goulian, and DeGrado, 2015b). The molecular principles of activation may therefore 482 be conserved between both subtypes of BceS-like kinases. 483 To further investigate the activation mechanism of BceS, we sought to force each of the 484 conformational changes via amino acid substitutions. Charge repulsion at the core of the DHp 485 domain to trigger helical rotation consistently led to kinase activation without a physiological 486 stimulus (Fig. 3). However, we were unable to artificially trigger activation through 487 substitutions that should have induced piston displacements in the membrane (Fig. S6) or 488 induced the active conformation of the HAMP domain ( Fig. 4 and S4). This is in contrast to 489 studies of other kinases where such substitutions least partially raises the basal activity 490 and it might be speculated that this might be relatively easy to mimic through amino acid 493 substitutions. In contrast, BceS-like proteins are subject to control by the transporter. It is 494 conceivable that in such a kinase small changes in individual domains are insufficient for 495 artificial activation, unless the DHp domain is directly manipulated. This may be supported 496 by the MD simulations, which showed that the wild-type DHp domain does not by itself 497 sample the conformational space observed in the active variant (Fig. S3), implying the 498 transporter may be required to provide the necessary force for transition to the active state. 499 Transporters are commonly used for accessory signaling control, but BceAB is unusual in 500 that it is an activator of signaling, while most others act as repressors (Piepenbreier et al., 501 2017). We previously showed that transporter and kinase constitutively form a signaling 502 complex in the membrane, irrespective of the inducer, bacitracin (Dintner et al., 2014). To 503 activate the kinase proportionally to transport activity, the transporter must therefore be able 504 to directly influence kinase activity within this complex. When we first tested this in vitro, we 505 noted that BceAB reduced kinase autophosphorylation in the OFF state, but could not explain 506 the observation at the time (Dintner et al., 2014). Here, we now showed that the transporter 507 indeed not only controls the ON but also the OFF state of BceS (Fig. 6). We propose that 508 through controlling both states of the kinase, the transporter is able to exert the close control 509 needed to explain how signaling is maintained in proportion to transport activity over more 510 than three orders of magnitude in antibiotic concentrations (Fritz et al., 2015). A crucial 511 feature of the flux-sensing mechanism is an immediate reduction of signaling if transport 512 activity is low, either because the antibiotic threat is over or because of sufficient 513 detoxification capacity (Fritz et al., 2015). The active switching off of BceS by BceAB 514 described here explains how this is realized on the molecular level. Complete control also 515 allows the transporter to actively prevent spurious signaling by the two-component system. 516 Finally, our mutagenesis analysis of BceAB revealed that evolution of the transporter itself 517 appears to have actively avoided signaling at very low antibiotic concentrations, highlighting 518 yet another level of cost reduction. 519 Taken together, we here present a molecular level model for activation of antibiotic 520 resistance based directly on the cell's need for protection, using a flux-sensing signaling 521 strategy that employs multiple layers of active energy conservation to minimize fitness costs 522 of resistance. The system is centered on two-component signaling, one of the most important 523 regulatory features of bacteria. Our findings should therefore have wide relevance and serve 524 as a model for understanding how fine-tuned gene expression in bacteria can achieve an 525 optimal cost-benefit balance. 526

EXPERIMENTAL PROCEDURES 527
Bacterial strains and growth conditions. All strains used in this study are listed in table S1 528 and were routinely grown in Lysogeny Broth (LB; 10 g l -1 tryptone, 5 g l -1 yeast extract, 5 g l -529 1 NaCl) at 37 ºC with aeration. Solid media contained 15 g l -1 agar. Selective media contained 530 chloramphenicol (5 µg ml -1 ), macrolide-lincosamide-streptogramin B (erythromycin; 1 µg 531 ml -1 , lincomycin; 25 µg ml -1 ), spectinomycin (100 µg ml -1 ), kanamycin (10 µg ml -1 ), or 532 ampicillin (100 µg ml -1 ). 533 DNA manipulation and strain construction. All primers and plasmids used in this study 534 are given in table S1. Site-directed mutagenesis was carried out according to the 535 manufacturer's instructions for the QuikChange II site-directed mutagenesis kit (Agilent 536 Technologies), or by PCR-overlap extension (Ho et al., 1989). Bacillus subtilis 537 transformations were carried out via natural competence as previously described (Harwood 538 and Cutting, 1990). DNA for transformation was either provided as linearised plasmid DNA, 539 or as genomic DNA from a B. subtilis strain already carrying the desired genetic construct. 540 To allow manipulation of both bceS and bceAB sequences, an unmarked deletion strain 541 lacking all three genes (DbceSAB, SGB377) was constructed as described previously (Arnaud 542 et al., 2004), employing plasmid pAK102. Strains lacking bceS only (DbceS; TMB1036) or 543 bceAB only (bceAB::kan; TMB035 (Rietkötter et al., 2008)) were available from previous 544 work. Signaling activity was assessed in strains harbouring a chromosomally integrated 545 transcriptional reporter construct of the target promoter PbceA, fused to either luxABCDE 546 (pSDlux101 (Kallenberg et al., 2013)) or lacZ (pER603 (Rietkötter et al., 2008)). Details of 547 the assays used are given below. reporter was employed. The bceS complementation construct was made cysteine-free via 611 C45S/C198S/C259S substitutions, followed by introduction of single cysteines to be tested 612 (strain SGB936 and derivatives). The same was done in the E115K/K116E variant to assess 613 transmembrane topology in the active state (strain SGB951 and derivatives). Cultures were 614 grown to mid-exponential phase as described above. Cells were harvested, washed twice with 615 50 mM HEPES buffer (pH 6.8) and resuspended in 5 ml of the same buffer. Water exposed 616 (i.e. cytoplasmic) cysteine residues were masked with 40 mM N-Ethylmaleimide (NEM) for 617 1.5 h at room temperature with shaking or left untreated. Masking reactions were stopped by 618 three washes in cold (4 ˚C) 50 mM HEPES buffer (pH 6.8). Cells were resuspended in 5 ml 619 of ice cold lysis buffer, sonicated and membranes collected as above. Membranes were 620 resuspended in 40 µl accessibility buffer (50 mM HEPES [pH 6.8], 1.3% (w/v) SDS). 621 Cysteines that had not been masked with NEM in the first step were labelled with 5 mM Ltd, 1:4000 dilution) was carried out in milk-PBST for 1h at room temperature, followed by 639 three final washes in PBST. Detection of proteins was by chemiluminescence using Pierce TM 640 optimised using SCWRL4 (Krivov et al., 2009). After MD equilibration of the dimer model 665 (see below), no Ramachandran outliers were present in the loop connecting the DHp domain 666 helices or elsewhere (only Gly262, located in a flexible loop in the ATPase domain, was just 667 outside the 'allowed region'). The N-terminus of the DHp domain of each monomer was 668 acetylated to cap the α-helix and prevent having a positive charge at the termini. ATP was 669 then docked into each ATP binding site using AutoDock Vina (exhaustiveness set to 32) 670 (Trott and Olson, 2010). The generated model of BceS was then manually modified to 671 construct the E115K/K116E variant. All residues were modelled in their standard protonation 672 states, with His108 and His124 singly protonated on the Nδ1 nitrogen, and all other His 673 singly protonated on the Nε2 nitrogen (as indicated by hydrogen bonding networks). Both 674 systems were solvated in an octahedral water box such that no protein/substrate atom was 675 within 10 Å of the box boundaries, with Na + or Cl − counter ions added as necessary to ensure 676 an overall neutral charge. MD simulations were performed using GPU accelerated Amber16, 677 with the ff14SB force field and TIP3P water model used to describe protein and water 678 molecules respectively. Parameters for ATP were taken from a study by Meagher and 679 colleagues (Meagher et al., 2003). Following a minimisation, heating, NVT and NPT 680 equilibration procedure (described below), Gaussian accelerated MD (GaMD) simulations 681 were done in the NVT ensemble at 300 K (Miao et al., 2015), with a boost potential applied to 682 both the total system potential energy as well as the dihedral potential energies (see also 683 GaMD settings in Table S3). To prepare for production GaMD simulations, a 12 ns long 684 conventional MD run was used to obtain maximum, minimum, average and standard 685 deviation of the system potential energies. Following this, a 66 ns long run was used to 686 update the above parameters (updated every 600 ps) to ensure converged GaMD production 687 parameters. Production GaMD simulations were then run for 4 ´ 500 ns, using different 688 random seeds for each replica. MD simulations were run using a 2 fs time step with the 689 SHAKE algorithm applied to any bond containing a hydrogen atom. An 8 Å direct space 690 non-bonded cut-off was applied with long-range electrostatics evaluated using the particle 691 mesh Ewald algorithm (Darden et al., 1993). Temperature was regulated using Langevin 692 temperature control (collision frequency of 1 ps −1 ). All MD and GaMD simulations were 693 performed with a 10 kcal mol -1 restraint on the Phi and Psi dihedrals between the acetyl and 694 distance measurements (to obtain PMFs) was performed using the 'PyReweighting' toolkit 699 (Miao et al., 2014).

700
To equilibrate structures for GaMD simulations, the following procedure was performed. 701 Minimisation of all hydrogen atoms and solvent molecules was done using 500 steps of 702 steepest descent followed by 500 steps of conjugate gradient. To keep all other atoms (i.e. the 703 protein heavy atoms) in place during the minimisation, 10 kcal mol −1 Å −1 positional restraints 704 were applied. Retaining the positional restraints on all protein heavy atoms, the system was 705 then heated rapidly from 50 K to 300 K in the NVT ensemble over the course of 200 ps. This 706 system was again minimised for a further 500 steps of steepest descent followed by 500 steps 707 of conjugate gradient, this time only applying positional restraints (of size 5 kcal mol −1 Å −1 ) 708 to the Cα carbon atoms. These Cα restraints were retained as the system was again heated 709 from 25 K to 300 K over the course of 50 ps in the NVT ensemble. Simulations were then 710 performed in the NPT ensemble (1 atm, 300 K), first gradually reducing the 5 kcal mol −1 Å −1 711 Cα carbon restraints over the course of 50 ps of simulation time. This was done in five steps 712 (5, 4, 3, 2, 1 kcal mol −1 Å −1 ) of 10 ps each. A final 20 ns long MD simulation was then 713 performed, in which no restraints were used, to equilibrate the box size. All dynamics steps 714 used the SHAKE algorithm. Simulations performed in the NVT ensemble used Langevin 715 temperature control (with a collision frequency of 1 ps −1 ) and a simulation time-step of 1 fs. 716 Simulations performed in the NPT ensemble again used Langevin temperature control 717 (collision frequency of 1 ps −1 ) and a Berendsen barostat (1 ps pressure relaxation time), with 718 a simulation time-step of 2 fs. 719 720 One of the main counteracting forces working against the development and spread of 736 antibiotic resistance is the associated cost in fitness, which can be minimized through careful 737 gene regulation. We here investigated fine-tuned signalling via a sensory complex consisting 738 of a transporter and a histidine kinase. We show how the kinase is activated upon antibiotic 739 attack, that the transporter exerts complete control over the kinase, and also has itself evolved 740 to conserve energy. 741 742