Fine control of metal concentrations is necessary for cells to discern zinc from cobalt

Bacteria possess transcription factors whose DNA-binding activity is altered upon binding to specific metals, but metal binding is not specific in vitro. Here we show that tight regulation of buffered intracellular metal concentrations is a prerequisite for metal specificity of Zur, ZntR, RcnR and FrmR in Salmonella Typhimurium. In cells, at non-inhibitory elevated concentrations, Zur and ZntR, only respond to Zn(II), RcnR to cobalt and FrmR to formaldehyde. However, in vitro all these sensors bind non-cognate metals, which alters DNA binding. We model the responses of these sensors to intracellular-buffered concentrations of Co(II) and Zn(II) based upon determined abundances, metal affinities and DNA affinities of each apo- and metalated sensor. The cognate sensors are modelled to respond at the lowest concentrations of their cognate metal, explaining specificity. However, other sensors are modelled to respond at concentrations only slightly higher, and cobalt or Zn(II) shock triggers mal-responses that match these predictions. Thus, perfect metal specificity is fine-tuned to a narrow range of buffered intracellular metal concentrations.

responds to metals in vitro 29,40 , and a single amino acid substitution generates an FrmR variant (FrmR E64H ), which can respond to Zn(II) and cobalt in vivo 29 . Increased sensitivity to Zn (II) of FrmR E64H is due to an~tenfold tighter Zn(II) affinity and 4-fold weaker DNA affinity of apo-FrmR E64H , relative to FrmR 29 . Thus, modest changes can generate a metal sensor from a non-metal sensor and this suggests that the cell may be poised close to thresholds for detecting and discerning between metals.  Allosteric mechanisms of Salmonella sensors and structural models based on Protein Data Bank files 4MTD for Zur (a), 4WLW for ZntR (b), 5LCY for both RcnR (c) and FrmR E64H (d) with identified DNA-binding sites (bold), upstream of target genes. The DNA sequences shown were used for fluorescence anisotropy and orange bars indicate the region amplified by end point PCR and quantitative PCR. Known or inferred ligands for effector binding are enlarged: Zur contains a Cys 4 -structural site and at least one sensory site. The dinuclear Zn(II) site of E. coli ZntR (PDB: 1Q08) is shown, noting that solution studies of Salmonella ZntR indicate a mononuclear site 5,29 . An RcnR Co(II) site has been proposed, which may also include Glu 32 70 . FrmR E64H and FrmR have overlapping sites for formaldehyde (Cys 35 , Pro 2 ) and metal binding (Cys 35 , His 60 and either His 64 for FrmR E64H or Glu 64 for FrmR) 40 . Cognate effectors are depicted When the transcription of genes encoding Ni(II) import and export was engineered to rely on sensors adjusted to respond at higher Ni(II) concentrations, the cellular Ni(II) content increased relatively little and instead the sensors ceased to respond 41 . Thus, the sensitivity of a DNA-binding metal sensor is tuned to a buffered concentration of its cognate ion, presumably to regulate mechanisms that prevent this buffer from becoming depleted or saturated with metal 41 . Factors known to influence metal detection by each sensor are metal affinity, DNA affinity, the allosteric mechanism connecting metal binding to DNA binding, plus the abundance of sensor protein 1,7,10,11,19,29,30,[41][42][43] . Several of these parameters have been measured for different members of a set of sensors for Zn(II), Ni(II) and Co(II) in a common cell (Synechocystis PCC 6803) 7,30,31,41 . The Ni(II) sensor, InrS, has the tightest affinity for Ni(II) while the Zn(II) sensor ZiaR has the greatest free energy coupling Zn(II) binding to DNA binding among the cells' set of sensors 7,30 . This illustrates how metal selectivity can be understood by comparing the relative properties of different metal sensors within a common cell. Here such observations are further rationalised by relating the sensitivities of metal sensors to buffered concentrations of the respective ions.
Ultimately, metal sensitivity of a sensor will be some function of all of the above factors operating together. However, quantitatively combining all of these factors presents a challenge. In this work, we measure these parameters and incorporate them into mathematical models in order to understand in vivo specificity of sensors to Zn(II) and Co(II). The computational methods are set out in a format to assist their use by others. Sensors are modelled to respond at lower-buffered concentrations of their cognate metal, compared to sensors for other effectors, explaining and correctly predicting metal specificity. However, sensors for other effectors are modelled to be only marginally less sensitive to the non-cognate metal, and indeed metal shock triggers predictable responses to non-cognate metals. Thus, we discover that tight regulation of buffered intracellular metal concentrations is a prerequisite for perfect metal specificity, rendering sensors vulnerable to dysregulation, with implications for the microbicidal action of metal fluxes.

Results
Sensors are selective at non-inhibitory concentrations. To compare the response of Zur, ZntR, RcnR, FrmR and FrmR E64H to Zn(II) and cobalt, Salmonella cells were cultured in minimal media (~4-5 h) supplemented with MNICs of each metal (giving ≤10% inhibition of growth), ensuring multiple cell division cycles in the presence of metal ( Supplementary Fig. 2). Transcript abundance was visualised by end point reverse-transcriptase PCR and enumerated by quantitative PCR (qPCR; Fig. 2a-e). Expression of znuA was repressed upon Zn(II) supplementation, but not upon cobalt supplementation, reflecting a selective response of Zur to Zn(II) (Fig. 2a). Under the same conditions, zntA transcripts accumulated in response to Zn(II) but not cobalt, confirming that ZntR-mediated activation of zntA expression was also specific for Zn(II) (Fig. 2b). Conversely, rcnA transcripts accumulated in response to cobalt but not Zn(II), confirming that repression of rcnA by RcnR was alleviated by cobalt, but not Zn (II) (Fig. 2c). The abundance of frm transcripts was monitored in Salmonella strains containing either FrmR or variant FrmR E64H29 . As expected, frm transcripts did not accumulate in response to cobalt or Zn(II) when regulated by FrmR, but surprisingly they also failed to respond when regulated by FrmR E64H (Fig. 2d, e), despite previous observations that this variant conferred cobalt and Zn(II)-dependent β-galactosidase activity 29 . Repression by both FrmR and FrmR E64H was exclusively alleviated by formaldehyde (Fig. 2f). Under these conditions, Zur, ZntR, RcnR and FrmR were selective for their cognate effector.
Multiple sensors respond to cobalt shock. Gene expression under the control of FrmR E64H was previously investigated via assays of β-galactosidase activity after short-term exposure to elevated cobalt: a cobalt shock 29,40 . In an attempt to reconcile apparent differences between these past data and Fig. 2e, a logarithmically growing culture was exposed to cobalt for 10 min and transcript abundance visualised by end point PCR and enumerated by quantitative PCR (Supplementary Figs. 6 and 7 and Fig. 3). A twofold change in transcript abundance was used as a threshold for sensor responsiveness as indicated by arrows on quantitative PCR graphs throughout. Higher cobalt concentrations could be used during such short term, compared to prolonged, metal exposures with only a modest effect on cell viability observed (~15% reduction at 5 μM CoCl 2 , Supplementary  Fig. 11a), while these higher cobalt concentrations are inhibitory during prolonged exposure (~30% growth reduction at 5 μM CoCl 2 , Supplementary Fig. 11b). Under these conditions, expression regulated by either RcnR or FrmR E64H was derepressed by cobalt, while repression mediated by FrmR was unaffected, consistent with FrmR E64H being a cobalt-sensing variant of FrmR. However, during this transient cobalt exposure, znuA expression was also partially repressed relative to the control, and zntA expression activated ( Fig. 3 and Supplementary  Fig. 6), identifying both Zur and ZntR as targets of cobalt mis-metalation. The affinity of FrmR E64H for Co(II) is 500-fold weaker than that of RcnR and for this reason its response to  Co(II) was previously considered enigmatic 29 . However, whereas RcnR is tuned to a buffered concentration of Co(II) in cells grown in non-inhibitory cobalt concentrations, it is now evident that FrmR E64H only responded during cobalt shock. Stepwise regulation of Salmonella sensors in response to increasing concentrations of cobalt shock occurred in the order RcnR, Zur, FrmR E64H , ZntR, followed by FrmR, which did not respond (Fig. 3). It is proposed that non-cobalt sensors responded under cobalt shock because the cytosolic buffer became transiently saturated and higher intracellular concentrations occurred. This is consistent with growth inhibition during prolonged exposure to these higher metal concentrations (Supplementary Fig. 11b).
ZntR and Zur bind Co(II) with sub-micromolar affinities. We have previously demonstrated that in vitro both ZntR and Zur bind Co(II) in sites that can be displaced by Zn(II) 29 . To determine their Co(II) affinities (hereafter affinity refers to a dissociation constant), both proteins were purified following overexpression in E. coli and confirmed to be ≥95% pure (Supplementary Fig. 12) and ≥90% reduced 29 . ZntR was ≥95% metal free, and Zur contained~1 molar eq. Zn(II) consistent with filling of a structural Zn(II) site identified in Zur and other Fur-family members 44 . Both proteins were competed against the fluorophore fura-2, which exhibits a decrease in fluorescence emission upon Co(II) binding and has been used to determine Co(II) affinities of metal sensor proteins including RcnR and FrmR E64H (Fig. 4a, b) 29,31 . ZntR binds two Co(II) ions per dimer 29 , and both sites were observed during competition with fura-2, with a combined affinity of 9.5 (±1.0) × 10 −8 M ( Fig. 4a and Table 1). Zur binds up to four Co(II) or Zn(II) ions per dimer in addition to the structural   and ZntR (9.8 μM) with Co(II). Solid line is a fit to a model describing competition from ZntR for one molar equivalent of Co(II). Dashed lines are simulated curves with K Co(II) tenfold tighter and tenfold weaker than the fitted value. b As in a but with fura-2 (14.6 μM) and Zur (9.8 μM) (n = 5). c Representative (n = 3) Zur absorbance spectra upon titration of Zur (52 μM) and EGTA (50 μM) with Co(II). d Binding isotherm at 350 nm for data shown in c. Solid line is a fit to a model describing competition from Zur for two molar equivalents of Co(II). Dashed lines are simulated curves with K Co (II) tenfold tighter and tenfold weaker than the fitted value ARTICLE NATURE COMMUNICATIONS | DOI: 10.1038/s41467-017-02085-z 4 Zn(II) sites (a total of 6:1 Me(II):Zur dimer) 29 . Competition with fura-2 did not distinguish between four Co(II)-binding events (Fig. 4b). In contrast, Zn(II) binding by Zur occurs with strong negative cooperativity: Sites 1-2 are tighter than site 3 by~120fold and site 3 is tighter than site 4 by~6000-fold 29 . The intense UV-visible absorption spectra of Co(II) binding to Zur (replicated here in Supplementary Fig. 13a) 29 , allowed observable competition with the spectrally silent chelator ethylene glycol tetraacetic acid (EGTA) ( Fig. 4c and Supplementary Fig. 13b), and confirmed the lack of detectable cooperativity of Co(II) binding to Zur. Data were fit to a model describing four combined Co(II)binding events per Zur dimer (Fig. 4d). The affinity of Zur for Co (II) was determined to be 1.5 (±0.6) × 10 −8 M averaged from both fura-2 and EGTA competition experiments ( Table 1). The affinity of ZntR for Co(II) was comparable to that of FrmR E64H , while the affinity of Zur was approximately one order of magnitude tighter than either ZntR or FrmR E64H .
Co(II) affects DNA binding by Zur, FrmR E64H and ZntR. Since Zur, FrmR E64H and ZntR responded to cobalt shock in vivo ( Fig. 3 and Supplementary Figs. 6 and 10), it seemed probable that Co(II) triggers allosteric responses which promote association of Zur with the znuA operator-promoter, dissociation of FrmR E64H from the frmRA operator-promoter, and activation by ZntR of the zntA operator-promoter. The degree to which metal binding is coupled to DNA binding can be described as the allosteric coupling-free energy (ΔG C ), which, in combination with metal affinity, contributes to metal selectivity 1,7,42 . Fluorescence anisotropy, using a fluorescently labelled dsDNA fragment containing the identified Zur-binding site upstream of znuA (Fig. 1a), was used to examine the effect of Co(II) on allostery. Initially, the stoichiometry of Zur binding to DNA was determined with saturating concentrations of Zn(II) (ensuring filling of exchangeable sites 1-4) and demonstrating that two Zn(II)-Zur dimers bind to the znuA operator-promoter sequence (Fig. 5a). E. coli Zur (93% identity to Salmonella Zur) binds to a similar target DNA sequence as two adjacent dimers with positive cooperativity 2 . The data for Salmonella Zur were fit to a model describing sequential binding of two Zur dimers to the znuA operator-promoter. The DNA affinity of both Zn(II)-Zur dimers was determined to be 5.4 (±1.8) × 10 −8 M ( Fig. 5b and Table 1).
Similar numbers of ZntR, RcnR and Zur protomers per cell. The number of copies of FrmR and FrmR E64H per cell was Table 1 Metal affinities, DNA affinities, allosteric coupling-free energies and abundance of Salmonella sensors All constants and abundances are means of at least triplicate determinations ('n' specified in figure legends) Table 1). Quantification of RcnR was initially challenging due to interfering species from the complex Salmonella cell lysates. RcnR was therefore enriched in a quantitative manner via partial purification. The abundance of RcnR was thus determined to be 22 (±2) tetramers per cell (Table 1).
Thermodynamic data predict the responses to cobalt. Fractional metal saturation of a sensor (determined by 1/K 1 alone, Fig. 6) has commonly been used as a surrogate measure of metal sensitivity 10 . However, metal binding and DNA binding are thermodynamically coupled such that DNA occupancy is not fully represented by 1/K 1 alone ( Fig. 6)  RcnR was thus modelled to respond at the lowest cobalt concentration explaining why this is the bona fide sensor of Co (II) (Fig. 7). Because the weak Co(II) affinity of FrmR precluded determination of the DNA affinity for Co(II)-FrmR, this was estimated from 1/K 4 for Zn(II)-FrmR and the fold-difference between 1/K 4 for the Co(II)-FrmR E64H and Zn(II)-FrmR E64H variant (Table 1) 29 . Fractional DNA occupancy by FrmR E64H did not reach that of FrmR, due to the weaker DNA affinity of apo-FrmR E64H relative to apo-FrmR (Fig. 7) 29 . The metalated form of MerR-family regulators (Fig. 1b), such as ZntR, activate expression by distorting their target promoter 48,49 , therefore, zntAPro bound by Co(II)-ZntR ((P•M)•D) was used to represent the active species. This implied a dynamic range close to that of FrmR E64H (Fig. 7). All of the other sensors were shown to be tuned above the cobalt sensitivity of RcnR, which would avoid mal-responses to Co(II). However, the margin for specificity was narrow such that Zur would also respond to Co(II) if the concentration became an order of magnitude greater than the set point for RcnR. To create the perfect metal selectivity observed in Fig. 2, there must be fine control of the intracellular cobalt concentration: Under cobalt shock, such control became imperfect (Fig. 3). Moreover, the observed sequence of activation of each sensor in response to increasing cobalt shock agreed with the order predicted by the thermodynamic models (Figs. 3 and 7): Noting that the modelled responses of FrmR E64H and ZntR overlapped (Fig. 7), although expression data indicated that the former was more sensitive to cobalt than the latter, perhaps because the effects of Co(II) on DNA binding do not fully reflect   Co(II)-FrmR E64H data were fit to a model describing a 2:1 FrmR E64H tetramer (non-dissociable):DNA stoichiometry and solid line is a simulated curve using the mean DNA affinity across the experiments shown 29,40 . Dashed grey line is a simulated curve using the published DNA affinity for apo-FrmR E64H 7 . e As a but with apo-ZntR and 2.5 μM zntAPro in 5 mM EDTA. Line is a linear fit to the first three data points predicting Δr obs = 0.0246 at 2:1 ZntR:zntAPro. f As b but with apo-ZntR and 10 nM zntAPro (n = 7) . g As f but with Zn(II)-ZntR (n = 3). h As f but with Co(II)-ZntR in 5 μM Co(II) (n = 4). Data in f, g and h were fit to a model describing 1:1 ZntR dimer (non-dissociable):DNA and lines are simulated curves using the mean DNA affinity across the experiments shown. For g and h, to determine the DNA affinity of the tightest binding event, data to 1500 nM ZntR monomer were used ZntR activation (Fig. 3 (Fig. 6), are known or can be estimated (detailed step-by-step instructions are described in the 'Methods' section).
The DNA affinity of Zur is tightened when all four exchangeable Zn(II) sites are filled (Fig. 5b), however, the weakest sites have Zn(II) affinities in the region of 10 −7 M. Using fluorescence anisotropy, we confirmed that filling of the tightest two sites alone (referred to as Zn(II) 2 -Zur; K Zn(II) = 6.4 (±0.4) × 10 −13 M, Table 1), was sufficient to induce an allosteric change that enabled Zur to bind to the znuA operator-promoter. The DNA affinity of Zn(II) 2 -Zur , was 4.1 (±1.0) × 10 −8 M (Fig. 8d, and Table 1). Using the Zn(II) affinities, DNA affinities and abundance of Zur, ZntR, RcnR, FrmR E64H and FrmR (Table 1), the fractional occupancy of their respective operator-promoters (with total protein or with Zn(II)-ZntR) were modelled as a function of buffered Zn(II) concentration using the same procedures as described for Co(II) (Fig. 9a). Analogous to the models for Co(II) (Fig. 7), the cognate sensors for Zn(II) were calculated to respond at the lowest buffered Zn(II) concentrations, once again explaining specificity, in this case for Zn(II) (Fig. 9a). Sensors for other effectors were tuned above this concentration, however the margin for specificity was again narrow such that RcnR would also respond to Zn(II) if the concentration became an order of magnitude greater than the set point for ZntR. To create the perfect metal selectivity shown in Fig. 2, as with Co(II), there must also be fine control of intracellular Zn(II) concentrations.
To investigate the response of each sensor to Zn(II) shock, Salmonella cells were exposed to increasing Zn(II) concentrations for 10 min (Fig. 9b, c and Supplementary Fig. 11c). The highest Zn(II) concentrations (80 and 100 μM) were inhibitory during prolonged exposure (Supplementary Fig. 11d). Under these conditions, the stepwise pattern of Zn(II)-responsive gene expression, either monitored by end point PCR or qPCR, again aligned with the thermodynamic models, such that Zur responded at the lowest Zn(II) concentrations, followed by ZntR, RcnR and FrmR E64H , and lastly FrmR which did not respond in vivo (Fig. 9a-c). Notably, the thermodynamic models were equivocal with respect to the relative sensitivities of RcnR and FrmR E64H to Zn(II) with the curves intersecting (Fig. 9a), and for these two sensors the order observed by quantitative PCR or end point PCR was indistinguishable (Fig. 9b, c), both were less sensitive than the bona fide Zn(II) sensors and more sensitive than FrmR. Selectivity is adapted to operate perfectly only when [D] [D]  Table 1. FrmR and FrmR E64H were normalised to the same scale. Derepression by RcnR, FrmR and FrmR E64H occurs as the fractional occupancy of their promoters decrease, co-repression by Zur occurs as occupancy of its promoter increases, and activation by ZntR occurs as the fractional occupancy of its promoter with metalated ZntR increases. Numbering reflects the order of response observed for each sensor in Fig. 3 NATURE COMMUNICATIONS | DOI: 10.1038/s41467-017-02085-z ARTICLE NATURE COMMUNICATIONS | 8: 1884 | DOI: 10.1038/s41467-017-02085-z | www.nature.com/naturecommunications cells exert fine control over intracellular metal concentrations and after Zn(II) shock, such control becomes imperfect. The modest differential between the fractional occupancy curves for the different sensors reveals that cells are on the cusp of mis-sensing Zn(II), as well as Co(II), when subjected to metal shocks (Figs. 7 and 9a).

Discussion
Our calculations of gene regulation at different intracellular Co(II) and Zn(II) concentrations explain metal selectivity in metal sensing in terms of equilibrium thermodynamics by using determined metal affinities, DNA affinities, coupling free energies and the number of sensor molecules per cell (Figs. 7 and 9a). At first inspection, these models seem incorrect by revealing that RcnR is inherently more sensitive to Zn(II) than to Co(II) by one to two orders of magnitude ( Supplementary Fig. 18), yet RcnR showed the opposite selectivity in cells during prolonged growth in elevated metal and is known to be a Co(II) sensor (Figs. 1 and 2) 29 . The explanation is that metal sensors are tuned to the buffered concentrations of their cognate metal 41 , and the buffered concentration of Zn(II), but not cobalt, is maintained below the set point for RcnR. The set points for the Zn(II) sensors ZntR and Zur reveal this lower buffered concentration for Zn(II) (Fig. 9a). In this context, metal specificity now becomes readily understandable by comparing the sensitivities for Zn(II) (Fig. 9a), and for Co(II) (Fig. 7), of the five sensors to reveal that the bona fide sensors are the most sensitive in the set. During prolonged growth in elevated Zn(II), the intracellular Zn(II) concentration must have been finely controlled to within about one order of magnitude in order to trigger Zur and ZntR, but not RcnR or FrmR E64H (Figs. 2 and 9a). This must be a buffered Zn(II) concentration, with associative metal transfer, since one hydrated ion per cell volume equates to~10 −9 M, which would be sufficient to trigger RcnR and FrmR E64H (Fig. 9a) 50 . Similarly, during prolonged growth in elevated Co(II), the intracellular buffered Co(II) concentration must also have been finely controlled to within about one order of magnitude in order to trigger RcnR but not Zur (Figs. 2 and 7). Thus, these metal sensors are adapted to discriminate perfectly between these inorganic elements only when metals are buffered, with associative metal transfer, and when metal concentrations are finely controlled.
The models predict that if the buffer becomes saturated then the Zn(II) sensors will respond to Co(II) and vice versa the Co(II) sensor will respond to Zn(II) (Figs. 7 and 9a). During Zn(II) shock, FrmR E64H and RcnR did respond, consistent with the Zn(II) concentration having transiently increased above the buffered concentration (Fig. 9b, c). Similarly during Co(II) shock Zur, FrmR E64H and ZntR responded (Fig. 3). For both metals, the order of the response to increasing metal shock ( Fig. 3 and 9b, c), correlated with the order predicted from the thermodynamic properties of the sensors (Figs. 7 and 9a), further validating the models and suggesting that even the shock responses are not solely determined by on-rates (kinetics). This also reveals that when assigning metal specificity to metal sensors by monitoring gene expression in cells exposed to exogenous metals, care should be taken to optimise the growth conditions and avoid saturation of the intracellular buffer. For sensors with more than one DNA target, multiple set points may exist to allow a graded response to changing metal demands as the cytosolic metal buffer becomes increasingly full. Intriguingly, Bacillus subtilis Zur has at least three set-points reflecting filling of its multiple Zn(II) sites and its varying DNA affinities on different operator-promoters, giving rise to three waves of Zn(II)-dependent gene expression 51 . These three waves could reflect different levels of saturation of the cytosolic buffer. Alternatively, evidence of mal-responses of sensors for other metals might indicate if one of the waves occurs when the buffer becomes fully Zn(II) saturated.
The crowded cytosol contains a multitude of sulphur, nitrogen and oxygen ligands associated with an array of metabolites and macromolecules, many of which can be readily organised into different metal-binding combinations and geometries. Such a polydisperse mixture will inevitably bind and buffer metals in the order of the Irving Williams series 41 . It is anticipated that cytosolfacing metal-binding sites of metal transporters will also be tuned to these buffered metal concentrations. For some metals, and in some organisms, the cytosolic buffer may be dominated by a single molecule such as glutathione or its substitutes such as bacillithiol, or a free amino acid such as histidine 41,[52][53][54][55] . Macromolecules with multiple labile metal sites, e.g., metallothioneins, may also be induced to expand the depth of the buffer when metals such as Zn(II) increase in abundance 56 . At present, it is unclear whether or not the buffer for either Co(II) or Zn(II) in Salmonella is dominated by a single, and potentially shared, molecule. Notably, mutants deficient in the synthesis of glutathione showed impaired detection of Co(II) and Zn(II) by FrmR E64H29 , but now we know that metal detection by this variant sensor only occurs during metal shock, presumably once components of the bona fide buffer have become saturated. In contrast, there is negligible effect of glutathione on Zn(II) sensing by Salmonella ZntR 29 , whereas Zn(II) sensing by B. subtilis CzrA is enhanced in the absence of bacillithiol 52 . It is noteworthy that Salmonella is at least an order of magnitude more sensitive to exogenous cobalt than Zn(II) (Supplementary Fig. 11). Unlike E. coli, Salmonella requires cobalt to synthesise cobalamin, vitamin B 12 , but this is only made under anaerobic conditions which perhaps renders Salmonella proteins susceptible to mismetalation by unwanted, un-sequestered, cobalt during aerobic growth. In eukaryotes, there is scope for diversity in buffered metal concentrations within different intracellular compartments (nucleus, organelles, vesicles, trans-Golgi network, endoplasmic reticulum, for example) and spectral probes have been developed to interrogate such concentrations [57][58][59] . Since metal-sensing transcriptional regulators can also report upon metal occupancy in vivo and are tuned to metal concentrations for example in the bacterial cytosol, then their metal sensitivities provide an alternative approach to interrogate the vital buffered metal concentrations: In short, the K 5 values at which each sensor responds using the calculations described here (Fig. 6, Supplementary Data 1, Supplementary Software and Methods section). In the same way that metal selectivity of metal sensors becomes comprehensible in the context of these values (Figs. 7 and 9a and Supplementary Fig. 18), so metalation of other metalloproteins should become understandable once a complement of cellular K 5 values for different metals have been calculated.
In conclusion, we discovered that perfect metal specificity in metal sensing was restricted to a finely controlled range of buffered metal concentrations, which were exceeded during metal shocks (Figs. 3, 7 and 9). These data support the prediction that bacteria are susceptible to the mis-sensing of metals and hence the notion that this vulnerability is exploited by immune systems. Metals, chelants and ionophores also have a long history of use as antimicrobials in medicine, agriculture and as preservatives 60,61 . However, the development of this wide spectrum of metal related treatments has largely been empirical, up to now.

Methods
Bacterial strains, DNA manipulations and growth conditions. S. enterica sv. Typhimurium strain SL1344 was used as wild type and strain LB5010a was used as a restriction-deficient modification proficient host for DNA manipulations. Both were a gift from J.S. Cavet (University of Manchester), and originally from the Salmonella Genetic Stock Centre 29 . Deletion derivative ΔfrmR (SL1344 lacking the frmR coding sequence) was generated previously 29 . E. coli strain DH5α was used for routine cloning and strain BL21(DE3) was used for recombinant protein overexpression (both from historical lab stocks). E. coli strains BW25113ΔzntR:: kan (JW3254-5) and BW25113ΔnikR::kan (JW3446-3) were originally from the Keio collection 62 (and a gift from D. Weinkove, Durham University). Kanamycin resistance cassettes were removed using the helper plasmid pCP20 carrying the FLP recombinase. Promoter-lacZ fusion constructs with the frmR operator-promoter and frmR or frmR E64H coding sequence upstream of lacZ, have been described previously 29 . Bacteria were cultured aerobically (shaking at 200 rpm) at 37°C in LB medium or M9 minimal medium, supplemented with thiamine (0.001%, w/v) and either L-histidine (20 μg ml −1 ) for Salmonella or 1 μM ferric citrate for E. coli. Kanamycin (25 μg ml −1 ), chloramphenicol (8 μg ml −1 ) and carbenicillin (100 μg ml −1 ) were added where appropriate. Maximum non-inhibitory concentrations (MNIC; defined as the maximum concentration which inhibited growth by~10 %) of CoCl 2 and ZnSO 4 were determined in M9 minimal medium supplemented with metal salt, following dilution of overnight cultures to an OD of 0.025 at 600 nm. For metal shock exposures, logarithmic cells were statically cooled to 25°C for 20 min followed by a 10 min exposure to CoCl 2 or ZnSO 4 before dilution in phosphate-buffered saline and enumeration on LB agar. Concentrations of metal salts were confirmed by ICP-MS.
Generation of E. coli BW25113 double-deletion mutants. BW25113Δzur::cat was generated by the λ Red method 63 , using plasmid pKD3 and primers 1 and 2 (Supplementary Table 4; hereafter all primer numbers relate to this table). Mutants  Table 1. Numbering reflects the order of response visualised for each sensor in Fig. 9b, c. b Representative (n = 3) transcript abundance following 10 min exposure of Salmonella to increasing [Zn(II)] assayed by end point PCR. Arrows identify the lowest observed exogenous [Zn(II)] at which each sensor appeared to respond. Data for control genes are presented in Supplementary Fig. 15 and full gel images in Supplementary Fig. 16. c Transcript abundance for the samples shown in b measured by qPCR (error bars are s.d.). Arrows represent a ≥twofold change in transcript abundance. Heat maps of qPCR data from three biological replicates are presented in Supplementary Fig. 17 were selected on LB medium supplemented with chloramphenicol. The Δzur::cat fragment was moved into strain BW25113ΔzntR (kan cassette removed) by P1 transduction. The chloramphenicol resistance cassette was removed and genotype confirmed by PCR using primers 3-6. P1 transducing lysate from BW25113ΔrcnR:: kan (JW2092-1, a gift from P. Chivers, Durham University) was used to move the ΔrcnR::kan fragment into BW25113ΔnikR (kan cassette removed). The kan cassette was removed and genotype confirmed by PCR using primers 7-10.
RNA extraction and reverse-transcriptase PCR. Expression mediated by FrmR and FrmR E64H was measured in Salmonella strain SL1344ΔfrmR containing either P frmRA -frmR or P frmRA -frmR E64H reporter constructs (generating SL1344 FrmR and SL1344 FrmRE64H , respectively) cultured in supplemented M9 minimal medium following dilution of overnight cultures to an OD of 0.025 at 600 nm. To enable direct comparison of metal sensor responses with FrmR E64H -mediated regulation, expression of rcnA, znuA and zntA was measured in SL1344FrmR E64H , with the exception of Supplementary Fig. 6, where SL1344 FrmR was used as a further control. The medium was supplemented with MNICs of metals or formaldehyde and grown to mid-logarithmic phase prior to assays. MNICs under these growth conditions were 0.25 μM CoCl 2 , 50 μM ZnSO 4 (described above; Supplementary  Fig. 11 Table 4). For end point PCR, fragments were subsequently resolved by agarose gel electrophoresis. Gels were imaged with a Gel-Doc XR + gel documentation system (Bio-Rad). Quantitative PCR analyses were performed in 20 μl reactions using 2 ng of cDNA as a template, 0.8 μM of the appropriate primer pairs and PowerUp™ SYBR ® Green Master Mix (ThermoFisher Scientific). Each sample was analysed in three technical replicates using a Rotor-Gene Q 2plex (Qiagen). The fold change in transcript level relative to control conditions was analysed using the 2 −ΔΔCT method with rpoD as the reference gene 64 . Trends were confirmed with biological replicates on three occasions.
Protein overexpression and purification. E. coli BL21(DE3) containing pETzntR, pETzur, pETfrmR E64H and pETrcnR was used to overexpress ZntR, Zur, FrmR E64H and RcnR, respectively 29 . Protein purification was conducted using a combination of Ni(II) affinity, gel filtration, heparin affinity and ion-exchange chromatography. 29 . Experimentally determined extinction coefficients were used to quantify purified proteins 29 . Proteins were confirmed to be ≥95% pure as assessed by SDS-PAGE ( Supplementary Fig. 12). Anaerobic protein stocks (maintained in an anaerobic chamber) were prepared as described and confirmed to be ≥95% metal free and ≥90% reduced 29 , with the exception of Zur which contained~1 molar equivalent of Zn(II) (per monomer) as purified. All in vitro experiments were carried out under anaerobic conditions using Chelex-treated and N 2 -purged buffers 29 .
Determination of metal affinities. All experiments were conducted in 100 mM NaCl, 400 mM KCl, 10 mM HEPES pH 7.0. For competition with fura-2, CoCl 2 was titrated into a mixed solution of protein and fura-2 and fluorescence emission was recorded at equilibrium at 510 nm (λ ex = 360 nm; T = 20°C) using a Cary Eclipse fluorescence spectrophotometer (Agilent Technologies) 29,31 . Fura-2 was quantified using the extinction coefficient ε 363 nm = 28,000 M −1 cm −1 31 . For competition with EGTA, CoCl 2 was titrated into a mixed solution of Zur and EGTA, and absorption spectra were recorded at equilibrium using a λ 35 UV-visible spectrophotometer (Perkin Elmer Life Sciences). Control experiments without EGTA were also performed. For competition with magfura-2 or quin-2, ZnCl 2 was titrated into a mixed solution of RcnR and magfura-2 or RcnR and quin-2, and absorbance was recorded at equilibrium at 325 nm (magfura-2) or 265 nm (quin-2). Magfura-2 and quin-2 were quantified using the extinction coefficients ε 369 nm = 22,000 M −1 cm −165 and ε 261 nm = 37,000 M −1 cm −166 , respectively. Competition data were fit to models described in figure legends and Table 1 using Dynafit to determine Co(II) and Zn(II) affinities 47  Fluorescence anisotropy. Fluorescently labelled double-stranded DNA probes, znuAPro and zntAPro were generated using oligonucleotides 23 (hexachlorofluorocein labelled) and 24 containing the identified Zur binding site upstream of znuA, or 25 (hexachlorofluorocein labelled) and 26 containing the identified ZntRbinding site upstream of zntA. frmRAPro and rcnAPro have been described previously 29,40 . All oligonucleotides are listed in Supplementary Experiments to determine the stoichiometry of binding of Zur to znuAPro and ZntR to zntAPro were performed as described 29 , by titration of Zn(II)-Zur into 1 μM znuAPro or apo-ZntR into 2.5 μM zntAPro. DNA-binding affinities were determined using 10 nM dsDNA probe. DNA affinities and coupling free energies (ΔG C ) were determined with Dynafit 29,40,47 . Models for RcnR and FrmR E64H have been described elsewhere 40 . Data for Zur were fit to a model describing sequential binding of two non-dissociable dimers to two sites on znuAPro. Data for ZntR were fit to a model describing binding of one non-dissociable dimer to zntAPro. The anisotropy change associated with a dimer binding to DNA was determined to be 0.025 by using Dynafit to simultaneously fit the data from apo-ZntR titrations (n = 7, Fig. 5f) and this value was then fixed to individually fit the data sets for apo-, Zn(II)-and Co(II)-ZntR and determine DNA affinities. Mean and standard deviation values were determined from at least triplicate analyses ('n' specified in figure legends).
Quantitation of protein abundance. E. coli strains BW25113ΔzntR/Δzur and BW25113ΔnikR/ΔrcnR, and Salmonella strain SL1344 were cultured to logarithmic phase in supplemented M9 minimal medium. Purified stocks of Zur, ZntR and RcnR were quantified by amino acid analysis (UC Davis). For ZntR and Zur quantitation, soluble cell lysates were prepared in 40 mM NaCl, 160 mM KCl, 10 mM EDTA, 10 mM DTT, 10 mM HEPES, pH 7.8, with addition of protease inhibitor mixture (Sigma), and post-sonication, the soluble cell lysate was syringe filtered (0.45-μm pore size), snap frozen in liquid N 2 , stored at −80°C, and thawed on ice before use 29 . Standard curve samples were prepared by dilution of purified protein stocks into cell lysates from BW25113ΔzntR/Δzur 29 . For RcnR quantitation, interference was observed using cell lysates directly, and an alternative approach was used. Following growth, harvested cells were stored at −20°C before being resuspended in 300 mM NaCl, 10 mM EDTA, 1 mM TCEP, 10 mM HEPES pH 7.0, 1 mM PMSF. Standard curves were prepared by dilution of purified RcnR into soluble cell lysates from BW25113ΔnikR/ΔrcnR. Cell lysates (standard curve and experimental samples) were enriched for RcnR using a 1 ml HiTrap Heparin column (GE Healthcare) equilibrated in 300 mM NaCl, 10 mM EDTA, 1 mM TCEP, 10 mM HEPES pH 7.0. Bound protein was washed in binding buffer, and eluted in a single step in the same buffer but with addition of 1 M NaCl. Aliquots were stored at −80°C. Heavy labelled peptides ([ 13 C 6 , 15 N 4 ]arginine residues; Thermo Fisher) were used as working internal standards (IS). Samples were prepared and analysed as described 29  affinities (1/K 3 and 1/K 4 ) plus cellular abundance of each sensor (P) and DNA target (D) ( Table 1) 47 . Where the standard deviations for the DNA affinities of Co (II)-and Zn(II)-bound proteins overlapped, average values generated by combining the data for both metals were used for 1/K 4 . These were: 3.6 × 10 −8 M for Co (II)-Zur and Zn(II) 2 -Zur, 1.4 × 10 −5 M for Co(II)-RcnR and Zn(II)-RcnR, and 4.7 × 10 −7 M for Co(II)-ZntR and Zn(II)-ZntR. To determine the amount of (P•M) •D for ZntR, the response for 'PD' was removed from the Dynafit script. A cell volume of 1 fl was used to calculate [P] total and [D] total from the number of protein assemblies per cell (i.e. dimers or tetramers) ( Table 1) and target DNA binding sites per cell (assumed to be 1 copy per cell for RcnR and ZntR, 4 copies per cell for Zur due to additional gene targets 2 , and 15 copies per cell for FrmR and FrmR E64H due to the presence of a low copy number reporter plasmid) 29