Generating a Metal-responsive Transcriptional Regulator to Test What Confers Metal Sensing in Cells*

Background: Metal-specific transcription has been correlated with the relative properties of a cells' set of metal sensors. Results: A one-residue substitution enabled a DNA-binding formaldehyde sensor to detect Zn(II) and cobalt. Conclusion: Weaker DNA affinity combined with tighter Zn(II) affinity enabled Zn(II) sensing with a smaller coupling free energy. Significance: Relative affinity determined the best sensor in the set for Zn(II) but not for cobalt.

Metal-sensing and DNA-binding transcriptional regulators are central to the machinery that optimizes buffered metal concentrations inside cells to enable correct protein metallation (1,2). In general, the tighter the K metal of a metal sensor, the lower the [buffered metal] (1). Fresh experimental approaches are needed to test hypotheses about the mechanisms determining which metal(s) each sensor detects. Uncertainty also remains about the nature of the exchangeable pools of different metals, including the major ligands and the precise buffered metal concentrations, and how these vary under different environmental conditions or between organisms.
Metal sensors tend to bind divalent metals with an order of affinity that matches the Irving Williams series, regardless of which metal(s) they detect in a cell (1)(2)(3). This raises questions about how a sub-set of sensors can detect the weaker binding metals in vivo (4 -6). One facet of the solution is that the kinetics of access to different metals can vary from sensor to sensor, for example due to interactions with specific donor molecules, including metallochaperones (1, 6 -8). Another part of the solution is that the allosteric mechanism connecting metal binding to DNA binding can be metal-selective (9 -12). Thus, a weaker binding metal can nonetheless be more effective at triggering the conformational changes that alter gene expression (10,13). For metal-dependent de-repressors and co-repressors the coupling free energy, ⌬G C metal-sensor⅐DNA , is typically larger for more effective metals (9). Unexpectedly, here we see how a metal can also become effective without increasing ⌬G C metal-sensor⅐DNA , if K DNA of the apo-form of a de-repressor is suitably weakened, to confer two mechanistic advantages in favor of Zn(II)-detection. Contrary to general dogma, here ⌬G C Zn(II)-sensor⅐DNA is smaller in the Zn(II)-sensing mutant relative to the nonsensing wild type protein.
In the course of a collaborative program to characterize the complement of metal sensors from Salmonella enterica serovar typhimurium strain SL1344 (hereafter referred to as Salmonella), we identified two genes encoding proteins with sequence similarity to members of the CsoR/RcnR family of DNA-binding and metal-responsive transcriptional de-repressors (14 -17). These are now shown to be Salmonella homo-logues of RcnR and FrmR. RcnR in Escherichia coli responds to cobalt and nickel, whereas CsoR, first discovered in Mycobacterium tuberculosis, responds to Cu(I) (15)(16)(17). Related metal sensors characterized from other bacteria detect the same metals (18 -27). Additionally, two homologues have been identified that respond to effectors other than metals, namely CstR from Staphylococcus aureus, which detects persulfide, plus E. coli FrmR (28 -30). CsoR forms a three helix bundle that assembles into tetramers (15). The sensory Cu(I) site exploits a conserved Cys-thiolate from the N-terminal end of helix ␣2 of one subunit in combination with an HXXXC motif from within helix ␣2Ј of a second subunit (Fig. 1A) (15). Three ligands in similar locations (with HXXXC replaced by HXXXH), along with additional ones from the N-terminal region of helix ␣1Ј, are recruited to the sensory metal site of RcnR (Fig. 1B) (17,31,32). A single residue variant of E. coli RcnR (H3E) also responds to Zn(II) (31).
In a global screen to discover the consequences of the readthrough of amber stop codons, E. coli FrmR (which has such a stop) emerged as the transcriptional repressor of the frmRAB operon (30). FrmA has formaldehyde dehydrogenase activity, and the operon was subsequently shown to respond to exogenous formaldehyde (30,33). This operon is de-repressed during anaerobic respiration using trimethylamine N-oxide as the terminal electron acceptor where endogenous formaldehyde is generated as a by-product of trimethylamine N-oxide demethylation (34). CO-releasing molecules and chloride treatments also trigger expression of the frm operon (35,36). There are no published studies of the Salmonella FrmR homologue. At least two potential metal ligands are retained in Salmonella FrmR, namely Cys at the N terminus of helix ␣2 but HXXXE (rather than HXXXH of paralogous Salmonella RcnR) at helix ␣2Ј (Fig.  1, A and B). Despite sequence similarity between FrmR and other CsoR/RcnR family members, whether or not (any)  (15,17). C, schematic representation of the frmRA operon (to scale) from Salmonella indicating the frmRA promoter, which includes a candidate FrmR-binding site and forms one strand of frmRAPro.
FrmRs de-repress gene expression in response to metals remains untested.
Recent studies have shown that relative affinity, relative allostery, and relative access determine the ability of metal sensors to respond selectively in vivo (1). This is exemplified by comparing metal affinities (K metal ) and metal-responsive allostery (⌬G C metal-sensor⅐DNA ) among multiple metal sensors, and for multiple metals in Synechocystis PCC 6803 (1,6,11,18). Thus, InrS responds to nickel in vivo and has a K Ni(II) that is substantially tighter than K Ni(II) of cobalt-sensing CoaR and Zn(II)-sensing ZiaR or Zur (a representative from each family of metal sensors that is present in this organism) (18). Provided the distribution of Ni(II) follows thermodynamic equilibrium predictions, as the [Ni(II)] rises, InrS will be the first to respond, de-repressing expression of nrsD (encoding a Ni(II)-efflux protein) and preventing [Ni(II)] from approaching K Ni(II) of the other sensors (18). For Zn(II), K Zn(II) of nickel-sensing InrS is similar to Zn(II)-sensing ZiaR, but crucially the allosteric mechanism of ZiaR is more responsive to Zn(II) compared with InrS (⌬G C Zn(II)-ZiaR⅐DNA Ͼ ⌬G C Zn(II)-InrS⅐DNA ) (11). Thus ZiaR will require a lower fractional Zn(II) occupancy than InrS to de-repress its target gene ziaA (encoding a Zn(II)-efflux ATPase). In this manner, ZiaR can prevent [Zn(II)] from exceeding the threshold where occupancy of DNA by InrS becomes sufficiently low for aberrant expression of nrsD to occur (11). In contrast, cobalt sensing does not correlate with relative affinity, and CoaR has the weakest K Co(II) in the set of sensors (6). There is evidence that the cobalt effector may be preferentially channelled to CoaR, and thus relative access has been invoked as the explanation for selective detection of cobalt (6). In summary, it is hypothesized that the sensor, which is triggered by a metal, is simply the most responsive within a cells' set of sensors, based upon relative affinity, relative allostery, and relative access (1). This hypothesis is now tested via a mutation conferring gain-of-metal sensing.
Here, the Co(II), Zn(II)-, and Cu(I)-binding affinities of Salmonella FrmR are determined and compared with equivalent data for the cognate sensors of these metals, namely Salmonella homologues of RcnR, Zn(II)-sensing ZntR and Zur, and Cu(I)sensing CueR. FrmR is found not to sense metals within cells, yet an E64H substitution (creating a Salmonella RcnR-like helix ␣2Ј HXXXH motif) gains responsiveness to cobalt and Zn(II) in vivo. By comparing the biochemical properties of Salmonella FrmR with FrmRE64H, and then relating these parameters to endogenous sensors for cobalt, Zn(II), and Cu(I), the relative properties that, in combination, enable metal sensing are identified.

Experimental Procedures
Bacterial Strains and DNA Manipulations-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). E. coli strain DH5␣ was used for routine cloning. Bacteria were cultured with shaking at 37°C in Luria-Bertani (LB) medium or M9 minimal medium (37), supplemented with thiamine (0.001% w/v) and L-histidine (20 g ml Ϫ1 ). Carbenicillin (100 g ml Ϫ1 ), kanamycin (50 g ml Ϫ1 ), and/or chloramphenicol (10 g ml Ϫ1 ) were added where appropriate. Cells were transformed to antibiotic resistance as described (37,38). All generated plasmid constructs were checked by sequence analysis. Primers are listed in supplemental Table S1.
Generation of Salmonella Deletion Mutants-Deletion derivatives of strain LB5010a were obtained using the Red method (38) using plasmid pKD3 and primers 1 and 2 for frmR or primers 3 and 4 for gshA. Mutagenesis was performed using strain LB5010a and selection of mutants achieved using LB medium supplemented with chloramphenicol. Mutations were subsequently moved to SL1344 or derivatives using P22 phage transduction and validated by PCR using primers 5 and 6 for frmR or primers 7 and 8 for gshA. The antibiotic resistance cassette from the ⌬frmR::cat locus was removed using the helper plasmid pCP20 carrying the FLP recombinase.
Generation of Promoter-lacZ Fusion Constructs and ␤-Galactosidase Assays-P frmRA or P frmRA -frmR was amplified from SL1344 genomic DNA using primer 9 and either primer 10 (for P frmRA ) or 11 (for P frmRA -frmR) and ligated into pGEM-T. Sitedirected mutagenesis to generate P frmRA -frmRE64H and P frmRA -frmR DOWN was conducted via the QuikChange protocol (Stratagene) using pGEM-P frmRA -frmR as template and primers 12-23. Codon optimization of the frmRE64H coding region to generate P frmRA -frmRE64H UP (supplemental Table  S2) was achieved using GeneArt Gene Synthesis (Life Technologies, Inc.) and optimization for Salmonella typhimurium. The rcnR-P rcnA region was amplified from SL1344 genomic DNA using primers 24 and 25. Digested fragments were cloned into the SmaI/BamHI site of pRS415 (39). P zntA cloned into pRS415 was provided by J. S. Cavet (University of Manchester). The resulting constructs were introduced into strain LB5010a prior to strain SL1344. ␤-Galactosidase assays were performed as described (40) in triplicate on at least three separate occasions. Overnight cultures were grown in M9 minimal medium, diluted 1:50 in fresh medium supplemented with maximum noninhibitory concentrations (MNIC 3 ; defined as the maximum concentration which inhibited growth by ϳ 10%) of metals, formaldehyde, EDTA, or N,N,NЈ,NЈ-tetrakis(2-pyridylmethyl)ethylenediamine (TPEN), and grown to mid-logarithmic phase prior to assays. For time course experiments, cells were grown to early logarithmic phase and statically cooled to 25°C for 20 min followed by a 2-h incubation in the presence of metal or formaldehyde. The metal salts used were MnCl 2 , C 6 ), and washed with 10 column volumes of the same buffer. ZntR and Zur were treated the same way, except using 5 mM NaCl, 1 mM EDTA, 5 mM DTT, 10 mM HEPES, pH 7.0, for ZntR, and 100 mM NaCl, 5 mM DTT, 1 mM EDTA, 10 mM HEPES pH 7.8 for Zur. FrmR, FrmRE64H, ZntR, and Zur were eluted in a single step using respective binding buffers plus 500 mM NaCl. RcnR was diluted to 100 mM NaCl, 10 mM EDTA, 10 mM DTT, 10 mM HEPES, pH 7.0, and applied to an equilibrated 5-ml HiTrap SP column (GE Healthcare), washed in the same buffer plus 200 mM NaCl, and eluted in 300 mM NaCl. CueR was expressed and purified as described previously (41). Anaerobic protein stocks were prepared by applying purified protein to a pre-equilibrated 1-ml HiTrap heparin column (diluting FrmR, FrmRE64H, ZntR, and Zur as described above and without dilution of RcnR for ZntR, Zur, RcnR, and CueR, respectively. It was noted that the absorbance spectra of FrmRE64H differed from FrmR (by exhibiting a shoulder at ϳ300 nm), except in two early preparations, and these were not used further. Reduced thiol and metal content were assayed as described previously (18), and all anaerobic protein samples (maintained in an anaerobic chamber) were Ն90% reduced and Ն95% metal-free, with the exception of Zur which contained ϳ1 M eq 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 as described previously (18). UV-visible Absorption Spectroscopy-Experiments were carried out in 100 mM NaCl, 400 mM KCl, and 10 mM HEPES, pH 7.0, for FrmR, FrmRE64H, ZntR, Zur, and CueR, with inclusion of 5% (v/v) glycerol for RcnR. Concentrations of metal stocks (CoCl 2 , NiCl 2 , CuCl, and ZnCl 2 ) were verified by ICP-MS. CuCl was prepared as described previously and confirmed to be Ͼ95% Cu(I) by titration against bathocuproine sulfonate (BCS) (42). CoCl 2 , NiCl 2 , or CuCl (Ͼ95% Cu(I)) were titrated into protein, or ZnCl 2 was titrated into protein pre-equilibrated with CoCl 2 , and the absorbance spectra were recorded at equilibrium using a 35 UV-visible spectrophotometer (Perkin-Elmer Life Sciences). Precipitation of ZntR was observed with further ZnCl 2 additions to Co(II)-ZntR than those shown.
Protein-Metal Migration by Size-exclusion Chromatography-FrmR, FrmRE64H, or Zur were incubated (60 min) with an excess of ZnCl 2 , CuCl (Ͼ95% Cu(I)) or EDTA (as stated) in 100 mM NaCl, 400 mM KCl, and 10 mM HEPES, pH 7.0, and an aliquot (0.5 ml) was resolved by size-exclusion chromatography (PD-10 Sephadex G25, GE Healthcare) in the same buffer conditions. Fractions (0.5 ml) were analyzed for metal by ICP-MS and protein by the Bradford assay using known concentrations of FrmR, FrmRE64H, or Zur as standards. Failure to recover all of the copper during experimentation with FrmR or FrmRE64H suggests (at least) some competition from and copper binding by the Sephadex matrix.
Protein-Chelator-Cu(I) Competitions-Experiments were carried out in 100 mM NaCl, 400 mM KCl, and 10 mM HEPES, pH 7.0. CuCl (Ͼ 95% Cu(I)) was titrated into a mixed solution of protein and BCA, and the absorbance at 562 nm was recorded at equilibrium. Data were fit to the models described in the figure legends and Table 1 footnotes using Dynafit to determine Cu(I)-binding constants (43). ␤ 2Cu(I) ϭ 1.58 ϫ 10 17 M Ϫ2 at pH 7.0 for BCA (48). For BCS, the absorbance at 483 nm was recorded following titration with CuCl (to generate a calibration curve) or following preincubation of BCS with CuCl (10 min) and addition of CueR. The absorbance at 483 nm was monitored to equilibrium. K Cu(I) of the tightest site of CueR was calculated using Equation 1 (48), CueR is expected to be a dimer with two metal-binding sites (49) that bind Cu(I) with negative cooperativity (41) (48). Fluorescence Spectroscopy-Experiments were carried out in 100 mM NaCl, 400 mM KCl, 10 mM HEPES, pH 7.0. ZnCl 2 was titrated into ZntR, and fluorescence emission spectra ( ex ϭ 280 nm, em ϭ 303 nm, and T ϭ 20°C) were recorded at equilibrium using a Cary Eclipse fluorescence spectrophotometer. Precipitation was observed with addition of more than 1.1 M eq ZnCl 2 .
Protein Quantification by Liquid Chromatography-Tandem Mass Spectrometry-Cellular lysates were prepared from logarithmic cultures grown in M9 minimal medium. Cell number was determined by enumeration on LB agar plates. Harvested cells were resuspended 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-m pore size), snap-frozen in liquid N 2 , stored at Ϫ80°C, and thawed on ice before use. Total protein was determined by the Bradford assay, using BSA as a standard. Purified stocks of FrmR or FrmRE64H were quantified by amino acid analysis (Proteomics Core Facility, University of California), stored at Ϫ80°C, and thawed on ice before dilution in PBS to 0.6 mg ml Ϫ1 . For standard curves, proteins were further diluted in soluble cell lysates from ⌬frmR cells to generate standard curve concentrations of 5, 10, 50, 250, 425, and 500 ng of 100 l Ϫ1 (which defined the limits for quantification). Aliquots were stored at Ϫ80°C. Working internal standards were prepared by dilution of labeled peptides ([ 13 C 6 , 15 N 4 ]arginine residues) GQVEALER[ 13 C 6 , 15 N 4 ], DEL-VSGETTPDQR[ 13 C 6 , 15 N 4 ], and DHLVSGETTPDQR[ 13 C 6 , 15 N 4 ] (Thermo Fisher) in 15% (v/v) acetonitrile with 0.1% (v/v) formic acid solution to obtain final concentrations of 313 fmol l Ϫ1 of each peptide. For experimental samples (soluble lysates from ⌬frmR cells containing the P frmRA -frmR construct and variants) and standard curve samples, 100 l was precipitated using 300 l of methanol (mixing at 2000 rpm for 1 min) before centrifugation (900 ϫ g, 5 min, room temperature). Pellets were suspended in 400 l of 200 mM NH 4 HCO 3 in 10% (v/v) methanol (mixing at 2000 rpm for 10 min) and 10 l internal standard added. Pellet digestion was performed with 10 l of trypsin (14 mg ml Ϫ1 ) and mixing (1000 rpm, 37°C, 16 h) and stopped with 10 l of 15% (v/v) formic acid. The digested samples were centrifuged (6000 ϫ g for 5 min at room temperature) to remove particulate material. Solvent was removed from clarified supernatants (50 -100 l) using a centrifugal evaporator (Thermo Scientific SpeedVac system). Samples were separated by gradient elution at 0.3 ml min Ϫ1 using a Zorbax Eclipse Plus C18 column (2.1 ϫ 150 mm, 3.5-m particles; Agilent Technologies) at 30°C. Mobile phase A and B consisted of 0.1% (v/v) formic acid in water and 0.1% (v/v) formic acid in acetonitrile, respectively. Aliquots (20 l) were applied to a 6500 triple quadrupole mass spectrometer (AB Sciex) operating in positive ionization mode. Acquisition methods used the following parameters: 5500 V ion spray voltage; 25 p.s.i. curtain gas; 60 p.s.i. source gas; 550°C interface heating temperature; 40 V declustering potential; 26 V collision energy; and 27 V collision cell exit potential. Scheduled multiple reaction monitoring was carried out with a 90-s multiple reaction monitoring detection window and 1.00-s target scan time. A quadratic 1/x 2 weighted regression model was used to perform standard calibration.
The coefficient of determination (R 2 ) was Ͼ0.990 for GQV-EALER in all validation runs.
Determining Intracellular [Glutathione]-Intracellular glutathione was measured using a glutathione assay kit (Sigma) according to the manufacturer's instructions. Cellular lysates were prepared from overnight cultures grown in M9 minimal medium, diluted 1:50 in fresh medium, and grown to early logarithmic phase, statically cooled to 25°C for 20 min, followed by 30-min incubation in the absence or presence of MNIC ZnSO 4 . Viable cells were enumerated on LB agar, and [glutathione] was calculated using a cell volume of 1 fl.
Fluorescence Anisotropy-Complementary single-stranded oligonucleotides 38 (hexachlorofluorescein-labeled) and 39 (containing the identified FrmR-binding site and flanking nucleotides, Fig. 1C) were annealed by heating 10 or 200 M of each strand in 10 mM HEPES, pH 7.0, 150 mM NaCl to 95°C, and cooled to room temperature overnight. For protein-DNA stoichiometry experiments, the fluorescently labeled and annealed probe (designated frmRAPro) was diluted to 2.5 M in 10 mM HEPES, pH 7.0, 60 mM NaCl, 240 mM KCl, and 5 mM EDTA and titrated with FrmR or FrmRE64H prepared in 100 mM NaCl, 400 mM KCl, 10 mM HEPES, pH 7.0, and 5 mM EDTA. For K DNA determination, frmRAPro was diluted to 10 nM, with addition of 5 mM EDTA or 5 M ZnCl 2 as required. FrmR or FrmRE64H was prepared as above with inclusion of 5 mM EDTA or 1.2 M eq of ZnCl 2 or CuCl (Ͼ95% Cu(I)) as appropriate. Changes in anisotropy (⌬r obs ) were measured using a modified Cary Eclipse fluorescence spectrophotometer (Agilent Technologies) fitted with polarizing filters ( ex ϭ 530 nm, em ϭ 570 nm, averaging time ϭ 20 s, replicates ϭ 5, and T ϭ 25°C) as described previously (11). Upon each addition, the cuvette was allowed to equilibrate for 5 min before recording data. Data were fit to the model described in the figure legends and Table 2 footnotes using Dynafit (43). For experiments with Cu(I)-or Zn(II)-FrmR or FrmRE64H, where DNA binding did not saturate, the average fitted ⌬r obs maximum value from apoprotein experiments was used in the script. The coupling free energy ⌬G C , linking DNA binding to metal binding, was calculated as described previously (11) using the following: ⌬G C ϭ ϪRTlnK C , where r ϭ 8.314 J K Ϫ1 mol Ϫ1 (gas constant), T ϭ 298.15 K (temperature at which experiment was conducted), and K C ϭ K DNA metal-protein /K DNA apoprotein (9). Mean ⌬G C values (and standard deviations) were calculated from the full set of (equally weighted) possible pairwise permutations of K C .
Fractional Occupancy Models-Fractional occupancy of the tightest metal-binding site of a sensor with metal as a function of buffered [metal], was determined using the following: () ϭ [metal] buffered /(K metal ϩ [metal] buffered ). K metal ϭ K D (tightest site) of sensor for metal, experimentally determined (K metal sensor ) ( Table 1) (48). For FrmR (and variants), K metal was additionally calculated for the DNA-bound form (K metal sensor⅐DNA ) from the coupling constant (K C ) (Fig. 10E). The concentration of apoand Zn(II)-protein at a given [Zn(II)] was calculated using the number of tetramers per cell (FrmR and variants; Fig. 9K), and a cell volume of 1 fl. Fractional DNA occupancies with apo-and Zn(II)-protein over a range of protein concentrations were modeled using Dynafit (43) (1:1 binding of tetramer/DNA; assuming the binding of one tetramer conferred repression) with K DNA (from Table 2) and [P frmRA ] as fixed parameters (sample Dynafit script is also shown in the supplemental material). [P frmRA ] was calculated assuming 15 copies cell Ϫ1 (due to the presence on low copy number reporter plasmid) and a cell volume of 1 fl. The response was set at 1/[P frmRA ]. The fractional occupancy of P frmRA with apo-and Zn(II)-protein was summed to give fractional occupancy of P frmRA at any given buffered [Zn(II)].

Results
CsoR/RcnR-like Repressor FrmR Solely Detects Formaldehyde and Not Metals-Despite similarity between FrmR and metalsensing transcriptional de-repressors, exposing Salmonella cultures to maximum noninhibitory concentrations of MnCl 2 , C 6 H 5 FeO 7 , CoCl 2 , NiSO 4 , CuSO 4 or ZnSO 4 does not de-repress expression from P frmRA -frmR fused to lacZ in ⌬frmR cells ( Fig. 2A). Exposure of cells to MNIC of formaldehyde does derepress expression from P frmRA -frmR ( Fig. 2A). The formaldehyde response was lost, and basal expression was elevated in cells harboring a similar construct (P frmRA ) devoid of frmR (Fig.  2, B and C). Thus, in common with E. coli FrmR (30), the Salmonella homologue represses expression from the frmRA operator-promoter with repression alleviated by formaldehyde, and here we show that repression by Salmonella FrmR is not alleviated by metals.
Substitution of FrmR Glu-64 for an RcnR Metal-Ligand Confers Zn(II) and Cobalt Detection in Cells-Replacement of FrmR residue 64 (glutamate) with histidine (a metal-ligand in RcnR; Fig. 1, A and B) generates a metal-sensing variant of FrmR (Fig.  2D). Repression is alleviated by CoCl 2 and ZnSO 4 in ⌬frmR cells containing P frmRA -frmRE64H (but not P frmRA -frmR) fused to lacZ (Fig. 2, D and E). MnCl 2 , C 6 H 5 FeO 7 , NiSO 4 , and CuSO 4 did not affect expression from P frmRA -frmRE64H (or P frmRA -frmR), although formaldehyde responsiveness was retained. Notably, metal-responsive family members RcnR and CsoR do respond to nickel and copper (15)(16)(17). In summary, a single residue change that mimics the metal-sensing site of RcnR is sufficient to create a detector of cellular Zn(II) and cobalt.
FrmRE64H and FrmR Both Bind Co(II), Cu(I), and Zn(II)-It was anticipated that the introduced histidine residue created a metal-binding site in FrmR. However, titration of FrmRE64H or FrmR with Co(II) results in the appearance of spectral features in the region of 330 nm, indicative of S3 Co(II) ligand-tometal charge transfer (LMCT) bands consistent with Co(II) binding to both proteins (Fig. 3, A and E). For FrmR and FrmRE64H, the intensities of the feature at saturation ϳ0.9 ϫ 10 3 M Ϫ1 cm Ϫ1 are consistent with a single thiolate ligand (50). The intensities of a second set of Co(II)-dependent features in the region of 600 nm, indicative of d-d transitions (50), suggest tetrahedral coordination geometry. Binding curves are linear up to 1 eq of Co(II), implying K Co(II) is too tight to estimate by this method (Fig. 3, A and E, insets). Cu(I)-dependent features similarly indicate tight binding of at least 1 eq of metal, and 1 eq of Cu(I) binds sufficiently tightly to co-migrate with either protein during size exclusion chromatography (Fig. 3, B, C, F, and G). One equivalent of Zn(II) (which is spectrally silent) also co-migrates with each protein during size exclusion chromatography (Fig. 3, D and H). Preliminary Ni(II)-binding experiments with FrmRE64H were ambiguous, but because no in vivo nickel response had been detected for FrmRE64H, Ni(II) affinities were not pursued.
Determination of K Zn(II) , K Co(II) , and K Cu(I) for FrmRE64H and FrmR-As the FrmRE64H variant, but not FrmR, responds to Zn(II) and cobalt in cells, it was anticipated that this substitution had succeeded in tightening the affinity for these metals. The chromophores mag fura-2 and quin-2 form 1:1 complexes with Zn(II) and undergo concomitant changes in absorbance upon metal binding, which can be used to monitor competition with proteins and hence to estimate protein K Zn(II) (11, 13, 44 -48, 51). Titration of 10.1 M or 12.2 M mag fura-2 with Zn(II) in the presence of FrmRE64H (18.8 M, monomer) or FrmR (20.4 M, monomer), respectively, gave negligible change in absorbance up to 0.5-0.75 eq of Zn(II) per protein monomer, implying competition with the chromophore for metal (Fig. 4,  A and B). At these protein concentrations, CsoR/RcnR family members exist as tetramers with four metal-binding sites per tetramer, and with some evidence of negative cooperativity between sites (12,17,18,52). A 1:1 stoichiometry equating to four Zn(II) per tetramer was observed for both FrmRE64H and FrmR (Fig. 3, D and H), but the fourth sites are too weak to compete with mag fura-2 (hence competition is complete after addition of ϳ24.2 and ϳ27.5 M Zn(II) (Fig. 4, A and B)). Data were fit to models describing tight binding of 3 M eq of Zn(II)/ tetramer, with dashed lines representing simulated curves describing K Zn1-3 10-fold tighter or 10-fold weaker than the calculated affinity (Fig. 4, A and B). For both proteins, this suggests K Zn1-3 at or approaching the tighter limit of the assay using mag fura-2 (K Zn(II) mag fura-2 ϭ 2.0 ϫ 10 Ϫ8 M). Competitions were therefore conducted with 13.4 or 14.1 M quin-2 (K Zn(II) quin-2 ϭ 3.7 ϫ 10 Ϫ12 M) and FrmRE64H (42.7 M, monomer) or FrmR (39.9 M, monomer), respectively (Fig. 4, C and  D). Again, data were fit to models describing binding of 3 M eq of Zn(II)/tetramer (as expected, the fourth sites did not show competition with quin-2) with dashed lines in Fig. 4, C and D, describing simulated curves for K Zn1-3 10-fold tighter or 10-fold weaker than the calculated affinity of the proteins. Mean values of K Zn1-3 2.33 (Ϯ0.3) ϫ 10 Ϫ11 M and 1.7 (Ϯ0.7) ϫ 10 Ϫ10 M for FrmRE64H and FrmR, respectively, are thus within the range of this assay (Fig. 4, C and D, and Table 1).
Cuprous affinities of both proteins were determined using BCA (␤ 2 ϭ 10 17.2 M Ϫ2 (48)) revealing competition in each case for 2 M eq of Cu(I) per monomer, but with greater competition and hence tighter affinity for FrmRE64H than FrmR (Table 1 and Fig. 4, I and J). The data were fit to models describing binding of eight Cu(I) ions per tetramer (see Table 1 footnotes for details), which for FrmR depart from simulated curves describing binding of the tightest two Cu(I) ions (K Cu1-2 ) 10-fold tighter and 10-fold weaker than the fitted value (Fig. 4I), giving FrmR K Cu1-2 ϭ 4.9 (Ϯ1.6) ϫ 10 Ϫ15 M (Table 1). In contrast, K Cu1-2 for FrmRE64H is too tight to measure by this assay. However, FrmRE64H does not significantly compete with 10 M BCS (␤ 2 ϭ 10 19.8 M Ϫ2 ) (48), with saturation of the BCS 2 Cu(I) complex observed at ϳ5 M CuCl (Fig. 4K). These data imply that FrmRE64H K Cu1-2 can only marginally depart from the value estimated using BCA (K Cu1-2 ϳ5 ϫ 10 Ϫ16 M) ( Table 1). It is noted that the final absorbance for the BCS 2 Cu(I) complex in the presence of protein was lower than predicted from its known extinction coefficient; hence, the possibility of a ternary complex cannot be ruled out. In summary, the two metals that FrmRE64H now detects, Co(II) and Zn(II), bind approximately an order of magnitude more tightly than to FrmR (Table 1).
Cognate K metal of Salmonella Zn(II), Cobalt, and Cu(I) Sensors ZntR, Zur, RcnR, and CueR-If metal sensing is dictated by relative affinity within the set of Salmonella metal sensors, the affinity of FrmRE64H for Zn(II) and Co(II) would need to become comparable with cellular sensors for these metals. Conversely, Cu(I) affinity would need to remain weaker than Cu(I)-sensing CueR making Cu(I) still undetectable (40,41,54). The Salmonella sensors for Zn(II) and Co(II) are confirmed here as ZntR, Zur, and RcnR (Fig. 5) (55,56). Expression is induced from P zntA and P rcnA in wild type cells exposed to MNIC ZnSO 4 and CoCl 2 , respectively (Fig. 5, A and B). Notably minimal media for this strain (SL1344) require histidine that may influence Ni(II) availability. Titration of ZntR with Co(II), as a spectral probe for Zn(II)-binding sites, generated features diagnostic for LMCTs and d-d transitions consistent with ϳ3 thiolate-Co(II) bonds per ZntR monomer and tetrahedral coordination geometry (Fig. 5C) (50). These features saturate at ϳ1 eq of Co(II) and are bleached by addition of ϳ1 eq of Zn(II) (Fig.  5D). Zn(II) (ϳ1 eq) also quenched ZntR auto-fluorescence (Fig.  5E). Salmonella ZntR is expected to be a dimer based on similarity to the E. coli homologue (49), implying a stoichiometry of two Zn(II) ions per dimer. Titrations of 18.6 M quin-2 and ZntR (16.0 M, monomer) with Zn(II) were fit to models describing detectable binding of two distinguishable Zn(II) ions per dimer (K Zn1 and K Zn2 ); the estimated mean values are shown in Table 1 (Fig. 6A). The optimized curves depart from simulated curves describing K Zn1 or K Zn2 10-fold tighter or 10-fold weaker than their fitted values, although K Zn2 does approach the simulated curve describing K Zn2 as 10-fold weaker (Fig. 6A). It remains possible that a higher Zn(II) stoichiometry may be achieved for Salmonella ZntR under some conditions (as observed for E. coli ZntR (49)). Importantly, we show here that ZntR binds only two Zn(II) ions per dimer with sufficient affinity to compete with quin-2.
Zur from Salmonella, and in common with other bacteria (57-60), contains a structural Zn(II) ion that remains associated with the protein (20 M, monomer) in the presence of excess (1 mM) EDTA (Fig. 5F). In the absence of EDTA, at least one further equivalent of Zn(II) binds sufficiently tightly to comigrate with the protein during size exclusion chromatography (Fig. 5F). Titration of apo-Zur (Zn(II)-saturated at the structural site) with Co(II) generated features diagnostic for LMCTs and d-d transitions consistent with two to three coordinating thiol groups, which saturate between 1.5 to 2 eq of Co(II) per monomer (Fig. 5G) (50). These features are bleached with addition of 1.5 to 2 eq of Zn(II) (Fig. 5H). Zur family members exist as dimers (57)(58)(59), and here data show there are at least three exchangeable sites per dimer that are accessible to both Co(II) and Zn(II). A total of 35.5 M Zn(II) is required to fully saturate Zur (11.7 M, monomer) and mag fura-2 (12.1 M), consistent with two monomer equivalents ((2 ϫ 11.7 M) ϩ 12.1 M ϭ 35.5 M) of exchangeable Zn(II) binding to Zur (І four sites per dimer) with sufficient affinity to show some competition with mag fura-2. Of these, an estimated three sites per dimer com-pletely withhold Zn(II) from mag fura-2 (Fig. 5I). The data in Fig. 5I were fit to a model describing four exchangeable sites per Zur dimer with dashed lines representing simulated curves describing K Zn4 10-fold tighter and 10-fold weaker than the fitted K Zn4 value, and a tighter limit for K Zn4 was estimated from replicate titrations (Table 1). To estimate K Zn1-2 and K Zn3 , quin-2 (9.6 M) and Zur (13.7 M, monomer) were titrated with Zn(II) and fit to models describing competition from 1.5 eq of Zn(II) per monomer (exchangeable sites 1-3 per dimer, but not site 4) with mean values for K Zn1-2 and K Zn3 shown in Table 1 (Fig. 6B). The optimized curve departs from simulated curves describing K Zn1-2 as 10-fold tighter or 10-fold weaker than the fitted value. K Zn3 departs from a simulated curve describing K Zn3 as 10-fold tighter, but it approaches a simulated curve describing K Zn3 as 10-fold weaker (Fig. 6B).
Titration of RcnR with Ni(II) or Co(II) generated spectral features that saturated at 1 eq of metal (Fig. 5, J and K). Ni(II)-RcnR demonstrated features Ͻ300 nm and weak d-d transitions consistent with a six coordinate octahedral Ni(II)-binding site, as seen for E. coli RcnR (17). An additional Co(II)-dependent feature at 314 nm appeared with time (Fig. 5L). Co(II)-dependent fluorescence quenching of fura-2 (13.2 M) in the presence of RcnR (18.4 M, monomer) was fit to a model describing competition from three sites per RcnR tetramer with two sites (K Co1-2 ) tighter than the third (K Co3 ) (Fig. 6C). The optimized curve departs from simulated curves describing K Co1-2 as 10-fold tighter or 10-fold weaker and K Co3 as 10-fold tighter than the respective fitted values. Mean values (generated from multiple titrations) for K Co1-2 and a range for K Co3 are shown in Table 1.

Sensor
Metal a Data were fit to a model describing Co(II) binding with equal affinity to four sites (K Co1-4 ) on an FrmR or FrmRE64H tetramer, determined by competition with BisTris (n ϭ 3). A weaker limit is defined for FrmRE64H. b Data were fit to a model describing Zn(II) binding with equal affinity to the first three sites (K Zn1-3 ) on an FrmR or FrmRE64H tetramer, determined by competition with quin-2 (n ϭ 3). c Data were fit to a model describing Cu(I) binding with equal affinity to the first two sites (K Cu1-2 ), with equal affinity to sites 3 and 4 (K Cu3-4 ), and with equal affinity to sites 5-8 (K Cu5-8 ) on an FrmR tetramer (with K Cu1-2 Ͻ K Cu3-4 Ͻ K Cu5-8 ), determined by competition with BCA (n ϭ 4). A tighter limit is defined for FrmR K Cu5-8 . d Fit to a model describing Co(II) binding to the first site (K Co1 ) on an FrmRE64H tetramer, determined by competition with fura-2 (n ϭ 3). e Data were fit to a model describing Cu(I) binding with equal affinity to the first two sites (K Cu1-2 ), with equal affinity to sites 3 and 4 (K Cu3-4 ), with equal affinity to sites 5 and 6 (K Cu5-6 ), and with equal affinity to sites 7 and 8 (K Cu7-8 ) on an FrmRE64H tetramer (with K Cu1-2 Ͻ K Cu3-4 Ͻ K Cu5-6 Ͻ K Cu7-8 ), determined by competition with BCA (n ϭ 4). A tighter limit is defined for FrmRE64H K Cu7-8 . f Approximation reflects the fact that sites 1 and 2 on an FrmRE64H tetramer outcompete BCA for Cu(I) but fail to compete with BCS (although the formation of a ternary complex cannot be ruled out). g Data were fit to a model describing Zn(II) binding to three sites (K Zn1-2 and K Zn3 ) on a Zur dimer (with the structural site already filled) with equal affinity to the first two sites, (K Zn1-2 ) and K Zn1-2 Ͻ K Zn3 , determined by competition with quin-2 (n ϭ 3). h Data were fit to a model describing Zn(II) binding to the fourth site (K Zn4 ) on a Zur dimer (with the structural site already filled), determined by competition with mag fura-2 (n ϭ 3). Only a tighter limit can be determined. i Data were fit to a model describing Zn(II) binding to two sites (K Zn1 and K Zn2 ) on a ZntR dimer (K Zn1 Ͻ K Zn2 ), determined by competition with quin-2 (n ϭ 3). j Data were fit to a model describing Co(II) binding to three sites (K Co1-2 and K Co3 ) on an RcnR tetramer with equal affinity to the first two sites (K Co1-2 ) and K Co1-2 Ͻ K Co3 , determined by competition with fura-2 (n ϭ 3). k Range represents the fact that RcnR exhibits linear absorbance features upon titration with Co(II) to 1 M eq per monomer, but site 3 does not sufficiently complete with fura-2 for Co(II). l Data were determined by competition with BCS (n ϭ 6) and describing binding of Cu(I) to the first site (K Cu1 ) on a CueR dimer.
Salmonella CueR out-competes a 10-fold molar excess of BCS (41), and here a 100-fold and then a 75-fold excess of BCS (the latter in Fig. 6D) were used to estimate K Cu1 (Table 1). In summary, the tightest exchangeable sites of the endogenous metal sensors are tighter for their cognate metals than either FrmR or FrmRE64H, in every case (Table 1). However, the dif-  Solid line describes competition from ZntR for 1 M eq of Zn(II) per monomer (two independent sites per dimer; K Zn1 and K Zn2 ). Dashed lines describe K Zn1 10-fold tighter and 10-fold weaker than fitted K Zn1 (K Zn2 fixed to fitted K Zn2 ). Dotted lines describe K Zn2 10-fold tighter and 10-fold weaker than fitted K Zn2 (K Zn1 fixed to fitted K Zn1 ). B, representative (n ϭ 3) quin-2 absorbance upon titration of quin-2 (9.6 M) and Zur (13.7 M, monomer) with ZnCl 2 . Solid line describes competition from Zur for 1.5 M eq of Zn(II) per monomer (three sites per dimer with two independent binding events: K Zn1-2 , and K Zn3 ). Dashed lines describe K Zn1-2 10-fold tighter and 10-fold weaker than fitted K Zn1-2 (K Zn3 fixed to fitted K Zn3 ). Dotted lines describe K Zn3 10-fold tighter and 10-fold weaker than the fitted K Zn3 (K Zn1-2 fixed to fitted K Zn1-2 ). C, representative (n ϭ 3) fura-2 fluorescence emission upon titration of fura-2 (13.2 M) and RcnR (18.4 M, monomer) with CoCl 2 . Solid line describes competition from RcnR for 0.75 M eq of Co(II) per monomer (three sites per tetramer, with two independent binding events: K Co1-2 and K Co3 ). Dashed lines describe K Co1-2 10-fold tighter and 10-fold weaker than the fitted K Co1-2 (K Co3 fixed to fitted K Co3 ). Dotted lines describe K Co3 10-fold tighter and 10-fold weaker than the fitted K Co3 (K Co1-2 fixed to fitted K Co1-2 ). ference in K Zn(II) between FrmRE64H and cognate Zn(II) sensors is the smallest.
Cognate Metal Sensors Out-compete FrmR for Metal-To confirm, or otherwise, that FrmR K metal is weaker than CueR K Cu(I) , ZntR K Zn(II) , and RcnR K Co(II) , pairwise competitions were conducted for the tightest metal-binding site in which metallated FrmR was incubated with apo-forms of the respective sensors. Cu(I)-FrmR co-migrates with copper following heparin affinity chromatography (Fig. 7A). However, after mixing Cu(I)-FrmR with apo-CueR (which can be differentially resolved), copper migrates with CueR (Fig. 7A). Likewise after mixing Zn(II)-FrmR with apo-ZntR, Zn(II) predominantly migrates (using different fractionation buffers to those in Fig.  7A) with ZntR (Ͼ 90% of control) (Fig. 7B). Diagnostic spectral features (d-d transitions) that discern Co(II)-FrmR, with tetrahedral binding geometry, from Co(II)-RcnR, with octahedral binding geometry, are lost upon addition of apo-RcnR to Co(II)-FrmR (Fig. 7C). Thus, in every case the cognate sensor out-competes FrmR confirming that FrmR K metal is weaker. Relative (to the cognate sensors) metal affinity could account for why wild type FrmR does not respond to metals within cells. Fig. 8A compares the calculated fractional occupancies of the tightest exchangeable sites (from K metal in Table 1) of FrmR and FrmRE64H for Zn(II), Cu(I), and Co(II) with the respective cognate Salmonella sensors, as a function of metal concentration. To detect Cu(I), FrmRE64H would require intracellularly buffered Cu(I) concentrations to rise ϳ3 orders of magnitude higher than necessary for detection by CueR, which could explain why FrmRE64H remains unresponsive to Cu(I). In contrast, partial Zn(II) occupancy of FrmRE64H will occur at Zn(II) concentrations below those required to saturate ZntR (Fig. 8A). Thus, theoretically, a 10-fold increase in K Zn(II) of FrmRE64H relative to FrmR may be sufficient to enable some Zn(II) detection within the cell.
Glutathione Enhances Metal Detection by FrmRE64H and RcnR-In addition to responding to Zn(II), the FrmRE64H variant also responds to cellular cobalt (Fig. 2D), yet K Co(II) for FrmRE64H is ϳ500-fold weaker than the endogenous cobalt sensor RcnR (Fig. 8A and Table 1). An ϳ10-fold increase in K Co(II) alone cannot readily explain why this variant of FrmR has become responsive to cobalt. Recent studies of the complement of metal sensors from a cyanobacterium concluded that the detection of Zn(II) and nickel matched predictions based upon equilibrium thermodynamics, but this was untrue for cobalt (6,11,18). In that system, a substantial kinetic component was invoked for the preferential distribution of cobalt to the cobalt sensor and away from sensors for other metals (6).
The possibility that glutathione is required for the detection of cobalt (and Zn(II)) by FrmRE64H was investigated in ⌬frmR/ ⌬gshA cells containing P frmRA -frmRE64H fused to lacZ (Fig. 8,  B and C). Cells lacking glutathione showed a negligible response to either metal. Previous studies of Zn(II) sensors have found that the low molecular weight thiol, bacillithiol, competes for metal thus reducing responses (61). ZntR-mediated expression in response to Zn(II) from the zntA promoter shows negligible difference in ⌬gshA cells compared with wild type (Fig. 8D). However, in common with regulation by FrmRE64H, the response of RcnR to cobalt was also reduced, but not lost, in cells missing glutathione (Fig. 8E). Thus, glutathione aids the   Table 1. B, ␤-galactosidase activity in ⌬frmR (filled circles) or ⌬frmR/⌬gshA (open circles) containing P frmRA -frmRE64H following exposure (2 h) of logarithmic cells to CoCl 2 . C, as B but with ZnSO 4 instead of CoCl 2 . D, ␤-galactosidase activity in Salmonella (wild type, defined earlier) (filled circles) or ⌬gshA (open circles) containing P zntA grown as described in C. E, ␤-galactosidase activity in wild type (filled circles) or ⌬gshA (open circles) containing rcnR-P rcnR following growth in conditions described in B. detection of cobalt by two different sensors but has varied effects on Zn(II) sensing.
Basal Repression by FrmRE64H Is Less than by FrmR-The tightening of K Zn(II) (and K Co(II) ) is modest suggesting that additional factors might contribute to the gain-of-metal detection by FrmRE64H (Figs. 2D and 8A and Table 1). It was noted that basal expression from the frmRA promoter is greater in cells containing FrmRE64H compared with wild type FrmR (Fig. 2, D  and E). Expression remains elevated in cultures treated with EDTA or the Zn(II) chelator TPEN, implying that this is not a response to basal levels of intracellular metal (Fig. 9, A and B). As a control, ZntR-mediated ␤-galactosidase expression from the zntA promoter does decline upon equivalent treatment with EDTA or TPEN (Fig. 9, C and D). Furthermore, because metal responsiveness from P frmRA -frmRE64H is affected by glutathione (Fig. 8, B and C), glutathione levels were measured but found not to be significantly altered between ⌬frmR cells expressing P frmRA -frmR or P frmRA -frmRE64H in either the presence (3.8 (Ϯ0.5) and 4.5 (Ϯ0.6) mM, respectively) or absence (4.4 (Ϯ0.8) and 3.3 (Ϯ0.4) mM, respectively) of added Zn(II).
Codon Optimization or De-optimization Alters FrmRE64H or FrmR Cell Ϫ1 but Does Not Switch Metal Perception-Loss of repression by FrmRE64H compared with FrmR could in theory be due to reduced protein abundance, for example due to impaired stability of the mutant protein. To test this suggestion, constructs were generated in which FrmRE64H codons were optimized for efficient translation (62,63), designated P frmRA -frmRE64H UP . Conversely, FrmR expression was de-optimized by introduction of rare arginine codons (62,63), designated P frmRA -frmR DOWN . This approach was chosen to alter abundance of the proteins while preserving the transcriptional architecture. Basal expression was enhanced in cells containing frmR DOWN and reduced in cells containing frmRE64H UP relative to the respective controls and yielding matched levels of basal lacZ expression by frmR DOWN versus frmRE64H UP (Fig.  9E). Moreover, the numbers of FrmRE64H and FrmR tetramers per cell, as determined by quantitative mass spectrometry, were indeed increased and decreased, respectively, in cells harboring the codon-altered variants (Fig. 9, G-K). Cells containing any of the variants, frmRE64H, frmRE64H UP , frmR, frmR DOWN , all showed enhanced expression following exposure to MNIC of formaldehyde, but crucially only the strains expressing FrmRE64H responded to Zn(II) and cobalt (Fig. 9F). Notably, the abundance of FrmRE64H is no less than FrmR (Fig. 9K), and an alternative explanation is needed for elevated basal expression in cells containing FrmRE64H.
⌬G C Zn(II)-FrmRE64H⅐DNA Is Less than ⌬G C Zn(II)-FrmR⅐DNA with Apo-FrmRE64H K DNA Being Weaker-Fluorescence anisotropy was used to monitor interactions between either FrmRE64H or FrmR and a fluorescently labeled double-stranded DNA fragment of the target operator-promoter, frmRAPro (Fig. 1C). DNA-protein stoichiometry was first determined by monitoring DNA binding to a relatively high concentration of frmRAPro (2.5 M) with saturation observed at ϳ20 M FrmRE64H or FrmR (monomer) consistent with binding of two tetramers (Fig. 10, A and B). A limiting concentration of frm-RAPro (10 nM) was subsequently titrated with apo-or Zn(II)saturated FrmRE64H or FrmR in the presence of 5 mM EDTA or 5 M Zn(II), respectively, and anisotropy data were fitted to models describing the binding of two nondissociable proteintetramers per DNA molecule (Fig. 10, C and D). The calculated FIGURE 9. Basal expression from P frmRA -frmRE64H is higher than P frmRA -frmR. ␤-Galactosidase activity in ⌬frmR containing P frmRA -frmR (filled circles) or P frmRA -frmRE64H (open circles) following growth to early exponential phase in the presence of EDTA (A) or TPEN (B). C, expression from P zntA in wild type Salmonella, grown as described in A, or D, as described in B. E, expression in ⌬frmR containing P frmRA -frmR (white bars), P frmRA -frmR DOWN (dashed white bars), P frmRA -frmRE64H (gray bars), or P frmRA -frmRE64H UP (dashed gray bars) following growth to early exponential phase, and F, following exposure (2 h) to Zn(II), Co(II) or formaldehyde, or untreated control. G-J, multiple reaction monitoring, quantitative MS of cell extracts. Representative (n ϭ 3) extracted LC-MS chromatograms of ion transitions detected in ⌬frmR containing P frmRA -frmR (G), P frmRA -frmR DOWN (H), P frmRA -frmRE64H (I), or P frmRA -frmRE64H UP (J). Transitions 451. 24 DNA binding affinities (n Ն 3) are shown in Table 2. K DNA was similarly determined for Cu(I)-FrmR (Table 2), but weak K Co(II) precluded equivalent K DNA estimations for Co(II)-saturated proteins ( Table 1). Metal binding weakens DNA binding, but unexpectedly this is true of wild type FrmR as well as FrmRE64H.
The degree to which metal binding allosterically inhibits DNA binding has previously been expressed as the coupling free energy (⌬G C ) calculated from the ratio of K DNA of apo-and holo-proteins and using a standard thermodynamic function (see Refs. 9,11,64 and the footnotes to Table 2). This yields Zn(II)-FrmR ⌬G C ϭ ϩ2.03 (Ϯ0.08) kcal mol Ϫ1 (⌬G C Zn(II)-FrmR⅐DNA ) and Zn(II)-FrmRE64H ⌬G C ϭ ϩ1.24 (Ϯ0.16) kcal mol Ϫ1 (⌬G C Zn(II)-FrmRE64H⅐DNA ) ( Table 2). Unexpectedly, this approach revealed that Zn(II) is less, not more, allosterically effective when binding to FrmRE64H than to FrmR, with the former having the smaller coupling free energy. However, inspection of the DNA binding curves (Fig. 10, C and D), and K DNA values ( Table 2), reveals that this results from apo-FrmRE64H having a weaker DNA affinity than apo-FrmR. These data explain the loss of basal repression by FrmRE64H. Importantly, despite a lesser ⌬G C , because K DNA of Zn(II)-FrmRE64H is not tighter than Zn(II)-FrmR (Table 2), at equivalent Zn(II) saturation DNA occupancy by FrmRE64H will still be less than FrmR, in effect rendering FrmRE64H more sensitive to de-repression. Moreover, assuming a closed system, coupled thermodynamic equilibria infer that any effect of metal binding on K DNA is reciprocated in an effect of DNA binding on K Zn(II) (Fig. 10E) (9,61,65,66). Thus a smaller ⌬G C Zn(II)-FrmRE64H⅐DNA means an even tighter K Zn(II) for DNAbound FrmRE64H relative to FrmR. The inferred K Zn(II) sensor⅐DNA (on-DNA) is 5.3 ϫ 10 Ϫ9 and 1.9 ϫ 10 Ϫ10 M for FrmR and FrmRE64H, respectively. A weaker K DNA thereby contributes in two ways to the mechanism enabling metal perception by the FrmRE64H variant, and overall, a tighter K Zn(II) plus a weaker K DNA act in combination to confer Zn(II) sensing.

Discussion
Substitution of one amino acid has created a metal sensor from the formaldehyde-responsive, DNA-binding transcriptional de-repressor FrmR (Fig. 2). Contrasting the biochemical properties of these two proteins (FrmR and FrmRE64H), along with endogenous Salmonella metal sensors (Figs. 3-7, 8A, and 10, A and B, and Tables 1 and 2), demonstrates what is required for metal sensing within cells. These data test (by gain-of-func- Data were fit to a model describing a 2:1 protein tetramer (nondissociable):DNA stoichiometry (binding with equal affinity), and lines represent simulated curves produced from the average K DNA determined across the experimental replicas shown. E, coupled thermodynamic equilibria (assuming a closed system) describing the relationship between FrmR tetramer (P), Zn(II) (Z), and P frmRA (D) (9,65,66). The coupling constant (K C ) is determined from the ratio K 4 /K 3   tion) theories that have been developed from correlations between the biochemical properties of various metal sensor proteins and the metals they detect (1,2,6,9,11,18). The single residue change in FrmRE64H tightens K Zn(II) by ϳ10-fold and weakens apo-K DNA by ϳ10-fold, and in combination these changes to metal binding and DNA binding make Zn(II) sensitivity comparable with endogenous Zn(II) sensors ZntR and Zur (Figs. 2, 8A, and 10, A and B). In common with recent studies of cobalt detection in other cells (6,67), relative access (a major kinetic contribution) is invoked to explain the gain of cobalt detection by FrmRE64H, a response that is assisted by glutathione (Fig. 8).
Unexpectedly, the native FrmR protein binds Co(II), Zn(II), and Cu(I) (Fig. 3 and 4). Moreover, Zn(II) and Cu(I) are shown by fluorescence anisotropy to be allosterically effective and able to weaken K DNA , thereby raising questions about why native FrmR does not normally de-repress gene expression in response to these metal ions ( Fig. 10C and Table 2). Crucially, by characterizing Salmonella ZntR and RcnR (Fig. 5, A-E and J-L), and by measuring cognate K metal of Salmonella Zn(II)sensing ZntR and Zur, Cu(I)-sensing CueR, and cobalt-sensing RcnR (Figs. 5, F-I, and 6, and Table 1), it becomes evident that in each case the respective metal affinity of FrmR is substantially weaker than each cognate sensor and it cannot compete (Figs. 7 and 8A). Values for K Zn(II) , K Co(II) , and K Cu(I) for Salmonella Zur, ZntR, RcnR, and CueR determined here are comparable with analogous sensors from some other organisms (Figs. 5 and 6) (1,11,17,28,64,68). The ability of FrmR to respond to metals in vitro but not within cells (Figs. 2, A and E, and 10C), coupled with relative K metal values (Table 1), provides another line of evidence that metal sensing within cells is a combined product of a set of sensors (1). The best sensor in the set is the one that responds to each element (1,11). In each case K metal for FrmR is substantially weaker than the respective K metal for the best in the set of sensors in Salmonella (Fig. 8A), and so it does not respond.
The E64H substitution was intended to create a metal-binding site more analogous to RcnR and indeed K Co(II) plus K Zn(II) and K Cu(I) are all tighter by ϳ10-fold compared with FrmR ( Fig.  4 and Table 1), but they all remain weaker than the respective cognate metal sensor (Table 1 and Figs. 6 and 8A). Nonetheless, for Zn(II) the affinity of FrmRE64H approaches that of known Zn(II) sensors such that there is overlap in fractional metal occupancy curves as a function of [Zn(II)] (Fig. 8A). The free energy of coupling of Zn(II) binding to DNA binding for FrmRE64H also changes relative to FrmR (Fig. 10, C and D, and Table 2). However, the change is the opposite of what might be predicted (1,9,11,64), with Zn(II) appearing to be less, not more, allosterically effective in the mutant protein (⌬G C Zn(II)-FrmRE64H⅐DNA Ͻ ⌬G C Zn(II)-FrmR⅐DNA ). Importantly, these values incorporate a much weaker K DNA for apo-FrmRE64H (Fig. 10D), which lowers overall promoter occupancy enhancing sensitivity to de-repression. Moreover, if regulation is dominated by metal binding to the DNA-protein complex to promote DNA dissociation, then the lesser ⌬G C of FrmRE64H infers an even tighter K Zn(II) (assuming a closed system (9, 61, 65, 66)) of the active DNA-bound species relative to FrmR (Fig. 10E).
Unlike for Zn(II), the enhanced K metal of FrmRE64H does not approach that of cognate sensors for Cu(I) or cobalt (Fig. 8A). Thus, relative affinity is consistent with the continued inability of FrmRE64H to detect Cu(I). However, the gain of cobalt sensing by FrmRE64H is enigmatic. In studies of the model cyanobacterium, Synechocystis PCC 6803, the detection of nickel and Zn(II) correlated with relative affinity and relative allostery within the set of sensors, but the detection of cobalt was attributed to relative access (1,6,11,18). Somehow, the cobalt effector was preferentially available to the cobalt sensor CoaR relative to sensors for other metals. Thus, although Zn(II) sensors ZiaR and Zur had tighter affinities for Co(II) than CoaR and both were (allosterically) responsive to Co(II) in vitro, neither ZiaR nor Zur responded to cobalt in the cell, whereas CoaR with weaker K Co(II) responded (6). Unlike Synechocystis CoaR, because FrmR has not evolved to detect cobalt, it is difficult to understand why cobalt should be channeled to FrmRE64H (Fig.  8A and Table 1). FrmR and cobalt-sensing RcnR do share common ancestry, and so interaction with a cobalt donor could perhaps be an evolutionary relic. Glutathione complexes are components of the buffered cellular pools for a number of metals (69). Because the substrates for formaldehyde dehydrogenase, FrmA, which is regulated by FrmR, are S-(hydroxymethyl) glutathione and S-nitrosoglutathione, it is also formally possible that FrmR can respond to glutathione adducts (33,70). Here, we see that cobalt and Zn(II) sensing by FrmRE64H is somehow assisted by glutathione (Fig. 8, B and C). This is opposite to what has previously been observed in the detection of cellular Zn(II) in other systems where the glutathione-substitute, bacillithiol, competes with Zn(II) sensors (61), and here we see a negligible effect of glutathione on Zn(II) sensing by ZntR (Fig. 8D). Whether glutathione aids the detection of cobalt due to cobalt binding and trafficking or due to redox effects on the oxidation state of cobalt or its ligands remains to be established.
Basal repression by FrmRE64H is less than FrmR, and this is explained by weaker K DNA of apo-FrmRE64H (Figs. 9 and 10, C and D). In pursuing the explanation for this phenotype, the abundance of both proteins was adjusted by optimizing or deoptimizing codons, an approach that preserves the transcriptional architecture. These changes were confirmed to increase and decrease the number of copies of FrmRE64H and FrmR per cell, respectively, with concomitant gain and loss of repression leading to matched levels of basal expression (Fig. 9, E-K). However, the magnitude of these changes in protein abundance alone was insufficient to switch FrmR into a metal sensor or to stop FrmRE64H from responding to Zn(II) or cobalt (Fig. 9F). Nonetheless, in theory, a change in relative protein abundance could alter metal competition with other sensors by mass action, and relative abundance should be added to the list of relative properties (affinity, allostery, and access) that determine which sensor is the best in the set to respond to a metal.
By how much do tighter K Zn(II) and weaker K DNA values of the apoprotein enhance the sensitivity of FrmRE64H to Zn(II), alone and in combination? By using the parameters set out in Tables 1 and 2, plus Fig. 9K, it has become possible to estimate fractional occupancy of the frmRA operator-promoter with repressor, either FrmR or FrmRE64H, as a function of [Zn(II)] (refer to "Experimental Procedures," supplemental material, and Fig. 10, F-I). First, a weaker K DNA of apo-FrmRE64H causes operator-promoter occupancy to be less than FrmR even in the absence of elevated Zn(II) (Fig. 10F), which explains the small but detectable (Figs. 2 and 9), basal de-repression. Individually, the determined tighter K Zn(II) or weaker apo-K DNA alone enhance the sensitivity of FrmRE64H to [Zn(II)] by ϳ1 order of magnitude (dotted and dashed lines on Fig. 10F), although in combination they increase sensitivity by ϳ2 orders of magnitude. If regulation is dominated by metal binding to (and promoting dissociation of) DNA-bound protein, then the inferred (weaker) K Zn(II) of the DNA-adduct becomes the relevant parameter (Fig. 10E). Under this regime (which assumes a closed system), the weaker K DNA of apo-FrmRE64H lessens ⌬G C Zn(II)-FrmRE64H⅐DNA and infers a tighter K Zn(II) of DNAbound FrmRE64H (on DNA) relative to FrmR. This, in combination with the measured tighter K Zn(II) , enhances sensitivity to [Zn(II)] by ϳ3 orders of magnitude (Fig. 10H).
There is ambiguity about the buffered concentrations of metals in cells (1). These values are important because relative metal availability influences metal occupancy by metalloproteins (71,72). Plausible limits on cellular buffered [Zn(II)] are defined by FrmRE64H, FrmR, FrmRE64H UP , and FrmR DOWN (Fig. 10, F-I). In the absence of elevated exogenous Zn(II), for FrmRE64H to fully repress, the buffered [Zn(II)] must be held below 10 Ϫ11 M, even if the inferred weaker (on DNA) K Zn(II) is applied to all molecules (Fig. 10H). This low (sub-nanomolar) value suggests that metalloproteins acquire competitive metals such as Zn(II) when there is no hydrated metal pool. These estimates of the buffered concentration of Zn(II) are consistent with the hypothesis that metalloproteins acquire Zn(II) via associative ligand exchange from a polydisperse buffer (1), rather than a hydrated pool of ions. This represents an associative cell biology of Zn(II).
Whether or not a significant pool of hydrated ions contributes to the metallation and hence regulation of FrmRE64H (and by inference other metal sensors) remains unresolved (73). One view is that metal sensors respond to hydrated ions at ϳ10 Ϫ9 M once the buffer is saturated (73). For FrmR DOWN to be unresponsive when cells are challenged with elevated exogenous Zn(II), the buffered [Zn(II)] must remain somewhere below 10 Ϫ8 M, even if the inferred weaker (on DNA) value for K Zn(II) is assigned to all FrmR molecules (Fig. 10I). This limit drops to 10 Ϫ10 M, if the determined (off DNA) K Zn(II) is used (Fig. 10G). Conversely, for FrmRE64H to respond, the buffered [Zn(II)] need only exceed 10 Ϫ11 M using the inferred weaker (on DNA) K Zn(II) (Fig. 10H). This places the intracellular [Zn(II)], at which FrmRE64H responds, somewhere within the range 10 Ϫ11 to 10 Ϫ8 M.
A long term aspiration is to gather analogous K metal , K DNA apoprotein , and K DNA metal-protein values, for cognate and noncognate metals, plus protein abundance for a cells' complement of metal sensors. In this manner, comparative models of sensor occupancy with metal (as in Fig. 8A) could be refined to more sophisticated and comparative models of promoter occupancy by repressors, as shown in Fig. 10, F-I. In turn, this should render transcriptional responses to metals predictable. In closing, the (subtle) biochemical changes, which in combination enable FrmRE64H to detect a sub-set of metals, support a view that (modest) differences in the relative properties of a cells' complement of sensors dictate which sensor is the best in the set to detect each metal inside cells.