Genetically Encoded Sensors to Elucidate Spatial Distribution of Cellular Zinc*

Transition metals are essential enzyme cofactors that are required for a wide range of cellular processes. Paradoxically, whereas metal ions are essential for numerous cellular processes, they are also toxic. Therefore cells must tightly regulate metal accumulation, transport, distribution, and export. Improved tools to interrogate metal ion availability and spatial distribution within living cells would greatly advance our understanding of cellular metal homeostasis. In this work, we present genetically encoded sensors for Zn2+ based on the principle of fluorescence resonance energy transfer. We also develop methodology to calibrate the probes within the cellular environment. To identify both sources of and sinks for Zn2+, these sensors are genetically targeted to specific locations within the cell, including cytosol, plasma membrane, and mitochondria. Localized probes reveal that mitochondria contain an elevated pool of Zn2+ under resting conditions that can be released into the cytosol upon glutamate stimulation of hippocampal neurons. We also observed that Zn2+ is taken up into mitochondria following glutamate/Zn2+ treatment and that there is heterogeneity in both the magnitude and kinetics of the response. Our results suggest that mitochondria serve as a source of and a sink for Zn2+ signals under different cellular conditions.

Although mammalian cells are known to concentrate transition metals, it is now well established that under resting conditions, "free" (e.g. unbound) metals are maintained at extremely low levels. Estimates of the total Zn 2ϩ concentration in mammalian cells typically range from 100 to 500 M (1); yet free Zn 2ϩ concentrations are tightly buffered by proteins such as metallothionein to maintain cytosolic Zn 2ϩ concentrations in the picomolar to nanomolar range (2)(3)(4)(5). However, there is emerging evidence that this static picture is dramatically altered by different cellular conditions, such as redox perturbations caused by oxidative stress (6,7) and cellular signals such as nitric oxide (8). Consequently, there is a pool of labile Zn 2ϩ that, if mobilized by cellular signals, would result in the generation of transient Zn 2ϩ signals. Recent studies suggest that these Zn 2ϩ signals influence critical biological processes, such as mitochondrial function (7,9,10). Elucidation of the sources and dynamics of these Zn 2ϩ signals would greatly advance our understanding of the interplay between metal regulation and cellular function.
There has been a huge effort in the past few years to develop sensitive and selective fluorescent probes to monitor Zn 2ϩ in biological systems. The majority of this work has focused on the generation of small molecule fluorescent indicators (reviewed by Que et al. (11)). Yet there are also examples of sensors based partially on Zn 2ϩ -binding proteins, such as carbonic anhydrase (12) and metallothionein (13), and peptide scaffolds (14). Although many of these sensors have begun to provide insight into Zn 2ϩ concentrations within cells, one limitation is that it is challenging to explicitly target them to subdomains within the cell. Localized probes are necessary to generate a complete picture of cellular Zn 2ϩ homeostasis in mammalian cells. For this reason, sensors that are genetically encoded (i.e. generated by translation of a nucleic acid sequence) are attractive platforms for engineering metal-specific sensors. Encoded sensors provide additional benefits such as retention of the sensor over days to weeks permitting long term imaging and the ability to systematically vary the sensor concentration to evaluate the extent to which the sensor perturbs resting Zn 2ϩ concentrations.
Here we present genetically encoded sensors designed with a "Zn 2ϩ -sensing domain" sandwiched between two fluorescent proteins. The fluorescent proteins are chosen so that they are capable of undergoing fluorescence resonance energy transfer (FRET). 2 Because the mechanism of FRET involves dipole-dipole coupling, it is exquisitely dependent on the distance and orientation of the fluorophores with respect to one another. Therefore, if the binding of Zn 2ϩ induces a conformational change in the sensor, it will alter the energy transfer between the two fluorescent proteins. The advantage of using FRET as the optical readout is that the donor emission will decrease and the acceptor emission will increase upon Zn 2ϩ binding. Hence, by taking the ratio of the acceptor to the donor emission, we can create a ratiometric sensor. These sensors are targeted to the cytosol, mitochondria, and plasma membrane by attachment of signal sequences and fusion to other proteins. These sensors reveal differences in the spatial distribution of Zn 2ϩ and highlight the power and utility of localized probes.

EXPERIMENTAL PROCEDURES
In Vitro Characterization-Details of sensor construction, protein purification, and buffered metal solutions are presented in the supplemental text. Purified sensor protein (0.5 M) was buffer exchanged into 10 mM MOPS, 100 mM KCl (pH 6.8) and titrated with Zn 2ϩ to obtain the apparent dissociation constant, KЈ d . We determined the KЈ d using two approaches. In one approach, the fluorescence intensity at 529 nm (emission maximum of YFP) was plotted as a function of Zn 2ϩ , and in the second approach, the change in FRET ratio (R Ϫ R min ) was plotted as a function of Zn 2ϩ , where the FRET ratio (R) is defined as the emission maximum of YFP divided by the emission maximum of CFP upon excitation of CFP, and R min is the minimum FRET ratio observed under Zn 2ϩ -free conditions. These measurements yielded KЈ d values within experimental error of one another. It should be noted that we measure the apparent K d because we are monitoring the conformational change that leads to FRET as opposed to directly measuring the Zn 2ϩ binding event. Mixed metal buffer solutions (15) were used to buffer Zn 2ϩ from 9 nM to 1.3 M (Zn 2ϩ /Sr 2ϩ /EGTA) and 2-134 M (Zn 2ϩ /Ca 2ϩ /EGTA). The Cys 2 His 2 sensor was purified and maintained under anaerobic conditions (see supplemental text for more details). A standard Ellman's assay (16) was used to quantify the percentage of reduced Cys. The observed FRET ratio change (R Ϫ R min ) for the Cys 2 His 2 sensor varied with the percentage of reduced Cys (typically between 50 and 80% of Cys were reduced). The lack of complete Cys reduction likely resulted from our inability to maintain a completely anoxic environment. Normalization to the amount of active, reduced protein yielded the maximal dynamic range for the Cys 2 His 2 sensor of 2.2-fold. Fluorescence measurements were made on a Safire-II fluorescence plate reader (Tecan). FRET ratios (R) were calculated for each titration point by dividing the YFP intensity ( max ϭ 529 nm) by CFP intensity ( max ϭ 475 nm) upon CFP excitation at 420 nm. The emission bandwidth was 10 nm.
Cellular Zn 2ϩ Measurements-The His 4 sensor was targeted to the extracellular surface of mammalian cells using pDisplay (Invitrogen). The Cys 2 His 2 sensor was targeted to the mitochondrial matrix, referred to as mito-Cys 2 His 2 , by attaching four tandem repeats of the cytochrome c oxidase signal sequence (MSVLTPLLLRGLTGSARRLPVPRAKIHSLGDP) to the N terminus of the sensor (17). Sensor constructs were transfected into HeLa cells using TransIt (Mirus) according to the manufacturer's instructions and imaged 48 h post-transfection. The imaging experiments were performed on an Axiovert 200M inverted fluorescence microscope (Zeiss) with a Cascade 512B CCD camera (Roper scientific), and equipped with CFP (430/24 excitation, 455 dichroic, 470/24 emission), YFP (495/10 excitation, 515 dichroic, 535/25 emission), and YFP FRET (430/24 excitation, 455 dichroic, 535/25 emission) filters controlled by a Lambda 10-3 filter changer (Sutter Instruments) and analyzed using Metafluor software (Universal Imaging). Experiments on cytosolically expressed sensors were collected at 40ϫ magnification, whereas mitochondrial experiments were collected at 100ϫ magnification.
The cells were imaged in phosphate-free HEPES-buffered Hanks' balanced salt solution. To determine the in situ KЈ d values, the cells were treated with 150 M TPEN to obtain the minimum FRET ratio (R min ) followed by a wash and the addition of increasing concentrations of a well defined Zn 2ϩ solution (Zn 2ϩ /Sr 2ϩ /EGTA-or Zn 2ϩ /Ca 2ϩ /EGTA-buffered solutions). For intracellularly expressed sensors (cytosol and mitochondria), the cells were treated with 150 M TPEN to obtain the minimum FRET ratio (R min ) followed by a wash and the addition of 15 M digitonin (Calbiochem) and a saturating Zn 2ϩ solution (1.5 mM) to establish the maximum FRET ratio (R max ). The digitonin was used to permeabilize the plasma membrane and facilitate Zn 2ϩ entry into the cell. The R max treatment was generally toxic to cells and therefore was always performed at the end of an experiment. For mitochondrial calibrations, the Zn 2ϩ ionophore pyrithione (25 M) was included along with digitonin to facilitate Zn 2ϩ entry across the mitochondrial membrane. Because for the Cys 2 His 2 sensor, the sensor concentration was similar in magnitude to the KЈ d , and the FRET ratio versus [Zn 2ϩ ] total was fit with the following expression (18), where [sensor] T was taken to be 6 M. This sensor concentration was determined experimentally as described below. Still, [sensor] T was varied from 0.5 to 10 M in the above fits with no change in the calculated K d . For the His 4 sensor, the data were fit to a one site binding equation.
Citrine YFP intensity was converted to sensor protein concentration using an established method (19). Briefly, a protein standard curve was generated by measuring the citrine intensities of purified sensor protein of known concentrations in a 50-m-tall rectangular glass capillary (VitroCom) using identical settings as the imaging experiments.
The FRET ratio (R) was calculated from background corrected FRET and CFP fluorescence images and corrected for photobleaching if necessary. The FRET ratios were converted to Zn 2ϩ concentrations using the following equation (20) along with experimentally derived R min , R max , and in situ KЈ d values.
It should be noted that R min and R max are obtained in each experiment and in each individual cell. Neuronal Culturing and Transfection-Primary hippocampal neurons were prepared from postnatal day 1 Sprague-Dawley rats, with slight modifications (21). Briefly, the hippocampi were dissociated and plated on 35-mm glassbottomed dishes coated with poly-D-lysine (Sigma) and laminin (Invitrogen) and grown in neurobasal A supplemented with B27 and Glutamax medium (Invitrogen). The neurons were transfected prior to plating by Amaxa nucleofection at 2,000,000 cells/transfection with 2 g of DNA/ manufacturer's recommendations (Amaxa) and grown to 9 days in vitro. The neurons were imaged at 9 days in vitro as above using 100ϫ magnification in a modified Tyrode's salts solution.

RESULTS
Sensor Design-The sensor design consists of two fluorescent proteins and a "sensing domain" that undergoes a conformational change upon Zn 2ϩ binding. A sensor schematic is presented in Fig. 1A. Binding of Zn 2ϩ to the sensing domain changes the distance and orientation between the cyan fluorescent protein (CFP) and the citrine fluorescent protein (YFP), leading to increased FRET from CFP to YFP.
The Zn 2ϩ -sensing domain is a canonical Cys 2 His 2 zinc finger from the mammalian transcription factor, Zif268. The zinc finger was selected because extensive structural data indicates that these domains are largely unstructured in the absence of metal ion and fold into a compact structure upon Zn 2ϩ binding (22). Three sensor variants were generated: a wild type zinc finger (Cys 2 His 2 ), a mutant zinc finger containing four histidines (His 4 ), and a mutant zinc finger containing four alanines (Ala 4 ) to abrogate Zn 2ϩ binding.
In Vitro Characterization-To characterize the sensors in vitro, the sensor protein was purified by affinity and size exclusion chromatography. Fig. 1B demonstrates that Zn 2ϩ binding to the His 4 sensor causes a decrease in CFP and an increase in YFP emission upon CFP excitation caused by an increase in FRET. The same spectral changes were observed for the Cys 2 His 2 sensor (data not shown). As expected, the sensor is ratiometric, and therefore the sensor response is presented as the FRET ratio (R). The dynamic range of the sensor is defined as R max /R min and was 4-fold for the His 4 sensor and 2.2-fold for the Cys 2 His 2 sensor, making these some of the more sensitive FRET sensors based on conformational change. For comparison, redesigned sensors for Ca 2ϩ exhibit a 5-fold ratio change (17,23), kinase sensors typically exhibit a 0.2-0.6-fold ratio change (24), and glutamate sensors exhibit a 0.25-fold ratio change (25). We suspect that the difference in the dynamic range between the His 4 and Cys 2 His 2 sensors results from slight variation in the overall geometry and hence orientation of the fluorescent proteins with respect to one another. There is some precedence for this because a systematic analysis of analogous Ca 2ϩ sensors revealed large changes in the dynamic range for sensors with different orientations but similar distances between the two fluorescent proteins (23).
To determine the apparent dissociation constant for Zn 2ϩ (KЈ d ), the FRET intensity and FRET ratio for each sensor were measured as a function of Zn 2ϩ concentration. These measurements yielded identical KЈ d values. The FRET ratio (R) binding curves are presented in Fig. 1 (C and D). The in vitro zinc affinities for the Cys 2 His 2 and His 4 sensors were determined to be 1.7 Ϯ 0.2 and 160 Ϯ 4 M, respectively. The Cys 2 His 2 sensor exhibited a lower Zn 2ϩ affinity than the isolated zinc finger (1.7 M versus 10 nM). We suspect that this is because attachment of the fluorescent proteins alters the apparent zinc affinity. As expected, mutation of Cys 2 His 2 to His 4 led to a reduction in Zn 2ϩ affinity, enabling us to generate both a high and low affinity sensor. Proteolytic removal of the His 6 tag did not alter the KЈ d of either sensor, indicating that the His tag does not participate in Zn 2ϩ binding. Mutation of Cys 2 His 2 to Ala 4 abrogates the sensor response, indicating that Zn 2ϩ binding to the Zn 2ϩ finger portion of the sensor is required for the observed FRET ratio change (Fig. 1D).
Although the cellular environment is generally reducing, cellular compartments may be oxidizing, or the cell may experience transient changes in redox potentials. The cysteines of the Cys 2 His 2 sensor are sensitive to oxidization. To test the effects of oxidation on the FRET ratio, we measured the in vitro FRET ratios of the reduced and oxidized Cys 2 His 2 sensor. R min was not affected by Cys 2 His 2 sensor oxidization; however, the oxidized form of the sensor was unable to achieve the maximum FRET ratio with the addition of saturating Zn 2ϩ (supplemental Fig. S1). These results indicate that artificial increases in FRET would not be generated by sensor oxidation, but that upon oxidation, the sensor will be less sensitive to Zn 2 , resulting in a decreased dynamic range.
For sensors to report Zn 2ϩ dynamics accurately within the complex environment of the cell, it is important that they be specific for Zn 2ϩ over other abundant divalent ions (Mg 2ϩ and Ca 2ϩ ) and biologically relevant transition metals (Cu 2ϩ , Mn 2ϩ , Fe 2ϩ , Fe 3ϩ , and Ni 2ϩ ). To determine the metal specificity, we examined whether other metals could elicit a FRET ratio change in the Cys 2 His 2 and His 4 sensors and/or interfere with the Zn 2ϩ response (supplemental Fig. S2). The majority of metals did not elicit a FRET change and did not interfere with the Zn 2ϩ response. The two exceptions were Fe 2ϩ and Cu 2ϩ . Fe 2ϩ caused a small perturbation of Zn 2ϩ binding to both sensors but only at the highest concentration tested (250 M). Fe 2ϩ itself did not cause an artificial FRET change. Cu 2ϩ interfered with Zn 2ϩ binding to the His 4 sensor such that the maximum Zn 2ϩ response could only be achieved at the lowest Cu 2ϩ concentration (5 M). The 5 M Cu 2ϩ level is still significantly higher than the estimated cytosolic free copper concentration of less than 10 Ϫ18 M or attomolar (26). Cu 2ϩ only bound to and elicited a FRET change at the highest concentration tested (250 M), suggesting that Cu 2ϩ signals would be unlikely to give false FRET increases. The metal specificity results suggest that metal cross-reactivity will not be a problem in the cellular environment. However, given that copper ions pose the greatest potential for interference, we further experimentally verified Zn 2ϩ selectivity over copper in cells (see below).
Cellular Expression and in Situ Characterization-Although genetically encoded FRET-based sensors for Zn 2ϩ have been reported previously, they have either not been demonstrated to function in cells (27-29) or were not used explicitly to measure cellular Zn 2ϩ (30). Therefore we sought to express our Cys 2 His 2 and His 4 sensors in mammalian cells and characterize the in situ response. We chose to target the low affinity His 4 sensor to the extracellular surface of the plasma membrane and the higher affinity Cys 2 His 2 sensor to both the cytosol and the mitochondrial matrix. Fig. 2A presents pseudo color FRET ratio images of HeLa cells expressing the cytosolic Cys 2 His 2 sensor. The addition of TPEN, a Zn 2ϩ -specific chelator, results in a decrease in the FRET ratio, whereas an addition of Zn 2ϩ results in an increase. Fig. 2 (B and C) demonstrates sensors targeted to mitochondria and the plasma membrane, respectively. Table 1 summarizes the sensors generated. To measure the apparent dissociation constant in situ, the cells were treated with TPEN to establish the minimum FRET ratio in zero Zn 2ϩ (R min ) followed by membrane permeabilization with digitonin and the addition of known concentrations of Zn 2ϩ . Fig. 2D   reports the full binding curves. The in situ Zn 2ϩ affinity of each sensor was comparable with that observed in vitro (KЈ d ϭ 1.5 Ϯ 0.2 M in situ versus 1.7 Ϯ 0.2 M in vitro for Cys 2 His 2 ; and 200 Ϯ 10 M in situ versus 160 Ϯ 4 M in vitro for His 4 ). As expected, the cytosolic Ala 4 sensor showed no significant FRET ratio change upon addition of Zn 2ϩ to cells. The dynamic range of both sensors was reduced in cells as compared with in vitro (0.25-fold in situ versus 2.2-fold for Cys 2 His 2 and 4-fold for His 4 in vitro), even though the affinity for Zn 2ϩ was unchanged. The dynamic range for Cys 2 His 2 was identical in the cytosol and mitochondria, suggesting that possible redox differences in these locations do not affect the sensor functionality and Zn 2ϩ measurements. We speculate that the reduced dynamic range in situ may be due to molecular crowding in the cellular environment, causing the fluorophores of the sensor to be closer together in the absence of Zn 2ϩ , resulting in an overall higher R min . This is supported by our observation that R min is generally higher in cells (ϳ4) than in purified protein (R min ϭ ϳ1.5) when measured under the same experimental conditions on the microscope. Although the dynamic range was reduced in cells, the sensors still yielded robust responses to changes in Zn 2ϩ levels and enabled visualization of Zn 2ϩ dynamics in neurons.
Next, we determined whether cellular Cu 1ϩ or Cu 2ϩ ions interfered with Zn 2ϩ measurements in cells. Resting FRET ratios were not affected by the addition of either neocuproine hydrochloride monohydrate (a Cu ϩ1 specific chelator) or bathocuproinedisulfonic acid disodium salt (a Cu ϩ1/ϩ2 chelator) but decreased to a minimum FRET ratio upon addition of TPEN, indicating that the in situ resting ratios were specific to Zn 2ϩ and not influenced by Cu 1ϩ or Cu 2ϩ (supplemental Fig. S3).
The mitochondrially targeted sensor (mito-Cys 2 His 2 ) exhibited a marked improvement in localization over the small molecule probe RhodZin-3 (Invitrogen; supplemental Fig. S4). Moreover, mito-Cys 2 His 2 stayed localized within mitochondria upon membrane depolarization, whereas RhodZin-3 leaked out into the cytosol under these conditions. The first step in characterizing cellular Zn 2ϩ homeostasis is to define the levels of Zn 2ϩ in different locations within the cell. There has been some controversy over whether mitochondria contain a labile pool of Zn 2ϩ that is elevated over the cytosol (9,31). To address this, we measured resting Zn 2ϩ concentrations in the cytosol and mitochondria using our Cys 2 His 2 sensor. Because the addition of a sensor may perturb cellular Zn 2ϩ levels, determinations of the "free" Zn 2ϩ concentrations should be conducted over a range of sensor concentrations, enabling mathematical extrapolation to the "free" [Zn 2ϩ ] at zero [sensor]. This variation in sensor concentrations is a natural consequence of transient transfection where the numbers of gene copies can vary from cell to cell. Tighter control of sensor expression will likely be possible with the use of inducible pro-moters, and the establishment of stably transfected cell lines. Fig. 3 (A and B) presents the concentration of Zn 2ϩ as a function of sensor concentration in the cytosol and mitochondria, respectively. For the cytosolic sensor citrine YFP intensity was converted to protein concentration using a protein standard curve (19) (supplemental Fig. S5). In mitochondria, the citrine intensity was not converted to a protein concentration because of the uncertainty in predicting mitochondrial volume.
We observe a direct correlation between sensor protein expression and the calculated Zn 2ϩ values in the cytosol, where higher sensor concentrations give rise to higher estimated Zn 2ϩ levels. A simple saturation model yields the best fit and projects  Control; does not bind Zn 2ϩ Cytosol a resting cytosolic Zn 2ϩ concentration of 180 nM when extrapolated to zero sensor concentration. For all of the subsequent experiments, we imaged cells with low (Ͻ5 M) expression levels to minimize perturbation of cellular Zn 2ϩ levels. We did not observe a strong correlation between citrine intensity and Zn 2ϩ levels within mitochondria. This is likely because the sensor expression was naturally low (hundreds of fluorescence counts instead of thousands), and the expression level did not vary significantly. The average Zn 2ϩ level within mitochondria was found to be 680 Ϯ 140 nM. Fig. 3C depicts representative time course experiments from both cytosol and mitochondria and directly illustrates the elevated resting Zn 2ϩ within mitochondria. These data support the hypothesis that mitochondria contain a pool of Zn 2ϩ that serve as a source of Zn 2ϩ signals, as suggested by Sensi et al. (9). Measurement of Mitochondrial Zn 2ϩ Dynamics in Hippocampal Neurons-A requisite feature of a useful sensor is the ability to observe dynamic fluctuations in living cells. Gluta-mate and Zn 2ϩ colocalize in presynaptic vesicles in hippocampal neurons, yet the physiological function of this vesicular Zn 2ϩ remains unclear (11). Nolan et al. (32) recently demonstrated that glutamate induces the uptake of exogenous Zn 2ϩ into the cytosol of hippocampal neurons. A natural extension of this work is to track the fate of Zn 2ϩ after entry. Because numerous studies have suggested that mitochondria may sequester Zn 2ϩ (8,31,33), we questioned whether transient Zn 2ϩ increases in the cytosol are transmitted to mitochondria in primary rat hippocampal neurons. Upon treatment with glutamate plus Zn 2ϩ , we observed an increase of Zn 2ϩ within mitochondria for the high affinity mito-Cys 2 His 2 sensor but not the lower affinity mito-His 4 sensor (Fig. 4). Supplemental Fig. S6 presents data traces of seven individual cells from three different experiments. Although there was some variability in the magnitude of the response (percentage of ratio change, i.e. R/R min , ϭ 3.2 Ϯ 0.8%, n ϭ 7), 100% of the cells measured showed an increase in mitochondrial Zn 2ϩ upon treatment with glutamate plus Zn 2ϩ . Interestingly, the transient mitochondrial Zn 2ϩ signals are dependent on the presence of extracellular Ca 2ϩ , because stimulation with glutamate and Zn 2ϩ in Ca 2ϩfree HEPES-buffered Hanks' balanced salt solution lead to no observable increase in mitochondrial Zn 2ϩ . The mechanism of Zn 2ϩ uptake is unclear, although the Ca 2ϩ uniporter has been shown to transport Zn 2ϩ (31). We are currently exploring the mechanism of uptake using inhibitors of different transport mechanisms. As depicted in Fig. 4 and supplemental movies, we observe heterogeneity in the magnitude and kinetics of Zn 2ϩ uptake. Fig. 4B depicts the FRET ratio as a function of time for four regions of interest, whereas Fig. 4C shows the actual images. The two cells responded at different rates, with the bottom cell peaking at ϳ20 s after stimulation, and the top cell peaking at ϳ40 s after stimulation. Likewise there is variability in the response within a given cell. Additionally we often observed transient hot spots (depicted by the white arrows) for one or two frames. These could correspond to individual mitochondria or small clusters of mitochondria. Although we are still exploring the cellular consequences of this heterogeneity, the experiments demonstrate the richness of information possible by subcellular analysis of Zn 2ϩ signals. Treatment with Zn 2ϩ alone showed no change to the FRET ratio in the mitochondria of hippocampal neurons. However, treatment of hippocampal neurons with glutamate alone resulted in a transient decrease in mitochondrial Zn 2ϩ (Fig. 5). The small molecule Zn 2ϩ indicator FluoZin-3-AM (Invitrogen) indicates that this mitochondrial Zn 2ϩ is released into the cytosol. Importantly, the decrease in mitochondrial Zn 2ϩ was not observed with the low affinity mito-His 4 sensor.
Glutamate-stimulated Zn 2ϩ release was slightly affected by extracellular Ca 2ϩ because glutamate treatment in Ca 2ϩ -free buffer altered the kinetics (but not the magnitude) of the response. Zinc release was measured in six individual cells from two separate experiments (supplemental Fig. S6), where the percentage of ratio change (R/R min ) was 3.5 Ϯ 1.0% (n ϭ 6), and every cell measured showed release of Zn 2ϩ upon treatment with glutamate alone. Glutamate stimulation of neurons leads to acidification followed by alkalinization of the cytosol. Given that fluorescent proteins, particularly the citrine YFP, can be pH-sensitive, we felt it was important to evaluate the pH sensitivity of the mitochondrial targeted sensor and the extent of acidification in mitochondria. As shown in Fig. 6A, the mito-Cys 2 His 2 sensor was unaffected by pH changes between 7.4 and 6.5, but the FRET ratio decreased at pH 6.0. The decrease in FRET ratio is likely caused by H ϩ quenching of citrine YFP fluorescence as the pK a of citrine is 5.7 (34). To characterize the pH changes within the mitochondrial matrix upon glutamate stimulation in the presence and absence of exogenous Zn 2ϩ , we targeted ecliptic pHluorin (35) to mitochondria. Following glutamate treatment in the presence and absence of Zn 2ϩ , the pH sensor was calibrated by adding H ϩ ionophores with buffers of known pH values. As seen in Fig. 6, glutamate stimulation lead to a slight acidification in mitochondria; the pH dropped from Ͼ7.4 to ϳ7.0. There is no observed effect on the mito-Cys 2 His 2 sensor in this pH range.

DISCUSSION
Transition metals such as Zn 2ϩ are both essential and toxic. The canonical view of metal homeostasis is that mammalian cells concentrate metal ions from their environment, but the vast majority of these metals are bound to proteins or cellular buffers. An alternate paradigm is emerging in which metals may be mobilized from labile pools, such as metallothionein or organelles. The possibility of transient metal "signals" is forcing us to re-evaluate cellular control of metal availability and its influence on cellular function. To obtain a comprehensive picture of Zn 2 regulation, it is necessary to define reservoirs of labile Zn 2ϩ as well as the fate of mobilized Zn 2ϩ .
In this work, we present genetically targeted Zn 2ϩ sensors and demonstrate their utility in monitoring the spatial distribution of Zn 2ϩ . Our mitochondrially targeted sensor reports an elevated resting Zn 2ϩ level in mitochondria relative to the cytosol. Moreover, treatment of hippocampal neurons with  glutamate results in transient release of this zinc into the cytosol, indicating that mitochondria can indeed serve as a source of Zn 2ϩ signals. This is consistent with the finding by Sensi et al. (9) that mitochondrial Zn 2ϩ pools may be mobilized independently of cytosolic pools. We also extended the work of Nolan et al. (32) and demonstrate that upon treatment with glutamate plus Zn 2ϩ , Zn 2ϩ is taken up into neurons and is sequestered into mitochondria, thus creating a transient Zn 2ϩ signal in mitochondria. Our results indicate that the role of mitochondria in modulating Zn 2ϩ signals is context-dependent; glutamate treatment alone leads to Zn 2ϩ release, whereas treatment with glutamate plus Zn 2ϩ leads to Zn 2ϩ uptake. This finding may have important biological implications because some hippocampal neurons contain only glutamate in presynaptic vesicles, whereas others contain both glutamate and Zn 2ϩ (4). Our results suggest that perhaps different neurons will elicit different cellular responses, although it is important to note that the cellular consequences of these Zn 2ϩ signals are still unclear.
There is a great challenge in accurately quantifying cellular Zn 2ϩ concentrations. In this work we demonstrate that cellular Zn 2ϩ levels are perturbed, even by expression of the relatively low affinity (K d ϭ ϳ1 M) Cys 2 His 2 sensor. This perturbation is not surprising given that our sensor concentration is also in the low micromolar range, and hence [sensor] is approximately equal to the K d . By measuring Zn 2ϩ levels as a function of [sensor], we were able to extrapolate to an estimate of 180 nM Zn 2ϩ in the cytosol. This estimate is higher than reports using other probes in which Zn 2ϩ levels are predicted to be in the picomolar to nanomolar range (2)(3)(4)(5). We suspect that all of these estimates, including our own, possess shortcomings that limit their accuracy. As our data indicate, measurements must be made at a range of sensor concentrations. This is particularly relevant for high affinity sensors, which when incorporated into cells at micromolar concentrations result in a situation in which the [sensor] is much greater than the K d . This is likely to lead to significant perturbation of the cellular free Zn 2ϩ levels. Although we account for this in our measurements, we believe our measurements overestimate Zn 2ϩ because our sensor is at the lower limit of its detection range. Still, we believe the relative estimates of cytosolic and mitochondrial Zn 2ϩ are fairly accurate given that the same sensor is used to monitor Zn 2ϩ in these different locations.
To overcome the shortcomings listed above and acquire accurate measurements of cellular Zn 2ϩ , we as a community will require a series of probes with different KЈ d values, where the [sensor] is kept much lower than the K d . We are currently working to place our sensors under control of an inducible promoter so that concentrations can be controlled more tightly and varied over a wider range for more accurate extrapolations. One advantage of our sensor design platform is that the fluorescent properties can be readily tuned by changing the fluorescent proteins. We are exploring different fluorescent protein combinations as well as optimizing the brightness of the proteins so that sensors can be incorporated into cells at lower expression levels in an effort to minimize the perturbation of cellular metals. In this regard, genetically encoded probes offer an advantage over small molecule indicators whose concentra-tion is difficult to control because cellular accumulation is a balance of uptake and extrusion (36). Moreover, accurately defining the concentration of small molecule probes whose intensity is modulated by Zn 2ϩ is challenging.
Our localized probe overcomes many of the limitations of RhodZin-3 as a mitochondrial Zn 2ϩ sensor. For example, RhodZin-3 exhibits suboptimal localization; it is lost from mitochondria upon membrane depolarization, and the readout is intensity-based rather than ratiometric. We attempted to use RhodZin-3 to examine mitochondrial Zn 2ϩ dynamics in neurons, but the localization was so poor that we could not obtain interpretable data (supplemental Fig. S3). Thus, the mitochondrial probe presented here represents a significant advance in our ability to monitor mitochondrial Zn 2ϩ .
However, the sensors presented here can still be improved. As illustrated in Fig. 6, they are quenched at low pH (6.0 and below), which would impede measurement of Zn 2ϩ in acidic compartments. The Cys 2 His 2 sensor relies on Cys to bind Zn 2ϩ , and thus its use is restricted to reducing environments. Lastly, although we were able to observe reproducible signals, the sensors would benefit from an expanded dynamic range.
In summary, we have developed genetically targeted sensors for Zn 2ϩ and used these to demonstrate that mitochondria contain a labile and releasable pool of Zn 2ϩ under resting conditions. In neurons, mitochondria can serve as a source of Zn 2ϩ signals by releasing Zn 2ϩ into the cytosol, as well as a sink by sequestering elevated cytosolic Zn 2ϩ . The observation of a mitochondrial pool of Zn 2ϩ raises the intriguing question of whether other organelles modulate Zn 2ϩ availability by serving as either sources or sinks for Zn 2ϩ .