Stable Interactions between Mitochondria and Endoplasmic Reticulum Allow Rapid Accumulation of Calcium in a Subpopulation of Mitochondria*

To better understand the functional role of the mitochondrial network in shaping the Ca2+ signals in living cells, we took advantage both of the newest genetically engineered green fluorescent protein-based Ca2+ sensors (“Cameleons,” “Camgaroos,” and “Pericams”) and of the classical Ca2+-sensitive photoprotein aequorin, all targeted to the mitochondrial matrix. The properties of the green fluorescent protein-based probes in terms of subcellular localization, photosensitivity, and Ca2+ affinity have been analyzed in detail. It is concluded that the ratiometric pericam is, at present, the most reliable mitochondrial Ca2+ probe for single cell studies, although this probe too is not devoid of problems. The results obtained with ratiometric pericam in single cells, combined with those obtained at the population level with aequorin, provide strong evidence demonstrating that the close vicinity of mitochondria to the Ca2+ release channels (and thus responsible for the fast uptake of Ca2+ by mitochondria upon receptor activation) are highly stable in time, suggesting the existence of specific interactions between mitochondria and the endoplasmic reticulum.

Calcium is arguably the most versatile player within the cell. This second messenger is directly or indirectly involved in a myriad of processes that span virtually all physiological aspects of a cell, including its birth, health, disease, and death. From a generic point of view, calcium was often seen to exert its action exclusively through changes in cytosolic free calcium concentration ([Ca 2ϩ ] c ). 1 In recent years, these global changes have been dissected into regional variations, and different organelles have seen their importance accrued. In particular, in practically all cell types investigated, it has been found that the speed and amplitude of mitochondrial Ca 2ϩ uptake depend not only on the amplitude of the [Ca 2ϩ ] c rise, but also on the source of Ca 2ϩ and the mechanisms through which the Ca 2ϩ increase is elicited. Specifically, fast increase of mitochondrial Ca 2ϩ concentration ([Ca 2ϩ ] m ) can be triggered by either Ca 2ϩ mobilization from stores or Ca 2ϩ influx from the medium, or both, depending on the cell type. Because mitochondrial Ca 2ϩ uptake takes place exclusively through the so-called calcium uniporter, and because this uniporter has a low affinity for Ca 2ϩ , it was not clear how the relatively low, average [Ca 2ϩ ] c increase elicited by physiological stimuli could efficiently drive the observed rise in [Ca 2ϩ ] m . The concept of high Ca 2ϩ microdomains was thus developed to explain the very rapid Ca 2ϩ uptake by mitochondria under physiological conditions. The hypothesis predicts that the fast [Ca 2ϩ ] m increases depend on the close vicinity of these organelles to Ca 2ϩ channels, where the Ca 2ϩ concentration is sufficiently high to drive an efficient uptake by the low affinity mitochondrial uniporter (1,2). It is worth noting that this microdomain model does not purport to explain all mitochondrial Ca 2ϩ uptake; rather, the model provides a conceptual framework to explain the fast [Ca 2ϩ ] m rises, erstwhile paradoxical.
Given the motility of both ER and mitochondria, and the requirement for regions of close proximity between the two organelles to ensure a highly efficient mitochondrial Ca 2ϩ uptake, it becomes of paramount importance to understand how this spatial arrangement is obtained. From a structural point of view, the ER is a highly convoluted reticular network, while mitochondria have historically been erroneously depicted as "fuse-like" individual organelles that pepper the cytoplasm. The notion that also mitochondria form a network within the cytoplasm is practically half a century old, but only recently has it been more widely accredited, with in vivo studies that have revealed a highly dynamic network that continuously undergoes multiple fusion and fission processes (2)(3)(4)(5). Essentially, two major scenarios can be envisaged. On one hand, the vicinity of mitochondria to Ca 2ϩ release sites may be a stochastic event, because of the abundance and motility of both organelles. On the other, transient or permanent interactions may exist to keep specific mitochondrial subpopulations close to sites where Ca 2ϩ reaches high concentrations. It has been suggested that stable mitochondria-ER interactions might occur in adrenal medullary cells (6), but this phenomenon has not been analyzed in detail. In the present study, we adopted a bipartite approach to explore this issue. For single cell studies we employed the new GFP-based Ca 2ϩ probes selectively targeted to the mitochondrial matrix, whereas at the cell popu-lation level we took advantage of specific characteristics of aequorin, a Ca 2ϩ probe with an established track record.

EXPERIMENTAL PROCEDURES
Generation of Constructs-cDNA encoding the N-terminal part (comprising the first 36 amino acids) of subunit VIII of human cytochrome c oxidase was fused, in-frame, to cDNA encoding DsRed (Clontech) to generate mt-DsRed. Mitochondrially targeted versions of Cameleon and split Cameleons were generated similarly. Cameleon, split Cameleons, and mtCamgaroo-2, and mt-and nuPericamR were generous gifts from Roger Y. Tsien and Atsushi Miyawaki, respectively. Details of all constructs are available upon request.
Cell Cultures and Transfection-HeLa cells were grown in Dulbecco's modified Eagle's medium containing 10% fetal calf serum, supplemented with L-glutamine (2 mM), penicillin (100 units/ml), and streptomycin (100 g/ml), in a humidified atmosphere containing 5% CO 2 . For transient transfections, cells were seeded onto glass coverslips (24-and 13-mm diameter for single cell imaging and for aequorin measurements, respectively). Transfections were performed at 50 -70% confluence with the calcium phosphate method, using a total of 4 or 10 g of DNA for small or large coverslips, respectively.
Cell Imaging-Cells expressing fluorescent probes were observed 36 h after transfection on an inverted fluorescence microscope (Zeiss Axioplan), with an oil immersion objective (ϫ63, N.A. 1.40). Excitation light at appropriate wavelengths was produced by a monochromator (Polychrome II, TILL Photonics, Martinsried, Germany): 440 nm for cameleon and CFP-CaM, 500 nm for M13-YFP and camgaroo, and 415 and 490 nm for PericamR. Dichroic beam splitters were 455DRLP, 525DRLP, and 505DRLP, respectively. Emission filters were 480DF30 (for CFP) and 545DF35 (for YFP) in the case of cameleon, HQ520LP in the case of camgaroo, and 535RDF45 in the case of PericamR; when using cameleon, the emission filters were alternated using a filter wheel (Lambda 10 -2, Sutter Instruments, San Rafael, CA). Filters and dichroic beam splitters were purchased from Omega Optical and Chroma Technologies (Brattleboro, VT). Images were acquired using a cooled CCD camera (Imago, TILL Photonics) attached to a 12-bit frame grabber. Synchronization of the monochromator and CCD camera was performed through a control unit using TILLvisION version 4.0 (TILL Photonics); this software was also used for image analysis. Additional image analyses employed the public domain ImageJ program (developed at the United States National Institutes of Health by Wayne Rasband and available on the Internet. 2 For co-localization studies, confocal planes were obtained using a Bio-Rad MRC1024ES system: for mtPericamR, the 488-nm line of a Kr-Ar laser was used together with a 522DF35 emission filter, whereas for mtDsRed, the 568-nm line and a 605DF32 emission filter were used. Unless otherwise specified, F measurements refer to average pixel intensities in regions covering Ͼ50% of the total mitochondrial or nuclear area. Whereas camgaroos respond to Ca 2ϩ changes only with a change in F, ratiometric pericam shows an antiparallel behavior in response to Ca 2ϩ , i.e. upon Ca 2ϩ binding, F upon excitation at 415 nm decreases, whereas F upon excitation at 490 nm increases. In a few experiments, however, the 415-and 490-nm signals did not behave in a perfectly antiparallel way. Such a behavior was random and most often because of either photobleaching or movement. In these cases the ratio of the two signals nicely corrects for the artifact. In a few cases, however, we noticed a drop of the 490-nm signal not accompanied by an increase in the 415-nm fluorescence that were not attributable to the above artifacts. The reason for this rare behavior is presently unknown, but, if it is observed, it can be verified by monitoring the fluorescence at the isosbestic point (470 nm).
Perfusion with histamine (100 M) was employed to trigger Ca 2ϩ release from intracellular stores. Perfusion with cyclopiazonic acid (20 M) was employed to inhibit ER Ca 2ϩ -ATPase.
For cytoskeleton disruption, cells were first challenged with histamine and then incubated in KRB, in the presence of colchicine (ICN, Milan, Italy; 10 g/ml) or cytochalasin D (ICN; 10 M), or both, for 30 min. The second histamine stimulus was applied in the presence of the same drug(s).
All experimental procedures and incubations were carried out at 37°C. All chemicals were of analytical or highest available grade and, unless otherwise stated, were acquired from Sigma. Data shown represent typical results obtained in at least five independent experiments; numerical data are presented as mean Ϯ S.D.; statistical significance was calculated by applying Student's t test.

Choice of a Mitochondrial Ca 2ϩ Indicator for Single Cell
Studies-The positively charged fluorescent indicator rhod-2 is often considered the indicator of choice to measure changes in mitochondrial [Ca 2ϩ ] at the single cell level. Although this probe has been extensively used by different groups, its subcellular localization is far from ideal; most likely because of this, its use has given rise to partially contradictory results in different cell types (7)(8)(9)(10). The search for alternative calcium indicators has prompted the development of a number of genetically encoded fluorescent probes. These probes, however, have yet to be thoroughly characterized. Below we consider the three families of fluorescent indicators that have been recently introduced, i.e. Cameleons, Camgaroos, and Pericams.
Cameleon is a GFP-based Ca 2ϩ indicator, which was initially introduced as a probe for Ca 2ϩ in the cytoplasm, ER, and nucleus (11). A mitochondrial cameleon has been generated by fusing at the N terminus of the cytosolic construct the targeting sequence of subunit VIII of human cytochrome c oxidase (12). This particular mitochondrial targeting presequence has been repeatedly shown to deliver fused proteins in an highly efficient and selective manner to the mitochondrial matrix (13,14). In a series of elegant experiments (12), it was observed that a cameleon targeted to mitochondria by this strategy only localizes efficiently in cells that express very low levels of the construct. We also confirmed that this fusion protein fails to localize correctly and transfected cells exhibit fluorescence throughout the cytoplasm; only rarely was a selective labeling of mitochondria observed.
An attempt to bypass this problem was to use the so-called split cameleon (11), in which two functional constituents of the cameleon are expressed individually. The two halves of the indicator were targeted to the mitochondrial matrix independently, using the same presequence as before. In this case, the M13-YFP moiety was efficiently delivered exclusively to the mitochondrial matrix (Fig. 1A, left panel); targeting of the CFP-CaM part was again deficient, although, with ϳ50% of the fluorescence present throughout the cell body (Fig. 1A, right panel). The reasons for this behavior are not clear at present. Because of these difficulties, no further attempt to characterize mtCameleons was undertaken.
Camgaroo is an insertional mutant of GFP, sensitive to Ca 2ϩ (15). A mitochondrial version of this probe (mtCamgaroo-2; Ref. 16) possesses an N-terminal presequence that effectively targets it to the mitochondrial matrix (Fig. 1B), although in strongly fluorescent cells a residual signal is often present throughout the cell cytoplasm. Results obtained by Tsien and co-workers (16) indicated that mtCamgaroo-2 can be used to monitor changes in [Ca 2ϩ ] m . Surprisingly, in our hands, HeLa cells expressing mtCamgaroo-2 transiently showed marginal, often undetected, changes in fluorescence when challenged with histamine (see below). The reason for this discrepancy lies in the illumination protocol used. When cells were illuminated continuously, a very rapid drop (to ϳ50% of the initial F) was observed; this drop was not due to photobleaching because a short recovery (30 s) in the dark resulted in almost complete recovery of the initial F ( Fig. 2A, upper panel). Importantly, the photoconverted form of the probe was almost completely insensitive to Ca 2ϩ changes, as indicated by the lack of increase in F upon histamine addition after ϳ10 s of continuous illumination ( Fig. 2A, upper panel). If, on the contrary, the time of illumination was reduced to a minimum and at least 1 s between two successive illuminations were allowed, a clear increase in F was revealed by mtCamgaroo-2 upon histamine challenge ( Fig.  2A, lower panel), on average about 20% of the initial value. Under the same conditions, a non-Ca 2ϩ -sensitive, but pH-sensitive, YFP gave no change in signal ( Fig. 2A, lower panel).
Using mtCamgaroo-2, we observed a substantial subcellular heterogeneity in peak Ca 2ϩ increases in single HeLa cells challenged with histamine ( Fig. 2B), in agreement with previous studies with rhod-2 or mtCameleons (7,12). With probes such as rhod-2 or mtCamgaroo-2 (that respond to alterations in [Ca 2ϩ ] simply by changing F), changes in size and shape of organelles, as well as organelle movement, may all create artifacts that confound both quantitative and qualitative analyses (17).
To overcome these limitations, we shifted our attention to the most recent genetically targeted Ca 2ϩ probe family (the Pericams; Ref. 18). Fig. 1C shows the typical staining pattern of a HeLa cell transiently expressing a mitochondrially targeted ratiometric pericam (mtPericamR). The staining is highly specific and completely overlaps with that of a mitochondrially targeted red variant of GFP (mtDsRed). In addition, the nuF is almost indistinguishable from the background of non-transfected cells.
The use of mtPericamR greatly diminishes the problem of photoconversion that plagues mtCamgaroo-2. Upon almost continuous illumination there is still an initial, rapid decrease in F at both wavelengths (but only ϳ15%); as in the case of mtCamgaroo-2, the initial F is recovered after brief periods of non-illumination (data not shown). The significant difference in relation to mtCamgaroo-2 is that this probe retains its Ca 2ϩ sensitivity even during periods of rapid data acquisition (see below). FIG. 2. Ca 2؉ measurement using mtCamgaroo-2. HeLa cells were transfected with mtCamgaroo-2 (or mtYFP) and incubated in KRB (see "Experimental Procedures"). A, in the upper panel a single cell was illuminated for 60 ms at 500 nm and images acquired every 80 ms; arrows indicate "light on" (1) and "light off " (2); the non-illumination period (//) was 30 s. In the lower panel the round symbols refer to a cell expressing mtCamgaroo-2 and the triangles to a cell transfected with mtYFP. In this case, cells were similarly illuminated for 60 ms, but images acquired only every second (for mtCamgaroo-2) or every 2 s (for mtYFP). The concentration of histamine was 100 M. B, the cells were illuminated with the low-frequency protocol (1 image every 2 s). On the left the pseudocolor-rendered images of ratio values (F/F 0 ) acquired at rest (1) and at the peak response following the histamine challenge (2); F is the fluorescence at each data point and F 0 is an average of the fluorescence of 5 images before the application of the stimulus. The graph represents the kinetic changes of the four regions indicated on the image; 1 and 2 indicate the time of acquisition of the two images.
Characterization of mtPericamR-HeLa cells expressing mt-PericamR and challenged with histamine show a subcellular heterogeneous response, as reported previously using other non-ratiometric indicators, confirming that the differences between mitochondrial populations are real and not artifacts of movement. Fig. 3A shows pseudocolor-rendered ratios of a HeLa cell before and after a histamine challenge. When different subsets of mitochondria are selected for analyses, a clear heterogeneity is observed (Fig. 3B).
We next addressed the problem of the relationship between mitochondrial and cytoplasmic Ca 2ϩ changes in the same cell. Given that it has been repeatedly demonstrated in HeLa cells that the nucleus behaves essentially as the cytoplasm, we cotransfected these cells with a variant of PericamR targeted to the nucleus (nuPericamR) and mtPericamR. Under resting conditions, the ratio value was higher in the latter. This most likely reflects a difference in resting pH, rather than in [Ca 2ϩ ] (see below). Upon a histamine challenge both compartments showed similar kinetics: a sharp rise in the F ratio, followed by a declining plateau, often accompanied by oscillations (Fig. 3C). Quantitatively, it was surprising to note that this increase was larger in the nucleus than in the mitochondria. This result is in sharp contrast with previous data obtained with aequorin, rhod-2, and mtCameleon, with all these probes, the increase in [Ca 2ϩ ] has been found to be larger within mitochondria than in the nucleus (and cytoplasm).
To verify whether this contradiction was because of a different affinity of the probe for Ca 2ϩ in the two compartments, we carried out a calibration in situ with the widely used ionomycin and high Ca 2ϩ method (e.g. Ref. 16). As shown in Fig. 4A (upper panel), both compartments exhibit a sharp increase in the ratio upon histamine stimulation, which returns to basal levels after the agonist is washed away; addition of Ca 2ϩ in the presence of the ionophore also causes an increase in the ratio, this time clearly biphasic: a first rapid increase, followed by a larger and slower rise. However, analysis of the two individual wavelengths shows distinct pictures in the case of histamine and ionomycin. In the former, the increase in ratio is because of a small increase in F upon excitation at 490 nm and a larger decrease in F upon excitation at 415 nm (Fig. 4A, lower panel). In the latter, the first rise is similar to that observed for histamine; the second rise, however, is fully because of an increase in F upon excitation at 490 nm, whereas the 415-nm signal remains essentially unchanged. Given that alkalinization increases the F of PericamR at 490 nm (18), the possibility of a pH artifact in the effect of ionomycin was thus considered. Cells expressing nuPericamR and mtPericamR were challenged with NH 4 Cl to increase the pH inside the cells without C, HeLa cells were cotransfected with mtPericamR and nuPericamR. The blue and black traces represent, respectively, the kinetic changes of the 490/415 nm ratio over large mitochondrial and nuclear regions of a single representative cell. When present, the frequency of the oscillatory pattern is reduced when the agonist concentration is lowered to 5 M.
grossly changing the [Ca 2ϩ ] (verified with fura-2). An apparent increase in [Ca 2ϩ ] was observed in both compartments (Fig. 4B, upper panel), but this was due simply to an increase in the 490-nm signal (Fig. 4B, lower panel). A further confirmation that the ionomycin effect is because of a pH artifact is provided by monitoring F at the isosbestic point (470 nm). As expected, no fluorescence increase was observed at this wavelength in the case of histamine, whereas both ionomycin and NH 4 Cl caused a major increase in F at this wavelength (data not shown).
To obtain an in situ measurement of the Ca 2ϩ affinity of mtPericamR, we employed the following protocol. Cells were first treated with digitonin to permeabilize the plasma membrane and then with the mitochondrial uncoupler FCCP to equilibrate matrix pH with that in the medium. The external medium, without added Ca 2ϩ and supplemented with 100 M EGTA, was buffered at pH 8 to mimic the conditions of the mitochondrial matrix in intact cells. Under these conditions, given the absence of a mitochondrial membrane potential and of a pH gradient, matrix [Ca 2ϩ ] rapidly establishes an equilibrium with the extracellular [Ca 2ϩ ] via the uniporter. Indeed, further addition of ionomycin at any [Ca 2ϩ ] resulted in no appreciable change in F. Increasing concentrations of CaCl 2 in the medium resulted in increased F ratios (Fig. 5A) with no change at the isosbestic point. The apparent Kd Ј of mtPericamR for Ca 2ϩ was thus determined to be ϳ11 M (Fig. 5B), much higher than the 1.3 M calculated in vitro (18). Using this calibration procedure we could conclude that the mean increase in mitochondrial [Ca 2ϩ ] at the peak of a histamine challenge is about 10 M with some regions reaching values well over 50 M (see, for example, Fig. 3A, where ratio values of 2.5 and 3.4 correspond to 10 and 50 M, respectively, according to the calibration shown in Fig. 5, A and B).
It was more difficult to perform the same calibration procedure in the nucleus, because the digitonin treatment (100 M, as used for calibrating mtPericamR) generally provoked the release of the nuclear probe. We tried to permeabilize the cells with S. aureus ␣-toxin (19) but the results were unsatisfactory, given that very high concentrations of the toxin (ϳ100 g/ml) resulted only in partial release of a trapped fluorescence probe such as fura-2. However, at lower digitonin concentrations (20 M), complete release of fura-2 was obtained, whereas in the majority of the cells a large part of nuPericamR remained localized in the nucleus. Accordingly, the Kd Ј of nuPericamR in situ was calculated to be ϳ2.5 M, similar to that calculated in vitro.
The kinetics of the mitochondrial Ca 2ϩ increase with respect to that in the nucleus were analyzed in the experiments presented in Fig. 6. The high rate of data acquisition permitted by the use of PericamR enabled us to acquire data points (F ratio upon excitation at two independent wavelengths) every 180 -200 ms. Detailed analyses of the ratio increases immediately following histamine stimulation revealed different types of heterogeneity. In some cells, the mitochondrial response time was homogeneous, but with different rates of Ca 2ϩ accumulation (Fig. 6A). In others, mitochondria have similar rates of Ca 2ϩ uptake, but different delays in relation to the nuclear [Ca 2ϩ ] rise (data not shown). In general, the heterogeneity observed among different mitochondrial populations, in terms of the lag time between nucleoplasmic and mitochondrial Ca 2ϩ increases, ranged from not appreciable (less than 200 ms) to well above 400 ms. On the other hand, in accordance with previously published results (18), the peak response of the mitochondria was reached with some delay (3-5 s) with respect to that in the nucleus (1-2 s).
Stochastic or Regulated ER-Mitochondria Interactions?-Experiments presented above and previously published data confirm a substantial heterogeneity in the mitochondrial response to the Ca 2ϩ increases elicited by an agent such as histamine (that by producing InsP 3 , mobilizes Ca 2ϩ from intracellular stores). The accepted interpretation for such heterogeneity (and for the speed and amplitude of mitochondrial Ca 2ϩ increases) is that some of the organelles are very close to the InsP 3 -gated channels. In this way, they are exposed not to the mean increases in cytoplasmic [Ca 2ϩ ], but rather to the microdomains of much higher [Ca 2ϩ ] that are formed close to the channels themselves. The still unanswered, but key question, is whether this vicinity is simply a stochastic event of two organelles that are densely packed within the cell (whereby the random movement and reorganization of both networks ensure that some mitochondria are always close to some InsP 3 receptors at any given time), or whether more stable associations occur between the two organelles. To address this question, a protocol of double stimulation with histamine was devised. The rationale is as follows. If the vicinity of mitochondria to the InsP 3 receptors is stochastic, one would expect that if two similar Ca 2ϩ releases from stores are triggered, the mitochondria that respond maximally to each release will be different in each case. Conversely, if there is a stable association between the two organelles at specific sites, the same subset of mitochondria will show a maximal response in two successive stimulations.
Preliminary to such an approach is the demonstration that under the chosen experimental conditions there is no desensitization of the overall mitochondrial Ca 2ϩ uptake upon two successive histamine stimulations (7). To analyze this aspect in detail, we transfected cells with both nu-and mtPericamR, and the cells were subjected to a double histamine challenge, spaced in time by 3, 7, 10, and 15 min. When the average responses of the mitochondria and of the nucleus were analyzed, no differences were observed (Fig. 6B) between the first and the second histamine response. Noteworthy, in some of the cells analyzed we observed a reduction (Ͼ10%) in the second response; in these cases, however, a reduced mitochondrial [Ca 2ϩ ] increase to the second histamine challenge was most commonly accompanied by a reduced nuclear response (data not shown).
The constant mobility of the mitochondrial network rendered the proposed analysis problematic. Indeed, when single organelles eliciting a maximal response were selected at any given time point and compared with subsequent frames, it was impossible to determine whether differences observed were because of modified responses of the organelles, or to their movement. Fig. 6C exemplifies this problem. First, pixels with maximal responses to the first stimulus were selected and color-coded as red (panel 1). The pixels of maximal response to the second stimulus were again selected and now color-coded as green (panel 2). Merging of panels 1 and 2 reveals that the pixels of maximal response that colocalize perfectly (color-coded as yellow) represent only a small fraction of the total. This analysis, however, is biased by the fact that the organelles are in continuous movement and accordingly the same mitochondrion could be in a slightly different position at different moments in time.
If a less detailed analysis was carried out, for example by identifying distinct mitochondrial regions, each comprising 3-5% of the total mitochondrial area, the regions that exhibited larger rises upon the first stimulus were generally those that responded maximally upon the second stimulus (data not shown). This type of analysis, however, appears not accurate enough to clearly distinguish whether there is indeed a stable interaction between specific ER regions and individual mitochondria.
Subcellular Heterogeneity in Mitochondrial Ca 2ϩ as Revealed by Targeted Aequorin-To further address the problem of stochastic versus specific ER-mitochondria interactions, we took advantage of another selective mitochondrial Ca 2ϩ probe, aequorin. Aequorin is best suited to monitor changes in cell populations, but some of its characteristics, often overlooked under routine use, makes it ideal to address the problem of stochastic versus regulated ER-mitochondria interactions. In fact, when in contact with Ca 2ϩ , aequorin emits light and in the process is transformed into a Ca 2ϩ -insensitive, non-luminescent form; in practical terms, the probe is "consumed" in the presence of Ca 2ϩ . We took advantage of this characteristic to address the question of whether associations between mitochondria and the ER are maintained over time. The rationale is as follows. Triggering Ca 2ϩ release from the ER exposes a subset of the mitochondrial network to microdomains of high [Ca 2ϩ ]; this subpopulation will uptake a large amount of Ca 2ϩ and aequorin will be selectively and markedly consumed. When a second ER Ca 2ϩ release event is triggered, two scenarios can be envisaged. If ER and mitochondria are tightly associated, then the mitochondrial regions that undergo massive uptake of Ca 2ϩ will be essentially the same; if, on the other hand, both organelles have free mobility, then random regions of mito- chondria will uptake this Ca 2ϩ . In terms of aequorin response, these two situations will generate radically different data. If the former is true, in the highly responding population of mitochondria the percentage of aequorin molecules capable of releasing photons will be greatly reduced (having been largely consumed during the exposure to the first Ca 2ϩ burst), and the emitted light will be drastically diminished. On the other hand, if the latter is true, essentially the same quantity of aequorin molecules will be available to release photons, and the light signal generated should be comparable with that observed dur-ing the first Ca 2ϩ burst. The calibrated Ca 2ϩ signal (that takes into account the average consumption of aequorin) would appear drastically reduced upon the second histamine challenge in the first case and essentially unmodified in the second. As shown in Fig. 7A, the amplitude of the second peak, 7 min after the first, was reduced by ϳ70%. For comparison, the kinetics of [Ca 2ϩ ] c changes (observed in a parallel batch of cells, transfected with cytAequorin) is also presented; in this case the intensity of the second response is only marginally diminished.
It should be noted that the experiments carried out with ; the number of cells analyzed ranged from 12 to 52, with no more than 3 cells per independent experiment; no significant difference between the first and second challenge was observed at any time point. C, HeLa cells expressing mtPericamR were subjected to two 30-s applications of 100 M histamine, applied 7 min apart. Pixels of maximal response (top 30% of response range) were selected from the ratio images after the first (1) and second (2) histamine challenges (shown in red and green, respectively). The overlay of the two selections is presented in panel 3. mtPericamR (see above) exclude the possibility that the dramatic drop in the mitochondrial Ca 2ϩ response is because of a desensitization of the mitochondrial uniporter. To address this issue in more detail, we took advantage of another particularity of the aequorin system. The expressed apoaequorin becomes fully functional (and, thus, capable of emitting light, and being consumed, in the presence of Ca 2ϩ ) only after being covalently linked to coelenterazine. This latter process is generally referred to as "reconstitution" and is usually carried out over a period of hours, during which cells expressing apoaequorin are bathed in medium containing coelenterazine. We ascertained that the minimum loading time that still ensures a reliable light signal could be lowered to 5 min and consequently developed the following experimental protocol. HeLa cells expressing wild-type apoaequorin, but without added coelenterazine, were challenged with histamine (no light output was obviously detected under these conditions, and aequorin is not consumed); histamine was washed away and coelenterazine loaded acutely for 5 min. Using this short reconstitution protocol the mitochondria are exposed to two consecutive fast Ca 2ϩ increases, but the selective consumption of aequorin during the first peak is avoided.
As shown in Fig. 7B, the amplitude of the mitochondrial response because of a second histamine challenge applied 9 min after the first, and measured by the freshly reconstituted aequorin, was almost 80% of that observed during a first challenge. Under the same conditions, the response of cells reconstituted with the standard protocol (i.e. incubation with coelenterazine for two hours prior to stimulation) was about 45% of the first. These data further support the notion that the drastic reduction of the [Ca 2ϩ ] increase observed upon a second challenge with histamine is largely because of a selective consumption of the probe and not to the desensitization of the mitochondrial uniporter.
To characterize in more detail the dynamics of aequorin responses both in the cytosol and in the mitochondrial matrix, measurements were made at different time points. As shown in Fig. 7C, when the second histamine pulse was applied 5 min after the second, the response was diminished in both compartments, but more drastically within mitochondria. From here onwards, the two compartments behaved very distinctively. Whereas the cytosolic response recovered to practically normal levels within 10 -15 min, the mitochondrial response took much longer to recover. Indeed, even when the second histamine pulse was delivered 90 min after the first, the peak amplitude still showed a 15% reduction.
To test whether the stability of the association between the two organelles depends on cytoskeleton integrity, the cells were treated with cytochalasin D or colchicine or both. The amplitude of the Ca 2ϩ response (30 min after the first challenge) in control experiments (7.49 Ϯ 2.59 M; n ϭ 8), and upon treatment with cytochalasin D (6.39 Ϯ 1.85 M; n ϭ 7), colchicine (6.37 Ϯ 1.54 M; n ϭ 10), or both (7.97 Ϯ 1.26 M; n ϭ 5), did not reveal significant differences. expressing mtAequorin were challenged twice with histamine with an interval of 9 min. The graph displays the response to the second stimulus in the case of short (black) and standard (gray) reconstitution protocols (expressed as a percentage of the first histamine response in each case) and are significantly different (**, p Ͻ 0.01), thus excluding the possibility of desensitization of the mitochondrial uniporter. The data represents mean Ϯ S.D. of nine and eight independent experiments, respectively. C, HeLa cells expressing cytAequorin or mtAequorin were challenged twice with histamine, at varying intervals. The graph displays the intensity of the second peak response (as a percentage of the first), expressed as mean Ϯ S.D.; significant differences between cytosolic and mitochondrial values are indicated by ** (p Ͻ 0.01) and * (p Ͼ 0.05). Taken together, the experiments carried out both at the single cell and at the population level make a strong case in favor of the existence of a stable, highly responsive, subpopulation of mitochondria. It could be argued that this highly responsive mitochondrial subpopulation takes up Ca 2ϩ more efficiently than the rest not because of its vicinity to the Ca 2ϩ release sites, but because its membrane potential (and thus the driving force for Ca 2ϩ accumulation) is higher (20). Were this the case, the same subpopulation of mitochondria would accumulate Ca 2ϩ with high efficiency, irrespective of the mechanism leading to the increase in [Ca 2ϩ ] c . To test this hypothesis, cytosolic Ca 2ϩ was increased in two different ways: by triggering efflux from the ER (via the InsP 3 receptors), and by activating influx across the plasma membrane (through storeoperated Ca 2ϩ channels).
The results presented in Fig. 8, panel A, show that the peak amplitude of the mitochondrial Ca 2ϩ uptake, elicited by Ca 2ϩ readdition to cells treated with cyclopiazonic acid (an inhibitor of the sarcoendoplasmic reticulum Ca 2ϩ -ATPase; the Ca 2ϩ influx is thus through the capacitative pathway), was hardly changed by a pre-challenge with histamine. Similarly, the Ca 2ϩ peak because of mobilization from stores (i.e. elicited by histamine in Ca 2ϩ free medium) was virtually unaffected by a previous activation of Ca 2ϩ influx (panel B). In a series of experiments, the peak because of capacitative Ca 2ϩ influx following a histamine challenge was ϳ87% of that elicited without any histamine challenge (3.87 Ϯ 0.45 versus 4.45 Ϯ 0.23 M; n ϭ 4 and 3, respectively). Similarly, the peak because of Ca 2ϩ mobilization showed a negligible difference to controls when elicited after the activation of capacitative influx (9.62 Ϯ 1.05 versus 9.42 Ϯ 1.55 M; n ϭ 4 and 3, respectively). For comparison, in the inset of Fig. 8, panel A, the major apparent reduction observed in the same batch of cells in a double histamine challenge is presented. Of interest, a double activation of Ca 2ϩ influx through the capacitative pathway resulted in a substantial apparent reduction (Ͼ50%) of the second response; this phenomenon was not reversed even when the second challenge was delayed by as much as 15-20 min (data not shown).

DISCUSSION
Since the initial discovery of rapid and efficient Ca 2ϩ uptake by mitochondria in intact cells, numerous fluorescent probes have been introduced to investigate in situ the properties of the uptake and release mechanisms, and the impact of mitochondrial Ca 2ϩ handling on the overall Ca 2ϩ signaling characteristics. The search for a reliable mitochondrial Ca 2ϩ indicator has been particularly intense in the last few years. None of the available probes appear ideal and all have different advantages and disadvantages. The use of rhod-2 is straightforward and the molecule can be loaded into virtually any cell. However, this dye is plagued by two major problems: it is partially mistargeted (depending on the cell type and the loading protocol), and responds to Ca 2ϩ changes simply with changes in F (i.e. it is not ratiometric). An additional problem that has been rarely (or not at all) addressed is the Ca 2ϩ buffering capacity of rhod-2. The possible influence of the dye in reducing the amplitude of [Ca 2ϩ ] m increases should be carefully considered.
Molecularly engineered GFP-based fluorescent Ca 2ϩ probes (Cameleons, Camgaroos, and Pericams), on the other hand, are theoretically superior to rhod-2 in terms of selectivity of localization and Ca 2ϩ buffering capacity. But they are not troublefree either: transfection is required (some cell types could therefore be more difficult to use), and the functional characterization of these probes as mitochondrial Ca 2ϩ indicators is still in its primordial stages.
We observed that mtCameleon has unexpected major problems of localization. Only in cells with very low levels of expression (and not always even in these) a selective targeting to the mitochondria was observed. We feel that this limitation is so important that we decided not to characterize this probe further.
Unlike mtCameleon, mtCamgaroo-2 is well directed to the matrix, although not as efficiently as mtGFP or mtAequorin. In fact, whereas the levels of mtGFP or mtAequorin in the nucleus and cytosol are below detection level, in the case of mtCamgaroo-2 the nuclear signal is 5-20% of that measured in mitochondria. The reasons for the partial mistargeting of calmodulin-containing GFP fusion proteins is presently unknown. Yet partial mistargeting is not the major problem of mtCamgaroo-2 as a Ca 2ϩ indicator, because the photoconversion phenomenon is by far more troublesome. Photoconversion is a known property of wild type GFP (21), but it has not been reported for GFP mutants. In mtCamgaroo2, the worse effect of this reversible photoconversion is that not only the intensity of the signal drops rapidly, but the residual fluorescence of the probe is insensitive to Ca 2ϩ . The reason why this effect was not noted by Griesbeck et al. (16) is probably because of the very slow sampling rates used during their experiments. These problems, allied to the fact that mtCamgaroo-2 responds to Ca 2ϩ increases only with an enhancement of fluorescence made us conclude that this probe is far from being an ideal Ca 2ϩ indicator, at least for mitochondria.
Ratiometric pericam is the latest addition to the growing family of GFP-based Ca 2ϩ indicators. Albeit not devoid of problems, at the moment this appears to be the less problematic probe for monitoring [Ca 2ϩ ] m at the single cell level. The advantages of PericamR are manifold. First, it exhibits very good targeting, even better than mtCamgaroo-2 (the residual nuclear signal is usually less than 10% of that in mitochondria). Second, it can be used in ratiometric mode. Third, even though some reversible photoconversion also occurs with this probe, it is far less than that observed with mtCamgaroo-2. Importantly, the residual fluorescent signal remains sensitive to Ca 2ϩ . Fourth, although we have not calculated accurately the matrix concentration of mtPericamR, the level of recombinant protein reached with other constructs rarely exceeds tenths of micromolar, whereas fluorescent probes usually reach concentrations of tens of micromolar (22). The main disadvantages that we have encountered with mtPericamR are its pH sensitivity and the effect of the local environment on the Ca 2ϩ affinity. As to the former, it seems to be an intrinsic and unavoidable characteristic of all insertional mutants of GFP. As to the latter, the low affinity of mtPericamR for Ca 2ϩ , once determined, turns out to be a serendipitous bonus that allows the investigation of a range of [Ca 2ϩ ] more appropriate for mitochondria.
Regarding the conclusions that can be drawn concerning the general response of mitochondria to Ca 2ϩ mobilization in cells, the data obtained with mtPericamR at the single cell level are in good agreement with those obtained with mtCameleon and aequorin. There are, however, a few major discrepancies with data obtained in the same cell system with rhod-2. Overall, we note the following four points. First, in agreement with data obtained with aequorin and mtCameleon, the average increase in mitochondrial Ca 2ϩ upon a histamine challenge is in the 10 M range, higher than the ϳ3 M calculated with rhod-2 (7). The simplest explanation of this difference is the extra buffering capacity provided by the fluorescent indicator, although differences in the calibration procedure cannot be excluded. Second, consistent with data obtained with all the indicators, the peak mitochondrial Ca 2ϩ is reached with a substantial delay (3-5 s) with respect to the nucleus and the cytoplasm. Third, in agreement with the results obtained with all other probes, also mtPericamR indicates that there is a substantial heterogeneity in [Ca 2ϩ ] m during InsP 3 -induced Ca 2ϩ release. Fourth, unlike the data obtained with rhod-2 in HeLa cells, [Ca 2ϩ ] m appears to oscillate in synchrony with the cytosol, even if the mitochondrial Ca 2ϩ oscillations are somehow attenuated. The group of Demaurex has also found mitochondrial Ca 2ϩ oscillations in HeLa cells using mtCameleon 3 and similar repetitive mitochondrial Ca 2ϩ increases mirroring repetitive cytosolic Ca 2ϩ changes have been reported in other model systems (8,9,(23)(24)(25).
Last, but not least, we found no evidence for substantial mitochondrial desensitization upon repetitive histamine challenges. These observations are also in agreement with data obtained by us and others in permeabilized cells (26,27). We note the discrepancy in relation to recently published data (7); in this case, the authors reported a strong desensitization of the mitochondrial Ca 2ϩ uniporter upon repetitive histamine pulses. The reason for this discrepancy is currently unknown, but may lie in the illumination protocol. Indeed, it is known that dyes closely related of rhod-2 (such as fura-2 and indo-1) undergo a photoisomerization process upon laser illumination, with the formation of fluorescent, but Ca 2ϩ -insensitive compounds (28,29). To our knowledge there is only one other report, in pancreatic ␤-cells, suggesting a possible desensitization of the mitochondrial Ca 2ϩ uniporter (27). In this latter case, however, desensitization was only marginal and required prolonged exposure of mitochondria to high Ca 2ϩ levels for up to 5 min.
From the cell biology point of view the most important conclusions concern the stability of the mitochondria-ER interactions. The high Ca 2ϩ microdomain model is widely accepted to explain the large and rapid Ca 2ϩ uptake that follows Ca 2ϩ mobilization from the ER. Until now it has not been clear whether the vicinity of mitochondria to the Ca 2ϩ channels is simply accidental or whether the two organelles are kept close to one another, particularly in the regions where Ca 2ϩ channels are located. The data obtained enable us to conclude that the mitochondrial subpopulation that is exposed to high Ca 2ϩ microdomains is in stable association with the ER regions where Ca 2ϩ release occurs maximally. These conclusions are based on the following three lines of evidence. First, the mtPericamR data show that the regions of highly responding mitochondria do not significantly change their intracellular distribution over prolonged periods. We note that the spatial resolution is not yet sufficient to monitor Ca 2ϩ changes at the single organelle level, but it is possible to obtain reliable regional data within each cell analyzed. Second, the apparent response of aequorin to a second challenge with histamine is apparently reduced much more than expected from the reduction in the cytosolic response, even 30 min after a first challenge with the stimulus. This in turn indicates that the mitochondria with burnt aequorin exchange very slowly with the rest of the population. Third, an initial challenge with histamine, while apparently dramatically inhibiting the Ca 2ϩ response to the same agent delivered several minutes later, does not significantly reduce the mitochondrial Ca 2ϩ peak elicited by activating the capacitative Ca 2ϩ influx, revealing that Ca 2ϩ mobilization from stores (and thus InsP 3 receptors) and Ca 2ϩ influx through the plasma membrane impinge on two distinct mitochondrial populations. These latter experiments demonstrate, in addition, that the mitochondria that undergo the major increase in [Ca 2ϩ ] (and thus burn their aequorin content) when Ca 2ϩ comes from the extracellular medium are similarly stably localized and undergo very slow exchange with the group of organelles that are mostly sensitive to Ca 2ϩ mobilization from stores.
The emphasis here is on the rapid Ca 2ϩ uptake, but it should be mentioned that a slow Ca 2ϩ accumulation in mitochondria can also occur and it may have important biological functions (7,30,31). Obviously the slow Ca 2ϩ uptake does not require the strict association with the Ca 2ϩ release sites, but rather depends on the bulk increase in [Ca 2ϩ ].
Our proposed model is based on dynamic, functional aspects of cell physiology and it fits well with previously determined static, structural aspects. Indeed, the structural proximity between mitochondria and ER has been well documented, both in animal and plant tissues; several ultrastructural studies have shown that parts of the ER are intimately associated with the mitochondrial outer membrane (32)(33)(34)(35)(36)(37). It should be noted that this structural organization, by placing the mitochondria stably and closely associated to the ER Ca 2ϩ release sites, maxi-mizes the efficiency of information transfer between the two organelles. As such, it most likely plays a prominent role in second messenger signal transduction. The question now is to determine not only the nature of the molecular components that allow this structural interaction to occur, but also to understand their relationship to the Ca 2ϩ release machinery of the ER.