Routes of Ca2+ Shuttling during Ca2+ Oscillations

Background: Ca2+ oscillations in mesothelial cells depend on Ca2+ influx. Results: However, the “lanthanum insulation method” renders the oscillations independent of extracellular Ca2+. Conclusion: Multiple pathways of Ca2+ shuttling are operating simultaneously during Ca2+ oscillations. Significance: Experimental and mathematical approaches shed light on the mechanism of Ca2+ oscillations.

The calcium ion (Ca 2ϩ ) is a universal intracellular messenger that controls a diverse range of cellular processes including cell proliferation, apoptosis, fertilization, neurotransmitter release, and heartbeat among many others (1). Ca 2ϩ pumps in the plasma membrane (plasma membrane Ca 2ϩ -ATPase) and in endoplasmic reticulum (ER) 2 membranes (SERCA) are responsible for the low cytosolic (c cyt ) and nuclear free Ca 2ϩ concentrations (c nucl ) (50 -100 nM) compared with the free Ca 2ϩ concentrations in the extracellular space (1-2 mM) and the ER lumen (c ER ) (100 -500 M). At rest, the free Ca 2ϩ concentration in the mitochondrial matrix (c mito ) is close to the resting c cyt , but it rises to 20 -30 M during stimulation, e.g. in motor nerve terminals in Drosophila melanogaster (2). Cell activation in a wide range of cell types results in Ca 2ϩ oscillations and in transient waves of increased c cyt (3)(4)(5)(6). These oscillations (or waves) are not restricted to c cyt , but also c nucl (7), c ER (8), and c mito show Ca 2ϩ oscillations (9). The spatial extent of the oscillatory Ca 2ϩ signal is also important. (i) In astrocytes, the area of Ca 2ϩ oscillations is sometimes restricted to only one protrusion regulating the release of gliotransmitters; i.e. different oscillatory frequencies can coexist at the same time within the same cell (10). (ii) In Xenopus laevis oocytes, regenerative spiral waves of release of free Ca 2ϩ spread through the entire cell (11). (iii) Intercellular Ca 2ϩ waves spreading via gap junctions occur in rat liver epithelial cells upon mechanical stimulation (12).
In cells maintained in vitro, serum starvation followed by readministration leads to intracellular Ca 2ϩ signals, most often in the form of oscillations (13,14). The precise mechanism(s) leading to these oscillations is poorly understood because serum contains a large number of known and as yet unidentified growth factors and mitogenic compounds, all potentially participating in this oscillatory activity (15). In Swiss 3T3 cells, serum-induced Ca 2ϩ changes are essential but not sufficient to induce NF-B activation and subsequent DNA synthesis (16). In some cell types, Ca 2ϩ oscillations even persist in the absence of Ca 2ϩ influx across the plasma membrane (3,4), whereas in others, Ca 2ϩ oscillations strictly depend on Ca 2ϩ influx (5,8).
Mitochondria influence cytosolic Ca 2ϩ oscillations in at least two ways. First, mitochondria produce ATP, which is required for SERCA and plasma membrane Ca 2ϩ -ATPase function, that results in Ca 2ϩ extrusion and thus lowering of c cyt . Second, during c cyt oscillations, c mito also manifests oscillations, indicative of a role of mitochondria in shaping and/or modulating c cyt oscillations (9). Ca 2ϩ uptake into the mitochondria is determined by both the large negative voltage (Ϫ150 to Ϫ180 mV) across the inner membrane that results from the proton pumping by the respiratory chain and the Ca 2ϩ concentration gradient between the cytoplasm and matrix (17). The mitochondrial calcium uniporter (MCU) is the key player responsible for the uptake of Ca 2ϩ by mitochondria (18). The MCU has a rather low Ca 2ϩ affinity and operates over a micromolar range of cytosolic Ca 2ϩ .
To address these questions, we performed lanthanum (La 3ϩ ) insulation experiments where both the Ca 2ϩ influx and efflux across the plasma membrane are blocked (19). We hypothesized that under these experimental conditions mitochondria serving as a Ca 2ϩ store/source might substitute for this function normally exerted by the extracellular space. Using a genetically encoded Ca 2ϩ indicator targeted to the mitochondria, we managed to verify this assumption in vitro. In addition, we investigated the effects of the following compounds on c cyt oscillations and mitochondrial Ca 2ϩ handling: the proton uncoupler carbonyl cyanide m-chlorophenylhydrazone (CCCP), the mitochondrial Na ϩ /Ca 2ϩ blocker CGP-37157, the mitochondrial MCU blocker Ru-360, and finally the "Ca 2ϩbuffering" protein calretinin. Based on the experimental findings, we built a mathematical model for Ca 2ϩ oscillations taking into account the various processes implicated in these oscillations.

Materials and Methods
Reagents-Thapsigargin, LaCl 3 , and EGTA were purchased from Sigma-Aldrich. CGP-37157 and BAPTA-AM were obtained from Tocris Bioscience (Bristol, UK). Ru-360 was purchased from Calbiochem, and Rhodamine 123 was from Invitrogen. EGTA-AM and tetramethylrhodamine methyl ester were purchased from AAT Bioquest (Sunnyvale, CA). CGP-37157 was dissolved in pure ethanol as 100 mM stock solutions. Thapsigargin and Rhodamine 123 were dissolved as 100 mM stock solutions in DMSO. BAPTA-AM and EGTA-AM were dissolved as 30 mM stock solutions in DMSO. LaCl 3 was dissolved in double distilled water. The final concentrations of the solvents were Ͻ0.1% in all experimental solutions. At these concentrations, the solvents did not modify the evoked Ca 2ϩ signals in control experiments (data not shown).
Plasmids and Lentiviral Infection-For the generation of cell lines stably expressing the Ca 2ϩ indicator proteins GCaMP3 (Addgene plasmid 22692 (22)) and mito-CAR-GECO1 (Addgene plasmid 46022 (23)), the lentiviral expression vector pLVTHM (Addgene plasmid 12247 (24)) was used. The GFP cassette in pLVTHM was replaced with cDNAs coding for the respective Ca 2ϩ indicator proteins. Briefly, pGCaMP3 was produced in SCS110 dam Ϫ bacteria and digested with AfeI and XbaI, and the fragment was inserted into the PmeI and SpeI sites of the backbone of pLVTHM to produce the final plasmid pLV-GCaMP3. The expression plasmid CMV-mito-CAR-GECO1 was used as template for the production of a DNA fragment coding for mito-CAR-GECO1. The required DNA fragment was synthesized by PCR using the following primers pairs: 5Ј-TAG CGT TTA AAC GGG CCC TC-3Ј and 5Ј-GAG AAC TAG TTT ACT TCG CTG TCA TCA TTT GTA C-3Ј. The amplicon was digested with PmeI and SpeI and inserted into the unique sites of the pLVTHM vector to produce the final pLV-mito-CAR-GECO1 plasmid. Calretinin overexpression was achieved by the help of a lentiviral system. Briefly, the GFP cassette in pLVTHM was replaced with the human CALB2 cDNA coding for full-length calretinin using the previously described expression plasmid RSV-CALB2-neo (25) as template. The DNA fragment coding for full-length calretinin was synthesized by PCR using the primers PmeI-CALB2 (5Ј-AGT CGT TTA AAC ATG GCT GGC CCG CAG CAG CAG-3Ј) and SpeI-CALB2 (AGT CAC TAG TTT ACA TGG GGG GCT CGC TGC A-3Ј). The amplicon was digested with PmeI and SpeI and inserted into the unique sites of the pLVTHM vector to produce the final pLV-CALB2 plasmid. We also generated a lentivirus expressing calretinin (CR) tagged with the enhanced blue fluorescent protein (EBFP) allowing for the easy identification of cells overexpressing EBFP-CR. For this, the pLV-EBFP2-nuc plasmid (Addgene plasmid 36085) and CMV-CALB2-neo were used. The DNA fragment coding for full-length calretinin was synthesized by PCR using the primers XhoI-CALB2 (5Ј-GAG ACT CGA GTA GCT GGC CCG CAG CAG C-5Ј) XbaI-CALB2 (5Ј-GAG ATC TAG ATT ACA TGG GGG GCT CGC TGC A-3Ј). The amplicon was digested with XhoI and XbaI and inserted into the unique sites of the pLV-EBFP2-nuc vector to produce the final pLV-EBFP2-CR plasmid. As a control plasmid coding for EBFP only (pLV-EBFP2-X), the nuclear localization signal was removed in the plasmid pLV-EBFP2-nuc by deleting an XhoI fragment. All lentiviral plasmids were verified by restriction enzyme digestion and sequencing. Lentivirus was produced by the calcium phosphate transfection method using HEK293T cells and three plasmids: one of the expression plasmids (e.g. pLV-GCaMP3 or pLVmito-CAR-GECO1), the envelope plasmid (pMD2G-VSVG, Addgene plasmid 12259), and the packaging plasmid (psPAX2, Addgene plasmid 12260). Virus-containing supernatants were collected after 48 and 72 h, filtered, aliquoted, and frozen at Ϫ80°C (26). MCU expression was knocked down in prMC expressing GCaMP3 and mito-CAR-GECO1 using Mission lentiviral transduction particles (Sigma-Aldrich) TRCN0000267404 and TRCN0000265169. Mission transduction particles directed toward human parvalbumin (TRCN000056549) and non-infected cells served as controls. Infected cells were selected using 2 g/ml puromycin dihydrochloride (Sigma-Aldrich) for 1 week. MCU transcript knockdown was verified by qRT-PCR analysis.
Calcium Imaging-prMC were isolated as described before (27) and grown on collagen-coated glass bottom 35-mm dishes (MatTek Corp., Ashland, MA). The buffer solution (Hepesbuffered saline) used for Ca 2ϩ imaging experiments contained 120 mM NaCl; 5.4 mM KCl, 0.8 mM Mg 2 Cl, 20 mM Hepes, 1 mM CaCl 2 , and 10 mM glucose, pH 7.4 (adjusted by NaOH). In the low Ca 2ϩ solution, CaCl 2 was replaced with an equimolar concentration of NaCl. The drugs (thapsigargin, FCS, and EGTA) were added to the solutions and remained in the solution until the end of the experiments. In some experiments, cells were pretreated either with 250 M CGP-37157 or with 10 M Ru-360 for 30 min at 37°C. Cells were loaded either with 30 M BAPTA-AM or 30 M EGTA-AM for 15 min at 37°C. We used a DMI6000 inverted confocal microscope integrated to a Leica TCS-SP5 work station to examine fluorescence signals indirectly, reporting c cyt or c mito . The following excitation wavelengths were used to illuminate the fluorophores: 488 nm for GCaMP3 and 561 nm for mito-CAR-GECO1. Fluorescence emissions were recorded with a 20ϫ objective and bandpass filters of 505-550 nm for GCaMP3 and 584 -683 nm for mito-CAR-GECO1. Fluorescence images for c cyt or c mito measurements were collected every 3 s. Circle-shaped regions of interest (ROIs) were placed inside the cytoplasmic area of cells. The fluorescence values were calculated after background subtraction (fluorescence intensity of regions without cells). Fluorescence intensity values were normalized in each experiment to the averaged basal value preceding the treatment period. A bleaching correction was carried out when the baseline was not stable. LAS-AF (Leica) and Prism5 (GraphPad Software, Inc., San Diego, CA) software were used for data analysis.
ATP Measurements-PrMC were starved in serum-free DMEM supplemented with 1% penicillin-streptomycin for 24 h and distributed into 15 centrifuge tubes (50,000 cells/tube) in 50 l of Hepes-buffered saline (ϩCa 2ϩ ). 1% FCS was added to 11 tubes, and 6 min later, 100 nM CCCP was added to five tubes. During the experiment, lysis buffer was added into each tube one after another with a delay of 1 min. ATP levels were determined using the ATP bioluminescence assay kit HS II according to the manufacturer's protocol (Roche Applied Science) with a microplate luminometer (PerkinElmer Life Sciences).

Mitochondrial Membrane Potential (⌬⌿) Measurements-
Mito-CAR-GECO1-expressing prMC were seeded on glass bottom Petri dishes and incubated with 10 M Rhodamine 123 for 20 min at room temperature. Cells were washed three times with Hepes-buffered saline (ϩCa 2ϩ ). During the recording using the confocal microscope, a 488-nm excitation wavelength was used to illuminate Rhodamine 123. Fluorescence emissions were recorded with a 20ϫ objective and bandpass filters of 505-550 nm for Rhodamine 123. The distribution of Rhodamine 123 between the mitochondrial matrix and cytosol is proportional to the mitochondrial membrane potential. As the mitochondrial network is distributed within the entire cytoplasmic space, circle-shaped ROIs were randomly assigned to the cytoplasmic region for the fluorescence intensity measurements. The signal intensity is proportional to the amount of Rhodamine 123 dye incorporated by mitochondria in this ROI. For the normalization and thus the measurement of Rhodamine 123 released by mitochondria, an ROI within the nuclear region not containing mitochondria was selected, and the fluorescence intensity in this ROI was determined. The relative (rel.) ⌬⌿ was calculated according to the following equation.
where F mito and F nucl are the fluorescence intensity of Rhodamine 123 in the mitochondrial and nuclear regions, respectively. The mitochondrial membrane potential was additionally measured with tetramethylrhodamine methyl ester. For these measurements, cells were preincubated with 50 nM tetramethylrhodamine methyl ester for 30 min. Estimation of the Intracellular Calretinin Concentration by Western Blot Analysis-Protein samples were isolated from cultured prMC. Cells were grown in 25-cm 2 flasks and harvested at near confluence. Total proteins were extracted with ice-cold radioimmune precipitation assay buffer. Serial dilutions of protein extracts (50, 5, 0.5, and 0.005 g) from each cell culture sample as well as 40 ng of purified human recombinant calretinin were loaded onto SDS-polyacrylamide gels (12.5%). After separation, proteins were transferred onto nitrocellulose membranes (Bio-Rad) and incubated overnight at 4°C with the calretinin-specific antibody CR7699/4 (Swant, Marly, Switzerland) at a dilution of 1:10,000. Rabbit secondary antibody linked to horseradish peroxidase (Sigma-Aldrich) was diluted at 1:10,000, and membranes were incubated for 4 h. For the detection, the chemiluminescent reagent Luminata Classico Forte (EMD Millipore Corp., Billerica, MA) was used. Chemiluminescent and normal illumination digital images were recorded on a system from Cell Biosciences (Santa Clara, CA). Area densities of calretinin bands were measured with ImageJ software. From the density curves, the cell protein concentration corresponding to 40 ng of calretinin was determined. This allowed determination of calretinin or more precisely that of the fusion protein EBFP-calretinin content in g/mg of total protein. Based on previous estimation of a protein concentration of about 0.2 g/ml (28) in mammalian cells, the intracellular concentration of EBFP-calretinin was estimated.
Frequency Determination and Amplitude Scan-Computerized peak recognition for frequency and amplitude analyses was realized via the Microsoft Excel 2010 environment as described before (8); normalized recordings from Ͼ30 oscillating prMC were evaluated. The oscillation frequency as well as the average amplitude was determined for three time windows: 1-5, 5-9, and 9 -13 min after serum administration.
Mathematical Simulation-To build the mathematical model, we considered four compartments: the extracellular space, cytoplasm, mitochondrial matrix, and ER lumen (Fig. 1). A fifth element placed within the cytoplasm in some simulations was the presence of a Ca 2ϩ buffer. Membrane junctions between the ER and the plasma membrane ensured that the functional unit components (Ca 2ϩ channels and pumps) are concentrated spatially in a very small space (29). Similarly close contacts were also assumed to exist between mitochondria and ER (30). One oscillatory unit represents an inositol trisphosphate receptor (InsP 3 R) cluster and its surrounding. We presumed that changes in c cyt , c ER , and c mito of the entire cell were similar to that of individual units, i.e. spatially homogenous. In our view, this simplification is acceptable because the oscillations are slow and the cell size is small. In this case, the spatial diffusion of Ca 2ϩ rapidly equilibrates the putative spatial differences and thus synchronizes the functions of individual functional units (31). In a cell with a 10-m diameter, the diffusion is estimated to equilibrate spatial heterogeneity in c cyt in less than 0.1 s (32). However, because Ca 2ϩ waves not only depend on Ca 2ϩ diffusion but also on the action of Ca 2ϩ pumps and channels, the Ca 2ϩ wave is ϳ10 times slower (33). Although our model is a minimal deterministic point model and cannot produce the stochastic and spatial phenomena of the Ca 2ϩ oscillations, it is a useful tool to illuminate the observed characteristics of the mitochondrial Ca 2ϩ handling. Our aim was to build the most simple model still able to produce the experimentally observed phenomena.
Ca 2ϩ transports across the plasma membrane were defined as J IN and J EFF , and the transports across the ER membrane were termed J SERCA and J EREFF , respectively. J IN includes Ca 2ϩ channels in the plasma membrane, e.g. voltage-gated Ca 2ϩ channels, transient receptor potential channels, store-operated channels, P2X purinoreceptors, hyperpolarization-activated cyclic nucleotide-gated channels, etc. The J EFF represents the pumps involved in Ca 2ϩ extrusion, plasma membrane Ca 2ϩ -ATPases and Na ϩ /Ca 2ϩ exchangers. The SERCA pumps transport Ca 2ϩ from the cytoplasm to the ER, whereas the J EREFF represents the ER channels involved in emptying the ER, ryanodine receptor and InsP 3 R. The function of the mitochondrial exchangers (J MEXC ) and the mitochondrial calcium uniporter (J MCU ) are responsible for the Ca 2ϩ transport across the mitochondrial inner membrane (see Fig. 1).
We denote by c cyt the Ca 2ϩ concentration (in nM) in the cytosol and by c ER that in the lumen of the ER. Mitochondrial matrix free concentration is denoted by c mito . The equations for the model are as follows.
where J IN is the flux of Ca 2ϩ ions entering the cell, J EFF is the Ca 2ϩ flux pumped out of the cell, J SERCA denotes the Ca 2ϩ flux pumped from the cytosol to ER, J EREFF is the flux of Ca 2ϩ passing from the ER to the cytosol, J MCU denotes the function of MCU, J MEXC displays the function of mitochondrial Ca 2ϩ exchangers (mitochondrial Na ϩ /Ca 2ϩ and H ϩ /Ca 2ϩ exchangers), and finally J ERLEAK represents a small flux of Ca 2ϩ diffusing from the ER to the cytosol (all values in nM/s). The constant ␥ is the ratio between the changes in c cyt and c ER caused by the same quantity of Ca 2ϩ ions transported through the ER membrane. This value is derived from the difference in the effective volume of the ER lumen and the cytoplasm and from the different fraction of free and protein-bound Ca 2ϩ in these compartments (34). The value of the ␥ parameter was estimated experimentally (8).
The quantity of Ca 2ϩ pumped out of the cell through the plasma membrane increases as a function of the Ca 2ϩ concentration in the cytosol. The Na ϩ /Ca 2ϩ exchangers have low Ca 2ϩ affinity but high capacity for Ca 2ϩ transport, whereas the plasma membrane Ca 2ϩ -ATPases have a high Ca 2ϩ affinity but a low transport capacity. Although the individual components of extrusion systems are usually modeled by Hill equations (35), the overall flux can be simulated by a simple linear equation (36) The dominant intracellular Ca 2ϩ release (J EREFF ) and uptake (J SERCA ) systems are localized in ER membranes. A small constant leak (J ERLEAK ) occurs independently of Ca 2ϩ channels. The functions of the mitochondrial Ca 2ϩ exchangers (Na ϩ /Ca 2ϩ exchanger and H ϩ /Ca 2ϩ exchanger) (J MEXC ) and the mitochondrial calcium uniporter (J MCU ) are responsible for the Ca 2ϩ transport across the mitochondrial inner membrane; intracellular Ca 2ϩ buffers (CaBP) such as calretinin acting as a transient cytosolic Ca 2ϩ store/source modulate temporal aspects of c cyt and consequently affect Ca 2ϩ fluxes across all membranes (plasma membrane, ER, and mitochondria (mito.)).
based on the experimental results of Herrington et al. (37).
where r e1 and r e2 are two positive constants.
SERCAs pump the Ca 2ϩ ions from the cytosol to the ER. The quantity of the transported Ca 2ϩ ions depends on c cyt levels. We assume a linear relationship because the ER influx is also composed of different SERCA pumps with different K d values (38). Nevertheless, our model can also work when J EFF and/or J SERCA is simulated with the conventional Hill equations.
where r s1 and r s2 are two positive constants.
Ca 2ϩ ions are released from the ER to the cytosol through InsP 3 R and ryanodine receptor. Because we found experimentally that ryanodine receptor does not play a role in seruminduced oscillations in mesothelial cells (8) similarly to other non-excitable cells (39), we focused on InsP 3 R. In our model, InsP 3 R is influenced both by c cyt and by c ER but without an allosteric regulation between the two. InsP 3 R has Ca 2ϩ binding sites not only on the cytoplasmic side but also on the luminal side (40). Experimental data show that an increase in inositol trisphosphate (InsP 3 ) concentration causes a significant Ca 2ϩ release from the ER in the absence of cytosolic Ca 2ϩ (c cyt ϭ 0) (41). Moreover, the effects of luminal Ca 2ϩ do not affect the cytosolic binding sites (42,43). Therefore we modeled InsP 3 R function as the sum of two individual contributions.
with positive constants and r i1 .
We introduced the dependence of InsP 3 R on the [InsP 3 ], which has an influence both on J cytdep and on J ERdep . According to the experimental data from several studies (44 -46), elevating c IP3 mainly changes the mean and the maximum ( and r i,max ) of the bell-shaped curve of c cyt dependence. Nevertheless, based on the experimental data presented (47,48), elevating c IP3 also has an effect on the loading of the ER. Increased c IP3 reduces the amount of the stored Ca 2ϩ ions. We simulated this effect by changing the r i2 parameter.
where K b is the half-saturation constant of InsP 3 R for InsP 3 and c IP3 represents the InsP 3 R sensitivity to the inositol trisphosphate molecule, which was taken equal to inositol trisphosphate concentration in M. max , min , r im,min , r im,max , r i2,min , and r i2,max are positive constants. The parameter J ERLEAK accounts for a Ca 2ϩ flux from the ER to the cytoplasm independently of known Ca 2ϩ channels, and this parameter is assumed to represent a small constant value (49).
The outer membrane of mitochondria is freely permeable for Ca 2ϩ ions, but the inner mitochondrial membrane provides a barrier. The constant is the ratio between the changes in c cyt and c ER caused by the same quantity of Ca 2ϩ ions transported through the mitochondrial inner membrane. This value is derived from the difference in the effective volume of the mitochondrial matrix and the cytoplasm and from the different fraction of free and protein-bound Ca 2ϩ in these compartments.
There is a fast Ca 2ϩ influx into the mitochondria matrix if c cyt reaches a certain value. This fast influx is attributable to the function of MCU. We used Hill equations with a very high Hill coefficient as was done in the work of Marhl et al. (50). For simplicity, we did not take into account the changes in mitochondrial transmembrane potential and in mitochondrial volume during the Ca 2ϩ oscillations in line with Marhl et al. (50), but we have to consider it during a protonophore treatment. The passage of calcium ions through the MCU requires the large membrane potential difference generated by the action of the electron transport chain (51).
where r ⌿ and v MCU,max are positive constants, K d,MCU is the dissociation constant of MCU for Ca 2ϩ ions, and H is the Hill coefficient. In our model, J MCU has a constant basal activity. That ensures that mitochondria can store a small amount of Ca 2ϩ ions, which are released into the cytoplasm immediately after the collapse of the mitochondrial membrane potential.
Higher ⌬⌿ means increased Ca 2ϩ uptake but slower mitochondrial Ca 2ϩ release.
To simulate the function of mitochondrial exchangers (Na ϩ / Ca 2ϩ and H ϩ /Ca 2ϩ exchangers), we consider that both will transport Ca 2ϩ ions with a low velocity when there is a concentration gradient between the two sides of the mitochondrial inner membrane. For the simplicity, we neglected the changes in sodium and proton concentrations during the Ca 2ϩ oscillations. Depending on the calcium concentration gradient, the exchangers can work in both directions.
where r m1 and r m2 are positive constants.
The Ca 2ϩ influx across the plasma membrane is composed of passive leakage and the agonist-activated fluxes: the capacitive (store-operated channel-dependent) and the non-capacitive (arachidonate-or diacylglycerol-regulated) Ca 2ϩ influx (52). We simulated the changes in J IN starting from the beginning of the administration of serum (t 1 ) using the following equations.
We simulated the changes in c IP3 from the beginning of the administration of serum (t 1 ) with the following equations. The resting c IP3 was set to 15 nM (53).
To simulate the effect of calretinin, we neglected its fast kinetics. Because this protein is considered as a fast Ca 2ϩ buffer (54), calretinin reaches the Ca 2ϩ steady state in a few milliseconds, which is much faster than our observed Ca 2ϩ changes lasting for a few seconds. The fast kinetics of calretinin plays an important role at the mouth of voltage-gated Ca 2ϩ channels in excitable cells (55) where fast and large changes in Ca 2ϩ concentrations are expected.
where c CR is the concentration of calretinin in the cytoplasm and is the average number of the Ca 2ϩ binding sites of calretinin occupied by Ca 2ϩ . Calretinin has four high affinity Ca 2ϩ binding sites and one low affinity binding site. The binding kinetics of the Ca 2ϩ binding sites were simulated with Hill equations.
where K d1 is the dissociation constant for the high affinity Ca 2ϩ binding sites, K d2 is the dissociation constant for low affinity Ca 2ϩ binding site, and h is the Hill coefficient for the high affinity binding sites. Among the high affinity Ca 2ϩ binding sites, there is a positive cooperativity (h Ͼ 1). The values for the parameters came from the study of Faas et al. (54).
The values of each parameter are listed in Table 1. The initial values of parameters are derived either from our experiments in primary mesothelial cells or from fitting to experimental data previously reported in the above mentioned articles. The presented values are the result of the sequential fitting of the initial values to our in situ recordings. All computations of the model were implemented in the Microsoft Excel 2010 environment. The model system was discretized with a temporal resolution of 0.1 s (supplemental Excel document). There were no significant differ-ences in the solution of the differential equations if we increased the temporal resolution (not shown). For visualization, Prism5 (GraphPad Software, Inc.) software was used.

Results
Characterizing Ca 2ϩ Fluctuations in Mitochondria of Primary Mouse Mesothelial Cells-In the absence of serum, prMC did not show Ca 2ϩ oscillations as reported before (8). However, in a small fraction of cells (2-3%), isolated arrhythmic mitochondrial increases in c mito were present without detectable changes in c cyt ( Fig. 2A). The addition of 1% FCS to the cell culture medium containing prMC that were grown in the absence of serum for 24 h resulted in a sudden rise of c cyt lasting, on average, for ϳ40 s followed by Ca 2ϩ oscillations (Fig. 2,  B-D). The percentage of prMC responding to serum readministration with Ca 2ϩ oscillations was in the order of 70%. Nonoscillatory cells showed only an initial single Ca 2ϩ transient or a so-called peak-plateau response (53). A wide range of different oscillatory patterns in c cyt was present in a supposedly homogenous population of mesothelial cells. Most cells displayed long period (Ͼ10 min) baseline spiking oscillations with various frequencies of one spike per 3 min (Fig. 2B) up to 10 per min (Fig.  2D); also maximal spike amplitudes varied between individual cells. The baseline spiking oscillations represent discrete Ca 2ϩ transients starting from a constant basal c cyt level (Fig. 2, B and  C). Sinusoidal oscillation is a term for a continuous fluctuation in c cyt starting from a c cyt value that is higher than the resting (basal) c cyt (Fig. 2D). Most probably, sinusoidal oscillations are the result of high frequency overlapping baseline spiking oscillations (56). In prMC maintained in cell culture medium for longer periods (Ͼ10 passages), the percentage of the cells showing sinusoidal oscillations was increased as exemplified in Fig.  2D. However, the percentage of cells showing oscillatory activity was rather low at higher passages (ϳ20 -40% at passages Ͼ10). The average frequency of the baseline spiking oscillations was found to be 15, 13, and 13 mHz in the following time segments: 1-5, 5-9, and 9 -13 min after serum administration, respectively. The average amplitude of spikes (-fold increase in GCaMP3 fluorescence intensity) was found to be 2.66, 2.53, and 2.53 in the above mentioned time segments. By using two Ca 2ϩ indicators targeted to either the mitochondrial matrix (mito-CAR-GECO1) or the cytoplasm (GCaMP3), we simultaneously monitored changes in c mito and c cyt , respectively. The initial serum-induced rise in c cyt was paralleled by a rapid rise in c mito that reached the peak value within 30 s after the addition of 1% FCS (Fig. 2, B-D); from then on, c mito decreased continuously until it reached its initial basal value, generally within the time span of 15 min monitored in most experiments. High frequency oscillations in c cyt resulted in continuous elevation in c mito (Fig.  2D). The rate of decay in c mito was also rather variable between cells, and oscillations in c cyt did not stop when c mito had reached its basal levels. In some cells during the decreasing phase in c mito , small fluctuations (short rises in c mito ) coincided with the c cyt spikes but with a small delay (e.g. shown in Fig. 2B, inset).
For the modeling, we took into account our previous results where basal and maximal c cyt values during Ca 2ϩ spikes in prMC were found to be 100 and 200 -300 nM, respectively (8).
Similarly, the values for the resting c ER were taken as 150 -250 M, and the values after serum readministration were taken as 100 -150 M (8). The pattern of c ER changes is best described as a sawtooth wave (8). These data were incorporated to build the mathematical model where Ca 2ϩ concentrations in all compartments (c cyt , c mito , and c ER ) were calculated and fitted to one c cyt recording (Fig. 2E). The model accurately recapitulated the experimental findings, in particular with respect to c mito , which had not been modeled in our previous study (8). The pattern in c mito is best described as a sudden rise after serum readministration followed by a rather smooth decay phase with small humps (increases in c mito ) as the result of the oscillatory Ca 2ϩ spikes.

Modulation of Mitochondrial Ca 2ϩ Transport Mechanisms Affects Ca 2ϩ
Oscillations-In the next series of experiments and modeling simulations, we investigated how altering mitochondrial function, mostly in relation to Ca 2ϩ handling, affects c cyt oscillations. The simulation showed that inhibition of the mitochondrial Ca 2ϩ release (J MEXC ) or mitochondrial Ca 2ϩ uptake (J MCU ) during Ca 2ϩ oscillations decreased the oscillation frequency (Fig. 3, A and B). The experimental verification of our predictions was hampered by the absence of pharmacological mitochondrially targeted compounds that immediately reach the mitochondrial inner membrane when added to the recording solution. Briefly, after serum readministration, the addition of CGP-37157 (50 M), a nonspecific blocker of the mitochondrial Na ϩ /Ca 2ϩ exchanger, had no effect on the patterns of Ca 2ϩ oscillations in either c cyt or c mito (data not shown). However, some cells pretreated with CGP-37157 for 30 min displayed stairlike increases in c mito (Fig. 3C), an effect predicted from our model (see Fig. 3A, right part (c mito )). Ruthenium compounds, e.g. ruthenium red and Ru-360, are potent and effective blockers of MCU in isolated mitochondria, but their usefulness for intact cells is limited by their poor membrane permeability and selectivity (57). Pretreatment of cells with 10 M Ru-360 reduced the average oscillation frequency (approximately a 30% decrease during each time segment) and the initial mitochondrial Ca 2ϩ uptake (Fig. 3D). However, it is currently still unclear whether in intact cells Ru-360 acts uniquely by the inhibition of the mitochondrial Ca 2ϩ uptake or additionally by the inhibition of the extracellular Ca 2ϩ influx. Therefore, we also used a molecular approach, i.e. down-regulation of MCU by shRNA to decrease the mitochondrial Ca 2ϩ uptake. The down-regulation of MCU mRNA levels by 60 -90% as determined by qRT-PCR resulted in a 40 -60% decrease in the initial mitochondrial Ca 2ϩ uptake (Fig. 3E). This led to an ϳ20% reduction in the c cyt oscillation frequency calculated by frequency scan analysis (Fig. 3F) in line with the predictions from our model (Fig. 3B). In all time windows (bins of 3 s), the oscillation frequency was lower in prMC where MCU had been down-regulated. Thus, both approaches (Ru-360 and shMCU)  NOVEMBER 20, 2015 • VOLUME 290 • NUMBER 47

JOURNAL OF BIOLOGICAL CHEMISTRY 28221
underscore the importance of mitochondria in Ca 2ϩ oscillations.
CCCP is an inhibitor of oxidative phosphorylation by acting as a protonophore; i.e. it allows H ϩ to cross the inner mitochondrial membrane, resulting in the collapse of ⌬⌿. During Ca 2ϩ oscillations, ⌬⌿ was slightly increased (more negative), but it collapsed immediately after CCCP (100 M) treatment (Fig.  4A). The collapse of the membrane potential after addition of 10 or 100 M CCCP was also confirmed by using the tetramethylrhodamine methyl ester indicator dye (data not shown).
When applied during Ca 2ϩ oscillations resulting from serum readministration, CCCP blocked Ca 2ϩ oscillations at 100 M but not at 10 M (data not shown). An immediate drop in c mito was observed after CCCP treatment (Fig. 4B) followed by a continuous elevation in c mito in some cells (ϳ20%) but not in others (Fig. 4, A and C). In a few cases (ϳ5% of cells), administration of ATP (1 M) partially reverted the CCCP-induced oscillation stop (Fig. 4C). However, addition of ATP to the recording solution in the absence of serum was equally able to evoke Ca 2ϩ oscillations in some cells (data not shown). The reason for this FIGURE 4. Effect of the proton uncoupler CCCP on Ca 2؉ oscillations. A, serum addition to prMC led to a rapid rise in c mito (red trace) followed by a gradual decay. Addition of CCCP (100 M) resulted in a rapid collapse of the membrane potential as evidenced by measuring Rhodamine 123 fluorescence signals (green trace). As the result of CCCP application, c mito immediately returned to basal levels because the slightly elevated c mito level (compared with the c cyt level) could not be maintained when the driving force for mitochondrial Ca 2ϩ uptake was eliminated. B-H, the role of the membrane potential and ATP treatment was investigated in greater detail, and one representative recording for each experimental condition is depicted. B, the addition of CCCP (100 M) at t ϭ 9 min blocked the serum-induced cytosolic Ca 2ϩ oscillations. CCCP treatment led to a rapid fall in c mito ; in some cells, a slow increase in c mito occurred afterward (B), whereas in others, c mito remained low (not shown). C, administration of ATP (1 M) reestablished the CCCP-inhibited Ca 2ϩ oscillations in some cells. D, addition of CCCP (100 M) prior to serum readministration lowered the resting c mito (also reflected by the simultaneous small increase in c cyt ). However, from then onward, c cyt dropped to levels lower then the basal c cyt before CCCP treatment. The serum readministration at t ϭ 8 min evoked only a single Ca 2ϩ transient. E, at a lower CCCP concentration (10 M), addition of 1% FCS at t ϭ 6 min resulted in elevations in c cyt and c mito followed by a few oscillations in c cyt . The small amount of Ca 2ϩ ions taken up by mitochondria (red trace) during Ca 2ϩ spikes (green trace) was released back to the cytosol almost immediately. F, in this simulation experiment, the mitochondrial membrane potential was switched off (⌬⌿ ϭ 0) at t ϭ 3 min; this resulted in a slight increase in c cyt and a decrease in c mito to values lower than basal c mito as observed experimentally in D and E. Serum readministration evoked oscillatory activity in c cyt and c mito . G, after the CCCP-induced collapse in ⌬⌿, ER Ca 2ϩ release by thapsigargin (5 M) resulted in mitochondrial Ca 2ϩ uptake (red trace) independent of ⌬⌿. A representative recording displays the simultaneous changes in c mito (red) and c cyt (green). H, during FCS-induced Ca 2ϩ oscillations, ATP levels in prMC were increased but dropped quickly after CCCP administration. The panel shows the mean ϩ S.D. of three independent experiments. Error bars represent S.D. effect is currently unknown; ATP might act on receptors on the surface of prMC but was also shown to cross the plasma membrane and to have an impact from the intracellular side (58). Application of CCCP before serum administration led to an immediate fall in the basal c mito , reaching a new plateau 1-2 min later; the fall in c mito was accompanied by a visible small rise in c cyt in ϳ20% of cells, indicative of a release of mitochondrial Ca 2ϩ to the cytosolic compartment. In addition, CCCP also decreased the basal level in c cyt (both at 100 and 10 M), signifying that also the plasma membrane potential was affected. Serum readministration following CCCP (100 M) treatment was still able to briefly elevate both c cyt and c mito , but the increase in c mito was smaller compared with cells not treated with CCCP; moreover, c mito returned quickly to the level reached after CCCP addition, i.e. not to basal levels before treatment (Fig. 4D). Serum readministration after treatment with the lower CCCP concentration (10 M) evoked low amplitude Ca 2ϩ oscillations, and the mitochondrial Ca 2ϩ rise during a cytosolic Ca 2ϩ spike was small, and c mito immediately returned to levels before serum administration but to lower levels than the basal c mito before CCCP administration (Fig. 4E). The model also correctly predicted that the collapse in ⌬⌿ (at t ϭ 3 min) resulted in a lower c mito . Serum administration (modeled as increasing J IN and c IP3 ) led to an increase in c cyt and c mito followed by oscillations in c cyt and c mito (Fig. 4, compare F with the experimental recording shown in E). To provide more evidence for the presence of ⌬⌿-independent mitochondrial Ca 2ϩ uptake as shown in Fig. 4D, we induced Ca 2ϩ release from the ER by thapsigargin after CCCP administration (Fig. 4G). We observed a rise not only as expected in c cyt but in parallel also in c mito , confirming the existence of a ⌬⌿-independent mitochondrial Ca 2ϩ uptake. Addition of 1% FCS also resulted in an increase in the intracellular ATP concentration that lasted during the entire period of Ca 2ϩ oscillations. Shortly after the collapse of ⌬⌿ induced by CCCP, an immediate fall in ATP levels was observed (Fig. 4H). Overall, our findings indicate that the oscillation stop induced by the protonophore CCCP is not exclusively the result of the decreased mitochondrial Ca 2ϩ uptake but also mediated via CCCP-induced changes in plasmalemmal Ca 2ϩ influx and decreased ATP production.

The Role of Ca 2ϩ Influx on Ca 2ϩ Oscillations and on Mitochondrial Ca 2ϩ Handling-A decrease in extracellular [Ca 2ϩ
] by the addition of 0.25 mM EGTA to the extracellular solution resulted in a reduction in the oscillation frequency (Fig. 5A). In this condition, i.e. when c cyt oscillations were not blocked completely, the amplitude of the Ca 2ϩ signals was not affected, and the pattern of mitochondrial Ca 2ϩ release/uptake was not affected (Fig. 5A). This could be accurately modeled in our simulation (Fig. 5B). When oscillations were induced by the addition of 1% FCS to the Ca 2ϩ -containing recording solutions ([Ca 2ϩ ] o Ϸ 1 mM), decreasing [Ca 2ϩ ] o to Ͻ1 M by the addition of 10 mM EGTA at t ϭ 9 min resulted in an immediate stop of the oscillations, indicating the necessity of Ca 2ϩ influx for the oscillations in c cyt (Fig. 5C). Removal of the extracellular Ca 2ϩ had no visible effect on the decay curve of c mito , and basal levels were reached at the end of the observation period (15 min). When the serum readministration was carried out in the "Ca 2ϩ -free" condition, most prMC did not show any response in c cyt . In ϳ5% of prMC, an initial small rise in c cyt was observed but without signs of Ca 2ϩ oscillations in support of the hypothesis that extracellular Ca 2ϩ is essential for the sustained oscillations (Fig. 5E). Interestingly, different results were obtained in the Ca 2ϩ -free condition when both Ca 2ϩ influx and efflux across the plasma membrane were blocked by the addition of 1 mM La 3ϩ , the so-called lanthanum insulation (19,59), prior to the serum readministration. After serum addition, an immediate rise in c cyt and c mito was detected; although c cyt decayed to basal levels within the next 2 min, c mito remained elevated and did not show the typical decay curve as seen e.g. in Fig. 5A, C, and E. Moreover, long lasting but slow oscillations in c cyt were observable (Fig. 5G), and at each cytosolic Ca 2ϩ spike, a corresponding spike in c mito occurred. This indicates that during the La 3ϩ insulation a considerable amount of Ca 2ϩ ions released from the ER, leading to the transient increase in c cyt , is taken up by mitochondria as evidenced by the mitochondrial Ca 2ϩ spikes (Fig. 5G). Thus, blocking the Ca 2ϩ efflux across the plasma membrane leads to a shuttling of the Ca 2ϩ ions between the ER and mitochondria, leading to these slow oscillations. Of note, the mitochondria remain in a rather Ca 2ϩ -loaded state because Ca 2ϩ cannot be transported out of the cell. We estimate that the mitochondrial Ca 2ϩ uptake and release velocities likely determine the frequency of Ca 2ϩ oscillations. La 3ϩ -induced blocking of the Ca 2ϩ transport across the plasma membrane at a time point when serum-induced Ca 2ϩ oscillations were ongoing led to a complete block of the oscillations (Fig. 5,  I and K). In some cases, La 3ϩ treatment caused a final longer lasting Ca 2ϩ spike (Fig. 5K), whereas in other prMC, La 3ϩ completely blocked any further spikes (Fig. 5I). In all cases, the mathematical model could truthfully recapitulate the experimental findings by changing the parameters J IN and J EFF at different time points (Fig. 5, D, F, H, J, and L). Of note during La 3ϩ insulations, the width (duration) of Ca 2ϩ spikes was wider (longer) both in vitro and in silico. Moreover, the La 3ϩ -evoked oscillation block in the presence of extracellular Ca 2ϩ ([Ca 2ϩ ] o Ϸ 1 mM) is, according to our model, mostly due to the decreased levels of Ca 2ϩ ions present in the different cell compartments; i.e. the sum of c cyt ϩ c ER ϩ c mito is smaller than the sum prior to agonist administration.

Effect of the Intracellular Buffer Calretinin on Ca 2ϩ
Oscillations-Based on previous findings that human mesothelioma in vivo, mesothelioma cells in vitro, and reactive mesothelial cells express calretinin (60, 61), we hypothesized that prMC also might express this protein and that its presence might affect the Ca 2ϩ oscillations. However, calretinin protein expression levels in prMC were found to be below the detection limit of our Western blot analysis (8), i.e. lower than ϳ100 nM and thus unlikely to affect the Ca 2ϩ oscillations as the result of the Ca 2ϩ -buffering capacity of calretinin. In support of this assumption, oscillation patterns (frequency, amplitude, and duration) in prMC from either wild type or calretinin knockout (CRϪ/Ϫ) mice were indistinguishable (data not shown). However, to mimic the situation of calretinin-expressing reactive mesothelial cells and to investigate the putative role of calretinin in those cells, we overexpressed a fusion protein consisting of EBFP separated from full-length calretinin by a small linker peptide by infection of prMC with the appropriate lenti-virus. We estimated in a semiquantitative way by Western blot analyses the expression levels of EBFP-calretinin. The expression level was found to be ϳ75 pg of EBFP-calretinin/g of total protein, leading to an estimated upper concentration of 250 M calretinin. The EBFP tag on calretinin served as a marker for the distinction of the two populations with or without calretinin (Fig. 6A). The percentage of infected cells was usually higher than 90%. The fraction of prMC showing Ca 2ϩ oscillations (ϳ10 -20%) was considerably lower than in non-infected cells not expressing calretinin (ϳ60 -70%) and moreover was restricted to cells showing faint blue fluorescence, i.e. low EBFP-calretinin expression levels. In the oscillating EBFP-calretinin-expressing prMC, the Ca 2ϩ spike amplitudes were smaller, and the half-width of Ca 2ϩ spikes (duration) was increased (Fig. 6B). The largest effect caused by EBFP-calretinin was the reduction of the amplitude of the first Ca 2ϩ spike after serum readministration (Fig. 6C); on average it was half the size compared with the situation without calretinin. Likely as the consequence of the reduction in c cyt , the increase in c mito also was clearly diminished (Fig. 6D). The frequency of oscillation slightly decreased (ϳ10 -20% reduction in each time segment). Our model simulations incorporating calretinin with the known Ca 2ϩ binding characteristics (54) showed similar modifications: a decrease both in the amplitudes of c cyt spikes and in the amount of mitochondrial Ca 2ϩ uptake (Fig. 6E). In our model, an increase in calretinin concentration resulted in an increase of the oscillation frequency, a prediction not supported by our experimental findings. One reason may be that calretinin, in addition to its buffering capacity, might act as a Ca 2ϩ sensor in prMC. We had previously shown that calretinin is able to directly modify the activity of a Ca 2ϩ channel (60), and direct targets for calretinin implicated in Ca 2ϩ transportation might also be present in prMC. As a control to exclude that observed effects were mediated by the EBFP part of the fusion protein, prMC were infected with the lentivirus LV-EBFP2-X leading to the expression of EBFP only. No differences in the Ca 2ϩ oscillations patterns were observed between cells expressing EBFP and non-infected control cells (data not shown). Based on the fact that Ca 2ϩ oscillations in EBFP-calretinin-expressing cells were limited to those with faint fluorescence, we reasoned that the concentration in these cells was ϳ10-fold (e.g. 25 M) lower than the global concentration (250 M) estimated from Western blot analyses. Thus, we tested whether the commonly used synthetic Ca 2ϩ chelators BAPTA and EGTA, which have different properties (e.g. K d , on-rate constant (k on ), and diffusion coefficient (D)) than calretinin, were able to recapitulate the effects of calretinin. In prMC loaded with BAPTA-AM (30 M), serum readministration evoked a slow and prolonged c cyt elevation paralleled by a minute increase in c mito (Fig. 6F). Most importantly, the initial rise in c cyt as also seen in EBFP-calretinin-expressing prMC (Fig. 6C) was completely abolished. In contrast, after EGTA-AM loading serum, readministration induced a short spike both in c cyt and c mito (Fig. 6E) followed by a rapid return to essentially baseline levels. No Ca 2ϩ oscillations were observed in both cases. This further indicates that the properties of calretinin are clearly distinct from those of either BAPTA or EGTA.

Discussion
Characteristics of mitochondrial Ca 2ϩ transport have not been examined in detail in most cell types. The main reason why we know relatively little about mitochondrial Ca 2ϩ handling is because the molecular identity of the channels involved in mitochondrial transport have only recently been discovered (18,62,63), and specifically targeted, pH-and ⌬⌿-insensitive Ca 2ϩ indicators are only currently available (23). Nevertheless, there are few models for Ca 2ϩ oscillations where the function of mitochondrial Ca 2ϩ uptake has been taken into account (64).
Our experiments affirm previous data that mitochondria, even at the resting state, are able to transport and store Ca 2ϩ ions (65). The fast release of the stored Ca 2ϩ from the mitochondria due to the decrease/collapse of the membrane potential indicates that the strongly negative ⌬⌿ ensures a constant Ca 2ϩ uptake into the mitochondria. This uptake is in a steadystate equilibrium with the constant Ca 2ϩ efflux mediated by the mitochondrial exchangers (66), and the efflux is an electrogenic process (67). The electrochemical proton gradient across the inner mitochondrial is used to remove the excess Ca 2ϩ ions (68). Our recordings show that this basal steady-state mitochondrial Ca 2ϩ concentration can fluctuate, showing "spontaneous" mitochondrial Ca 2ϩ spikes. Most probably this is mediated by an endogenous MCU activator that has not been identified at the molecular level yet. Ca 2ϩ transients in c cyt were previously reported to evoke an increase in c mito , activating both cytoplasmic (19) and mitochondrial enzymes (2). Thus, Ca 2ϩ transients observed selectively in c mito in some prMC ( Fig.  2A) might allow for the autonomous activation of mitochondrial enzymes. The Ca 2ϩ ions causing the mitochondrial spike are likely to originate from the cytosolic compartment; however, our results indicate that the amount of Ca 2ϩ ions responsible for the increase in c mito was not sufficient to be detected as a decrease in c cyt . Alternatively, at basal conditions, the equilibrium level of c cyt might be regulated by a rather rapid constant exchange of Ca 2ϩ ions among the cytosol, the extracellular space, and/or the ER compartment.
The Ca 2ϩ oscillation models usually differ in how they simulate the functions of InsP 3 R, the channel that transports Ca 2ϩ ions from the ER to the cytosol. The "c cyt /[InsP 3 ]" models (for a review, see Ref. 69) postulate that the InsP 3 R has a binding site for InsP 3 , an activating binding site for Ca 2ϩ , and an inhibiting binding site for Ca 2ϩ . In these models, all binding sites are localized on the cytoplasmic side, and the function of InsP 3 does not depend on c ER . Binding of Ca 2ϩ to the activating site and of InsP 3 to the InsP 3 binding site opens the channel, whereas Ca 2ϩ binding to the inhibiting site closes the InP 3 R. Moreover, the binding of Ca 2ϩ to the inhibiting site occurs rather slowly and with a lower affinity as compared with the activating site, subsequently resulting in oscillations in c cyt . In these models, the InsP 3 concentration uniquely determines the oscillation frequency (70). In the "store loading" models (also called "c cyt /c ER " models), the function of InsP 3 R depends not only on c cyt but also on c ER . In these models, the Ca 2ϩ influx across the plasma membrane plays a critical role in determining the oscillation frequency (8,71,72). At a constant [InsP 3 ], the duration of the interspike period is determined by the velocity of cellular Ca 2ϩ replenishment, which is manifested as a continuous ER loading together with a constant basal c cyt . The experimentally observable sawtooth wave oscillations in c ER during the cytoplasmic baseline spiking oscillations are an important argument in favor of the store loading theory (8). However, the store loading-based models cannot cope with the fact that in some cells the Ca 2ϩ oscillations do not depend on Ca 2ϩ influx across the plasma membrane. Our experiments and modeling studies revealed that the incorporation of mitochondria as an additional Ca 2ϩ source/store in the store loadingbased models considerably augments the quality of the simulations. That is, the modeling predictions are more congruent with the experimental findings, which allows for a better mechanistic understanding. The mitochondrial Ca 2ϩ transport enables the store loading-based models also to display Ca 2ϩ oscillation in the absence of extracellular Ca 2ϩ .
The simulation of the La 3ϩ insulation was previously endeavored by Sneyd et al. (73). Although their model does not contain mitochondria and moreover c cyt is continuously decreasing during the oscillations, their model reveals important aspects of the Ca 2ϩ oscillations, namely their dependence on the total Ca 2ϩ load of the cell. In their model, the cell has a high resting Ca 2ϩ ; upon agonist stimulation, the activation of plasma membrane Ca 2ϩ -ATPases causes a net loss of Ca 2ϩ from the cells even though the Ca 2ϩ influx is augmented after stimulation (73). A similar phenomenon is also observed in our model; the total cellular Ca 2ϩ content (c cyt ϩ c ER ϩ c mito ) determines the response to the La 3ϩ insulation; blocking of the Ca 2ϩ influx and efflux results in an oscillation stop that can either occur after a final Ca 2ϩ spike or directly after La 3ϩ addition, i.e. without a change in c cyt . In contrast to the previous model (73), basal c cyt levels during the interspike phase of the oscillations remain constant. This is in line with the experiments carried out by us and others (74).
Shuttling of Ca 2ϩ ions between the ER and mitochondria was experimentally demonstrated in the study of Ishii et al. (9). They reported that in HeLa cells the cycles of ER/mitochondrion shuttling are repeated until c mito has reached the basal level prior to the stimulation. In our study with prMC, we observed Ca 2ϩ oscillation even (i) when c mito had reached its basal levels or (ii) if c mito had been considerably lowered by CCCP administration. One has to keep in mind that CCCP also results in the collapse of the plasma membrane potential (75), which subsequently reduces the plasmalemmal Ca 2ϩ influx (76). Thus, one reason for the CCCP-evoked stop in oscillations might be a disturbed Ca 2ϩ influx. Moreover, the CCCP-mediated drop in ATP production likely leading to an impairment of the ER Ca 2ϩ transport might also contribute to the oscillation arrest (77); i.e. the effects of protonophores are not exclusively attributed to the reduced mitochondrial Ca 2ϩ uptake as was proposed in earlier studies (9). When CCCP was administered before serum, it caused a Ca 2ϩ transient due to the mitochondrial release, which was followed by a period of lower resting c cyt . A lower c cyt is a sign of the reduced Ca 2ϩ influx (resting plasmalemmal Ca 2ϩ leakage). There was a similar decrease in resting c cyt when the extracellular free Ca 2ϩ was chelated by EGTA (data not shown).
The Ca 2ϩ influx across the plasma membrane is important to sustain the Ca 2ϩ oscillations in prMC (78) but not in HeLa cells (9). The different dependence of these cell types on extracellular Ca 2ϩ for the oscillations might be the result of differences in the contribution/importance of the various Ca 2ϩ shuttling pathways between ER and mitochondria on the one hand and between ER and the extracellular space on the other. Our results indicate that plasmalemmal Ca 2ϩ extrusion systems and mitochondrial Ca 2ϩ uptake channels compete for the Ca 2ϩ ions released from the ER. We hypothesize that in some cells, such as prMC and HEK cells (74), the shuttling between the extracellular space and the ER dominates over the shuttling between mitochondria and the ER. However, in HeLa cells and hepatocytes, the ER/mitochondrion shuttling prevails. This might explain why Ca 2ϩ oscillations in some cells are strongly dependent on extracellular Ca 2ϩ ions but not in others.
Another often neglected aspect about "Ca 2ϩ shuttling" pathways is the contribution of cytosolic Ca 2ϩ buffers present at rather high concentrations in the cytosol of some cell types. They are expected to modulate the Ca 2ϩ shuttling among all compartments, extracellular space, ER, and mitochondria, as well as to transiently affect c cyt (Fig. 7). A strong interdependence between cytoplasmic Ca 2ϩ buffers and mitochondria has FIGURE 7. Contribution of Ca 2؉ signaling toolkit components to serum-induced Ca 2؉ oscillations in prMC. A, in unperturbed (control) prMC in vitro, Ca 2ϩ oscillations are primarily the result of the interplay between Ca 2ϩ from the extracellular space and the ER with some minor contributions of mitochondria. The arrows indicate the shuttling of Ca 2ϩ ions between the different compartments (the thicker the arrow, the more important is this pathway). In prMC, expression of calretinin is virtually absent, excluding an important role of this protein with respect to mobile Ca 2ϩ buffering. B, if cells are subjected to La 3ϩ insulation excluding the exchange of Ca 2ϩ ions via the plasma membrane, the repetitive Ca 2ϩ exchange between the ER and mitochondria allows for the generation of Ca 2ϩ oscillations. C, the addition of the mobile Ca 2ϩ buffer calretinin as observed in reactive mesothelial cells and mesothelioma cells affects Ca 2ϩ oscillations. High expression levels (Ͼ1 M in our model) completely block oscillations; lower levels (Ϸ0.5 M) reduce the amplitude of c cyt as well as of c mito during Ca 2ϩ oscillations; i.e. calretinin competes with mitochondria, thus reducing the shuttling of Ca 2ϩ ions between the ER and mitochondria.
been demonstrated before. The expression levels of parvalbumin, a Ca 2ϩ -buffering protein with slow binding kinetics, and the mitochondrial volume in fast twitch muscle cells and in parvalbumin-expressing neurons are inversely regulated (Ref. 79, and for more details, see Ref. 80). In our study, we observed that overexpression of calretinin modifies Ca 2ϩ signals and associated oscillations. It reduces the amount of Ca 2ϩ ions shuttling both between the ER and mitochondria and between the ER and the cytoplasm. Our model predicts that at calretinin concentrations Ͼ1 M Ca 2ϩ oscillations should be blocked in prMC. This is in apparent contradiction with the experimental results where oscillations still existed in EBFP-calretinin-expressing cells likely expressing levels higher than 1 M (Fig. 6). However, in our modeling, the Ca 2ϩ microdomain was not considered, and Ca 2ϩ binding characteristics of calretinin (e.g. K d and k on ) might be different in the cytosol of prMC than the parameters determined in vitro (34). Furthermore, adaptation/ compensation mechanisms might be induced in prMC overexpressing calretinin that would still allow for the generation of Ca 2ϩ oscillations.
Of relevance, calretinin reduced the mitochondrial Ca 2ϩ uptake and Ca 2ϩ accumulation. In human malignant mesothelioma, mostly of the epithelioid type, calretinin is overexpressed (81). This might cause changes, e.g. a delay or blocking of apoptotic/necrotic processes (78,82). Thus, the increased calretinin expression in mesothelioma cells and moreover in certain colon cancer (83) and derived cell lines (84) might be correlated or causally linked to the increased resistance of these tumor cells to the apoptotic/necrotic signals either occurring in healthy physiological conditions or resulting from treatment with chemotherapy drugs such as oxaliplatin or 5-fluorouracil (85). In support, colon cancer cells resistant to aurora kinase inhibitors are characterized by higher calretinin expression levels (86). Moreover, down-regulation of calretinin by lentiviral infection induces apoptosis in mesothelioma cell lines in vitro via an intrinsic mitochondrion-mediated pathway (87). Also down-regulation of calretinin in colon cancer cells is associated with cell growth arrest and increased apoptosis (88).
Author Contributions-L. P. designed the study, performed the experiments with simulations, and wrote the paper. W. B. provided assistance, contributed to lentivirus production and cloning (CALB2), and performed qRT-PCR. B. S. secured funding, analyzed data, and wrote the paper.