NCX3 regulates mitochondrial Ca²⁺ handling through the AKAP121-anchored signaling complex and prevents hypoxia-induced neuronal death

Mitochondria play a pivotal role in neuronal Ca²⁺ signaling by sensing and shaping cytosolic Ca²⁺ transients. In turn, intra-mitochondrial Ca²⁺ also has a central role in neuronal metabolism, regulating the activity of matrix dehydrogenases involved in ATP production (Denton, 2009). By contrast, mitochondrial Ca²⁺ (Ca²⁺_m) overload causes the opening of the permeability transition pore (PTP), thereby influencing mitochondrial control of programmed neuronal death (Denton and McCormack, 1980; Gunter et al., 2000; Kroemer, 1997). Moreover, Ca²⁺_m overload, occurring during ischemia, leads to mitochondrial dysfunction and ultimately neuronal death (Starkov et al., 2004). Thus, the mitochondrial Ca²⁺ influx and efflux pathways might have a relevant role in maintaining cytosolic and mitochondrial Ca²⁺ homeostasis. The existence of a mitochondrial Na⁺/Ca²⁺ exchanger, first postulated in 1974 (Carafoli et al., 1974), has been primarily investigated from a functional point of view. Its identity and localization have been a matter of debate for the past three decades. Recently, among the Ca²⁺-efflux mechanisms, NCLX, a member of the CCX superfamily, has been found on the cristae of the inner mitochondrial membrane (IMM), where it mediates Na⁺- and Li⁺-dependent Ca²⁺ efflux from the mitochondrial matrix (Palty et al., 2010). By contrast, mounting evidence has suggested that the outer mitochondrial membrane (OMM) is not passively permeable to Ca²⁺ fluxes into the cytoplasm, but rather plays a key role in controlling mitochondrial function and Ca²⁺ cycling (Báthori et al., 2006; Jonas et al., 1999; Szabadkai and Duchen, 2008). More specifically, the OMM serves as a permeability barrier not only to Ca²⁺ influx but also to Ca²⁺ efflux (Baines et al., 2007; Crompton et al., 2002; Moran et al., 1992). This evidence suggests that an additional Na⁺-dependent Ca²⁺ extruding mechanism operates between the intermembrane space and the cytoplasmic compartment.

In the present study, we explored the possibility that a member of the NCX family, coded by nuclear DNA, as is the case for the majority of mitochondrial proteins, might be present on mitochondria (Stojanovski et al., 2003). Moreover, we also investigated whether an interaction between NCX and other OMM proteins could constitute a possible mechanism that is involved in the regulation of mitochondrial Ca²⁺ efflux and neuronal survival. We focused particularly on AKAP121 (also known as AKAP1), a member of the protein kinase A (PKA) anchoring protein (AKAP) family that controls mitochondrial metabolism and neuronal survival (Carlucci et al., 2008b) by interacting with several mitochondrial proteins and biochemical signaling pathways. Briefly, AKAP121, as well as its human homolog AKAP149, besides having an RNA-binding motif (KH domain) that binds several mRNAs of nuclear-encoded mitochondrial proteins, also assembles an mRNA and PKA complex on mitochondria. This multicomponent system, which represents an important signal crossroad regulating translation and mitochondrial protein import, also encompasses cAMP phosphodiesterase (PDE4A), Ser/Thr phosphatase (PP1), and an Src-associated tyrosine phosphatase (PTPD1).

Here, we demonstrate for the first time that the nuclear encoded NCX3 isoform (1) is located on the outer mitochondrial membrane of neurons; (2) colocalizes and immunoprecipitates with AKAP121; (3) extrudes Ca²⁺ from mitochondria through AKAP121 interaction in a PKA-mediated manner, both under normoxia and hypoxia; and (4) improves cell survival when it works in the Ca²⁺-efflux mode at the level of the outer mitochondrial membrane.

We found that the NCX3 protein, apart from being present at the plasma membrane (PM), as previously described (Secondo et al., 2007), was also localized on the mitochondrial membranes. By contrast, NCX1 and NCX2 isoforms were not detected (Fig. 1A). Experiments performed with whole mitochondria or with mitochondrial fractions of baby hamster kidney (BHK) NCX3-transfected cells revealed that NCX3 was located on the OMM (Fig. 1B,C). Indeed, when the OMM was removed with trypsin (20 µg/ml) (Sardanelli et al., 2006), NCX3 immunoreactivity disappeared. Importantly, immunoreactivity for voltage-dependent anion channel protein 1 (VDAC) and AKAP121, two proteins specific to the outer mitochondrial membrane, were undetectable after treatment of fractionated mitochondria with trypsin. Conversely, Mn-SOD immunoreactivity, a marker of the mitochondrial matrix (MM), and COX-IV, a marker of the IMM, remained unaffected (Fig. 1B). Similar results were obtained by treating mitochondria with digitonin (500 µM) (Fig. 1B). Interestingly, confocal microscopy showed that some NCX3 immunoreactivity in the cell overlapped with Mito-Tracker- or Mito-RFP-stained mitochondria, along neurites of cortical neurons (Fig. 2A, panels e, f and g). Furthermore, double immunofluorescence experiments with NCX3 and Mito-RFP indicated that, whereas Mito-RFP stained the entire mitochondria, the NCX3 signal localized in punctuate spots, which mostly aligned along mitochondria edges (arrows, Fig. 2A, panel f). This punctuate distribution pattern, which we observed with both the anti-NCX3 and the anti-FLAG antibody in cells transfected with FLAG-tagged NCX3 (NCX3^F), is similar to that described for other proteins located on the mitochondria (Margineantu et al., 2002). The calculated percentage of NCX3 immunoreactivity in mitochondria corresponded to 25% of the total NCX3 signal (Fig. 2A, panel h). Immunoelectron microscopy, performed both in NCX3-transfected BHK cells (Fig. 1C) and in neurons (Fig. 2B) demonstrated that NCX3 labeling was detected on the PM, endoplasmic reticulum (ER) and mitochondria (Fig. 1C), whereas it was undetectable in other compartments such as Golgi stacks and endosomes (Table 1). These findings were supported by immunocytochemistry experiments performed in NCX3^F-transfected BHK cells (supplementary material Fig. S1). In particular, NCX3 labeling on mitochondria was mainly on the OMM (77.6% in BHK cells and 90.7% in neurons, Fig. 1C). Fig. 1D shows that the mitochondrial fraction was not contaminated by other membranes, as immunoblotting detected the absence of calnexin, LAMP1 and GM130, markers of the ER, lysosomes, and Golgi compartments, respectively, as well as IP₃R3, a specific marker of the mitochondrial-associated membrane (MAM) (Hood et al., 2003; Liu et al., 2012; Mendes et al., 2005; Szabadkai et al., 2006). Consistently, mitochondria obtained from whole brain extracts of Ncx3^+/+ (Slc8a3^+/+) mice displayed an obvious NCX3 immunoreactivity (Fig. 2C). Conversely, mitochondria obtained from Ncx3^−/− (Slc8a3^–/–) mice failed to reveal any NCX3 immunoreactivity (Fig. 2B).

As we had demonstrated that NCX3 was localized on the OMM, we characterized its biochemical properties. In particular, we hypothesized that AKAP121, exclusively localized on the mitochondria, might play a role in the regulation of mitochondrial NCX3 (mNCX3) activity. To test this hypothesis, wild-type BKH cells (BHK-Wt), which endogenously express AKAP121, were transiently co-transfected with NCX3^F and AKAP121. In these cells, immunofluorescence experiments revealed that the NCX3^F signal was associated with a large number of mitochondria (Fig. 3A). Immunoprecipitation with anti-FLAG and anti-AKAP121 antibodies demonstrated that NCX3 and AKAP121 form a stable complex in transiently NCX3^F+AKAP121 co-transfected BHK cells (Fig. 3B). In accordance, confocal experiments using anti-FLAG and anti-AKAP121 antibodies revealed that NCX3^F immunosignal was colocalized with AKAP121 (Fig. 3C, arrowheads).

To demonstrate that the molecular interaction between NCX3 and AKAP121 might be responsible for Ca²⁺_m handling, we coexpressed NCX3^F and cytosolic (cytoAEQ) or mitochondria-targeted (mtAEQ) aequorin-based Ca²⁺ probes in BHK cells, endogenously expressing AKAP121, and evaluated global and organellar Ca²⁺ responses to agonist stimulation. After reconstitution with aequorin cofactor coelenterazine, cells were challenged with histamine (100 µM), and luminescence was measured and converted into a value for [Ca²⁺]. As reported in Fig. 4A, the mitochondrial Ca²⁺ concentration ([Ca²⁺]_m) was lower in cells transfected with NCX3^F than that measured in BHK-Wt cells. Interestingly, when the cells were transiently co-transfected with NCX3^F and siRNA against AKAP121 (siAKAP121), [Ca²⁺]_m was higher compared to cells transiently co-transfected with NCX3^F (Fig. 4A,B). To exclude the possibility that AKAP121 might affect Ca²⁺_m through its ability to modify mitochondrial membrane potential and not NCX3 activity, further experiments measuring mitochondrial membrane potential were performed in BHK-Wt, and in NCX3^F, siAKAP121, and NCX3^F+siAKAP121 transiently transfected cells. The results obtained demonstrated that in BHK cells transfected with siAKAP121, mitochondrial membrane potential was lower compared to BHK cells transfected with NCX3^F or NCX3^F+siAKAP121 (Fig. 4A, inset), thus suggesting that the effect of AKAP121 on [Ca²⁺]_m is independent of changes in mitochondrial membrane potential but depends on its ability to modulate NCX3 activity. Indeed, when NCX3^F and siAKAP121 are co-transfected, the mitochondrial membrane is hyperpolarized and [Ca²⁺]_m is higher than when NCX3^F is transfected alone (Fig. 4A and inset). Similar results were obtained when the catalytic subunit of PKA was inhibited by the PKI construct, which contains a PKA pseudophosphorylation site (data not shown). The specificity of siRNA for AKAP121 is reported in Fig. 4C. To demonstrate that [Ca²⁺]_m is not indirectly affected by the changes in cytosolic Ca²⁺ level due to variation in PM NCX3 expression, we measured cytosolic Ca²⁺ concentrations in each of the above-mentioned experimental conditions and we did not detect any differences (Fig. 4D). The capacity of NCX3 to enhance mitochondrial Ca²⁺ efflux independently of PM NCX3 was also confirmed in digitonin-permeabilized BHK cells transiently transfected with NCX3^F and mtAEQ (Fig. 4F). Finally, to exclude any possible effect of NCX3 expression in the ER, the same experiments were performed in the presence of thapsigargin. These experiments demonstrated that in cells pretreated with thapsigargin for 15 minutes, the effect of histamine on [Ca²⁺]_m was the same as that observed in the absence of thapsigargin, thus suggesting that the inhibition of SERCA does not affect [Ca²⁺]_m (Fig. 4G). This evidence suggests that the activation of non-selective plasma membrane cationic channels elicited by histamine increases the intracellular Ca²⁺ concentration ([Ca²⁺]_i) in these cells, as previously reported (Zanner et al., 2002).

To demonstrate the role of endogenous NCX3 in the modulation of Ca²⁺_m extrusion, cortical neurons treated with siRNA against NCX3 (siNCX3) and siAKAP121 were exposed to thapsigargin (Tg, 1 µM, 100 seconds) to allow ER Ca²⁺ depletion (Secondo et al., 2000) and consequently [Ca²⁺]_m loading (Baumgartner et al., 2009). Cells were then treated with the mitochondrial uncoupler FCCP (300 nM for 1 minute) to induce mitochondrial depolarization and Ca²⁺ extrusion (Medler and Gleason, 2002). After thapsigargin treatment, the Ca²⁺ release into cytoplasm upon FCCP exposure was measured as [Ca²⁺]_i increase. In particular, when FCCP-induced Ca²⁺ release into the cytoplasm is low, a higher activity of the Ca²⁺_m efflux pathway is occurring and vice versa. As reported in Fig. 5A–C, siNCX3 or siAKAP121 induced an enhancement of [Ca²⁺]_m compared to untreated neurons. Interestingly, [Ca²⁺]_i levels were significantly affected only by siNCX3, thus demonstrating the specificity of siAKAP121 in the modulation of mNCX3 activity (Fig. 5D).

The functional role of endogenous mNCX3 was supported by quantitative colocalization analysis of NCX3 with MitoTracker and Mito-RFP in cortical neurons exposed to oxygen and glucose deprivation (OGD, 3 hours) followed by reoxygenation (Rx, 24 hours). As shown in Fig. 6A,B, double-labeling experiments with anti-NCX3 antibody in MitoTracker-stained and in Mito-RFP-infected neurons revealed that NCX3 mitochondrial immunosignal decreased during OGD and returned to the basal level after Rx. Western blotting on isolated mitochondria supported this finding (Fig. 6C). In accordance with these results, [Ca²⁺]_m significantly increased when neurons were exposed to OGD, whereas it decreased following OGD-Rx (Fig. 6D). Interestingly, when NCX3 was knocked down with siRNA, an impairment in Ca²⁺_m extrusion was recorded under both basal and OGD-Rx conditions, whereas no alteration occurred in Ca²⁺_m extrusion during OGD, a condition in which the NCX3 mitochondrial immunosignal decreased.

The [Ca²⁺]_m increase elicited by chemical hypoxia in BHK-Wt cells was reduced when these cells were transfected with NCX3^F, thus showing that mNCX3 works as a Ca²⁺ efflux pathway (Fig. 6E). Moreover, the silencing of endogenous AKAP121 in NCX3^F-transfected BHK cells reduced the Ca²⁺ efflux activity of the mNCX, as demonstrated by the [Ca²⁺]_m increase caused by chemical hypoxia (Fig. 6E). In agreement with these results, cell survival following chemical hypoxia was higher in BHK cells transfected with NCX3^F and NCX3^F plus AKAP121 (Fig. 6F). Moreover, the silencing of constitutively expressed AKAP121 completely reverted the prosurvival effect exerted by NCX3^F (Fig. 6F). By contrast, the silencing of endogenous AKAP121 in wild-type BHK cells not expressing NCX3 (third column of Fig. 6F) had no protective action.

This study provides evidence that the nuclear-encoded NCX3 is the only Na⁺/Ca²⁺ exchanger isoform localized within the OMM of neurons, where it plays a relevant role in controlling Ca²⁺_m homeostasis under both basal and hypoxic conditions. Further results also revealed that mNCX3 colocalizes with AKAP121 – a member of PKA anchoring protein expressed on the OMM – and that this interaction modulates mNCX3 activity. This finding led to the hypothesis that the interplay between mNCX3 and AKAP121 regulates [Ca²⁺]_m and contributes to cell survival. Indeed, when the constitutively expressed AKAP121 was silenced, [Ca²⁺]_m increases and the prosurvival effect exerted by these two proteins was reverted.

An interesting finding emerging from our studies is that, among the three nuclear-encoded Na⁺/Ca²⁺ exchanger isoforms, only NCX3 was localized on mitochondria. In particular, mNCX3 was specifically localized on the OMM, as its lysis completely eliminated mNCX3 immunoreactivity. In this regard, an even more compelling result is that ablation of the gene encoding NCX3 (Slc8a3) induced not only the disappearance of the protein from the OMM but also the accumulation of Ca²⁺_m in cortical neurons. Immunocytochemistry and electron microscopy experiments performed in both neurons and NCX3^F-transfected BHK cells further support the hypothesis that, among the subcellular compartments, NCX3 is localized mainly to mitochondrial. Moreover, in neurons, the NCX3 mitochondrial localization was also particularly evident along the neurites and in the neuropils close to the PM, where ATP is necessary to drive the activity of those proteins involved in the regulation of ionic homeostasis (Blaustein et al., 2002; Guerini et al., 2005; Lytton, 2007).

Although a recent paper reports that all three NCX isoforms, NCX1, NCX2 and NCX3, can be detected on mitochondria in both neurons and glial cells (Gobbi et al., 2007), in this study, NCX2 expression was detected by an anti-NCX2 antibody, which, in fact, cross-reacts strongly and not specifically with an unidentified glial protein different from NCX2 (Thurneysen et al., 2002). Additionally, the molecular mass of the NCX2 protein detected with this anti-NCX2 antibody differs from those reported in the literature (Boscia et al., 2009; Secondo et al., 2007; Sirabella et al., 2009). Regarding the presence of NCX1 protein on mitochondria reported by Gobbi et al. (Gobbi et al., 2007), it should be underlined that in their study the exchanger isoform was detectable at very low levels, most likely because detection occurred only after immunoprecipitation. More convincingly, our data demonstrated that even when NCX1 and NCX2 were overexpressed in BHK cells, no immunoreactivity was detected for either one of these proteins in the whole mitochondrial fraction. In a very recent study, another component of the NCX family, NCLX, has been identified in neurons at the level of the IMM, particularly within the cristae (Palty et al., 2010). Moreover, this study revealed that this Li⁺-sensitive protein, which is both phylogenetically and functionally distinct from NCX and NCKX family members also participates in the mitochondrial Na⁺/Ca²⁺ exchange activity in neuronal cells. This novel finding fully agrees with our own results, in that we reasoned that the Na⁺/Ca²⁺ exchange in mitochondria requires two consecutive steps. The first, operated by the Na⁺-sensitive NCLX, mediates Ca²⁺ transport from the matrix to the intermembrane space, and the second, operated by mNCX3, promotes Ca²⁺ efflux from the intermembrane space to the cytosol. This interpretation is in line with the recent physiological role attributed to the OMM in the control of Ca²⁺_m cycling. Indeed, the outer surface of the membrane is not a passive permeable membrane as previously considered (Babcock et al., 1997; Gunter and Pfeiffer, 1990), but it does constitute a permeability barrier not only to Ca²⁺ influx but also to Ca²⁺ efflux (Baines et al., 2007; Jonas et al., 1999; Moran et al., 1992; Szabadkai and Duchen, 2008). In addition, it has been recently demonstrated that VDAC, localized on OMM and whose opening depends on OMM potential and cytosolic Ca²⁺ levels, promotes Ca²⁺ influx from the cytosol to the intermembrane space (Báthori et al., 2006). In the light of these findings, it is conceivable that VDAC, together with NCLX, could generate a driving force for NCX3 to extrude Ca²⁺ from the intermembrane space to the cytosol by increasing [Ca²⁺] in the interspace compartment.

Another novel finding emerging in this paper is the physical, biochemical and functional interaction between mNCX3 and the PKA anchoring protein AKAP121 in cortical neurons. We previously demonstrated that AKAP121 regulates the activity of the components belonging to the mitochondrial respiratory chain, thus promoting ΔΨ_m hyperpolarization and improving the oxidative synthesis of ATP in a PKA-dependent manner (Livigni et al., 2006). Similarly, in the present study, immunoprecipitation assays and confocal microscopy experiments revealed that mNCX3 interacts with AKAP121, thus suggesting the involvement of the anchoring protein in promoting mNCX3 Ca²⁺ extrusion activity under basal and hypoxic conditions. This finding suggests that the endogenous AKAP121 plays a role in the regulation of mNCX3 efflux activity, most likely mediated by its ability to anchor PKA on mitochondria, given that the silencing of AKAP121 or the inhibition of the kinase by the co-transfection of a specific PKI construct, which contains a PKA pseudophosphorylation site, abolished the effects of AKAP121 on [Ca²⁺]_m.

The physiological importance of this interaction is even more evident given its role in preventing hypoxic neuronal death. Indeed, the silencing of AKAP121 significantly reduced mNCX3 Ca²⁺-efflux activity and, consequently, increased cell death during hypoxia. Accordingly, we reasoned that the activation of mNCX3 by AKAP121-anchored PKA on the OMM regulates Ca²⁺_m handling, thus boosting mitochondrial metabolism and, in the process, cellular survival. These results reveal a close link with our previous data demonstrating that, in BHK cells, NCX3 significantly contributes to the maintenance of [Ca²⁺]_i homeostasis during experimental conditions mimicking ischemia, thereby preventing ΔΨ_m collapse and cell death (Secondo et al., 2007). Moreover, the results obtained in neurons exposed to OGD and OGD-Rx demonstrated that the endogenous mNCX3 also plays a relevant role in the regulation of Ca²⁺_m extrusion. Indeed, when neurons were exposed to OGD, [Ca²⁺]_m significantly increased while NCX3 expression was reduced. By contrast, when neurons were exposed to OGD-Rx, [Ca²⁺]_m decreased, whereas NCX3 expression returned to the basal values. Interestingly, when NCX3 was knocked down, an impairment in Ca²⁺_m extrusion was recorded under both basal and OGD-Rx conditions, whereas no alteration in Ca²⁺_m extrusion occurred during OGD. These results might be related to changes in the expression of endogenous NCX3 and AKAP121 during OGD and OGD-Rx (Carlucci et al., 2008a; Sirabella et al., 2009). Indeed, we have already shown in a previous study that during hypoxia NCX3 and AKAP121 undergo a rapid degradation through the ubiquitin-proteosome pathway in neuronal culture. In particular, for AKAP121, the responsible ubiquitin ligase is SIAH2, a hypoxia-induced protein. The resultant decrease in AKAP121 significantly reduces mitochondrial activity (Carlucci et al., 2008a). These experiments strongly suggest that the mitochondrial-localized fraction of NCX3 is responsible for the cellular events observed. However, it should be underlined that NCX3 expression in mitochondria of neurons exposed to OGD-Rx is higher compared to that observed in neurons in basal conditions. Therefore, it is possible to speculate that NCX3 might translocate to mitochondria as needed, thus contributing to improved mitochondrial function during OGD-Rx.

The identification of NCX3 isoform as a molecular target of PKA on the OMM of neurons represents the first evidence for a functional relationship between AKAP121 and those mitochondrial proteins able to regulate Ca²⁺_m efflux. This finding might have important physiological and pathophysiological implications. Specifically, a considerable crosstalk between the bioenergetic function and Ca²⁺ homeostasis occurs within the mitochondria, for Ca²⁺ is necessary to activate mitochondrial oxidative metabolism and to promote mitochondrial respiration (Denton, 2009; Denton and McCormack, 1980; McCormack et al., 1990). However, if [Ca²⁺]_m increases over its buffering capacity, ATP production will decrease, causing the mitochondrial PTP to open, and, eventually, cells to die by apoptosis (Jeong and Seol, 2008; Kim et al., 2003; Krieger and Duchen, 2002). Recently, Ghandi et al. have demonstrated that PINK1, a mitochondrial kinase whose gene mutation is involved in Parkinson's disease, participates in the regulation of an unidentified mitochondrial Na⁺/Ca²⁺ exchanger. This hypothesis, based on the observation that mitochondria from neurons with PINK1 knocked-out display Ca²⁺ accumulation due to the impairment of Ca²⁺ efflux mechanisms, suggests that the accumulation of Ca²⁺_m is indeed responsible for increased neuronal vulnerability (Gandhi et al., 2009).

In summary, we have identified NCX3 as the putative mNCX isoform and demonstrated that mNCX3, by interacting with AKAP121 on the outer mitochondrial membrane of neurons, might participate in the control of Ca²⁺_m efflux and cell survival in a PKA-sensitive manner.

Wild-type and BHK cells stably transfected with canine cardiac NCX1, or rat brain NCX2 or NCX3 (Linck et al., 1998) were grown on plastic dishes in a mix of DMEM and Ham's F12 media (1∶1) (Invitrogen) supplemented with 5% FBS, 100 U/ml penicillin and 100 µg/ml streptomycin (Sigma). For confocal and Ca²⁺ imaging experiments, cells were plated on glass coverslips (Fisher) coated with poly-D-lysine (100 µg/ml) (Sigma), and used at least 24 hours after seeding.

Mixed cultures of cortical neurons from 2–4-day-old Wistar rat pups (Charles River), were prepared as previously described (Abramov et al., 2007). Primary cultures from Ncx3^−/− (Slc8a3^–/–) mice were obtained as previously reported (Molinaro et al., 2008; Sokolow et al., 2004). The experiments on primary cortical neurons were performed according to the procedures described in experimental protocols approved by Ethical Committee of the ‘Federico II’ University of Naples.

For immunocytochemistry, confocal and Ca²⁺ imaging experiments, cells were plated on glass coverslips coated with poly-D-lysine (100 µg/ml).

Mouse pCEP4-AKAP121 cDNA was a gift from Dr C. Rubin (Albert Einstein College of Medicine, NY). Vectors encoding the CMV promoter and AKAP121 protein are as previously described (Affaitati et al., 2003). OmicLink Expression M14 vector expressing NCX3^F was engineered by ligating an oligonucleotide encoding a 3XFLAG epitope to the C-terminus of mouse NCX3 (Slc8a3) cDNA (GeneCopeia). No deletion mutants of NCX3 were used. To knock-down NCX3, the 18 nucleotides sequence spanning from +124 to +142, referring to transcriptional start site (+1), of rat NCX3 cDNA (GenBank accession no. U53420) was inserted in the mammalian expression vector pSUPER.retro.puro (OligoEngine). All these constructs were transiently transfected using Lipofectamine 2000 (Invitrogen). The siRNAs were transiently transfected using Lipofectamine 2000 (Invitrogen) at a final concentration of 250 pmol/ml of culture medium.

BHK-Wt and NCX3-transfected cells were washed in phosphate-buffered saline medium, and trypsinized with 1 ml of trypsin solution per 10-cm plate. The trypsinization was stopped by adding 6 ml of growth medium. As reported in supplementary material Fig. S2A, the cell suspension was collected in a 15-ml falcon tube and pelleted by centrifugation at 112 g for 3 minutes. The pellet was resuspended in 300 µl of homogenization buffer solution (Buffer A) containing the following (mM): 250 mannitol, 0.5 EGTA, 5 HEPES (pH 7.4), 1.5 MgCl₂, 0.1% aprotinin, 0.7 mg/ml pepstatin, and 1 µg/ml leupeptin, and gently disrupted by passing ten times through a 26-gauge needle. The suspension was centrifuged at 500 g for 5 minutes at 4°C. After the first centrifugation, pellets, corresponding to the fraction containing membranes but not intracellular organelles including mitochondria, were separated from supernatants and measured for protein concentrations. The supernatant, was centrifuged at 500 g for 5 minutes at 4°C. After this centrifugation the pellet was discarded, and the supernatant corresponding to the cytosolic fraction containing the organelles, was further centrifuged at 19,000 g for 10 minutes at 4°C in order to separate the mitochondrial from the cytosolic fraction. Supernatants (Cytosol) were then removed and assessed for protein content. Next, the pellets containing mitochondria were lyzed in 50 µl of lysis buffer containing (mM): 20 Tris-HCl (pH 7.5), 10 NaF, 150 NaCl, 1 PMSF, 1% Nonidet P-40, 1 Na₃VO₄, 0.1% aprotinin, 0.7 mg/ml pepstatin, and 1 µl/mg leupeptin, and kept on ice for 15 minutes. Finally, samples were purified again by centrifugation (18,000 g, 10 minutes) and supernatants (Mitochondria) were assessed for protein content by Bradford's assay (Bradford, 1976). The three fractions obtained were used for western blotting (WB). All the procedure was carried out at 4°C to minimize the action of proteases and phospholipases.

The purity of the mitochondrial preparation was assessed by evaluating the expression of the proteins GM131, a Golgi marker, calnexin, an ER marker, LAMP1, a lysosome marker, (Hood et al., 2003; Liu et al., 2012; Szabadkai et al., 2006), and the IP3R3 receptor, a MAM marker (Mendes et al., 2005) on the fractions relative to membranes, cytosol and mitochondria.

The Percoll gradient method was used to purify mitochondria from tissues (Hovius et al., 1990). Ncx3^+/+ and Ncx3^−/− mice were killed by decapitation according to the experimental procedures approved by the Ethical Committee of the ‘Federico II’ University of Naples and the brains were rapidly removed. Each brain was washed twice in cold PBS in order to remove the blood and was cut into small pieces using scissors. The brain pieces were transferred into a 50-ml glass/teflon Potter Elvehjem homogenizer and the homogenization buffer solution (Buffer A) was added in the ratio of 5 ml of buffer per gram of tissue. The homogenization was performed by using a teflon pestle with a slow movement to preserve the integrity of mitochondria. The homogenate was passed through a 22-gauge needle once and then through a 26-gauge needle five times to accomplish the homogenization of the brain. Subsequently, it was centrifuged four times at 500 g for 5 minutes and once again at 18,000 g for 10 min. The supernatant was removed and assayed for protein content. As depicted in supplementary material Fig. S2B, the pellet containing crude mitochondria was suspended in 1 ml buffer A and then stratified in 3 ml of Percoll medium containing the following (mM): 30% (v/v) Percoll, 250 mannitol, 0.5 EGTA, 5 HEPES (pH 7.4), and centrifuged at 95,000 g for 30 min at 4°C in a Beckman 60Ti rotor. The fraction containing pure mitochondria, lying under the Percoll band (supplementary material Fig. 2B, dark grey), was removed, suspended in buffer A, and then centrifuged twice (7000 g for 10 minutes). The pellet obtained was re-suspended in a 50 µl lysis buffer, kept on ice for 10 minutes, and centrifuged (18,000 g for 10 minutes). The supernatant was finally assessed for protein content and used for western blot analysis. All procedures were carried out at 4°C to minimize the action of proteases and phospholipases.

50 µg of protein samples were analyzed on 8% SDS-PAGE and electrotransferred to Hybond ECL nitrocellulose paper (Amersham). Membranes were first blocked with 5% non-fat dry milk in 0.1% Tween-20 (TBS-T; 2 mM Tris–HCl, 50 mM NaCl, pH 7.5) for 2 hours at room temperature. They were then incubated overnight at 4°C in the blocked buffer with the 1∶1000 anti-NCX1 (rabbit polyclonal antibody, Swant), 1∶1000 anti-NCX2 (rabbit polyclonal antibody, Alpha Diagnostic), and 1∶5000 anti-NCX3 rabbit polyclonal antibody (kindly provided by Kenneth D. Philipson, UCLA, Los Angeles, California, USA), and subsequently incubated with the secondary antibodies for 1 hour (1∶5000; Amersham). Immunoreactive bands were detected with the ECL kit (Amersham). Discrimination among the distinct types of extracts was ensured by running parallel western blots with the endogenous protein tubulin (localized to the PM), Mn-SOD (localized to the MM), VDAC and AKAP121 (localized to the OMM) (Cardone et al., 2004; Ginsberg et al., 2003; Liu et al., 2012), and COX-IV (localized to the IMM) (Mattei et al., 2011).

BHK wild-type cells, BHK cells stably transfected with the NCX3 isoform and cortical neurons were cultured on glass coverslips [BHK cells for 48 hours and cortical neurons for 11 days in vitro (DIV)]. The cells were rinsed twice in cold 0.01M PBS at pH 7.4 and fixed at room temperature in 4% (w/v) paraformaldehyde for 20 minutes. Following three washes in PBS, cells were blocked in PBS containing 10% FBS and the following antibodies: anti-NCX3 (rabbit, kindly supplied by Kenneth D., Philipson, UCLA, Los Angeles, California, USA, dilution 1∶4000), anti-FLAG (mouse, Sigma, dilution 1∶500), and anti-AKAP121 (rabbit, Carlucci et al., 2008a), anti-LAMP1 (kindly supplied by Polishchuk EV, dilution 1∶200), anti-VAP-A (kindly supplied by Maria Antonietta De Matteis, Telethon Institute of Genetics and Medicine, Naples, Italy, dilution 1∶300). They were then incubated overnight at 4°C. Next, slides were washed in PBS, incubated with anti-rabbit Cy2-conjugated antibody (Jackson, dilution 1∶200) and anti-mouse Cy3-conjugated antibody (Jackson, dilution 1∶200) for 1 hour at room temperature under dark conditions, and washed again with PBS. Finally, they were mounted with a SlowFade™ Antifade Kit (Molecular Probes-Invitrogen) and analyzed by confocal microscopy (Boscia et al., 2012). The cells were infected with CellLight Reagents BacMam 2.0 Mito-RFP (Molecular probes) according to the manufacturer's protocol to stain mitochondria.

Cells were analyzed for colocalization between Mito (red) and NCX3 (green) by using the ‘co-localization highlighter’ plug-in for ImageJ Software (NIH, Bethesda, MA, USA). Before colocalization analysis, threshold settings for each image were determined, and quantification was achieved by counting the number of NCX3 and Mito colocalized points (white) per microscope field. Results were expressed as a percentage of colocalization (Molinaro et al., 2011).

Ncx3^+/+, Ncx3^−/− cortical neurons, and wild-type and transfected BHK cells were first fixed with a mixture of 4% paraformaldehyde and 0.05% glutaraldehyde and then labeled with a polyclonal antibody against NCX3 using the Nanogold protocol. Next, they were embedded in Epon-812 and finally cut in thin sections (Polishchuk et al., 2003). Electron microscopy images were acquired from the thin sections using a JEOL JEM-1011 electron microscope (Tokyo, Japan) equipped with a Morada CCD digital camera.

Surface gold density (arbitrary units, AU) was estimated according to the method of Griffiths and Hoppeler (Griffiths and Hoppeler, 1986). Following this method, a morphometric grid with definitive size (100 nm) was applied to all images acquired and then the number of gold particles present on the membranes of interest (mitochondria, ER, PM, Golgi stacks and endosomes) was counted. Next, the number of intersections between the organelle membranes and the grid was calculated. In the case of mitochondria, only the outer membrane was counted. Finally, the number of gold particles was divided by the number of intersections to give the gold density value. At total of 30 structures were analyzed for each compartment of interest. Surface gold density for NCX3 in mitochondria corresponded to 0.19 AU in neurons and to 0.21 AU in BHK NCX3-transfected cells; surface gold density for NCX3 in ER corresponded to 0.24 AU in neurons and to 0.32 AU in BHK NCX3-transfected cells; the plasma membrane surface gold density for NCX3 corresponded to 0.23 AU in neurons and to 0.98 AU in BHK NCX3-transfected cells (Fig. 1C; Fig. 2B; Table 1).

Transfected cytoAEQ and mtAEQ were reconstituted by incubating BHK cells for 2 hours with a 5 µM concentration of the wild-type aequorin prosthetic group coelenterazine (Molecular Probes) in KRB (Krebs–Ringer modified buffer: 125 mM NaCl, 5 mM KCl, 1 mM Na₃PO₄, 1 mM MgSO₄, 5.5 mM glucose and 20 mM Hepes, pH 7.4) supplemented with 1 mM CaCl₂ at 37°C. During the experiment, the cells were continuously perfused with KRB medium. Agonist was added to the same medium. In our experiments, we induced a rise in [Ca²⁺] by stimulating the cells with histamine 100 µM. All aequorin measurements were terminated by lyzing the cells with 100 µM digitonin in a hypotonic Ca²⁺-rich solution (10 mM CaCl₂ in H₂O) thus discharging the remaining AEQ pool. The aequorin luminescence data were calibrated off-line into [Ca²⁺] values using a computer algorithm based on the Ca²⁺ response curve of wild-type aequorin as previously reported (Brini et al., 1995; Rizzuto et al., 1995). In the experiments with permeabilized cells, a buffer mimicking the cytosolic ionic composition [intracellular buffer (IB)] was employed: 140 mM KCl, 10 mM NaCl, 1 mM K₃PO₄, 5.5 mM glucose, 2 mM MgSO₄, 1 mM ATP, 2 mM Na⁺ succinate, and 20 mM Hepes (pH 7.5 at 37°C) supplemented with EGTA 1 mM. BHK cells were permeabilized by a 5-minute incubation with 20 µM digitonin (added to IB with EGTA) before luminescence measurements (Rapizzi et al., 2002). Changes in [Ca²⁺]_m and [Ca²⁺]_cyto were followed as previously reported (Staiano et al., 2009) and the area under the curve (AUC) was calculated using the trapezoidal rule (Bailer and Piegorsch, 1990). The rate of mitochondrial Ca²⁺ release was measured as the time of recovery from the histamine-induced Ca²⁺_m peak to the basal level. These data were reported as the ratio between the time of recovery in each group (NCX3^F-transfected BHK cells and BHK cells transfected with NCX3^F+siAKAP) and that measured in BHK-Wt cells.

Mitochondrial membrane potential was assessed using the fluorescent dye tetramethyl rhodamine ethyl ester (TMRE) in the ‘redistribution mode’. Cells were loaded with TMRE (20 nM) for 30 minutes in a medium containing 156 mM NaCl, 3 mM KCl, 2 mM MgSO₄, 1.25 mM KH₂PO₄, 2 mM CaCl₂, 10 mM glucose and 10 mM Hepes. The pH value was adjusted to 7.35 with NaOH. At the end of the incubation period, cells were washed in the same medium containing TMRE (20 nM) and allowed to equilibrate. A decline in the mitochondrial fluorescence intensity was indicative of mitochondrial membrane depolarization (Livigni et al., 2006).

Confocal images were obtained using a Zeiss inverted 510 confocal laser scanning microscopy and a 63× oil immersion objective. The illumination intensity of the 543 Xenon laser, used to excite TMRE fluorescence, was kept to a minimum of 0.5% of laser output to avoid phototoxicity.

[Ca²⁺]_i was measured by single-cell computer-assisted videoimaging as previously described (Secondo et al., 2007). Briefly, cells, grown on glass coverslips, were loaded with 10 µM Fura-2 acetoxymethyl ester (AM) (Fura-2AM) for 30 minutes at room temperature in normal Krebs solution containing (in mM) 5.5 KCl, 160 NaCl, 1.2 MgCl₂, 1.5 CaCl₂, 10 glucose and 10 Hepes–NaOH, pH 7.4. At the end of the Fura-2-AM loading period, the coverslips were placed into a perfusion chamber (Medical System, Co. Greenvale, NY, USA) mounted onto a Zeiss Axiovert 200 microscope (Carl Zeiss, Germany). The experiments were carried out with a digital imaging system composed of a MicroMax 512BFT cooled CCD camera (Princeton Instruments, Trenton, NJ, USA), LAMBDA 10-2 filter wheeler (Sutter Instruments, Novato, CA, USA), and Meta-Morph/MetaFluor Imaging System software (Universal Imaging,West Chester, PA, USA). All the results are presented as cytosolic Ca²⁺ concentration, assuming that the K_d for FURA-2 was 224 nM. The amount of Ca²⁺ extruded in the cytoplasm upon FCCP exposure was measured as [Ca²⁺]_i elevations. This method is widely used to measure [Ca²⁺]_m content (Medler and Gleason, 2002). The rate of [Ca²⁺]_i increase after Tg+FCCP exposure was calculated as the peak of [Ca²⁺]_i against time, averaged over three measurements, and expressed as mean±s.e.m.

Cells were homogenized in lysis buffer containing (mM) the following: 50 Tris-HCl pH 7.4, 150 NaCl, 1 EDTA, 1% Triton X-100, 100 NaF, 100 Na₃VO₄, 5 µg/ml aprotinin, 10 µg/ml leupeptin and 2 µg/ml pepstatin. The lysates were cleared by centrifugation (15,000 g, 15 min). Two milligrams of cell lysate was immunoprecipitated with either anti-FLAG mouse antibody (1∶100), anti AKAP121 (1∶200) or non immune IgG antibody. One hundred micrograms either of total cell lysate or of immunoprecipitates was resolved by SDS-PAGE gel and transferred to nitrocellulose membrane. Immunoblot analysis was performed using anti AKAP121 or anti FLAG (Cardone et al., 2004; Carlucci et al., 2008a).

Chemical hypoxia was reproduced by adding 5 µg/ml oligomycin plus 2 mM 2-deoxyglucose to cells in glucose-free medium for 45 minutes. The medium was composed of (in mM): 145 NaCl, 5.5 KCl, 1.2 MgCl₂, 1.5 CaCl₂, and 10 Hepes, pH 7.4, as already described (Secondo et al., 2007). Control cells were exposed for the same amount of time to normal Krebs medium composed of (in mM): 145 NaCl, 5.5 KCl, 1.2 MgCl₂, 1.5 CaCl₂, 10 glucose and 10 Hepes, pH 7.4. Reoxygenation was performed in 1× MEM plus 25 mM NaHCO₃ and 22 mM dextrose at 37°C.

In cortical neurons, OGD insult was reproduced by exposing cells for 3 hours to a medium containing (in mM): 116 NaCl, 5.4 KCl, 0.8 MgSO₄, 26.2 NaHCO₃, 1 NaH₂PO₄, 1.8 CaCl₂, 0.01 glycine and 0.001 w/v Phenol Red (Carlucci et al., 2008a; Sirabella et al., 2009). Hypoxic conditions were maintained using a hypoxia chamber (Billups Rothemberg Inc. Del Mar.) (temperature 37°C, under an atmosphere of 5% CO₂ and 95% N₂). Reoxygenation was obtained by incubating the cells for 24 hours in the presence of normal levels of glucose and oxygen.

To quantify cell death after the experimental procedures, cells were washed with normal Krebs and stained with 36 µM fluorescein diacetate (FDA) and 7 µM propidium iodide (PI) for 5 minutes at 37°C in PBS (Secondo et al., 2007). Cell death was determined by the ratio of the number of PI-positive cells to PI+FDA-positive cells (Wei et al., 2000).

Data were generated from three independent experiments. Ca²⁺ measurements were performed at least in 20 cells for each of the three independent experimental sessions. Data are expressed as mean±s.e.m. Statistical analysis was performed with ANOVA followed by Newman-Keul's test. Statistical significance was accepted at the 95% confidence level (P≤0.05).

We thank Paola Merolla for the editorial revision, Vincenzo Grillo for technical support, and Telethon Electron Microscopy Core Facility and Integrated Microscopy Facility for electron microscopy assistance. We also thank Maria Antonietta De Matteis (Telethon Institute of Genetics and Medicine, Naples) for the anti-VAP-A antibody. We are particularly grateful to Michael Duchen and Gyorgy Szabadkai of University College London for the AEQ constructs.