Apart from its known function, the plasma membrane Ca²⁺ATPase can regulate Ca²⁺ signaling by controlling phosphatidylinositol 4,5-bisphosphate levels

The plasma membrane Ca²⁺ ATPase (PMCA) belongs to the P-type ATPase family. Its major role is to remove excess Ca²⁺ from the cytosol and maintain intracellular Ca²⁺ homeostasis essential for the living cell (Brini and Carafoli, 2009; Strehler et al., 2007). The primary structure of PMCA shows about 30% identity with that of the sarco/endoplasmic reticulum Ca²⁺ ATPase (SERCA), and it has highly conserved sequence motifs that are common to most P-type ATPases (Verma et al., 1988). Information about the tertiary structure is not available for the PMCAs because it is difficult to crystallize them; therefore, crystal structures of other members of P-type ATPases (Bublitz et al., 2010; Toyoshima et al., 2000) are used to make structural predictions (Bublitz et al., 2011). According to these predictions, the overall cytoplasmic domain arrangement of the PMCAs is similar to that of SERCA except that PMCAs have two additional relatively large inserts; one connects the A domain with transmembrane helix 3, whereas the other is a long C-terminal tail that has binding sites for Ca²⁺-calmodulin and other regulatory elements.

There are four isoforms of PMCA (PMCA1–PMCA4, also known as ATP2B1–ATP2B4). Alternative splicing at splice sites ‘A’ (variants x, y, z, w) and ‘C’ (variants a, b, c) result in more than 20 PMCA variants (Strehler and Zacharias, 2001). According to their activation kinetics with Ca²⁺-CaM, the PMCAs are being classified as ‘slow’ and ‘fast’ pumps (Caride et al., 2001). It has long been known that, besides the activation with Ca²⁺-CaM, the activity of the PMCA is also increased by acidic phospholipids. Compared with other acidic lipids, however, lower concentrations of PtdIns(4,5)P₂ are required for their activation (Choquette et al., 1984; Lehotsky et al., 1992; Missiaen et al., 1989a; Missiaen et al., 1989b). Early proteolysis (Enyedi et al., 1987; Zvaritch et al., 1990) and peptide binding (Brodin et al., 1992; Filoteo et al., 1992) studies identified an acidic lipid binding sequence, the so-called AL region, close to transmembrane segment M3. This sequence has eight candidate residues that, if exposed, could provide positively charged side-chains for the electrostatic interaction with the lipids. Recent studies have shown that deletions and/or mutations within/or close to this putative lipid-binding region activate the pump in the same manner as do acidic lipids (Brini et al., 2010; Pinto and Adamo, 2002).

PtdIns(4,5)P₂ is a second messenger of the phospholipase C (PLC) signaling pathway (Berridge, 2009) and regulates many cellular functions, including membrane targeting (Schill and Anderson, 2009), cytoskeletal reorganization (Moss, 2012), endocytosis and exocytosis (Koch and Holt, 2012), as well as a variety of ion channels and receptors (Lemonnier et al., 2008; Michailidis et al., 2007; Michailidis et al., 2011; Suh and Hille, 2008; Trebak et al., 2009; Zhao et al., 2007). PtdIns(4,5)P₂ also interacts with many different proteins through either unstructured basic-residue-rich regions or more-structured domains, such as the pleckstrin homology (PH), Tubby, Fab1, YOTB, Vac1 and EEA1 (FYVE); Phox homology (PX), epsin N-terminal homology (ENTH), clathrin assembly lymphoid myeloid (CALM); phosphotyrosine binding (PTB), and 4.1, ezrin, radixin, moiesin (FERM) domains, and regulates their activity and/or localization (Balla, 2005; McLaughlin and Murray, 2005; McLaughlin et al., 2002).

In the last three decades we have learned how the overall level of PtdIns(4,5)P₂ in the membrane is modulated by phosphoinositol phosphate kinases and phospholipase C. However, it has become evident that specific membrane microdomains of high PtdIns(4,5)P₂ levels are formed by complexes of proteins with PtdIns(4,5)P₂. Such proteins have been called ‘PIPmodulins’ (Gamper and Shapiro, 2007; Lanier and Gertler, 2000). In the brain the myristoylated-alanine-rich C-kinase substrate (MARCKS) proteins have been implicated as major collectors of PtdIns(4,5)P₂ molecules (Kwiatkowska, 2010). Recent experiments have suggested that syntaxin-1A could also induce PtdIns(4,5)P₂ segregation and that PtdIns(4,5)P₂ could sequester syntaxin-1A into distinct lipid–protein clusters (van den Bogaart et al., 2011).

Here, we show for the first time that PMCAs can serve as PIPmodulins because they attract PtdIns(4,5)P₂ molecules and protect them from excessive depletion by phospholipase C. We also show that PMCA expression decreases receptor-mediated inositol 1,4,5-triphosphate (InsP₃) formation and, hence, Ca²⁺ release from intracellular pools. Molecular dynamics simulation revealed a region of positively charged side chains that appears to interact more strongly with PtdIns(4,5)P₂ than does the previously described AL region.

Previously a fluorescence resonance energy transfer (FRET)-based direct binding assay has suggested that there is a strong interaction between the purified erythrocyte PMCA pump and PtdIns(4,5)P₂ (Lehotsky et al., 1992). Here, we wanted to see how PMCA affected PtdIns(4,5)P₂ signaling in live cells. In order to study this we used three species of PMCA and two types of HeLa cells in five distinct experimental setups: (1) sh-HeLa cells, in which the endogenous PMCA4 was knocked down to achieve the lowest possible PMCA level (supplementary material Fig. S1); (2) sh-HeLa cells overexpressing PMCA2wb, the ‘w’ splice variant of PMCA2b, which is a rapidly activated ‘fast’ PMCA isoform encoded by the gene ATP2B2 (Caride et al., 2001); (3) wild-type HeLa cells, with small amounts of endogenous PMCA4 and PMCA1; (4) wild-type HeLa cells overexpressing PMCA4b-LA, which is a localization mutant of the slowly activated PMCA4b encoded by gene ATP2B4 [the designation LA indicates mutation of three leucine residues to alanine residues at the extreme C-terminus of PMCA4b at positions 1167–1169 (highlighted in red letters on a gray background in the sequence of supplementary material Fig. S4) – this mutant had the same enzymatic characteristics as the wild-type PMCA4b pump, but showed enhanced plasma membrane localization (Antalffy et al., 2013)]; and (5) wild-type HeLa cells overexpressing a non-functional (dead) PMCA4b-(D⁶⁷²E)-LA mutant (hereafter referred to as dPMCA4b-LA; the location of the D⁶⁷²E mutation is shown in supplementary material Fig. S4, as a white letter on purple). Given that this mutant cannot transport Ca²⁺ (Adamo et al., 1995), it allowed us to separate the Ca²⁺ extrusion from the PtdIns(4,5)P₂ binding characteristics of the pump. This set of five conditions, covered the range from cells almost lacking PMCAs (1 and 3), through cells containing intact but inactive PMCAs (5), cells containing large amounts of a ‘slow’ PMCA (4), to cells containing large amounts of a ‘fast’ PMCA (2). The relative levels of PMCA expression in the different cell types and the different experimental setups are shown in the supplementary material Fig. S1. It is important to note that the expression levels of the different PMCA constructs are comparable, and ∼100 times higher than that of the endogenous pumps in wild-type HeLa cells.

In our studies, we also used four fluorescent reporter molecules. GFP or mCherry was covalently attached to PMCA and allowed visualization of the PMCA location. The third reporter molecule, PH_PLCδ1–RFP [which has the pleckstrin homology (PH) domain of PLCδ1 fused to RFP] (Szentpetery et al., 2009; Várnai and Balla, 2008) is a sensor of PtdIns(4,5)P₂ and of InsP₃ molecules. We refer to this construct as the PtdIns(4,5)P₂ sensor. The fourth reporter, GCaMP2, is a sensor of the intracellular Ca²⁺ concentration.

Fig. 1 shows how the intracellular distribution of the PtdIns(4,5)P₂ sensor was affected by expression of PMCA when Ca²⁺ was introduced into the cell. Before introduction of Ca²⁺ by addition of ionomycin, the PtdIns(4,5)P₂ sensor localized mainly in the plasma membrane either when expressed alone or together with PMCA2wb or PMCA4b-LA. The graphs show the PtdIns(4,5)P₂ sensor (red) and PMCA (green) signal intensity profiles along the line of interest marked on the confocal images.

Ionomycin treatment increases [Ca²⁺]_i, which leads to the activation of PLC and then PtdIns(4,5)P₂ hydrolysis, resulting in the translocation of the PtdIns(4,5)P₂ sensor from the plasma membrane to the cytosol (Szentpetery et al., 2009). When the PtdIns(4,5)P₂ sensor moves from the plasma membrane to the cytosol, this reflects both the reduction of PtdIns(4,5)P₂ in the plasma membrane and the generation of InsP₃ in the cytoplasm (Hirose et al., 1999). Fig. 1B shows that PMCAs dramatically reduce the movement of the PtdIns(4,5)P₂ sensor during ionomycin treatment. In the absence of PMCA, the PtdIns(4,5)P₂ sensor was distributed homogeneously over the entire cell, whereas in cells expressing GFP–PMCA4b-LA or GFP–PMCA2wb, a high level of the PtdIns(4,5)P₂ sensor signal was detected in the plasma membrane even after the ionomycin treatment (see the graphs in Fig. 1).

To study further the effect of PMCAs on the distribution of PtdIns(4,5)P₂, cells were exposed to increasing concentrations of ionomycin (Fig. 2). Images in Fig. 2A show the location of the PtdIns(4,5)P₂ sensor before treatments and at 5 minutes after the addition of 2, 5 and 10 µM ionomycin, as indicated. The images of Fig. 2A compare the effect of PMCA4b-LA (bottom) with control wild-type HeLa cells (middle) and cells in which PMCA4 is suppressed (sh-HeLa cells; top). The graphs of Fig. 2B show time courses of the changes in the cytosolic fluorescence intensities of the PtdIns(4,5)P₂ sensor, and the bar graph in Fig. 2C shows the frequency of cells responding to a given ionomycin concentration with complete translocation of the sensor. Experiments of this sort allowed us to classify cells according to their sensitivity to ionomycin. In Fig. 2B, the signal measured in sh-HeLa cells are shown in gray color. The light grey curve shows the averaged responses of highly ionomycin-sensitive cells (18% of total sh-HeLa cells) in which addition of 2 µM ionomycin induced full translocation of the PtdIns(4,5)P₂ sensor from the plasma membrane to the cytosol; 24% of cells (Fig. 2B, medium-gray curve) responded to 5 µM ionomycin and the remaining cells (Fig. 2B, dark gray curve) responded to 10 µM ionomycin resulting in full PtdIns(4,5)P₂ sensor translocation in all sh-HeLa cells (see also supplementary material Movie 1). Expression of GFP–PMCA2wb in sh-HeLa cells protected the PtdIns(4,5)P₂ pool, so that the sensor remained in the plasma membrane region during the ionomycin treatment even at high (10 µM) ionomycin concentration (Fig. 2B, red curve). Similarly, ionomycin could not destroy PtdIns(4,5)P₂, and the PtdIns(4,5)P₂ sensor remained in the plasma membrane when GFP–PMCA4b-LA was expressed in wild-type HeLa cells (Fig. 2B, blue curve; supplementary material Movie 2).

To show the effect of an inactive PMCA in sequestering and protecting the PtdIns(4,5)P₂ molecules in the plasma membrane, we used the GFP-tagged inactive dPMCA4b-LA mutant. Fig. 2B,C shows that in HeLa cells expressing the inactive pump, high (10 µM) ionomycin concentration was required to induce substantial movement of the PtdIns(4,5)P₂ sensor into the cytosol, whereas lower ionomycin concentrations caused translocation of the PtdIns(4,5)P₂ sensor in only a few cells (Fig. 2C). In contrast, 27% of wild-type HeLa cells expressing the endogenous low amounts of PMCA4b and PMCA1b responded to 5 µM ionomycin. Unlike sh-HeLa cells, however, none of the wild-type HeLa cells responded to 2 µM ionomycin, indicating that the endogenous pumps also provide some protection from PLC. Addition of 10 µM ionomycin, by contrast, induced translocation of the sensor in all wild-type or sh-HeLa cells. From these experiments, it is evident that functional PMCAs can protect the PtdIns(4,5)P₂ pool from an excess Ca²⁺ overload and PLC activation. The comparison of the dPMCA4b-LA-expressing cells with the wild-type HeLa cells in Fig. 2C suggests that even the inactive PMCA mutant could provide a significant shielding of PtdIns(4,5)P₂ from hydrolysis.

We also examined how the different PMCA variants affected the Ca²⁺ transients during the ionomycin treatments (Fig. 3) by repeating the same experiments as above using the GCaMP2 Ca²⁺ sensor to monitor [Ca²⁺]_i and mCherry-tagged PMCAs (supplementary material Fig. S1). In sh-HeLa cells, ionomycin induced long lasting Ca²⁺ increases and a high final signal value (Fig. 3B, black curve). In cells expressing mCherry–PMCA2wb, addition of ionomycin induced only small and short-lived Ca²⁺ signals (Fig. 3B, red curve) explaining why the PtdIns(4,5)P₂ sensor was not translocated in the corresponding experiments. Expression of the mCherry–PMCA4b-LA pump resulted in much larger ionomycin-evoked Ca²⁺ signals (Fig. 3B, blue curve) than in the PMCA2wb-expressing cells, although they were somewhat less pronounced than in the sh-HeLa cells. In cells expressing the inactive dPMCA4b-LA, the [Ca²⁺]_i response upon ionomycin stimulation (Fig. 3B, green curve) was greater than that measured in PMCA4b-LA expressing cells, as expected. The response of the dPMCA4b-LA-expressing cells was similar to that of the wild-type HeLa cells in this respect (data not shown).

To study further the role of the PMCAs in PtdIns(4,5)P₂ signaling we induced activation of PLC through G-protein-coupled receptors. In this case, initial activation of PLC does not need an increase in [Ca²⁺]_i. Instead, PLC is activated through G-proteins to hydrolyze PtdIns(4,5)P₂ and release InsP₃, which then stimulates Ca²⁺ release from the endoplasmic reticulum (ER) (Berridge, 2009). An example of the type of protocol used in our experiments is shown in supplementary material Fig. S2. First, the PLC signaling pathways were activated with a saturating concentration of extracellular ATP. After stimulation, purinergic receptors quickly desensitize (Mundell and Kelly, 2011), therefore we used histamine, another G-protein-coupled agonist, to test how effectively the cells can respond to a second stimulus. Both receptor stimuli were added in a Ca²⁺-free medium to show only the InsP₃-mediated signal. It is clear from supplementary material Fig. S2 that in response to ATP stimulation only a fraction (10–20%; see also Fig. 4A1,B1) of total PtdIns(4,5)P₂ is degraded, inducing robust Ca²⁺ release from the internal stores in good accordance with recent findings (Dickson et al., 2013). After 5 minutes, Ca²⁺ was added back to the medium to allow the entry of extracellular Ca²⁺ into the cells through channels that open in response to the low concentration of Ca²⁺ in the lumen of the ER (store-operated Ca²⁺ entry, SOCE) (Bird et al., 2008). Finally, we added a high concentration of ionomycin to evoke a maximal response of the system.

Next, we examined how PMCAs affected InsP₃ formation and the Ca²⁺ signal in the absence of external Ca²⁺. In both HeLa and sh-HeLa cells, addition of ATP resulted in a sustained increase in the cytosolic intensity of the PtdIns(4,5)P₂ sensor indicating a substantial release of InsP₃ (Fig. 4A1,B1, black curves). In good correlation with the elevated InsP₃ concentration, the corresponding Ca²⁺ signaling experiments in Fig. 4A2,B2 (black curves) showed a rapid Ca²⁺ release from the internal store followed by a slowly descending phase. The subsequent histamine stimulus resulted in only minor peaks of both InsP₃ and Ca²⁺ superimposed on the first one suggesting that the freely available PtdIns(4,5)P₂ and/or ER resources were largely depleted. The shape of the Ca²⁺ signals was similar in both wild-type HeLa and sh-HeLa cells (Fig. 4A2,B2) and the area under the ATP peaks was not significantly different between the two types of cells (Fig. 4C1).

In contrast, in HeLa cells overexpressing PMCA4b-LA, ATP caused a much shorter Ca²⁺ peak followed by a short histamine-induced secondary response (Fig. 4B2, blue curve). Consistent with this, the translocation of the PtdIns(4,5)P₂ sensor was also less pronounced (Fig. 4B1), and the signal decayed after the ATP and histamine stimuli, indicating quick relocation of the sensor to the plasma membrane in the PMCA4b-LA-expressing cells. Strikingly, the Ca²⁺ signals in cells expressing PMCA4b-LA or the inactive dPMCA4b-LA were almost identical during the early period of agonist stimulation in the absence of extracellular Ca²⁺ (Fig. 4B2,C1). The fact that Ca²⁺ returned to the baseline level in case of the inactive dPMCA4b-LA underlines the importance of other mechanisms (the SERCA pump and/or mitochondria) in eliminating Ca²⁺ from the cytosol. The signals of the PtdIns(4,5)P₂ sensor were also similar, although the cells with the inactive dPMCA4b-LA showed a slightly higher peak after the first stimulus than the cells with the active pump (Fig. 4B1). These experiments suggest that during the early period of agonist stimulation, the slow PMCA4 variant did not act through its Ca²⁺-removing ability but rather through its ability to bind PtdIns(4,5)P₂ making it less accessible to PLC. To fulfill this role it is likely that a relatively large amount of the pump is required, which is accomplished by the transient transfection in our experiments (supplementary material Fig. S1B).

The PtdIns(4,5)P₂ sensor signal was similar in cells expressing either the ‘fast’ PMCA2wb or the ‘slow’ PMCA4b-LA pumps (Fig. 4A1,B1), suggesting that the PtdIns(4,5)P₂-binding capacity of both pumps was similar. The Ca²⁺ transients, however, returned to the baseline level more quickly in cells expressing the ‘fast’ PMCA2wb (supplementary material Fig. S3A), indicating that in the ‘slow’ PMCA4b-LA-expressing cells it was mostly the PtdIns(4,5)P₂-binding ability that contributed to the attenuation of the InsP₃-mediated Ca²⁺ signal, whereas in the case of the ‘fast’ PMCA2wb-expressing cells it was both the PtdIns(4,5)P₂-binding and pumping function that contributed. Control experiments shown in supplementary material Fig. S3B, show that expression of the PtdIns(4,5)P₂ sensor alone had no effect on the Ca²⁺ signal, suggesting that the sensor does not interfere with PtdIns(4,5)P₂ hydrolysis, as expected (Szentpetery et al., 2009; van der Wal et al., 2001).

Fig. 4 also shows that the Ca²⁺-pumping activity of PMCA4b-LA was essential in removing Ca²⁺ after Ca²⁺ entry through the plasma membrane Ca²⁺ channels. Adding Ca²⁺ back to the PMCA4b-LA-expressing cells after ER store depletion resulted in a Ca²⁺ peak with smaller amplitude and shorter period than that observed in wild type HeLa cells. Here, we found a pronounced difference between cells expressing the active or the inactive dPMCA4b-LA; in the dPMCA4b-LA-expressing cells the Ca²⁺ curve had a long lasting decay phase similarly to the control cells (Fig. 4B2,C2). These results are in accordance with previous findings on the role of the PMCA in shaping Ca²⁺-entry-mediated responses (Bautista et al., 2002; Bautista and Lewis, 2004; Ritchie et al., 2012; Snitsarev and Taylor, 1999).

Next, we wanted to understand how PMCAs provide such a protecting environment for the PtdIns(4,5)P₂ molecules. The model of PMCA in the E1 conformation (see Materials and Methods) shows a collar of positively charged residues (Arg and Lys) at the cytoplasmic face of the membrane (Fig. 5A). Fig. 5B–D shows the surface electrostatic potential representations of B, the same molecule; C, a hypothetical molecule missing the PMCA-specific inserts (these include the AL region and the complete C-terminal tail); and D, the E1 conformation of SERCA (PDB 1T5S). These maps show a nearly continuous positive surface potential around the neck (the ‘blue collar’) of the PMCA that must arise from a cluster of basic residues. The previously identified lipid-binding (AL) domain was a stretch of 25 amino acids, which were contiguous in the primary sequence of PMCA. In contrast, the ‘blue collar’ is composed of residues from all parts of the stalk regions of PMCA, which is also present in the hypothetical PMCA molecule missing the AL region (Fig. 5C). The blue collar residues are assembled into a coherent structure when the secondary and tertiary structures of PMCA are formed. Fig. 5D shows SERCA, which lacks nearly all basic residues in the stalk region (supplementary material Figs S4, S5) and therefore, also the blue collar.

In order to observe if and where the PtdIns(4,5)P₂ molecules bind to the blue collar, 15 molecular dynamics runs were carried out; each run lasted 10 nanoseconds and contained five PtdIns(4,5)P₂ molecules, for a total of 75 PtdIns(4,5)P₂ molecules. Of these 75 molecules, 20 bound to one residue, 24 bound to more than one amino acid residue and 31 remained unbound. A PtdIns(4,5)P₂ molecule was considered bound when it remained within 4.0 Å of a PMCA residue for many nanoseconds. Each run showed a different pattern of binding because of the randomness of the original PtdIns(4,5)P₂ positions, but these runs showed four pockets of positively charged residues that bound PtdIns(4,5)P₂.

The images in Fig. 6A,B show the four binding pockets; left to right, they are K173 (red), R993 (green), K980 (purple) and R914 (blue). Of the 24 multiply bound PtdIns(4,5)P₂ molecules, eight bound to pocket R993, seven to R914, five to K980, three to K173 and one in an unusual way to residues K365 and K909. The interaction energies of the PtdIns(4,5)P₂ molecules with their binding pockets are summarized in Fig. 6D. Given that it was only possible to do 10-nanosecond runs, it is probable that these PtdIns(4,5)P₂ molecules are not yet in their most favorable positions relative to the binding pocket. This point of view is reinforced by the observation that the positions of PtdIns(4,5)P₂ molecules in the different runs are different. For comparison, the interaction energy was calculated for a PtdIns(4,5)P₂-binding site from a K⁺ channel (PDB 3SPI) (Hansen et al., 2011). Given that this structure was that of a crystal, we would expect the PtdIns(4,5)P₂ and the binding pocket to be closer to their most favorable position. This is probably why the interaction energy is stronger for the 3SPI channel than the strongest interaction in PMCA.

The strongest interaction energy for a binding of PtdIns(4,5)P₂ to R993 was −2550 kJ/mol, about 62% of the −4091 kJ/mol found for binding of PtdIns(4,5)P₂ to its site in the K⁺ channel (Hansen et al., 2011). The strength of the interaction of PtdIns(4,5)P₂ with R993 shows that its binding is probably biologically significant.

Table 1 shows that PtdIns(4,5)P₂ binds to amino acid residues more strongly than other phospholipids do. PtdIns(4,5)P₂ also binds more frequently to multiple residues. Not only does PtdIns(4,5)P₂ have a higher negative charge than the other lipids, but its negative charge is in two locations, the P1 phosphorus atom and the P4 and P5 phosphorus atoms. The P1 atom is closer to the surface of the membrane, where it can bind to the lower part of the blue collar, whereas the P4 and P5 atoms are further from the membrane surface and can bind to the upper parts of the blue collar.

This charge distribution of PtdIns(4,5)P₂ contributes especially strongly to its binding to pocket R993. The alignments of supplementary material Figs S4 and S5 show the alpha carbon of residue R993 in the middle of transmembrane helix M9. In the three-dimensional model, residue R993 reaches towards the cytoplasm, so that the charged guanido group is in the amphiphilic boundary layer between the hydrophobic core of the membrane and the cytoplasm. Here, it is positioned so that it will interact with the P1 charged group of PtdIns(4,5)P₂ and other anionic lipids, as shown in Fig. 6C.

All of the studies reported here were performed with overexpressed PMCA and with models. Nevertheless, they have provided useful information about the binding of PtdIns(4,5)P₂ to PMCA and suggest new roles for PMCA in managing the Ca²⁺ signal. It has long been known that PMCA pumps Ca²⁺ out of cells and that PtdIns(4,5)P₂ stimulates that activity. By using a PtdIns(4,5)P₂ sensor we found that expression of PMCA protects PtdIns(4,5)P₂ molecules from hydrolysis by PLC and that the protection operates by two mechanisms. One requires only an intact PMCA (no activity is required); the second requires a fully active PMCA. The presence of a buffering mechanism by which inactive PMCA acts suggests that PMCAs must bind PtdIns(4,5)P₂ and create a PtdIns(4,5)P₂ pool, which is probably in equilibrium with free PtdIns(4,5)P₂ molecules in the plasma membrane. Hence, the retention of the PtdIns(4,5)P₂ sensor in the plasma membrane upon expression of the inactive PMCA is due to the binding of the sensor to the free PtdIns(4,5)P₂ molecules, which are maintained in equilibrium with the PtdIns(4,5)P₂ bound by the PMCA. The role of such a PtdIns(4,5)P₂-buffering mechanism has been described recently, as the MARCKS protein can regulate the activity of TRPC1 channel through buffering PtdIns(4,5)P₂ molecules in the vicinity of the channel (Shi et al., 2013).

In general, cells express small amounts of PMCAs; red blood cells, for example, express only 35 copies per square micron so is much less abundant than plasma membrane PtdIns(4,5)P₂ molecules [about 5000–10,000 per square micron (Suh and Hille, 2008)]. We suggest that buffering of PtdIns(4,5)P₂ by the PMCAs might have a substantial impact on PtdIns(4,5)P₂ signaling when a large amount of pump is being expressed at discrete locations, like the dendritic spines or the stereocilia of hair cells (Burette et al., 2010; Zhao et al., 2012). We must emphasize, however, that even a small amount of active PMCA could be protective if it is located near the PtdIns(4,5)P₂ pool. The PMCA–PtdIns(4,5)P₂ pool could have a great importance in PtdIns(4,5)P₂-related processes by (1) preventing complete depletion of PtdIns(4,5)P₂ and (2) encouraging the formation of extensive PtdIns(4,5)P₂-rich areas in specific membrane locales where PMCA is abundant.

Calculation of the electrostatic potential at the surface of a PMCA4b model disclosed a positively charged electrostatic patch around the stalk region of the molecule that we called the blue collar. Molecular dynamics simulation experiments revealed four PtdIns(4,5)P₂-binding pockets formed from the blue collar residues of PMCA4b. On the basis of these findings we propose that PtdIns(4,5)P₂ interacts not only through non-specific electrostatic interactions with the previously described lipid-binding AL sequence (Brodin et al., 1992; Filoteo et al., 1992) but also through specific interactions by fitting its phosphatidylinositol head-group into binding pockets. Specific pocket binding of the PtdIns(4,5)P₂ head is also used by other molecules. X-ray crystal structures of the Kir2.2 (Hansen et al., 2011) and GIRK2 K⁺ (Whorton and MacKinnon, 2011) channels have identified highly specific phosphatidylinositol-binding regions at an interface between the transmembrane domain and the cytoplasmic domains of these molecules. In addition, several phosphoinositide-binding signal molecules use structured domains to achieve specific recognition of different phosphoinositide head groups (Kutateladze, 2010).

Supplementary material Figs S4–S6 show the relationship of the human PMCA4b to other PMCAs and SERCA. Supplementary material Fig. S5 shows that the four genes that encode for human PMCA retain almost all of the residues that make up the blue collar. Of the residues that comprise the binding sites (highlighted in cyan), the only exceptions are R993 (missing in human PMCA3 and human PMCA1), R1052 (present only in human PMCA4) and K89 (missing in human PMCA3). This figure also shows that SERCA1a does not retain any corresponding residues, except for R438, R439 and K442. In many of the corresponding positions, SERCA1a has acidic residues instead of basic ones. These relationships indicate that the PtdIns(4,5)P₂-binding sites of PMCA have special functions not shared by SERCA.

Supplementary material Fig. S6 shows that the PtdIns(4,5)P₂-binding sites are conserved in PMCAs throughout the animal kingdom. From humans to fish and insects, the binding residues are almost always retained. However, in other organisms (plants and prokaryotes) PMCA-type molecules have very different structures not only in the blue collar region, but also in most other parts of the molecule. Supplementary material Fig. S4 shows the alignment of the whole SERCA and PMCA sequences, which were used for constructing the model.

The properties of PMCA reported here suggest that it might function as a unique PIPmodulin that mediates PtdIns(4,5)P₂ clustering at specific locations. Given that PMCA binds PtdIns(4,5)P₂ and provides a low Ca²⁺ milieu; it might form a ‘Ca²⁺-less PMCA–PtdIns(4,5)P₂ pool’ (CPPP) in the cell membrane. The known PIPmodulins, such as the myristoylated alanine-rich C-kinase substrate (MARCKS), growth-associated protein-43 (GAP43) and cytoskeleton-associated protein (CAP23), also sequester PtdIns(4,5)P₂ molecules (Laux et al., 2000), but when they bind Ca²⁺-calmodulin these PIPmodulins release PtdIns(4,5)P₂. In contrast, as is seen in Figs 1, 2, PMCA keeps PtdIns(4,5)P₂ in the plasma membrane even in the presence of elevated Ca²⁺. PtdIns(4,5)P₂ also enhances the activity of the PMCAs; this would be expected to reduce the Ca²⁺ concentration in the vicinity of the CPPP, reducing the activity of PLC and stabilizing the PMCA-bound PtdIns(4,5)P₂ compartment.

Moreover, there are certain plasma membrane compartments with high PtdIns(4,5)P₂ content where the members of the MARCKS family are not expressed and the putative PIPmodulin is likely to be PMCA. Three-dimensional simulations suggest that local sequestration of PtdIns(4,5)P₂ in the spine heads of Purkinje neurons is responsible for the observed InsP₃ signaling (Brown et al., 2008). MARCKS, GAP43 or CAP23 are not found in Purkinje cells (ConsoleBram et al., 1996; Ouimet et al., 1990) but PMCA2 is concentrated in the spine heads of Purkinje neurons (Burette et al., 2010). Another example where both PtdIns(4,5)P₂ and PMCA are abundant in the same specific compartment is the stereocilia of hair cells in the inner ear. In these regions of the hair cells, ∼2000 PMCA2 molecules are gathered in a square micron (Yamoah et al., 1998). If we take the assumption that about 5000–10,000 PtdIns(4,5)P₂ molecules are in a square micron of the plasma membrane we might assume that most of them are sequestered by the large amount of PMCA in the stereocilia. In good accordance with this hypothesis, recent studies have demonstrated that PtdIns(4,5)P₂ is concentrated at the tips of stereocilia together with the PMCA2 protein (Zhao et al., 2012). Further experiments have demonstrated that a specific asymmetric distribution of PtdIns(4,5)P₂ (Hirono et al., 2004) and PMCA2 (Kozel et al., 1998) is essential for the appropriate function of hair cells, underlining the importance of a putative CPPP in hearing.

In summary, we suggest that by protecting PtdIns(4,5)P₂ molecules, PMCA can mediate the cellular concentrations of two messenger molecules, InsP₃ and Ca²⁺, and, hence, might play a substantial role in the complex interplay between local PtdIns(4,5)P₂ and Ca²⁺ signaling. In addition, by binding PtdIns(4,5)P₂, PMCAs might affect many cellular functions regulated by PtdIns(4,5)P₂, such as the activity of channels and pumps, remodeling of the actin cytoskeleton, endocytosis and migration. Finally, imbalances of the PtdIns(4,5)P₂ and PtdIns(3,4,5)P₃ phospholipids might cause defects in cell functions leading to cancer and other diseases. Therefore, controlling the level of these molecules by use of PMCAs might protect cells against such diseases.

Fugene HD Transfection Reagent was obtained from Roche Applied Science (Mannheim, Germany). All other chemicals used were of reagent grade.

The mammalian expression construct for PH_PLCδ1–RFP was a kind gift from Péter Várnai (Semmelweis University, Hungary) and is as described previously (Várnai and Balla, 2008). pN1-GCaMP2 plasmid was a kind gift from Junichi Nakai, RIKEN Brain Science Institute, Saitama, Japan (Nakai and Ohkura, 2003). Generation of the pEGFP-PMCA2wb plasmid was described previously (Chicka and Strehler, 2003). The mCherry–PMCA2wb construct was created by replacing the EGFP fragment of the pEGFP-PMCA2wb plasmid with the mCherry coding sequence from the pmCherry-C1 vector (Clontech) using the AgeI-KpnI restriction sites. The pEGFP-PMCA4b-L^1167–1169A plasmid was constructed as described previously (Antalffy et al., 2013). To create the mCherry-PMCA4b-L^1167–1169A construct, site-directed mutagenesis, introducing the L^1167–1169A triple mutation was performed on the mCherry-PMCA4b template plasmid (Antalffy et al., 2013) using the QuikChange II Site-Directed Mutagenesis Kit (Stratagene) with the following primers: forward, 5′-CTAAGTTTGGGACTAGGGTGGCAGCGGCGGATGGTGAGGTCACTCCATATGCC-3′; reverse, 5′-GGCATATGGAGTGACCTCACCATCCGCCGCTGCCACCCTAGTCCCAAACTTAG-3′. pEGFP-PMCA4b-D⁶⁷²E-L^1167–1169A and mCherry-PMCA4b-D⁶⁷²E-L^1167–1169A constructs were created by introducing the D⁶⁷²E point mutation to the PMCA4b-LA plasmids using the QuikChange II Site-Directed Mutagenesis Kit (Stratagene). PCR primers were as follows: forward, 5′-GGTGGGCATTGAGGAGCCTGTGCGCCCAGAG-3′; reverse, 5′-CTCTGGGCGCACAGGCTCCTCAATGCCCACC-3′.

The sh-HeLa cell line was generated by using the PMCA4 shRNA plasmid (Santa Cruz Biotechnology, sc-42602-SH). HeLa cells were transfected with PMCA4 shRNA plasmid or with control shRNA plasmid-A (sc-108060) in a six-well plate by using shRNA plasmid transfection medium (sc-108062) and shRNA plasmid transfection reagent according to the manufacturer's protocol. At 48 hours after transfection cells were selected by puromycin dihydrochloride (2 µg/ml) (sc-108071) in DMEM for 2 weeks and medium was changed to fresh medium with puromycin every 2–3 days. After selection, cell clones were prepared using cloning rings. PMCA4b expression was tested by western blot analysis. From the most efficient clone, the sh-HeLa cell line was generated.

HeLa cells (obtained from ECACC) or sh-HeLa cells were grown in DMEM supplemented with 10% FBS, 100 U/ml penicillin, 100 µg/ml streptomycin and 2 mM glutamine under 5% CO₂ at 37°C. At 1 day prior to transfection, cells were seeded into a Lab-Tek^TM II Chambered Slide (Nalge Nunc Int.) at a density of 4×10⁴–6×10⁴ cells. Transfection using the appropriate combination of plasmids, described in supplementary material Fig. S1, was carried out with FuGene HD (Roche) according to the protocol of the manufacturer. This procedure yielded about 60–70% transfected cells in the cell culture. Cells were inspected 24 hours after transfection.

HeLa or sh-HeLa cells were seeded into a Lab-Tek^TM II Chambered Slide (Nalge Nunc Int.) and transfected with PH_PLCδ1–RFP alone or together with GFP-tagged PMCA constructs for the PtdIns(4,5)P₂ translocation experiments; or with GCaMP2 alone or together with mCherry-tagged PMCA constructs for Ca²⁺ imaging (supplementary material Fig. S1). Prior to the measurement of sensitivity of cells to ionomycin, the culture medium was replaced with Hank's buffered salt solution (HBSS) supplemented with 0.9 mM MgCl₂, 2 mM CaCl₂ and 20 mM Hepes (pH 7.4). Cells were sequentially exposed to 2 µM, 5 µM and 10 µM ionomycin at 0, 6 and 12 minutes. In order to study activation of the PLC pathway by G-protein-receptor-coupled agonists the culture medium was replaced with nominally Ca²⁺-free HBSS supplemented with 0.9 mM MgCl₂, 100 µM EGTA, 100 µM CaCl₂ and 20 mM Hepes pH 7.4. Cells were stimulated by the consecutive additions of 100 µM ATP and 100 µM histamine (at 0 and 2 minutes) in the absence of Ca²⁺ (InsP₃-mediated signal). After 5 minutes incubation, the extracellular medium was replenished by 2 mM Ca²⁺ and the store-operated Ca²⁺ entry (SOCE signal) was monitored for 5 more minutes. Finally, the maximal response was induced by the addition of 10 µM ionomycin.

Images were taken with an Olympus IX-81 laser scanning confocal microscope and Fluoview FV500 software using an Olympus PLAPO 60× (1.4) oil immersion objective. For PH_PLCδ1–RFP imaging, cells were excited at 543 nm and emission was collected above 560 nm. For GCaMP2 imaging, cells were excited at 488 nm and emission was collected between 505 and 535 nm. Regions of interest were selected in the cytosol, nuclei were omitted. GFP-tagged PMCAs were illuminated at 488 nm and emission between 505 and 535 nm was recorded. mCherry-tagged PMCAs were illuminated at 543 nm and emission above 560 nm was recorded. Images were acquired every 0.3–1.2 seconds, the z-resolution was 0.5 µm. Time-lapse sequences were recorded with the Fluoview Tiempo (v4.3) time course software at room temperature. The relative fluorescence was calculated as F/F₀ (where F₀ was the average initial fluorescence). GraphPad Prism4 software was used to analyze the experimental data.

Data presented in Fig. 4 are means of at least three independent experiments (±s.e.m.) and statistical significance was determined by Student's t-test. For statistical analysis of data in Fig. 2, cells were scored according to the ionomycin concentration at which full translocation of the PtdIns(4,5)P₂ sensor occurred. Statistical significance of image scores was determined by the χ² test. Statistical analyses were performed using Excel software (Microsoft) and GraphPad Prism 4 software (GraphPad Software Inc.).

The Ca²⁺ pump of rabbit sarcoplasmic reticulum SERCA1a has been crystallized at several stages of its catalytic cycle (Toyoshima et al., 2004; Moller et al., 2010) and has considerable resemblance to PMCA4b. Supplementary material Fig. S4 shows an alignment of these two pumps. Study of this figure shows that PMCA is longer than SERCA, the extra protein in PMCA being at the N- and C-terminal domains and in a long insert marked as I. Not including these three regions, the PMCA sequence contained 31% residues identical to SERCA, as well as numerous similar residues. In addition to sequence identities, proper positioning of the hydrophobic transmembrane regions assures an accurate alignment (PMCA and SERCA have ten transmembrane regions each).

The structures of the SERCA AMPPCP-Ca2 (PDB 1T5S and 1VFP) and of the N-domains of the Na⁺/K⁺ ATPase (PDB 1MO7 and 1MO8) were aligned with PMCA4b and a model of PMCAE1 was made. The alignment of PMCA4b with SERCA1a was the same as is shown in supplementary material Fig. S4. Modeller 8v2 was then used to create the PMCA model covering residues 1–1058. It has been reported that F1094 crosslinks to the peptide C537-ALLGFVT in the N domain (Carafoli et al., 1992; Falchetto et al., 1991; Falchetto et al., 1992). Accordingly, residues 1059–1094 (part of the C-terminus) were added next, with F1094 placed near C537. This caused the C-terminus to go away from the membrane interface, up over the cytoplasmic top of PMCA. The remainder of the C-terminus (residues 1095–1205) was modeled using Tasser (Roy et al., 2010) and attached to the end of PMCA.

In general, the subsequent work was performed using CHARMM (Brooks et al., 1983; http://www.charmm.org/) and CHARMM-GUI (Jo et al., 2008; http://www.charmm-gui.org/). The PMCA was put into a hexagonal box with a lipid bilayer, TIP3 water and enough K⁺ and Cl⁻ ions to make a neutral system with 0.15 M salt. The lipid composition was chosen to approximate that of the plasma membrane of 3T3 fibroblasts (Pankov et al., 2006). The outer leaflet of the membrane contained 17 cholesterol molecules (CHL-13%), 82 of palmitoyloleoylphosphatidylcholine (POPC, 63%) and 31 of stearoylarachidonylphosphatidylcholine (SAPC, 24% representing sphingomyelin). The inner (cytoplasmic) leaflet contained 17 cholesterol molecules (CHL, 13%), 31 of palmitoyloleoylphosphatidylcholine (POPC, 24%), 4 of palmitoyloleoylphoshatidic acid (PA, 3%), 42 of palmitoyloleoylphosphatidylethanolamine (POPE-32%), 16 of palmitoyloleylphosphatidylserine (POPS, 13%) 15 of stearoylarachidonylphosphatidylinositol (PI, 12%) and 5 of stearoylarachidonylphosphatidylinositolbisphosphate [PtdIns(4,5)P₂, 3%].

After minimization and heating, the molecular dynamics studies were performed on this hexagonal box containing the 1205 residues of PMCA, 250 molecules of membrane lipid, 32074 TIP3 water molecules, 132 K⁺ ions and 63 Cl⁻ ions at 310° K. The box was 104 Å on each side and its initial length was 155 Å. The early stages of the dynamics run were performed at constant pressure and temperature; the dimensions of the box were allowed to vary to keep the pressure constant. After the initial fluctuations in the dimensions had diminished, the dimensions of the box were fixed and the run carried on at constant volume and temperature.

The electrostatic and van der Waals interaction energies between various atoms were calculated by CHARMM. These energies are calculated directly from the physical formulas, and so are the total energies rather than the energies with respect to a standard reference state. They also lack a PV term, but this does not affect the validity of the results, because the electrostatic energy is clearly the dominant term, and the one that changes the most.

In order to keep the molecular dynamics run times to reasonable values, a large portion of the atoms in the model were fixed in space. After the initial phase was running smoothly, all atoms in the cytoplasmic half of the prism were fixed, as well as the backbone of the protein, the side chains and waters in the extracellular part of the prism. This left the side chains of the collar, the lipids of the membrane and nearby waters free to move.

Also because of time limitations, it was necessary to place the PtdIns(4,5)P₂ molecules near PMCA in order to observe the interactions within the 10 nanoseconds of simulation. The five PtdIns(4,5)P₂ molecules were placed at random starting positions near the PMCA in order to sample all the possible binding sites. 15 runs of 10 nanoseconds each were performed, and the interactions of lipid with the protein observed in each.

Some variations from these conditions occurred. In the first two runs, the PtdIns(4,5)P₂ was not placed near PMCA, and only one interaction was observed. In some of the early runs, the side chains of the blue collar were not allowed to move.

We are grateful to Krisztina Lór for excellent technical assistance, to Junichi Nakai (RIKEN Brain Science Institute, Saitama, Japan) for providing the pN1-GCaMP2 plasmid, to Péter Várnai (Semmelweis University, Hungary) for the PH_PLCδ1–RFP plasmid and to Robert Katona (Institute of Genetics, Biological Research Center of the Hungarian Academy of Sciences) for the pcDNA3-mCherry plasmid. The SB-CAG-Amaxa GFP vector was a gift of Tamás Orbán (Institute of Molecular Pharmacology, Research Centre for Natural Sciences, Hungarian Academy of Sciences).