Group IV Phospholipase A2α Controls the Formation of Inter-Cisternal Continuities Involved in Intra-Golgi Transport

The enzyme phospholipase A2 (cPLA2α) is involved in the formation of intercisternal tubules that mediate transport of proteins within the Golgi complex.


Introduction
After their synthesis in the endoplasmic reticulum (ER), cargo proteins move to the Golgi complex. This unique structure comprises numerous compact stacks of cisternae that are laterally interconnected into the Golgi ''ribbon'' through tubular-reticular networks (''non-compact zones'' [1]). Cargo proteins then traverse the Golgi cisternal subcompartments (where they are glycosylated), and at the trans-Golgi face they are sorted and delivered to their further destinations via large tubular/pleiomorphic carriers [2][3][4][5][6].
Despite significant advances in recent years, important aspects of the organization of intra-Golgi trafficking remain unclear. Three main intra-Golgi transport models have been traditionally considered for higher eukaryotes: trafficking by anterograde vesicles [7], trafficking by compartment maturation-progression [8][9][10][11], and trafficking by diffusion via tubular continuities joining different Golgi cisternae [12][13][14][15][16][17][18]. Among these, the cisternal progression model has recently gained broad consensus, although the nature of the intermediates involved in this mechanism (vesicles or intercisternal tubules) remains unclear. In addition, more recently, a new model has been proposed by which cargo molecules mix rapidly throughout the stack (compatible with cargo diffusion via intercisternal continuities) and then partition into specialized export domains before leaving the Golgi complex [19].
Thus far, intra-Golgi transport models, the last, diffusion via tubular continuities, has received the least attention [12][13][14][15][16][17], and even the very existence of intercisternal tubules has long remained an issue of debate [20]. The main reasons for this uncertainty have probably been the technical difficulties of detecting the convoluted structures of the intercisternal tubules by traditional electron microscopy (EM) and the traffic-related dynamics of these tubules [18,21]. Over the last few years, however, tomography studies have indicated that tubules can be shown to join successive Golgi cisternae in animal cells, and that they form specifically under conditions of active trafficking [17,18,21]. Thus, although intercisternal tubules have still not been completely characterized (for instance, quantitative data on their frequency remain scarce [18,22]), they can be considered as potential players in intra-Golgi trafficking. It is therefore of interest to understand and manipulate their molecular mechanisms of formation.
The formation of an intercisternal tubule is likely to depend on several molecular events, including mechanical deformation by specialized proteins and changes in the distribution/geometry of the lipids within the membrane bilayer [23,24]. These latter can be induced in several ways, including the formation of local spontaneous positive membrane curvature via the generation of lysolipids and fatty acids by PLA 2 activity [24,25]. Indeed, a role for PLA 2 in membrane shaping has been suggested in vitro [26], as well as in a series of in vivo studies that have indicated that chemical blockers of PLA 2 suppress tubule formation in several endomembrane compartments and inhibit the related trafficking steps (see Brown et al. [27] for review).
Here, we show that transport through the Golgi complex coincides with the rapid recruitment of a specific molecular PLA 2 isoform onto Golgi membranes: Group IVA, Ca 2+ -dependent cytosolic PLA 2 (cPLA 2 a). The activity of cPLA 2 a is required for the formation of intercisternal connections in the Golgi. In addition, we show that treatments that inhibit cPLA 2 a and suppress intercisternal tubule formation also block intra-Golgi trafficking of several cargo proteins. These data identify cPLA 2 a as a component of the machinery underlying intercisternal tubular continuities and support a role for these continuities in intra-Golgi transport.

Results
As indicated above, PLA 2 inhibitors have been reported to suppress membranous tubules that extend from different cellular compartments, including the Golgi complex [28,29]. We thus sought to identify the Golgi-associated PLA 2 isoform that might serve as the molecular target of these inhibitors and might be involved in the regulation of Golgi-associated tubular structures. The superfamily of PLA 2 enzymes consists of 15 groups comprising secretory and cytosolic enzymes, with the latter divided into Ca 2+ -sensitive (cPLA 2 s, or Group IV PLA 2 s) and Ca 2+ -insensitive (Group VI, VII, and VIII PLA 2 s) isoforms [30]. Among those that are Ca 2+ sensitive, one of the Group IV isoforms, cPLA 2 a, that is normally cytosolic has been reported to associate preferentially with the Golgi complex upon moderate increases in cytosolic Ca 2+ concentrations [31][32][33][34]. This affinity for Golgi membranes prompted us to examine whether cPLA 2 a might itself be involved in intercisternal tubule formation and intra-Golgi trafficking.
Traffic Induces Binding of cPLA 2 a to the Golgi Complex in a Ca 2+ -Dependent Fashion Initially, we asked whether cPLA 2 a is recruited to the Golgi complex during activation of transport through this organelle. We first generated a procollagen-I (PC-I) traffic pulse in human fibroblasts (HFs) using the 40-32uC transport synchronization protocol [8,35]. During the 40uC block, PC-I was diffusely distributed in the ER and cPLA 2 a was mostly cytosolic ( Figure 1A). After releasing the block, PC-I reached the Golgi complex, where its levels continued to increase for 30 min ( Figure 1B). At the same time, remarkably, cPLA 2 a was partially recruited to the Golgi complex ( Figure 1B). This cPLA 2 a increase on the Golgi complex lasted for at least 30 min. Similar experiments were carried out using the same 40-32uC transport synchronization protocol [35] in other cell types (HeLa and MDCK cells) with the temperature sensitive mutant of the vesicular somatitis virus glycoprotein (VSVG) as cargo, a well-characterized trafficking marker [35,36]. VSVG can be expressed by either transfecting or infecting cells with vesicular somatitis virus (VSV). In both cases, the generation of a traffic pulse induced the recruitment of cPLA 2 a to the Golgi complex ( Figure 1F and 1G), as seen with PC-I. We further examined the distribution of cPLA 2 a in the Golgi by immuno-EM. We fixed VSV-infected cells expressing cPLA 2 a-GFP both during the 40uC block and after releasing the block, and labelled the cells with an anti-GFP antibody (ab). Most of the cPLA 2 a-GFP was dispersed in the cytosol during the 40uC traffic block, with very few gold particles on the ER and the Golgi complex. In contrast, the cells fixed 30 min after releasing the block (i.e., during a traffic wave) showed significant amounts of cPLA 2 a-GFP at the rims of the Golgi cisternae and on rim-associated tubules ( Figure 1H-1J). Some cPLA 2 a-GFP labelling was still seen in the cytosol and at low levels in the ER, but not on other intracellular membranes. Notably, this distribution is consistent with an action of cPLA 2 a in the formation of tubules from cisternal rims (see below).
To verify that this cPLA 2 a recruitment to the Golgi complex was due solely to changes in membrane transport rather than to the temperature changes required for inducing and releasing the trafficking block, we infected HeLa cells with VSV (as above) and then treated cells for 2 h with cycloheximide at 40uC (to deplete the cells of cargo and thus reduce transport [18]). Under these conditions, the temperature shift from 40uC to 32uC did not induce any significant changes in cPLA 2 a localization (unpublished data).
We next investigated whether recruitment of cPLA 2 a to the Golgi complex could be seen also under physiological steady-state trafficking conditions, rather than only during traffic waves. Indeed, HFs showed detectable levels of cPLA 2 a recruitment to the Golgi complex also at steady-state ( Figure 1C). Moreover, when PC-I transport was inhibited using the 40uC temperature block, cPLA 2 a lost its association with the Golgi complex, and when trafficking resumed at 32uC, cPLA 2 a ''rebounded'' to high levels on the Golgi complex ( Figure 1A and 1B). Quantification of cPLA 2 a levels on the Golgi under these conditions is shown in Figure 1E. Recruitment of cPLA 2 a to the Golgi at steady-state was detected also in other cell types (HeLa, MDCK, and HepG2). In particular, in liver hepatoma HepG2 cells, which are professional secretor cells that release up to 20 different soluble serum proteins [37], the Golgi showed relatively high PLA 2 levels ( Figure 1D). Thus, collectively, these data indicate that both ''pulsed'' and steady-state trafficking through the Golgi complex induces the association of cPLA 2 a with the Golgi membranes.
What is the mechanism of cPLA 2 a recruitment to the Golgi complex? The binding of cPLA 2 a to membranes has been shown to be due both to the Ca 2+ -binding C2 domain and to the catalytic domain of this protein. The C2 domain is required for initiating membrane association, in a process that depends exclusively on the intracellular Ca 2+ concentration ([Ca 2+ ] i ) and has thus been proposed to function as a calcium sensor [32], while the catalytic domain prolongs this membrane binding of cPLA 2 a even after the [Ca 2+ ] i has returned to lower levels [32]. We monitored the in vivo dynamics of the calcium-sensor C2 domain of cPLA 2 a fused with GFP (C2-GFP [32]) during the 40-32uC traffic pulse. Figure 1K shows that C2-GFP had a diffuse cytosolic pattern when most of the cargo protein (VSVG-YFP) was arrested in the ER. However, as soon as VSVG started to concentrate within the Golgi complex after the release of the traffic block (after 150-250 s), a significant portion of C2-GFP moved from the cytosol to the Golgi complex ( Figure 1L and 1M; Video S1), and then gradually returned to the cytosol. Under the same conditions, the full-length cPLA 2 a protein shifted to the Golgi complex and remained there for over 30 min (see above). This behaviour of C2-GFP and cPLA 2 a-GFP suggests that the trafficking induces a transient increase in [Ca 2+ ] i , which in turn initiates cPLA 2 a recruitment to the Golgi complex [32], an interaction that would then be prolonged by the catalytic portion . cPLA 2 a is recruited to Golgi membranes upon arrival of cargo from the ER. (A, B) HFs were incubated for 3 h at 40uC (A) and then shifted to 32uC for 30 min in the presence of ascorbic acid (B), or HFs (C) and HepG2 cells (D) were grown under steady-state conditions. After fixing, the cells were double labelled with antibodies against PC-I and cPLA 2 a and examined under the confocal microscope. (A) When PC-I was trapped in the ER after the 40uC block, cPLA 2 a enrichment was not detected in the Golgi area. (B) Arrival of PC-I at the Golgi complex after release of the temperature block induced cPLA 2 a binding to the Golgi complex. (C, D) Under steady-state conditions, the cells were fixed, labelled with antibodies against cPLA 2 a and either PC-I (C) or giantin (D). Confocal microscopy reveals that here some cPLA 2 a can be seen in the Golgi area together with PC-I (C) and giantin (D). (E) Fluorescence intensity of cPLA 2 a in the Golgi region (defined as giantin-positive area) was quantified and normalized to cPLA 2 a intensity in the cytosol of HFs. Plot shows that Golgi/cytosol ratio of cPLA 2 a (mean6SD; n = 20 cells) decreases in cells subjected to 40uC transport block, but increases over the levels detected at steady-state conditions upon block release (at 32uC). (F, G) HeLa cells were infected with VSV and kept at 40uC for 3 h to accumulate VSVG in the ER. The cells were then fixed (F) or shifted to 32uC for 30 min to allow VSVG to exit from the ER (G), and prepared for confocal microscopy. Immunofluorescence labelling shows a diffuse pattern of cPLA 2 a when VSVG is blocked in the ER (F). In contrast, cPLA 2 a undergoes recruitment to the Golgi membranes as soon as VSVG arrives at the Golgi complex from the ER (G). (H) HeLa cells were transfected with cDNA encoding full length cPLA 2 a fused with GFP (cPLA 2 a-GFP), infected with VSVG, and subjected to the 40uC block. The cells were then fixed 30 min after the 40uC block release, to allow VSVG to reach the Golgi complex, and processed for immuno-gold EM with an anti-GFP ab, to reveal cPLA 2 a-GFP localization. After activation of VSVG transport, a cPLA 2 a-GFP signal was detected at the rims of the Golgi cisternae and flanking tubular structures (arrows). (I, J) Quantification of gold labelling (mean6SD; n = 30 stacks) at the Golgi complex (see Materials and Methods) shows most of the cPLA 2 a-GFP is bound to the tubular profiles (I) and the rims of the cisternae (J). (K, L) HeLa cells were co-transfected with the cDNAs encoding VSVG-YFP and the C2 domain of cPLA 2 a fused with GFP (C2-GFP), kept at 40uC for 3 h (K), and observed in vivo under the confocal microscope during VSVG-YFP release from the ER. Images extracted from the time-lapse sequence show that C2-GFP moves from the cytosol (K) to the Golgi membranes (L) when VSVG appears within the Golgi area. (M) HeLa cells were transfected and incubated as above (K, L), and then observed under the confocal microscope and fixed during VSVG release when detectable amounts of C2-GFP started to appear in the perinuclear area. Further staining with antigiantin antibodies reveals overlap of C2-GFP perinuclear signal with giantin labelling (arrows). Scale bar, 7 mm (A-D, F, G), 160 nm (H), 9 mm (K-M). doi:10.1371/journal.pbio.1000194.g001 of cPLA 2 a. While this increase remains to be further defined and clarified, it could be related to the high lumenal Ca 2+ concentrations in the Golgi complex [38] and the localization in the Golgi area of the signalling machinery that is involved in the release of Ca 2+ from intracellular stores [38,39]. Thus, the specific recruitment of cPLA 2 a to the Golgi complex can be explained by a local traffic-induced Ca 2+ release from the Golgi complex and probably, additionally, by an intrinsic affinity of cPLA 2 a for the Golgi (possibly due to its affinity for lipids and proteins that are enriched in Golgi membranes [40][41][42]). Potentially related to this, Ca 2+ released from the Golgi cisternae has been shown to be required for intra-Golgi transport [43].
In summary, traffic moving through the Golgi complex triggers the recruitment of cPLA 2 a to the rims of the Golgi cisternae, possibly through a signalling mechanism [44] that might induce the local Ca 2+ increase that is required for cPLA 2 a binding to membranes [32].
Inhibiting cPLA 2 a Reduces Inter-Cisternal Tubules and Suppresses Intra-Golgi Transport We next examined whether cPLA 2 a is involved in the formation of intercisternal tubules. Tubules can interconnect the stacked Golgi cisternae in at least two ways: tubular non-compact zones can join adjacent stacks ''longitudinally,'' to form the continuous Golgi ribbon [1], and tubules can link the Golgi cisternae in the cis-trans (''vertical'') direction within the same stack, as shown by EM tomography and stereoscopy [12,18,21]. To examine the role of cPLA 2 a here, we sought to inhibit/deplete cPLA 2 a by a variety of approaches, while monitoring the presence/formation of Golgi tubules. HeLa cells were first exposed to siRNAs directed against cPLA 2 a. This resulted in a decrease in cPLA 2 a levels, as evaluated by immunofluorescence (Figure 2A and 2B), western blotting ( Figure 2C), and cPLA 2 a activity assays under basal and elevated Ca 2+ conditions ( Figure 2D). In these cells, growth was partially inhibited (50%-70%) in the last 24 h of exposure to the siRNAs; however, cell viability did not appear to be affected. In these cPLA 2 a-silenced cells, the Golgi ribbon was disassembled into numerous fragments that remained perinuclear ( Figure 2E, asterisks, 2F), as has been previously described upon application of PLA 2 inhibitors [28,29]. EM showed that this was due to a suppression of the longitudinal tubular elements ( Figure 2G-2I) of the non-compact zones [1], which resulted in the breakdown of the Golgi ribbon into separate stacks ( Figure 2H). We then investigated whether this cPLA 2 a deficit also affects vertical intercisternal connections, which are presumably more relevant to cis-trans transport, using EM tomography (which is required to fully reconstruct these tubular structures [18,21]). This showed that tubules connecting different cisternae were present within individual Golgi stacks in these cells ( Figure 3A and 3B, arrows; Video S2; see also below), as has been previously reported for other cell types [18], and that these tubules were almost completely suppressed by RNA interference (RNAi) of cPLA 2 a ( Figure 3C and 3D; Video S3). Other tools that specifically inhibit cPLA 2 a had similar effects: both microinjection of an ab against the catalytic domain of cPLA 2 a (see below) and treatment with the selective inhibitors of cPLA 2 a catalytic activity pyrrophenone and pyrrolidine (not shown) [45,46] induced a significant fragmentation of the Golgi ribbon corresponding to a reduction in the tubular structures at the EM level (not shown). Of note, the tubules in the non-compact zones and the vertical intercisternal continuities always responded in the same way to a cPLA 2 a deficit, suggesting that they both depend on the activity of cPLA 2 a. Instead, other intracellular tubular structures (such as those of endosomal origin, for example) were not affected by cPLA 2 a depletion (not shown).
We also examined the effects of enhancing the levels of cPLA 2 a by its overexpression. Remarkably, this treatment caused an overall increase in the Golgi tubular elements that was sometimes accompanied by a partial loss of the stack structure ( Figure 2J), further supporting the concept that PLA 2 a promotes Golgi tubulation.
A further point was whether the tubulating effects of cPLA 2 a arise from the formation of lysolipids, which can create positive membrane curvature directly [24,27], or whether they are mediated by the formation of arachidonic-acid (AA) metabolites, perhaps via their signalling function. To test the latter possibility, we used chemical inhibitors to block the main metabolic enzymes of AA [47], the cyclooxygenases (using 50 mm indomethacin and 5 mm ibuprofen for 30 min) and the lipoxygenases (using 10 mm ketokenazol for 30 min). These agents had no influence on Golgi tubule formation (unpublished data), consistent with previously reported observations [28]. Moreover, the addition of AA to cPLA 2 a-siRNAs-treated cells did not counteract the ''antitubular'' effects of the cPLA 2 a deficit. Altogether, these data suggest that cPLA 2 a is required to support tubulation and that its effects are mediated via the formation of lysolipids [24,25,27,28].
We thus turned to investigate whether disassembly of intercisternal connections affects transport of cargo proteins across the Golgi complex. To test this, we first suppressed intercisternal connections by silencing cPLA 2 a, and then monitored the effects of this treatment on intra-Golgi transport using the VSVGsynchronized transport assay. We thus infected cPLA 2 a-depleted cells with VSV and accumulated VSVG in the ER at 40uC, and then we released the traffic block at 32uC. VSVG reached the Golgi complex apparently normally, but then it accumulated at the cis-Golgi pole (also inducing a moderate swelling of the cis cisternae) and did not proceed through the Golgi complex ( Figure 4A-4E). Compatible results were obtained using biochemical transport assays (see Materials and Methods) ( Figure 4F and 4G). Rescuing the cPLA 2 a activity in cPLA 2 a-siRNAs-treated cells by microinjection of recombinant cPLA 2 a resulted in the reactivation of VSVG trafficking ( Figure 4H). Further along this line, an arrest of VSVG in the Golgi complex was seen also when the cells were transfected with a dominant-negative cPLA 2 a mutant [48] (Figure 4I), and when cPLA 2 a was acutely inhibited by the microinjection of antibodies against the catalytic portion of cPLA 2 a ( Figure 5) or by specific inhibitors (not shown). Notably, these inhibitory effects were marked but not complete (up to 60%-80%; see Figure 4), possibly explaining the lack of visible toxicity (at least within our experimental time frame).
A possible concern here is that these experiments were carried out using a synchronization protocol that involves the sudden arrival of a large cargo load at the Golgi complex, raising the possibility that cPLA 2 a-dependent connections might have a role only under conditions of Golgi overload. We therefore examined whether the inhibition of cPLA 2 a would cause similar effects during more ''physiological'' non-synchronized trafficking. To this end, we infected cPLA 2 a-silenced cells with VSV and kept them at 32uC, to allow VSVG to continuously exit the ER. As was seen during the 40-32uC synchronized pulse, VSVG reached the Golgi complex but did not traverse it (see below), indicating that the role of cPLA 2 a in intra-Golgi trafficking is not limited to conditions of cargo overload.
The transport of other cargo proteins was also examined using cPLA 2 a silencing. Ablation of cPLA 2 a in HFs resulted in a strong delay of PC-I transport at the level of the Golgi complex (unpublished data). The use of cyclooxygenase and lipooxygenase inhibitors had no effects on trafficking, and AA addition did not reverse the transport block in cPLA 2 a-siRNAs-treated cells, again in parallel with the above effects on Golgi tubules. Thus, the induction of a cPLA 2 a deficit and the attendant disassembly of intercisternal connections appear to be associated with the inhibition of intra-Golgi trafficking of different classes of cargo proteins.
A further question is which type of connection is required for trafficking. Our data indicate that the inhibition of cPLA 2 a disrupts the ''vertical'' (cis-trans) and ''longitudinal'' (inter-stack) tubular connections equally (see Figures 2 and 3). Thus, although it may appear logical to assume that the vertical connections are those relevant for transport, a role for the longitudinal connections cannot be formally excluded by the above data. To resolve this issue, we used a Golgi system that contains vertical, but lacks horizontal, connections, and tested whether transport through such a system is sensitive to the ablation of cPLA 2 a. A Golgi complex with only vertical connections can be generated by nocodazol (NZ) treatment, which results in the fragmentation of the Golgi ribbon into isolated (non-horizontally connected) stacks [18]. Any transport-relevant role of longitudinal connections can be excluded in this system. Control and cPLA 2 a-deficient cells were subjected to 3 h of 30 mM NZ treatment and processed for EM tomography. As expected, vertical connections were revealed in stacks of control cells ( Figure 6A; Video S4), while in silenced cells, these connections were almost absent ( Figure 6B; Video S5). To test the efficiency of transport under these conditions, the cells were infected with VSVG and exposed to NZ during the 40uC block, and then shifted to 32uC to activate transport. Figure 6C shows that in control cells, VSVG moved to the plasma membrane first through GM130-positive and then through TGN46-positive compartments of the Golgi stacks, as has been previously reported [18]. In contrast, in cPLA 2 a-silenced cells, VSVG was retained in the Golgi stacks, where it showed strong overlap with the cis-Golgi  were triple labelled with anti-VSVG, anti-cPLA 2 a, and anti-TGN46 antibodies (B, C) or prepared for immuno-EM using the nanogold protocol (D, E). In control cells, VSVG showed good colocalization with TGN46 in the Golgi area (B, inset); this colocalization was poor in cPLA 2 a-siRNAs-treated cells (C, inset). In control cells, EM showed little VSVG in the cis portion of the stack (7.8%), with most (61%) within trans-Golgi compartments (D, arrows), post-Golgi carriers (D, filled arrowhead), and at the plasma membrane (D, empty arrowhead). In cPLA 2 a-siRNAs-treated cells, most of the VSVG (62%) remained within the swollen cis portion of the stack (E, arrows). (F, G) Control and cPLA 2 a-siRNAs-treated HeLa cells infected with VSV were metabolically labelled with [ 35 S]-methionine and chased at 32uC. At the indicated times, the cells were solubilized and digested with endoglycosidase H (Endo-H), which cleaves sugar chains built on the proteins early in the secretory pathway (i.e., before their processing by the medial Golgi enzyme mannosidase-II, which convert sugars into an Endo-H resistant form). The cell lysates were then separated by SDS-PAGE, and the gels scanned (F). The percentages of the Endo-H-resistant form of VSVG with respect to the total amounts of VSVG were quantified (G) using a FUJIFILM imager. The data indicate that VSVG processing to its Endo-H resistant form (which occurs in the medial Golgi) was reduced when cPLA 2 a was silenced. (H) cPLA 2 a-siRNAs-treated cells were infected with VSV, microinjected with recombinant cPLA 2 a during the 40uC block, and fixed 45 min after the block release at 32uC. The cells were then stained with anti-VSVG and anti-cPLA 2 a antibodies and observed under the confocal microscope. VSVG was delivered to the plasma membrane after cPLA 2 a microinjection (arrows) but remained in the Golgi in noninjected cells (asterisks). (I) HeLa cells were transfected with VSVG-GFP and the dominant-negative cPLA 2 a(1-522) isoform, subjected to a 40uC block, and fixed 45 min after the temperature shift to 32uC. marker (GM130) even 60 min after release from the ER ( Figure 6D). Thus, the disruption of vertical intercisternal bridges by cPLA 2 a silencing inhibited the progression of cargo across NZinduced isolated stacks, which are devoid of horizontal connections. This provides evidence that it is the intercisternal connections of the vertical type that are required for intra-Golgi transport (see Discussion).
The Catalytic Activity of cPLA 2 a Is Required to Support Intra-Golgi Transport Our experiments with specific cPLA 2 chemical inhibitors (see above) taken together with already published observations [27] suggest that the changes in lipid geometry during Golgi tubulation require PLA 2 catalytic activity. Nevertheless, given that the C2 domain of cPLA 2 a inserts deep into the membrane bilayer [49] and could therefore be classified among the membrane-bending protein modules [50], we wanted to determine whether it is indeed the catalytic activity of cPLA 2 a, rather than the insertion of this enzyme into the Golgi membranes, that is responsible for the generation of tubules and transport across the Golgi.
We first tested whether cPLA 2 a maintains its ability to bind to the Golgi complex in the presence of chemical inhibitors that suppress intra-Golgi transport. VSV-infected HeLa cells were treated with pyrrolidine, and the localization of cPLA 2 a was monitored during a VSVG traffic pulse. cPLA 2 a translocated to the Golgi complex to the same extent in control and inhibitortreated cells, but only the latter showed VSVG retention at the Golgi ( Figure 7A and 7B). Second, we examined the effects of two cPLA 2 a mutants, cPLA 2 a (1-522) and cPLA 2 a S228C , which lack PLA 2 catalytic activity and yet show normal binding to membranes [48,51]. cPLA 2 a (1-522) is a deletion mutant that lacks an amino acid (Asp549) essential for enzyme activity [52]. Of note,   . The catalytic activity of cPLA 2 a is required to support intra-Golgi transport. (A, B) HeLa cells were co-transfected with cPLA 2 a-GFP and VSVG-Cherry and exposed to the 40uC block to accumulate VSVG-Cherry within the ER. Pyrrolidine (1 mM; Pyr) was added to the cells (B) 15 min before block release. Then cells were shifted to 32uC in the absence (A) or in the presence (B) of pyrrolidine for 45 min, fixed, and investigated under confocal microscopy. In control cells, VSVG-Cherry was detected at the cell surface (A) and its delivery to the PM was inhibited in pyrrolidinetreated cells (B) ,while cPLA 2 a-GFP recruitment to the Golgi membranes was not affected by inhibitor treatment (compare A and B). (C-G) Control (C) and cPLA 2 a-silenced (D-G) mouse MC3T3 cells were transfected with VSVG-Cherry alone (C, D) or in combination with the following human cPLA 2 a constructs: cPLA 2 a-GFP (E), cPLA 2 a (1-522) (F), cPLA 2 a S228C (G). The cells were subjected to 40uC block, then shifted to 32uC (to activate VSVG-Cherry exit from the ER) for 45 min, fixed, and stained with an anti-cPLA 2 a ab. Confocal microscopy revealed efficient delivery of VSVG-Cherry to the PM in control cells (C), while in silenced cells, most of the VSVG-Cherry remained within the fragmented Golgi complex (D). Transfection of wild-type cPLA 2 a-GFP rescued VSVG-Cherry delivery to the cell surface in cPLA 2 a-silenced cells, while expression of the catalytically inactive mutants cPLA 2 a (1-522) (F) or cPLA 2 a S228C (G) did not reactivate VSVG-Cherry transport. (H) PCCL3 cells were loaded with [ 3 H]-AA (see Materials and Methods). Part of the loaded cells was infected with VSV. Then infected and noninfected (control) cells were exposed to the 40uC block, washed, and shifted to 32uC. Fresh medium was added to the cells for 3 min time intervals and then collected. The cPLA 2 a (1-522) can be produced endogenously by caspase-mediated cleavage at Asp 522 during apoptosis [48], and it can act as a dominant-negative mutant of cPLA 2 a [48]. The cPLA 2 a S228C mutant contains a single point mutation in the active site, again resulting in a complete loss of cPLA 2 a enzymatic activity [51].
Each mutant was transfected into cPLA 2 a-silenced cells and compared to wild-type cPLA 2 a for its ability to rescue the transport block induced by cPLA 2 a ablation and to translocate to the Golgi complex. For the transport experiments, mouse cells were transfected with VSVG carrying a red fluorescent tag (VSVG-Cherry) and subjected to the transport-synchronization protocol. VSVG-Cherry was efficiently delivered to the surface in control cells ( Figure 7C) but not in cPLA 2 a-silenced cells, as expected ( Figure 7D). The cPLA 2 a-silenced cells were then transfected with either the mutants or the wild-type cPLA 2 a. While the latter efficiently rescued VSVG-Cherry transport to the cell surface ( Figure 7E), as expected, neither cPLA 2 a (1-522) nor cPLA 2 a S228C modified the transport block ( Figure 7F and 7G). However, both cPLA 2 a mutants translocated to the Golgi complex as efficiently as wild-type cPLA 2 a ( Figure 7F and 7G). Therefore, these collective results indicate that the catalytic activity of cPLA 2 a, rather than the ability of this enzyme to translocate to Golgi membranes, is required to support transport across the Golgi complex.
Finally, we sought to directly monitor the increase in cPLA 2 a activation that based on the above data should occur during cargo trafficking through the Golgi complex. First, we used a classical PLA 2 activity assay based on the release of [ 3 H]-AA from AAprelabelled cells [53]. A potential problem here is that during cargo trafficking, only a fraction of the total cellular cPLA 2 a is bound to the Golgi complex (which represents, in turn, less than 5% of the cellular membranes). Thus, the increase in AA release over basal values might be very small. To overcome these problems, we used two approaches.
For the first, in addition to HeLa cells, we used a cell line that has been previously characterized in our laboratory to be an efficient AA releaser (PCCL3 cells) [46]. Both cell types were loaded with [ 3 H]-AA, infected with VSV, and subjected to a 40-32uC transport synchronization protocol. When they were shifted from 40uC to the permissive temperature of 32uC, the VSVG expressing PCCL3 cells showed a modest but statistically significant increase in AA release over that seen in control cells ( Figure 7H). This increase coincided in time with VSVG transit through the Golgi complex. HeLa cells showed a trend in the same direction, which, however, was not statistically significant. To overcome this difficulty with HeLa cells, we used here a second approach based on the fluorogenic phosphatidylcholine analogue (bis-BODIPY FL C 11 -PC) as a sensor of local changes in PLA 2 activity [54]. The hydrolysis of this lipid by PLA 2 enzymes results in generation of fluorescent products by fluorescence dequenching [54]. After loading the bis-BODIPY FL C 11 -PC, HeLa cells at steady-state showed a diffuse (ER-like) fluorescent signal in the cell periphery (indicating ongoing PLA 2 activity), and a clearer signal in the perinuclear (Golgi) area (as assessed by confocal microscopy, Figure 7I). In control experiments, the application of the calcium ionophore ionomycin, which strongly stimulates cPLA 2 a [31], markedly increased this signal both at the cell periphery and in the perinuclear region, while the cPLA 2 a inhibitor pyrrolidine reduced overall BODIPY fluorescence ( Figure 7J-7L), indicating that the probe functions as expected under our conditions. Then, the cells were exposed to the 40-32uC VSVG synchronized traffic pulse. During the 40uC block ( Figure 7M) the cells did not show any significant concentrating of fluorescence signal in the Golgi area (consistent with the lack of transport through the Golgi and of cPLA 2 a recruitment). When the traffic block was released, the fluorescence signal increased selectively in the Golgi area, to nearly 2-fold the control ( Figure 7N and 7O).
Thus, taken together, the AA release and the microscopy data suggest that the catalytic activity of cPLA 2 a increases during the passage of cargo through the Golgi complex and that this activity is required for transport across the Golgi stack.

Suppression of cPLA 2 a Activity Does Not Inhibit Golgi Vesicle Formation
A series of control experiments was then carried out. In the first, we asked whether the inhibition of cPLA 2 a activity might have an effect on the Golgi COPI vesicles. We thus inhibited/depleted cells of cPLA 2 a and examined the features of the Golgi vesicles as well as on the dynamics of the COPI machinery in these cells. cPLA 2 a silencing affects neither the number nor the morphology of Golgi vesicles ( Figure 2H and 2I). We also blocked vesicle fusion with their target membranes and monitored the kinetics of vesicle accumulation as an indicator of the rate of vesicle formation. This was achieved via inhibition of aSNAP (one of the main membrane fusion factors) by incubating permeabilized cells with an L294A aSNAP mutant that blocks fusion [55,56]. L294A aSNAP induced an accumulation of Golgi vesicles, as expected. This accumulation was the same in the absence and presence of the cPLA 2 a inhibitor ( Figure S1A-S1E). Also, the machinery responsible for COPI vesicle formation was not affected by the cPLA 2 a inhibitor, as judged by the COPI and ARF1 dynamics of association with Golgi membranes in live cells ( Figure S1F-S1K). Finally, we looked at the effects of cPLA 2 a silencing on a known COPI-vesicledependent trafficking step: the recycling of the KDEL receptor (KDELR) from the Golgi complex to the ER. For this, we used a well-characterized assay based on a KDELR-VSVG chimera [57]. This assay indicated that the KDELR recycles from the Golgi complex to the ER equally well in control and cPLA 2 a-siRNAstreated cells (see below, Figure S2A-S2D), again indicating that the COPI machinery is not inhibited by a cPLA 2 a deficit.
Thus, treatments that block cPLA 2 a suppress the formation of intercisternal tubules while having no apparent inhibitory effects on Golgi-associated COPI vesicles (which presumably rely mostly on coat proteins for their curvature; [23]).

Specificity of the Effects of cPLA 2 a Inhibition on Different Trafficking Steps
We also examined the specificity of the effects of cPLA 2 a on several transport steps. Since cPLA 2 a is recruited selectively to the Golgi complex upon activation of transport and its inactivation selectively suppresses Golgi tubule formation, its effects on trafficking should be restricted to the Golgi complex. In contrast, some relatively nonspecific inhibitors of many PLA 2 isoforms used previously, such as ONO [58], have been reported to block transport at multiple segments of the exocytic and endocytic transport pathways that rely on tubular transport intermediates (reviewed in Brown et al. [27]). To address this apparent discrepancy, we compared the effects of cPLA 2 a RNAi and of pyrrophenone with those reported for ONO [28,29].
First, we examined the effects of silencing cPLA 2 a on several transport steps. These included retrograde transport of the KDELR from the Golgi complex to the ER (as above; Figure  S2A-S2D) plus endocytosis and recycling of transferrin to the plasma membrane ( Figure S2E and S2F), and endocytosis of wheat-germ agglutinin lectin uptake and its transport to the trans-Golgi network (TGN) (Figure S2G and S2H). None of these steps was affected by cPLA 2 a silencing. We also examined the TGN-to-plasma-membrane transport of VSVG after a 20uC transport block (at this temperature, VSVG is arrested and accumulates both in the TGN and in the medial-trans Golgi cisternae [4]). When the 20uC block was released in inhibitor-treated cells, a large fraction of VSVG reached the plasma membrane normally (presumably from the TGN), while the remaining fraction (presumably residing in the medial-trans cisternae) remained trapped in the Golgi complex (not shown), consistent with an effect of cPLA 2 a silencing on intra-Golgi trafficking (see above) and with a lack of effect on TGN-to-plasma-membrane transport. Also, transport from the ER to the Golgi was not affected by cPLA 2 a silencing, as shown above (see Figure 4). Furthermore, the labelling of different proteins that reside in the endocytic compartments ( Figure S2I-S2L) or at the ER/Golgi interface ( Figure S2M-S2P) showed no significant changes after this specific cPLA 2 a silencing. Thus, these data confirm the selectivity of the cPLA 2 a role in intra-Golgi trafficking.
We then compared the above effects with those of ONO, a relatively nonspecific drug that inhibits several PLA 2 isoforms [27]. ONO inhibited the transport of VSVG ( Figure S3A-S3F, S3K, and S3L) and PC-I ( Figure S3G-S3J). Moreover, ONO induced the structural changes expected of PLA 2 inhibition; namely, fragmentation of the Golgi ribbon ( Figure S4A and S4B) and suppression of Golgi-associated tubular elements ( Figure S4C-S4E), including intercisternal connections (as revealed by EM tomography) ( Figure S4F and S4G; Videos S6 and S7). These effects were reversible, as ONO wash-out resulted in a rapid reappearance of bridges connecting cisternae within the stack ( Figure S4H-S4K; Video S8), which coincided with reactivation of transport through the Golgi complex ( Figure S3M-S3O). These effects mimic those due to cPLA 2 a inhibition; however, in addition to these, ONO had effects that were not seen in cPLA 2 a-silenced cells, including the suppression of tubular structures in transferrincontaining early endosomes (not shown). Moreover, ONO has been shown by others to suppress the recycling of transferrin from the endosomes to the plasma-membrane [59], as well as retrograde transport from the Golgi complex back to the ER [60]. It is possible that as an inhibitor of many PLA 2 isoforms, ONO has generalized effects on different membrane tubules because these depend on activities of different PLA 2 enzymes [27], while cPLA 2 a is Golgi specific (see Discussion).

Other PLA 2 Enzymes Support Trafficking in Cells Derived from cPLA 2 a Knock-Out (KO) Mice
An apparent difficulty in this study is that cPLA 2 a KO mice show reduced fertility but, surprisingly, no other major phenotypes [61,62]. We examined whether cells obtained from these mice show transport abnormalities. To test this, immortalized lung fibroblasts (IMLFs) from cPLA 2 a KO mice (IMLFs 2/2 ) or control mice (IMLFs +/+ ) were infected with VSV and subjected to the 40-32uC transport synchronization protocol ( Figure S5). No significant differences in VSVG transport were detected between IMLFs +/+ and IMLFs 2/2 ( Figure S5A, S5C, S5E, S5G), except that while intra-Golgi transport in IMLFs from control mice showed the expected inhibition in the presence of the specific cPLA 2 a inhibitor pyrrophenone, the IMLFs from KO mice were insensitive ( Figure S5B, S5D, S5F, and S5H). This suggests that the latter cells had developed an adaptive mechanism to compensate for the loss of cPLA 2 a (and, incidentally, confirms the specificity of pyrrophenone). We considered the possibility that this mechanism might be based on other PLA 2 s. To verify this, we used siRNAs to screen for the roles of all of the cytosolic PLA 2 proteins from Groups IV, VI, VII, and VIII of the PLA 2 superfamily [30] in VSVG transport in these cPLA 2 a KO cells ( Figure 8). The efficiency of siRNA delivery was checked with fluorescent siGLO. Among these siRNAs, only those silencing the Ca 2+ -independent Group VIIIA (GVIIIA)-PLA 2 inhibited VSVG transport, resulting in accumulation of cargo within the Golgi complex ( Figure 8L and 8P). Strikingly, this PLA 2 isoform has been detected by others at the Golgi membranes and appears to be important for maintenance of the tubular elements of the Golgi complex (W. Brown, personal communication). We also characterized the effects of the GVIIIA-PLA 2 ablation in IMLFs 2/2 and saw clear similarities with the effects of cPLA 2 a silencing in ''normal'' cells. The Golgi remained perinuclear, but the ribbon underwent fragmentation (i.e., it exhibited numerous breaks; Figure 8P), presumably due to the loss of the tubular elements connecting cisternae. VSVG reached the Golgi complex normally, but to a large extent remained trapped in the Golgi complex, where it showed substantial overlap with the cis-Golgi marker GM130 ( Figure 8P) (i.e., it remained in the cis-Golgi) and showed a marked delay of protein progression across the stack. It is thus likely that GVIIIA-PLA 2 is responsible for compensating for the cPLA 2 a deficit in KO cells (or animals) and for supporting transport through the Golgi complex in these cells.
We also asked whether and to what extend GVIIIA-PLA 2 is involved in the regulation of transport in other cell lines with normal levels of cPLA 2 a by evaluating transport both in control IMLFs +/+ and in HeLa cells depleted in GVIIIA-PLA 2 . The silencing of GVIIIA-PLA 2 inhibited VSVG transport less markedly than that of cPLA 2 a, and the double knock-down of these enzymes was slightly more effective than that of cPLA 2 a depletion alone (Figure 9), indicating that GVIIIA-PLA 2 has a subsidiary role in normal cells.
Notably, these experiments were carried out under conditions of both synchronized (high load) and non-synchronized (low load) trafficking with similar results. Also notably, the traffic inhibition was marked (up to 80%), but not complete, even in double KD cells. This could be due either to incomplete silencing or to some further compensatory effects, or also to the presence of redundant transport mechanisms. As a consequence, silenced cells can survive, although their rate of growth was significantly decreased (by 60%-80%) in the last 24 h with the siRNAs, probably reflecting this inhibition of their secretory trafficking.
Altogether, these observations indicate that cPLA 2 a is the main regulator of transport across the Golgi complex under different conditions of cargo load. GVIIIA-PLA 2 has a minor role in intra-Golgi transport when cPLA 2 a is normally expressed, but appears to be able to compensate for the lack of cPLA 2 a to support trafficking in cPLA 2 a KO mice.

Discussion
It has become clear in recent years that cells have at their disposal a vast repertoire of protein-based and lipid-based mechanisms for the bending of their membranes. The former, which have been more extensively characterized, include coat proteins such as clathrin, COPI and COPII complexes, as well as BAR proteins [23,24,50], while the lipids include substrates and products of phospholipases, acyltransferases, phospholipid transfer proteins, and flippases [24,50]. The main finding in this study is that in the Golgi complex, a specific PLA 2 isoform, namely Group IV cPLA 2 a, is required for the formation of the intercisternal tubules that appear to be involved in intra-Golgi trafficking.
The simplest explanation for the role of cPLA 2 a in Golgi tubulation is that this enzyme can induce the rapid accumulation of wedge-like lysolipids at the cisternal rims, resulting in a local increase in spontaneous positive membrane curvature, and hence However, RNAi of GVIIIA-PLA 2 induced accumulation of VSVG within the Golgi complex of IMLFs (L). Notably, VSVG always overlapped strongly with GM130 (see inset in P) in cells incubated with GVIIIA-PLA 2 -specific siRNAs, suggesting that VSVG is trapped within the cis-Golgi compartment. (R) IMLF 2/2 and control lung fibroblasts from mice expressing cPLA 2 a (IMLFs +/+ ) were incubated with siRNAs specific for either cPLA 2 a or GVIIIA-PLA 2 and control siRNAs for 72 h. The cells were then infected with VSV, subjected to the 40uC block with its further release for 60 min and fixed. VSVG was detected at the surface of IMLFs 2/2 and IMLFs +/+ using an ab against its ectodomain. Afterwards, the cells were permeabilized and incubated again with an anti-VSVG ab to reveal the total pool of VSVG within the cell. Then fluorescense intensities of surface and total VSVG were evaluated, expressed as a ratio, and normalized to the control. This quantification reveals that cPLA 2 a silencing significantly inhibits VSVG transport in IMLFs +/+ (but not in IMLFs 2/2 ), while GVIIIA-PLA2 RNAi strongly affects VSVG transport in IMLFs 2/2 and only slightly in IMLFs +/+ . Scale bar, 18 mm (A-O), 7.2 mm (P). doi:10.1371/journal.pbio.1000194.g008 in tubulation (Figure 10). In addition to curvature, the generation of an intercisternal tubular continuity presumably requires the assembly of the fusion machinery at the tip of the budding tubule, for its connecting with a neighbouring cisterna. This, in turn, is likely to involve the formation of an ARF/COPI coat to recruit these fusion proteins into this bud [18,63,64]. Therefore, a simple model that fits our observations is that the cPLA 2 a-generated lysolipids help to create and stabilize the curvature of the necks of COPI buds. This might prevent the fission of such buds into vesicles, allowing these buds to dock and fuse with the next cisternae, creating an intercisternal continuity. It is also possible that the cPLA 2 a-generated lysolipids favour the elongation of buds into tubules, as suggested by the observation that overexpression of cPLA 2 a induces tubulation of the stack structure ( Figure 2J). Notably, for these events to occur, diffusion of the lysolipids away from their site of synthesis should be limited by a diffusion barrier at the Golgi rims (perhaps similar to the molecular fences described at the plasma membrane [65]). This fence-like role could involve the COPI coat that resides at the rims of the cisternae. Future studies will elucidate these further components of the tubulation machinery. As noted, the inhibition of cPLA 2 a and the attendant abrogation of the tubules are associated with the arrest of transit through the Golgi complex, indicating that the cPLA 2 a-dependent intra-Golgi tubules are involved in trafficking. Interestingly, a similar association between intra-Golgi tubules and traffic has been reported for the effects of dicumarol. This drug is Figure 9. VSVG transport efficiency in HeLa cells after depletion of cPLA 2 a and/or GVIIIA-PLA 2 . HeLa cells were incubated with control siRNAs or siRNAs specific for cPLA 2 a, GVIIIA-PLA 2 , or both for 72 h. (A) Cells exposed to these siRNAs were infected with VSV and kept for 6 h at 32uC to allow continuous synthesis and transport of VSVG through the secretory pathway (''steady-state'' conditions). Then the cells were fixed and VSVG at the cell surface was stained with an ab against its ectodomain. Afterwards, the cells were permeabilized and incubated again with an anti-VSVG ab to reveal the total pool of VSVG within the cell. Scale bar, 10 mm. (B) Cells were treated as in A or subjected to the 40uC block and then shifted to the permissive temperature of 32uC for 60 min (''traffic wave'' conditions), fixed, and labelled for surface and total VSVG as described above. Then the fluorescence intensities of surface and total VSVG were evaluated, expressed as a ratio, and normalized to the control. This quantification reveals that under both ''steady-state'' and ''traffic-wave'' conditions, cPLA 2 a silencing inhibits VSVG transport much more strongly than GVIIIA-PLA 2 depletion (see also images in panel A). RNAi of both PLA 2 enzymes, in turn, shows a slightly stronger inhibitory effect on VSVG transport over cPLA 2 a silencing alone. (C) Cells incubated with different siRNAs (as described above) were subjected to western blotting with antibodies against either cPLA 2 a or GVIIIA-PLA 2 or GAPDH, as indicated. doi:10.1371/journal.pbio.1000194.g009 an activator of the fission-inducing protein CtBP1/BARS, and it suppresses horizontal Golgi tubules (and thus, presumably, also the vertical intercisternal tubules, which, however, were not directly examined) and inhibits intra-Golgi trafficking [66]. Thus, suppressing the tubules appears to result in the arrest of trafficking independently of the molecular mechanisms that underlie tubule disruption.
How do cPLA 2 a-dependent tubules support intra-Golgi trafficking, and which are the tubules-longitudinal or vertical-that are involved in intra-Golgi transport? The simplest hypothesis is that Golgi tubules allow the intercisternal diffusion of molecules crucial for trafficking. Both vertical and longitudinal tubular elements are cPLA 2 a-dependent. The former are much less abundant; however, since traffic requires movement along the cis-trans (i.e., vertical) axis, they must be functionally crucial (as also supported by investigations in NZ-treated cells). Moreover, while relatively infrequent and difficult to detect [18,21,22], vertical tubules have the potential to be a very efficient means of intra-Golgi transit due to the great speed of diffusion over short distances (microns in the Golgi complex). Thus, very few vertical tubules per stack can be sufficient to achieve rapid diffusion between the cis and trans compartments of the Golgi complex (see Figure S6). This may be the case even when the connections are fewer than those required for complete intra-stack connectivity. For instance, in the Golgi ribbon, where stacks are connected to each other by horizontal membrane bridges, gaps in one stack might be compensated for by connections in neighbouring stacks (in this sense, also horizontal tubules may contribute to cistrans diffusion; Figure S6). Connections might be transient; here, again, very few connections need to be present in a stack at any given time to support intra-Golgi diffusion if they rapidly open and close between cisternae ( Figure S6). Given the above, the question arises as to the role of intercisternal diffusion in intra-Golgi trafficking of cargo proteins that cross the Golgi by cisternal maturation/progression, such as PC-I and VSVG [35]. The simplest hypothesis is that these tubules allow retrograde movement of Golgi membranes and resident proteins (e.g., enzymes), which is required for maturation [18,21]. According to this scheme (also discussed elsewhere [18]), during cisternal progression, the Golgi enzymes diffuse through the intercisternal continuities and explore the Golgi space, where they partition according to their physicochemical properties into those cisternae that have their most favourable composition. This partitioning is driven by a physicochemical gradient that is maintained across the stack at all times, possibly by the input of compositionally different intermediate compartment membrane into the cis cisternae and of endosomal membrane into the trans-Golgi. Thus, the arrival of intermediate compartment membranes at the cis pole (accompanied by consumption at the trans) promotes both enzyme backflow and cisternal progression (above), resulting in the synchronization of these two events and the maintenance of Golgi polarity. Clearly, this model requires more work to fully test it experimentally, but at this stage, it provides a logical explanation of the observations. At the same time, it should be noted that while our data point to a crucial role for tubules, complementary transport mechanisms cannot be excluded. For instance, if Golgi tubules indeed arise from the stabilization of COPI vesicles, as proposed above, it is possible that trafficking might switch between vesicular [67,68] and connectionmediated modes, with one or the other mechanism prevailing, depending on the cell type and the functional state.
How ''general'' is the requirement for cPLA 2 a-dependent tubules in trafficking? Our findings show that the role of cPLA 2 a, is very specific for intra-Golgi tubular structures and trafficking. However, other (non-Golgi) tubulation-dependent transport steps have been reported to be blocked by PLA 2 inhibitors. For instance, ONO (a rather nonspecific inhibitor of many PLA 2 isoforms) has been shown by us and others to also suppress non-Golgi tubules and to block non-Golgi transport steps that appear to be dependent on tubular intermediates, including endosome-to-plasma-membrane recycling of transferrin [59] and retrograde transport from the Golgi complex to the ER [60]. It is thus possible that these transport steps [59,60] are regulated by other PLA 2 isoforms that are located in the different organelles. For instance, cPLA 2 b and cPLA 2 e have been reported to be located to the early [69] and late [70] endocytic compartments, respectively, and may be involved in the regulation of specific steps of endocytosis that are carried out via tubular carriers and that require PLA 2 activity [27,59]. If this is the case, then the PLA 2 family in general (through different PLA 2 isoforms), rather than cPLA 2 a itself, could underlie a membrane-bending mechanism based on the induction of spontaneous membrane curvature [24] that is involved in tubulation and trafficking at the different levels of cellular membranes in mammals [27]. A further observation that is most probably related to these considerations is that mice knocked out for cPLA 2 a have reduced fertility, but do not show any other major phenotypes [61,62], and that secretory transport and Golgi morphology in IMLFs obtained from KO mice [71] are normal. This appears to be because cPLA 2 a in KO cells is functionally replaced by GVIIIA-PLA 2 (Figure 8). GVIIIA-PLA 2 also partially localizes at the Golgi complex, where it appears to control tubulation processes (W. Brown, personal communication). Whether the mechanisms of action of this enzyme in trafficking are similar to those of cPLA 2 a is unclear at this time. At the mechanistic level, the properties of GVIIIA-PLA 2 are not well defined in vivo. Although GVIIIA-PLA 2 has been shown to have specificity in vitro towards PAF-like lipids, its endogenous substrates remain unknown [72]. Thus, the precise metabolic reactions by which GVIIIA-PLA 2 supports tubulation remain to be determined.
In conclusion, the activity of cPLA 2 a appears to be an important mechanism for the formation of Golgi tubules in mammalian cells [24]. For other tubulation events (in other organelles or cell types), as noted, this role of cPLA 2 a might be taken on by other PLA 2 isoforms or even other phospholipases (yeast). Nevertheless, the identification of cPLA 2 a as a player in Golgi tubulation is a key finding, in that it reveals that generation of lysolipids is an important event in the formation of cellular tubules, and it should open the way towards the unravelling of further components of the tubulation machinery. It is now important to elucidate the mode of action of the intercisternal tubular connections and to define their underlying molecular machinery as well as the relationships of these tubules with other key players in intra-Golgi trafficking [73,74]. The efficiency of cPLA 2 a knock-down was evaluated by either western blotting or immunofluorescense. Infection of cells with VSV was performed as described previously [18]. For VSVG rescue experiments, HeLa cells were incubated with oligonucleotide #4 (directed against amino acids 649-655) and then transfected with mouse full length cPLA 2 a or the human cPLA 2 a (1-522) mutant. Alternatively, mouse MC3T3 cells silenced for cPLA 2 a with pool of three oligonucleotides were transfected with full length human cPLA 2 a or human cPLA 2 a (1-522) or cPLA 2 a S228C mutants.

Cell Microinjection
VSV-infected HFs were microinjected with 4 mg/ml anti-cPLA 2 a ab in the presence of FITC or TRITC dextrans during the course of the 40uC block, using an Eppendorf transjector 5246 (Eppendorf, Milan, Italy). They were then shifted to 32uC and processed for confocal microscopy. Similarly, VSV-infected cPLA 2 a-silenced HeLa cells were injected with 2 mg/ml recombinant cPLA 2 a protein in resque experiments.

Treatment with the aSNAP Mutant
HeLa cells were permeabilized with streptolysin-O and incubated with the recombinant L294A aSNAP (4 mg/ml) protein in the presence of cytosol and an ATP-regenerating system, as described in Kweon et al. [56].

Immunofluorescence, Confocal Microscopy, and Live-Cell Imaging
For immunofluorescence analyses, the cells were fixed with 4% paraformaldehyde and permeabilised in 0.02% saponin, 0.5% BSA, and 50 mM ammonium chloride prior to their incubation with the primary and secondary antibodies of interest. The cells were mounted in mowiol and examined on a Zeiss LSM 510 META confocal microscope (Carl Zeiss, Gottingen, Germany). All confocal images were obtained using the necessary filter sets for GFP, Alexa 488, and Alexa 546 using a Zeiss Plan-Neofluor 636oil immersion objective (NA 1.4), with the pinhole set to one Airey unit. Overlap between different markers was quantified using the ''Co-localization'' module of LSM 3.2 software (Zeiss). Time-lapse images were obtained using a Zeiss LSM510 META confocal microscope. Cells co-expressing GFP and YFP fusion proteins were observed at 32uC, as in the case of VSVG-GFP in 20 mM HEPES buffered DMEM. Temperature was controlled with a Nevtek air stream stage incubator (Burnsville, VA, USA). GFP molecules were excited with the 488 nm line of a krypton-argon laser and imaged using the lscan mode of the META detector. Confocal digital images were collected using a Zeiss Plan-Neofluor 636 oil immersion objective (NA 1.4) and GFP and YFP fluorescence was unmixed using LSM 3.2 software. Selective photobleaching in the regions of interest within the cell was carried out on the Zeiss LSM510 using 100 consecutive scans with a 488 nm laser line at full power. Average fluorescence intensities within regions of interests were quantified using the LSM 3.2 software.

Electron Microscopy
For routine EM and EM tomography, the cells were fixed with 1% glutaraldehyde. EM tomography and morphometry on thin sections and tomograms were all performed as described previously [18]. Vertical intercisternal connections were defined as tubular elements, which link different cisternae within the same stack (or cisternae located at different levels across neighbouring stacks). They show a clearly visible lumen (with minimal diameter of 10 nm at the narrowest point along their length) through at least two subsequent tomogram slices. The number of connections for each stack was calculated in each single tomogram of 200 nm thick single sections and expressed as ''connections per stack in section.'' For pre-embedding gold labelling, cells were fixed with 4% formaldehyde and 0.1% glutaraldehyde, washed, incubated with the primary ab overnight, and then with Nanogold conjugated Fab fragments of the secondary antibodies (Nanoprobes) for 2 h. The Nanogold particles were developed using the Gold-enhance kit. For visualization of the trans-Golgi, NRK cells expressing ST-HRP were fixed as indicated above, washed, and incubated with a mixture of DAB and H 2 O 2 , as previously described [3]. Both HRP and gold-labelled cells were embedded in Epon and sectioned. EM images were acquired from thin sections under a Philips Tecnai-12 electron microscope (Philips, Einhoven, The Netherlands) using an ULTRA VIEW CCD digital camera (Soft Imaging Systems GmbH, Munster, Germany). Quantification of gold particles was carried out using the AnalySIS software (Soft Imaging Systems GmbH, Munster, Germany).

Endo-H Resistance Assay
To determine Endo-H resistance, cells were initially infected with VSV for 1 h and 32uC. The excess virus was then washed off, and the cells were incubated in DMEM containing 10% HEPES for 2 h at 32uC. The cells were then washed three times with PBS and starved in DMEM without methionine and cysteine for 30 min at 32uC. The cells were then pulsed for 5 min with 200 Ci/ml [ 35 S]-methionine in DMEM without methionine and cysteine. To stop the pulse, 10 ml 0.25 m methionine in complete DMEM was added, and the cells incubated for 2 min at 32uC. A subset of the samples was then transferred to ice and this was considered Time 0. Other samples were washed with complete medium and chased for the indicated times at 32uC. At the end of the chase, the cells were washed once in PBS and lysed in 1 ml lysis buffer (70 mM Tris [pH 7.4], 150 mM NaCl, 0.5% SDS, 1% Triton X-100, 1 mM EDTA, and 1 mM PMSF) and incubated for 60-90 min on ice. The lysates were centrifuged, and the supernatants were incubated with an anti-VSVG ab overnight at 4uC. The immune complexes were pulled down using Protein A Sepharose. After washing off the unbound material, the protein was eluted by boiling in Endo-H buffer (0.1 m sodium citrate [pH 5.5], 0.5% SDS, and 1% beta-mercaptoethanol) for 3-4 min. The eluates were then divided into two tubes and one was incubated with 40 U Endo-H overnight. The samples were then boiled in SDS-PAGE sample buffer and resolved on an 8% acrylamide gel, using standard procedures. The gels were then scanned and the percentages of the Endo-H resistant form of VSVG with respect to the total amounts of VSVG were quantified using a FUJIFILM imager or ImageJ software.

Arachidonic-Acid-Release Assays
For the AA-release assays, HeLa or PCCL3 cells were labelled for 18 h in growth medium with 0.1 mCi/ml [ 3 H]-arachidonic acid. The [ 3 H]-arachidonic acid released into the medium was quantified in triplicates, as described previously [46]; the radioactivity released from cells is expressed as percentages of the total incorporated radioactivity. [ 3 H]-arachidonic acid release was quantified over a period of 15 min in experiments with ionomycin stimulation, and over 3 min intervals in VSVG transport experiments.

Visualization of PLA 2 Activity with Fluorogenic Substrate
Visualization of PLA 2 activity in HeLa cells was performed using 1,2-bis-(4,4-difluoro-5,7-dimethyl-4-bora-3a, 4a-diaza-sindacene-3-undecanoyl)-sn-glycero-3-phosphocholine, which is known also as bis-BODIPY FL C 11 -PC. Liposomes containing bis-BODIPY FL C 11 -PC were prepared according to the manufacture instructions and incubated with VSV-infected HeLa cells 1 h prior to the release of the 40uC block. The cells were fixed at the end of the 40uC block or at different time intervals after the temperature shift to 32uC; they were then immunolabelled for VSVG and GM130 and examined under confocal microscopy. The intensity of the fluorescent signals derived from bis-BODIPY FL C 11 -PC hydrolysis was quantified in the Golgi area using the LSM 3.2 software. within the bleached areas at the same rate in both control and ONOtreated cells, indicating that the PLA 2 inhibitor ONO did not affect Arf1 turnover at the Golgi complex. (I-K) HeLa cells were incubated either with 5 mM ONO for 30 min (I) or with 10 mg/ml BFA for 15 min (J); alternatively, they were first treated with 5 mM ONO for 30 min, and then 10 mg/ml BFA was added for an additional 15 min (K). The cells were then fixed and stained for galactosyltransferase (GalT) and b-COP. Alone, the PLA 2 inhibitor ONO did not affect b-COP association with the Golgi complex (I) and did not prevent BFAinduced displacement of b-COP from the Golgi membranes (compare J and K). Interestingly, BFA treatment did not cause disassembly of the Golgi complex in ONO-treated cells, although b-COP was detached from the Golgi membranes (K). Scale bar: 300 nm of the temperature block (I, J). The cells were then incubated without (I) and with (J) 5 mM ONO at 37uC for 45 min in medium with ascorbic acid (to allow PC-I folding and exit from the ER), fixed and double labelled with anti-PC-I and anti-giantin antibodies. While in the control cells PC-I was efficiently exported towards the plasma membrane within transport carriers (I, arrows), ONO treatment induced accumulation of PC-I in the Golgi area (J). (K, L) HFs infected with VSV were metabolically labelled with [ 35 S]-methionine and then chased at 32uC in the absence and presence of 5 mM ONO. At the indicated times, the cells were solubilized and digested with endoglycosidase H (Endo-H), which cleaves sugar chains built on the proteins early in the secretory pathway (i.e., before their processing by medial Golgi enzyme mannosidase-II, which convert sugars into endo-H resistant forms). The cell lysates were then separated by SDS-PAGE, and the gels scanned (K). The percentages of the Endo-H resistant form of VSVG with respect to the total amounts of VSVG were quantified (L) using a FUJIFILM imager. The data indicate that VSVG processing to its Endo-H resistant form (which occurs in the medial Golgi) was reduced when PLA 2 activity was inhibited. (M-O) HFs were infected with VSV, and kept at 40uC for 3 h to accumulate VSVG in the ER. The cells were then incubated in the presence of 5 mM ONO at 32uC for 45 min to allow VSVG exit from the ER. Then the cells were fixed immediately (M) or incubated for an additional 15 min (N) or 45 min (O) in fresh ONO-free medium to allow the recovery of PLA 2 activity. Immunofluorescent labelling of these ONO-treated cells (M) reveals accumulation of VSVG within isolated Golgi stacks labelled with an anti-giantin ab. Shortly after ONO washout (N), partial reconnection of the Golgi stacks into a ribbon (N, arrow) coincided with the reactivation of transport and the appearance of VSVG at the cell surface (N, arrowheads); cells still with disconnected Golgi stacks (N, asterisk) did not show VSVG at the plasma membrane. Complete recovery of PLA 2 activity resulted in the reassembly of the Golgi stacks into a highly connected structure and the exit of the cargo to the plasma membrane (O). Scale bar: 8.  Video S2 EM tomography reveals tubular bridges between cisternae located at the different levels of the Golgi stack in HeLa cells.