Golgi membrane fission requires the CtBP1-S/BARS-induced activation of lysophosphatidic acid acyltransferase δ

Membrane fission is an essential cellular process by which continuous membranes split into separate parts. We have previously identified CtBP1-S/BARS (BARS) as a key component of a protein complex that is required for fission of several endomembranes, including basolateral post-Golgi transport carriers. Assembly of this complex occurs at the Golgi apparatus, where BARS binds to the phosphoinositide kinase PI4KIIIβ through a 14-3-3γ dimer, as well as to ARF and the PKD and PAK kinases. We now report that, when incorporated into this complex, BARS binds to and activates a trans-Golgi lysophosphatidic acid (LPA) acyltransferase type δ (LPAATδ) that converts LPA into phosphatidic acid (PA); and that this reaction is essential for fission of the carriers. LPA and PA have unique biophysical properties, and their interconversion might facilitate the fission process either directly or indirectly (via recruitment of proteins that bind to PA, including BARS itself).

M embrane fission consists of a series of molecular rearrangements by which a tubular or neck-like bilayer joining two membranous compartments undergoes constriction and splits in two parts without leakage of contents. Fission is required for fundamental cellular processes such as the formation of transport vesicles during membrane traffic, organelle partitioning, cell division and in general for the maintenance of the compartmental organization of endomembranes. The mechanisms of fission have been studied intensely during the last decade, and multiple pathways leading to fission have been documented or proposed [1][2][3][4][5] . The best characterized fission processes are based on constriction and destabilization of membranes by the mechano-enzyme dynamin [6][7][8][9][10] , shallow membrane insertion of amphipathic protein domains 2,11,12 and phase separation of lipid domains 3,13 . Nevertheless, key aspects of the lipid rearrangements leading to membrane fission remain elusive, and further analysis is required.
We have identified the protein CtBP1-S/BARS (henceforth, BARS) as a key player in the fission of post-Golgi tubular/ pleiomorphic carriers 5,[14][15][16] , macropinosomes 17,18 , COPIdependent transport vesicles [19][20][21] and in the Golgi ribbon partitioning during mitosis 22,23 . BARS (brefeldin A ADPribosylation substrate) is a member of the C-terminal-binding protein (CtBP) family, which evolutionarily derives from an ancestral dehydrogenase by gene duplication and functional differentiation into proteins involved in transcription, membrane transport, microtubule organization and synaptic transmission 16 . BARS itself is a dual-function protein that controls fission in the cytoplasm and gene transcription in the nucleus 16,24 . Structurally, BARS closely resembles the D-hydroxyacid dehydrogenases 25 and features a 'classical' NAD(H)-binding Rossman fold 26 , which regulates the interconversion of BARS between a monomeric and a dimeric conformation depending on binding to NAD(H) and/or other ligands to the Rossman domain 16,17,[26][27][28] . This conversion is critical for function because BARS can drive fission as a monomer, while it is fission-incompetent as a dimer 17,19,26,28 .
The mechanism of action of BARS in fission has been studied mostly in the context of the process of basolateral post-Golgi carrier formation [14][15][16] . Here BARS assembles into a complex that includes ARF, frequenin (also known as NCS-1), the phosphoinositide kinase PI4KIIIb, 14-3-3g and the kinases PKD and PAK, and functions to couple the budding of carriers with fission 15,16 . To induce fission, BARS must bind to 14-3-3g through a phosphorylated serine (Ser147) in its dimerization surface 15,16 (see also below). This binding locks BARS in its monomeric fission-competent conformation. However, how the 14-3-3g-bound BARS leads to the lipid rearrangements leading to fission remains unclear.
We have previously proposed that BARS-dependent fission involves a lysophosphatidic acid (LPA) acyltransferase (LPAAT) activity, based on the following observations: an LPAAT activity is present in liver Golgi membranes which, on addition of suitable substrates, generates phosphatidic acid (PA) 5 ; PA production induces, or correlates with, the fission of Golgi membranes 5 ; the addition of BARS to these Golgi membranes stimulates both PA production and membrane fission 5 ; treatments that inhibit the formation of monomeric fission-competent BARS inhibit both BARS-stimulated LPAAT activity and membrane fission 5,19,28 . We have also shown that recombinant BARS is associated with a slow LPAAT activity 5 (see below) which is inhibited by anti-BARS antibodies, and proposed that this activity is ascribable to BARS itself 5 . This activity, however, was later shown not to be intrinsic to BARS 29 . The simplest interpretation of these collective findings is that BARS binds to and stimulates an endogenous LPAAT, and that this reaction is involved in membrane fission.
Here we have examined this hypothesis. There are 11 known lysophospholipid acyltransferases (LPLATs), four of which have been cloned and shown to transfer fatty acids from acyl-CoA to the sn-2 position of LPA to form PA (LPAATa, b, g and d), while others have mixed specificities for LPA and glycerol phosphate 30,31 . We find that (i) BARS interacts with LPAAT type d (LPAATd, also known as AGPAT4); (ii) this LPAAT localizes to the trans-Golgi and to post-Golgi carrier precursors at the trans-Golgi Network (TGN); (iii) the catalytic activity of LPAATd is essential for Golgi carrier fission; (iv) BARS potently stimulates LPAATd, and this stimulation is essential for carrier fission; (v) BARS needs to be incorporated into the PI4KIIIb-14-3-3g dimer-BARS complex 15 to stimulate LPAATd and induce fission. BARS thus appears to function as an adaptor/regulator protein that binds to and stimulates LPAATd, to induce LPA-PA conversion and carrier fission. LPA and PA have unique biophysical properties that can markedly affect the organization of lipid bilayers 32 . Their interconversion might facilitate the fission process either directly or through the action of PA-binding proteins, including BARS itself 21 .

Results
LPAATd localizes at the Golgi and binds directly to BARS. To examine whether BARS interacts with an LPAAT, we first sought to identify the LPAATs that localize to the Golgi, as most of the BARS-dependent fission reactions occur in this organelle 15,16 . We Flag-tagged and expressed the available mammalian LPAATs and inspected their localization by immunofluorescence microscopy. LPAATg, LPAATd and LPAATZ localized to both the Golgi and the endoplasmic reticulum (ER) (Fig. 1a), while LPAATb and LPAATe localized to the ER and mitochondria, respectively (Supplementary Fig. 1; see also refs 30,31). We thus examined whether BARS interacts with the Golgi LPAATs by co-expressing BARS with each of these transferases and testing for co-immunoprecipitation of the two proteins. BARS co-precipitated with LPAATg and LPAATd (and vice versa), but not with LPAATZ (Fig. 1b,c). LPAATg has been shown to reside at the cis-Golgi and regulate Golgi structure and retrograde transport to the ER 33,34 , while LPAATd has no known Golgi-related function to date. We asked whether the LPAATd location might be compatible with a role in post-Golgi traffic by immuno-electron microscopy of the Flag-tagged protein (Fig. 1d). Tagged LPAATd localized preferentially in the trans-Golgi and in the TGN (Fig. 1d), and to a lesser extent in the ER in the cell periphery. We also examined the distribution of the endogenous protein using specific antibodies and immunofluorescence microscopy (Fig. 2). The endogenous LPAATd showed a localization that was very similar but not identical to that of the tagged protein, in that it was visualized mostly in the elongating tubules emanating from the TGN (Fig. 2), and gave a weaker signal in the cell periphery compared with the tagged protein. Of note, while this manuscript was being revised, it was reported that LPAATd localizes to the mitochondrial outer membrane in murine cells 35 . This differs from our findings that LPAATd localizes in the Golgi complex. However, cases of dual localization of the same transmembrane protein to the secretory pathway and to outer mitochondrial membrane have been reported several times [36][37][38] and the mechanism of this dual localization is partially understood 39 . We thus considered that LPAATd might localize both to the Golgi and the mitochondria, but that only the Golgi LPAATd can be stained using the low titre antibody conditions optimized for specificity in our study, while LPAATd in mitochondria might require a higher antibody titre for detection. We find that this is indeed the case, in all of the cell types tested ( Supplementary Fig. 2a,b). Moreover, both the Golgi and mitochondrial signals were abolished by LPAATd depletion, that is, are specific ( Supplementary Fig. 2c). We conclude that LPAATd can localize both in the secretory pathway and to a lower extent in mitochondria in most or all cells, though we do not exclude that the LPAATd distribution might be partially cell type-dependent.
We then asked whether BARS binds directly to LPAATd by performing pull-down experiments with the two purified recombinant proteins. After coincubation, recombinant BARS and LPAATd (prepared as described in Methods) co-precipitated efficiently (Fig. 3a). We also tested for binding between BARS and the 152-251aa portion of LPAATd, a cytosolic segment exposed at the protein surface (see models in ref. 30). BARS co-precipitated also with this fragment (Fig. 3a). These data indicate direct binding between the two proteins.
Finally, we examined whether LPAATd binds selectively with the monomeric fission-competent form of BARS 15,16 . As noted, BARS shifts between monomeric and dimeric conformations depending on ligand binding to its Rossman fold. Dimerization is promoted by the binding either to NAD(H) 19,[26][27][28] or to the Brefeldin A-ADP-ribosylated conjugate (BAC), an ADP-ribosylated metabolite of brefeldin A, which locks BARS in the dimeric fission-inactive conformation 28 . Either BARS alone or BAC-bound BARS were added to lysates from cells overexpressing LPAATd. BARS and LPAATd interacted efficiently as judged from co-precipitation (see above), as expected, and this interaction was markedly reduced by the binding of BARS with BAC ( Fig. 3b) (NAD(H) had similar effects, not shown), indicating that LPAATd preferentially binds monomeric BARS.
We conclude that the LPAATd is located in the elongating tubules emanating from the TGN (Fig. 2) and binds directly and selectively the monomeric fission-active form of BARS (Fig. 3).
Bacterial LPAAT competes with LPAATd for BARS binding. As noted, recombinant BARS purified from Escherichia coli is associated with low levels of LPAAT activity. Since this activity belongs to the bacterial enzyme 5,29 (see also Supplementary  Fig. 3), we carried out pull-down experiments to examine whether BARS and E. coli LPAAT might bind to each other and do so in a specific manner. His-tagged E. coli LPAAT (which can be prepared in soluble form) 40 showed strong binding with BARS ( Supplementary Fig. 4a), and this binding was abolished by pre-treatment of BARS with BAC or NAD(H) ( Supplementary  Fig. 4b,c). This indicates that, like the mammalian enzyme (Fig. 3b), the E. coli LPAAT binds selectively to monomeric BARS ( Supplementary Fig. 4b,c). We also asked whether E. coli LPAAT competes with mammalian LPAATd for binding to BARS. This was the case (though the effect was partial; Supplementary  Fig. 4d), suggesting that the mammalian and bacterial LPAATs bind to the same BARS domain, and that the BARS-binding surfaces of the mammalian LPAATd and E. coli LPAAT are conserved. Considering the evolutionary distance between the two LPAATs, this is somewhat surprising. It is conceivable that the evolutionary ancestors of LPAATd and BARS (probably an LPAAT and a D-hydroxyacid dehydrogenase, respectively) 40 , which are fundamental metabolic enzymes, were interactors in an ancient metabolic multi-enzyme complex 41 , and that this interaction was maintained through evolution in different functional contexts.
Regardless, these data explain our previous observations that recombinant BARS from E. coli is associated with an LPAAT activity 5 , which, as we now know, belongs to the E. coli enzyme. They also provides a potentially useful tool (the soluble bacterial enzyme) for in vitro reconstitution of BARS-dependent fission.
LPAATd is required for post-Golgi carrier fission. The TGN localization and the interaction of LPAATd with BARS persuaded us to investigate the role of this LPAAT isoform in the BARS-dependent formation of post-Golgi carriers 14 . LPAATd was silenced with specific small-interfering RNAs (siRNAs), and the formation of carriers, as well as the rate of transport to the plasma membrane was monitored using the temperaturesensitive vesicular stomatitis virus G protein (ts045-VSVG; henceforth, VSVG) 42 as a traffic marker. The exit of VSVG out of the Golgi can be synchronized by accumulating VSVG in the TGN at 20°C and then shifting the temperature to 32°C 14 . The formation of VSVG-containing carriers from the TGN was visualized by immunofluorescence microscopy and quantified by inhibiting the fusion of these carriers with the plasma membrane with tannic acid 14,15,43 , which results in accumulating carriers close to the cell surface. The depletion of LPAATd markedly reduced the formation of VSVG-containing carriers (Fig. 4a), whereas the depletion of LPAATg, which is located at the cis-Golgi and is involved in Golgi-to-ER traffic 33,34 , had no effect (Fig. 4b). LPAATd depletion had no effect on cell viability, growth and morphology for the duration of the experiment and longer.
To determine whether the reduction in VSVG-containing carriers was due to inhibition of carrier budding or of fission, we monitored carrier formation in living cells expressing VSVG-green fluorescent protein (VSVG-GFP). While control cells exhibited dynamic tubules that formed and underwent multiple fission events to generate a large number of free-moving carriers, which then fused with the plasma membrane, LPAATd-depleted cells showed several long (410 mm) tubular extensions that contained VSVG-GFP. These tubules represent carrier precursors that elongate out of the Golgi but do not detach to form mature transport intermediates 14,15 (Supplementary Movies 1 and 2). When they (rarely) did detach, however, they could be seen to move towards, and fuse with, the plasma membrane (Supplementary Movie 2). This phenotype was similar to that induced by expressing BARS dominant-negative mutants or by depleting BARS 14,15 or by depleting 14-3-3g, a key These findings indicate an essential role for LPAATd in the fission of tubular carriers exiting the Golgi complex.
We also determined the role of LPAATd in other traffic steps. We first examined retrograde traffic from the Golgi to the ER (which is known to require LPAATg) 33,34 , by monitoring the well-characterized retrograde transport marker VSVG-KDELR (a fusion of VSVG with the KDEL receptor) 45 . The transport of VSVG-KDELR was not affected by LPAATd depletion (Fig. 4c). Second, since BARS controls the fission of basolateral but not apical carriers 14 , we examined the role of LPAATd in Golgi export of the apical cargo p75 (ref. 46) in comparison with export of the LDL receptor (a basolateral cargo, like VSVG) 47 . Depletion of LPAATd inhibited export of the LDL receptor (Fig. 4d), but not that of p75 (Fig. 4e). Finally, we probed a soluble basolateral cargo, the stably expressed constitutively secreted GFP-tagged variant of the human growth hormone (hGH) 48 . Depletion of LPAATd strongly inhibited export of hGH-FM-GFP from the Golgi to the plasma membrane ( Fig. 4f). Therefore, like BARS, LPAATd appears to be selectively required for the fission of basolateral carriers.
The LPAATd activity is needed for post-Golgi carrier fission. To examine whether the enzymatic activity of LPAATd 49 is required for fission, we first set-up an assay to determine such activity. We prepared and incubated post-nuclear supernatants with the acyl donor [1-14 C]-oleoyl-CoA and the acyl acceptor oleoyl-LPA, with [1-14 C]-PA measured as the reaction product (see Methods and Fig. 5a). These are standard conditions used for LPAAT assays 49,50 , as attempts to purify the LPAAT enzymes result in activity loss 49,50 . Extracts from control cells showed an efficient LPAAT activity, which was suppressed by the general LPAAT inhibitor CI-976 ( Fig. 5b) 34,51 . A difficulty with these extracts is that they contain multiple LPAATs. We therefore designed conditions to selectively determine the LPAATd activity ( Fig. 5a), based on suppressing or overexpressing this enzyme. Extracts from LPAATd-depleted cells (Fig. 5a) or treatment of control extracts with a specific affinity-purified antibody against LPAATd ( Fig. 5b) (see Methods for antibody characterization) showed a reproducibly lower (E25%) activity than in controls ( Fig. 5a,b), confirming that LPAATd is responsible for a fraction of the total LPAAT activity, and in line with the presence of other LPAATs such as the abundant glycerolipid synthetic enzymes LPAATa and LPAATb 30,52 . Extracts from LPAATd-overexpressing cells showed a E40% increase in LPAAT activity over controls ( Fig. 5a-c), and this increase was completely inhibited by antibodies against LPAATd (Fig. 5b), down to the levels found in the absence of LPAATd (Fig. 5a).
LPAATd silencing or overexpression did not affect the cellular levels of other LPAATs ( Supplementary Fig. 6). Similar data were obtained using [1-14 C]-palmitoyl-CoA as acyl donor and arachidonoyl-LPA as acyl acceptor (extracts from LPAATd-overexpressing cells showed a E45% increase in LPAAT activity over controls). We thus define the LPAATddependent activity (or LPAATd activity) as the activity value of LPAATd-overexpressing extracts (as measured using concentrations of substrates below the K m values, see below) minus the activity value of LPAATd-depleted (or antibodytreated) extracts ( Fig. 5a,b; see dashed line and Methods). The maximum reaction rate V max and Michaelis-Menten constant K m of this LPAATd activity were 38 ± 3 nmol min À 1 per mg 2 protein and 58±18 mM, respectively, for oleoyl-CoA, and 38 ± 1 nmol min À 1 per mg 2 protein and 29 ± 1 mM, respectively, for oleoyl-LPA (data are means±s.d. of three independent experiments). These rates are comparable to those reported for LPAATg 53 . Importantly, they are potentially sufficient, depending on substrate availability, to change the PA concentrations in the TGN rapidly and substantially.
Finally, we asked whether the LPAATd catalytic activity is required for post-Golgi carrier fission. We generated a singlepoint mutant (LPAATd H96V ) in the conserved acyltransferase catalytic site of LPAATd (NHX 4 D) 30,54 . Overexpressed LPAATd H96V was devoid of LPAAT activity (Fig. 5c), confirming that LPAATd is a canonical LPAAT family member 30,54 . We then depleted cells of LPAATd, with the consequent inhibition of the post-Golgi transport of VSVG (see Fig. 4a and Supplementary Movie 2) and expressed either a siRNA-resistant variant of LPAATd or of the catalytically dead LPAATd H96V mutant. Only the wild-type LPAATd rescued carrier formation, while LPAATd H96V was completely inactive (Fig. 5d). These data indicate that the catalytic activity of LPAATd, and hence most likely the formation of PA from LPA, is necessary for post-Golgi carrier formation.

Activation of LPAATd by BARS is required for carrier fission.
Since BARS and LPAATd interact directly (Fig. 3a) and are required for post-Golgi carrier fission (see refs 14,15, and Supplementary Movies 2 and 3, respectively), we asked whether BARS regulates the enzymatic activity of LPAATd, and whether this regulation is required for carrier fission. We first silenced BARS and measured the LPAATd activity in cell extracts. BARS depletion abolished the LPAATd activity (Fig. 6a). We then re-expressed BARS in BARS-silenced cells, using a siRNAresistant replacement BARS construct ( Supplementary Fig. 7a). This nearly completely restored the LPAATd activity (Fig. 6a). As a specificity control, the above BARS manipulations did not affect the cellular levels of LPAATd ( Supplementary Fig. 7a) or of other LPAATs ( Supplementary Fig. 7b). These results indicate that LPAATd requires BARS to express its activity.
We then sought to manipulate the BARS levels acutely in in vitro lysates to exclude transcriptional or compensatory effects that might arise in siRNA-depletion experiments 16,24 . We prepared extracts from LPAATd-expressing and BARS-depleted cells, where LPAATd is inactive (Fig. 6a), and added immunopurified BARS to the assay mixture to a final BARS concentration of 5 mg ml À 1 (comparable to the levels of endogenous BARS) 14 . Under these conditions, BARS completely restored the LPAATd-dependent activity (Fig. 6b) (of note, the LPAAT activity associated with immunopurified BARS was quantitatively negligible) 5,29 . We also added BARS to control LPAATd-expressing extracts. This treatment only slightly stimulated the LPAATd-dependent activity (Fig. 6b), suggesting that endogenous BARS is sufficient to activate LPAATd, at least in extracts from quiescent cells (that is, in cells not subjected to a traffic pulse; see below). As a further control, we used extracts from cells depleted of both BARS and LPAATδ− independent activity A n t i-P r e im I g G A n t i-L P A A T δ  LPAATd (Fig. 6b). Here addition of immunopurified BARS had no effect on the LPAAT activity, suggesting that other LPAAT isoforms are not detectably stimulated by BARS, at least under our experimental conditions. Further along this line, we sought to inhibit BARS by adding to the LPAAT assay mixture a characterized neutralizing affinitypurified anti-BARS antibody which, when microinjected into cells, inhibits carrier fission 14,15,55 . This antibody inhibited the LPAATd-dependent activity, while preimmune-IgG addition had no effect (Fig. 6c). Moreover, a pre-treatment of the assay mixture with BAC, which locks BARS in its dimeric fission-incompetent conformation and inhibits the BARS-LPAATd binding (see above and Fig. 3b) 28 , reduced the LPAATd activity (Fig. 6d), supporting the role of BARS in LPAATd activation and indicating that monomeric BARS is required for LPAATd to express its enzymatic activity.
We further examined the relationship between LPAATd activity, BARS and carrier fission by expressing suitable BARS mutants. Previously, we have characterized two single-point mutants, BARS D355A and BARS S147A that have dominantnegative effects on carrier fission in living cells 14,15 . We tested these mutants in the LPAATd activity assay by co-expressing each of them with LPAATd. Both nearly completely inhibited the LPAATd-dependent activity (Fig. 6e), again without affecting LPAATd expression levels ( Supplementary Fig. 7c). As a control, we tested the effects of overexpressing wild-type BARS (of note, wild-type BARS and the dominant-negative mutants showed comparable expression levels in these experiments; see Supplementary Fig. 7c). Overexpressed BARS did not have significant effects on the LPAATd activity in extracts from 'quiescent' cells (Fig. 6e), again suggesting that basal BARS levels are sufficient to support LPAATd activity under these conditions (see above). We noted, however, that BARS is recruited to the Golgi during a traffic pulse, which suggests that active traffic increases the requirement for BARS 15 , possibly to stimulate LPAATd. Indeed, in extracts prepared during a traffic pulse, the overexpression of BARS stimulated the LPAATd activity over control levels (Fig. 6f). Moreover, in these extracts, the expression of the fission-active BARS S147D mutant that mimics the activatory phosphorylation of BARS on Ser147 15,17,18 stimulated the LPAATd activity to an even greater extent (Fig. 6f). These collective data are consistent with the idea that LPAATd is activated during a traffic pulse and that this activation requires higher (local) levels of BARS than those present in quiescent cells. Testing this possibility further, we examined whether the interaction between BARS and LPAATd might be enhanced during a pulse. Co-precipitation experiments indicate that BARS co-precipitates more efficiently with LPAATd under pulse conditions ( Supplementary Fig. 8).
We also tested the role of the PI4KIIIb-14-3-3g dimer-BARS complex in LPAATd activation. As noted, within this complex, 14-3-3g binds to phosphorylated Ser147 in the BARS dimerization interface and is necessary for Golgi carrier fission 15,16 . The LPAATd activity of cell extracts was markedly suppressed by 14-3-3g depletion (Fig. 7a), while depletion of other 14-3-3 isoforms had no effect ( Fig. 7b and see also Supplementary Fig. 9). Moreover, addition to cell extracts of a characterized affinity-purified anti-14-3-3g antibody 15 also suppressed the LPAATd activity (Fig. 7c). These data indicate that 14-3-3g is required for LPAATd activity, presumably because it stabilizes BARS in its monomeric fission-competent conformation.
Finally, we repeated a few of the above experiments using a Golgi-membrane-enriched fraction (Supplementary Fig. 10a). Here the LPAATd overexpression increased LPAAT activity by E60% over controls ( Supplementary Fig. 10b), while LPAATd-depletion or treatment of the membranes with the anti-LPAATd antibody (see also Fig. 5a,b) reduced the activity by E35% compared with controls ( Supplementary Fig. 10b,c), in line with the notion that Golgi membranes are enriched in LPAATd over total cell extracts (compare with Fig. 5a). Also, as seen with total extracts (Fig. 6), the LPAATd activity in Golgi membranes was nearly abolished by BARS depletion or inhibition by the anti-BARS antibody ( Supplementary Fig. 10d).
In summary, a number of experimental results based on BARS silencing or overexpression, or on the use of BARS activatory or dominant-negative mutants and of anti-BARS antibodies and inhibitors, or on manipulations of the BARS-containing complex, converge towards the conclusion that the BARS-induced LPAATd activation controls the fission of Golgi carriers. The stimulation of LPAATd by BARS is potent, and appears to occur rapidly, most likely via the direct physical interaction between BARS and LPAATd during assembly of the BARS protein complex that is required for carrier formation.
The BARS-LPAATd interaction occurs at the Golgi complex. The effects of BARS and LPAATd on the fission of carriers emanating from the Golgi suggest that this BARS-LPAATd interaction occurs at this organelle. To verify this notion, we first re-examined the Golgi localization of BARS and of the other components of the complex, 14-3-3g and PI4KIIIb, focusing, in particular, on the carrier precursors elongating out of the Golgi, where fission takes place. Similar to LPAATd (Fig. 2), these proteins were all seen to localize at the TGN (Fig. 8a) and on the VSVG-containing tubular carriers that form during synchronized exit from the TGN (Fig. 8b). Next, we directly examined whether the interaction between BARS and LPAATd occurs at the Golgi using an approach based on Förster resonance energy transfer (FRET) and fluorescence lifetime imaging microscopy (FLIM). Non-targeting + LPAATδ siRNAs 14-3-3γ + LPAATδ siRNAs 14-3-3ζ + LPAATδ siRNAs 14-3-3ε + LPAATδ siRNAs 14-3-3τ + LPAATδ siRNAs 14-3-3β + LPAATδ A n t i-P r e im I g G A n t i-1 4 -3 -3 γ I g G This technique reveals the co-presence of suitable donor and acceptor fluorophores within the same complex (that is, at a distance of r8 nm). We expressed E 0 GFP-BARS as the FRET donor, and tdTomato-LPAATd as the acceptor, and the donor lifetime was measured to assess FRET under steady state and traffic pulse conditions. A comparison of the donor decay lifetime in control cells (cells not expressing the acceptor) and in cells co-expressing donor and acceptor showed in the latter case a marked reduction of donor lifetime at the Golgi (Fig. 8c,d), indicating that FRET, and hence interaction, between the fluorescent partners takes place (Fig. 8e) (Fig. 8c-e), consistent with the enhanced BARS-LPAATd interaction observed during a pulse (Supplementary Fig. 8).
These results indicate that LPAATd is in complex with BARS in vivo at the Golgi and that it colocalizes at the TGN with BARS, 14-3-3g and PI4KIIIb, in agreement with the BARS-LPAATd co-precipitation and interaction data (Figs 1 and  3 and Supplementary Fig. 8).

Discussion
The main finding of this study is that BARS induces fission of post-Golgi basolateral carriers by interacting directly with, and activating, the enzyme LPAATd, a member of the acyltransferase family, which localizes at the TGN and converts LPA into PA. This indicates that the LPA-PA conversion plays a role in BARS-dependent fission. Notably, the metabolism of PA has been implicated in various aspects of membrane dynamics by other groups, albeit generally based on indirect evidence 18,21,34,[56][57][58] . Based on current and previous results, we propose the following working model for basolateral carrier formation. ARF initiates the process by recruiting and activating PI4KIIIb at the Golgi. This produces a local increase in phosphatidylinositol 4-phosphate (PtdIns4P) 59 , which supports the budding of tubular carrier precursors, most likely through recruitment of PtdIns4P-binding proteins 16 . Carrier precursor budding is assisted by phospholipase A 2 (PLA 2 ) via production of positively curved lysolipids, including LPA, which facilitate the bending of membranes into tubules 2,33,34,60,61 . Concomitantly, BARS assembles with 14-3-3g and PI4KIIIb into a complex where 14-3-3g binds to the BARS dimerization surface, and thus keeps BARS in a monomeric fission-competent conformation 15,16 . BARS then activates LPAATd at the TGN and on the elongating tubular carrier precursors, where the PLA 2 -generated LPA is converted into PA by LPAATd, leading to carrier fission, in a process that might be aided by the property of BARS itself and ARF to insert into lipid membranes 21,62 .
LPA and PA have biophysical properties that may be relevant for fission. PA has a highly charged headgroup close to the glycerol backbone, a tendency to form intramolecular and intermolecular hydrogen bonds and segregate into microdomains. Because PA can act as a binding site for proteins bearing amphipathic/hydrophobic surfaces 56 , which can lead to fission through the shallow hydrophobic insertion mechanism 11 , its above properties might cause PA-binding proteins to insert into membranes at the optimal depth and extent for causing fission. Also of note is that BARS itself 21 has been reported to bind PA. One can thus hypothesize that a runaway process might take place at the tubular precursor surface by which BARS-bound LPAATd converts LPA into PA, causing more BARS molecules to be recruited to the membrane and bind to LPAATd, and hence more LPA to be converted into PA (similar to the positive-feedback Sec7-Arf1 loop whereby Sec7 activates Arf1 to recruit more Arf1 to activate more Sec7 and so on) 63 , until the local concentrations of these molecules become high enough to induce fission. In addition, under physiological conditions LPA and PA have strongly positive and negative spontaneous curvatures, respectively 32 ; thus the conversion from LPA to PA may generate negatively curved PA microdomains within overall positively curved membrane areas, which might lead to membrane destabilization. And, finally, the formation of PA might lead to other potential fission-related lipid-based mechanisms, such as the enzymatic conversion of PA into diacylglycerol 57,64,65 .
The precise role of the lipid and protein players considered in this study must now be defined. Most of the key components involved in BARS-dependent fission are available in pure form, and the possibility to reconstitute this fission pathway in artificial membranes using known components appears to be within reach.
For BARS immunoprecipitation, 500 mg lysate protein from these HeLa cells was brought to 0.2% (v/v) Triton X-100 (final concentration), and incubated with 3 mg anti-BARS polyclonal antibody (overnight, 4°C, shaking) 55 . Then 50 ml protein A Sepharose beads were added for a further 1 h of incubation (4°C, shaking). For LPAAT immunoprecipitation, 1.2 mg lysate protein from the HeLa cells was brought to 0.2% (v/v) Triton X-100 (final concentration), and incubated with 40 ml anti-FLAG M2 affinity-gel-purified antibody (2 h, 4°C, shaking). For BARS immunoprecipitation in the presence of EcLPAAT, 0.8 mg lysate protein from the COS7 cells was brought to 0.2% (v/v) Triton X-100 (final concentration) and incubated with 160 mg purified EcLPAAT (2 h, 4°C, shaking) and then incubated with 3 mg anti-BARS polyclonal antibody (overnight, 4°C, shaking). The immune complexes were collected by centrifugation (500g, 5 min, 4°C). After three washes with lysis buffer with 0.2% Triton X-100, and twice with lysis buffer without Triton X-100, the bound protein was eluted from the protein A Sepharose beads or from anti-FLAG M2 affinity-gel-purified antibody by boiling (10 min) in 100 ml Laemmli buffer. Thirty micrograms of input and 70% of the eluted proteins were separated by 10% SDS-PAGE, and subjected to WB via transfer to nitrocellulose membranes (Millipore).
For histidine pull-down from LPAATd-Flag expressing cell lysate, His-BARS (20 mg) was incubated for 3 h at 37°C with buffer alone (20 mM Tris, pH 7.4, 10 mM sucrose) or with 120 mM HPLC-purified BAC, to allow binding of BAC to His-BARS. The reaction mixture was stopped on ice, and 1 mg lysate protein from LPAATd-Flag expressing cells was incubated with each sample (2 h, 4°C, shaking). Then, 30 ml Ni-NTA agarose beads were added, and the samples were incubated (1 h, 4°C, shaking). The beads were then washed three times with lysis buffer at pH 8.0 supplemented with 0.2% (v/v) Triton X-100 and 20 mM imidazole, by centrifugation (700g, 5 min), and then twice with lysis buffer at pH 8.0 without Triton X-100 but supplemented with 20 mM imidazole. The bound protein was eluted from the Ni-NTA agarose beads by boiling (10 min) in 100 ml Laemmli buffer. Twenty micrograms of input and 50% of the eluted proteins were separated by 10% SDS-PAGE, and subjected to WB via transfer to nitrocellulose membranes.
For GST pull-down with LPAATd, 3 mg of His-BARS were incubated with 1.5 mg GST as control or with 4 mg GST-LPAATd full length or with 3 mg GST-LPAATd (aa 152-251) in GST incubation buffer (50 mM Tris, pH 8.0, 100 mM KCl, 0.2% Triton X-100, protease inhibitors; 2 h, 4°C, shaking). Then, 30 ml glutathione sepharose beads were added for a further incubation (1 h, 4°C, shaking). The beads were then washed five times with GST incubation buffer, by centrifugation (500g, 5 min). The bound protein was eluted from the glutathione sepharose beads with GST elution buffer (100 mM Tris, pH 8.0, 20 mM glutathione, 5 mM dithiothreitol). The 10% and 50% of the unbound and eluted proteins, respectively, were separated by 10% SDS-PAGE, and subjected to WB via transfer to nitrocellulose membranes.
For GST pull-down with EcLPAAT, 3 mg of EcLPAAT were incubated with 3 mg GST as control or with 5 mg GST-BARS in GST incubation buffer (20 mM Tris, pH 8.0, 1 mM EDTA, 0.2% Triton X-100, 100 mM KCl) (overnight, 4°C, shaking). Then, 30 ml glutathione sepharose beads were added for a further incubation (1 h, 4°C, shaking). The beads were then washed five times with GST incubation buffer, by centrifugation (500g, 5 min). The bound protein was eluted from the glutathione sepharose beads with GST elution buffer (100 mM Tris, pH 8.0, 20 mM glutathione, 5 mM dithiothreitol). Fifty per cent of the eluted proteins were separated by 10% SDS-PAGE, and subjected to WB via transfer to nitrocellulose membranes. For the GST pull-down with BAC-treated BARS, 5 mg GST-BARS was initially incubated with buffer alone (20 mM Tris, pH 7.4, 10 mM sucrose) or with 120 mM HPLC-purified BAC 28 (3 h, 37°C), to allow binding of BAC to GST-BARS in the GST incubation buffer. For the GST pull-down with NAD-treated BARS, 5 mg GST-BARS was initially incubated with 100 mM NAD þ in GST incubation buffer (1 h, room temperature). Full-scan images of all WB data are reported in Supplementary Figs 11-13.
Transport protocols and light and wide-field microscopy. For the TGN-exit assay of VSVG, the cells were transfected with VSVG-GFP cDNA or infected with VSV, then incubated for 2 h at 40°C, followed by 2 h at 20°C (with 100 mg ml À 1 cycloheximide) to accumulate VSVG in the Golgi complex. For the TGN-exit assay of p75, the cells were transfected with p75-GFP cDNA, incubated overnight in complete medium and for 2 h at 20°C (with 100 mg ml À 1 cycloheximide). To quantify VSVG-containing carriers, the TGN-exit assay was carried out in the presence of 0.5% tannic acid that inhibits fusion of the mature carriers to the plasma membrane and therefore leads to the accumulation of these carriers near the cell surface. The cells were then fixed and labelled with the P5D4 anti-VSVG antibody and the anti-TGN46 antibody. The fluorescent spots (0.5-2.0 mm) stained by both antibodies (at least 80% of VSVG-containing carriers) were then counted. The p75-GFP-containing carriers were counted as above, although without the co-staining for TGN46. All values are means±s.d. from three independent experiments. For the COPI transport assay, the cells were transfected with VSVG-ts045-KDELR-myc and then equilibrated at the permissive temperature of 32°C for 3 h with the subsequent redistribution of the misfolded VSVG-KDELR chimera into the ER by shifting to the non-permissive temperature of 40°C. The cells were immunostained with anti-VSVG antibody and the percentage of the cells with the ER staining was quantifed 19 . For the transport of the endocytosis-defective LDL-GFP receptor (LDLR Y18A ), the cells were transfected with LDLR Y18A -GFP constructs and then equilibrated 2 h at 20°C to accumulate the protein in the TGN with the subsequent shifting to 37°C to allow for synchronized release of the proteins from the TGN 66 . For the hGH-GFP-FM transport, the HeLa cells stably transfected with hGH-FM-GFP were treated with the DD-solubilizer at 37°C to induce the release of the protein from the ER to the PM 48 . The CI-976 treatment was performed during the VSVG TGN-exit assay, at 50 mM for 10 min, before the 32°C temperature-release block and during the 32°C temperature-release block. The anti-LPAATd antibody (1 mg ml À 1 ) was microinjected 1 h after the beginning of the 20°C incubation in the VSVG transport assay, and after 1 h of recovery the cells were then processed for wide-field microscopy. Wide-field microscopy was performed as described previously 15 , with some modifications. COS7 cells were transfected with siRNAs for LPAATd (Smart Pool, Lipofectamine 2000), and after 48 h the cells were transfected with VSVG-cyan fluorescent protein (VSVG-CFP; overnight, 40°C) and then incubated with 100 mg ml À 1 cycloheximide (3 h, 20°C). The cells were then shifted to 32°C (with continued cycloheximide), and followed by fast video microscopy (see ref 15). For CI-976 treatment, COS7 cells were treated with 50 mM CI-976 for 10 min before the shift to 32°C.
FLIM measurements. The DNA coding for E 0 GFP-BARS and LPAATd-tdTomato were generated using the Restriction Free (RF) cloning procedure 67 using the following two pairs of primers: 5 0 -cacgctggccggagagccttatgtcaggcgtccgacct-3 0 and 5 0 -tttaaacgggccctctagagcactacaactggtcagtcgta-3 0 ; and 5 0 -ctaccggactcagatctcgagatggacc tcgcgggac-3 0 and 5 0 -cccttgctcaccatggtggcctgattcagtttctggttg-3 0 , respectively. BARS and LPAATd genes were amplified as follows: a single denaturation step (95°C, 5 min), followed by 35 cycles PCR (95°C, 1 min; 50°C, 30 s; 72°C, 2 min) and a final elongation step of 5 min (72°C), in a final volume of 50 ml (including 100 ng of template DNA, 0.4 mM of each primer, 200 mM of each dNTPs, 1 Â Pfu buffer and 2 U Pfu Turbo Cx hotstart DNA polymerase, Agilent). Then the amplified genes were purified using the Agilent Strataprep PCR purification kit and used as megaprimer in the second PCR for gene cloning. LPAATd was cloned into tdTomato-N1 vector, fused to the N-terminal of tdTomato sequence, and BARS was cloned into pcDNA3.1 þ vector carrying the E 0 GFP sequence 68 fused to the C-terminal of E 0 GFP sequence. The RF reactions were carried out as follows: a single denaturation step (95°C, 5 min) was performed followed by 20 cycles PCR (95°C, 1 min; 50°C, 30 s; 72°C, 7 min) and a final elongation step of 10 min at 72°C. RF reactions were performed in a final volume of 50 ml (including 50 ng of target DNA, 300 ng of PCR mega-primer products, 200 mM of each dNTPs, 1 Â Pfu buffer and 5 U Pfu Turbo Cx hotstart DNA polymerase). Thirty microlitre of each PCR mix were digested using 20 U DpnI (37°C, 2 h), and 10 ml of each aliquot were used to transform E. coli Top10 strain (Invitrogen). Colony PCR screening were used to search positive clones by using specific primers for BARS and LPAATd. The cloned fragments were then completely sequenced and verified.
COS7 cells were transiently transfected with 0.4 mg E 0 GFP-BARS and 0.4 mg LPAATd-tdTomato using TransIT-LT1 Transfection Reagent, according to manufacturer's instructions. Sixteen hours after transfection, the cells were subjected to the VSVG TGN-exit assay. After 10 min of the 32°C temperaturerelease block, the cells were fixed with 4% (w/v) PFA in 20 mM Hepes pH 7.4, for 10 min at room temperature. The samples were then treated with 100 mM glycine in 20 mM Hepes for 30 min, washed three times with 20 mM Hepes and mounted in glycerol/20 mM Hepes (1:1).
The lifetime values were extracted from the fitting of decay curves obtained from the donor emission (E 0 GFP-BARS) on the Golgi area, and global and pixel kinetic data analyses are computed by a commercially available software package SymPho Time, Version 5.3.2.2 (PicoQuant GmbH, Berlin, Germany). Time-resolved images were acquired by the confocal microscope Leica SP5 II equipped with the White Light Laser (both Leica Microsystems GmbH, Mannheim, Germany) and the TCSPC measurements were performed with the SMD Upgrade Kit (PicoQuant GmbH). The excitation light was pulsed at 80 MHz and the donor fluorophore was excited at 475 nm and acquired from 495 to 530 nm. The measurements were carried out in fixed cells, for ease and speed of acquisition of a statistically reliable number of samples. The donor lifetime values were extracted from the fitting of decay curves obtained from the donor emission (E 0 GFP-BARS) in the Golgi area, and both global and pixel kinetic data analyses were computed by the commercially available software package SymPho Time, Version 5.3.2.2. Changes in donor lifetime due to FRET were assessed by comparing the donor lifetime in cells expressing only the donor with the lifetime in cells co-transfected with donor and acceptor and fixed either under steady-state condition or during a traffic pulse.
Cryo-immunogold electron microscopy. HeLa cells were transiently transfected with 8 mg plasmid DNA encoding Flag-LPAATd for 24 h (using TransIT-LT1) and then fixed with 2% formaldehyde and 0.2% glutaraldehyde in 100 mM Phosphate buffer pH 7.4. The cells were pelleted by centrifugation, embedded in 12% gelatin, cooled on ice and cut into 1-mm 3 cubes at 4°C. The cubes were immersed in 2.3 M sucrose at 4°C overnight, frozen in liquid nitrogen and cut using Leica EM FC7 ultramicrotome 69 . Thin sections (50 nm) were picked up in a mix of 2% methylcellulose and 2.3 M sucrose (1:1) and incubated with the rabbit anti-LPAATd (1:10) and the mouse anti-Golgin-97 (Life Technologies, catalogue number: A21270, 1:50) antibodies. Each of these incubations was followed by incubations with Protein A gold; 10 nm Gold particles for anti-LPAATd; and 15 nm Gold particles for anti-Golgin-97. After labelling, the sections were treated with 1% glutaraldehyde and embedded in methylcellulose uranyl acetate. Electron microscopy images were acquired using FEI Tecnai-12 electron microscope.
Transfections with siRNAs. COS7 and HeLa cells were transfected with a non-targeting siRNA or with 150 nm of a Smart Pool of LPAATd/M-009283 or LPAATg/M-008620 siRNAs, for 72 h (except for BARS and 14-3-3s siRNAs, where 100 nm of a Smart Pool was used for 48 h, as previously described in ref. 15) using Lipofectamine 2000, according to manufacturer's instructions. The efficiency of interference was assessed by WB. The treatment with Smart Pool siRNAs for LPAATg (M-008620) and for LPAATd (M-009283) specifically reduces the endogenous protein levels of LPAATg and LPAATd (by WB), respectively, without affecting the levels of other tested LPAATs (our unpublished data). Alternatively, COS7 cells were transfected with the siRNAs (as above) in combination for the last 16 h with VSVG-CFP, VSVG-GFP, VSVG-ts045-KDELR-myc, LDLR Y18A -GFP or p75-GFP, and then subjected to the specified Golgi-transport assay 14,15,66 . For the rescue experiments, COS7 cells were transfected with siRNAs for LPAATd/ D-009283-03 (5 0 -GCACACGGUUCACGGAGAA-3 0 , Dharmacon) for 48 h, and transfected for a further 24 h (using TransIT-LT1) with Flag-LPAATd wt or Flag-LPAATd H96V (both encode an siRNA-resistant silent mutation), followed by infection with VSV for the TGN-exit assay.
In vitro acyltransferase assay. HeLa cells (1 Â 10 6 ) in 10-cm Petri dishes were transiently transfected with 8 mg plasmid DNA encoding Flag-LPAATd wt or Flag-LPAATd H96V for 48 h (using TransIT-LT1). Alternatively, the HeLa cells were transfected with siRNAs (as above) in combination with Flag-LPAATd wt for 48 h (using Lipofectamine 2000). The cells were washed three times with PBS, harvested as 250 ml per dish in homogenization buffer (100 mM Tris, pH 7.4, 5 mM NaCl, 3 mM MgCl 2 ) supplemented with the protease inhibitor mixture, and homogenized (6 pulses, 30% amplitude; Branson Digital Sonifier). The lysate was centrifuged at 600g for 10 min at 4°C, and the post-nuclear supernatant fraction was used in the acyltransferase assay.
Golgi membranes from HeLa cells treated as described above were obtained as described previously 70 , with some modifications. HeLa cells were washed with PBS, harvested by trypsinization and pelleted (400g, 5 min, 4°C). The cell pellets were resuspended in 1:4 vol. homogenization buffer (250 mM sucrose, 100 mM Tris, pH 7.4) supplemented with the protease inhibitor mixture. The cells were then passed six times each direction (12 passes total) through the Balch homogenizer with a 7.988 mm diameter tungsten-carbide ball bearing (clearance, 12 mm) using constant manual pressure. Following homogenization, the samples (E 850 ml in 0.25 M sucrose, 10 mM Tris-HCl, pH 7.4) were adjusted to 1.4 M sucrose by addition of ice-cold 2.3 M sucrose containing 10 mM Tris-HCl (pH 7.4). Then, 1 mM Na 2 EDTA was added from a 100 mM stock solution, and the samples were vortexed vigorously to ensure uniform mixing. The suspension (E 2.8 ml) was layered onto sucrose gradients (700 ml 2.0 M and 1.4 ml 1.6 M sucrose, from the bottom of the tubes for a swing-out rotor) and covered with 4.2 ml 1.2 M and 2.8 ml 0.8 M sucrose, from the suspension fraction. After centrifugation (90,000g, 2.5 h, 4°C), the Golgi-enriched fraction was collected at the 0.8/1.2 M sucrose interface, aliquoted, frozen in liquid nitrogen and stored at À 80°C.
For CI-976 treatment, the post-nuclear fraction from HeLa cells was incubated with 50 mM CI-976 for 30 min at 25°C, followed by the addition of the LPAAT reaction buffer (as above). For anti-LPAATd or anti-BARS antibody treatment, the post-nuclear or Golgi membrane fraction from HeLa cells was incubated with 100 ng anti-LPAATd or anti-BARS affinity-purified polyclonal antibody for 30 min at 25°C, followed by addition of the LPAAT reaction buffer (as above). For immunopurified BARS treatment, the post-nuclear fraction from HeLa cells was incubated with 500 ng immunoprecipitated BARS purified from rat-brain cytosol with anti-BARS-IgG crosslinked matrice, as described in ref. 15 for 30 min at 25°C, followed by addition of the LPAAT reaction buffer (as above).
In these experiments, LPAATd-dependent activity (or LPAATd activity) is defined as the activity value of LPAATd-overexpressing extracts minus the activity value of LPAATd depleted (or antibody-treated) extracts (see also Fig. 5a). In Figs 5-7 and Supplementary Fig. 10, the LPAATd-independent activity (that is, derived from LPAATd-depleted or antibody-treated extracts) is indicated with a dashed line. The LPAATd-independent activity was reproducibly E50% of the total activity in LPAATd-overexpressing extracts (as evaluated in 430 independent experiments) and E30% of the total activity in purified Golgi membranes from LPAATd-overexpressing cells (as evaluated in 420 independent experiments).
Size-exclusion chromatography. For the fast protein liquid chromatography, 600 mg purified recombinant GST or GST-BARS was applied to a Sephacryl S-200 High Resolution HiPrep 16/60 (Amersham Pharmacia) gel filtration column equilibrated with PBS buffer (4°C; flow rate, 0.3 ml min À 1 ), with 1 ml fractions collected using an AKTA Fast Protein Liquid Chromatography (FPLC) system. The eluted protein was detected by monitoring absorbance at 280 nm, and 10 ml of the collected fractions was separated on 10% SDS-PAGE gels, and analysed by silver staining. Eighty microlitres of each of these fractions was then subjected to the LPAAT assay.
Ion-exchange chromatography. For the fast protein liquid chromatography, 600 mg purified recombinant GST-BARS was applied to a MonoQ column (HR5/5 Pharmacia LKB) equilibrated with buffer A (25 mM Tris, pH 8.00, 50 mM KCl; 4°C; flow rate, 1 ml min À 1 ), with 0.5 ml fractions collected using an AKTA FPLC system. After washing the MonoQ column with buffer A, the protein was eluted with a 20 ml gradient of 0.05-1 M KCl, and 10 ml of the collected fractions were separated on 10% SDS-PAGE gels and analysed by silver staining. Eighty microlitres of each of these fractions was then subjected to the LPAAT assay.
Data availability. The authors declare that the data supporting the findings of this study are available within the article and its Supplementary Information files or are available from the corresponding authors upon request. 70. Balch, W. E., Dunphy, W. G., Braell, W. A. & Rothman, J. E. Reconstitution of the transport of protein between successive compartments of the Golgi measured by the coupled incorporation of N-acetylglucosamine. Cell 39, 405-416 (1984).