Very-long-chain fatty acids are required for cell plate formation during cytokinesis in Arabidopsis thaliana

Very-long-chain fatty acids (VLCFAs) are acyl chains greater than 18 carbons in length, and are now known to be membrane components in all eukaryotic cells (Bach and Faure, 2010; Schneiter et al., 2004; Schneiter et al., 1996). VLCFAs are synthesised by the elongation of C16 and C18 precursor fatty acids by the sequential addition of C2 moieties from malonyl-CoA, a reaction catalysed by the ER membrane-bound microsomal elongase complex. This elongase complex carries out four distinct and sequential enzymatic reactions. Chain elongation is initiated by the condensation of the C16/18-CoA with malonyl-CoA to form 3-keto-acyl-CoA. In animals and yeasts, this reaction is carried out by the keto-acyl-CoA synthase (KCS or condensing enzymes), which is encoded by the multigenic ELO gene family (Paul et al., 2006). In plants, the KCS activity is encoded by two structurally unrelated gene families: in Arabidopsis there is a large family of FAE1-like genes as well four members of the unrelated ELO-like class of condensing enzymes (Dunn et al., 2004). Both forms of KCS generate 3-keto-acyl-CoA, which is then reduced by the keto-acyl-CoA reductase (KCR) to form 3-hydroxy-acyl-CoA, which is in turn dehydrated by the 3-hydroxy-acyl-CoA dehydratase (PAS2) into an enoyl-CoA (Bach et al., 2008). Finally, this enoyl-CoA is reduced by the enoyl-CoA reductase (ECR) to generate an acyl chain extended by two carbons (Nugteren, 1965).

Depending on their chain length and their level of unsaturation, VLCFAs show a wide range of physiological and structural functions, making them crucial for many biological processes such as cell expansion, cell proliferation or differentiation (Bach and Faure, 2010).

VLCFAs are commonly found in neutral lipids such as triacylglycerols (TAGs) or storage lipids in adipose tissue or seeds, but in animals they also have a structural role in skin-barrier function as protective lipids (Westerberg et al., 2004) or epicuticular waxes and suberin, which prevent desiccation and physical damage in plants. Furthermore, VLCFAs are also found in phospholipids and sphingolipids, where they probably determine the properties of cell membranes. Different subcellular membranes have a discrete lipid composition, and even within the same membrane the lipid distribution differs between the two sides of the lipid bilayer and displays different degrees of lateral organization (Lingwood and Simons, 2010; van Meer, 1989). Fatty acyl chain length was also found to be crucial for membrane curvature. In yeast, the isolation of a conditional acyl-CoA carboxylase mutant demonstrated that a reduced amount of 26:0-containing lipids was linked to defects in nuclear pore formation. In particular, C26-phosphatidylinositol was found to be essential for stabilizing highly curved membrane structures (Schneiter et al., 2004). Sphingolipids, sterols and saturated phospholipids also demonstrate domain-forming properties that allow them to trigger the formation of microdomains (so-called ‘lipid rafts’) involved in the sorting of certain plasma membrane proteins, cell stability at the cell surface and endocytosis (Borner et al., 2005; Ikonen, 2001; Mongrand et al., 2004; Wang and Chang, 2002). VLCFAs were recently shown to be necessary for coordinating plasma membrane deformation and forming specialized membrane domains required for actin ring constriction and stability at the cortex (Szafer-Glusman et al., 2008). Interestingly, a loss-of-function mutation in BOND, a Drosophila member of the Elovl KCS family led to cytokinesis defects in spermatocytes. In particular, defects in the synthesis of VLCFAs affected the contraction of the actin ring, as well as the stabilization of the midzone microtubules, resulting in a dramatic block of cleavage furrow ingression in early telophase (Szafer-Glusman et al., 2008).

Importantly, cytokinesis in vascular plants is strikingly different to that in other higher eukaryotes because the separation of the two daughter cells requires a new structure (called the cell plate), the formation of which requires the supply of new material to form plasma membrane and cell wall. In Arabidopsis roots, cell plate formation can be completed in less than 30 minutes and requires the creation of one-third of the original cell surface area (Jürgens and Pacher, 2004). In these conditions, rapid delivery of new material to a precise subcellular location implies specific targeting and rapid vesicular trafficking. Plant cells control cytokinesis by building the phragmoplast, which includes microtubules and actin-based structures that allow unidirectional vesicle delivery and fusion, generating the cell plate. Most of these vesicles are derived from the endoplasmic reticulum (ER) and Golgi complex (Alberts et al., 2002; Reichardt et al., 2007), but endocytosis was also suggested to contribute significantly to the accumulation of plasma membrane proteins in the nascent cell plate (Dhonukshe et al., 2006). Recently, sterol-dependent endocytosis was found to restrict lateral diffusion of the cytokinesis-specific syntaxin KNOLLE and to maintain its correct localization at the plane of division (Boutte et al., 2009).

Although it is agreed that vesicle fusions leading to the tubulovesicular and tubular networks necessary for plant cell plate formation and growth require SNARE proteins and small Rab GTPase (for reviews, see Lipka et al., 2007; Molendijk et al., 2004), the role of endomembrane lipids remains unclear. Previously, PAS2 protein was characterized as the 3-acyl-CoA dehydratase involved in the elongation of the VLCFAs. The absence of PAS2 led to a reduction of VLCFAs in three major lipid classes (triacylglycerol, epicuticular waxes, sphingolipids) resulting in profound developmental defects throughout the plant (Bach et al., 2008). Interestingly, the pas2 mutant displayed abnormal cell division patterns and ectopic expression of cell cycle markers, suggesting that PAS2 activity is involved in cell division (Bellec et al., 2002). However, the precise role of VLCFAs and VLCFA-containing lipids during cell division remained to be determined.

Here, we demonstrate that the pas2 mutant displays reduced root growth that was caused by a delay in cell plate establishment during cytokinesis. Live-cell imaging associated with ultrastructure analysis confirmed that the cell plate structure was altered and that endomembrane dynamics was modified. These defects were correlated with the reduction of very long fatty acyl chains into phosphatidylethanolamine and phosphatidylserine phospholipid pools, suggesting that the acyl chain length in endomembranes are crucial for cell plate formation and cytokinesis.

The pas2-1 mutant was initially characterized by ectopic cell proliferation and competence for differentiated tissue to divide in the shoot (Bellec et al., 2002; Faure et al., 1998), but also by its shorter primary root (supplementary material Fig. S1A). The role of PAS2 during root development was first investigated by analysing its expression of the pPAS2:PAS2-GFP construct. This construct was fully functional because it complemented all pas2-1 developmental defects (supplementary material Fig. S2A). Stable expression of pPAS2:PAS2-GFP in several independent transgenic lines showed that PAS2 was expressed in the root tip (supplementary material Fig. S2B). Primary root length of 5-day-old pas-2-1 root was 50% of that in the wild type (supplementary material Fig. S1A), which could result from altered cell division or elongation. Root growth has the advantage of spatially uncoupling cell division, elongation and differentiation, with a division zone at the tip, followed by an elongation zone that precedes the differentiated part consisting of the rest of the root. We therefore measured the total length and the number of cells of the cell division zone. pas2-1 roots exhibited shorter division zone (supplementary material Fig. S1B), but the cortical cell number was similar to that in the wild type (29.25±2.11 and 31.43±2.20, respectively) (supplementary material Fig. S1C), suggesting a slight reduction in cell size in the mutant. A similar effect could also be seen in the elongation zone, which was slightly shorter in the pas2-1 mutant compared with the wild type (supplementary material Fig. S2C). Nevertheless, the reduction in the cell division and elongation zones could not fully account for the reduced root growth. Alternatively, reduced root growth could be caused by a delay in cell division. To investigate cell division frequency, we monitored the expression of the mitotic cyclin CYCB1 (pCYCB1:db-GUS) in pas2-1 primary root tip. The number of GUS-positive (as an indicator of mitosis) cells was higher in the pas2-1 mutant (13.64±3.20) than in the wild type (8.63±2.25), showing that although root growth is delayed in the mutant, the pas2-1 root has a higher number of dividing cells (supplementary material Fig. S1D).

Because the pas2-1 mutation did not cause an increase in the number of cells in the division zone, the apparent enhanced mitotic activity might result from an early entry into mitosis or extended mitotic phases. We investigated whether the relative distribution of mitotic stages was modified by the pas2-1 mutation. To determine precisely the different phases of mitosis, phragmoplast, cell plate and nuclear DNA were observed in fixed pas2-1 and wild-type roots. The microtubule network was labelled with anti-tubulin antibodies and the cell plate was defined with antibodies against the syntaxin KNOLLE, which specifically labels the cell plate during cytokinesis (Lauber et al., 1997). Chromatin and condensed chromosomes were counterstained with DAPI. In prophase, metaphase or anaphase, the KNOLLE protein was associated with dispersed punctuate structures inside cells (supplementary material Fig. S3A–C). Then, from the early telophase to the end of cytokinesis, KNOLLE was accumulated in the mid plane of the cell to form the cell plate in association with the expanding annular phragmoplast microtubules (supplementary material Fig. S3D–F). In the pas2-1 mutant, microtubules organized in the mitotic spindle and phragmoplast in a very similar pattern to that in the wild type (supplementary material Fig. S3G,H).

The different stages of mitosis were quantified (with the caveat that it was sometimes difficult to discriminate between the late stages, i.e. early telophase, telophase and cytokinesis, in the pas2-1 mutant because the KNOLLE-labelled cell plate was often observed in cells at an undefined mitotic stage, see below). To get a quantitative estimate, all the mitotic stages showing a clear KNOLLE-labelled cell plate were combined into one experimental group spanning the early telophase, telophase or cytokinesis stages (Fig. 1). The relative number of cells at the early stages of mitotis (prophase, metaphase and anaphase) was slightly lower in the pas2-1 mutant compared with the wild type. By contrast, pas2-1 roots showed a high proportion of cells in the late mitotic stages (6.74%) compared with the wild type (4.42%), suggesting that the lack of VLCFAs impaired cell plate progression.

The enhanced incidence of pas2-1 cells in late mitosis could result from delayed establishment of the new cell plate. To investigate the kinetics of cell plate formation, we used the fluorescent lipophilic styryl dye FM4-64, which is rapidly targeted to the emerging cell plate in early telophase (Dettmer et al., 2006).

Real-time kinetics analysis of cell plate formation was carried out by imaging pRabA2a:YFP-RabA2a-expressing roots stained with FM4-64. RabA2a was previously described as a small GTPase that labels both early endosomes in interphasic cells and the cell plate during cytokinesis (Chow et al., 2008). The first steps of cell plate formation consist of very rapid aggregation of RabA2a vesicles into a plate structure, and no difference could be distinguished between wild-type cells and the pas2-1 mutant (Fig. 2A). Once the cell plate is formed, it extends until it reaches the lateral membranes (Fig. 2B). Cell plate expansion kinetics showed that the cell plate required more time to complete in the pas2-1 mutant compared with the wild type (Fig. 2B,C). Cell plate kinetics was measured based on the ratio of cell plate length per cell width of dividing root tip cells (Fig. 2C). In the wild type, the cell plate was nearly completed in 30–40 minutes, whereas the pas2-1 cell plate required more than 120 minutes to establish (Fig. 2B,C). In contrast to the wild-type cell plate, which grew constantly with a velocity of 13 nm/minute, the pas2-1 cell plate completion showed two distinct growth phases: for at least 50 minutes, cell plate extension did not progress significantly, it then resumed growth and gradually completed cytokinesis at a velocity of 5 nm/minute (50% of the wild-type value).

In about 20% of cases, cell plate extension in pas2-1 was eventually stopped before completion, regressing to form heterogeneous aggregates labelled by both YFP–RabA2a and FM4-64 (Fig. 3A,B and supplementary material Fig. S4A–C). In addition, 13% of pas2-1 root cells showed aggregation of YFP–RabA2a, which corresponds to the mitotic index in the mutant (supplementary material Fig. S1D). Aggregates were occasionally directly associated with the extremity of the cell plate, suggesting that they were formed at the point of vesicular fusion (Fig. 3C). This cytokinetic defect in pas2-1 cells was not associated with the aggregation of the prevacuolar marker GFP–RabF2b, suggesting some specificity to early endosome compartments (supplementary material Fig. S5A). We also occasionally observed fragmented cell plates in pas2-1 cells, which eventually regressed, leaving cell wall stubs at the junction sites with the plasma membrane (supplementary material Fig. S4D). Usually, these cell wall stubs were correlated with binucleate or multinucleate cells caused by successive rounds of mitosis and nuclear division without completed cytokinesis. Binucleated cells were never observed in the pas2-1 mutant, probably because VLCFA depletion was not sufficient for complete cell plate arrest. Indeed, pas2-1 cell plate formation could resume and eventually progress to completion (supplementary material Fig. S4E). This observation would also explain the low number of cells with defective cell plates. Indeed, only 4.5% of pas2-1 mitotic cells showed abnormal cell plates. However, inhibition of elongase activity with the ketoacylsynthase inhibitor flufenacet led to a cytokinetic defect and binucleated cells (supplementary material Fig. S5B,C). Flufenacet was previously shown to specifically inhibit several KCS and to lead to post-genital organ fusion, a hallmark of elongase mutants (Trenkamp et al., 2004). Flufenacet treatment of Arabidopsis seedlings showed that VLCFA elongation was indeed inhibited, with a reduction in C22, C24 and C26 fatty acids (supplementary material Fig. S6). Interestingly, C20:1 and C20:2 levels were increased, suggesting that flufenacet is targetting KCS involved in C20+ elongation. As for pas2-1, flufenacet induced YFP–RabA2a aggregation, but also ectopic accumulation in mature membrane (supplementary material Fig. S5C,D). In the conditions of YFP–RabA2a aggregation, no effect of flufenacet was observed on GFP–RabF2b, SNX1–GFP and VHA1a–GFP markers, indicating that prevacuolar and late endosomes compartments, as well as TGN, were not globally modified by VLCFA depletion (supplementary material Fig. S5E).

VLCFA depletion induced by pas2-1 mutant or by flufenacet appeared to perturb preferentially YFP-RabA2a compartments during cytokinesis but also in interphasic cells (Fig. 3C, top and Fig. 4C). To test whether these aggregates labelled by YFP–RabA2a and observed in both dividing and interphasic cells were related to cell plates in regression, we also monitored the distribution of KNOLLE. YFP–RabA2a aggregates were always labelled by KNOLLE during metaphase and anaphase–telophase when the cell plate is normally emerging (Fig. 4A,B). Interestingly, the aggregates labelled with YFP–RabA2a and KNOLLE were either associated with the expanding cell plate or scattered along a pseudo-cell-plate line, suggesting that these aggregates are directly related to the formation of the cell plate (Fig. 4A,B). In contrast to dividing cells, no KNOLLE aggregates could be observed in interphasic cells, suggesting that KNOLLE could still undergo post-mitotic turnover even in aggregates (Fig. 4C). Nonetheless, KNOLLE-labelled ectopic cell plates could occasionally be found in interphasic pas2-1 cells, indicating that misorganized cell plate structure could remain after cell division is completed (Fig. 4D). Interestingly, similar cytokinetic defects with ectopic KNOLLE-labelled cell plate could be observed in the other VLCFA-defective pasticcino mutant pas3-1 (supplementary material Fig. S3I).

To gain insights into the cause of cell plate delay or fragmentation, transmission electron microscopy imaging was carried out on pas2-1 and wild-type root tips. Electron micrographs of pas2-1 dividing cells clearly show that the mutation interferes with several aspects of membrane fusion of cell plate material. In the early stage of cell division, both wild-type and pas2-1 cells showed an accumulation of vesicle fusion in the equatorial plane, forming an electron-dense tubulovesicular network typical of the early events of cell plate formation (data not shown) (Samuels et al., 1995). However, in the subsequent stages, pas2-1 cell plate development showed significant modifications compared with that in the wild type. In control cells, the cell plate changed from a tubulovesicular network into a more open, smooth and straight tubular network (Fig. 5A). At the same time, the lumen of the network became less electron dense, probably because of the accumulation of new cell wall components (callose, cellulose). By contrast, in pas2-1 cells at late telophase, a large gap of cytoplasm between the tubular elements was often observed, even at a stage where both nuclei were recondensed. The conversion of the cell plate from the tubulo-vesicular network into this typical tubular network was also modified (Fig. 5B). The tubular network was thinner than in control cells, exhibited a ‘wavy’ profile, and maintained electron-dense characteristics from the first stage of tubulovesicular network formation. Despite these defects, the pas2-1 cell plate was able to resume growth and reached completion, with electron-dense material that eventually disappeared, similarly to that in control cells. However, the pas2-1 mature cell plate still exhibited this wavy profile (characteristic of an early stage), whereas control cell plates showed a straight structure (Fig. 5C,D). Even after completion, cell plate material still anchored at the plasma membrane was observed, probably as a result of an abortive cell plate (Fig. 5E). The presence of numerous autophagosomal structures in these cells suggests that there is intense membrane-recycling activity to dispose of the ectopic membrane (Fig. 5E). Another unusual feature was the presence of electron-dense vesicular and tubular structures within the cytoplasm (Fig. 5F, arrows). They were often observed in cells having already completed their cytokinetic process and were reminiscent of the tubulovesicular structures that usually form at the initial stage of cell plate development.

The delay in cell plate formation, as well as the aggregation of the cell plate markers, could indicate that vesicular trafficking to the new membrane is modified by the depletion of VLCFAs associated with the pas2-1 mutation. Cell plate formation requires not only the secretory pathway to transport the different cargoes to the newly formed membrane, but also endocytosis to recycle plasma membrane components (Dhonukshe et al., 2006; Reichardt et al., 2007). The involvement of endocytosis in the pas2-1 cytokinesis defect was suggested by the fact that at the end of mitosis, KNOLLE was often found mislocalized in lateral membranes of the dividing cell (Fig. 4E). Similar labelling could also be observed in the other pasticcino mutants (supplementary material Fig. S3J). Interestingly, this labelling of lateral membranes was reminiscent of what was previously observed in the endocytosis mutants arf1 and drp1A or in tyrphostin-A23-treated wild-type cells (Boutte et al., 2006). Sterol depletion in the cpi-1 mutant similarly to VLCFA depletion in pas2-1 cells, also resulted in KNOLLE distribution in lateral membranes (Boutte et al., 2006). To confirm the effect of VLCFA depletion on endocytosis, FM4-64 pulse-chase experiments were carried out in pas2-1 and wild-type root tips. FM4-64 is usually used as an endocytic marker because it internalizes and rapidly labels endosomal compartments and the cell plate (Geldner et al., 2003). After a short labelling, FM4-64 was washed away and its internalization was monitored to determine the endocytic time when clear FM4-64 labelling was visible inside a cortical cell (Fig. 6A). In contrast to the wild type, where the average endocytic time was 5 minutes, pas2-1 cortical cells required about 20 minutes to internalize FM4-64, indicating that endocytosis was strongly delayed in the pas2-1 mutant (Fig. 6B).

To investigate the involvement of endosomal trafficking in the pas2-1 mutant, we used the anterograde transport inhibitor brefeldin A (BFA), which targets the ADP ribosylation factor and guanine nucleotide exchange factors (ARF-GEFs) leading to reversible inhibition of vesicle secretory trafficking (Geldner et al., 2003; Renault et al., 2002). BFA inhibits in particular endosomal recycling to the plasma membrane of several proteins such as PIN FORMED 1 (PIN1) (Steinmann et al., 1999). BFA application results in the accumulation of PIN1 in large endosomal and Golgi aggregates referred as BFA compartments (Boutte et al., 2006). The sensitivity of PIN1 distribution to different concentrations of BFA was investigated in wild-type and pas2-1 primary roots. BFA was applied to wild-type and pas2-1 seedlings for 30 minutes and the occurrence of PIN1–GFP in large BFA compartments was observed in the stele (Fig. 7A). At 50 μM BFA, both wild-type and pas2-1 primary roots showed saturating formation of BFA compartments labelled with PIN1–GFP (Fig. 7B). However, at 25 μM BFA, less than 50% of pas2-1 cells were BFA responsive compared with 75% in the wild type, indicating that PIN1 trafficking is less sensitive to BFA in the pas2-1 background. Similar finding was also observed with YFP–RabA2a (Fig. 7C) indicating that both recycling membrane proteins, similarly to resident early endosomes were more resistant to BFA in the context of pas2-1. However, the Golgi marker QUASIMODO2 (QUA2–GFP), which does not cycle with the plasma membrane, showed similar BFA sensitivity in pas2-1 and wild-type cells (Mouille et al., 2007). VLCFA depletion could modify BFA sensitivity by inhibiting endocytosis or by enhancing trafficking through early endosomal compartments. We investigated the latter effect in the pas2-1 mutant by measuring YFP–RabA2a recovery at the cell plate after photobleaching (supplementary material Fig. S7A). The maximum recovery of YFP–RabA2a fluorescence at the cell plate was reached after about 3 minutes in both wild-type and pas2-1 cells (supplementary material Fig. S7B). No obvious difference could be seen in the kinetic recovery, indicating that VLCFA depletion did not impair YFP–RabA2a trafficking from endosomal pools to the expanding cell plate. Similarly, treatment with flufenacet did not modify the apoplastic distribution of the secreted LeAGP1 marker (Estevez et al., 2006), confirming the minor contribution of VLCFA to the secretory pathway (supplementary material Fig. S7C).

Finally, we investigated whether the different defects in cell plate formation and endomembrane dynamics were correlated with changes in membrane lipid composition. Biological bilayer membranes consist of polar lipids, predominantly glycerolipids associated with sterols and sphingolipids. Among the glycerolipids, the phospholipids are the most representative. It has been postulated that VLCFAs, through their involvement in structural membrane lipids, can influence membrane dynamics (membrane bending and fusion, membrane microdomains) (Schneiter et al., 2004). In a previous study, we reported that the levels of sphingolipids with very long acyl chains were strongly reduced in pas2-1 seedlings compared with levels in the wild type (Bach et al., 2008). Here, we compared the levels of phospholipids in pas2-1 and wild-type cells (Fig. 8, supplementary material Figs S8 and S9). Phosphatidylethanolamine (PE) and phosphatidylserine (PS) are the two phospholipids that contain significant amounts of VLCFAs, and are known to be fusogenic phospholipids (Mima and Wickner, 2009; Moreau et al., 1992). In roots, VLCFAs are almost exclusively present in PE. VLCFA-containing PE levels (PE≥C40) were almost absent in pas2-1 mutant compared to the wild type and a similar effect could be observed in VLCFA-containing PS (supplementary material Fig. S8). The effects of pas2-1 on VLCFA-containing-phospholipids was also confirmed in shoots where the levels of both VLCFA PE and VLCFA PS were strongly reduced by 80% and 60%, respectively (supplementary material Fig. S9). Interestingly, in both roots and shoots, the reduction in VLCFA phospholipids was correlated with enhanced levels of long acyl chain PE and PS. A similar compensatory effect was also found for sterols. Total sterols represented 12.9±0.4% of total lipids (n=3) in pas2-1 seedlings compared with 8.8±0.3% (n=4) for the wild type, indicating that the reduction in VLCFAs is correlated with enhanced sterol levels.

Complete loss of PAS2 function leads to embryo lethality, demonstrating that very long acyl chains are essential for cell viability (Bach et al., 2008). However, weak alleles partially impairing PAS2 function still significantly reduced accumulation of VLCFAs and caused strong developmental defects (Bach et al., 2008; Bellec et al., 2002). We found that the pas2-1 reduction of VLCFAs levels led to specific accumulation of cells in late mitosis with delayed, unfinished or abnormal cell plates. When so delayed, cell plate expansion later resumed at a velocity lower than observed in the wild type. However, when cell plate expansion was arrested for more than 1 hour, it eventually regressed and collapsed, resulting in aggregation of specific cell plate proteins and endocytic markers. The direct involvement of VLCFAs in cell plate formation was confirmed by the presence of defective cytokinesis in the pasticcino1 and pasticcino3 mutants, both of which were recently characterized as VLCFA-defective mutants impaired in protein trafficking to the plasma membrane (Roudier et al., 2010). Moreover, cell plate defects and specific YFP–RabA2a aggregation was also confirmed in cells in which VLCFA synthesis was chemically inhibited.

Cell plate formation is a multistep process that starts during late anaphase and is completed by late telophase (Samuels et al., 1995; Van Damme et al., 2008). Reduction of VLCFA did not affect the targeting of vesicles to the equatorial plane or the first fusion events, which typically take place shortly after to generate the tubulo-vesicular network. However, several observations suggest that VLCFAs are essential later for cell plate expansion and the synthesis of a new plasma membrane and cell wall. First, pas2-1 cell plates combined a tubulo-vesicular network typical of the initial stage and an open tubular network typical of the telophase stage, indicating a delay of vesicle fusion into tubules. Similar structures have already been described as a hallmark of the degradation of abnormal cell plates (Hepler and Bonsignore, 1990). Second, the presence of wavy profiles in almost completed cell plates that were not always positioned in the expected perpendicular plane is very unusual because such an undulating pattern is typical of early phases of cell plate assembly. Flattening and stiffening seem to occur when cellulose starts to be deposited during the late tubular network stage (Samuels et al., 1995). Finally, the occurrence of fragmented cell plates and related materials in association with autophagic structures indicates that cell plate formation was aborted and the remaining structures recycled. Once the cell plate is formed and anchored, the maturation phase takes place with additional fusion events to close the fenestrate but also requires endocytosis to redefine membrane polarity and to remove excess cellulose and callose (Samuels et al., 1995; Segui-Simarro et al., 2004). Callose was postulated to mechanically stabilize the membrane networks, eventually flattening the structure into a plate-like shape (Samuels et al., 1995). The essential role of callose during cytokinesis was recently demonstrated with the functional analysis of the callose synthase GSL8 (Chen et al., 2009), where reduction of callose deposition at the cell plate associates with defective cell plate and the presence of cell wall stubs. Altogether, these results suggest that VLCFAs could be directly involved in the dynamics of cell plate expansion by providing the appropriate lipid environment for membrane fusion and recycling.

Mutation in PAS2 led to the reduction of VLCFAs, which are themselves components of several classes of lipids including cuticular waxes and triacylgycerols, and also of sphingolipids and phospholipids (Bach et al., 2008). Interestingly, perturbations of the levels of VLCFAs in glycerolipids and sphingolipids led to curved thylakoid membranes and to defective endocytic membrane traffic, respectively (Millar et al., 1998; Zheng et al., 2005). The root-specific lipid suberin, a VLCFA-containing extracellular biopolymer restricted to perivascular cells (Franke and Schreiber, 2007), has also been shown to have a role in root development. Moreover, the double mutant kcs20,daisy-1 and also the kcr1 RNAi lines showed specific reduction of VLCFAs in suberin and they exhibited reduced root growth and an abnormal root phenotype associated with an inhibition of lateral root initiation and root hair elongation (Beaudoin et al., 2009; Lee et al., 2009). None of these phenotypes was observed in the pas2-1 mutant, suggesting that cytokinesis defects observed in the root are not caused by defective suberin synthesis but most probably by the reduction of VLCFAs in phospholipids and sphingolipids. The involvement of VLCFAs in cytokinesis was confirmed by the occurrence of a defective cell plate, ectopic KNOLLE labeling and YFP–RabA2a aggregation in the other pas mutants, and also in cells treated with the specific elongase inhibitor flufenacet. Polar VLCFA-containing lipids are major constituents of plasma membrane (along with sterols). Previous studies of the cyclopropylsterol isomerase cpi1-1 mutant demonstrated that sterol membrane composition was essential for differential endocytosis during cell plate maturation (Boutte et al., 2009; Men et al., 2008). The pas2-1 mutation caused cytokinesis defects resembling those of cpi1-1 mutant because it too exhibited unfused KNOLLE-positive cell-plate-like structures and KNOLLE mislocalization on the lateral membrane of dividing cells. However, sterol levels were not reduced in pas2-1 cells, but instead were elevated, indicating that pas2-1 cell plate defects were not caused by sterol depletion, and that VLCFA and sterols underpin similar cellular processes. Such a compensatory increase of sterol levels was also observed in the VLCFA-depleted pas1 mutant and also in sphingolipid-deficient yeast mutants (Guan et al., 2009; Roudier et al., 2010).

Although it is clear that the secretory pathway is strongly involved in cell plate formation, the role of endocytosis remains a source of debate. Indeed, it has been reported that the styryl dye FM4-64 and fluid-phase markers such as Alexa Fluor 633 rapidly label the forming cell plate within minutes (Dhonukshe et al., 2006). The discovery of a BFA-resistant ARF-GEF GNOM-LIKE1 (GNL1) would explain the insensivity of the cell plate to BFA (Reichardt et al., 2007) because the gnl1 mutant seedlings treated with BFA (mimicking the gnl1,gn double mutant), showed KNOLLE blocked in the ER and cytokinesis defects resembling those in KNOLLE mutants (Teh and Moore, 2007). FM4-64 pulse-chase experiments in the pas2-1 mutant support the role of VLCFA during endocytosis at least in interphasic cells. Reduced endocytosis could also explain the relative insensitivity of PIN1–GFP and YFP–RabA2^a to BFA in pas2-1 cells. Endocytosis inhibition in pas2-1 would indeed reduce PIN1 recycling throughout the endosomal compartments and therefore could at least partially explain its reduced accumulation upon BFA inhibition. Finally, the pattern of KNOLLE localization in pas2-1 argues in favour of defective endocytosis during late cytokinesis. Indeed, KNOLLE is normally removed from the cell plate by endocytosis and targeted into multivesicular bodies or prevacuolar compartments in late mitotic cells (Dhonukshe et al., 2006; Reichardt et al., 2007; Teh and Moore, 2007). The finding that KNOLLE was retained in the ectopic cell plate in interphase cells and that it was mislocalized on lateral membrane in dividing cells are clear signatures of defective endocytosis (Boutte et al., 2009).

Most of the cytokinesis defects associated with PAS2-dependent VLCFA synthesis involved membrane dynamics associated with (most probably) endocytosis. However, the secretory pathway would be more important during cell plate expansion than initiation, as suggested by normal recovery of YFP–RabA2a after photobleaching in expanding pas2-1 cell plate. The presence of two growth phases in pas2-1 cell plate expansion could also be explained by the existence of two different processes: the first stages of cell plate emergence might require the provision of VLCFA-containing lipids provided by the secretory pathway, and later, the cell plate expansion would require mainly endocytosis of pre-existing plasma membrane material.

The most straightforward explanation would be that VLCFAs present in membrane lipids (phospholipids and sphingolipids) are necessary for vesicle transport, fusion and/or budding, as a consequence of their biophysical properties. In yeast, acyl chain length was found to be of crucial importance for such functions. In particular, C26-VLCFAs have been demonstrated to be necessary in protein trafficking and protein stability (Eisenkolb et al., 2002). A connection between VLCFA synthesis and membrane biophysical properties was provided by the observation that the conditional yeast acetyl-CoA carboxylase mutant acc1^ts,mtr7 was affected in the structure and the function of the nuclear envelope (Al-Feel et al., 2003). Crucially, the C26 phosphatidylinositol (but not sphingolipids or GPI anchors with similar VLCFAs) was shown to be critical for membrane bending, probably by stabilizing the highly negative curvature of the membrane (Schneiter et al., 2004). Similar findings were observed in animals, where membrane curvature of the plasma membrane during furrow ingression is dependent on the VLCFA content (Szafer-Glusman et al., 2008).

Finally, our data demonstrated that VCLFAs are essential for plant cell division by controlling endomembrane dynamics. Although the exact roles of the different VLCFA-containing lipid classes in the formation of the new cell plate remain to be determined, the availability of mutants in the different biosynthetic pathways and the development of specific subcellular markers will provide the appropriate tools to further dissect the structural role of these lipids during cytokinesis in plants.

Arabidopsis thaliana Columbia-0 accession (Col0), pas1-2, pas2-1 and pas3-1 heterozygous mutants (EMS alleles in Col0 background) were used for this study. The pas2-1 heterozygous line carrying the GUS fusion with cyclin B1 and its destruction box (db) pCYCB1:db-GUS (Doerner et al., 1996) was described previously (Harrar et al., 2003). The pPIN1:PIN1-GFP, pRab-A2:YFP-RabA2 and pSNX1:SNX1-GFP, p35S:GFP-RabF2b constructs (Vernoux et al., 2000; Chow et al., 2008) were crossed into pas2-1 mutant. pPAS2:PAS2-GFP construct (Bach et al., 2008) and were stably transferred into pas2-1 heterozygous line by Agrobacterium tumefaciens flower dip transformation. Representative lines were selected from several independent transformants for further analyses. Plants were grown in greenhouse on soil (Tref Substrates, Rotterdam, The Netherlands) and watered with Plant-Prod nutritive solution (Fertil, Boulogne Billancourt, France). Seedlings were germinated on vertically oriented Arabidopsis agar medium (Estelle and Sommerville, 1987) and grown under constant temperature 18°C, 16 hour light, 8 hour dark cycle and with 60% hygrometry. Analyses were carried out on 5-day-old seedlings. Statistical analysis for supplementary material Fig. S1 and Figs 1 and 2 was carried out by applying Student's t-test with P≤0.05.

GUS staining was carried out as described previously (Harrar et al., 2003). Analysis of the cell plate formation was performed on seedlings incubated for 5 minutes in 8 μM FM4-64 (Invitrogen) and immediately mounted on slides. To observe FM4-64 internalization, 5-day-old seedlings were stained in 8 μM FM-64 for 3 minutes and washed in water before imaging. BFA (Sigma) was applied to seedlings to final concentrations of 25 μM and 50 μM for 30 minutes. Treated seedlings were mounted with FM4-64 and imaged for 30 minutes. Seeds of fluorescent marker lines were germinated on Arabidopsis medium supplemented with 75 nM Flufenacet (Sigma) and 7-day-old roots were imaged.

Whole-mount preparations were carried out according to published results (Friml et al., 2003) and modified according to Bolte and colleagues (Bolte et al., 2007). Seedlings were mounted on slides with one drop of anti-fading solution (Vectashield) with DAPI (100 μM; Roche) and coverslips were placed on top. Slides were stored 3 days at 4°C and then visualized by confocal microscopy. The antibodies and dilutions used are listed below: rabbit anti-KNOLLE, 1:2000; mouse anti-α-tubulin (Invitrogen), 1:500. Alexa Fluor 488 goat anti-rabbit, Alexa Fluor 568 goat anti-mouse, Alexa Fluor 568 goat anti-rabbit, secondary antibodies were diluted at 1:500.

Root-tip cells were imaged with a Zeiss LSM710 confocal microscope using a 405 nm diode laser line exciting DAPI, a 488 nm argon laser line exciting Alexa Fluor 488 and a 561 nm diode laser line exciting Alexa Fluor 568. Fluorescence emission was detected between 410 and 480 nm for DAPI, 495–530 nm for Alexa Fluor 488 and 565–600 nm for Alexa Fluor 568. In multi-labelling studies, detection was performed in a sequential line-scanning mode with a line average of eight. Images were analysed using ZEN (Zeiss) and IMAGE J software.

Live-cell analysis of root-tip seedlings was performed on the Zeiss LSM710 confocal microscope. Fluorescence was recorded after a 488 nm (GFP) or 514 nm (YFP and FM4-64) excitation and a selective emission of 495–525 nm for GFP, 520–550 nm for YFP and 600–700 nm for FM4-64. Time-lapse analysis of YFP was carried out with low laser output and a large YFP emission bandwidth to minimize photobleaching. Some bleedthrough of FM4-64 staining could thus be observed in the YFP channel as shown by YFP-RabA2a plasma membrane labelling. Elongation zone length was defined as the distance from the last epidermal meristematic cell to the first differentiated cell with a root hair bulge. Division zone length was defined as the position from the quiescent centre to the position of the first rapidly elongating epidermal cell. Cell size of 5-day-old seedlings was measured with ImageJ, and the values were gathered and processed in Microsoft Excel.

Root tips (3 mm) were cut in 1-hexadecen, transferred to a 200 μm size cupule (Leica, ref. 16706897) containing 1-hexadecen and frozen with a high-pressure freezer apparatus (EMPACT2, Leica, France). Freeze substitution was carried out (Leica freeze substitution unit AFS2, Leica, France) in acetone supplemented with 2% osmium tetroxide warming up progressively from −90°C to −30°C (specimens were left at −90°C for 27 hours, then warmed up to −60°C over 15 hours). Specimens were placed in a −60°bath for 8 hours, before the next warm-up step to −30°C over 15 hours, where they remain for additional 8 hours. Root tips were finally infiltrated and embedded in epoxy resin (Low Viscosity Premix Kit-medium, Agar) at room temperature according to the manufacturer's instructions. For polymerization, they were placed in flat plastic moulds and polymerized for 17 hours at 60°C.

70 nm ultrathin sections (Ultracut UC6, Leica) were collected on formvar-coated copper grids and poststained with aqueous 2% uranyl acetate and lead citrate as described (Hawes and Satiat-Jeunemaitre, 2001). They were examined with a JEOL 1400 transmission electron microscope (Croissy, France) operating at 120 kV. Images were acquired using a post-column high-resolution (11 megapixel) high-speed camera (SC1000 Orius, Gatan).

Phospholipid analyses were performed at the Kansas Lipidomics Research Center, as described (Welti et al., 2002). For sterol analysis, total lipids were loaded onto HPTLC plates developed in hexane, ethylether and acetic acid (90:15:2; v/v/v) and separated into diacylglycerols (R_F 0.08), sterols (R_F 0.17), fatty alcohols (R_F 0.22) and free fatty acids (R_F 0.29).

We thank Ian Moore, Thierry Gaude, Karin Schumacher and Gerd Jürgens for respectively for the gifts of pRab-A2:YFP-Rab-A2a, p35S:GFP-RabF2b and p35S:VHAa1-GFP transgenic lines and for KNOLLE antibodies. We thank Jan Traas and Jose Estevez for the gift of pPIN1-PIN1:GFP and LeAGP1-GFP transgenic lines respectively. We thank Bruno Letarnec for taking care of the plants. Rothamsted Research receives grant-aided support from the BBSRC (UK). This work has used the IJPB cytology and imaging facility and the plant chemistry facility (supported by Region Ile de France and Conseil Général des Yvelines) as well as the Electron Microscopy facilities and Cell biology unit of the Imagif platform (CNRS, supported by the Conseil General de l'Essonne, www.imagif.cnrs.fr). The author responsible for distribution of materials presented in this article is Jean-Denis Faure.