Cellularisation in the endosperm of Arabidopsis thaliana is coupled to mitosis and shares multiple components with cytokinesis

Cytokinesis partitions the cytoplasm of a dividing cell and involves a precisely orchestrated interaction between the cytoskeleton and vesicle trafficking that leads to synthesis of new plasma membrane. In higher plants,a specialised cytoskeletal structure supports the de novo formation of a cell wall during cytokinesis. The components of this cellular assembly, the phragmoplast, include microtubules, actin microfilaments and Golgi-derived vesicles (Staehelin and Hepler,1996; Heese et al.,1998; Sylvester,2000). At the end of telophase, elements for membrane and cell wall synthesis are transported by the phragmoplast to the interzone where the cell plate forms a membrane-bound compartment enclosing the immature wall. Phragmoplast and cell plate expand laterally until the cell plate fuses with the parental plasma membrane and cell wall. At the onset of mitosis, the site of fusion is marked by a transient array of microtubule, the preprophase band(PPB).

The prominent role played by the microtubule cytoskeleton in this process is well supported by genetic evidence. Mutations in the Arabidopsis PILZ genes encoding the microtubule folding cofactor complex(Steinborn et al., 2002)severely disrupt microtubule organisation and result in a complete block of mitosis and cytokinesis both in embryo and endosperm(Mayer et al., 1999). Other mutations, such as tonneau1 and fass/tonneau2, more specifically affect organisation of cortical microtubules: fass/ton2embryos, for example, lack both the cortical array of parallel microtubules and the PPB (Traas et al.,1995; McClinton and Sung,1997). As a consequence, fass/ton2 cells undergo unordered expansion in all dimensions and form oblique irregular cell walls very unlike the well-ordered cell files that are representative ofArabidopsis embryos (Torres-Ruiz and Jürgens, 1994; Traas et al., 1995).

Several mutations that affect cytokinesis in the Arabidopsisembryo have been isolated, including knolle and keule(Mayer et al., 1991).KNOLLE encodes a member of the syntaxin family, a class of membrane-bound receptors required for docking and fusion of vesicles at the target membrane (Lukowitz et al.,1996). The KNOLLE protein only accumulates at mitosis and localises to the plane of division, indicating that its specific role is to mediate membrane fusion at the cell plate(Lauber et al., 1997). Thekeule (Assaad et al.,1996) and hinkel(Strompen et al., 2002)mutants display cytokinesis defects during embryo development similar to those in knolle. The KEULE gene encodes a member of the Sec1 family (Assaad et al., 2001). SEC1 proteins regulate vesicle trafficking and interact with syntaxins. E. coli-expressed KNOLLE binds KEULE from plant extracts in vitro,suggesting that the two proteins interact directly during cytokinesis. In support of this, the knolle keule double mutant displays a synthetic embryo-lethal phenotype, with cytokinesis completely prevented(Waizenegger et al., 2000). The HINKEL gene encodes a kinesin-like protein apparently involved in reorganisation of phragmoplast microtubules(Strompen et al., 2002). Additional mutations in genes non-allelic to KNOLLE, KEULE andHINKEL but with similar embryo-lethal phenotypes have recently been isolated. These mutants include runkel(Nacry et al., 2000), open house (allelic to findling; W. Lukowitz and C. Somerville,unpublished) (Nacry et al.,2000) and a novel allele of pleiade, first isolated as a root-specific cytokinesis mutant (Hauser and Bauer, 2000). The functions and products of the mutated genes are not known.

In flowering plants the pollen delivers two male gametes that fuse with the egg cell and the central cell present in the ovule. The ensuing fertilisation product of the central cell develops as the endosperm. In Arabidopsisthe early endosperm development is characterised by a series of synchronised nuclear divisions that are not followed by cytokinesis(Mansfield and Briarty,1990a). This results in the formation of a large multinucleate cell, a syncytium. The syncytial Arabidopsis endosperm contains three distinct domains (Brown et al.,1999). In this study we refer to these as the micropylar endosperm, surrounding the embryo at the anterior pole (MCE), the peripheral endosperm (PEN) and the chalazal endosperm (CZE) at the posterior pole,reflecting the arrangement along the anterior-posterior axis defined by the site of delivery of the sperm (anterior). While nuclei in the MCE share a common mass of cytoplasm, each nucleus in the PEN is associated with its individual cytoplasm surrounded by a dense array of microtubules in a structure termed a nucleocytoplasmic domain (NCD)(Brown et al., 1999). In MCE and PEN pseudo-synchronous mitoses take place at different paces(Boisnard-Lorig et al., 2001). At the globular stage of embryogenesis, the syncytial endosperm contains approximately 100 nuclei (Stage VIII). At this stage, the formation of cell walls that partition nuclei and cytoplasm into individual cells is initiated in the MCE. This process, termed cellularisation, was reported to proceed as a wave across the PEN towards the CZE(Mansfield and Briarty, 1990b;Brown et al., 1999). Cellularisation establishes one endosperm cell layer and precedes further cell division. In maize and in barley, complex microtubule rearrangements take place prior to endosperm cellularisation. During cellularisation so-called adventitious or cytoplasmic phragmoplasts are observed at the junctions of opposing ends of radial microtubules that encircle each nucleus(Brown et al., 1994). Similar adventitious phragmoplasts are seen during Arabidopsis endosperm cellularisation (Brown et al.,1999). Recently, detailed studies of endosperm cellularisation inArabidopsis have confirmed that cell wall formation is mediated by specialised types of phragmoplasts and cell plates(Otegui and Staehelin, 2000b). The syncytial-type cell plates go through a series of maturation steps that are similar, but not identical, to those observed in conventional cytokinesis(Otegui and Staehelin, 2000a). While cytological descriptions of endosperm cellularisation thus indicate its unique features, evidence for the accumulation of both KNOLLE(Lauber et al., 1997) and the dynamin-like protein ADL1A (Otegui et al.,2001) in both somatic- and syncytial-type cell plates hints at shared components in the two processes.

We report in vivo observation of cellularisation in the PEN ofArabidopsis and show that it follows nuclear division, as in conventional cytokinesis. We evaluate the differences and similarities between conventional cytokinesis and endosperm cellularisation using a genetic approach. We analyse the effect on endosperm cellularisation of sixArabidopsis mutations that affect cytokinesis in the embryo. This analysis shows that most embryo cytokinesis-defective mutants display similar defects in endosperm cellularisation. In addition, we characterise mutant alleles of a novel gene, SPÄTZLE, in which endosperm cellularisation is defective but cytokinesis is normal in the embryo and in plants homozygous for the mutation.

Ecotypes used as wild types were Landsberg erecta and WS. Alleles used in the Ler ecotype were fas/ton2(Torres-Ruiz and Jürgens,1994), knolle-AP6-16(Lukowitz et al., 1996),keule-Ap77 (Lukowitz et al.,1996) and runkel U381. Mutant alleles of other genes required for cytokinesis in the embryo runkel-7 and hinkelwere generated in the Ler ecotype at the Max Planck Institute in Tübingen while OX10, pleiade-4, and open house-1 were generated in the Ler ecotype at the Carnegie Institution. The lines DRU42 and DQB12 were generated in the WS ecotype in Versailles(Bechtold et al., 1993). The KS117 GFP marker line has been described elsewhere(Sørensen et al.,2001).

Individual Arabidopsis thaliana siliques were opened with two shallow longitudinal cuts on either side of the false septum. Siliques were stained with Schiff's reagent (Sigma) and embedded in LR-white (Sigma)according to the method of Braselton et al.(Braselton et al., 1996). All mutant lines were initially propagated as heterozygotes and produced siliques that contained both wild-type seeds and seeds that displayed the mutant phenotype. Seeds that originated from individual siliques were isolated in each preparation in order to be able to compare the mutant and wild-type development at corresponding stages. Confocal microscopy was performed using a Zeiss LSM-510 microscope using the 488 nm excitation line of an argon laser and a long pass emission filter at 510 nm.

Seeds of an endosperm marker line from the enhancer trap collection of J. Haseloff were placed in a small moist chamber as described previously(Boisnard-Lorig et al., 2001). Imaging of the GFP fluorescence (488 nm excitation; 510-550 nm emission) was performed for 12 to 24 hours using a time-lapse automated recording of confocal sections every 10 minutes. We used a × 16 objective n.a. 0.7. with oil immersion. 1024×1024 pixel confocal sections were processed using Photoshop 5.0 (Adobe). Movies were produced with Metamorph.

The homozygous mutant line DRU42 in WS ecotype was crossed to wild-type Landsberg erecta. F₂ seeds from F₁ progeny heterozygous for the DRU42 mutation were used to generate the mapping population. The phenotype of individual F₂ plants was determined and the wild-type plants were used to investigate recombination events with polymorphic markers (CAPS and SSLP). 49 plants were analysed for three markers per chromosome.

Previous analyses of Arabidopsis endosperm using microscopical observations of fixed tissues have shown that cellularisation begins at the MCE when the embryo has reached the late globular stage. This corresponds to the end of the stage VIII as defined by the number of mitotic rounds in the syncytial endosperm (Boisnard-Lorig et al.,2001). During the transition from the triangular to the early-heart stage of embryogenesis, cellularisation takes place in the PEN,apparently following a wave-like pattern initiated at the anterior pole(Mansfield and Briarty,1990b). At the completion of cellularisation, anticlinal walls are established in the entire PEN. The endosperm then comprises approx. 200 nuclei(stage IX) (Scott et al.,1998). At late heart stage, a near-synchronous round of mitosis(developmental stage X) followed by what appears to be a conventional cytokinesis allows the formation of periclinal walls, defining an outer peripheral cell layer and an inner layer that undergo another round of cell division (stage XI, Fig. 4A)(Brown et al., 1999).

To perform a detailed analysis of the timing and dynamics of endosperm cellularisation we have taken advantage of an Arabidopsis enhancer trap line that expresses a variant of the green fluorescent protein, mGFP5(Haseloff et al., 1997),uniformly in the developing endosperm(Boisnard-Lorig et al., 2001). mGFP5 is retained in the endoplasmic reticulum and during cytokinesis fluorescence from mGFP5 accumulates at the forming cell plate(Haseloff, 1999). This labelling permits in vivo observation, using confocal microscopy, of endosperm at the transition between the stages VIII and IX when PEN cellularisation takes place. At the end of stage VIII, cellularisation is complete at the micropylar pole around the embryo but the PEN remains syncytial(Fig. 1A, 0 minutes; MCE shown only in supplementary material). At this point the NCDs remain unchanged for an extended period, without any signs of cell wall formation or division. In the data presented here, this period is more than 3 hours(Fig. 1A,B). In multiple observations, each with a duration greater than 6 hours we did not observe any extension into the PEN of the cell walls present in the MCE. At the transition to stage IX, simultaneous nuclear divisions are initiated in the micropylar mitotic domain and propagate rapidly as a wave across the PEN(Fig. 1C; total duration approximately 40 minutes, the anterior pole is toward the top of the figure)as reported for earlier developmental stages(Boisnard-Lorig et al., 2001). In the already cellularised MCE, GFP fluorescence is detected in a small band between sister nuclei, about 15 minutes after karyokinesis (see movie). This band expands centrifugally as in conventional cytokinesis and delineates the nascent cell wall between the two daughter cells. In the syncytial PEN a GFP signal is first detected between post mitotic nuclei that still form a distinguishable pair, 30-40 minutes after telophase(Fig. 1D). Over a period of 60 minutes after karyokinesis, cellularisation becomes apparent throughout the peripheral endosperm (Fig. 1E). It was often difficult to observe more precisely the extension of the band of GFP fluorescence between individual nuclei because of growth of the seed that displaced the observed nuclei out of the focal plane. Initiation of PEN cellularisation was, however, repeatedly (n=13) observed immediately after the eighth wave of mitosis that crosses the endosperm from the anterior to the posterior pole. Cellularisation is completed without any further divisions occurring (Fig. 1F). Fortuitous observation of a developing seed fixed at a point during the advancement of the eighth mitotic round revealed further details of the first steps of cellularisation. Fig. 1G-J are periclinal views of dividing nuclei positioned along the posterior to anterior axis. The mitotic stage seen mostly frequently at the posterior (i.e. the more recently produced) is metaphase(Fig. 1G). Toward the anterior pole anaphase (Fig. 1H) and telophase (Fig. 1I) are observed successively, immediately followed by formation of a structure resembling a cell plate between separating sister nuclei(Fig. 1J). It thus appears that cellularisation follows a series of events initiated in conjunction with a synchronised division, where a cell plate is first formed between sister nuclei, followed rapidly by assembly of cell walls between non-sister nuclei throughout the PEN, resulting in a layer of cells at the periphery of the central vacuole.

To address the question of whether cellularisation and conventional cytokinesis are under the same genetic and molecular control, we analysed endosperm development in several cytokinesis mutants. Visual screens ofArabidopsis embryos or seedlings with altered morphology have identified numerous genes required for cytokinesis. Mutants with very similar phenotypes include KNOLLE (KN)(Mayer et al., 1991),KEULE (KEU) (Mayer et al., 1991), HINKEL (HIN)(Strompen et al., 2002),RUNKEL (RUK) (Nacry et al., 2000), OPEN HOUSE (OPN) (W. L. and C. S.,unpublished) and PLEIADE (PLE)(Hauser and Bauer, 2000).

The phenotype of knolle and keule mutant embryos is easily discernible from the wild type, which has ordered files of regularly shaped cells (Fig. 2A). Both mutants are characterised by the presence of cells and nuclei of irregular shapes and sizes in addition to giant multinucleate cells(Fig. 2C,E). Nevertheless, the endosperm development is affected differently by these two mutations. In allkeule seed analysed (n>30) cellularisation was indistinguishable from that in wild type, beginning at the MCE(Fig. 2C) and subsequently forming a layer of regularly positioned hexagonal cells in the PEN(Fig. 2D). Following cellularisation keule endosperm cells begin anticlinal division,forming a multilayered endosperm identical to that observed in wild-type seeds with early torpedo stage embryos. In contrast, knolle endosperm was not cellularised in the vast majority of seeds observed (n>50),with both MCE (Fig. 2E) and PEN(Fig. 2F) remaining syncytial. In approximately 10% of the seeds some degree of cellularisation was apparent,most frequently limited to the MCE. However, one case of cellularised PEN with multiple cell layers, containing multinucleated cells and irregularly shaped nuclei, was also observed. This is reminiscent of the phenotype variation observed in knolle embryos, containing a mixture of apparently normal and defective cells, with the difference that, being a synchronised event within each endosperm domain (MCE or PEN), defective cellularisation will be evident throughout the entire domain. Subsequently, individual cells might be affected by aberrant cytokinesis when additional cell layers are formed.

Double mutant knolle/keule embryos display a remarkable additive phenotype, where cytokinesis is completely absent(Fig. 2G)(Waizenegger et al., 2000). In accordance with the conclusion that KNOLLE is required for cellularisation, the endosperm of knolle/keule seeds remained syncytial (Fig. 2G,H). In contrast to the embryo, no new endosperm phenotype was observed and the endosperm phenotypes of knolle and keule/knolle mutants appeared identical.

Endosperm cellularisation in plants mutant for HINKEL, which encodes a kinesin-related protein, and for OPEN HOUSE for which the gene product has not yet been identified, follows the pattern ofknolle mutants. Although generally remaining syncytial(Fig. 3A-D), partial cellularisation, particularly in the MCE, was occasionally observed.

The endosperm phenotypes of runkel and pleiade are less severe than those of knolle, hinkel and open house, although cellularisation clearly requires these genes in order to proceed normally(Fig. 3E-H). Partial cellularisation of the MCE (Fig. 3E,G) was observed in almost all seeds analysed. Cellularisation in the PEN, albeit defective and displaying multinucleate cells(Fig. 3F), nuclei of different sizes (Fig. 3H) and incomplete cell walls (Fig. 3F,H) was likewise observed in most seeds. In pleiade the relatively mild cellularisation phenotype is accompanied by a correspondingly mild cytokinesis phenotype, with the embryo maintaining a correct morphology until cotyledon development (Fig. 3G). Additionally, the first isolated mutant alleles of this gene allow germination and plant development, with cytokinesis being affected only in the root(Hauser and Bauer, 2000).

A common feature of the cellularisation mutants described here is the high density of NCDs aligned along the peripheral cell wall in the PEN(Fig. 2F,H,Fig. 3B,D). This indicates that, in the absence of cellularisation, the syncytial PEN will continue divisions at least through nine mitoses.

The fass/ton2 mutation affects embryo morphogenesis by interfering with cell wall orientation and cell shape, resulting in compressed embryos with seemingly randomly arranged cells and abnormal tissue organisation(Torres-Ruiz and Jürgens,1994). Although endosperm cellularisation takes place infass/ton2 seeds, closer examination reveals differences between wild-type and fass/ton2 endosperm. The number of cell layers in completely cellularised seeds are frequently reduced from four in the wild type (Fig. 4A) to two or three in fass/ton2 (Fig. 4B). Observations at the cellularisation front, just after the first anticlinal division reveal that these differences are caused by the inability of the fass/ton2 mutant to correctly form periclinal cell walls separating newly divided sister nuclei(Fig. 4D). The layer of small peripheral cells is not continuous in fass/ton2 endosperm and cells of the inner layers of the PEN are longer than in the wild type possibly as a result of the absence of periclinal divisions(Fig. 4B,E).

In an attempt to identify mutants that specifically affect endosperm development but not embryonic development, we visually screened a subset of 150 lines from the Versailles collection of T-DNA insertion lines (20,000 lines pre-screened by L. L. for abnormal seed shape). Two lines, DRU42 and DQB12, were recovered that showed phenotypes with aberrant endosperm development but no apparent defect in the embryo. Complementation analyses revealed the two mutations to be allelic. Upon selfing, 24.7% (s.e.m.=0.7;n=500) of the seed from heterozygous plants displayed the mutant endosperm phenotype, indicating that the mutations are recessive and segregate in a mendelian fashion. In none of the lines was the mutant endosperm phenotype linked with the kanamycin resistance gene of the T-DNA, leading to the conclusion that the mutated gene is not tagged by a complete T-DNA. Mapping with PCR based markers indicated that the SPÄTZLE gene is located on chromosome 1 approximately 5 cM south of the SSLP marker ACC2.

As with the other mutants that we analysed, spätzle displayed a variable range of phenotypes. At late-heart stage of embryogenesis, when endosperm cellularisation is complete in wild type(Fig. 2A), no sign of cell wall formation could be detected in the most frequent mutant phenotype, either in the MCE or in the PEN (Fig. 5). Until endosperm stage VIII, spätzle endosperm develops normally and the PEN contains regularly organised NCDs(Fig. 5C), as in wild type just prior to cellularisation. The PEN undergoes at least one additional mitotic division resulting in a syncytium with increased density of NCDs(Fig. 5D), similar to that ofknolle, hinkel and open house as described above. Subsequently, the number of NCDs in the PEN continuously reduces while the size of individual nuclei increases. After stage VIII, we observed various degree of fusion of NCDs (Fig. 5E). Division of nuclei in these binucleate NCDs apparently leads to lack of nuclear separation upon mitosis(Fig. 5E) and ultimately to fusion of nuclei (Fig. 5F). This process can continue until the PEN contains a few giant NCDs, each with one or more giant multinucleolated nuclei(Fig. 5G). This pattern of endosperm development distinguishes spätzle from the other cytokinesis mutants where enlarged nuclei were always limited in frequency and accompanied by partial cellularisation of the PEN (e.g.Fig. 3H). At all stages analysed, in seeds displaying the spätzle endosperm phenotype embryos did not show any visible defects associated with cell division, such as incomplete cell walls, enlarged nuclei or multinucleate cells, which are typical of all the other cytokinesis mutants described here (Figs2 and3). Moreover, morphogenesis and pattern formation in spätzle embryos is normal (protoderm,vascular tissue, root and shoot embryonic meristems)(Fig. 5A). After torpedo stage,most spätzle embryos show retardation in overall growth(Fig. 5B), perhaps resulting from the inability of the defective endosperm to nurture the embryo. Mutant seeds, which can be recognised by their slightly shrunken appearance and slightly smaller size at maturity, are viable, can germinate on soil and give rise to homozygous spätzle plants. In those plants we could detect neither morphological defects nor growth retardation. Fertility was not altered, and homozygous spätzle plants produced 100% seeds with non-cellularised endosperm. Similar observations were made for bothspätzle alleles, DRU42 and DQB12. These observations strongly support the conclusion that the SPÄTZLE gene product is required specifically for cellularisation in the endosperm but not for embryogenesis.

We have previously shown that the FIS (FertilisationIndependent Seed) genes are required for correct anterior-posterior polarisation in the Arabidopsis endosperm and that, additionally, endosperm cellularisation is impaired in a fismutant background (Sørensen et al.,2001). To investigate whether a general connection exists between cellularisation and patterning, we analysed development of the CZE in two representative cytokinesis mutants. As in wild type(Fig. 6A), knolleposterior pole endosperm is characterised by the presence of a chalazal cyst and one or more nodules, both containing multiple nuclei in a common cytoplasm(Fig. 6C). Likewise,spätzle CZE morphology appears normal(Fig. 6E), but the cytological distinction between chalazal nodules and giant NCDs in the PEN is not as clear as in the wild type or as in knolle. In the two mutantsknolle and spätzle we introgressed the marker line(KS117) with an exclusive GFP expression at the posterior pole of the endosperm (Fig. 6B). The expression of the marker is restricted to the CZE in all seeds of bothknolle (Fig. 6D) andspätzle (Fig. 6F)siliques. This confirms the correct patterning of the endosperm in the absence of cellularisation in cytokinesis-defective mutants as well as inspätzle. Cellularisation therefore does not appear to be necessary for the expression of endosperm polarity.

In Drosophila the embryo develops initially as a syncytium and cellularises according to mechanisms distinct from those observed during cytokinesis. The analysis of cellularisation-defective mutants(Schweisguth et al., 1991;Schejter and Wieschaus, 1993;Miller and Kiehart, 1995) has shown that despite an apparent mechanistic difference at the cellular level,many components are shared between cellularisation and cytokinesis. We have undertaken a similar genetic approach in Arabidopsis. Six mutants that affect cytokinesis in the embryo were analysed and, with the exception ofkeule, found to be defective in endosperm cellularisation. Endosperm cellularisation does not take place or is incomplete while nuclear division is not directly affected. However, at later stages after partial cellularisation,larger nuclei are present in some larger cells, presumably due to coalescence of mitotic spindles.

Among the genes investigated, KNOLLE, KEULE and HINKELhave been cloned and found to encode proteins potentially involved in vesicle trafficking. HINKEL encodes a novel kinesin that is localised to the cell plate during cytokinesis in a pattern complementary to that of the KNOLLE protein (Strompen et al.,2002). KNOLLE is localised to forming cell plates in cellularising endosperm (Lauber et al.,1997), in agreement with the absence of cellularisation observed in knolle. Direct interaction between the KNOLLE syntaxin and the KEULE sec1-related protein has been demonstrated and it is therefore surprising that keule mutant endosperm has a wild-type appearance. It is not known whether KEULE is present in the endosperm. In contrast to the embryo, the endosperm of the knolle keule double mutant does not show an additive phenotype to that observed in knolle. It is thus likely that KEULE either is not expressed during endosperm cellularisation or that a functionally redundant gene exists. It has been suggested that each Sec1 protein interacts with several syntaxins(Sanderfoot et al., 2000). Our observation that KNOLLE but not KEULE is required for endosperm cellularisation suggests that the KNOLLE syntaxin is likely to interact with other Sec1 proteins than KEULE in the endosperm.

Absence of endosperm cellularisation has also been reported in mutants affected in structural elements such as microtubules. Mutations in thePILZ group of genes (Mayer et al., 1999) that includes ttn1 and ttn5(McElver et al., 2000;Tzafrir et al., 2002) affect the complex of tubulin-folding cofactors and cause disruption to the microtubule cytoskeleton (Steinborn et al., 2002). The absence of microtubules thus explains the lack of cytokinesis in the embryo and of cellularisation in the endosperm ofpilz mutants. Other titan mutants(Liu and Meinke, 1998),including titan-7 and -8 encoding chromosome scaffold proteins (Liu et al., 2002)show a non-cellularised endosperm with giant nuclei likely to be the indirect consequence of defective chromosome separation.

Hitherto, comparisons of cellularisation with conventional cytokinesis has been based on different cytological analyses of wild-type tissues, and although the similarities of the two processes has been acknowledged(Olsen, 2001) most reports have emphasised their differences (Otegui and Staehelin, 2000a; Nacry et al., 2000; Verma,2001). Our genetic analysis of the knolle, hinkel, open house,runkel and pleiade mutations directly involved in cytokinesis provides conclusive evidence that somatic cytokinesis and endosperm cellularisation share many components.

The PPB is a temporary assemblage of microtubules that marks the future site of connection of the cell plate to the cell wall in plant cytokinesis and results in an F-actin-free zone. PPBs have not been observed during endosperm cellularisation but only during the subsequent developmental phase of the barley aleurone layer (Brown et al.,1994). In Arabidopsis, both cellularisation and subsequent periclinal divisions take place without a discernible PPB(Brown et al., 1999). The PPB is lacking in fass/ton2 mutants and interestingly, fass/ton2cellularisation occurs normally in MCE and PEN but subsequent cytokinesis is impaired in the endosperm. This confirms that a PPB is not required for cellularisation, but, in contrast to earlier observations, it appears to indicate that the PPB is necessary for the first periclinal divisions to proceed correctly. Alternatively, the FASS gene product might be involved in an unknown process or structure directing periclinal divisions inArabidopsis endosperm.

In previous observations on fixed seeds cellularisation appeared to progress as a wave from the MCE towards the CZE(Mansfield and Briarty,1990b). Cellularisation was considered to be uncoupled from mitosis and no distinction was seen in the appearance of cell walls between sister or non-sister nuclei (Brown et al.,1999). In order to perform real time in vivo observations of this process we used the GFP variant mGFP5 that is targeted to the endoplasmic reticulum and accumulates around the forming cell plate in root meristem cells(Haseloff, 1999). Accumulation of RE on the side of cell plates in endosperm has been reported(Otegui et al., 2001) and this likely accounts for the labelling with mGFP5 that we observed. This property allowed us to visualise the time course of cellularisation in living seeds. Completely cellularised MCE was observed at the end of the globular stage of embryogenesis. At this stage, the endosperm contains approx. 100 nuclei (stage VIII). The wave of mitosis leading to stage IX is initiated in the MCE and propagates within 1 hour through the PEN, as observed already during syncytial divisions (Boisnard-Lorig et al.,2001). In vivo observations enabled us to establish that cellularisation in the PEN is taking place in conjunction with the 8th mitotic cycle. The cell walls first appear between sister nuclei, rapidly followed by walls completely surrounding individual NCDs, thereby defining cells. Although the connection between mitosis and cellularisation has not previously been observed in Arabidopsis, probably due to the short time scale (one hour) of these events, coincidence of cell plate initiation with nuclear division has been remarked on in Ranunculus sceleratus(Xu and van Lammeren, 1993). Indeed we observed typical intermediate structures with nascent cell walls forming between sister nuclei only in three fixed seeds amongst several thousands. Furthermore, our observations relate to the initial appearance of walls, not to the rate at which these might expand anticlinally to form the characteristic open-ended alveoli (Brown et al., 1999).

In Arabidopsis endosperm the link between cellularisation and mitosis reported in this study furthers the parallels between endosperm development and Drosophila embryogenesis. In Drosophila a series of rapid succession of pseudo-synchronous waves of mitosis cross the embryo until cycle 10. Then a marked and gradual increase in the length of the cell cycle culminates at cycle 14 in the cellularisation of the blastoderm(Edgar et al., 1986). InDrosophila, the signal that initiates blastoderm cellularisation relies on a nucleo-cytoplasmic threshold(Edgar et al., 1986). This might be the case as well in the Arabidopsis endosperm.

We have isolated two alleles of a new mutant, spätzle, that are characterised by the absence of cellularisation in the endosperm but do not show cytokinesis defects in the embryo. Mature seeds will germinate and the homozygous spätzle plant is indistinguishable from wild type. One possible explanation for the absence of defects in embryos and homozygous vegetative tissues is that SPÄTZLE encodes a gene involved in mechanisms common to cellularisation and cytokinesis, but functional redundancy prevents the appearance of an embryo phenotype, a situation that could mirror our interpretation of defects observed inkeule embryos but not in the keule endosperm (see discussion above). Alternatively, spätzle mutants might be affected in a process specific to cellularisation.

Much like spätzle, the fis mutants produce endosperm that does not cellularise although the embryo does not display any defect in cytokinesis, is viable in fis1/medea and in fis2 mutants(Sørensen et al., 2001)and can produce plants of wild-type appearance. The FIS genes,control multiple aspects of seed development including initiation of endosperm divisions (Ohad et al., 1999;Luo et al., 1999) and correct definition of the CZE (Sørensen et al., 2001). In contrast, endosperm polarity is clearly not affected in spätzle endosperm and the SPÄTZLE andFIS genes are probably not in the same genetic pathway.

Like spätzle, titan3 does not affect embryo development, with the mutant phenotype being restricted to the endosperm, where giant nuclei and NCDs are formed (Liu and Meinke,1998). However, ttn3 endosperm is capable of cellularisation that takes place at the same time as in the wild-type seeds. The defective cellularisation in spätzle endosperm is probably not a consequence of nuclei enlargement. The gene product ofSPÄTZLE is indeed required for and limited to the process of cellularisation, possibly being a structural protein specifically expressed in the endosperm. As an alternative hypothesis we propose that the primary defect in spätzle is related to the nucleo-cytoplasmic ratio or to its interpretation. Cloning the gene mutated in spätzle might allow distinction between these hypotheses.

We would like to thank M.-T. Hauser for sharing results prior to publication and Jean-Emmanuel Faure for critical reading and helpful comments on the manuscript. This work was carried out in the RDP laboratory directed by Christian Dumas and supported by the INRA, the CNRS, the University Lyon I and the Ecole Normale Supérieure de Lyon. L. L. is supported by the INRA in the Biologie des Semences laboratory directed by Michel Caboche and receives some support form the laboratory directed by Georges Pelletier. This work was partially supported by the HFSP Organisation (fellowship LT594/96 to W. L.),the Carnegie Institution of Washington (Barbara McClintock fellowship to W. L.), the NRI Competitive Grant Program/USDA (grant No. 00-35304-9394 to W. L. and C. S.). More specific funds came from a Marie Curie Training Fellowship(MBS), an EMBO short term fellowship (F. B.) and a Leibniz award from the Deutsche Forschungsgemeinschaft (G. J.).