CUPULIFORMIS establishes lateral organ boundaries in Antirrhinum

A characteristic feature of plant development is that it can be modified in response to external and internal signals throughout the life of the plant. This is possible because almost all adult organs are ultimately derived,postembryonically, from pools of undifferentiated stem cells, which are maintained in the meristems (Steeves and Sussex, 1989). In order to produce mature plants with a regular arrangement of lateral organs, the shoot apical meristem (SAM) and other meristems derived from the SAM, such as the floral meristems, need to be organised at several levels. The spatial distribution of lateral organs is established by the geometrical pattern of initiation of primordia on the flanks of the meristems (see Byrne et al.,2003). In addition to regulating the position of primordia initiation, each primordium acquires an identity to enable it to develop into the appropriate type of lateral organ. At least for floral organs, it appears that the primary determinants of identity are members of the MADS-box family of transcription factors (Parenicova et al., 2003). Surprisingly, it has also emerged that plants require another function to define the outer boundaries of lateral organs. The Petunia NO APICAL MERISTEM (NAM) gene is expressed at lateral organ boundaries (Souer et al.,1996), as are the Arabidopsis CUP-SHAPED COTYLEDON genes CUC1, CUC2 and CUC3(Souer et al., 1996; Aida et al., 1997; Takada et al., 2001). nam mutants and cuc1 cuc2 double mutants have fused cotyledons, and also show floral abnormalities including organ fusion(Souer et al., 1996; Aida et al., 1997; Takada et al., 2001). Fused cotyledons are also found in the cuc3 mutant, and in mutant combinations between cuc3 and cuc1, and/or cuc2(Vroemen et al., 2003). This suggests that these genes function to define the boundaries between some developing lateral organs, possibly by negatively regulating cell proliferation in the boundary regions.

Another phenotypic feature of nam and the cuc mutants is an inability to form a SAM during normal development. In nam mutants,the Petunia apical meristem fails to form in most seedlings, but a variable proportion eventually manage to produce an escape shoot with a functional SAM (Souer et al.,1996). Double and triple mutants involving cuc2 and cuc1, and/or cuc3, also lack a functional SAM(Vroemen et al., 2003). In the case of cuc1 cuc2 double mutants, shoots bearing a functional SAM have been regenerated from double mutant calli(Aida et al., 1997). SHOOT MERISTEMLESS (STM), which encodes a homeodomain-containing factor, is one of the key determinants of the Arabidopsis SAM(Long et al., 1996). As STM is not expressed in cuc1 cuc2 double mutant embryos, it is possible that CUC1 and CUC2 exert their influence on SAM formation by promoting the expression of STM(Aida et al., 1999). Indeed, CUC1 and CUC2 can even promote adventitious shoot formation on calli in an STM-dependent manner(Daimon et al., 2003). It has been suggested that a boundary region of reduced cell division is required for the establishment of a SAM, and that the CUC and STM genes participate in the formation of such a niche of cells from which meristems can be formed (Vroemen et al.,2003).

Neither the nam nor the cuc1 cuc2 mutants affect the boundaries of vegetative or inflorescence lateral organs outside the flower,raising the possibility that other genes are involved in the establishment of these boundaries (Souer et al.,1996). Here, we describe cupuliformis (cup)(Stubbe, 1966), a classical Antirrhinum mutant that shows extensive fusion of all lateral organs and a transitory defect in SAM initiation. We show that CUP encodes a NAC-domain transcription factor, related to NAM and the CUCs. In contrast to nam and cuc1 cuc2, cup mutants strongly affect all lateral organ boundaries, suggesting a greater redundancy of function for the boundary factors in Petunia and Arabidopsis. Furthermore, we demonstrate a direct interaction between CUP and a TCP-domain factor, TIC,which suggests a model for the establishment of organ boundaries in plants.

Two alleles of cupuliformis, cup^fert and cup^sem, were originally described by Stubbe(Stubbe, 1966), and the corresponding seeds were obtained from the Gatersleben Stock Centre. Molecular analysis of the two alleles subsequently demonstrated that they were identical. This is in agreement with our failure to detect differences in fertility, originally described for cup^fert and cup^sem, in the plants obtained. We conclude that cup^fert has been lost and that both stocks represent the cup^sem allele, which we now refer to as cup-1. A further allele, cup-2, was identified(Z.S.-S., unpublished) in an unrelated EMS mutagenesis experiment. Allelism was confirmed by crossing. The ratio of wild type to cup mutants in a cup-1/CUP×cup-2/CUP cross was 3.5:1(n=410). For electron microscopy, tissue was fixed overnight in 2.5%glutaraldehyde/0.1 M phosphate buffer, and incubated in Osmium tetroxide/0.1 M phosphate buffer before dehydration in a 20-100% ethanol series. Samples were then critical-point dried, mounted on stubs, sputter coated and viewed using a JEOL T20 scanning electron microscope. Embryos were dissected from mature seeds from selfed CUP/cup plants allowed to imbibe at 4°C overnight.

The full-length CUP cDNA was isolated by screening an Antirrhinum cDNA library using a mixed PCR probe generated by amplifying the conserved Arabidopsis NAC domain region by RT-PCR,using primers 5′-TCCCACGGACGAAGAGCTCATCAC-3′ and 5′-TGGAGCTTCTTGAGATGAAATGGTAAG-3′. The Genomewalker kit (Clontech)was used to determine the sequence within the 5′-untranslated region of TIC. Primer TIC1 (5′-GGAAGGATCCGAAGGAGGCGATGAT-3′), derived from this region, together with primer TIC-2(5′-ATAAGAATGCGGCCGCTCAAGAATGGTGACTTGTG-3′), derived from the 3′ sequence of the truncated TIC clone obtained in the two-hybrid screen, was used to obtain a full-length TIC transcript by RT-PCR. Two independently amplified PCR products were sequenced.

The full-length CUP gene, lacking the ATG start codon, was cloned into the DNA-binding-domain vector pGBT9 (Clontech) and transformed into yeast strain HF7c (Feilotter et al.,1994), to form the bait strain. The screen was carried out using a random-primed, whole-plant Antirrhinum cDNA library, as previously described (Davies et al.,1996). Positive colonies appearing within 10 days of the screen were confirmed by growth in the presence of 20 mM 3-amino-1,2,4-triazole(3AT). Plasmid DNA was recovered from the positive yeast colonies by transformation into E. coli, and interactions were verified by re-transformation of bait and prey into the yeast strains HF7c and SFY526(Bartel et al., 1993), as previously described (Davies et al.,1996).

A full-length GST-CUP fusion was produced by cloning a CUP PCR fragment, amplified using primers that introduce BamHI and NotI sites, into pGEX-4T-1 (Amersham Pharmacia Biotech) to form pGSTCUP. TIC was also cloned into pGEX-4T-1, between the same restriction sites, to form pGST-TIC. pGEX-4T-1, pGST-CUP and pGST-TIC were used to transform E. coli BL21 (DE3) cells (Stratagene). Proteins were produced and GST pulldowns carried out essentially as described(Causier et al., 2003), with the exception that following lysis the pellets were re-suspended into a solution of 1.5% N-Lauroylsarcosine, 25 mM Triethanolamine, 1 mM EDTA (pH 8.0), and incubated for 10 minutes at 4°C followed by a further centrifugation to remove cell debris.

Seedlings and inflorescences were embedded, sectioned and hybridised as described (Zachgo et al.,2000). A DIG-labelled antisense RNA probe was transcribed from a 600 bp PCR fragment containing the 3′ region of CUP, outside the conserved NAC-domain. A sense control probe was also tested and no signal obtained (not shown).

RT-PCR analysis was carried out using first strand cDNA generated from 0.5μg DNase treated total RNA, using oligo dT and Omniscript RT (Qiagen)according to the manufacturer's instructions. PCR was performed with RT reaction corresponding to 0.0125 μg total RNA using Expand High Fidelity PCR System (Roche) with the following oligos:

TICSMA, 5′-gaattcccgggggaaggaggcgatgatcat-3′;

TICBAM3, 5′-cgggatcctcaagaatggtgacttgt-3′;

CUPBAM5, 5′-ggaaggatccgagaattacaattgctac-3′;

CUPRI3, 5′-ggaagaattcctagtaaccccaaatacg-3′;

AMEF1, 5′-tgagaccaccaagtactactg-3′; and

AMEF2, 5′-caacattgtcaccgggaagag-3′.

All primers were confirmed to be specific for the gene to be amplified. Northern blots were performed as previously described(Sommer et al., 1990).

The two cotyledons of wild-type Antirrhinum are visible at later stages of embryo development as discrete organs, separated by a boundary region where organ growth does not occur(Fig. 1A). By contrast, the cotyledons formed in cup embryos arise as an uninterrupted ring of tissue (Fig. 1B). In wild-type seedlings the first pair of leaves forms shortly after germination, from primordia initiated on the flanks of the SAM(Fig. 1C). Initially, leaves are rarely produced in cup seedlings and they remain as hypocotyls topped by a cup-shaped structure, which can be almost totally closed at its apex (Fig. 1D). Longitudinal sections through the apices of cup mutants reveals an absence of meristematic tissue, although occasionally undifferentiated tissue or small leaves proliferate inside the cups (not shown). Development continues in cup mutants with the formation of new shoots arising from the hypocotyl (Fig. 1E,F,G). These shoots, and further shoots arising from them, also terminate in cup-shaped structures lacking a SAM (Fig. 1H). Over a period of time, most cup plants succeed in producing a shoot that is capable of maintaining a SAM, and which does not terminate in a cup-shaped structure. However, unlike nam and cuc1 cuc2 shoots, these escape shoots show a variety of phenotypic abnormalities. In wild-type plants, leaves and associated axillary meristems form at each node (Fig. 2A,E). In cup mutants the phyllotaxis of leaf production is disrupted, and fusions between neighbouring leaves are very common(Fig. 2C,F,G). As a consequence of leaf fusion, the nodes are surrounded by fused petiole tissue and axillary meristems are not usually formed (compare Fig. 2A with Fig. 2B). The absence of axillary meristems, confirmed by serial sections through nodes, results in plants lacking side shoots, but side shoots can form after removal of the shoot apex (not shown).

Inflorescence development in Antirrhinum is characterised by a change in phyllotaxis, and flowers develop from a spiral of floral meristems produced on the flanks of the inflorescence meristem, in the axils of small leaf-like bracts. This phase of development produces the most spectacular aspects of the cup mutant phenotype. The phyllotaxis of cupmutant inflorescences also changes to spiral, but floral meristems are rarely formed. Instead the inflorescences of cup mutants consist of long shoots ringed by continuous spirals of fused bracts(Fig. 2D). Gradually, the organisation of the SAM is lost and the shoots become increasingly fasciated(Fig. 2H), often producing masses of inflorescence tissue covered with fused bracts. Flowers are only formed in one of the two cup alleles (cup-1),although they are rare, even in that allele. The low frequency of flower formation, and the disorganised and fasciated nature of the SAM, makes characterisation of the early stages of flower formation impossible. Morphological abnormalities are obvious as soon as flowers become apparent,although the abnormalities are highly variable(Fig. 2I-L). Sepals are fused to a greater extent than is observed in wild-type flowers, and an additional whorl consisting of a variable number of petaloid/sepaloid organs is often formed between the first and second whorls. Typically this whorl, which is absent from the wild-type flowers, contains a mean of 3.3 sepal/petal organs(n=61 flowers). The third whorl contains a mean of 5.7 petals(n=61) compared with 5 in the wild-type flower. The petals and stamens are fused and the anthers usually lack pollen, reducing male fertility. The gynoecium is often distorted, flattened, fasciated or hollow. Sometimes additional carpels are produced (mean of 2.2 compared with 2 in the wild type) and fusion between stamens and carpels is frequently observed. Fewer ovules are produced within the ovary and neighbouring ovules are frequently fused (Fig. 2M,N). cup-1 mutants are female sterile, with no seeds being formed whatever the source of pollen used. Radialisation of the normally zygomorphic Antirrhinum flowers is sometimes observed in cup flowers. No obvious morphological alterations are observed in the roots of cupmutant plants.

The cup phenotype can be characterised as a failure to form a SAM or axillary meristems and an inability to define lateral organ boundaries. There are clear differences between the phenotypes of cup mutants and organ boundary mutants in other species(Souer et al., 1996; Aida et al., 1997; Takada et al., 2001). The Petunia nam mutant has fused cotyledons and the SAM is absent. In contrast to cup, most nam seedlings die, but some continue to develop by forming an escape shoot. Unlike cup escape shoots,those nam escape shoots which go on to produce a mature plant, show normal vegetative and inflorescence development. The Arabidopsis cuc1 cuc2 double mutant also has fused cotyledons. In this case, the individual single mutants look normal, except for a tendency towards partial fusion of the cotyledons. cuc1 cuc2 double mutants do not form escape shoots naturally, but shoots can be induced by regeneration from mutant calli. As with nam escape shoots, cuc1 cuc2 shoots do not show the dramatic organ fusion and fasciation effects observed in cup escape shoots. Flowers produced from regenerated cuc1 cuc2 plants show fusion of sepals and stamens, petal loss and female sterility. Like cup flowers, cuc1 cuc2 flowers show a reduction in ovule number (Ishida et al., 2000). nam flowers are male and female sterile, contain an extra whorl of petals that become fused to the stamens, and have abnormal gynoecia with aberrant placental and ovule development. In summary, the embryonic and floral phenotypes of cuc1 cuc2, nam and cup are similar, but only cup mutants show the striking vegetative defects, which include repetitive generation of cup-tipped shoots from the hypocotyl, extensive fusion of leaves, loss of axillary meristems and fasciation.

NAM, CUC1, CUC2 and CUC3 all encode NAC-domain proteins(Aida et al., 1997). NAC-domain proteins form a large family of transcription factors, with more than 100 members in Arabidopsis. The similarity between embryonic and floral aspects of the cup, cuc1 cuc2 and nam phenotypes raised the possibility that the cup phenotype resulted from a defect in a related NAC-domain gene. To test this hypothesis, 25 AntirrhinumNAC-domain genes were isolated from a cDNA library using a mixture of heterologous NAC-domain genes as a probe (Accession Numbers: AJ568261-AJ568281 and AJ568337-AJ568340). The closest match to NAM and CUC2was tentatively named CUP (Fig. 3A), and was used as a probe on Southern blots from segregating cup populations. Both alleles of cup showed co-segregating polymorphisms, suggesting that they were caused by transposon insertions. PCR analysis with gene specific and generic transposon primers identified insertions of transposons in both alleles(Fig. 3B). Molecular and genetic analysis confirmed that escape shoots formed in each allele are not revertants. In the absence of reversion of either allele, co-segregation analysis was extended to 73 plants in three segregating populations. In all cases, the transposon insertion showed absolute cosegregation with the mutant phenotype. Northern blot analysis confirmed that the CUP gene is not expressed in either cup allele(Fig. 3C). Taken together, the similarity of CUP to NAM and the CUC factors, the presence of transposon insertions in both alleles of cup, the absolute co-segregation of the insertions with the mutant phenotype and the lack of CUP expression in the homozygous cup mutants indicates that loss of CUPexpression is responsible for the phenotypes observed in the cupmutants.

CUP expression was investigated by in situ hybridisation. Expression was detected throughout all stages of development. In young seedlings CUP is expressed in a band of cells surrounding the apex,and between the apex and young leaves and leaf primordia(Fig. 4A). On flowering, CUP expression is observed between the inflorescence apex and the bracts (Fig. 4B). As floral meristems form between the bracts and the inflorescence meristem, CUPis expressed between the floral meristem and the inflorescence meristem, and expression is lost between the floral meristem and the maturing bract. Floral organ primordia form on the floral meristem, and CUP expression is seen between the earliest formed sepal primordia and the floral meristem at late stage 3 (Carpenter et al.,1995) (Fig. 4C). By stage 5, the primordia in whorls 2, 3 and 4 are beginning to form, and CUP expression is observed at the boundaries of each primordium(Fig. 4D). CUPexpression thus separates primordia between whorls and within whorls(Fig. 4D,E). Later in flower development, during the maturing of the floral organs, CUP expression is seen in stamens and carpels. A band of CUP expression is observed between the sites of the developing locules of the anthers and, in carpels, CUP is expressed at the tip of the developing placenta(Fig. 4F). CUPexpression is lost in the anthers at later stages, as the locules form, but expression continues in the carpel. CUP expression is apparent between ovule primordia, with weaker transient expression within the primordia themselves (Fig. 4G). CUP expression is also observed in more mature ovules in a ring surrounding the nucellus, at the point of attachment of the inner integument(Fig. 4H). The expression patterns of NAM and the CUC genes have been previously reported (Souer et al., 1996; Aida et al., 1997; Ishida,2000; Takada et al., 2001) and show striking similarities to the expression pattern of CUP.

In summary, CUP expression is observed at the boundaries between all developing lateral organs. This expression includes boundaries between the same organ types, between organs in different floral whorls, and between developing organs and adjacent meristems. In contrast to nam and cuc1 cuc2, cup mutants show phenotypic abnormalities associated with all areas of observed gene expression.

Three different species, Petunia, Arabidopsis and Antirrhinum have now been shown to use related NAC-domain genes to establish the boundaries between lateral organs. In an attempt to understand how expression of these genes, in cells destined not to develop into organs,could result in the formation of discrete boundaries, a yeast two-hybrid screen was performed using CUP as bait. Eight putative CUP interactors were isolated from the 5×10⁶ colonies screened. DNA sequence analysis showed that six of these corresponded to known yeast two-hybrid false-positive sequences, leaving two candidate interactors. One of these cDNAs was unable to encode a protein because of stop codons in all reading frames and the other was found to encode a truncated member of the PCF subclade of the TCP family of transcription factors(Cubas et al., 1999), which we designated TIC (TCP-Interacting with CUP). A cDNA corresponding to full-length TIC was subsequently isolated (Accession Number:AJ580844). Full-length TIC encodes a 398 amino acid protein with a characteristic TCP-domain between residues 76-130. The interaction between CUP and full-length TIC was confirmed by GST pulldown(Fig. 5A). Interaction between CUP and TIC was detectable when either partner was expressed as a GST fusion. In addition, interaction was also detected between labelled TIC and the TIC-GST fusion protein, suggesting that TIC is also capable of interacting with itself. Homodimerisation and heterodimerisation between TCP-domain factors has previously been reported(Kosugi and Ohashi, 2002). TIC expression appears to be too low to be detected by in situ hybridisation, as no signal could be detected. However, RT-PCR analysis confirmed that TIC expression overlaps with that of CUP(Fig. 5B).

In theory, the boundaries of lateral organs could be defined either from within the initiated primordium itself, by external factors alone, or by a combination of both internal and external factors. In the first case it would be sufficient to confer an organ identity on a cell or group of cells. In this way, organ boundaries would be determined by the site of initiation of lateral primordia and the patterns of cell division within them. However, at least in plants, this appears not to be the case. Mutants in several species have now demonstrated a requirement for an external boundary function. It is not clear that such a requirement exists in animal development. To our knowledge, no external boundary function has yet been demonstrated in, for example, the establishment of limb buds. If this represents a true mechanistic difference between animal and plant development, the requirement for fixed external boundaries in plant development could be a consequence of the fact that organ primordia arise in the peripheral zone of the SAM, where cells are actively dividing and differentiating.

It appears that the determination of lateral organ boundaries in plants is controlled by a class of plant-specific transcription factors: the NAC-domain factors. Although NAC-domain genes form a large gene family, little is known about their functions. The results presented here support the idea that members of this gene family are functionally redundant. The Antirrhinum cup mutant affects lateral organ boundaries throughout the plant,suggesting that the lack of vegetative and inflorescence phenotypes seen in nam and cuc1 cuc2 are due to redundancy within this family. Redundancy of gene function has been previously postulated for NAM,both to explain the discrepancy between its expression pattern and sites of phenotypic abnormality (Souer et al.,1996), and because co-suppression experiments, in which several NAC-domain genes are silenced, result in a lack of lateral meristems(Souer et al., 1998). Redundancy is also apparent amongst the Arabidopsis CUC genes, but even in cuc1 cuc2 double mutants vegetative and inflorescence development is unaffected (Aida et al.,1997). Preliminary phenotypic observations have been reported to suggest that cuc3 plays a role in the establishment of boundaries in adult plants (Vroemen et al.,2003). It therefore appears that less redundancy exists for this function in Antirrhinum, where the loss of a single gene affects all lateral organ boundaries. Our low stringency heterologous screen for NAM/CUC related genes in Antirrhinum identified 25 different NAC-domain containing genes, but only CUP belonged to the NAM/CUC subclade of this large gene family. This further supports the view that less redundancy exists for this function in Antirrhinum than in Arabidopsis. However, as lateral meristems can occasionally form in cup mutants, and fusion is not observed between all lateral organs, it remains possible that another function is also involved in specifying lateral organ boundaries. This could be a member of the NAC-domain family or an unrelated factor.

The finding that CUP interacts with a TCP-domain factor suggests a model for the establishment of lateral organ boundaries. The best characterised members of the TCP-domain factor family are CYCLOIDEA (CYC) and TEOSINTE BRANCHED 1 (TB1) (Luo et al.,1996; Doebley et al.,1997). A common feature of both these TCP-domain factors is their role in the prevention of organ outgrowth(Cubas et al., 1999). CYC acts early in flower development to reduce the growth rate in the dorsal part of the floral meristem. Later, CYC activity prevents the growth of the dorsal stamen. Similarly in maize, TB1 activity prevents the outgrowth of lateral branches. Another AntirrhinumTCP-domain gene, CINCINNATA (CIN), provides further evidence for the connection between this gene family and the regulation of cell division (Nath et al., 2003). The leaves of cin mutants are wrinkled rather than flat because of a failure to restrict cell division in the leaf margins. During the development of leaves in wild-type plants, a front of cell cycle arrest moves progressively from the tip of the leaf towards the base. In cinmutants this cell cycle arrest front moves more slowly, resulting in excessive cell division before eventual arrest. The expression pattern of CINin developing leaves suggests that CIN might act to sensitise cells to the arrest signal (Nath et al.,2003). Thus, mutational studies in different species have identified TCP-domain genes as negative regulators of growth.

Although the mechanism by which these TCP-factors might control cell division is not yet understood, the other founder members of this family, PCF1 and PCF2, could provide a clue. PCF1 and PCF2 were identified in a screen for factors which bind to sites in the promoter of the rice proliferating cell nuclear antigen (PCNA)gene, which itself has a role in DNA-synthesis and cell cycle control(Kosugi and Ohashi, 1997). More recently an Arabidopsis TCP-domain protein, At-TCP20, was shown to bind to a motif in the promoter of Arabidopsis PCNA, and in the promoters of other genes regulated during the G1-S phase transition(Trémousaygue et al.,2003). All three of these TCP-domain factors, which belong to the same PCF subclade of TCP-domain genes as TIC, are suggested to activate expression in meristematic cells, but it is also possible that interaction with other factors could result in repression (or lack of activation) of target genes. This leads us to postulate a model whereby the NAC-domain factor CUP, expressed in specific boundary domains surrounding all lateral organs,interacts with the more widely expressed TCP-domain factor TIC, resulting in either repression, or failure of activation, of genes which promote cell division. As a consequence of this interaction, cell division is reduced between developing meristems and primordia, resulting in the formation of discrete lateral organs with defined boundaries.

The interaction between CUP and the TCP-domain factor TIC also provides a possible explanation for the radialisation seen in cup flowers. Radialisation of the flower is found in cyc and dichotoma(dich) mutants (Luo et al.,1999), which result from defects in TCP-domain encoding genes. The radialisation found in cup mutants could be explained in a variety of ways, including the direct interaction of CUP and CYC, and/or DICH, or an alteration in the stoichiometry of interactions within the TCP family following the removal of a TCP-factor interactor. Alternatively, radialisation might be a consequence of the loss of organ boundaries and breakdown in organisation of the floral meristem. Preliminary yeast two-hybrid experiments suggest that CUP is indeed capable of interacting with CYC (J.L. and B.D.,unpublished). As we have discovered a direct protein-protein interaction between members of the NAC-domain and TCP-domain families it is possible that other members of these families also interact.

The NAC-domain genes NAM, CUC1, CUC2, CUC3 and CUP act to establish boundaries between lateral organs. The extreme phenotype observed in cup mutants supports the idea that further redundancy will be discovered amongst these genes in Petunia and Arabidopsis,and suggests that members of this gene family regulate the separation of all lateral organs in plants. The interaction between NAC-domain factors and TCP-factors provides a model to explain how boundaries might be established,but this is unlikely to be the whole picture. Several other genes, such as LATERAL ORGAN BOUNDARIES (LOB), are also expressed at boundaries (Shuai et al.,2002). Further work is required to assess the contribution of all these genes to the establishment of organs boundaries and meristem formation.

We acknowledge Mr Adrian Hick for help with the SEM analysis and Dr Hans Sommer for the Antirrhinum cDNA library. We thank Dr M. Kieffer for helpful discussions. I.W., J.L. and B.C. were funded by the BBSRC. Collaboration between B.D. and Z.S.-S. is supported by a British Council ARC project grant.