The structural homology of the daffodil corona has remained a source of debate throughout the history of botany. Over the years it has been separately referred to as a modified petal stipule, stamen and tepal. Here we provide insights from anatomy and molecular studies to clarify the early developmental stages and position of corona initiation in Narcissus bulbocodium. We demonstrate that the corona initiates as six separate anlagen from hypanthial tissue between the stamens and perianth. Scanning electron microscope images and serial sections demonstrate that corona initiation occurs late in development, after the other floral whorls are fully developed. To define more precisely the identity of the floral structures, daffodil orthologues of the ABC floral organ identity genes were isolated and expression patterns were examined in perianth, stamens, carpel, hypanthial tube and corona tissue. Coupled with in situ hybridisation experiments, these analyses showed that the expression pattern of the C-class gene NbAGAMOUS in the corona is more similar to that of the stamens than that of the tepals. In combination, our results demonstrate that the corona of the daffodil N. bulbocodium exhibits stamen-like identity, develops independently from the orthodox floral whorls and is best interpreted as a late elaboration of the region between the petals and stamens associated with epigyny and the hypanthium.
If you can't find a tool you're looking for, please click the link at the top of the page to "Go to old article view". Alternatively, view our Knowledge Base articles for additional help. Your feedback is important to us, so please let us know if you have comments or ideas for improvement.
The flower has been widely implicated in the adaptive success of the angiosperms (Stebbins, 1981; Crane et al., 1995; Magallón and Sanderson, 2001; Crepet and Niklas, 2009). The bewildering array of floral morphologies (Judd et al., 1999; Endress, 2001) emphasises the significance of developmental plasticity as a driver of evolutionary change and diversification (Magallón and Sanderson, 2001; Endress, 2006). While the canonical floral structure consists of four whorls – an outermost whorl comprising sepals, a second of petals, a third of stamens and an innermost whorl of carpels – modification of this basic ground plan through processes such as reduction, fusion and homeosis are commonplace (Satler, 1988; Coen, 1991; Endress, 2006). Petaloid monocots, such as tulips and hyacinths, exhibit deviation from the typical floral ground plan in that the two outer whorls of the flower have similar identity, and the organs are collectively termed tepals rather than sepals and petals (Remizowa et al., 2010). Such deviations from the common bauplan necessitate a consideration of both positional homology (Owen, 1843; Patterson, 1982) and floral organ identity (Remane, 1952) when interpreting the diversity in floral anatomy among angiosperms.
Early genetic studies in Arabidopsis thaliana and Antirrhinum majus identified three classes of homeotic genes that are differentially expressed throughout the four floral whorls. The ABC model of floral development holds that the individual and joint actions of A-, B- and C-class genes confer floral organ identity, with A-class specifying sepals, A- and B-class together specifying petals, B- and C-class together specifying stamens, and C-class specifying carpels (Bowman et al., 1991b; Coen and Meyerowitz, 1991). In Arabidopsis, A-class function is encoded by APETALA1 (AP1) and AP2, B-class by AP3 and PISTILLATA (PI), and C-class by AGAMOUS (AG). With the exception of AP2, the ABC genes all encode MADS-box transcription factors. The ABC model has since been extended to include the SEPALLATA (SEP) MADS-box genes, which are most closely related to the AP1 gene family. SEP genes are expressed prior to the onset of organ identity genes and are required to establish the floral context in which the organ identity genes can function (Pelaz et al., 2000; Ditta et al., 2004). Functionally, SEP proteins form complexes with B- and C-class proteins to regulate gene expression and promote organ identity (Honma and Goto, 2001; Immink et al., 2009). The phenotypes of A-, B-, and C-class mutants – in which floral organs are absent or mis-specified – have led to a conceptual framework whereby the MADS-box transcription factors act as developmental switches that activate organ specification programs. These programs appear to be well conserved among angiosperms, not only because of consistent mutant phenotypes between Arabidopsis, Antirrhinum, rice and maize (Sommer et al., 1990; Yanofsky et al., 1990; Ambrose et al., 2000; Nagasawa et al., 2003), but also because the patterns of ABC gene expression in non-model angiosperms generally correspond with floral anatomy (Theissen and Melzer, 2007; Litt and Kramer, 2010). As a result, the expression patterns of ABC orthologues have been used to infer the homology of floral structures in unusual or non-canonical floral morphologies (Ambrose et al., 2000; Whipple et al., 2004; Álvarez-Buylla et al., 2010; Hemingway et al., 2011).
Flowers of the genus Narcissus (the daffodils) possess a distinct fifth organ termed the corona, a cup-shaped or crown-like structure located between the tepals and the stamens. The corona has a homoplastic distribution among the Amaryllidaceae, suggesting that this structure has evolved several times independently (Figure 1a). The identity and homology of the daffodil corona has been a subject of debate for over a century. For example, some have interpreted the corona as an outgrowth of tepals equivalent to ligules of leaves (Döll, 1857) or confluent petal stipules (Smith, 1866) and others as a modification of stamens (Masters, 1865; Goebel, 1933). In an influential paper Agnes Arber stated that ‘the corona of Narcissus has no connexion with the stamens, but is split off from the inner surface of the perianth tube’ (Arber, 1937, pp. 300–301). Some of the controversy regarding the interpretation of the daffodil corona has arisen from comparison with the coronas observed in several other genera of Amaryllidaceae, including Eucharis, Hymenocallis and Pancratium. In most of these genera, the anthers are positioned at the apex of the corona (Figure 1b), suggesting that the corona arises from growth associated with the staminal filaments. In contrast, the anthers of daffodils are attached at or towards the base of the flower rather than at the apex of the corona (Figure 1c). Such morphological differences have led to the distinction between staminal and perianthal coronas within Amaryllidaceae, dependent on the extent and position of the corona (Table S1 in Supporting Information). Arber (1937) and others (Chen, 1971) highlighted the differences in arrangement and orientation of vascular bundles in taxa with perianthal and staminal coronas. Staminal coronas have vascular bundles restricted to the stamens with no vasculature in the corona itself. In contrast, the corona of a daffodil has numerous bundles which have inverted orientation from the bundle serving the perianth. Considering Arber's (1937) assertion that the daffodil corona is derived from the perianth tube, the question of whether the corona is staminal in origin or an evolutionarily novel structure remains unresolved.
Given the variety of floral forms among the Amaryllidaceae, the debate over the homology of the corona cannot be resolved on the basis of comparative anatomy alone. Here, we present a detailed anatomical study of early floral development in the hoop-coat daffodil, Narcissus bulbocodium. Combined with expression analysis of floral organ identity genes, we show that despite its petaloid appearance, the daffodil corona shares greater identity with stamens than it does with tepals. Our observations imply that staminal and petaloid coronas may share common structural identity through unique modifications of pre-existing developmental programs.
The corona initiates late in floral development
Although daffodils flower conspicuously in the spring, the transition from vegetative to reproductive growth occurs during the quiescent phase at the end of the previous growing season, such that the floral bud is fully formed within the dormant bulb (Noy-Porat et al., 2009). As such, we dissected buds from dormant bulbs to establish the relative timing of floral organ development.
At the earliest stage, the floral meristem is identifiable as a flattened apex that has initiated bracts, but floral organs have not yet developed (Figure S1a). The successive appearance of floral organ primordia then starts from the meristem periphery and continues inwards towards the central apical region (Figure S1b), later resulting in the individually distinguishable organs of whorls 1–4 (Figure S1c). By this stage, the main floral organs (tepals, stamens and carpels) are fully differentiated, but there is no evidence of a corona (Figure 2a,b). At a slightly later stage, however, the corona becomes visible as an undulating sheet of tissue that emerges between the tepals and the stamens (Figure 2c,d). Notably, the ridges of the corona coincide with the inter-staminal gaps (Figure 2d), suggesting that the corona either initiates or grows unevenly around its circumference. From these observations we conclude that the corona differentiates late in floral development, after the main floral ground plan is fully formed.
To define the timing and location of corona emergence more precisely, we studied serial sections of flowers at a stage shortly after full differentiation of the main floral whorls. The first visible evidence for the initiation of the corona is small foci of densely stained cells, or anlagen, on the inner surface of the hypanthium between the tepals and the insertion point of the stamens (Figure 3a). (The hypanthium is a specialised structure formed by fusions at the base of the floral whorls (Figure S1c) (Kaplan, 1967; Chen, 1971; Soltis and Hufford, 2002).) These anlagen later develop into primordia that protrude from the hypanthial tube (Figure 3b). While examining serial longitudinal sections of flowers at this stage, we noticed that the corona primordia are not visible in regions that are adjacent to a staminal filament, but instead are prominent in regions between filaments (Figure 3c,d). In transverse sections, the anlagen become visible on the inner surface of the hypanthial tube opposite longitudinal invaginations of the hypanthium that correspond to the edges of adjacent tepals (Figure 3e). As the flower grows and the anlagen develop into coronal primordia, the invaginations spread laterally, forming a groove that begins to separate the hypanthium in two (Figure 3f). Finally, as the primordia extend vertically and laterally, the split completes and the corona forms a coherent ring of tissue between the tepals and stamens (Figure 3g, h). Together, these observations suggest that coronal primordia initiate at the six inter-staminal locations on the inner hypanthial wall at a vertical point that is between the stamens and the tepals. The initiation of growth at six points that later join together laterally is consistent with the ridged appearance of the corona during early growth phases (Figure 2d).
An important issue of interpretation concerns the nature of the hypanthium in epigynous flowers, where floral organs are attached near the top of the ovary. In these cases, the hypanthium has been variously characterised as either appendicular (outer floral whorls are adnate to one another in the ovary region and adnate to the ovary wall) or receptacular (following initiation of calyx, corolla, androecium and gynoecium primordia, peripheral upgrowth around the apical meristem lifts the base of the perianth and androecium to a level above the base of the ovary) (Kaplan, 1967; Soltis and Hufford, 2002). Along with the inferior ovaries in many other angiosperm taxa with inferior ovaries the hypanthial tissue of Narcissus has been characterised as appendicular (Chen, 1971), formed from the fusion of floral whorls outside of the carpels (Figure S1c). This is notable because it is from this tissue that the corona initiates.
The corona shares AGAMOUS gene expression patterns with the stamens and hypanthium, but not the tepals
To resolve the uncertainty over the homology of the daffodil corona, we cloned orthologous gene sequences of the major ABC classes of floral identity genes. We first generated multiple sequence alignments of monocot orthologues of the Arabidopsis genes AP1 (A-class), AP3 (B-class), PI (B-class) and AG (C-class) in order to design degenerate primers for PCR. Fragments of putative ABC orthologues were amplified from N. bulbocodium, sequenced, aligned with other members of each class and subjected to phylogenetic analysis. Four N. bulbocodium sequences were amplified, each of which clustered with one of the targeted sub-families of MADS-box proteins (Figures 4 and Supporting Data S1 and Table S3). Notably, these sequences consistently grouped with members from rice. While we cannot rule out the possibility of gene paralogy in Narcissus, as is apparent in rice and Vitis, these data support the conclusion that the identified sequences are probably N. bulbocodium orthologues.
To characterise gene expression patterns of the ABC orthologs in N. bulbocodium, we first carried out gel blot analysis with RNA extracted from the various organs of mature flowers that had emerged from the bulb in the spring. Transcripts of the C-class gene NbAG were detected at high levels in carpels and the corona, and to a lesser extent in stamens and hypanthial tissue at the base of the flower (Figure 5). As expected, NbAG transcripts were undetectable in the tepals. In contrast, the B-class transcripts NbPI and NbAP3 were not detected in the carpel, but were detected at high levels in the corona and at lower levels elsewhere, including the tepals. The presence of B-class transcripts in the outer whorl is consistent with the modified ABC model (van Tunen et al., 1993; Theissen et al., 2007) in which expansion of the B-class boundary to the outer whorl specifies the development of petaloid sepals as previously demonstrated in tulip and lily (Winter et al., 2002; Kanno et al., 2003, 2007) that are indistinguishable from the ‘true petals’ of whorl 2 (Krizek and Meyerowitz, 1996; Bowman, 1997). A-class NbAP1 transcripts were detected evenly throughout all whorls (Figure 5), which is consistent with what has been observed for AP1 lineage members outside the core eudicots (reviewed Litt and Kramer, 2010). Weak expression levels in the stamens and hypanthium of mature flower buds make it difficult to compare precisely the identity of these tissues with the corona. However, based primarily on the accumulation of NbAG transcripts in the corona but not in the tepals, these data indicate that the tissue identity of the corona is most like that of stamens, and is distinct from that of tepals. It is also notable that the hypanthium, from which the corona develops, shares a similar ABC gene expression profile with both the corona and stamens. This would be expected in the case of appendicular epigyny as the hypanthium is formed from the fusion of non-carpel floral whorls.
NbAGAMOUS is expressed in the corona during early development
Determining the identity of tissue types based on expression profiles of ABC genes in mature floral organs is possibly problematic, because organ identity is specified early in ontogeny and the corresponding ABC gene expression domains may later change (Kramer et al., 2007; Brockington et al., 2012). Therefore, because the presence of NbAG transcripts is potentially diagnostic of stamen-like coronal identity, we examined NbAG expression in situ in developing flower buds. At early stages of development, after the main whorls are established but prior to emergence of the corona, NbAG transcripts were detected in the carpel, staminal filaments and the hypanthium (Figure 6a). At this stage, NbAG transcripts are spatially restricted to vascular tissue and the epidermis in carpels and filaments, and to the vasculature in the hypanthium (Figure 6a,b). However, shortly after the point at which the corona emerges from the hypanthium and separates from the tepals, NbAG transcripts were abundant throughout the corona and carpel (Figure 6c). Crucially, compared with the corona, NbAG transcripts were barely detectable in the tepals, and were restricted to the vasculature alone (Figure 6d). Thus, the expression pattern of NbAG in developing flowers is consistent with that observed in the various organs of mature flowers (Figure 5), further implying that the corona has stamen-like tissue identity.
We have shown that in N. bulbocodium all of the canonical floral organs are well-established before the corona begins to develop (Arber, 1937; Chen, 1971). Initiation occurs from six anlagen (Figure 3a–f), situated between the second whorl tepals and stamens (Figure 7d), and outgrowths of tissue from these anlagen merge to form a coherent corona (Figures 2d and 3g,h). Overall, these observations demonstrate that the corona is neither a simple outgrowth nor a modified part of any existing whorl, but is instead a novel independent structure that develops from cells of the hypanthium. Given that the corona is initiated both spatially and temporally apart from the tepals and that transcripts of the C-class gene AGAMOUS accumulate in the corona, homology between the corona and tepals can most likely be rejected. However, because expression studies demonstrate that the corona expresses a qualitatively similar ABC gene profile combination as the stamens, the relationship between stamens and the corona is less straightforward. We propose that the corona of N. bulbocodium is best interpreted as a late elaboration of the region between the petals and stamens that is associated with epigyny and the hypanthium. This proposal is in agreement with observations of the corona in Velloziaceae, where coronal appendages were shown to be closely associated with the stamens and but also share vasculature with the tepals and develop late in ontogeny (Sajo et al., 2010). Our interpretation means that the ‘so-called’ staminal coronas, as seen in most genera of Amaryllidaceae (Figure 1c), reflect small differences in timing and position of corona elaboration. Indeed, this interpretation of the various forms of corona in Amaryllidaceae, as variations on the same theme, was clearly illustrated by (Goebel, 1933) (Figure S2). The opposite interpretation, based on observations of mature flower sections (Arber, 1937), namely, that the daffodil corona is an outgrowth of the perianth tube, holds that the stamens separate at a lower point on the hypanthium (Figure 7b). Such an interpretation suggests that the corona would be derived more readily from the perianth. However, when seen in serial sections from below, the separation of carpels, then stamens, then tepals represents nothing more than the incline of the hypanthial cup. Because the corona initiates between the stamens and tepals (Figures 2 and 3), by default it would appear to be derived from the perianth in this trajectory. In fact, Arber's interpretation that the corona was derived from the perianth tube is similar to our interpretation because the only difference between those views is a terminological one regarding the same tissue and whether it is a perianth tube or a hypanthial tube. If the hypanthial region is formed from adnation of non-carpel floral whorls to provide the outer wall of the epigynous ovary, then the hypanthium and the hypanthial tube would express a combination of genes found in the perianth and the stamens, which is consistent with what we report here.
In emergent flowers, we found that B- and C-class expression was relatively weak in the stamens compared to the corona (Figure 5). One possible explanation for this difference is that the anthers, which do not strongly express NbAG (Figure 6), account for much of the stamenal mass relative to the strongly AG-expressing filaments. The spatial domain of AG expression in the anthers is variable among species: while AG is expressed in anther walls and the filaments of mature flower buds in both Arabidopsis (Bowman et al., 1991a) and tomato (Pnueli et al., 1994), PLENA expression in antirrhinum stamens is limited to the central stomium and not the anther walls (Davies et al., 1999). Another plausible explanation for the weak B- and C-class expression in stamens relates to the seasonal development of the daffodil flower. Variation in absolute abundances of ABC transcripts between different floral organs and over developmental time in flowers with long quiescent periods is not without precedent. For example, in the woody legume Sophora tetraptera, StAP1 and StAG were most active in the winter stages of flower development. Expression of these transcripts varied 15-fold over the course of flower development, and notably were very weakly expressed in mature flowers (Song et al., 2008). We suggest that weak expression of B- and C-class genes in the mature stamens of Narcissus may partly reflect the early organ specification and development that occur during the autumn and winter. In addition, not only is the corona specified later in development than the stamens, but also the corona undergoes extensive post-specification growth between bud emergence in the spring and when the flower is fully open. Thus, we suspect that the relatively high transcriptional activity of floral identity genes in the corona relative to the other floral organs reflects the late burst of coronal growth in the spring.
Recent research on passion flowers (Passiflora; Hemingway et al., 2011), which also have a corona, show similarities and differences with the results reported here for Narcissus. They differ in that the corona of passion flowers is a composite structure of some developmental complexity. It also forms from a ring of tissue rather than separate anlagen as in Narcissus. However, the passion flower corona initiates between the perianth and the stamens late in floral development, after the formation of the petals, stamens and carpels as in Narcissus and Vellozia (Sajo et al., 2010). The corona in Passiflora also expresses homologues of B- and C-class genes as does Narcissus. It seems that the corona of both Passiflora and Narcissus represents a late elaboration of the stamens and/or hypanthium that expresses AG even though the structures are not fertile and are markedly petaloid. This suggests a convergence of phenotype whereby petaloid tissue can be produced by different genetic pathways even within the same flower.
Dormant bulbs of N. bulbocodium var. conspicuus (the hoop petticoat daffodil) were obtained from Peter Nyssen bulbs (http://www.peternyssen.com/) and stored at 4°C in the dark before dissection for floral buds. To obtain leaf and floral tissue for nucleic acid isolation, bulbs were planted in soil in September 2005 and grown outdoors in Oxford, UK. Tissue was harvested in April 2006.
Tissue embedding and sectioning
For histological samples, excised buds were fixed in FAA (3.7% formaldehyde, 5% glacial acetic acid, 50% ethanol) overnight at room temperature (20-22°C). The tissue was dehydrated through a graded ethanol series, stained with 0.1% eosin in 95% ethanol, and embedded in Paraplast Plus (Leica, http://www.leica.com/) as recommended by the manufacturer. Embedded samples were sectioned into an 8 μm-thick ribbon, placed on microscopy slides, floated on distilled water and gently heated to adhere the sections to the glass. Sections were pre-treated with 100% Histo-Clear (Fisher Scientific, UK, http://www.fisher.co.uk/) followed by 100% ethanol. They were then stained with Safranin O, cleared with ethanol and counterstained with Fast Green FCF (Ruzin, 1999). The slides were left to dry for 2 h. To mount the sections, Entellan (Merck Millipore, http://www.merckmillipore.com/) was mixed with Histo-Clear in the ratio 3:1 (v/v) and used to seal a cover slip on to the slide. The slides were then left for 24 h to dry.
Scanning electron microscopy
Flower buds were carefully dissected from whole bulbs, fixed in FAA, and dehydrated through a graded ethanol series to 100% ethanol. They were then critical-point dried with a Leica Microsystems (http://www.leica-microsystems.com) CPD-030 critical point dryer, mounted onto stubs, coated with platinum using an Quorumtech (http://www.quorumtech.com) K550 sputter-coater, and examined using a JEOL JSM-5510 microscope (http://www.jeol.com/).
In situ hybridisation
A DNA fragment encompassing 282 bp of the NbAG coding region plus 260 bp of the 3′ untranslated region (UTR) was cloned bidirectionally into the pGEM-T Easy vector (Promega, http://www.promega.com/). Antisense and sense riboprobes were synthesised and labeled in vitro using the T7 promoter and digoxigenin-11-UTP (Roche, http://www.roche.com/). Riboprobes were generated in a 25-μl reaction containing 0.1 mm digoxigenin-UTP, 0.5 mm (each) ATP, CTP and GTP, and 1 μg plasmid DNA template. Riboprobes were hydrolysed in 200 mm carbonate buffer (pH 10.2) for 60 min at 60°C, precipitated, resuspended in 50 μl water and assayed for label incorporation.
Excised flowers were fixed with 4% paraformaldehyde, 0.1% Tween-20 and 0.1% Triton X-100 overnight at 4°C. Samples were embedded and sectioned as above. Section pre-treatment, hybridisation and slide washing were performed as described by Jackson (1991), with additional modifications as described by Lincoln et al. (1994). Blocking steps and detection of labelled transcripts with anti-digoxigenin antibodies conjugated to alkaline phosphatase (Roche) were performed as described by Coen et al. (1990). Sense and antisense probes against NbAG were hybridised in parallel; no discernible signal was detected for the sense probe (Figure S3).
Cloning of N. bulbocodium ABC orthologues
Orthologous sequences were amplified in a single round of PCR using N. bulbocodium genomic DNA as a template with degenerate oligonucleotides, which were designed against conserved regions with the aid of alignments of multiple nucleotide sequences obtained by BLASTn searches of sequence databases. The alignments generated, including GenBank accession numbers and the relative position of the primers, are shown in Data S2. The PCR products were cloned and verified by sequencing. Sequences for PI, AP3 and AG were extended by 3′ rapid amplification of complementary DNA ends (RACE) using gene-specific nested primers. Narcissus bulbocodium cDNA was generated from 1 μg total RNA from unopened flower buds with SuperScript II reverse transcriptase (Life Technologies, http://www.lifetechnologies.com/) and an oligo(dT) anchor primer. The cDNA sequence for AP3 was further extended by anchor-ligation-mediated 5′ RACE using the FirstChoice® RLM-RACE kit (Life Technologies). Oligonucleotide sequences are detailed in Table S4.
Homologues of Arabidopsis AP1, AP3, PI and AG from multiple plant genomes were identified by BLASTn searches. A subset of results was chosen to select representatives from monocots and dicots. Alignments were generated with Muscle v.3.6 (Edgar, 2004) and phylogenies inferred with MrBayes v.3.1.2 (Ronquist and Huelsenbeck, 2003). Bayesian run parameters were as follows: the data were assigned a model using modeltest (Posada and Crandall, 1998) implemented in paup* (Swofford, 2003), using the Akaike information criterion (Akaike, 1973) to choose between models. Subsequently the analysis was run with a six-parameter nucleotide substitution matrix (nst = 6) and a gamma-shaped rate variation with a proportion of invariable sites. The analysis comprised two runs, each with four chains (three hot, one cold) run for 2 000 000 generations, sampling every 100 trees, and discarding the first 20% of trees as the burn-in. Both runs were considered to have reached convergence when the standard deviation of the split frequencies was lower than 0.01. A 50% majority rule consensus tree was constructed to obtain posterior probabilities as a measure of node support. For the MADS genes phylogeny based on amino acid data (Figure 4), the mixed models option (aamodelpr = mixed) was used, otherwise the run conditions were as above. Analyses using raxml (Stamatakis et al., 2008) corroborated the results from Bayesian analysis, showing only very minor differences in topology.
Total RNA from floral structures and leaves was isolated by guanidium thiocyanate–phenol–chloroform extraction (Chomczynski and Sacchi, 2006). The RNA was precipitated with 25% (v/v) isopropanol and 25% (v/v) salt solution (1.2 m NaCl and 0.8 m sodium citrate). Gel electrophoresis, blotting, hybridisation and washing were performed as described (Langdale et al., 1988). Signal was detected with a phosphorimager (Bio-Rad FX, http://www.bio-rad.com/). DNA fragments used to generate 32P-radiolabelled probes were subcloned into pGEM-T Easy and amplified by PCR; the corresponding regions are shown in Data S2.
Sequence data originating from this study are deposited with GenBank under accession numbers: NbAPETALA1, JX678973; NbAPETALA3, JX678974; NbPISTILLATA, JX678975, NbAGAMOUS, JX678976.
We thank the Gatsby Charitable Foundation for funding to RWS and the National Science Foundation (USA) for grants DEB-968787 and 0129179 to AWM. Thanks also to Chris Della Vedova for initial training of RWS and Jill Harrison for advice and discussion. We also thank Paula Rudall for comments on the first draft of the paper and two anonymous reviewers for helpful suggestions that improved the paper.