•Pigment stripes associated with veins (venation) is a common flower colour pattern. The molecular genetics and function of venation were investigated in the genus Antirrhinum, in which venation is determined by Venosa (encoding an R2R3MYB transcription factor).
•Pollinator preferences were measured by field tests with Antirrhinum majus. Venosa function was examined using in situ hybridization and transient overexpression. The origin of the venation trait was examined by molecular phylogenetics.
•Venation and full-red flower colouration provide a comparable level of advantage for pollinator attraction relative to palely pigmented or white lines. Ectopic expression of Venosa confers pigmentation outside the veins. Venosa transcript is produced only in small areas of the corolla between the veins and the adaxial epidermis. Phylogenetic analyses suggest that venation patterning is an ancestral trait in Antirrhinum. Different accessions of three species with full-red pigmentation with or without venation patterning have been found.
•Epidermal-specific venation is defined through overlapping expression domains of the MYB (myoblastoma) and bHLH (basic Helix-Loop-Helix) co-regulators of anthocyanin biosynthesis, with the bHLH providing epidermal specificity and Venosa vein specificity. Venation may be the ancestral trait, with full-red pigmentation a derived, polyphyletic trait. Venation patterning is probably not fixed once species evolve full-red floral pigmentation.
When a pollinator forages, it optimizes its success by returning to flowers of the species from which it has previously found food (Chittka et al., 1999). This requires the pollinator to learn and memorize specific floral signals, such as flower morphology, scent, colour and patterns of pigmentation. Simple changes in flower colour, as a result of single gene changes, have been shown to have dramatic effects on the prevalence of visits by alternative pollinators, specifically, bees, hummingbirds or moths (Bradshaw & Schemske, 2003; Hoballah et al., 2007). In many species, flower colour involves not only the corolla-specific production of pigments but also patterning of pigmentation within the corolla. Developmentally programmed pigmentation patterns are of interest with respect to the evolution and genetic basis of specific floral morphology for specialized plant–pollinator associations (pollination syndromes), and the promotion of reproductive barriers that maintain or encourage different plant forms within a sexually compatible population and/or promote plant speciation (Grant, 1949, 1994). Floral pigmentation usually involves pigment formation (commonly anthocyanin pigments) in a single layer of cells (the epidermis) within which all cells are capable of pigment biosynthesis and all have a common developmental origin. This means that, where pigmentation patterns form, a nonclonal mechanism is likely to be involved in signalling to near-identical neighbouring cells in the epidermis to influence the type and amount of anthocyanins that the cells produce. Pigmentation patterning can therefore also provide an excellent model for dissecting regulatory signalling in plants.
Pigmentation patterns may involve colour variation between petals or differential colouration between individual cells within the corolla, to generate spots, stripes, irregular blotches, or combinations of these (Supporting Information Fig. S1). Particularly common patterns are bi-coloured petals, providing a ‘bull’s-eye’, blotches or stripes resembling stamens, and venation (Lunau, 2000). Patterning may increase successful pollination both by increasing the number of pollinator visits and the effectiveness of each visit in transferring pollen and by providing guides for the location of pollen, nectar or preferable landing regions of the flower (Sasaki & Takahashi, 2002; Medel et al., 2003; Heuschen et al., 2005; Lunau et al., 2006; Ushimaru et al., 2007). Pollen guides may be particularly important for species with zygomorphic flowers (e.g. Orchidaceae, Fabaceae and Plantaginaceae), as the stamens often require specific positioning of the pollinator or may be hidden from view (Ushimaru et al., 2007, 2009). In the genus Antirrhinum (Plantaginaceae syn. Antirrhinaceae), for example, the yellow floral guides may help to compensate for the visual obscurity of the enclosed anthers and pollen.
The genus Antirrhinum provides an excellent model for the study of the role of pigmentation in pollinator preferences and the associated regulation of pigment biosynthesis. This genus has radiated recently and rapidly and its pollination system may have played some part in this diversification. Europe has 20 distinct species of Antirrhinum (Sutton, 1988), with considerable variation existing among species in the intensity and patterning of the anthocyanin pigments in their flowers. Many are acyanic (anthocyanin lacking), or very palely pigmented with anthocyanin. However, several of these palely pigmented species exhibit strong venation patterning (pigment stripes associated with the veins). Antirrhinum pollination principally involves bumblebees because the closed flower requires the weight of the bee to open it. Floral traits are likely to have evolved, at least in part, in accordance with the trichromatic visual system of this pollinator (Glover & Martin, 1998; Hempel de Ibarra et al., 2002). Given the recent and rapid radiation of the Antirrhinum genus, and the ability of Antirrhinum species to form hybrids, obtaining a clear picture of the phylogenetic relationships between the species within the genus is challenging. Morphological classification of Antirrhinum species is complex, and has led to different views on the separation of the taxa and rank assignment (Sutton, 1988). However, DNA sequences from the third intron of the nitrate reductase (Nia) gene may provide phylogenetic resolution at low taxonomic levels (Howarth & Baum, 2002) and offer a basis for understanding the relationship between speciation and flower colour.
Antirrhinum majus is also well established as a model system for molecular studies of flower colour. As in other plant species, specific members of the R2R3MYB and bHLH transcription factor families interact to activate genes in the anthocyanin biosynthetic pathway, with the further involvement of a WD-repeat (WDR) protein that promotes the activity of the MYB–bHLH regulatory complex. In A. majus, a number of mutants or variant alleles that affect the activity of regulatory loci have been described, including delila, Eluta, rosea, Venosa and mutabilis (Martin et al., 1991; Martin & Gerats, 1993). Delila has been shown to encode a bHLH factor that is required for the activation of genes late in the biosynthetic pathway (late biosynthesis genes, LBGs) (Martin et al., 1991; Goodrich et al., 1992). However, loss of DELILA function results in loss of pigmentation only in the corolla tube as a second bHLH gene, Mutabilis, acts redundantly with Delila in activation of the LBGs in the lobe (Schwinn et al., 2006; C. Martin et al., unpublished data). The genes encoding the R2R3MYB transcription factors involved in the pigmentation of flowers have also been identified (Fig. 1) (Schwinn et al., 2006). The Rosea locus comprises two closely linked genes encoding structurally related R2R3MYB proteins, Rosea1 and Rosea2. Activity of Rosea1 gives strong, full-red corolla pigmentation in both the adaxial (inner) and abaxial (outer) epidermis. Activity of Rosea2 gives weak pigmentation, principally in the adaxial epidermis of the corolla lobes. A third gene, Venosa, encodes a structurally related R2R3MYB protein that controls the production of anthocyanins in adaxial epidermal cells that overlie the veins of the corolla (venation), a phenotype clearly visible only when Rosea1 is inactive (Fig. 1b,d,e). In Ve+ plants strong venation in the corolla extends down into the tube, and forms part of the visual floral guide to the enclosed stamens and nectar reward. The developmental origins of the Antirrhinum corolla lobes may serve to strengthen this visual cue, because the veins radiate out from the pinched region at the top of the tube, so providing visual guides directing the pollinator to the central landing platform and the entrance to the flower. Differences in anthocyanin production in flowers among six species of the genus Antirrhinum have been shown to be attributable to variations in the activity of the Rosea and Venosa loci, and variations in anthocyanin-related R2R3MYB gene activity (subgroup 6; Stracke et al., 2001) are a primary source of natural variations in flower colour/pigment patterning in plants (Schwinn et al., 2006).
Despite their common occurrence, the molecular basis of developmentally programmed floral pigmentation patterning, such as stripes and spots, is not well characterized, and the collective activity of the Rosea1, Rosea2 and Venosa genes represents one of the best defined systems. However, although the Venosa gene has been shown to be associated with venation patterns (Schwinn et al., 2006), how the gene determines this phenotype is not known. Similarly, the mechanisms by which other loci control pigment patterns at the molecular level are not known. The timing of transcript abundance for a light-induced anthocyanin-related R2R3MYB of Asiatic hybrid lily (Lilium spp.), LhMYB6, was recently shown to correlate with the appearance of tepal pigmentation spots in this species (Yamagishi et al., 2010). Also, Myb gene expression was detected in the purple spots of Phalaenopsis orchid petals, but not the surrounding white petal regions (Ma et al., 2009). Together, these observations suggest that R2R3MYB genes may be responsible for many flower colour patterns. In petunia, at least five loci (Ve1, Ve2, Ve3, Fine venation and Anthocyanin12) affect the incidence of venation in the corolla (Wiering & de Vlaming, 1984; Martin & Gerats, 1993), while a single gene probably defines the near-absence of the large, central, red-purple pigment spot in flowers of Clarkia gracilis subsp. gracilis compared with subspecies sonomensis (Gottlieb & Ford, 1988). The only floral pigmentation patterns for which molecular mechanisms are understood are some irregular patterns, for example colour flecks from transposon activity (Itoh et al., 2002), and nonclonal patterns such as ‘Red Star’ and ‘picotee’ patterns of petunia flowers which are a result of endogenous short-interfering RNA instability mechanisms (Koseki et al., 2005; Olbricht et al., 2006; Saito et al., 2006).
We have undertaken a comprehensive study of the mechanism and function of one of the most common flower colour patterns in plants, venation. Using the model genus Antirrhinum, we have determined the molecular basis of venation patterning and the effectiveness of venation for pollinator attraction. Using phylogenetic analyses we have also investigated the evolution of venation in the genus Antirrhinum, and its relationship to full-red pigmentation.
Materials and Methods
Antirrhinum majus L. lines AA114 (venation; Venosa+/roseadorsea (Ve+/rosdor)) and AA104 (no venation; ve−/rosdor) were grown in a glasshouse with ambient environmental lighting. The roseadorsea allele limits background pigmentation to the abaxial epidermis of the upper region of the dorsal petal, giving venation against an acyanic background. For the emasculation experiments, 15 plants were propagated through cuttings from a single plant (CT128) that had a strong venation phenotype. For in situ hybridization, plants were grown in the field. The origins of the different Antirrhinum species accessions used in this work were Antirrhinum majus subsp. majus var. majus (Toulouse and Barcelona), Antirrhinum graniticum Rothm. and Antirrhinum latifolium Mill. (Marseilles and Pyrenean), all from the Institut für Kulturpflanzenforschung, Gatersleben, Germany, Antirrhinum graniticum, Antirrhinum molle L., Antirrhinum mollissimum Rothm., Antirrhinum meonanthemum Hoffgg. & Link, Antirrhinum barrelieri Boreau. and Antirrhinum australe Rothm., collected by R. K. Oyama (vouchers deposited at the Herbarium of the Arnold Arboretum of Harvard University), and A. majus stock JI522 from the John Innes Centre. Flowers of the different species and lines used are shown in Figs S2 and S3.
To generate the plasmid containing 35S:gVenosa, a genomic fragment of Ve+ was PCR amplified using the 5′-primer (5′-GCCGGTACCATGGGAAATAATCCT CTTGGA-GTAAGAAAAGGC-3′) and the 3′-primer (5′-GGGTT-AATCTACGT AGTCCGCAAAATCGAGCAATTC-3′), and the KpnI/SmaI digested product cloned into pART7 (Gleave, 1992). The control construct for biolistic experiments was pLc349 (Ludwig et al., 1989), containing 35SCaMV:maize Leaf color (35S:Lc). pBluescriptKS+ (Stratagene, La Jolla, CA, USA) was used for generating RNA transcribed probes from a XhoI/SmaI 3′-fragment of the Ve+ cDNA that lacked significant sequence similarity to Rosea1 or Rosea2, and the full-length chalcone synthase (Chs) cDNA fragment from A. majus. The 35S:GFP construct was pPN93.
Biolistic transient assays
Particle bombardment experiments were conducted as in Schwinn et al. (2006), except that 4 μg rather than 10 μg of DNA was used per bombardment, with controls of gold particles alone, 35S:Lc and 35S:GFP constructs. After bombardment, tissue was cultured on half-strength Murashige and Skoog (MS) medium under 20–50 μmol m−2 s−1 light from Osram 36W grolux fluorescent tubes (Impel New Zealand, Auckland, New Zealand) (16 h photoperiod) at 25°C. At least two flowers were used for each construct per experiment, and each experiment was repeated at least twice.
In situ RNA hybridization analysis
Tissues were fixed and embedded as in Jackson et al. (1991, 1992) and sectioned at 8 μm using a Sorvall JB-4 microtome (ThermoFisher Scientific, Auckland, New Zealand) (Table 1). Digoxigenin-labelled RNA probes were prepared by in vitro transcription using the pBluescript T7 and SP6 promoters and a DIG RNA-Labelling Kit (Roche, New Zealand). Chemical hydrolysis was used to reduce the size of the 1.3-kb Chs transcripts to fragments of approx. 0.2 kb. Hybridization and detection were as described in Eason et al. (1996). Sections were counter-stained with safranin O and fast green (Best et al., 2004).
Table 1. Seven species of Antirrhinum used for in situ RNA hybridization analysis, and their flower phenotypes with respect to venation
Bud stage (mm)2
1See Supporting Information Fig. S2 for examples of the flowers.
2The size of the bud used for in situ hybridization analysis (Figs 3, 4).
A. majus, Venosa roseadorsea
Magenta pigmentation in the abaxial epidermis of the dorsal lobes; strong venation
A. majus, Venosa roseadorsea
Magenta pigmentation in the abaxial epidermis of the dorsal lobes; no venation
Very pale or no magenta pigmentation; no venation
No background magenta pigmentation; strong venation
No background magenta pigmentation; strong venation
Background magenta pigmentation with venation
No background magenta pigmentation; some venation
No background magenta pigmentation; strong venation
Fragments corresponding to the third intron of the Nia gene were amplified by PCR using degenerate primers designed by Howarth & Baum (2002). Their sequences included flanking portions of the third exon at the 5′-end of the third intron and a few bases of the fourth exon at the 3′-end (NIA-5, 5′-AAGTACTGGTGTTGGTGYTTTTGGTC-3′ and NIA-3, 5′-GGCATGATGAACAACTGCTGGTTC-3′). Fragments corresponding to 83 bp of exon 5, c. 340 bp of intron 5 and 469 bp of exon 6 of the dihydroflavonol 4-reductase (Dfr) gene from the different Antirrhinum species were amplified using the forward oligonucleotide primer DFRiF (5′-CTTGTTTGAGTATCCTAAGGCAGAAGG-AAG-3′) and the reverse primer DFRSTOP (5′-CTAGA-TTCTGCCATCAGTATGATCGTTTGC-3′). Fragments of DNA encoding the first intron of the Venosa gene were amplified using the forward primer VeExon1F (5′-GCAA-TGCATAGAGAAGTATGGGGAAGGTAAG-3′), which ends 26 bp upstream of the intron donor splice site, and the reverse primer VeExon2R (5′-GATAATTTAACCATCT-CATCCTGCAACTCTTC-3′), which ends 15 bp downstream of the intron acceptor splice site. PCR reactions were performed in a total volume of 50 μl, containing 200 ng of genomic DNA, 0.1 μM of each primer, 0.5 mM of dNTPs, and 2.5 U of AmpliTaq polymerase. Amplification involved one cycle of denaturation at 95°C for 5 min, and 40 cycles of 94°C for 1 min, 58°C for 1 min and 72°C for 1 min. These cycles were followed by one cycle of elongation at 72°C for 10 min. DNA fragments were cloned into pGEM-T-EASY and sequenced on each strand, and sequences were assembled in Lasergene (DNASTAR, Madison, WI, USA). Consensus sequences were aligned with the prank program (Loytynoja & Goldman, 2008).
Maximum likelihood analysis using the dnaml program of the phylip package (Felsenstein, 2004) was performed with an alignment of the third intron in the Nia gene and with intron 5 and exon 6 of the Dfr gene for 11 Antirrhinum accessions and an outgroup species, Misopates orontium. The transition : transversion rate ratio parameter for the F84 model of evolution and the alpha parameter for the discrete gamma model were calculated in the baseml program of the paml package (Yang, 2007) and the values set in the dnaml program. The other options used in the dnaml program were option S (set to ‘no’), global rearrangements (set to ‘yes’– default) and option J (set to 10 jumbles). The log likelihood value for the tree using the F84 model was −1188, which was statistically more significant than values obtained using either the JC69 or the K80 model with the likelihood ratio test. To estimate the confidence of each tree node, the run of the dnaml program was repeated with 1000 bootstrap data sets (option M) with option J set to 1 and a consensus tree generated using the phylip consense program.
Array test for nonrandom pollinator foraging
Differences in pollinator behaviour between flower morphs were assessed in pair-wise comparisons in the field at the John Innes Centre, Norwich, UK. Forty-eight plants (24 of each morph) were planted 80 cm from each other. No other Antirrhinum plants were present within 50 m of the test arrays. Plants were randomized within the array using the edgar program (J. K. M Brown, http://www.jic.bbsrc.ac.uk/services/statistics/edgar.htm). Flower phenotypes tested were: red against white, ivory, pink, and venation. Full-red plants were wild-type stock JI522, which is highly pigmented relative to wild accessions of A. majus (compare Fig. S2 with S3). Ivory-flowered plants were a mutabilis/delila (mut/del) line, which also lacks flavanone 3-hydroxylase activity (Schwinn et al., 2006). Pink-flowered plants were roseacolorata (roscol), and the venation line was Ve+ in the rosdor background. White-flowered plants carried the nivea mutation in which no flavonoids are produced, including the colourless flavones. All the mutations had been backcrossed to JI522 for at least three generations, and were considered isogenic for comparisons with the JI522 control.
Observations were carried out each day from 09:00 to 17:00 h for 3 wk, with each array being observed for a total of at least 30 h, and the observations being equally distributed among arrays with respect to time of day. Each visit and the identity of the visitor was recorded, along with the identity of the individual plant visited, the number of flowers visited on each plant and the time of day. Only those bouts in which more than five plants were visited were used in subsequent analyses. All foraging bouts in an array were first tested for heterogeneity of preference among bouts for red-flowered plants following the procedure of Jones (1997), which allowed discrimination between a general preference of pollinators for one plant morph displayed over several foraging bouts and preferences for one morph displayed during a single bout that could be changed in subsequent bouts.
Development and location of anthocyanin stripes in Venosa
Sectioning of the corolla of A. majus had shown that in Ve+ lines the pigmented stripes are associated closely with the veins and that anthocyanins are produced only in the adaxial epidermis (Fig. 1e) (Schwinn et al., 2006). To determine when the anthocyanin stripes are produced, pigmentation was observed in dissected flowers of A. majus carrying the Ve+ allele in the roscol background (Fig. 1d). Venation patterning was first visible when the flower bud had reached 8 mm in length, and appeared first at the base of the tube in the adaxial epidermis (i.e. the inside of the tube). Subsequently, pigmentation intensified and venation stripes also appeared in the lobes. In the tube, venation consisted of parallel lines along both sides of the tube, with no pigmentation in the top and base regions. The size and intensity of the stripes correlated with the size of the veins. Nonvenation pigmentation in flowers of Rosea1 lines appeared earlier, first being visible in buds of 4 mm in length.
Action of Venosa in determining venation
Biolistic introduction of 35S:gVenosa into the corolla of a Ve+ line gave rise to pigmented cells in the normally nonpigmented abaxial epidermis and also among the nonpigmented epidermal cells that occur between the pigment stripes of the adaxial epidermis (Fig. 2a). No new pigmented foci occurred when negative controls of 35S:Lc (bHLH; Fig. 2b) or 35S:GFP were introduced. These data suggested that the stripes of pigmentation in Ve+ lines resulted from specific expression of the Venosa gene in the cells of the epidermis overlying the veins.
Venation does not require the presence of the anthers or stigma
Emasculation was used to test whether signals originating from the anthers influence venation pigmentation in A. majus, as anther-produced signals, specifically gibberellin, can promote petal development (Olszewski et al., 2002; Hu et al., 2008) and pigmentation (Weiss, 2000). Flower buds emasculated at between 4 and 7 mm in length failed to develop further (Fig. 1f), and the corolla gradually senesced. Flower buds emasculated at between 7 and 10 mm developed more slowly than normal and had a smaller final size, but had normal venation pigmentation (Fig. 1g). When the bud length was > 10 mm, emasculation did not alter the development of the flower or corolla pigmentation. The occurrence of occasional aberrant flowers in the Venosa line of A. majus during hot weather allowed examination of the effect of the absence of all sexual organs (Fig. 1h). The aberrant flowers contained only the two dorsal petals and the sepals. The petals that did develop were much smaller than usual but had normal venation pigmentation.
Venosa transcript accumulates in a vein-associated pattern in the corolla
In situ hybridization analysis of the tube region of Ve+ corolla (line AA114; Ve+/rosdor) with a Venosa cDNA probe detected small patches of expression in cells on the adaxial side of the corolla around the veins (Fig. 3). When entire tubes were observed in section (Fig. 3a,b), signal was observed on both lateral sides of the tube but not in the top or base. This distribution of transcript matched the location of the pigment stripes in the tube (Fig. 3a). At higher magnifications the transcript was detected in a wedge of cells between the xylem and the adaxial epidermis, in both the epidermal and subepidermal cells (Fig. 3b). A similar hybridization signal was detected in the lobes (data not shown). Significantly, anthocyanin accumulation was detected only in the adaxial epidermal cells overlying the veins and not in the mesophyll tissue beneath.
As a negative tissue control for the hybridization, tube tissue from A. majus line AA104 (ve−/rosdor) which lacks venation was probed with Venosa. Although there was weak background signal on the vein elements themselves, the vein-associated signal observed for the Ve+ line was not detected (Fig. 3c). This can be seen more clearly by comparing the magnified portion of the tube in which the edges of the tissue have been outlined (Fig. 3c) with the images in Fig. 3(b). Chs was used as a positive control for the in situ analysis, because the activity of this enzyme is required for the biosynthesis of all flavonoids, including the colourless flavones that are present in the petals. The Chs antisense probe gave the expected strong hybridization pattern in both the adaxial and abaxial epidermal layers of the corolla (Fig. 3d). No significant signal was detected using negative controls (Fig. 3d).
Venation and Venosa expression within the genus Antirrhinum
Many of the European species of Antirrhinum have pale or acyanic flowers with red pigmentation in a venation pattern as a result of the activity of the Venosa gene. Such species include A. meonanthemum, A. molle, A. mollissimum and A. latifolium (Fig. S2). This is in contrast to the full-red phenotype of A. majus, A. barrelieri and A. australe and the pale/acyanic phenotype without venation pigmentation of A. graniticum (Schwinn et al., 2006). A faint venation pattern of pigmentation is apparent in some but not all accessions of A. majus, A. barrelieri and A. australe (Fig. S3), although this trait is hard to discern visually against the full-red background pigmentation in flowers of these species. In situ hybridization detected vein-associated Venosa transcript in the petals of five species that have venation pigmentation, but not in the one species (A. graniticum) that lacks venation (Fig. 4).
Phylogenetic relationships among Antirrhinum species with different floral pigmentation patterns
The origin of the venation trait within the Antirrhinum genus was examined by establishing the phylogenetic relationships among the species of interest with respect to their floral pigmentation patterns using the sequences of the third intron of the Nia gene. As a very variable marker, Nia coalesces more quickly than other plant markers and hence can be used to uncover the relationships between closely related taxa that may have diverged only very recently (Howarth & Baum, 2002).
The alignment of the Nia sequences provided a topology that separated the different species of the genus Antirrhinum (Fig. 5a). Bootstrap values did not provide statistical support for this topology, but this reflects the low level of information available in the form of sequence differences, even within the hypervarible region of the Nia gene, rather than implying that the topology is wrong. To gather further support for the phylogeny we added c. 870 bp of sequence of the gene encoding Dfr from each Antirrhinum species. This sequence included the fifth intron and the sixth exon of each Dfr gene and contained a microsatellite present in the fifth intron (Fig. 5c). As microsatellites evolve more rapidly than single nucleotide changes, and may serve to confound phylogenies based on single nucleotide polymorphisms (SNPs) (Zhang & Hewitt, 2003), we did not use the variable microsatellite sequences from the alignments to construct the trees. Addition of the Dfr sequences did not significantly increase the bootstrap support for the phylogeny, although the phylogeny based on the combined Nia plus Dfr sequences gave a similar topology to that based on Nia sequences alone (compare Fig. 5b with 5a). This attested to the usefulness of the Nia sequences for phylogenetic analyses. The problem of the lack of robust resolution of the species is also apparent in other recently published phylogenies for the genus (Vargas et al., 2009) and probably reflects the high levels of hybridization that occur between inter-fertile species which confound resolution with strong statistical support. Given the rapid and recent radiation of the species in the genus Antirrhinum (Vargas et al. (2009) calculated the time of divergence of Misopates orontium and A. majus as 8 million yr ago and that of A. molle and A. majus as 1 million yr ago) and the evidence for significant gene flow between species in the genus (Whibley et al., 2006; Vargas et al., 2009), it may be impossible to obtain increased statistical support for the phylogeny. However, the phylogeny does support the view that pale floral pigmentation with a strong venation pattern is an ancestral trait in the genus. First, this phenotype is shown by the outgroup M. orontium. Second, this phenotype is shown by the species that align basally, A. latifolium, A. molle and A. meonanthemum. Third, the full-red species (A. majus, A. barrelieri and A. australe) appear to have been derived from ancestors that probably had pale floral pigmentation with venation patterning; two species with full-red phenotypes cluster with species that have retained the pale pigmentation with venation patterning phenotype (A. mollissimum with A. majus and A. barrelieri) and the other, A. australe, appears to have shared a common ancestor with palely pigmented species with venation pigmentation (A. molle and A. latifolium). These data support the idea that ‘full-red’ floral colouring is a derived trait, an idea also supported by the recent phylogenies based on both plastid and nuclear DNA polymorphisms within the genus (Vargas et al., 2009). Indeed, our data suggest that ‘full-red’ may be polyphyletic within the genus, having evolved independently in A. majus/A. barrelieri and A. australe, an idea supported by the geographical separation between populations of A. majus and A. australe (Vargas et al., 2009). This conclusion is supported by the phylogenies based on both plastid and nuclear DNA polymorphisms, although A. barrelieri was not included in those analyses (Vargas et al., 2009).
Analysis of Venosa alleles in different species within the genus Antirrhinum
To investigate the evolution of venation pigmentation within the genus, we aligned the sequences of the first intron of the Venosa gene from each species (Fig. S5a). The intron of the Venosa gene from A. majus was truncated compared with the sequences from the other species (200 bp instead of c. 540 bp) and only the 5′-end of the intron showed homology to sequences from the other species, precluding the inclusion of this sequence in constructing the phylogenetic tree. The phylogenetic relationships of the species based on the sequence of the first intron of the Venosa gene (Fig. S5b) were remarkably similar to those in the phylogenies based on Nia alone or Nia plus Dfr sequences, except that A. barrelieri and A. australe share very closely related Venosa intron sequences. This was noteworthy because the A. barrelieri accession analysed is homozygous for functional Ve+ alleles whereas the A. australe accession is homozygous for ve− alleles (Schwinn et al., 2006). Analysis of the Venosa sequences suggested that loss of venation is a derived trait, and may be polyphyletic within the section (i.e. independently derived in the full-red species A. majus and A. australe and in the acyanic A. graniticum).
Pollinators show preferences for different flower colour morphs of A. majus
The most likely selective advantage associated with full-red flowers is improved attractiveness to pollinators, and field trials with Antirrhinum majus showed higher fruit set associated with red- vs white-flowered morphs (Glover & Martin, 1998). Improved attractiveness to pollinators could drive the evolution of full-red flowers from more palely coloured ancestors, with or without venation. To test for any preferences for full-red flowers compared with pale or venation-patterned morphs, we used near-isogenic lines of A. majus that differed in their genetic constitution affecting floral pigmentation and patterning. Full-red plants (Ros+/ve−; containing both flavones and anthocyanins) were compared with white (niv−; no flavones or anthocyanins), ivory (mut/del; flavones but no anthocyanins), pale red (roscol/ve−; flavones and weak anthocyanins) and very pale red with venation patterning (rosdor/Ve+−; flavones and venation anthocyanins) plants in four pair-wise comparisons of pollinator visitations in randomized plots. In all four arrays, bumblebees foraged nonrandomly with regard to flower colour (Table S1). Tests of significant variation found discernable assortative visitation between the two different morphs in each array. No consistent differences were observed with regard to the three types of pollinator (bumblebees: Bombus terrestris, Bombus muscorum and Bombus lapidarius) or the period of day in which the observations were made.
The mean number of each type of plant visited over all the bouts indicated a strong preference towards red colour (Fig. 6a, Table S1). The strongest preference for red was found in the comparison of red with ivory (inc−) and red with white (niv−): 80% and 76% of visitations in all the bouts were made to red flowers, respectively. When presented with the pink alternative (roscol), pollinator preference for the red morph was still statistically significant (P = 0.027), with 64% of red plants visited over all bouts. The array including flowers with venation (Ve+ in the rosdor background) was the only one for which preferences for the red (55%) or the alternative morph (45%) were not statistically different. Furthermore, the overall number of flowers visited for each morph in each array indicates that pollinators not only visited the red morph more often than the pink, ivory or white, but also visited significantly more flowers per plant for the red-flowered plants, while there was again no significant difference between the full-red and the pale red with venation pigmentation morphs (Fig. 6b).
Complex floral pigmentation patterns, comprising spots and/or stripes, are common in nature. Despite their widespread occurrence, both their function in pollination biology and their molecular mechanisms of patterning are poorly understood. We present here an analysis of one of the most common floral pigment patterns: venation, the production of pigment in stripes associated with the veins. We have shown that venation of anthocyanin pigment in A. majus is defined by patterned expression of the Venosa gene (R2R3MYB transcription factor), in combination with the epidermal production of other factors, probably its bHLH co-regulators. Furthermore, phylogenetic analysis suggests that venation is probably the ancestral trait in the Antirrhinum genus, and the pattern defined by the action of the Venosa gene provides advantages for pollinator attraction compared with pale or acyanic floral colouration, but that these advantages are probably lost when the background pigmentation is full red.
In the genus Antirrhinum, venation patterning of the adaxial epidermis is prevalent in several species. In A. majus, venation occurs when Ve+ is active in a background lacking activity of Rosea1, so that the pigment stripes are seen on an otherwise acyanic or palely pigmented corolla. Ectopic expression of Venosa using biolistics demonstrated that lack of Venosa transcript production is responsible for the acyanic phenotype of the abaxial epidermis and the cells between the veins in the adaxial epidermis (Fig. 2). Therefore, we predicted that venation is defined by the specific expression of Ve+ in the epidermal cells above the veins. In situ hybridization analysis indeed showed that Venosa transcript is produced only in small regions associated with the veins. However, transcript is not limited to the epidermis, but rather is present in a ‘wedge’ of cells between the vein and the adaxial epidermis. Venosa can activate anthocyanin biosynthesis in A. majus when partnered with an appropriate bHLH factor (in the corolla tube this is Delila and in the lobe Delila or Mutabilis) and probably a WDR protein. Delila transcript has been shown to accumulate in an epidermal-specific pattern in the corolla (Goodrich et al., 1992; Jackson et al., 1992), while the WDR protein is produced constitutively in other species and may move to neighbouring cells (Walker et al., 1999; Bouyer et al., 2008). Thus, the epidermal-specific venation pattern is probably determined by overlapping of the expression domains of the MYB and bHLH factors involved in combinatorial control of anthocyanin production in the genus Antirrhinum (Fig. 7). There are probably further regulatory domains overlapping these venation pattern domains. For example, Venosa transcript (and venation pigmentation) is associated only with the veins on the lateral sides of the tube and not those of the top and base of the tube (Fig. 3a,b). Such multiple and overlapping signalling domains for the different transcription factor members of the regulatory complex may be key to enabling the differential regulation of pigmentation between developmentally similar neighbouring cells within the epidermis. The characterization of R3MYB repressor proteins involved in regulation of anthocyanin biosynthesis (Dubos et al., 2008; Matsui et al., 2008) offers additional potential players that may determine the spatial control of pigmentation patterns. However, the ability of 35S:gVenosa, introduced biolistically, to induce pigmentation between the veins argues against the action of a repressor of Venosa in the unpigmented regions (Fig. 2).
The vein-associated production of Venosa transcript was consistent with the presence or absence of venation pigmentation. No Venosa transcript was detected in corolla sections from an A. majus line with a nonfunctional venosa allele (ve−/rosdor) or in A. graniticum, a species that lacks venation. Vein-associated Venosa transcript was detected in corolla sections of all the species showing venation, and a correlation between the strength of venation and transcript abundance was observed. For example, A. majus, A. mollissimum and A. barrelieri had the strongest transcript signal and have the strongest venation, and A. latifolium had the weakest signal and has the weakest venation. These data support the previous proposal that the MYB genes are the principal determinates of both the intensity and the distribution of anthocyanin pigmentation in the flowers of different species belonging to the genus Antirrhinum (Schwinn et al., 2006).
The Venosa transcript pattern is suggestive of a directional signal originating in the veins. What this signal may be is not clear. The identity of the signal cannot be determined clonally, because the epidermal and mesophyll layers of plants are clonally distinct. A regulatory domain could possibly be established during the formation of the petal and the patterning of veins. However, venation pigmentation is a relatively late developmental event, with pigment first present in the tube in buds approx. 8 mm in length, after the time of establishment of many of the boundaries in the corolla (Coen & Carpenter, 1986). From our data it seems likely that a translocated signal originating in the veins determines Venosa gene expression. The endogenous and environmental signals that control floral pigmentation have been examined in only a few species and are generally not well understood, but at least light, gibberellin and carbohydrate have been shown to be required in some species (Weiss, 2000; Farzad et al., 2002, 2003). The emasculation tests and studies of aberrant flowers (Fig. 1) suggest that anther-derived signals promote petal expansion in A. majus but are probably not required for pigmentation. Based on data from other species (Weiss, 2000; Olszewski et al., 2002; Hu et al., 2008), in vitro supplementation experiments (data not shown) and its role in petal expansion, the principal anther-derived signal is likely to be gibberellin. The requirement for gibberellin for petal expansion is probably limited to the phase of rapid corolla expansion, as emasculation of buds > 10 mm in length did not affect petal development. Carbohydrate induces anthocyanin production in Arabidopsis thaliana via the R2R3MYB factor PRODUCTION OF ANTHOCYANIN PIGMENT1 (Teng et al., 2005) but is unlikely to be unloaded in a polar direction to the adaxial epidermis. Testing the influence of carbohydrate on A. majus venation pigmentation proved difficult, because of the strong association between general petal development and carbohydrate supply (Y. Shang, unpublished observation). An alternative potential signal molecule that has a key role in formation and differentiation of the vascular network (Fukuda, 2004), and which may have a polar translocation pattern, is auxin (Scarpella et al., 2006). Isolation and analysis of a 2.4-kb promoter sequence for Venosa (HM212794) identified two copies of a sequence similar to the auxin response element (TGTCTC or TGTC(A/C)C) at −308 bp and −1045 bp (relative to the initiating ATG). No gibberellin or carbohydrate regulatory elements were identified.
Given that venation is one of the most common floral pigmentation patterns, some advantage for pollination might be predicted. The field-based tests we conducted with the natural bumblebee pollinators of A. majus showed that, in contrast to flowers with predominantly white or ivory colour, the Venosa phenotype in A. majus attracted pollinator visits at a similar rate to the full-red Rosea1 flower phenotype (Fig. 6). In fact, in wild species, the frequency of visits of bumblebee pollinators to full-red and venation-patterned morphs might be even closer than demonstrated in our field trials, because we had to use a very strongly coloured full-red line of A. majus for the field trials in order to maintain a common genetic background with the white, ivory, pale pink and venation patterned lines available. Wild accessions of A. majus and other ‘full-red’ species accumulate significantly less anthocyanin in their petals than the laboratory line JI522 (compare JI522 in Fig. S2 with wild A. majus accessions in Fig. S3).
Phylogenetic analysis of different species within the European section of the genus Antirrhinum was undertaken using sequences from the third intron of Nia and from intron 5 to the stop codon of the Dfr gene. The occurrence of inter-specific hybrids generating morphologically intermediate specimens, reported in the wild where overlapping populations occur (Mather, 1947), has been a major obstacle in obtaining a clear picture of the phylogenetic relationships among the species within the genus Antirrhinum. Studies using chloroplastic (Gubitz et al., 2003) or nuclear (Vargas et al., 2004) DNA markers have provided little resolution as a result of insufficient numbers of informative characters, although the combination of plastid and nuclear markers has proved more informative (Vargas et al., 2009). The Nia and Dfr sequences were variable enough to give resolution between the different species accessions sampled. The analysis suggested that species with full-red floral pigmentation have been derived from more ancestral forms with paler background floral pigmentation, and may have evolved on more than one occasion within the section. By contrast, the venation pattern of pigmentation on a pale or acyanic background appears to be the ancestral trait. While venation patterning is present in all accessions of all species with pale background pigmentation except A. graniticum, venation patterning is not present in all accessions of full-red species, including A. majus, A. barrelieri and A. australe. This suggests that, when full-red pigmentation evolves in a species, the selection pressure for venation patterning is relaxed, and the trait may be lost in some populations. This possibility is supported by the field observations of pollinator visits (Fig. 6).
Venation is one of the most common pigmentation patterns in plants and, as shown in field observations, can assist in attraction of pollinators to the flower. We have shown that venation in the model genus Antirrhinum is a result of intersecting spatial regulation domains among the MBW transcriptional regulators. Given the conservation of the MBW regulation system across species, and the proposed involvement of R2R3MYBs in defining other floral pigmentation patterns (Ma et al., 2009; Yamagishi et al., 2010), it is likely that Venosa-like genes will underpin venation in many species and that venation will provide excellent systems for dissection of the upstream signalling pathways controlling floral pigmentation patterning.
We thank Stephanie Hares for undertaking the field observations and the Nuffield Organisation for funding her summer studentship; Rosemary Carpenter and Ryan K. Oyama for seed of the species accessions; Simon Deroles for assistance with the manuscript; Susan Wessler for pLC349; Andrew Gleave for pART7; Simon Coupe for pPN93; the Marsden Fund of New Zealand for supporting K.S., K.D. and Y.S.; the New Zealand Foundation for Research, Science, and Technology contracts C02X0203 and C02X0701 for supporting K.S. and K.D.; the John Innes Foundation for funding the PhD studentship of J.V. and the BBSRC for support of C.M. and P.B. through the core strategic grant awarded to JIC.