• Multiple evolutionary shifts in floral symmetry and stamen number have occurred in the snapdragon (Antirrhinum majus) family Veronicaceae. In Mohavea, Veronica and Gratiola there have been independent evolutionary reductions in stamen number and modifications to corolla shape. It is hypothesized that changes in the regulation of homologs of snapdragon dorsal flower identity genes CYCLOIDEA (CYC) and RADIALIS (RAD) underlie these floral transitions.
• CYC-like and RAD-like genes from Veronica montana and Gratiola officinalis were cloned and sequenced, compared with homologs from other Veronicaceae species using phylogenetic analysis, and their expression was investigated by reverse transcriptase-polymerase chain reaction (RT-PCR) and in situ hybridization.
• VmCYC1, GoCYC1, GoCYC2 and RAD-like genes are expressed exclusively in the dorsal region of floral meristems and developing flowers. Their expression patterns do not correlate with patterns of stamen arrest. VmCYC2 and GoCYC3 are expressed in both vegetative and floral tissues, with VmCYC2 being most abundant in all regions of the floral meristem and all petals.
• These results support conservation of the floral symmetry gene network for Veronicaceae RAD-like and some CYC-like paralogs, suggest regulatory evolution of other CYC-like genes following gene duplication and implicate different genetic mechanisms underlying dorsal versus ventral stamen abortion within Veronica and Gratiola.
Bilaterally symmetrical (zygomorphic or monosymmetric) flowers have evolved many times across angiosperms, and evolutionary transitions from radial to bilateral symmetry are considered to have increased associations with diverse and reliable pollinators through more specialized plant-pollinator interactions (Neal et al., 1998; Sargent, 2004). In Veronicaceae (Lamiales) the majority of species have zygomorphic flowers (Reeves & Olmstead, 1998). However, modification of bilateral symmetry or reversal to near radial symmetry has occurred independently along different lineages. These morphological shifts are generally associated with changes in stamen/staminode number and/or petal shape (Endress, 1999), and often correlate with pollinator shifts (Neal et al., 1998).
The genetic basis underlying patterns of petal symmetry and stamen abortion is best understood in the model species snapdragon (Antirrhinum majus, Antirrhineae, Veronicaceae). Snapdragon has bilaterally symmetrical flowers typical of Veronicaceae, with five sepals, five petals, four stamens, one staminode and two carpels. Petals are differentiated into three types: two dorsal (adaxial or upper lip), two lateral and one ventral (abaxial or lower lip). The dorsal and lateral petals develop asymmetrically from the petal midline and the dorsal stamen aborts early in development to form the staminode. The mature ventral petal is bilaterally symmetrical, and acts as a landing platform for pollination by nectar-foraging bees (Kampny, 1995; Cubas, 2004). Four genes are known to interact to control dorsiventral symmetry in snapdragon: CYCLOIDEA (CYC), DICHOTOMA (DICH), RADIALIS (RAD) and DIVARICATA (DIV). CYC and DICH encode proteins within the TCP family of transcription factors and are involved in promoting dorsal flower identity (Luo et al., 1996; Cubas et al., 1999; Luo et al., 1999; Costa et al., 2005). RAD and DIV encode MYB-like transcription factors that promote dorsal and ventral identity, respectively (Almeida et al., 1997; Galego & Almeida, 2002; Corley et al., 2005; Costa et al., 2005).
The TCP gene family is functionally diverse. Phylogenetic studies show that CYC and DICH are members of the CYC2 ECE clade (Howarth & Donoghue, 2005, 2006), a lineage characterized by an arginine-rich R domain (Cubas et al., 1999) and a history of extensive gene duplication. CYC and DICH originated from a duplication event at the base of Antirrhineae (Gübitz et al., 2003; Hileman & Baum, 2003) and have overlapping functions. Both genes are expressed in a dorsal-specific manner in the floral meristem, where they repress growth and control organ number (Luo et al., 1996, 1999). At later stages of development CYC and DICH positively regulate cell division in the dorsal petal by upregulating RAD (Corley et al., 2005; Costa et al., 2005). CYC also negatively regulates cell division in the stamen whorl by directly downregulating cell cycle genes, such as CYCLIN D3b (Gaudin et al., 2000). In addition, asymmetric DICH expression across dorsal petals at later stages causes internal petal asymmetry (Luo et al., 1999; Gaudin et al., 2000). In turn, RAD negatively regulates DIV protein in the dorsal and lateral regions of the flower, restricting the ventral identity developmental program (Almeida et al., 1997; Galego & Almeida, 2002). Double cyc:dich mutants have fully ventralized flowers that often have extra petals and stamens (Luo et al., 1996). Ventralization is largely caused by loss of negative and/or positive regulation of DIV and cell cycle genes, respectively, in the dorsal and lateral regions (Almeida et al., 1997; Galego & Almeida, 2002; Gaudin et al., 2000). In cyc and rad single mutants there is a loss of dorsal petal identity and the dorsal staminode either develops as a fertile stamen (cyc) or is a slightly longer staminode than in wild-type (rad) (Corley et al., 2005). By contrast, mutations in dich only affect internal dorsal petal asymmetry, and div mutants have ventral petals that adopt lateral identity (Luo et al., 1996, 1999; Almeida et al., 1997).
Interspecific variation in corolla shape and stamen number is found in all major Veronicaceae lineages. Are changes in the regulation of CYC/DICH genes and their downstream targets responsible? Comparative analyses of gene expression between snapdragon and its close relative desert ghost flower (Mohavea confertiflora, Antirrhineae, Veronicaceae) suggest that they might be (Hileman et al., 2003). Unlike snapdragon, petals of desert ghost flower lack internal asymmetry and only the ventral stamens develop fully. McCYC1 and McCYC2 are expressed in a similar manner to CYC in the petal whorl, but are expressed in both dorsal and lateral staminodes in desert ghost flower, correlating with arrest of these organs. McDICH1 and McDICH2 expression also positively correlates with patterns of stamen arrest. In addition, McDICH genes are not expressed in late stages of petal development, correlating with the loss of dorsal petal internal asymmetry, which is established late in snapdragon petal development (Hileman et al., 2003). The fact that expression of McCYC and McDICH genes is only expanded into lateral regions of the stamen whorl, but is dorsally restricted in the petal whorl, shows that the regulation of CYC-like genes can be uncoupled in different whorls. This hypothesis is further supported by the backpetals mutant of snapdragon, where specific mutations in the CYC promoter result in ectopic CYC expression within lateral and ventral petals, but not stamens (Luo et al., 1999). These observations are important in light of the fact that reduction in stamen number is not always correlated with changes in petal morphology.
Similarly to desert ghost flower, Gratiola officinalis (Gratioleae, Veronicaceae) and Veronica montana (Veroniceae, Veronicaceae) flowers develop only two functional stamens (Fig. 1). However, unlike desert ghost flower, the fertile stamens of both species are in the lateral position (Fig. 1b–e), due to dorsal and ventral stamen arrest during organ development. Unlike snapdragon and G. officinalis, Veronica flowers entirely lack staminodes (Kampny et al., 1993; Endress, 1999; Fig. 1d). In Veronica, flowers have two dorsal petals that are fully fused early in development to form a single organ (Kampny et al., 1993; Endress, 1999) (Fig. 1b). At maturity, the dorsal petal is much larger than the lateral and ventral petals, however, all four petals are similar in shape and are internally symmetrical (Fig. 1b).
We hypothesize that G. officinalis and V. montana dorsal petal identity is specified by CYC- and RAD-like genes, but that the derived pattern of stamen arrest results from changes in expression of CYC-like genes and their putative downstream targets exclusively in the third whorl. Specifically, we predict that V. montana and G. officinalis CYC-like genes are expressed in the ventral region of the third whorl where stamen arrest occurs during flower development (Fig. 1d,e). To test this, we isolated homologs of the symmetry network genes from both species and compared their expression. Our data support conservation of the floral symmetry gene network in Veronicaceae for the specification of dorsal flower identity, but implicate different genetic mechanisms underlying dorsal versus ventral stamen abortion within Veronica and Gratiola.
Materials and Methods
Plant materials and growth conditions
Seed of G. officinalis L. and V. montana L. were obtained from B and T World Seeds (http://www.b-and-t-world-seeds.com/). Material from both species was collected from plants grown in a glasshouse at the University of Kansas, Lawrence, USA, under long-day conditions. Silica-dried leaf material of Bacopa caroliniana (Walt.) B. L. Rob. was kindly provided by Amy Litt at the New York Botanical Garden.
Scanning electron microscopy
Developing G. officinalis inflorescences were fixed in FAA (3.7% formaldehyde, 50% ethanol, 5% acetic acid) or Karnovsky's fixative (Electron Microscopy Sciences, Hatfield, PA, USA), dehydrated in an ethanol series, and dissected as necessary to reveal internal floral organs. Tissues were chemically dried by washing in a hexamethyldisilizane (HMDS) series (25%, 50%, 75%, 100% HMDS in ethanol), and air-dried under vacuum. Specimens were mounted on stubs, sputter-coated with gold, and viewed with a D. Leo field emission scanning electron microscope.
Inflorescences of G. officinalis were fixed, dehydrated, and embedded in Paraplast blocks for sectioning. Wax blocks were sectioned, cleared, rehydrated, and stained with aqueous 0.1% toluidine blue. Stained material was dehydrated and photographed using a Leica DM5000B microscope and a Leica DFC 300FX camera.
Cloning and phylogenetic analysis
CYC- and RAD-like genes were amplified with polymerase chain reaction (PCR) from G. officinalis, B. caroliniana and V. montana cDNA and/or genomic DNA using degenerate primers designed from an alignment of available Genbank asterid sequences (see the Supporting Information, Table S1). Multiple primer sets were used, PCR products were cloned into TOPO-TA (Invitrogen, Carlsbad, CA, USA) and multiple clones were sequenced per primer set in an attempt to find all symmetry network gene paralogs in the genome. Sampling was considered complete when primer sets amplified nonCYC-like TCP genes and nonRAD-like MYB genes, and when no additional CYC-like or RAD-like paralogs were found. To determine the relationships among paralogs, CYC and RAD amino acid sequences were aligned separately with other asterid sequences in macclade (Maddison & Maddison, 2003). Nucleotide sequences were used to generate maximum likelihood (ML) trees in paup 4.0b10 (Swofford, 2001) under the GTR + I + Γ model of molecular evolution, as recommended by mrmodeltest 2.2 (Nylander, 2004), using 10 random addition sequences. Parsimony and likelihood bootstrap values were obtained using 1000 and 100 replicates, respectively, in paup 4.0b10 (Swofford, 2001).
Reverse transcriptase (RT)-PCR
Total RNA was extracted from various G. officinalis and V. montana tissues using the RNeasy Plant Mini Kit and RNase-Free DNase Set from Qiagen (Valencia, CA, USA). Approximately 500 ng was added to a 15 µl cDNA reaction to synthesize cDNA using the iScript Select cDNA Synthesis Kit (Bio-Rad, Hercules, CA, USA). To determine expression levels, gene-specific primers were designed to amplify c. 150–400 bp of the open reading frame (Table S1), and Actin was used as a loading control as described in Prasad et al. (2001). For each primer pair, PCR amplifications were initially run for 28–34 cycles to confirm linearity.
In situ hybridization
Gene-specific probe templates spanning the protein coding region from the 3′ end of the TCP domain to the end of the coding region were generated for GoCYC1, GoCYC2, GoCYC3, VmCYC1 and VmCYC2 using the primers GoCYC1sF, GoCYC2sF, GoCYC3sF, VmCYC1F and VmCYC2F, respectively (Table S1), in combination with the reverse primer P2 (Vieira et al., 1999). Probe templates spanning most of the protein coding region to the end of the 3′-UTR were generated for GoCYC3, GoRAD and VmRAD using the primers GoCYC2sF, GoRADF and VmRADF, respectively, in combination with the polyTQT primer (Table S1). DNA fragments of G. officinalis Histone 4 (GoH4) were isolated using primers hist-H4-F (5′-GTCTGGTCGTGGAAAGGGAGGCAAGGG-3′) and hist-H4-R (5′-TTAACCGCCAAATCCATAGAGTCC-3′). All PCR products for probe generation were cloned into the TOPO-TA vector (Invitrogen), and confirmed by sequencing. Probes were c. 300- to 900-bp long. Sense and antisense riboprobes were generated using T7 and T3 RNA Taq polymerase (Roche, Indianapolis, IN, USA) according to the manufacturer's instructions. Probe hydrolysis followed Jackson (1991) to yield fragments c. 150 bp long. In situ hybridization was performed on transverse and longitudinal sections of multiple inflorescences as in Jackson (1991) with modifications described in Preston & Kellogg (2007). In situ images were documented using a Leica DM5000B microscope attached to a Leica DFC 300FX camera. Photographs were imported into Adobe Photoshop and adjusted for contrast, brightness, and color balance.
Gratiola officinalis flower development
Flower development has been extensively documented for Veronica (Kampny et al., 1993; Endress, 1999), but little is known about the developmental process of Gratiola. Thus, to better understand floral organ development in G. officinalis, we studied flower buds at different developmental stages using light microscopy and scanning electron microscopy (SEM). Serial sections and SEM revealed early asymmetric development along the dorsiventral flower axis (Fig. 2). However, unlike Veronica and snapdragon, where dorsal organ development is delayed relative to ventral organ development, G. officinalis dorsal floral organs developed before ventral organs.
Floral development began in the axils of two lateral bracts (Fig. 2a), with initiation of the dorsal sepal, closely followed by development of the ventral sepals (Fig. 2b). After limited dorsal and ventral sepal outgrowth, the two lateral sepals were initiated (Fig. 2c) followed by initiation of the stamens, then petals (Fig. 2d). Petals emerged in a dorsal to ventral sequence (Fig. 2d) followed by development of the gynoecium (Fig. 2e). At early stages of stamen development, the dorsal stamen arrested, leaving a very reduced staminode (Fig. 2d). At this stage ventral petal and stamen development continued to lag behind that of the dorsal and lateral petals, and lateral stamens (Fig. 2d–f). Subsequently, both the lateral and reduced ventral stamens continued to differentiate distinct filaments and anthers, before ventral stamen arrest around the time of petal enlargement (Fig. 2e). The majority of ventral petal enlargement occurred just before and following anthesis, resulting in an elongated lip characteristic of mature flowers (Fig. 1c).
Material stained with toluidine blue was studied to distinguish between cessation of cell proliferation and cell death in arresting staminodes. The nuclei of the significantly reduced dorsal and lateral staminodes showed dark blue staining, similar to that observed in fully developed stamens (Fig. 2f; data not shown). Together with the development series, this supports the hypothesis that G. officinalis stamen arrest occurs through a gradual reduction in the rate of cell division, consistent with the mechanism described for snapdragon dorsal stamen arrest (Gaudin et al., 2000). Antisense histone H4 in situ hybridization was also used to determine the relative timing of petal enlargement. At late stages of flower development, histone H4 was detectable only in the outer margins of the lateral and ventral petals, consistent with relatively late expansion of these petals relative to the partially fused dorsal petals (Fig. 2g).
Independent duplications of CYC but not RAD genes
To establish the role of snapdragon floral symmetry gene homologs in the evolution of V. montana and G. officinalis floral symmetry and stamen number, CYC- and RAD-like genes were isolated from both species (Genbank accessions FJ649687–FJ649688; FJ649691–FJ649696) and their phylogenetic relationships determined. Extensive screening of putative homologs revealed two copies of CYC in V. montana, three copies of CYC in G. officinalis, and one copy of RAD in each species (Fig. 3). All CYC genes contained the conserved TCP and R domain characteristic of the CYC2 ECE clade of TCP genes (Howarth & Donoghue, 2005, 2006). VmRAD and GoRAD both contained one conserved MYB I domain, as described for AmRAD. Comparison of VmRAD genomic and cDNA sequences revealed that the gene is composed of two exons and one intron. However, unlike AmRAD, the VmRAD intron lacks the CYC-consensus binding site described in Costa et al. (2005).
Maximum likelihood analysis placed both VmCYC1 and VmCYC2 in a well-supported clade with Digitalis purpurea (foxglove) CYC sequences, in accord with the species phylogeny (Fig. 3a). However, rather than showing a sister relationship, the two V. montana CYC genes were both sister to D. purpurea genes, suggesting a gene duplication event predating the split between these species (Fig. 3a). The same analysis placed GoCYC genes in a clade sister to CYC genes from snapdragon and close relatives, although this relationship was not well-supported (Fig. 3a). To better determine the timing of the GoCYC gene duplications, two distinct CYC genes from Bacopa caroliniana (Genbank accessions FJ649689–FJ649690), another member of tribe Gratioleae, were cloned and sequenced. The well-supported relationship of Gratioleae CYC genes revealed a duplication event that predated the divergence of Gratiola and Bacopa, followed by two more recent duplication events following their divergence (Fig. 3a). Similar phylogenetic analyses for RAD genes supported the monophyly of the Veronicaceae RAD genes in our sample (Fig. 3b). In contrast to the species phylogeny, the most parsimonious and maximum likelihood trees showed a sister relationship between RAD genes from Gratiola and Antirrhinum, with Veronica RAD genes sister to them both. However, none of these relationships were well-supported (Fig. 3b).
CYC-like gene expression
To determine if CYC-like gene expression is correlated with dorsal petal identity and patterns of stamen abortion, gene expression analyses based on gene sequences isolated in this study were carried out on V. montana and G. officinalis tissues. Actin-standardized RT-PCR analyses showed that VmCYC1 was only expressed in floral tissues (Fig. 4a). Transcripts were most abundant in pre-anthesis flowers, with apparent downregulation post-anthesis (Fig. 4a). In situ hybridization using a VmCYC1 mRNA riboprobe supported a similar pattern of expression within flowers. At the floral meristem stage, VmCYC1 was restricted to the dorsal region (Fig. 5a–d). This was observed in both transverse (Fig. 5a,b) and longitudinal (Fig. 5d) sections. During sepal and lateral stamen development (petal primordia emerge very late in Veronica flower development) VmCYC1 was expressed in the dorsal petal initials (Fig. 5e). At this stage there was no expression in the ventral region where the ventral stamens are presumably being inhibited from developing (Fig. 5c,e). During later stages of development, when all organs are visible, expression was restricted to the dorsal petal (Fig. 5f). Gene transcripts were undetectable in sepals, stamens, gynoecia and subtending leafy bracts.
In contrast to VmCYC1, VmCYC2 transcripts were not confined to inflorescence tissues. The RT-PCR analyses also revealed low levels of expression within stems and leaves (Fig. 4a). Furthermore, in floral tissues, VmCYC2 mRNA was elevated in post-anthesis flowers compared with pre-anthesis flowers (Fig. 4a). In situ hybridization showed that VmCYC2 was not restricted to the dorsal side of floral meristems of developing flowers (Fig. 5g–k). Instead, it was expressed across the floral meristem (Fig. 5g), and at later stages, equally within the dorsal, lateral and ventral petals (Fig. 5h–k). Control sections probed with sense riboprobes of V. montana CYC-like genes showed no significant staining (Fig. 5l,m).
Actin-standardized RT-PCR analyses of GoCYC1 and GoCYC2 revealed a dorsiventral gradient of expression across mid to late stage petals and stamens, but no expression within vegetative tissues (Fig. 4b). As in snapdragon, expression for both genes was highest in the dorsal region of the flower. However, there was some weaker expression of GoCYC2 in lateral petals + stamens, and of GoCYC1 and GoCYC2 in the gynoecium (Fig. 4b). In situ hybridization using antisense mRNA riboprobes was consistent with these results. In early flower development, both GoCYC1 and GoCYC2 were expressed exclusively in the dorsal region (Fig. 6a; data not shown). During petal and stamen development, transcripts were only detectable in the dorsal petals and staminode (Fig. 6b,d,e). Little to no hybridization was visible in the lateral and ventral petals, lateral stamens, ventral staminodes, gynoecium, or the two subtending lateral bracts (Fig. 6b–e). Expression was similar for both genes in the dorsal staminode (Fig. 6b,d). However, GoCYC2 expression appeared to be higher in the dorsal petals than GoCYC1 (Fig. 6b,d). This was apparent in both transverse (Fig. 6b,d) and longitudinal (Fig. 6e; data not shown) sections. No detectable expression was observed using sense control mRNA riboprobes (Fig. 6g,h).
Unlike GoCYC1 and GoCYC2, RT-PCR analyses showed GoCYC3 expression in both vegetative and inflorescence tissues (Fig. 4b). However, like GoCYC1 and GoCYC2, GoCYC3 expression within the flower was highest in the dorsal petals + staminode, as well as in the gynoecium (Fig. 4b). Comparison of RT-PCR for all three genes demonstrated a significantly lower level of GoCYC3 in all tissues. This was supported by in situ hybridization. No clear staining was detectable in any developmental stage for inflorescences hybridized with an antisense GoCYC3 riboprobe (Fig. 6f; data not shown). As in the case of GoCYC1 and GoCYC2, no detectable expression was observed using sense control mRNA riboprobes (Fig. 6i).
RAD-like gene expression
In early to mid stages of flower development, expression of RAD-like genes from both V. montana and G. officinalis was very similar to that observed for VmCYC1 and GoCYC1/GoCYC2, respectively. Analysis by RT-PCR revealed a high level of VmRAD expression in pre-anthesis flowers, with reduced levels of expression in post-anthesis flowers (Fig. 4a). No expression was detectable in vegetative tissues (Fig. 4a). More detailed analyses using in situ hybridization found VmRAD expression in the dorsal region of floral meristems (Fig. 7a,d). At later stages of development, VmRAD was restricted to the dorsal petal, with no apparent mRNA accumulation in the ventral regions (Fig. 7b–f). Similarly, GoRAD expression was detected most strongly in the dorsal region of early to mid-stage developing flowers (Fig. 7g–i). Transverse and longitudinal sections showed GoRAD expression in the dorsal petals and dorsal staminode, with significantly reduced expression in the lateral and ventral petals and staminodes (Fig. 7g–i). Unlike VmRAD, RT-PCR analyses in mid- to late-stage flowers revealed a different pattern of GoRAD expression. At this stage, rather than being found exclusively in the dorsal petals + staminode, GoRAD was also expressed in the lateral petals + stamens, ventral petal + staminodes, and in gynoecia (Fig. 4b). This more diffuse pattern of RAD expression at late stages of development is similar to that observed in dissected snapdragon flowers post anthesis (data not shown), suggesting a loss of spatial regulation in maturing flowers of both snapdragon and G. officinalis.
The floral symmetry gene network may be largely conserved across Veronicaceae. In early flower development, expression patterns of V. montana and G. officinalis floral symmetry gene homologs are largely similar to patterns of CYC/DICH and RAD expression in snapdragon. In snapdragon, early dorsal expression of these floral symmetry genes has been proposed to constrain organ number. This is based on the fact that cyc mutants often develop extra petals and stamens in the dorsal region of the flower (Luo et al., 1996). We have shown that VmCYC1, GoCYC1 and GoCYC2 are exclusively expressed in the dorsal region of V. montana and G. officinalis at early stages of development. This may suggest a conserved role for these genes in constraining organ number. Furthermore, both VmRAD and GoRAD are expressed in a similar manner to their CYC gene counterparts, suggesting conservation of the CYC-RAD regulatory interaction across Veronicaceae. This is despite an apparent lack of a CYC-like binding site in the first intron of VmRAD, which is disrupted in several of the well-characterized snapdragon rad mutants (Corley et al., 2005). However it is unknown whether there are CYC-like binding sites upstream of the VmRAD coding region, as have been found for AmRAD (Costa et al., 2005).
In addition to a role in organ number determination, early asymmetric expression of the snapdragon dorsal floral symmetry genes has been proposed to delay organ development in the dorsal region of the flower. This is shown by the loss of asynchronous sepal and petal organ initiation in snapdragon cyc mutants (Luo et al., 1996). Our study found a similar positive correlation between VmCYC1 gene expression and organ growth. However, a detailed analysis of floral development in G. officinalis, revealed a negative correlation between GoCYC1/GoCYC2 expression and early organ growth. Since snapdragon and V. montana are more closely related to each other than to G. officinalis, this may indicate that the early role of CYC-like genes in constraining dorsal organ growth is a derived function within Veronicaceae. Alternatively, this function may have been lost in the lineage leading to G. officinalis, or may have been modified to increase organ growth, similar to the late role of the CYC-like gene in dorsal petal development of snapdragon (Luo et al., 1996). These last explanations may be more likely given that CYC-like genes have been implicated in early floral organ growth suppression in divergent taxa outside Veronicaceae, including Bournea leptophylla (Gesneriaceae) and Arabidopsis thaliana (Brassicaceae) (Cubas et al., 2001; Zhou et al., 2008).
In addition to their early role in flower development, a second, and more striking, role for CYC-like and RAD-like gene products in dorsal petal and stamen morphogenesis has been shown. For example, cyc:dich double mutants, rad and Lcyc mutants of snapdragon and toadflax (Linaria vulgaris), respectively, develop peloric (radialized) flowers in which dorsal and lateral petals develop with ventral identity, and a fertile dorsal stamen develops in place of the wild-type staminode (Luo et al., 1996, 1999; Cubas et al., 1999; Corley et al., 2005). These data illustrate that CYC, DICH and RAD gene products are necessary to specify dorsal identity in these bilaterally symmetrical flowers. Expression data from this study supports a similar role for CYC-like and RAD-like genes in V. montana and G. officinalis. At early stages of floral organ morphogenesis, VmCYC1 and VmRAD are expressed in the dorsal region of the flower, becoming detectable only in the dorsal petal at late stages of development (Figs 5, 7). Similarly, GoCYC1, GoCYC2 and GoRAD are expressed in the dorsal petals and dorsal staminode during early to late stages of G. officinalis flower development, with much weaker expression in sepals, lateral and ventral petals, and gynoecium (Figs 6, 7). A similar result was recently found for B. leptophylla (Zhou et al., 2008).
CYC-like gene duplication and divergence
Duplicated genes contribute to the raw genetic material on which selection can act and, therefore, may be important for the evolution of form (Ohno, 1970; Force et al., 1999; Hughes, 1999). Genetic redundancy following gene duplication events promotes relaxed selection on one or both paralogous genes, resulting in loss of gene function (nonfunctionalization), partitioning of ancestral gene function (subfunctionalization) or gain of function (neofunctionalization) (Ohno, 1970; Nei & Roychoudhury, 1973; Force et al., 1999; Lynch et al., 2001). Previous studies have demonstrated that duplication events in the CYC-like gene lineage have resulted in functional divergence. For example, snapdragon CYC and DICH are only partially redundant in function (Luo et al., 1999). CYC alone is responsible for defining organ number in the dorsal region of the flower, whereas DICH alone is involved in shaping internal asymmetry during development of the dorsal petals (Luo et al., 1996, 1999). In the lineages leading to both V. montana and G. officinalis, duplication events independent of the CYC/DICH duplication, have occurred in the CYC gene family, giving rise to two and three CYC-like paralogs, respectively (Fig. 3). Analyses comparing VmCYC genes show that VmCYC2 is expressed more broadly than VmCYC1, and in different regions of the flower. Furthermore, VmCYC2 expression is not confined to the inflorescence; it is also found in vegetative tissues (Fig. 4a). Unlike GoCYC1 and GoCYC2, GoCYC3 is also expressed in vegetative tissues, but the expression level is low in vegetative and inflorescence organs alike (Fig. 4b). These divergent patterns of expression may suggest some loss of function for VmCYC2 and GoCYC3 in specifying dorsal identity. Future analyses of gene function will be important to determine whether these regulatory changes can be implicated in novel aspects of developmental patterning.
CYC and RAD gene expression is not correlated with the derived pattern of stamen arrest in V. montana or G. officinalis
Data from this study are not consistent with the hypothesis that CYC and/or RAD genes have been coopted for ventral stamen abortion in V. montana or G. officinalis. Hileman et al. (2003) demonstrated that in desert ghost flower CYC and DICH orthologs are expressed in lateral, as well as dorsal staminodes. Although not tested functionally, this positive correlation between CYC/DICH expression and stamen arrest suggested a genetic mechanism by which patterns of stamen arrest could evolve within Lamiales. A similar mechanism was recently hypothesized to explain derived patterns of lateral stamen abortion in Chirita heterotricha (Gesneriaceae) (Gao et al., 2008a). In our study, expression of CYC and RAD genes in V. montana and G. officinalis did not positively correlate with patterns of stamen abortion. VmCYC1 and VmRAD were expressed in the dorsal region; no expression was observed in the ventral region at any stage of development (Figs 5, 7). Similarly, GoCYC1, GoCYC2 and GoRAD were all expressed in the dorsal staminode, but there was no discernable expression in the ventral staminodes (Figs 6, 7). Although VmCYC2 and GoCYC3 have a derived expression pattern compared with VmCYC1, and GoCYC1 and GoCYC2, respectively, the difference was found generally across the flower, indicating a possible loss of function in specifying dorsal identity. Unlike snapdragon DICH (Luo et al., 1999), none of the V. montana or G. officinalis CYC-like genes showed an asymmetric pattern of expression within the dorsal petals. This further suggests that the role of DICH in determining internal organ asymmetry is not shared by homologous genes in other Veronicaceae species, and may have evolved more recently in the lineage leading to snapdragon.
Implications for understanding evolution of stamen abortion in Lamiales
Nearly all species of Lamiales develop bilaterally symmetrical flowers with four to five stamens (Endress, 1998; Reeves & Olmstead, 1998). However, there have been many evolutionary modifications to this floral plan. Independent shifts in stamen number have occurred in Veronicaceae, as well as other families in the order (Endress, 1998; Reeves & Olmstead, 1998). Dorsal stamen arrest in snapdragon and toadflax is known to be under the control of CYC (Luo et al., 1996; Cubas et al., 1999), and expression analyses from this and other studies suggest a similar role for CYC homologs in other species of Lamiales (Hileman et al., 2003; Gao et al., 2008a; Zhou et al., 2008). Like other TCP genes, CYC genes are postulated to negatively regulate stamen growth by directly regulating genes involved in cell division (Gaudin et al., 2000). Thus, changes in CYC expression have been hypothesized to underlie derived patterns of lateral and/or ventral stamen abortion across Lamiales.
Evidence that CYC genes may regulate lateral stamen growth comes from functional analyses in snapdragon, and gene expression analyses in desert ghost flower and Chirita heterotricha (Luo et al., 1996; Hileman et al., 2003; Gao et al., 2008a). Lateral stamens of snapdragon are reduced in size compared with ventral stamens, suggesting a gradient of effect for CYC along the dorsiventral axis, similar to that in the petal whorl (Luo et al., 1996; for review see Kalisz et al., 2006). An even more striking difference in the size of lateral versus ventral stamens is evident in desert ghost flower and C. heterotricha, where CYC expression is much more obvious within the lateral staminodes and staminodes + petals, respectively (Hileman et al., 2003; Gao et al., 2008a). By contrast, results from our study suggest a different genetic mechanism for ventral stamen abortion, at least in V. montana and G. officinalis. Given that CYC genes may regulate cell cycle genes directly, this may imply recruitment of an alternative genetic pathway affecting cell cycle regulation within the ventral region of the flower.
Apart from the floral symmetry gene network, genetic pathways known to be involved in stamen abortion have best been characterized in species that develop at least some unisexual flowers, such as Zea mays (maize) (for review see Irish & Nelson, 1989). In maize, stamens are initiated in all florets. However, at later stages of development, stamen development is arrested in the female inflorescence (ear), resulting in florets that are functionally female (Cheng et al., 1983). Maize mutants, such as the dwarf mutants and anther ear 1 (An1), have revealed an important role for the gibberellin biosynthetic pathway in stamen abortion (Phinney, 1956; Coe et al., 1989; Bensen et al., 1995). Specifically, mutant plants that produce low levels of gibberellic acid (GA) are unable to repress stamen abortion, resulting in the development of masculinized ears (Dellaporta & Calderon-Urrea, 1994). Although stamen abortion occurs uniformly across ear florets, maize develop both female and male (tassel) inflorescences on the same plants. This suggests some mechanism for spatially defined regulation of the gibberellin (GA) pathway. Furthermore, in A. thaliana and tomato (Solanum lycopersicon, Solanaceae), specific DELLA-domain proteins act as repressors of GA biosynthesis, in some cases resulting in reduced fertility owing to reduced pollen development and anther filament elongation (Jacobsen & Olszewski, 1991; Cheng et al., 2004; Tyler et al., 2004; Yu et al., 2004; reviewed in Fleet & Sun, 2005; Gao et al., 2008b). Thus, genes involved in the GA biosynthesis pathway may be promising alternative candidates for stamen abortion in Lamiales, particularly in cases where these derived patterns cannot be explained by regulatory changes in CYC-like genes. Alternatively, nonCYC-like TCP genes, and non RAD-like MYB genes may have been independently coopted to regulate cell cycle genes leading to ventral stamen abortion in these species.
We thank Heather Shinogle for training and assistance on the scanning electron microscope, Laryssa Baldridge for assistance with gene isolation, and two anonymous reviewers for comments on an earlier version of the manuscript. This work was supported by NSF (IOS-0616025) to L.C.H.