Transference of function shapes organ identity in the dove tree inflorescence


Author for correspondence:
Koen Geuten
Tel: +32 0 16321541


  • An important evolutionary mechanism shaping the biodiversity of flowering plants is the transfer of function from one plant organ to another. To investigate whether and how transference of function is associated with the remodeling of the floral organ identity program we studied Davidia involucrata, a species with conspicuous, petaloid bracts subtending a contracted inflorescence with reduced flowers.
  • A detailed ontogeny enabled the interpretation of expression patterns of B-, C- and E-class homeotic MADS-box genes using qRT-PCR and in situ hybridization techniques. We investigated protein–protein interactions using yeast two-hybrid assays.
  • Although loss of organs does not appear to have affected organ identity in the retained organs of the reduced flowers of D. involucrata, the bracts express the B-class TM6 (Tomato MADS box gene 6) and GLOBOSA homologs, but not DEFICIENS, and the C-class AGAMOUS homolog, representing a subset of genes also involved in stamen identity.
  • Our results may illustrate how petal identity can be partially transferred outside the flower by expressing a subset of stamen identity genes. This adds to the molecular mechanisms explaining the diversity of plant reproductive morphology.


Our understanding of the development of angiosperm flowers is mainly the result of extensive molecular and genetic studies in a few model plants. Yet little is known about the developmental and genetic mechanisms that shape morphologies that are different from those of the model species. A suitable strategy for studying these cases involves investigating regulatory genes known in model plants to control a certain trait, in species that display a variation of this trait. For instance, it was proposed on the basis of the well-known ABC model (Schwarz-Sommer et al., 1990; Coen & Meyerowitz, 1991) that petal-like characters – generally described as petaloidy – when present in organs other than petals could be controlled by the heterotopic expression of the genes controlling petal development (van Tunen et al., 1993; Bowman, 1997; Albert et al., 1998). This hypothesis is supported by constitutive expression experiments in Arabidopsis: the combined expression of SEPALLATA (SEP), APETALA3 (AP3) and PISTILLATA (PI) transforms leaf-like organs into petal-like organs (Honma & Goto, 2001; Pelaz et al., 2001). The model originally proposed by van Tunen et al. (1993) has been tested for first-whorl petaloid floral organs, mainly in species belonging to basal angiosperms (Jaramillo & Kramer, 2004; Kim et al., 2005; Chanderbali et al., 2006), monocots (Tzeng & Yang, 2001; Kanno et al., 2003; Park et al., 2003, 2004; Tsai et al., 2004, 2005; Nakamura et al., 2005; Xu et al., 2006) or basal eudicots (Kramer et al., 2003; Di Stilio et al., 2005, 2009). In most of these species, first- and second-whorl organs appear identical and a bipartite perianth with green sepals and attractive petals is not present. The results from these analyses corroborate the hypothesis overall, yet several exceptions have been described. In species of genera such as Tulipa gesneriana, Agapanthus praecox ssp. orientalis, Dendrobium crumenatum or Ranunculales–representatives, petaloidy in the first-whorl organs is indeed associated with heterotopic expression of B genes (Kanno et al., 2003; Kramer et al., 2003; Nakamura et al., 2005; Xu et al., 2006). In other species, however, often closely related to these, no correlation between heterotopic petaloidy and heterotopic expression of AP3 and PI genes has been observed (e.g. Aristolochia manshuriensis, Asparagus officinalis (L.) and Lilium longiflorum) (Tzeng & Yang, 2001; Park et al., 2003, 2004; Jaramillo & Kramer, 2004; Tsai et al., 2004). This suggests that a mechanism of maintained heterotopic expression of both AP3 and PI homologs is not applicable to all cases of heterotopic petaloidy.

The petal identity program is thought to have undergone several changes in the diversification of angiosperms, with a major one at the base of core eudicots when the bipartite perianth characteristic of this group originated (reviewed in Irish, 2009). Because petal identity is best understood in core eudicot model plants, we might expect this knowledge to be best applied to cases of heterotopic petaloidy within core eudicots. However, studies of heterotopic petaloidy within core eudicots are very limited in number. The only published study thus far suggests that in Impatienshawkeri (Balsaminaceae, Ericales, basal asterids) the petaloid identity program could be controlled by genes other than AP3 and PI, as a SEPALLATA3-like (SEP3) gene is expressed in the spurred petaloid sepal of Impatiens while the relative expression level of AP3 and PI homologs in these organs is low late in development (Geuten et al., 2006). The diversity of petaloidy, which can be observed at the developmental and morphological level, also suggests that the molecular mechanisms establishing this trait can be expected to be diverse. Petal identity is composed of a number of morphological characters. It is generally associated with conical or elongate epidermal cells (Weberling, 1989). These are usually ornamented with ridges of thickened cuticle (Whitney & Glover, 2007) and can produce aromatic compounds (Dudareva & Pichersky, 2000). Characteristic of petals is also the absence of chlorophyll (Mara et al., 2010), but specific pigments can be produced to provide color (reviewed in Grotewold, 2006). Organs other than petals can appear petaloid by expressing traits commonly associated with petal identity. These organs can be part of the flower, such as sepals or stamens, but in other cases this function is transferred to extra-floral organs, such as bracts. Such bracts frequently develop in combination with strongly reduced inflorescences in Anthurium sp. (Araceae), Justicia brandegeana (Acanthaceae), Euphorbia poinsettia (Euphorbiaceae) and Cornus sp. (Cornaceae) (Kramer & Jaramillo, 2005).

In the small order Cornales, a number of species develop such petaloid bracts. The most widely cultivated examples are dogwoods belonging to the genus Cornus. Another species illustrating heterotopic petaloidy is the dove tree, or handkerchief tree, Davidia involucrata. Phylogenetically, Davidia belongs to the family Nyssaceae, also in the order Cornales, which is placed at the base of the asterid clade. Davidia involucrata is best known for its intriguing pair of white bracts (‘the wings of the dove’) that subtend a remarkable reproductive structure. The latter is so condensed that individual flowers cannot be discerned. This complexity resulted in problematic morphological descriptions and controversial homology interpretations, for instance different interpretations of what comprises male flowers in the inflorescence and controversy over the presence of sepals (Harms, 1898; Horne, 1909; Hutchinson, 1959; Engler, 1964; Cronquist, 1981; Takhtajan, 1997). By studying D. involucrata we aim to answer two questions: ‘Can floral organ identity be functionally transferred outside the flower to extrafloral bracts and in which form?’ and ‘How is the floral organ identity program expressed when several floral organs are missing?’ Both questions relate to the more general question of plasticity in genetic mechanisms controlling organ identity and how this shapes variations in morphology and creates floral diversity. With these aims, we provide a detailed ontogeny of the reproductive development of D. involucrata, which allows us to interpret gene expression patterns of organ identity genes in both early and late development. As only a subset of B-class genes in combination with the C-class AGAMOUS (AG) homolog are expressed in the petaloid bract of D. involucrata, our data seem to suggest that developmental pathways can be modified when partially transferred in evolution from one organ to another.

Materials and Methods

Scanning electron microscopy

The material, collected from three different plants in the National Botanical Garden of Belgium and in the Arboretum in Wespelaar (Belgium), was fixed in FAA (70% ethanol:acetic acid:40% formaldehyde, 90 : 5 : 5). Inflorescence buds were dissected in 70% ethanol under a Wild M3 stereomicroscope (Wild Heerbrugg Ltd, Heerbrugg, Switzerland). This material was washed twice in 70% ethanol and dehydrated in a 1 : 1 mixture of 70% ethanol and dimethoxymethan (DMM) for 5 min and in pure DMM for 20 min. After critical-point drying (CPD 030; BAL-TEC AG, Balzers, Liechtenstein), the dried material was mounted on aluminum stubs using Leit-C and coated with gold (SPI Module Sputter Coater; Spi Supplies, West Chester, PA, USA) before observation with a JEOL JSM-6360 SEM (Jeol Ltd, Tokyo, Japan).

Cloning of MADS-box genes

Inflorescences were frozen in liquid nitrogen and stored at −80°C. Total RNA was isolated using the Invisorb Spin Plant RNA Mini Kit (Invitek, Berlin, DE) or Trizol (Invitrogen, Carlsbad, California, USA). The mRNA was reverse-transcribed into cDNA using Avian Myeloblastosis Virus (AMV) reverse transcriptase (Promega, Madison, Wisconsin, USA) with an oligo-dT primer (Kramer et al., 1998). We adopted the priority rule for naming the cloned genes. According to this rule the name of the first published sequence (date of acceptance of publication) is used. DiAGAMOUS (DiAG) was amplified using a degenerate forward primer, RQVT (5′-CGRCARGTGACSTTCTSCAARCG -3′), and a PCR program taken from the literature (Kramer et al., 1998; Winter et al., 1999). To amplify DiAGAMOUS-like9 (DiAGL9), a degenerate forward primer (AGL9f: 5′- GAGCTYTCBGTYCTHTGYGAYGCKGAG-3′) was constructed based on a SEPALLATA-alignment of core eudicot species and a different PCR program (35 cycles of 30 s of denaturation at 95°C, 30 s of annealing at 55°C and 60 s of extension at 72°C) was used. DiAGAMOUS-like2 (DiAGL2), DiFLORAL BINDING PROTEIN9-1 and 2 (DiFBP9-1/2) and DiDEFICIENS (DiDEF) were isolated using specific primer combinations based on sequences of closely related species. For DiAGL2 and DiFBP9-1/2, the primer combination AGL234f (5′- TGTGAYGCKGARGTTGCYCTCATCAT-3′) and AGL 234r (5′- AAGCATCCAMCCMGGRATRSASCCATT-3′) was used (annealing at 53°C). DEFf (5′-TGAATACATCAGTC- CTTCCA-3′) in combination with DEFr (5′- AACTAACCCA- TAGTGTGGATC-3′) was used to isolate DiDEF with a 50°C annealing temperature. All PCR amplifications were carried out using Taq DNA polymerase (Invitrogen). PCR products were gel-purified using the Nucleospin Extract 2 kit (Macherey-Nagel, Düren, DE) and cloned into the pGEM-T vector (Promega). After transformation, between 50 and 100 white clones were checked for inserts in a PCR reaction using the same primers and program. Plasmid DNA for selected clones was extracted with a Nucleospin Plasmid kit (Macherey-Nagel) or a PureYield Plasmid Miniprep System kit (Promega). The plasmid inserts were sequenced using T7 and SP6 universal primers and the BigDye Terminator 1.1 kit (Applied Biosystems, Foster City, CA, USA) on an Applied Biosystems 310 sequencer or the plasmids were sent for sequencing (MacroGen Inc., Seoul, Korea). For each gene, 20–50 clones were characterized. New sequences were deposited in GenBank with accession numbers JN705939–JN705945.

Phylogenetic analysis

After using the Blast search engine at the National Center for Biotechnology Information (NCBI) website for preliminary identification of the newly isolated sequences, we generated data matrices (See Supporting Information Table S2 for a list of species used in the phylogenetic analysis, with abbreviations) with selected representatives from B-, C- and E-class MADS-box subfamilies to identify our sequences as true AGL2 (DiAGL2), AGL9 (DiAGL9), FBP9 (DiFBP9-1 and 2), AG (DiAG) and DEF (DiDEF) orthologs. Taxon sampling was designed to include members of different MADS-box gene subfamilies and species of basal angiosperms, basal eudicots and core eudicots, and was depending on the availability of sequences. The obtained nucleotide sequences were manually aligned for each subfamily using MacClade4 (Maddison & Maddison, 2003). After alignment, nucleotide sequences were analyzed using MrBayes 3.1.2 (Huelsenbeck & Ronquist, 2001) and phyml (Guindon & Gascuel, 2003). Modeltest 3.06 (Posada & Crandall, 2001) selected the GTR+I+G substitution model using the Akaike information criterion for each data set. MrBayes was run for 3 million generations and every 100 generations one tree was saved. Searches quickly reached stationarity at c. 40 000 generations. This number was considered the ‘burn-in period’ and the trees generated from the 40 000 generations were excluded when the consensus phylo-geny was constructed. phyml was used for maximum likelihood inference of the matrices. Node support was estimated by bootstrap analyses with 150 replicates. The most likely tree was used and Bayesian posterior probability (BPP) values (> 90) and bootstrap values (> 70) were plotted on the tree. The resulting trees can be found in Supporting Information Figs S1–3.

qRT-PCR quantification of gene expression

To examine the expression patterns of selected MADS-box genes using qRT-PCR, young and mature inflorescences of D. involucrata were dissected and frozen in liquid nitrogen. We define ‘young’ as the early-bud stadium in which the inflorescences are c. 2–4 mm and the bracts are still green and covered with a dense layer of one- or two-celled hairs. The ‘mature’ stadium was collected several weeks after the ‘young’ stadium, when the inflorescences are c. 6–10 mm. This stadium is found immediately before anthesis, when the bracts are lighter green and the layer of hairs is already less dense. RNA was extracted using the Invisorb Spin Plant RNA Mini Kit (Invitek) or Trizol (Invitrogen) and each RNA sample was DNase-treated using TURBO DNA-free (Ambion, Austin, TX, USA). To confirm the absence of contaminating genomic DNA in the total RNA, we used a PCR reaction (40 cycles) with actin primers (data not shown). Total RNA was reverse-transcribed using AMV reverse transcriptase (Promega) and the oligo-dT and random primers. qPCR was performed on a StepOne Plus cycler (Applied Biosystems) using Fast SYBR Green Master Mix (Applied Biosystems). Primers were designed using the Primer Express software (Applied Biosystems). The data presented here are the average of two technical replicates with standard error of the mean, and two biological replicates are shown. The biological replicates represent different inflorescence buds at the same stage of development, collected at the same time from two different trees. All samples are normalized against ACTIN (DiACTIN), which was cloned using a specific primer pair. Data were analyzed using the delta CT-method (Applied Biosystems User Bulletin No. 2). The ACTIN sequence was deposited in GenBank (JN705945). For all primers used, see Table S1.

In situ hybridization

Antisense and sense (used as control) probes were in vitro transcribed using T7 RNA polymerase (Fermentas, Burlington, CA, USA) in the presence of digoxigenin-labeled UTP (Roche, Basel, Switzerland) from a PCR-amplified template that included a T7 promoter in the primer sequence. Tissues were fixed overnight in cold 4% paraformaldehyde buffer, paraffin embedded and sectioned at 8 μm. Sections were mounted on Probe-On-Plus slides (Fisher Scientific, Pittsburgh, PA, USA). Prehybridization, hybridization and detection were essentially performed following Carr & Irish (1997). Sense (control) and antisense experiments were conducted simultaneously. Primers used are listed in Table S1.


To investigate whether the transition from green to white in the bracts of D. involucrata is associated with loss of chlorophyll, we performed spectrophotometric measurements. Vegetative leaves, young bracts and fully grown bracts were frozen in liquid nitrogen. The same weight of ground material for every sample was extracted for a few hours in the dark with cold, pure acetone. Following extraction, samples were vortexed and spun down for 5 min to remove cellular debris. Absorbance was measured at 665 nm (chlorophyll a) and 470 nm (chlorophyll b) (Genesys 5; Spectronic Instruments, Rochester, NY, USA).

Yeast two-hybrid assays

The partial cDNA sequences were cloned in-frame with the yeast GAL4 galactosidase transcriptional activation domain in vector pGADT7 or with the GAL4 DNA-binding domain in vector pGBKT7. Constructs were co-transformed into the yeast AH109 strain. Activation of the beta-galactosidase reporter was monitored using the o-nitrophenyl-b-D-galactosid (ONPG) substrate, measured at 420 nm and converted into beta-Gal units using the cell density measured at 550 nm.


The inflorescence head of Davidia involucrata resembles a flower

The reproductive organs of the tree species D. involucrata are present in bud in autumn and open in spring. The most notable feature of these reproductive structures is the presence of two, rarely three, creamy white bracts that surround the male or bisexual globose reproductive structure (andromonoecious) (Fig. 1). When the structure is male, it consists of numerous dark purple stamens (Fig. 1a–c); if it is bisexual, one greenish female reproductive structure on which a whorl of stamens can be observed is obliquely situated on the globose head and is surrounded by additional stamen clusters (Fig. 1d–f). As only part of the style and the carpel tips are elevated above the stamen insertion, the ovary is inferior (Fig. 1d). Any indication of a perianth associated with the male or female organs is completely lacking, while the function of attracting pollinators is transferred to the petaloid bracts, causing the reproductive unit to resemble and function as a single flower.

Figure 1.

(a) Inflorescence of Davidia involucrata. (b–c) Inflorescence containing male flowers subtended by two petaloid bracts. (d–f) Inflorescence containing male flowers and one hermaphrodite flower. b, bract; hf, hermaphrodite flower; mf, male flowers.

Petaloidy of bracts is associated with loss of chlorophyll and a characteristic epidermal cell pattern

In young stages, five or six morphologically identical leaf-like organs enclose the reproductive structure (Fig. 2a). These are green and covered with a dense layer of one- or two-celled hairs. The upper two, or very rarely three, eventually develop into white, petaloid bracts that can differ in size. This color change is associated with the loss of chlorophyll (Fig. 2b) and the development of a puzzle-shaped epidermal cell pattern (Fig. 2c,d) that differs from that of the other three or four leaves, which are much more polygonal-shaped, especially on the adaxial side (Fig. 2e,f). No conical cell type, often characteristic of petals, was observed, illustrating the difference between the petaloid bracts and petals. The epidermal hairs are shed during development, starting from the base and ending at the top of the leaf (Fig. 2g,h). This shedding is observed in both bracts and leaves (Fig. 2g–j). Therefore, a functional association of petaloidy with the shedding of epidermal hairs seems unlikely.

Figure 2.

Development of the bracts and vegetative leaves of Davidia involucrata. (a) Inflorescence structure of D. involucrata. (b) Absorbance of chlorophyll in vegetative leaves, and young and old bracts (pale gray bars, chlorophyll b; dark gray bars, chlorophyll a). (c–d) Puzzle-shaped cell pattern on the adaxial (c) and abaxial (d) sides of the bracts. (e–f) Epidermal cell pattern on the adaxial (e) and abaxial (f) sides of the vegetative leaves. (g–h) Shedding of epidermal hairs on the bracts. (i–j) Shedding of epidermal hairs on the vegetative leaves. b, bract; mf, male flowers; vl, vegetative leaves.

Stamen clusters can be interpreted as male perianthless flowers

During early developmental stages of the globose reproductive structure, numerous bulges initiate on its surface (Fig. 3a,b). Subsequently, these bulges develop stamen primordia in irregular circles and ellipses in random numbers between two and seven (Fig. 3c,d). These clusters of stamens are distinct in young developmental stages, but they become difficult to discern later in development (Fig. 3e,f). The fact that in ontogeny stamens develop in clusters on a defined outgrowth of the receptacle suggests homology of these stamen clusters with male flowers. Therefore, the reproductive structures bearing these flowers can be considered to be inflorescences. Anthers of mature stamens are dithecal and tetrasporangiate, they have a pointed connective and open by longitudinal slits (Fig. 3g). No rudiments of a perianth or a gynoecium are present in any of the observed developmental stages (Fig. 3h).

Figure 3.

Development of the male flowers of Davidia involucrata. (a) Globose reproductive structure with male flower primordia. (b) Detail of male flower primordia before the initiation of stamen promordia. (c) Male flowers with initiation of stamen primordia. (d) Development of stamen primordia on the male flowers. (e) Mature stamens. (f) Side-view of mature male flowers. Stamens have been removed. (g) Dithecal, tetrasporangiate anther. (h) Mature male flower with all but one stamen removed. mf, male flower (primordia); re, receptacle; s, stamen (primordia).

The bisexual structure can be interpreted as a single epigynous perianthless flower

More or less simultaneously with the development of the male flowers, in some inflorescence heads, a gynoecium primordium develops as a large cylindrical structure toward the side covered by the larger bract (Fig. 4a). On the flanks of this primordium, 20–35 stamen primordia initiate to form an epigynous whorl (Fig. 4a,b). This stamen whorl defines the boundary between the congenitally fused ovary-receptacle tissue (below) and carpel tissue (above) (Fig. 4b,c). Subsequently, a variable number of six to eight carpel tips become visible (Fig. 4b,c). They extend and fuse to form a solitary, ridged style, ending in a stigma with as many lobes as there are locules present in the inferior ovary (Fig. 4d,e). Within the ovary, the septa fuse basally, forming a central column on which the ovules are attached on axillary placentas (Fig. 4b,e,g). In each of the locules, only one unitegmic ovule develops (Fig. 4e–g). In young stages, the nucellus is pointing upward with the funiculus attached to the central column (Fig. 4e,f). Subsequently, the integument and the nucellus curve inward and the micropyle points toward the center (Fig. 4f). As a result of the subsequent growth of the ovary and the central column, the ovules tumble backward and start growing downward (Fig. 4g), resulting in the anatropy of the ovules (Fig. 4h,i). A few ovules eventually abort and only one to five remain after anthesis. During these rotations, unicellular intra-ovarian hairs originate underneath each developing ovule and grow up into the stylar canal (Fig. 4j). The same type of hairs can be observed around the base of this bisexual flower (Fig. 4k). Similar to the male flowers, a perianth is missing, although small lobes surrounding the stamens may refer to a rudimentary perianth (Fig. 4l). Together, these morphological and developmental observations of the bisexual structure indicate homology with a single epigynous perianthless flower.

Figure 4.

Development of the hermaphrodite flower of Davidia involucrata. (a–c) Early developmental stages of the hermaphrodite flower with the initiation of an epigynous whorl of stamens. (d) Stigma. (e) Axillary placentation of unitegmic ovule primorida. (f) Apical view of unitegmic ovules. (g) Developing ovules tumbling backward. (h–i) Mature anatrope ovules. (j–k) Intra-ovarian trichomes originate underneath the developing ovule and are also observed round the base of the hermaphrodite flower. In (k), the hermaphrodite flower has been removed. (l) Little lobes surrounding the stamens. c, carpel; f, funiculus; i, integument; iot, intra-ovarian trichome; n, nucellus; o, ovule; pl, placenta; r, receptacle (fused to inferior ovary tissue); s, stamen (primordia); st, stigma.

Cloned MADS-box cDNAs are members of the homeotic B, C and E lineages of MADS-domain transcription factors

To determine which homeotic MADS-domain transcription factors function in the inflorescence of D. involucrata, we cloned MADS-box genes belonging to the known floral homeotic lineages. Previously, two GLOBOSA-like genes, which resulted from a recent duplication event (DiGLO1 and DiGLO2; Viaene et al., 2009), and a TM6 homolog (DiTM6; Viaene et al., 2009) were identified as B-class gene representatives. We further isolated a third B-class gene, from here on termed DiDEFICIENS (DiDEF). Phylogenetic analysis places this gene inside the euAP3 clade (Fig. S1). In addition, we isolated DiAGAMOUS (DiAG) and identified it as a member of the euAG clade (Fig. S2). Finally, four SEPALLATA-like genes were identified belonging to different lineages within the SEPALLATA phylogeny (Zahn et al., 2005). These four members of the E-class lineage were identified as DiAGL9 in the SEPALLATA3 clade, DiAGL2 in the SEP1-2 clade and DiFBP9-1 and DiFBP9-2 in the FBP9 clade (Fig. S3). These latter two probably originate from a recent duplication event (Fig. S3).

qRT-PCR reveals expression of class-B, -C and -E floral homeotic genes in male and perfect flowers during anthesis

To understand where and when the identified MADS-box genes function in the inflorescence and floral development of D. involucrata, we first used quantitative RT-PCR. Relative gene expression levels of the identified class-B, -C and -E homeotic genes were measured in both young (Fig. 5) and mature dissected floral organs and bracts (Fig. S4). We found that, in these late stages of development, DiAGL2 was the only one of the studied genes (DiAGL2, DiAG, DiDEF, DiTM6, DiGLO1, DiGLO2, DiFBP9-1, DiFBP9-2 and DiAGL9) to show significantly elevated expression in the bracts relative to expression in the flowers (Figs 5, S4). Further, as could be predicted from studies of B-, C- and E-class genes in model plants (Carpenter & Coen, 1990; Sommer et al., 1990; Yanofsky et al., 1990; Jack et al., 1992; Tröbner et al., 1992; Bradley et al., 1993; Flanagan & Ma, 1994; Goto & Meyerowitz, 1994; Savidge et al., 1995; Mandel & Yanofsky, 1998; Ditta et al., 2004; Vandenbussche et al., 2004; De Martino et al., 2006), all studied genes show significant expression in the male and perfect flowers (Figs 5a–i, S4a–i). However, the B-class genes DiTM6, DiGLO1 and DiGLO2 were more strongly expressed in the male flowers (Figs 5c–e, S4c–e). Taking biological variability into account, we observed on average comparable expression levels for all other investigated MADS-box genes in both male and bisexual flowers (Figs 5a,b,f–i, S4a,b,f–i).

Figure 5.

Expression of DiAGAMOUS (DiAG) (a), DiDEFICIENS (DiDEF) (b), DiTomato MADS box gene 6 (DiTM6) (c), DiGLOBOSA1 (DiGLO1) (d), DiGLOBOSA2 (DiGLO2) (e), DiAGAMOUS-like2 (DiAGL2) (f), DiFLORAL BINDING PROTEIN9-1 (DiFBP9-1) (g), DiFBP9-2 (h) and DiAGL9 (i) in young developmental stages of Davidia involucrata using qRT-PCR. The y-axis shows the relative expression to ACTIN using the delta CT method. Gray and black bars represent biological replicates: see the section ‘qRT-PCR quantification of gene expression’ for further details.

In situ hybridization reveals expression of DiTM6, DiGLO and DiAG in the early developmental stages of the bracts of D. involucrata

To investigate the expression patterns of MADS-box genes in the primordial stages of floral organ development we used in situ hybridization. In the stamen clusters, in situ data were consistent with the qRT-PCR results. We detected expression of all class-B genes (DiTM6, DiDEF, DiGLO1/2,Fig. 6b,d,f), a class-E gene (DiAGL9; Fig. 6c) and the class-C gene (DiAG; Fig. 6e) in stamen primordia or early stages of stamen development, similar to the expression of B, C and E genes in model plants (Yanofsky et al., 1990; Goto & Meyerowitz, 1994; Jack et al., 1992; Mandel & Yanofsky, 1998; Rijpkema et al., 2006). Interestingly, expression was always confined to the primordia and the developing filaments or anthers and was never observed in the base of the stamen clusters, suggesting that the base of the clusters is receptacle flower tissue on which stamen primordia develop. This interpretation would be consistent with the stamen clusters being male flowers.

Figure 6.

In situ hybridization data in male flowers (a–f), hermaphrodite flowers (g–l) and bracts (m–r) of Davidia involucrata for DiTomato MADS box gene 6 (DiTM6) (b, j, o), DiAGAMOUS-like9 (DiAGL9) (c, l, r), DiDEFICIENS (DiDEF) (d, h, q), DiAGAMOUS (DiAG) (e, k, p), and DiGLOBOSA1/2 (DiGLO1/2)) (f, i, n). a, anther; b, bract; c, carpel; o, ovules.

In the developing bisexual flower, expression of each of the B, C and E genes was again detected in the stamens localized on the inferior ovary (Fig. 6h–l). In addition, expression of DiTM6, DiAG and DiAGL9 was detected weakly in the carpel tissue above the whorl of stamens but was relatively strong in the developing ovules (Fig. 6j–l). Also these expression patterns are consistent with expression patterns observed in model plants, as TM6 is regulated like a C-class rather than a B-class gene (De Martino et al., 2006; Rijpkema et al., 2006; Geuten & Irish, 2010).

Unexpected expression patterns were observed in the developing bracts of D. involucrata. Two class-B genes (DiGLO1/2 and DiTM6) were clearly detectable in the young bracts, while the relative expression of these genes was low in qRT-PCR experiments (Fig. 6n–o). Also for the class-C gene (DiAG) we clearly detected expression using in situ hybridization in bracts (Fig. 6p). This suggests that, although expression of these genes is activated, it is not strongly maintained in this organ. By contrast, DiAGL2, another SEP homolog, which we did not investigate using in situ hybridization, showed strong expression in bracts (Fig. 5f). We did not observe expression of the SEP3 homolog DiAGL9 or of the class-B gene DiDEF in early developmental stages (Fig. 6q–r), while a clear signal was observed in male floral organs (Fig. 6c–d), suggesting that this undetected expression indeed reflects absence of activation or an expression level below the level that we were able to detect. All these expression patterns were consistently observed in each inflorescence bud examined.

In support of these signals early in development reflecting gene expression, rather than aspecific binding, is the fact that a sense probe of DiTM6 did not result in any signal (Fig. S5). The fact that, in these early developmental stages of bract development, some genes appear to be expressed, while others are not expressed in the bracts but are expressed in other organs, further validates the specificity of the obtained signals. The combination of the in situ hybridization data and relative expression levels obtained using qRT-PCR are indicative of a relative decrease of the expression of DiGLO1/2, DiAG and DiTM6 during bract development.

The genes co-expressed in D. involucrata reproductive development interact at the protein level

We found that the MADS-box transcription factors DiTM6, DiGLO1/2 and DiAG are co-expressed in the early stages of the developing bract of D. involucrata. In addition, strong expression of DiAGL2 was observed in fully developed bracts. Because MADS-domain transcription factors are required to interact at the protein level to bind DNA and control downstream targets, we tested these proteins in addition to DiAGL9 in yeast two-hybrid assays (Table 1). For the class-B proteins, we found that DiTM6 strongly activates the reporter in combination with either DiGLO1 or DiGLO2. No interaction was found between any of the B proteins and the C-class protein DiAG in our assays, as could be expected from previous studies (e.g. Honma & Goto, 2001). None of the B-class genes was able to form homodimers, which is consistent with observations in other eudicots (e.g. Winter et al., 2002). The SEPALLATA3-like protein DiAGL9 autoactivates the reporter, probably as a result of a C-terminal activation domain, but this is not the case for DiAGL2. Both E-class proteins have the capacity to homodimerize and also heterodimerize with each other and with DiAG. The B-class protein DiGLO1 heterodimerizes with DiAGL9 and DiAGL2 in both directions, while DiGLO2 only does so when fused to the GAL4 activation domain, which may indicate that these proteins have acquired somewhat different functions.

Table 1.   Protein interaction as characterized by yeast two-hybrid assays
  1. na, not analyzed.

  2. AD, activation domain; AG, AGAMOUS; AGL, AGAMOUS-like; BD, binding domain; GLO, GLOBOSA; TM6, Tomato MADS box gene 6.


The data from the yeast two-hybrid assays we performed are in close agreement with data acquired for model plant species, suggesting that the transfer of petaloidy toward the bracts in D. involucrata was associated in evolution mainly with spatial changes in expression patterns, rather than with changes in protein interaction specificity.


Loss of organs does not affect organ identity in retained organs in D. involucrata flowers

Organ identity is established through the overlapping expression of organ identity genes that typically function in establishing multiple identitities. Therefore, loss of floral organs may affect organ identity in retained organs. Both male and bisexual flowers of D. involucrata lack sepals and petals, and the gynoecium is lost in the male flowers. No ontogenetic evidence was found that points to rudimentary organs. Considering this floral morphology, we asked whether and how this affected the organ identity program expressed in the retained floral organs of D. involucrata.

In agreement with expression patterns and functional characterization in other plants, stamens in male and bisexual flowers express DiDEF, DiTM6, DiGLOBOSA and DiAG genes (Yanofsky et al., 1990; Goto & Meyerowitz, 1994; Jack et al., 1994; Rijpkema et al., 2006). Also for the female reproductive structure of the bisexual flower, the expression patterns of the female organ identity genes are consistent with observations of orthologous genes in other species (e.g. Bowman et al., 1991; De Martino et al., 2006; Rijpkema et al., 2006; Geuten & Irish, 2010). Although organ identity genes are expressed weakly in the carpel tissue above the stamen whorl, expression of the female organ identity genes DiTM6 and DiAG is strong in the developing ovules. Because the gynoecium is inferior, no clear distinction can be observed between carpels and receptacle tissue below the whorl of stamens and, except for DiAGL9, this tissue does not seem to express organ identity genes.

Taken together, although it seemed plausible that loss of floral organs might also influence the floral organ identity programs in other organs, our findings suggest that the reduced flowers of D. involucrata express organ identity genes corresponding to the identity of the developing floral organs. Thus, loss of organs does not appear to have affected organ identity in the retained organs of D. involucrata flowers.

The petaloid bracts express B- and C-class genes early in development

In D. involucrata, the bracts that subtend the inflorescence with reduced flowers are petaloid in the sense that they acquire their characteristic white color through the loss of chlorophyll, develop a specific epidermal cell pattern and become UV-absorbing (Burr & Barthlott, 1993). Their function is to facilitate pollinator attraction and to protect pollen from rain (Sun et al., 2008). We found that these bracts express a subset of organ identity genes early in development as determined by in situ hybridization: we observed expression of genes orthologous to GLOBOSA, TM6 and AGAMOUS. Based on functional characterization in model plants and as confirmed by early (using in situ hybridization) and late (using qRT-PCR) expression in D. involucrata stamen development, this combination of genes could be expected to be involved in the establishment of stamen identity, rather than petal identity. However, two experimental observations clearly distinguish the organ identity program of early bract development from early stamen development. First, while organ identity gene expression is maintained in the stamens, it is not in the bracts. In model plants, it has been established that elaborate feedback regulation exists to maintain the expression of organ identity genes throughout developmental stages (e.g. Schwarz-Sommer et al., 1992; Kaufmann et al., 2009), but such feedback regulation appears not to be functional in D. involucrata bracts.

Nevertheless, as the different targets of these gene products are often required at different developmental stages it has previously been suggested that the organ identity gene products are required throughout much of floral development (Bowman et al., 1989; Carpenter & Coen, 1990). This implies that the MADS-box gene products directly orchestrate the expression of different suites of genes at different times in development (Irish, 2011). However, in D. involucrata bracts, the MADS-box genes are only expressed in the early developmental stages. Therefore, only a subset of possible targets can be activated, possibly resulting in a partial petaloid phenotype. Indeed, while the bract color refers to petals, no typical petaloid epidermal cell pattern with conical cells was observed.

Secondly, DiDEF is not expressed early in bract development, while it is expressed throughout stamen development. This absence of DiDEF gene expression in bracts might provide an explanation for the absence of maintained organ identity gene expression in these organs by disrupting proper feedback regulation. In addition, in the absence of DiDEF, DiTM6 may acquire a more important role, without competition with DiDEF for certain targets. This may result in the gene products of DiAG, DiGLO1/2 and DiTM6 forming a protein complex that is responsible for establishing petaloid characteristics in the bracts.

Together, our data suggest that petaloidy of the bracts of D. involucrata, a species in which petals are completely absent, may be established early in development by a subset of organ identity genes more similar to a combination establishing stamen identity than to a combination establishing petal identity. However, although the co-expression of the genes in D. involucrata bracts and their interaction at the protein level seem to suggest an early role in bract development, we cannot exclude the possibility that the transient activity of these genes might also reveal they do not play a role in petaloid bract development.

Evolution of the petaloid bracts in D. involucrata

The reproductive diversity in the Nyssaceae–Mastixiaceae clade within Cornales may illustrate the evolutionary steps through which heterotopic petaloidy associated with perianth loss could have originated in D. involucrata (Fig. 7).

Figure 7.

Fifty per cent majority rule tree summarizing data from Xiang et al. (2011) resulting from phylogenetic analysis of combined sequences of nuclear 26 SrDNA and six cpDNA regions (rbcL, matK, ndhF, atpB, trnH-K and trnl-F) using MrBayes. The 50% rule trees with model partitions and the best tree from maximum likelihood analysis show relationships identical to this tree with similar support values (Xiang et al., 2011) All nodes are supported by posterior probability and bootstrap values of > 95%. The phylogenetic position of Amersinia is based on Manchester et al. (1999). Diagrams of the reproductive diversity in the Nyssoid–Mastixioid clade illustrate how heterotopic petaloidy in Davidia might have originated. In comparison with Mastixiaceae and other Cornales, the flower individuality within the Nyssaceae decreases because of loss of floral organs. In both Nyssa and Camptotheca, petals are reduced or absent, respectively, and the stamens are the most conspicuous floral organs. In Davidia, and maybe also in Amersinia, in the absence of petals, the inflorescences are even more condensed and resemble single flowers subtended by large petaloid bracts to compensate for the loss of floral attraction. Although inflorescences with only male flowers in Amersinia have not been found, it seems likely these have also existed (Manchester et al., 1999).

In the Mastixiaceae, and in general in the Cornales outside the Nyssaceae, the individual flowers are usually distinct without obvious organ reductions. In addition, the individual flowers have generally showy petals and are often organized in attractive inflorescences (Takhtajan, 2009). Within the Nyssaceae, however, flower individuality is further reduced and associated with the loss of floral organs. The overall greenish flowers of the genus Nyssa develop a minute calyx and greenish petals, but a nectary is present to reward pollinators. The female, male or bisexual flowers are clustered in dense inflorescences and female flowers in some species do not develop petals (Kubitzki, 2003; Takhtajan, 2009). In male and perfect flowers, the yellowish stamens appear to be the most conspicuous floral organs. In the genus Camptotheca the inflorescence is further reduced to a globular head with staminate or perfect flowers. The petals are again small, light green and deciduous (Kubitzki, 2003). It appears therefore that Camptotheca depends on the conspicuous white stamen filaments to attract pollinators. Also in D. involucrata, purple-colored UV-absorbing stamens are the most conspicuous floral organs and a perianth or a nectariferous disk is completely absent (Burr & Barthlott, 1993). In this species, large bracts developed to aid in attracting pollinators before the divergence of Davidia and the fossil genus Amersinia. In the latter extinct genus, four or five basal bracts developed rather than two (Manchester et al., 1999). As Amersinia also has a weakly developed perianth, the bracts might have appeared petaloid, as in Davidia. In both Davidia and Amersinia, the reduction in perianth organs is correlated with a globose inflorescence that resembles a flower.

Taken the findings of this and previous studies together, it appears that, in the Nyssaceae in general, the function of attracting pollinators is transferred from petals to stamens in combination with a reduction at both the floral and inflorescence levels. Moreover, in Davidia, and possibly also in Amersinia, it seems that bracts have become conspicuous petaloid organs to compensate for the loss of floral attraction as a result of severe reduction at both the floral and inflorescence levels.

Given the fact that in Nyssaceae petals are reduced or absent and have transferred their function to attract pollinators to stamens, a possible scenario is that, previous to the diversification of the Nyssaceae, petaloidy was transferred from petals to stamens. In the lineage leading to Davidia, the bracts would have received their petaloid characteristics from the colored stamens, resulting in expression patterns reflecting stamen identity, rather than petal identity. Yet we cannot exclude the scenario in which the origins of the conspicuous stamens and bracts are not linked in evolution and are therefore independent evolutionary events.

Conserved pathways may be modulated in diverse ways to establish petaloidy in different species

The occurrence of petaloidy in organs other than petals can frequently be observed in angiosperms and appears to have evolved multiple times in their diversification (Tzeng & Yang, 2001; Kanno et al., 2003; Kramer et al., 2003; Park et al., 2003, 2004; Jaramillo & Kramer, 2004; Tsai et al., 2004, 2005; Di Stilio et al., 2005; Kim et al., 2005; Nakamura et al., 2005; Chanderbali et al., 2006; Geuten et al., 2006; Xu et al., 2006). This may suggest that a single required mechanism was recruited multiple times to establish petaloidy or that different molecular mechanisms have the potential to establish petaloidy outside petals. The available experimental evidence is suggestive of an intermediate scenario in which conserved pathways can be recruited to establish petaloidy, but diverse modulations of these pathways can be expected in different species (Tzeng & Yang, 2001; Kanno et al., 2003; Kramer et al., 2003; Park et al., 2003, 2004; Jaramillo & Kramer, 2004; Tsai et al., 2004, 2005; Di Stilio et al., 2005; Nakamura et al., 2005; Geuten et al., 2006; Xu et al., 2006).

Little is known about the mechanisms that transfer petaloidy to organs outside the flower, as is the case in D. involucrata. Given the apparent diversity in mechanisms to transfer petaloid characteristics from one floral organ to another, it would not be surprising if extrafloral heterotopic petaloidy were established by a similar range of molecular mechanisms. The correlation we observed between heterotopic petaloidy and the combined heterotopic expression of genes normally taking part in stamen identity suggests that the expression patterns of DiAG, DiGLO and DiTM6 do reflect a function. A possible evolutionary scenario would be that, in Davidia, in the absence of petals, a protein complex of DiAG, DiGLO and DiTM6 acquired regulatory control over genes responsible for establishing specific petaloid characters, such as loss of chlorophyll. The marked differences from true stamen identity in petaloid bracts, such as the absence of expression maintenance and the absence of DiDEF expression, may illustrate how such an identity can be modulated when transferred in evolution outside the flower.


This work was supported by the Fund for Scientific Research-Flanders (Belgium) (FWO G.0418.08). D.V. acknowledges a fellowship from the Fonds voor Wetenschappelijk Onderzoek Vlaanderen and T.V. acknowledges a postdoctoral fellowship from K.U. Leuven. We would like to thank Viviane Leyman, Elke Bellefroid and Dirk de Meyere from the National Botanical Garden of Belgium and Koen Camelbeke from the Arboretum in Wespelaar for the collection of fresh flower material. We also thank Anja Vandeperre, Nathalie Geerts and Hilde Huyghe for technical assistance and support. K.G., D.V. and T.V. designed the study. D.V. and T.V. carried out the research. P.C. provided helpful comments on the floral ontogeny. The manuscript was written by D.V., K.G. and T.V.