Petal-specific subfunctionalization of an APETALA3 paralog in the Ranunculales and its implications for petal evolution


  • Bharti Sharma,

    1. Department of Organismic and Evolutionary Biology, Harvard University, 16 Divinity Ave., Cambridge, MA 02138, USA
    Search for more papers by this author
  • Chunce Guo,

    1. State Key Laboratory of Systematic and Evolutionary Biology, Institute of Botany, Chinese Academy of Science, 20 Nanxincun, Xiangshan, Beijing 100093, China
    Search for more papers by this author
  • Hongzhi Kong,

    1. State Key Laboratory of Systematic and Evolutionary Biology, Institute of Botany, Chinese Academy of Science, 20 Nanxincun, Xiangshan, Beijing 100093, China
    Search for more papers by this author
  • Elena M. Kramer

    1. Department of Organismic and Evolutionary Biology, Harvard University, 16 Divinity Ave., Cambridge, MA 02138, USA
    Search for more papers by this author

Author for correspondence:
Elena M. Kramer
Tel: +1 617 496 3460


  • The petals of the lower eudicot family Ranunculaceae are thought to have been derived many times independently from stamens. However, investigation of the genetic basis of their identity has suggested an alternative hypothesis: that they share a commonly inherited petal identity program. This theory is based on the fact that an ancient paralogous lineage of APETALA3 (AP3) in the Ranunculaceae appears to have a conserved, petal-specific expression pattern.
  • Here, we have used a combination of approaches, including RNAi, comparative gene expression and molecular evolutionary studies, to understand the function of this petal-specific AP3 lineage.
  • Functional analysis of the Aquilegia locus AqAP3-3 has demonstrated that the paralog is required for petal identity with little contribution to the identity of the other floral organs. Expanded expression studies and analyses of molecular evolutionary patterns provide further evidence that orthologs of AqAP3-3 are primarily expressed in petals and are under higher purifying selection across the family than the other AP3 paralogs.
  • Taken together, these findings suggest that the AqAP3-3 lineage underwent progressive subfunctionalization within the order Ranunculales, ultimately yielding a specific role in petal identity that has probably been conserved, in stark contrast with the multiple independent origins predicted by botanical theories.


Angiosperm petals are wonderfully diverse from a morphological standpoint. For over a century, this diversity has led botanists to conclude that second whorl petals – true petals – have evolved many times independently, in some cases from sterile bracts and, in others, from stamens (Prantl, 1888; Hiepko, 1965; Bierhorst, 1971; Takhtajan, 1980). With the elucidation of the genetic basis of floral organ identity, the ABC model, plant biologists acquired a set of tools that we hoped would shed light on the complex evolutionary history of petals. This model outlines the genetic code for each organ type: A alone for sepals, A + B for petals, B + C for stamens and C alone for carpels (Coen & Meyerowitz, 1991). The identification of the loci corresponding to these codes demonstrated that the majority were members of the type II MADS box transcription factors (Becker & Theissen, 2003). In Arabidopsis, this included the class A locus APETALA1 (AP1), the class B loci APETALA3 (AP3) and PISTILLATA (PI), and the C class gene AGAMOUS (AG) (reviewed by Causier et al., 2010). Numerous studies across the angiosperms have found that stamen and carpel identities are highly conserved, whereas sepal identity and A function in general are not (reviewed in Causier et al., 2010; Litt & Kramer, 2010). Petal identity falls in between, with evidence both for and against conservation (reviewed in Jaramillo & Kramer, 2007; Litt & Kramer, 2010). In the core eudicots, it appears that B gene homologs are required for petal identity in asterids and rosids, but further study is needed in the Caryophyllales, which have several instances of independently derived petals (Brockington et al., 2009). In the monocot grasses, B gene homologs function in lodicules in what appears to be a modified petal identity program (Ambrose et al., 2000; Whipple et al., 2007). There is also evidence for AP3/PI homologs promoting the identity of both the first and second whorls in monocots with entirely petaloid perianths (Kanno et al., 2007; Mondragon-Palomino & Theissen, 2008). That being said, investigation of the petaloid organs in Persea (Lauraceae) uncovered significant AG homolog expression, which may reflect their independent derivation from stamens (Chanderbali et al., 2006).

We must ask whether the widespread role of AP3/PI homologs in petal identity reflects a commonly inherited genetic program or represents convergence caused by the repeated modification of stamens into petals, which could involve repeated co-option of AP3 and PI (Kramer & Irish, 2000; Kramer & Jaramillo, 2005; Jaramillo & Kramer, 2007; Irish, 2009). For most groups, there is essentially no clear method for differentiating between these two possibilities, but one potential exception is the lower eudicot family Ranunculaceae. The genera of this family typically have petaloid sepals, but vary with regard to the presence of sterile, nectiferous petals in the second whorl (Tamura, 1993). This variability with regard to the presence or absence of the petals, combined with other features, led botanists to propose that the petals in this group were derived from stamens many times independently (reviewed in Rasmussen et al., 2009). In seeking to test this hypothesis, Kramer et al. (2003) identified homologs of AP3 and PI from across the Ranunculaceae and discovered that the gene lineages had experienced many independent gene duplication events, including two relatively ancient AP3 duplications that gave rise to three paralogous lineages, termed AP3-I, AP3-II and AP3-III (Kramer et al., 2003). The last of these, the AP3-III lineage, is of particular interest because it appeared to be expressed only in species bearing petals, with the one exception in the apetalous Anemone. This dataset was further expanded to demonstrate that AP3-III orthologs are commonly petal specific in their expression across members of the Ranunculaceae and even in several representatives of its sister family Berberidaceae (Rasmussen et al., 2009). Rather than supporting the botanical theories, these findings suggest that many petals across the Ranunculaceae and Berberidaceae express a commonly inherited identity program involving AP3-III orthologs. In order to invoke independent petal derivations, one would have to assume that this same AP3 paralog was repeatedly recruited to a petal-specific function, in spite of the fact that there are three AP3 paralogs present and no particular reason why any of them should become petal-specific.

Although this argument is reasonably compelling, it lacks functional data to establish whether AP3-III paralogs are, in fact, critical to petal identity. The new genetic model genus Aquilegia (columbine), a member of the Ranunculaceae, is well positioned to investigate this question. The flowers of Aquilegia have five types of floral organs: an outermost whorl of petaloid sepals, a second whorl of spurred petals, multiple whorls of fertile stamens, a single inner whorl of sterile staminodia and the innermost whorl of carpels (Fig. 1a–d; Kramer et al., 2007; Kramer, 2009). Detailed studies of the three known Aquilegia AP3 paralogs revealed evidence for complex expression patterns (Kramer et al., 2007), which are suggestive of temporal and spatial subfunctionalization (Force et al., 1999). AqAP3-1 is initially expressed across the floral meristem, but becomes preferentially expressed in staminodia. At early stages, AqAP3-2 is only detected in stamens and staminodia, but then turns off in staminodia and comes on in petals. By contrast, AqAP3-3 is only detected in petals from before their inception through late stages of spur development. AqPI expression encompasses all of these domains and, in addition, AqAP3-1/2 and AqPI are all detected in late sepals. Silencing of the AqPI locus using the RNAi technique virus-induced gene silencing (VIGS) confirmed that the petals, stamens and staminodia are under the control of the B gene homologs (Kramer et al., 2007). This conclusion can be broadly applied to all four loci because, similar to the case in Arabidopsis (Jack et al., 1994; Riechmann et al., 1996a), AqPI appears to function as a heterodimer with the three AP3 paralogs (Kramer et al., 2007).

Figure 1.

Phenotypes of wild-type and TRV2-AqAP33-AqANS-silenced flowers and floral organs. (a) Wild-type flower with one sepal and two petals removed. (b) Wild-type sepal. (c) Wild-type petal. (d) Wild-type stamen. (e) Strongly silenced flower. (f) Flower in (e) with sepals removed. Note lanceolate apices (compare with petals in a, c, i). Central organ has a narrow base (arrowhead), but adjacent organ has a slight bulge (arrow), indicating partial spur development. (g) Flower in (e) with sepals and petals removed to display unaffected stamens. (h) Flower with chimeric silencing, several sepals removed. Note the lack of spur development and narrow base of the second whorl organ (arrowhead). (i) Same flower as in (h) rotated to show weakly transformed second whorl organs. One additional sepal removed relative to photograph in (h). (j) Strongly transformed second whorl organ. (k) Moderately transformed second whorl organ with slight spur formation (arrow). (l) Scanning electron micrograph of the central laminar region of the adaxial epidermis of TRV2-AqANS-silenced sepal. (m) Scanning electron micrograph of a comparable region of the adaxial epidermis of TRV2-AqANS-silenced petal limb. (n) Scanning electron micrograph of the central laminar region of the adaxial epidermis of TRV2-AqAP33-AqANS-silenced sepal. (o) Scanning electron micrograph of the same region of the adaxial epidermis of TRV2-AqAP33-AqANS-silenced second whorl organ. Note the strong development of papillated cells. Bars: (a–g) 5 mm; (h–k) 20 μm.

In the current study, we have applied VIGS to the petal-specific AqAP3-3 in order to determine whether the gene is essential to petal identity. This silencing also appears to have affected a recently identified paralog of AqAP3-3, a locus termed AqAP3-3b. We have further investigated the Ranunculid AP3 and PI homologs, including new expression studies and analyses of the molecular evolutionary patterns among the AP3 paralogs. The results of these studies provide further support for the hypothesis that petal identity is controlled by AP3-III and is probably conserved across the Ranunculaceae and possibly Berberidaceae. The larger implications of this hypothesis are discussed relative to previous studies in Papaver and our broader understanding of the conservation of petal identity.

Materials and Methods

Virus-induced gene silencing

The Aquilegia VIGS protocol and construction of the TRV2-AqANS positive control plasmid have been described previously (Gould & Kramer, 2007; Kramer et al., 2007). To make the TRV2-AqAP33-AqANS construct, we PCR amplified a 230-bp fragment of AqAP3-3 using primers that added EcoRI and Xba sites to the respective 5′ and 3′ ends of the PCR product (5′ AGAATTCCTGGATGAGTCTGTGAAACTTGTTCGG, 5′ AATCTAGATGTTGGCTTGGTTGCACACGATAAGA). This PCR product was used to produce the TRV2-AqAP33-AqANS construct in a manner similar to that described by Kramer et al. (2007). Seventy-five Aquilegia coerulea E. James ‘Origami’ plants at the four to six true leaf stage were vernalized at 4°C for 3 wk; 1 d after the plants had been removed from vernalization, they were treated as described for seedlings in Gould & Kramer (2007). Twenty-five control plants were treated with TRV1 and TRV2-AqANS. Flowers showing any AqANS silencing were photo-documented and, on maturation, the flowers were dissected. All individual perianth organs were photographed using a Kontron Elektronik (Poway, CA, USA) ProgRes 3012 digital camera mounted on a Leica (Buffalo Grove, IL, USA) WILD M10 dissecting microscope (Harvard Imaging Center, Harvard University, Cambridge, MA, USA). For every flower showing silencing, a selection of organs from each whorl were either frozen at −80°C for subsequent RNA analysis or fixed in freshly prepared, ice-cold formalin–acetic acid–alcohol for scanning electron microscopy (SEM) analysis. This process was also repeated for several unsilenced flowers from the TRV2-AqAP33-AqANS cohort, as well as flowers that were treated with TRV2-AqANS as controls. SEM analysis and light microscopy was performed as described in Kramer et al. (2007).

Wild-type expression of AqAP3-3b

For Fig. 2(c,d), total RNA and cDNA were prepared from wild-type floral organs as described in Kramer et al. (2003) and Rasmussen et al. (2009). Petal sample EP corresponds to stages P1–P2 from Kramer et al. (2007), and LP corresponds to stages P3–P5. Reverse transcription-polymerase chain reaction (RT-PCR) using locus-specific primers for AqAP3-3, AqAP3-3b and ACTIN was conducted as described in Rasmussen et al. (2009) (see Supporting Information Table S1 for primer sequences). Ten nanograms of plasmid DNA containing the AqAP3-3b cDNA were used as a positive control template to confirm that the PCRs were viable. Quantitative real-time PCR (qRT-PCR) was further used to study relative gene expression levels between AqAP3-3 and the newly annotated AqAP3-3b. qRT-PCRs were carried out using PerfeCTa® SYBR® Green FastMix®, Low ROX™ for qPCR (Quanta BioSciences, Inc., Gaithersburg, MD, USA) in the Stratagene (Santa Clara, CA, USA) Mx3005P QPCR System. A list of primers is included in Table S1. Gene expression was calculated relative to AqIPP2 (isopentyl pyrophosphate:dimethylallyl pyrophosphate isomerase) using the ΔΔCT method (Livak & Schmittgen, 2001). Three biological replicates were performed. The fold expression level is reported relative to AqAP3-3 expression in inflorescences or petals, as noted in the text.

Figure 2.

Sequence, structure and expression patterns of new Aquilegia AP3 paralog AqAP3-3b. (a) Alignment of the AqAP3-3 and AqAP3-3b predicted protein sequences. (b) Genomic structure of the two loci. Black boxes indicate coding region; white boxes indicate untranslated region. (c) Reverse transcription-polymerase chain reaction (RT-PCR) analysis of AqAP3-3 and AqAP3-3b using 25 and 30 cycles. Early and late petal stages were analyzed (Pet1 and Pet2, respectively). cDNA from: Sep, sepals; EP, early petals; LP, late petals; Sta, stamens; Std, staminodia; Car, carpels; Inf, inflorescence; +, plasmid control. (d) Quantitative real-time PCR analysis of AqAP3-3 (black bars) and AqAP3-3b (gray bars) in early inflorescences and dissected petals (Pet) and stamens (Sta). In the former, the fold expression level is relative to AqAP3-3 expression in inflorescences, whereas, in the latter, it is relative to AqAP3-3 expression in petals.

Expression analysis of VIGS-treated organs

Total RNA and cDNA were prepared as described above from floral organs of treated but unsilenced flowers (referred to as ‘un’), as well as floral organs from strongly silenced TRV2-AqAP33-AqANS flowers. qRT-PCR was carried out as described above with the primers listed in Table S1. Several different sample types were examined. We determined AqAP3-3 expression in 23 individual TRV2-AqAP33-AqANS-silenced second whorl organs, which were selected to represent the observed range of phenotypes. Each sample was analyzed in three technical replicates. These values were averaged across all of the silenced organs and compared with a pooled RNA sample from treated but unsilenced second whorl organs (Fig. 3a). We also analyzed these data considering only eight strongly transformed second whorl organs (Fig. 3b). For Fig. 3(c), RNA from individual moderate (e.g. Fig. 1k) or strongly (e.g. Fig. 1j) silenced second whorl organs was analyzed in three technical replicates for both AqAP3-3 and AqAP3-3b expression. These expression levels were compared with those for treated but unsilenced organs. In order to determine expression levels in the stamens (Fig. 3d), we extracted RNA from a pool of stamens from flowers that appeared to be strongly silenced (e.g. Fig. 1g). This sample was analyzed in three technical replicates and compared with stamens from flowers that were treated but unsilenced. For the RT-PCR in Fig. 3(e), pools of first (w1) and second (w2) organs were collected from flowers that were treated with TRV2-AqAP33-AqANS but unsilenced (un), strongly silenced by TRV2-AqANS or strongly silenced by TRV2-AqAP33-AqANS. Three independent w2 samples were examined for TRV2-AqAP33-AqANS. Total RNA and cDNA preparation and locus-specific RT-PCR were performed as described in Kramer et al. (2007). Primers are listed in Table S1.

Figure 3.

Locus-specific quantitative real-time polymerase chain reaction (qRT-PCR) (a–d) and reverse transcription-polymerase chain reaction (RT-PCR) (e) on RNA prepared from organs of virus-induced gene silencing (VIGS)-treated flowers. (a) Average fold change in AqAP3-3 expression in pooled unsilenced (un) and 23 individual AqAP3-3-silenced (sil) second whorl organs. For the silenced organs, the bar represents the average fold change and the hash marks represent the individual values. (b) Average fold change in AqAP3-3 expression among eight strongly transformed second whorl (w2) organs. Error bars ± SD among the eight organs. (c) Fold change in AqAP3-3 (dark gray bars) and AqAP3-3b (light gray bars) expression in individual unsilenced (un), moderately silenced (mod) and strongly AqAP3-silenced (str) second whorl (w2) organs. All values are relative to AqAP3-3 expression levels in unsilenced second whorl organs. Errors bars ± SD between technical replicates. (d) Fold change in AqAP3-3 (dark gray bars) and AqAP3-3b (light gray bars) expression as measured by qRT-PCR on pooled stamen RNA from unsilenced (un) and AqAP3-3-silenced (sil) flowers. All values are relative to AqAP3-3 expression levels in unsilenced stamens. Errors bars ± SD between technical replicates. (e) RT-PCR results for all Aquilegia B gene homologs, AqSEP1, AqANS and ACTIN. RNA samples from three different types of flowers were tested: unsilenced flowers that were treated with TRV2-AqAP33-AqANS (UN), strongly silenced TRV2-AqANS flowers, and strongly silenced TRV2-AqAP33-AqANS flowers (TRV2-AqAP3-3). RNA was prepared from pooled first whorl (w1) and second whorl (w2) organs. Multiple pools of second whorl organs were tested for TRV2-AqAP33-AqANS. Thirty cycles were used to visualize AqAP3-3b expression, but all other loci were amplified for 25 cycles.

Plant materials for additional AP3 and PI characterization

A broad developmental range of floral tissue was obtained from the following taxa: Euptelea pleiosperma Hook. f. et Thomson (Eupteleaceae, C. Liu SX2006001), Circaeaster agrestis Maximowicz (Circaeasteraceae, C. Guo SX2007011), Kingdonia uniflora B. Balfour & W. W. Smith (Circaeasteraceae, C. Guo SX2007023), Cocculus trilobus DC (Menispermaceae, AA 108-2001), Nigella sativa L. (Ranunculaceae, B. Sharma AA2010001), Actaea asiatica H. Hara (Ranunculaceae, C. Guo SX2007043), Hepatica henryi Oliver (Ranunculaceae, C. Guo SX2007041) and Adonis vernalis L. (Ranunculaceae, E. Kramer AA 127-2004).

Identification of AP3 and PI homologs

New AP3 and PI homologs were identified as described in Kramer et al. (2003) and Rasmussen et al. (2009), including AqAP3-3b, which was annotated from the A. coerulea‘Origami’ genome sequence. Sequences are deposited in GenBank under accession numbers HQ647367HQ647380 and HQ694788HQ694802.

Phylogenetic and maximum likelihood tests of selection

The 17 new AP3 homologs identified in this study were added to a full-length nucleotide dataset previously assembled (Rasmussen et al., 2009). This dataset contains a total of 81 Ranunculales homologs, nine eudicot homologs, nine core eudicot homologs (both euAP3 and TM6 lineages) and four magnoliid homologs (see Table S2 for all accession information and Notes S1 for the final alignment). The magnoliid dicot sequences were used to root the phylogeny and all non-Ranunculales sequences are represented by the label ‘Outgroup’ in Fig. 4. The alignment was initially compiled using CLUSTALW and subsequently refined manually using MacClade 4.06 (Notes S1; Maddison & Maddison, 2000). Maximum likelihood phylogenetic analyses were performed using RaxML (Stamatakis et al., 2005) as implemented by the CIPRES portal (Miller et al., 2009). We used Modeltest (Posada & Crandall, 1998) to determine the simplest and most appropriate evolutionary model for our dataset employing the Akaike information criterion (AIC). The model selected was a general time-reversible model (GTR) with a proportion of invariable sites (I) and a gamma approximation for the rate of variation among sites (Γ). Branch support was estimated by performing 200 replicates of fast bootstrapping (Stamatakis et al., 2008) using the same parameters as in the original analysis. A similar analysis was conducted on a nucleotide alignment of all available Ranunculales PI homologs using the Euptelea homolog EupPI as the designated outgroup (Fig. S1 and Notes S2). This homolog was chosen because Euptelea appears to be sister to all other Ranunculales (Soltis et al., 2003).

Figure 4.

Maximum likelihood (ML) phylogeny of the AP3 lineage of Ranunculales. ML bootstrap values (above 50%) are placed at the nodes. Stars indicate inferred gene duplication events. Brackets on the right denote family membership and gene lineage. B, Berberidaceae; C, Circaeasteraceae; E, Eupteleaceae; L, Lardizabalaceae; M, Menispermaceae; P, Papaveraceae; R, Ranunculaceae. Inset at the bottom shows established phylogenetic relationships of families based on Soltis et al. (2003). Outgroup sequences are described in the Materials and Methods section.

We used a tree-based maximum likelihood approach, as implemented in PAML (Yang, 1997), to test for changes in selection constraints in the AP3 paralogous lineages. We compared a one-ratio model, which assumes the same ratio (ω) for all branches, with two-ratio models that assume a different ratio for a designated AP3 paralogous clade relative to the remaining sequences. This analysis was repeated for each of the three paralogous lineages, AP3-I, AP3-II and AP3-III, with an emphasis on the Ranunculaceae + Berberidaceae (Ran + Ber) sequences. The test was conducted for the entire genes and also for each of the functional domains defined for AP3 genes (Riechmann et al., 1996a,b). These analyses on the M, IK and C domains were performed in order to evaluate whether there was a difference in selection pressure on one or more of the distinct regions.

Expression studies

Total RNA extraction, cDNA synthesis and locus-specific RT-PCR were performed as described in Kramer et al. (2003) and Rasmussen et al. (2009). Primers could not be easily designed to distinguish between HhPI-1 and HhPI-2, and so these loci were amplified together in a single reaction. Primer sequences are given in Table S1 (Stellari et al., 2004).


VIGS of AqAP3-3 and AqAP3-3b in Aquilegia

We used VIGS via the Tobacco rattle virus (TRV) system to knock down AqAP3-3 expression in A. coerulea. A 230-bp fragment from the variable C-terminal region was used to specifically target this locus relative to the other two known AP3 paralogs, AqAP3-1 and AqAP3-2. However, during the course of the experiment, we discovered a second AP3-III lineage member in the Aquilegia 8x genome sequence (Joint Genome Institute, 2010). The region used for the TRV2 construct has 84% identity between AqAP3-3 and this new locus, AqAP3-3b. Aquilegia ANTHOCYANIDIN SYNTHASE (AqANS) was simultaneously targeted from the same TRV2 construct in order to recognize plants with strong silencing. We obtained 32 plants that showed appreciable amounts of AqANS silencing. The only observed perturbations of floral morphology affected the second whorl organs (Fig. 1). As expected with VIGS, we recovered a range of phenotypes, from reduced petals to complete petal-into-sepal transformations. The transformation of petals into sepals was evidenced by variable reduction in spur development and the presence of lanceolate apices that contrasted with the rounded apices of normal petals (compare Fig. 1a,c with Fig. 1f,h–k). In seven plants, the petal phenotype was limited to spur reduction, similar to that seen in Fig. 1(i). In 23 plants, comprising 35 flowers, we obtained moderate to strong silencing. Within the second whorl, phenotypes varied from petals with stunted spurs to full transformation into sepals (Fig. 1h–k). Two plants produced a total of five flowers that had very strong silencing (Fig. 1e,f). In these flowers, all of the second whorl organs had lanceolate apices typical of sepals, but the basal attachment points varied from narrow with a flat lamina (Fig. 1f,j) to broader with a bulged lamina, indicating some residual spur development (Fig. 1f,k). SEM analysis of the central laminar regions of the adaxial epidermal surface revealed that the transformed second whorl organs were characterized by papillated cell types that are normally present on the adaxial surface of Aquilegia sepals, but not observed in comparable regions of the petal limbs (Fig. 1l–o). In contrast with the strong second whorl phenotypes, the morphology of the sepals, stamens, staminodia and carpels was unaffected (Fig. 1e,g–i, data not shown).

Before we could properly characterize the gene expression patterns in the silenced flowers, we had to establish the wild-type distribution of AqAP3-3b. This locus shows only 83% amino acid identity with AqAP3-3 and has a 42-bp deletion in the C-terminal region (Fig. 2a,b). We conducted RT-PCR and qRT-PCR on RNA from dissected wild-type floral organs and found that expression is quite low relative to AqAP3-3 (Fig. 2c,d). Expression was similarly found to be very low in early inflorescences (Fig. 2c,d), probably explaining why this locus escaped previous detection.

We used qRT-PCR to assess the down-regulation of AqAP3-3 and AqAP3-3b in TRV2-AqAP33-AqANS-treated floral organs relative to organs from flowers that were similarly treated but unsilenced (un) (Fig. 3). These experiments demonstrated that AqAP3-3 silencing ranged from 1.25- to 50-fold, with an average of five-fold down-regulation across 23 organs selected to represent the range and frequency of phenotypes (Fig. 3a). When eight samples from strongly transformed organs (e.g. Fig. 1i) were considered, down-regulation of AqAP3-3 was found to be > 10-fold (Fig. 3b). We also compared silencing of AqAP3-3 and AqAP3-3b in individual second whorl organs exhibiting moderate (e.g. Fig. 1k) and strong (e.g. Fig. 1j) transformation. The decreased expression of AqAP3-3b in strongly transformed organs (Fig. 3c) suggests that both AqAP3-3 and AqAP3-3b may have been targeted by the TRV2 construct. Likewise, both AqAP3-3 and AqAP3-3b were very strongly silenced in stamens from flowers that showed strong transformation in the second whorl (Fig. 3d). This is an important point because weak AqAP3-3/3b expression can be detected in wild-type stamens at late developmental stages (Fig. 2c,d), and we wished to rule out any potential function in the stamens. It should be noted that expression in Fig. 3(d) is normalized relative to wild-type levels of AqAP3-3 expression in stamens, but this level itself is c. 30% of that in wild-type petals (Fig. 2d). Next, we used RT-PCR to examine the expression of the other AP3 homologs and AqPI in the TRV2-AqAP33-AqANS-treated plants relative to plants that were similarly treated but unsilenced (UN) or treated with TRV2-AqANS. The results confirmed the down-regulation of AqAP3-3 only in TRV2-AqAP33-AqANS-treated plants and showed no similar down-regulation of the other B gene homologs (Fig. 3e). We also examined the expression of AqSEP1, which has previously been found to be specifically expressed in Aquilegia sepals at late stages. In TRV2-AqPI-AqANS-treated plants, AqSEP1 expression expanded into the transformed second whorl (Kramer et al., 2007), but, in the current AqAP3-3 experiment, it was only weakly detected in two of the samples of transformed second whorl organs (Fig. 3d).

Phylogenetic analyses of Ranunculid AP3 and PI lineages

To complement the functional studies and in an effort to expand our sampling of B gene homologs across the Ranunculales, we characterized new AP3 and PI loci from the families Eupteleaceae, Circaeasteraceae, Menispermaceae and Ranunculaceae. This analysis yielded 17 new AP3 homologs and nine new PI homologs from eight different genera, and has now sampled every family of the Ranunculales (Table S2). Phylogenetic analyses were conducted on AP3 and PI nucleotide alignments using the maximum likelihood method as implemented by RAxML (Stamatakis et al., 2008). The AP3 analysis recovered three deeply conserved, paralogous lineages, termed AP3-I, AP3-II and AP3-III (Fig. 4). Each of these clades contains representatives from most, if not all, families of the Ranunculales with the exception of the Euptelea homologs, which are placed as sister to all three lineages, although with weak statistical support. The three paralogous clades themselves have moderate support, but their relationships to each other have no support. Within each lineage, loci from the same family are all grouped together with strong support and the relationships of these family-specific clades are roughly consistent with the established phylogenetic relationships of the sampled taxa (Soltis et al., 2003), with some poorly supported exceptions. This suggests that there were two duplication events predating the diversification of most of the order, followed by a few much more recent duplications. These include one each in the AP3-III lineage of Helleborus and Aquilegia, and one in the AP3-II lineage predating the split of Aquilegia and Thalictrum, although this second AP3-II gene appears to have been lost from the Aquilegia genome (Joint Genome Institute, 2010). As noted above, all three AP3 lineages were not recovered in all taxa. In particular, the AP3-II lineage has not yet been recovered from any representative of the Papaveraceae or Circaeasteraceae, and the AP3-III lineage is yet to be found in the Lardizabalaceae. In contrast with the pattern recovered for AP3, the PI homologs all cluster by taxonomic families, indicating the absence of ancient duplication events (Fig. S1). There is, however, strong evidence for multiple recent duplication events, at least 11 in the Ranunculaceae alone.

Expression studies of AP3 and PI homologs

We have also expanded the AP3 and PI homolog expression dataset in the Ranunculales. From the Ranunculaceae, we examined expression patterns in Helleborus hydrida and Nigella sativa, which are of particular interest because of their highly modified petals (Fig. 5). As observed in other members of the family, the AP3-III ortholog is strongly expressed in petals and is not detected at significant levels in the other floral organs (Fig. 5d,h). In Helleborus, the other B homologs show narrow expression patterns, with strongest expression being in the stamens (Fig. 5d). In contrast, in Nigella, the other loci are expressed very broadly (Fig. 5h), even being detected in the carpels, which is probably a result of the common expression of B gene homologs in ovules (reviewed in Kramer & Irish, 2000). By contrast, the Menispermaceae member Cocculus exhibits expression of the AP3-I ortholog CctAP3-1 in petals and stamens, low levels of the AP3-II ortholog CctAP3-2 in stamens, and broad expression of the AP3-III ortholog CctAP3-3 (Fig. 5k). These findings are consistent with the previous study (Rasmussen et al., 2009) in detecting petal-specific AP3-III expression in Ranunculaceae, but not in Menispermaceae.

Figure 5.

Flowers and floral organs with corresponding expression studies. (a–d) Helleborus hybrida (Ranunculaceae). (a) Anthesis stage flower. Arrow indicates petal. (b) Petaloid sepal. (c) Petal (left) and stamen (right) from pre-anthesis stage flower. (d) Reverse transcription-polymerase chain reaction (RT-PCR) expression results. Closely related PISTILLATA (PI) paralogs were not distinguished (HhPI-1/2). (e–h) Nigella sativa (Ranunculaceae). (e) Anthesis stage flower. Arrow indicates petal. (f) Petaloid sepal. (g) Petal (left) and stamen (right) from pre-anthesis stage flower. (h) RT-PCR expression results. (i–k) Cocculus trilobus (Menispermaceae). Photographs courtesy of Dr Yannick Staedler. (i) Inflorescence. (j) Male anthesis stage flower. Five inner strap-like petals are indicated by asterisks. Sixth petal is hidden by sepal on lower left side. (k) RT-PCR expression results. Expression was only analyzed in male flowers. Positive control reactions were performed using cDNA. Car, carpel; Pet, petal; Sep, sepal; Sta, stamen. Bars: (a–c, e–g, i, j) 5 mm.

Patterns of molecular evolution across the AP3 paralogous lineages

Gene duplication events are often associated with shifts in gene function and patterns of molecular evolution across paralogs (Yang, 1994, 1998). In the Ranunculid AP3 homologs, we have observed relatively conserved petal-specific expression of the AP3-III lineage members from the Ranunculaceae and Berberidaceae, whereas the other AP3 homologs are much more variable (Fig. 5; Rasmussen et al., 2009). For this reason, we wanted to determine whether the levels of purifying selection differ between each of the Ranunculaceae + Berberidaceae (Ran + Ber) clades across the three paralogous AP3 lineages. We used one- and two-model tests as implemented by PAML (Yang, 2000) to compare the relative likelihoods that selection has been acting equivalently across all three lineages, or shows distinct patterns in each of the three Ran + Ber clades. First, we performed these tests on the entire length of the AP3 dataset (Table 1). Not surprisingly, all of the tests revealed strong purifying selection acting on these sequences. There were differences, however, in the degree of purifying selection. The Ran + Ber clades of AP3-I and AP3-II showed a relaxed degree of purifying selection, although with differing degrees of significance. By contrast, the AP3-III lineage showed much stronger purifying selection with a probability of < 0.001. In order to gain a better understanding of how these findings might differ across the different domains of the type II MADS box genes, we repeated the tests on three distinct sequence compartments: the MADS domain of 135 nucleotides (nt) (note, the first 45 nt of the MADS domain are missing because of the position of the PCR primer used to obtain the sequences); the I and K domains of 258 nt; and the C-terminal domain of 222 nt. As expected, the DNA-binding MADS domain showed the strongest purifying selection, followed by the I + K and then the C region (Table 1), but each Ran + Ber clade displayed a different pattern of selection. For the Ran + Ber AP3-I lineage, the I + K domain test revealed a highly significant relaxation of purifying selection, but no differences for M or C. The Ran + Ber AP3-II lineage showed the opposite pattern, with significant relaxation of purifying selection in the M and C domains. Finally, the Ran + Ber AP3-III lineage showed greater purifying selection in all three regions, but this was only significant for M and I + K.

Table 1.   Molecular evolution of the paralogous APETALA3 (AP3) lineages
ModelsComplete lociMADS domainIntervening and keratin-like domainsC-terminal domain
  1. ω, dN : dS ratio for single-model analysis; ωB, dN : dS ratio for background (all unlabeled) branches; ωF, dN : dS ratio for foreground (labeled) branches; dN, number of nonsynonymous changes; dS, number of synonymous changes; LRT, Likelihood Ratio Test.

  2. Significance: *, < 0.05; **, < 0.01; ***, < 0.001.

−loge likelihood24 550.54524 547.8244811.7064811.1619613.6729605.6509896.6909896.558
ω or ωB0.251720.24170.147660.141450.249840.221780.316570.32098
−loge likelihood24 550.54524 543.9214811.7064806.5389613.6729612.4489896.6909894.549
ω or ωB0.251720.236160.147660.128880.249840.239290.316570.29796
−loge likelihood24 550.54524 540.6294811.7064807.2649613.6729608.7549896.6909895.462
ω or ωB0.251720.269630.147660.163930.249840.270440.316570.32953


Botanical theories have long held that the petals of the Ranunculaceae have been derived from outer stamens many times independently within the family (Prantl, 1888; Worsdell, 1903; Tamura, 1963, 1965; Hiepko, 1965). Based on gene expression patterns, Rasmussen et al. (2009) instead hypothesized that many of these petals shared a commonly inherited petal identity program. This work left several questions unanswered, however, including the critical issue of whether AP3-III orthologs are actually essential to petal identity. The current study has demonstrated that this is the case. Furthermore, expanded sampling has extended our comparative expression dataset to nine members of the Ranunculaceae and two members of the Berberidaceae in which the AP3-III orthologs are primarily expressed in petals. Consistent with this conserved expression pattern, representatives of the AP3-III clade show increased purifying selection, particularly in the functionally critical M and I + K domains. These results provide insights into both the genetic basis of petal identity in the emerging model system Aquilegia and into the evolution of petal identity in the Ranunculaceae.

Petal identity in Aquilegia

Previous studies of Aquilegia have found that each AP3 paralog has a distinct expression domain, which is petal-specific for AqAP3-3 (Kramer et al., 2007). The other paralogs are expressed in the petal, but only for relatively narrow time frames – either very early (AqAP3-1) or at later stages (AqAP3-2). Further expression analyses in a homeotic mutant with petal-to-sepal transformation (clematiflora or clm) detected no AqAP3-3 expression, and found that AqAP3-1 and AqAP3-2 expression in the second whorl was similar to that normally observed in sepals, which means that they are only expressed in late developmental stages (Kramer et al., 2007). Obvious lesions in the clm AqAP3-3 locus have not been identified; therefore, we applied locus-specific RNAi to knock down the expression of AqAP3-3. This yielded clear evidence that AqAP3-3, with possible contribution from the recently discovered AqAP3-3b, is essential to petal identity, but plays no critical role in sepal, stamen, staminodium or carpel identity. Although AqAP3-3b is expressed at much lower levels, the fact that the strongest phenotypes appear to be associated with the down-regulation of both paralogs may indicate some degree of residual function for AqAP3-3b. At present, we cannot determine exactly how AqAP3-1 and AqAP3-2 may contribute to petal identity or elaboration, but we can say that their detected expression is not sufficient to rescue petal identity. Unfortunately, the nature of VIGS means that AqAP3-3-silenced organs or flowers cannot be identified until relatively late in development. Therefore, we cannot characterize AqAP3-1 and AqAP3-2 expression throughout the development of the transformed second whorl organs. Based on the findings of the clm study, it seems reasonable to predict that AqAP3-1 and AqAP3-2 are expressed in the transformed second whorl in a pattern more similar to sepals, but even their normal petal expression is not constitutive (Kramer et al., 2007). Thus, there appear to be two general scenarios for the division of petal identity function among Aquilegia AP3 paralogs. First, the requirement for AqAP3-3 in petal identity could reflect a biochemical differentiation, such that AqAP3-1 and AqAP3-2 cannot substitute for the other paralog. Alternatively, it may be that their only differentiation is regulatory and AqAP3-1 and AqAP3-2 cannot rescue petal identity because they are not expressed at the correct time or concentration. Given that heterologous expression of AP3 homologs in Arabidopsis has yielded inconsistent results (e.g. Lamb & Irish, 2003; Whipple et al., 2004), the resolution of this question will await the ongoing development of transgenics in Aquilegia.

Implications for petal evolution in the Ranunculales

The important questions regarding the AP3-III lineage are: when did this petal-specific function evolve and what does that tell us about the evolution of petals in the Ranunculaceae? Every aspect of our data – the common petal-specific expression of AP3-III orthologs, the absence of their expression from apetalous taxa and the requirement for AqAP3-3 function to establish petal identity – argues for a commonly inherited petal identity program across much of the Ranunculaceae and potentially Berberidaceae as well. We also see more stringent purifying selection across petal-specific AP3-III orthologs, which could be consistent with a more constrained developmental function. By contrast, AP3-I and AP3-II orthologs exhibit relaxed purifying selection that is consistent with their highly variable expression patterns (Rasmussen et al., 2009). This evidence for process homology has profound implications for long-held botanical hypotheses regarding the evolution of petals in both families. Explicit hypotheses concerning the repeated derivation of petals from stamens in the Ranunculaceae are engrained into all morphological, developmental and phylogenetic treatments of the family (Tamura, 1965; Kosuge & Tamura, 1989; Hoot, 1991; Kosuge, 1994; Erbar et al., 1998). Despite their undeniable diversity, the second whorl sterile structures of Ranunculaceae, which we will simply term ‘petals’, share certain features, most notably the presence of nectaries (Tamura, 1993). Other traits that have been used to associate them with stamens, such as their shared phyllotaxy and the similar appearances of the primordia, are quite common for petals across the angiosperms (Takhtajan, 1991), and therefore we do not see why they should distinguish these petals for special theories of derivation. Thus, in the light of our new findings, we believe that the long-standing interpretations of the morphological and developmental data must largely be set aside. Instead, our hypothesis is that the ancestor of Ranunculaceae had petals whose identity was controlled by an AP3-III-dependent program. This program appears to have been lost in many independent cases during the diversification of the family.

It is important to note, however, that reconstruction of the presence of petals across the Ranunculaceae (Rasmussen et al., 2009) indicates that there is still at least one case in which petals evolved from an apetalous ancestor – the two unrelated Clematis sections Atragene and Naraveliocarpa. Members of these sections have petals with distinctly different morphologies, and they appear to be separately nested within the otherwise apetalous genus Clematis (Miikeda et al., 2006). Previous studies of one Atragene member indicated that it does express an AP3-III ortholog, whereas an apetalous species from the Integrifoliae does not; however, Clematis is a large genus with over 300 species, and so further study of this phenotype is required to determine whether these petals depend on an AP3-III program. If they do, it would raise interesting questions as to how this program could have been turned off and then back on again within the genus. In addition, characterization of AP3-III homolog expression in more genera may provide insight into other potential cases of independent petal evolution in the family.

The case of Clematis highlights the larger mechanistic question of how petal identity has apparently been lost so many times across the Ranunculaceae and Berberidaceae. Simple homeotic transformation of the petals into stamens via an outward shift of AG expression would achieve this end, but we know that, in most apetalous taxa, AP3-III expression has been lost altogether (Kramer et al., 2003; Rasmussen et al., 2009). One possibility is that there is direct negative regulation of AP3-III by the stamen identity program, but the genetic interactions between AqAP3-3 and AqAG remain to be tested. Another question is: what has happened to the genomic AP3-III loci in apetalous taxa that no longer express the lineage? One might predict a neutral decay of the genes once petal identity is lost, but this will require investigation of the genomic regions from apetalous taxa. At the very least, the absence of AP3-III expression in most apetalous taxa examined to date (seven of eight) argues against the possibility that the lineage was in some way predisposed towards being independently recruited to control petal identity. Such genetic parallelism certainly has precedent, most notably with the repeated recruitment of CYCLOIDEA-like TCP genes to control floral zygomorphy in the core eudicots (reviewed in Preston & Hileman, 2009). Under such a parallelism model AP3-III orthologs would be expected to be expressed in apetalous taxa, perhaps in the outer stamens. This does not appear to be the case, however, a fact that is in some ways the strongest evidence arguing for a commonly inherited petal identity program.

Having largely rejected the hypothesis of many independent derivations of Ranunculaceae petals, it is still reasonable to ask why they are so diverse and even bizarre in some cases (e.g. Fig. 5g). We suggest that the solution lies in the first whorl, where the sepals are commonly showy and petaloid. This transference of function (Baum & Donoghue, 2002) of the pollinator attraction role to the outer perianth members may have released morphological constraints on the second whorl petals, thereby allowing them to diversify in form or to be lost more readily. Petaloidy of the sepals is common throughout the Ranunculid order and accompanied by frequent variation in the presence and morphology of the petals (Worsdell, 1903; Tamura, 1993). It remains unknown how petaloidy of the sepals is promoted at the genetic level, but studies in Aquilegia have demonstrated that the B gene homologs are not essential to their identity or the production of their papillated epidermal cells, although the loci may still contribute to color production (Kramer et al., 2007).

It is also interesting to consider when the AP3-III petal identity program may have evolved. As the two examined members of the family Menispermaceae do not show petal-specific expression, we currently have to conclude that the pattern evolved in the last common ancestor of Ranunculaceae + Berberidaceae. However, it will be very interesting to examine expression of the AP3-III ortholog in Kingdonia (Circaeasteraceae), which has nectiferous sterile organs in the second whorl (Wu & Kubitzki, 1993), and further study is necessary of the Lardizabalaceae, where AP3-III representatives have recently been isolated (H. Kong, unpublished). In addition, we can ask whether the evolution of petal-specific expression in AP3-III represents a simple subfunctionalization or was perhaps associated with a truly novel derivation of petal identity. Previous work performed in the Papaveraceae may shed some light on this question, as the poppy AP3-III ortholog has been functionally investigated (Drea et al., 2007). This locus, PapAP3-1 (represented by PnAP3-1 in Fig. 4), is expressed in both petals and stamens, whereas the paralogous AP3-II lineage member, PapAP3-2, is only detected at early stages in stamens, with later expression extending to petals (Kramer & Irish, 1999; Drea et al., 2007). VIGS showed that PapAP3-1 is essential to petal identity with some redundant contribution to stamen identity, and PapAP3-2 plays a major role in stamen identity with little or no contribution to that of petals (Drea et al., 2007). The fact that PapAP3-1 primarily functions in petal identity raises the possibility that this role is deeply conserved in AP3-III across the order. This would suggest that the petal-specific AP3-III expression seen in Ranunculaceae and Berberidaceae represents the ultimate product of a progressive subfunctionalization process acting on an ancestral petal identity function over tens of millions of years. Of course, it remains possible that these AP3-III-associated petal identity programs evolved independently; therefore, further functional data in diverse Ranunculales would be ideal to test the prediction that AP3-III orthologs should be involved in petal identity even in taxa in which their expression domains are broad. As these conclusions take advantage of the fortuitous presence of the Ranunculid AP3-III lineage, the inferential power of this dataset reaches its limit at the base of the order, outside of which we can make no specific conclusions regarding the conservation of petal identity. Regardless, the growing evidence against the independent derivation of petals in the Ranunculaceae should lead us to re-evaluate such long-held typologies in the light of molecular genetic and modern phylogenetic data.


E.M.K. would like to thank Sarah Mathews for extensive assistance with the molecular evolutionary analyses. H.Z.K. would like to thank Yi Ren for help in collecting plant materials. We also thank Sarah Mathews, members of the Kramer and Kong laboratories and three anonymous reviewers for comments on the manuscript. This work was supported by National Science Foundation award IOS-0720240 to E.M.K. and Chinese Academy of Sciences grant KSCX2-YW-R-135 to H.Z.K. SEM analysis was conducted at Harvard’s Center for Nanoscale Systems supported by NSF Infrastructure Grant 0099916.