Sub- and neo-functionalization of APETALA3 paralogs have contributed to the evolution of novel floral organ identity in Aquilegia (columbine, Ranunculaceae)


Author for correspondence:

Elena M. Kramer

Tel: +1 617 496 3460



  • Previous studies of the lower eudicot model Aquilegia have revealed differential expression patterns of two APETALA3 (AP3) paralogs that appear to coincide with the development of a distinct fifth floral organ type, the staminodium. The AqAP3-1 locus quickly becomes limited to the staminodia while AqAP3-2 becomes stamen-specific.
  • We used transient RNAi-based methods to silence each of these loci individually and in combination, followed by detailed studies of the resultant morphologies and the effects on gene expression patterns.
  • Silencing of AqAP3-1 had a strong effect on the staminodia, causing transformation into carpeloid organs, while silencing of AqAP3-2 only affected the stamens, resulting in sterility, stunting or weak transformation towards carpel identity. Much more dramatic phenotypes were obtained in the doubly silenced flowers, where all stamens and staminodia were transformed into carpels. Quantitative reverse-transcription polymerase chain reaction analyses of B gene homolog expression in these flowers are consistent with complex patterns of regulatory feedback among the loci.
  • These findings suggest that the presence of ancient AP3 paralogs in the Ranunculaceae has facilitated the recent evolution of a novel organ identity program in Aquilegia. Specifically, it appears that downregulation of AqAP3-2 in the innermost whorl of stamens was a critical step in the evolution of elaborated sterile organs in this position.


The genetic basis of organ identity in standard floral organs – the sepals, petals, stamens, and carpels – has been well studied in multiple angiosperm models (reviewed in Causier et al., 2010; Litt & Kramer, 2010; Rijpkema et al., 2010). This program is summarized in the well-known ABC model, which holds that over-lapping gene activities create combinatorial codes for each of the four organ identities: A alone encoding sepals; A + B, petals; B + C, stamens; and C alone, carpels (Coen & Meyerowitz, 1991). Although the conservation, and perhaps even existence, of A function has not been supported, it generally appears that the B and C class functions are broadly conserved, albeit with interesting exceptions (reviewed in Causier et al., 2010; Litt & Kramer, 2010; Rijpkema et al., 2010). Beyond the conservation of standard organ identity is the question of how new floral organ identities arise. The best understood example of this phenomenon is probably the lodicules of grasses, which represent a modification of the pre-existing petal identity program (Whipple et al., 2007 and references therein). Another striking case is the de novo derivation of petaloid organs from outer stamens in the Caryophyllales (Brockington et al., 2012). Both of these instances, however, still result in a total of four organ types, so an outstanding issue is how floral organ identity is established in taxa that produce more than four types of floral organs. Most commonly, these fifth organ types are thought to be evolutionarily derived from pre-existing stamens, therefore termed staminodia, and range enormously in morphology and function (Walker-Larsen & Harder, 2000). Currently, the only functionally tractable model for the genetic study of such novel organs is the lower eudicot Aquilegia (columbine), a member of the Ranunculaceae with numerous available genetic and genomic tools (Kramer, 2009). These flowers produce five petaloid sepals in the first whorl, five spurred petals in the second whorl, four to seven whorls of ten stamens, a single whorl of ten sterile staminodia, and four to six unfused carpels (Fig. 1a–f). Compared with the stamens, the staminodia lack anthers and are laterally expanded with ruffled margins (Fig. 1e). Although their ecological function remains unknown, it is clear that the staminodia represent a completely distinct, fifth organ type (Tucker & Hodges, 2005; Kramer, 2009).

In a first effort to characterize the genetic basis of staminodium identity, homologs of the B class MADS box transcription factors APETALA3 (AP3) and PISTILLATA (PI) were characterized in Aquilegia (Kramer et al., 2007). The first discovery of note was that there are, in fact, four copies of the AP3 lineage, termed AqAP3-1, AqAP3-2, AqAP3-3 and AqAP3-3b, although expression of AqAP3-3b is so low that we will not consider it further (Kramer et al., 2007; Sharma et al., 2011). AqAP3-1 and AqPI are broadly expressed in the presumptive petals, stamens and staminodia before these primordia even initiate. At the same time, AqAP3-3 is detected only in the petal whorl but AqAP3-2 expression in the stamens and staminodia does not become apparent until somewhat later, after the primordia begin to form. These patterns remain constant until roughly the point at which carpel primordia initiate. This marks a clear transition when AqAP3-2 begins to decline in the staminodia but remains on in the stamens, while AqAP3-1 expression declines in the petals and stamens but not the staminodia, leading to a largely mutually exclusive expression pattern that persists through anthesis. AqAP3-2 expression also increases in the petals during later stages after the stamens have clearly differentiated locules. Throughout this period, AqAP3-3 remains constant in the petals while AqPI is expressed in petals, stamens and staminodia. The broad overlap of AqPI expression relative to the more restricted domains of the AP3 paralogs is consistent with the general finding that AP3 and PI homologs act as obligate heterodimers (reviewed in Immink et al., 2010) and, more specifically, that the Aquilegia B gene homologs similarly form heterodimers (Kramer et al., 2007).

These complex expression patterns suggest that both subfunctionalization and neofunctionalization (sensu Force et al., 1999) have occurred among the Aquilegia AP3 paralogs. The ancestral expression domain of AP3 homologs in petals and stamens appears to have been subfunctionalized such that AqAP3-3 is petal-specific and AqAP3-2 is largely stamen-specific, while AqAP3-1 may have experienced neofunctionalization in assuming a role in the novel staminodium identity program. In this regard, it is important to note that studies in other ranunculids show that orthologs of AqAP3-1 and AqAP3-2 are typically broadly expressed in petals and stamens, indicating that their differential expression in Aquilegia must be recently derived. The first test of this hypothesis was RNAi-based downregulation of AqPI using the virus-induced gene silencing (VIGS) (Kramer et al., 2007) technique. This confirmed that petal, stamen and staminodium identity is completely dependent on AqPI, which supports the conclusion that the protein is required for each AP3-containing dimer (Kramer et al., 2007). Silencing of AqPI also resulted in downregulation of AqAP3-2 and AqAP3-3 but not AqAP3-1 (Kramer et al., 2007), which we speculated could indicate a dependency of AqAP3-2 and AqAP3-3 expression on functional B protein heterodimers (Kramer et al., 2007). Such autoregulation and cross-regulation is common among B gene homologs across the core eudicots (reviewed in Rijpkema et al., 2010). Further silencing of AqAP3-3 confirmed that this locus is required only for petal identity and does not play a significant role in any other organ identity program (Sharma et al., 2011).

The current study builds on this previous work by examining the phenotypes of flowers silenced for AqAP3-1 or AqAP3-2 alone as well as simultaneous silencing of AqAP3-1/2. Our results suggest that AqAP3-1 is, in fact, essential to staminodium identity while silencing of AqAP3-2 has no effect on these organs. By contrast, AqAP3-2 is required for normal stamen development but AqAP3-1 also contributes to the establishment of stamen identity; complete stamen to carpel transformation is only observed when both loci are silenced. Although these data seem straightforward enough, quantitative reverse-transcription polymerase chain reaction (qRT-PCR) analysis of the expression of the other B gene homologs in each case reveals evidence for complex regulatory feedback interactions. Overall, our current picture of floral organ identity in Aquilegia reveals an intricate evolutionary history by which ancient gene duplications laid the foundation for a stepwise series of subfunctionalization and neofunctionalization that facilitated the evolution of a completely new floral organ identity.

Materials and Methods

Virus-induced gene silencing

The Aquilegia VIGS protocol and construction of the TRV2-AqANS positive control plasmid has been described previously (Gould & Kramer, 2007; Kramer et al., 2007). To make TRV2-AqAP31-AqANS construct, we PCR amplified a 290 bp fragment of AqAP3-1 using primers that added EcoRI and Xba sites to the 5′ and 3′ end of the PCR products (see the Supporting Information, Table S1). Similarly for TRV2–AqAP32–AqANS, a 202 bp fragment was amplified with primers that added BamHI and KpnI sites (Table S1). The TRV2–AqAP31–AqAP32AqANS construct was made using the same regions as in the individual constructs. The regions used to prepare the VIGS constructs share c. 51% similarity (Fig. S1) and there are no contiguous 21-nt stretches of identity (the longest stretch of identity is 11 nt). For each treatment, including the control TRV2–AqANS, 100 Aquilegia coerulea ‘Origami’ plants at the four to six true leaf stage were vernalized at 4°C for 3 wk and then were treated as described in Gould & Kramer (2007). Flowers showing any AqANS silencing were photo-documented and, upon maturation, the flowers were dissected. All individual organs were photographed using a Kontron Elektronik ProgRes 3012 digital camera mounted on a Leica WILD M10 dissecting microscope (Harvard Imaging Center; Cambridge, MA, USA). For every flower showing silencing, a selection of organs from each whorl was either frozen at −80°C for subsequent RNA analysis or fixed in freshly prepared, ice cold Formalin/Alcohol/Acetic Acid (FAA) for scanning electron microscopy (SEM) analysis. Both SEM analysis and light microscopy were performed as described in Kramer et al. (2007).

Expression analysis of wild type and VIGS-treated organs

RNA was extracted from wild type (untreated), control treated (AqANS-silenced) and experimental treated organs using the RNeasy Mini Kit (Qiagen, Valencia, CA, USA). For the analysis of wild-type expression (Fig. 1), organs of the same type were pooled. For the silenced samples (Fig. 3), organs of the same type with the same phenotype from the same flower were pooled. For example, for the AqAP3-1 study (Fig. 3a), the S1–4 samples were derived from four different strongly affected flowers in which the staminodia were largely to entirely transformed into carpels. Total RNA was DNAsed to eliminate any genomic contamination using TURBO DNase (Ambion, Life Technologies, Green Island, NY, USA). cDNA was prepared as described in Kramer et al. (2007). A qRT-PCR was performed using PerfeCTa qPCR FastMix , Low ROX (Quanta Biosciences Inc., Gaithersburg, MD, USA) in the Stratagene Mx3005P QPCR system to study the relative expression of AqAP3-1, AqAP3-2, AqAP3-3, AqPI, AqAG1 and AqAG2 via the inline image method (Livak & Schmittgen, 2001). AqIPP2 (isopentyl pyrophosphate:dimethylallyl pyrophosphate isomerase) expression was used for value normalization. All the primers used are listed in Table S1. These primers were designed using the MacVector sequence analysis software package (Cary, NC, USA) and all primer pairs (synthesized by Integrated DNA Technologies, Coralville, IA, USA) were tested and optimized to confirm comparable amplification efficiencies. All primer pairs were further designed to span intron positions and qRT-PCR reactions were run out after amplification in order to further confirm that no genomic products were present. Multiple biological replicates (at least three but as many as five) were examined in most cases except when individual samples are shown, (Fig. 3a,d,g) in which case three technical replicates were examined (see text and figure legend). All error bars represent standard deviation and unpaired Student's t-tests were used to determine the statistical significance of differences between experimental and control values.


Expression profile of B and C gene homologs in wild type organs

Although in situ hybridization has previously been used to study the expression dynamics of Aquilegia's B gene homologs, VIGS requires us to limit our analysis to relatively late-stage floral organs. Thus, in order to set a baseline for subsequent analyses of gene expression, we determined the relative expression levels of AqAP3-1, AqAP3-2, AqAP3-3, AqPI, AqAG1 and AqAG2 across all floral organs from wild-type flowers using qRT-PCR (Figs 1g, S2a). We have previously shown that the fourth AP3 paralog, AqAP3-3b, is expressed at very low levels and was, therefore, excluded from the entire study. The four to seven stamen whorls were divided into inner and outer pools, where outer corresponds to the first two to three whorls and inner to the later two to four whorls. In general, the lower expression of all genes across the ‘outer’ stamens likely reflects the fact that these stamens mature earlier than the inner and are, therefore, developmentally more advanced. Consistent with the expression patterns observed at early stages (Kramer et al., 2007), AqAP3-1 is most highly expressed in staminodia, AqAP3-2 is most highly expressed in stamens and AqAP3-3 is most highly expressed in petals. AqPI, a cofactor for all three AP3 paralogs, is expressed in petals, stamens and staminodia. Weaker expression of the AP3 homologs is detected in the petaloid sepals,. Given that the staminodia are derived from a floral region normally associated with C function, we also analysed the two Aquilegia homologs of the C gene AGAMOUS (AG), AqAG1 and AqAG2. The loci are both highly expressed in carpels but AqAG1 is more strongly expressed in stamens and staminodia than AqAG2 (Fig. S2a).

Figure 1.

Wild-type morphology and late-stage gene expression. (a) Wild-type anthesis-stage flower of Aquilegia coerulea ‘Origami Red and White’. (b) Pre-anthesis stage flower with perianth removed. Multiple whorls of stamens surround a single whorl of white, ruffled staminodia and a single whorl of innermost carpels. (c) Control TRV2-AqANS silenced flower. (d) Single wild-type stamen. (e) Staminodial whorl showing late stage adhesion. (f) Carpel whorl. Bars: (a–c) 1 cm; (d–f) 5 mm. (g) Quantitative reverse-transcription polymerase chain reaction (qRT-PCR) expression patterns of all Aquilegia B gene homologs from dissected pre-anthesis stage sepals (SEP), petals (PET), outer stamens (OUT STA), inner stamens (IN STA), staminodia (STD), and carpels (CAR). Expression levels are normalized to carpel values. Error bars ± SD.

Characterization of AqAP3-1 silenced plants

We used VIGS via the Tobacco Rattle Virus (TRV) system to knock down AqAP3-1 expression in A. coerulea. In addition, all VIGS constructs contained a fragment of Aquilegia ANTHOCYANIDIN SYNTHASE (AqANS) to help identify plants experiencing silencing. AqANS-silencing alone causes flowers to be white or pale pink but they are completely morphologically normal (Fig. 1c). As expected with VIGS, which is a knock-down rather than a knock-out technique, we observed a range of phenotypes in the AqAP3-1-silenced flowers. In the strongest phenotype, all of the staminodia exhibited dramatic transformation into carpels (Fig. 2a–g) and, in some of these flowers, the innermost stamens were misshapen with carpeloid anthers or partial necrosis of the anthers (Figs 2c, S3a). We base our assessment of carpeloidy on the presence of multiple carpel-associated features that are not normally observed in any other floral organs, including photosynthetic tissue, development of dense trichomes and the presence of style-like extensions at the apex of the organ (see Figs 1f and 2d for wild-type carpels). By contrast, staminodia are strictly colorless, lack trichomes and have an undulating laminar surface (Figs 1e, 2e). The stamens are similarly lacking in trichomes or photosynthetic tissue and do not normally have protruding connectives (Fig. 1d). Furthermore, strong carpel transformation of the staminodia often resulted in the development of ovules (Fig. S3f). Less severe phenotypes exhibited transformation of only some of the staminodia as well as chimerism in which only the distal portions of the organs were affected (Fig. 2f,g). These less severe phenotypes were divided into moderate and weak classes in which the moderate comprised flowers with six to eight chimeric staminodia and the weak class had five or fewer chimeric organs. Out of the 100 plants treated, the strong phenotype was observed in 10 plants with 14 flowers while 18 plants with 25 flowers showed moderate to mild silencing. None of these flowers showed developmental perturbations outside the staminodia or innermost stamens.

Figure 2.

Morphology of VIGS-silenced floral organs. (a–g) TRV2–AqAP31AqANS-treated flowers. (a) Entire flower. (b) Flower with perianth removed, stigmas are indicated with arrows. (c) Flower with perianth and almost all stamens removed. One aberrant inner stamen is shown (asterisk). (d) SEM of wild type carpel. (e) SEM of wild type staminodium. (f) SEM of chimeric staminodium/carpel. (g) Close-up of chimeric organ in (f), focusing on junction between staminodial and carpeloid tissue. (h–m) TRV2–AqAP32AqANS-treated flowers. Note that some flowers do not have dual silencing of AqANS (a,l). (h) Entire flower. (i) Closer view of flower with several sepals and petals removed: arrows indicate necrotic, sterile stamens; arrowheads indicate stamens with elongated connectives. (j) Another flower with entire perianth removed: arrows indicate necrotic, sterile stamens; arrowheads indicate stamens with elongated connectives; asterisks indicate outer stamens that appear to be partially transformed into petal-like structures. (k) Carpeloid-stamen with flattened anther, elongated connective and trichomes on the filament (arrows). (l) Severe flower with several sepals and petals removed. (m) Same flower as in (l) with entire perianth removed. All stamens are reduced to naked filaments. (n–s) TRV2–AqAP31AqAP32AqANS treated flowers. (n) Entire flower. (o) Flower with several sepals and petals removed. (p) Flower with entire perianth removed. (q) Transformed carpel. (r) Scanning electron microscopy (SEM) of transformed carpel. (s) Close-up of open ovary in (r). Bars: (a,d–f,h,i,l,n,o,r) 1 cm; (b,c,j,m,p) 5 mm; (g,s) 200 μm; (k,q) 2.5 mm.

In order to assess the degrees and specificity of silencing, qRT-PCR was performed on seven biological replicates of AqAP3-1-silenced staminodia, representing a range of severe, moderate and weak phenotypes as well as control AqANS-silenced organs (Fig. 3a). Consistent with the morphological severity, AqAP3-1 downregulation was strongest in the severe class (> 90%), followed by the moderate (80–90%) and low (50–60%) classes. Next, we examined AqAP3-2 and AqPI expression in the transformed staminodia. Interestingly, AqAP3-2 expression levels were also lower, again tracking with the severity of the phenotype, but AqPI expression was unaffected (Fig. 3b). In order to determine whether this was a general pattern, we analysed expression of all three AP3 homologs across the other floral organs – sepals, petals, stamens and carpels – from strongly silenced flowers, with AqAP3-3 only examined in petals (Fig. 3c). Consistent with the results in staminodia, AqAP3-1 was strongly downregulated in all organs (90–50%) accompanied by similar downregulation of AqAP3-2 (70–50%; differences are significant for all organs (< 0.005). The expression of AqAP3-3 in silenced petals was only reduced by c. 10% compared with control petals but this was significant with < 0.01. Finally, we determined the expression of AqAG1 and AqAG2 in the same samples as shown in Fig. 3(b) and found that expression levels of both loci in the former staminodia were increased to levels similar to that of carpels, consistent with the degree of transformation in the organs (Fig. S2b). The levels of both AqAG1 and AqAG2 differ significantly from their wild-type values in staminodia (< 0.0001) but most values are not distinguishable from those of wild-type carpels. The exceptions are AqAG1 in moderately silenced organs, which is slightly but significantly higher than wild-type carpel values, and the intermediate expression of AqAG2 in weakly silenced staminodia, which is significantly different from both wild-type staminodia and carpels.

Figure 3.

Quantitative reverse-transcription polymerase chain reaction (qRT-PCR) assessment of silenced tissue with standard deviations. (a–c) TRV2–AqAP31AqANS-treated organs. (a) AqAP3-1 expression levels in control (C) staminodia (STD), stamens (STA) and carpels (CAR) compared with staminodia with strong (S), moderate (M) and weak (W) phenotypes. Multiple samples were tested in replicate for each phenotype. All values are normalized to the control staminodia values. (b) AqAP3-2 and AqPI expression levels in control (C) staminodia (STD), stamens (STA) and carpels (CAR) compared with staminodia with strong (S), moderate (M) and weak (W) phenotypes. Multiple biological replicates of each phenotype class were averaged together. All values are normalized to the control staminodia values. (c) AqAP3-1, AqAP3-2 and AqAP3-3 expression levels in control (C) sepals (SEP), petals (PET), stamens (STA) and carpels (CAR) compared with organs dissected from flowers with strong staminodium phenotypes (SIL). Multiple biological replicates were averaged and normalized to the control values for each organ type. (d–f) TRV2-AqAP32-AqANS treated organs. (d) AqAP3-2 expression levels in control (C) stamens (STA) compared with stamens with strong (S) phenotypes. Multiple samples (S1-7) were tested in replicate. All values are normalized to the control stamen values. (e) AqAP3-1 and AqPI expression levels in control (C) stamens (STA) compared with silenced stamens (SIL STA). Multiple biological replicates of the silenced stamens were averaged together. All values are normalized to the control stamen values. (f) AqAP3-1, AqAP3-2 and AqAP3-3 expression levels in control (C) sepals (SEP), petals (PET), staminodia (STD), and carpels (CAR) compared with organs dissected from flowers with strong stamen phenotypes (SIL). Multiple biological replicates were averaged and normalized to the control values for each organ type. (g–i) TRV2-AqAP31-AqAP32-AqANS treated organs. (g) AqAP3-1 and AqAP3-2 expression levels in control (C) staminodia (STD), stamens (STA) and carpels (CAR) compared with pooled transformed stamens and staminodia. Multiple samples (S1–7) were tested in replicate. All values are normalized to the control staminodium values. (h) AqPI expression levels in control (C) staminodia (STD), stamens (STA) and carpels (CAR) compared with silenced stamens and staminodia (SIL ST/SD). Multiple biological replicates of the silenced stamens/staminodia were averaged together. All values are normalized to the control staminodium values. (i) AqAP3-1, AqAP3-2 and AqAP3-3 expression levels in control (C) sepals (SEP), petals (PET) and carpels (CAR) compared with organs dissected from flowers with strong stamen/staminodia phenotypes (SIL). Multiple biological replicates were averaged and normalized to the control values for each organ type. All vertical axes represent relative (rel.) expression levels, all asterisks indicate expression levels that are significantly different from the respective control organs and all error bars ± SD.

Characterization of AqAP3-2 silenced plants

AqAP3-2 was targeted for silencing using a 202 bp fragment of the C-terminal domain. The AqAP3-2-silenced plants exhibited a wide range of phenotypes that strictly affected the stamen whorls (Figs 3h,i,l, S3c,d). The most common phenotype, which we would term mild to moderate, involved various disruptions of normal stamen development, including widespread anther reduction and necrosis, the presence of trichomes on the filaments, development of green, extended connectives and occasional transformation of the outermost stamens into petal-like organs (Figs 2h–k, S3b,c, g–j). The most severe phenotype resulted in complete elimination of the anthers and extreme stunting of filament development (Fig. 2l,m). It should be noted in regard to Fig. 2(h,l) that in some cases (c. 10–20% of flowers), dual silencing of AqAP3-2 and AqANS does not occur. This phenomenon has been observed in other studies using multiple markers (Kramer et al., 2007; Blein et al., 2008) and appears to be a product of the somewhat stochastic nature of VIGS. Obviously, flowers that appear to be morphologically aberrant but are not silenced for AqANS are always confirmed using qRT-PCR on the main target locus, in this case AqAP3-2 (see later). Some of the moderate stamen phenotypes, particularly the presence of trichomes on the filaments and development of green, elongated connectives, could be associated with a partial transformation to carpel identity but no dramatic transformations were observed. Out of the 100 plants treated, the severe phenotype was observed in seven plants with 10 flowers while 16 plants with 25 flowers showed moderate to mild silencing (only four of these flowers showed transformation of outer stamens towards petal identity).

The qRT-PCR technique was used to assess AqAP3-2 silencing in seven samples of affected stamens, three of which displayed complete anther loss (severe) while four had necrotic or partially transformed stamens (moderate). All of the samples showed > 80% silencing of AqAP3-2, with strongest silencing in the severe phenotype samples (Fig. 3d). When AqAP3-1 and AqPI expression was examined in these samples, we discovered that AqAP3-1 is actually upregulated c. 3-fold (significant at < 0.0001) while AqPI is unaffected (Fig. 3e). This apparent upregulation of AqAP3-1 was not significant in the sepals, petals, staminodia and carpels, but a similar trend appeared to be present (Fig. 3f). Analysis of the other floral organs did confirm broad downregulation of AqAP3-2 (all significant at < 0.005) as well as reduction of AqAP3-3 expression in petals (c. 60%, significant at < 0.005). Both AqAG1 and AqAG2 were upregulated in the AqAP3-2-silenced stamens, possibly consistent with mild transformation toward carpel identity (Fig. S2c; both significant at < 0.005).

Characterization of AqAP3-1 and AqAP3-2 double-silenced plants

We created a TRV2 construct containing both AqAP3-1 and AqAP3-2 VIGS fragments in order to determine the degree of functional redundancy between the two loci. Unlike the first two analyses, the double silencing yielded highly consistent, uniformly severe phenotypes: all stamens and staminodia were strongly transformed into carpels (Figs 2n–s, S3e). These organs displayed all the hallmarks of carpels, including differentiation into styles and ovaries with ovules. However, they often displayed incomplete closure along with some distorted morphology (Fig. 2q–s). The sepals and innermost carpels were unaffected but petal development appeared stunted, with nectar spurs that reached only 50% of normal length (Fig. 2n,o). Out of the 100 plants treated, the phenotype was observed in 15 plants with 26 flowers.

For expression analyses, stamens and staminodia that were transformed to carpels were pooled together because no phenotypic difference was observed among the organs. Across seven biological replicates of severely transformed carpels, strong downregulation (75–95%) was detected for both AqAP3-1 and AqAP3-2 (Fig. 3g). Interestingly, AqPI expression was unaffected compared with wild type stamens or staminodia, rather than decreasing to the low AqPI expression levels normally observed in carpels (Fig. 3h). Downregulation of AqAP3-1 and AqAP3-2 was also observed in the sepals, petals and innermost carpels (Fig. 3i). In the petals, AqAP3-3 expression was reduced more strongly than had been observed in the single-locus silencing experiments (c. 85%, significantly different at < 0.0001). Expression of AqAG1 and AqAG2 was much greater in the transformed organs than in normal stamens and staminodia, consistent with their new identity as carpels (Fig. S2d; both values differ significantly from wild type stamens or staminodia at < 0.0001 but do not differ significantly from wild type carpels).


Each of the constructs tested resulted in highly distinct phenotypes. Targeting of AqAP3-1 primarily affected staminodia, which were transformed towards carpel identity, with some perturbation of inner stamen identity as well. Silencing of AqAP3-2 strictly affected stamens, causing sterility because of late anther necrosis, severe reduction of anther development, or partial transformation toward carpel identity. By contrast, dual silencing produced a very dramatic phenotype in which all of the stamens and staminodia were strongly transformed towards carpel identity. There was no effect on petal development in either of the singly silenced cohorts but double silencing resulted in stunted petal spurs that failed to completely elongate. Analyses of gene expression patterns within the context of these phenotypes yielded complex results. Our targeted silencing of AqAP3-1 also produced downregulation of AqAP3-2, but had only weak effects on AqAP3-3 and none on AqPI. By contrast, targeted silencing of AqAP3-2 resulted in upregulation of AqAP3-1 in staminodia and moderate downregulation of AqAP3-3 in petals, but no effect on AqPI. Similarly, dual silencing of AqAP3-1/2 resulted in strong reduction in AqAP3-3 expression but not in AqPI expression.

The primary issue in interpreting these results is determining whether the reduced expression of AqAP3-2 in the TRV2-AqAP3-1-treated plants is most likely caused by off-target silencing or, alternatively, by genetic interactions among the paralogs. We favor the latter explanation for a number of reasons. First, both the decline in AqAP3-2 expression in TRV2–AqAP3-1-treated plants and that of AqAP3-3 in TRV2–AqAP3-2-treated plants are consistent with the previous findings in AqPI-silenced plants, where AqAP3-2 and AqAP3-3 expression decreased (Kramer et al., 2007). In this case, off-target silencing between AqPI and the distantly related AP3 paralogs is highly unlikely but autoregulation and cross-regulation is common among B gene homologs (reviewed in Goto & Meyerowitz, 1994; Liu & Mara, 2010). Second, if the issue was off-target silencing owing to siRNA amplification, one would expect that it would be a general response and affect both TRV2–AqAP3-1 and TRV2–AqAP3-2. This is not the case because AqAP3-1 expression appears to increase in the AqAP3-2-silenced plants. Third, if AqAP3-2 was being targeted by VIGS in TRV2–AqAP3-1-treated plants, we might expect the phenotypes to be equivalent to those of TRV2–AqAP3-1/2. Instead, these phenotypes differ dramatically with the former primarily affecting staminodia while the latter affects all of the stamens and staminodia. Alternatively, if the loss of AqAP3-2 was caused by regulatory interactions, the different phenotypes could be explained by differences in the relative timing of repression. Lastly, while it is theoretically possible that signal amplification could lead to siRNA production from regions outside the original target fragment, several studies have revealed high levels of fidelity for VIGS (reviewed in Burch-Smith et al., 2004), including when working with the closely related MADS box genes (Liu et al., 2004; Drea et al., 2007; Sharma et al., 2011). In light of all of these facts, we believe that the simplest explanation for the collective phenotypic and expression data is that the AP3 paralogs exhibit both positive and negative cross-regulation.

Based on this, we can expand our hypotheses of regulatory interactions among the Aquilegia B gene homologs (Fig. 4). It seems that maintained expression of AqAP3-2 is particularly dependent on full function of AqAP3-1. As AqAP3-1 expression in the stamens is normally weak at late stages, we would propose that either this low level of AqAP3-1 expression is required for persistent AqAP3-2 activation or the strong, early AqAP3-1 expression potentiates long-term expression of AqAP3-2. Given the distinct phenotypes of AqAP3-1-, AqAP3-2- and AqAP3-1/2-silenced flowers, we assume that AqAP3-1 is not essential to AqAP3-2 initiation but rather its maintenance. Conversely, AqAP3-2 appears to feed back negatively onto AqAP3-1, certainly in the stamens and possibly also in the other floral organs. It is important to note that even if there is a degree of off-target silencing in these experiments, the data still suggest negative regulation of AqAP3-1 by AqAP3-2. At the same time, late AqAP3-2 expression in petals is important for strong AqAP3-3 expression, likely with contributions from AqAP3-1 (double AqAP3-1/2 silencing affects AqAP3-3 expression more than silencing of AqAP3-2 alone). Alternatively, it may simply be that AqAP3-2 silencing was more effective in the doubly silenced plants, leading to stronger reduction in AqAP3-3. As AqPI would be expected to participate in all of these interactions through heterodimerization with the AP3 proteins, the loss of AqAP3-2 and AqAP3-3 in AqPI-silenced flowers (Kramer et al., 2007) is consistent with the model, although we cannot rule out additional autoregulatory feedback for AqAP3-2 and AqAP3-3. Interestingly, neither AqAP3-1 nor AqAP3-2 expression was affected when AqAP3-3 was silenced (Sharma et al., 2011) and, at least so far, no positive feedback onto AqPI has been observed in any of the AP3 silencing experiments (this study, Sharma et al., 2011).

Figure 4.

A modified ABC model for Aquilegia petals, stamens and staminodia based on the expression and apparent functional roles of each Aquilegia AP3 paralog as well as AqPI. Dashed orange lines indicate regulatory feedback interactions hypothesized based on changes of paralog expression in different silenced backgrounds. Note that AqPI is inferred to participate in all of the AP3 paralog interactions because of their demonstrated heterodimerization (Kramer et al., 2007). PET, petals; STA, stamens; STD, staminodia.

Beyond these potential regulatory interactions, it appears clear that the novel staminodium identity program is dependent on AqAP3-1 function but does not especially require AqAP3-2. Conversely, in stamens, AqAP3-1 expression seems to be sufficient to prevent the development of carpel identity but is not capable of promoting proper stamen development. It may even be that the effect in inner stamens in AqAP3-1-silenced flowers results more from the loss of AqAP3-2 than from AqAP3-1 itself. Thus AqAP3-1 and AqAP3-2 contribute differentially to staminodium and stamen identity, respectively, but what is the basis of the contrast? Is it simply because of their differential expression or are they biochemically differentiated as well? Obviously, B gene homologs typically control both petal and stamen identity but the separate programs result from the presence or absence of other factors, such as C gene homologs (Coen & Meyerowitz, 1991). In Aquilegia, the C gene homologs AqAG1 and AqAG2 appear to be expressed in both stamens and staminodia, although there may be reduced expression in the latter (Kramer et al., 2007), which bears further investigation. There are cases where duplications in the B lineage, particularly AP3, have led to subfunctionalization between petal and stamen identity, a situation best characterized in petunia. In that case, although there are distinct expression patterns and dimerization preferences for the two AP3 paralogs, they are both biochemically capable of promoting petal identity (Vandenbussche et al., 2004; Rijpkema et al., 2006). Conversely, paralogs of the C gene AGAMOUS (AG) in Antirrhinum are subfunctionalized both in terms of their expression and biochemical ability to promote stamen vs carpel identity (Causier et al., 2005; Airoldi et al., 2010). In Aquilegia AqAP3-2-silenced flowers, we see that even though AqAP3-1 is upregulated, it is still incapable of rescuing normal stamen development. It may be that the degree of upregulation is not enough to rescue the absence of AqAP3-2, but we must consider the possibility that the loci are not biochemically equivalent. This being said, it is also true that AqAP3-1 expression alone in AqAP3-2-silenced stamens does not appear to be sufficient to convert these organs to full staminodium identity, indicating that other factors specific to the staminodium whorl are required (or even higher AqAP3-1 expression). Protein interaction studies to date have found that both AqAP3-1 and AqAP3-2 interact with AqPI, although the AqAP3-1/AqPI dimerization is particularly strong (Kramer et al., 2007). Addressing the question of biochemical differentiation, as well as testing our hypotheses of regulatory interactions, will require the use of transgenic studies in Aquilegia. In addition, a thorough investigation of AqAG1 and AqAG2 may shed light on the full genetic identity program that distinguishes staminodia from stamens.

We now have a picture of organ identity in Aquilegia that reveals complex subfunctionalization and neofunctionalization, which have evolved along different time-scales. The petal-specific AqAP3-3 paralog is subfunctionalized relative to ancestral AP3 function and appears to be both necessary and sufficient for establishing petal identity (Rasmussen et al., 2009; Sharma et al., 2011). The lack of normal petal development in AqAP3-1/2-silenced plants may result from the reduction in AqAP3-3 expression, but it is also possible that the other AP3 paralogs have some important contribution to petal development. Regardless, all evidence suggests that orthologs of AqAP3-3 are similarly subfunctionalized to a specific role in petal identity across the Ranunculaceae, Berberidaceae and possibly even the Lardizabalaceae (Kramer et al., 2003; Rasmussen et al., 2009; Sharma et al., 2011; Hu et al., 2012), indicating that an ancient subfunctionalization has been broadly maintained. The next key question will be understanding how this expression pattern is controlled. Although a petal-specific element has been characterized in the Arabidopsis AP3 promoter (Hill et al., 1998), it remains to be seen whether the AqAP3-3 subfunctionalization relies on a similar element.

The other two AP3 paralogs have experienced both subfunctionalization and neofunctionalization: neither locus is essential to petal identity, both act to promote stamen identity, with AqAP3-2 being the crucial player, and only AqAP3-1 is essential to staminodium identity. Again, independent of any potential off-target silencing, the phenotypes resulting from the different TRV2 constructs are completely distinct and support this model. Furthermore, there is every reason to believe that the distinct respective roles of AqAP3-1 and AqAP3-2 in staminodium and stamen identity were critical factors in the evolution of this novel organ identity in Aquilegia. It is interesting to note that although the paralogous lineages defined by these two genes are quite ancient, predating the ancestor of the bulk of the Ranunculales (Rasmussen et al., 2009; Sharma et al., 2011), staminodia between the stamens and carpels are recently evolved in the tribe Thalictroideae (Tucker & Hodges, 2005). The only other genus to possess staminodia-like structures is Semiaquilegia, the sister genus to Aquilegia, which has sterile organs in the same position that are more irregular in number and morphology than the broad, elaborated staminodia of Aquilegia (Tucker & Hodges, 2005). In fact, the Semiaquilegia organs bear considerable resemblance to the sterile filaments obtained in strong AqAP3-2-silenced flowers. This suggests a model in which silencing of AqAP3-2 in the innermost whorl of stamens may have been the first step in the evolution of the novel organs. Essentially, the ancestor of Aquilegia Semiaquilegia could have possessed whorls of fertile stamens that expressed AqAP3-1 and AqAP3-2 together, but in this novel inner whorl, transient AqAP3-2 expression with persistent AqAP3-1 produced sterile organs that ultimately evolved into the full staminodium phenotype of Aquilegia. Even in the fertile stamen whorls of Aquilegia, it appears that the innermost stamens are more sensitive to AqAP3-1 silencing than the outer ones, suggesting a degree of differentiation within the morphologically homogeneous stamen whorls. One possibility is that this partitioning of AqAP3-1 and AqAP3-2 function – with one paralog more responsible for outer stamens while the other primarily controls inner stamens – may have been further exaggerated to yield the novel staminodial whorl. Examination of AqAP3-1 and AqAP3-2 ortholog expression in Semiaquilegia and other related genera will be critical to understanding this evolutionary process, as will exploring the regulatory mechanisms controlling expression of the two paralogs.

What is clear is that the existence of the ancient AP3-1 and AqAP3-2 paralogs, which are otherwise commonly expressed in both petals and stamens (Rasmussen et al., 2009; Sharma et al., 2011), has facilitated the evolution of a whole new organ identity program in the lineage leading to Aquilegia. This involved a series of genetic innovations, including the definition of a new regulatory domain within the previously homogeneous stamen region and the acquisition of negative feedback by AqAP3-2 onto AqAP3-1, which is unusual for B gene homologs. Further studies will be necessary to determine whether gene duplication similarly underlies identity in other taxa that produce more than four types of floral organs, but the common presence of paralogs among the floral MADS box genes (reviewed in Litt & Kramer, 2010; Rijpkema et al., 2010) means that raw material for such neofunctionalization is broadly available. For example, an independent set of AP3 duplicates has been implicated in the evolution of the novel petaloid organ types seen in orchids, but this hypothesis is yet to be functionally tested (Mondragon-Palomino & Theissen, 2011). Of course, the association between gene duplication and morphological or physiological novelty has been well documented in both plants and animals (for review see Flagel & Wendel, 2009; Hall & Kerney, 2012). In Aquilegia, the next goal is to elucidate the mechanistic basis of this process and determine just how many genetic changes were required downstream of the AP3 paralogs to achieve the final elaboration of the staminodia.


We thank Sarah Mathews, members of the Kramer lab, and three anonymous reviewers for comments on the manuscript. This work was supported by NSF award IOS-0720240 to E.M.K. The SEM analysis was conducted at Harvard's Center for Nanoscale Systems supported by NSF Infrastructure Grant 0099916.