Developmental and molecular characterization of novel staminodes in Aquilegia

Abstract

Background and Aims

The ranunculid model system Aquilegia is notable for the presence of a fifth type of floral organ, the staminode, which appears to be the result of sterilization and modification of the two innermost whorls of stamens. Previous studies have found that the genetic basis for the identity of this new organ is the result of sub- and neofunctionalization of floral organ identity gene paralogues; however, we do not know the extent of developmental and molecular divergence between stamens and staminodes.

Methods

We used histological techniques to describe the development of the Aquilegia coerulea ‘Origami’ staminode relative to the stamen filament. These results have been compared with four other Aquilegia species and the closely related genera Urophysa and Semiaquilegia. As a complement, RNA sequencing has been conducted at two developmental stages to investigate the molecular divergence of the stamen filaments and staminodes in A. coerulea ‘Origami’.

Key Results

Our developmental study has revealed novel features of staminode development, most notably a physical interaction along the lateral margin of adjacent organs that appears to mediate their adhesion. In addition, patterns of abaxial/adaxial differentiation are observed in staminodes but not stamen filaments, including asymmetric lignification of the adaxial epidermis in the staminodes. The comparative transcriptomics are consistent with the observed lignification of staminodes and indicate that stamen filaments are radialized due to overexpression of adaxial identity, while the staminodes are expanded due to the balanced presence of abaxial identity.

Conclusions

These findings suggest a model in which the novel staminode identity programme interacts with the abaxial/adaxial identity pathways to produce two whorls of laterally expanded organs that are highly differentiated along their abaxial/adaxial axis. While the ecological function of Aquilegia staminodes remains to be determined, these data are consistent with a role in protecting the early carpels from herbivory and/or pathogens.

INTRODUCTION

To paraphrase Goethe, all lateral organs are fundamentally leaves (Goethe, 1790), but their morphology varies depending on what identity programme is expressed in the developing organ primordia. How do new organ identity programmes arise? This process is of interest in the context of the flower, which normally has four identity types – sepals, petals, stamens and carpels – but can have more. There are a wide range of examples of novel floral organs across the angiosperms but, unfortunately, most of these are either relatively ancient (e.g. the origin of sepals and petals, or the origin of the lodicule) and/or are found in taxa that are not genetically tractable (e.g. staminodes in Loasaceae). Although stamens themselves are thought to have only evolved once (Endress, 1994), the large numbers of stamens present in many flowers create a kind of raw material for evolutionary modification and the expression of novel morphologies. Staminodes are stamens that have been modified such that they are sterile (Walker-Larsen and Harder, 2000). This can occur due to reduction of the male whorls in female flowers, as is seen in moneocious or dioecious plants, or in cases of stamen abortion in the context of zygomorphy (e.g. the dorsal stamen of Antirrhinum). Over time, such non-functional organs are likely to become vestigial and even completely lost. If, however, sterilized stamens evolve a new role, they may become highly elaborated and can perform various functions, frequently related to pollination (Walker-Larsen and Harder, 2000). These roles are diverse and can include pollinator attraction, provision of attractants/rewards, prevention of self-pollination or facilitation of pollen removal and receipt (Walker-Larsen and Harder, 2000; Endress and Matthews, 2006). Staminodes have evolved many times independently across the angiosperms, including in the lower eudicot genus Aquilegia (Walker-Larsen and Harder, 2000).

Aquilegia is a genus of approx. 70 species found in the eudicot family Ranunculaceae (Munz, 1946). These species are diverse in the geography and ecology of their habitats as well as their morphology (including flower orientation, petal spur length and colour), are widely interfertile, and are the product of a recent adaptive radiation (Hodges and Arnold, 1994). Their flowers have five distinct types of floral organ: sepals, petals, stamens, staminodes and carpels (Fig. 1A). The staminodes are arranged in two successive whorls of five organs that are immediately adjacent to the carpels. Similar sterile organs are also observed in the sister genera of Aquilegia, Semiaquilegia and Urophysa, placing their estimated origin in the last common ancestor of these three taxa, approx. 20–22 million years ago (Mya) (Bastida et al., 2010). Staminode primordia strongly resemble those of stamens and are arranged on the same orthostichies (Tucker and Hodges, 2005). At maturity, individual Aquilegia staminodes clearly diverge morphologically from stamens in that they lack anthers and each organ is composed of a laterally expanded, ruffled lamina (Fig. 1B; Kramer et al., 2007). It has been inferred that the lamina is specifically derived from the stamen filament, as suggested by the occasional appearance of weakly chimeric organs in which an otherwise wild-type staminode is terminated by a reduced, sterile anther (Fig. 1C). This expanded lamina may be critical to a novel developmental feature of the staminodes: they appear to undergo late post-genital fusion along their lateral margins to form a continuous sheath (Fig. 1B, E, F; Sharma et al., 2014). This covering remains surrounding the gynoecium after the outer floral organs abscise (Fig. 1D–F), but is finally lost as the fruits swell and mature (Fig. 1G). Interestingly, given that stamens tend to be highly numerous in Aquilegia and most of its close relatives (Drummond and Hutchinson, 1920; Tucker and Hodges, 2005; Ren et al. 2011; Zhao et al., 2016), these staminodes are likely to have evolved in a context in which pollen limitation was not a major factor. In contrast, carpel number is relatively stable at 4–5 per flower across both the clade bearing staminodes (Aquilegia + Semiaquilegia + Urophysa) and its sister clade (Isopyrum + Enemion) (Zhai et al., 2019), further suggesting that the staminodes are derived from stamens rather than carpels.

Fig. 1.

Aquilegia coerulea ‘Origami’ flowers and staminodes. (A) Aquilegia floral diagram. (B) Staminode sheath comprised of six adhered organs. (C) Flower showing two staminode/stamen chimeras (arrowheads). (D) Stage 11a flower. Top, undissected bud; centre, bud with three sepals, two petals and three orthostichies of stamens removed to reveal staminode whorl; bottom, magnified view of the staminode whorl. (E) Stage 11c flower. Top, undissected flower; centre, flower with three sepals, two petals and three orthostichies of stamens removed to reveal the staminode whorl; bottom, magnified view of the staminode whorl surrounding the carpels. (F) Stage 13b flower. Top, undissected flower; centre, flower with two sepals, three petals and three orthostichies of stamens removed to reveal the staminode whorl; bottom, magnified view of the staminode whorl surrounding the carpels. (G) Early stage 15 flower with split staminode sheath. sep = sepal; pet = petal; sta = stamen; std = staminode; car = carpel. Scale bars: (B, C) = 5 mm, (D) = 1 mm; (E–G) = 1 cm.

Previous work has focused on identifying the genetic basis of organ identity in Aquilegia staminodes. In the context of the canonical ABC model, it is difficult to account for the production of a fifth organ identity that represents an entire whorl (as opposed to the interaction of identity and zygomorphy that produces the dorsal staminode of Antirrhinum; Luo et al., 1999). Stamen identity, which is a highly conserved aspect of the ABC model, is conferred by the combinatorial function of the B genes APETALA3 (AP3) and PISTILLATA (PI) together with the C gene AGAMOUS (AG) (reviewed in Litt and Kramer, 2010). In Aquilegia, paralogues of APETALA3 exhibit evidence of both sub- and neofunctionalization, which is critical to staminode identity (Sharma and Kramer, 2013). Prior to the diversification of the Ranunculales, duplication events gave rise to three paralogous APETALA3 lineages: AP3-1, AP3-2 and AP3-3 (Kramer et al., 2003). It appears that members of the AP3-3 lineage underwent an early sub-functionalization event to function specifically in petal identity, which has been confirmed by broad comparative gene expression and functional studies (Sharma et al., 2011; Zhang et al., 2013). In contrast, the expression patterns of the AP3-1 and AP3-2 homologues are quite variable across the Ranunculaceae (Kramer et al., 2003) but, in the lineage leading to Aquilegia, the respective representatives AqAP3-1 and AqAP3-2 experienced sub- and neofunctionalization (Sharma et al., 2011; Sharma and Kramer, 2013). Although both paralogues are initially expressed across the stamen and staminode primordia, at the stage when carpels initiate, their expression becomes differentiated such that AqAP3-2 remains expressed in stamens while AqAP3-1 becomes concentrated in the staminodes. Virus-induced gene silencing of the loci separately and together has been used to tease apart the distinct functions of these paralogues (Sharma and Kramer, 2013). Silencing of AqAP3-1 results in the strong transformation of staminodes into carpels, but the stamens are only weakly affected, which indicates that AqAP3-1 is essential to staminode identity while AqAP3-2 expression alone is sufficient to promote normal stamen development. In contrast, AqAP3-2 silencing does not affect the staminodes but the stamens develop as sterilized filaments. It is only when both loci are silenced together that all the stamens and staminodes are transformed into carpels, as would be expected in a B class mutant. Thus, the fifth organ identity programme of the staminodes appears to be derived from modification of the pre-existing stamen identity programme and results from the differential expression of multiple B gene paralogues, yielding a unique developmental identity.

Having now determined the upstream genetic programme controlling staminode identity, many questions remain regarding the basis of their developmental elaboration and ecological function. In order to better understand the developmental programmes that are differentially expressed between the two organ types, we compared the development and mature morphology of staminodes and stamen filaments across members of the Aquilegia genus and between Aquilegia, Semiaquilegia and Urophysa. We further used comparative transcriptomics of Aquilegia × coerulea ‘Origami’ to characterize the molecular pathways downstream of organ identity that are responsible for the divergent morphologies of the staminode and stamen filament. Differentially expressed loci were confirmed using in situ hybridization, which revealed critical differences at both early and late developmental stages. Taken together, these findings suggest significant differences in the regulation of abaxial/adaxial identity as well as divergence in terminal cell type differentiation that may implicate a role in defence rather than in pollination biology for these staminodes.

MATERIALS AND METHODS

Materials

Seeds for Aquilegia × coerulea ‘Origami Red and White’ were obtained from Swallowtail Seeds (Santa Rosa, CA, USA), and germinated and grown under long-day conditions (16 h light at 18 °C, 8 h dark at 13 °C). Plants with 4–6 true leaves were then transferred to vernalization conditions (16 h light at 4 °C and 8 h dark at 4 °C) for 4 weeks to induce flowering. Aquilegia alpina, A. flabellata and A. vulgaris plants in flower were obtained from Russell’s Garden Center (Wayland, MA, USA), and A. canadensis plants in flower were obtained from Garden in the Woods (Framingham, MA, USA). Semiaquilegia adaoxoides buds and flowers were obtained from Dr Scott Hodges, University of Calfornia, Santa Barbara, CA, USA. Urophysa rockii seed was obtained from Drs Hongzhi Kong and Rui Zhang, Chinese Academy of Science, Beijing China, and grown under the same conditions as described above for Aquilegia. All species identifications were confirmed by keying out traits based on Munz (1946).

Histology

Developing inflorescences and mature flowers at various stages of development were collected from wild-type plants and fixed under vacuum for 30 min in freshly prepared, ice-cold FAA. Tissue was incubated on a shaker overnight for 16–18 h at 4 °C, dehydrated in an ethanol and Citrasolv series at 20 °C, embedded in Paraplast and stored at 4 °C (Kramer, 2005). Tissue was sectioned into 8 µm thick sections using a Jung Histocut microtome. At least five flowers per species were sectioned, and slides were stained with 0.5 mg mL^–1 toluidine blue O. All slides were imaged using a Zeiss Axio Scan.Z1 slide scanner (Harvard Imaging Center). The cell counter plugin of ImageJ was used to count individual cells from histological sections (Rueden et al., 2017). Image J was also used to measure the width of transverse sections of 20 stamens and 20 staminodes from five flowers at maturity. To visualize lignification in staminodes, sections were stained in 1 % Safranin O solution for 2 h, and then counterstained with Fast Green for 30 s. Sections were imaged using a Zeiss LSM700 Confocal Microscope at the Arnold Arboretum of Harvard University under white light and with a red fluorescence filter with 488 nm excitation and 590 nm long-pass barrier emission wavelengths. Phloroglucinol staining was performed on freshly collected staminodes and stamen filaments according to the method outlined in Ruzin (1999).

RNA sequencing

Quadruplet replicates of staminodes and stamen filaments at two stages of floral development (stages 11a/b and 13a as defined in Min and Kramer, 2017; Fig. 1D–F) were removed and immediately flash-frozen in liquid nitrogen. Each biological replicate contained tissue from multiple flowers, with each bioreplicate drawn from different plants. Total RNA was isolated using the Qiagen RNeasy Plant Mini kit (Qiagen, Valencia, CA, USA), including a repeated elution step. RNA quality was analysed using a NanoDrop 8000 spectrophotometer (Thermofisher) and an Agilent Bioanalyzer. RNA samples with a minimum 2.09 260/280, and an RNA integrity number of 7.5 were used as input using the Apollo Prep X PolyA8 mRNA 200 bp beta V1 protocol for single-end library generation. All libraries were quality confirmed for correct size distributions by TapeStation, quantified by QBIT and quantitative PCR using the ABI prism 7900 and Kappa quantification kit, and pooled before running on one Illumina HiSeq2500 lane. Raw read data were deposited to the Sequence Read Archive (SRA) with sample accessions and read counts provided in Supplementary data Table S1.

We aligned reads to the A. coerulea ‘Origami’ V3 genome (JGI) using TOPHAT v.2.0.13. We acquired read counts for 30 023 Aquilegia gene models using HTSeq (Anders et al., 2015), and used the BLASTx algorithm to identify their Arabidopsis thaliana orthologues. Differential expression between tissue samples was analysed in RStudio using EdgeR, and 4153 gene models were identified with log-fold changes values greater than ±1. Lists of genes with significant differences between sample types are included in Supplementary data Tables S2 and S3. Gene Ontology (GO) enrichment was analysed using Panther (Supplementary data Table S4), with a false discovery rate (FDR) cut-off of 0.05 (Mi et al., 2016).

YABBY gene tree construction

Amino acid sequences were obtained from Phytozome (JGI) and aligned using CLUSTALW (Larkin et al., 2007) as implemented in MacVector 15.1.5 (Gary, NC, USA) and then adjusted by hand to remove uninformative sequences. A maximum likelihood phylogenetic tree was constructed using RAxML (Randomized Axelerated Maximum Likelihood) v.8.0.0 (Stamatakis, 2014) with the default model of amino acid substitution as implemented on the CIPRES V.3.3 platform (Miller et al., 2011). The resultant tree was displayed by FigTree v1.4.2 (http://tree.bio.ed.ac.uk/software/figtree/).

In situ hybridization

Fixation, embedding and sectioning of floral inflorescences were conducted as described above. For probe template selection, mRNA sequences for AqFIL and AqCRC were obtained from Phytozome (JGI). DNA templates for RNA probe synthesis were produced by PCR amplification of 200–300 bp fragments from cDNA clones of each locus (AqFIL and AqCRC; see Supplementary data Table S5 for primers) and as outlined in Puzey et al. (2012) for AqHIS4. Fragments were cloned using the TOPO-TA plasmid vector (Invitrogen). Digoxigenin-labelled RNA probes were prepared from linearized template plasmids and alkaline hydrolysed to 150 bp. In situ hybridization was performed using previously described methods (Kramer, 2005). Sections were imaged on the Zeiss AxioImager microscope at the Arnold Arboretum of Harvard University. The numbers of AqHIS4-positive cells in staminodes and stamen filaments of stage 8 flower buds were counted manually using ImageJ. Sixteen staminodes and their adjacent stamen filaments from five different sections were counted, and Student’s t-test was used to compare the number of AqHIS4-positive cells in staminodes and stamens.

RESULTS

Two whorls of staminodes with marginal curling

The development of the pentamerous, actinomorphic flowers of Aquilegia has been divided into 16 stages that span floral meristem initiation to fruit maturation (Ballerini and Kramer, 2011; Min and Kramer, 2017). The floral organs are arranged in whorls that each contain five organs (Tucker and Hodges, 2005). A single whorl each of sepals and petals surround 8–14 whorls of stamens, two whorls of staminodes and one whorl of carpels (Fig. 1A). The stamens are arranged in antesepalous and antepetalous vertical ranks (orthostichies), yielding ten stamen orthostichies, which each have either a sepal or a petal at their base. Each stamen orthostichy has a staminode in the uppermost position, so the two whorls of staminodes can be likewise termed antesepalous or antepetalous. The ten staminode primordia initiate as individual organs during the transition to stage 7, but by the beginning of stage 13 they appear to be adhered to one another, forming a continuous ruffled sheath (Fig. 1B).

We performed toluidine blue O staining on transversely sectioned flowers from three developmental stages: stage 11a buds, stage 11b/c buds just before flower opening and stage 13b open flowers. This allowed us to visualize the lateral expansion of the staminodes relative to the narrow stamen filaments (Fig. 2). It also revealed a new developmental feature that differentiates staminodes from stamens: the lateral margins of the antesepalous staminodes curl typically inward towards the carpels, clasping the lateral margins of the antepetalous staminodes, which themselves generally curl in the opposite direction, thereby creating interlocking margins (Fig. 2D–F). This creates a ‘hand-holding’ effect, which is seen in all of the examined developmental stages of staminodes and is consistent from the base of the organs to just below their tips. The staminodes remain unfused at their apices at all developmental stages (e.g. Fig. 1B, D). Occasionally, individual staminodes break the pattern such that instead of curling in the same direction at both margins, they curl in different directions. This is observable in Fig. 3D in which several staminodes curl outward on one side but inward on the other (see outlined staminode primordia in the lower left of the panel). This was very rarely observed in A. coerulea ‘Origami’, but see also below regarding other species.

Fig. 2.

Aquilegia coerulea ‘Origami’ toluidine blue-stained histological sections from three stages of development. (A, D, G) Transverse sections of stage 11a bud. (B, E, H) Transverse sections of stage 11b bud. (C, F, I) Transverse sections of stage 13b flower. (A–C) Continuous staminodial whorls surround the carpels with stamen filaments visible outside the sheath. (D–F) The same images as (A–C) with original colour removed and the staminodes false coloured to highlight both their identity and their alternate marginal curling behaviour. Antesepalous staminodes are coloured pink, while antepetalous staminodes are coloured blue. (G–I) Magnified views of individual staminodes showing marginal junctions between adjacent staminodes and morphologically distinct stamen filaments from the same orthostichy (arrowheads). Scale bars: (A–F) = 500 µm; (G–I) = 200 µm.

Fig. 3.

Cell counting and AqHIS4 in situ hybridization in A. coerulea ‘Origami’. (A) Transverse section of a stage 13b flower stained by calcofluor white showing an adjacent stamen filament (sta) and staminode (std). (B) The same image from (A) processed without colour. Coloured dots represent cells counted using Image J. Light blue dots are abaxial stamen filament epidermal cells; dark blue, adaxial stamen filament epidermal cells; pink, abaxial staminode epidermal cells; and magenta, adaxial staminode epidermal cells. (C) Cell counts from 20 staminodes and adjacent stamens from six flowers. The four classes of cell counts were found to be statistically differentiated by one-way ANOVA, and Tukey HSD was used to determine how the classes differed from one another. This test identified three distinct classes: abaxial staminode cell counts (a), adaxial staminode cell counts (b) and the two stamen cell counts (c). The two staminode classes are significantly different at P < 0.05 while the staminode vs. the stamen classes were significantly different at P < 0.01. (D) Transverse section of stage 8 bud hybridized with the AqHIS4 probe. Cell divisions, as indicated by AqHIS4 expression, are more common in staminode primordia than in stamen primordia. Carpel primordia are outlined in dashed green lines; stamens in blue dashed lines; and selected staminodes in pink dashed lines. Note that the highlighted staminode primordia represent examples of atypical marginal curling, as discussed in the text. (E) Close-up of stamen primordium (blue outline) and staminode primordium (pink outline). (F) Counts of AqHIS4-positive cells in five transverse sections of staminode and stamen primordia from stage 8 flower buds. Staminode cell counts are higher than stamen counts as indicated by Student’s t-test at P < 0.001. Scale bars: (A, D) = 200 µm; (E) = 100 µm.

Staminodes, like stamen filaments, each have a single vascular bundle (Fig. 2G–I). In early stage 11a flower buds, sections of the base of the staminodes show that, like the stamens, they are filled with mesophyll cells (Supplementary data Fig. S1A). However, moving apically, much less mesophyll is present (Fig. 2A, D, G; Supplementary data Fig. S1B–D). This lack of mesophyll becomes even more obvious in stage 11b/c organs (Fig. 2B, E, H; Supplementary data Fig. S1F–H) and, by maturity, the staminodes consist of only two epidermal cell layers enclosing the single vascular bundle (Fig. 2C, F, I; Supplementary data Fig. S1J–L). It is unclear based on this histology whether the decrease in mesophyll is primarily due to active programmed cell death processes or a failure of proliferation/expansion in the mesophyll cells to keep pace with the rapidly expanding epidermal cells.

At maturity, the staminodes of A. coerulea ‘Origami’ are on average three times as broad as the narrow stamen filaments (Supplementary data Fig. S2). We counted the number of epidermal cells arrayed horizontally at a point roughly midway along the proximal/distal length of the staminodes and stamen filaments (Fig. 3A–C). These cell counts confirmed that there are 2–3 times more cells arrayed laterally in staminodes than in the stamen filaments, and also found that there are a significantly higher number of cells (approx. 25 % more) across the adaxial epidermal surface of the staminodes than the abaxial surface. In situ hybridization of AqHISTONE4 (AqHIS4) was used to visualize the relative number of dividing cells in early stage staminodes and stamen filaments (Fig. 3D–F). This revealed that cell division appears to persist for a longer time in staminodes than in stamen filaments (Fig. 3F).

Lignification

In our histological sections of mature A. coerulea ‘Origami’, we identified a band of bright blue/green staining in the adaxial periclinal walls of the adaxial epidermal layer of the staminodes (Fig. 4B), a pattern that was not observed in the stamens (Fig. 4A). This colour of toluidine blue O staining is typical of lignin (O’Brien et al., 1964) and is also observed in the vasculature (Fig. 4A, B). Phloroglucinol staining of fresh tissue confirmed lignification in the staminodes, while there was no staining of stamen filaments (Fig. 4C, D). This pattern is also clearly observable via epifluorescence microscopy (Fig. 4E–H). It appears that the lignification develops quite late as it is not present at stage 11b (Fig. 4E) but is clearly apparent by stage 13b (Fig. 4F). Interestingly, in the staminodes themselves, this staining appeared to be reduced in the cells at the lateral margins that participate in adhesion (Fig. 4G, H).

Fig. 4.

Lignin visualization in A. coerulea ‘Origami’. (A) Toluidine blue-stained transverse section of a stage 11b flower bud showing lignification of xylem cells in the staminode and stamen filament (arrowheads), but no evidence in the abaxial (ab) or adaxial (ad) epidermal layers. (B) Transverse section of a staminode from a stage 13 flower showing asymmetric lignification of the adaxial (ad) epidermis, but not the abaxial (ab) epidermis. Lignification is also visible in the xylem (arrowhead). (C) Phloroglucinol staining in fresh staminodes from a stage 13 flower. (D) Lack of phloroglucinol staining in a stamen filament from the same flower as in (A). (E–H) Toluidine blue-stained sections. Left column, light micrograph; centre column, fluorescent micrograph; and right column, merged image. (E) Transverse section of a stage 11b flower. (F) Transverse section of a stage 13b flower. (G) Magnified view of a single staminode from a stage 13b flower. (H) Magnified view of the junction of two adjacent staminodes from a stage 13b flower. The asymmetric lignification does not extend through the region of adherence between two adjacent staminodes (arrow). Scale bars: (A, B) = 1 mm; (C, D) = 10 µm, (D) = 100 µm; (E–E''), (G–G'') = 200 µm; (F–F'') = 500 µm; (H–H'') = 50 μm.

Comparative histology

To complement our study of the predominantly New World A. coerulea ‘Origami’, we collected multiple developmental stages of flowers from both Old World (A. alpina, A. flabellata and A. vulgaris) and New World (A. canadensis) species, as well as from the sister genera Semiaquilegia adoxoides and Urophysa rockii (Fig. 5). All of the staminodes from Aquilegia species as well as U. rockii displayed the marginal curling (Fig. 5A–E, G). Although the staminodes in S. adoxoides were identifiable due to their two cell layers and flattened morphology, they were not present in a continuous whorl, being dispersed within a whorl of stamens, and generally did not show marginal curling (Fig. 5F). Asymmetric adaxial lignification was observed in every Aquilegia species but was not seen in S. adoxoides or U. rockii, indicating that lignin enrichment is unique to Aquilegia. A significantly higher number of laterally arrayed adaxial epidermal cells were also found in each of the Aquilegia species (Supplementary data Fig. S3). We observed occasional gaps between adjacent stage 13b staminodes in A. alpina, A. vulgaris and A. canadensis flowers (Fig. 5A'', C'', D''), suggesting that tight lateral adhesion may not be uniform across the genus.

Fig. 5.

Histology of Aquilegia, Semiaquilegia and Urophysa staminodes. (A) A. canadensis. (B) A. coerulea ‘Origami’. (C) A. alpina. (D) A. vulgaris. (E) A. flabellata. (F). S. adoxoides. (G) U. rockii. (A–G) Mature flowers. (A'–G') Transverse sections of the androecium, staminodial whorl and gynoecium, positioned roughly midway between the receptacle and staminode apices of mature flowers. The arrow in F' indicates a solitary staminode. (A''–E'') Magnified view of single staminode showing lignification of the adaxial epidermis. (F'' and G'') Staminodes and stamen filaments (asterisks) of S. adoxoides or U. rockii. No lignification is seen in the staminodes. (A'''–E''') Magnified view of staminode junctions showing adhesion in some cases (B''', E''') but lack of adhesion in others (A''', C''', D'''). Scale bars: (A–E), (G) = 1 cm; (F) = 5 mm; (A'–G') = 500 µm; (A''–G'') = 50 µm; (A'''–E ''') = 200 µm.

Staminode vs. stamen filament transcriptomics

In order to investigate the molecular mechanisms underlying the development of the staminodes, we conducted RNA sequencing of staminodes and stamen filaments from A. coerulea ‘Origami’ at two developmental stages: stage 11a, which precedes lateral staminode adhesion; and stage 13b, which is after lateral adhesion. Four biological replicates were collected for each of the four sample types. In order to identify genes that were specifically up- or downregulated in each of the four samples, we fit generalized linear models to the expression count data. BLASTx was used to identify the arabidopsis homologues of our differentially expressed genes. The GO enrichment analysis revealed three categories as enriched in both staminode samples: secondary cell wall biogenesis; xylan metabolic process; and lignin biosynthesis. Notable GO terms that were enriched in staminodes at stage 11a included monoterpene and suberin biosynthetic processes. Notable GO categories that were enriched in staminodes at stage 13b include lignin biosynthetic processes, defence responses to wounding and cell wall biosynthesis/organization (Supplementary data Fig. S4; Table S4). These results are all consistent with our observations regarding enriched lignification of the staminodes. Additionally, two GO categories enriched in staminodes at stage 13b were related to catabolic processes, and included genes with functions related to programmed cell death, which may implicate programmed cell death in the decrease in mesophyll by stage 13b. Although none of these GO terms suggests an immediate mechanism for the late developing lateral adhesion, we note that expression of both HOTHEAD and DEFECTIVE IN CUTICULAR RIDGES, two loci known to be involved in cuticle development and organ adhesion in arabidopsis (Krolikowski et al., 2003; Panikashvili et al, 2009), were enriched in stage 11a staminodes relative to stamen filaments (Supplementary data Table S6).

In order to identify key loci in these pathways, we sub-sampled only the transcription factors from our lists of differentially expressed genes. We were particularly interested in identifying transcription factors involved in patterning abaxial/adaxial polarity, which is known to control laminar expansion and has been implicated in other instances of differences between stamens and staminodes (De Almeida et al., 2014). We identified the abaxial identity transcription factor genes FILAMENTOUS FLOWER/YABBY1 (AqFIL) and CRABS CLAW (AqCRC) as upregulated in both staminode stages and stage 11a staminode samples, respectively, while the adaxial transcription factor gene REVOLUTA (AqREV) was enriched in stage 11a stamen filaments (Supplementary data Fig. S5). The two YABBY family members AqFIL and AqCRC were selected for further study.

The fully sequenced Aquilegia genome contains five members of the YABBY transcription factor family (Supplementary data Fig. S6). In situ hybridization studies were used to fully characterize the expression patterns of AqCRC and AqFIL. Limited expression data have been previously reported for AqCRC (Lee et al., 2005) and our study confirmed these findings across a wider range of developmental stages. Strong AqCRC expression is observed in the carpel primordia from their inception (Supplementary data Fig. S7A) and persists in these organs, especially their distal and abaxial regions, throughout development (Supplementary data Fig. S7). No clear differential expression of AqCRC was detected, however, between the developing stamens or staminodes. This reflects the fact that although 128 times higher AqCRC expression was detected in stage 11a staminodes, the overall read count for this gene was still very low. In contrast, AqFIL showed much broader expression in developing floral primordia (Fig. 6). Expression is first observed in the emerging sepal primordia (Fig. 6A, B) but is quickly detected in all initiating primordia and across the dome of the floral meristem (Fig. 6C–E). All floral organ primordia initially show abaxial AqFIL expression throughout their length (Fig. 6E). As the stamens differentiate into anthers and filaments, expression of AqFIL becomes restricted to the abaxial anther and is lost from the filament (Fig. 6F). In contrast, expression persists in the staminodes throughout their length until stage 10 (Fig. 6F–H), and afterwards it becomes localized to the abaxial epidermis of the basal portion of the staminode (Fig. 6H). Strong abaxial AqFIL expression was also observed in the carpels throughout their development (Fig. 6E–G).

Fig. 6.

In situ hybridization of AqFIL in early floral developmental stages of A. coerulea ‘Origami’. (A and B) Stage 3 floral meristems in which the sepals are just emerging. AqFIL is strongly expressed on the abaxial surface of the organs. (C and D) Stage 4–6 buds, as the petal and stamen primordia are initiating. AqFIL is expressed in each arising floral organ. (E) Stage 7 buds, when the carpel primordia (asterisks) appear adjacent to the staminode primordia (arrowheads). (F) Stage 8 buds, during which the floral organ primordia are differentiating. AqFIL expression is restricted to the distal region of the stamens (sta) but is seen throughout the length of the staminodes (arrowheads) as well as the carpels (car). (G) Stage 9 buds, with staminodes (arrowheads) that are clearly differentiated from the stamens (asterisks). AqFIL is still detectable along the abaxial epidermis of the staminodes. (H) Stage 10 floral bud with adjacent stamens and staminode (arrowhead). AqFIL is still detectable along the abaxial epidermis of the staminodes. (I) Stage 7 bud hybridized with a sense probe as a negative control. Arrowheads = staminodes; sta = stamens; pet = petals; asterisks = carpels. Scale bars: (A–D) = 100 µm; (E–I) = 200 µm.

DISCUSSION

Novel floral organ identities appear to most commonly evolve through modification of pre-existing genetic programmes, frequently from stamens (reviewed in Kramer, 2019). When their primordia first emerge, the elaborate staminodes of Aquilegia are essentially indistinguishable from those of stamens, but they rapidly diverge developmentally, most notably in their lateral expansion and the failure to differentiate anthers (Tucker and Hodges, 2005). The high similarity between the organ identity programmes of Aquilegia stamens and staminodes is consistent with the derivation of the latter from the former (Sharma and Kramer, 2013), but many aspects of staminode development and biology remain unknown. The current study has uncovered a series of developmental and molecular differences in staminodes relative to stamens that have both evolutionary and ecological implications.

Abaxial/adaxial differentiation

Previous studies have found that the initial developmental divergence of staminode and stamen primordia closely coincides with the establishment of differential expression of the AqAP3-1 and AqAP3-2 organ identity genes (Kramer et al., 2007). AqAP3-1 expression persists only in staminodes, while AqAP3-2 becomes concentrated in the stamens. Functional studies have demonstrated that targeted silencing of AqAP3-2 results in the loss of anther differentiation in stamens, which develop as naked, narrow filaments, while silencing of AqAP3-1 results in the loss of staminode identity as the organs are transformed into carpels (Sharma and Kramer, 2013). Full transformation of the stamens into carpels only occurs when both loci are targeted for silencing. These findings support a model in which both AqAP3-1 and -2 contribute to stamen identity, although AqAP3-2 is clearly the major player, but only AqAP3-1 is required for staminode identity. Several aspects of our new findings implicate differential expression and function of the abaxial/adaxial identity pathways as being regulated by the distinct AqAP3-1 and AqAP3-2 identity programmes.

In other model systems, laminar expansion has been shown to depend on balanced expression of the adaxial/abaxial genetic programmes such that loss of either identity produces narrow, radialized organs (Waites and Hudson, 1995; Eshed et al., 2001). Many studies of novel lateral organ morphologies, such as unifacial leaves or complex three-dimensional leaf forms, have found that they are produced by reorganization of abaxial and adaxial identity, to generate either radialization or novel axes of laminar expansion (reviewed by Fukushima and Hasebe, 2014). Perhaps the most common example of such reorganization is the typical angiosperm stamen, with its narrow basal filament and terminal tetrasporangiate anther (reviewed by Toriba et al., 2011). Detailed studies in rice have found that stamen primordia initially express abaxial and adaxial identity markers in their normal domains throughout the proximal/distal axis of the organ, but, as differentiation occurs, these domains are rapidly rearranged (Toriba et al., 2010). In the proximal filament, adaxial gene expression is lost such that the narrow structure is abaxialized. In contrast, the distal anther acquires 4-fold transverse symmetry with abaxial identity expressed in the medial domains of the anther while adaxial identity occupies the lateral margins, which creates four abaxial/adaxial interfaces along which the pollen sacs expand and differentiate. This aspect of stamen development is actually not as well characterized in A. thaliana, but expression studies of the abaxial factor gene FIL reveal a loss of abaxial identity in the filament (Siegfried et al., 1999), which would be consistent with radialization by adaxialization.

In Aquilegia, our RNA sequencing has identified higher relative expression of the adaxial identity gene AqREV in stamen filaments (Supplementary data Fig. S5), suggesting that they are effectively adaxialized. In the staminodes, it appears that the balance of abaxial/adaxial identity has been re-established via maintained expression of AqREV combined with increased relative expression of abaxial identity YABBY transcription factor family members, especially AqFIL. The primary effect of this difference would be predicted to be laminar expansion in the staminodes relative to the stamen filaments, which is in agreement with our observation of a longer period of cell division in the staminode primordia and a higher number of lateral cells in staminodes vs. stamen filaments (Fig. 3). These results in Aquilegia contrast with an independent case of laminar staminode evolution in the Zingiberales, in which the ancestral stamen filaments seem to be abaxialized via overexpression of YABBY genes and the derivation of the staminodes is associated with upregulation of adaxial factors (De Almeida et al., 2014; Tian et al., 2018). These two separate instances of the derivation of laminar staminodes both use similar development logic – radialized stamen filaments become laminar though re-establishment of a balance of abaxial/adaxial identity – but the details of the changes are different. It appears that the two monocots characterized to date have abaxialized stamen filaments (Toriba et al., 2010; De Almeida et al., 2014) but, in A. thaliana, FIL is downregulated in the stamen filament (Siegfried et al., 1999), suggesting adaxialization similar to Aquilegia. It will be interesting to see whether these differences in the molecular mechanism underlying stamen filament radialization are consistent among eudicots (adaxialized) and monocots (abaxialized). Such a finding could reflect the independent derivations of radial filaments in the two large clades. Alternatively, it may be that the mechanism underlying stamen filament radialization is highly labile, even among closely related taxa, due to developmental system drift (True and Haag, 2001).

Having established the broad lamina of the Aquilegia staminode, the adaxial/abaxial identity pathways further promote cell type differentiation between the two epidermal surfaces of the staminodes. First, a higher number of cells are observed across the lateral adaxial surface vs. the abaxial surface (Fig. 3; Supplementary data Fig. S3). Consistent with this, a previous study that examined Aquilegia floral epidermal cell types found that abaxial staminode cells tended to be wider than the highly elongated adaxial cells (Kramer et al., 2007). Secondly, the adaxial epidermal cells exhibit a distinct asymmetric pattern of lignification on their adaxial periclinal walls (Figs 4 and 5). Thus, the two staminode epidermal layers are distinguished in cell number, shape and terminal differentiation, differences that may contribute to the rugose surface of the organs and, ultimately, to their ecological function (see below). Overall, these patterns suggest that staminode morphology is driven by two major factors: (1) the absence of AqAP3-2 expression, which sterilizes the organs; and (2) the interaction of AqAP3-1 expression and abaxial/adaxial identity, which leads to lateral expansion and the differentiation of the two epidermal surfaces.

Lateral interactions and adhesion

Aquilegia staminodes have previously been treated as two identical whorls, but we now appreciate that there are subtle differences between the whorls due to consistent patterns of physical interactions along their lateral margins. Antesepalous staminodes tend to curl inward towards the carpels, while the antepetalous organs curl slightly outward towards the stamen whorl. This creates a physical interlocking along the lateral margins of the organs. It is still an open question as to whether the curling interactions are pre-programmed (i.e. potentially due to slight differences in identity) or if they are induced by physical interactions (or possibly some combination of both). On the one hand, the patterns we observed were highly consistent, potentially suggesting a pre-programmed difference, but the occasional exceptions (e.g. Fig. 3D) leave room for contributions from induced mechanisms. We attempted to remove individual staminodes to see if it affected the curling of adjacent organs, but several factors complicated the experiment. Perhaps most significantly, the curling occurs so early in development that it is very difficult to remove organs before it has initiated. Furthermore, when unfused staminodes are processed for any kind of imaging, they have a tendency to shrivel up, making accurate assessment of the curling phenotype impossible. A closely related feature is the physical adherence we observed at late developmental stages. Neither the histological studies nor the RNA sequencing results revealed an obvious mechanism for this adherence, and measuring its strength was confounded by the fragility of the organs themselves. Notably, two genes involved in cuticle biosynthesis, HOTHEAD and DEFECTIVE IN CUTICLE FORMATION, were enriched in stage 11a staminodes, implicating cuticle modification as a potential mechanism for adherence (Krolikowski et al., 2003; Yeats and Rose, 2013). Understanding the details of the curling and adherence phenomena will require further study, potentially using more detailed microscopy and transcriptomic approaches.

Evolutionary and ecological implications

It appears that inner sterilized stamens evolved in the ancestor of Aquilegia, Semiaquilegia and Urophysa, along with laminar expansion and disorganized marginal curling. However, the exact phylogenetic relationships between the three genera are not well supported (Bastida et al., 2010; Zhai et al., 2019), so it remains to be determined whether the trajectory of staminode evolution involved secondary reduction in Semiaquilegia or whether their sparse presence in this genus represents the ancestral condition. Regardless, it is clear that in the lineage leading to Aquilegia, the staminodes evolved consistent marginal curling patterns, lateral adhesion and asymmetric lignification of the adaxial epidermis (Fig. 5).

Do Aquilegia staminodes have an ecological function? The observed patterns of developmental elaboration are both relatively conserved within the genus and quite complex, making it seem unlikely that they evolved neutrally, but we cannot rule out this possibility. Although many functional staminodes in other angiosperm lineages play key roles in pollination, we do not believe that this is the case here. The staminodes are almost entirely hidden by the stamens during anthesis, and Aquilegia flowers are well supplied with other organs for pollinator attraction (Fig. 1). Moreover, by the time of anthesis, the innermost anthers extend well above the staminodes, making a role in preventing self-pollination improbable. By stage 14, all of the floral organs except the staminodes have fully abscised from the flower. The staminodes detach from the receptacle at their bases but remain encircling the carpels due to their lateral adherence until the expanding mature fruits finally swell and tear them apart. Based on this, it has been hypothesized that the role of staminodes in Aquilegia is to somehow protect the early developing fruits (Brayshaw, 1989). This would be consistent with their rugose abaxial surface, lateral adhesion and lignification patterns. In particular, the secondary compound lignin can be deployed for structural support as well as herbivore and/or pathogen resistance (Barcelo, 1997; Miedes et al., 2014). To our knowledge, the only system in which intracellular asymmetric lignin deposition has been well characterized is in Cardamine hirsuta, where the cell wall thickenings facilitate explosive seed dispersal (Hofhuis et al., 2016; Monniaux and Hay, 2016). Obviously, this ecological function is not related to our system, so the exact reason for the pattern remains a mystery. Overall, we believe that the combined traits of interlocking, adherent lateral margins and adaxial lignification, together with some evidence for enrichment in defence responses in the transcriptomic data, is consistent with a hypothesis that Aquilegia staminodes play a role in protecting the early development carpels and/or fruits, possibly from small herbivores or pathogen infection.

SUPPLEMENTARY DATA

Supplementary data are available online at https://academic.oup.com/aob and consist of the following. Figure S1: histological sections of Aquilegia coerulea ‘Origami’ stamens, staminodes and carpels. Figure S2: width of mature staminodes and stamen filaments at 50 % of length. Figure S3: boxplots showing the number of cells present in transverse transects of the abaxial or adaxial epidermal surfaces of staminodes and adjacent stamen filaments of Aquilegia, Semiaquliegia and Urophysa. Figure S4: heatmap showing relative expression levels of homologues of genes known to participate in the lignin biosynthesis pathway. Figure S5: heatmap showing relative expression levels of homologues of genes known to participate in abaxial/adaxial identity of lateral organs. Figure S6: relationships among subfamilies of YAB genes. Figure S7: in situ hybridization of AqCRC in wild-type floral meristems. Table S1: RNA sequencing sample read counts. Table S2: differentially expressed genes specific to each sample at LogFC 1. Table S3: dfferentially expressed genes identified when comparing staminode and stamen filament samples. Table S4: enriched Gene Ontology biological process categories. Table S5: PCR primers. Table S6: differentially expressed Aquilegia gene loci and arabidopsis homologues.

FUNDING

This work was supported by the Department of Organismic and Evolutionary Biology, Harvard University.